Adding upstream version 4.19.249.upstream/4.19.249

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

28 files changed, 16974 insertions, 0 deletions

diff --git a/drivers/gpu/drm/vc4/Kconfig b/drivers/gpu/drm/vc4/Kconfig
new file mode 100644
index 000000000..fdae18aea
--- /dev/null
+++ b/drivers/gpu/drm/vc4/Kconfig
@@ -0,0 +1,29 @@
+config DRM_VC4
+	tristate "Broadcom VC4 Graphics"
+	depends on ARCH_BCM || ARCH_BCM2835 || COMPILE_TEST
+	depends on DRM
+	depends on SND && SND_SOC
+	depends on COMMON_CLK
+	select DRM_KMS_HELPER
+	select DRM_KMS_CMA_HELPER
+	select DRM_GEM_CMA_HELPER
+	select DRM_PANEL_BRIDGE
+	select SND_PCM
+	select SND_PCM_ELD
+	select SND_SOC_GENERIC_DMAENGINE_PCM
+	select DRM_MIPI_DSI
+	help
+	  Choose this option if you have a system that has a Broadcom
+	  VC4 GPU, such as the Raspberry Pi or other BCM2708/BCM2835.
+
+	  This driver requires that "avoid_warnings=2" be present in
+	  the config.txt for the firmware, to keep it from smashing
+	  our display setup.
+
+config DRM_VC4_HDMI_CEC
+       bool "Broadcom VC4 HDMI CEC Support"
+       depends on DRM_VC4
+       select CEC_CORE
+       help
+	  Choose this option if you have a Broadcom VC4 GPU
+	  and want to use CEC.
diff --git a/drivers/gpu/drm/vc4/Makefile b/drivers/gpu/drm/vc4/Makefile
new file mode 100644
index 000000000..b303703bc
--- /dev/null
+++ b/drivers/gpu/drm/vc4/Makefile
@@ -0,0 +1,29 @@
+# SPDX-License-Identifier: GPL-2.0
+# Please keep these build lists sorted!
+
+# core driver code
+vc4-y := \
+	vc4_bo.o \
+	vc4_crtc.o \
+	vc4_drv.o \
+	vc4_dpi.o \
+	vc4_dsi.o \
+	vc4_fence.o \
+	vc4_kms.o \
+	vc4_gem.o \
+	vc4_hdmi.o \
+	vc4_vec.o \
+	vc4_hvs.o \
+	vc4_irq.o \
+	vc4_perfmon.o \
+	vc4_plane.o \
+	vc4_render_cl.o \
+	vc4_trace_points.o \
+	vc4_txp.o \
+	vc4_v3d.o \
+	vc4_validate.o \
+	vc4_validate_shaders.o
+
+vc4-$(CONFIG_DEBUG_FS) += vc4_debugfs.o
+
+obj-$(CONFIG_DRM_VC4)  += vc4.o
diff --git a/drivers/gpu/drm/vc4/vc4_bo.c b/drivers/gpu/drm/vc4/vc4_bo.c
new file mode 100644
index 000000000..1e28ff981
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_bo.c
@@ -0,0 +1,1107 @@
+/*
+ *  Copyright © 2015 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+/**
+ * DOC: VC4 GEM BO management support
+ *
+ * The VC4 GPU architecture (both scanout and rendering) has direct
+ * access to system memory with no MMU in between.  To support it, we
+ * use the GEM CMA helper functions to allocate contiguous ranges of
+ * physical memory for our BOs.
+ *
+ * Since the CMA allocator is very slow, we keep a cache of recently
+ * freed BOs around so that the kernel's allocation of objects for 3D
+ * rendering can return quickly.
+ */
+
+#include <linux/dma-buf.h>
+
+#include "vc4_drv.h"
+#include "uapi/drm/vc4_drm.h"
+
+static const char * const bo_type_names[] = {
+	"kernel",
+	"V3D",
+	"V3D shader",
+	"dumb",
+	"binner",
+	"RCL",
+	"BCL",
+	"kernel BO cache",
+};
+
+static bool is_user_label(int label)
+{
+	return label >= VC4_BO_TYPE_COUNT;
+}
+
+static void vc4_bo_stats_dump(struct vc4_dev *vc4)
+{
+	int i;
+
+	for (i = 0; i < vc4->num_labels; i++) {
+		if (!vc4->bo_labels[i].num_allocated)
+			continue;
+
+		DRM_INFO("%30s: %6dkb BOs (%d)\n",
+			 vc4->bo_labels[i].name,
+			 vc4->bo_labels[i].size_allocated / 1024,
+			 vc4->bo_labels[i].num_allocated);
+	}
+
+	mutex_lock(&vc4->purgeable.lock);
+	if (vc4->purgeable.num)
+		DRM_INFO("%30s: %6zdkb BOs (%d)\n", "userspace BO cache",
+			 vc4->purgeable.size / 1024, vc4->purgeable.num);
+
+	if (vc4->purgeable.purged_num)
+		DRM_INFO("%30s: %6zdkb BOs (%d)\n", "total purged BO",
+			 vc4->purgeable.purged_size / 1024,
+			 vc4->purgeable.purged_num);
+	mutex_unlock(&vc4->purgeable.lock);
+}
+
+#ifdef CONFIG_DEBUG_FS
+int vc4_bo_stats_debugfs(struct seq_file *m, void *unused)
+{
+	struct drm_info_node *node = (struct drm_info_node *)m->private;
+	struct drm_device *dev = node->minor->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int i;
+
+	mutex_lock(&vc4->bo_lock);
+	for (i = 0; i < vc4->num_labels; i++) {
+		if (!vc4->bo_labels[i].num_allocated)
+			continue;
+
+		seq_printf(m, "%30s: %6dkb BOs (%d)\n",
+			   vc4->bo_labels[i].name,
+			   vc4->bo_labels[i].size_allocated / 1024,
+			   vc4->bo_labels[i].num_allocated);
+	}
+	mutex_unlock(&vc4->bo_lock);
+
+	mutex_lock(&vc4->purgeable.lock);
+	if (vc4->purgeable.num)
+		seq_printf(m, "%30s: %6zdkb BOs (%d)\n", "userspace BO cache",
+			   vc4->purgeable.size / 1024, vc4->purgeable.num);
+
+	if (vc4->purgeable.purged_num)
+		seq_printf(m, "%30s: %6zdkb BOs (%d)\n", "total purged BO",
+			   vc4->purgeable.purged_size / 1024,
+			   vc4->purgeable.purged_num);
+	mutex_unlock(&vc4->purgeable.lock);
+
+	return 0;
+}
+#endif
+
+/* Takes ownership of *name and returns the appropriate slot for it in
+ * the bo_labels[] array, extending it as necessary.
+ *
+ * This is inefficient and could use a hash table instead of walking
+ * an array and strcmp()ing.  However, the assumption is that user
+ * labeling will be infrequent (scanout buffers and other long-lived
+ * objects, or debug driver builds), so we can live with it for now.
+ */
+static int vc4_get_user_label(struct vc4_dev *vc4, const char *name)
+{
+	int i;
+	int free_slot = -1;
+
+	for (i = 0; i < vc4->num_labels; i++) {
+		if (!vc4->bo_labels[i].name) {
+			free_slot = i;
+		} else if (strcmp(vc4->bo_labels[i].name, name) == 0) {
+			kfree(name);
+			return i;
+		}
+	}
+
+	if (free_slot != -1) {
+		WARN_ON(vc4->bo_labels[free_slot].num_allocated != 0);
+		vc4->bo_labels[free_slot].name = name;
+		return free_slot;
+	} else {
+		u32 new_label_count = vc4->num_labels + 1;
+		struct vc4_label *new_labels =
+			krealloc(vc4->bo_labels,
+				 new_label_count * sizeof(*new_labels),
+				 GFP_KERNEL);
+
+		if (!new_labels) {
+			kfree(name);
+			return -1;
+		}
+
+		free_slot = vc4->num_labels;
+		vc4->bo_labels = new_labels;
+		vc4->num_labels = new_label_count;
+
+		vc4->bo_labels[free_slot].name = name;
+		vc4->bo_labels[free_slot].num_allocated = 0;
+		vc4->bo_labels[free_slot].size_allocated = 0;
+
+		return free_slot;
+	}
+}
+
+static void vc4_bo_set_label(struct drm_gem_object *gem_obj, int label)
+{
+	struct vc4_bo *bo = to_vc4_bo(gem_obj);
+	struct vc4_dev *vc4 = to_vc4_dev(gem_obj->dev);
+
+	lockdep_assert_held(&vc4->bo_lock);
+
+	if (label != -1) {
+		vc4->bo_labels[label].num_allocated++;
+		vc4->bo_labels[label].size_allocated += gem_obj->size;
+	}
+
+	vc4->bo_labels[bo->label].num_allocated--;
+	vc4->bo_labels[bo->label].size_allocated -= gem_obj->size;
+
+	if (vc4->bo_labels[bo->label].num_allocated == 0 &&
+	    is_user_label(bo->label)) {
+		/* Free user BO label slots on last unreference.
+		 * Slots are just where we track the stats for a given
+		 * name, and once a name is unused we can reuse that
+		 * slot.
+		 */
+		kfree(vc4->bo_labels[bo->label].name);
+		vc4->bo_labels[bo->label].name = NULL;
+	}
+
+	bo->label = label;
+}
+
+static uint32_t bo_page_index(size_t size)
+{
+	return (size / PAGE_SIZE) - 1;
+}
+
+static void vc4_bo_destroy(struct vc4_bo *bo)
+{
+	struct drm_gem_object *obj = &bo->base.base;
+	struct vc4_dev *vc4 = to_vc4_dev(obj->dev);
+
+	lockdep_assert_held(&vc4->bo_lock);
+
+	vc4_bo_set_label(obj, -1);
+
+	if (bo->validated_shader) {
+		kfree(bo->validated_shader->uniform_addr_offsets);
+		kfree(bo->validated_shader->texture_samples);
+		kfree(bo->validated_shader);
+		bo->validated_shader = NULL;
+	}
+
+	reservation_object_fini(&bo->_resv);
+
+	drm_gem_cma_free_object(obj);
+}
+
+static void vc4_bo_remove_from_cache(struct vc4_bo *bo)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(bo->base.base.dev);
+
+	lockdep_assert_held(&vc4->bo_lock);
+	list_del(&bo->unref_head);
+	list_del(&bo->size_head);
+}
+
+static struct list_head *vc4_get_cache_list_for_size(struct drm_device *dev,
+						     size_t size)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	uint32_t page_index = bo_page_index(size);
+
+	if (vc4->bo_cache.size_list_size <= page_index) {
+		uint32_t new_size = max(vc4->bo_cache.size_list_size * 2,
+					page_index + 1);
+		struct list_head *new_list;
+		uint32_t i;
+
+		new_list = kmalloc_array(new_size, sizeof(struct list_head),
+					 GFP_KERNEL);
+		if (!new_list)
+			return NULL;
+
+		/* Rebase the old cached BO lists to their new list
+		 * head locations.
+		 */
+		for (i = 0; i < vc4->bo_cache.size_list_size; i++) {
+			struct list_head *old_list =
+				&vc4->bo_cache.size_list[i];
+
+			if (list_empty(old_list))
+				INIT_LIST_HEAD(&new_list[i]);
+			else
+				list_replace(old_list, &new_list[i]);
+		}
+		/* And initialize the brand new BO list heads. */
+		for (i = vc4->bo_cache.size_list_size; i < new_size; i++)
+			INIT_LIST_HEAD(&new_list[i]);
+
+		kfree(vc4->bo_cache.size_list);
+		vc4->bo_cache.size_list = new_list;
+		vc4->bo_cache.size_list_size = new_size;
+	}
+
+	return &vc4->bo_cache.size_list[page_index];
+}
+
+static void vc4_bo_cache_purge(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	mutex_lock(&vc4->bo_lock);
+	while (!list_empty(&vc4->bo_cache.time_list)) {
+		struct vc4_bo *bo = list_last_entry(&vc4->bo_cache.time_list,
+						    struct vc4_bo, unref_head);
+		vc4_bo_remove_from_cache(bo);
+		vc4_bo_destroy(bo);
+	}
+	mutex_unlock(&vc4->bo_lock);
+}
+
+void vc4_bo_add_to_purgeable_pool(struct vc4_bo *bo)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(bo->base.base.dev);
+
+	mutex_lock(&vc4->purgeable.lock);
+	list_add_tail(&bo->size_head, &vc4->purgeable.list);
+	vc4->purgeable.num++;
+	vc4->purgeable.size += bo->base.base.size;
+	mutex_unlock(&vc4->purgeable.lock);
+}
+
+static void vc4_bo_remove_from_purgeable_pool_locked(struct vc4_bo *bo)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(bo->base.base.dev);
+
+	/* list_del_init() is used here because the caller might release
+	 * the purgeable lock in order to acquire the madv one and update the
+	 * madv status.
+	 * During this short period of time a user might decide to mark
+	 * the BO as unpurgeable, and if bo->madv is set to
+	 * VC4_MADV_DONTNEED it will try to remove the BO from the
+	 * purgeable list which will fail if the ->next/prev fields
+	 * are set to LIST_POISON1/LIST_POISON2 (which is what
+	 * list_del() does).
+	 * Re-initializing the list element guarantees that list_del()
+	 * will work correctly even if it's a NOP.
+	 */
+	list_del_init(&bo->size_head);
+	vc4->purgeable.num--;
+	vc4->purgeable.size -= bo->base.base.size;
+}
+
+void vc4_bo_remove_from_purgeable_pool(struct vc4_bo *bo)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(bo->base.base.dev);
+
+	mutex_lock(&vc4->purgeable.lock);
+	vc4_bo_remove_from_purgeable_pool_locked(bo);
+	mutex_unlock(&vc4->purgeable.lock);
+}
+
+static void vc4_bo_purge(struct drm_gem_object *obj)
+{
+	struct vc4_bo *bo = to_vc4_bo(obj);
+	struct drm_device *dev = obj->dev;
+
+	WARN_ON(!mutex_is_locked(&bo->madv_lock));
+	WARN_ON(bo->madv != VC4_MADV_DONTNEED);
+
+	drm_vma_node_unmap(&obj->vma_node, dev->anon_inode->i_mapping);
+
+	dma_free_wc(dev->dev, obj->size, bo->base.vaddr, bo->base.paddr);
+	bo->base.vaddr = NULL;
+	bo->madv = __VC4_MADV_PURGED;
+}
+
+static void vc4_bo_userspace_cache_purge(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	mutex_lock(&vc4->purgeable.lock);
+	while (!list_empty(&vc4->purgeable.list)) {
+		struct vc4_bo *bo = list_first_entry(&vc4->purgeable.list,
+						     struct vc4_bo, size_head);
+		struct drm_gem_object *obj = &bo->base.base;
+		size_t purged_size = 0;
+
+		vc4_bo_remove_from_purgeable_pool_locked(bo);
+
+		/* Release the purgeable lock while we're purging the BO so
+		 * that other people can continue inserting things in the
+		 * purgeable pool without having to wait for all BOs to be
+		 * purged.
+		 */
+		mutex_unlock(&vc4->purgeable.lock);
+		mutex_lock(&bo->madv_lock);
+
+		/* Since we released the purgeable pool lock before acquiring
+		 * the BO madv one, the user may have marked the BO as WILLNEED
+		 * and re-used it in the meantime.
+		 * Before purging the BO we need to make sure
+		 * - it is still marked as DONTNEED
+		 * - it has not been re-inserted in the purgeable list
+		 * - it is not used by HW blocks
+		 * If one of these conditions is not met, just skip the entry.
+		 */
+		if (bo->madv == VC4_MADV_DONTNEED &&
+		    list_empty(&bo->size_head) &&
+		    !refcount_read(&bo->usecnt)) {
+			purged_size = bo->base.base.size;
+			vc4_bo_purge(obj);
+		}
+		mutex_unlock(&bo->madv_lock);
+		mutex_lock(&vc4->purgeable.lock);
+
+		if (purged_size) {
+			vc4->purgeable.purged_size += purged_size;
+			vc4->purgeable.purged_num++;
+		}
+	}
+	mutex_unlock(&vc4->purgeable.lock);
+}
+
+static struct vc4_bo *vc4_bo_get_from_cache(struct drm_device *dev,
+					    uint32_t size,
+					    enum vc4_kernel_bo_type type)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	uint32_t page_index = bo_page_index(size);
+	struct vc4_bo *bo = NULL;
+
+	size = roundup(size, PAGE_SIZE);
+
+	mutex_lock(&vc4->bo_lock);
+	if (page_index >= vc4->bo_cache.size_list_size)
+		goto out;
+
+	if (list_empty(&vc4->bo_cache.size_list[page_index]))
+		goto out;
+
+	bo = list_first_entry(&vc4->bo_cache.size_list[page_index],
+			      struct vc4_bo, size_head);
+	vc4_bo_remove_from_cache(bo);
+	kref_init(&bo->base.base.refcount);
+
+out:
+	if (bo)
+		vc4_bo_set_label(&bo->base.base, type);
+	mutex_unlock(&vc4->bo_lock);
+	return bo;
+}
+
+/**
+ * vc4_gem_create_object - Implementation of driver->gem_create_object.
+ * @dev: DRM device
+ * @size: Size in bytes of the memory the object will reference
+ *
+ * This lets the CMA helpers allocate object structs for us, and keep
+ * our BO stats correct.
+ */
+struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_bo *bo;
+
+	bo = kzalloc(sizeof(*bo), GFP_KERNEL);
+	if (!bo)
+		return NULL;
+
+	bo->madv = VC4_MADV_WILLNEED;
+	refcount_set(&bo->usecnt, 0);
+	mutex_init(&bo->madv_lock);
+	mutex_lock(&vc4->bo_lock);
+	bo->label = VC4_BO_TYPE_KERNEL;
+	vc4->bo_labels[VC4_BO_TYPE_KERNEL].num_allocated++;
+	vc4->bo_labels[VC4_BO_TYPE_KERNEL].size_allocated += size;
+	mutex_unlock(&vc4->bo_lock);
+	bo->resv = &bo->_resv;
+	reservation_object_init(bo->resv);
+
+	return &bo->base.base;
+}
+
+struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t unaligned_size,
+			     bool allow_unzeroed, enum vc4_kernel_bo_type type)
+{
+	size_t size = roundup(unaligned_size, PAGE_SIZE);
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct drm_gem_cma_object *cma_obj;
+	struct vc4_bo *bo;
+
+	if (size == 0)
+		return ERR_PTR(-EINVAL);
+
+	/* First, try to get a vc4_bo from the kernel BO cache. */
+	bo = vc4_bo_get_from_cache(dev, size, type);
+	if (bo) {
+		if (!allow_unzeroed)
+			memset(bo->base.vaddr, 0, bo->base.base.size);
+		return bo;
+	}
+
+	cma_obj = drm_gem_cma_create(dev, size);
+	if (IS_ERR(cma_obj)) {
+		/*
+		 * If we've run out of CMA memory, kill the cache of
+		 * CMA allocations we've got laying around and try again.
+		 */
+		vc4_bo_cache_purge(dev);
+		cma_obj = drm_gem_cma_create(dev, size);
+	}
+
+	if (IS_ERR(cma_obj)) {
+		/*
+		 * Still not enough CMA memory, purge the userspace BO
+		 * cache and retry.
+		 * This is sub-optimal since we purge the whole userspace
+		 * BO cache which forces user that want to re-use the BO to
+		 * restore its initial content.
+		 * Ideally, we should purge entries one by one and retry
+		 * after each to see if CMA allocation succeeds. Or even
+		 * better, try to find an entry with at least the same
+		 * size.
+		 */
+		vc4_bo_userspace_cache_purge(dev);
+		cma_obj = drm_gem_cma_create(dev, size);
+	}
+
+	if (IS_ERR(cma_obj)) {
+		DRM_ERROR("Failed to allocate from CMA:\n");
+		vc4_bo_stats_dump(vc4);
+		return ERR_PTR(-ENOMEM);
+	}
+	bo = to_vc4_bo(&cma_obj->base);
+
+	/* By default, BOs do not support the MADV ioctl. This will be enabled
+	 * only on BOs that are exposed to userspace (V3D, V3D_SHADER and DUMB
+	 * BOs).
+	 */
+	bo->madv = __VC4_MADV_NOTSUPP;
+
+	mutex_lock(&vc4->bo_lock);
+	vc4_bo_set_label(&cma_obj->base, type);
+	mutex_unlock(&vc4->bo_lock);
+
+	return bo;
+}
+
+int vc4_dumb_create(struct drm_file *file_priv,
+		    struct drm_device *dev,
+		    struct drm_mode_create_dumb *args)
+{
+	int min_pitch = DIV_ROUND_UP(args->width * args->bpp, 8);
+	struct vc4_bo *bo = NULL;
+	int ret;
+
+	if (args->pitch < min_pitch)
+		args->pitch = min_pitch;
+
+	if (args->size < args->pitch * args->height)
+		args->size = args->pitch * args->height;
+
+	bo = vc4_bo_create(dev, args->size, false, VC4_BO_TYPE_DUMB);
+	if (IS_ERR(bo))
+		return PTR_ERR(bo);
+
+	bo->madv = VC4_MADV_WILLNEED;
+
+	ret = drm_gem_handle_create(file_priv, &bo->base.base, &args->handle);
+	drm_gem_object_put_unlocked(&bo->base.base);
+
+	return ret;
+}
+
+static void vc4_bo_cache_free_old(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	unsigned long expire_time = jiffies - msecs_to_jiffies(1000);
+
+	lockdep_assert_held(&vc4->bo_lock);
+
+	while (!list_empty(&vc4->bo_cache.time_list)) {
+		struct vc4_bo *bo = list_last_entry(&vc4->bo_cache.time_list,
+						    struct vc4_bo, unref_head);
+		if (time_before(expire_time, bo->free_time)) {
+			mod_timer(&vc4->bo_cache.time_timer,
+				  round_jiffies_up(jiffies +
+						   msecs_to_jiffies(1000)));
+			return;
+		}
+
+		vc4_bo_remove_from_cache(bo);
+		vc4_bo_destroy(bo);
+	}
+}
+
+/* Called on the last userspace/kernel unreference of the BO.  Returns
+ * it to the BO cache if possible, otherwise frees it.
+ */
+void vc4_free_object(struct drm_gem_object *gem_bo)
+{
+	struct drm_device *dev = gem_bo->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_bo *bo = to_vc4_bo(gem_bo);
+	struct list_head *cache_list;
+
+	/* Remove the BO from the purgeable list. */
+	mutex_lock(&bo->madv_lock);
+	if (bo->madv == VC4_MADV_DONTNEED && !refcount_read(&bo->usecnt))
+		vc4_bo_remove_from_purgeable_pool(bo);
+	mutex_unlock(&bo->madv_lock);
+
+	mutex_lock(&vc4->bo_lock);
+	/* If the object references someone else's memory, we can't cache it.
+	 */
+	if (gem_bo->import_attach) {
+		vc4_bo_destroy(bo);
+		goto out;
+	}
+
+	/* Don't cache if it was publicly named. */
+	if (gem_bo->name) {
+		vc4_bo_destroy(bo);
+		goto out;
+	}
+
+	/* If this object was partially constructed but CMA allocation
+	 * had failed, just free it. Can also happen when the BO has been
+	 * purged.
+	 */
+	if (!bo->base.vaddr) {
+		vc4_bo_destroy(bo);
+		goto out;
+	}
+
+	cache_list = vc4_get_cache_list_for_size(dev, gem_bo->size);
+	if (!cache_list) {
+		vc4_bo_destroy(bo);
+		goto out;
+	}
+
+	if (bo->validated_shader) {
+		kfree(bo->validated_shader->uniform_addr_offsets);
+		kfree(bo->validated_shader->texture_samples);
+		kfree(bo->validated_shader);
+		bo->validated_shader = NULL;
+	}
+
+	/* Reset madv and usecnt before adding the BO to the cache. */
+	bo->madv = __VC4_MADV_NOTSUPP;
+	refcount_set(&bo->usecnt, 0);
+
+	bo->t_format = false;
+	bo->free_time = jiffies;
+	list_add(&bo->size_head, cache_list);
+	list_add(&bo->unref_head, &vc4->bo_cache.time_list);
+
+	vc4_bo_set_label(&bo->base.base, VC4_BO_TYPE_KERNEL_CACHE);
+
+	vc4_bo_cache_free_old(dev);
+
+out:
+	mutex_unlock(&vc4->bo_lock);
+}
+
+static void vc4_bo_cache_time_work(struct work_struct *work)
+{
+	struct vc4_dev *vc4 =
+		container_of(work, struct vc4_dev, bo_cache.time_work);
+	struct drm_device *dev = vc4->dev;
+
+	mutex_lock(&vc4->bo_lock);
+	vc4_bo_cache_free_old(dev);
+	mutex_unlock(&vc4->bo_lock);
+}
+
+int vc4_bo_inc_usecnt(struct vc4_bo *bo)
+{
+	int ret;
+
+	/* Fast path: if the BO is already retained by someone, no need to
+	 * check the madv status.
+	 */
+	if (refcount_inc_not_zero(&bo->usecnt))
+		return 0;
+
+	mutex_lock(&bo->madv_lock);
+	switch (bo->madv) {
+	case VC4_MADV_WILLNEED:
+		if (!refcount_inc_not_zero(&bo->usecnt))
+			refcount_set(&bo->usecnt, 1);
+		ret = 0;
+		break;
+	case VC4_MADV_DONTNEED:
+		/* We shouldn't use a BO marked as purgeable if at least
+		 * someone else retained its content by incrementing usecnt.
+		 * Luckily the BO hasn't been purged yet, but something wrong
+		 * is happening here. Just throw an error instead of
+		 * authorizing this use case.
+		 */
+	case __VC4_MADV_PURGED:
+		/* We can't use a purged BO. */
+	default:
+		/* Invalid madv value. */
+		ret = -EINVAL;
+		break;
+	}
+	mutex_unlock(&bo->madv_lock);
+
+	return ret;
+}
+
+void vc4_bo_dec_usecnt(struct vc4_bo *bo)
+{
+	/* Fast path: if the BO is still retained by someone, no need to test
+	 * the madv value.
+	 */
+	if (refcount_dec_not_one(&bo->usecnt))
+		return;
+
+	mutex_lock(&bo->madv_lock);
+	if (refcount_dec_and_test(&bo->usecnt) &&
+	    bo->madv == VC4_MADV_DONTNEED)
+		vc4_bo_add_to_purgeable_pool(bo);
+	mutex_unlock(&bo->madv_lock);
+}
+
+static void vc4_bo_cache_time_timer(struct timer_list *t)
+{
+	struct vc4_dev *vc4 = from_timer(vc4, t, bo_cache.time_timer);
+
+	schedule_work(&vc4->bo_cache.time_work);
+}
+
+struct reservation_object *vc4_prime_res_obj(struct drm_gem_object *obj)
+{
+	struct vc4_bo *bo = to_vc4_bo(obj);
+
+	return bo->resv;
+}
+
+struct dma_buf *
+vc4_prime_export(struct drm_device *dev, struct drm_gem_object *obj, int flags)
+{
+	struct vc4_bo *bo = to_vc4_bo(obj);
+	struct dma_buf *dmabuf;
+	int ret;
+
+	if (bo->validated_shader) {
+		DRM_DEBUG("Attempting to export shader BO\n");
+		return ERR_PTR(-EINVAL);
+	}
+
+	/* Note: as soon as the BO is exported it becomes unpurgeable, because
+	 * noone ever decrements the usecnt even if the reference held by the
+	 * exported BO is released. This shouldn't be a problem since we don't
+	 * expect exported BOs to be marked as purgeable.
+	 */
+	ret = vc4_bo_inc_usecnt(bo);
+	if (ret) {
+		DRM_ERROR("Failed to increment BO usecnt\n");
+		return ERR_PTR(ret);
+	}
+
+	dmabuf = drm_gem_prime_export(dev, obj, flags);
+	if (IS_ERR(dmabuf))
+		vc4_bo_dec_usecnt(bo);
+
+	return dmabuf;
+}
+
+vm_fault_t vc4_fault(struct vm_fault *vmf)
+{
+	struct vm_area_struct *vma = vmf->vma;
+	struct drm_gem_object *obj = vma->vm_private_data;
+	struct vc4_bo *bo = to_vc4_bo(obj);
+
+	/* The only reason we would end up here is when user-space accesses
+	 * BO's memory after it's been purged.
+	 */
+	mutex_lock(&bo->madv_lock);
+	WARN_ON(bo->madv != __VC4_MADV_PURGED);
+	mutex_unlock(&bo->madv_lock);
+
+	return VM_FAULT_SIGBUS;
+}
+
+int vc4_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+	struct drm_gem_object *gem_obj;
+	unsigned long vm_pgoff;
+	struct vc4_bo *bo;
+	int ret;
+
+	ret = drm_gem_mmap(filp, vma);
+	if (ret)
+		return ret;
+
+	gem_obj = vma->vm_private_data;
+	bo = to_vc4_bo(gem_obj);
+
+	if (bo->validated_shader && (vma->vm_flags & VM_WRITE)) {
+		DRM_DEBUG("mmaping of shader BOs for writing not allowed.\n");
+		return -EINVAL;
+	}
+
+	if (bo->madv != VC4_MADV_WILLNEED) {
+		DRM_DEBUG("mmaping of %s BO not allowed\n",
+			  bo->madv == VC4_MADV_DONTNEED ?
+			  "purgeable" : "purged");
+		return -EINVAL;
+	}
+
+	/*
+	 * Clear the VM_PFNMAP flag that was set by drm_gem_mmap(), and set the
+	 * vm_pgoff (used as a fake buffer offset by DRM) to 0 as we want to map
+	 * the whole buffer.
+	 */
+	vma->vm_flags &= ~VM_PFNMAP;
+
+	/* This ->vm_pgoff dance is needed to make all parties happy:
+	 * - dma_mmap_wc() uses ->vm_pgoff as an offset within the allocated
+	 *   mem-region, hence the need to set it to zero (the value set by
+	 *   the DRM core is a virtual offset encoding the GEM object-id)
+	 * - the mmap() core logic needs ->vm_pgoff to be restored to its
+	 *   initial value before returning from this function because it
+	 *   encodes the  offset of this GEM in the dev->anon_inode pseudo-file
+	 *   and this information will be used when we invalidate userspace
+	 *   mappings  with drm_vma_node_unmap() (called from vc4_gem_purge()).
+	 */
+	vm_pgoff = vma->vm_pgoff;
+	vma->vm_pgoff = 0;
+	ret = dma_mmap_wc(bo->base.base.dev->dev, vma, bo->base.vaddr,
+			  bo->base.paddr, vma->vm_end - vma->vm_start);
+	vma->vm_pgoff = vm_pgoff;
+
+	if (ret)
+		drm_gem_vm_close(vma);
+
+	return ret;
+}
+
+int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma)
+{
+	struct vc4_bo *bo = to_vc4_bo(obj);
+
+	if (bo->validated_shader && (vma->vm_flags & VM_WRITE)) {
+		DRM_DEBUG("mmaping of shader BOs for writing not allowed.\n");
+		return -EINVAL;
+	}
+
+	return drm_gem_cma_prime_mmap(obj, vma);
+}
+
+void *vc4_prime_vmap(struct drm_gem_object *obj)
+{
+	struct vc4_bo *bo = to_vc4_bo(obj);
+
+	if (bo->validated_shader) {
+		DRM_DEBUG("mmaping of shader BOs not allowed.\n");
+		return ERR_PTR(-EINVAL);
+	}
+
+	return drm_gem_cma_prime_vmap(obj);
+}
+
+struct drm_gem_object *
+vc4_prime_import_sg_table(struct drm_device *dev,
+			  struct dma_buf_attachment *attach,
+			  struct sg_table *sgt)
+{
+	struct drm_gem_object *obj;
+	struct vc4_bo *bo;
+
+	obj = drm_gem_cma_prime_import_sg_table(dev, attach, sgt);
+	if (IS_ERR(obj))
+		return obj;
+
+	bo = to_vc4_bo(obj);
+	bo->resv = attach->dmabuf->resv;
+
+	return obj;
+}
+
+int vc4_create_bo_ioctl(struct drm_device *dev, void *data,
+			struct drm_file *file_priv)
+{
+	struct drm_vc4_create_bo *args = data;
+	struct vc4_bo *bo = NULL;
+	int ret;
+
+	/*
+	 * We can't allocate from the BO cache, because the BOs don't
+	 * get zeroed, and that might leak data between users.
+	 */
+	bo = vc4_bo_create(dev, args->size, false, VC4_BO_TYPE_V3D);
+	if (IS_ERR(bo))
+		return PTR_ERR(bo);
+
+	bo->madv = VC4_MADV_WILLNEED;
+
+	ret = drm_gem_handle_create(file_priv, &bo->base.base, &args->handle);
+	drm_gem_object_put_unlocked(&bo->base.base);
+
+	return ret;
+}
+
+int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data,
+		      struct drm_file *file_priv)
+{
+	struct drm_vc4_mmap_bo *args = data;
+	struct drm_gem_object *gem_obj;
+
+	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
+	if (!gem_obj) {
+		DRM_DEBUG("Failed to look up GEM BO %d\n", args->handle);
+		return -EINVAL;
+	}
+
+	/* The mmap offset was set up at BO allocation time. */
+	args->offset = drm_vma_node_offset_addr(&gem_obj->vma_node);
+
+	drm_gem_object_put_unlocked(gem_obj);
+	return 0;
+}
+
+int
+vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data,
+			   struct drm_file *file_priv)
+{
+	struct drm_vc4_create_shader_bo *args = data;
+	struct vc4_bo *bo = NULL;
+	int ret;
+
+	if (args->size == 0)
+		return -EINVAL;
+
+	if (args->size % sizeof(u64) != 0)
+		return -EINVAL;
+
+	if (args->flags != 0) {
+		DRM_INFO("Unknown flags set: 0x%08x\n", args->flags);
+		return -EINVAL;
+	}
+
+	if (args->pad != 0) {
+		DRM_INFO("Pad set: 0x%08x\n", args->pad);
+		return -EINVAL;
+	}
+
+	bo = vc4_bo_create(dev, args->size, true, VC4_BO_TYPE_V3D_SHADER);
+	if (IS_ERR(bo))
+		return PTR_ERR(bo);
+
+	bo->madv = VC4_MADV_WILLNEED;
+
+	if (copy_from_user(bo->base.vaddr,
+			     (void __user *)(uintptr_t)args->data,
+			     args->size)) {
+		ret = -EFAULT;
+		goto fail;
+	}
+	/* Clear the rest of the memory from allocating from the BO
+	 * cache.
+	 */
+	memset(bo->base.vaddr + args->size, 0,
+	       bo->base.base.size - args->size);
+
+	bo->validated_shader = vc4_validate_shader(&bo->base);
+	if (!bo->validated_shader) {
+		ret = -EINVAL;
+		goto fail;
+	}
+
+	/* We have to create the handle after validation, to avoid
+	 * races for users to do doing things like mmap the shader BO.
+	 */
+	ret = drm_gem_handle_create(file_priv, &bo->base.base, &args->handle);
+
+ fail:
+	drm_gem_object_put_unlocked(&bo->base.base);
+
+	return ret;
+}
+
+/**
+ * vc4_set_tiling_ioctl() - Sets the tiling modifier for a BO.
+ * @dev: DRM device
+ * @data: ioctl argument
+ * @file_priv: DRM file for this fd
+ *
+ * The tiling state of the BO decides the default modifier of an fb if
+ * no specific modifier was set by userspace, and the return value of
+ * vc4_get_tiling_ioctl() (so that userspace can treat a BO it
+ * received from dmabuf as the same tiling format as the producer
+ * used).
+ */
+int vc4_set_tiling_ioctl(struct drm_device *dev, void *data,
+			 struct drm_file *file_priv)
+{
+	struct drm_vc4_set_tiling *args = data;
+	struct drm_gem_object *gem_obj;
+	struct vc4_bo *bo;
+	bool t_format;
+
+	if (args->flags != 0)
+		return -EINVAL;
+
+	switch (args->modifier) {
+	case DRM_FORMAT_MOD_NONE:
+		t_format = false;
+		break;
+	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
+		t_format = true;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
+	if (!gem_obj) {
+		DRM_DEBUG("Failed to look up GEM BO %d\n", args->handle);
+		return -ENOENT;
+	}
+	bo = to_vc4_bo(gem_obj);
+	bo->t_format = t_format;
+
+	drm_gem_object_put_unlocked(gem_obj);
+
+	return 0;
+}
+
+/**
+ * vc4_get_tiling_ioctl() - Gets the tiling modifier for a BO.
+ * @dev: DRM device
+ * @data: ioctl argument
+ * @file_priv: DRM file for this fd
+ *
+ * Returns the tiling modifier for a BO as set by vc4_set_tiling_ioctl().
+ */
+int vc4_get_tiling_ioctl(struct drm_device *dev, void *data,
+			 struct drm_file *file_priv)
+{
+	struct drm_vc4_get_tiling *args = data;
+	struct drm_gem_object *gem_obj;
+	struct vc4_bo *bo;
+
+	if (args->flags != 0 || args->modifier != 0)
+		return -EINVAL;
+
+	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
+	if (!gem_obj) {
+		DRM_DEBUG("Failed to look up GEM BO %d\n", args->handle);
+		return -ENOENT;
+	}
+	bo = to_vc4_bo(gem_obj);
+
+	if (bo->t_format)
+		args->modifier = DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED;
+	else
+		args->modifier = DRM_FORMAT_MOD_NONE;
+
+	drm_gem_object_put_unlocked(gem_obj);
+
+	return 0;
+}
+
+int vc4_bo_cache_init(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int i;
+
+	/* Create the initial set of BO labels that the kernel will
+	 * use.  This lets us avoid a bunch of string reallocation in
+	 * the kernel's draw and BO allocation paths.
+	 */
+	vc4->bo_labels = kcalloc(VC4_BO_TYPE_COUNT, sizeof(*vc4->bo_labels),
+				 GFP_KERNEL);
+	if (!vc4->bo_labels)
+		return -ENOMEM;
+	vc4->num_labels = VC4_BO_TYPE_COUNT;
+
+	BUILD_BUG_ON(ARRAY_SIZE(bo_type_names) != VC4_BO_TYPE_COUNT);
+	for (i = 0; i < VC4_BO_TYPE_COUNT; i++)
+		vc4->bo_labels[i].name = bo_type_names[i];
+
+	mutex_init(&vc4->bo_lock);
+
+	INIT_LIST_HEAD(&vc4->bo_cache.time_list);
+
+	INIT_WORK(&vc4->bo_cache.time_work, vc4_bo_cache_time_work);
+	timer_setup(&vc4->bo_cache.time_timer, vc4_bo_cache_time_timer, 0);
+
+	return 0;
+}
+
+void vc4_bo_cache_destroy(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int i;
+
+	del_timer(&vc4->bo_cache.time_timer);
+	cancel_work_sync(&vc4->bo_cache.time_work);
+
+	vc4_bo_cache_purge(dev);
+
+	for (i = 0; i < vc4->num_labels; i++) {
+		if (vc4->bo_labels[i].num_allocated) {
+			DRM_ERROR("Destroying BO cache with %d %s "
+				  "BOs still allocated\n",
+				  vc4->bo_labels[i].num_allocated,
+				  vc4->bo_labels[i].name);
+		}
+
+		if (is_user_label(i))
+			kfree(vc4->bo_labels[i].name);
+	}
+	kfree(vc4->bo_labels);
+}
+
+int vc4_label_bo_ioctl(struct drm_device *dev, void *data,
+		       struct drm_file *file_priv)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct drm_vc4_label_bo *args = data;
+	char *name;
+	struct drm_gem_object *gem_obj;
+	int ret = 0, label;
+
+	if (!args->len)
+		return -EINVAL;
+
+	name = strndup_user(u64_to_user_ptr(args->name), args->len + 1);
+	if (IS_ERR(name))
+		return PTR_ERR(name);
+
+	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
+	if (!gem_obj) {
+		DRM_ERROR("Failed to look up GEM BO %d\n", args->handle);
+		kfree(name);
+		return -ENOENT;
+	}
+
+	mutex_lock(&vc4->bo_lock);
+	label = vc4_get_user_label(vc4, name);
+	if (label != -1)
+		vc4_bo_set_label(gem_obj, label);
+	else
+		ret = -ENOMEM;
+	mutex_unlock(&vc4->bo_lock);
+
+	drm_gem_object_put_unlocked(gem_obj);
+
+	return ret;
+}
diff --git a/drivers/gpu/drm/vc4/vc4_crtc.c b/drivers/gpu/drm/vc4/vc4_crtc.c
new file mode 100644
index 000000000..5615ceb15
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_crtc.c
@@ -0,0 +1,1254 @@
+/*
+ * Copyright (C) 2015 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+/**
+ * DOC: VC4 CRTC module
+ *
+ * In VC4, the Pixel Valve is what most closely corresponds to the
+ * DRM's concept of a CRTC.  The PV generates video timings from the
+ * encoder's clock plus its configuration.  It pulls scaled pixels from
+ * the HVS at that timing, and feeds it to the encoder.
+ *
+ * However, the DRM CRTC also collects the configuration of all the
+ * DRM planes attached to it.  As a result, the CRTC is also
+ * responsible for writing the display list for the HVS channel that
+ * the CRTC will use.
+ *
+ * The 2835 has 3 different pixel valves.  pv0 in the audio power
+ * domain feeds DSI0 or DPI, while pv1 feeds DS1 or SMI.  pv2 in the
+ * image domain can feed either HDMI or the SDTV controller.  The
+ * pixel valve chooses from the CPRMAN clocks (HSM for HDMI, VEC for
+ * SDTV, etc.) according to which output type is chosen in the mux.
+ *
+ * For power management, the pixel valve's registers are all clocked
+ * by the AXI clock, while the timings and FIFOs make use of the
+ * output-specific clock.  Since the encoders also directly consume
+ * the CPRMAN clocks, and know what timings they need, they are the
+ * ones that set the clock.
+ */
+
+#include <drm/drm_atomic.h>
+#include <drm/drm_atomic_helper.h>
+#include <drm/drm_crtc_helper.h>
+#include <linux/clk.h>
+#include <drm/drm_fb_cma_helper.h>
+#include <linux/component.h>
+#include <linux/of_device.h>
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+struct vc4_crtc_state {
+	struct drm_crtc_state base;
+	/* Dlist area for this CRTC configuration. */
+	struct drm_mm_node mm;
+	bool feed_txp;
+	bool txp_armed;
+};
+
+static inline struct vc4_crtc_state *
+to_vc4_crtc_state(struct drm_crtc_state *crtc_state)
+{
+	return (struct vc4_crtc_state *)crtc_state;
+}
+
+#define CRTC_WRITE(offset, val) writel(val, vc4_crtc->regs + (offset))
+#define CRTC_READ(offset) readl(vc4_crtc->regs + (offset))
+
+#define CRTC_REG(reg) { reg, #reg }
+static const struct {
+	u32 reg;
+	const char *name;
+} crtc_regs[] = {
+	CRTC_REG(PV_CONTROL),
+	CRTC_REG(PV_V_CONTROL),
+	CRTC_REG(PV_VSYNCD_EVEN),
+	CRTC_REG(PV_HORZA),
+	CRTC_REG(PV_HORZB),
+	CRTC_REG(PV_VERTA),
+	CRTC_REG(PV_VERTB),
+	CRTC_REG(PV_VERTA_EVEN),
+	CRTC_REG(PV_VERTB_EVEN),
+	CRTC_REG(PV_INTEN),
+	CRTC_REG(PV_INTSTAT),
+	CRTC_REG(PV_STAT),
+	CRTC_REG(PV_HACT_ACT),
+};
+
+static void vc4_crtc_dump_regs(struct vc4_crtc *vc4_crtc)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(crtc_regs); i++) {
+		DRM_INFO("0x%04x (%s): 0x%08x\n",
+			 crtc_regs[i].reg, crtc_regs[i].name,
+			 CRTC_READ(crtc_regs[i].reg));
+	}
+}
+
+#ifdef CONFIG_DEBUG_FS
+int vc4_crtc_debugfs_regs(struct seq_file *m, void *unused)
+{
+	struct drm_info_node *node = (struct drm_info_node *)m->private;
+	struct drm_device *dev = node->minor->dev;
+	int crtc_index = (uintptr_t)node->info_ent->data;
+	struct drm_crtc *crtc;
+	struct vc4_crtc *vc4_crtc;
+	int i;
+
+	i = 0;
+	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
+		if (i == crtc_index)
+			break;
+		i++;
+	}
+	if (!crtc)
+		return 0;
+	vc4_crtc = to_vc4_crtc(crtc);
+
+	for (i = 0; i < ARRAY_SIZE(crtc_regs); i++) {
+		seq_printf(m, "%s (0x%04x): 0x%08x\n",
+			   crtc_regs[i].name, crtc_regs[i].reg,
+			   CRTC_READ(crtc_regs[i].reg));
+	}
+
+	return 0;
+}
+#endif
+
+bool vc4_crtc_get_scanoutpos(struct drm_device *dev, unsigned int crtc_id,
+			     bool in_vblank_irq, int *vpos, int *hpos,
+			     ktime_t *stime, ktime_t *etime,
+			     const struct drm_display_mode *mode)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct drm_crtc *crtc = drm_crtc_from_index(dev, crtc_id);
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	u32 val;
+	int fifo_lines;
+	int vblank_lines;
+	bool ret = false;
+
+	/* preempt_disable_rt() should go right here in PREEMPT_RT patchset. */
+
+	/* Get optional system timestamp before query. */
+	if (stime)
+		*stime = ktime_get();
+
+	/*
+	 * Read vertical scanline which is currently composed for our
+	 * pixelvalve by the HVS, and also the scaler status.
+	 */
+	val = HVS_READ(SCALER_DISPSTATX(vc4_crtc->channel));
+
+	/* Get optional system timestamp after query. */
+	if (etime)
+		*etime = ktime_get();
+
+	/* preempt_enable_rt() should go right here in PREEMPT_RT patchset. */
+
+	/* Vertical position of hvs composed scanline. */
+	*vpos = VC4_GET_FIELD(val, SCALER_DISPSTATX_LINE);
+	*hpos = 0;
+
+	if (mode->flags & DRM_MODE_FLAG_INTERLACE) {
+		*vpos /= 2;
+
+		/* Use hpos to correct for field offset in interlaced mode. */
+		if (VC4_GET_FIELD(val, SCALER_DISPSTATX_FRAME_COUNT) % 2)
+			*hpos += mode->crtc_htotal / 2;
+	}
+
+	/* This is the offset we need for translating hvs -> pv scanout pos. */
+	fifo_lines = vc4_crtc->cob_size / mode->crtc_hdisplay;
+
+	if (fifo_lines > 0)
+		ret = true;
+
+	/* HVS more than fifo_lines into frame for compositing? */
+	if (*vpos > fifo_lines) {
+		/*
+		 * We are in active scanout and can get some meaningful results
+		 * from HVS. The actual PV scanout can not trail behind more
+		 * than fifo_lines as that is the fifo's capacity. Assume that
+		 * in active scanout the HVS and PV work in lockstep wrt. HVS
+		 * refilling the fifo and PV consuming from the fifo, ie.
+		 * whenever the PV consumes and frees up a scanline in the
+		 * fifo, the HVS will immediately refill it, therefore
+		 * incrementing vpos. Therefore we choose HVS read position -
+		 * fifo size in scanlines as a estimate of the real scanout
+		 * position of the PV.
+		 */
+		*vpos -= fifo_lines + 1;
+
+		return ret;
+	}
+
+	/*
+	 * Less: This happens when we are in vblank and the HVS, after getting
+	 * the VSTART restart signal from the PV, just started refilling its
+	 * fifo with new lines from the top-most lines of the new framebuffers.
+	 * The PV does not scan out in vblank, so does not remove lines from
+	 * the fifo, so the fifo will be full quickly and the HVS has to pause.
+	 * We can't get meaningful readings wrt. scanline position of the PV
+	 * and need to make things up in a approximative but consistent way.
+	 */
+	vblank_lines = mode->vtotal - mode->vdisplay;
+
+	if (in_vblank_irq) {
+		/*
+		 * Assume the irq handler got called close to first
+		 * line of vblank, so PV has about a full vblank
+		 * scanlines to go, and as a base timestamp use the
+		 * one taken at entry into vblank irq handler, so it
+		 * is not affected by random delays due to lock
+		 * contention on event_lock or vblank_time lock in
+		 * the core.
+		 */
+		*vpos = -vblank_lines;
+
+		if (stime)
+			*stime = vc4_crtc->t_vblank;
+		if (etime)
+			*etime = vc4_crtc->t_vblank;
+
+		/*
+		 * If the HVS fifo is not yet full then we know for certain
+		 * we are at the very beginning of vblank, as the hvs just
+		 * started refilling, and the stime and etime timestamps
+		 * truly correspond to start of vblank.
+		 *
+		 * Unfortunately there's no way to report this to upper levels
+		 * and make it more useful.
+		 */
+	} else {
+		/*
+		 * No clue where we are inside vblank. Return a vpos of zero,
+		 * which will cause calling code to just return the etime
+		 * timestamp uncorrected. At least this is no worse than the
+		 * standard fallback.
+		 */
+		*vpos = 0;
+	}
+
+	return ret;
+}
+
+static void vc4_crtc_destroy(struct drm_crtc *crtc)
+{
+	drm_crtc_cleanup(crtc);
+}
+
+static void
+vc4_crtc_lut_load(struct drm_crtc *crtc)
+{
+	struct drm_device *dev = crtc->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	u32 i;
+
+	/* The LUT memory is laid out with each HVS channel in order,
+	 * each of which takes 256 writes for R, 256 for G, then 256
+	 * for B.
+	 */
+	HVS_WRITE(SCALER_GAMADDR,
+		  SCALER_GAMADDR_AUTOINC |
+		  (vc4_crtc->channel * 3 * crtc->gamma_size));
+
+	for (i = 0; i < crtc->gamma_size; i++)
+		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_r[i]);
+	for (i = 0; i < crtc->gamma_size; i++)
+		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_g[i]);
+	for (i = 0; i < crtc->gamma_size; i++)
+		HVS_WRITE(SCALER_GAMDATA, vc4_crtc->lut_b[i]);
+}
+
+static void
+vc4_crtc_update_gamma_lut(struct drm_crtc *crtc)
+{
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	struct drm_color_lut *lut = crtc->state->gamma_lut->data;
+	u32 length = drm_color_lut_size(crtc->state->gamma_lut);
+	u32 i;
+
+	for (i = 0; i < length; i++) {
+		vc4_crtc->lut_r[i] = drm_color_lut_extract(lut[i].red, 8);
+		vc4_crtc->lut_g[i] = drm_color_lut_extract(lut[i].green, 8);
+		vc4_crtc->lut_b[i] = drm_color_lut_extract(lut[i].blue, 8);
+	}
+
+	vc4_crtc_lut_load(crtc);
+}
+
+static u32 vc4_get_fifo_full_level(u32 format)
+{
+	static const u32 fifo_len_bytes = 64;
+	static const u32 hvs_latency_pix = 6;
+
+	switch (format) {
+	case PV_CONTROL_FORMAT_DSIV_16:
+	case PV_CONTROL_FORMAT_DSIC_16:
+		return fifo_len_bytes - 2 * hvs_latency_pix;
+	case PV_CONTROL_FORMAT_DSIV_18:
+		return fifo_len_bytes - 14;
+	case PV_CONTROL_FORMAT_24:
+	case PV_CONTROL_FORMAT_DSIV_24:
+	default:
+		return fifo_len_bytes - 3 * hvs_latency_pix;
+	}
+}
+
+/*
+ * Returns the encoder attached to the CRTC.
+ *
+ * VC4 can only scan out to one encoder at a time, while the DRM core
+ * allows drivers to push pixels to more than one encoder from the
+ * same CRTC.
+ */
+static struct drm_encoder *vc4_get_crtc_encoder(struct drm_crtc *crtc)
+{
+	struct drm_connector *connector;
+	struct drm_connector_list_iter conn_iter;
+
+	drm_connector_list_iter_begin(crtc->dev, &conn_iter);
+	drm_for_each_connector_iter(connector, &conn_iter) {
+		if (connector->state->crtc == crtc) {
+			drm_connector_list_iter_end(&conn_iter);
+			return connector->encoder;
+		}
+	}
+	drm_connector_list_iter_end(&conn_iter);
+
+	return NULL;
+}
+
+static void vc4_crtc_config_pv(struct drm_crtc *crtc)
+{
+	struct drm_encoder *encoder = vc4_get_crtc_encoder(crtc);
+	struct vc4_encoder *vc4_encoder = to_vc4_encoder(encoder);
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	struct drm_crtc_state *state = crtc->state;
+	struct drm_display_mode *mode = &state->adjusted_mode;
+	bool interlace = mode->flags & DRM_MODE_FLAG_INTERLACE;
+	u32 pixel_rep = (mode->flags & DRM_MODE_FLAG_DBLCLK) ? 2 : 1;
+	bool is_dsi = (vc4_encoder->type == VC4_ENCODER_TYPE_DSI0 ||
+		       vc4_encoder->type == VC4_ENCODER_TYPE_DSI1);
+	u32 format = is_dsi ? PV_CONTROL_FORMAT_DSIV_24 : PV_CONTROL_FORMAT_24;
+
+	/* Reset the PV fifo. */
+	CRTC_WRITE(PV_CONTROL, 0);
+	CRTC_WRITE(PV_CONTROL, PV_CONTROL_FIFO_CLR | PV_CONTROL_EN);
+	CRTC_WRITE(PV_CONTROL, 0);
+
+	CRTC_WRITE(PV_HORZA,
+		   VC4_SET_FIELD((mode->htotal -
+				  mode->hsync_end) * pixel_rep,
+				 PV_HORZA_HBP) |
+		   VC4_SET_FIELD((mode->hsync_end -
+				  mode->hsync_start) * pixel_rep,
+				 PV_HORZA_HSYNC));
+	CRTC_WRITE(PV_HORZB,
+		   VC4_SET_FIELD((mode->hsync_start -
+				  mode->hdisplay) * pixel_rep,
+				 PV_HORZB_HFP) |
+		   VC4_SET_FIELD(mode->hdisplay * pixel_rep, PV_HORZB_HACTIVE));
+
+	CRTC_WRITE(PV_VERTA,
+		   VC4_SET_FIELD(mode->crtc_vtotal - mode->crtc_vsync_end,
+				 PV_VERTA_VBP) |
+		   VC4_SET_FIELD(mode->crtc_vsync_end - mode->crtc_vsync_start,
+				 PV_VERTA_VSYNC));
+	CRTC_WRITE(PV_VERTB,
+		   VC4_SET_FIELD(mode->crtc_vsync_start - mode->crtc_vdisplay,
+				 PV_VERTB_VFP) |
+		   VC4_SET_FIELD(mode->crtc_vdisplay, PV_VERTB_VACTIVE));
+
+	if (interlace) {
+		CRTC_WRITE(PV_VERTA_EVEN,
+			   VC4_SET_FIELD(mode->crtc_vtotal -
+					 mode->crtc_vsync_end - 1,
+					 PV_VERTA_VBP) |
+			   VC4_SET_FIELD(mode->crtc_vsync_end -
+					 mode->crtc_vsync_start,
+					 PV_VERTA_VSYNC));
+		CRTC_WRITE(PV_VERTB_EVEN,
+			   VC4_SET_FIELD(mode->crtc_vsync_start -
+					 mode->crtc_vdisplay,
+					 PV_VERTB_VFP) |
+			   VC4_SET_FIELD(mode->crtc_vdisplay, PV_VERTB_VACTIVE));
+
+		/* We set up first field even mode for HDMI.  VEC's
+		 * NTSC mode would want first field odd instead, once
+		 * we support it (to do so, set ODD_FIRST and put the
+		 * delay in VSYNCD_EVEN instead).
+		 */
+		CRTC_WRITE(PV_V_CONTROL,
+			   PV_VCONTROL_CONTINUOUS |
+			   (is_dsi ? PV_VCONTROL_DSI : 0) |
+			   PV_VCONTROL_INTERLACE |
+			   VC4_SET_FIELD(mode->htotal * pixel_rep / 2,
+					 PV_VCONTROL_ODD_DELAY));
+		CRTC_WRITE(PV_VSYNCD_EVEN, 0);
+	} else {
+		CRTC_WRITE(PV_V_CONTROL,
+			   PV_VCONTROL_CONTINUOUS |
+			   (is_dsi ? PV_VCONTROL_DSI : 0));
+	}
+
+	CRTC_WRITE(PV_HACT_ACT, mode->hdisplay * pixel_rep);
+
+	CRTC_WRITE(PV_CONTROL,
+		   VC4_SET_FIELD(format, PV_CONTROL_FORMAT) |
+		   VC4_SET_FIELD(vc4_get_fifo_full_level(format),
+				 PV_CONTROL_FIFO_LEVEL) |
+		   VC4_SET_FIELD(pixel_rep - 1, PV_CONTROL_PIXEL_REP) |
+		   PV_CONTROL_CLR_AT_START |
+		   PV_CONTROL_TRIGGER_UNDERFLOW |
+		   PV_CONTROL_WAIT_HSTART |
+		   VC4_SET_FIELD(vc4_encoder->clock_select,
+				 PV_CONTROL_CLK_SELECT) |
+		   PV_CONTROL_FIFO_CLR |
+		   PV_CONTROL_EN);
+}
+
+static void vc4_crtc_mode_set_nofb(struct drm_crtc *crtc)
+{
+	struct drm_device *dev = crtc->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
+	struct drm_display_mode *mode = &crtc->state->adjusted_mode;
+	bool interlace = mode->flags & DRM_MODE_FLAG_INTERLACE;
+	bool debug_dump_regs = false;
+
+	if (debug_dump_regs) {
+		DRM_INFO("CRTC %d regs before:\n", drm_crtc_index(crtc));
+		vc4_crtc_dump_regs(vc4_crtc);
+	}
+
+	if (vc4_crtc->channel == 2) {
+		u32 dispctrl;
+		u32 dsp3_mux;
+
+		/*
+		 * SCALER_DISPCTRL_DSP3 = X, where X < 2 means 'connect DSP3 to
+		 * FIFO X'.
+		 * SCALER_DISPCTRL_DSP3 = 3 means 'disable DSP 3'.
+		 *
+		 * DSP3 is connected to FIFO2 unless the transposer is
+		 * enabled. In this case, FIFO 2 is directly accessed by the
+		 * TXP IP, and we need to disable the FIFO2 -> pixelvalve1
+		 * route.
+		 */
+		if (vc4_state->feed_txp)
+			dsp3_mux = VC4_SET_FIELD(3, SCALER_DISPCTRL_DSP3_MUX);
+		else
+			dsp3_mux = VC4_SET_FIELD(2, SCALER_DISPCTRL_DSP3_MUX);
+
+		dispctrl = HVS_READ(SCALER_DISPCTRL) &
+			   ~SCALER_DISPCTRL_DSP3_MUX_MASK;
+		HVS_WRITE(SCALER_DISPCTRL, dispctrl | dsp3_mux);
+	}
+
+	if (!vc4_state->feed_txp)
+		vc4_crtc_config_pv(crtc);
+
+	HVS_WRITE(SCALER_DISPBKGNDX(vc4_crtc->channel),
+		  SCALER_DISPBKGND_AUTOHS |
+		  SCALER_DISPBKGND_GAMMA |
+		  (interlace ? SCALER_DISPBKGND_INTERLACE : 0));
+
+	/* Reload the LUT, since the SRAMs would have been disabled if
+	 * all CRTCs had SCALER_DISPBKGND_GAMMA unset at once.
+	 */
+	vc4_crtc_lut_load(crtc);
+
+	if (debug_dump_regs) {
+		DRM_INFO("CRTC %d regs after:\n", drm_crtc_index(crtc));
+		vc4_crtc_dump_regs(vc4_crtc);
+	}
+}
+
+static void require_hvs_enabled(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	WARN_ON_ONCE((HVS_READ(SCALER_DISPCTRL) & SCALER_DISPCTRL_ENABLE) !=
+		     SCALER_DISPCTRL_ENABLE);
+}
+
+static void vc4_crtc_atomic_disable(struct drm_crtc *crtc,
+				    struct drm_crtc_state *old_state)
+{
+	struct drm_device *dev = crtc->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	u32 chan = vc4_crtc->channel;
+	int ret;
+	require_hvs_enabled(dev);
+
+	/* Disable vblank irq handling before crtc is disabled. */
+	drm_crtc_vblank_off(crtc);
+
+	CRTC_WRITE(PV_V_CONTROL,
+		   CRTC_READ(PV_V_CONTROL) & ~PV_VCONTROL_VIDEN);
+	ret = wait_for(!(CRTC_READ(PV_V_CONTROL) & PV_VCONTROL_VIDEN), 1);
+	WARN_ONCE(ret, "Timeout waiting for !PV_VCONTROL_VIDEN\n");
+
+	if (HVS_READ(SCALER_DISPCTRLX(chan)) &
+	    SCALER_DISPCTRLX_ENABLE) {
+		HVS_WRITE(SCALER_DISPCTRLX(chan),
+			  SCALER_DISPCTRLX_RESET);
+
+		/* While the docs say that reset is self-clearing, it
+		 * seems it doesn't actually.
+		 */
+		HVS_WRITE(SCALER_DISPCTRLX(chan), 0);
+	}
+
+	/* Once we leave, the scaler should be disabled and its fifo empty. */
+
+	WARN_ON_ONCE(HVS_READ(SCALER_DISPCTRLX(chan)) & SCALER_DISPCTRLX_RESET);
+
+	WARN_ON_ONCE(VC4_GET_FIELD(HVS_READ(SCALER_DISPSTATX(chan)),
+				   SCALER_DISPSTATX_MODE) !=
+		     SCALER_DISPSTATX_MODE_DISABLED);
+
+	WARN_ON_ONCE((HVS_READ(SCALER_DISPSTATX(chan)) &
+		      (SCALER_DISPSTATX_FULL | SCALER_DISPSTATX_EMPTY)) !=
+		     SCALER_DISPSTATX_EMPTY);
+
+	/*
+	 * Make sure we issue a vblank event after disabling the CRTC if
+	 * someone was waiting it.
+	 */
+	if (crtc->state->event) {
+		unsigned long flags;
+
+		spin_lock_irqsave(&dev->event_lock, flags);
+		drm_crtc_send_vblank_event(crtc, crtc->state->event);
+		crtc->state->event = NULL;
+		spin_unlock_irqrestore(&dev->event_lock, flags);
+	}
+}
+
+void vc4_crtc_txp_armed(struct drm_crtc_state *state)
+{
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
+
+	vc4_state->txp_armed = true;
+}
+
+static void vc4_crtc_update_dlist(struct drm_crtc *crtc)
+{
+	struct drm_device *dev = crtc->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
+
+	if (crtc->state->event) {
+		unsigned long flags;
+
+		crtc->state->event->pipe = drm_crtc_index(crtc);
+
+		WARN_ON(drm_crtc_vblank_get(crtc) != 0);
+
+		spin_lock_irqsave(&dev->event_lock, flags);
+
+		if (!vc4_state->feed_txp || vc4_state->txp_armed) {
+			vc4_crtc->event = crtc->state->event;
+			crtc->state->event = NULL;
+		}
+
+		HVS_WRITE(SCALER_DISPLISTX(vc4_crtc->channel),
+			  vc4_state->mm.start);
+
+		spin_unlock_irqrestore(&dev->event_lock, flags);
+	} else {
+		HVS_WRITE(SCALER_DISPLISTX(vc4_crtc->channel),
+			  vc4_state->mm.start);
+	}
+}
+
+static void vc4_crtc_atomic_enable(struct drm_crtc *crtc,
+				   struct drm_crtc_state *old_state)
+{
+	struct drm_device *dev = crtc->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
+	struct drm_display_mode *mode = &crtc->state->adjusted_mode;
+
+	require_hvs_enabled(dev);
+
+	/* Enable vblank irq handling before crtc is started otherwise
+	 * drm_crtc_get_vblank() fails in vc4_crtc_update_dlist().
+	 */
+	drm_crtc_vblank_on(crtc);
+	vc4_crtc_update_dlist(crtc);
+
+	/* Turn on the scaler, which will wait for vstart to start
+	 * compositing.
+	 * When feeding the transposer, we should operate in oneshot
+	 * mode.
+	 */
+	HVS_WRITE(SCALER_DISPCTRLX(vc4_crtc->channel),
+		  VC4_SET_FIELD(mode->hdisplay, SCALER_DISPCTRLX_WIDTH) |
+		  VC4_SET_FIELD(mode->vdisplay, SCALER_DISPCTRLX_HEIGHT) |
+		  SCALER_DISPCTRLX_ENABLE |
+		  (vc4_state->feed_txp ? SCALER_DISPCTRLX_ONESHOT : 0));
+
+	/* When feeding the transposer block the pixelvalve is unneeded and
+	 * should not be enabled.
+	 */
+	if (!vc4_state->feed_txp)
+		CRTC_WRITE(PV_V_CONTROL,
+			   CRTC_READ(PV_V_CONTROL) | PV_VCONTROL_VIDEN);
+}
+
+static enum drm_mode_status vc4_crtc_mode_valid(struct drm_crtc *crtc,
+						const struct drm_display_mode *mode)
+{
+	/* Do not allow doublescan modes from user space */
+	if (mode->flags & DRM_MODE_FLAG_DBLSCAN) {
+		DRM_DEBUG_KMS("[CRTC:%d] Doublescan mode rejected.\n",
+			      crtc->base.id);
+		return MODE_NO_DBLESCAN;
+	}
+
+	return MODE_OK;
+}
+
+static int vc4_crtc_atomic_check(struct drm_crtc *crtc,
+				 struct drm_crtc_state *state)
+{
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
+	struct drm_device *dev = crtc->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct drm_plane *plane;
+	unsigned long flags;
+	const struct drm_plane_state *plane_state;
+	struct drm_connector *conn;
+	struct drm_connector_state *conn_state;
+	u32 dlist_count = 0;
+	int ret, i;
+
+	/* The pixelvalve can only feed one encoder (and encoders are
+	 * 1:1 with connectors.)
+	 */
+	if (hweight32(state->connector_mask) > 1)
+		return -EINVAL;
+
+	drm_atomic_crtc_state_for_each_plane_state(plane, plane_state, state)
+		dlist_count += vc4_plane_dlist_size(plane_state);
+
+	dlist_count++; /* Account for SCALER_CTL0_END. */
+
+	spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
+	ret = drm_mm_insert_node(&vc4->hvs->dlist_mm, &vc4_state->mm,
+				 dlist_count);
+	spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
+	if (ret)
+		return ret;
+
+	for_each_new_connector_in_state(state->state, conn, conn_state, i) {
+		if (conn_state->crtc != crtc)
+			continue;
+
+		/* The writeback connector is implemented using the transposer
+		 * block which is directly taking its data from the HVS FIFO.
+		 */
+		if (conn->connector_type == DRM_MODE_CONNECTOR_WRITEBACK) {
+			state->no_vblank = true;
+			vc4_state->feed_txp = true;
+		} else {
+			state->no_vblank = false;
+			vc4_state->feed_txp = false;
+		}
+
+		break;
+	}
+
+	return 0;
+}
+
+static void vc4_crtc_atomic_flush(struct drm_crtc *crtc,
+				  struct drm_crtc_state *old_state)
+{
+	struct drm_device *dev = crtc->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
+	struct drm_plane *plane;
+	struct vc4_plane_state *vc4_plane_state;
+	bool debug_dump_regs = false;
+	bool enable_bg_fill = false;
+	u32 __iomem *dlist_start = vc4->hvs->dlist + vc4_state->mm.start;
+	u32 __iomem *dlist_next = dlist_start;
+
+	if (debug_dump_regs) {
+		DRM_INFO("CRTC %d HVS before:\n", drm_crtc_index(crtc));
+		vc4_hvs_dump_state(dev);
+	}
+
+	/* Copy all the active planes' dlist contents to the hardware dlist. */
+	drm_atomic_crtc_for_each_plane(plane, crtc) {
+		/* Is this the first active plane? */
+		if (dlist_next == dlist_start) {
+			/* We need to enable background fill when a plane
+			 * could be alpha blending from the background, i.e.
+			 * where no other plane is underneath. It suffices to
+			 * consider the first active plane here since we set
+			 * needs_bg_fill such that either the first plane
+			 * already needs it or all planes on top blend from
+			 * the first or a lower plane.
+			 */
+			vc4_plane_state = to_vc4_plane_state(plane->state);
+			enable_bg_fill = vc4_plane_state->needs_bg_fill;
+		}
+
+		dlist_next += vc4_plane_write_dlist(plane, dlist_next);
+	}
+
+	writel(SCALER_CTL0_END, dlist_next);
+	dlist_next++;
+
+	WARN_ON_ONCE(dlist_next - dlist_start != vc4_state->mm.size);
+
+	if (enable_bg_fill)
+		/* This sets a black background color fill, as is the case
+		 * with other DRM drivers.
+		 */
+		HVS_WRITE(SCALER_DISPBKGNDX(vc4_crtc->channel),
+			  HVS_READ(SCALER_DISPBKGNDX(vc4_crtc->channel)) |
+			  SCALER_DISPBKGND_FILL);
+
+	/* Only update DISPLIST if the CRTC was already running and is not
+	 * being disabled.
+	 * vc4_crtc_enable() takes care of updating the dlist just after
+	 * re-enabling VBLANK interrupts and before enabling the engine.
+	 * If the CRTC is being disabled, there's no point in updating this
+	 * information.
+	 */
+	if (crtc->state->active && old_state->active)
+		vc4_crtc_update_dlist(crtc);
+
+	if (crtc->state->color_mgmt_changed) {
+		u32 dispbkgndx = HVS_READ(SCALER_DISPBKGNDX(vc4_crtc->channel));
+
+		if (crtc->state->gamma_lut) {
+			vc4_crtc_update_gamma_lut(crtc);
+			dispbkgndx |= SCALER_DISPBKGND_GAMMA;
+		} else {
+			/* Unsetting DISPBKGND_GAMMA skips the gamma lut step
+			 * in hardware, which is the same as a linear lut that
+			 * DRM expects us to use in absence of a user lut.
+			 */
+			dispbkgndx &= ~SCALER_DISPBKGND_GAMMA;
+		}
+		HVS_WRITE(SCALER_DISPBKGNDX(vc4_crtc->channel), dispbkgndx);
+	}
+
+	if (debug_dump_regs) {
+		DRM_INFO("CRTC %d HVS after:\n", drm_crtc_index(crtc));
+		vc4_hvs_dump_state(dev);
+	}
+}
+
+static int vc4_enable_vblank(struct drm_crtc *crtc)
+{
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+
+	CRTC_WRITE(PV_INTEN, PV_INT_VFP_START);
+
+	return 0;
+}
+
+static void vc4_disable_vblank(struct drm_crtc *crtc)
+{
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+
+	CRTC_WRITE(PV_INTEN, 0);
+}
+
+static void vc4_crtc_handle_page_flip(struct vc4_crtc *vc4_crtc)
+{
+	struct drm_crtc *crtc = &vc4_crtc->base;
+	struct drm_device *dev = crtc->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(crtc->state);
+	u32 chan = vc4_crtc->channel;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev->event_lock, flags);
+	if (vc4_crtc->event &&
+	    (vc4_state->mm.start == HVS_READ(SCALER_DISPLACTX(chan)) ||
+	     vc4_state->feed_txp)) {
+		drm_crtc_send_vblank_event(crtc, vc4_crtc->event);
+		vc4_crtc->event = NULL;
+		drm_crtc_vblank_put(crtc);
+	}
+	spin_unlock_irqrestore(&dev->event_lock, flags);
+}
+
+void vc4_crtc_handle_vblank(struct vc4_crtc *crtc)
+{
+	crtc->t_vblank = ktime_get();
+	drm_crtc_handle_vblank(&crtc->base);
+	vc4_crtc_handle_page_flip(crtc);
+}
+
+static irqreturn_t vc4_crtc_irq_handler(int irq, void *data)
+{
+	struct vc4_crtc *vc4_crtc = data;
+	u32 stat = CRTC_READ(PV_INTSTAT);
+	irqreturn_t ret = IRQ_NONE;
+
+	if (stat & PV_INT_VFP_START) {
+		CRTC_WRITE(PV_INTSTAT, PV_INT_VFP_START);
+		vc4_crtc_handle_vblank(vc4_crtc);
+		ret = IRQ_HANDLED;
+	}
+
+	return ret;
+}
+
+struct vc4_async_flip_state {
+	struct drm_crtc *crtc;
+	struct drm_framebuffer *fb;
+	struct drm_framebuffer *old_fb;
+	struct drm_pending_vblank_event *event;
+
+	struct vc4_seqno_cb cb;
+};
+
+/* Called when the V3D execution for the BO being flipped to is done, so that
+ * we can actually update the plane's address to point to it.
+ */
+static void
+vc4_async_page_flip_complete(struct vc4_seqno_cb *cb)
+{
+	struct vc4_async_flip_state *flip_state =
+		container_of(cb, struct vc4_async_flip_state, cb);
+	struct drm_crtc *crtc = flip_state->crtc;
+	struct drm_device *dev = crtc->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct drm_plane *plane = crtc->primary;
+
+	vc4_plane_async_set_fb(plane, flip_state->fb);
+	if (flip_state->event) {
+		unsigned long flags;
+
+		spin_lock_irqsave(&dev->event_lock, flags);
+		drm_crtc_send_vblank_event(crtc, flip_state->event);
+		spin_unlock_irqrestore(&dev->event_lock, flags);
+	}
+
+	drm_crtc_vblank_put(crtc);
+	drm_framebuffer_put(flip_state->fb);
+
+	/* Decrement the BO usecnt in order to keep the inc/dec calls balanced
+	 * when the planes are updated through the async update path.
+	 * FIXME: we should move to generic async-page-flip when it's
+	 * available, so that we can get rid of this hand-made cleanup_fb()
+	 * logic.
+	 */
+	if (flip_state->old_fb) {
+		struct drm_gem_cma_object *cma_bo;
+		struct vc4_bo *bo;
+
+		cma_bo = drm_fb_cma_get_gem_obj(flip_state->old_fb, 0);
+		bo = to_vc4_bo(&cma_bo->base);
+		vc4_bo_dec_usecnt(bo);
+		drm_framebuffer_put(flip_state->old_fb);
+	}
+
+	kfree(flip_state);
+
+	up(&vc4->async_modeset);
+}
+
+/* Implements async (non-vblank-synced) page flips.
+ *
+ * The page flip ioctl needs to return immediately, so we grab the
+ * modeset semaphore on the pipe, and queue the address update for
+ * when V3D is done with the BO being flipped to.
+ */
+static int vc4_async_page_flip(struct drm_crtc *crtc,
+			       struct drm_framebuffer *fb,
+			       struct drm_pending_vblank_event *event,
+			       uint32_t flags)
+{
+	struct drm_device *dev = crtc->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct drm_plane *plane = crtc->primary;
+	int ret = 0;
+	struct vc4_async_flip_state *flip_state;
+	struct drm_gem_cma_object *cma_bo = drm_fb_cma_get_gem_obj(fb, 0);
+	struct vc4_bo *bo = to_vc4_bo(&cma_bo->base);
+
+	/* Increment the BO usecnt here, so that we never end up with an
+	 * unbalanced number of vc4_bo_{dec,inc}_usecnt() calls when the
+	 * plane is later updated through the non-async path.
+	 * FIXME: we should move to generic async-page-flip when it's
+	 * available, so that we can get rid of this hand-made prepare_fb()
+	 * logic.
+	 */
+	ret = vc4_bo_inc_usecnt(bo);
+	if (ret)
+		return ret;
+
+	flip_state = kzalloc(sizeof(*flip_state), GFP_KERNEL);
+	if (!flip_state) {
+		vc4_bo_dec_usecnt(bo);
+		return -ENOMEM;
+	}
+
+	drm_framebuffer_get(fb);
+	flip_state->fb = fb;
+	flip_state->crtc = crtc;
+	flip_state->event = event;
+
+	/* Make sure all other async modesetes have landed. */
+	ret = down_interruptible(&vc4->async_modeset);
+	if (ret) {
+		drm_framebuffer_put(fb);
+		vc4_bo_dec_usecnt(bo);
+		kfree(flip_state);
+		return ret;
+	}
+
+	/* Save the current FB before it's replaced by the new one in
+	 * drm_atomic_set_fb_for_plane(). We'll need the old FB in
+	 * vc4_async_page_flip_complete() to decrement the BO usecnt and keep
+	 * it consistent.
+	 * FIXME: we should move to generic async-page-flip when it's
+	 * available, so that we can get rid of this hand-made cleanup_fb()
+	 * logic.
+	 */
+	flip_state->old_fb = plane->state->fb;
+	if (flip_state->old_fb)
+		drm_framebuffer_get(flip_state->old_fb);
+
+	WARN_ON(drm_crtc_vblank_get(crtc) != 0);
+
+	/* Immediately update the plane's legacy fb pointer, so that later
+	 * modeset prep sees the state that will be present when the semaphore
+	 * is released.
+	 */
+	drm_atomic_set_fb_for_plane(plane->state, fb);
+
+	vc4_queue_seqno_cb(dev, &flip_state->cb, bo->seqno,
+			   vc4_async_page_flip_complete);
+
+	/* Driver takes ownership of state on successful async commit. */
+	return 0;
+}
+
+static int vc4_page_flip(struct drm_crtc *crtc,
+			 struct drm_framebuffer *fb,
+			 struct drm_pending_vblank_event *event,
+			 uint32_t flags,
+			 struct drm_modeset_acquire_ctx *ctx)
+{
+	if (flags & DRM_MODE_PAGE_FLIP_ASYNC)
+		return vc4_async_page_flip(crtc, fb, event, flags);
+	else
+		return drm_atomic_helper_page_flip(crtc, fb, event, flags, ctx);
+}
+
+static struct drm_crtc_state *vc4_crtc_duplicate_state(struct drm_crtc *crtc)
+{
+	struct vc4_crtc_state *vc4_state, *old_vc4_state;
+
+	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
+	if (!vc4_state)
+		return NULL;
+
+	old_vc4_state = to_vc4_crtc_state(crtc->state);
+	vc4_state->feed_txp = old_vc4_state->feed_txp;
+
+	__drm_atomic_helper_crtc_duplicate_state(crtc, &vc4_state->base);
+	return &vc4_state->base;
+}
+
+static void vc4_crtc_destroy_state(struct drm_crtc *crtc,
+				   struct drm_crtc_state *state)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(crtc->dev);
+	struct vc4_crtc_state *vc4_state = to_vc4_crtc_state(state);
+
+	if (vc4_state->mm.allocated) {
+		unsigned long flags;
+
+		spin_lock_irqsave(&vc4->hvs->mm_lock, flags);
+		drm_mm_remove_node(&vc4_state->mm);
+		spin_unlock_irqrestore(&vc4->hvs->mm_lock, flags);
+
+	}
+
+	drm_atomic_helper_crtc_destroy_state(crtc, state);
+}
+
+static void
+vc4_crtc_reset(struct drm_crtc *crtc)
+{
+	if (crtc->state)
+		vc4_crtc_destroy_state(crtc, crtc->state);
+
+	crtc->state = kzalloc(sizeof(struct vc4_crtc_state), GFP_KERNEL);
+	if (crtc->state)
+		crtc->state->crtc = crtc;
+}
+
+static const struct drm_crtc_funcs vc4_crtc_funcs = {
+	.set_config = drm_atomic_helper_set_config,
+	.destroy = vc4_crtc_destroy,
+	.page_flip = vc4_page_flip,
+	.set_property = NULL,
+	.cursor_set = NULL, /* handled by drm_mode_cursor_universal */
+	.cursor_move = NULL, /* handled by drm_mode_cursor_universal */
+	.reset = vc4_crtc_reset,
+	.atomic_duplicate_state = vc4_crtc_duplicate_state,
+	.atomic_destroy_state = vc4_crtc_destroy_state,
+	.gamma_set = drm_atomic_helper_legacy_gamma_set,
+	.enable_vblank = vc4_enable_vblank,
+	.disable_vblank = vc4_disable_vblank,
+};
+
+static const struct drm_crtc_helper_funcs vc4_crtc_helper_funcs = {
+	.mode_set_nofb = vc4_crtc_mode_set_nofb,
+	.mode_valid = vc4_crtc_mode_valid,
+	.atomic_check = vc4_crtc_atomic_check,
+	.atomic_flush = vc4_crtc_atomic_flush,
+	.atomic_enable = vc4_crtc_atomic_enable,
+	.atomic_disable = vc4_crtc_atomic_disable,
+};
+
+static const struct vc4_crtc_data pv0_data = {
+	.hvs_channel = 0,
+	.encoder_types = {
+		[PV_CONTROL_CLK_SELECT_DSI] = VC4_ENCODER_TYPE_DSI0,
+		[PV_CONTROL_CLK_SELECT_DPI_SMI_HDMI] = VC4_ENCODER_TYPE_DPI,
+	},
+};
+
+static const struct vc4_crtc_data pv1_data = {
+	.hvs_channel = 2,
+	.encoder_types = {
+		[PV_CONTROL_CLK_SELECT_DSI] = VC4_ENCODER_TYPE_DSI1,
+		[PV_CONTROL_CLK_SELECT_DPI_SMI_HDMI] = VC4_ENCODER_TYPE_SMI,
+	},
+};
+
+static const struct vc4_crtc_data pv2_data = {
+	.hvs_channel = 1,
+	.encoder_types = {
+		[PV_CONTROL_CLK_SELECT_DPI_SMI_HDMI] = VC4_ENCODER_TYPE_HDMI,
+		[PV_CONTROL_CLK_SELECT_VEC] = VC4_ENCODER_TYPE_VEC,
+	},
+};
+
+static const struct of_device_id vc4_crtc_dt_match[] = {
+	{ .compatible = "brcm,bcm2835-pixelvalve0", .data = &pv0_data },
+	{ .compatible = "brcm,bcm2835-pixelvalve1", .data = &pv1_data },
+	{ .compatible = "brcm,bcm2835-pixelvalve2", .data = &pv2_data },
+	{}
+};
+
+static void vc4_set_crtc_possible_masks(struct drm_device *drm,
+					struct drm_crtc *crtc)
+{
+	struct vc4_crtc *vc4_crtc = to_vc4_crtc(crtc);
+	const struct vc4_crtc_data *crtc_data = vc4_crtc->data;
+	const enum vc4_encoder_type *encoder_types = crtc_data->encoder_types;
+	struct drm_encoder *encoder;
+
+	drm_for_each_encoder(encoder, drm) {
+		struct vc4_encoder *vc4_encoder;
+		int i;
+
+		/* HVS FIFO2 can feed the TXP IP. */
+		if (crtc_data->hvs_channel == 2 &&
+		    encoder->encoder_type == DRM_MODE_ENCODER_VIRTUAL) {
+			encoder->possible_crtcs |= drm_crtc_mask(crtc);
+			continue;
+		}
+
+		vc4_encoder = to_vc4_encoder(encoder);
+		for (i = 0; i < ARRAY_SIZE(crtc_data->encoder_types); i++) {
+			if (vc4_encoder->type == encoder_types[i]) {
+				vc4_encoder->clock_select = i;
+				encoder->possible_crtcs |= drm_crtc_mask(crtc);
+				break;
+			}
+		}
+	}
+}
+
+static void
+vc4_crtc_get_cob_allocation(struct vc4_crtc *vc4_crtc)
+{
+	struct drm_device *drm = vc4_crtc->base.dev;
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	u32 dispbase = HVS_READ(SCALER_DISPBASEX(vc4_crtc->channel));
+	/* Top/base are supposed to be 4-pixel aligned, but the
+	 * Raspberry Pi firmware fills the low bits (which are
+	 * presumably ignored).
+	 */
+	u32 top = VC4_GET_FIELD(dispbase, SCALER_DISPBASEX_TOP) & ~3;
+	u32 base = VC4_GET_FIELD(dispbase, SCALER_DISPBASEX_BASE) & ~3;
+
+	vc4_crtc->cob_size = top - base + 4;
+}
+
+static int vc4_crtc_bind(struct device *dev, struct device *master, void *data)
+{
+	struct platform_device *pdev = to_platform_device(dev);
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_crtc *vc4_crtc;
+	struct drm_crtc *crtc;
+	struct drm_plane *primary_plane, *cursor_plane, *destroy_plane, *temp;
+	const struct of_device_id *match;
+	int ret, i;
+
+	vc4_crtc = devm_kzalloc(dev, sizeof(*vc4_crtc), GFP_KERNEL);
+	if (!vc4_crtc)
+		return -ENOMEM;
+	crtc = &vc4_crtc->base;
+
+	match = of_match_device(vc4_crtc_dt_match, dev);
+	if (!match)
+		return -ENODEV;
+	vc4_crtc->data = match->data;
+
+	vc4_crtc->regs = vc4_ioremap_regs(pdev, 0);
+	if (IS_ERR(vc4_crtc->regs))
+		return PTR_ERR(vc4_crtc->regs);
+
+	/* For now, we create just the primary and the legacy cursor
+	 * planes.  We should be able to stack more planes on easily,
+	 * but to do that we would need to compute the bandwidth
+	 * requirement of the plane configuration, and reject ones
+	 * that will take too much.
+	 */
+	primary_plane = vc4_plane_init(drm, DRM_PLANE_TYPE_PRIMARY);
+	if (IS_ERR(primary_plane)) {
+		dev_err(dev, "failed to construct primary plane\n");
+		ret = PTR_ERR(primary_plane);
+		goto err;
+	}
+
+	drm_crtc_init_with_planes(drm, crtc, primary_plane, NULL,
+				  &vc4_crtc_funcs, NULL);
+	drm_crtc_helper_add(crtc, &vc4_crtc_helper_funcs);
+	vc4_crtc->channel = vc4_crtc->data->hvs_channel;
+	drm_mode_crtc_set_gamma_size(crtc, ARRAY_SIZE(vc4_crtc->lut_r));
+	drm_crtc_enable_color_mgmt(crtc, 0, false, crtc->gamma_size);
+
+	/* We support CTM, but only for one CRTC at a time. It's therefore
+	 * implemented as private driver state in vc4_kms, not here.
+	 */
+	drm_crtc_enable_color_mgmt(crtc, 0, true, crtc->gamma_size);
+
+	/* Set up some arbitrary number of planes.  We're not limited
+	 * by a set number of physical registers, just the space in
+	 * the HVS (16k) and how small an plane can be (28 bytes).
+	 * However, each plane we set up takes up some memory, and
+	 * increases the cost of looping over planes, which atomic
+	 * modesetting does quite a bit.  As a result, we pick a
+	 * modest number of planes to expose, that should hopefully
+	 * still cover any sane usecase.
+	 */
+	for (i = 0; i < 8; i++) {
+		struct drm_plane *plane =
+			vc4_plane_init(drm, DRM_PLANE_TYPE_OVERLAY);
+
+		if (IS_ERR(plane))
+			continue;
+
+		plane->possible_crtcs = drm_crtc_mask(crtc);
+	}
+
+	/* Set up the legacy cursor after overlay initialization,
+	 * since we overlay planes on the CRTC in the order they were
+	 * initialized.
+	 */
+	cursor_plane = vc4_plane_init(drm, DRM_PLANE_TYPE_CURSOR);
+	if (!IS_ERR(cursor_plane)) {
+		cursor_plane->possible_crtcs = drm_crtc_mask(crtc);
+		crtc->cursor = cursor_plane;
+	}
+
+	vc4_crtc_get_cob_allocation(vc4_crtc);
+
+	CRTC_WRITE(PV_INTEN, 0);
+	CRTC_WRITE(PV_INTSTAT, PV_INT_VFP_START);
+	ret = devm_request_irq(dev, platform_get_irq(pdev, 0),
+			       vc4_crtc_irq_handler, 0, "vc4 crtc", vc4_crtc);
+	if (ret)
+		goto err_destroy_planes;
+
+	vc4_set_crtc_possible_masks(drm, crtc);
+
+	for (i = 0; i < crtc->gamma_size; i++) {
+		vc4_crtc->lut_r[i] = i;
+		vc4_crtc->lut_g[i] = i;
+		vc4_crtc->lut_b[i] = i;
+	}
+
+	platform_set_drvdata(pdev, vc4_crtc);
+
+	return 0;
+
+err_destroy_planes:
+	list_for_each_entry_safe(destroy_plane, temp,
+				 &drm->mode_config.plane_list, head) {
+		if (destroy_plane->possible_crtcs == drm_crtc_mask(crtc))
+		    destroy_plane->funcs->destroy(destroy_plane);
+	}
+err:
+	return ret;
+}
+
+static void vc4_crtc_unbind(struct device *dev, struct device *master,
+			    void *data)
+{
+	struct platform_device *pdev = to_platform_device(dev);
+	struct vc4_crtc *vc4_crtc = dev_get_drvdata(dev);
+
+	vc4_crtc_destroy(&vc4_crtc->base);
+
+	CRTC_WRITE(PV_INTEN, 0);
+
+	platform_set_drvdata(pdev, NULL);
+}
+
+static const struct component_ops vc4_crtc_ops = {
+	.bind   = vc4_crtc_bind,
+	.unbind = vc4_crtc_unbind,
+};
+
+static int vc4_crtc_dev_probe(struct platform_device *pdev)
+{
+	return component_add(&pdev->dev, &vc4_crtc_ops);
+}
+
+static int vc4_crtc_dev_remove(struct platform_device *pdev)
+{
+	component_del(&pdev->dev, &vc4_crtc_ops);
+	return 0;
+}
+
+struct platform_driver vc4_crtc_driver = {
+	.probe = vc4_crtc_dev_probe,
+	.remove = vc4_crtc_dev_remove,
+	.driver = {
+		.name = "vc4_crtc",
+		.of_match_table = vc4_crtc_dt_match,
+	},
+};
diff --git a/drivers/gpu/drm/vc4/vc4_debugfs.c b/drivers/gpu/drm/vc4/vc4_debugfs.c
new file mode 100644
index 000000000..7a0003de7
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_debugfs.c
@@ -0,0 +1,40 @@
+/*
+ *  Copyright © 2014 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/seq_file.h>
+#include <linux/circ_buf.h>
+#include <linux/ctype.h>
+#include <linux/debugfs.h>
+#include <drm/drmP.h>
+
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+static const struct drm_info_list vc4_debugfs_list[] = {
+	{"bo_stats", vc4_bo_stats_debugfs, 0},
+	{"dpi_regs", vc4_dpi_debugfs_regs, 0},
+	{"dsi1_regs", vc4_dsi_debugfs_regs, 0, (void *)(uintptr_t)1},
+	{"hdmi_regs", vc4_hdmi_debugfs_regs, 0},
+	{"vec_regs", vc4_vec_debugfs_regs, 0},
+	{"txp_regs", vc4_txp_debugfs_regs, 0},
+	{"hvs_regs", vc4_hvs_debugfs_regs, 0},
+	{"crtc0_regs", vc4_crtc_debugfs_regs, 0, (void *)(uintptr_t)0},
+	{"crtc1_regs", vc4_crtc_debugfs_regs, 0, (void *)(uintptr_t)1},
+	{"crtc2_regs", vc4_crtc_debugfs_regs, 0, (void *)(uintptr_t)2},
+	{"v3d_ident", vc4_v3d_debugfs_ident, 0},
+	{"v3d_regs", vc4_v3d_debugfs_regs, 0},
+};
+
+#define VC4_DEBUGFS_ENTRIES ARRAY_SIZE(vc4_debugfs_list)
+
+int
+vc4_debugfs_init(struct drm_minor *minor)
+{
+	return drm_debugfs_create_files(vc4_debugfs_list, VC4_DEBUGFS_ENTRIES,
+					minor->debugfs_root, minor);
+}
diff --git a/drivers/gpu/drm/vc4/vc4_dpi.c b/drivers/gpu/drm/vc4/vc4_dpi.c
new file mode 100644
index 000000000..f18581297
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_dpi.c
@@ -0,0 +1,402 @@
+/*
+ * Copyright (C) 2016 Broadcom Limited
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published by
+ * the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+/**
+ * DOC: VC4 DPI module
+ *
+ * The VC4 DPI hardware supports MIPI DPI type 4 and Nokia ViSSI
+ * signals.  On BCM2835, these can be routed out to GPIO0-27 with the
+ * ALT2 function.
+ */
+
+#include <drm/drm_atomic_helper.h>
+#include <drm/drm_bridge.h>
+#include <drm/drm_crtc_helper.h>
+#include <drm/drm_edid.h>
+#include <drm/drm_of.h>
+#include <drm/drm_panel.h>
+#include <linux/clk.h>
+#include <linux/component.h>
+#include <linux/of_graph.h>
+#include <linux/of_platform.h>
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+#define DPI_C			0x00
+# define DPI_OUTPUT_ENABLE_MODE		BIT(16)
+
+/* The order field takes the incoming 24 bit RGB from the pixel valve
+ * and shuffles the 3 channels.
+ */
+# define DPI_ORDER_MASK			VC4_MASK(15, 14)
+# define DPI_ORDER_SHIFT		14
+# define DPI_ORDER_RGB			0
+# define DPI_ORDER_BGR			1
+# define DPI_ORDER_GRB			2
+# define DPI_ORDER_BRG			3
+
+/* The format field takes the ORDER-shuffled pixel valve data and
+ * formats it onto the output lines.
+ */
+# define DPI_FORMAT_MASK		VC4_MASK(13, 11)
+# define DPI_FORMAT_SHIFT		11
+/* This define is named in the hardware, but actually just outputs 0. */
+# define DPI_FORMAT_9BIT_666_RGB	0
+/* Outputs 00000000rrrrrggggggbbbbb */
+# define DPI_FORMAT_16BIT_565_RGB_1	1
+/* Outputs 000rrrrr00gggggg000bbbbb */
+# define DPI_FORMAT_16BIT_565_RGB_2	2
+/* Outputs 00rrrrr000gggggg00bbbbb0 */
+# define DPI_FORMAT_16BIT_565_RGB_3	3
+/* Outputs 000000rrrrrrggggggbbbbbb */
+# define DPI_FORMAT_18BIT_666_RGB_1	4
+/* Outputs 00rrrrrr00gggggg00bbbbbb */
+# define DPI_FORMAT_18BIT_666_RGB_2	5
+/* Outputs rrrrrrrrggggggggbbbbbbbb */
+# define DPI_FORMAT_24BIT_888_RGB	6
+
+/* Reverses the polarity of the corresponding signal */
+# define DPI_PIXEL_CLK_INVERT		BIT(10)
+# define DPI_HSYNC_INVERT		BIT(9)
+# define DPI_VSYNC_INVERT		BIT(8)
+# define DPI_OUTPUT_ENABLE_INVERT	BIT(7)
+
+/* Outputs the signal the falling clock edge instead of rising. */
+# define DPI_HSYNC_NEGATE		BIT(6)
+# define DPI_VSYNC_NEGATE		BIT(5)
+# define DPI_OUTPUT_ENABLE_NEGATE	BIT(4)
+
+/* Disables the signal */
+# define DPI_HSYNC_DISABLE		BIT(3)
+# define DPI_VSYNC_DISABLE		BIT(2)
+# define DPI_OUTPUT_ENABLE_DISABLE	BIT(1)
+
+/* Power gate to the device, full reset at 0 -> 1 transition */
+# define DPI_ENABLE			BIT(0)
+
+/* All other registers besides DPI_C return the ID */
+#define DPI_ID			0x04
+# define DPI_ID_VALUE		0x00647069
+
+/* General DPI hardware state. */
+struct vc4_dpi {
+	struct platform_device *pdev;
+
+	struct drm_encoder *encoder;
+
+	void __iomem *regs;
+
+	struct clk *pixel_clock;
+	struct clk *core_clock;
+};
+
+#define DPI_READ(offset) readl(dpi->regs + (offset))
+#define DPI_WRITE(offset, val) writel(val, dpi->regs + (offset))
+
+/* VC4 DPI encoder KMS struct */
+struct vc4_dpi_encoder {
+	struct vc4_encoder base;
+	struct vc4_dpi *dpi;
+};
+
+static inline struct vc4_dpi_encoder *
+to_vc4_dpi_encoder(struct drm_encoder *encoder)
+{
+	return container_of(encoder, struct vc4_dpi_encoder, base.base);
+}
+
+#define DPI_REG(reg) { reg, #reg }
+static const struct {
+	u32 reg;
+	const char *name;
+} dpi_regs[] = {
+	DPI_REG(DPI_C),
+	DPI_REG(DPI_ID),
+};
+
+#ifdef CONFIG_DEBUG_FS
+int vc4_dpi_debugfs_regs(struct seq_file *m, void *unused)
+{
+	struct drm_info_node *node = (struct drm_info_node *)m->private;
+	struct drm_device *dev = node->minor->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_dpi *dpi = vc4->dpi;
+	int i;
+
+	if (!dpi)
+		return 0;
+
+	for (i = 0; i < ARRAY_SIZE(dpi_regs); i++) {
+		seq_printf(m, "%s (0x%04x): 0x%08x\n",
+			   dpi_regs[i].name, dpi_regs[i].reg,
+			   DPI_READ(dpi_regs[i].reg));
+	}
+
+	return 0;
+}
+#endif
+
+static const struct drm_encoder_funcs vc4_dpi_encoder_funcs = {
+	.destroy = drm_encoder_cleanup,
+};
+
+static void vc4_dpi_encoder_disable(struct drm_encoder *encoder)
+{
+	struct vc4_dpi_encoder *vc4_encoder = to_vc4_dpi_encoder(encoder);
+	struct vc4_dpi *dpi = vc4_encoder->dpi;
+
+	clk_disable_unprepare(dpi->pixel_clock);
+}
+
+static void vc4_dpi_encoder_enable(struct drm_encoder *encoder)
+{
+	struct drm_device *dev = encoder->dev;
+	struct drm_display_mode *mode = &encoder->crtc->mode;
+	struct vc4_dpi_encoder *vc4_encoder = to_vc4_dpi_encoder(encoder);
+	struct vc4_dpi *dpi = vc4_encoder->dpi;
+	struct drm_connector_list_iter conn_iter;
+	struct drm_connector *connector = NULL, *connector_scan;
+	u32 dpi_c = DPI_ENABLE | DPI_OUTPUT_ENABLE_MODE;
+	int ret;
+
+	/* Look up the connector attached to DPI so we can get the
+	 * bus_format.  Ideally the bridge would tell us the
+	 * bus_format we want, but it doesn't yet, so assume that it's
+	 * uniform throughout the bridge chain.
+	 */
+	drm_connector_list_iter_begin(dev, &conn_iter);
+	drm_for_each_connector_iter(connector_scan, &conn_iter) {
+		if (connector_scan->encoder == encoder) {
+			connector = connector_scan;
+			break;
+		}
+	}
+	drm_connector_list_iter_end(&conn_iter);
+
+	if (connector && connector->display_info.num_bus_formats) {
+		u32 bus_format = connector->display_info.bus_formats[0];
+
+		switch (bus_format) {
+		case MEDIA_BUS_FMT_RGB888_1X24:
+			dpi_c |= VC4_SET_FIELD(DPI_FORMAT_24BIT_888_RGB,
+					       DPI_FORMAT);
+			break;
+		case MEDIA_BUS_FMT_BGR888_1X24:
+			dpi_c |= VC4_SET_FIELD(DPI_FORMAT_24BIT_888_RGB,
+					       DPI_FORMAT);
+			dpi_c |= VC4_SET_FIELD(DPI_ORDER_BGR, DPI_ORDER);
+			break;
+		case MEDIA_BUS_FMT_RGB666_1X24_CPADHI:
+			dpi_c |= VC4_SET_FIELD(DPI_FORMAT_18BIT_666_RGB_2,
+					       DPI_FORMAT);
+			break;
+		case MEDIA_BUS_FMT_RGB666_1X18:
+			dpi_c |= VC4_SET_FIELD(DPI_FORMAT_18BIT_666_RGB_1,
+					       DPI_FORMAT);
+			break;
+		case MEDIA_BUS_FMT_RGB565_1X16:
+			dpi_c |= VC4_SET_FIELD(DPI_FORMAT_16BIT_565_RGB_3,
+					       DPI_FORMAT);
+			break;
+		default:
+			DRM_ERROR("Unknown media bus format %d\n", bus_format);
+			break;
+		}
+	} else {
+		/* Default to 24bit if no connector found. */
+		dpi_c |= VC4_SET_FIELD(DPI_FORMAT_24BIT_888_RGB, DPI_FORMAT);
+	}
+
+	if (mode->flags & DRM_MODE_FLAG_NHSYNC)
+		dpi_c |= DPI_HSYNC_INVERT;
+	else if (!(mode->flags & DRM_MODE_FLAG_PHSYNC))
+		dpi_c |= DPI_HSYNC_DISABLE;
+
+	if (mode->flags & DRM_MODE_FLAG_NVSYNC)
+		dpi_c |= DPI_VSYNC_INVERT;
+	else if (!(mode->flags & DRM_MODE_FLAG_PVSYNC))
+		dpi_c |= DPI_VSYNC_DISABLE;
+
+	DPI_WRITE(DPI_C, dpi_c);
+
+	ret = clk_set_rate(dpi->pixel_clock, mode->clock * 1000);
+	if (ret)
+		DRM_ERROR("Failed to set clock rate: %d\n", ret);
+
+	ret = clk_prepare_enable(dpi->pixel_clock);
+	if (ret)
+		DRM_ERROR("Failed to set clock rate: %d\n", ret);
+}
+
+static enum drm_mode_status vc4_dpi_encoder_mode_valid(struct drm_encoder *encoder,
+						       const struct drm_display_mode *mode)
+{
+	if (mode->flags & DRM_MODE_FLAG_INTERLACE)
+		return MODE_NO_INTERLACE;
+
+	return MODE_OK;
+}
+
+static const struct drm_encoder_helper_funcs vc4_dpi_encoder_helper_funcs = {
+	.disable = vc4_dpi_encoder_disable,
+	.enable = vc4_dpi_encoder_enable,
+	.mode_valid = vc4_dpi_encoder_mode_valid,
+};
+
+static const struct of_device_id vc4_dpi_dt_match[] = {
+	{ .compatible = "brcm,bcm2835-dpi", .data = NULL },
+	{}
+};
+
+/* Sets up the next link in the display chain, whether it's a panel or
+ * a bridge.
+ */
+static int vc4_dpi_init_bridge(struct vc4_dpi *dpi)
+{
+	struct device *dev = &dpi->pdev->dev;
+	struct drm_panel *panel;
+	struct drm_bridge *bridge;
+	int ret;
+
+	ret = drm_of_find_panel_or_bridge(dev->of_node, 0, 0,
+					  &panel, &bridge);
+	if (ret) {
+		/* If nothing was connected in the DT, that's not an
+		 * error.
+		 */
+		if (ret == -ENODEV)
+			return 0;
+		else
+			return ret;
+	}
+
+	if (panel)
+		bridge = drm_panel_bridge_add(panel, DRM_MODE_CONNECTOR_DPI);
+
+	return drm_bridge_attach(dpi->encoder, bridge, NULL);
+}
+
+static int vc4_dpi_bind(struct device *dev, struct device *master, void *data)
+{
+	struct platform_device *pdev = to_platform_device(dev);
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	struct vc4_dpi *dpi;
+	struct vc4_dpi_encoder *vc4_dpi_encoder;
+	int ret;
+
+	dpi = devm_kzalloc(dev, sizeof(*dpi), GFP_KERNEL);
+	if (!dpi)
+		return -ENOMEM;
+
+	vc4_dpi_encoder = devm_kzalloc(dev, sizeof(*vc4_dpi_encoder),
+				       GFP_KERNEL);
+	if (!vc4_dpi_encoder)
+		return -ENOMEM;
+	vc4_dpi_encoder->base.type = VC4_ENCODER_TYPE_DPI;
+	vc4_dpi_encoder->dpi = dpi;
+	dpi->encoder = &vc4_dpi_encoder->base.base;
+
+	dpi->pdev = pdev;
+	dpi->regs = vc4_ioremap_regs(pdev, 0);
+	if (IS_ERR(dpi->regs))
+		return PTR_ERR(dpi->regs);
+
+	if (DPI_READ(DPI_ID) != DPI_ID_VALUE) {
+		dev_err(dev, "Port returned 0x%08x for ID instead of 0x%08x\n",
+			DPI_READ(DPI_ID), DPI_ID_VALUE);
+		return -ENODEV;
+	}
+
+	dpi->core_clock = devm_clk_get(dev, "core");
+	if (IS_ERR(dpi->core_clock)) {
+		ret = PTR_ERR(dpi->core_clock);
+		if (ret != -EPROBE_DEFER)
+			DRM_ERROR("Failed to get core clock: %d\n", ret);
+		return ret;
+	}
+	dpi->pixel_clock = devm_clk_get(dev, "pixel");
+	if (IS_ERR(dpi->pixel_clock)) {
+		ret = PTR_ERR(dpi->pixel_clock);
+		if (ret != -EPROBE_DEFER)
+			DRM_ERROR("Failed to get pixel clock: %d\n", ret);
+		return ret;
+	}
+
+	ret = clk_prepare_enable(dpi->core_clock);
+	if (ret)
+		DRM_ERROR("Failed to turn on core clock: %d\n", ret);
+
+	drm_encoder_init(drm, dpi->encoder, &vc4_dpi_encoder_funcs,
+			 DRM_MODE_ENCODER_DPI, NULL);
+	drm_encoder_helper_add(dpi->encoder, &vc4_dpi_encoder_helper_funcs);
+
+	ret = vc4_dpi_init_bridge(dpi);
+	if (ret)
+		goto err_destroy_encoder;
+
+	dev_set_drvdata(dev, dpi);
+
+	vc4->dpi = dpi;
+
+	return 0;
+
+err_destroy_encoder:
+	drm_encoder_cleanup(dpi->encoder);
+	clk_disable_unprepare(dpi->core_clock);
+	return ret;
+}
+
+static void vc4_dpi_unbind(struct device *dev, struct device *master,
+			   void *data)
+{
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	struct vc4_dpi *dpi = dev_get_drvdata(dev);
+
+	drm_of_panel_bridge_remove(dev->of_node, 0, 0);
+
+	drm_encoder_cleanup(dpi->encoder);
+
+	clk_disable_unprepare(dpi->core_clock);
+
+	vc4->dpi = NULL;
+}
+
+static const struct component_ops vc4_dpi_ops = {
+	.bind   = vc4_dpi_bind,
+	.unbind = vc4_dpi_unbind,
+};
+
+static int vc4_dpi_dev_probe(struct platform_device *pdev)
+{
+	return component_add(&pdev->dev, &vc4_dpi_ops);
+}
+
+static int vc4_dpi_dev_remove(struct platform_device *pdev)
+{
+	component_del(&pdev->dev, &vc4_dpi_ops);
+	return 0;
+}
+
+struct platform_driver vc4_dpi_driver = {
+	.probe = vc4_dpi_dev_probe,
+	.remove = vc4_dpi_dev_remove,
+	.driver = {
+		.name = "vc4_dpi",
+		.of_match_table = vc4_dpi_dt_match,
+	},
+};
diff --git a/drivers/gpu/drm/vc4/vc4_drv.c b/drivers/gpu/drm/vc4/vc4_drv.c
new file mode 100644
index 000000000..868dd1ef3
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_drv.c
@@ -0,0 +1,413 @@
+/*
+ * Copyright (C) 2014-2015 Broadcom
+ * Copyright (C) 2013 Red Hat
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+/**
+ * DOC: Broadcom VC4 Graphics Driver
+ *
+ * The Broadcom VideoCore 4 (present in the Raspberry Pi) contains a
+ * OpenGL ES 2.0-compatible 3D engine called V3D, and a highly
+ * configurable display output pipeline that supports HDMI, DSI, DPI,
+ * and Composite TV output.
+ *
+ * The 3D engine also has an interface for submitting arbitrary
+ * compute shader-style jobs using the same shader processor as is
+ * used for vertex and fragment shaders in GLES 2.0.  However, given
+ * that the hardware isn't able to expose any standard interfaces like
+ * OpenGL compute shaders or OpenCL, it isn't supported by this
+ * driver.
+ */
+
+#include <linux/clk.h>
+#include <linux/component.h>
+#include <linux/device.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/of_platform.h>
+#include <linux/platform_device.h>
+#include <linux/pm_runtime.h>
+#include <drm/drm_fb_cma_helper.h>
+#include <drm/drm_fb_helper.h>
+
+#include "uapi/drm/vc4_drm.h"
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+#define DRIVER_NAME "vc4"
+#define DRIVER_DESC "Broadcom VC4 graphics"
+#define DRIVER_DATE "20140616"
+#define DRIVER_MAJOR 0
+#define DRIVER_MINOR 0
+#define DRIVER_PATCHLEVEL 0
+
+/* Helper function for mapping the regs on a platform device. */
+void __iomem *vc4_ioremap_regs(struct platform_device *dev, int index)
+{
+	struct resource *res;
+	void __iomem *map;
+
+	res = platform_get_resource(dev, IORESOURCE_MEM, index);
+	map = devm_ioremap_resource(&dev->dev, res);
+	if (IS_ERR(map)) {
+		DRM_ERROR("Failed to map registers: %ld\n", PTR_ERR(map));
+		return map;
+	}
+
+	return map;
+}
+
+static int vc4_get_param_ioctl(struct drm_device *dev, void *data,
+			       struct drm_file *file_priv)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct drm_vc4_get_param *args = data;
+	int ret;
+
+	if (args->pad != 0)
+		return -EINVAL;
+
+	switch (args->param) {
+	case DRM_VC4_PARAM_V3D_IDENT0:
+		ret = pm_runtime_get_sync(&vc4->v3d->pdev->dev);
+		if (ret < 0)
+			return ret;
+		args->value = V3D_READ(V3D_IDENT0);
+		pm_runtime_mark_last_busy(&vc4->v3d->pdev->dev);
+		pm_runtime_put_autosuspend(&vc4->v3d->pdev->dev);
+		break;
+	case DRM_VC4_PARAM_V3D_IDENT1:
+		ret = pm_runtime_get_sync(&vc4->v3d->pdev->dev);
+		if (ret < 0)
+			return ret;
+		args->value = V3D_READ(V3D_IDENT1);
+		pm_runtime_mark_last_busy(&vc4->v3d->pdev->dev);
+		pm_runtime_put_autosuspend(&vc4->v3d->pdev->dev);
+		break;
+	case DRM_VC4_PARAM_V3D_IDENT2:
+		ret = pm_runtime_get_sync(&vc4->v3d->pdev->dev);
+		if (ret < 0)
+			return ret;
+		args->value = V3D_READ(V3D_IDENT2);
+		pm_runtime_mark_last_busy(&vc4->v3d->pdev->dev);
+		pm_runtime_put_autosuspend(&vc4->v3d->pdev->dev);
+		break;
+	case DRM_VC4_PARAM_SUPPORTS_BRANCHES:
+	case DRM_VC4_PARAM_SUPPORTS_ETC1:
+	case DRM_VC4_PARAM_SUPPORTS_THREADED_FS:
+	case DRM_VC4_PARAM_SUPPORTS_FIXED_RCL_ORDER:
+	case DRM_VC4_PARAM_SUPPORTS_MADVISE:
+	case DRM_VC4_PARAM_SUPPORTS_PERFMON:
+		args->value = true;
+		break;
+	default:
+		DRM_DEBUG("Unknown parameter %d\n", args->param);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int vc4_open(struct drm_device *dev, struct drm_file *file)
+{
+	struct vc4_file *vc4file;
+
+	vc4file = kzalloc(sizeof(*vc4file), GFP_KERNEL);
+	if (!vc4file)
+		return -ENOMEM;
+
+	vc4_perfmon_open_file(vc4file);
+	file->driver_priv = vc4file;
+	return 0;
+}
+
+static void vc4_close(struct drm_device *dev, struct drm_file *file)
+{
+	struct vc4_file *vc4file = file->driver_priv;
+
+	vc4_perfmon_close_file(vc4file);
+	kfree(vc4file);
+}
+
+static const struct vm_operations_struct vc4_vm_ops = {
+	.fault = vc4_fault,
+	.open = drm_gem_vm_open,
+	.close = drm_gem_vm_close,
+};
+
+static const struct file_operations vc4_drm_fops = {
+	.owner = THIS_MODULE,
+	.open = drm_open,
+	.release = drm_release,
+	.unlocked_ioctl = drm_ioctl,
+	.mmap = vc4_mmap,
+	.poll = drm_poll,
+	.read = drm_read,
+	.compat_ioctl = drm_compat_ioctl,
+	.llseek = noop_llseek,
+};
+
+static const struct drm_ioctl_desc vc4_drm_ioctls[] = {
+	DRM_IOCTL_DEF_DRV(VC4_SUBMIT_CL, vc4_submit_cl_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_WAIT_SEQNO, vc4_wait_seqno_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_WAIT_BO, vc4_wait_bo_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_CREATE_BO, vc4_create_bo_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_MMAP_BO, vc4_mmap_bo_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_CREATE_SHADER_BO, vc4_create_shader_bo_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_GET_HANG_STATE, vc4_get_hang_state_ioctl,
+			  DRM_ROOT_ONLY),
+	DRM_IOCTL_DEF_DRV(VC4_GET_PARAM, vc4_get_param_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_SET_TILING, vc4_set_tiling_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_GET_TILING, vc4_get_tiling_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_LABEL_BO, vc4_label_bo_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_GEM_MADVISE, vc4_gem_madvise_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_PERFMON_CREATE, vc4_perfmon_create_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_PERFMON_DESTROY, vc4_perfmon_destroy_ioctl, DRM_RENDER_ALLOW),
+	DRM_IOCTL_DEF_DRV(VC4_PERFMON_GET_VALUES, vc4_perfmon_get_values_ioctl, DRM_RENDER_ALLOW),
+};
+
+static struct drm_driver vc4_drm_driver = {
+	.driver_features = (DRIVER_MODESET |
+			    DRIVER_ATOMIC |
+			    DRIVER_GEM |
+			    DRIVER_HAVE_IRQ |
+			    DRIVER_RENDER |
+			    DRIVER_PRIME |
+			    DRIVER_SYNCOBJ),
+	.lastclose = drm_fb_helper_lastclose,
+	.open = vc4_open,
+	.postclose = vc4_close,
+	.irq_handler = vc4_irq,
+	.irq_preinstall = vc4_irq_preinstall,
+	.irq_postinstall = vc4_irq_postinstall,
+	.irq_uninstall = vc4_irq_uninstall,
+
+	.get_scanout_position = vc4_crtc_get_scanoutpos,
+	.get_vblank_timestamp = drm_calc_vbltimestamp_from_scanoutpos,
+
+#if defined(CONFIG_DEBUG_FS)
+	.debugfs_init = vc4_debugfs_init,
+#endif
+
+	.gem_create_object = vc4_create_object,
+	.gem_free_object_unlocked = vc4_free_object,
+	.gem_vm_ops = &vc4_vm_ops,
+
+	.prime_handle_to_fd = drm_gem_prime_handle_to_fd,
+	.prime_fd_to_handle = drm_gem_prime_fd_to_handle,
+	.gem_prime_import = drm_gem_prime_import,
+	.gem_prime_export = vc4_prime_export,
+	.gem_prime_res_obj = vc4_prime_res_obj,
+	.gem_prime_get_sg_table	= drm_gem_cma_prime_get_sg_table,
+	.gem_prime_import_sg_table = vc4_prime_import_sg_table,
+	.gem_prime_vmap = vc4_prime_vmap,
+	.gem_prime_vunmap = drm_gem_cma_prime_vunmap,
+	.gem_prime_mmap = vc4_prime_mmap,
+
+	.dumb_create = vc4_dumb_create,
+
+	.ioctls = vc4_drm_ioctls,
+	.num_ioctls = ARRAY_SIZE(vc4_drm_ioctls),
+	.fops = &vc4_drm_fops,
+
+	.name = DRIVER_NAME,
+	.desc = DRIVER_DESC,
+	.date = DRIVER_DATE,
+	.major = DRIVER_MAJOR,
+	.minor = DRIVER_MINOR,
+	.patchlevel = DRIVER_PATCHLEVEL,
+};
+
+static int compare_dev(struct device *dev, void *data)
+{
+	return dev == data;
+}
+
+static void vc4_match_add_drivers(struct device *dev,
+				  struct component_match **match,
+				  struct platform_driver *const *drivers,
+				  int count)
+{
+	int i;
+
+	for (i = 0; i < count; i++) {
+		struct device_driver *drv = &drivers[i]->driver;
+		struct device *p = NULL, *d;
+
+		while ((d = bus_find_device(&platform_bus_type, p, drv,
+					    (void *)platform_bus_type.match))) {
+			put_device(p);
+			component_match_add(dev, match, compare_dev, d);
+			p = d;
+		}
+		put_device(p);
+	}
+}
+
+static void vc4_kick_out_firmware_fb(void)
+{
+	struct apertures_struct *ap;
+
+	ap = alloc_apertures(1);
+	if (!ap)
+		return;
+
+	/* Since VC4 is a UMA device, the simplefb node may have been
+	 * located anywhere in memory.
+	 */
+	ap->ranges[0].base = 0;
+	ap->ranges[0].size = ~0;
+
+	drm_fb_helper_remove_conflicting_framebuffers(ap, "vc4drmfb", false);
+	kfree(ap);
+}
+
+static int vc4_drm_bind(struct device *dev)
+{
+	struct platform_device *pdev = to_platform_device(dev);
+	struct drm_device *drm;
+	struct vc4_dev *vc4;
+	int ret = 0;
+
+	dev->coherent_dma_mask = DMA_BIT_MASK(32);
+
+	vc4 = devm_kzalloc(dev, sizeof(*vc4), GFP_KERNEL);
+	if (!vc4)
+		return -ENOMEM;
+
+	drm = drm_dev_alloc(&vc4_drm_driver, dev);
+	if (IS_ERR(drm))
+		return PTR_ERR(drm);
+	platform_set_drvdata(pdev, drm);
+	vc4->dev = drm;
+	drm->dev_private = vc4;
+
+	ret = vc4_bo_cache_init(drm);
+	if (ret)
+		goto dev_put;
+
+	drm_mode_config_init(drm);
+
+	vc4_gem_init(drm);
+
+	ret = component_bind_all(dev, drm);
+	if (ret)
+		goto gem_destroy;
+
+	vc4_kick_out_firmware_fb();
+
+	ret = drm_dev_register(drm, 0);
+	if (ret < 0)
+		goto unbind_all;
+
+	vc4_kms_load(drm);
+
+	return 0;
+
+unbind_all:
+	component_unbind_all(dev, drm);
+gem_destroy:
+	vc4_gem_destroy(drm);
+	drm_mode_config_cleanup(drm);
+	vc4_bo_cache_destroy(drm);
+dev_put:
+	drm_dev_put(drm);
+	return ret;
+}
+
+static void vc4_drm_unbind(struct device *dev)
+{
+	struct drm_device *drm = dev_get_drvdata(dev);
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+
+	drm_dev_unregister(drm);
+
+	drm_fb_cma_fbdev_fini(drm);
+
+	drm_mode_config_cleanup(drm);
+
+	drm_atomic_private_obj_fini(&vc4->ctm_manager);
+
+	drm_dev_put(drm);
+}
+
+static const struct component_master_ops vc4_drm_ops = {
+	.bind = vc4_drm_bind,
+	.unbind = vc4_drm_unbind,
+};
+
+static struct platform_driver *const component_drivers[] = {
+	&vc4_hdmi_driver,
+	&vc4_vec_driver,
+	&vc4_dpi_driver,
+	&vc4_dsi_driver,
+	&vc4_txp_driver,
+	&vc4_hvs_driver,
+	&vc4_crtc_driver,
+	&vc4_v3d_driver,
+};
+
+static int vc4_platform_drm_probe(struct platform_device *pdev)
+{
+	struct component_match *match = NULL;
+	struct device *dev = &pdev->dev;
+
+	vc4_match_add_drivers(dev, &match,
+			      component_drivers, ARRAY_SIZE(component_drivers));
+
+	return component_master_add_with_match(dev, &vc4_drm_ops, match);
+}
+
+static int vc4_platform_drm_remove(struct platform_device *pdev)
+{
+	component_master_del(&pdev->dev, &vc4_drm_ops);
+
+	return 0;
+}
+
+static const struct of_device_id vc4_of_match[] = {
+	{ .compatible = "brcm,bcm2835-vc4", },
+	{ .compatible = "brcm,cygnus-vc4", },
+	{},
+};
+MODULE_DEVICE_TABLE(of, vc4_of_match);
+
+static struct platform_driver vc4_platform_driver = {
+	.probe		= vc4_platform_drm_probe,
+	.remove		= vc4_platform_drm_remove,
+	.driver		= {
+		.name	= "vc4-drm",
+		.of_match_table = vc4_of_match,
+	},
+};
+
+static int __init vc4_drm_register(void)
+{
+	int ret;
+
+	ret = platform_register_drivers(component_drivers,
+					ARRAY_SIZE(component_drivers));
+	if (ret)
+		return ret;
+
+	return platform_driver_register(&vc4_platform_driver);
+}
+
+static void __exit vc4_drm_unregister(void)
+{
+	platform_unregister_drivers(component_drivers,
+				    ARRAY_SIZE(component_drivers));
+	platform_driver_unregister(&vc4_platform_driver);
+}
+
+module_init(vc4_drm_register);
+module_exit(vc4_drm_unregister);
+
+MODULE_ALIAS("platform:vc4-drm");
+MODULE_DESCRIPTION("Broadcom VC4 DRM Driver");
+MODULE_AUTHOR("Eric Anholt <eric@anholt.net>");
+MODULE_LICENSE("GPL v2");
diff --git a/drivers/gpu/drm/vc4/vc4_drv.h b/drivers/gpu/drm/vc4/vc4_drv.h
new file mode 100644
index 000000000..bd6ef1f31
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_drv.h
@@ -0,0 +1,822 @@
+/*
+ * Copyright (C) 2015 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/mm_types.h>
+#include <linux/reservation.h>
+#include <drm/drmP.h>
+#include <drm/drm_encoder.h>
+#include <drm/drm_gem_cma_helper.h>
+#include <drm/drm_atomic.h>
+#include <drm/drm_syncobj.h>
+
+#include "uapi/drm/vc4_drm.h"
+
+/* Don't forget to update vc4_bo.c: bo_type_names[] when adding to
+ * this.
+ */
+enum vc4_kernel_bo_type {
+	/* Any kernel allocation (gem_create_object hook) before it
+	 * gets another type set.
+	 */
+	VC4_BO_TYPE_KERNEL,
+	VC4_BO_TYPE_V3D,
+	VC4_BO_TYPE_V3D_SHADER,
+	VC4_BO_TYPE_DUMB,
+	VC4_BO_TYPE_BIN,
+	VC4_BO_TYPE_RCL,
+	VC4_BO_TYPE_BCL,
+	VC4_BO_TYPE_KERNEL_CACHE,
+	VC4_BO_TYPE_COUNT
+};
+
+/* Performance monitor object. The perform lifetime is controlled by userspace
+ * using perfmon related ioctls. A perfmon can be attached to a submit_cl
+ * request, and when this is the case, HW perf counters will be activated just
+ * before the submit_cl is submitted to the GPU and disabled when the job is
+ * done. This way, only events related to a specific job will be counted.
+ */
+struct vc4_perfmon {
+	/* Tracks the number of users of the perfmon, when this counter reaches
+	 * zero the perfmon is destroyed.
+	 */
+	refcount_t refcnt;
+
+	/* Number of counters activated in this perfmon instance
+	 * (should be less than DRM_VC4_MAX_PERF_COUNTERS).
+	 */
+	u8 ncounters;
+
+	/* Events counted by the HW perf counters. */
+	u8 events[DRM_VC4_MAX_PERF_COUNTERS];
+
+	/* Storage for counter values. Counters are incremented by the HW
+	 * perf counter values every time the perfmon is attached to a GPU job.
+	 * This way, perfmon users don't have to retrieve the results after
+	 * each job if they want to track events covering several submissions.
+	 * Note that counter values can't be reset, but you can fake a reset by
+	 * destroying the perfmon and creating a new one.
+	 */
+	u64 counters[0];
+};
+
+struct vc4_dev {
+	struct drm_device *dev;
+
+	struct vc4_hdmi *hdmi;
+	struct vc4_hvs *hvs;
+	struct vc4_v3d *v3d;
+	struct vc4_dpi *dpi;
+	struct vc4_dsi *dsi1;
+	struct vc4_vec *vec;
+	struct vc4_txp *txp;
+
+	struct vc4_hang_state *hang_state;
+
+	/* The kernel-space BO cache.  Tracks buffers that have been
+	 * unreferenced by all other users (refcounts of 0!) but not
+	 * yet freed, so we can do cheap allocations.
+	 */
+	struct vc4_bo_cache {
+		/* Array of list heads for entries in the BO cache,
+		 * based on number of pages, so we can do O(1) lookups
+		 * in the cache when allocating.
+		 */
+		struct list_head *size_list;
+		uint32_t size_list_size;
+
+		/* List of all BOs in the cache, ordered by age, so we
+		 * can do O(1) lookups when trying to free old
+		 * buffers.
+		 */
+		struct list_head time_list;
+		struct work_struct time_work;
+		struct timer_list time_timer;
+	} bo_cache;
+
+	u32 num_labels;
+	struct vc4_label {
+		const char *name;
+		u32 num_allocated;
+		u32 size_allocated;
+	} *bo_labels;
+
+	/* Protects bo_cache and bo_labels. */
+	struct mutex bo_lock;
+
+	/* Purgeable BO pool. All BOs in this pool can have their memory
+	 * reclaimed if the driver is unable to allocate new BOs. We also
+	 * keep stats related to the purge mechanism here.
+	 */
+	struct {
+		struct list_head list;
+		unsigned int num;
+		size_t size;
+		unsigned int purged_num;
+		size_t purged_size;
+		struct mutex lock;
+	} purgeable;
+
+	uint64_t dma_fence_context;
+
+	/* Sequence number for the last job queued in bin_job_list.
+	 * Starts at 0 (no jobs emitted).
+	 */
+	uint64_t emit_seqno;
+
+	/* Sequence number for the last completed job on the GPU.
+	 * Starts at 0 (no jobs completed).
+	 */
+	uint64_t finished_seqno;
+
+	/* List of all struct vc4_exec_info for jobs to be executed in
+	 * the binner.  The first job in the list is the one currently
+	 * programmed into ct0ca for execution.
+	 */
+	struct list_head bin_job_list;
+
+	/* List of all struct vc4_exec_info for jobs that have
+	 * completed binning and are ready for rendering.  The first
+	 * job in the list is the one currently programmed into ct1ca
+	 * for execution.
+	 */
+	struct list_head render_job_list;
+
+	/* List of the finished vc4_exec_infos waiting to be freed by
+	 * job_done_work.
+	 */
+	struct list_head job_done_list;
+	/* Spinlock used to synchronize the job_list and seqno
+	 * accesses between the IRQ handler and GEM ioctls.
+	 */
+	spinlock_t job_lock;
+	wait_queue_head_t job_wait_queue;
+	struct work_struct job_done_work;
+
+	/* Used to track the active perfmon if any. Access to this field is
+	 * protected by job_lock.
+	 */
+	struct vc4_perfmon *active_perfmon;
+
+	/* List of struct vc4_seqno_cb for callbacks to be made from a
+	 * workqueue when the given seqno is passed.
+	 */
+	struct list_head seqno_cb_list;
+
+	/* The memory used for storing binner tile alloc, tile state,
+	 * and overflow memory allocations.  This is freed when V3D
+	 * powers down.
+	 */
+	struct vc4_bo *bin_bo;
+
+	/* Size of blocks allocated within bin_bo. */
+	uint32_t bin_alloc_size;
+
+	/* Bitmask of the bin_alloc_size chunks in bin_bo that are
+	 * used.
+	 */
+	uint32_t bin_alloc_used;
+
+	/* Bitmask of the current bin_alloc used for overflow memory. */
+	uint32_t bin_alloc_overflow;
+
+	struct work_struct overflow_mem_work;
+
+	int power_refcount;
+
+	/* Mutex controlling the power refcount. */
+	struct mutex power_lock;
+
+	struct {
+		struct timer_list timer;
+		struct work_struct reset_work;
+	} hangcheck;
+
+	struct semaphore async_modeset;
+
+	struct drm_modeset_lock ctm_state_lock;
+	struct drm_private_obj ctm_manager;
+};
+
+static inline struct vc4_dev *
+to_vc4_dev(struct drm_device *dev)
+{
+	return (struct vc4_dev *)dev->dev_private;
+}
+
+struct vc4_bo {
+	struct drm_gem_cma_object base;
+
+	/* seqno of the last job to render using this BO. */
+	uint64_t seqno;
+
+	/* seqno of the last job to use the RCL to write to this BO.
+	 *
+	 * Note that this doesn't include binner overflow memory
+	 * writes.
+	 */
+	uint64_t write_seqno;
+
+	bool t_format;
+
+	/* List entry for the BO's position in either
+	 * vc4_exec_info->unref_list or vc4_dev->bo_cache.time_list
+	 */
+	struct list_head unref_head;
+
+	/* Time in jiffies when the BO was put in vc4->bo_cache. */
+	unsigned long free_time;
+
+	/* List entry for the BO's position in vc4_dev->bo_cache.size_list */
+	struct list_head size_head;
+
+	/* Struct for shader validation state, if created by
+	 * DRM_IOCTL_VC4_CREATE_SHADER_BO.
+	 */
+	struct vc4_validated_shader_info *validated_shader;
+
+	/* normally (resv == &_resv) except for imported bo's */
+	struct reservation_object *resv;
+	struct reservation_object _resv;
+
+	/* One of enum vc4_kernel_bo_type, or VC4_BO_TYPE_COUNT + i
+	 * for user-allocated labels.
+	 */
+	int label;
+
+	/* Count the number of active users. This is needed to determine
+	 * whether we can move the BO to the purgeable list or not (when the BO
+	 * is used by the GPU or the display engine we can't purge it).
+	 */
+	refcount_t usecnt;
+
+	/* Store purgeable/purged state here */
+	u32 madv;
+	struct mutex madv_lock;
+};
+
+static inline struct vc4_bo *
+to_vc4_bo(struct drm_gem_object *bo)
+{
+	return (struct vc4_bo *)bo;
+}
+
+struct vc4_fence {
+	struct dma_fence base;
+	struct drm_device *dev;
+	/* vc4 seqno for signaled() test */
+	uint64_t seqno;
+};
+
+static inline struct vc4_fence *
+to_vc4_fence(struct dma_fence *fence)
+{
+	return (struct vc4_fence *)fence;
+}
+
+struct vc4_seqno_cb {
+	struct work_struct work;
+	uint64_t seqno;
+	void (*func)(struct vc4_seqno_cb *cb);
+};
+
+struct vc4_v3d {
+	struct vc4_dev *vc4;
+	struct platform_device *pdev;
+	void __iomem *regs;
+	struct clk *clk;
+};
+
+struct vc4_hvs {
+	struct platform_device *pdev;
+	void __iomem *regs;
+	u32 __iomem *dlist;
+
+	/* Memory manager for CRTCs to allocate space in the display
+	 * list.  Units are dwords.
+	 */
+	struct drm_mm dlist_mm;
+	/* Memory manager for the LBM memory used by HVS scaling. */
+	struct drm_mm lbm_mm;
+	spinlock_t mm_lock;
+
+	struct drm_mm_node mitchell_netravali_filter;
+};
+
+struct vc4_plane {
+	struct drm_plane base;
+};
+
+static inline struct vc4_plane *
+to_vc4_plane(struct drm_plane *plane)
+{
+	return (struct vc4_plane *)plane;
+}
+
+enum vc4_scaling_mode {
+	VC4_SCALING_NONE,
+	VC4_SCALING_TPZ,
+	VC4_SCALING_PPF,
+};
+
+struct vc4_plane_state {
+	struct drm_plane_state base;
+	/* System memory copy of the display list for this element, computed
+	 * at atomic_check time.
+	 */
+	u32 *dlist;
+	u32 dlist_size; /* Number of dwords allocated for the display list */
+	u32 dlist_count; /* Number of used dwords in the display list. */
+
+	/* Offset in the dlist to various words, for pageflip or
+	 * cursor updates.
+	 */
+	u32 pos0_offset;
+	u32 pos2_offset;
+	u32 ptr0_offset;
+
+	/* Offset where the plane's dlist was last stored in the
+	 * hardware at vc4_crtc_atomic_flush() time.
+	 */
+	u32 __iomem *hw_dlist;
+
+	/* Clipped coordinates of the plane on the display. */
+	int crtc_x, crtc_y, crtc_w, crtc_h;
+	/* Clipped area being scanned from in the FB. */
+	u32 src_x, src_y;
+
+	u32 src_w[2], src_h[2];
+
+	/* Scaling selection for the RGB/Y plane and the Cb/Cr planes. */
+	enum vc4_scaling_mode x_scaling[2], y_scaling[2];
+	bool is_unity;
+	bool is_yuv;
+
+	/* Offset to start scanning out from the start of the plane's
+	 * BO.
+	 */
+	u32 offsets[3];
+
+	/* Our allocation in LBM for temporary storage during scaling. */
+	struct drm_mm_node lbm;
+
+	/* Set when the plane has per-pixel alpha content or does not cover
+	 * the entire screen. This is a hint to the CRTC that it might need
+	 * to enable background color fill.
+	 */
+	bool needs_bg_fill;
+};
+
+static inline struct vc4_plane_state *
+to_vc4_plane_state(struct drm_plane_state *state)
+{
+	return (struct vc4_plane_state *)state;
+}
+
+enum vc4_encoder_type {
+	VC4_ENCODER_TYPE_NONE,
+	VC4_ENCODER_TYPE_HDMI,
+	VC4_ENCODER_TYPE_VEC,
+	VC4_ENCODER_TYPE_DSI0,
+	VC4_ENCODER_TYPE_DSI1,
+	VC4_ENCODER_TYPE_SMI,
+	VC4_ENCODER_TYPE_DPI,
+};
+
+struct vc4_encoder {
+	struct drm_encoder base;
+	enum vc4_encoder_type type;
+	u32 clock_select;
+};
+
+static inline struct vc4_encoder *
+to_vc4_encoder(struct drm_encoder *encoder)
+{
+	return container_of(encoder, struct vc4_encoder, base);
+}
+
+struct vc4_crtc_data {
+	/* Which channel of the HVS this pixelvalve sources from. */
+	int hvs_channel;
+
+	enum vc4_encoder_type encoder_types[4];
+};
+
+struct vc4_crtc {
+	struct drm_crtc base;
+	const struct vc4_crtc_data *data;
+	void __iomem *regs;
+
+	/* Timestamp at start of vblank irq - unaffected by lock delays. */
+	ktime_t t_vblank;
+
+	/* Which HVS channel we're using for our CRTC. */
+	int channel;
+
+	u8 lut_r[256];
+	u8 lut_g[256];
+	u8 lut_b[256];
+	/* Size in pixels of the COB memory allocated to this CRTC. */
+	u32 cob_size;
+
+	struct drm_pending_vblank_event *event;
+};
+
+static inline struct vc4_crtc *
+to_vc4_crtc(struct drm_crtc *crtc)
+{
+	return (struct vc4_crtc *)crtc;
+}
+
+#define V3D_READ(offset) readl(vc4->v3d->regs + offset)
+#define V3D_WRITE(offset, val) writel(val, vc4->v3d->regs + offset)
+#define HVS_READ(offset) readl(vc4->hvs->regs + offset)
+#define HVS_WRITE(offset, val) writel(val, vc4->hvs->regs + offset)
+
+struct vc4_exec_info {
+	/* Sequence number for this bin/render job. */
+	uint64_t seqno;
+
+	/* Latest write_seqno of any BO that binning depends on. */
+	uint64_t bin_dep_seqno;
+
+	struct dma_fence *fence;
+
+	/* Last current addresses the hardware was processing when the
+	 * hangcheck timer checked on us.
+	 */
+	uint32_t last_ct0ca, last_ct1ca;
+
+	/* Kernel-space copy of the ioctl arguments */
+	struct drm_vc4_submit_cl *args;
+
+	/* This is the array of BOs that were looked up at the start of exec.
+	 * Command validation will use indices into this array.
+	 */
+	struct drm_gem_cma_object **bo;
+	uint32_t bo_count;
+
+	/* List of BOs that are being written by the RCL.  Other than
+	 * the binner temporary storage, this is all the BOs written
+	 * by the job.
+	 */
+	struct drm_gem_cma_object *rcl_write_bo[4];
+	uint32_t rcl_write_bo_count;
+
+	/* Pointers for our position in vc4->job_list */
+	struct list_head head;
+
+	/* List of other BOs used in the job that need to be released
+	 * once the job is complete.
+	 */
+	struct list_head unref_list;
+
+	/* Current unvalidated indices into @bo loaded by the non-hardware
+	 * VC4_PACKET_GEM_HANDLES.
+	 */
+	uint32_t bo_index[2];
+
+	/* This is the BO where we store the validated command lists, shader
+	 * records, and uniforms.
+	 */
+	struct drm_gem_cma_object *exec_bo;
+
+	/**
+	 * This tracks the per-shader-record state (packet 64) that
+	 * determines the length of the shader record and the offset
+	 * it's expected to be found at.  It gets read in from the
+	 * command lists.
+	 */
+	struct vc4_shader_state {
+		uint32_t addr;
+		/* Maximum vertex index referenced by any primitive using this
+		 * shader state.
+		 */
+		uint32_t max_index;
+	} *shader_state;
+
+	/** How many shader states the user declared they were using. */
+	uint32_t shader_state_size;
+	/** How many shader state records the validator has seen. */
+	uint32_t shader_state_count;
+
+	bool found_tile_binning_mode_config_packet;
+	bool found_start_tile_binning_packet;
+	bool found_increment_semaphore_packet;
+	bool found_flush;
+	uint8_t bin_tiles_x, bin_tiles_y;
+	/* Physical address of the start of the tile alloc array
+	 * (where each tile's binned CL will start)
+	 */
+	uint32_t tile_alloc_offset;
+	/* Bitmask of which binner slots are freed when this job completes. */
+	uint32_t bin_slots;
+
+	/**
+	 * Computed addresses pointing into exec_bo where we start the
+	 * bin thread (ct0) and render thread (ct1).
+	 */
+	uint32_t ct0ca, ct0ea;
+	uint32_t ct1ca, ct1ea;
+
+	/* Pointer to the unvalidated bin CL (if present). */
+	void *bin_u;
+
+	/* Pointers to the shader recs.  These paddr gets incremented as CL
+	 * packets are relocated in validate_gl_shader_state, and the vaddrs
+	 * (u and v) get incremented and size decremented as the shader recs
+	 * themselves are validated.
+	 */
+	void *shader_rec_u;
+	void *shader_rec_v;
+	uint32_t shader_rec_p;
+	uint32_t shader_rec_size;
+
+	/* Pointers to the uniform data.  These pointers are incremented, and
+	 * size decremented, as each batch of uniforms is uploaded.
+	 */
+	void *uniforms_u;
+	void *uniforms_v;
+	uint32_t uniforms_p;
+	uint32_t uniforms_size;
+
+	/* Pointer to a performance monitor object if the user requested it,
+	 * NULL otherwise.
+	 */
+	struct vc4_perfmon *perfmon;
+};
+
+/* Per-open file private data. Any driver-specific resource that has to be
+ * released when the DRM file is closed should be placed here.
+ */
+struct vc4_file {
+	struct {
+		struct idr idr;
+		struct mutex lock;
+	} perfmon;
+};
+
+static inline struct vc4_exec_info *
+vc4_first_bin_job(struct vc4_dev *vc4)
+{
+	return list_first_entry_or_null(&vc4->bin_job_list,
+					struct vc4_exec_info, head);
+}
+
+static inline struct vc4_exec_info *
+vc4_first_render_job(struct vc4_dev *vc4)
+{
+	return list_first_entry_or_null(&vc4->render_job_list,
+					struct vc4_exec_info, head);
+}
+
+static inline struct vc4_exec_info *
+vc4_last_render_job(struct vc4_dev *vc4)
+{
+	if (list_empty(&vc4->render_job_list))
+		return NULL;
+	return list_last_entry(&vc4->render_job_list,
+			       struct vc4_exec_info, head);
+}
+
+/**
+ * struct vc4_texture_sample_info - saves the offsets into the UBO for texture
+ * setup parameters.
+ *
+ * This will be used at draw time to relocate the reference to the texture
+ * contents in p0, and validate that the offset combined with
+ * width/height/stride/etc. from p1 and p2/p3 doesn't sample outside the BO.
+ * Note that the hardware treats unprovided config parameters as 0, so not all
+ * of them need to be set up for every texure sample, and we'll store ~0 as
+ * the offset to mark the unused ones.
+ *
+ * See the VC4 3D architecture guide page 41 ("Texture and Memory Lookup Unit
+ * Setup") for definitions of the texture parameters.
+ */
+struct vc4_texture_sample_info {
+	bool is_direct;
+	uint32_t p_offset[4];
+};
+
+/**
+ * struct vc4_validated_shader_info - information about validated shaders that
+ * needs to be used from command list validation.
+ *
+ * For a given shader, each time a shader state record references it, we need
+ * to verify that the shader doesn't read more uniforms than the shader state
+ * record's uniform BO pointer can provide, and we need to apply relocations
+ * and validate the shader state record's uniforms that define the texture
+ * samples.
+ */
+struct vc4_validated_shader_info {
+	uint32_t uniforms_size;
+	uint32_t uniforms_src_size;
+	uint32_t num_texture_samples;
+	struct vc4_texture_sample_info *texture_samples;
+
+	uint32_t num_uniform_addr_offsets;
+	uint32_t *uniform_addr_offsets;
+
+	bool is_threaded;
+};
+
+/**
+ * _wait_for - magic (register) wait macro
+ *
+ * Does the right thing for modeset paths when run under kdgb or similar atomic
+ * contexts. Note that it's important that we check the condition again after
+ * having timed out, since the timeout could be due to preemption or similar and
+ * we've never had a chance to check the condition before the timeout.
+ */
+#define _wait_for(COND, MS, W) ({ \
+	unsigned long timeout__ = jiffies + msecs_to_jiffies(MS) + 1;	\
+	int ret__ = 0;							\
+	while (!(COND)) {						\
+		if (time_after(jiffies, timeout__)) {			\
+			if (!(COND))					\
+				ret__ = -ETIMEDOUT;			\
+			break;						\
+		}							\
+		if (W && drm_can_sleep())  {				\
+			msleep(W);					\
+		} else {						\
+			cpu_relax();					\
+		}							\
+	}								\
+	ret__;								\
+})
+
+#define wait_for(COND, MS) _wait_for(COND, MS, 1)
+
+/* vc4_bo.c */
+struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size);
+void vc4_free_object(struct drm_gem_object *gem_obj);
+struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t size,
+			     bool from_cache, enum vc4_kernel_bo_type type);
+int vc4_dumb_create(struct drm_file *file_priv,
+		    struct drm_device *dev,
+		    struct drm_mode_create_dumb *args);
+struct dma_buf *vc4_prime_export(struct drm_device *dev,
+				 struct drm_gem_object *obj, int flags);
+int vc4_create_bo_ioctl(struct drm_device *dev, void *data,
+			struct drm_file *file_priv);
+int vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data,
+			       struct drm_file *file_priv);
+int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data,
+		      struct drm_file *file_priv);
+int vc4_set_tiling_ioctl(struct drm_device *dev, void *data,
+			 struct drm_file *file_priv);
+int vc4_get_tiling_ioctl(struct drm_device *dev, void *data,
+			 struct drm_file *file_priv);
+int vc4_get_hang_state_ioctl(struct drm_device *dev, void *data,
+			     struct drm_file *file_priv);
+int vc4_label_bo_ioctl(struct drm_device *dev, void *data,
+		       struct drm_file *file_priv);
+vm_fault_t vc4_fault(struct vm_fault *vmf);
+int vc4_mmap(struct file *filp, struct vm_area_struct *vma);
+struct reservation_object *vc4_prime_res_obj(struct drm_gem_object *obj);
+int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma);
+struct drm_gem_object *vc4_prime_import_sg_table(struct drm_device *dev,
+						 struct dma_buf_attachment *attach,
+						 struct sg_table *sgt);
+void *vc4_prime_vmap(struct drm_gem_object *obj);
+int vc4_bo_cache_init(struct drm_device *dev);
+void vc4_bo_cache_destroy(struct drm_device *dev);
+int vc4_bo_stats_debugfs(struct seq_file *m, void *arg);
+int vc4_bo_inc_usecnt(struct vc4_bo *bo);
+void vc4_bo_dec_usecnt(struct vc4_bo *bo);
+void vc4_bo_add_to_purgeable_pool(struct vc4_bo *bo);
+void vc4_bo_remove_from_purgeable_pool(struct vc4_bo *bo);
+
+/* vc4_crtc.c */
+extern struct platform_driver vc4_crtc_driver;
+int vc4_crtc_debugfs_regs(struct seq_file *m, void *arg);
+bool vc4_crtc_get_scanoutpos(struct drm_device *dev, unsigned int crtc_id,
+			     bool in_vblank_irq, int *vpos, int *hpos,
+			     ktime_t *stime, ktime_t *etime,
+			     const struct drm_display_mode *mode);
+void vc4_crtc_handle_vblank(struct vc4_crtc *crtc);
+void vc4_crtc_txp_armed(struct drm_crtc_state *state);
+
+/* vc4_debugfs.c */
+int vc4_debugfs_init(struct drm_minor *minor);
+
+/* vc4_drv.c */
+void __iomem *vc4_ioremap_regs(struct platform_device *dev, int index);
+
+/* vc4_dpi.c */
+extern struct platform_driver vc4_dpi_driver;
+int vc4_dpi_debugfs_regs(struct seq_file *m, void *unused);
+
+/* vc4_dsi.c */
+extern struct platform_driver vc4_dsi_driver;
+int vc4_dsi_debugfs_regs(struct seq_file *m, void *unused);
+
+/* vc4_fence.c */
+extern const struct dma_fence_ops vc4_fence_ops;
+
+/* vc4_gem.c */
+void vc4_gem_init(struct drm_device *dev);
+void vc4_gem_destroy(struct drm_device *dev);
+int vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
+			struct drm_file *file_priv);
+int vc4_wait_seqno_ioctl(struct drm_device *dev, void *data,
+			 struct drm_file *file_priv);
+int vc4_wait_bo_ioctl(struct drm_device *dev, void *data,
+		      struct drm_file *file_priv);
+void vc4_submit_next_bin_job(struct drm_device *dev);
+void vc4_submit_next_render_job(struct drm_device *dev);
+void vc4_move_job_to_render(struct drm_device *dev, struct vc4_exec_info *exec);
+int vc4_wait_for_seqno(struct drm_device *dev, uint64_t seqno,
+		       uint64_t timeout_ns, bool interruptible);
+void vc4_job_handle_completed(struct vc4_dev *vc4);
+int vc4_queue_seqno_cb(struct drm_device *dev,
+		       struct vc4_seqno_cb *cb, uint64_t seqno,
+		       void (*func)(struct vc4_seqno_cb *cb));
+int vc4_gem_madvise_ioctl(struct drm_device *dev, void *data,
+			  struct drm_file *file_priv);
+
+/* vc4_hdmi.c */
+extern struct platform_driver vc4_hdmi_driver;
+int vc4_hdmi_debugfs_regs(struct seq_file *m, void *unused);
+
+/* vc4_vec.c */
+extern struct platform_driver vc4_vec_driver;
+int vc4_vec_debugfs_regs(struct seq_file *m, void *unused);
+
+/* vc4_txp.c */
+extern struct platform_driver vc4_txp_driver;
+int vc4_txp_debugfs_regs(struct seq_file *m, void *unused);
+
+/* vc4_irq.c */
+irqreturn_t vc4_irq(int irq, void *arg);
+void vc4_irq_preinstall(struct drm_device *dev);
+int vc4_irq_postinstall(struct drm_device *dev);
+void vc4_irq_uninstall(struct drm_device *dev);
+void vc4_irq_reset(struct drm_device *dev);
+
+/* vc4_hvs.c */
+extern struct platform_driver vc4_hvs_driver;
+void vc4_hvs_dump_state(struct drm_device *dev);
+int vc4_hvs_debugfs_regs(struct seq_file *m, void *unused);
+
+/* vc4_kms.c */
+int vc4_kms_load(struct drm_device *dev);
+
+/* vc4_plane.c */
+struct drm_plane *vc4_plane_init(struct drm_device *dev,
+				 enum drm_plane_type type);
+u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist);
+u32 vc4_plane_dlist_size(const struct drm_plane_state *state);
+void vc4_plane_async_set_fb(struct drm_plane *plane,
+			    struct drm_framebuffer *fb);
+
+/* vc4_v3d.c */
+extern struct platform_driver vc4_v3d_driver;
+int vc4_v3d_debugfs_ident(struct seq_file *m, void *unused);
+int vc4_v3d_debugfs_regs(struct seq_file *m, void *unused);
+int vc4_v3d_get_bin_slot(struct vc4_dev *vc4);
+
+/* vc4_validate.c */
+int
+vc4_validate_bin_cl(struct drm_device *dev,
+		    void *validated,
+		    void *unvalidated,
+		    struct vc4_exec_info *exec);
+
+int
+vc4_validate_shader_recs(struct drm_device *dev, struct vc4_exec_info *exec);
+
+struct drm_gem_cma_object *vc4_use_bo(struct vc4_exec_info *exec,
+				      uint32_t hindex);
+
+int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec);
+
+bool vc4_check_tex_size(struct vc4_exec_info *exec,
+			struct drm_gem_cma_object *fbo,
+			uint32_t offset, uint8_t tiling_format,
+			uint32_t width, uint32_t height, uint8_t cpp);
+
+/* vc4_validate_shader.c */
+struct vc4_validated_shader_info *
+vc4_validate_shader(struct drm_gem_cma_object *shader_obj);
+
+/* vc4_perfmon.c */
+void vc4_perfmon_get(struct vc4_perfmon *perfmon);
+void vc4_perfmon_put(struct vc4_perfmon *perfmon);
+void vc4_perfmon_start(struct vc4_dev *vc4, struct vc4_perfmon *perfmon);
+void vc4_perfmon_stop(struct vc4_dev *vc4, struct vc4_perfmon *perfmon,
+		      bool capture);
+struct vc4_perfmon *vc4_perfmon_find(struct vc4_file *vc4file, int id);
+void vc4_perfmon_open_file(struct vc4_file *vc4file);
+void vc4_perfmon_close_file(struct vc4_file *vc4file);
+int vc4_perfmon_create_ioctl(struct drm_device *dev, void *data,
+			     struct drm_file *file_priv);
+int vc4_perfmon_destroy_ioctl(struct drm_device *dev, void *data,
+			      struct drm_file *file_priv);
+int vc4_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
+				 struct drm_file *file_priv);
diff --git a/drivers/gpu/drm/vc4/vc4_dsi.c b/drivers/gpu/drm/vc4/vc4_dsi.c
new file mode 100644
index 000000000..0c607eb33
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_dsi.c
@@ -0,0 +1,1744 @@
+/*
+ * Copyright (C) 2016 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published by
+ * the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+/**
+ * DOC: VC4 DSI0/DSI1 module
+ *
+ * BCM2835 contains two DSI modules, DSI0 and DSI1.  DSI0 is a
+ * single-lane DSI controller, while DSI1 is a more modern 4-lane DSI
+ * controller.
+ *
+ * Most Raspberry Pi boards expose DSI1 as their "DISPLAY" connector,
+ * while the compute module brings both DSI0 and DSI1 out.
+ *
+ * This driver has been tested for DSI1 video-mode display only
+ * currently, with most of the information necessary for DSI0
+ * hopefully present.
+ */
+
+#include <drm/drm_atomic_helper.h>
+#include <drm/drm_crtc_helper.h>
+#include <drm/drm_edid.h>
+#include <drm/drm_mipi_dsi.h>
+#include <drm/drm_of.h>
+#include <drm/drm_panel.h>
+#include <linux/clk.h>
+#include <linux/clk-provider.h>
+#include <linux/completion.h>
+#include <linux/component.h>
+#include <linux/dmaengine.h>
+#include <linux/i2c.h>
+#include <linux/of_address.h>
+#include <linux/of_platform.h>
+#include <linux/pm_runtime.h>
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+#define DSI_CMD_FIFO_DEPTH  16
+#define DSI_PIX_FIFO_DEPTH 256
+#define DSI_PIX_FIFO_WIDTH   4
+
+#define DSI0_CTRL		0x00
+
+/* Command packet control. */
+#define DSI0_TXPKT1C		0x04 /* AKA PKTC */
+#define DSI1_TXPKT1C		0x04
+# define DSI_TXPKT1C_TRIG_CMD_MASK	VC4_MASK(31, 24)
+# define DSI_TXPKT1C_TRIG_CMD_SHIFT	24
+# define DSI_TXPKT1C_CMD_REPEAT_MASK	VC4_MASK(23, 10)
+# define DSI_TXPKT1C_CMD_REPEAT_SHIFT	10
+
+# define DSI_TXPKT1C_DISPLAY_NO_MASK	VC4_MASK(9, 8)
+# define DSI_TXPKT1C_DISPLAY_NO_SHIFT	8
+/* Short, trigger, BTA, or a long packet that fits all in CMDFIFO. */
+# define DSI_TXPKT1C_DISPLAY_NO_SHORT		0
+/* Primary display where cmdfifo provides part of the payload and
+ * pixelvalve the rest.
+ */
+# define DSI_TXPKT1C_DISPLAY_NO_PRIMARY		1
+/* Secondary display where cmdfifo provides part of the payload and
+ * pixfifo the rest.
+ */
+# define DSI_TXPKT1C_DISPLAY_NO_SECONDARY	2
+
+# define DSI_TXPKT1C_CMD_TX_TIME_MASK	VC4_MASK(7, 6)
+# define DSI_TXPKT1C_CMD_TX_TIME_SHIFT	6
+
+# define DSI_TXPKT1C_CMD_CTRL_MASK	VC4_MASK(5, 4)
+# define DSI_TXPKT1C_CMD_CTRL_SHIFT	4
+/* Command only.  Uses TXPKT1H and DISPLAY_NO */
+# define DSI_TXPKT1C_CMD_CTRL_TX	0
+/* Command with BTA for either ack or read data. */
+# define DSI_TXPKT1C_CMD_CTRL_RX	1
+/* Trigger according to TRIG_CMD */
+# define DSI_TXPKT1C_CMD_CTRL_TRIG	2
+/* BTA alone for getting error status after a command, or a TE trigger
+ * without a previous command.
+ */
+# define DSI_TXPKT1C_CMD_CTRL_BTA	3
+
+# define DSI_TXPKT1C_CMD_MODE_LP	BIT(3)
+# define DSI_TXPKT1C_CMD_TYPE_LONG	BIT(2)
+# define DSI_TXPKT1C_CMD_TE_EN		BIT(1)
+# define DSI_TXPKT1C_CMD_EN		BIT(0)
+
+/* Command packet header. */
+#define DSI0_TXPKT1H		0x08 /* AKA PKTH */
+#define DSI1_TXPKT1H		0x08
+# define DSI_TXPKT1H_BC_CMDFIFO_MASK	VC4_MASK(31, 24)
+# define DSI_TXPKT1H_BC_CMDFIFO_SHIFT	24
+# define DSI_TXPKT1H_BC_PARAM_MASK	VC4_MASK(23, 8)
+# define DSI_TXPKT1H_BC_PARAM_SHIFT	8
+# define DSI_TXPKT1H_BC_DT_MASK		VC4_MASK(7, 0)
+# define DSI_TXPKT1H_BC_DT_SHIFT	0
+
+#define DSI0_RXPKT1H		0x0c /* AKA RX1_PKTH */
+#define DSI1_RXPKT1H		0x14
+# define DSI_RXPKT1H_CRC_ERR		BIT(31)
+# define DSI_RXPKT1H_DET_ERR		BIT(30)
+# define DSI_RXPKT1H_ECC_ERR		BIT(29)
+# define DSI_RXPKT1H_COR_ERR		BIT(28)
+# define DSI_RXPKT1H_INCOMP_PKT		BIT(25)
+# define DSI_RXPKT1H_PKT_TYPE_LONG	BIT(24)
+/* Byte count if DSI_RXPKT1H_PKT_TYPE_LONG */
+# define DSI_RXPKT1H_BC_PARAM_MASK	VC4_MASK(23, 8)
+# define DSI_RXPKT1H_BC_PARAM_SHIFT	8
+/* Short return bytes if !DSI_RXPKT1H_PKT_TYPE_LONG */
+# define DSI_RXPKT1H_SHORT_1_MASK	VC4_MASK(23, 16)
+# define DSI_RXPKT1H_SHORT_1_SHIFT	16
+# define DSI_RXPKT1H_SHORT_0_MASK	VC4_MASK(15, 8)
+# define DSI_RXPKT1H_SHORT_0_SHIFT	8
+# define DSI_RXPKT1H_DT_LP_CMD_MASK	VC4_MASK(7, 0)
+# define DSI_RXPKT1H_DT_LP_CMD_SHIFT	0
+
+#define DSI0_RXPKT2H		0x10 /* AKA RX2_PKTH */
+#define DSI1_RXPKT2H		0x18
+# define DSI_RXPKT1H_DET_ERR		BIT(30)
+# define DSI_RXPKT1H_ECC_ERR		BIT(29)
+# define DSI_RXPKT1H_COR_ERR		BIT(28)
+# define DSI_RXPKT1H_INCOMP_PKT		BIT(25)
+# define DSI_RXPKT1H_BC_PARAM_MASK	VC4_MASK(23, 8)
+# define DSI_RXPKT1H_BC_PARAM_SHIFT	8
+# define DSI_RXPKT1H_DT_MASK		VC4_MASK(7, 0)
+# define DSI_RXPKT1H_DT_SHIFT		0
+
+#define DSI0_TXPKT_CMD_FIFO	0x14 /* AKA CMD_DATAF */
+#define DSI1_TXPKT_CMD_FIFO	0x1c
+
+#define DSI0_DISP0_CTRL		0x18
+# define DSI_DISP0_PIX_CLK_DIV_MASK	VC4_MASK(21, 13)
+# define DSI_DISP0_PIX_CLK_DIV_SHIFT	13
+# define DSI_DISP0_LP_STOP_CTRL_MASK	VC4_MASK(12, 11)
+# define DSI_DISP0_LP_STOP_CTRL_SHIFT	11
+# define DSI_DISP0_LP_STOP_DISABLE	0
+# define DSI_DISP0_LP_STOP_PERLINE	1
+# define DSI_DISP0_LP_STOP_PERFRAME	2
+
+/* Transmit RGB pixels and null packets only during HACTIVE, instead
+ * of going to LP-STOP.
+ */
+# define DSI_DISP_HACTIVE_NULL		BIT(10)
+/* Transmit blanking packet only during vblank, instead of allowing LP-STOP. */
+# define DSI_DISP_VBLP_CTRL		BIT(9)
+/* Transmit blanking packet only during HFP, instead of allowing LP-STOP. */
+# define DSI_DISP_HFP_CTRL		BIT(8)
+/* Transmit blanking packet only during HBP, instead of allowing LP-STOP. */
+# define DSI_DISP_HBP_CTRL		BIT(7)
+# define DSI_DISP0_CHANNEL_MASK		VC4_MASK(6, 5)
+# define DSI_DISP0_CHANNEL_SHIFT	5
+/* Enables end events for HSYNC/VSYNC, not just start events. */
+# define DSI_DISP0_ST_END		BIT(4)
+# define DSI_DISP0_PFORMAT_MASK		VC4_MASK(3, 2)
+# define DSI_DISP0_PFORMAT_SHIFT	2
+# define DSI_PFORMAT_RGB565		0
+# define DSI_PFORMAT_RGB666_PACKED	1
+# define DSI_PFORMAT_RGB666		2
+# define DSI_PFORMAT_RGB888		3
+/* Default is VIDEO mode. */
+# define DSI_DISP0_COMMAND_MODE		BIT(1)
+# define DSI_DISP0_ENABLE		BIT(0)
+
+#define DSI0_DISP1_CTRL		0x1c
+#define DSI1_DISP1_CTRL		0x2c
+/* Format of the data written to TXPKT_PIX_FIFO. */
+# define DSI_DISP1_PFORMAT_MASK		VC4_MASK(2, 1)
+# define DSI_DISP1_PFORMAT_SHIFT	1
+# define DSI_DISP1_PFORMAT_16BIT	0
+# define DSI_DISP1_PFORMAT_24BIT	1
+# define DSI_DISP1_PFORMAT_32BIT_LE	2
+# define DSI_DISP1_PFORMAT_32BIT_BE	3
+
+/* DISP1 is always command mode. */
+# define DSI_DISP1_ENABLE		BIT(0)
+
+#define DSI0_TXPKT_PIX_FIFO		0x20 /* AKA PIX_FIFO */
+
+#define DSI0_INT_STAT		0x24
+#define DSI0_INT_EN		0x28
+# define DSI1_INT_PHY_D3_ULPS		BIT(30)
+# define DSI1_INT_PHY_D3_STOP		BIT(29)
+# define DSI1_INT_PHY_D2_ULPS		BIT(28)
+# define DSI1_INT_PHY_D2_STOP		BIT(27)
+# define DSI1_INT_PHY_D1_ULPS		BIT(26)
+# define DSI1_INT_PHY_D1_STOP		BIT(25)
+# define DSI1_INT_PHY_D0_ULPS		BIT(24)
+# define DSI1_INT_PHY_D0_STOP		BIT(23)
+# define DSI1_INT_FIFO_ERR		BIT(22)
+# define DSI1_INT_PHY_DIR_RTF		BIT(21)
+# define DSI1_INT_PHY_RXLPDT		BIT(20)
+# define DSI1_INT_PHY_RXTRIG		BIT(19)
+# define DSI1_INT_PHY_D0_LPDT		BIT(18)
+# define DSI1_INT_PHY_DIR_FTR		BIT(17)
+
+/* Signaled when the clock lane enters the given state. */
+# define DSI1_INT_PHY_CLOCK_ULPS	BIT(16)
+# define DSI1_INT_PHY_CLOCK_HS		BIT(15)
+# define DSI1_INT_PHY_CLOCK_STOP	BIT(14)
+
+/* Signaled on timeouts */
+# define DSI1_INT_PR_TO			BIT(13)
+# define DSI1_INT_TA_TO			BIT(12)
+# define DSI1_INT_LPRX_TO		BIT(11)
+# define DSI1_INT_HSTX_TO		BIT(10)
+
+/* Contention on a line when trying to drive the line low */
+# define DSI1_INT_ERR_CONT_LP1		BIT(9)
+# define DSI1_INT_ERR_CONT_LP0		BIT(8)
+
+/* Control error: incorrect line state sequence on data lane 0. */
+# define DSI1_INT_ERR_CONTROL		BIT(7)
+/* LPDT synchronization error (bits received not a multiple of 8. */
+
+# define DSI1_INT_ERR_SYNC_ESC		BIT(6)
+/* Signaled after receiving an error packet from the display in
+ * response to a read.
+ */
+# define DSI1_INT_RXPKT2		BIT(5)
+/* Signaled after receiving a packet.  The header and optional short
+ * response will be in RXPKT1H, and a long response will be in the
+ * RXPKT_FIFO.
+ */
+# define DSI1_INT_RXPKT1		BIT(4)
+# define DSI1_INT_TXPKT2_DONE		BIT(3)
+# define DSI1_INT_TXPKT2_END		BIT(2)
+/* Signaled after all repeats of TXPKT1 are transferred. */
+# define DSI1_INT_TXPKT1_DONE		BIT(1)
+/* Signaled after each TXPKT1 repeat is scheduled. */
+# define DSI1_INT_TXPKT1_END		BIT(0)
+
+#define DSI1_INTERRUPTS_ALWAYS_ENABLED	(DSI1_INT_ERR_SYNC_ESC | \
+					 DSI1_INT_ERR_CONTROL |	 \
+					 DSI1_INT_ERR_CONT_LP0 | \
+					 DSI1_INT_ERR_CONT_LP1 | \
+					 DSI1_INT_HSTX_TO |	 \
+					 DSI1_INT_LPRX_TO |	 \
+					 DSI1_INT_TA_TO |	 \
+					 DSI1_INT_PR_TO)
+
+#define DSI0_STAT		0x2c
+#define DSI0_HSTX_TO_CNT	0x30
+#define DSI0_LPRX_TO_CNT	0x34
+#define DSI0_TA_TO_CNT		0x38
+#define DSI0_PR_TO_CNT		0x3c
+#define DSI0_PHYC		0x40
+# define DSI1_PHYC_ESC_CLK_LPDT_MASK	VC4_MASK(25, 20)
+# define DSI1_PHYC_ESC_CLK_LPDT_SHIFT	20
+# define DSI1_PHYC_HS_CLK_CONTINUOUS	BIT(18)
+# define DSI0_PHYC_ESC_CLK_LPDT_MASK	VC4_MASK(17, 12)
+# define DSI0_PHYC_ESC_CLK_LPDT_SHIFT	12
+# define DSI1_PHYC_CLANE_ULPS		BIT(17)
+# define DSI1_PHYC_CLANE_ENABLE		BIT(16)
+# define DSI_PHYC_DLANE3_ULPS		BIT(13)
+# define DSI_PHYC_DLANE3_ENABLE		BIT(12)
+# define DSI0_PHYC_HS_CLK_CONTINUOUS	BIT(10)
+# define DSI0_PHYC_CLANE_ULPS		BIT(9)
+# define DSI_PHYC_DLANE2_ULPS		BIT(9)
+# define DSI0_PHYC_CLANE_ENABLE		BIT(8)
+# define DSI_PHYC_DLANE2_ENABLE		BIT(8)
+# define DSI_PHYC_DLANE1_ULPS		BIT(5)
+# define DSI_PHYC_DLANE1_ENABLE		BIT(4)
+# define DSI_PHYC_DLANE0_FORCE_STOP	BIT(2)
+# define DSI_PHYC_DLANE0_ULPS		BIT(1)
+# define DSI_PHYC_DLANE0_ENABLE		BIT(0)
+
+#define DSI0_HS_CLT0		0x44
+#define DSI0_HS_CLT1		0x48
+#define DSI0_HS_CLT2		0x4c
+#define DSI0_HS_DLT3		0x50
+#define DSI0_HS_DLT4		0x54
+#define DSI0_HS_DLT5		0x58
+#define DSI0_HS_DLT6		0x5c
+#define DSI0_HS_DLT7		0x60
+
+#define DSI0_PHY_AFEC0		0x64
+# define DSI0_PHY_AFEC0_DDR2CLK_EN		BIT(26)
+# define DSI0_PHY_AFEC0_DDRCLK_EN		BIT(25)
+# define DSI0_PHY_AFEC0_LATCH_ULPS		BIT(24)
+# define DSI1_PHY_AFEC0_IDR_DLANE3_MASK		VC4_MASK(31, 29)
+# define DSI1_PHY_AFEC0_IDR_DLANE3_SHIFT	29
+# define DSI1_PHY_AFEC0_IDR_DLANE2_MASK		VC4_MASK(28, 26)
+# define DSI1_PHY_AFEC0_IDR_DLANE2_SHIFT	26
+# define DSI1_PHY_AFEC0_IDR_DLANE1_MASK		VC4_MASK(27, 23)
+# define DSI1_PHY_AFEC0_IDR_DLANE1_SHIFT	23
+# define DSI1_PHY_AFEC0_IDR_DLANE0_MASK		VC4_MASK(22, 20)
+# define DSI1_PHY_AFEC0_IDR_DLANE0_SHIFT	20
+# define DSI1_PHY_AFEC0_IDR_CLANE_MASK		VC4_MASK(19, 17)
+# define DSI1_PHY_AFEC0_IDR_CLANE_SHIFT		17
+# define DSI0_PHY_AFEC0_ACTRL_DLANE1_MASK	VC4_MASK(23, 20)
+# define DSI0_PHY_AFEC0_ACTRL_DLANE1_SHIFT	20
+# define DSI0_PHY_AFEC0_ACTRL_DLANE0_MASK	VC4_MASK(19, 16)
+# define DSI0_PHY_AFEC0_ACTRL_DLANE0_SHIFT	16
+# define DSI0_PHY_AFEC0_ACTRL_CLANE_MASK	VC4_MASK(15, 12)
+# define DSI0_PHY_AFEC0_ACTRL_CLANE_SHIFT	12
+# define DSI1_PHY_AFEC0_DDR2CLK_EN		BIT(16)
+# define DSI1_PHY_AFEC0_DDRCLK_EN		BIT(15)
+# define DSI1_PHY_AFEC0_LATCH_ULPS		BIT(14)
+# define DSI1_PHY_AFEC0_RESET			BIT(13)
+# define DSI1_PHY_AFEC0_PD			BIT(12)
+# define DSI0_PHY_AFEC0_RESET			BIT(11)
+# define DSI1_PHY_AFEC0_PD_BG			BIT(11)
+# define DSI0_PHY_AFEC0_PD			BIT(10)
+# define DSI1_PHY_AFEC0_PD_DLANE3		BIT(10)
+# define DSI0_PHY_AFEC0_PD_BG			BIT(9)
+# define DSI1_PHY_AFEC0_PD_DLANE2		BIT(9)
+# define DSI0_PHY_AFEC0_PD_DLANE1		BIT(8)
+# define DSI1_PHY_AFEC0_PD_DLANE1		BIT(8)
+# define DSI_PHY_AFEC0_PTATADJ_MASK		VC4_MASK(7, 4)
+# define DSI_PHY_AFEC0_PTATADJ_SHIFT		4
+# define DSI_PHY_AFEC0_CTATADJ_MASK		VC4_MASK(3, 0)
+# define DSI_PHY_AFEC0_CTATADJ_SHIFT		0
+
+#define DSI0_PHY_AFEC1		0x68
+# define DSI0_PHY_AFEC1_IDR_DLANE1_MASK		VC4_MASK(10, 8)
+# define DSI0_PHY_AFEC1_IDR_DLANE1_SHIFT	8
+# define DSI0_PHY_AFEC1_IDR_DLANE0_MASK		VC4_MASK(6, 4)
+# define DSI0_PHY_AFEC1_IDR_DLANE0_SHIFT	4
+# define DSI0_PHY_AFEC1_IDR_CLANE_MASK		VC4_MASK(2, 0)
+# define DSI0_PHY_AFEC1_IDR_CLANE_SHIFT		0
+
+#define DSI0_TST_SEL		0x6c
+#define DSI0_TST_MON		0x70
+#define DSI0_ID			0x74
+# define DSI_ID_VALUE		0x00647369
+
+#define DSI1_CTRL		0x00
+# define DSI_CTRL_HS_CLKC_MASK		VC4_MASK(15, 14)
+# define DSI_CTRL_HS_CLKC_SHIFT		14
+# define DSI_CTRL_HS_CLKC_BYTE		0
+# define DSI_CTRL_HS_CLKC_DDR2		1
+# define DSI_CTRL_HS_CLKC_DDR		2
+
+# define DSI_CTRL_RX_LPDT_EOT_DISABLE	BIT(13)
+# define DSI_CTRL_LPDT_EOT_DISABLE	BIT(12)
+# define DSI_CTRL_HSDT_EOT_DISABLE	BIT(11)
+# define DSI_CTRL_SOFT_RESET_CFG	BIT(10)
+# define DSI_CTRL_CAL_BYTE		BIT(9)
+# define DSI_CTRL_INV_BYTE		BIT(8)
+# define DSI_CTRL_CLR_LDF		BIT(7)
+# define DSI0_CTRL_CLR_PBCF		BIT(6)
+# define DSI1_CTRL_CLR_RXF		BIT(6)
+# define DSI0_CTRL_CLR_CPBCF		BIT(5)
+# define DSI1_CTRL_CLR_PDF		BIT(5)
+# define DSI0_CTRL_CLR_PDF		BIT(4)
+# define DSI1_CTRL_CLR_CDF		BIT(4)
+# define DSI0_CTRL_CLR_CDF		BIT(3)
+# define DSI0_CTRL_CTRL2		BIT(2)
+# define DSI1_CTRL_DISABLE_DISP_CRCC	BIT(2)
+# define DSI0_CTRL_CTRL1		BIT(1)
+# define DSI1_CTRL_DISABLE_DISP_ECCC	BIT(1)
+# define DSI0_CTRL_CTRL0		BIT(0)
+# define DSI1_CTRL_EN			BIT(0)
+# define DSI0_CTRL_RESET_FIFOS		(DSI_CTRL_CLR_LDF | \
+					 DSI0_CTRL_CLR_PBCF | \
+					 DSI0_CTRL_CLR_CPBCF |	\
+					 DSI0_CTRL_CLR_PDF | \
+					 DSI0_CTRL_CLR_CDF)
+# define DSI1_CTRL_RESET_FIFOS		(DSI_CTRL_CLR_LDF | \
+					 DSI1_CTRL_CLR_RXF | \
+					 DSI1_CTRL_CLR_PDF | \
+					 DSI1_CTRL_CLR_CDF)
+
+#define DSI1_TXPKT2C		0x0c
+#define DSI1_TXPKT2H		0x10
+#define DSI1_TXPKT_PIX_FIFO	0x20
+#define DSI1_RXPKT_FIFO		0x24
+#define DSI1_DISP0_CTRL		0x28
+#define DSI1_INT_STAT		0x30
+#define DSI1_INT_EN		0x34
+/* State reporting bits.  These mostly behave like INT_STAT, where
+ * writing a 1 clears the bit.
+ */
+#define DSI1_STAT		0x38
+# define DSI1_STAT_PHY_D3_ULPS		BIT(31)
+# define DSI1_STAT_PHY_D3_STOP		BIT(30)
+# define DSI1_STAT_PHY_D2_ULPS		BIT(29)
+# define DSI1_STAT_PHY_D2_STOP		BIT(28)
+# define DSI1_STAT_PHY_D1_ULPS		BIT(27)
+# define DSI1_STAT_PHY_D1_STOP		BIT(26)
+# define DSI1_STAT_PHY_D0_ULPS		BIT(25)
+# define DSI1_STAT_PHY_D0_STOP		BIT(24)
+# define DSI1_STAT_FIFO_ERR		BIT(23)
+# define DSI1_STAT_PHY_RXLPDT		BIT(22)
+# define DSI1_STAT_PHY_RXTRIG		BIT(21)
+# define DSI1_STAT_PHY_D0_LPDT		BIT(20)
+/* Set when in forward direction */
+# define DSI1_STAT_PHY_DIR		BIT(19)
+# define DSI1_STAT_PHY_CLOCK_ULPS	BIT(18)
+# define DSI1_STAT_PHY_CLOCK_HS		BIT(17)
+# define DSI1_STAT_PHY_CLOCK_STOP	BIT(16)
+# define DSI1_STAT_PR_TO		BIT(15)
+# define DSI1_STAT_TA_TO		BIT(14)
+# define DSI1_STAT_LPRX_TO		BIT(13)
+# define DSI1_STAT_HSTX_TO		BIT(12)
+# define DSI1_STAT_ERR_CONT_LP1		BIT(11)
+# define DSI1_STAT_ERR_CONT_LP0		BIT(10)
+# define DSI1_STAT_ERR_CONTROL		BIT(9)
+# define DSI1_STAT_ERR_SYNC_ESC		BIT(8)
+# define DSI1_STAT_RXPKT2		BIT(7)
+# define DSI1_STAT_RXPKT1		BIT(6)
+# define DSI1_STAT_TXPKT2_BUSY		BIT(5)
+# define DSI1_STAT_TXPKT2_DONE		BIT(4)
+# define DSI1_STAT_TXPKT2_END		BIT(3)
+# define DSI1_STAT_TXPKT1_BUSY		BIT(2)
+# define DSI1_STAT_TXPKT1_DONE		BIT(1)
+# define DSI1_STAT_TXPKT1_END		BIT(0)
+
+#define DSI1_HSTX_TO_CNT	0x3c
+#define DSI1_LPRX_TO_CNT	0x40
+#define DSI1_TA_TO_CNT		0x44
+#define DSI1_PR_TO_CNT		0x48
+#define DSI1_PHYC		0x4c
+
+#define DSI1_HS_CLT0		0x50
+# define DSI_HS_CLT0_CZERO_MASK		VC4_MASK(26, 18)
+# define DSI_HS_CLT0_CZERO_SHIFT	18
+# define DSI_HS_CLT0_CPRE_MASK		VC4_MASK(17, 9)
+# define DSI_HS_CLT0_CPRE_SHIFT		9
+# define DSI_HS_CLT0_CPREP_MASK		VC4_MASK(8, 0)
+# define DSI_HS_CLT0_CPREP_SHIFT	0
+
+#define DSI1_HS_CLT1		0x54
+# define DSI_HS_CLT1_CTRAIL_MASK	VC4_MASK(17, 9)
+# define DSI_HS_CLT1_CTRAIL_SHIFT	9
+# define DSI_HS_CLT1_CPOST_MASK		VC4_MASK(8, 0)
+# define DSI_HS_CLT1_CPOST_SHIFT	0
+
+#define DSI1_HS_CLT2		0x58
+# define DSI_HS_CLT2_WUP_MASK		VC4_MASK(23, 0)
+# define DSI_HS_CLT2_WUP_SHIFT		0
+
+#define DSI1_HS_DLT3		0x5c
+# define DSI_HS_DLT3_EXIT_MASK		VC4_MASK(26, 18)
+# define DSI_HS_DLT3_EXIT_SHIFT		18
+# define DSI_HS_DLT3_ZERO_MASK		VC4_MASK(17, 9)
+# define DSI_HS_DLT3_ZERO_SHIFT		9
+# define DSI_HS_DLT3_PRE_MASK		VC4_MASK(8, 0)
+# define DSI_HS_DLT3_PRE_SHIFT		0
+
+#define DSI1_HS_DLT4		0x60
+# define DSI_HS_DLT4_ANLAT_MASK		VC4_MASK(22, 18)
+# define DSI_HS_DLT4_ANLAT_SHIFT	18
+# define DSI_HS_DLT4_TRAIL_MASK		VC4_MASK(17, 9)
+# define DSI_HS_DLT4_TRAIL_SHIFT	9
+# define DSI_HS_DLT4_LPX_MASK		VC4_MASK(8, 0)
+# define DSI_HS_DLT4_LPX_SHIFT		0
+
+#define DSI1_HS_DLT5		0x64
+# define DSI_HS_DLT5_INIT_MASK		VC4_MASK(23, 0)
+# define DSI_HS_DLT5_INIT_SHIFT		0
+
+#define DSI1_HS_DLT6		0x68
+# define DSI_HS_DLT6_TA_GET_MASK	VC4_MASK(31, 24)
+# define DSI_HS_DLT6_TA_GET_SHIFT	24
+# define DSI_HS_DLT6_TA_SURE_MASK	VC4_MASK(23, 16)
+# define DSI_HS_DLT6_TA_SURE_SHIFT	16
+# define DSI_HS_DLT6_TA_GO_MASK		VC4_MASK(15, 8)
+# define DSI_HS_DLT6_TA_GO_SHIFT	8
+# define DSI_HS_DLT6_LP_LPX_MASK	VC4_MASK(7, 0)
+# define DSI_HS_DLT6_LP_LPX_SHIFT	0
+
+#define DSI1_HS_DLT7		0x6c
+# define DSI_HS_DLT7_LP_WUP_MASK	VC4_MASK(23, 0)
+# define DSI_HS_DLT7_LP_WUP_SHIFT	0
+
+#define DSI1_PHY_AFEC0		0x70
+
+#define DSI1_PHY_AFEC1		0x74
+# define DSI1_PHY_AFEC1_ACTRL_DLANE3_MASK	VC4_MASK(19, 16)
+# define DSI1_PHY_AFEC1_ACTRL_DLANE3_SHIFT	16
+# define DSI1_PHY_AFEC1_ACTRL_DLANE2_MASK	VC4_MASK(15, 12)
+# define DSI1_PHY_AFEC1_ACTRL_DLANE2_SHIFT	12
+# define DSI1_PHY_AFEC1_ACTRL_DLANE1_MASK	VC4_MASK(11, 8)
+# define DSI1_PHY_AFEC1_ACTRL_DLANE1_SHIFT	8
+# define DSI1_PHY_AFEC1_ACTRL_DLANE0_MASK	VC4_MASK(7, 4)
+# define DSI1_PHY_AFEC1_ACTRL_DLANE0_SHIFT	4
+# define DSI1_PHY_AFEC1_ACTRL_CLANE_MASK	VC4_MASK(3, 0)
+# define DSI1_PHY_AFEC1_ACTRL_CLANE_SHIFT	0
+
+#define DSI1_TST_SEL		0x78
+#define DSI1_TST_MON		0x7c
+#define DSI1_PHY_TST1		0x80
+#define DSI1_PHY_TST2		0x84
+#define DSI1_PHY_FIFO_STAT	0x88
+/* Actually, all registers in the range that aren't otherwise claimed
+ * will return the ID.
+ */
+#define DSI1_ID			0x8c
+
+/* General DSI hardware state. */
+struct vc4_dsi {
+	struct platform_device *pdev;
+
+	struct mipi_dsi_host dsi_host;
+	struct drm_encoder *encoder;
+	struct drm_bridge *bridge;
+
+	void __iomem *regs;
+
+	struct dma_chan *reg_dma_chan;
+	dma_addr_t reg_dma_paddr;
+	u32 *reg_dma_mem;
+	dma_addr_t reg_paddr;
+
+	/* Whether we're on bcm2835's DSI0 or DSI1. */
+	int port;
+
+	/* DSI channel for the panel we're connected to. */
+	u32 channel;
+	u32 lanes;
+	u32 format;
+	u32 divider;
+	u32 mode_flags;
+
+	/* Input clock from CPRMAN to the digital PHY, for the DSI
+	 * escape clock.
+	 */
+	struct clk *escape_clock;
+
+	/* Input clock to the analog PHY, used to generate the DSI bit
+	 * clock.
+	 */
+	struct clk *pll_phy_clock;
+
+	/* HS Clocks generated within the DSI analog PHY. */
+	struct clk_fixed_factor phy_clocks[3];
+
+	struct clk_hw_onecell_data *clk_onecell;
+
+	/* Pixel clock output to the pixelvalve, generated from the HS
+	 * clock.
+	 */
+	struct clk *pixel_clock;
+
+	struct completion xfer_completion;
+	int xfer_result;
+};
+
+#define host_to_dsi(host) container_of(host, struct vc4_dsi, dsi_host)
+
+static inline void
+dsi_dma_workaround_write(struct vc4_dsi *dsi, u32 offset, u32 val)
+{
+	struct dma_chan *chan = dsi->reg_dma_chan;
+	struct dma_async_tx_descriptor *tx;
+	dma_cookie_t cookie;
+	int ret;
+
+	/* DSI0 should be able to write normally. */
+	if (!chan) {
+		writel(val, dsi->regs + offset);
+		return;
+	}
+
+	*dsi->reg_dma_mem = val;
+
+	tx = chan->device->device_prep_dma_memcpy(chan,
+						  dsi->reg_paddr + offset,
+						  dsi->reg_dma_paddr,
+						  4, 0);
+	if (!tx) {
+		DRM_ERROR("Failed to set up DMA register write\n");
+		return;
+	}
+
+	cookie = tx->tx_submit(tx);
+	ret = dma_submit_error(cookie);
+	if (ret) {
+		DRM_ERROR("Failed to submit DMA: %d\n", ret);
+		return;
+	}
+	ret = dma_sync_wait(chan, cookie);
+	if (ret)
+		DRM_ERROR("Failed to wait for DMA: %d\n", ret);
+}
+
+#define DSI_READ(offset) readl(dsi->regs + (offset))
+#define DSI_WRITE(offset, val) dsi_dma_workaround_write(dsi, offset, val)
+#define DSI_PORT_READ(offset) \
+	DSI_READ(dsi->port ? DSI1_##offset : DSI0_##offset)
+#define DSI_PORT_WRITE(offset, val) \
+	DSI_WRITE(dsi->port ? DSI1_##offset : DSI0_##offset, val)
+#define DSI_PORT_BIT(bit) (dsi->port ? DSI1_##bit : DSI0_##bit)
+
+/* VC4 DSI encoder KMS struct */
+struct vc4_dsi_encoder {
+	struct vc4_encoder base;
+	struct vc4_dsi *dsi;
+};
+
+static inline struct vc4_dsi_encoder *
+to_vc4_dsi_encoder(struct drm_encoder *encoder)
+{
+	return container_of(encoder, struct vc4_dsi_encoder, base.base);
+}
+
+#define DSI_REG(reg) { reg, #reg }
+static const struct {
+	u32 reg;
+	const char *name;
+} dsi0_regs[] = {
+	DSI_REG(DSI0_CTRL),
+	DSI_REG(DSI0_STAT),
+	DSI_REG(DSI0_HSTX_TO_CNT),
+	DSI_REG(DSI0_LPRX_TO_CNT),
+	DSI_REG(DSI0_TA_TO_CNT),
+	DSI_REG(DSI0_PR_TO_CNT),
+	DSI_REG(DSI0_DISP0_CTRL),
+	DSI_REG(DSI0_DISP1_CTRL),
+	DSI_REG(DSI0_INT_STAT),
+	DSI_REG(DSI0_INT_EN),
+	DSI_REG(DSI0_PHYC),
+	DSI_REG(DSI0_HS_CLT0),
+	DSI_REG(DSI0_HS_CLT1),
+	DSI_REG(DSI0_HS_CLT2),
+	DSI_REG(DSI0_HS_DLT3),
+	DSI_REG(DSI0_HS_DLT4),
+	DSI_REG(DSI0_HS_DLT5),
+	DSI_REG(DSI0_HS_DLT6),
+	DSI_REG(DSI0_HS_DLT7),
+	DSI_REG(DSI0_PHY_AFEC0),
+	DSI_REG(DSI0_PHY_AFEC1),
+	DSI_REG(DSI0_ID),
+};
+
+static const struct {
+	u32 reg;
+	const char *name;
+} dsi1_regs[] = {
+	DSI_REG(DSI1_CTRL),
+	DSI_REG(DSI1_STAT),
+	DSI_REG(DSI1_HSTX_TO_CNT),
+	DSI_REG(DSI1_LPRX_TO_CNT),
+	DSI_REG(DSI1_TA_TO_CNT),
+	DSI_REG(DSI1_PR_TO_CNT),
+	DSI_REG(DSI1_DISP0_CTRL),
+	DSI_REG(DSI1_DISP1_CTRL),
+	DSI_REG(DSI1_INT_STAT),
+	DSI_REG(DSI1_INT_EN),
+	DSI_REG(DSI1_PHYC),
+	DSI_REG(DSI1_HS_CLT0),
+	DSI_REG(DSI1_HS_CLT1),
+	DSI_REG(DSI1_HS_CLT2),
+	DSI_REG(DSI1_HS_DLT3),
+	DSI_REG(DSI1_HS_DLT4),
+	DSI_REG(DSI1_HS_DLT5),
+	DSI_REG(DSI1_HS_DLT6),
+	DSI_REG(DSI1_HS_DLT7),
+	DSI_REG(DSI1_PHY_AFEC0),
+	DSI_REG(DSI1_PHY_AFEC1),
+	DSI_REG(DSI1_ID),
+};
+
+static void vc4_dsi_dump_regs(struct vc4_dsi *dsi)
+{
+	int i;
+
+	if (dsi->port == 0) {
+		for (i = 0; i < ARRAY_SIZE(dsi0_regs); i++) {
+			DRM_INFO("0x%04x (%s): 0x%08x\n",
+				 dsi0_regs[i].reg, dsi0_regs[i].name,
+				 DSI_READ(dsi0_regs[i].reg));
+		}
+	} else {
+		for (i = 0; i < ARRAY_SIZE(dsi1_regs); i++) {
+			DRM_INFO("0x%04x (%s): 0x%08x\n",
+				 dsi1_regs[i].reg, dsi1_regs[i].name,
+				 DSI_READ(dsi1_regs[i].reg));
+		}
+	}
+}
+
+#ifdef CONFIG_DEBUG_FS
+int vc4_dsi_debugfs_regs(struct seq_file *m, void *unused)
+{
+	struct drm_info_node *node = (struct drm_info_node *)m->private;
+	struct drm_device *drm = node->minor->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	int dsi_index = (uintptr_t)node->info_ent->data;
+	struct vc4_dsi *dsi = (dsi_index == 1 ? vc4->dsi1 : NULL);
+	int i;
+
+	if (!dsi)
+		return 0;
+
+	if (dsi->port == 0) {
+		for (i = 0; i < ARRAY_SIZE(dsi0_regs); i++) {
+			seq_printf(m, "0x%04x (%s): 0x%08x\n",
+				   dsi0_regs[i].reg, dsi0_regs[i].name,
+				   DSI_READ(dsi0_regs[i].reg));
+		}
+	} else {
+		for (i = 0; i < ARRAY_SIZE(dsi1_regs); i++) {
+			seq_printf(m, "0x%04x (%s): 0x%08x\n",
+				   dsi1_regs[i].reg, dsi1_regs[i].name,
+				   DSI_READ(dsi1_regs[i].reg));
+		}
+	}
+
+	return 0;
+}
+#endif
+
+static void vc4_dsi_encoder_destroy(struct drm_encoder *encoder)
+{
+	drm_encoder_cleanup(encoder);
+}
+
+static const struct drm_encoder_funcs vc4_dsi_encoder_funcs = {
+	.destroy = vc4_dsi_encoder_destroy,
+};
+
+static void vc4_dsi_latch_ulps(struct vc4_dsi *dsi, bool latch)
+{
+	u32 afec0 = DSI_PORT_READ(PHY_AFEC0);
+
+	if (latch)
+		afec0 |= DSI_PORT_BIT(PHY_AFEC0_LATCH_ULPS);
+	else
+		afec0 &= ~DSI_PORT_BIT(PHY_AFEC0_LATCH_ULPS);
+
+	DSI_PORT_WRITE(PHY_AFEC0, afec0);
+}
+
+/* Enters or exits Ultra Low Power State. */
+static void vc4_dsi_ulps(struct vc4_dsi *dsi, bool ulps)
+{
+	bool non_continuous = dsi->mode_flags & MIPI_DSI_CLOCK_NON_CONTINUOUS;
+	u32 phyc_ulps = ((non_continuous ? DSI_PORT_BIT(PHYC_CLANE_ULPS) : 0) |
+			 DSI_PHYC_DLANE0_ULPS |
+			 (dsi->lanes > 1 ? DSI_PHYC_DLANE1_ULPS : 0) |
+			 (dsi->lanes > 2 ? DSI_PHYC_DLANE2_ULPS : 0) |
+			 (dsi->lanes > 3 ? DSI_PHYC_DLANE3_ULPS : 0));
+	u32 stat_ulps = ((non_continuous ? DSI1_STAT_PHY_CLOCK_ULPS : 0) |
+			 DSI1_STAT_PHY_D0_ULPS |
+			 (dsi->lanes > 1 ? DSI1_STAT_PHY_D1_ULPS : 0) |
+			 (dsi->lanes > 2 ? DSI1_STAT_PHY_D2_ULPS : 0) |
+			 (dsi->lanes > 3 ? DSI1_STAT_PHY_D3_ULPS : 0));
+	u32 stat_stop = ((non_continuous ? DSI1_STAT_PHY_CLOCK_STOP : 0) |
+			 DSI1_STAT_PHY_D0_STOP |
+			 (dsi->lanes > 1 ? DSI1_STAT_PHY_D1_STOP : 0) |
+			 (dsi->lanes > 2 ? DSI1_STAT_PHY_D2_STOP : 0) |
+			 (dsi->lanes > 3 ? DSI1_STAT_PHY_D3_STOP : 0));
+	int ret;
+	bool ulps_currently_enabled = (DSI_PORT_READ(PHY_AFEC0) &
+				       DSI_PORT_BIT(PHY_AFEC0_LATCH_ULPS));
+
+	if (ulps == ulps_currently_enabled)
+		return;
+
+	DSI_PORT_WRITE(STAT, stat_ulps);
+	DSI_PORT_WRITE(PHYC, DSI_PORT_READ(PHYC) | phyc_ulps);
+	ret = wait_for((DSI_PORT_READ(STAT) & stat_ulps) == stat_ulps, 200);
+	if (ret) {
+		dev_warn(&dsi->pdev->dev,
+			 "Timeout waiting for DSI ULPS entry: STAT 0x%08x",
+			 DSI_PORT_READ(STAT));
+		DSI_PORT_WRITE(PHYC, DSI_PORT_READ(PHYC) & ~phyc_ulps);
+		vc4_dsi_latch_ulps(dsi, false);
+		return;
+	}
+
+	/* The DSI module can't be disabled while the module is
+	 * generating ULPS state.  So, to be able to disable the
+	 * module, we have the AFE latch the ULPS state and continue
+	 * on to having the module enter STOP.
+	 */
+	vc4_dsi_latch_ulps(dsi, ulps);
+
+	DSI_PORT_WRITE(STAT, stat_stop);
+	DSI_PORT_WRITE(PHYC, DSI_PORT_READ(PHYC) & ~phyc_ulps);
+	ret = wait_for((DSI_PORT_READ(STAT) & stat_stop) == stat_stop, 200);
+	if (ret) {
+		dev_warn(&dsi->pdev->dev,
+			 "Timeout waiting for DSI STOP entry: STAT 0x%08x",
+			 DSI_PORT_READ(STAT));
+		DSI_PORT_WRITE(PHYC, DSI_PORT_READ(PHYC) & ~phyc_ulps);
+		return;
+	}
+}
+
+static u32
+dsi_hs_timing(u32 ui_ns, u32 ns, u32 ui)
+{
+	/* The HS timings have to be rounded up to a multiple of 8
+	 * because we're using the byte clock.
+	 */
+	return roundup(ui + DIV_ROUND_UP(ns, ui_ns), 8);
+}
+
+/* ESC always runs at 100Mhz. */
+#define ESC_TIME_NS 10
+
+static u32
+dsi_esc_timing(u32 ns)
+{
+	return DIV_ROUND_UP(ns, ESC_TIME_NS);
+}
+
+static void vc4_dsi_encoder_disable(struct drm_encoder *encoder)
+{
+	struct vc4_dsi_encoder *vc4_encoder = to_vc4_dsi_encoder(encoder);
+	struct vc4_dsi *dsi = vc4_encoder->dsi;
+	struct device *dev = &dsi->pdev->dev;
+
+	drm_bridge_disable(dsi->bridge);
+	vc4_dsi_ulps(dsi, true);
+	drm_bridge_post_disable(dsi->bridge);
+
+	clk_disable_unprepare(dsi->pll_phy_clock);
+	clk_disable_unprepare(dsi->escape_clock);
+	clk_disable_unprepare(dsi->pixel_clock);
+
+	pm_runtime_put(dev);
+}
+
+/* Extends the mode's blank intervals to handle BCM2835's integer-only
+ * DSI PLL divider.
+ *
+ * On 2835, PLLD is set to 2Ghz, and may not be changed by the display
+ * driver since most peripherals are hanging off of the PLLD_PER
+ * divider.  PLLD_DSI1, which drives our DSI bit clock (and therefore
+ * the pixel clock), only has an integer divider off of DSI.
+ *
+ * To get our panel mode to refresh at the expected 60Hz, we need to
+ * extend the horizontal blank time.  This means we drive a
+ * higher-than-expected clock rate to the panel, but that's what the
+ * firmware does too.
+ */
+static bool vc4_dsi_encoder_mode_fixup(struct drm_encoder *encoder,
+				       const struct drm_display_mode *mode,
+				       struct drm_display_mode *adjusted_mode)
+{
+	struct vc4_dsi_encoder *vc4_encoder = to_vc4_dsi_encoder(encoder);
+	struct vc4_dsi *dsi = vc4_encoder->dsi;
+	struct clk *phy_parent = clk_get_parent(dsi->pll_phy_clock);
+	unsigned long parent_rate = clk_get_rate(phy_parent);
+	unsigned long pixel_clock_hz = mode->clock * 1000;
+	unsigned long pll_clock = pixel_clock_hz * dsi->divider;
+	int divider;
+
+	/* Find what divider gets us a faster clock than the requested
+	 * pixel clock.
+	 */
+	for (divider = 1; divider < 8; divider++) {
+		if (parent_rate / divider < pll_clock) {
+			divider--;
+			break;
+		}
+	}
+
+	/* Now that we've picked a PLL divider, calculate back to its
+	 * pixel clock.
+	 */
+	pll_clock = parent_rate / divider;
+	pixel_clock_hz = pll_clock / dsi->divider;
+
+	adjusted_mode->clock = pixel_clock_hz / 1000;
+
+	/* Given the new pixel clock, adjust HFP to keep vrefresh the same. */
+	adjusted_mode->htotal = adjusted_mode->clock * mode->htotal /
+				mode->clock;
+	adjusted_mode->hsync_end += adjusted_mode->htotal - mode->htotal;
+	adjusted_mode->hsync_start += adjusted_mode->htotal - mode->htotal;
+
+	return true;
+}
+
+static void vc4_dsi_encoder_enable(struct drm_encoder *encoder)
+{
+	struct drm_display_mode *mode = &encoder->crtc->state->adjusted_mode;
+	struct vc4_dsi_encoder *vc4_encoder = to_vc4_dsi_encoder(encoder);
+	struct vc4_dsi *dsi = vc4_encoder->dsi;
+	struct device *dev = &dsi->pdev->dev;
+	bool debug_dump_regs = false;
+	unsigned long hs_clock;
+	u32 ui_ns;
+	/* Minimum LP state duration in escape clock cycles. */
+	u32 lpx = dsi_esc_timing(60);
+	unsigned long pixel_clock_hz = mode->clock * 1000;
+	unsigned long dsip_clock;
+	unsigned long phy_clock;
+	int ret;
+
+	ret = pm_runtime_get_sync(dev);
+	if (ret) {
+		DRM_ERROR("Failed to runtime PM enable on DSI%d\n", dsi->port);
+		return;
+	}
+
+	if (debug_dump_regs) {
+		DRM_INFO("DSI regs before:\n");
+		vc4_dsi_dump_regs(dsi);
+	}
+
+	/* Round up the clk_set_rate() request slightly, since
+	 * PLLD_DSI1 is an integer divider and its rate selection will
+	 * never round up.
+	 */
+	phy_clock = (pixel_clock_hz + 1000) * dsi->divider;
+	ret = clk_set_rate(dsi->pll_phy_clock, phy_clock);
+	if (ret) {
+		dev_err(&dsi->pdev->dev,
+			"Failed to set phy clock to %ld: %d\n", phy_clock, ret);
+	}
+
+	/* Reset the DSI and all its fifos. */
+	DSI_PORT_WRITE(CTRL,
+		       DSI_CTRL_SOFT_RESET_CFG |
+		       DSI_PORT_BIT(CTRL_RESET_FIFOS));
+
+	DSI_PORT_WRITE(CTRL,
+		       DSI_CTRL_HSDT_EOT_DISABLE |
+		       DSI_CTRL_RX_LPDT_EOT_DISABLE);
+
+	/* Clear all stat bits so we see what has happened during enable. */
+	DSI_PORT_WRITE(STAT, DSI_PORT_READ(STAT));
+
+	/* Set AFE CTR00/CTR1 to release powerdown of analog. */
+	if (dsi->port == 0) {
+		u32 afec0 = (VC4_SET_FIELD(7, DSI_PHY_AFEC0_PTATADJ) |
+			     VC4_SET_FIELD(7, DSI_PHY_AFEC0_CTATADJ));
+
+		if (dsi->lanes < 2)
+			afec0 |= DSI0_PHY_AFEC0_PD_DLANE1;
+
+		if (!(dsi->mode_flags & MIPI_DSI_MODE_VIDEO))
+			afec0 |= DSI0_PHY_AFEC0_RESET;
+
+		DSI_PORT_WRITE(PHY_AFEC0, afec0);
+
+		DSI_PORT_WRITE(PHY_AFEC1,
+			       VC4_SET_FIELD(6,  DSI0_PHY_AFEC1_IDR_DLANE1) |
+			       VC4_SET_FIELD(6,  DSI0_PHY_AFEC1_IDR_DLANE0) |
+			       VC4_SET_FIELD(6,  DSI0_PHY_AFEC1_IDR_CLANE));
+	} else {
+		u32 afec0 = (VC4_SET_FIELD(7, DSI_PHY_AFEC0_PTATADJ) |
+			     VC4_SET_FIELD(7, DSI_PHY_AFEC0_CTATADJ) |
+			     VC4_SET_FIELD(6, DSI1_PHY_AFEC0_IDR_CLANE) |
+			     VC4_SET_FIELD(6, DSI1_PHY_AFEC0_IDR_DLANE0) |
+			     VC4_SET_FIELD(6, DSI1_PHY_AFEC0_IDR_DLANE1) |
+			     VC4_SET_FIELD(6, DSI1_PHY_AFEC0_IDR_DLANE2) |
+			     VC4_SET_FIELD(6, DSI1_PHY_AFEC0_IDR_DLANE3));
+
+		if (dsi->lanes < 4)
+			afec0 |= DSI1_PHY_AFEC0_PD_DLANE3;
+		if (dsi->lanes < 3)
+			afec0 |= DSI1_PHY_AFEC0_PD_DLANE2;
+		if (dsi->lanes < 2)
+			afec0 |= DSI1_PHY_AFEC0_PD_DLANE1;
+
+		afec0 |= DSI1_PHY_AFEC0_RESET;
+
+		DSI_PORT_WRITE(PHY_AFEC0, afec0);
+
+		DSI_PORT_WRITE(PHY_AFEC1, 0);
+
+		/* AFEC reset hold time */
+		mdelay(1);
+	}
+
+	ret = clk_prepare_enable(dsi->escape_clock);
+	if (ret) {
+		DRM_ERROR("Failed to turn on DSI escape clock: %d\n", ret);
+		return;
+	}
+
+	ret = clk_prepare_enable(dsi->pll_phy_clock);
+	if (ret) {
+		DRM_ERROR("Failed to turn on DSI PLL: %d\n", ret);
+		return;
+	}
+
+	hs_clock = clk_get_rate(dsi->pll_phy_clock);
+
+	/* Yes, we set the DSI0P/DSI1P pixel clock to the byte rate,
+	 * not the pixel clock rate.  DSIxP take from the APHY's byte,
+	 * DDR2, or DDR4 clock (we use byte) and feed into the PV at
+	 * that rate.  Separately, a value derived from PIX_CLK_DIV
+	 * and HS_CLKC is fed into the PV to divide down to the actual
+	 * pixel clock for pushing pixels into DSI.
+	 */
+	dsip_clock = phy_clock / 8;
+	ret = clk_set_rate(dsi->pixel_clock, dsip_clock);
+	if (ret) {
+		dev_err(dev, "Failed to set pixel clock to %ldHz: %d\n",
+			dsip_clock, ret);
+	}
+
+	ret = clk_prepare_enable(dsi->pixel_clock);
+	if (ret) {
+		DRM_ERROR("Failed to turn on DSI pixel clock: %d\n", ret);
+		return;
+	}
+
+	/* How many ns one DSI unit interval is.  Note that the clock
+	 * is DDR, so there's an extra divide by 2.
+	 */
+	ui_ns = DIV_ROUND_UP(500000000, hs_clock);
+
+	DSI_PORT_WRITE(HS_CLT0,
+		       VC4_SET_FIELD(dsi_hs_timing(ui_ns, 262, 0),
+				     DSI_HS_CLT0_CZERO) |
+		       VC4_SET_FIELD(dsi_hs_timing(ui_ns, 0, 8),
+				     DSI_HS_CLT0_CPRE) |
+		       VC4_SET_FIELD(dsi_hs_timing(ui_ns, 38, 0),
+				     DSI_HS_CLT0_CPREP));
+
+	DSI_PORT_WRITE(HS_CLT1,
+		       VC4_SET_FIELD(dsi_hs_timing(ui_ns, 60, 0),
+				     DSI_HS_CLT1_CTRAIL) |
+		       VC4_SET_FIELD(dsi_hs_timing(ui_ns, 60, 52),
+				     DSI_HS_CLT1_CPOST));
+
+	DSI_PORT_WRITE(HS_CLT2,
+		       VC4_SET_FIELD(dsi_hs_timing(ui_ns, 1000000, 0),
+				     DSI_HS_CLT2_WUP));
+
+	DSI_PORT_WRITE(HS_DLT3,
+		       VC4_SET_FIELD(dsi_hs_timing(ui_ns, 100, 0),
+				     DSI_HS_DLT3_EXIT) |
+		       VC4_SET_FIELD(dsi_hs_timing(ui_ns, 105, 6),
+				     DSI_HS_DLT3_ZERO) |
+		       VC4_SET_FIELD(dsi_hs_timing(ui_ns, 40, 4),
+				     DSI_HS_DLT3_PRE));
+
+	DSI_PORT_WRITE(HS_DLT4,
+		       VC4_SET_FIELD(dsi_hs_timing(ui_ns, lpx * ESC_TIME_NS, 0),
+				     DSI_HS_DLT4_LPX) |
+		       VC4_SET_FIELD(max(dsi_hs_timing(ui_ns, 0, 8),
+					 dsi_hs_timing(ui_ns, 60, 4)),
+				     DSI_HS_DLT4_TRAIL) |
+		       VC4_SET_FIELD(0, DSI_HS_DLT4_ANLAT));
+
+	/* T_INIT is how long STOP is driven after power-up to
+	 * indicate to the slave (also coming out of power-up) that
+	 * master init is complete, and should be greater than the
+	 * maximum of two value: T_INIT,MASTER and T_INIT,SLAVE.  The
+	 * D-PHY spec gives a minimum 100us for T_INIT,MASTER and
+	 * T_INIT,SLAVE, while allowing protocols on top of it to give
+	 * greater minimums.  The vc4 firmware uses an extremely
+	 * conservative 5ms, and we maintain that here.
+	 */
+	DSI_PORT_WRITE(HS_DLT5, VC4_SET_FIELD(dsi_hs_timing(ui_ns,
+							    5 * 1000 * 1000, 0),
+					      DSI_HS_DLT5_INIT));
+
+	DSI_PORT_WRITE(HS_DLT6,
+		       VC4_SET_FIELD(lpx * 5, DSI_HS_DLT6_TA_GET) |
+		       VC4_SET_FIELD(lpx, DSI_HS_DLT6_TA_SURE) |
+		       VC4_SET_FIELD(lpx * 4, DSI_HS_DLT6_TA_GO) |
+		       VC4_SET_FIELD(lpx, DSI_HS_DLT6_LP_LPX));
+
+	DSI_PORT_WRITE(HS_DLT7,
+		       VC4_SET_FIELD(dsi_esc_timing(1000000),
+				     DSI_HS_DLT7_LP_WUP));
+
+	DSI_PORT_WRITE(PHYC,
+		       DSI_PHYC_DLANE0_ENABLE |
+		       (dsi->lanes >= 2 ? DSI_PHYC_DLANE1_ENABLE : 0) |
+		       (dsi->lanes >= 3 ? DSI_PHYC_DLANE2_ENABLE : 0) |
+		       (dsi->lanes >= 4 ? DSI_PHYC_DLANE3_ENABLE : 0) |
+		       DSI_PORT_BIT(PHYC_CLANE_ENABLE) |
+		       ((dsi->mode_flags & MIPI_DSI_CLOCK_NON_CONTINUOUS) ?
+			0 : DSI_PORT_BIT(PHYC_HS_CLK_CONTINUOUS)) |
+		       (dsi->port == 0 ?
+			VC4_SET_FIELD(lpx - 1, DSI0_PHYC_ESC_CLK_LPDT) :
+			VC4_SET_FIELD(lpx - 1, DSI1_PHYC_ESC_CLK_LPDT)));
+
+	DSI_PORT_WRITE(CTRL,
+		       DSI_PORT_READ(CTRL) |
+		       DSI_CTRL_CAL_BYTE);
+
+	/* HS timeout in HS clock cycles: disabled. */
+	DSI_PORT_WRITE(HSTX_TO_CNT, 0);
+	/* LP receive timeout in HS clocks. */
+	DSI_PORT_WRITE(LPRX_TO_CNT, 0xffffff);
+	/* Bus turnaround timeout */
+	DSI_PORT_WRITE(TA_TO_CNT, 100000);
+	/* Display reset sequence timeout */
+	DSI_PORT_WRITE(PR_TO_CNT, 100000);
+
+	/* Set up DISP1 for transferring long command payloads through
+	 * the pixfifo.
+	 */
+	DSI_PORT_WRITE(DISP1_CTRL,
+		       VC4_SET_FIELD(DSI_DISP1_PFORMAT_32BIT_LE,
+				     DSI_DISP1_PFORMAT) |
+		       DSI_DISP1_ENABLE);
+
+	/* Ungate the block. */
+	if (dsi->port == 0)
+		DSI_PORT_WRITE(CTRL, DSI_PORT_READ(CTRL) | DSI0_CTRL_CTRL0);
+	else
+		DSI_PORT_WRITE(CTRL, DSI_PORT_READ(CTRL) | DSI1_CTRL_EN);
+
+	/* Bring AFE out of reset. */
+	if (dsi->port == 0) {
+	} else {
+		DSI_PORT_WRITE(PHY_AFEC0,
+			       DSI_PORT_READ(PHY_AFEC0) &
+			       ~DSI1_PHY_AFEC0_RESET);
+	}
+
+	vc4_dsi_ulps(dsi, false);
+
+	drm_bridge_pre_enable(dsi->bridge);
+
+	if (dsi->mode_flags & MIPI_DSI_MODE_VIDEO) {
+		DSI_PORT_WRITE(DISP0_CTRL,
+			       VC4_SET_FIELD(dsi->divider,
+					     DSI_DISP0_PIX_CLK_DIV) |
+			       VC4_SET_FIELD(dsi->format, DSI_DISP0_PFORMAT) |
+			       VC4_SET_FIELD(DSI_DISP0_LP_STOP_PERFRAME,
+					     DSI_DISP0_LP_STOP_CTRL) |
+			       DSI_DISP0_ST_END |
+			       DSI_DISP0_ENABLE);
+	} else {
+		DSI_PORT_WRITE(DISP0_CTRL,
+			       DSI_DISP0_COMMAND_MODE |
+			       DSI_DISP0_ENABLE);
+	}
+
+	drm_bridge_enable(dsi->bridge);
+
+	if (debug_dump_regs) {
+		DRM_INFO("DSI regs after:\n");
+		vc4_dsi_dump_regs(dsi);
+	}
+}
+
+static ssize_t vc4_dsi_host_transfer(struct mipi_dsi_host *host,
+				     const struct mipi_dsi_msg *msg)
+{
+	struct vc4_dsi *dsi = host_to_dsi(host);
+	struct mipi_dsi_packet packet;
+	u32 pkth = 0, pktc = 0;
+	int i, ret;
+	bool is_long = mipi_dsi_packet_format_is_long(msg->type);
+	u32 cmd_fifo_len = 0, pix_fifo_len = 0;
+
+	mipi_dsi_create_packet(&packet, msg);
+
+	pkth |= VC4_SET_FIELD(packet.header[0], DSI_TXPKT1H_BC_DT);
+	pkth |= VC4_SET_FIELD(packet.header[1] |
+			      (packet.header[2] << 8),
+			      DSI_TXPKT1H_BC_PARAM);
+	if (is_long) {
+		/* Divide data across the various FIFOs we have available.
+		 * The command FIFO takes byte-oriented data, but is of
+		 * limited size. The pixel FIFO (never actually used for
+		 * pixel data in reality) is word oriented, and substantially
+		 * larger. So, we use the pixel FIFO for most of the data,
+		 * sending the residual bytes in the command FIFO at the start.
+		 *
+		 * With this arrangement, the command FIFO will never get full.
+		 */
+		if (packet.payload_length <= 16) {
+			cmd_fifo_len = packet.payload_length;
+			pix_fifo_len = 0;
+		} else {
+			cmd_fifo_len = (packet.payload_length %
+					DSI_PIX_FIFO_WIDTH);
+			pix_fifo_len = ((packet.payload_length - cmd_fifo_len) /
+					DSI_PIX_FIFO_WIDTH);
+		}
+
+		WARN_ON_ONCE(pix_fifo_len >= DSI_PIX_FIFO_DEPTH);
+
+		pkth |= VC4_SET_FIELD(cmd_fifo_len, DSI_TXPKT1H_BC_CMDFIFO);
+	}
+
+	if (msg->rx_len) {
+		pktc |= VC4_SET_FIELD(DSI_TXPKT1C_CMD_CTRL_RX,
+				      DSI_TXPKT1C_CMD_CTRL);
+	} else {
+		pktc |= VC4_SET_FIELD(DSI_TXPKT1C_CMD_CTRL_TX,
+				      DSI_TXPKT1C_CMD_CTRL);
+	}
+
+	for (i = 0; i < cmd_fifo_len; i++)
+		DSI_PORT_WRITE(TXPKT_CMD_FIFO, packet.payload[i]);
+	for (i = 0; i < pix_fifo_len; i++) {
+		const u8 *pix = packet.payload + cmd_fifo_len + i * 4;
+
+		DSI_PORT_WRITE(TXPKT_PIX_FIFO,
+			       pix[0] |
+			       pix[1] << 8 |
+			       pix[2] << 16 |
+			       pix[3] << 24);
+	}
+
+	if (msg->flags & MIPI_DSI_MSG_USE_LPM)
+		pktc |= DSI_TXPKT1C_CMD_MODE_LP;
+	if (is_long)
+		pktc |= DSI_TXPKT1C_CMD_TYPE_LONG;
+
+	/* Send one copy of the packet.  Larger repeats are used for pixel
+	 * data in command mode.
+	 */
+	pktc |= VC4_SET_FIELD(1, DSI_TXPKT1C_CMD_REPEAT);
+
+	pktc |= DSI_TXPKT1C_CMD_EN;
+	if (pix_fifo_len) {
+		pktc |= VC4_SET_FIELD(DSI_TXPKT1C_DISPLAY_NO_SECONDARY,
+				      DSI_TXPKT1C_DISPLAY_NO);
+	} else {
+		pktc |= VC4_SET_FIELD(DSI_TXPKT1C_DISPLAY_NO_SHORT,
+				      DSI_TXPKT1C_DISPLAY_NO);
+	}
+
+	/* Enable the appropriate interrupt for the transfer completion. */
+	dsi->xfer_result = 0;
+	reinit_completion(&dsi->xfer_completion);
+	DSI_PORT_WRITE(INT_STAT, DSI1_INT_TXPKT1_DONE | DSI1_INT_PHY_DIR_RTF);
+	if (msg->rx_len) {
+		DSI_PORT_WRITE(INT_EN, (DSI1_INTERRUPTS_ALWAYS_ENABLED |
+					DSI1_INT_PHY_DIR_RTF));
+	} else {
+		DSI_PORT_WRITE(INT_EN, (DSI1_INTERRUPTS_ALWAYS_ENABLED |
+					DSI1_INT_TXPKT1_DONE));
+	}
+
+	/* Send the packet. */
+	DSI_PORT_WRITE(TXPKT1H, pkth);
+	DSI_PORT_WRITE(TXPKT1C, pktc);
+
+	if (!wait_for_completion_timeout(&dsi->xfer_completion,
+					 msecs_to_jiffies(1000))) {
+		dev_err(&dsi->pdev->dev, "transfer interrupt wait timeout");
+		dev_err(&dsi->pdev->dev, "instat: 0x%08x\n",
+			DSI_PORT_READ(INT_STAT));
+		ret = -ETIMEDOUT;
+	} else {
+		ret = dsi->xfer_result;
+	}
+
+	DSI_PORT_WRITE(INT_EN, DSI1_INTERRUPTS_ALWAYS_ENABLED);
+
+	if (ret)
+		goto reset_fifo_and_return;
+
+	if (ret == 0 && msg->rx_len) {
+		u32 rxpkt1h = DSI_PORT_READ(RXPKT1H);
+		u8 *msg_rx = msg->rx_buf;
+
+		if (rxpkt1h & DSI_RXPKT1H_PKT_TYPE_LONG) {
+			u32 rxlen = VC4_GET_FIELD(rxpkt1h,
+						  DSI_RXPKT1H_BC_PARAM);
+
+			if (rxlen != msg->rx_len) {
+				DRM_ERROR("DSI returned %db, expecting %db\n",
+					  rxlen, (int)msg->rx_len);
+				ret = -ENXIO;
+				goto reset_fifo_and_return;
+			}
+
+			for (i = 0; i < msg->rx_len; i++)
+				msg_rx[i] = DSI_READ(DSI1_RXPKT_FIFO);
+		} else {
+			/* FINISHME: Handle AWER */
+
+			msg_rx[0] = VC4_GET_FIELD(rxpkt1h,
+						  DSI_RXPKT1H_SHORT_0);
+			if (msg->rx_len > 1) {
+				msg_rx[1] = VC4_GET_FIELD(rxpkt1h,
+							  DSI_RXPKT1H_SHORT_1);
+			}
+		}
+	}
+
+	return ret;
+
+reset_fifo_and_return:
+	DRM_ERROR("DSI transfer failed, resetting: %d\n", ret);
+
+	DSI_PORT_WRITE(TXPKT1C, DSI_PORT_READ(TXPKT1C) & ~DSI_TXPKT1C_CMD_EN);
+	udelay(1);
+	DSI_PORT_WRITE(CTRL,
+		       DSI_PORT_READ(CTRL) |
+		       DSI_PORT_BIT(CTRL_RESET_FIFOS));
+
+	DSI_PORT_WRITE(TXPKT1C, 0);
+	DSI_PORT_WRITE(INT_EN, DSI1_INTERRUPTS_ALWAYS_ENABLED);
+	return ret;
+}
+
+static int vc4_dsi_host_attach(struct mipi_dsi_host *host,
+			       struct mipi_dsi_device *device)
+{
+	struct vc4_dsi *dsi = host_to_dsi(host);
+
+	dsi->lanes = device->lanes;
+	dsi->channel = device->channel;
+	dsi->mode_flags = device->mode_flags;
+
+	switch (device->format) {
+	case MIPI_DSI_FMT_RGB888:
+		dsi->format = DSI_PFORMAT_RGB888;
+		dsi->divider = 24 / dsi->lanes;
+		break;
+	case MIPI_DSI_FMT_RGB666:
+		dsi->format = DSI_PFORMAT_RGB666;
+		dsi->divider = 24 / dsi->lanes;
+		break;
+	case MIPI_DSI_FMT_RGB666_PACKED:
+		dsi->format = DSI_PFORMAT_RGB666_PACKED;
+		dsi->divider = 18 / dsi->lanes;
+		break;
+	case MIPI_DSI_FMT_RGB565:
+		dsi->format = DSI_PFORMAT_RGB565;
+		dsi->divider = 16 / dsi->lanes;
+		break;
+	default:
+		dev_err(&dsi->pdev->dev, "Unknown DSI format: %d.\n",
+			dsi->format);
+		return 0;
+	}
+
+	if (!(dsi->mode_flags & MIPI_DSI_MODE_VIDEO)) {
+		dev_err(&dsi->pdev->dev,
+			"Only VIDEO mode panels supported currently.\n");
+		return 0;
+	}
+
+	return 0;
+}
+
+static int vc4_dsi_host_detach(struct mipi_dsi_host *host,
+			       struct mipi_dsi_device *device)
+{
+	return 0;
+}
+
+static const struct mipi_dsi_host_ops vc4_dsi_host_ops = {
+	.attach = vc4_dsi_host_attach,
+	.detach = vc4_dsi_host_detach,
+	.transfer = vc4_dsi_host_transfer,
+};
+
+static const struct drm_encoder_helper_funcs vc4_dsi_encoder_helper_funcs = {
+	.disable = vc4_dsi_encoder_disable,
+	.enable = vc4_dsi_encoder_enable,
+	.mode_fixup = vc4_dsi_encoder_mode_fixup,
+};
+
+static const struct of_device_id vc4_dsi_dt_match[] = {
+	{ .compatible = "brcm,bcm2835-dsi1", (void *)(uintptr_t)1 },
+	{}
+};
+
+static void dsi_handle_error(struct vc4_dsi *dsi,
+			     irqreturn_t *ret, u32 stat, u32 bit,
+			     const char *type)
+{
+	if (!(stat & bit))
+		return;
+
+	DRM_ERROR("DSI%d: %s error\n", dsi->port, type);
+	*ret = IRQ_HANDLED;
+}
+
+/*
+ * Initial handler for port 1 where we need the reg_dma workaround.
+ * The register DMA writes sleep, so we can't do it in the top half.
+ * Instead we use IRQF_ONESHOT so that the IRQ gets disabled in the
+ * parent interrupt contrller until our interrupt thread is done.
+ */
+static irqreturn_t vc4_dsi_irq_defer_to_thread_handler(int irq, void *data)
+{
+	struct vc4_dsi *dsi = data;
+	u32 stat = DSI_PORT_READ(INT_STAT);
+
+	if (!stat)
+		return IRQ_NONE;
+
+	return IRQ_WAKE_THREAD;
+}
+
+/*
+ * Normal IRQ handler for port 0, or the threaded IRQ handler for port
+ * 1 where we need the reg_dma workaround.
+ */
+static irqreturn_t vc4_dsi_irq_handler(int irq, void *data)
+{
+	struct vc4_dsi *dsi = data;
+	u32 stat = DSI_PORT_READ(INT_STAT);
+	irqreturn_t ret = IRQ_NONE;
+
+	DSI_PORT_WRITE(INT_STAT, stat);
+
+	dsi_handle_error(dsi, &ret, stat,
+			 DSI1_INT_ERR_SYNC_ESC, "LPDT sync");
+	dsi_handle_error(dsi, &ret, stat,
+			 DSI1_INT_ERR_CONTROL, "data lane 0 sequence");
+	dsi_handle_error(dsi, &ret, stat,
+			 DSI1_INT_ERR_CONT_LP0, "LP0 contention");
+	dsi_handle_error(dsi, &ret, stat,
+			 DSI1_INT_ERR_CONT_LP1, "LP1 contention");
+	dsi_handle_error(dsi, &ret, stat,
+			 DSI1_INT_HSTX_TO, "HSTX timeout");
+	dsi_handle_error(dsi, &ret, stat,
+			 DSI1_INT_LPRX_TO, "LPRX timeout");
+	dsi_handle_error(dsi, &ret, stat,
+			 DSI1_INT_TA_TO, "turnaround timeout");
+	dsi_handle_error(dsi, &ret, stat,
+			 DSI1_INT_PR_TO, "peripheral reset timeout");
+
+	if (stat & (DSI1_INT_TXPKT1_DONE | DSI1_INT_PHY_DIR_RTF)) {
+		complete(&dsi->xfer_completion);
+		ret = IRQ_HANDLED;
+	} else if (stat & DSI1_INT_HSTX_TO) {
+		complete(&dsi->xfer_completion);
+		dsi->xfer_result = -ETIMEDOUT;
+		ret = IRQ_HANDLED;
+	}
+
+	return ret;
+}
+
+/**
+ * vc4_dsi_init_phy_clocks - Exposes clocks generated by the analog
+ * PHY that are consumed by CPRMAN (clk-bcm2835.c).
+ * @dsi: DSI encoder
+ */
+static int
+vc4_dsi_init_phy_clocks(struct vc4_dsi *dsi)
+{
+	struct device *dev = &dsi->pdev->dev;
+	const char *parent_name = __clk_get_name(dsi->pll_phy_clock);
+	static const struct {
+		const char *dsi0_name, *dsi1_name;
+		int div;
+	} phy_clocks[] = {
+		{ "dsi0_byte", "dsi1_byte", 8 },
+		{ "dsi0_ddr2", "dsi1_ddr2", 4 },
+		{ "dsi0_ddr", "dsi1_ddr", 2 },
+	};
+	int i;
+
+	dsi->clk_onecell = devm_kzalloc(dev,
+					sizeof(*dsi->clk_onecell) +
+					ARRAY_SIZE(phy_clocks) *
+					sizeof(struct clk_hw *),
+					GFP_KERNEL);
+	if (!dsi->clk_onecell)
+		return -ENOMEM;
+	dsi->clk_onecell->num = ARRAY_SIZE(phy_clocks);
+
+	for (i = 0; i < ARRAY_SIZE(phy_clocks); i++) {
+		struct clk_fixed_factor *fix = &dsi->phy_clocks[i];
+		struct clk_init_data init;
+		int ret;
+
+		/* We just use core fixed factor clock ops for the PHY
+		 * clocks.  The clocks are actually gated by the
+		 * PHY_AFEC0_DDRCLK_EN bits, which we should be
+		 * setting if we use the DDR/DDR2 clocks.  However,
+		 * vc4_dsi_encoder_enable() is setting up both AFEC0,
+		 * setting both our parent DSI PLL's rate and this
+		 * clock's rate, so it knows if DDR/DDR2 are going to
+		 * be used and could enable the gates itself.
+		 */
+		fix->mult = 1;
+		fix->div = phy_clocks[i].div;
+		fix->hw.init = &init;
+
+		memset(&init, 0, sizeof(init));
+		init.parent_names = &parent_name;
+		init.num_parents = 1;
+		if (dsi->port == 1)
+			init.name = phy_clocks[i].dsi1_name;
+		else
+			init.name = phy_clocks[i].dsi0_name;
+		init.ops = &clk_fixed_factor_ops;
+
+		ret = devm_clk_hw_register(dev, &fix->hw);
+		if (ret)
+			return ret;
+
+		dsi->clk_onecell->hws[i] = &fix->hw;
+	}
+
+	return of_clk_add_hw_provider(dev->of_node,
+				      of_clk_hw_onecell_get,
+				      dsi->clk_onecell);
+}
+
+static int vc4_dsi_bind(struct device *dev, struct device *master, void *data)
+{
+	struct platform_device *pdev = to_platform_device(dev);
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	struct vc4_dsi *dsi = dev_get_drvdata(dev);
+	struct vc4_dsi_encoder *vc4_dsi_encoder;
+	struct drm_panel *panel;
+	const struct of_device_id *match;
+	dma_cap_mask_t dma_mask;
+	int ret;
+
+	match = of_match_device(vc4_dsi_dt_match, dev);
+	if (!match)
+		return -ENODEV;
+
+	dsi->port = (uintptr_t)match->data;
+
+	vc4_dsi_encoder = devm_kzalloc(dev, sizeof(*vc4_dsi_encoder),
+				       GFP_KERNEL);
+	if (!vc4_dsi_encoder)
+		return -ENOMEM;
+	vc4_dsi_encoder->base.type = VC4_ENCODER_TYPE_DSI1;
+	vc4_dsi_encoder->dsi = dsi;
+	dsi->encoder = &vc4_dsi_encoder->base.base;
+
+	dsi->regs = vc4_ioremap_regs(pdev, 0);
+	if (IS_ERR(dsi->regs))
+		return PTR_ERR(dsi->regs);
+
+	if (DSI_PORT_READ(ID) != DSI_ID_VALUE) {
+		dev_err(dev, "Port returned 0x%08x for ID instead of 0x%08x\n",
+			DSI_PORT_READ(ID), DSI_ID_VALUE);
+		return -ENODEV;
+	}
+
+	/* DSI1 has a broken AXI slave that doesn't respond to writes
+	 * from the ARM.  It does handle writes from the DMA engine,
+	 * so set up a channel for talking to it.
+	 */
+	if (dsi->port == 1) {
+		dsi->reg_dma_mem = dma_alloc_coherent(dev, 4,
+						      &dsi->reg_dma_paddr,
+						      GFP_KERNEL);
+		if (!dsi->reg_dma_mem) {
+			DRM_ERROR("Failed to get DMA memory\n");
+			return -ENOMEM;
+		}
+
+		dma_cap_zero(dma_mask);
+		dma_cap_set(DMA_MEMCPY, dma_mask);
+		dsi->reg_dma_chan = dma_request_chan_by_mask(&dma_mask);
+		if (IS_ERR(dsi->reg_dma_chan)) {
+			ret = PTR_ERR(dsi->reg_dma_chan);
+			if (ret != -EPROBE_DEFER)
+				DRM_ERROR("Failed to get DMA channel: %d\n",
+					  ret);
+			return ret;
+		}
+
+		/* Get the physical address of the device's registers.  The
+		 * struct resource for the regs gives us the bus address
+		 * instead.
+		 */
+		dsi->reg_paddr = be32_to_cpup(of_get_address(dev->of_node,
+							     0, NULL, NULL));
+	}
+
+	init_completion(&dsi->xfer_completion);
+	/* At startup enable error-reporting interrupts and nothing else. */
+	DSI_PORT_WRITE(INT_EN, DSI1_INTERRUPTS_ALWAYS_ENABLED);
+	/* Clear any existing interrupt state. */
+	DSI_PORT_WRITE(INT_STAT, DSI_PORT_READ(INT_STAT));
+
+	if (dsi->reg_dma_mem)
+		ret = devm_request_threaded_irq(dev, platform_get_irq(pdev, 0),
+						vc4_dsi_irq_defer_to_thread_handler,
+						vc4_dsi_irq_handler,
+						IRQF_ONESHOT,
+						"vc4 dsi", dsi);
+	else
+		ret = devm_request_irq(dev, platform_get_irq(pdev, 0),
+				       vc4_dsi_irq_handler, 0, "vc4 dsi", dsi);
+	if (ret) {
+		if (ret != -EPROBE_DEFER)
+			dev_err(dev, "Failed to get interrupt: %d\n", ret);
+		return ret;
+	}
+
+	dsi->escape_clock = devm_clk_get(dev, "escape");
+	if (IS_ERR(dsi->escape_clock)) {
+		ret = PTR_ERR(dsi->escape_clock);
+		if (ret != -EPROBE_DEFER)
+			dev_err(dev, "Failed to get escape clock: %d\n", ret);
+		return ret;
+	}
+
+	dsi->pll_phy_clock = devm_clk_get(dev, "phy");
+	if (IS_ERR(dsi->pll_phy_clock)) {
+		ret = PTR_ERR(dsi->pll_phy_clock);
+		if (ret != -EPROBE_DEFER)
+			dev_err(dev, "Failed to get phy clock: %d\n", ret);
+		return ret;
+	}
+
+	dsi->pixel_clock = devm_clk_get(dev, "pixel");
+	if (IS_ERR(dsi->pixel_clock)) {
+		ret = PTR_ERR(dsi->pixel_clock);
+		if (ret != -EPROBE_DEFER)
+			dev_err(dev, "Failed to get pixel clock: %d\n", ret);
+		return ret;
+	}
+
+	ret = drm_of_find_panel_or_bridge(dev->of_node, 0, 0,
+					  &panel, &dsi->bridge);
+	if (ret) {
+		/* If the bridge or panel pointed by dev->of_node is not
+		 * enabled, just return 0 here so that we don't prevent the DRM
+		 * dev from being registered. Of course that means the DSI
+		 * encoder won't be exposed, but that's not a problem since
+		 * nothing is connected to it.
+		 */
+		if (ret == -ENODEV)
+			return 0;
+
+		return ret;
+	}
+
+	if (panel) {
+		dsi->bridge = devm_drm_panel_bridge_add(dev, panel,
+							DRM_MODE_CONNECTOR_DSI);
+		if (IS_ERR(dsi->bridge))
+			return PTR_ERR(dsi->bridge);
+	}
+
+	/* The esc clock rate is supposed to always be 100Mhz. */
+	ret = clk_set_rate(dsi->escape_clock, 100 * 1000000);
+	if (ret) {
+		dev_err(dev, "Failed to set esc clock: %d\n", ret);
+		return ret;
+	}
+
+	ret = vc4_dsi_init_phy_clocks(dsi);
+	if (ret)
+		return ret;
+
+	if (dsi->port == 1)
+		vc4->dsi1 = dsi;
+
+	drm_encoder_init(drm, dsi->encoder, &vc4_dsi_encoder_funcs,
+			 DRM_MODE_ENCODER_DSI, NULL);
+	drm_encoder_helper_add(dsi->encoder, &vc4_dsi_encoder_helper_funcs);
+
+	ret = drm_bridge_attach(dsi->encoder, dsi->bridge, NULL);
+	if (ret) {
+		dev_err(dev, "bridge attach failed: %d\n", ret);
+		return ret;
+	}
+	/* Disable the atomic helper calls into the bridge.  We
+	 * manually call the bridge pre_enable / enable / etc. calls
+	 * from our driver, since we need to sequence them within the
+	 * encoder's enable/disable paths.
+	 */
+	dsi->encoder->bridge = NULL;
+
+	pm_runtime_enable(dev);
+
+	return 0;
+}
+
+static void vc4_dsi_unbind(struct device *dev, struct device *master,
+			   void *data)
+{
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	struct vc4_dsi *dsi = dev_get_drvdata(dev);
+
+	if (dsi->bridge)
+		pm_runtime_disable(dev);
+
+	vc4_dsi_encoder_destroy(dsi->encoder);
+
+	if (dsi->port == 1)
+		vc4->dsi1 = NULL;
+}
+
+static const struct component_ops vc4_dsi_ops = {
+	.bind   = vc4_dsi_bind,
+	.unbind = vc4_dsi_unbind,
+};
+
+static int vc4_dsi_dev_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	struct vc4_dsi *dsi;
+	int ret;
+
+	dsi = devm_kzalloc(dev, sizeof(*dsi), GFP_KERNEL);
+	if (!dsi)
+		return -ENOMEM;
+	dev_set_drvdata(dev, dsi);
+
+	dsi->pdev = pdev;
+
+	/* Note, the initialization sequence for DSI and panels is
+	 * tricky.  The component bind above won't get past its
+	 * -EPROBE_DEFER until the panel/bridge probes.  The
+	 * panel/bridge will return -EPROBE_DEFER until it has a
+	 * mipi_dsi_host to register its device to.  So, we register
+	 * the host during pdev probe time, so vc4 as a whole can then
+	 * -EPROBE_DEFER its component bind process until the panel
+	 * successfully attaches.
+	 */
+	dsi->dsi_host.ops = &vc4_dsi_host_ops;
+	dsi->dsi_host.dev = dev;
+	mipi_dsi_host_register(&dsi->dsi_host);
+
+	ret = component_add(&pdev->dev, &vc4_dsi_ops);
+	if (ret) {
+		mipi_dsi_host_unregister(&dsi->dsi_host);
+		return ret;
+	}
+
+	return 0;
+}
+
+static int vc4_dsi_dev_remove(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	struct vc4_dsi *dsi = dev_get_drvdata(dev);
+
+	component_del(&pdev->dev, &vc4_dsi_ops);
+	mipi_dsi_host_unregister(&dsi->dsi_host);
+
+	return 0;
+}
+
+struct platform_driver vc4_dsi_driver = {
+	.probe = vc4_dsi_dev_probe,
+	.remove = vc4_dsi_dev_remove,
+	.driver = {
+		.name = "vc4_dsi",
+		.of_match_table = vc4_dsi_dt_match,
+	},
+};
diff --git a/drivers/gpu/drm/vc4/vc4_fence.c b/drivers/gpu/drm/vc4/vc4_fence.c
new file mode 100644
index 000000000..580214e21
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_fence.c
@@ -0,0 +1,48 @@
+/*
+ * Copyright © 2017 Broadcom
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#include "vc4_drv.h"
+
+static const char *vc4_fence_get_driver_name(struct dma_fence *fence)
+{
+	return "vc4";
+}
+
+static const char *vc4_fence_get_timeline_name(struct dma_fence *fence)
+{
+	return "vc4-v3d";
+}
+
+static bool vc4_fence_signaled(struct dma_fence *fence)
+{
+	struct vc4_fence *f = to_vc4_fence(fence);
+	struct vc4_dev *vc4 = to_vc4_dev(f->dev);
+
+	return vc4->finished_seqno >= f->seqno;
+}
+
+const struct dma_fence_ops vc4_fence_ops = {
+	.get_driver_name = vc4_fence_get_driver_name,
+	.get_timeline_name = vc4_fence_get_timeline_name,
+	.signaled = vc4_fence_signaled,
+};
diff --git a/drivers/gpu/drm/vc4/vc4_gem.c b/drivers/gpu/drm/vc4/vc4_gem.c
new file mode 100644
index 000000000..7910b9ace
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_gem.c
@@ -0,0 +1,1377 @@
+/*
+ * Copyright © 2014 Broadcom
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/pm_runtime.h>
+#include <linux/device.h>
+#include <linux/io.h>
+#include <linux/sched/signal.h>
+#include <linux/dma-fence-array.h>
+
+#include "uapi/drm/vc4_drm.h"
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+#include "vc4_trace.h"
+
+static void
+vc4_queue_hangcheck(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	mod_timer(&vc4->hangcheck.timer,
+		  round_jiffies_up(jiffies + msecs_to_jiffies(100)));
+}
+
+struct vc4_hang_state {
+	struct drm_vc4_get_hang_state user_state;
+
+	u32 bo_count;
+	struct drm_gem_object **bo;
+};
+
+static void
+vc4_free_hang_state(struct drm_device *dev, struct vc4_hang_state *state)
+{
+	unsigned int i;
+
+	for (i = 0; i < state->user_state.bo_count; i++)
+		drm_gem_object_put_unlocked(state->bo[i]);
+
+	kfree(state);
+}
+
+int
+vc4_get_hang_state_ioctl(struct drm_device *dev, void *data,
+			 struct drm_file *file_priv)
+{
+	struct drm_vc4_get_hang_state *get_state = data;
+	struct drm_vc4_get_hang_state_bo *bo_state;
+	struct vc4_hang_state *kernel_state;
+	struct drm_vc4_get_hang_state *state;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	unsigned long irqflags;
+	u32 i;
+	int ret = 0;
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	kernel_state = vc4->hang_state;
+	if (!kernel_state) {
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		return -ENOENT;
+	}
+	state = &kernel_state->user_state;
+
+	/* If the user's array isn't big enough, just return the
+	 * required array size.
+	 */
+	if (get_state->bo_count < state->bo_count) {
+		get_state->bo_count = state->bo_count;
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		return 0;
+	}
+
+	vc4->hang_state = NULL;
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+
+	/* Save the user's BO pointer, so we don't stomp it with the memcpy. */
+	state->bo = get_state->bo;
+	memcpy(get_state, state, sizeof(*state));
+
+	bo_state = kcalloc(state->bo_count, sizeof(*bo_state), GFP_KERNEL);
+	if (!bo_state) {
+		ret = -ENOMEM;
+		goto err_free;
+	}
+
+	for (i = 0; i < state->bo_count; i++) {
+		struct vc4_bo *vc4_bo = to_vc4_bo(kernel_state->bo[i]);
+		u32 handle;
+
+		ret = drm_gem_handle_create(file_priv, kernel_state->bo[i],
+					    &handle);
+
+		if (ret) {
+			state->bo_count = i;
+			goto err_delete_handle;
+		}
+		bo_state[i].handle = handle;
+		bo_state[i].paddr = vc4_bo->base.paddr;
+		bo_state[i].size = vc4_bo->base.base.size;
+	}
+
+	if (copy_to_user(u64_to_user_ptr(get_state->bo),
+			 bo_state,
+			 state->bo_count * sizeof(*bo_state)))
+		ret = -EFAULT;
+
+err_delete_handle:
+	if (ret) {
+		for (i = 0; i < state->bo_count; i++)
+			drm_gem_handle_delete(file_priv, bo_state[i].handle);
+	}
+
+err_free:
+	vc4_free_hang_state(dev, kernel_state);
+	kfree(bo_state);
+
+	return ret;
+}
+
+static void
+vc4_save_hang_state(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct drm_vc4_get_hang_state *state;
+	struct vc4_hang_state *kernel_state;
+	struct vc4_exec_info *exec[2];
+	struct vc4_bo *bo;
+	unsigned long irqflags;
+	unsigned int i, j, k, unref_list_count;
+
+	kernel_state = kcalloc(1, sizeof(*kernel_state), GFP_KERNEL);
+	if (!kernel_state)
+		return;
+
+	state = &kernel_state->user_state;
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	exec[0] = vc4_first_bin_job(vc4);
+	exec[1] = vc4_first_render_job(vc4);
+	if (!exec[0] && !exec[1]) {
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		return;
+	}
+
+	/* Get the bos from both binner and renderer into hang state. */
+	state->bo_count = 0;
+	for (i = 0; i < 2; i++) {
+		if (!exec[i])
+			continue;
+
+		unref_list_count = 0;
+		list_for_each_entry(bo, &exec[i]->unref_list, unref_head)
+			unref_list_count++;
+		state->bo_count += exec[i]->bo_count + unref_list_count;
+	}
+
+	kernel_state->bo = kcalloc(state->bo_count,
+				   sizeof(*kernel_state->bo), GFP_ATOMIC);
+
+	if (!kernel_state->bo) {
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		return;
+	}
+
+	k = 0;
+	for (i = 0; i < 2; i++) {
+		if (!exec[i])
+			continue;
+
+		for (j = 0; j < exec[i]->bo_count; j++) {
+			bo = to_vc4_bo(&exec[i]->bo[j]->base);
+
+			/* Retain BOs just in case they were marked purgeable.
+			 * This prevents the BO from being purged before
+			 * someone had a chance to dump the hang state.
+			 */
+			WARN_ON(!refcount_read(&bo->usecnt));
+			refcount_inc(&bo->usecnt);
+			drm_gem_object_get(&exec[i]->bo[j]->base);
+			kernel_state->bo[k++] = &exec[i]->bo[j]->base;
+		}
+
+		list_for_each_entry(bo, &exec[i]->unref_list, unref_head) {
+			/* No need to retain BOs coming from the ->unref_list
+			 * because they are naturally unpurgeable.
+			 */
+			drm_gem_object_get(&bo->base.base);
+			kernel_state->bo[k++] = &bo->base.base;
+		}
+	}
+
+	WARN_ON_ONCE(k != state->bo_count);
+
+	if (exec[0])
+		state->start_bin = exec[0]->ct0ca;
+	if (exec[1])
+		state->start_render = exec[1]->ct1ca;
+
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+
+	state->ct0ca = V3D_READ(V3D_CTNCA(0));
+	state->ct0ea = V3D_READ(V3D_CTNEA(0));
+
+	state->ct1ca = V3D_READ(V3D_CTNCA(1));
+	state->ct1ea = V3D_READ(V3D_CTNEA(1));
+
+	state->ct0cs = V3D_READ(V3D_CTNCS(0));
+	state->ct1cs = V3D_READ(V3D_CTNCS(1));
+
+	state->ct0ra0 = V3D_READ(V3D_CT00RA0);
+	state->ct1ra0 = V3D_READ(V3D_CT01RA0);
+
+	state->bpca = V3D_READ(V3D_BPCA);
+	state->bpcs = V3D_READ(V3D_BPCS);
+	state->bpoa = V3D_READ(V3D_BPOA);
+	state->bpos = V3D_READ(V3D_BPOS);
+
+	state->vpmbase = V3D_READ(V3D_VPMBASE);
+
+	state->dbge = V3D_READ(V3D_DBGE);
+	state->fdbgo = V3D_READ(V3D_FDBGO);
+	state->fdbgb = V3D_READ(V3D_FDBGB);
+	state->fdbgr = V3D_READ(V3D_FDBGR);
+	state->fdbgs = V3D_READ(V3D_FDBGS);
+	state->errstat = V3D_READ(V3D_ERRSTAT);
+
+	/* We need to turn purgeable BOs into unpurgeable ones so that
+	 * userspace has a chance to dump the hang state before the kernel
+	 * decides to purge those BOs.
+	 * Note that BO consistency at dump time cannot be guaranteed. For
+	 * example, if the owner of these BOs decides to re-use them or mark
+	 * them purgeable again there's nothing we can do to prevent it.
+	 */
+	for (i = 0; i < kernel_state->user_state.bo_count; i++) {
+		struct vc4_bo *bo = to_vc4_bo(kernel_state->bo[i]);
+
+		if (bo->madv == __VC4_MADV_NOTSUPP)
+			continue;
+
+		mutex_lock(&bo->madv_lock);
+		if (!WARN_ON(bo->madv == __VC4_MADV_PURGED))
+			bo->madv = VC4_MADV_WILLNEED;
+		refcount_dec(&bo->usecnt);
+		mutex_unlock(&bo->madv_lock);
+	}
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	if (vc4->hang_state) {
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		vc4_free_hang_state(dev, kernel_state);
+	} else {
+		vc4->hang_state = kernel_state;
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+	}
+}
+
+static void
+vc4_reset(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	DRM_INFO("Resetting GPU.\n");
+
+	mutex_lock(&vc4->power_lock);
+	if (vc4->power_refcount) {
+		/* Power the device off and back on the by dropping the
+		 * reference on runtime PM.
+		 */
+		pm_runtime_put_sync_suspend(&vc4->v3d->pdev->dev);
+		pm_runtime_get_sync(&vc4->v3d->pdev->dev);
+	}
+	mutex_unlock(&vc4->power_lock);
+
+	vc4_irq_reset(dev);
+
+	/* Rearm the hangcheck -- another job might have been waiting
+	 * for our hung one to get kicked off, and vc4_irq_reset()
+	 * would have started it.
+	 */
+	vc4_queue_hangcheck(dev);
+}
+
+static void
+vc4_reset_work(struct work_struct *work)
+{
+	struct vc4_dev *vc4 =
+		container_of(work, struct vc4_dev, hangcheck.reset_work);
+
+	vc4_save_hang_state(vc4->dev);
+
+	vc4_reset(vc4->dev);
+}
+
+static void
+vc4_hangcheck_elapsed(struct timer_list *t)
+{
+	struct vc4_dev *vc4 = from_timer(vc4, t, hangcheck.timer);
+	struct drm_device *dev = vc4->dev;
+	uint32_t ct0ca, ct1ca;
+	unsigned long irqflags;
+	struct vc4_exec_info *bin_exec, *render_exec;
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+
+	bin_exec = vc4_first_bin_job(vc4);
+	render_exec = vc4_first_render_job(vc4);
+
+	/* If idle, we can stop watching for hangs. */
+	if (!bin_exec && !render_exec) {
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		return;
+	}
+
+	ct0ca = V3D_READ(V3D_CTNCA(0));
+	ct1ca = V3D_READ(V3D_CTNCA(1));
+
+	/* If we've made any progress in execution, rearm the timer
+	 * and wait.
+	 */
+	if ((bin_exec && ct0ca != bin_exec->last_ct0ca) ||
+	    (render_exec && ct1ca != render_exec->last_ct1ca)) {
+		if (bin_exec)
+			bin_exec->last_ct0ca = ct0ca;
+		if (render_exec)
+			render_exec->last_ct1ca = ct1ca;
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		vc4_queue_hangcheck(dev);
+		return;
+	}
+
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+
+	/* We've gone too long with no progress, reset.  This has to
+	 * be done from a work struct, since resetting can sleep and
+	 * this timer hook isn't allowed to.
+	 */
+	schedule_work(&vc4->hangcheck.reset_work);
+}
+
+static void
+submit_cl(struct drm_device *dev, uint32_t thread, uint32_t start, uint32_t end)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	/* Set the current and end address of the control list.
+	 * Writing the end register is what starts the job.
+	 */
+	V3D_WRITE(V3D_CTNCA(thread), start);
+	V3D_WRITE(V3D_CTNEA(thread), end);
+}
+
+int
+vc4_wait_for_seqno(struct drm_device *dev, uint64_t seqno, uint64_t timeout_ns,
+		   bool interruptible)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int ret = 0;
+	unsigned long timeout_expire;
+	DEFINE_WAIT(wait);
+
+	if (vc4->finished_seqno >= seqno)
+		return 0;
+
+	if (timeout_ns == 0)
+		return -ETIME;
+
+	timeout_expire = jiffies + nsecs_to_jiffies(timeout_ns);
+
+	trace_vc4_wait_for_seqno_begin(dev, seqno, timeout_ns);
+	for (;;) {
+		prepare_to_wait(&vc4->job_wait_queue, &wait,
+				interruptible ? TASK_INTERRUPTIBLE :
+				TASK_UNINTERRUPTIBLE);
+
+		if (interruptible && signal_pending(current)) {
+			ret = -ERESTARTSYS;
+			break;
+		}
+
+		if (vc4->finished_seqno >= seqno)
+			break;
+
+		if (timeout_ns != ~0ull) {
+			if (time_after_eq(jiffies, timeout_expire)) {
+				ret = -ETIME;
+				break;
+			}
+			schedule_timeout(timeout_expire - jiffies);
+		} else {
+			schedule();
+		}
+	}
+
+	finish_wait(&vc4->job_wait_queue, &wait);
+	trace_vc4_wait_for_seqno_end(dev, seqno);
+
+	return ret;
+}
+
+static void
+vc4_flush_caches(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	/* Flush the GPU L2 caches.  These caches sit on top of system
+	 * L3 (the 128kb or so shared with the CPU), and are
+	 * non-allocating in the L3.
+	 */
+	V3D_WRITE(V3D_L2CACTL,
+		  V3D_L2CACTL_L2CCLR);
+
+	V3D_WRITE(V3D_SLCACTL,
+		  VC4_SET_FIELD(0xf, V3D_SLCACTL_T1CC) |
+		  VC4_SET_FIELD(0xf, V3D_SLCACTL_T0CC) |
+		  VC4_SET_FIELD(0xf, V3D_SLCACTL_UCC) |
+		  VC4_SET_FIELD(0xf, V3D_SLCACTL_ICC));
+}
+
+static void
+vc4_flush_texture_caches(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	V3D_WRITE(V3D_L2CACTL,
+		  V3D_L2CACTL_L2CCLR);
+
+	V3D_WRITE(V3D_SLCACTL,
+		  VC4_SET_FIELD(0xf, V3D_SLCACTL_T1CC) |
+		  VC4_SET_FIELD(0xf, V3D_SLCACTL_T0CC));
+}
+
+/* Sets the registers for the next job to be actually be executed in
+ * the hardware.
+ *
+ * The job_lock should be held during this.
+ */
+void
+vc4_submit_next_bin_job(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_exec_info *exec;
+
+again:
+	exec = vc4_first_bin_job(vc4);
+	if (!exec)
+		return;
+
+	vc4_flush_caches(dev);
+
+	/* Only start the perfmon if it was not already started by a previous
+	 * job.
+	 */
+	if (exec->perfmon && vc4->active_perfmon != exec->perfmon)
+		vc4_perfmon_start(vc4, exec->perfmon);
+
+	/* Either put the job in the binner if it uses the binner, or
+	 * immediately move it to the to-be-rendered queue.
+	 */
+	if (exec->ct0ca != exec->ct0ea) {
+		submit_cl(dev, 0, exec->ct0ca, exec->ct0ea);
+	} else {
+		struct vc4_exec_info *next;
+
+		vc4_move_job_to_render(dev, exec);
+		next = vc4_first_bin_job(vc4);
+
+		/* We can't start the next bin job if the previous job had a
+		 * different perfmon instance attached to it. The same goes
+		 * if one of them had a perfmon attached to it and the other
+		 * one doesn't.
+		 */
+		if (next && next->perfmon == exec->perfmon)
+			goto again;
+	}
+}
+
+void
+vc4_submit_next_render_job(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_exec_info *exec = vc4_first_render_job(vc4);
+
+	if (!exec)
+		return;
+
+	/* A previous RCL may have written to one of our textures, and
+	 * our full cache flush at bin time may have occurred before
+	 * that RCL completed.  Flush the texture cache now, but not
+	 * the instructions or uniforms (since we don't write those
+	 * from an RCL).
+	 */
+	vc4_flush_texture_caches(dev);
+
+	submit_cl(dev, 1, exec->ct1ca, exec->ct1ea);
+}
+
+void
+vc4_move_job_to_render(struct drm_device *dev, struct vc4_exec_info *exec)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	bool was_empty = list_empty(&vc4->render_job_list);
+
+	list_move_tail(&exec->head, &vc4->render_job_list);
+	if (was_empty)
+		vc4_submit_next_render_job(dev);
+}
+
+static void
+vc4_update_bo_seqnos(struct vc4_exec_info *exec, uint64_t seqno)
+{
+	struct vc4_bo *bo;
+	unsigned i;
+
+	for (i = 0; i < exec->bo_count; i++) {
+		bo = to_vc4_bo(&exec->bo[i]->base);
+		bo->seqno = seqno;
+
+		reservation_object_add_shared_fence(bo->resv, exec->fence);
+	}
+
+	list_for_each_entry(bo, &exec->unref_list, unref_head) {
+		bo->seqno = seqno;
+	}
+
+	for (i = 0; i < exec->rcl_write_bo_count; i++) {
+		bo = to_vc4_bo(&exec->rcl_write_bo[i]->base);
+		bo->write_seqno = seqno;
+
+		reservation_object_add_excl_fence(bo->resv, exec->fence);
+	}
+}
+
+static void
+vc4_unlock_bo_reservations(struct drm_device *dev,
+			   struct vc4_exec_info *exec,
+			   struct ww_acquire_ctx *acquire_ctx)
+{
+	int i;
+
+	for (i = 0; i < exec->bo_count; i++) {
+		struct vc4_bo *bo = to_vc4_bo(&exec->bo[i]->base);
+
+		ww_mutex_unlock(&bo->resv->lock);
+	}
+
+	ww_acquire_fini(acquire_ctx);
+}
+
+/* Takes the reservation lock on all the BOs being referenced, so that
+ * at queue submit time we can update the reservations.
+ *
+ * We don't lock the RCL the tile alloc/state BOs, or overflow memory
+ * (all of which are on exec->unref_list).  They're entirely private
+ * to vc4, so we don't attach dma-buf fences to them.
+ */
+static int
+vc4_lock_bo_reservations(struct drm_device *dev,
+			 struct vc4_exec_info *exec,
+			 struct ww_acquire_ctx *acquire_ctx)
+{
+	int contended_lock = -1;
+	int i, ret;
+	struct vc4_bo *bo;
+
+	ww_acquire_init(acquire_ctx, &reservation_ww_class);
+
+retry:
+	if (contended_lock != -1) {
+		bo = to_vc4_bo(&exec->bo[contended_lock]->base);
+		ret = ww_mutex_lock_slow_interruptible(&bo->resv->lock,
+						       acquire_ctx);
+		if (ret) {
+			ww_acquire_done(acquire_ctx);
+			return ret;
+		}
+	}
+
+	for (i = 0; i < exec->bo_count; i++) {
+		if (i == contended_lock)
+			continue;
+
+		bo = to_vc4_bo(&exec->bo[i]->base);
+
+		ret = ww_mutex_lock_interruptible(&bo->resv->lock, acquire_ctx);
+		if (ret) {
+			int j;
+
+			for (j = 0; j < i; j++) {
+				bo = to_vc4_bo(&exec->bo[j]->base);
+				ww_mutex_unlock(&bo->resv->lock);
+			}
+
+			if (contended_lock != -1 && contended_lock >= i) {
+				bo = to_vc4_bo(&exec->bo[contended_lock]->base);
+
+				ww_mutex_unlock(&bo->resv->lock);
+			}
+
+			if (ret == -EDEADLK) {
+				contended_lock = i;
+				goto retry;
+			}
+
+			ww_acquire_done(acquire_ctx);
+			return ret;
+		}
+	}
+
+	ww_acquire_done(acquire_ctx);
+
+	/* Reserve space for our shared (read-only) fence references,
+	 * before we commit the CL to the hardware.
+	 */
+	for (i = 0; i < exec->bo_count; i++) {
+		bo = to_vc4_bo(&exec->bo[i]->base);
+
+		ret = reservation_object_reserve_shared(bo->resv);
+		if (ret) {
+			vc4_unlock_bo_reservations(dev, exec, acquire_ctx);
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
+/* Queues a struct vc4_exec_info for execution.  If no job is
+ * currently executing, then submits it.
+ *
+ * Unlike most GPUs, our hardware only handles one command list at a
+ * time.  To queue multiple jobs at once, we'd need to edit the
+ * previous command list to have a jump to the new one at the end, and
+ * then bump the end address.  That's a change for a later date,
+ * though.
+ */
+static int
+vc4_queue_submit(struct drm_device *dev, struct vc4_exec_info *exec,
+		 struct ww_acquire_ctx *acquire_ctx,
+		 struct drm_syncobj *out_sync)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_exec_info *renderjob;
+	uint64_t seqno;
+	unsigned long irqflags;
+	struct vc4_fence *fence;
+
+	fence = kzalloc(sizeof(*fence), GFP_KERNEL);
+	if (!fence)
+		return -ENOMEM;
+	fence->dev = dev;
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+
+	seqno = ++vc4->emit_seqno;
+	exec->seqno = seqno;
+
+	dma_fence_init(&fence->base, &vc4_fence_ops, &vc4->job_lock,
+		       vc4->dma_fence_context, exec->seqno);
+	fence->seqno = exec->seqno;
+	exec->fence = &fence->base;
+
+	if (out_sync)
+		drm_syncobj_replace_fence(out_sync, exec->fence);
+
+	vc4_update_bo_seqnos(exec, seqno);
+
+	vc4_unlock_bo_reservations(dev, exec, acquire_ctx);
+
+	list_add_tail(&exec->head, &vc4->bin_job_list);
+
+	/* If no bin job was executing and if the render job (if any) has the
+	 * same perfmon as our job attached to it (or if both jobs don't have
+	 * perfmon activated), then kick ours off.  Otherwise, it'll get
+	 * started when the previous job's flush/render done interrupt occurs.
+	 */
+	renderjob = vc4_first_render_job(vc4);
+	if (vc4_first_bin_job(vc4) == exec &&
+	    (!renderjob || renderjob->perfmon == exec->perfmon)) {
+		vc4_submit_next_bin_job(dev);
+		vc4_queue_hangcheck(dev);
+	}
+
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+
+	return 0;
+}
+
+/**
+ * vc4_cl_lookup_bos() - Sets up exec->bo[] with the GEM objects
+ * referenced by the job.
+ * @dev: DRM device
+ * @file_priv: DRM file for this fd
+ * @exec: V3D job being set up
+ *
+ * The command validator needs to reference BOs by their index within
+ * the submitted job's BO list.  This does the validation of the job's
+ * BO list and reference counting for the lifetime of the job.
+ */
+static int
+vc4_cl_lookup_bos(struct drm_device *dev,
+		  struct drm_file *file_priv,
+		  struct vc4_exec_info *exec)
+{
+	struct drm_vc4_submit_cl *args = exec->args;
+	uint32_t *handles;
+	int ret = 0;
+	int i;
+
+	exec->bo_count = args->bo_handle_count;
+
+	if (!exec->bo_count) {
+		/* See comment on bo_index for why we have to check
+		 * this.
+		 */
+		DRM_DEBUG("Rendering requires BOs to validate\n");
+		return -EINVAL;
+	}
+
+	exec->bo = kvmalloc_array(exec->bo_count,
+				    sizeof(struct drm_gem_cma_object *),
+				    GFP_KERNEL | __GFP_ZERO);
+	if (!exec->bo) {
+		DRM_ERROR("Failed to allocate validated BO pointers\n");
+		return -ENOMEM;
+	}
+
+	handles = kvmalloc_array(exec->bo_count, sizeof(uint32_t), GFP_KERNEL);
+	if (!handles) {
+		ret = -ENOMEM;
+		DRM_ERROR("Failed to allocate incoming GEM handles\n");
+		goto fail;
+	}
+
+	if (copy_from_user(handles, u64_to_user_ptr(args->bo_handles),
+			   exec->bo_count * sizeof(uint32_t))) {
+		ret = -EFAULT;
+		DRM_ERROR("Failed to copy in GEM handles\n");
+		goto fail;
+	}
+
+	spin_lock(&file_priv->table_lock);
+	for (i = 0; i < exec->bo_count; i++) {
+		struct drm_gem_object *bo = idr_find(&file_priv->object_idr,
+						     handles[i]);
+		if (!bo) {
+			DRM_DEBUG("Failed to look up GEM BO %d: %d\n",
+				  i, handles[i]);
+			ret = -EINVAL;
+			break;
+		}
+
+		drm_gem_object_get(bo);
+		exec->bo[i] = (struct drm_gem_cma_object *)bo;
+	}
+	spin_unlock(&file_priv->table_lock);
+
+	if (ret)
+		goto fail_put_bo;
+
+	for (i = 0; i < exec->bo_count; i++) {
+		ret = vc4_bo_inc_usecnt(to_vc4_bo(&exec->bo[i]->base));
+		if (ret)
+			goto fail_dec_usecnt;
+	}
+
+	kvfree(handles);
+	return 0;
+
+fail_dec_usecnt:
+	/* Decrease usecnt on acquired objects.
+	 * We cannot rely on  vc4_complete_exec() to release resources here,
+	 * because vc4_complete_exec() has no information about which BO has
+	 * had its ->usecnt incremented.
+	 * To make things easier we just free everything explicitly and set
+	 * exec->bo to NULL so that vc4_complete_exec() skips the 'BO release'
+	 * step.
+	 */
+	for (i-- ; i >= 0; i--)
+		vc4_bo_dec_usecnt(to_vc4_bo(&exec->bo[i]->base));
+
+fail_put_bo:
+	/* Release any reference to acquired objects. */
+	for (i = 0; i < exec->bo_count && exec->bo[i]; i++)
+		drm_gem_object_put_unlocked(&exec->bo[i]->base);
+
+fail:
+	kvfree(handles);
+	kvfree(exec->bo);
+	exec->bo = NULL;
+	return ret;
+}
+
+static int
+vc4_get_bcl(struct drm_device *dev, struct vc4_exec_info *exec)
+{
+	struct drm_vc4_submit_cl *args = exec->args;
+	void *temp = NULL;
+	void *bin;
+	int ret = 0;
+	uint32_t bin_offset = 0;
+	uint32_t shader_rec_offset = roundup(bin_offset + args->bin_cl_size,
+					     16);
+	uint32_t uniforms_offset = shader_rec_offset + args->shader_rec_size;
+	uint32_t exec_size = uniforms_offset + args->uniforms_size;
+	uint32_t temp_size = exec_size + (sizeof(struct vc4_shader_state) *
+					  args->shader_rec_count);
+	struct vc4_bo *bo;
+
+	if (shader_rec_offset < args->bin_cl_size ||
+	    uniforms_offset < shader_rec_offset ||
+	    exec_size < uniforms_offset ||
+	    args->shader_rec_count >= (UINT_MAX /
+					  sizeof(struct vc4_shader_state)) ||
+	    temp_size < exec_size) {
+		DRM_DEBUG("overflow in exec arguments\n");
+		ret = -EINVAL;
+		goto fail;
+	}
+
+	/* Allocate space where we'll store the copied in user command lists
+	 * and shader records.
+	 *
+	 * We don't just copy directly into the BOs because we need to
+	 * read the contents back for validation, and I think the
+	 * bo->vaddr is uncached access.
+	 */
+	temp = kvmalloc_array(temp_size, 1, GFP_KERNEL);
+	if (!temp) {
+		DRM_ERROR("Failed to allocate storage for copying "
+			  "in bin/render CLs.\n");
+		ret = -ENOMEM;
+		goto fail;
+	}
+	bin = temp + bin_offset;
+	exec->shader_rec_u = temp + shader_rec_offset;
+	exec->uniforms_u = temp + uniforms_offset;
+	exec->shader_state = temp + exec_size;
+	exec->shader_state_size = args->shader_rec_count;
+
+	if (copy_from_user(bin,
+			   u64_to_user_ptr(args->bin_cl),
+			   args->bin_cl_size)) {
+		ret = -EFAULT;
+		goto fail;
+	}
+
+	if (copy_from_user(exec->shader_rec_u,
+			   u64_to_user_ptr(args->shader_rec),
+			   args->shader_rec_size)) {
+		ret = -EFAULT;
+		goto fail;
+	}
+
+	if (copy_from_user(exec->uniforms_u,
+			   u64_to_user_ptr(args->uniforms),
+			   args->uniforms_size)) {
+		ret = -EFAULT;
+		goto fail;
+	}
+
+	bo = vc4_bo_create(dev, exec_size, true, VC4_BO_TYPE_BCL);
+	if (IS_ERR(bo)) {
+		DRM_ERROR("Couldn't allocate BO for binning\n");
+		ret = PTR_ERR(bo);
+		goto fail;
+	}
+	exec->exec_bo = &bo->base;
+
+	list_add_tail(&to_vc4_bo(&exec->exec_bo->base)->unref_head,
+		      &exec->unref_list);
+
+	exec->ct0ca = exec->exec_bo->paddr + bin_offset;
+
+	exec->bin_u = bin;
+
+	exec->shader_rec_v = exec->exec_bo->vaddr + shader_rec_offset;
+	exec->shader_rec_p = exec->exec_bo->paddr + shader_rec_offset;
+	exec->shader_rec_size = args->shader_rec_size;
+
+	exec->uniforms_v = exec->exec_bo->vaddr + uniforms_offset;
+	exec->uniforms_p = exec->exec_bo->paddr + uniforms_offset;
+	exec->uniforms_size = args->uniforms_size;
+
+	ret = vc4_validate_bin_cl(dev,
+				  exec->exec_bo->vaddr + bin_offset,
+				  bin,
+				  exec);
+	if (ret)
+		goto fail;
+
+	ret = vc4_validate_shader_recs(dev, exec);
+	if (ret)
+		goto fail;
+
+	/* Block waiting on any previous rendering into the CS's VBO,
+	 * IB, or textures, so that pixels are actually written by the
+	 * time we try to read them.
+	 */
+	ret = vc4_wait_for_seqno(dev, exec->bin_dep_seqno, ~0ull, true);
+
+fail:
+	kvfree(temp);
+	return ret;
+}
+
+static void
+vc4_complete_exec(struct drm_device *dev, struct vc4_exec_info *exec)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	unsigned long irqflags;
+	unsigned i;
+
+	/* If we got force-completed because of GPU reset rather than
+	 * through our IRQ handler, signal the fence now.
+	 */
+	if (exec->fence) {
+		dma_fence_signal(exec->fence);
+		dma_fence_put(exec->fence);
+	}
+
+	if (exec->bo) {
+		for (i = 0; i < exec->bo_count; i++) {
+			struct vc4_bo *bo = to_vc4_bo(&exec->bo[i]->base);
+
+			vc4_bo_dec_usecnt(bo);
+			drm_gem_object_put_unlocked(&exec->bo[i]->base);
+		}
+		kvfree(exec->bo);
+	}
+
+	while (!list_empty(&exec->unref_list)) {
+		struct vc4_bo *bo = list_first_entry(&exec->unref_list,
+						     struct vc4_bo, unref_head);
+		list_del(&bo->unref_head);
+		drm_gem_object_put_unlocked(&bo->base.base);
+	}
+
+	/* Free up the allocation of any bin slots we used. */
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	vc4->bin_alloc_used &= ~exec->bin_slots;
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+
+	/* Release the reference we had on the perf monitor. */
+	vc4_perfmon_put(exec->perfmon);
+
+	mutex_lock(&vc4->power_lock);
+	if (--vc4->power_refcount == 0) {
+		pm_runtime_mark_last_busy(&vc4->v3d->pdev->dev);
+		pm_runtime_put_autosuspend(&vc4->v3d->pdev->dev);
+	}
+	mutex_unlock(&vc4->power_lock);
+
+	kfree(exec);
+}
+
+void
+vc4_job_handle_completed(struct vc4_dev *vc4)
+{
+	unsigned long irqflags;
+	struct vc4_seqno_cb *cb, *cb_temp;
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	while (!list_empty(&vc4->job_done_list)) {
+		struct vc4_exec_info *exec =
+			list_first_entry(&vc4->job_done_list,
+					 struct vc4_exec_info, head);
+		list_del(&exec->head);
+
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		vc4_complete_exec(vc4->dev, exec);
+		spin_lock_irqsave(&vc4->job_lock, irqflags);
+	}
+
+	list_for_each_entry_safe(cb, cb_temp, &vc4->seqno_cb_list, work.entry) {
+		if (cb->seqno <= vc4->finished_seqno) {
+			list_del_init(&cb->work.entry);
+			schedule_work(&cb->work);
+		}
+	}
+
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+}
+
+static void vc4_seqno_cb_work(struct work_struct *work)
+{
+	struct vc4_seqno_cb *cb = container_of(work, struct vc4_seqno_cb, work);
+
+	cb->func(cb);
+}
+
+int vc4_queue_seqno_cb(struct drm_device *dev,
+		       struct vc4_seqno_cb *cb, uint64_t seqno,
+		       void (*func)(struct vc4_seqno_cb *cb))
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int ret = 0;
+	unsigned long irqflags;
+
+	cb->func = func;
+	INIT_WORK(&cb->work, vc4_seqno_cb_work);
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	if (seqno > vc4->finished_seqno) {
+		cb->seqno = seqno;
+		list_add_tail(&cb->work.entry, &vc4->seqno_cb_list);
+	} else {
+		schedule_work(&cb->work);
+	}
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+
+	return ret;
+}
+
+/* Scheduled when any job has been completed, this walks the list of
+ * jobs that had completed and unrefs their BOs and frees their exec
+ * structs.
+ */
+static void
+vc4_job_done_work(struct work_struct *work)
+{
+	struct vc4_dev *vc4 =
+		container_of(work, struct vc4_dev, job_done_work);
+
+	vc4_job_handle_completed(vc4);
+}
+
+static int
+vc4_wait_for_seqno_ioctl_helper(struct drm_device *dev,
+				uint64_t seqno,
+				uint64_t *timeout_ns)
+{
+	unsigned long start = jiffies;
+	int ret = vc4_wait_for_seqno(dev, seqno, *timeout_ns, true);
+
+	if ((ret == -EINTR || ret == -ERESTARTSYS) && *timeout_ns != ~0ull) {
+		uint64_t delta = jiffies_to_nsecs(jiffies - start);
+
+		if (*timeout_ns >= delta)
+			*timeout_ns -= delta;
+	}
+
+	return ret;
+}
+
+int
+vc4_wait_seqno_ioctl(struct drm_device *dev, void *data,
+		     struct drm_file *file_priv)
+{
+	struct drm_vc4_wait_seqno *args = data;
+
+	return vc4_wait_for_seqno_ioctl_helper(dev, args->seqno,
+					       &args->timeout_ns);
+}
+
+int
+vc4_wait_bo_ioctl(struct drm_device *dev, void *data,
+		  struct drm_file *file_priv)
+{
+	int ret;
+	struct drm_vc4_wait_bo *args = data;
+	struct drm_gem_object *gem_obj;
+	struct vc4_bo *bo;
+
+	if (args->pad != 0)
+		return -EINVAL;
+
+	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
+	if (!gem_obj) {
+		DRM_DEBUG("Failed to look up GEM BO %d\n", args->handle);
+		return -EINVAL;
+	}
+	bo = to_vc4_bo(gem_obj);
+
+	ret = vc4_wait_for_seqno_ioctl_helper(dev, bo->seqno,
+					      &args->timeout_ns);
+
+	drm_gem_object_put_unlocked(gem_obj);
+	return ret;
+}
+
+/**
+ * vc4_submit_cl_ioctl() - Submits a job (frame) to the VC4.
+ * @dev: DRM device
+ * @data: ioctl argument
+ * @file_priv: DRM file for this fd
+ *
+ * This is the main entrypoint for userspace to submit a 3D frame to
+ * the GPU.  Userspace provides the binner command list (if
+ * applicable), and the kernel sets up the render command list to draw
+ * to the framebuffer described in the ioctl, using the command lists
+ * that the 3D engine's binner will produce.
+ */
+int
+vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
+		    struct drm_file *file_priv)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_file *vc4file = file_priv->driver_priv;
+	struct drm_vc4_submit_cl *args = data;
+	struct drm_syncobj *out_sync = NULL;
+	struct vc4_exec_info *exec;
+	struct ww_acquire_ctx acquire_ctx;
+	struct dma_fence *in_fence;
+	int ret = 0;
+
+	if ((args->flags & ~(VC4_SUBMIT_CL_USE_CLEAR_COLOR |
+			     VC4_SUBMIT_CL_FIXED_RCL_ORDER |
+			     VC4_SUBMIT_CL_RCL_ORDER_INCREASING_X |
+			     VC4_SUBMIT_CL_RCL_ORDER_INCREASING_Y)) != 0) {
+		DRM_DEBUG("Unknown flags: 0x%02x\n", args->flags);
+		return -EINVAL;
+	}
+
+	if (args->pad2 != 0) {
+		DRM_DEBUG("Invalid pad: 0x%08x\n", args->pad2);
+		return -EINVAL;
+	}
+
+	exec = kcalloc(1, sizeof(*exec), GFP_KERNEL);
+	if (!exec) {
+		DRM_ERROR("malloc failure on exec struct\n");
+		return -ENOMEM;
+	}
+
+	mutex_lock(&vc4->power_lock);
+	if (vc4->power_refcount++ == 0) {
+		ret = pm_runtime_get_sync(&vc4->v3d->pdev->dev);
+		if (ret < 0) {
+			mutex_unlock(&vc4->power_lock);
+			vc4->power_refcount--;
+			kfree(exec);
+			return ret;
+		}
+	}
+	mutex_unlock(&vc4->power_lock);
+
+	exec->args = args;
+	INIT_LIST_HEAD(&exec->unref_list);
+
+	ret = vc4_cl_lookup_bos(dev, file_priv, exec);
+	if (ret)
+		goto fail;
+
+	if (args->perfmonid) {
+		exec->perfmon = vc4_perfmon_find(vc4file,
+						 args->perfmonid);
+		if (!exec->perfmon) {
+			ret = -ENOENT;
+			goto fail;
+		}
+	}
+
+	if (args->in_sync) {
+		ret = drm_syncobj_find_fence(file_priv, args->in_sync,
+					     &in_fence);
+		if (ret)
+			goto fail;
+
+		/* When the fence (or fence array) is exclusively from our
+		 * context we can skip the wait since jobs are executed in
+		 * order of their submission through this ioctl and this can
+		 * only have fences from a prior job.
+		 */
+		if (!dma_fence_match_context(in_fence,
+					     vc4->dma_fence_context)) {
+			ret = dma_fence_wait(in_fence, true);
+			if (ret) {
+				dma_fence_put(in_fence);
+				goto fail;
+			}
+		}
+
+		dma_fence_put(in_fence);
+	}
+
+	if (exec->args->bin_cl_size != 0) {
+		ret = vc4_get_bcl(dev, exec);
+		if (ret)
+			goto fail;
+	} else {
+		exec->ct0ca = 0;
+		exec->ct0ea = 0;
+	}
+
+	ret = vc4_get_rcl(dev, exec);
+	if (ret)
+		goto fail;
+
+	ret = vc4_lock_bo_reservations(dev, exec, &acquire_ctx);
+	if (ret)
+		goto fail;
+
+	if (args->out_sync) {
+		out_sync = drm_syncobj_find(file_priv, args->out_sync);
+		if (!out_sync) {
+			ret = -EINVAL;
+			goto fail;
+		}
+
+		/* We replace the fence in out_sync in vc4_queue_submit since
+		 * the render job could execute immediately after that call.
+		 * If it finishes before our ioctl processing resumes the
+		 * render job fence could already have been freed.
+		 */
+	}
+
+	/* Clear this out of the struct we'll be putting in the queue,
+	 * since it's part of our stack.
+	 */
+	exec->args = NULL;
+
+	ret = vc4_queue_submit(dev, exec, &acquire_ctx, out_sync);
+
+	/* The syncobj isn't part of the exec data and we need to free our
+	 * reference even if job submission failed.
+	 */
+	if (out_sync)
+		drm_syncobj_put(out_sync);
+
+	if (ret)
+		goto fail;
+
+	/* Return the seqno for our job. */
+	args->seqno = vc4->emit_seqno;
+
+	return 0;
+
+fail:
+	vc4_complete_exec(vc4->dev, exec);
+
+	return ret;
+}
+
+void
+vc4_gem_init(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	vc4->dma_fence_context = dma_fence_context_alloc(1);
+
+	INIT_LIST_HEAD(&vc4->bin_job_list);
+	INIT_LIST_HEAD(&vc4->render_job_list);
+	INIT_LIST_HEAD(&vc4->job_done_list);
+	INIT_LIST_HEAD(&vc4->seqno_cb_list);
+	spin_lock_init(&vc4->job_lock);
+
+	INIT_WORK(&vc4->hangcheck.reset_work, vc4_reset_work);
+	timer_setup(&vc4->hangcheck.timer, vc4_hangcheck_elapsed, 0);
+
+	INIT_WORK(&vc4->job_done_work, vc4_job_done_work);
+
+	mutex_init(&vc4->power_lock);
+
+	INIT_LIST_HEAD(&vc4->purgeable.list);
+	mutex_init(&vc4->purgeable.lock);
+}
+
+void
+vc4_gem_destroy(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	/* Waiting for exec to finish would need to be done before
+	 * unregistering V3D.
+	 */
+	WARN_ON(vc4->emit_seqno != vc4->finished_seqno);
+
+	/* V3D should already have disabled its interrupt and cleared
+	 * the overflow allocation registers.  Now free the object.
+	 */
+	if (vc4->bin_bo) {
+		drm_gem_object_put_unlocked(&vc4->bin_bo->base.base);
+		vc4->bin_bo = NULL;
+	}
+
+	if (vc4->hang_state)
+		vc4_free_hang_state(dev, vc4->hang_state);
+}
+
+int vc4_gem_madvise_ioctl(struct drm_device *dev, void *data,
+			  struct drm_file *file_priv)
+{
+	struct drm_vc4_gem_madvise *args = data;
+	struct drm_gem_object *gem_obj;
+	struct vc4_bo *bo;
+	int ret;
+
+	switch (args->madv) {
+	case VC4_MADV_DONTNEED:
+	case VC4_MADV_WILLNEED:
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	if (args->pad != 0)
+		return -EINVAL;
+
+	gem_obj = drm_gem_object_lookup(file_priv, args->handle);
+	if (!gem_obj) {
+		DRM_DEBUG("Failed to look up GEM BO %d\n", args->handle);
+		return -ENOENT;
+	}
+
+	bo = to_vc4_bo(gem_obj);
+
+	/* Only BOs exposed to userspace can be purged. */
+	if (bo->madv == __VC4_MADV_NOTSUPP) {
+		DRM_DEBUG("madvise not supported on this BO\n");
+		ret = -EINVAL;
+		goto out_put_gem;
+	}
+
+	/* Not sure it's safe to purge imported BOs. Let's just assume it's
+	 * not until proven otherwise.
+	 */
+	if (gem_obj->import_attach) {
+		DRM_DEBUG("madvise not supported on imported BOs\n");
+		ret = -EINVAL;
+		goto out_put_gem;
+	}
+
+	mutex_lock(&bo->madv_lock);
+
+	if (args->madv == VC4_MADV_DONTNEED && bo->madv == VC4_MADV_WILLNEED &&
+	    !refcount_read(&bo->usecnt)) {
+		/* If the BO is about to be marked as purgeable, is not used
+		 * and is not already purgeable or purged, add it to the
+		 * purgeable list.
+		 */
+		vc4_bo_add_to_purgeable_pool(bo);
+	} else if (args->madv == VC4_MADV_WILLNEED &&
+		   bo->madv == VC4_MADV_DONTNEED &&
+		   !refcount_read(&bo->usecnt)) {
+		/* The BO has not been purged yet, just remove it from
+		 * the purgeable list.
+		 */
+		vc4_bo_remove_from_purgeable_pool(bo);
+	}
+
+	/* Save the purged state. */
+	args->retained = bo->madv != __VC4_MADV_PURGED;
+
+	/* Update internal madv state only if the bo was not purged. */
+	if (bo->madv != __VC4_MADV_PURGED)
+		bo->madv = args->madv;
+
+	mutex_unlock(&bo->madv_lock);
+
+	ret = 0;
+
+out_put_gem:
+	drm_gem_object_put_unlocked(gem_obj);
+
+	return ret;
+}
diff --git a/drivers/gpu/drm/vc4/vc4_hdmi.c b/drivers/gpu/drm/vc4/vc4_hdmi.c
new file mode 100644
index 000000000..116166266
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_hdmi.c
@@ -0,0 +1,1520 @@
+/*
+ * Copyright (C) 2015 Broadcom
+ * Copyright (c) 2014 The Linux Foundation. All rights reserved.
+ * Copyright (C) 2013 Red Hat
+ * Author: Rob Clark <robdclark@gmail.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published by
+ * the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+/**
+ * DOC: VC4 Falcon HDMI module
+ *
+ * The HDMI core has a state machine and a PHY.  On BCM2835, most of
+ * the unit operates off of the HSM clock from CPRMAN.  It also
+ * internally uses the PLLH_PIX clock for the PHY.
+ *
+ * HDMI infoframes are kept within a small packet ram, where each
+ * packet can be individually enabled for including in a frame.
+ *
+ * HDMI audio is implemented entirely within the HDMI IP block.  A
+ * register in the HDMI encoder takes SPDIF frames from the DMA engine
+ * and transfers them over an internal MAI (multi-channel audio
+ * interconnect) bus to the encoder side for insertion into the video
+ * blank regions.
+ *
+ * The driver's HDMI encoder does not yet support power management.
+ * The HDMI encoder's power domain and the HSM/pixel clocks are kept
+ * continuously running, and only the HDMI logic and packet ram are
+ * powered off/on at disable/enable time.
+ *
+ * The driver does not yet support CEC control, though the HDMI
+ * encoder block has CEC support.
+ */
+
+#include <drm/drm_atomic_helper.h>
+#include <drm/drm_crtc_helper.h>
+#include <drm/drm_edid.h>
+#include <linux/clk.h>
+#include <linux/component.h>
+#include <linux/i2c.h>
+#include <linux/of_address.h>
+#include <linux/of_gpio.h>
+#include <linux/of_platform.h>
+#include <linux/pm_runtime.h>
+#include <linux/rational.h>
+#include <sound/dmaengine_pcm.h>
+#include <sound/pcm_drm_eld.h>
+#include <sound/pcm_params.h>
+#include <sound/soc.h>
+#include "media/cec.h"
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+#define HSM_CLOCK_FREQ 163682864
+#define CEC_CLOCK_FREQ 40000
+#define CEC_CLOCK_DIV  (HSM_CLOCK_FREQ / CEC_CLOCK_FREQ)
+
+/* HDMI audio information */
+struct vc4_hdmi_audio {
+	struct snd_soc_card card;
+	struct snd_soc_dai_link link;
+	int samplerate;
+	int channels;
+	struct snd_dmaengine_dai_dma_data dma_data;
+	struct snd_pcm_substream *substream;
+};
+
+/* General HDMI hardware state. */
+struct vc4_hdmi {
+	struct platform_device *pdev;
+
+	struct drm_encoder *encoder;
+	struct drm_connector *connector;
+
+	struct vc4_hdmi_audio audio;
+
+	struct i2c_adapter *ddc;
+	void __iomem *hdmicore_regs;
+	void __iomem *hd_regs;
+	int hpd_gpio;
+	bool hpd_active_low;
+
+	struct cec_adapter *cec_adap;
+	struct cec_msg cec_rx_msg;
+	bool cec_tx_ok;
+	bool cec_irq_was_rx;
+
+	struct clk *pixel_clock;
+	struct clk *hsm_clock;
+};
+
+#define HDMI_READ(offset) readl(vc4->hdmi->hdmicore_regs + offset)
+#define HDMI_WRITE(offset, val) writel(val, vc4->hdmi->hdmicore_regs + offset)
+#define HD_READ(offset) readl(vc4->hdmi->hd_regs + offset)
+#define HD_WRITE(offset, val) writel(val, vc4->hdmi->hd_regs + offset)
+
+/* VC4 HDMI encoder KMS struct */
+struct vc4_hdmi_encoder {
+	struct vc4_encoder base;
+	bool hdmi_monitor;
+	bool limited_rgb_range;
+	bool rgb_range_selectable;
+};
+
+static inline struct vc4_hdmi_encoder *
+to_vc4_hdmi_encoder(struct drm_encoder *encoder)
+{
+	return container_of(encoder, struct vc4_hdmi_encoder, base.base);
+}
+
+/* VC4 HDMI connector KMS struct */
+struct vc4_hdmi_connector {
+	struct drm_connector base;
+
+	/* Since the connector is attached to just the one encoder,
+	 * this is the reference to it so we can do the best_encoder()
+	 * hook.
+	 */
+	struct drm_encoder *encoder;
+};
+
+static inline struct vc4_hdmi_connector *
+to_vc4_hdmi_connector(struct drm_connector *connector)
+{
+	return container_of(connector, struct vc4_hdmi_connector, base);
+}
+
+#define HDMI_REG(reg) { reg, #reg }
+static const struct {
+	u32 reg;
+	const char *name;
+} hdmi_regs[] = {
+	HDMI_REG(VC4_HDMI_CORE_REV),
+	HDMI_REG(VC4_HDMI_SW_RESET_CONTROL),
+	HDMI_REG(VC4_HDMI_HOTPLUG_INT),
+	HDMI_REG(VC4_HDMI_HOTPLUG),
+	HDMI_REG(VC4_HDMI_MAI_CHANNEL_MAP),
+	HDMI_REG(VC4_HDMI_MAI_CONFIG),
+	HDMI_REG(VC4_HDMI_MAI_FORMAT),
+	HDMI_REG(VC4_HDMI_AUDIO_PACKET_CONFIG),
+	HDMI_REG(VC4_HDMI_RAM_PACKET_CONFIG),
+	HDMI_REG(VC4_HDMI_HORZA),
+	HDMI_REG(VC4_HDMI_HORZB),
+	HDMI_REG(VC4_HDMI_FIFO_CTL),
+	HDMI_REG(VC4_HDMI_SCHEDULER_CONTROL),
+	HDMI_REG(VC4_HDMI_VERTA0),
+	HDMI_REG(VC4_HDMI_VERTA1),
+	HDMI_REG(VC4_HDMI_VERTB0),
+	HDMI_REG(VC4_HDMI_VERTB1),
+	HDMI_REG(VC4_HDMI_TX_PHY_RESET_CTL),
+	HDMI_REG(VC4_HDMI_TX_PHY_CTL0),
+
+	HDMI_REG(VC4_HDMI_CEC_CNTRL_1),
+	HDMI_REG(VC4_HDMI_CEC_CNTRL_2),
+	HDMI_REG(VC4_HDMI_CEC_CNTRL_3),
+	HDMI_REG(VC4_HDMI_CEC_CNTRL_4),
+	HDMI_REG(VC4_HDMI_CEC_CNTRL_5),
+	HDMI_REG(VC4_HDMI_CPU_STATUS),
+	HDMI_REG(VC4_HDMI_CPU_MASK_STATUS),
+
+	HDMI_REG(VC4_HDMI_CEC_RX_DATA_1),
+	HDMI_REG(VC4_HDMI_CEC_RX_DATA_2),
+	HDMI_REG(VC4_HDMI_CEC_RX_DATA_3),
+	HDMI_REG(VC4_HDMI_CEC_RX_DATA_4),
+	HDMI_REG(VC4_HDMI_CEC_TX_DATA_1),
+	HDMI_REG(VC4_HDMI_CEC_TX_DATA_2),
+	HDMI_REG(VC4_HDMI_CEC_TX_DATA_3),
+	HDMI_REG(VC4_HDMI_CEC_TX_DATA_4),
+};
+
+static const struct {
+	u32 reg;
+	const char *name;
+} hd_regs[] = {
+	HDMI_REG(VC4_HD_M_CTL),
+	HDMI_REG(VC4_HD_MAI_CTL),
+	HDMI_REG(VC4_HD_MAI_THR),
+	HDMI_REG(VC4_HD_MAI_FMT),
+	HDMI_REG(VC4_HD_MAI_SMP),
+	HDMI_REG(VC4_HD_VID_CTL),
+	HDMI_REG(VC4_HD_CSC_CTL),
+	HDMI_REG(VC4_HD_FRAME_COUNT),
+};
+
+#ifdef CONFIG_DEBUG_FS
+int vc4_hdmi_debugfs_regs(struct seq_file *m, void *unused)
+{
+	struct drm_info_node *node = (struct drm_info_node *)m->private;
+	struct drm_device *dev = node->minor->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(hdmi_regs); i++) {
+		seq_printf(m, "%s (0x%04x): 0x%08x\n",
+			   hdmi_regs[i].name, hdmi_regs[i].reg,
+			   HDMI_READ(hdmi_regs[i].reg));
+	}
+
+	for (i = 0; i < ARRAY_SIZE(hd_regs); i++) {
+		seq_printf(m, "%s (0x%04x): 0x%08x\n",
+			   hd_regs[i].name, hd_regs[i].reg,
+			   HD_READ(hd_regs[i].reg));
+	}
+
+	return 0;
+}
+#endif /* CONFIG_DEBUG_FS */
+
+static void vc4_hdmi_dump_regs(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(hdmi_regs); i++) {
+		DRM_INFO("0x%04x (%s): 0x%08x\n",
+			 hdmi_regs[i].reg, hdmi_regs[i].name,
+			 HDMI_READ(hdmi_regs[i].reg));
+	}
+	for (i = 0; i < ARRAY_SIZE(hd_regs); i++) {
+		DRM_INFO("0x%04x (%s): 0x%08x\n",
+			 hd_regs[i].reg, hd_regs[i].name,
+			 HD_READ(hd_regs[i].reg));
+	}
+}
+
+static enum drm_connector_status
+vc4_hdmi_connector_detect(struct drm_connector *connector, bool force)
+{
+	struct drm_device *dev = connector->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	if (vc4->hdmi->hpd_gpio) {
+		if (gpio_get_value_cansleep(vc4->hdmi->hpd_gpio) ^
+		    vc4->hdmi->hpd_active_low)
+			return connector_status_connected;
+		cec_phys_addr_invalidate(vc4->hdmi->cec_adap);
+		return connector_status_disconnected;
+	}
+
+	if (drm_probe_ddc(vc4->hdmi->ddc))
+		return connector_status_connected;
+
+	if (HDMI_READ(VC4_HDMI_HOTPLUG) & VC4_HDMI_HOTPLUG_CONNECTED)
+		return connector_status_connected;
+	cec_phys_addr_invalidate(vc4->hdmi->cec_adap);
+	return connector_status_disconnected;
+}
+
+static void vc4_hdmi_connector_destroy(struct drm_connector *connector)
+{
+	drm_connector_unregister(connector);
+	drm_connector_cleanup(connector);
+}
+
+static int vc4_hdmi_connector_get_modes(struct drm_connector *connector)
+{
+	struct vc4_hdmi_connector *vc4_connector =
+		to_vc4_hdmi_connector(connector);
+	struct drm_encoder *encoder = vc4_connector->encoder;
+	struct vc4_hdmi_encoder *vc4_encoder = to_vc4_hdmi_encoder(encoder);
+	struct drm_device *dev = connector->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int ret = 0;
+	struct edid *edid;
+
+	edid = drm_get_edid(connector, vc4->hdmi->ddc);
+	cec_s_phys_addr_from_edid(vc4->hdmi->cec_adap, edid);
+	if (!edid)
+		return -ENODEV;
+
+	vc4_encoder->hdmi_monitor = drm_detect_hdmi_monitor(edid);
+
+	if (edid && edid->input & DRM_EDID_INPUT_DIGITAL) {
+		vc4_encoder->rgb_range_selectable =
+			drm_rgb_quant_range_selectable(edid);
+	}
+
+	drm_connector_update_edid_property(connector, edid);
+	ret = drm_add_edid_modes(connector, edid);
+	kfree(edid);
+
+	return ret;
+}
+
+static const struct drm_connector_funcs vc4_hdmi_connector_funcs = {
+	.detect = vc4_hdmi_connector_detect,
+	.fill_modes = drm_helper_probe_single_connector_modes,
+	.destroy = vc4_hdmi_connector_destroy,
+	.reset = drm_atomic_helper_connector_reset,
+	.atomic_duplicate_state = drm_atomic_helper_connector_duplicate_state,
+	.atomic_destroy_state = drm_atomic_helper_connector_destroy_state,
+};
+
+static const struct drm_connector_helper_funcs vc4_hdmi_connector_helper_funcs = {
+	.get_modes = vc4_hdmi_connector_get_modes,
+};
+
+static struct drm_connector *vc4_hdmi_connector_init(struct drm_device *dev,
+						     struct drm_encoder *encoder)
+{
+	struct drm_connector *connector;
+	struct vc4_hdmi_connector *hdmi_connector;
+
+	hdmi_connector = devm_kzalloc(dev->dev, sizeof(*hdmi_connector),
+				      GFP_KERNEL);
+	if (!hdmi_connector)
+		return ERR_PTR(-ENOMEM);
+	connector = &hdmi_connector->base;
+
+	hdmi_connector->encoder = encoder;
+
+	drm_connector_init(dev, connector, &vc4_hdmi_connector_funcs,
+			   DRM_MODE_CONNECTOR_HDMIA);
+	drm_connector_helper_add(connector, &vc4_hdmi_connector_helper_funcs);
+
+	connector->polled = (DRM_CONNECTOR_POLL_CONNECT |
+			     DRM_CONNECTOR_POLL_DISCONNECT);
+
+	connector->interlace_allowed = 1;
+	connector->doublescan_allowed = 0;
+
+	drm_connector_attach_encoder(connector, encoder);
+
+	return connector;
+}
+
+static void vc4_hdmi_encoder_destroy(struct drm_encoder *encoder)
+{
+	drm_encoder_cleanup(encoder);
+}
+
+static const struct drm_encoder_funcs vc4_hdmi_encoder_funcs = {
+	.destroy = vc4_hdmi_encoder_destroy,
+};
+
+static int vc4_hdmi_stop_packet(struct drm_encoder *encoder,
+				enum hdmi_infoframe_type type)
+{
+	struct drm_device *dev = encoder->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	u32 packet_id = type - 0x80;
+
+	HDMI_WRITE(VC4_HDMI_RAM_PACKET_CONFIG,
+		   HDMI_READ(VC4_HDMI_RAM_PACKET_CONFIG) & ~BIT(packet_id));
+
+	return wait_for(!(HDMI_READ(VC4_HDMI_RAM_PACKET_STATUS) &
+			  BIT(packet_id)), 100);
+}
+
+static void vc4_hdmi_write_infoframe(struct drm_encoder *encoder,
+				     union hdmi_infoframe *frame)
+{
+	struct drm_device *dev = encoder->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	u32 packet_id = frame->any.type - 0x80;
+	u32 packet_reg = VC4_HDMI_RAM_PACKET(packet_id);
+	uint8_t buffer[VC4_HDMI_PACKET_STRIDE];
+	ssize_t len, i;
+	int ret;
+
+	WARN_ONCE(!(HDMI_READ(VC4_HDMI_RAM_PACKET_CONFIG) &
+		    VC4_HDMI_RAM_PACKET_ENABLE),
+		  "Packet RAM has to be on to store the packet.");
+
+	len = hdmi_infoframe_pack(frame, buffer, sizeof(buffer));
+	if (len < 0)
+		return;
+
+	ret = vc4_hdmi_stop_packet(encoder, frame->any.type);
+	if (ret) {
+		DRM_ERROR("Failed to wait for infoframe to go idle: %d\n", ret);
+		return;
+	}
+
+	for (i = 0; i < len; i += 7) {
+		HDMI_WRITE(packet_reg,
+			   buffer[i + 0] << 0 |
+			   buffer[i + 1] << 8 |
+			   buffer[i + 2] << 16);
+		packet_reg += 4;
+
+		HDMI_WRITE(packet_reg,
+			   buffer[i + 3] << 0 |
+			   buffer[i + 4] << 8 |
+			   buffer[i + 5] << 16 |
+			   buffer[i + 6] << 24);
+		packet_reg += 4;
+	}
+
+	HDMI_WRITE(VC4_HDMI_RAM_PACKET_CONFIG,
+		   HDMI_READ(VC4_HDMI_RAM_PACKET_CONFIG) | BIT(packet_id));
+	ret = wait_for((HDMI_READ(VC4_HDMI_RAM_PACKET_STATUS) &
+			BIT(packet_id)), 100);
+	if (ret)
+		DRM_ERROR("Failed to wait for infoframe to start: %d\n", ret);
+}
+
+static void vc4_hdmi_set_avi_infoframe(struct drm_encoder *encoder)
+{
+	struct vc4_hdmi_encoder *vc4_encoder = to_vc4_hdmi_encoder(encoder);
+	struct drm_crtc *crtc = encoder->crtc;
+	const struct drm_display_mode *mode = &crtc->state->adjusted_mode;
+	union hdmi_infoframe frame;
+	int ret;
+
+	ret = drm_hdmi_avi_infoframe_from_display_mode(&frame.avi, mode, false);
+	if (ret < 0) {
+		DRM_ERROR("couldn't fill AVI infoframe\n");
+		return;
+	}
+
+	drm_hdmi_avi_infoframe_quant_range(&frame.avi, mode,
+					   vc4_encoder->limited_rgb_range ?
+					   HDMI_QUANTIZATION_RANGE_LIMITED :
+					   HDMI_QUANTIZATION_RANGE_FULL,
+					   vc4_encoder->rgb_range_selectable,
+					   false);
+
+	vc4_hdmi_write_infoframe(encoder, &frame);
+}
+
+static void vc4_hdmi_set_spd_infoframe(struct drm_encoder *encoder)
+{
+	union hdmi_infoframe frame;
+	int ret;
+
+	ret = hdmi_spd_infoframe_init(&frame.spd, "Broadcom", "Videocore");
+	if (ret < 0) {
+		DRM_ERROR("couldn't fill SPD infoframe\n");
+		return;
+	}
+
+	frame.spd.sdi = HDMI_SPD_SDI_PC;
+
+	vc4_hdmi_write_infoframe(encoder, &frame);
+}
+
+static void vc4_hdmi_set_audio_infoframe(struct drm_encoder *encoder)
+{
+	struct drm_device *drm = encoder->dev;
+	struct vc4_dev *vc4 = drm->dev_private;
+	struct vc4_hdmi *hdmi = vc4->hdmi;
+	union hdmi_infoframe frame;
+	int ret;
+
+	ret = hdmi_audio_infoframe_init(&frame.audio);
+
+	frame.audio.coding_type = HDMI_AUDIO_CODING_TYPE_STREAM;
+	frame.audio.sample_frequency = HDMI_AUDIO_SAMPLE_FREQUENCY_STREAM;
+	frame.audio.sample_size = HDMI_AUDIO_SAMPLE_SIZE_STREAM;
+	frame.audio.channels = hdmi->audio.channels;
+
+	vc4_hdmi_write_infoframe(encoder, &frame);
+}
+
+static void vc4_hdmi_set_infoframes(struct drm_encoder *encoder)
+{
+	vc4_hdmi_set_avi_infoframe(encoder);
+	vc4_hdmi_set_spd_infoframe(encoder);
+}
+
+static void vc4_hdmi_encoder_disable(struct drm_encoder *encoder)
+{
+	struct drm_device *dev = encoder->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_hdmi *hdmi = vc4->hdmi;
+	int ret;
+
+	HDMI_WRITE(VC4_HDMI_RAM_PACKET_CONFIG, 0);
+
+	HDMI_WRITE(VC4_HDMI_TX_PHY_RESET_CTL, 0xf << 16);
+	HD_WRITE(VC4_HD_VID_CTL,
+		 HD_READ(VC4_HD_VID_CTL) & ~VC4_HD_VID_CTL_ENABLE);
+
+	clk_disable_unprepare(hdmi->pixel_clock);
+
+	ret = pm_runtime_put(&hdmi->pdev->dev);
+	if (ret < 0)
+		DRM_ERROR("Failed to release power domain: %d\n", ret);
+}
+
+static void vc4_hdmi_encoder_enable(struct drm_encoder *encoder)
+{
+	struct drm_display_mode *mode = &encoder->crtc->state->adjusted_mode;
+	struct vc4_hdmi_encoder *vc4_encoder = to_vc4_hdmi_encoder(encoder);
+	struct drm_device *dev = encoder->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_hdmi *hdmi = vc4->hdmi;
+	bool debug_dump_regs = false;
+	bool hsync_pos = mode->flags & DRM_MODE_FLAG_PHSYNC;
+	bool vsync_pos = mode->flags & DRM_MODE_FLAG_PVSYNC;
+	bool interlaced = mode->flags & DRM_MODE_FLAG_INTERLACE;
+	u32 pixel_rep = (mode->flags & DRM_MODE_FLAG_DBLCLK) ? 2 : 1;
+	u32 verta = (VC4_SET_FIELD(mode->crtc_vsync_end - mode->crtc_vsync_start,
+				   VC4_HDMI_VERTA_VSP) |
+		     VC4_SET_FIELD(mode->crtc_vsync_start - mode->crtc_vdisplay,
+				   VC4_HDMI_VERTA_VFP) |
+		     VC4_SET_FIELD(mode->crtc_vdisplay, VC4_HDMI_VERTA_VAL));
+	u32 vertb = (VC4_SET_FIELD(0, VC4_HDMI_VERTB_VSPO) |
+		     VC4_SET_FIELD(mode->crtc_vtotal - mode->crtc_vsync_end,
+				   VC4_HDMI_VERTB_VBP));
+	u32 vertb_even = (VC4_SET_FIELD(0, VC4_HDMI_VERTB_VSPO) |
+			  VC4_SET_FIELD(mode->crtc_vtotal -
+					mode->crtc_vsync_end -
+					interlaced,
+					VC4_HDMI_VERTB_VBP));
+	u32 csc_ctl;
+	int ret;
+
+	ret = pm_runtime_get_sync(&hdmi->pdev->dev);
+	if (ret < 0) {
+		DRM_ERROR("Failed to retain power domain: %d\n", ret);
+		return;
+	}
+
+	ret = clk_set_rate(hdmi->pixel_clock,
+			   mode->clock * 1000 *
+			   ((mode->flags & DRM_MODE_FLAG_DBLCLK) ? 2 : 1));
+	if (ret) {
+		DRM_ERROR("Failed to set pixel clock rate: %d\n", ret);
+		return;
+	}
+
+	ret = clk_prepare_enable(hdmi->pixel_clock);
+	if (ret) {
+		DRM_ERROR("Failed to turn on pixel clock: %d\n", ret);
+		return;
+	}
+
+	HDMI_WRITE(VC4_HDMI_SW_RESET_CONTROL,
+		   VC4_HDMI_SW_RESET_HDMI |
+		   VC4_HDMI_SW_RESET_FORMAT_DETECT);
+
+	HDMI_WRITE(VC4_HDMI_SW_RESET_CONTROL, 0);
+
+	/* PHY should be in reset, like
+	 * vc4_hdmi_encoder_disable() does.
+	 */
+	HDMI_WRITE(VC4_HDMI_TX_PHY_RESET_CTL, 0xf << 16);
+
+	HDMI_WRITE(VC4_HDMI_TX_PHY_RESET_CTL, 0);
+
+	if (debug_dump_regs) {
+		DRM_INFO("HDMI regs before:\n");
+		vc4_hdmi_dump_regs(dev);
+	}
+
+	HD_WRITE(VC4_HD_VID_CTL, 0);
+
+	HDMI_WRITE(VC4_HDMI_SCHEDULER_CONTROL,
+		   HDMI_READ(VC4_HDMI_SCHEDULER_CONTROL) |
+		   VC4_HDMI_SCHEDULER_CONTROL_MANUAL_FORMAT |
+		   VC4_HDMI_SCHEDULER_CONTROL_IGNORE_VSYNC_PREDICTS);
+
+	HDMI_WRITE(VC4_HDMI_HORZA,
+		   (vsync_pos ? VC4_HDMI_HORZA_VPOS : 0) |
+		   (hsync_pos ? VC4_HDMI_HORZA_HPOS : 0) |
+		   VC4_SET_FIELD(mode->hdisplay * pixel_rep,
+				 VC4_HDMI_HORZA_HAP));
+
+	HDMI_WRITE(VC4_HDMI_HORZB,
+		   VC4_SET_FIELD((mode->htotal -
+				  mode->hsync_end) * pixel_rep,
+				 VC4_HDMI_HORZB_HBP) |
+		   VC4_SET_FIELD((mode->hsync_end -
+				  mode->hsync_start) * pixel_rep,
+				 VC4_HDMI_HORZB_HSP) |
+		   VC4_SET_FIELD((mode->hsync_start -
+				  mode->hdisplay) * pixel_rep,
+				 VC4_HDMI_HORZB_HFP));
+
+	HDMI_WRITE(VC4_HDMI_VERTA0, verta);
+	HDMI_WRITE(VC4_HDMI_VERTA1, verta);
+
+	HDMI_WRITE(VC4_HDMI_VERTB0, vertb_even);
+	HDMI_WRITE(VC4_HDMI_VERTB1, vertb);
+
+	HD_WRITE(VC4_HD_VID_CTL,
+		 (vsync_pos ? 0 : VC4_HD_VID_CTL_VSYNC_LOW) |
+		 (hsync_pos ? 0 : VC4_HD_VID_CTL_HSYNC_LOW));
+
+	csc_ctl = VC4_SET_FIELD(VC4_HD_CSC_CTL_ORDER_BGR,
+				VC4_HD_CSC_CTL_ORDER);
+
+	if (vc4_encoder->hdmi_monitor &&
+	    drm_default_rgb_quant_range(mode) ==
+	    HDMI_QUANTIZATION_RANGE_LIMITED) {
+		/* CEA VICs other than #1 requre limited range RGB
+		 * output unless overridden by an AVI infoframe.
+		 * Apply a colorspace conversion to squash 0-255 down
+		 * to 16-235.  The matrix here is:
+		 *
+		 * [ 0      0      0.8594 16]
+		 * [ 0      0.8594 0      16]
+		 * [ 0.8594 0      0      16]
+		 * [ 0      0      0       1]
+		 */
+		csc_ctl |= VC4_HD_CSC_CTL_ENABLE;
+		csc_ctl |= VC4_HD_CSC_CTL_RGB2YCC;
+		csc_ctl |= VC4_SET_FIELD(VC4_HD_CSC_CTL_MODE_CUSTOM,
+					 VC4_HD_CSC_CTL_MODE);
+
+		HD_WRITE(VC4_HD_CSC_12_11, (0x000 << 16) | 0x000);
+		HD_WRITE(VC4_HD_CSC_14_13, (0x100 << 16) | 0x6e0);
+		HD_WRITE(VC4_HD_CSC_22_21, (0x6e0 << 16) | 0x000);
+		HD_WRITE(VC4_HD_CSC_24_23, (0x100 << 16) | 0x000);
+		HD_WRITE(VC4_HD_CSC_32_31, (0x000 << 16) | 0x6e0);
+		HD_WRITE(VC4_HD_CSC_34_33, (0x100 << 16) | 0x000);
+		vc4_encoder->limited_rgb_range = true;
+	} else {
+		vc4_encoder->limited_rgb_range = false;
+	}
+
+	/* The RGB order applies even when CSC is disabled. */
+	HD_WRITE(VC4_HD_CSC_CTL, csc_ctl);
+
+	HDMI_WRITE(VC4_HDMI_FIFO_CTL, VC4_HDMI_FIFO_CTL_MASTER_SLAVE_N);
+
+	if (debug_dump_regs) {
+		DRM_INFO("HDMI regs after:\n");
+		vc4_hdmi_dump_regs(dev);
+	}
+
+	HD_WRITE(VC4_HD_VID_CTL,
+		 HD_READ(VC4_HD_VID_CTL) |
+		 VC4_HD_VID_CTL_ENABLE |
+		 VC4_HD_VID_CTL_UNDERFLOW_ENABLE |
+		 VC4_HD_VID_CTL_FRAME_COUNTER_RESET);
+
+	if (vc4_encoder->hdmi_monitor) {
+		HDMI_WRITE(VC4_HDMI_SCHEDULER_CONTROL,
+			   HDMI_READ(VC4_HDMI_SCHEDULER_CONTROL) |
+			   VC4_HDMI_SCHEDULER_CONTROL_MODE_HDMI);
+
+		ret = wait_for(HDMI_READ(VC4_HDMI_SCHEDULER_CONTROL) &
+			       VC4_HDMI_SCHEDULER_CONTROL_HDMI_ACTIVE, 1000);
+		WARN_ONCE(ret, "Timeout waiting for "
+			  "VC4_HDMI_SCHEDULER_CONTROL_HDMI_ACTIVE\n");
+	} else {
+		HDMI_WRITE(VC4_HDMI_RAM_PACKET_CONFIG,
+			   HDMI_READ(VC4_HDMI_RAM_PACKET_CONFIG) &
+			   ~(VC4_HDMI_RAM_PACKET_ENABLE));
+		HDMI_WRITE(VC4_HDMI_SCHEDULER_CONTROL,
+			   HDMI_READ(VC4_HDMI_SCHEDULER_CONTROL) &
+			   ~VC4_HDMI_SCHEDULER_CONTROL_MODE_HDMI);
+
+		ret = wait_for(!(HDMI_READ(VC4_HDMI_SCHEDULER_CONTROL) &
+				 VC4_HDMI_SCHEDULER_CONTROL_HDMI_ACTIVE), 1000);
+		WARN_ONCE(ret, "Timeout waiting for "
+			  "!VC4_HDMI_SCHEDULER_CONTROL_HDMI_ACTIVE\n");
+	}
+
+	if (vc4_encoder->hdmi_monitor) {
+		u32 drift;
+
+		WARN_ON(!(HDMI_READ(VC4_HDMI_SCHEDULER_CONTROL) &
+			  VC4_HDMI_SCHEDULER_CONTROL_HDMI_ACTIVE));
+		HDMI_WRITE(VC4_HDMI_SCHEDULER_CONTROL,
+			   HDMI_READ(VC4_HDMI_SCHEDULER_CONTROL) |
+			   VC4_HDMI_SCHEDULER_CONTROL_VERT_ALWAYS_KEEPOUT);
+
+		HDMI_WRITE(VC4_HDMI_RAM_PACKET_CONFIG,
+			   VC4_HDMI_RAM_PACKET_ENABLE);
+
+		vc4_hdmi_set_infoframes(encoder);
+
+		drift = HDMI_READ(VC4_HDMI_FIFO_CTL);
+		drift &= VC4_HDMI_FIFO_VALID_WRITE_MASK;
+
+		HDMI_WRITE(VC4_HDMI_FIFO_CTL,
+			   drift & ~VC4_HDMI_FIFO_CTL_RECENTER);
+		HDMI_WRITE(VC4_HDMI_FIFO_CTL,
+			   drift | VC4_HDMI_FIFO_CTL_RECENTER);
+		usleep_range(1000, 1100);
+		HDMI_WRITE(VC4_HDMI_FIFO_CTL,
+			   drift & ~VC4_HDMI_FIFO_CTL_RECENTER);
+		HDMI_WRITE(VC4_HDMI_FIFO_CTL,
+			   drift | VC4_HDMI_FIFO_CTL_RECENTER);
+
+		ret = wait_for(HDMI_READ(VC4_HDMI_FIFO_CTL) &
+			       VC4_HDMI_FIFO_CTL_RECENTER_DONE, 1);
+		WARN_ONCE(ret, "Timeout waiting for "
+			  "VC4_HDMI_FIFO_CTL_RECENTER_DONE");
+	}
+}
+
+static enum drm_mode_status
+vc4_hdmi_encoder_mode_valid(struct drm_encoder *crtc,
+			    const struct drm_display_mode *mode)
+{
+	/*
+	 * As stated in RPi's vc4 firmware "HDMI state machine (HSM) clock must
+	 * be faster than pixel clock, infinitesimally faster, tested in
+	 * simulation. Otherwise, exact value is unimportant for HDMI
+	 * operation." This conflicts with bcm2835's vc4 documentation, which
+	 * states HSM's clock has to be at least 108% of the pixel clock.
+	 *
+	 * Real life tests reveal that vc4's firmware statement holds up, and
+	 * users are able to use pixel clocks closer to HSM's, namely for
+	 * 1920x1200@60Hz. So it was decided to have leave a 1% margin between
+	 * both clocks. Which, for RPi0-3 implies a maximum pixel clock of
+	 * 162MHz.
+	 *
+	 * Additionally, the AXI clock needs to be at least 25% of
+	 * pixel clock, but HSM ends up being the limiting factor.
+	 */
+	if (mode->clock > HSM_CLOCK_FREQ / (1000 * 101 / 100))
+		return MODE_CLOCK_HIGH;
+
+	return MODE_OK;
+}
+
+static const struct drm_encoder_helper_funcs vc4_hdmi_encoder_helper_funcs = {
+	.mode_valid = vc4_hdmi_encoder_mode_valid,
+	.disable = vc4_hdmi_encoder_disable,
+	.enable = vc4_hdmi_encoder_enable,
+};
+
+/* HDMI audio codec callbacks */
+static void vc4_hdmi_audio_set_mai_clock(struct vc4_hdmi *hdmi)
+{
+	struct drm_device *drm = hdmi->encoder->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	u32 hsm_clock = clk_get_rate(hdmi->hsm_clock);
+	unsigned long n, m;
+
+	rational_best_approximation(hsm_clock, hdmi->audio.samplerate,
+				    VC4_HD_MAI_SMP_N_MASK >>
+				    VC4_HD_MAI_SMP_N_SHIFT,
+				    (VC4_HD_MAI_SMP_M_MASK >>
+				     VC4_HD_MAI_SMP_M_SHIFT) + 1,
+				    &n, &m);
+
+	HD_WRITE(VC4_HD_MAI_SMP,
+		 VC4_SET_FIELD(n, VC4_HD_MAI_SMP_N) |
+		 VC4_SET_FIELD(m - 1, VC4_HD_MAI_SMP_M));
+}
+
+static void vc4_hdmi_set_n_cts(struct vc4_hdmi *hdmi)
+{
+	struct drm_encoder *encoder = hdmi->encoder;
+	struct drm_crtc *crtc = encoder->crtc;
+	struct drm_device *drm = encoder->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	const struct drm_display_mode *mode = &crtc->state->adjusted_mode;
+	u32 samplerate = hdmi->audio.samplerate;
+	u32 n, cts;
+	u64 tmp;
+
+	n = 128 * samplerate / 1000;
+	tmp = (u64)(mode->clock * 1000) * n;
+	do_div(tmp, 128 * samplerate);
+	cts = tmp;
+
+	HDMI_WRITE(VC4_HDMI_CRP_CFG,
+		   VC4_HDMI_CRP_CFG_EXTERNAL_CTS_EN |
+		   VC4_SET_FIELD(n, VC4_HDMI_CRP_CFG_N));
+
+	/*
+	 * We could get slightly more accurate clocks in some cases by
+	 * providing a CTS_1 value.  The two CTS values are alternated
+	 * between based on the period fields
+	 */
+	HDMI_WRITE(VC4_HDMI_CTS_0, cts);
+	HDMI_WRITE(VC4_HDMI_CTS_1, cts);
+}
+
+static inline struct vc4_hdmi *dai_to_hdmi(struct snd_soc_dai *dai)
+{
+	struct snd_soc_card *card = snd_soc_dai_get_drvdata(dai);
+
+	return snd_soc_card_get_drvdata(card);
+}
+
+static int vc4_hdmi_audio_startup(struct snd_pcm_substream *substream,
+				  struct snd_soc_dai *dai)
+{
+	struct vc4_hdmi *hdmi = dai_to_hdmi(dai);
+	struct drm_encoder *encoder = hdmi->encoder;
+	struct vc4_dev *vc4 = to_vc4_dev(encoder->dev);
+	int ret;
+
+	if (hdmi->audio.substream && hdmi->audio.substream != substream)
+		return -EINVAL;
+
+	hdmi->audio.substream = substream;
+
+	/*
+	 * If the HDMI encoder hasn't probed, or the encoder is
+	 * currently in DVI mode, treat the codec dai as missing.
+	 */
+	if (!encoder->crtc || !(HDMI_READ(VC4_HDMI_RAM_PACKET_CONFIG) &
+				VC4_HDMI_RAM_PACKET_ENABLE))
+		return -ENODEV;
+
+	ret = snd_pcm_hw_constraint_eld(substream->runtime,
+					hdmi->connector->eld);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static int vc4_hdmi_audio_set_fmt(struct snd_soc_dai *dai, unsigned int fmt)
+{
+	return 0;
+}
+
+static void vc4_hdmi_audio_reset(struct vc4_hdmi *hdmi)
+{
+	struct drm_encoder *encoder = hdmi->encoder;
+	struct drm_device *drm = encoder->dev;
+	struct device *dev = &hdmi->pdev->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	int ret;
+
+	ret = vc4_hdmi_stop_packet(encoder, HDMI_INFOFRAME_TYPE_AUDIO);
+	if (ret)
+		dev_err(dev, "Failed to stop audio infoframe: %d\n", ret);
+
+	HD_WRITE(VC4_HD_MAI_CTL, VC4_HD_MAI_CTL_RESET);
+	HD_WRITE(VC4_HD_MAI_CTL, VC4_HD_MAI_CTL_ERRORF);
+	HD_WRITE(VC4_HD_MAI_CTL, VC4_HD_MAI_CTL_FLUSH);
+}
+
+static void vc4_hdmi_audio_shutdown(struct snd_pcm_substream *substream,
+				    struct snd_soc_dai *dai)
+{
+	struct vc4_hdmi *hdmi = dai_to_hdmi(dai);
+
+	if (substream != hdmi->audio.substream)
+		return;
+
+	vc4_hdmi_audio_reset(hdmi);
+
+	hdmi->audio.substream = NULL;
+}
+
+/* HDMI audio codec callbacks */
+static int vc4_hdmi_audio_hw_params(struct snd_pcm_substream *substream,
+				    struct snd_pcm_hw_params *params,
+				    struct snd_soc_dai *dai)
+{
+	struct vc4_hdmi *hdmi = dai_to_hdmi(dai);
+	struct drm_encoder *encoder = hdmi->encoder;
+	struct drm_device *drm = encoder->dev;
+	struct device *dev = &hdmi->pdev->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	u32 audio_packet_config, channel_mask;
+	u32 channel_map, i;
+
+	if (substream != hdmi->audio.substream)
+		return -EINVAL;
+
+	dev_dbg(dev, "%s: %u Hz, %d bit, %d channels\n", __func__,
+		params_rate(params), params_width(params),
+		params_channels(params));
+
+	hdmi->audio.channels = params_channels(params);
+	hdmi->audio.samplerate = params_rate(params);
+
+	HD_WRITE(VC4_HD_MAI_CTL,
+		 VC4_HD_MAI_CTL_RESET |
+		 VC4_HD_MAI_CTL_FLUSH |
+		 VC4_HD_MAI_CTL_DLATE |
+		 VC4_HD_MAI_CTL_ERRORE |
+		 VC4_HD_MAI_CTL_ERRORF);
+
+	vc4_hdmi_audio_set_mai_clock(hdmi);
+
+	audio_packet_config =
+		VC4_HDMI_AUDIO_PACKET_ZERO_DATA_ON_SAMPLE_FLAT |
+		VC4_HDMI_AUDIO_PACKET_ZERO_DATA_ON_INACTIVE_CHANNELS |
+		VC4_SET_FIELD(0xf, VC4_HDMI_AUDIO_PACKET_B_FRAME_IDENTIFIER);
+
+	channel_mask = GENMASK(hdmi->audio.channels - 1, 0);
+	audio_packet_config |= VC4_SET_FIELD(channel_mask,
+					     VC4_HDMI_AUDIO_PACKET_CEA_MASK);
+
+	/* Set the MAI threshold.  This logic mimics the firmware's. */
+	if (hdmi->audio.samplerate > 96000) {
+		HD_WRITE(VC4_HD_MAI_THR,
+			 VC4_SET_FIELD(0x12, VC4_HD_MAI_THR_DREQHIGH) |
+			 VC4_SET_FIELD(0x12, VC4_HD_MAI_THR_DREQLOW));
+	} else if (hdmi->audio.samplerate > 48000) {
+		HD_WRITE(VC4_HD_MAI_THR,
+			 VC4_SET_FIELD(0x14, VC4_HD_MAI_THR_DREQHIGH) |
+			 VC4_SET_FIELD(0x12, VC4_HD_MAI_THR_DREQLOW));
+	} else {
+		HD_WRITE(VC4_HD_MAI_THR,
+			 VC4_SET_FIELD(0x10, VC4_HD_MAI_THR_PANICHIGH) |
+			 VC4_SET_FIELD(0x10, VC4_HD_MAI_THR_PANICLOW) |
+			 VC4_SET_FIELD(0x10, VC4_HD_MAI_THR_DREQHIGH) |
+			 VC4_SET_FIELD(0x10, VC4_HD_MAI_THR_DREQLOW));
+	}
+
+	HDMI_WRITE(VC4_HDMI_MAI_CONFIG,
+		   VC4_HDMI_MAI_CONFIG_BIT_REVERSE |
+		   VC4_SET_FIELD(channel_mask, VC4_HDMI_MAI_CHANNEL_MASK));
+
+	channel_map = 0;
+	for (i = 0; i < 8; i++) {
+		if (channel_mask & BIT(i))
+			channel_map |= i << (3 * i);
+	}
+
+	HDMI_WRITE(VC4_HDMI_MAI_CHANNEL_MAP, channel_map);
+	HDMI_WRITE(VC4_HDMI_AUDIO_PACKET_CONFIG, audio_packet_config);
+	vc4_hdmi_set_n_cts(hdmi);
+
+	return 0;
+}
+
+static int vc4_hdmi_audio_trigger(struct snd_pcm_substream *substream, int cmd,
+				  struct snd_soc_dai *dai)
+{
+	struct vc4_hdmi *hdmi = dai_to_hdmi(dai);
+	struct drm_encoder *encoder = hdmi->encoder;
+	struct drm_device *drm = encoder->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+
+	switch (cmd) {
+	case SNDRV_PCM_TRIGGER_START:
+		vc4_hdmi_set_audio_infoframe(encoder);
+		HDMI_WRITE(VC4_HDMI_TX_PHY_CTL0,
+			   HDMI_READ(VC4_HDMI_TX_PHY_CTL0) &
+			   ~VC4_HDMI_TX_PHY_RNG_PWRDN);
+		HD_WRITE(VC4_HD_MAI_CTL,
+			 VC4_SET_FIELD(hdmi->audio.channels,
+				       VC4_HD_MAI_CTL_CHNUM) |
+			 VC4_HD_MAI_CTL_ENABLE);
+		break;
+	case SNDRV_PCM_TRIGGER_STOP:
+		HD_WRITE(VC4_HD_MAI_CTL,
+			 VC4_HD_MAI_CTL_DLATE |
+			 VC4_HD_MAI_CTL_ERRORE |
+			 VC4_HD_MAI_CTL_ERRORF);
+		HDMI_WRITE(VC4_HDMI_TX_PHY_CTL0,
+			   HDMI_READ(VC4_HDMI_TX_PHY_CTL0) |
+			   VC4_HDMI_TX_PHY_RNG_PWRDN);
+		break;
+	default:
+		break;
+	}
+
+	return 0;
+}
+
+static inline struct vc4_hdmi *
+snd_component_to_hdmi(struct snd_soc_component *component)
+{
+	struct snd_soc_card *card = snd_soc_component_get_drvdata(component);
+
+	return snd_soc_card_get_drvdata(card);
+}
+
+static int vc4_hdmi_audio_eld_ctl_info(struct snd_kcontrol *kcontrol,
+				       struct snd_ctl_elem_info *uinfo)
+{
+	struct snd_soc_component *component = snd_kcontrol_chip(kcontrol);
+	struct vc4_hdmi *hdmi = snd_component_to_hdmi(component);
+
+	uinfo->type = SNDRV_CTL_ELEM_TYPE_BYTES;
+	uinfo->count = sizeof(hdmi->connector->eld);
+
+	return 0;
+}
+
+static int vc4_hdmi_audio_eld_ctl_get(struct snd_kcontrol *kcontrol,
+				      struct snd_ctl_elem_value *ucontrol)
+{
+	struct snd_soc_component *component = snd_kcontrol_chip(kcontrol);
+	struct vc4_hdmi *hdmi = snd_component_to_hdmi(component);
+
+	memcpy(ucontrol->value.bytes.data, hdmi->connector->eld,
+	       sizeof(hdmi->connector->eld));
+
+	return 0;
+}
+
+static const struct snd_kcontrol_new vc4_hdmi_audio_controls[] = {
+	{
+		.access = SNDRV_CTL_ELEM_ACCESS_READ |
+			  SNDRV_CTL_ELEM_ACCESS_VOLATILE,
+		.iface = SNDRV_CTL_ELEM_IFACE_PCM,
+		.name = "ELD",
+		.info = vc4_hdmi_audio_eld_ctl_info,
+		.get = vc4_hdmi_audio_eld_ctl_get,
+	},
+};
+
+static const struct snd_soc_dapm_widget vc4_hdmi_audio_widgets[] = {
+	SND_SOC_DAPM_OUTPUT("TX"),
+};
+
+static const struct snd_soc_dapm_route vc4_hdmi_audio_routes[] = {
+	{ "TX", NULL, "Playback" },
+};
+
+static const struct snd_soc_component_driver vc4_hdmi_audio_component_drv = {
+	.controls		= vc4_hdmi_audio_controls,
+	.num_controls		= ARRAY_SIZE(vc4_hdmi_audio_controls),
+	.dapm_widgets		= vc4_hdmi_audio_widgets,
+	.num_dapm_widgets	= ARRAY_SIZE(vc4_hdmi_audio_widgets),
+	.dapm_routes		= vc4_hdmi_audio_routes,
+	.num_dapm_routes	= ARRAY_SIZE(vc4_hdmi_audio_routes),
+	.idle_bias_on		= 1,
+	.use_pmdown_time	= 1,
+	.endianness		= 1,
+	.non_legacy_dai_naming	= 1,
+};
+
+static const struct snd_soc_dai_ops vc4_hdmi_audio_dai_ops = {
+	.startup = vc4_hdmi_audio_startup,
+	.shutdown = vc4_hdmi_audio_shutdown,
+	.hw_params = vc4_hdmi_audio_hw_params,
+	.set_fmt = vc4_hdmi_audio_set_fmt,
+	.trigger = vc4_hdmi_audio_trigger,
+};
+
+static struct snd_soc_dai_driver vc4_hdmi_audio_codec_dai_drv = {
+	.name = "vc4-hdmi-hifi",
+	.playback = {
+		.stream_name = "Playback",
+		.channels_min = 2,
+		.channels_max = 8,
+		.rates = SNDRV_PCM_RATE_32000 | SNDRV_PCM_RATE_44100 |
+			 SNDRV_PCM_RATE_48000 | SNDRV_PCM_RATE_88200 |
+			 SNDRV_PCM_RATE_96000 | SNDRV_PCM_RATE_176400 |
+			 SNDRV_PCM_RATE_192000,
+		.formats = SNDRV_PCM_FMTBIT_IEC958_SUBFRAME_LE,
+	},
+};
+
+static const struct snd_soc_component_driver vc4_hdmi_audio_cpu_dai_comp = {
+	.name = "vc4-hdmi-cpu-dai-component",
+};
+
+static int vc4_hdmi_audio_cpu_dai_probe(struct snd_soc_dai *dai)
+{
+	struct vc4_hdmi *hdmi = dai_to_hdmi(dai);
+
+	snd_soc_dai_init_dma_data(dai, &hdmi->audio.dma_data, NULL);
+
+	return 0;
+}
+
+static struct snd_soc_dai_driver vc4_hdmi_audio_cpu_dai_drv = {
+	.name = "vc4-hdmi-cpu-dai",
+	.probe  = vc4_hdmi_audio_cpu_dai_probe,
+	.playback = {
+		.stream_name = "Playback",
+		.channels_min = 1,
+		.channels_max = 8,
+		.rates = SNDRV_PCM_RATE_32000 | SNDRV_PCM_RATE_44100 |
+			 SNDRV_PCM_RATE_48000 | SNDRV_PCM_RATE_88200 |
+			 SNDRV_PCM_RATE_96000 | SNDRV_PCM_RATE_176400 |
+			 SNDRV_PCM_RATE_192000,
+		.formats = SNDRV_PCM_FMTBIT_IEC958_SUBFRAME_LE,
+	},
+	.ops = &vc4_hdmi_audio_dai_ops,
+};
+
+static const struct snd_dmaengine_pcm_config pcm_conf = {
+	.chan_names[SNDRV_PCM_STREAM_PLAYBACK] = "audio-rx",
+	.prepare_slave_config = snd_dmaengine_pcm_prepare_slave_config,
+};
+
+static int vc4_hdmi_audio_init(struct vc4_hdmi *hdmi)
+{
+	struct snd_soc_dai_link *dai_link = &hdmi->audio.link;
+	struct snd_soc_card *card = &hdmi->audio.card;
+	struct device *dev = &hdmi->pdev->dev;
+	const __be32 *addr;
+	int ret;
+
+	if (!of_find_property(dev->of_node, "dmas", NULL)) {
+		dev_warn(dev,
+			 "'dmas' DT property is missing, no HDMI audio\n");
+		return 0;
+	}
+
+	/*
+	 * Get the physical address of VC4_HD_MAI_DATA. We need to retrieve
+	 * the bus address specified in the DT, because the physical address
+	 * (the one returned by platform_get_resource()) is not appropriate
+	 * for DMA transfers.
+	 * This VC/MMU should probably be exposed to avoid this kind of hacks.
+	 */
+	addr = of_get_address(dev->of_node, 1, NULL, NULL);
+	hdmi->audio.dma_data.addr = be32_to_cpup(addr) + VC4_HD_MAI_DATA;
+	hdmi->audio.dma_data.addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
+	hdmi->audio.dma_data.maxburst = 2;
+
+	ret = devm_snd_dmaengine_pcm_register(dev, &pcm_conf, 0);
+	if (ret) {
+		dev_err(dev, "Could not register PCM component: %d\n", ret);
+		return ret;
+	}
+
+	ret = devm_snd_soc_register_component(dev, &vc4_hdmi_audio_cpu_dai_comp,
+					      &vc4_hdmi_audio_cpu_dai_drv, 1);
+	if (ret) {
+		dev_err(dev, "Could not register CPU DAI: %d\n", ret);
+		return ret;
+	}
+
+	/* register component and codec dai */
+	ret = devm_snd_soc_register_component(dev, &vc4_hdmi_audio_component_drv,
+				     &vc4_hdmi_audio_codec_dai_drv, 1);
+	if (ret) {
+		dev_err(dev, "Could not register component: %d\n", ret);
+		return ret;
+	}
+
+	dai_link->name = "MAI";
+	dai_link->stream_name = "MAI PCM";
+	dai_link->codec_dai_name = vc4_hdmi_audio_codec_dai_drv.name;
+	dai_link->cpu_dai_name = dev_name(dev);
+	dai_link->codec_name = dev_name(dev);
+	dai_link->platform_name = dev_name(dev);
+
+	card->dai_link = dai_link;
+	card->num_links = 1;
+	card->name = "vc4-hdmi";
+	card->dev = dev;
+	card->owner = THIS_MODULE;
+
+	/*
+	 * Be careful, snd_soc_register_card() calls dev_set_drvdata() and
+	 * stores a pointer to the snd card object in dev->driver_data. This
+	 * means we cannot use it for something else. The hdmi back-pointer is
+	 * now stored in card->drvdata and should be retrieved with
+	 * snd_soc_card_get_drvdata() if needed.
+	 */
+	snd_soc_card_set_drvdata(card, hdmi);
+	ret = devm_snd_soc_register_card(dev, card);
+	if (ret)
+		dev_err(dev, "Could not register sound card: %d\n", ret);
+
+	return ret;
+
+}
+
+#ifdef CONFIG_DRM_VC4_HDMI_CEC
+static irqreturn_t vc4_cec_irq_handler_thread(int irq, void *priv)
+{
+	struct vc4_dev *vc4 = priv;
+	struct vc4_hdmi *hdmi = vc4->hdmi;
+
+	if (hdmi->cec_irq_was_rx) {
+		if (hdmi->cec_rx_msg.len)
+			cec_received_msg(hdmi->cec_adap, &hdmi->cec_rx_msg);
+	} else if (hdmi->cec_tx_ok) {
+		cec_transmit_done(hdmi->cec_adap, CEC_TX_STATUS_OK,
+				  0, 0, 0, 0);
+	} else {
+		/*
+		 * This CEC implementation makes 1 retry, so if we
+		 * get a NACK, then that means it made 2 attempts.
+		 */
+		cec_transmit_done(hdmi->cec_adap, CEC_TX_STATUS_NACK,
+				  0, 2, 0, 0);
+	}
+	return IRQ_HANDLED;
+}
+
+static void vc4_cec_read_msg(struct vc4_dev *vc4, u32 cntrl1)
+{
+	struct cec_msg *msg = &vc4->hdmi->cec_rx_msg;
+	unsigned int i;
+
+	msg->len = 1 + ((cntrl1 & VC4_HDMI_CEC_REC_WRD_CNT_MASK) >>
+					VC4_HDMI_CEC_REC_WRD_CNT_SHIFT);
+	for (i = 0; i < msg->len; i += 4) {
+		u32 val = HDMI_READ(VC4_HDMI_CEC_RX_DATA_1 + i);
+
+		msg->msg[i] = val & 0xff;
+		msg->msg[i + 1] = (val >> 8) & 0xff;
+		msg->msg[i + 2] = (val >> 16) & 0xff;
+		msg->msg[i + 3] = (val >> 24) & 0xff;
+	}
+}
+
+static irqreturn_t vc4_cec_irq_handler(int irq, void *priv)
+{
+	struct vc4_dev *vc4 = priv;
+	struct vc4_hdmi *hdmi = vc4->hdmi;
+	u32 stat = HDMI_READ(VC4_HDMI_CPU_STATUS);
+	u32 cntrl1, cntrl5;
+
+	if (!(stat & VC4_HDMI_CPU_CEC))
+		return IRQ_NONE;
+	hdmi->cec_rx_msg.len = 0;
+	cntrl1 = HDMI_READ(VC4_HDMI_CEC_CNTRL_1);
+	cntrl5 = HDMI_READ(VC4_HDMI_CEC_CNTRL_5);
+	hdmi->cec_irq_was_rx = cntrl5 & VC4_HDMI_CEC_RX_CEC_INT;
+	if (hdmi->cec_irq_was_rx) {
+		vc4_cec_read_msg(vc4, cntrl1);
+		cntrl1 |= VC4_HDMI_CEC_CLEAR_RECEIVE_OFF;
+		HDMI_WRITE(VC4_HDMI_CEC_CNTRL_1, cntrl1);
+		cntrl1 &= ~VC4_HDMI_CEC_CLEAR_RECEIVE_OFF;
+	} else {
+		hdmi->cec_tx_ok = cntrl1 & VC4_HDMI_CEC_TX_STATUS_GOOD;
+		cntrl1 &= ~VC4_HDMI_CEC_START_XMIT_BEGIN;
+	}
+	HDMI_WRITE(VC4_HDMI_CEC_CNTRL_1, cntrl1);
+	HDMI_WRITE(VC4_HDMI_CPU_CLEAR, VC4_HDMI_CPU_CEC);
+
+	return IRQ_WAKE_THREAD;
+}
+
+static int vc4_hdmi_cec_adap_enable(struct cec_adapter *adap, bool enable)
+{
+	struct vc4_dev *vc4 = cec_get_drvdata(adap);
+	/* clock period in microseconds */
+	const u32 usecs = 1000000 / CEC_CLOCK_FREQ;
+	u32 val = HDMI_READ(VC4_HDMI_CEC_CNTRL_5);
+
+	val &= ~(VC4_HDMI_CEC_TX_SW_RESET | VC4_HDMI_CEC_RX_SW_RESET |
+		 VC4_HDMI_CEC_CNT_TO_4700_US_MASK |
+		 VC4_HDMI_CEC_CNT_TO_4500_US_MASK);
+	val |= ((4700 / usecs) << VC4_HDMI_CEC_CNT_TO_4700_US_SHIFT) |
+	       ((4500 / usecs) << VC4_HDMI_CEC_CNT_TO_4500_US_SHIFT);
+
+	if (enable) {
+		HDMI_WRITE(VC4_HDMI_CEC_CNTRL_5, val |
+			   VC4_HDMI_CEC_TX_SW_RESET | VC4_HDMI_CEC_RX_SW_RESET);
+		HDMI_WRITE(VC4_HDMI_CEC_CNTRL_5, val);
+		HDMI_WRITE(VC4_HDMI_CEC_CNTRL_2,
+			 ((1500 / usecs) << VC4_HDMI_CEC_CNT_TO_1500_US_SHIFT) |
+			 ((1300 / usecs) << VC4_HDMI_CEC_CNT_TO_1300_US_SHIFT) |
+			 ((800 / usecs) << VC4_HDMI_CEC_CNT_TO_800_US_SHIFT) |
+			 ((600 / usecs) << VC4_HDMI_CEC_CNT_TO_600_US_SHIFT) |
+			 ((400 / usecs) << VC4_HDMI_CEC_CNT_TO_400_US_SHIFT));
+		HDMI_WRITE(VC4_HDMI_CEC_CNTRL_3,
+			 ((2750 / usecs) << VC4_HDMI_CEC_CNT_TO_2750_US_SHIFT) |
+			 ((2400 / usecs) << VC4_HDMI_CEC_CNT_TO_2400_US_SHIFT) |
+			 ((2050 / usecs) << VC4_HDMI_CEC_CNT_TO_2050_US_SHIFT) |
+			 ((1700 / usecs) << VC4_HDMI_CEC_CNT_TO_1700_US_SHIFT));
+		HDMI_WRITE(VC4_HDMI_CEC_CNTRL_4,
+			 ((4300 / usecs) << VC4_HDMI_CEC_CNT_TO_4300_US_SHIFT) |
+			 ((3900 / usecs) << VC4_HDMI_CEC_CNT_TO_3900_US_SHIFT) |
+			 ((3600 / usecs) << VC4_HDMI_CEC_CNT_TO_3600_US_SHIFT) |
+			 ((3500 / usecs) << VC4_HDMI_CEC_CNT_TO_3500_US_SHIFT));
+
+		HDMI_WRITE(VC4_HDMI_CPU_MASK_CLEAR, VC4_HDMI_CPU_CEC);
+	} else {
+		HDMI_WRITE(VC4_HDMI_CPU_MASK_SET, VC4_HDMI_CPU_CEC);
+		HDMI_WRITE(VC4_HDMI_CEC_CNTRL_5, val |
+			   VC4_HDMI_CEC_TX_SW_RESET | VC4_HDMI_CEC_RX_SW_RESET);
+	}
+	return 0;
+}
+
+static int vc4_hdmi_cec_adap_log_addr(struct cec_adapter *adap, u8 log_addr)
+{
+	struct vc4_dev *vc4 = cec_get_drvdata(adap);
+
+	HDMI_WRITE(VC4_HDMI_CEC_CNTRL_1,
+		   (HDMI_READ(VC4_HDMI_CEC_CNTRL_1) & ~VC4_HDMI_CEC_ADDR_MASK) |
+		   (log_addr & 0xf) << VC4_HDMI_CEC_ADDR_SHIFT);
+	return 0;
+}
+
+static int vc4_hdmi_cec_adap_transmit(struct cec_adapter *adap, u8 attempts,
+				      u32 signal_free_time, struct cec_msg *msg)
+{
+	struct vc4_dev *vc4 = cec_get_drvdata(adap);
+	u32 val;
+	unsigned int i;
+
+	for (i = 0; i < msg->len; i += 4)
+		HDMI_WRITE(VC4_HDMI_CEC_TX_DATA_1 + i,
+			   (msg->msg[i]) |
+			   (msg->msg[i + 1] << 8) |
+			   (msg->msg[i + 2] << 16) |
+			   (msg->msg[i + 3] << 24));
+
+	val = HDMI_READ(VC4_HDMI_CEC_CNTRL_1);
+	val &= ~VC4_HDMI_CEC_START_XMIT_BEGIN;
+	HDMI_WRITE(VC4_HDMI_CEC_CNTRL_1, val);
+	val &= ~VC4_HDMI_CEC_MESSAGE_LENGTH_MASK;
+	val |= (msg->len - 1) << VC4_HDMI_CEC_MESSAGE_LENGTH_SHIFT;
+	val |= VC4_HDMI_CEC_START_XMIT_BEGIN;
+
+	HDMI_WRITE(VC4_HDMI_CEC_CNTRL_1, val);
+	return 0;
+}
+
+static const struct cec_adap_ops vc4_hdmi_cec_adap_ops = {
+	.adap_enable = vc4_hdmi_cec_adap_enable,
+	.adap_log_addr = vc4_hdmi_cec_adap_log_addr,
+	.adap_transmit = vc4_hdmi_cec_adap_transmit,
+};
+#endif
+
+static int vc4_hdmi_bind(struct device *dev, struct device *master, void *data)
+{
+	struct platform_device *pdev = to_platform_device(dev);
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = drm->dev_private;
+	struct vc4_hdmi *hdmi;
+	struct vc4_hdmi_encoder *vc4_hdmi_encoder;
+	struct device_node *ddc_node;
+	u32 value;
+	int ret;
+
+	hdmi = devm_kzalloc(dev, sizeof(*hdmi), GFP_KERNEL);
+	if (!hdmi)
+		return -ENOMEM;
+
+	vc4_hdmi_encoder = devm_kzalloc(dev, sizeof(*vc4_hdmi_encoder),
+					GFP_KERNEL);
+	if (!vc4_hdmi_encoder)
+		return -ENOMEM;
+	vc4_hdmi_encoder->base.type = VC4_ENCODER_TYPE_HDMI;
+	hdmi->encoder = &vc4_hdmi_encoder->base.base;
+
+	hdmi->pdev = pdev;
+	hdmi->hdmicore_regs = vc4_ioremap_regs(pdev, 0);
+	if (IS_ERR(hdmi->hdmicore_regs))
+		return PTR_ERR(hdmi->hdmicore_regs);
+
+	hdmi->hd_regs = vc4_ioremap_regs(pdev, 1);
+	if (IS_ERR(hdmi->hd_regs))
+		return PTR_ERR(hdmi->hd_regs);
+
+	hdmi->pixel_clock = devm_clk_get(dev, "pixel");
+	if (IS_ERR(hdmi->pixel_clock)) {
+		DRM_ERROR("Failed to get pixel clock\n");
+		return PTR_ERR(hdmi->pixel_clock);
+	}
+	hdmi->hsm_clock = devm_clk_get(dev, "hdmi");
+	if (IS_ERR(hdmi->hsm_clock)) {
+		DRM_ERROR("Failed to get HDMI state machine clock\n");
+		return PTR_ERR(hdmi->hsm_clock);
+	}
+
+	ddc_node = of_parse_phandle(dev->of_node, "ddc", 0);
+	if (!ddc_node) {
+		DRM_ERROR("Failed to find ddc node in device tree\n");
+		return -ENODEV;
+	}
+
+	hdmi->ddc = of_find_i2c_adapter_by_node(ddc_node);
+	of_node_put(ddc_node);
+	if (!hdmi->ddc) {
+		DRM_DEBUG("Failed to get ddc i2c adapter by node\n");
+		return -EPROBE_DEFER;
+	}
+
+	/* This is the rate that is set by the firmware.  The number
+	 * needs to be a bit higher than the pixel clock rate
+	 * (generally 148.5Mhz).
+	 */
+	ret = clk_set_rate(hdmi->hsm_clock, HSM_CLOCK_FREQ);
+	if (ret) {
+		DRM_ERROR("Failed to set HSM clock rate: %d\n", ret);
+		goto err_put_i2c;
+	}
+
+	ret = clk_prepare_enable(hdmi->hsm_clock);
+	if (ret) {
+		DRM_ERROR("Failed to turn on HDMI state machine clock: %d\n",
+			  ret);
+		goto err_put_i2c;
+	}
+
+	/* Only use the GPIO HPD pin if present in the DT, otherwise
+	 * we'll use the HDMI core's register.
+	 */
+	if (of_find_property(dev->of_node, "hpd-gpios", &value)) {
+		enum of_gpio_flags hpd_gpio_flags;
+
+		hdmi->hpd_gpio = of_get_named_gpio_flags(dev->of_node,
+							 "hpd-gpios", 0,
+							 &hpd_gpio_flags);
+		if (hdmi->hpd_gpio < 0) {
+			ret = hdmi->hpd_gpio;
+			goto err_unprepare_hsm;
+		}
+
+		hdmi->hpd_active_low = hpd_gpio_flags & OF_GPIO_ACTIVE_LOW;
+	}
+
+	vc4->hdmi = hdmi;
+
+	/* HDMI core must be enabled. */
+	if (!(HD_READ(VC4_HD_M_CTL) & VC4_HD_M_ENABLE)) {
+		HD_WRITE(VC4_HD_M_CTL, VC4_HD_M_SW_RST);
+		udelay(1);
+		HD_WRITE(VC4_HD_M_CTL, 0);
+
+		HD_WRITE(VC4_HD_M_CTL, VC4_HD_M_ENABLE);
+	}
+	pm_runtime_enable(dev);
+
+	drm_encoder_init(drm, hdmi->encoder, &vc4_hdmi_encoder_funcs,
+			 DRM_MODE_ENCODER_TMDS, NULL);
+	drm_encoder_helper_add(hdmi->encoder, &vc4_hdmi_encoder_helper_funcs);
+
+	hdmi->connector = vc4_hdmi_connector_init(drm, hdmi->encoder);
+	if (IS_ERR(hdmi->connector)) {
+		ret = PTR_ERR(hdmi->connector);
+		goto err_destroy_encoder;
+	}
+#ifdef CONFIG_DRM_VC4_HDMI_CEC
+	hdmi->cec_adap = cec_allocate_adapter(&vc4_hdmi_cec_adap_ops,
+					      vc4, "vc4",
+					      CEC_CAP_TRANSMIT |
+					      CEC_CAP_LOG_ADDRS |
+					      CEC_CAP_PASSTHROUGH |
+					      CEC_CAP_RC, 1);
+	ret = PTR_ERR_OR_ZERO(hdmi->cec_adap);
+	if (ret < 0)
+		goto err_destroy_conn;
+	HDMI_WRITE(VC4_HDMI_CPU_MASK_SET, 0xffffffff);
+	value = HDMI_READ(VC4_HDMI_CEC_CNTRL_1);
+	value &= ~VC4_HDMI_CEC_DIV_CLK_CNT_MASK;
+	/*
+	 * Set the logical address to Unregistered and set the clock
+	 * divider: the hsm_clock rate and this divider setting will
+	 * give a 40 kHz CEC clock.
+	 */
+	value |= VC4_HDMI_CEC_ADDR_MASK |
+		 (4091 << VC4_HDMI_CEC_DIV_CLK_CNT_SHIFT);
+	HDMI_WRITE(VC4_HDMI_CEC_CNTRL_1, value);
+	ret = devm_request_threaded_irq(dev, platform_get_irq(pdev, 0),
+					vc4_cec_irq_handler,
+					vc4_cec_irq_handler_thread, 0,
+					"vc4 hdmi cec", vc4);
+	if (ret)
+		goto err_delete_cec_adap;
+	ret = cec_register_adapter(hdmi->cec_adap, dev);
+	if (ret < 0)
+		goto err_delete_cec_adap;
+#endif
+
+	ret = vc4_hdmi_audio_init(hdmi);
+	if (ret)
+		goto err_destroy_encoder;
+
+	return 0;
+
+#ifdef CONFIG_DRM_VC4_HDMI_CEC
+err_delete_cec_adap:
+	cec_delete_adapter(hdmi->cec_adap);
+err_destroy_conn:
+	vc4_hdmi_connector_destroy(hdmi->connector);
+#endif
+err_destroy_encoder:
+	vc4_hdmi_encoder_destroy(hdmi->encoder);
+err_unprepare_hsm:
+	clk_disable_unprepare(hdmi->hsm_clock);
+	pm_runtime_disable(dev);
+err_put_i2c:
+	put_device(&hdmi->ddc->dev);
+
+	return ret;
+}
+
+static void vc4_hdmi_unbind(struct device *dev, struct device *master,
+			    void *data)
+{
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = drm->dev_private;
+	struct vc4_hdmi *hdmi = vc4->hdmi;
+
+	cec_unregister_adapter(hdmi->cec_adap);
+	vc4_hdmi_connector_destroy(hdmi->connector);
+	vc4_hdmi_encoder_destroy(hdmi->encoder);
+
+	clk_disable_unprepare(hdmi->hsm_clock);
+	pm_runtime_disable(dev);
+
+	put_device(&hdmi->ddc->dev);
+
+	vc4->hdmi = NULL;
+}
+
+static const struct component_ops vc4_hdmi_ops = {
+	.bind   = vc4_hdmi_bind,
+	.unbind = vc4_hdmi_unbind,
+};
+
+static int vc4_hdmi_dev_probe(struct platform_device *pdev)
+{
+	return component_add(&pdev->dev, &vc4_hdmi_ops);
+}
+
+static int vc4_hdmi_dev_remove(struct platform_device *pdev)
+{
+	component_del(&pdev->dev, &vc4_hdmi_ops);
+	return 0;
+}
+
+static const struct of_device_id vc4_hdmi_dt_match[] = {
+	{ .compatible = "brcm,bcm2835-hdmi" },
+	{}
+};
+
+struct platform_driver vc4_hdmi_driver = {
+	.probe = vc4_hdmi_dev_probe,
+	.remove = vc4_hdmi_dev_remove,
+	.driver = {
+		.name = "vc4_hdmi",
+		.of_match_table = vc4_hdmi_dt_match,
+	},
+};
diff --git a/drivers/gpu/drm/vc4/vc4_hvs.c b/drivers/gpu/drm/vc4/vc4_hvs.c
new file mode 100644
index 000000000..5d8c749c9
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_hvs.c
@@ -0,0 +1,277 @@
+/*
+ * Copyright (C) 2015 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+/**
+ * DOC: VC4 HVS module.
+ *
+ * The Hardware Video Scaler (HVS) is the piece of hardware that does
+ * translation, scaling, colorspace conversion, and compositing of
+ * pixels stored in framebuffers into a FIFO of pixels going out to
+ * the Pixel Valve (CRTC).  It operates at the system clock rate (the
+ * system audio clock gate, specifically), which is much higher than
+ * the pixel clock rate.
+ *
+ * There is a single global HVS, with multiple output FIFOs that can
+ * be consumed by the PVs.  This file just manages the resources for
+ * the HVS, while the vc4_crtc.c code actually drives HVS setup for
+ * each CRTC.
+ */
+
+#include <linux/component.h>
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+#define HVS_REG(reg) { reg, #reg }
+static const struct {
+	u32 reg;
+	const char *name;
+} hvs_regs[] = {
+	HVS_REG(SCALER_DISPCTRL),
+	HVS_REG(SCALER_DISPSTAT),
+	HVS_REG(SCALER_DISPID),
+	HVS_REG(SCALER_DISPECTRL),
+	HVS_REG(SCALER_DISPPROF),
+	HVS_REG(SCALER_DISPDITHER),
+	HVS_REG(SCALER_DISPEOLN),
+	HVS_REG(SCALER_DISPLIST0),
+	HVS_REG(SCALER_DISPLIST1),
+	HVS_REG(SCALER_DISPLIST2),
+	HVS_REG(SCALER_DISPLSTAT),
+	HVS_REG(SCALER_DISPLACT0),
+	HVS_REG(SCALER_DISPLACT1),
+	HVS_REG(SCALER_DISPLACT2),
+	HVS_REG(SCALER_DISPCTRL0),
+	HVS_REG(SCALER_DISPBKGND0),
+	HVS_REG(SCALER_DISPSTAT0),
+	HVS_REG(SCALER_DISPBASE0),
+	HVS_REG(SCALER_DISPCTRL1),
+	HVS_REG(SCALER_DISPBKGND1),
+	HVS_REG(SCALER_DISPSTAT1),
+	HVS_REG(SCALER_DISPBASE1),
+	HVS_REG(SCALER_DISPCTRL2),
+	HVS_REG(SCALER_DISPBKGND2),
+	HVS_REG(SCALER_DISPSTAT2),
+	HVS_REG(SCALER_DISPBASE2),
+	HVS_REG(SCALER_DISPALPHA2),
+	HVS_REG(SCALER_OLEDOFFS),
+	HVS_REG(SCALER_OLEDCOEF0),
+	HVS_REG(SCALER_OLEDCOEF1),
+	HVS_REG(SCALER_OLEDCOEF2),
+};
+
+void vc4_hvs_dump_state(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(hvs_regs); i++) {
+		DRM_INFO("0x%04x (%s): 0x%08x\n",
+			 hvs_regs[i].reg, hvs_regs[i].name,
+			 HVS_READ(hvs_regs[i].reg));
+	}
+
+	DRM_INFO("HVS ctx:\n");
+	for (i = 0; i < 64; i += 4) {
+		DRM_INFO("0x%08x (%s): 0x%08x 0x%08x 0x%08x 0x%08x\n",
+			 i * 4, i < HVS_BOOTLOADER_DLIST_END ? "B" : "D",
+			 readl((u32 __iomem *)vc4->hvs->dlist + i + 0),
+			 readl((u32 __iomem *)vc4->hvs->dlist + i + 1),
+			 readl((u32 __iomem *)vc4->hvs->dlist + i + 2),
+			 readl((u32 __iomem *)vc4->hvs->dlist + i + 3));
+	}
+}
+
+#ifdef CONFIG_DEBUG_FS
+int vc4_hvs_debugfs_regs(struct seq_file *m, void *unused)
+{
+	struct drm_info_node *node = (struct drm_info_node *)m->private;
+	struct drm_device *dev = node->minor->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(hvs_regs); i++) {
+		seq_printf(m, "%s (0x%04x): 0x%08x\n",
+			   hvs_regs[i].name, hvs_regs[i].reg,
+			   HVS_READ(hvs_regs[i].reg));
+	}
+
+	return 0;
+}
+#endif
+
+/* The filter kernel is composed of dwords each containing 3 9-bit
+ * signed integers packed next to each other.
+ */
+#define VC4_INT_TO_COEFF(coeff) (coeff & 0x1ff)
+#define VC4_PPF_FILTER_WORD(c0, c1, c2)				\
+	((((c0) & 0x1ff) << 0) |				\
+	 (((c1) & 0x1ff) << 9) |				\
+	 (((c2) & 0x1ff) << 18))
+
+/* The whole filter kernel is arranged as the coefficients 0-16 going
+ * up, then a pad, then 17-31 going down and reversed within the
+ * dwords.  This means that a linear phase kernel (where it's
+ * symmetrical at the boundary between 15 and 16) has the last 5
+ * dwords matching the first 5, but reversed.
+ */
+#define VC4_LINEAR_PHASE_KERNEL(c0, c1, c2, c3, c4, c5, c6, c7, c8,	\
+				c9, c10, c11, c12, c13, c14, c15)	\
+	{VC4_PPF_FILTER_WORD(c0, c1, c2),				\
+	 VC4_PPF_FILTER_WORD(c3, c4, c5),				\
+	 VC4_PPF_FILTER_WORD(c6, c7, c8),				\
+	 VC4_PPF_FILTER_WORD(c9, c10, c11),				\
+	 VC4_PPF_FILTER_WORD(c12, c13, c14),				\
+	 VC4_PPF_FILTER_WORD(c15, c15, 0)}
+
+#define VC4_LINEAR_PHASE_KERNEL_DWORDS 6
+#define VC4_KERNEL_DWORDS (VC4_LINEAR_PHASE_KERNEL_DWORDS * 2 - 1)
+
+/* Recommended B=1/3, C=1/3 filter choice from Mitchell/Netravali.
+ * http://www.cs.utexas.edu/~fussell/courses/cs384g/lectures/mitchell/Mitchell.pdf
+ */
+static const u32 mitchell_netravali_1_3_1_3_kernel[] =
+	VC4_LINEAR_PHASE_KERNEL(0, -2, -6, -8, -10, -8, -3, 2, 18,
+				50, 82, 119, 155, 187, 213, 227);
+
+static int vc4_hvs_upload_linear_kernel(struct vc4_hvs *hvs,
+					struct drm_mm_node *space,
+					const u32 *kernel)
+{
+	int ret, i;
+	u32 __iomem *dst_kernel;
+
+	ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS);
+	if (ret) {
+		DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
+			  ret);
+		return ret;
+	}
+
+	dst_kernel = hvs->dlist + space->start;
+
+	for (i = 0; i < VC4_KERNEL_DWORDS; i++) {
+		if (i < VC4_LINEAR_PHASE_KERNEL_DWORDS)
+			writel(kernel[i], &dst_kernel[i]);
+		else {
+			writel(kernel[VC4_KERNEL_DWORDS - i - 1],
+			       &dst_kernel[i]);
+		}
+	}
+
+	return 0;
+}
+
+static int vc4_hvs_bind(struct device *dev, struct device *master, void *data)
+{
+	struct platform_device *pdev = to_platform_device(dev);
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = drm->dev_private;
+	struct vc4_hvs *hvs = NULL;
+	int ret;
+	u32 dispctrl;
+
+	hvs = devm_kzalloc(&pdev->dev, sizeof(*hvs), GFP_KERNEL);
+	if (!hvs)
+		return -ENOMEM;
+
+	hvs->pdev = pdev;
+
+	hvs->regs = vc4_ioremap_regs(pdev, 0);
+	if (IS_ERR(hvs->regs))
+		return PTR_ERR(hvs->regs);
+
+	hvs->dlist = hvs->regs + SCALER_DLIST_START;
+
+	spin_lock_init(&hvs->mm_lock);
+
+	/* Set up the HVS display list memory manager.  We never
+	 * overwrite the setup from the bootloader (just 128b out of
+	 * our 16K), since we don't want to scramble the screen when
+	 * transitioning from the firmware's boot setup to runtime.
+	 */
+	drm_mm_init(&hvs->dlist_mm,
+		    HVS_BOOTLOADER_DLIST_END,
+		    (SCALER_DLIST_SIZE >> 2) - HVS_BOOTLOADER_DLIST_END);
+
+	/* Set up the HVS LBM memory manager.  We could have some more
+	 * complicated data structure that allowed reuse of LBM areas
+	 * between planes when they don't overlap on the screen, but
+	 * for now we just allocate globally.
+	 */
+	drm_mm_init(&hvs->lbm_mm, 0, 96 * 1024);
+
+	/* Upload filter kernels.  We only have the one for now, so we
+	 * keep it around for the lifetime of the driver.
+	 */
+	ret = vc4_hvs_upload_linear_kernel(hvs,
+					   &hvs->mitchell_netravali_filter,
+					   mitchell_netravali_1_3_1_3_kernel);
+	if (ret)
+		return ret;
+
+	vc4->hvs = hvs;
+
+	dispctrl = HVS_READ(SCALER_DISPCTRL);
+
+	dispctrl |= SCALER_DISPCTRL_ENABLE;
+
+	/* Set DSP3 (PV1) to use HVS channel 2, which would otherwise
+	 * be unused.
+	 */
+	dispctrl &= ~SCALER_DISPCTRL_DSP3_MUX_MASK;
+	dispctrl |= VC4_SET_FIELD(2, SCALER_DISPCTRL_DSP3_MUX);
+
+	HVS_WRITE(SCALER_DISPCTRL, dispctrl);
+
+	return 0;
+}
+
+static void vc4_hvs_unbind(struct device *dev, struct device *master,
+			   void *data)
+{
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = drm->dev_private;
+
+	if (vc4->hvs->mitchell_netravali_filter.allocated)
+		drm_mm_remove_node(&vc4->hvs->mitchell_netravali_filter);
+
+	drm_mm_takedown(&vc4->hvs->dlist_mm);
+	drm_mm_takedown(&vc4->hvs->lbm_mm);
+
+	vc4->hvs = NULL;
+}
+
+static const struct component_ops vc4_hvs_ops = {
+	.bind   = vc4_hvs_bind,
+	.unbind = vc4_hvs_unbind,
+};
+
+static int vc4_hvs_dev_probe(struct platform_device *pdev)
+{
+	return component_add(&pdev->dev, &vc4_hvs_ops);
+}
+
+static int vc4_hvs_dev_remove(struct platform_device *pdev)
+{
+	component_del(&pdev->dev, &vc4_hvs_ops);
+	return 0;
+}
+
+static const struct of_device_id vc4_hvs_dt_match[] = {
+	{ .compatible = "brcm,bcm2835-hvs" },
+	{}
+};
+
+struct platform_driver vc4_hvs_driver = {
+	.probe = vc4_hvs_dev_probe,
+	.remove = vc4_hvs_dev_remove,
+	.driver = {
+		.name = "vc4_hvs",
+		.of_match_table = vc4_hvs_dt_match,
+	},
+};
diff --git a/drivers/gpu/drm/vc4/vc4_irq.c b/drivers/gpu/drm/vc4/vc4_irq.c
new file mode 100644
index 000000000..4cd2ccfe1
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_irq.c
@@ -0,0 +1,290 @@
+/*
+ * Copyright © 2014 Broadcom
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+/**
+ * DOC: Interrupt management for the V3D engine
+ *
+ * We have an interrupt status register (V3D_INTCTL) which reports
+ * interrupts, and where writing 1 bits clears those interrupts.
+ * There are also a pair of interrupt registers
+ * (V3D_INTENA/V3D_INTDIS) where writing a 1 to their bits enables or
+ * disables that specific interrupt, and 0s written are ignored
+ * (reading either one returns the set of enabled interrupts).
+ *
+ * When we take a binning flush done interrupt, we need to submit the
+ * next frame for binning and move the finished frame to the render
+ * thread.
+ *
+ * When we take a render frame interrupt, we need to wake the
+ * processes waiting for some frame to be done, and get the next frame
+ * submitted ASAP (so the hardware doesn't sit idle when there's work
+ * to do).
+ *
+ * When we take the binner out of memory interrupt, we need to
+ * allocate some new memory and pass it to the binner so that the
+ * current job can make progress.
+ */
+
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+#define V3D_DRIVER_IRQS (V3D_INT_OUTOMEM | \
+			 V3D_INT_FLDONE | \
+			 V3D_INT_FRDONE)
+
+DECLARE_WAIT_QUEUE_HEAD(render_wait);
+
+static void
+vc4_overflow_mem_work(struct work_struct *work)
+{
+	struct vc4_dev *vc4 =
+		container_of(work, struct vc4_dev, overflow_mem_work);
+	struct vc4_bo *bo = vc4->bin_bo;
+	int bin_bo_slot;
+	struct vc4_exec_info *exec;
+	unsigned long irqflags;
+
+	bin_bo_slot = vc4_v3d_get_bin_slot(vc4);
+	if (bin_bo_slot < 0) {
+		DRM_ERROR("Couldn't allocate binner overflow mem\n");
+		return;
+	}
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+
+	if (vc4->bin_alloc_overflow) {
+		/* If we had overflow memory allocated previously,
+		 * then that chunk will free when the current bin job
+		 * is done.  If we don't have a bin job running, then
+		 * the chunk will be done whenever the list of render
+		 * jobs has drained.
+		 */
+		exec = vc4_first_bin_job(vc4);
+		if (!exec)
+			exec = vc4_last_render_job(vc4);
+		if (exec) {
+			exec->bin_slots |= vc4->bin_alloc_overflow;
+		} else {
+			/* There's nothing queued in the hardware, so
+			 * the old slot is free immediately.
+			 */
+			vc4->bin_alloc_used &= ~vc4->bin_alloc_overflow;
+		}
+	}
+	vc4->bin_alloc_overflow = BIT(bin_bo_slot);
+
+	V3D_WRITE(V3D_BPOA, bo->base.paddr + bin_bo_slot * vc4->bin_alloc_size);
+	V3D_WRITE(V3D_BPOS, bo->base.base.size);
+	V3D_WRITE(V3D_INTCTL, V3D_INT_OUTOMEM);
+	V3D_WRITE(V3D_INTENA, V3D_INT_OUTOMEM);
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+}
+
+static void
+vc4_irq_finish_bin_job(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_exec_info *next, *exec = vc4_first_bin_job(vc4);
+
+	if (!exec)
+		return;
+
+	vc4_move_job_to_render(dev, exec);
+	next = vc4_first_bin_job(vc4);
+
+	/* Only submit the next job in the bin list if it matches the perfmon
+	 * attached to the one that just finished (or if both jobs don't have
+	 * perfmon attached to them).
+	 */
+	if (next && next->perfmon == exec->perfmon)
+		vc4_submit_next_bin_job(dev);
+}
+
+static void
+vc4_cancel_bin_job(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_exec_info *exec = vc4_first_bin_job(vc4);
+
+	if (!exec)
+		return;
+
+	/* Stop the perfmon so that the next bin job can be started. */
+	if (exec->perfmon)
+		vc4_perfmon_stop(vc4, exec->perfmon, false);
+
+	list_move_tail(&exec->head, &vc4->bin_job_list);
+	vc4_submit_next_bin_job(dev);
+}
+
+static void
+vc4_irq_finish_render_job(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_exec_info *exec = vc4_first_render_job(vc4);
+	struct vc4_exec_info *nextbin, *nextrender;
+
+	if (!exec)
+		return;
+
+	vc4->finished_seqno++;
+	list_move_tail(&exec->head, &vc4->job_done_list);
+
+	nextbin = vc4_first_bin_job(vc4);
+	nextrender = vc4_first_render_job(vc4);
+
+	/* Only stop the perfmon if following jobs in the queue don't expect it
+	 * to be enabled.
+	 */
+	if (exec->perfmon && !nextrender &&
+	    (!nextbin || nextbin->perfmon != exec->perfmon))
+		vc4_perfmon_stop(vc4, exec->perfmon, true);
+
+	/* If there's a render job waiting, start it. If this is not the case
+	 * we may have to unblock the binner if it's been stalled because of
+	 * perfmon (this can be checked by comparing the perfmon attached to
+	 * the finished renderjob to the one attached to the next bin job: if
+	 * they don't match, this means the binner is stalled and should be
+	 * restarted).
+	 */
+	if (nextrender)
+		vc4_submit_next_render_job(dev);
+	else if (nextbin && nextbin->perfmon != exec->perfmon)
+		vc4_submit_next_bin_job(dev);
+
+	if (exec->fence) {
+		dma_fence_signal_locked(exec->fence);
+		dma_fence_put(exec->fence);
+		exec->fence = NULL;
+	}
+
+	wake_up_all(&vc4->job_wait_queue);
+	schedule_work(&vc4->job_done_work);
+}
+
+irqreturn_t
+vc4_irq(int irq, void *arg)
+{
+	struct drm_device *dev = arg;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	uint32_t intctl;
+	irqreturn_t status = IRQ_NONE;
+
+	barrier();
+	intctl = V3D_READ(V3D_INTCTL);
+
+	/* Acknowledge the interrupts we're handling here. The binner
+	 * last flush / render frame done interrupt will be cleared,
+	 * while OUTOMEM will stay high until the underlying cause is
+	 * cleared.
+	 */
+	V3D_WRITE(V3D_INTCTL, intctl);
+
+	if (intctl & V3D_INT_OUTOMEM) {
+		/* Disable OUTOMEM until the work is done. */
+		V3D_WRITE(V3D_INTDIS, V3D_INT_OUTOMEM);
+		schedule_work(&vc4->overflow_mem_work);
+		status = IRQ_HANDLED;
+	}
+
+	if (intctl & V3D_INT_FLDONE) {
+		spin_lock(&vc4->job_lock);
+		vc4_irq_finish_bin_job(dev);
+		spin_unlock(&vc4->job_lock);
+		status = IRQ_HANDLED;
+	}
+
+	if (intctl & V3D_INT_FRDONE) {
+		spin_lock(&vc4->job_lock);
+		vc4_irq_finish_render_job(dev);
+		spin_unlock(&vc4->job_lock);
+		status = IRQ_HANDLED;
+	}
+
+	return status;
+}
+
+void
+vc4_irq_preinstall(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	init_waitqueue_head(&vc4->job_wait_queue);
+	INIT_WORK(&vc4->overflow_mem_work, vc4_overflow_mem_work);
+
+	/* Clear any pending interrupts someone might have left around
+	 * for us.
+	 */
+	V3D_WRITE(V3D_INTCTL, V3D_DRIVER_IRQS);
+}
+
+int
+vc4_irq_postinstall(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	/* Enable both the render done and out of memory interrupts. */
+	V3D_WRITE(V3D_INTENA, V3D_DRIVER_IRQS);
+
+	return 0;
+}
+
+void
+vc4_irq_uninstall(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	/* Disable sending interrupts for our driver's IRQs. */
+	V3D_WRITE(V3D_INTDIS, V3D_DRIVER_IRQS);
+
+	/* Clear any pending interrupts we might have left. */
+	V3D_WRITE(V3D_INTCTL, V3D_DRIVER_IRQS);
+
+	/* Finish any interrupt handler still in flight. */
+	disable_irq(dev->irq);
+
+	cancel_work_sync(&vc4->overflow_mem_work);
+}
+
+/** Reinitializes interrupt registers when a GPU reset is performed. */
+void vc4_irq_reset(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	unsigned long irqflags;
+
+	/* Acknowledge any stale IRQs. */
+	V3D_WRITE(V3D_INTCTL, V3D_DRIVER_IRQS);
+
+	/*
+	 * Turn all our interrupts on.  Binner out of memory is the
+	 * only one we expect to trigger at this point, since we've
+	 * just come from poweron and haven't supplied any overflow
+	 * memory yet.
+	 */
+	V3D_WRITE(V3D_INTENA, V3D_DRIVER_IRQS);
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	vc4_cancel_bin_job(dev);
+	vc4_irq_finish_render_job(dev);
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+}
diff --git a/drivers/gpu/drm/vc4/vc4_kms.c b/drivers/gpu/drm/vc4/vc4_kms.c
new file mode 100644
index 000000000..f4d8a730e
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_kms.c
@@ -0,0 +1,449 @@
+/*
+ * Copyright (C) 2015 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+/**
+ * DOC: VC4 KMS
+ *
+ * This is the general code for implementing KMS mode setting that
+ * doesn't clearly associate with any of the other objects (plane,
+ * crtc, HDMI encoder).
+ */
+
+#include <drm/drm_crtc.h>
+#include <drm/drm_atomic.h>
+#include <drm/drm_atomic_helper.h>
+#include <drm/drm_crtc_helper.h>
+#include <drm/drm_plane_helper.h>
+#include <drm/drm_fb_helper.h>
+#include <drm/drm_fb_cma_helper.h>
+#include <drm/drm_gem_framebuffer_helper.h>
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+struct vc4_ctm_state {
+	struct drm_private_state base;
+	struct drm_color_ctm *ctm;
+	int fifo;
+};
+
+static struct vc4_ctm_state *to_vc4_ctm_state(struct drm_private_state *priv)
+{
+	return container_of(priv, struct vc4_ctm_state, base);
+}
+
+static struct vc4_ctm_state *vc4_get_ctm_state(struct drm_atomic_state *state,
+					       struct drm_private_obj *manager)
+{
+	struct drm_device *dev = state->dev;
+	struct vc4_dev *vc4 = dev->dev_private;
+	struct drm_private_state *priv_state;
+	int ret;
+
+	ret = drm_modeset_lock(&vc4->ctm_state_lock, state->acquire_ctx);
+	if (ret)
+		return ERR_PTR(ret);
+
+	priv_state = drm_atomic_get_private_obj_state(state, manager);
+	if (IS_ERR(priv_state))
+		return ERR_CAST(priv_state);
+
+	return to_vc4_ctm_state(priv_state);
+}
+
+static struct drm_private_state *
+vc4_ctm_duplicate_state(struct drm_private_obj *obj)
+{
+	struct vc4_ctm_state *state;
+
+	state = kmemdup(obj->state, sizeof(*state), GFP_KERNEL);
+	if (!state)
+		return NULL;
+
+	__drm_atomic_helper_private_obj_duplicate_state(obj, &state->base);
+
+	return &state->base;
+}
+
+static void vc4_ctm_destroy_state(struct drm_private_obj *obj,
+				  struct drm_private_state *state)
+{
+	struct vc4_ctm_state *ctm_state = to_vc4_ctm_state(state);
+
+	kfree(ctm_state);
+}
+
+static const struct drm_private_state_funcs vc4_ctm_state_funcs = {
+	.atomic_duplicate_state = vc4_ctm_duplicate_state,
+	.atomic_destroy_state = vc4_ctm_destroy_state,
+};
+
+/* Converts a DRM S31.32 value to the HW S0.9 format. */
+static u16 vc4_ctm_s31_32_to_s0_9(u64 in)
+{
+	u16 r;
+
+	/* Sign bit. */
+	r = in & BIT_ULL(63) ? BIT(9) : 0;
+
+	if ((in & GENMASK_ULL(62, 32)) > 0) {
+		/* We have zero integer bits so we can only saturate here. */
+		r |= GENMASK(8, 0);
+	} else {
+		/* Otherwise take the 9 most important fractional bits. */
+		r |= (in >> 23) & GENMASK(8, 0);
+	}
+
+	return r;
+}
+
+static void
+vc4_ctm_commit(struct vc4_dev *vc4, struct drm_atomic_state *state)
+{
+	struct vc4_ctm_state *ctm_state = to_vc4_ctm_state(vc4->ctm_manager.state);
+	struct drm_color_ctm *ctm = ctm_state->ctm;
+
+	if (ctm_state->fifo) {
+		HVS_WRITE(SCALER_OLEDCOEF2,
+			  VC4_SET_FIELD(vc4_ctm_s31_32_to_s0_9(ctm->matrix[0]),
+					SCALER_OLEDCOEF2_R_TO_R) |
+			  VC4_SET_FIELD(vc4_ctm_s31_32_to_s0_9(ctm->matrix[3]),
+					SCALER_OLEDCOEF2_R_TO_G) |
+			  VC4_SET_FIELD(vc4_ctm_s31_32_to_s0_9(ctm->matrix[6]),
+					SCALER_OLEDCOEF2_R_TO_B));
+		HVS_WRITE(SCALER_OLEDCOEF1,
+			  VC4_SET_FIELD(vc4_ctm_s31_32_to_s0_9(ctm->matrix[1]),
+					SCALER_OLEDCOEF1_G_TO_R) |
+			  VC4_SET_FIELD(vc4_ctm_s31_32_to_s0_9(ctm->matrix[4]),
+					SCALER_OLEDCOEF1_G_TO_G) |
+			  VC4_SET_FIELD(vc4_ctm_s31_32_to_s0_9(ctm->matrix[7]),
+					SCALER_OLEDCOEF1_G_TO_B));
+		HVS_WRITE(SCALER_OLEDCOEF0,
+			  VC4_SET_FIELD(vc4_ctm_s31_32_to_s0_9(ctm->matrix[2]),
+					SCALER_OLEDCOEF0_B_TO_R) |
+			  VC4_SET_FIELD(vc4_ctm_s31_32_to_s0_9(ctm->matrix[5]),
+					SCALER_OLEDCOEF0_B_TO_G) |
+			  VC4_SET_FIELD(vc4_ctm_s31_32_to_s0_9(ctm->matrix[8]),
+					SCALER_OLEDCOEF0_B_TO_B));
+	}
+
+	HVS_WRITE(SCALER_OLEDOFFS,
+		  VC4_SET_FIELD(ctm_state->fifo, SCALER_OLEDOFFS_DISPFIFO));
+}
+
+static void
+vc4_atomic_complete_commit(struct drm_atomic_state *state)
+{
+	struct drm_device *dev = state->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	drm_atomic_helper_wait_for_fences(dev, state, false);
+
+	drm_atomic_helper_wait_for_dependencies(state);
+
+	drm_atomic_helper_commit_modeset_disables(dev, state);
+
+	vc4_ctm_commit(vc4, state);
+
+	drm_atomic_helper_commit_planes(dev, state, 0);
+
+	drm_atomic_helper_commit_modeset_enables(dev, state);
+
+	drm_atomic_helper_fake_vblank(state);
+
+	drm_atomic_helper_commit_hw_done(state);
+
+	drm_atomic_helper_wait_for_flip_done(dev, state);
+
+	drm_atomic_helper_cleanup_planes(dev, state);
+
+	drm_atomic_helper_commit_cleanup_done(state);
+
+	drm_atomic_state_put(state);
+
+	up(&vc4->async_modeset);
+}
+
+static void commit_work(struct work_struct *work)
+{
+	struct drm_atomic_state *state = container_of(work,
+						      struct drm_atomic_state,
+						      commit_work);
+	vc4_atomic_complete_commit(state);
+}
+
+/**
+ * vc4_atomic_commit - commit validated state object
+ * @dev: DRM device
+ * @state: the driver state object
+ * @nonblock: nonblocking commit
+ *
+ * This function commits a with drm_atomic_helper_check() pre-validated state
+ * object. This can still fail when e.g. the framebuffer reservation fails. For
+ * now this doesn't implement asynchronous commits.
+ *
+ * RETURNS
+ * Zero for success or -errno.
+ */
+static int vc4_atomic_commit(struct drm_device *dev,
+			     struct drm_atomic_state *state,
+			     bool nonblock)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int ret;
+
+	if (state->async_update) {
+		ret = down_interruptible(&vc4->async_modeset);
+		if (ret)
+			return ret;
+
+		ret = drm_atomic_helper_prepare_planes(dev, state);
+		if (ret) {
+			up(&vc4->async_modeset);
+			return ret;
+		}
+
+		drm_atomic_helper_async_commit(dev, state);
+
+		drm_atomic_helper_cleanup_planes(dev, state);
+
+		up(&vc4->async_modeset);
+
+		return 0;
+	}
+
+	/* We know for sure we don't want an async update here. Set
+	 * state->legacy_cursor_update to false to prevent
+	 * drm_atomic_helper_setup_commit() from auto-completing
+	 * commit->flip_done.
+	 */
+	state->legacy_cursor_update = false;
+	ret = drm_atomic_helper_setup_commit(state, nonblock);
+	if (ret)
+		return ret;
+
+	INIT_WORK(&state->commit_work, commit_work);
+
+	ret = down_interruptible(&vc4->async_modeset);
+	if (ret)
+		return ret;
+
+	ret = drm_atomic_helper_prepare_planes(dev, state);
+	if (ret) {
+		up(&vc4->async_modeset);
+		return ret;
+	}
+
+	if (!nonblock) {
+		ret = drm_atomic_helper_wait_for_fences(dev, state, true);
+		if (ret) {
+			drm_atomic_helper_cleanup_planes(dev, state);
+			up(&vc4->async_modeset);
+			return ret;
+		}
+	}
+
+	/*
+	 * This is the point of no return - everything below never fails except
+	 * when the hw goes bonghits. Which means we can commit the new state on
+	 * the software side now.
+	 */
+
+	BUG_ON(drm_atomic_helper_swap_state(state, false) < 0);
+
+	/*
+	 * Everything below can be run asynchronously without the need to grab
+	 * any modeset locks at all under one condition: It must be guaranteed
+	 * that the asynchronous work has either been cancelled (if the driver
+	 * supports it, which at least requires that the framebuffers get
+	 * cleaned up with drm_atomic_helper_cleanup_planes()) or completed
+	 * before the new state gets committed on the software side with
+	 * drm_atomic_helper_swap_state().
+	 *
+	 * This scheme allows new atomic state updates to be prepared and
+	 * checked in parallel to the asynchronous completion of the previous
+	 * update. Which is important since compositors need to figure out the
+	 * composition of the next frame right after having submitted the
+	 * current layout.
+	 */
+
+	drm_atomic_state_get(state);
+	if (nonblock)
+		queue_work(system_unbound_wq, &state->commit_work);
+	else
+		vc4_atomic_complete_commit(state);
+
+	return 0;
+}
+
+static struct drm_framebuffer *vc4_fb_create(struct drm_device *dev,
+					     struct drm_file *file_priv,
+					     const struct drm_mode_fb_cmd2 *mode_cmd)
+{
+	struct drm_mode_fb_cmd2 mode_cmd_local;
+
+	/* If the user didn't specify a modifier, use the
+	 * vc4_set_tiling_ioctl() state for the BO.
+	 */
+	if (!(mode_cmd->flags & DRM_MODE_FB_MODIFIERS)) {
+		struct drm_gem_object *gem_obj;
+		struct vc4_bo *bo;
+
+		gem_obj = drm_gem_object_lookup(file_priv,
+						mode_cmd->handles[0]);
+		if (!gem_obj) {
+			DRM_DEBUG("Failed to look up GEM BO %d\n",
+				  mode_cmd->handles[0]);
+			return ERR_PTR(-ENOENT);
+		}
+		bo = to_vc4_bo(gem_obj);
+
+		mode_cmd_local = *mode_cmd;
+
+		if (bo->t_format) {
+			mode_cmd_local.modifier[0] =
+				DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED;
+		} else {
+			mode_cmd_local.modifier[0] = DRM_FORMAT_MOD_NONE;
+		}
+
+		drm_gem_object_put_unlocked(gem_obj);
+
+		mode_cmd = &mode_cmd_local;
+	}
+
+	return drm_gem_fb_create(dev, file_priv, mode_cmd);
+}
+
+/* Our CTM has some peculiar limitations: we can only enable it for one CRTC
+ * at a time and the HW only supports S0.9 scalars. To account for the latter,
+ * we don't allow userland to set a CTM that we have no hope of approximating.
+ */
+static int
+vc4_ctm_atomic_check(struct drm_device *dev, struct drm_atomic_state *state)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_ctm_state *ctm_state = NULL;
+	struct drm_crtc *crtc;
+	struct drm_crtc_state *old_crtc_state, *new_crtc_state;
+	struct drm_color_ctm *ctm;
+	int i;
+
+	for_each_oldnew_crtc_in_state(state, crtc, old_crtc_state, new_crtc_state, i) {
+		/* CTM is being disabled. */
+		if (!new_crtc_state->ctm && old_crtc_state->ctm) {
+			ctm_state = vc4_get_ctm_state(state, &vc4->ctm_manager);
+			if (IS_ERR(ctm_state))
+				return PTR_ERR(ctm_state);
+			ctm_state->fifo = 0;
+		}
+	}
+
+	for_each_oldnew_crtc_in_state(state, crtc, old_crtc_state, new_crtc_state, i) {
+		if (new_crtc_state->ctm == old_crtc_state->ctm)
+			continue;
+
+		if (!ctm_state) {
+			ctm_state = vc4_get_ctm_state(state, &vc4->ctm_manager);
+			if (IS_ERR(ctm_state))
+				return PTR_ERR(ctm_state);
+		}
+
+		/* CTM is being enabled or the matrix changed. */
+		if (new_crtc_state->ctm) {
+			/* fifo is 1-based since 0 disables CTM. */
+			int fifo = to_vc4_crtc(crtc)->channel + 1;
+
+			/* Check userland isn't trying to turn on CTM for more
+			 * than one CRTC at a time.
+			 */
+			if (ctm_state->fifo && ctm_state->fifo != fifo) {
+				DRM_DEBUG_DRIVER("Too many CTM configured\n");
+				return -EINVAL;
+			}
+
+			/* Check we can approximate the specified CTM.
+			 * We disallow scalars |c| > 1.0 since the HW has
+			 * no integer bits.
+			 */
+			ctm = new_crtc_state->ctm->data;
+			for (i = 0; i < ARRAY_SIZE(ctm->matrix); i++) {
+				u64 val = ctm->matrix[i];
+
+				val &= ~BIT_ULL(63);
+				if (val > BIT_ULL(32))
+					return -EINVAL;
+			}
+
+			ctm_state->fifo = fifo;
+			ctm_state->ctm = ctm;
+		}
+	}
+
+	return 0;
+}
+
+static int
+vc4_atomic_check(struct drm_device *dev, struct drm_atomic_state *state)
+{
+	int ret;
+
+	ret = vc4_ctm_atomic_check(dev, state);
+	if (ret < 0)
+		return ret;
+
+	return drm_atomic_helper_check(dev, state);
+}
+
+static const struct drm_mode_config_funcs vc4_mode_funcs = {
+	.output_poll_changed = drm_fb_helper_output_poll_changed,
+	.atomic_check = vc4_atomic_check,
+	.atomic_commit = vc4_atomic_commit,
+	.fb_create = vc4_fb_create,
+};
+
+int vc4_kms_load(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_ctm_state *ctm_state;
+	int ret;
+
+	sema_init(&vc4->async_modeset, 1);
+
+	/* Set support for vblank irq fast disable, before drm_vblank_init() */
+	dev->vblank_disable_immediate = true;
+
+	ret = drm_vblank_init(dev, dev->mode_config.num_crtc);
+	if (ret < 0) {
+		dev_err(dev->dev, "failed to initialize vblank\n");
+		return ret;
+	}
+
+	dev->mode_config.max_width = 2048;
+	dev->mode_config.max_height = 2048;
+	dev->mode_config.funcs = &vc4_mode_funcs;
+	dev->mode_config.preferred_depth = 24;
+	dev->mode_config.async_page_flip = true;
+	dev->mode_config.allow_fb_modifiers = true;
+
+	drm_modeset_lock_init(&vc4->ctm_state_lock);
+
+	ctm_state = kzalloc(sizeof(*ctm_state), GFP_KERNEL);
+	if (!ctm_state)
+		return -ENOMEM;
+	drm_atomic_private_obj_init(&vc4->ctm_manager, &ctm_state->base,
+				    &vc4_ctm_state_funcs);
+
+	drm_mode_config_reset(dev);
+
+	if (dev->mode_config.num_connector)
+		drm_fb_cma_fbdev_init(dev, 32, 0);
+
+	drm_kms_helper_poll_init(dev);
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/vc4/vc4_packet.h b/drivers/gpu/drm/vc4/vc4_packet.h
new file mode 100644
index 000000000..0f31cc065
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_packet.h
@@ -0,0 +1,399 @@
+/*
+ * Copyright © 2014 Broadcom
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#ifndef VC4_PACKET_H
+#define VC4_PACKET_H
+
+#include "vc4_regs.h" /* for VC4_MASK, VC4_GET_FIELD, VC4_SET_FIELD */
+
+enum vc4_packet {
+	VC4_PACKET_HALT = 0,
+	VC4_PACKET_NOP = 1,
+
+	VC4_PACKET_FLUSH = 4,
+	VC4_PACKET_FLUSH_ALL = 5,
+	VC4_PACKET_START_TILE_BINNING = 6,
+	VC4_PACKET_INCREMENT_SEMAPHORE = 7,
+	VC4_PACKET_WAIT_ON_SEMAPHORE = 8,
+
+	VC4_PACKET_BRANCH = 16,
+	VC4_PACKET_BRANCH_TO_SUB_LIST = 17,
+
+	VC4_PACKET_STORE_MS_TILE_BUFFER = 24,
+	VC4_PACKET_STORE_MS_TILE_BUFFER_AND_EOF = 25,
+	VC4_PACKET_STORE_FULL_RES_TILE_BUFFER = 26,
+	VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER = 27,
+	VC4_PACKET_STORE_TILE_BUFFER_GENERAL = 28,
+	VC4_PACKET_LOAD_TILE_BUFFER_GENERAL = 29,
+
+	VC4_PACKET_GL_INDEXED_PRIMITIVE = 32,
+	VC4_PACKET_GL_ARRAY_PRIMITIVE = 33,
+
+	VC4_PACKET_COMPRESSED_PRIMITIVE = 48,
+	VC4_PACKET_CLIPPED_COMPRESSED_PRIMITIVE = 49,
+
+	VC4_PACKET_PRIMITIVE_LIST_FORMAT = 56,
+
+	VC4_PACKET_GL_SHADER_STATE = 64,
+	VC4_PACKET_NV_SHADER_STATE = 65,
+	VC4_PACKET_VG_SHADER_STATE = 66,
+
+	VC4_PACKET_CONFIGURATION_BITS = 96,
+	VC4_PACKET_FLAT_SHADE_FLAGS = 97,
+	VC4_PACKET_POINT_SIZE = 98,
+	VC4_PACKET_LINE_WIDTH = 99,
+	VC4_PACKET_RHT_X_BOUNDARY = 100,
+	VC4_PACKET_DEPTH_OFFSET = 101,
+	VC4_PACKET_CLIP_WINDOW = 102,
+	VC4_PACKET_VIEWPORT_OFFSET = 103,
+	VC4_PACKET_Z_CLIPPING = 104,
+	VC4_PACKET_CLIPPER_XY_SCALING = 105,
+	VC4_PACKET_CLIPPER_Z_SCALING = 106,
+
+	VC4_PACKET_TILE_BINNING_MODE_CONFIG = 112,
+	VC4_PACKET_TILE_RENDERING_MODE_CONFIG = 113,
+	VC4_PACKET_CLEAR_COLORS = 114,
+	VC4_PACKET_TILE_COORDINATES = 115,
+
+	/* Not an actual hardware packet -- this is what we use to put
+	 * references to GEM bos in the command stream, since we need the u32
+	 * int the actual address packet in order to store the offset from the
+	 * start of the BO.
+	 */
+	VC4_PACKET_GEM_HANDLES = 254,
+} __attribute__ ((__packed__));
+
+#define VC4_PACKET_HALT_SIZE						1
+#define VC4_PACKET_NOP_SIZE						1
+#define VC4_PACKET_FLUSH_SIZE						1
+#define VC4_PACKET_FLUSH_ALL_SIZE					1
+#define VC4_PACKET_START_TILE_BINNING_SIZE				1
+#define VC4_PACKET_INCREMENT_SEMAPHORE_SIZE				1
+#define VC4_PACKET_WAIT_ON_SEMAPHORE_SIZE				1
+#define VC4_PACKET_BRANCH_SIZE						5
+#define VC4_PACKET_BRANCH_TO_SUB_LIST_SIZE				5
+#define VC4_PACKET_STORE_MS_TILE_BUFFER_SIZE				1
+#define VC4_PACKET_STORE_MS_TILE_BUFFER_AND_EOF_SIZE			1
+#define VC4_PACKET_STORE_FULL_RES_TILE_BUFFER_SIZE			5
+#define VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER_SIZE			5
+#define VC4_PACKET_STORE_TILE_BUFFER_GENERAL_SIZE			7
+#define VC4_PACKET_LOAD_TILE_BUFFER_GENERAL_SIZE			7
+#define VC4_PACKET_GL_INDEXED_PRIMITIVE_SIZE				14
+#define VC4_PACKET_GL_ARRAY_PRIMITIVE_SIZE				10
+#define VC4_PACKET_COMPRESSED_PRIMITIVE_SIZE				1
+#define VC4_PACKET_CLIPPED_COMPRESSED_PRIMITIVE_SIZE			1
+#define VC4_PACKET_PRIMITIVE_LIST_FORMAT_SIZE				2
+#define VC4_PACKET_GL_SHADER_STATE_SIZE					5
+#define VC4_PACKET_NV_SHADER_STATE_SIZE					5
+#define VC4_PACKET_VG_SHADER_STATE_SIZE					5
+#define VC4_PACKET_CONFIGURATION_BITS_SIZE				4
+#define VC4_PACKET_FLAT_SHADE_FLAGS_SIZE				5
+#define VC4_PACKET_POINT_SIZE_SIZE					5
+#define VC4_PACKET_LINE_WIDTH_SIZE					5
+#define VC4_PACKET_RHT_X_BOUNDARY_SIZE					3
+#define VC4_PACKET_DEPTH_OFFSET_SIZE					5
+#define VC4_PACKET_CLIP_WINDOW_SIZE					9
+#define VC4_PACKET_VIEWPORT_OFFSET_SIZE					5
+#define VC4_PACKET_Z_CLIPPING_SIZE					9
+#define VC4_PACKET_CLIPPER_XY_SCALING_SIZE				9
+#define VC4_PACKET_CLIPPER_Z_SCALING_SIZE				9
+#define VC4_PACKET_TILE_BINNING_MODE_CONFIG_SIZE			16
+#define VC4_PACKET_TILE_RENDERING_MODE_CONFIG_SIZE			11
+#define VC4_PACKET_CLEAR_COLORS_SIZE					14
+#define VC4_PACKET_TILE_COORDINATES_SIZE				3
+#define VC4_PACKET_GEM_HANDLES_SIZE					9
+
+/* Number of multisamples supported. */
+#define VC4_MAX_SAMPLES							4
+/* Size of a full resolution color or Z tile buffer load/store. */
+#define VC4_TILE_BUFFER_SIZE			(64 * 64 * 4)
+
+/** @{
+ * Bits used by packets like VC4_PACKET_STORE_TILE_BUFFER_GENERAL and
+ * VC4_PACKET_TILE_RENDERING_MODE_CONFIG.
+*/
+#define VC4_TILING_FORMAT_LINEAR    0
+#define VC4_TILING_FORMAT_T         1
+#define VC4_TILING_FORMAT_LT        2
+/** @} */
+
+/** @{
+ *
+ * low bits of VC4_PACKET_STORE_FULL_RES_TILE_BUFFER and
+ * VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER.
+ */
+#define VC4_LOADSTORE_FULL_RES_EOF                     BIT(3)
+#define VC4_LOADSTORE_FULL_RES_DISABLE_CLEAR_ALL       BIT(2)
+#define VC4_LOADSTORE_FULL_RES_DISABLE_ZS              BIT(1)
+#define VC4_LOADSTORE_FULL_RES_DISABLE_COLOR           BIT(0)
+
+/** @{
+ *
+ * low bits of VC4_PACKET_STORE_FULL_RES_TILE_BUFFER and
+ * VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER.
+ */
+#define VC4_LOADSTORE_FULL_RES_EOF                     BIT(3)
+#define VC4_LOADSTORE_FULL_RES_DISABLE_CLEAR_ALL       BIT(2)
+#define VC4_LOADSTORE_FULL_RES_DISABLE_ZS              BIT(1)
+#define VC4_LOADSTORE_FULL_RES_DISABLE_COLOR           BIT(0)
+
+/** @{
+ *
+ * byte 2 of VC4_PACKET_STORE_TILE_BUFFER_GENERAL and
+ * VC4_PACKET_LOAD_TILE_BUFFER_GENERAL (low bits of the address)
+ */
+
+#define VC4_LOADSTORE_TILE_BUFFER_EOF                  BIT(3)
+#define VC4_LOADSTORE_TILE_BUFFER_DISABLE_FULL_VG_MASK BIT(2)
+#define VC4_LOADSTORE_TILE_BUFFER_DISABLE_FULL_ZS      BIT(1)
+#define VC4_LOADSTORE_TILE_BUFFER_DISABLE_FULL_COLOR   BIT(0)
+
+/** @} */
+
+/** @{
+ *
+ * byte 0-1 of VC4_PACKET_STORE_TILE_BUFFER_GENERAL and
+ * VC4_PACKET_LOAD_TILE_BUFFER_GENERAL
+ */
+#define VC4_STORE_TILE_BUFFER_DISABLE_VG_MASK_CLEAR BIT(15)
+#define VC4_STORE_TILE_BUFFER_DISABLE_ZS_CLEAR     BIT(14)
+#define VC4_STORE_TILE_BUFFER_DISABLE_COLOR_CLEAR  BIT(13)
+#define VC4_STORE_TILE_BUFFER_DISABLE_SWAP         BIT(12)
+
+#define VC4_LOADSTORE_TILE_BUFFER_FORMAT_MASK      VC4_MASK(9, 8)
+#define VC4_LOADSTORE_TILE_BUFFER_FORMAT_SHIFT     8
+#define VC4_LOADSTORE_TILE_BUFFER_RGBA8888         0
+#define VC4_LOADSTORE_TILE_BUFFER_BGR565_DITHER    1
+#define VC4_LOADSTORE_TILE_BUFFER_BGR565           2
+/** @} */
+
+/** @{
+ *
+ * byte 0 of VC4_PACKET_STORE_TILE_BUFFER_GENERAL and
+ * VC4_PACKET_LOAD_TILE_BUFFER_GENERAL
+ */
+#define VC4_STORE_TILE_BUFFER_MODE_MASK            VC4_MASK(7, 6)
+#define VC4_STORE_TILE_BUFFER_MODE_SHIFT           6
+#define VC4_STORE_TILE_BUFFER_MODE_SAMPLE0         (0 << 6)
+#define VC4_STORE_TILE_BUFFER_MODE_DECIMATE_X4     (1 << 6)
+#define VC4_STORE_TILE_BUFFER_MODE_DECIMATE_X16    (2 << 6)
+
+/** The values of the field are VC4_TILING_FORMAT_* */
+#define VC4_LOADSTORE_TILE_BUFFER_TILING_MASK      VC4_MASK(5, 4)
+#define VC4_LOADSTORE_TILE_BUFFER_TILING_SHIFT     4
+
+#define VC4_LOADSTORE_TILE_BUFFER_BUFFER_MASK      VC4_MASK(2, 0)
+#define VC4_LOADSTORE_TILE_BUFFER_BUFFER_SHIFT     0
+#define VC4_LOADSTORE_TILE_BUFFER_NONE             0
+#define VC4_LOADSTORE_TILE_BUFFER_COLOR            1
+#define VC4_LOADSTORE_TILE_BUFFER_ZS               2
+#define VC4_LOADSTORE_TILE_BUFFER_Z                3
+#define VC4_LOADSTORE_TILE_BUFFER_VG_MASK          4
+#define VC4_LOADSTORE_TILE_BUFFER_FULL             5
+/** @} */
+
+#define VC4_INDEX_BUFFER_U8                        (0 << 4)
+#define VC4_INDEX_BUFFER_U16                       (1 << 4)
+
+/* This flag is only present in NV shader state. */
+#define VC4_SHADER_FLAG_SHADED_CLIP_COORDS         BIT(3)
+#define VC4_SHADER_FLAG_ENABLE_CLIPPING            BIT(2)
+#define VC4_SHADER_FLAG_VS_POINT_SIZE              BIT(1)
+#define VC4_SHADER_FLAG_FS_SINGLE_THREAD           BIT(0)
+
+/** @{ byte 2 of config bits. */
+#define VC4_CONFIG_BITS_EARLY_Z_UPDATE             BIT(1)
+#define VC4_CONFIG_BITS_EARLY_Z                    BIT(0)
+/** @} */
+
+/** @{ byte 1 of config bits. */
+#define VC4_CONFIG_BITS_Z_UPDATE                   BIT(7)
+/** same values in this 3-bit field as PIPE_FUNC_* */
+#define VC4_CONFIG_BITS_DEPTH_FUNC_SHIFT           4
+#define VC4_CONFIG_BITS_COVERAGE_READ_LEAVE        BIT(3)
+
+#define VC4_CONFIG_BITS_COVERAGE_UPDATE_NONZERO    (0 << 1)
+#define VC4_CONFIG_BITS_COVERAGE_UPDATE_ODD        (1 << 1)
+#define VC4_CONFIG_BITS_COVERAGE_UPDATE_OR         (2 << 1)
+#define VC4_CONFIG_BITS_COVERAGE_UPDATE_ZERO       (3 << 1)
+
+#define VC4_CONFIG_BITS_COVERAGE_PIPE_SELECT       BIT(0)
+/** @} */
+
+/** @{ byte 0 of config bits. */
+#define VC4_CONFIG_BITS_RASTERIZER_OVERSAMPLE_NONE (0 << 6)
+#define VC4_CONFIG_BITS_RASTERIZER_OVERSAMPLE_4X   (1 << 6)
+#define VC4_CONFIG_BITS_RASTERIZER_OVERSAMPLE_16X  (2 << 6)
+
+#define VC4_CONFIG_BITS_AA_POINTS_AND_LINES        BIT(4)
+#define VC4_CONFIG_BITS_ENABLE_DEPTH_OFFSET        BIT(3)
+#define VC4_CONFIG_BITS_CW_PRIMITIVES              BIT(2)
+#define VC4_CONFIG_BITS_ENABLE_PRIM_BACK           BIT(1)
+#define VC4_CONFIG_BITS_ENABLE_PRIM_FRONT          BIT(0)
+/** @} */
+
+/** @{ bits in the last u8 of VC4_PACKET_TILE_BINNING_MODE_CONFIG */
+#define VC4_BIN_CONFIG_DB_NON_MS                   BIT(7)
+
+#define VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_MASK       VC4_MASK(6, 5)
+#define VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_SHIFT      5
+#define VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_32         0
+#define VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_64         1
+#define VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_128        2
+#define VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_256        3
+
+#define VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_MASK  VC4_MASK(4, 3)
+#define VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_SHIFT 3
+#define VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_32    0
+#define VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_64    1
+#define VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_128   2
+#define VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_256   3
+
+#define VC4_BIN_CONFIG_AUTO_INIT_TSDA              BIT(2)
+#define VC4_BIN_CONFIG_TILE_BUFFER_64BIT           BIT(1)
+#define VC4_BIN_CONFIG_MS_MODE_4X                  BIT(0)
+/** @} */
+
+/** @{ bits in the last u16 of VC4_PACKET_TILE_RENDERING_MODE_CONFIG */
+#define VC4_RENDER_CONFIG_DB_NON_MS                BIT(12)
+#define VC4_RENDER_CONFIG_EARLY_Z_COVERAGE_DISABLE BIT(11)
+#define VC4_RENDER_CONFIG_EARLY_Z_DIRECTION_G      BIT(10)
+#define VC4_RENDER_CONFIG_COVERAGE_MODE            BIT(9)
+#define VC4_RENDER_CONFIG_ENABLE_VG_MASK           BIT(8)
+
+/** The values of the field are VC4_TILING_FORMAT_* */
+#define VC4_RENDER_CONFIG_MEMORY_FORMAT_MASK       VC4_MASK(7, 6)
+#define VC4_RENDER_CONFIG_MEMORY_FORMAT_SHIFT      6
+
+#define VC4_RENDER_CONFIG_DECIMATE_MODE_1X         (0 << 4)
+#define VC4_RENDER_CONFIG_DECIMATE_MODE_4X         (1 << 4)
+#define VC4_RENDER_CONFIG_DECIMATE_MODE_16X        (2 << 4)
+
+#define VC4_RENDER_CONFIG_FORMAT_MASK              VC4_MASK(3, 2)
+#define VC4_RENDER_CONFIG_FORMAT_SHIFT             2
+#define VC4_RENDER_CONFIG_FORMAT_BGR565_DITHERED   0
+#define VC4_RENDER_CONFIG_FORMAT_RGBA8888          1
+#define VC4_RENDER_CONFIG_FORMAT_BGR565            2
+
+#define VC4_RENDER_CONFIG_TILE_BUFFER_64BIT        BIT(1)
+#define VC4_RENDER_CONFIG_MS_MODE_4X               BIT(0)
+
+#define VC4_PRIMITIVE_LIST_FORMAT_16_INDEX         (1 << 4)
+#define VC4_PRIMITIVE_LIST_FORMAT_32_XY            (3 << 4)
+#define VC4_PRIMITIVE_LIST_FORMAT_TYPE_POINTS      (0 << 0)
+#define VC4_PRIMITIVE_LIST_FORMAT_TYPE_LINES       (1 << 0)
+#define VC4_PRIMITIVE_LIST_FORMAT_TYPE_TRIANGLES   (2 << 0)
+#define VC4_PRIMITIVE_LIST_FORMAT_TYPE_RHT         (3 << 0)
+
+enum vc4_texture_data_type {
+	VC4_TEXTURE_TYPE_RGBA8888 = 0,
+	VC4_TEXTURE_TYPE_RGBX8888 = 1,
+	VC4_TEXTURE_TYPE_RGBA4444 = 2,
+	VC4_TEXTURE_TYPE_RGBA5551 = 3,
+	VC4_TEXTURE_TYPE_RGB565 = 4,
+	VC4_TEXTURE_TYPE_LUMINANCE = 5,
+	VC4_TEXTURE_TYPE_ALPHA = 6,
+	VC4_TEXTURE_TYPE_LUMALPHA = 7,
+	VC4_TEXTURE_TYPE_ETC1 = 8,
+	VC4_TEXTURE_TYPE_S16F = 9,
+	VC4_TEXTURE_TYPE_S8 = 10,
+	VC4_TEXTURE_TYPE_S16 = 11,
+	VC4_TEXTURE_TYPE_BW1 = 12,
+	VC4_TEXTURE_TYPE_A4 = 13,
+	VC4_TEXTURE_TYPE_A1 = 14,
+	VC4_TEXTURE_TYPE_RGBA64 = 15,
+	VC4_TEXTURE_TYPE_RGBA32R = 16,
+	VC4_TEXTURE_TYPE_YUV422R = 17,
+};
+
+#define VC4_TEX_P0_OFFSET_MASK                     VC4_MASK(31, 12)
+#define VC4_TEX_P0_OFFSET_SHIFT                    12
+#define VC4_TEX_P0_CSWIZ_MASK                      VC4_MASK(11, 10)
+#define VC4_TEX_P0_CSWIZ_SHIFT                     10
+#define VC4_TEX_P0_CMMODE_MASK                     VC4_MASK(9, 9)
+#define VC4_TEX_P0_CMMODE_SHIFT                    9
+#define VC4_TEX_P0_FLIPY_MASK                      VC4_MASK(8, 8)
+#define VC4_TEX_P0_FLIPY_SHIFT                     8
+#define VC4_TEX_P0_TYPE_MASK                       VC4_MASK(7, 4)
+#define VC4_TEX_P0_TYPE_SHIFT                      4
+#define VC4_TEX_P0_MIPLVLS_MASK                    VC4_MASK(3, 0)
+#define VC4_TEX_P0_MIPLVLS_SHIFT                   0
+
+#define VC4_TEX_P1_TYPE4_MASK                      VC4_MASK(31, 31)
+#define VC4_TEX_P1_TYPE4_SHIFT                     31
+#define VC4_TEX_P1_HEIGHT_MASK                     VC4_MASK(30, 20)
+#define VC4_TEX_P1_HEIGHT_SHIFT                    20
+#define VC4_TEX_P1_ETCFLIP_MASK                    VC4_MASK(19, 19)
+#define VC4_TEX_P1_ETCFLIP_SHIFT                   19
+#define VC4_TEX_P1_WIDTH_MASK                      VC4_MASK(18, 8)
+#define VC4_TEX_P1_WIDTH_SHIFT                     8
+
+#define VC4_TEX_P1_MAGFILT_MASK                    VC4_MASK(7, 7)
+#define VC4_TEX_P1_MAGFILT_SHIFT                   7
+# define VC4_TEX_P1_MAGFILT_LINEAR                 0
+# define VC4_TEX_P1_MAGFILT_NEAREST                1
+
+#define VC4_TEX_P1_MINFILT_MASK                    VC4_MASK(6, 4)
+#define VC4_TEX_P1_MINFILT_SHIFT                   4
+# define VC4_TEX_P1_MINFILT_LINEAR                 0
+# define VC4_TEX_P1_MINFILT_NEAREST                1
+# define VC4_TEX_P1_MINFILT_NEAR_MIP_NEAR          2
+# define VC4_TEX_P1_MINFILT_NEAR_MIP_LIN           3
+# define VC4_TEX_P1_MINFILT_LIN_MIP_NEAR           4
+# define VC4_TEX_P1_MINFILT_LIN_MIP_LIN            5
+
+#define VC4_TEX_P1_WRAP_T_MASK                     VC4_MASK(3, 2)
+#define VC4_TEX_P1_WRAP_T_SHIFT                    2
+#define VC4_TEX_P1_WRAP_S_MASK                     VC4_MASK(1, 0)
+#define VC4_TEX_P1_WRAP_S_SHIFT                    0
+# define VC4_TEX_P1_WRAP_REPEAT                    0
+# define VC4_TEX_P1_WRAP_CLAMP                     1
+# define VC4_TEX_P1_WRAP_MIRROR                    2
+# define VC4_TEX_P1_WRAP_BORDER                    3
+
+#define VC4_TEX_P2_PTYPE_MASK                      VC4_MASK(31, 30)
+#define VC4_TEX_P2_PTYPE_SHIFT                     30
+# define VC4_TEX_P2_PTYPE_IGNORED                  0
+# define VC4_TEX_P2_PTYPE_CUBE_MAP_STRIDE          1
+# define VC4_TEX_P2_PTYPE_CHILD_IMAGE_DIMENSIONS   2
+# define VC4_TEX_P2_PTYPE_CHILD_IMAGE_OFFSETS      3
+
+/* VC4_TEX_P2_PTYPE_CUBE_MAP_STRIDE bits */
+#define VC4_TEX_P2_CMST_MASK                       VC4_MASK(29, 12)
+#define VC4_TEX_P2_CMST_SHIFT                      12
+#define VC4_TEX_P2_BSLOD_MASK                      VC4_MASK(0, 0)
+#define VC4_TEX_P2_BSLOD_SHIFT                     0
+
+/* VC4_TEX_P2_PTYPE_CHILD_IMAGE_DIMENSIONS */
+#define VC4_TEX_P2_CHEIGHT_MASK                    VC4_MASK(22, 12)
+#define VC4_TEX_P2_CHEIGHT_SHIFT                   12
+#define VC4_TEX_P2_CWIDTH_MASK                     VC4_MASK(10, 0)
+#define VC4_TEX_P2_CWIDTH_SHIFT                    0
+
+/* VC4_TEX_P2_PTYPE_CHILD_IMAGE_OFFSETS */
+#define VC4_TEX_P2_CYOFF_MASK                      VC4_MASK(22, 12)
+#define VC4_TEX_P2_CYOFF_SHIFT                     12
+#define VC4_TEX_P2_CXOFF_MASK                      VC4_MASK(10, 0)
+#define VC4_TEX_P2_CXOFF_SHIFT                     0
+
+#endif /* VC4_PACKET_H */
diff --git a/drivers/gpu/drm/vc4/vc4_perfmon.c b/drivers/gpu/drm/vc4/vc4_perfmon.c
new file mode 100644
index 000000000..437e7a27f
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_perfmon.c
@@ -0,0 +1,188 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2018 Broadcom
+ */
+
+/**
+ * DOC: VC4 V3D performance monitor module
+ *
+ * The V3D block provides 16 hardware counters which can count various events.
+ */
+
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+#define VC4_PERFMONID_MIN	1
+#define VC4_PERFMONID_MAX	U32_MAX
+
+void vc4_perfmon_get(struct vc4_perfmon *perfmon)
+{
+	if (perfmon)
+		refcount_inc(&perfmon->refcnt);
+}
+
+void vc4_perfmon_put(struct vc4_perfmon *perfmon)
+{
+	if (perfmon && refcount_dec_and_test(&perfmon->refcnt))
+		kfree(perfmon);
+}
+
+void vc4_perfmon_start(struct vc4_dev *vc4, struct vc4_perfmon *perfmon)
+{
+	unsigned int i;
+	u32 mask;
+
+	if (WARN_ON_ONCE(!perfmon || vc4->active_perfmon))
+		return;
+
+	for (i = 0; i < perfmon->ncounters; i++)
+		V3D_WRITE(V3D_PCTRS(i), perfmon->events[i]);
+
+	mask = GENMASK(perfmon->ncounters - 1, 0);
+	V3D_WRITE(V3D_PCTRC, mask);
+	V3D_WRITE(V3D_PCTRE, V3D_PCTRE_EN | mask);
+	vc4->active_perfmon = perfmon;
+}
+
+void vc4_perfmon_stop(struct vc4_dev *vc4, struct vc4_perfmon *perfmon,
+		      bool capture)
+{
+	unsigned int i;
+
+	if (WARN_ON_ONCE(!vc4->active_perfmon ||
+			 perfmon != vc4->active_perfmon))
+		return;
+
+	if (capture) {
+		for (i = 0; i < perfmon->ncounters; i++)
+			perfmon->counters[i] += V3D_READ(V3D_PCTR(i));
+	}
+
+	V3D_WRITE(V3D_PCTRE, 0);
+	vc4->active_perfmon = NULL;
+}
+
+struct vc4_perfmon *vc4_perfmon_find(struct vc4_file *vc4file, int id)
+{
+	struct vc4_perfmon *perfmon;
+
+	mutex_lock(&vc4file->perfmon.lock);
+	perfmon = idr_find(&vc4file->perfmon.idr, id);
+	vc4_perfmon_get(perfmon);
+	mutex_unlock(&vc4file->perfmon.lock);
+
+	return perfmon;
+}
+
+void vc4_perfmon_open_file(struct vc4_file *vc4file)
+{
+	mutex_init(&vc4file->perfmon.lock);
+	idr_init(&vc4file->perfmon.idr);
+}
+
+static int vc4_perfmon_idr_del(int id, void *elem, void *data)
+{
+	struct vc4_perfmon *perfmon = elem;
+
+	vc4_perfmon_put(perfmon);
+
+	return 0;
+}
+
+void vc4_perfmon_close_file(struct vc4_file *vc4file)
+{
+	mutex_lock(&vc4file->perfmon.lock);
+	idr_for_each(&vc4file->perfmon.idr, vc4_perfmon_idr_del, NULL);
+	idr_destroy(&vc4file->perfmon.idr);
+	mutex_unlock(&vc4file->perfmon.lock);
+}
+
+int vc4_perfmon_create_ioctl(struct drm_device *dev, void *data,
+			     struct drm_file *file_priv)
+{
+	struct vc4_file *vc4file = file_priv->driver_priv;
+	struct drm_vc4_perfmon_create *req = data;
+	struct vc4_perfmon *perfmon;
+	unsigned int i;
+	int ret;
+
+	/* Number of monitored counters cannot exceed HW limits. */
+	if (req->ncounters > DRM_VC4_MAX_PERF_COUNTERS ||
+	    !req->ncounters)
+		return -EINVAL;
+
+	/* Make sure all events are valid. */
+	for (i = 0; i < req->ncounters; i++) {
+		if (req->events[i] >= VC4_PERFCNT_NUM_EVENTS)
+			return -EINVAL;
+	}
+
+	perfmon = kzalloc(sizeof(*perfmon) + (req->ncounters * sizeof(u64)),
+			  GFP_KERNEL);
+	if (!perfmon)
+		return -ENOMEM;
+
+	for (i = 0; i < req->ncounters; i++)
+		perfmon->events[i] = req->events[i];
+
+	perfmon->ncounters = req->ncounters;
+
+	refcount_set(&perfmon->refcnt, 1);
+
+	mutex_lock(&vc4file->perfmon.lock);
+	ret = idr_alloc(&vc4file->perfmon.idr, perfmon, VC4_PERFMONID_MIN,
+			VC4_PERFMONID_MAX, GFP_KERNEL);
+	mutex_unlock(&vc4file->perfmon.lock);
+
+	if (ret < 0) {
+		kfree(perfmon);
+		return ret;
+	}
+
+	req->id = ret;
+	return 0;
+}
+
+int vc4_perfmon_destroy_ioctl(struct drm_device *dev, void *data,
+			      struct drm_file *file_priv)
+{
+	struct vc4_file *vc4file = file_priv->driver_priv;
+	struct drm_vc4_perfmon_destroy *req = data;
+	struct vc4_perfmon *perfmon;
+
+	mutex_lock(&vc4file->perfmon.lock);
+	perfmon = idr_remove(&vc4file->perfmon.idr, req->id);
+	mutex_unlock(&vc4file->perfmon.lock);
+
+	if (!perfmon)
+		return -EINVAL;
+
+	vc4_perfmon_put(perfmon);
+	return 0;
+}
+
+int vc4_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
+				 struct drm_file *file_priv)
+{
+	struct vc4_file *vc4file = file_priv->driver_priv;
+	struct drm_vc4_perfmon_get_values *req = data;
+	struct vc4_perfmon *perfmon;
+	int ret;
+
+	mutex_lock(&vc4file->perfmon.lock);
+	perfmon = idr_find(&vc4file->perfmon.idr, req->id);
+	vc4_perfmon_get(perfmon);
+	mutex_unlock(&vc4file->perfmon.lock);
+
+	if (!perfmon)
+		return -EINVAL;
+
+	if (copy_to_user(u64_to_user_ptr(req->values_ptr), perfmon->counters,
+			 perfmon->ncounters * sizeof(u64)))
+		ret = -EFAULT;
+	else
+		ret = 0;
+
+	vc4_perfmon_put(perfmon);
+	return ret;
+}
diff --git a/drivers/gpu/drm/vc4/vc4_plane.c b/drivers/gpu/drm/vc4/vc4_plane.c
new file mode 100644
index 000000000..39e608271
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_plane.c
@@ -0,0 +1,1015 @@
+/*
+ * Copyright (C) 2015 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+/**
+ * DOC: VC4 plane module
+ *
+ * Each DRM plane is a layer of pixels being scanned out by the HVS.
+ *
+ * At atomic modeset check time, we compute the HVS display element
+ * state that would be necessary for displaying the plane (giving us a
+ * chance to figure out if a plane configuration is invalid), then at
+ * atomic flush time the CRTC will ask us to write our element state
+ * into the region of the HVS that it has allocated for us.
+ */
+
+#include <drm/drm_atomic.h>
+#include <drm/drm_atomic_helper.h>
+#include <drm/drm_fb_cma_helper.h>
+#include <drm/drm_plane_helper.h>
+
+#include "uapi/drm/vc4_drm.h"
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+static const struct hvs_format {
+	u32 drm; /* DRM_FORMAT_* */
+	u32 hvs; /* HVS_FORMAT_* */
+	u32 pixel_order;
+} hvs_formats[] = {
+	{
+		.drm = DRM_FORMAT_XRGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
+		.pixel_order = HVS_PIXEL_ORDER_ABGR,
+	},
+	{
+		.drm = DRM_FORMAT_ARGB8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
+		.pixel_order = HVS_PIXEL_ORDER_ABGR,
+	},
+	{
+		.drm = DRM_FORMAT_ABGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
+		.pixel_order = HVS_PIXEL_ORDER_ARGB,
+	},
+	{
+		.drm = DRM_FORMAT_XBGR8888, .hvs = HVS_PIXEL_FORMAT_RGBA8888,
+		.pixel_order = HVS_PIXEL_ORDER_ARGB,
+	},
+	{
+		.drm = DRM_FORMAT_RGB565, .hvs = HVS_PIXEL_FORMAT_RGB565,
+		.pixel_order = HVS_PIXEL_ORDER_XRGB,
+	},
+	{
+		.drm = DRM_FORMAT_BGR565, .hvs = HVS_PIXEL_FORMAT_RGB565,
+		.pixel_order = HVS_PIXEL_ORDER_XBGR,
+	},
+	{
+		.drm = DRM_FORMAT_ARGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
+		.pixel_order = HVS_PIXEL_ORDER_ABGR,
+	},
+	{
+		.drm = DRM_FORMAT_XRGB1555, .hvs = HVS_PIXEL_FORMAT_RGBA5551,
+		.pixel_order = HVS_PIXEL_ORDER_ABGR,
+	},
+	{
+		.drm = DRM_FORMAT_RGB888, .hvs = HVS_PIXEL_FORMAT_RGB888,
+		.pixel_order = HVS_PIXEL_ORDER_XRGB,
+	},
+	{
+		.drm = DRM_FORMAT_BGR888, .hvs = HVS_PIXEL_FORMAT_RGB888,
+		.pixel_order = HVS_PIXEL_ORDER_XBGR,
+	},
+	{
+		.drm = DRM_FORMAT_YUV422,
+		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
+		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
+	},
+	{
+		.drm = DRM_FORMAT_YVU422,
+		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE,
+		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
+	},
+	{
+		.drm = DRM_FORMAT_YUV420,
+		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
+		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
+	},
+	{
+		.drm = DRM_FORMAT_YVU420,
+		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE,
+		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
+	},
+	{
+		.drm = DRM_FORMAT_NV12,
+		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
+		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
+	},
+	{
+		.drm = DRM_FORMAT_NV21,
+		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE,
+		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
+	},
+	{
+		.drm = DRM_FORMAT_NV16,
+		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
+		.pixel_order = HVS_PIXEL_ORDER_XYCBCR,
+	},
+	{
+		.drm = DRM_FORMAT_NV61,
+		.hvs = HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE,
+		.pixel_order = HVS_PIXEL_ORDER_XYCRCB,
+	},
+};
+
+static const struct hvs_format *vc4_get_hvs_format(u32 drm_format)
+{
+	unsigned i;
+
+	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
+		if (hvs_formats[i].drm == drm_format)
+			return &hvs_formats[i];
+	}
+
+	return NULL;
+}
+
+static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
+{
+	if (dst > src)
+		return VC4_SCALING_PPF;
+	else if (dst < src)
+		return VC4_SCALING_TPZ;
+	else
+		return VC4_SCALING_NONE;
+}
+
+static bool plane_enabled(struct drm_plane_state *state)
+{
+	return state->fb && state->crtc;
+}
+
+static struct drm_plane_state *vc4_plane_duplicate_state(struct drm_plane *plane)
+{
+	struct vc4_plane_state *vc4_state;
+
+	if (WARN_ON(!plane->state))
+		return NULL;
+
+	vc4_state = kmemdup(plane->state, sizeof(*vc4_state), GFP_KERNEL);
+	if (!vc4_state)
+		return NULL;
+
+	memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
+
+	__drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
+
+	if (vc4_state->dlist) {
+		vc4_state->dlist = kmemdup(vc4_state->dlist,
+					   vc4_state->dlist_count * 4,
+					   GFP_KERNEL);
+		if (!vc4_state->dlist) {
+			kfree(vc4_state);
+			return NULL;
+		}
+		vc4_state->dlist_size = vc4_state->dlist_count;
+	}
+
+	return &vc4_state->base;
+}
+
+static void vc4_plane_destroy_state(struct drm_plane *plane,
+				    struct drm_plane_state *state)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
+	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
+
+	if (vc4_state->lbm.allocated) {
+		unsigned long irqflags;
+
+		spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
+		drm_mm_remove_node(&vc4_state->lbm);
+		spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
+	}
+
+	kfree(vc4_state->dlist);
+	__drm_atomic_helper_plane_destroy_state(&vc4_state->base);
+	kfree(state);
+}
+
+/* Called during init to allocate the plane's atomic state. */
+static void vc4_plane_reset(struct drm_plane *plane)
+{
+	struct vc4_plane_state *vc4_state;
+
+	WARN_ON(plane->state);
+
+	vc4_state = kzalloc(sizeof(*vc4_state), GFP_KERNEL);
+	if (!vc4_state)
+		return;
+
+	plane->state = &vc4_state->base;
+	plane->state->alpha = DRM_BLEND_ALPHA_OPAQUE;
+	vc4_state->base.plane = plane;
+}
+
+static void vc4_dlist_write(struct vc4_plane_state *vc4_state, u32 val)
+{
+	if (vc4_state->dlist_count == vc4_state->dlist_size) {
+		u32 new_size = max(4u, vc4_state->dlist_count * 2);
+		u32 *new_dlist = kmalloc_array(new_size, 4, GFP_KERNEL);
+
+		if (!new_dlist)
+			return;
+		memcpy(new_dlist, vc4_state->dlist, vc4_state->dlist_count * 4);
+
+		kfree(vc4_state->dlist);
+		vc4_state->dlist = new_dlist;
+		vc4_state->dlist_size = new_size;
+	}
+
+	vc4_state->dlist[vc4_state->dlist_count++] = val;
+}
+
+/* Returns the scl0/scl1 field based on whether the dimensions need to
+ * be up/down/non-scaled.
+ *
+ * This is a replication of a table from the spec.
+ */
+static u32 vc4_get_scl_field(struct drm_plane_state *state, int plane)
+{
+	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
+
+	switch (vc4_state->x_scaling[plane] << 2 | vc4_state->y_scaling[plane]) {
+	case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
+		return SCALER_CTL0_SCL_H_PPF_V_PPF;
+	case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
+		return SCALER_CTL0_SCL_H_TPZ_V_PPF;
+	case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
+		return SCALER_CTL0_SCL_H_PPF_V_TPZ;
+	case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
+		return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
+	case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
+		return SCALER_CTL0_SCL_H_PPF_V_NONE;
+	case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
+		return SCALER_CTL0_SCL_H_NONE_V_PPF;
+	case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
+		return SCALER_CTL0_SCL_H_NONE_V_TPZ;
+	case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
+		return SCALER_CTL0_SCL_H_TPZ_V_NONE;
+	default:
+	case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
+		/* The unity case is independently handled by
+		 * SCALER_CTL0_UNITY.
+		 */
+		return 0;
+	}
+}
+
+static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
+{
+	struct drm_plane *plane = state->plane;
+	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
+	struct drm_framebuffer *fb = state->fb;
+	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
+	u32 subpixel_src_mask = (1 << 16) - 1;
+	u32 format = fb->format->format;
+	int num_planes = fb->format->num_planes;
+	u32 h_subsample = 1;
+	u32 v_subsample = 1;
+	int i;
+
+	for (i = 0; i < num_planes; i++)
+		vc4_state->offsets[i] = bo->paddr + fb->offsets[i];
+
+	/* We don't support subpixel source positioning for scaling. */
+	if ((state->src_x & subpixel_src_mask) ||
+	    (state->src_y & subpixel_src_mask) ||
+	    (state->src_w & subpixel_src_mask) ||
+	    (state->src_h & subpixel_src_mask)) {
+		return -EINVAL;
+	}
+
+	vc4_state->src_x = state->src_x >> 16;
+	vc4_state->src_y = state->src_y >> 16;
+	vc4_state->src_w[0] = state->src_w >> 16;
+	vc4_state->src_h[0] = state->src_h >> 16;
+
+	vc4_state->crtc_x = state->crtc_x;
+	vc4_state->crtc_y = state->crtc_y;
+	vc4_state->crtc_w = state->crtc_w;
+	vc4_state->crtc_h = state->crtc_h;
+
+	vc4_state->x_scaling[0] = vc4_get_scaling_mode(vc4_state->src_w[0],
+						       vc4_state->crtc_w);
+	vc4_state->y_scaling[0] = vc4_get_scaling_mode(vc4_state->src_h[0],
+						       vc4_state->crtc_h);
+
+	vc4_state->is_unity = (vc4_state->x_scaling[0] == VC4_SCALING_NONE &&
+			       vc4_state->y_scaling[0] == VC4_SCALING_NONE);
+
+	if (num_planes > 1) {
+		vc4_state->is_yuv = true;
+
+		h_subsample = drm_format_horz_chroma_subsampling(format);
+		v_subsample = drm_format_vert_chroma_subsampling(format);
+		vc4_state->src_w[1] = vc4_state->src_w[0] / h_subsample;
+		vc4_state->src_h[1] = vc4_state->src_h[0] / v_subsample;
+
+		vc4_state->x_scaling[1] =
+			vc4_get_scaling_mode(vc4_state->src_w[1],
+					     vc4_state->crtc_w);
+		vc4_state->y_scaling[1] =
+			vc4_get_scaling_mode(vc4_state->src_h[1],
+					     vc4_state->crtc_h);
+
+		/* YUV conversion requires that horizontal scaling be enabled
+		 * on the UV plane even if vc4_get_scaling_mode() returned
+		 * VC4_SCALING_NONE (which can happen when the down-scaling
+		 * ratio is 0.5). Let's force it to VC4_SCALING_PPF in this
+		 * case.
+		 */
+		if (vc4_state->x_scaling[1] == VC4_SCALING_NONE)
+			vc4_state->x_scaling[1] = VC4_SCALING_PPF;
+	} else {
+		vc4_state->is_yuv = false;
+		vc4_state->x_scaling[1] = VC4_SCALING_NONE;
+		vc4_state->y_scaling[1] = VC4_SCALING_NONE;
+	}
+
+	/* No configuring scaling on the cursor plane, since it gets
+	   non-vblank-synced updates, and scaling requires requires
+	   LBM changes which have to be vblank-synced.
+	 */
+	if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
+		return -EINVAL;
+
+	/* Clamp the on-screen start x/y to 0.  The hardware doesn't
+	 * support negative y, and negative x wastes bandwidth.
+	 */
+	if (vc4_state->crtc_x < 0) {
+		for (i = 0; i < num_planes; i++) {
+			u32 cpp = fb->format->cpp[i];
+			u32 subs = ((i == 0) ? 1 : h_subsample);
+
+			vc4_state->offsets[i] += (cpp *
+						  (-vc4_state->crtc_x) / subs);
+		}
+		vc4_state->src_w[0] += vc4_state->crtc_x;
+		vc4_state->src_w[1] += vc4_state->crtc_x / h_subsample;
+		vc4_state->crtc_x = 0;
+	}
+
+	if (vc4_state->crtc_y < 0) {
+		for (i = 0; i < num_planes; i++) {
+			u32 subs = ((i == 0) ? 1 : v_subsample);
+
+			vc4_state->offsets[i] += (fb->pitches[i] *
+						  (-vc4_state->crtc_y) / subs);
+		}
+		vc4_state->src_h[0] += vc4_state->crtc_y;
+		vc4_state->src_h[1] += vc4_state->crtc_y / v_subsample;
+		vc4_state->crtc_y = 0;
+	}
+
+	return 0;
+}
+
+static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
+{
+	u32 scale, recip;
+
+	scale = (1 << 16) * src / dst;
+
+	/* The specs note that while the reciprocal would be defined
+	 * as (1<<32)/scale, ~0 is close enough.
+	 */
+	recip = ~0 / scale;
+
+	vc4_dlist_write(vc4_state,
+			VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
+			VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
+	vc4_dlist_write(vc4_state,
+			VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
+}
+
+static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
+{
+	u32 scale = (1 << 16) * src / dst;
+
+	vc4_dlist_write(vc4_state,
+			SCALER_PPF_AGC |
+			VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
+			VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
+}
+
+static u32 vc4_lbm_size(struct drm_plane_state *state)
+{
+	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
+	/* This is the worst case number.  One of the two sizes will
+	 * be used depending on the scaling configuration.
+	 */
+	u32 pix_per_line = max(vc4_state->src_w[0], (u32)vc4_state->crtc_w);
+	u32 lbm;
+
+	if (!vc4_state->is_yuv) {
+		if (vc4_state->is_unity)
+			return 0;
+		else if (vc4_state->y_scaling[0] == VC4_SCALING_TPZ)
+			lbm = pix_per_line * 8;
+		else {
+			/* In special cases, this multiplier might be 12. */
+			lbm = pix_per_line * 16;
+		}
+	} else {
+		/* There are cases for this going down to a multiplier
+		 * of 2, but according to the firmware source, the
+		 * table in the docs is somewhat wrong.
+		 */
+		lbm = pix_per_line * 16;
+	}
+
+	lbm = roundup(lbm, 32);
+
+	return lbm;
+}
+
+static void vc4_write_scaling_parameters(struct drm_plane_state *state,
+					 int channel)
+{
+	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
+
+	/* Ch0 H-PPF Word 0: Scaling Parameters */
+	if (vc4_state->x_scaling[channel] == VC4_SCALING_PPF) {
+		vc4_write_ppf(vc4_state,
+			      vc4_state->src_w[channel], vc4_state->crtc_w);
+	}
+
+	/* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
+	if (vc4_state->y_scaling[channel] == VC4_SCALING_PPF) {
+		vc4_write_ppf(vc4_state,
+			      vc4_state->src_h[channel], vc4_state->crtc_h);
+		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
+	}
+
+	/* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
+	if (vc4_state->x_scaling[channel] == VC4_SCALING_TPZ) {
+		vc4_write_tpz(vc4_state,
+			      vc4_state->src_w[channel], vc4_state->crtc_w);
+	}
+
+	/* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
+	if (vc4_state->y_scaling[channel] == VC4_SCALING_TPZ) {
+		vc4_write_tpz(vc4_state,
+			      vc4_state->src_h[channel], vc4_state->crtc_h);
+		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
+	}
+}
+
+/* Writes out a full display list for an active plane to the plane's
+ * private dlist state.
+ */
+static int vc4_plane_mode_set(struct drm_plane *plane,
+			      struct drm_plane_state *state)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
+	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
+	struct drm_framebuffer *fb = state->fb;
+	u32 ctl0_offset = vc4_state->dlist_count;
+	const struct hvs_format *format = vc4_get_hvs_format(fb->format->format);
+	u64 base_format_mod = fourcc_mod_broadcom_mod(fb->modifier);
+	int num_planes = drm_format_num_planes(format->drm);
+	bool mix_plane_alpha;
+	bool covers_screen;
+	u32 scl0, scl1, pitch0;
+	u32 lbm_size, tiling;
+	unsigned long irqflags;
+	u32 hvs_format = format->hvs;
+	int ret, i;
+
+	ret = vc4_plane_setup_clipping_and_scaling(state);
+	if (ret)
+		return ret;
+
+	/* Allocate the LBM memory that the HVS will use for temporary
+	 * storage due to our scaling/format conversion.
+	 */
+	lbm_size = vc4_lbm_size(state);
+	if (lbm_size) {
+		if (!vc4_state->lbm.allocated) {
+			spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
+			ret = drm_mm_insert_node_generic(&vc4->hvs->lbm_mm,
+							 &vc4_state->lbm,
+							 lbm_size, 32, 0, 0);
+			spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
+		} else {
+			WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
+		}
+	}
+
+	if (ret)
+		return ret;
+
+	/* SCL1 is used for Cb/Cr scaling of planar formats.  For RGB
+	 * and 4:4:4, scl1 should be set to scl0 so both channels of
+	 * the scaler do the same thing.  For YUV, the Y plane needs
+	 * to be put in channel 1 and Cb/Cr in channel 0, so we swap
+	 * the scl fields here.
+	 */
+	if (num_planes == 1) {
+		scl0 = vc4_get_scl_field(state, 0);
+		scl1 = scl0;
+	} else {
+		scl0 = vc4_get_scl_field(state, 1);
+		scl1 = vc4_get_scl_field(state, 0);
+	}
+
+	switch (base_format_mod) {
+	case DRM_FORMAT_MOD_LINEAR:
+		tiling = SCALER_CTL0_TILING_LINEAR;
+		pitch0 = VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH);
+		break;
+
+	case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED: {
+		/* For T-tiled, the FB pitch is "how many bytes from
+		 * one row to the next, such that pitch * tile_h ==
+		 * tile_size * tiles_per_row."
+		 */
+		u32 tile_size_shift = 12; /* T tiles are 4kb */
+		u32 tile_h_shift = 5; /* 16 and 32bpp are 32 pixels high */
+		u32 tiles_w = fb->pitches[0] >> (tile_size_shift - tile_h_shift);
+
+		tiling = SCALER_CTL0_TILING_256B_OR_T;
+
+		pitch0 = (VC4_SET_FIELD(0, SCALER_PITCH0_TILE_Y_OFFSET) |
+			  VC4_SET_FIELD(0, SCALER_PITCH0_TILE_WIDTH_L) |
+			  VC4_SET_FIELD(tiles_w, SCALER_PITCH0_TILE_WIDTH_R));
+		break;
+	}
+
+	case DRM_FORMAT_MOD_BROADCOM_SAND64:
+	case DRM_FORMAT_MOD_BROADCOM_SAND128:
+	case DRM_FORMAT_MOD_BROADCOM_SAND256: {
+		uint32_t param = fourcc_mod_broadcom_param(fb->modifier);
+
+		/* Column-based NV12 or RGBA.
+		 */
+		if (fb->format->num_planes > 1) {
+			if (hvs_format != HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE) {
+				DRM_DEBUG_KMS("SAND format only valid for NV12/21");
+				return -EINVAL;
+			}
+			hvs_format = HVS_PIXEL_FORMAT_H264;
+		} else {
+			if (base_format_mod == DRM_FORMAT_MOD_BROADCOM_SAND256) {
+				DRM_DEBUG_KMS("SAND256 format only valid for H.264");
+				return -EINVAL;
+			}
+		}
+
+		switch (base_format_mod) {
+		case DRM_FORMAT_MOD_BROADCOM_SAND64:
+			tiling = SCALER_CTL0_TILING_64B;
+			break;
+		case DRM_FORMAT_MOD_BROADCOM_SAND128:
+			tiling = SCALER_CTL0_TILING_128B;
+			break;
+		case DRM_FORMAT_MOD_BROADCOM_SAND256:
+			tiling = SCALER_CTL0_TILING_256B_OR_T;
+			break;
+		default:
+			break;
+		}
+
+		if (param > SCALER_TILE_HEIGHT_MASK) {
+			DRM_DEBUG_KMS("SAND height too large (%d)\n", param);
+			return -EINVAL;
+		}
+
+		pitch0 = VC4_SET_FIELD(param, SCALER_TILE_HEIGHT);
+		break;
+	}
+
+	default:
+		DRM_DEBUG_KMS("Unsupported FB tiling flag 0x%16llx",
+			      (long long)fb->modifier);
+		return -EINVAL;
+	}
+
+	/* Control word */
+	vc4_dlist_write(vc4_state,
+			SCALER_CTL0_VALID |
+			VC4_SET_FIELD(SCALER_CTL0_RGBA_EXPAND_ROUND, SCALER_CTL0_RGBA_EXPAND) |
+			(format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
+			(hvs_format << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
+			VC4_SET_FIELD(tiling, SCALER_CTL0_TILING) |
+			(vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
+			VC4_SET_FIELD(scl0, SCALER_CTL0_SCL0) |
+			VC4_SET_FIELD(scl1, SCALER_CTL0_SCL1));
+
+	/* Position Word 0: Image Positions and Alpha Value */
+	vc4_state->pos0_offset = vc4_state->dlist_count;
+	vc4_dlist_write(vc4_state,
+			VC4_SET_FIELD(state->alpha >> 8, SCALER_POS0_FIXED_ALPHA) |
+			VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
+			VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
+
+	/* Position Word 1: Scaled Image Dimensions. */
+	if (!vc4_state->is_unity) {
+		vc4_dlist_write(vc4_state,
+				VC4_SET_FIELD(vc4_state->crtc_w,
+					      SCALER_POS1_SCL_WIDTH) |
+				VC4_SET_FIELD(vc4_state->crtc_h,
+					      SCALER_POS1_SCL_HEIGHT));
+	}
+
+	/* Don't waste cycles mixing with plane alpha if the set alpha
+	 * is opaque or there is no per-pixel alpha information.
+	 * In any case we use the alpha property value as the fixed alpha.
+	 */
+	mix_plane_alpha = state->alpha != DRM_BLEND_ALPHA_OPAQUE &&
+			  fb->format->has_alpha;
+
+	/* Position Word 2: Source Image Size, Alpha */
+	vc4_state->pos2_offset = vc4_state->dlist_count;
+	vc4_dlist_write(vc4_state,
+			VC4_SET_FIELD(fb->format->has_alpha ?
+				      SCALER_POS2_ALPHA_MODE_PIPELINE :
+				      SCALER_POS2_ALPHA_MODE_FIXED,
+				      SCALER_POS2_ALPHA_MODE) |
+			(mix_plane_alpha ? SCALER_POS2_ALPHA_MIX : 0) |
+			(fb->format->has_alpha ? SCALER_POS2_ALPHA_PREMULT : 0) |
+			VC4_SET_FIELD(vc4_state->src_w[0], SCALER_POS2_WIDTH) |
+			VC4_SET_FIELD(vc4_state->src_h[0], SCALER_POS2_HEIGHT));
+
+	/* Position Word 3: Context.  Written by the HVS. */
+	vc4_dlist_write(vc4_state, 0xc0c0c0c0);
+
+
+	/* Pointer Word 0/1/2: RGB / Y / Cb / Cr Pointers
+	 *
+	 * The pointers may be any byte address.
+	 */
+	vc4_state->ptr0_offset = vc4_state->dlist_count;
+	for (i = 0; i < num_planes; i++)
+		vc4_dlist_write(vc4_state, vc4_state->offsets[i]);
+
+	/* Pointer Context Word 0/1/2: Written by the HVS */
+	for (i = 0; i < num_planes; i++)
+		vc4_dlist_write(vc4_state, 0xc0c0c0c0);
+
+	/* Pitch word 0 */
+	vc4_dlist_write(vc4_state, pitch0);
+
+	/* Pitch word 1/2 */
+	for (i = 1; i < num_planes; i++) {
+		if (hvs_format != HVS_PIXEL_FORMAT_H264) {
+			vc4_dlist_write(vc4_state,
+					VC4_SET_FIELD(fb->pitches[i],
+						      SCALER_SRC_PITCH));
+		} else {
+			vc4_dlist_write(vc4_state, pitch0);
+		}
+	}
+
+	/* Colorspace conversion words */
+	if (vc4_state->is_yuv) {
+		vc4_dlist_write(vc4_state, SCALER_CSC0_ITR_R_601_5);
+		vc4_dlist_write(vc4_state, SCALER_CSC1_ITR_R_601_5);
+		vc4_dlist_write(vc4_state, SCALER_CSC2_ITR_R_601_5);
+	}
+
+	if (vc4_state->x_scaling[0] != VC4_SCALING_NONE ||
+	    vc4_state->x_scaling[1] != VC4_SCALING_NONE ||
+	    vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
+	    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
+		/* LBM Base Address. */
+		if (vc4_state->y_scaling[0] != VC4_SCALING_NONE ||
+		    vc4_state->y_scaling[1] != VC4_SCALING_NONE) {
+			vc4_dlist_write(vc4_state, vc4_state->lbm.start);
+		}
+
+		if (num_planes > 1) {
+			/* Emit Cb/Cr as channel 0 and Y as channel
+			 * 1. This matches how we set up scl0/scl1
+			 * above.
+			 */
+			vc4_write_scaling_parameters(state, 1);
+		}
+		vc4_write_scaling_parameters(state, 0);
+
+		/* If any PPF setup was done, then all the kernel
+		 * pointers get uploaded.
+		 */
+		if (vc4_state->x_scaling[0] == VC4_SCALING_PPF ||
+		    vc4_state->y_scaling[0] == VC4_SCALING_PPF ||
+		    vc4_state->x_scaling[1] == VC4_SCALING_PPF ||
+		    vc4_state->y_scaling[1] == VC4_SCALING_PPF) {
+			u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
+						   SCALER_PPF_KERNEL_OFFSET);
+
+			/* HPPF plane 0 */
+			vc4_dlist_write(vc4_state, kernel);
+			/* VPPF plane 0 */
+			vc4_dlist_write(vc4_state, kernel);
+			/* HPPF plane 1 */
+			vc4_dlist_write(vc4_state, kernel);
+			/* VPPF plane 1 */
+			vc4_dlist_write(vc4_state, kernel);
+		}
+	}
+
+	vc4_state->dlist[ctl0_offset] |=
+		VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
+
+	/* crtc_* are already clipped coordinates. */
+	covers_screen = vc4_state->crtc_x == 0 && vc4_state->crtc_y == 0 &&
+			vc4_state->crtc_w == state->crtc->mode.hdisplay &&
+			vc4_state->crtc_h == state->crtc->mode.vdisplay;
+	/* Background fill might be necessary when the plane has per-pixel
+	 * alpha content or a non-opaque plane alpha and could blend from the
+	 * background or does not cover the entire screen.
+	 */
+	vc4_state->needs_bg_fill = fb->format->has_alpha || !covers_screen ||
+				   state->alpha != DRM_BLEND_ALPHA_OPAQUE;
+
+	return 0;
+}
+
+/* If a modeset involves changing the setup of a plane, the atomic
+ * infrastructure will call this to validate a proposed plane setup.
+ * However, if a plane isn't getting updated, this (and the
+ * corresponding vc4_plane_atomic_update) won't get called.  Thus, we
+ * compute the dlist here and have all active plane dlists get updated
+ * in the CRTC's flush.
+ */
+static int vc4_plane_atomic_check(struct drm_plane *plane,
+				  struct drm_plane_state *state)
+{
+	struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
+
+	vc4_state->dlist_count = 0;
+
+	if (plane_enabled(state))
+		return vc4_plane_mode_set(plane, state);
+	else
+		return 0;
+}
+
+static void vc4_plane_atomic_update(struct drm_plane *plane,
+				    struct drm_plane_state *old_state)
+{
+	/* No contents here.  Since we don't know where in the CRTC's
+	 * dlist we should be stored, our dlist is uploaded to the
+	 * hardware with vc4_plane_write_dlist() at CRTC atomic_flush
+	 * time.
+	 */
+}
+
+u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist)
+{
+	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
+	int i;
+
+	vc4_state->hw_dlist = dlist;
+
+	/* Can't memcpy_toio() because it needs to be 32-bit writes. */
+	for (i = 0; i < vc4_state->dlist_count; i++)
+		writel(vc4_state->dlist[i], &dlist[i]);
+
+	return vc4_state->dlist_count;
+}
+
+u32 vc4_plane_dlist_size(const struct drm_plane_state *state)
+{
+	const struct vc4_plane_state *vc4_state =
+		container_of(state, typeof(*vc4_state), base);
+
+	return vc4_state->dlist_count;
+}
+
+/* Updates the plane to immediately (well, once the FIFO needs
+ * refilling) scan out from at a new framebuffer.
+ */
+void vc4_plane_async_set_fb(struct drm_plane *plane, struct drm_framebuffer *fb)
+{
+	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
+	struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
+	uint32_t addr;
+
+	/* We're skipping the address adjustment for negative origin,
+	 * because this is only called on the primary plane.
+	 */
+	WARN_ON_ONCE(plane->state->crtc_x < 0 || plane->state->crtc_y < 0);
+	addr = bo->paddr + fb->offsets[0];
+
+	/* Write the new address into the hardware immediately.  The
+	 * scanout will start from this address as soon as the FIFO
+	 * needs to refill with pixels.
+	 */
+	writel(addr, &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
+
+	/* Also update the CPU-side dlist copy, so that any later
+	 * atomic updates that don't do a new modeset on our plane
+	 * also use our updated address.
+	 */
+	vc4_state->dlist[vc4_state->ptr0_offset] = addr;
+}
+
+static void vc4_plane_atomic_async_update(struct drm_plane *plane,
+					  struct drm_plane_state *state)
+{
+	struct vc4_plane_state *vc4_state = to_vc4_plane_state(plane->state);
+
+	if (plane->state->fb != state->fb) {
+		vc4_plane_async_set_fb(plane, state->fb);
+		drm_atomic_set_fb_for_plane(plane->state, state->fb);
+	}
+
+	swap(plane->state->fb, state->fb);
+	/* Set the cursor's position on the screen.  This is the
+	 * expected change from the drm_mode_cursor_universal()
+	 * helper.
+	 */
+	plane->state->crtc_x = state->crtc_x;
+	plane->state->crtc_y = state->crtc_y;
+
+	/* Allow changing the start position within the cursor BO, if
+	 * that matters.
+	 */
+	plane->state->src_x = state->src_x;
+	plane->state->src_y = state->src_y;
+
+	/* Update the display list based on the new crtc_x/y. */
+	vc4_plane_atomic_check(plane, plane->state);
+
+	/* Note that we can't just call vc4_plane_write_dlist()
+	 * because that would smash the context data that the HVS is
+	 * currently using.
+	 */
+	writel(vc4_state->dlist[vc4_state->pos0_offset],
+	       &vc4_state->hw_dlist[vc4_state->pos0_offset]);
+	writel(vc4_state->dlist[vc4_state->pos2_offset],
+	       &vc4_state->hw_dlist[vc4_state->pos2_offset]);
+	writel(vc4_state->dlist[vc4_state->ptr0_offset],
+	       &vc4_state->hw_dlist[vc4_state->ptr0_offset]);
+}
+
+static int vc4_plane_atomic_async_check(struct drm_plane *plane,
+					struct drm_plane_state *state)
+{
+	/* No configuring new scaling in the fast path. */
+	if (plane->state->crtc_w != state->crtc_w ||
+	    plane->state->crtc_h != state->crtc_h ||
+	    plane->state->src_w != state->src_w ||
+	    plane->state->src_h != state->src_h)
+		return -EINVAL;
+
+	return 0;
+}
+
+static int vc4_prepare_fb(struct drm_plane *plane,
+			  struct drm_plane_state *state)
+{
+	struct vc4_bo *bo;
+	struct dma_fence *fence;
+	int ret;
+
+	if (!state->fb)
+		return 0;
+
+	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
+
+	fence = reservation_object_get_excl_rcu(bo->resv);
+	drm_atomic_set_fence_for_plane(state, fence);
+
+	if (plane->state->fb == state->fb)
+		return 0;
+
+	ret = vc4_bo_inc_usecnt(bo);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static void vc4_cleanup_fb(struct drm_plane *plane,
+			   struct drm_plane_state *state)
+{
+	struct vc4_bo *bo;
+
+	if (plane->state->fb == state->fb || !state->fb)
+		return;
+
+	bo = to_vc4_bo(&drm_fb_cma_get_gem_obj(state->fb, 0)->base);
+	vc4_bo_dec_usecnt(bo);
+}
+
+static const struct drm_plane_helper_funcs vc4_plane_helper_funcs = {
+	.atomic_check = vc4_plane_atomic_check,
+	.atomic_update = vc4_plane_atomic_update,
+	.prepare_fb = vc4_prepare_fb,
+	.cleanup_fb = vc4_cleanup_fb,
+	.atomic_async_check = vc4_plane_atomic_async_check,
+	.atomic_async_update = vc4_plane_atomic_async_update,
+};
+
+static void vc4_plane_destroy(struct drm_plane *plane)
+{
+	drm_plane_helper_disable(plane, NULL);
+	drm_plane_cleanup(plane);
+}
+
+static bool vc4_format_mod_supported(struct drm_plane *plane,
+				     uint32_t format,
+				     uint64_t modifier)
+{
+	/* Support T_TILING for RGB formats only. */
+	switch (format) {
+	case DRM_FORMAT_XRGB8888:
+	case DRM_FORMAT_ARGB8888:
+	case DRM_FORMAT_ABGR8888:
+	case DRM_FORMAT_XBGR8888:
+	case DRM_FORMAT_RGB565:
+	case DRM_FORMAT_BGR565:
+	case DRM_FORMAT_ARGB1555:
+	case DRM_FORMAT_XRGB1555:
+		switch (fourcc_mod_broadcom_mod(modifier)) {
+		case DRM_FORMAT_MOD_LINEAR:
+		case DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED:
+		case DRM_FORMAT_MOD_BROADCOM_SAND64:
+		case DRM_FORMAT_MOD_BROADCOM_SAND128:
+			return true;
+		default:
+			return false;
+		}
+	case DRM_FORMAT_NV12:
+	case DRM_FORMAT_NV21:
+		switch (fourcc_mod_broadcom_mod(modifier)) {
+		case DRM_FORMAT_MOD_LINEAR:
+		case DRM_FORMAT_MOD_BROADCOM_SAND64:
+		case DRM_FORMAT_MOD_BROADCOM_SAND128:
+		case DRM_FORMAT_MOD_BROADCOM_SAND256:
+			return true;
+		default:
+			return false;
+		}
+	case DRM_FORMAT_YUV422:
+	case DRM_FORMAT_YVU422:
+	case DRM_FORMAT_YUV420:
+	case DRM_FORMAT_YVU420:
+	case DRM_FORMAT_NV16:
+	case DRM_FORMAT_NV61:
+	default:
+		return (modifier == DRM_FORMAT_MOD_LINEAR);
+	}
+}
+
+static const struct drm_plane_funcs vc4_plane_funcs = {
+	.update_plane = drm_atomic_helper_update_plane,
+	.disable_plane = drm_atomic_helper_disable_plane,
+	.destroy = vc4_plane_destroy,
+	.set_property = NULL,
+	.reset = vc4_plane_reset,
+	.atomic_duplicate_state = vc4_plane_duplicate_state,
+	.atomic_destroy_state = vc4_plane_destroy_state,
+	.format_mod_supported = vc4_format_mod_supported,
+};
+
+struct drm_plane *vc4_plane_init(struct drm_device *dev,
+				 enum drm_plane_type type)
+{
+	struct drm_plane *plane = NULL;
+	struct vc4_plane *vc4_plane;
+	u32 formats[ARRAY_SIZE(hvs_formats)];
+	u32 num_formats = 0;
+	int ret = 0;
+	unsigned i;
+	static const uint64_t modifiers[] = {
+		DRM_FORMAT_MOD_BROADCOM_VC4_T_TILED,
+		DRM_FORMAT_MOD_BROADCOM_SAND128,
+		DRM_FORMAT_MOD_BROADCOM_SAND64,
+		DRM_FORMAT_MOD_BROADCOM_SAND256,
+		DRM_FORMAT_MOD_LINEAR,
+		DRM_FORMAT_MOD_INVALID
+	};
+
+	vc4_plane = devm_kzalloc(dev->dev, sizeof(*vc4_plane),
+				 GFP_KERNEL);
+	if (!vc4_plane)
+		return ERR_PTR(-ENOMEM);
+
+	for (i = 0; i < ARRAY_SIZE(hvs_formats); i++) {
+		/* Don't allow YUV in cursor planes, since that means
+		 * tuning on the scaler, which we don't allow for the
+		 * cursor.
+		 */
+		if (type != DRM_PLANE_TYPE_CURSOR ||
+		    hvs_formats[i].hvs < HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE) {
+			formats[num_formats++] = hvs_formats[i].drm;
+		}
+	}
+	plane = &vc4_plane->base;
+	ret = drm_universal_plane_init(dev, plane, 0,
+				       &vc4_plane_funcs,
+				       formats, num_formats,
+				       modifiers, type, NULL);
+
+	drm_plane_helper_add(plane, &vc4_plane_helper_funcs);
+
+	drm_plane_create_alpha_property(plane);
+
+	return plane;
+}
diff --git a/drivers/gpu/drm/vc4/vc4_qpu_defines.h b/drivers/gpu/drm/vc4/vc4_qpu_defines.h
new file mode 100644
index 000000000..f4e795a0d
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_qpu_defines.h
@@ -0,0 +1,279 @@
+/*
+ * Copyright © 2014 Broadcom
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#ifndef VC4_QPU_DEFINES_H
+#define VC4_QPU_DEFINES_H
+
+enum qpu_op_add {
+	QPU_A_NOP,
+	QPU_A_FADD,
+	QPU_A_FSUB,
+	QPU_A_FMIN,
+	QPU_A_FMAX,
+	QPU_A_FMINABS,
+	QPU_A_FMAXABS,
+	QPU_A_FTOI,
+	QPU_A_ITOF,
+	QPU_A_ADD = 12,
+	QPU_A_SUB,
+	QPU_A_SHR,
+	QPU_A_ASR,
+	QPU_A_ROR,
+	QPU_A_SHL,
+	QPU_A_MIN,
+	QPU_A_MAX,
+	QPU_A_AND,
+	QPU_A_OR,
+	QPU_A_XOR,
+	QPU_A_NOT,
+	QPU_A_CLZ,
+	QPU_A_V8ADDS = 30,
+	QPU_A_V8SUBS = 31,
+};
+
+enum qpu_op_mul {
+	QPU_M_NOP,
+	QPU_M_FMUL,
+	QPU_M_MUL24,
+	QPU_M_V8MULD,
+	QPU_M_V8MIN,
+	QPU_M_V8MAX,
+	QPU_M_V8ADDS,
+	QPU_M_V8SUBS,
+};
+
+enum qpu_raddr {
+	QPU_R_FRAG_PAYLOAD_ZW = 15, /* W for A file, Z for B file */
+	/* 0-31 are the plain regfile a or b fields */
+	QPU_R_UNIF = 32,
+	QPU_R_VARY = 35,
+	QPU_R_ELEM_QPU = 38,
+	QPU_R_NOP,
+	QPU_R_XY_PIXEL_COORD = 41,
+	QPU_R_MS_REV_FLAGS = 42,
+	QPU_R_VPM = 48,
+	QPU_R_VPM_LD_BUSY,
+	QPU_R_VPM_LD_WAIT,
+	QPU_R_MUTEX_ACQUIRE,
+};
+
+enum qpu_waddr {
+	/* 0-31 are the plain regfile a or b fields */
+	QPU_W_ACC0 = 32, /* aka r0 */
+	QPU_W_ACC1,
+	QPU_W_ACC2,
+	QPU_W_ACC3,
+	QPU_W_TMU_NOSWAP,
+	QPU_W_ACC5,
+	QPU_W_HOST_INT,
+	QPU_W_NOP,
+	QPU_W_UNIFORMS_ADDRESS,
+	QPU_W_QUAD_XY, /* X for regfile a, Y for regfile b */
+	QPU_W_MS_FLAGS = 42,
+	QPU_W_REV_FLAG = 42,
+	QPU_W_TLB_STENCIL_SETUP = 43,
+	QPU_W_TLB_Z,
+	QPU_W_TLB_COLOR_MS,
+	QPU_W_TLB_COLOR_ALL,
+	QPU_W_TLB_ALPHA_MASK,
+	QPU_W_VPM,
+	QPU_W_VPMVCD_SETUP, /* LD for regfile a, ST for regfile b */
+	QPU_W_VPM_ADDR, /* LD for regfile a, ST for regfile b */
+	QPU_W_MUTEX_RELEASE,
+	QPU_W_SFU_RECIP,
+	QPU_W_SFU_RECIPSQRT,
+	QPU_W_SFU_EXP,
+	QPU_W_SFU_LOG,
+	QPU_W_TMU0_S,
+	QPU_W_TMU0_T,
+	QPU_W_TMU0_R,
+	QPU_W_TMU0_B,
+	QPU_W_TMU1_S,
+	QPU_W_TMU1_T,
+	QPU_W_TMU1_R,
+	QPU_W_TMU1_B,
+};
+
+enum qpu_sig_bits {
+	QPU_SIG_SW_BREAKPOINT,
+	QPU_SIG_NONE,
+	QPU_SIG_THREAD_SWITCH,
+	QPU_SIG_PROG_END,
+	QPU_SIG_WAIT_FOR_SCOREBOARD,
+	QPU_SIG_SCOREBOARD_UNLOCK,
+	QPU_SIG_LAST_THREAD_SWITCH,
+	QPU_SIG_COVERAGE_LOAD,
+	QPU_SIG_COLOR_LOAD,
+	QPU_SIG_COLOR_LOAD_END,
+	QPU_SIG_LOAD_TMU0,
+	QPU_SIG_LOAD_TMU1,
+	QPU_SIG_ALPHA_MASK_LOAD,
+	QPU_SIG_SMALL_IMM,
+	QPU_SIG_LOAD_IMM,
+	QPU_SIG_BRANCH
+};
+
+enum qpu_mux {
+	/* hardware mux values */
+	QPU_MUX_R0,
+	QPU_MUX_R1,
+	QPU_MUX_R2,
+	QPU_MUX_R3,
+	QPU_MUX_R4,
+	QPU_MUX_R5,
+	QPU_MUX_A,
+	QPU_MUX_B,
+
+	/* non-hardware mux values */
+	QPU_MUX_IMM,
+};
+
+enum qpu_cond {
+	QPU_COND_NEVER,
+	QPU_COND_ALWAYS,
+	QPU_COND_ZS,
+	QPU_COND_ZC,
+	QPU_COND_NS,
+	QPU_COND_NC,
+	QPU_COND_CS,
+	QPU_COND_CC,
+};
+
+enum qpu_pack_mul {
+	QPU_PACK_MUL_NOP,
+	/* replicated to each 8 bits of the 32-bit dst. */
+	QPU_PACK_MUL_8888 = 3,
+	QPU_PACK_MUL_8A,
+	QPU_PACK_MUL_8B,
+	QPU_PACK_MUL_8C,
+	QPU_PACK_MUL_8D,
+};
+
+enum qpu_pack_a {
+	QPU_PACK_A_NOP,
+	/* convert to 16 bit float if float input, or to int16. */
+	QPU_PACK_A_16A,
+	QPU_PACK_A_16B,
+	/* replicated to each 8 bits of the 32-bit dst. */
+	QPU_PACK_A_8888,
+	/* Convert to 8-bit unsigned int. */
+	QPU_PACK_A_8A,
+	QPU_PACK_A_8B,
+	QPU_PACK_A_8C,
+	QPU_PACK_A_8D,
+
+	/* Saturating variants of the previous instructions. */
+	QPU_PACK_A_32_SAT, /* int-only */
+	QPU_PACK_A_16A_SAT, /* int or float */
+	QPU_PACK_A_16B_SAT,
+	QPU_PACK_A_8888_SAT,
+	QPU_PACK_A_8A_SAT,
+	QPU_PACK_A_8B_SAT,
+	QPU_PACK_A_8C_SAT,
+	QPU_PACK_A_8D_SAT,
+};
+
+enum qpu_unpack_r4 {
+	QPU_UNPACK_R4_NOP,
+	QPU_UNPACK_R4_F16A_TO_F32,
+	QPU_UNPACK_R4_F16B_TO_F32,
+	QPU_UNPACK_R4_8D_REP,
+	QPU_UNPACK_R4_8A,
+	QPU_UNPACK_R4_8B,
+	QPU_UNPACK_R4_8C,
+	QPU_UNPACK_R4_8D,
+};
+
+#define QPU_MASK(high, low) \
+	((((uint64_t)1 << ((high) - (low) + 1)) - 1) << (low))
+
+#define QPU_GET_FIELD(word, field) \
+	((uint32_t)(((word)  & field ## _MASK) >> field ## _SHIFT))
+
+#define QPU_SIG_SHIFT                   60
+#define QPU_SIG_MASK                    QPU_MASK(63, 60)
+
+#define QPU_UNPACK_SHIFT                57
+#define QPU_UNPACK_MASK                 QPU_MASK(59, 57)
+
+/**
+ * If set, the pack field means PACK_MUL or R4 packing, instead of normal
+ * regfile a packing.
+ */
+#define QPU_PM                          ((uint64_t)1 << 56)
+
+#define QPU_PACK_SHIFT                  52
+#define QPU_PACK_MASK                   QPU_MASK(55, 52)
+
+#define QPU_COND_ADD_SHIFT              49
+#define QPU_COND_ADD_MASK               QPU_MASK(51, 49)
+#define QPU_COND_MUL_SHIFT              46
+#define QPU_COND_MUL_MASK               QPU_MASK(48, 46)
+
+#define QPU_BRANCH_COND_SHIFT           52
+#define QPU_BRANCH_COND_MASK            QPU_MASK(55, 52)
+
+#define QPU_BRANCH_REL                  ((uint64_t)1 << 51)
+#define QPU_BRANCH_REG                  ((uint64_t)1 << 50)
+
+#define QPU_BRANCH_RADDR_A_SHIFT        45
+#define QPU_BRANCH_RADDR_A_MASK         QPU_MASK(49, 45)
+
+#define QPU_SF                          ((uint64_t)1 << 45)
+
+#define QPU_WADDR_ADD_SHIFT             38
+#define QPU_WADDR_ADD_MASK              QPU_MASK(43, 38)
+#define QPU_WADDR_MUL_SHIFT             32
+#define QPU_WADDR_MUL_MASK              QPU_MASK(37, 32)
+
+#define QPU_OP_MUL_SHIFT                29
+#define QPU_OP_MUL_MASK                 QPU_MASK(31, 29)
+
+#define QPU_RADDR_A_SHIFT               18
+#define QPU_RADDR_A_MASK                QPU_MASK(23, 18)
+#define QPU_RADDR_B_SHIFT               12
+#define QPU_RADDR_B_MASK                QPU_MASK(17, 12)
+#define QPU_SMALL_IMM_SHIFT             12
+#define QPU_SMALL_IMM_MASK              QPU_MASK(17, 12)
+
+#define QPU_ADD_A_SHIFT                 9
+#define QPU_ADD_A_MASK                  QPU_MASK(11, 9)
+#define QPU_ADD_B_SHIFT                 6
+#define QPU_ADD_B_MASK                  QPU_MASK(8, 6)
+#define QPU_MUL_A_SHIFT                 3
+#define QPU_MUL_A_MASK                  QPU_MASK(5, 3)
+#define QPU_MUL_B_SHIFT                 0
+#define QPU_MUL_B_MASK                  QPU_MASK(2, 0)
+
+#define QPU_WS                          ((uint64_t)1 << 44)
+
+#define QPU_OP_ADD_SHIFT                24
+#define QPU_OP_ADD_MASK                 QPU_MASK(28, 24)
+
+#define QPU_LOAD_IMM_SHIFT              0
+#define QPU_LOAD_IMM_MASK               QPU_MASK(31, 0)
+
+#define QPU_BRANCH_TARGET_SHIFT         0
+#define QPU_BRANCH_TARGET_MASK          QPU_MASK(31, 0)
+
+#endif /* VC4_QPU_DEFINES_H */
diff --git a/drivers/gpu/drm/vc4/vc4_regs.h b/drivers/gpu/drm/vc4/vc4_regs.h
new file mode 100644
index 000000000..d6864fa4b
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_regs.h
@@ -0,0 +1,1051 @@
+/*
+ *  Copyright © 2014-2015 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef VC4_REGS_H
+#define VC4_REGS_H
+
+#include <linux/bitops.h>
+
+#define VC4_MASK(high, low) ((u32)GENMASK(high, low))
+/* Using the GNU statement expression extension */
+#define VC4_SET_FIELD(value, field)					\
+	({								\
+		uint32_t fieldval = (value) << field##_SHIFT;		\
+		WARN_ON((fieldval & ~field##_MASK) != 0);		\
+		fieldval & field##_MASK;				\
+	 })
+
+#define VC4_GET_FIELD(word, field) (((word) & field##_MASK) >>		\
+				    field##_SHIFT)
+
+#define V3D_IDENT0   0x00000
+# define V3D_EXPECTED_IDENT0 \
+	((2 << 24) | \
+	('V' << 0) | \
+	('3' << 8) | \
+	 ('D' << 16))
+
+#define V3D_IDENT1   0x00004
+/* Multiples of 1kb */
+# define V3D_IDENT1_VPM_SIZE_MASK                      VC4_MASK(31, 28)
+# define V3D_IDENT1_VPM_SIZE_SHIFT                     28
+# define V3D_IDENT1_NSEM_MASK                          VC4_MASK(23, 16)
+# define V3D_IDENT1_NSEM_SHIFT                         16
+# define V3D_IDENT1_TUPS_MASK                          VC4_MASK(15, 12)
+# define V3D_IDENT1_TUPS_SHIFT                         12
+# define V3D_IDENT1_QUPS_MASK                          VC4_MASK(11, 8)
+# define V3D_IDENT1_QUPS_SHIFT                         8
+# define V3D_IDENT1_NSLC_MASK                          VC4_MASK(7, 4)
+# define V3D_IDENT1_NSLC_SHIFT                         4
+# define V3D_IDENT1_REV_MASK                           VC4_MASK(3, 0)
+# define V3D_IDENT1_REV_SHIFT                          0
+
+#define V3D_IDENT2   0x00008
+#define V3D_SCRATCH  0x00010
+#define V3D_L2CACTL  0x00020
+# define V3D_L2CACTL_L2CCLR                            BIT(2)
+# define V3D_L2CACTL_L2CDIS                            BIT(1)
+# define V3D_L2CACTL_L2CENA                            BIT(0)
+
+#define V3D_SLCACTL  0x00024
+# define V3D_SLCACTL_T1CC_MASK                         VC4_MASK(27, 24)
+# define V3D_SLCACTL_T1CC_SHIFT                        24
+# define V3D_SLCACTL_T0CC_MASK                         VC4_MASK(19, 16)
+# define V3D_SLCACTL_T0CC_SHIFT                        16
+# define V3D_SLCACTL_UCC_MASK                          VC4_MASK(11, 8)
+# define V3D_SLCACTL_UCC_SHIFT                         8
+# define V3D_SLCACTL_ICC_MASK                          VC4_MASK(3, 0)
+# define V3D_SLCACTL_ICC_SHIFT                         0
+
+#define V3D_INTCTL   0x00030
+#define V3D_INTENA   0x00034
+#define V3D_INTDIS   0x00038
+# define V3D_INT_SPILLUSE                              BIT(3)
+# define V3D_INT_OUTOMEM                               BIT(2)
+# define V3D_INT_FLDONE                                BIT(1)
+# define V3D_INT_FRDONE                                BIT(0)
+
+#define V3D_CT0CS    0x00100
+#define V3D_CT1CS    0x00104
+#define V3D_CTNCS(n) (V3D_CT0CS + 4 * n)
+# define V3D_CTRSTA      BIT(15)
+# define V3D_CTSEMA      BIT(12)
+# define V3D_CTRTSD      BIT(8)
+# define V3D_CTRUN       BIT(5)
+# define V3D_CTSUBS      BIT(4)
+# define V3D_CTERR       BIT(3)
+# define V3D_CTMODE      BIT(0)
+
+#define V3D_CT0EA    0x00108
+#define V3D_CT1EA    0x0010c
+#define V3D_CTNEA(n) (V3D_CT0EA + 4 * (n))
+#define V3D_CT0CA    0x00110
+#define V3D_CT1CA    0x00114
+#define V3D_CTNCA(n) (V3D_CT0CA + 4 * (n))
+#define V3D_CT00RA0  0x00118
+#define V3D_CT01RA0  0x0011c
+#define V3D_CTNRA0(n) (V3D_CT00RA0 + 4 * (n))
+#define V3D_CT0LC    0x00120
+#define V3D_CT1LC    0x00124
+#define V3D_CTNLC(n) (V3D_CT0LC + 4 * (n))
+#define V3D_CT0PC    0x00128
+#define V3D_CT1PC    0x0012c
+#define V3D_CTNPC(n) (V3D_CT0PC + 4 * (n))
+
+#define V3D_PCS      0x00130
+# define V3D_BMOOM       BIT(8)
+# define V3D_RMBUSY      BIT(3)
+# define V3D_RMACTIVE    BIT(2)
+# define V3D_BMBUSY      BIT(1)
+# define V3D_BMACTIVE    BIT(0)
+
+#define V3D_BFC      0x00134
+#define V3D_RFC      0x00138
+#define V3D_BPCA     0x00300
+#define V3D_BPCS     0x00304
+#define V3D_BPOA     0x00308
+#define V3D_BPOS     0x0030c
+#define V3D_BXCF     0x00310
+#define V3D_SQRSV0   0x00410
+#define V3D_SQRSV1   0x00414
+#define V3D_SQCNTL   0x00418
+#define V3D_SRQPC    0x00430
+#define V3D_SRQUA    0x00434
+#define V3D_SRQUL    0x00438
+#define V3D_SRQCS    0x0043c
+#define V3D_VPACNTL  0x00500
+#define V3D_VPMBASE  0x00504
+#define V3D_PCTRC    0x00670
+#define V3D_PCTRE    0x00674
+# define V3D_PCTRE_EN	BIT(31)
+#define V3D_PCTR(x)  (0x00680 + ((x) * 8))
+#define V3D_PCTRS(x) (0x00684 + ((x) * 8))
+#define V3D_DBGE     0x00f00
+#define V3D_FDBGO    0x00f04
+#define V3D_FDBGB    0x00f08
+#define V3D_FDBGR    0x00f0c
+#define V3D_FDBGS    0x00f10
+#define V3D_ERRSTAT  0x00f20
+
+#define PV_CONTROL				0x00
+# define PV_CONTROL_FORMAT_MASK			VC4_MASK(23, 21)
+# define PV_CONTROL_FORMAT_SHIFT		21
+# define PV_CONTROL_FORMAT_24			0
+# define PV_CONTROL_FORMAT_DSIV_16		1
+# define PV_CONTROL_FORMAT_DSIC_16		2
+# define PV_CONTROL_FORMAT_DSIV_18		3
+# define PV_CONTROL_FORMAT_DSIV_24		4
+
+# define PV_CONTROL_FIFO_LEVEL_MASK		VC4_MASK(20, 15)
+# define PV_CONTROL_FIFO_LEVEL_SHIFT		15
+# define PV_CONTROL_CLR_AT_START		BIT(14)
+# define PV_CONTROL_TRIGGER_UNDERFLOW		BIT(13)
+# define PV_CONTROL_WAIT_HSTART			BIT(12)
+# define PV_CONTROL_PIXEL_REP_MASK		VC4_MASK(5, 4)
+# define PV_CONTROL_PIXEL_REP_SHIFT		4
+# define PV_CONTROL_CLK_SELECT_DSI		0
+# define PV_CONTROL_CLK_SELECT_DPI_SMI_HDMI	1
+# define PV_CONTROL_CLK_SELECT_VEC		2
+# define PV_CONTROL_CLK_SELECT_MASK		VC4_MASK(3, 2)
+# define PV_CONTROL_CLK_SELECT_SHIFT		2
+# define PV_CONTROL_FIFO_CLR			BIT(1)
+# define PV_CONTROL_EN				BIT(0)
+
+#define PV_V_CONTROL				0x04
+# define PV_VCONTROL_ODD_DELAY_MASK		VC4_MASK(22, 6)
+# define PV_VCONTROL_ODD_DELAY_SHIFT		6
+# define PV_VCONTROL_ODD_FIRST			BIT(5)
+# define PV_VCONTROL_INTERLACE			BIT(4)
+# define PV_VCONTROL_DSI			BIT(3)
+# define PV_VCONTROL_COMMAND			BIT(2)
+# define PV_VCONTROL_CONTINUOUS			BIT(1)
+# define PV_VCONTROL_VIDEN			BIT(0)
+
+#define PV_VSYNCD_EVEN				0x08
+
+#define PV_HORZA				0x0c
+# define PV_HORZA_HBP_MASK			VC4_MASK(31, 16)
+# define PV_HORZA_HBP_SHIFT			16
+# define PV_HORZA_HSYNC_MASK			VC4_MASK(15, 0)
+# define PV_HORZA_HSYNC_SHIFT			0
+
+#define PV_HORZB				0x10
+# define PV_HORZB_HFP_MASK			VC4_MASK(31, 16)
+# define PV_HORZB_HFP_SHIFT			16
+# define PV_HORZB_HACTIVE_MASK			VC4_MASK(15, 0)
+# define PV_HORZB_HACTIVE_SHIFT			0
+
+#define PV_VERTA				0x14
+# define PV_VERTA_VBP_MASK			VC4_MASK(31, 16)
+# define PV_VERTA_VBP_SHIFT			16
+# define PV_VERTA_VSYNC_MASK			VC4_MASK(15, 0)
+# define PV_VERTA_VSYNC_SHIFT			0
+
+#define PV_VERTB				0x18
+# define PV_VERTB_VFP_MASK			VC4_MASK(31, 16)
+# define PV_VERTB_VFP_SHIFT			16
+# define PV_VERTB_VACTIVE_MASK			VC4_MASK(15, 0)
+# define PV_VERTB_VACTIVE_SHIFT			0
+
+#define PV_VERTA_EVEN				0x1c
+#define PV_VERTB_EVEN				0x20
+
+#define PV_INTEN				0x24
+#define PV_INTSTAT				0x28
+# define PV_INT_VID_IDLE			BIT(9)
+# define PV_INT_VFP_END				BIT(8)
+# define PV_INT_VFP_START			BIT(7)
+# define PV_INT_VACT_START			BIT(6)
+# define PV_INT_VBP_START			BIT(5)
+# define PV_INT_VSYNC_START			BIT(4)
+# define PV_INT_HFP_START			BIT(3)
+# define PV_INT_HACT_START			BIT(2)
+# define PV_INT_HBP_START			BIT(1)
+# define PV_INT_HSYNC_START			BIT(0)
+
+#define PV_STAT					0x2c
+
+#define PV_HACT_ACT				0x30
+
+#define SCALER_DISPCTRL                         0x00000000
+/* Global register for clock gating the HVS */
+# define SCALER_DISPCTRL_ENABLE			BIT(31)
+# define SCALER_DISPCTRL_DSP2EISLUR		BIT(15)
+# define SCALER_DISPCTRL_DSP1EISLUR		BIT(14)
+# define SCALER_DISPCTRL_DSP3_MUX_MASK		VC4_MASK(19, 18)
+# define SCALER_DISPCTRL_DSP3_MUX_SHIFT		18
+
+/* Enables Display 0 short line and underrun contribution to
+ * SCALER_DISPSTAT_IRQDISP0.  Note that short frame contributions are
+ * always enabled.
+ */
+# define SCALER_DISPCTRL_DSP0EISLUR		BIT(13)
+# define SCALER_DISPCTRL_DSP2EIEOLN		BIT(12)
+# define SCALER_DISPCTRL_DSP2EIEOF		BIT(11)
+# define SCALER_DISPCTRL_DSP1EIEOLN		BIT(10)
+# define SCALER_DISPCTRL_DSP1EIEOF		BIT(9)
+/* Enables Display 0 end-of-line-N contribution to
+ * SCALER_DISPSTAT_IRQDISP0
+ */
+# define SCALER_DISPCTRL_DSP0EIEOLN		BIT(8)
+/* Enables Display 0 EOF contribution to SCALER_DISPSTAT_IRQDISP0 */
+# define SCALER_DISPCTRL_DSP0EIEOF		BIT(7)
+
+# define SCALER_DISPCTRL_SLVRDEIRQ		BIT(6)
+# define SCALER_DISPCTRL_SLVWREIRQ		BIT(5)
+# define SCALER_DISPCTRL_DMAEIRQ		BIT(4)
+# define SCALER_DISPCTRL_DISP2EIRQ		BIT(3)
+# define SCALER_DISPCTRL_DISP1EIRQ		BIT(2)
+/* Enables interrupt generation on the enabled EOF/EOLN/EISLUR
+ * bits and short frames..
+ */
+# define SCALER_DISPCTRL_DISP0EIRQ		BIT(1)
+/* Enables interrupt generation on scaler profiler interrupt. */
+# define SCALER_DISPCTRL_SCLEIRQ		BIT(0)
+
+#define SCALER_DISPSTAT                         0x00000004
+# define SCALER_DISPSTAT_COBLOW2		BIT(29)
+# define SCALER_DISPSTAT_EOLN2			BIT(28)
+# define SCALER_DISPSTAT_ESFRAME2		BIT(27)
+# define SCALER_DISPSTAT_ESLINE2		BIT(26)
+# define SCALER_DISPSTAT_EUFLOW2		BIT(25)
+# define SCALER_DISPSTAT_EOF2			BIT(24)
+
+# define SCALER_DISPSTAT_COBLOW1		BIT(21)
+# define SCALER_DISPSTAT_EOLN1			BIT(20)
+# define SCALER_DISPSTAT_ESFRAME1		BIT(19)
+# define SCALER_DISPSTAT_ESLINE1		BIT(18)
+# define SCALER_DISPSTAT_EUFLOW1		BIT(17)
+# define SCALER_DISPSTAT_EOF1			BIT(16)
+
+# define SCALER_DISPSTAT_RESP_MASK		VC4_MASK(15, 14)
+# define SCALER_DISPSTAT_RESP_SHIFT		14
+# define SCALER_DISPSTAT_RESP_OKAY		0
+# define SCALER_DISPSTAT_RESP_EXOKAY		1
+# define SCALER_DISPSTAT_RESP_SLVERR		2
+# define SCALER_DISPSTAT_RESP_DECERR		3
+
+# define SCALER_DISPSTAT_COBLOW0		BIT(13)
+/* Set when the DISPEOLN line is done compositing. */
+# define SCALER_DISPSTAT_EOLN0			BIT(12)
+/* Set when VSTART is seen but there are still pixels in the current
+ * output line.
+ */
+# define SCALER_DISPSTAT_ESFRAME0		BIT(11)
+/* Set when HSTART is seen but there are still pixels in the current
+ * output line.
+ */
+# define SCALER_DISPSTAT_ESLINE0		BIT(10)
+/* Set when the the downstream tries to read from the display FIFO
+ * while it's empty.
+ */
+# define SCALER_DISPSTAT_EUFLOW0		BIT(9)
+/* Set when the display mode changes from RUN to EOF */
+# define SCALER_DISPSTAT_EOF0			BIT(8)
+
+/* Set on AXI invalid DMA ID error. */
+# define SCALER_DISPSTAT_DMA_ERROR		BIT(7)
+/* Set on AXI slave read decode error */
+# define SCALER_DISPSTAT_IRQSLVRD		BIT(6)
+/* Set on AXI slave write decode error */
+# define SCALER_DISPSTAT_IRQSLVWR		BIT(5)
+/* Set when SCALER_DISPSTAT_DMA_ERROR is set, or
+ * SCALER_DISPSTAT_RESP_ERROR is not SCALER_DISPSTAT_RESP_OKAY.
+ */
+# define SCALER_DISPSTAT_IRQDMA			BIT(4)
+# define SCALER_DISPSTAT_IRQDISP2		BIT(3)
+# define SCALER_DISPSTAT_IRQDISP1		BIT(2)
+/* Set when any of the EOF/EOLN/ESFRAME/ESLINE bits are set and their
+ * corresponding interrupt bit is enabled in DISPCTRL.
+ */
+# define SCALER_DISPSTAT_IRQDISP0		BIT(1)
+/* On read, the profiler interrupt.  On write, clear *all* interrupt bits. */
+# define SCALER_DISPSTAT_IRQSCL			BIT(0)
+
+#define SCALER_DISPID                           0x00000008
+#define SCALER_DISPECTRL                        0x0000000c
+#define SCALER_DISPPROF                         0x00000010
+#define SCALER_DISPDITHER                       0x00000014
+#define SCALER_DISPEOLN                         0x00000018
+#define SCALER_DISPLIST0                        0x00000020
+#define SCALER_DISPLIST1                        0x00000024
+#define SCALER_DISPLIST2                        0x00000028
+#define SCALER_DISPLSTAT                        0x0000002c
+#define SCALER_DISPLISTX(x)			(SCALER_DISPLIST0 +	\
+						 (x) * (SCALER_DISPLIST1 - \
+							SCALER_DISPLIST0))
+
+#define SCALER_DISPLACT0                        0x00000030
+#define SCALER_DISPLACT1                        0x00000034
+#define SCALER_DISPLACT2                        0x00000038
+#define SCALER_DISPLACTX(x)			(SCALER_DISPLACT0 +	\
+						 (x) * (SCALER_DISPLACT1 - \
+							SCALER_DISPLACT0))
+
+#define SCALER_DISPCTRL0                        0x00000040
+# define SCALER_DISPCTRLX_ENABLE		BIT(31)
+# define SCALER_DISPCTRLX_RESET			BIT(30)
+/* Generates a single frame when VSTART is seen and stops at the last
+ * pixel read from the FIFO.
+ */
+# define SCALER_DISPCTRLX_ONESHOT		BIT(29)
+/* Processes a single context in the dlist and then task switch,
+ * instead of an entire line.
+ */
+# define SCALER_DISPCTRLX_ONECTX		BIT(28)
+/* Set to have DISPSLAVE return 2 16bpp pixels and no status data. */
+# define SCALER_DISPCTRLX_FIFO32		BIT(27)
+/* Turns on output to the DISPSLAVE register instead of the normal
+ * FIFO.
+ */
+# define SCALER_DISPCTRLX_FIFOREG		BIT(26)
+
+# define SCALER_DISPCTRLX_WIDTH_MASK		VC4_MASK(23, 12)
+# define SCALER_DISPCTRLX_WIDTH_SHIFT		12
+# define SCALER_DISPCTRLX_HEIGHT_MASK		VC4_MASK(11, 0)
+# define SCALER_DISPCTRLX_HEIGHT_SHIFT		0
+
+#define SCALER_DISPBKGND0                       0x00000044
+# define SCALER_DISPBKGND_AUTOHS		BIT(31)
+# define SCALER_DISPBKGND_INTERLACE		BIT(30)
+# define SCALER_DISPBKGND_GAMMA			BIT(29)
+# define SCALER_DISPBKGND_TESTMODE_MASK		VC4_MASK(28, 25)
+# define SCALER_DISPBKGND_TESTMODE_SHIFT	25
+/* Enables filling the scaler line with the RGB value in the low 24
+ * bits before compositing.  Costs cycles, so should be skipped if
+ * opaque display planes will cover everything.
+ */
+# define SCALER_DISPBKGND_FILL			BIT(24)
+
+#define SCALER_DISPSTAT0                        0x00000048
+# define SCALER_DISPSTATX_MODE_MASK		VC4_MASK(31, 30)
+# define SCALER_DISPSTATX_MODE_SHIFT		30
+# define SCALER_DISPSTATX_MODE_DISABLED		0
+# define SCALER_DISPSTATX_MODE_INIT		1
+# define SCALER_DISPSTATX_MODE_RUN		2
+# define SCALER_DISPSTATX_MODE_EOF		3
+# define SCALER_DISPSTATX_FULL			BIT(29)
+# define SCALER_DISPSTATX_EMPTY			BIT(28)
+# define SCALER_DISPSTATX_FRAME_COUNT_MASK	VC4_MASK(17, 12)
+# define SCALER_DISPSTATX_FRAME_COUNT_SHIFT	12
+# define SCALER_DISPSTATX_LINE_MASK		VC4_MASK(11, 0)
+# define SCALER_DISPSTATX_LINE_SHIFT		0
+
+#define SCALER_DISPBASE0                        0x0000004c
+/* Last pixel in the COB (display FIFO memory) allocated to this HVS
+ * channel.  Must be 4-pixel aligned (and thus 4 pixels less than the
+ * next COB base).
+ */
+# define SCALER_DISPBASEX_TOP_MASK		VC4_MASK(31, 16)
+# define SCALER_DISPBASEX_TOP_SHIFT		16
+/* First pixel in the COB (display FIFO memory) allocated to this HVS
+ * channel.  Must be 4-pixel aligned.
+ */
+# define SCALER_DISPBASEX_BASE_MASK		VC4_MASK(15, 0)
+# define SCALER_DISPBASEX_BASE_SHIFT		0
+
+#define SCALER_DISPCTRL1                        0x00000050
+#define SCALER_DISPBKGND1                       0x00000054
+#define SCALER_DISPBKGNDX(x)			(SCALER_DISPBKGND0 +        \
+						 (x) * (SCALER_DISPBKGND1 - \
+							SCALER_DISPBKGND0))
+#define SCALER_DISPSTAT1                        0x00000058
+#define SCALER_DISPSTATX(x)			(SCALER_DISPSTAT0 +        \
+						 (x) * (SCALER_DISPSTAT1 - \
+							SCALER_DISPSTAT0))
+#define SCALER_DISPBASE1                        0x0000005c
+#define SCALER_DISPBASEX(x)			(SCALER_DISPBASE0 +        \
+						 (x) * (SCALER_DISPBASE1 - \
+							SCALER_DISPBASE0))
+#define SCALER_DISPCTRL2                        0x00000060
+#define SCALER_DISPCTRLX(x)			(SCALER_DISPCTRL0 +        \
+						 (x) * (SCALER_DISPCTRL1 - \
+							SCALER_DISPCTRL0))
+#define SCALER_DISPBKGND2                       0x00000064
+#define SCALER_DISPSTAT2                        0x00000068
+#define SCALER_DISPBASE2                        0x0000006c
+#define SCALER_DISPALPHA2                       0x00000070
+#define SCALER_GAMADDR                          0x00000078
+# define SCALER_GAMADDR_AUTOINC			BIT(31)
+/* Enables all gamma ramp SRAMs, not just those of CRTCs with gamma
+ * enabled.
+ */
+# define SCALER_GAMADDR_SRAMENB			BIT(30)
+
+#define SCALER_OLEDOFFS                         0x00000080
+/* Clamps R to [16,235] and G/B to [16,240]. */
+# define SCALER_OLEDOFFS_YUVCLAMP               BIT(31)
+
+/* Chooses which display FIFO the matrix applies to. */
+# define SCALER_OLEDOFFS_DISPFIFO_MASK          VC4_MASK(25, 24)
+# define SCALER_OLEDOFFS_DISPFIFO_SHIFT         24
+# define SCALER_OLEDOFFS_DISPFIFO_DISABLED      0
+# define SCALER_OLEDOFFS_DISPFIFO_0             1
+# define SCALER_OLEDOFFS_DISPFIFO_1             2
+# define SCALER_OLEDOFFS_DISPFIFO_2             3
+
+/* Offsets are 8-bit 2s-complement. */
+# define SCALER_OLEDOFFS_RED_MASK               VC4_MASK(23, 16)
+# define SCALER_OLEDOFFS_RED_SHIFT              16
+# define SCALER_OLEDOFFS_GREEN_MASK             VC4_MASK(15, 8)
+# define SCALER_OLEDOFFS_GREEN_SHIFT            8
+# define SCALER_OLEDOFFS_BLUE_MASK              VC4_MASK(7, 0)
+# define SCALER_OLEDOFFS_BLUE_SHIFT             0
+
+/* The coefficients are S0.9 fractions. */
+#define SCALER_OLEDCOEF0                        0x00000084
+# define SCALER_OLEDCOEF0_B_TO_R_MASK           VC4_MASK(29, 20)
+# define SCALER_OLEDCOEF0_B_TO_R_SHIFT          20
+# define SCALER_OLEDCOEF0_B_TO_G_MASK           VC4_MASK(19, 10)
+# define SCALER_OLEDCOEF0_B_TO_G_SHIFT          10
+# define SCALER_OLEDCOEF0_B_TO_B_MASK           VC4_MASK(9, 0)
+# define SCALER_OLEDCOEF0_B_TO_B_SHIFT          0
+
+#define SCALER_OLEDCOEF1                        0x00000088
+# define SCALER_OLEDCOEF1_G_TO_R_MASK           VC4_MASK(29, 20)
+# define SCALER_OLEDCOEF1_G_TO_R_SHIFT          20
+# define SCALER_OLEDCOEF1_G_TO_G_MASK           VC4_MASK(19, 10)
+# define SCALER_OLEDCOEF1_G_TO_G_SHIFT          10
+# define SCALER_OLEDCOEF1_G_TO_B_MASK           VC4_MASK(9, 0)
+# define SCALER_OLEDCOEF1_G_TO_B_SHIFT          0
+
+#define SCALER_OLEDCOEF2                        0x0000008c
+# define SCALER_OLEDCOEF2_R_TO_R_MASK           VC4_MASK(29, 20)
+# define SCALER_OLEDCOEF2_R_TO_R_SHIFT          20
+# define SCALER_OLEDCOEF2_R_TO_G_MASK           VC4_MASK(19, 10)
+# define SCALER_OLEDCOEF2_R_TO_G_SHIFT          10
+# define SCALER_OLEDCOEF2_R_TO_B_MASK           VC4_MASK(9, 0)
+# define SCALER_OLEDCOEF2_R_TO_B_SHIFT          0
+
+/* Slave addresses for DMAing from HVS composition output to other
+ * devices.  The top bits are valid only in !FIFO32 mode.
+ */
+#define SCALER_DISPSLAVE0                       0x000000c0
+#define SCALER_DISPSLAVE1                       0x000000c9
+#define SCALER_DISPSLAVE2                       0x000000d0
+# define SCALER_DISPSLAVE_ISSUE_VSTART          BIT(31)
+# define SCALER_DISPSLAVE_ISSUE_HSTART          BIT(30)
+/* Set when the current line has been read and an HSTART is required. */
+# define SCALER_DISPSLAVE_EOL                   BIT(26)
+/* Set when the display FIFO is empty. */
+# define SCALER_DISPSLAVE_EMPTY                 BIT(25)
+/* Set when there is RGB data ready to read. */
+# define SCALER_DISPSLAVE_VALID                 BIT(24)
+# define SCALER_DISPSLAVE_RGB_MASK              VC4_MASK(23, 0)
+# define SCALER_DISPSLAVE_RGB_SHIFT             0
+
+#define SCALER_GAMDATA                          0x000000e0
+#define SCALER_DLIST_START                      0x00002000
+#define SCALER_DLIST_SIZE                       0x00004000
+
+#define VC4_HDMI_CORE_REV			0x000
+
+#define VC4_HDMI_SW_RESET_CONTROL		0x004
+# define VC4_HDMI_SW_RESET_FORMAT_DETECT	BIT(1)
+# define VC4_HDMI_SW_RESET_HDMI			BIT(0)
+
+#define VC4_HDMI_HOTPLUG_INT			0x008
+
+#define VC4_HDMI_HOTPLUG			0x00c
+# define VC4_HDMI_HOTPLUG_CONNECTED		BIT(0)
+
+/* 3 bits per field, where each field maps from that corresponding MAI
+ * bus channel to the given HDMI channel.
+ */
+#define VC4_HDMI_MAI_CHANNEL_MAP		0x090
+
+#define VC4_HDMI_MAI_CONFIG			0x094
+# define VC4_HDMI_MAI_CONFIG_FORMAT_REVERSE		BIT(27)
+# define VC4_HDMI_MAI_CONFIG_BIT_REVERSE		BIT(26)
+# define VC4_HDMI_MAI_CHANNEL_MASK_MASK			VC4_MASK(15, 0)
+# define VC4_HDMI_MAI_CHANNEL_MASK_SHIFT		0
+
+/* Last received format word on the MAI bus. */
+#define VC4_HDMI_MAI_FORMAT			0x098
+
+#define VC4_HDMI_AUDIO_PACKET_CONFIG		0x09c
+# define VC4_HDMI_AUDIO_PACKET_ZERO_DATA_ON_SAMPLE_FLAT		BIT(29)
+# define VC4_HDMI_AUDIO_PACKET_ZERO_DATA_ON_INACTIVE_CHANNELS	BIT(24)
+# define VC4_HDMI_AUDIO_PACKET_FORCE_SAMPLE_PRESENT		BIT(19)
+# define VC4_HDMI_AUDIO_PACKET_FORCE_B_FRAME			BIT(18)
+# define VC4_HDMI_AUDIO_PACKET_B_FRAME_IDENTIFIER_MASK		VC4_MASK(13, 10)
+# define VC4_HDMI_AUDIO_PACKET_B_FRAME_IDENTIFIER_SHIFT		10
+/* If set, then multichannel, otherwise 2 channel. */
+# define VC4_HDMI_AUDIO_PACKET_AUDIO_LAYOUT			BIT(9)
+/* If set, then AUDIO_LAYOUT overrides audio_cea_mask */
+# define VC4_HDMI_AUDIO_PACKET_FORCE_AUDIO_LAYOUT		BIT(8)
+# define VC4_HDMI_AUDIO_PACKET_CEA_MASK_MASK			VC4_MASK(7, 0)
+# define VC4_HDMI_AUDIO_PACKET_CEA_MASK_SHIFT			0
+
+#define VC4_HDMI_RAM_PACKET_CONFIG		0x0a0
+# define VC4_HDMI_RAM_PACKET_ENABLE		BIT(16)
+
+#define VC4_HDMI_RAM_PACKET_STATUS		0x0a4
+
+#define VC4_HDMI_CRP_CFG			0x0a8
+/* When set, the CTS_PERIOD counts based on MAI bus sync pulse instead
+ * of pixel clock.
+ */
+# define VC4_HDMI_CRP_USE_MAI_BUS_SYNC_FOR_CTS	BIT(26)
+/* When set, no CRP packets will be sent. */
+# define VC4_HDMI_CRP_CFG_DISABLE		BIT(25)
+/* If set, generates CTS values based on N, audio clock, and video
+ * clock.  N must be divisible by 128.
+ */
+# define VC4_HDMI_CRP_CFG_EXTERNAL_CTS_EN	BIT(24)
+# define VC4_HDMI_CRP_CFG_N_MASK		VC4_MASK(19, 0)
+# define VC4_HDMI_CRP_CFG_N_SHIFT		0
+
+/* 20-bit fields containing CTS values to be transmitted if !EXTERNAL_CTS_EN */
+#define VC4_HDMI_CTS_0				0x0ac
+#define VC4_HDMI_CTS_1				0x0b0
+/* 20-bit fields containing number of clocks to send CTS0/1 before
+ * switching to the other one.
+ */
+#define VC4_HDMI_CTS_PERIOD_0			0x0b4
+#define VC4_HDMI_CTS_PERIOD_1			0x0b8
+
+#define VC4_HDMI_HORZA				0x0c4
+# define VC4_HDMI_HORZA_VPOS			BIT(14)
+# define VC4_HDMI_HORZA_HPOS			BIT(13)
+/* Horizontal active pixels (hdisplay). */
+# define VC4_HDMI_HORZA_HAP_MASK		VC4_MASK(12, 0)
+# define VC4_HDMI_HORZA_HAP_SHIFT		0
+
+#define VC4_HDMI_HORZB				0x0c8
+/* Horizontal pack porch (htotal - hsync_end). */
+# define VC4_HDMI_HORZB_HBP_MASK		VC4_MASK(29, 20)
+# define VC4_HDMI_HORZB_HBP_SHIFT		20
+/* Horizontal sync pulse (hsync_end - hsync_start). */
+# define VC4_HDMI_HORZB_HSP_MASK		VC4_MASK(19, 10)
+# define VC4_HDMI_HORZB_HSP_SHIFT		10
+/* Horizontal front porch (hsync_start - hdisplay). */
+# define VC4_HDMI_HORZB_HFP_MASK		VC4_MASK(9, 0)
+# define VC4_HDMI_HORZB_HFP_SHIFT		0
+
+#define VC4_HDMI_FIFO_CTL			0x05c
+# define VC4_HDMI_FIFO_CTL_RECENTER_DONE	BIT(14)
+# define VC4_HDMI_FIFO_CTL_USE_EMPTY		BIT(13)
+# define VC4_HDMI_FIFO_CTL_ON_VB		BIT(7)
+# define VC4_HDMI_FIFO_CTL_RECENTER		BIT(6)
+# define VC4_HDMI_FIFO_CTL_FIFO_RESET		BIT(5)
+# define VC4_HDMI_FIFO_CTL_USE_PLL_LOCK		BIT(4)
+# define VC4_HDMI_FIFO_CTL_INV_CLK_XFR		BIT(3)
+# define VC4_HDMI_FIFO_CTL_CAPTURE_PTR		BIT(2)
+# define VC4_HDMI_FIFO_CTL_USE_FULL		BIT(1)
+# define VC4_HDMI_FIFO_CTL_MASTER_SLAVE_N	BIT(0)
+# define VC4_HDMI_FIFO_VALID_WRITE_MASK		0xefff
+
+#define VC4_HDMI_SCHEDULER_CONTROL		0x0c0
+# define VC4_HDMI_SCHEDULER_CONTROL_MANUAL_FORMAT BIT(15)
+# define VC4_HDMI_SCHEDULER_CONTROL_IGNORE_VSYNC_PREDICTS BIT(5)
+# define VC4_HDMI_SCHEDULER_CONTROL_VERT_ALWAYS_KEEPOUT	BIT(3)
+# define VC4_HDMI_SCHEDULER_CONTROL_HDMI_ACTIVE	BIT(1)
+# define VC4_HDMI_SCHEDULER_CONTROL_MODE_HDMI	BIT(0)
+
+#define VC4_HDMI_VERTA0				0x0cc
+#define VC4_HDMI_VERTA1				0x0d4
+/* Vertical sync pulse (vsync_end - vsync_start). */
+# define VC4_HDMI_VERTA_VSP_MASK		VC4_MASK(24, 20)
+# define VC4_HDMI_VERTA_VSP_SHIFT		20
+/* Vertical front porch (vsync_start - vdisplay). */
+# define VC4_HDMI_VERTA_VFP_MASK		VC4_MASK(19, 13)
+# define VC4_HDMI_VERTA_VFP_SHIFT		13
+/* Vertical active lines (vdisplay). */
+# define VC4_HDMI_VERTA_VAL_MASK		VC4_MASK(12, 0)
+# define VC4_HDMI_VERTA_VAL_SHIFT		0
+
+#define VC4_HDMI_VERTB0				0x0d0
+#define VC4_HDMI_VERTB1				0x0d8
+/* Vertical sync pulse offset (for interlaced) */
+# define VC4_HDMI_VERTB_VSPO_MASK		VC4_MASK(21, 9)
+# define VC4_HDMI_VERTB_VSPO_SHIFT		9
+/* Vertical pack porch (vtotal - vsync_end). */
+# define VC4_HDMI_VERTB_VBP_MASK		VC4_MASK(8, 0)
+# define VC4_HDMI_VERTB_VBP_SHIFT		0
+
+#define VC4_HDMI_CEC_CNTRL_1			0x0e8
+/* Set when the transmission has ended. */
+# define VC4_HDMI_CEC_TX_EOM			BIT(31)
+/* If set, transmission was acked on the 1st or 2nd attempt (only one
+ * retry is attempted).  If in continuous mode, this means TX needs to
+ * be filled if !TX_EOM.
+ */
+# define VC4_HDMI_CEC_TX_STATUS_GOOD		BIT(30)
+# define VC4_HDMI_CEC_RX_EOM			BIT(29)
+# define VC4_HDMI_CEC_RX_STATUS_GOOD		BIT(28)
+/* Number of bytes received for the message. */
+# define VC4_HDMI_CEC_REC_WRD_CNT_MASK		VC4_MASK(27, 24)
+# define VC4_HDMI_CEC_REC_WRD_CNT_SHIFT		24
+/* Sets continuous receive mode.  Generates interrupt after each 8
+ * bytes to signal that RX_DATA should be consumed, and at RX_EOM.
+ *
+ * If disabled, maximum 16 bytes will be received (including header),
+ * and interrupt at RX_EOM.  Later bytes will be acked but not put
+ * into the RX_DATA.
+ */
+# define VC4_HDMI_CEC_RX_CONTINUE		BIT(23)
+# define VC4_HDMI_CEC_TX_CONTINUE		BIT(22)
+/* Set this after a CEC interrupt. */
+# define VC4_HDMI_CEC_CLEAR_RECEIVE_OFF		BIT(21)
+/* Starts a TX.  Will wait for appropriate idel time before CEC
+ * activity. Must be cleared in between transmits.
+ */
+# define VC4_HDMI_CEC_START_XMIT_BEGIN		BIT(20)
+# define VC4_HDMI_CEC_MESSAGE_LENGTH_MASK	VC4_MASK(19, 16)
+# define VC4_HDMI_CEC_MESSAGE_LENGTH_SHIFT	16
+/* Device's CEC address */
+# define VC4_HDMI_CEC_ADDR_MASK			VC4_MASK(15, 12)
+# define VC4_HDMI_CEC_ADDR_SHIFT		12
+/* Divides off of HSM clock to generate CEC bit clock. */
+/* With the current defaults the CEC bit clock is 40 kHz = 25 usec */
+# define VC4_HDMI_CEC_DIV_CLK_CNT_MASK		VC4_MASK(11, 0)
+# define VC4_HDMI_CEC_DIV_CLK_CNT_SHIFT		0
+
+/* Set these fields to how many bit clock cycles get to that many
+ * microseconds.
+ */
+#define VC4_HDMI_CEC_CNTRL_2			0x0ec
+# define VC4_HDMI_CEC_CNT_TO_1500_US_MASK	VC4_MASK(30, 24)
+# define VC4_HDMI_CEC_CNT_TO_1500_US_SHIFT	24
+# define VC4_HDMI_CEC_CNT_TO_1300_US_MASK	VC4_MASK(23, 17)
+# define VC4_HDMI_CEC_CNT_TO_1300_US_SHIFT	17
+# define VC4_HDMI_CEC_CNT_TO_800_US_MASK	VC4_MASK(16, 11)
+# define VC4_HDMI_CEC_CNT_TO_800_US_SHIFT	11
+# define VC4_HDMI_CEC_CNT_TO_600_US_MASK	VC4_MASK(10, 5)
+# define VC4_HDMI_CEC_CNT_TO_600_US_SHIFT	5
+# define VC4_HDMI_CEC_CNT_TO_400_US_MASK	VC4_MASK(4, 0)
+# define VC4_HDMI_CEC_CNT_TO_400_US_SHIFT	0
+
+#define VC4_HDMI_CEC_CNTRL_3			0x0f0
+# define VC4_HDMI_CEC_CNT_TO_2750_US_MASK	VC4_MASK(31, 24)
+# define VC4_HDMI_CEC_CNT_TO_2750_US_SHIFT	24
+# define VC4_HDMI_CEC_CNT_TO_2400_US_MASK	VC4_MASK(23, 16)
+# define VC4_HDMI_CEC_CNT_TO_2400_US_SHIFT	16
+# define VC4_HDMI_CEC_CNT_TO_2050_US_MASK	VC4_MASK(15, 8)
+# define VC4_HDMI_CEC_CNT_TO_2050_US_SHIFT	8
+# define VC4_HDMI_CEC_CNT_TO_1700_US_MASK	VC4_MASK(7, 0)
+# define VC4_HDMI_CEC_CNT_TO_1700_US_SHIFT	0
+
+#define VC4_HDMI_CEC_CNTRL_4			0x0f4
+# define VC4_HDMI_CEC_CNT_TO_4300_US_MASK	VC4_MASK(31, 24)
+# define VC4_HDMI_CEC_CNT_TO_4300_US_SHIFT	24
+# define VC4_HDMI_CEC_CNT_TO_3900_US_MASK	VC4_MASK(23, 16)
+# define VC4_HDMI_CEC_CNT_TO_3900_US_SHIFT	16
+# define VC4_HDMI_CEC_CNT_TO_3600_US_MASK	VC4_MASK(15, 8)
+# define VC4_HDMI_CEC_CNT_TO_3600_US_SHIFT	8
+# define VC4_HDMI_CEC_CNT_TO_3500_US_MASK	VC4_MASK(7, 0)
+# define VC4_HDMI_CEC_CNT_TO_3500_US_SHIFT	0
+
+#define VC4_HDMI_CEC_CNTRL_5			0x0f8
+# define VC4_HDMI_CEC_TX_SW_RESET		BIT(27)
+# define VC4_HDMI_CEC_RX_SW_RESET		BIT(26)
+# define VC4_HDMI_CEC_PAD_SW_RESET		BIT(25)
+# define VC4_HDMI_CEC_MUX_TP_OUT_CEC		BIT(24)
+# define VC4_HDMI_CEC_RX_CEC_INT		BIT(23)
+# define VC4_HDMI_CEC_CLK_PRELOAD_MASK		VC4_MASK(22, 16)
+# define VC4_HDMI_CEC_CLK_PRELOAD_SHIFT		16
+# define VC4_HDMI_CEC_CNT_TO_4700_US_MASK	VC4_MASK(15, 8)
+# define VC4_HDMI_CEC_CNT_TO_4700_US_SHIFT	8
+# define VC4_HDMI_CEC_CNT_TO_4500_US_MASK	VC4_MASK(7, 0)
+# define VC4_HDMI_CEC_CNT_TO_4500_US_SHIFT	0
+
+/* Transmit data, first byte is low byte of the 32-bit reg.  MSB of
+ * each byte transmitted first.
+ */
+#define VC4_HDMI_CEC_TX_DATA_1			0x0fc
+#define VC4_HDMI_CEC_TX_DATA_2			0x100
+#define VC4_HDMI_CEC_TX_DATA_3			0x104
+#define VC4_HDMI_CEC_TX_DATA_4			0x108
+#define VC4_HDMI_CEC_RX_DATA_1			0x10c
+#define VC4_HDMI_CEC_RX_DATA_2			0x110
+#define VC4_HDMI_CEC_RX_DATA_3			0x114
+#define VC4_HDMI_CEC_RX_DATA_4			0x118
+
+#define VC4_HDMI_TX_PHY_RESET_CTL		0x2c0
+
+#define VC4_HDMI_TX_PHY_CTL0			0x2c4
+# define VC4_HDMI_TX_PHY_RNG_PWRDN		BIT(25)
+
+/* Interrupt status bits */
+#define VC4_HDMI_CPU_STATUS			0x340
+#define VC4_HDMI_CPU_SET			0x344
+#define VC4_HDMI_CPU_CLEAR			0x348
+# define VC4_HDMI_CPU_CEC			BIT(6)
+# define VC4_HDMI_CPU_HOTPLUG			BIT(0)
+
+#define VC4_HDMI_CPU_MASK_STATUS		0x34c
+#define VC4_HDMI_CPU_MASK_SET			0x350
+#define VC4_HDMI_CPU_MASK_CLEAR			0x354
+
+#define VC4_HDMI_GCP(x)				(0x400 + ((x) * 0x4))
+#define VC4_HDMI_RAM_PACKET(x)			(0x400 + ((x) * 0x24))
+#define VC4_HDMI_PACKET_STRIDE			0x24
+
+#define VC4_HD_M_CTL				0x00c
+/* Debug: Current receive value on the CEC pad. */
+# define VC4_HD_CECRXD				BIT(9)
+/* Debug: Override CEC output to 0. */
+# define VC4_HD_CECOVR				BIT(8)
+# define VC4_HD_M_REGISTER_FILE_STANDBY		(3 << 6)
+# define VC4_HD_M_RAM_STANDBY			(3 << 4)
+# define VC4_HD_M_SW_RST			BIT(2)
+# define VC4_HD_M_ENABLE			BIT(0)
+
+#define VC4_HD_MAI_CTL				0x014
+/* Set when audio stream is received at a slower rate than the
+ * sampling period, so MAI fifo goes empty.  Write 1 to clear.
+ */
+# define VC4_HD_MAI_CTL_DLATE			BIT(15)
+# define VC4_HD_MAI_CTL_BUSY			BIT(14)
+# define VC4_HD_MAI_CTL_CHALIGN			BIT(13)
+# define VC4_HD_MAI_CTL_WHOLSMP			BIT(12)
+# define VC4_HD_MAI_CTL_FULL			BIT(11)
+# define VC4_HD_MAI_CTL_EMPTY			BIT(10)
+# define VC4_HD_MAI_CTL_FLUSH			BIT(9)
+/* If set, MAI bus generates SPDIF (bit 31) parity instead of passing
+ * through.
+ */
+# define VC4_HD_MAI_CTL_PAREN			BIT(8)
+# define VC4_HD_MAI_CTL_CHNUM_MASK		VC4_MASK(7, 4)
+# define VC4_HD_MAI_CTL_CHNUM_SHIFT		4
+# define VC4_HD_MAI_CTL_ENABLE			BIT(3)
+/* Underflow error status bit, write 1 to clear. */
+# define VC4_HD_MAI_CTL_ERRORE			BIT(2)
+/* Overflow error status bit, write 1 to clear. */
+# define VC4_HD_MAI_CTL_ERRORF			BIT(1)
+/* Single-shot reset bit.  Read value is undefined. */
+# define VC4_HD_MAI_CTL_RESET			BIT(0)
+
+#define VC4_HD_MAI_THR				0x018
+# define VC4_HD_MAI_THR_PANICHIGH_MASK		VC4_MASK(29, 24)
+# define VC4_HD_MAI_THR_PANICHIGH_SHIFT		24
+# define VC4_HD_MAI_THR_PANICLOW_MASK		VC4_MASK(21, 16)
+# define VC4_HD_MAI_THR_PANICLOW_SHIFT		16
+# define VC4_HD_MAI_THR_DREQHIGH_MASK		VC4_MASK(13, 8)
+# define VC4_HD_MAI_THR_DREQHIGH_SHIFT		8
+# define VC4_HD_MAI_THR_DREQLOW_MASK		VC4_MASK(5, 0)
+# define VC4_HD_MAI_THR_DREQLOW_SHIFT		0
+
+/* Format header to be placed on the MAI data. Unused. */
+#define VC4_HD_MAI_FMT				0x01c
+
+/* Register for DMAing in audio data to be transported over the MAI
+ * bus to the Falcon core.
+ */
+#define VC4_HD_MAI_DATA				0x020
+
+/* Divider from HDMI HSM clock to MAI serial clock.  Sampling period
+ * converges to N / (M + 1) cycles.
+ */
+#define VC4_HD_MAI_SMP				0x02c
+# define VC4_HD_MAI_SMP_N_MASK			VC4_MASK(31, 8)
+# define VC4_HD_MAI_SMP_N_SHIFT			8
+# define VC4_HD_MAI_SMP_M_MASK			VC4_MASK(7, 0)
+# define VC4_HD_MAI_SMP_M_SHIFT			0
+
+#define VC4_HD_VID_CTL				0x038
+# define VC4_HD_VID_CTL_ENABLE			BIT(31)
+# define VC4_HD_VID_CTL_UNDERFLOW_ENABLE	BIT(30)
+# define VC4_HD_VID_CTL_FRAME_COUNTER_RESET	BIT(29)
+# define VC4_HD_VID_CTL_VSYNC_LOW		BIT(28)
+# define VC4_HD_VID_CTL_HSYNC_LOW		BIT(27)
+
+#define VC4_HD_CSC_CTL				0x040
+# define VC4_HD_CSC_CTL_ORDER_MASK		VC4_MASK(7, 5)
+# define VC4_HD_CSC_CTL_ORDER_SHIFT		5
+# define VC4_HD_CSC_CTL_ORDER_RGB		0
+# define VC4_HD_CSC_CTL_ORDER_BGR		1
+# define VC4_HD_CSC_CTL_ORDER_BRG		2
+# define VC4_HD_CSC_CTL_ORDER_GRB		3
+# define VC4_HD_CSC_CTL_ORDER_GBR		4
+# define VC4_HD_CSC_CTL_ORDER_RBG		5
+# define VC4_HD_CSC_CTL_PADMSB			BIT(4)
+# define VC4_HD_CSC_CTL_MODE_MASK		VC4_MASK(3, 2)
+# define VC4_HD_CSC_CTL_MODE_SHIFT		2
+# define VC4_HD_CSC_CTL_MODE_RGB_TO_SD_YPRPB	0
+# define VC4_HD_CSC_CTL_MODE_RGB_TO_HD_YPRPB	1
+# define VC4_HD_CSC_CTL_MODE_CUSTOM		3
+# define VC4_HD_CSC_CTL_RGB2YCC			BIT(1)
+# define VC4_HD_CSC_CTL_ENABLE			BIT(0)
+
+#define VC4_HD_CSC_12_11			0x044
+#define VC4_HD_CSC_14_13			0x048
+#define VC4_HD_CSC_22_21			0x04c
+#define VC4_HD_CSC_24_23			0x050
+#define VC4_HD_CSC_32_31			0x054
+#define VC4_HD_CSC_34_33			0x058
+
+#define VC4_HD_FRAME_COUNT			0x068
+
+/* HVS display list information. */
+#define HVS_BOOTLOADER_DLIST_END                32
+
+enum hvs_pixel_format {
+	/* 8bpp */
+	HVS_PIXEL_FORMAT_RGB332 = 0,
+	/* 16bpp */
+	HVS_PIXEL_FORMAT_RGBA4444 = 1,
+	HVS_PIXEL_FORMAT_RGB555 = 2,
+	HVS_PIXEL_FORMAT_RGBA5551 = 3,
+	HVS_PIXEL_FORMAT_RGB565 = 4,
+	/* 24bpp */
+	HVS_PIXEL_FORMAT_RGB888 = 5,
+	HVS_PIXEL_FORMAT_RGBA6666 = 6,
+	/* 32bpp */
+	HVS_PIXEL_FORMAT_RGBA8888 = 7,
+
+	HVS_PIXEL_FORMAT_YCBCR_YUV420_3PLANE = 8,
+	HVS_PIXEL_FORMAT_YCBCR_YUV420_2PLANE = 9,
+	HVS_PIXEL_FORMAT_YCBCR_YUV422_3PLANE = 10,
+	HVS_PIXEL_FORMAT_YCBCR_YUV422_2PLANE = 11,
+	HVS_PIXEL_FORMAT_H264 = 12,
+	HVS_PIXEL_FORMAT_PALETTE = 13,
+	HVS_PIXEL_FORMAT_YUV444_RGB = 14,
+	HVS_PIXEL_FORMAT_AYUV444_RGB = 15,
+};
+
+/* Note: the LSB is the rightmost character shown.  Only valid for
+ * HVS_PIXEL_FORMAT_RGB8888, not RGB888.
+ */
+#define HVS_PIXEL_ORDER_RGBA			0
+#define HVS_PIXEL_ORDER_BGRA			1
+#define HVS_PIXEL_ORDER_ARGB			2
+#define HVS_PIXEL_ORDER_ABGR			3
+
+#define HVS_PIXEL_ORDER_XBRG			0
+#define HVS_PIXEL_ORDER_XRBG			1
+#define HVS_PIXEL_ORDER_XRGB			2
+#define HVS_PIXEL_ORDER_XBGR			3
+
+#define HVS_PIXEL_ORDER_XYCBCR			0
+#define HVS_PIXEL_ORDER_XYCRCB			1
+#define HVS_PIXEL_ORDER_YXCBCR			2
+#define HVS_PIXEL_ORDER_YXCRCB			3
+
+#define SCALER_CTL0_END				BIT(31)
+#define SCALER_CTL0_VALID			BIT(30)
+
+#define SCALER_CTL0_SIZE_MASK			VC4_MASK(29, 24)
+#define SCALER_CTL0_SIZE_SHIFT			24
+
+#define SCALER_CTL0_TILING_MASK			VC4_MASK(21, 20)
+#define SCALER_CTL0_TILING_SHIFT		20
+#define SCALER_CTL0_TILING_LINEAR		0
+#define SCALER_CTL0_TILING_64B			1
+#define SCALER_CTL0_TILING_128B			2
+#define SCALER_CTL0_TILING_256B_OR_T		3
+
+#define SCALER_CTL0_ALPHA_MASK                  BIT(19)
+#define SCALER_CTL0_HFLIP                       BIT(16)
+#define SCALER_CTL0_VFLIP                       BIT(15)
+
+#define SCALER_CTL0_KEY_MODE_MASK		VC4_MASK(18, 17)
+#define SCALER_CTL0_KEY_MODE_SHIFT		17
+#define SCALER_CTL0_KEY_DISABLED		0
+#define SCALER_CTL0_KEY_LUMA_OR_COMMON_RGB	1
+#define SCALER_CTL0_KEY_MATCH			2 /* turn transparent */
+#define SCALER_CTL0_KEY_REPLACE			3 /* replace with value from key mask word 2 */
+
+#define SCALER_CTL0_ORDER_MASK			VC4_MASK(14, 13)
+#define SCALER_CTL0_ORDER_SHIFT			13
+
+#define SCALER_CTL0_RGBA_EXPAND_MASK		VC4_MASK(12, 11)
+#define SCALER_CTL0_RGBA_EXPAND_SHIFT		11
+#define SCALER_CTL0_RGBA_EXPAND_ZERO		0
+#define SCALER_CTL0_RGBA_EXPAND_LSB		1
+#define SCALER_CTL0_RGBA_EXPAND_MSB		2
+#define SCALER_CTL0_RGBA_EXPAND_ROUND		3
+
+#define SCALER_CTL0_SCL1_MASK			VC4_MASK(10, 8)
+#define SCALER_CTL0_SCL1_SHIFT			8
+
+#define SCALER_CTL0_SCL0_MASK			VC4_MASK(7, 5)
+#define SCALER_CTL0_SCL0_SHIFT			5
+
+#define SCALER_CTL0_SCL_H_PPF_V_PPF		0
+#define SCALER_CTL0_SCL_H_TPZ_V_PPF		1
+#define SCALER_CTL0_SCL_H_PPF_V_TPZ		2
+#define SCALER_CTL0_SCL_H_TPZ_V_TPZ		3
+#define SCALER_CTL0_SCL_H_PPF_V_NONE		4
+#define SCALER_CTL0_SCL_H_NONE_V_PPF		5
+#define SCALER_CTL0_SCL_H_NONE_V_TPZ		6
+#define SCALER_CTL0_SCL_H_TPZ_V_NONE		7
+
+/* Set to indicate no scaling. */
+#define SCALER_CTL0_UNITY			BIT(4)
+
+#define SCALER_CTL0_PIXEL_FORMAT_MASK		VC4_MASK(3, 0)
+#define SCALER_CTL0_PIXEL_FORMAT_SHIFT		0
+
+#define SCALER_POS0_FIXED_ALPHA_MASK		VC4_MASK(31, 24)
+#define SCALER_POS0_FIXED_ALPHA_SHIFT		24
+
+#define SCALER_POS0_START_Y_MASK		VC4_MASK(23, 12)
+#define SCALER_POS0_START_Y_SHIFT		12
+
+#define SCALER_POS0_START_X_MASK		VC4_MASK(11, 0)
+#define SCALER_POS0_START_X_SHIFT		0
+
+#define SCALER_POS1_SCL_HEIGHT_MASK		VC4_MASK(27, 16)
+#define SCALER_POS1_SCL_HEIGHT_SHIFT		16
+
+#define SCALER_POS1_SCL_WIDTH_MASK		VC4_MASK(11, 0)
+#define SCALER_POS1_SCL_WIDTH_SHIFT		0
+
+#define SCALER_POS2_ALPHA_MODE_MASK		VC4_MASK(31, 30)
+#define SCALER_POS2_ALPHA_MODE_SHIFT		30
+#define SCALER_POS2_ALPHA_MODE_PIPELINE		0
+#define SCALER_POS2_ALPHA_MODE_FIXED		1
+#define SCALER_POS2_ALPHA_MODE_FIXED_NONZERO	2
+#define SCALER_POS2_ALPHA_MODE_FIXED_OVER_0x07	3
+#define SCALER_POS2_ALPHA_PREMULT		BIT(29)
+#define SCALER_POS2_ALPHA_MIX			BIT(28)
+
+#define SCALER_POS2_HEIGHT_MASK			VC4_MASK(27, 16)
+#define SCALER_POS2_HEIGHT_SHIFT		16
+
+#define SCALER_POS2_WIDTH_MASK			VC4_MASK(11, 0)
+#define SCALER_POS2_WIDTH_SHIFT			0
+
+/* Color Space Conversion words.  Some values are S2.8 signed
+ * integers, except that the 2 integer bits map as {0x0: 0, 0x1: 1,
+ * 0x2: 2, 0x3: -1}
+ */
+/* bottom 8 bits of S2.8 contribution of Cr to Blue */
+#define SCALER_CSC0_COEF_CR_BLU_MASK		VC4_MASK(31, 24)
+#define SCALER_CSC0_COEF_CR_BLU_SHIFT		24
+/* Signed offset to apply to Y before CSC. (Y' = Y + YY_OFS) */
+#define SCALER_CSC0_COEF_YY_OFS_MASK		VC4_MASK(23, 16)
+#define SCALER_CSC0_COEF_YY_OFS_SHIFT		16
+/* Signed offset to apply to CB before CSC (Cb' = Cb - 128 + CB_OFS). */
+#define SCALER_CSC0_COEF_CB_OFS_MASK		VC4_MASK(15, 8)
+#define SCALER_CSC0_COEF_CB_OFS_SHIFT		8
+/* Signed offset to apply to CB before CSC (Cr' = Cr - 128 + CR_OFS). */
+#define SCALER_CSC0_COEF_CR_OFS_MASK		VC4_MASK(7, 0)
+#define SCALER_CSC0_COEF_CR_OFS_SHIFT		0
+#define SCALER_CSC0_ITR_R_601_5			0x00f00000
+#define SCALER_CSC0_ITR_R_709_3			0x00f00000
+#define SCALER_CSC0_JPEG_JFIF			0x00000000
+
+/* S2.8 contribution of Cb to Green */
+#define SCALER_CSC1_COEF_CB_GRN_MASK		VC4_MASK(31, 22)
+#define SCALER_CSC1_COEF_CB_GRN_SHIFT		22
+/* S2.8 contribution of Cr to Green */
+#define SCALER_CSC1_COEF_CR_GRN_MASK		VC4_MASK(21, 12)
+#define SCALER_CSC1_COEF_CR_GRN_SHIFT		12
+/* S2.8 contribution of Y to all of RGB */
+#define SCALER_CSC1_COEF_YY_ALL_MASK		VC4_MASK(11, 2)
+#define SCALER_CSC1_COEF_YY_ALL_SHIFT		2
+/* top 2 bits of S2.8 contribution of Cr to Blue */
+#define SCALER_CSC1_COEF_CR_BLU_MASK		VC4_MASK(1, 0)
+#define SCALER_CSC1_COEF_CR_BLU_SHIFT		0
+#define SCALER_CSC1_ITR_R_601_5			0xe73304a8
+#define SCALER_CSC1_ITR_R_709_3			0xf2b784a8
+#define SCALER_CSC1_JPEG_JFIF			0xea34a400
+
+/* S2.8 contribution of Cb to Red */
+#define SCALER_CSC2_COEF_CB_RED_MASK		VC4_MASK(29, 20)
+#define SCALER_CSC2_COEF_CB_RED_SHIFT		20
+/* S2.8 contribution of Cr to Red */
+#define SCALER_CSC2_COEF_CR_RED_MASK		VC4_MASK(19, 10)
+#define SCALER_CSC2_COEF_CR_RED_SHIFT		10
+/* S2.8 contribution of Cb to Blue */
+#define SCALER_CSC2_COEF_CB_BLU_MASK		VC4_MASK(19, 10)
+#define SCALER_CSC2_COEF_CB_BLU_SHIFT		10
+#define SCALER_CSC2_ITR_R_601_5			0x00066204
+#define SCALER_CSC2_ITR_R_709_3			0x00072a1c
+#define SCALER_CSC2_JPEG_JFIF			0x000599c5
+
+#define SCALER_TPZ0_VERT_RECALC			BIT(31)
+#define SCALER_TPZ0_SCALE_MASK			VC4_MASK(28, 8)
+#define SCALER_TPZ0_SCALE_SHIFT			8
+#define SCALER_TPZ0_IPHASE_MASK			VC4_MASK(7, 0)
+#define SCALER_TPZ0_IPHASE_SHIFT		0
+#define SCALER_TPZ1_RECIP_MASK			VC4_MASK(15, 0)
+#define SCALER_TPZ1_RECIP_SHIFT			0
+
+/* Skips interpolating coefficients to 64 phases, so just 8 are used.
+ * Required for nearest neighbor.
+ */
+#define SCALER_PPF_NOINTERP			BIT(31)
+/* Replaes the highest valued coefficient with one that makes all 4
+ * sum to unity.
+ */
+#define SCALER_PPF_AGC				BIT(30)
+#define SCALER_PPF_SCALE_MASK			VC4_MASK(24, 8)
+#define SCALER_PPF_SCALE_SHIFT			8
+#define SCALER_PPF_IPHASE_MASK			VC4_MASK(6, 0)
+#define SCALER_PPF_IPHASE_SHIFT			0
+
+#define SCALER_PPF_KERNEL_OFFSET_MASK		VC4_MASK(13, 0)
+#define SCALER_PPF_KERNEL_OFFSET_SHIFT		0
+#define SCALER_PPF_KERNEL_UNCACHED		BIT(31)
+
+/* PITCH0/1/2 fields for raster. */
+#define SCALER_SRC_PITCH_MASK			VC4_MASK(15, 0)
+#define SCALER_SRC_PITCH_SHIFT			0
+
+/* PITCH0/1/2 fields for tiled (SAND). */
+#define SCALER_TILE_SKIP_0_MASK			VC4_MASK(18, 16)
+#define SCALER_TILE_SKIP_0_SHIFT		16
+#define SCALER_TILE_HEIGHT_MASK			VC4_MASK(15, 0)
+#define SCALER_TILE_HEIGHT_SHIFT		0
+
+/* PITCH0 fields for T-tiled. */
+#define SCALER_PITCH0_TILE_WIDTH_L_MASK		VC4_MASK(22, 16)
+#define SCALER_PITCH0_TILE_WIDTH_L_SHIFT	16
+#define SCALER_PITCH0_TILE_LINE_DIR		BIT(15)
+#define SCALER_PITCH0_TILE_INITIAL_LINE_DIR	BIT(14)
+/* Y offset within a tile. */
+#define SCALER_PITCH0_TILE_Y_OFFSET_MASK	VC4_MASK(13, 7)
+#define SCALER_PITCH0_TILE_Y_OFFSET_SHIFT	7
+#define SCALER_PITCH0_TILE_WIDTH_R_MASK		VC4_MASK(6, 0)
+#define SCALER_PITCH0_TILE_WIDTH_R_SHIFT	0
+
+#endif /* VC4_REGS_H */
diff --git a/drivers/gpu/drm/vc4/vc4_render_cl.c b/drivers/gpu/drm/vc4/vc4_render_cl.c
new file mode 100644
index 000000000..273984f71
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_render_cl.c
@@ -0,0 +1,661 @@
+/*
+ * Copyright © 2014-2015 Broadcom
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+/**
+ * DOC: Render command list generation
+ *
+ * In the V3D hardware, render command lists are what load and store
+ * tiles of a framebuffer and optionally call out to binner-generated
+ * command lists to do the 3D drawing for that tile.
+ *
+ * In the VC4 driver, render command list generation is performed by the
+ * kernel instead of userspace.  We do this because validating a
+ * user-submitted command list is hard to get right and has high CPU overhead,
+ * while the number of valid configurations for render command lists is
+ * actually fairly low.
+ */
+
+#include "uapi/drm/vc4_drm.h"
+#include "vc4_drv.h"
+#include "vc4_packet.h"
+
+struct vc4_rcl_setup {
+	struct drm_gem_cma_object *color_read;
+	struct drm_gem_cma_object *color_write;
+	struct drm_gem_cma_object *zs_read;
+	struct drm_gem_cma_object *zs_write;
+	struct drm_gem_cma_object *msaa_color_write;
+	struct drm_gem_cma_object *msaa_zs_write;
+
+	struct drm_gem_cma_object *rcl;
+	u32 next_offset;
+
+	u32 next_write_bo_index;
+};
+
+static inline void rcl_u8(struct vc4_rcl_setup *setup, u8 val)
+{
+	*(u8 *)(setup->rcl->vaddr + setup->next_offset) = val;
+	setup->next_offset += 1;
+}
+
+static inline void rcl_u16(struct vc4_rcl_setup *setup, u16 val)
+{
+	*(u16 *)(setup->rcl->vaddr + setup->next_offset) = val;
+	setup->next_offset += 2;
+}
+
+static inline void rcl_u32(struct vc4_rcl_setup *setup, u32 val)
+{
+	*(u32 *)(setup->rcl->vaddr + setup->next_offset) = val;
+	setup->next_offset += 4;
+}
+
+/*
+ * Emits a no-op STORE_TILE_BUFFER_GENERAL.
+ *
+ * If we emit a PACKET_TILE_COORDINATES, it must be followed by a store of
+ * some sort before another load is triggered.
+ */
+static void vc4_store_before_load(struct vc4_rcl_setup *setup)
+{
+	rcl_u8(setup, VC4_PACKET_STORE_TILE_BUFFER_GENERAL);
+	rcl_u16(setup,
+		VC4_SET_FIELD(VC4_LOADSTORE_TILE_BUFFER_NONE,
+			      VC4_LOADSTORE_TILE_BUFFER_BUFFER) |
+		VC4_STORE_TILE_BUFFER_DISABLE_COLOR_CLEAR |
+		VC4_STORE_TILE_BUFFER_DISABLE_ZS_CLEAR |
+		VC4_STORE_TILE_BUFFER_DISABLE_VG_MASK_CLEAR);
+	rcl_u32(setup, 0); /* no address, since we're in None mode */
+}
+
+/*
+ * Calculates the physical address of the start of a tile in a RCL surface.
+ *
+ * Unlike the other load/store packets,
+ * VC4_PACKET_LOAD/STORE_FULL_RES_TILE_BUFFER don't look at the tile
+ * coordinates packet, and instead just store to the address given.
+ */
+static uint32_t vc4_full_res_offset(struct vc4_exec_info *exec,
+				    struct drm_gem_cma_object *bo,
+				    struct drm_vc4_submit_rcl_surface *surf,
+				    uint8_t x, uint8_t y)
+{
+	return bo->paddr + surf->offset + VC4_TILE_BUFFER_SIZE *
+		(DIV_ROUND_UP(exec->args->width, 32) * y + x);
+}
+
+/*
+ * Emits a PACKET_TILE_COORDINATES if one isn't already pending.
+ *
+ * The tile coordinates packet triggers a pending load if there is one, are
+ * used for clipping during rendering, and determine where loads/stores happen
+ * relative to their base address.
+ */
+static void vc4_tile_coordinates(struct vc4_rcl_setup *setup,
+				 uint32_t x, uint32_t y)
+{
+	rcl_u8(setup, VC4_PACKET_TILE_COORDINATES);
+	rcl_u8(setup, x);
+	rcl_u8(setup, y);
+}
+
+static void emit_tile(struct vc4_exec_info *exec,
+		      struct vc4_rcl_setup *setup,
+		      uint8_t x, uint8_t y, bool first, bool last)
+{
+	struct drm_vc4_submit_cl *args = exec->args;
+	bool has_bin = args->bin_cl_size != 0;
+
+	/* Note that the load doesn't actually occur until the
+	 * tile coords packet is processed, and only one load
+	 * may be outstanding at a time.
+	 */
+	if (setup->color_read) {
+		if (args->color_read.flags &
+		    VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
+			rcl_u8(setup, VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER);
+			rcl_u32(setup,
+				vc4_full_res_offset(exec, setup->color_read,
+						    &args->color_read, x, y) |
+				VC4_LOADSTORE_FULL_RES_DISABLE_ZS);
+		} else {
+			rcl_u8(setup, VC4_PACKET_LOAD_TILE_BUFFER_GENERAL);
+			rcl_u16(setup, args->color_read.bits);
+			rcl_u32(setup, setup->color_read->paddr +
+				args->color_read.offset);
+		}
+	}
+
+	if (setup->zs_read) {
+		if (args->zs_read.flags &
+		    VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
+			rcl_u8(setup, VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER);
+			rcl_u32(setup,
+				vc4_full_res_offset(exec, setup->zs_read,
+						    &args->zs_read, x, y) |
+				VC4_LOADSTORE_FULL_RES_DISABLE_COLOR);
+		} else {
+			if (setup->color_read) {
+				/* Exec previous load. */
+				vc4_tile_coordinates(setup, x, y);
+				vc4_store_before_load(setup);
+			}
+
+			rcl_u8(setup, VC4_PACKET_LOAD_TILE_BUFFER_GENERAL);
+			rcl_u16(setup, args->zs_read.bits);
+			rcl_u32(setup, setup->zs_read->paddr +
+				args->zs_read.offset);
+		}
+	}
+
+	/* Clipping depends on tile coordinates having been
+	 * emitted, so we always need one here.
+	 */
+	vc4_tile_coordinates(setup, x, y);
+
+	/* Wait for the binner before jumping to the first
+	 * tile's lists.
+	 */
+	if (first && has_bin)
+		rcl_u8(setup, VC4_PACKET_WAIT_ON_SEMAPHORE);
+
+	if (has_bin) {
+		rcl_u8(setup, VC4_PACKET_BRANCH_TO_SUB_LIST);
+		rcl_u32(setup, (exec->tile_alloc_offset +
+				(y * exec->bin_tiles_x + x) * 32));
+	}
+
+	if (setup->msaa_color_write) {
+		bool last_tile_write = (!setup->msaa_zs_write &&
+					!setup->zs_write &&
+					!setup->color_write);
+		uint32_t bits = VC4_LOADSTORE_FULL_RES_DISABLE_ZS;
+
+		if (!last_tile_write)
+			bits |= VC4_LOADSTORE_FULL_RES_DISABLE_CLEAR_ALL;
+		else if (last)
+			bits |= VC4_LOADSTORE_FULL_RES_EOF;
+		rcl_u8(setup, VC4_PACKET_STORE_FULL_RES_TILE_BUFFER);
+		rcl_u32(setup,
+			vc4_full_res_offset(exec, setup->msaa_color_write,
+					    &args->msaa_color_write, x, y) |
+			bits);
+	}
+
+	if (setup->msaa_zs_write) {
+		bool last_tile_write = (!setup->zs_write &&
+					!setup->color_write);
+		uint32_t bits = VC4_LOADSTORE_FULL_RES_DISABLE_COLOR;
+
+		if (setup->msaa_color_write)
+			vc4_tile_coordinates(setup, x, y);
+		if (!last_tile_write)
+			bits |= VC4_LOADSTORE_FULL_RES_DISABLE_CLEAR_ALL;
+		else if (last)
+			bits |= VC4_LOADSTORE_FULL_RES_EOF;
+		rcl_u8(setup, VC4_PACKET_STORE_FULL_RES_TILE_BUFFER);
+		rcl_u32(setup,
+			vc4_full_res_offset(exec, setup->msaa_zs_write,
+					    &args->msaa_zs_write, x, y) |
+			bits);
+	}
+
+	if (setup->zs_write) {
+		bool last_tile_write = !setup->color_write;
+
+		if (setup->msaa_color_write || setup->msaa_zs_write)
+			vc4_tile_coordinates(setup, x, y);
+
+		rcl_u8(setup, VC4_PACKET_STORE_TILE_BUFFER_GENERAL);
+		rcl_u16(setup, args->zs_write.bits |
+			(last_tile_write ?
+			 0 : VC4_STORE_TILE_BUFFER_DISABLE_COLOR_CLEAR));
+		rcl_u32(setup,
+			(setup->zs_write->paddr + args->zs_write.offset) |
+			((last && last_tile_write) ?
+			 VC4_LOADSTORE_TILE_BUFFER_EOF : 0));
+	}
+
+	if (setup->color_write) {
+		if (setup->msaa_color_write || setup->msaa_zs_write ||
+		    setup->zs_write) {
+			vc4_tile_coordinates(setup, x, y);
+		}
+
+		if (last)
+			rcl_u8(setup, VC4_PACKET_STORE_MS_TILE_BUFFER_AND_EOF);
+		else
+			rcl_u8(setup, VC4_PACKET_STORE_MS_TILE_BUFFER);
+	}
+}
+
+static int vc4_create_rcl_bo(struct drm_device *dev, struct vc4_exec_info *exec,
+			     struct vc4_rcl_setup *setup)
+{
+	struct drm_vc4_submit_cl *args = exec->args;
+	bool has_bin = args->bin_cl_size != 0;
+	uint8_t min_x_tile = args->min_x_tile;
+	uint8_t min_y_tile = args->min_y_tile;
+	uint8_t max_x_tile = args->max_x_tile;
+	uint8_t max_y_tile = args->max_y_tile;
+	uint8_t xtiles = max_x_tile - min_x_tile + 1;
+	uint8_t ytiles = max_y_tile - min_y_tile + 1;
+	uint8_t xi, yi;
+	uint32_t size, loop_body_size;
+	bool positive_x = true;
+	bool positive_y = true;
+
+	if (args->flags & VC4_SUBMIT_CL_FIXED_RCL_ORDER) {
+		if (!(args->flags & VC4_SUBMIT_CL_RCL_ORDER_INCREASING_X))
+			positive_x = false;
+		if (!(args->flags & VC4_SUBMIT_CL_RCL_ORDER_INCREASING_Y))
+			positive_y = false;
+	}
+
+	size = VC4_PACKET_TILE_RENDERING_MODE_CONFIG_SIZE;
+	loop_body_size = VC4_PACKET_TILE_COORDINATES_SIZE;
+
+	if (args->flags & VC4_SUBMIT_CL_USE_CLEAR_COLOR) {
+		size += VC4_PACKET_CLEAR_COLORS_SIZE +
+			VC4_PACKET_TILE_COORDINATES_SIZE +
+			VC4_PACKET_STORE_TILE_BUFFER_GENERAL_SIZE;
+	}
+
+	if (setup->color_read) {
+		if (args->color_read.flags &
+		    VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
+			loop_body_size += VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER_SIZE;
+		} else {
+			loop_body_size += VC4_PACKET_LOAD_TILE_BUFFER_GENERAL_SIZE;
+		}
+	}
+	if (setup->zs_read) {
+		if (args->zs_read.flags &
+		    VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
+			loop_body_size += VC4_PACKET_LOAD_FULL_RES_TILE_BUFFER_SIZE;
+		} else {
+			if (setup->color_read &&
+			    !(args->color_read.flags &
+			      VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES)) {
+				loop_body_size += VC4_PACKET_TILE_COORDINATES_SIZE;
+				loop_body_size += VC4_PACKET_STORE_TILE_BUFFER_GENERAL_SIZE;
+			}
+			loop_body_size += VC4_PACKET_LOAD_TILE_BUFFER_GENERAL_SIZE;
+		}
+	}
+
+	if (has_bin) {
+		size += VC4_PACKET_WAIT_ON_SEMAPHORE_SIZE;
+		loop_body_size += VC4_PACKET_BRANCH_TO_SUB_LIST_SIZE;
+	}
+
+	if (setup->msaa_color_write)
+		loop_body_size += VC4_PACKET_STORE_FULL_RES_TILE_BUFFER_SIZE;
+	if (setup->msaa_zs_write)
+		loop_body_size += VC4_PACKET_STORE_FULL_RES_TILE_BUFFER_SIZE;
+
+	if (setup->zs_write)
+		loop_body_size += VC4_PACKET_STORE_TILE_BUFFER_GENERAL_SIZE;
+	if (setup->color_write)
+		loop_body_size += VC4_PACKET_STORE_MS_TILE_BUFFER_SIZE;
+
+	/* We need a VC4_PACKET_TILE_COORDINATES in between each store. */
+	loop_body_size += VC4_PACKET_TILE_COORDINATES_SIZE *
+		((setup->msaa_color_write != NULL) +
+		 (setup->msaa_zs_write != NULL) +
+		 (setup->color_write != NULL) +
+		 (setup->zs_write != NULL) - 1);
+
+	size += xtiles * ytiles * loop_body_size;
+
+	setup->rcl = &vc4_bo_create(dev, size, true, VC4_BO_TYPE_RCL)->base;
+	if (IS_ERR(setup->rcl))
+		return PTR_ERR(setup->rcl);
+	list_add_tail(&to_vc4_bo(&setup->rcl->base)->unref_head,
+		      &exec->unref_list);
+
+	/* The tile buffer gets cleared when the previous tile is stored.  If
+	 * the clear values changed between frames, then the tile buffer has
+	 * stale clear values in it, so we have to do a store in None mode (no
+	 * writes) so that we trigger the tile buffer clear.
+	 */
+	if (args->flags & VC4_SUBMIT_CL_USE_CLEAR_COLOR) {
+		rcl_u8(setup, VC4_PACKET_CLEAR_COLORS);
+		rcl_u32(setup, args->clear_color[0]);
+		rcl_u32(setup, args->clear_color[1]);
+		rcl_u32(setup, args->clear_z);
+		rcl_u8(setup, args->clear_s);
+
+		vc4_tile_coordinates(setup, 0, 0);
+
+		rcl_u8(setup, VC4_PACKET_STORE_TILE_BUFFER_GENERAL);
+		rcl_u16(setup, VC4_LOADSTORE_TILE_BUFFER_NONE);
+		rcl_u32(setup, 0); /* no address, since we're in None mode */
+	}
+
+	rcl_u8(setup, VC4_PACKET_TILE_RENDERING_MODE_CONFIG);
+	rcl_u32(setup,
+		(setup->color_write ? (setup->color_write->paddr +
+				       args->color_write.offset) :
+		 0));
+	rcl_u16(setup, args->width);
+	rcl_u16(setup, args->height);
+	rcl_u16(setup, args->color_write.bits);
+
+	for (yi = 0; yi < ytiles; yi++) {
+		int y = positive_y ? min_y_tile + yi : max_y_tile - yi;
+		for (xi = 0; xi < xtiles; xi++) {
+			int x = positive_x ? min_x_tile + xi : max_x_tile - xi;
+			bool first = (xi == 0 && yi == 0);
+			bool last = (xi == xtiles - 1 && yi == ytiles - 1);
+
+			emit_tile(exec, setup, x, y, first, last);
+		}
+	}
+
+	BUG_ON(setup->next_offset != size);
+	exec->ct1ca = setup->rcl->paddr;
+	exec->ct1ea = setup->rcl->paddr + setup->next_offset;
+
+	return 0;
+}
+
+static int vc4_full_res_bounds_check(struct vc4_exec_info *exec,
+				     struct drm_gem_cma_object *obj,
+				     struct drm_vc4_submit_rcl_surface *surf)
+{
+	struct drm_vc4_submit_cl *args = exec->args;
+	u32 render_tiles_stride = DIV_ROUND_UP(exec->args->width, 32);
+
+	if (surf->offset > obj->base.size) {
+		DRM_DEBUG("surface offset %d > BO size %zd\n",
+			  surf->offset, obj->base.size);
+		return -EINVAL;
+	}
+
+	if ((obj->base.size - surf->offset) / VC4_TILE_BUFFER_SIZE <
+	    render_tiles_stride * args->max_y_tile + args->max_x_tile) {
+		DRM_DEBUG("MSAA tile %d, %d out of bounds "
+			  "(bo size %zd, offset %d).\n",
+			  args->max_x_tile, args->max_y_tile,
+			  obj->base.size,
+			  surf->offset);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int vc4_rcl_msaa_surface_setup(struct vc4_exec_info *exec,
+				      struct drm_gem_cma_object **obj,
+				      struct drm_vc4_submit_rcl_surface *surf)
+{
+	if (surf->flags != 0 || surf->bits != 0) {
+		DRM_DEBUG("MSAA surface had nonzero flags/bits\n");
+		return -EINVAL;
+	}
+
+	if (surf->hindex == ~0)
+		return 0;
+
+	*obj = vc4_use_bo(exec, surf->hindex);
+	if (!*obj)
+		return -EINVAL;
+
+	exec->rcl_write_bo[exec->rcl_write_bo_count++] = *obj;
+
+	if (surf->offset & 0xf) {
+		DRM_DEBUG("MSAA write must be 16b aligned.\n");
+		return -EINVAL;
+	}
+
+	return vc4_full_res_bounds_check(exec, *obj, surf);
+}
+
+static int vc4_rcl_surface_setup(struct vc4_exec_info *exec,
+				 struct drm_gem_cma_object **obj,
+				 struct drm_vc4_submit_rcl_surface *surf,
+				 bool is_write)
+{
+	uint8_t tiling = VC4_GET_FIELD(surf->bits,
+				       VC4_LOADSTORE_TILE_BUFFER_TILING);
+	uint8_t buffer = VC4_GET_FIELD(surf->bits,
+				       VC4_LOADSTORE_TILE_BUFFER_BUFFER);
+	uint8_t format = VC4_GET_FIELD(surf->bits,
+				       VC4_LOADSTORE_TILE_BUFFER_FORMAT);
+	int cpp;
+	int ret;
+
+	if (surf->flags & ~VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
+		DRM_DEBUG("Extra flags set\n");
+		return -EINVAL;
+	}
+
+	if (surf->hindex == ~0)
+		return 0;
+
+	*obj = vc4_use_bo(exec, surf->hindex);
+	if (!*obj)
+		return -EINVAL;
+
+	if (is_write)
+		exec->rcl_write_bo[exec->rcl_write_bo_count++] = *obj;
+
+	if (surf->flags & VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES) {
+		if (surf == &exec->args->zs_write) {
+			DRM_DEBUG("general zs write may not be a full-res.\n");
+			return -EINVAL;
+		}
+
+		if (surf->bits != 0) {
+			DRM_DEBUG("load/store general bits set with "
+				  "full res load/store.\n");
+			return -EINVAL;
+		}
+
+		ret = vc4_full_res_bounds_check(exec, *obj, surf);
+		if (ret)
+			return ret;
+
+		return 0;
+	}
+
+	if (surf->bits & ~(VC4_LOADSTORE_TILE_BUFFER_TILING_MASK |
+			   VC4_LOADSTORE_TILE_BUFFER_BUFFER_MASK |
+			   VC4_LOADSTORE_TILE_BUFFER_FORMAT_MASK)) {
+		DRM_DEBUG("Unknown bits in load/store: 0x%04x\n",
+			  surf->bits);
+		return -EINVAL;
+	}
+
+	if (tiling > VC4_TILING_FORMAT_LT) {
+		DRM_DEBUG("Bad tiling format\n");
+		return -EINVAL;
+	}
+
+	if (buffer == VC4_LOADSTORE_TILE_BUFFER_ZS) {
+		if (format != 0) {
+			DRM_DEBUG("No color format should be set for ZS\n");
+			return -EINVAL;
+		}
+		cpp = 4;
+	} else if (buffer == VC4_LOADSTORE_TILE_BUFFER_COLOR) {
+		switch (format) {
+		case VC4_LOADSTORE_TILE_BUFFER_BGR565:
+		case VC4_LOADSTORE_TILE_BUFFER_BGR565_DITHER:
+			cpp = 2;
+			break;
+		case VC4_LOADSTORE_TILE_BUFFER_RGBA8888:
+			cpp = 4;
+			break;
+		default:
+			DRM_DEBUG("Bad tile buffer format\n");
+			return -EINVAL;
+		}
+	} else {
+		DRM_DEBUG("Bad load/store buffer %d.\n", buffer);
+		return -EINVAL;
+	}
+
+	if (surf->offset & 0xf) {
+		DRM_DEBUG("load/store buffer must be 16b aligned.\n");
+		return -EINVAL;
+	}
+
+	if (!vc4_check_tex_size(exec, *obj, surf->offset, tiling,
+				exec->args->width, exec->args->height, cpp)) {
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int
+vc4_rcl_render_config_surface_setup(struct vc4_exec_info *exec,
+				    struct vc4_rcl_setup *setup,
+				    struct drm_gem_cma_object **obj,
+				    struct drm_vc4_submit_rcl_surface *surf)
+{
+	uint8_t tiling = VC4_GET_FIELD(surf->bits,
+				       VC4_RENDER_CONFIG_MEMORY_FORMAT);
+	uint8_t format = VC4_GET_FIELD(surf->bits,
+				       VC4_RENDER_CONFIG_FORMAT);
+	int cpp;
+
+	if (surf->flags != 0) {
+		DRM_DEBUG("No flags supported on render config.\n");
+		return -EINVAL;
+	}
+
+	if (surf->bits & ~(VC4_RENDER_CONFIG_MEMORY_FORMAT_MASK |
+			   VC4_RENDER_CONFIG_FORMAT_MASK |
+			   VC4_RENDER_CONFIG_MS_MODE_4X |
+			   VC4_RENDER_CONFIG_DECIMATE_MODE_4X)) {
+		DRM_DEBUG("Unknown bits in render config: 0x%04x\n",
+			  surf->bits);
+		return -EINVAL;
+	}
+
+	if (surf->hindex == ~0)
+		return 0;
+
+	*obj = vc4_use_bo(exec, surf->hindex);
+	if (!*obj)
+		return -EINVAL;
+
+	exec->rcl_write_bo[exec->rcl_write_bo_count++] = *obj;
+
+	if (tiling > VC4_TILING_FORMAT_LT) {
+		DRM_DEBUG("Bad tiling format\n");
+		return -EINVAL;
+	}
+
+	switch (format) {
+	case VC4_RENDER_CONFIG_FORMAT_BGR565_DITHERED:
+	case VC4_RENDER_CONFIG_FORMAT_BGR565:
+		cpp = 2;
+		break;
+	case VC4_RENDER_CONFIG_FORMAT_RGBA8888:
+		cpp = 4;
+		break;
+	default:
+		DRM_DEBUG("Bad tile buffer format\n");
+		return -EINVAL;
+	}
+
+	if (!vc4_check_tex_size(exec, *obj, surf->offset, tiling,
+				exec->args->width, exec->args->height, cpp)) {
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec)
+{
+	struct vc4_rcl_setup setup = {0};
+	struct drm_vc4_submit_cl *args = exec->args;
+	bool has_bin = args->bin_cl_size != 0;
+	int ret;
+
+	if (args->min_x_tile > args->max_x_tile ||
+	    args->min_y_tile > args->max_y_tile) {
+		DRM_DEBUG("Bad render tile set (%d,%d)-(%d,%d)\n",
+			  args->min_x_tile, args->min_y_tile,
+			  args->max_x_tile, args->max_y_tile);
+		return -EINVAL;
+	}
+
+	if (has_bin &&
+	    (args->max_x_tile > exec->bin_tiles_x ||
+	     args->max_y_tile > exec->bin_tiles_y)) {
+		DRM_DEBUG("Render tiles (%d,%d) outside of bin config "
+			  "(%d,%d)\n",
+			  args->max_x_tile, args->max_y_tile,
+			  exec->bin_tiles_x, exec->bin_tiles_y);
+		return -EINVAL;
+	}
+
+	ret = vc4_rcl_render_config_surface_setup(exec, &setup,
+						  &setup.color_write,
+						  &args->color_write);
+	if (ret)
+		return ret;
+
+	ret = vc4_rcl_surface_setup(exec, &setup.color_read, &args->color_read,
+				    false);
+	if (ret)
+		return ret;
+
+	ret = vc4_rcl_surface_setup(exec, &setup.zs_read, &args->zs_read,
+				    false);
+	if (ret)
+		return ret;
+
+	ret = vc4_rcl_surface_setup(exec, &setup.zs_write, &args->zs_write,
+				    true);
+	if (ret)
+		return ret;
+
+	ret = vc4_rcl_msaa_surface_setup(exec, &setup.msaa_color_write,
+					 &args->msaa_color_write);
+	if (ret)
+		return ret;
+
+	ret = vc4_rcl_msaa_surface_setup(exec, &setup.msaa_zs_write,
+					 &args->msaa_zs_write);
+	if (ret)
+		return ret;
+
+	/* We shouldn't even have the job submitted to us if there's no
+	 * surface to write out.
+	 */
+	if (!setup.color_write && !setup.zs_write &&
+	    !setup.msaa_color_write && !setup.msaa_zs_write) {
+		DRM_DEBUG("RCL requires color or Z/S write\n");
+		return -EINVAL;
+	}
+
+	return vc4_create_rcl_bo(dev, exec, &setup);
+}
diff --git a/drivers/gpu/drm/vc4/vc4_trace.h b/drivers/gpu/drm/vc4/vc4_trace.h
new file mode 100644
index 000000000..deafb3292
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_trace.h
@@ -0,0 +1,63 @@
+/*
+ * Copyright (C) 2015 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#if !defined(_VC4_TRACE_H_) || defined(TRACE_HEADER_MULTI_READ)
+#define _VC4_TRACE_H_
+
+#include <linux/stringify.h>
+#include <linux/types.h>
+#include <linux/tracepoint.h>
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM vc4
+#define TRACE_INCLUDE_FILE vc4_trace
+
+TRACE_EVENT(vc4_wait_for_seqno_begin,
+	    TP_PROTO(struct drm_device *dev, uint64_t seqno, uint64_t timeout),
+	    TP_ARGS(dev, seqno, timeout),
+
+	    TP_STRUCT__entry(
+			     __field(u32, dev)
+			     __field(u64, seqno)
+			     __field(u64, timeout)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->dev = dev->primary->index;
+			   __entry->seqno = seqno;
+			   __entry->timeout = timeout;
+			   ),
+
+	    TP_printk("dev=%u, seqno=%llu, timeout=%llu",
+		      __entry->dev, __entry->seqno, __entry->timeout)
+);
+
+TRACE_EVENT(vc4_wait_for_seqno_end,
+	    TP_PROTO(struct drm_device *dev, uint64_t seqno),
+	    TP_ARGS(dev, seqno),
+
+	    TP_STRUCT__entry(
+			     __field(u32, dev)
+			     __field(u64, seqno)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->dev = dev->primary->index;
+			   __entry->seqno = seqno;
+			   ),
+
+	    TP_printk("dev=%u, seqno=%llu",
+		      __entry->dev, __entry->seqno)
+);
+
+#endif /* _VC4_TRACE_H_ */
+
+/* This part must be outside protection */
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH ../../drivers/gpu/drm/vc4
+#include <trace/define_trace.h>
diff --git a/drivers/gpu/drm/vc4/vc4_trace_points.c b/drivers/gpu/drm/vc4/vc4_trace_points.c
new file mode 100644
index 000000000..e6278f257
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_trace_points.c
@@ -0,0 +1,14 @@
+/*
+ * Copyright (C) 2015 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include "vc4_drv.h"
+
+#ifndef __CHECKER__
+#define CREATE_TRACE_POINTS
+#include "vc4_trace.h"
+#endif
diff --git a/drivers/gpu/drm/vc4/vc4_txp.c b/drivers/gpu/drm/vc4/vc4_txp.c
new file mode 100644
index 000000000..3b56558a5
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_txp.c
@@ -0,0 +1,483 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright © 2018 Broadcom
+ *
+ * Authors:
+ *	Eric Anholt <eric@anholt.net>
+ *	Boris Brezillon <boris.brezillon@bootlin.com>
+ */
+
+#include <drm/drm_atomic_helper.h>
+#include <drm/drm_fb_cma_helper.h>
+#include <drm/drm_crtc_helper.h>
+#include <drm/drm_edid.h>
+#include <drm/drm_panel.h>
+#include <drm/drm_writeback.h>
+#include <linux/clk.h>
+#include <linux/component.h>
+#include <linux/of_graph.h>
+#include <linux/of_platform.h>
+#include <linux/pm_runtime.h>
+
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+/* Base address of the output.  Raster formats must be 4-byte aligned,
+ * T and LT must be 16-byte aligned or maybe utile-aligned (docs are
+ * inconsistent, but probably utile).
+ */
+#define TXP_DST_PTR		0x00
+
+/* Pitch in bytes for raster images, 16-byte aligned.  For tiled, it's
+ * the width in tiles.
+ */
+#define TXP_DST_PITCH		0x04
+/* For T-tiled imgaes, DST_PITCH should be the number of tiles wide,
+ * shifted up.
+ */
+# define TXP_T_TILE_WIDTH_SHIFT		7
+/* For LT-tiled images, DST_PITCH should be the number of utiles wide,
+ * shifted up.
+ */
+# define TXP_LT_TILE_WIDTH_SHIFT	4
+
+/* Pre-rotation width/height of the image.  Must match HVS config.
+ *
+ * If TFORMAT and 32-bit, limit is 1920 for 32-bit and 3840 to 16-bit
+ * and width/height must be tile or utile-aligned as appropriate.  If
+ * transposing (rotating), width is limited to 1920.
+ *
+ * Height is limited to various numbers between 4088 and 4095.  I'd
+ * just use 4088 to be safe.
+ */
+#define TXP_DIM			0x08
+# define TXP_HEIGHT_SHIFT		16
+# define TXP_HEIGHT_MASK		GENMASK(31, 16)
+# define TXP_WIDTH_SHIFT		0
+# define TXP_WIDTH_MASK			GENMASK(15, 0)
+
+#define TXP_DST_CTRL		0x0c
+/* These bits are set to 0x54 */
+#define TXP_PILOT_SHIFT			24
+#define TXP_PILOT_MASK			GENMASK(31, 24)
+/* Bits 22-23 are set to 0x01 */
+#define TXP_VERSION_SHIFT		22
+#define TXP_VERSION_MASK		GENMASK(23, 22)
+
+/* Powers down the internal memory. */
+# define TXP_POWERDOWN			BIT(21)
+
+/* Enables storing the alpha component in 8888/4444, instead of
+ * filling with ~ALPHA_INVERT.
+ */
+# define TXP_ALPHA_ENABLE		BIT(20)
+
+/* 4 bits, each enables stores for a channel in each set of 4 bytes.
+ * Set to 0xf for normal operation.
+ */
+# define TXP_BYTE_ENABLE_SHIFT		16
+# define TXP_BYTE_ENABLE_MASK		GENMASK(19, 16)
+
+/* Debug: Generate VSTART again at EOF. */
+# define TXP_VSTART_AT_EOF		BIT(15)
+
+/* Debug: Terminate the current frame immediately.  Stops AXI
+ * writes.
+ */
+# define TXP_ABORT			BIT(14)
+
+# define TXP_DITHER			BIT(13)
+
+/* Inverts alpha if TXP_ALPHA_ENABLE, chooses fill value for
+ * !TXP_ALPHA_ENABLE.
+ */
+# define TXP_ALPHA_INVERT		BIT(12)
+
+/* Note: I've listed the channels here in high bit (in byte 3/2/1) to
+ * low bit (in byte 0) order.
+ */
+# define TXP_FORMAT_SHIFT		8
+# define TXP_FORMAT_MASK		GENMASK(11, 8)
+# define TXP_FORMAT_ABGR4444		0
+# define TXP_FORMAT_ARGB4444		1
+# define TXP_FORMAT_BGRA4444		2
+# define TXP_FORMAT_RGBA4444		3
+# define TXP_FORMAT_BGR565		6
+# define TXP_FORMAT_RGB565		7
+/* 888s are non-rotated, raster-only */
+# define TXP_FORMAT_BGR888		8
+# define TXP_FORMAT_RGB888		9
+# define TXP_FORMAT_ABGR8888		12
+# define TXP_FORMAT_ARGB8888		13
+# define TXP_FORMAT_BGRA8888		14
+# define TXP_FORMAT_RGBA8888		15
+
+/* If TFORMAT is set, generates LT instead of T format. */
+# define TXP_LINEAR_UTILE		BIT(7)
+
+/* Rotate output by 90 degrees. */
+# define TXP_TRANSPOSE			BIT(6)
+
+/* Generate a tiled format for V3D. */
+# define TXP_TFORMAT			BIT(5)
+
+/* Generates some undefined test mode output. */
+# define TXP_TEST_MODE			BIT(4)
+
+/* Request odd field from HVS. */
+# define TXP_FIELD			BIT(3)
+
+/* Raise interrupt when idle. */
+# define TXP_EI				BIT(2)
+
+/* Set when generating a frame, clears when idle. */
+# define TXP_BUSY			BIT(1)
+
+/* Starts a frame.  Self-clearing. */
+# define TXP_GO				BIT(0)
+
+/* Number of lines received and committed to memory. */
+#define TXP_PROGRESS		0x10
+
+#define TXP_READ(offset) readl(txp->regs + (offset))
+#define TXP_WRITE(offset, val) writel(val, txp->regs + (offset))
+
+struct vc4_txp {
+	struct platform_device *pdev;
+
+	struct drm_writeback_connector connector;
+
+	void __iomem *regs;
+};
+
+static inline struct vc4_txp *encoder_to_vc4_txp(struct drm_encoder *encoder)
+{
+	return container_of(encoder, struct vc4_txp, connector.encoder);
+}
+
+static inline struct vc4_txp *connector_to_vc4_txp(struct drm_connector *conn)
+{
+	return container_of(conn, struct vc4_txp, connector.base);
+}
+
+#define TXP_REG(reg) { reg, #reg }
+static const struct {
+	u32 reg;
+	const char *name;
+} txp_regs[] = {
+	TXP_REG(TXP_DST_PTR),
+	TXP_REG(TXP_DST_PITCH),
+	TXP_REG(TXP_DIM),
+	TXP_REG(TXP_DST_CTRL),
+	TXP_REG(TXP_PROGRESS),
+};
+
+#ifdef CONFIG_DEBUG_FS
+int vc4_txp_debugfs_regs(struct seq_file *m, void *unused)
+{
+	struct drm_info_node *node = (struct drm_info_node *)m->private;
+	struct drm_device *dev = node->minor->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_txp *txp = vc4->txp;
+	int i;
+
+	if (!txp)
+		return 0;
+
+	for (i = 0; i < ARRAY_SIZE(txp_regs); i++) {
+		seq_printf(m, "%s (0x%04x): 0x%08x\n",
+			   txp_regs[i].name, txp_regs[i].reg,
+			   TXP_READ(txp_regs[i].reg));
+	}
+
+	return 0;
+}
+#endif
+
+static int vc4_txp_connector_get_modes(struct drm_connector *connector)
+{
+	struct drm_device *dev = connector->dev;
+
+	return drm_add_modes_noedid(connector, dev->mode_config.max_width,
+				    dev->mode_config.max_height);
+}
+
+static enum drm_mode_status
+vc4_txp_connector_mode_valid(struct drm_connector *connector,
+			     struct drm_display_mode *mode)
+{
+	struct drm_device *dev = connector->dev;
+	struct drm_mode_config *mode_config = &dev->mode_config;
+	int w = mode->hdisplay, h = mode->vdisplay;
+
+	if (w < mode_config->min_width || w > mode_config->max_width)
+		return MODE_BAD_HVALUE;
+
+	if (h < mode_config->min_height || h > mode_config->max_height)
+		return MODE_BAD_VVALUE;
+
+	return MODE_OK;
+}
+
+static const u32 drm_fmts[] = {
+	DRM_FORMAT_RGB888,
+	DRM_FORMAT_BGR888,
+	DRM_FORMAT_XRGB8888,
+	DRM_FORMAT_XBGR8888,
+	DRM_FORMAT_ARGB8888,
+	DRM_FORMAT_ABGR8888,
+	DRM_FORMAT_RGBX8888,
+	DRM_FORMAT_BGRX8888,
+	DRM_FORMAT_RGBA8888,
+	DRM_FORMAT_BGRA8888,
+};
+
+static const u32 txp_fmts[] = {
+	TXP_FORMAT_RGB888,
+	TXP_FORMAT_BGR888,
+	TXP_FORMAT_ARGB8888,
+	TXP_FORMAT_ABGR8888,
+	TXP_FORMAT_ARGB8888,
+	TXP_FORMAT_ABGR8888,
+	TXP_FORMAT_RGBA8888,
+	TXP_FORMAT_BGRA8888,
+	TXP_FORMAT_RGBA8888,
+	TXP_FORMAT_BGRA8888,
+};
+
+static int vc4_txp_connector_atomic_check(struct drm_connector *conn,
+					struct drm_connector_state *conn_state)
+{
+	struct drm_crtc_state *crtc_state;
+	struct drm_gem_cma_object *gem;
+	struct drm_framebuffer *fb;
+	int i;
+
+	if (!conn_state->writeback_job || !conn_state->writeback_job->fb)
+		return 0;
+
+	crtc_state = drm_atomic_get_new_crtc_state(conn_state->state,
+						   conn_state->crtc);
+
+	fb = conn_state->writeback_job->fb;
+	if (fb->width != crtc_state->mode.hdisplay ||
+	    fb->height != crtc_state->mode.vdisplay) {
+		DRM_DEBUG_KMS("Invalid framebuffer size %ux%u\n",
+			      fb->width, fb->height);
+		return -EINVAL;
+	}
+
+	for (i = 0; i < ARRAY_SIZE(drm_fmts); i++) {
+		if (fb->format->format == drm_fmts[i])
+			break;
+	}
+
+	if (i == ARRAY_SIZE(drm_fmts))
+		return -EINVAL;
+
+	gem = drm_fb_cma_get_gem_obj(fb, 0);
+
+	/* Pitch must be aligned on 16 bytes. */
+	if (fb->pitches[0] & GENMASK(3, 0))
+		return -EINVAL;
+
+	vc4_crtc_txp_armed(crtc_state);
+
+	return 0;
+}
+
+static void vc4_txp_connector_atomic_commit(struct drm_connector *conn,
+					struct drm_connector_state *conn_state)
+{
+	struct vc4_txp *txp = connector_to_vc4_txp(conn);
+	struct drm_gem_cma_object *gem;
+	struct drm_display_mode *mode;
+	struct drm_framebuffer *fb;
+	u32 ctrl;
+	int i;
+
+	if (WARN_ON(!conn_state->writeback_job ||
+		    !conn_state->writeback_job->fb))
+		return;
+
+	mode = &conn_state->crtc->state->adjusted_mode;
+	fb = conn_state->writeback_job->fb;
+
+	for (i = 0; i < ARRAY_SIZE(drm_fmts); i++) {
+		if (fb->format->format == drm_fmts[i])
+			break;
+	}
+
+	if (WARN_ON(i == ARRAY_SIZE(drm_fmts)))
+		return;
+
+	ctrl = TXP_GO | TXP_EI |
+	       VC4_SET_FIELD(0xf, TXP_BYTE_ENABLE) |
+	       VC4_SET_FIELD(txp_fmts[i], TXP_FORMAT);
+
+	if (fb->format->has_alpha)
+		ctrl |= TXP_ALPHA_ENABLE;
+	else
+		/*
+		 * If TXP_ALPHA_ENABLE isn't set and TXP_ALPHA_INVERT is, the
+		 * hardware will force the output padding to be 0xff.
+		 */
+		ctrl |= TXP_ALPHA_INVERT;
+
+	gem = drm_fb_cma_get_gem_obj(fb, 0);
+	TXP_WRITE(TXP_DST_PTR, gem->paddr + fb->offsets[0]);
+	TXP_WRITE(TXP_DST_PITCH, fb->pitches[0]);
+	TXP_WRITE(TXP_DIM,
+		  VC4_SET_FIELD(mode->hdisplay, TXP_WIDTH) |
+		  VC4_SET_FIELD(mode->vdisplay, TXP_HEIGHT));
+
+	TXP_WRITE(TXP_DST_CTRL, ctrl);
+
+	drm_writeback_queue_job(&txp->connector, conn_state->writeback_job);
+}
+
+static const struct drm_connector_helper_funcs vc4_txp_connector_helper_funcs = {
+	.get_modes = vc4_txp_connector_get_modes,
+	.mode_valid = vc4_txp_connector_mode_valid,
+	.atomic_check = vc4_txp_connector_atomic_check,
+	.atomic_commit = vc4_txp_connector_atomic_commit,
+};
+
+static enum drm_connector_status
+vc4_txp_connector_detect(struct drm_connector *connector, bool force)
+{
+	return connector_status_connected;
+}
+
+static void vc4_txp_connector_destroy(struct drm_connector *connector)
+{
+	drm_connector_unregister(connector);
+	drm_connector_cleanup(connector);
+}
+
+static const struct drm_connector_funcs vc4_txp_connector_funcs = {
+	.detect = vc4_txp_connector_detect,
+	.fill_modes = drm_helper_probe_single_connector_modes,
+	.destroy = vc4_txp_connector_destroy,
+	.reset = drm_atomic_helper_connector_reset,
+	.atomic_duplicate_state = drm_atomic_helper_connector_duplicate_state,
+	.atomic_destroy_state = drm_atomic_helper_connector_destroy_state,
+};
+
+static void vc4_txp_encoder_disable(struct drm_encoder *encoder)
+{
+	struct vc4_txp *txp = encoder_to_vc4_txp(encoder);
+
+	if (TXP_READ(TXP_DST_CTRL) & TXP_BUSY) {
+		unsigned long timeout = jiffies + msecs_to_jiffies(1000);
+
+		TXP_WRITE(TXP_DST_CTRL, TXP_ABORT);
+
+		while (TXP_READ(TXP_DST_CTRL) & TXP_BUSY &&
+		       time_before(jiffies, timeout))
+			;
+
+		WARN_ON(TXP_READ(TXP_DST_CTRL) & TXP_BUSY);
+	}
+
+	TXP_WRITE(TXP_DST_CTRL, TXP_POWERDOWN);
+}
+
+static const struct drm_encoder_helper_funcs vc4_txp_encoder_helper_funcs = {
+	.disable = vc4_txp_encoder_disable,
+};
+
+static irqreturn_t vc4_txp_interrupt(int irq, void *data)
+{
+	struct vc4_txp *txp = data;
+
+	TXP_WRITE(TXP_DST_CTRL, TXP_READ(TXP_DST_CTRL) & ~TXP_EI);
+	vc4_crtc_handle_vblank(to_vc4_crtc(txp->connector.base.state->crtc));
+	drm_writeback_signal_completion(&txp->connector, 0);
+
+	return IRQ_HANDLED;
+}
+
+static int vc4_txp_bind(struct device *dev, struct device *master, void *data)
+{
+	struct platform_device *pdev = to_platform_device(dev);
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	struct vc4_txp *txp;
+	int ret, irq;
+
+	irq = platform_get_irq(pdev, 0);
+	if (irq < 0)
+		return irq;
+
+	txp = devm_kzalloc(dev, sizeof(*txp), GFP_KERNEL);
+	if (!txp)
+		return -ENOMEM;
+
+	txp->pdev = pdev;
+
+	txp->regs = vc4_ioremap_regs(pdev, 0);
+	if (IS_ERR(txp->regs))
+		return PTR_ERR(txp->regs);
+
+	drm_connector_helper_add(&txp->connector.base,
+				 &vc4_txp_connector_helper_funcs);
+	ret = drm_writeback_connector_init(drm, &txp->connector,
+					   &vc4_txp_connector_funcs,
+					   &vc4_txp_encoder_helper_funcs,
+					   drm_fmts, ARRAY_SIZE(drm_fmts));
+	if (ret)
+		return ret;
+
+	ret = devm_request_irq(dev, irq, vc4_txp_interrupt, 0,
+			       dev_name(dev), txp);
+	if (ret)
+		return ret;
+
+	dev_set_drvdata(dev, txp);
+	vc4->txp = txp;
+
+	return 0;
+}
+
+static void vc4_txp_unbind(struct device *dev, struct device *master,
+			   void *data)
+{
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	struct vc4_txp *txp = dev_get_drvdata(dev);
+
+	vc4_txp_connector_destroy(&txp->connector.base);
+
+	vc4->txp = NULL;
+}
+
+static const struct component_ops vc4_txp_ops = {
+	.bind   = vc4_txp_bind,
+	.unbind = vc4_txp_unbind,
+};
+
+static int vc4_txp_probe(struct platform_device *pdev)
+{
+	return component_add(&pdev->dev, &vc4_txp_ops);
+}
+
+static int vc4_txp_remove(struct platform_device *pdev)
+{
+	component_del(&pdev->dev, &vc4_txp_ops);
+	return 0;
+}
+
+static const struct of_device_id vc4_txp_dt_match[] = {
+	{ .compatible = "brcm,bcm2835-txp" },
+	{ /* sentinel */ },
+};
+
+struct platform_driver vc4_txp_driver = {
+	.probe = vc4_txp_probe,
+	.remove = vc4_txp_remove,
+	.driver = {
+		.name = "vc4_txp",
+		.of_match_table = vc4_txp_dt_match,
+	},
+};
diff --git a/drivers/gpu/drm/vc4/vc4_v3d.c b/drivers/gpu/drm/vc4/vc4_v3d.c
new file mode 100644
index 000000000..e47e29426
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_v3d.c
@@ -0,0 +1,470 @@
+/*
+ * Copyright (c) 2014 The Linux Foundation. All rights reserved.
+ * Copyright (C) 2013 Red Hat
+ * Author: Rob Clark <robdclark@gmail.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published by
+ * the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <linux/clk.h>
+#include <linux/component.h>
+#include <linux/pm_runtime.h>
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+#ifdef CONFIG_DEBUG_FS
+#define REGDEF(reg) { reg, #reg }
+static const struct {
+	uint32_t reg;
+	const char *name;
+} vc4_reg_defs[] = {
+	REGDEF(V3D_IDENT0),
+	REGDEF(V3D_IDENT1),
+	REGDEF(V3D_IDENT2),
+	REGDEF(V3D_SCRATCH),
+	REGDEF(V3D_L2CACTL),
+	REGDEF(V3D_SLCACTL),
+	REGDEF(V3D_INTCTL),
+	REGDEF(V3D_INTENA),
+	REGDEF(V3D_INTDIS),
+	REGDEF(V3D_CT0CS),
+	REGDEF(V3D_CT1CS),
+	REGDEF(V3D_CT0EA),
+	REGDEF(V3D_CT1EA),
+	REGDEF(V3D_CT0CA),
+	REGDEF(V3D_CT1CA),
+	REGDEF(V3D_CT00RA0),
+	REGDEF(V3D_CT01RA0),
+	REGDEF(V3D_CT0LC),
+	REGDEF(V3D_CT1LC),
+	REGDEF(V3D_CT0PC),
+	REGDEF(V3D_CT1PC),
+	REGDEF(V3D_PCS),
+	REGDEF(V3D_BFC),
+	REGDEF(V3D_RFC),
+	REGDEF(V3D_BPCA),
+	REGDEF(V3D_BPCS),
+	REGDEF(V3D_BPOA),
+	REGDEF(V3D_BPOS),
+	REGDEF(V3D_BXCF),
+	REGDEF(V3D_SQRSV0),
+	REGDEF(V3D_SQRSV1),
+	REGDEF(V3D_SQCNTL),
+	REGDEF(V3D_SRQPC),
+	REGDEF(V3D_SRQUA),
+	REGDEF(V3D_SRQUL),
+	REGDEF(V3D_SRQCS),
+	REGDEF(V3D_VPACNTL),
+	REGDEF(V3D_VPMBASE),
+	REGDEF(V3D_PCTRC),
+	REGDEF(V3D_PCTRE),
+	REGDEF(V3D_PCTR(0)),
+	REGDEF(V3D_PCTRS(0)),
+	REGDEF(V3D_PCTR(1)),
+	REGDEF(V3D_PCTRS(1)),
+	REGDEF(V3D_PCTR(2)),
+	REGDEF(V3D_PCTRS(2)),
+	REGDEF(V3D_PCTR(3)),
+	REGDEF(V3D_PCTRS(3)),
+	REGDEF(V3D_PCTR(4)),
+	REGDEF(V3D_PCTRS(4)),
+	REGDEF(V3D_PCTR(5)),
+	REGDEF(V3D_PCTRS(5)),
+	REGDEF(V3D_PCTR(6)),
+	REGDEF(V3D_PCTRS(6)),
+	REGDEF(V3D_PCTR(7)),
+	REGDEF(V3D_PCTRS(7)),
+	REGDEF(V3D_PCTR(8)),
+	REGDEF(V3D_PCTRS(8)),
+	REGDEF(V3D_PCTR(9)),
+	REGDEF(V3D_PCTRS(9)),
+	REGDEF(V3D_PCTR(10)),
+	REGDEF(V3D_PCTRS(10)),
+	REGDEF(V3D_PCTR(11)),
+	REGDEF(V3D_PCTRS(11)),
+	REGDEF(V3D_PCTR(12)),
+	REGDEF(V3D_PCTRS(12)),
+	REGDEF(V3D_PCTR(13)),
+	REGDEF(V3D_PCTRS(13)),
+	REGDEF(V3D_PCTR(14)),
+	REGDEF(V3D_PCTRS(14)),
+	REGDEF(V3D_PCTR(15)),
+	REGDEF(V3D_PCTRS(15)),
+	REGDEF(V3D_DBGE),
+	REGDEF(V3D_FDBGO),
+	REGDEF(V3D_FDBGB),
+	REGDEF(V3D_FDBGR),
+	REGDEF(V3D_FDBGS),
+	REGDEF(V3D_ERRSTAT),
+};
+
+int vc4_v3d_debugfs_regs(struct seq_file *m, void *unused)
+{
+	struct drm_info_node *node = (struct drm_info_node *)m->private;
+	struct drm_device *dev = node->minor->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(vc4_reg_defs); i++) {
+		seq_printf(m, "%s (0x%04x): 0x%08x\n",
+			   vc4_reg_defs[i].name, vc4_reg_defs[i].reg,
+			   V3D_READ(vc4_reg_defs[i].reg));
+	}
+
+	return 0;
+}
+
+int vc4_v3d_debugfs_ident(struct seq_file *m, void *unused)
+{
+	struct drm_info_node *node = (struct drm_info_node *)m->private;
+	struct drm_device *dev = node->minor->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	uint32_t ident1 = V3D_READ(V3D_IDENT1);
+	uint32_t nslc = VC4_GET_FIELD(ident1, V3D_IDENT1_NSLC);
+	uint32_t tups = VC4_GET_FIELD(ident1, V3D_IDENT1_TUPS);
+	uint32_t qups = VC4_GET_FIELD(ident1, V3D_IDENT1_QUPS);
+
+	seq_printf(m, "Revision:   %d\n",
+		   VC4_GET_FIELD(ident1, V3D_IDENT1_REV));
+	seq_printf(m, "Slices:     %d\n", nslc);
+	seq_printf(m, "TMUs:       %d\n", nslc * tups);
+	seq_printf(m, "QPUs:       %d\n", nslc * qups);
+	seq_printf(m, "Semaphores: %d\n",
+		   VC4_GET_FIELD(ident1, V3D_IDENT1_NSEM));
+
+	return 0;
+}
+#endif /* CONFIG_DEBUG_FS */
+
+static void vc4_v3d_init_hw(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+
+	/* Take all the memory that would have been reserved for user
+	 * QPU programs, since we don't have an interface for running
+	 * them, anyway.
+	 */
+	V3D_WRITE(V3D_VPMBASE, 0);
+}
+
+int vc4_v3d_get_bin_slot(struct vc4_dev *vc4)
+{
+	struct drm_device *dev = vc4->dev;
+	unsigned long irqflags;
+	int slot;
+	uint64_t seqno = 0;
+	struct vc4_exec_info *exec;
+
+try_again:
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	slot = ffs(~vc4->bin_alloc_used);
+	if (slot != 0) {
+		/* Switch from ffs() bit index to a 0-based index. */
+		slot--;
+		vc4->bin_alloc_used |= BIT(slot);
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		return slot;
+	}
+
+	/* Couldn't find an open slot.  Wait for render to complete
+	 * and try again.
+	 */
+	exec = vc4_last_render_job(vc4);
+	if (exec)
+		seqno = exec->seqno;
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+
+	if (seqno) {
+		int ret = vc4_wait_for_seqno(dev, seqno, ~0ull, true);
+
+		if (ret == 0)
+			goto try_again;
+
+		return ret;
+	}
+
+	return -ENOMEM;
+}
+
+/**
+ * vc4_allocate_bin_bo() - allocates the memory that will be used for
+ * tile binning.
+ *
+ * The binner has a limitation that the addresses in the tile state
+ * buffer that point into the tile alloc buffer or binner overflow
+ * memory only have 28 bits (256MB), and the top 4 on the bus for
+ * tile alloc references end up coming from the tile state buffer's
+ * address.
+ *
+ * To work around this, we allocate a single large buffer while V3D is
+ * in use, make sure that it has the top 4 bits constant across its
+ * entire extent, and then put the tile state, tile alloc, and binner
+ * overflow memory inside that buffer.
+ *
+ * This creates a limitation where we may not be able to execute a job
+ * if it doesn't fit within the buffer that we allocated up front.
+ * However, it turns out that 16MB is "enough for anybody", and
+ * real-world applications run into allocation failures from the
+ * overall CMA pool before they make scenes complicated enough to run
+ * out of bin space.
+ */
+static int vc4_allocate_bin_bo(struct drm_device *drm)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	struct vc4_v3d *v3d = vc4->v3d;
+	uint32_t size = 16 * 1024 * 1024;
+	int ret = 0;
+	struct list_head list;
+
+	/* We may need to try allocating more than once to get a BO
+	 * that doesn't cross 256MB.  Track the ones we've allocated
+	 * that failed so far, so that we can free them when we've got
+	 * one that succeeded (if we freed them right away, our next
+	 * allocation would probably be the same chunk of memory).
+	 */
+	INIT_LIST_HEAD(&list);
+
+	while (true) {
+		struct vc4_bo *bo = vc4_bo_create(drm, size, true,
+						  VC4_BO_TYPE_BIN);
+
+		if (IS_ERR(bo)) {
+			ret = PTR_ERR(bo);
+
+			dev_err(&v3d->pdev->dev,
+				"Failed to allocate memory for tile binning: "
+				"%d. You may need to enable CMA or give it "
+				"more memory.",
+				ret);
+			break;
+		}
+
+		/* Check if this BO won't trigger the addressing bug. */
+		if ((bo->base.paddr & 0xf0000000) ==
+		    ((bo->base.paddr + bo->base.base.size - 1) & 0xf0000000)) {
+			vc4->bin_bo = bo;
+
+			/* Set up for allocating 512KB chunks of
+			 * binner memory.  The biggest allocation we
+			 * need to do is for the initial tile alloc +
+			 * tile state buffer.  We can render to a
+			 * maximum of ((2048*2048) / (32*32) = 4096
+			 * tiles in a frame (until we do floating
+			 * point rendering, at which point it would be
+			 * 8192).  Tile state is 48b/tile (rounded to
+			 * a page), and tile alloc is 32b/tile
+			 * (rounded to a page), plus a page of extra,
+			 * for a total of 320kb for our worst-case.
+			 * We choose 512kb so that it divides evenly
+			 * into our 16MB, and the rest of the 512kb
+			 * will be used as storage for the overflow
+			 * from the initial 32b CL per bin.
+			 */
+			vc4->bin_alloc_size = 512 * 1024;
+			vc4->bin_alloc_used = 0;
+			vc4->bin_alloc_overflow = 0;
+			WARN_ON_ONCE(sizeof(vc4->bin_alloc_used) * 8 !=
+				     bo->base.base.size / vc4->bin_alloc_size);
+
+			break;
+		}
+
+		/* Put it on the list to free later, and try again. */
+		list_add(&bo->unref_head, &list);
+	}
+
+	/* Free all the BOs we allocated but didn't choose. */
+	while (!list_empty(&list)) {
+		struct vc4_bo *bo = list_last_entry(&list,
+						    struct vc4_bo, unref_head);
+
+		list_del(&bo->unref_head);
+		drm_gem_object_put_unlocked(&bo->base.base);
+	}
+
+	return ret;
+}
+
+#ifdef CONFIG_PM
+static int vc4_v3d_runtime_suspend(struct device *dev)
+{
+	struct vc4_v3d *v3d = dev_get_drvdata(dev);
+	struct vc4_dev *vc4 = v3d->vc4;
+
+	vc4_irq_uninstall(vc4->dev);
+
+	drm_gem_object_put_unlocked(&vc4->bin_bo->base.base);
+	vc4->bin_bo = NULL;
+
+	clk_disable_unprepare(v3d->clk);
+
+	return 0;
+}
+
+static int vc4_v3d_runtime_resume(struct device *dev)
+{
+	struct vc4_v3d *v3d = dev_get_drvdata(dev);
+	struct vc4_dev *vc4 = v3d->vc4;
+	int ret;
+
+	ret = vc4_allocate_bin_bo(vc4->dev);
+	if (ret)
+		return ret;
+
+	ret = clk_prepare_enable(v3d->clk);
+	if (ret != 0)
+		return ret;
+
+	vc4_v3d_init_hw(vc4->dev);
+
+	/* We disabled the IRQ as part of vc4_irq_uninstall in suspend. */
+	enable_irq(vc4->dev->irq);
+	vc4_irq_postinstall(vc4->dev);
+
+	return 0;
+}
+#endif
+
+static int vc4_v3d_bind(struct device *dev, struct device *master, void *data)
+{
+	struct platform_device *pdev = to_platform_device(dev);
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	struct vc4_v3d *v3d = NULL;
+	int ret;
+
+	v3d = devm_kzalloc(&pdev->dev, sizeof(*v3d), GFP_KERNEL);
+	if (!v3d)
+		return -ENOMEM;
+
+	dev_set_drvdata(dev, v3d);
+
+	v3d->pdev = pdev;
+
+	v3d->regs = vc4_ioremap_regs(pdev, 0);
+	if (IS_ERR(v3d->regs))
+		return PTR_ERR(v3d->regs);
+
+	vc4->v3d = v3d;
+	v3d->vc4 = vc4;
+
+	v3d->clk = devm_clk_get(dev, NULL);
+	if (IS_ERR(v3d->clk)) {
+		int ret = PTR_ERR(v3d->clk);
+
+		if (ret == -ENOENT) {
+			/* bcm2835 didn't have a clock reference in the DT. */
+			ret = 0;
+			v3d->clk = NULL;
+		} else {
+			if (ret != -EPROBE_DEFER)
+				dev_err(dev, "Failed to get V3D clock: %d\n",
+					ret);
+			return ret;
+		}
+	}
+
+	if (V3D_READ(V3D_IDENT0) != V3D_EXPECTED_IDENT0) {
+		DRM_ERROR("V3D_IDENT0 read 0x%08x instead of 0x%08x\n",
+			  V3D_READ(V3D_IDENT0), V3D_EXPECTED_IDENT0);
+		return -EINVAL;
+	}
+
+	ret = clk_prepare_enable(v3d->clk);
+	if (ret != 0)
+		return ret;
+
+	ret = vc4_allocate_bin_bo(drm);
+	if (ret) {
+		clk_disable_unprepare(v3d->clk);
+		return ret;
+	}
+
+	/* Reset the binner overflow address/size at setup, to be sure
+	 * we don't reuse an old one.
+	 */
+	V3D_WRITE(V3D_BPOA, 0);
+	V3D_WRITE(V3D_BPOS, 0);
+
+	vc4_v3d_init_hw(drm);
+
+	ret = drm_irq_install(drm, platform_get_irq(pdev, 0));
+	if (ret) {
+		DRM_ERROR("Failed to install IRQ handler\n");
+		return ret;
+	}
+
+	pm_runtime_set_active(dev);
+	pm_runtime_use_autosuspend(dev);
+	pm_runtime_set_autosuspend_delay(dev, 40); /* a little over 2 frames. */
+	pm_runtime_enable(dev);
+
+	return 0;
+}
+
+static void vc4_v3d_unbind(struct device *dev, struct device *master,
+			   void *data)
+{
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+
+	pm_runtime_disable(dev);
+
+	drm_irq_uninstall(drm);
+
+	/* Disable the binner's overflow memory address, so the next
+	 * driver probe (if any) doesn't try to reuse our old
+	 * allocation.
+	 */
+	V3D_WRITE(V3D_BPOA, 0);
+	V3D_WRITE(V3D_BPOS, 0);
+
+	vc4->v3d = NULL;
+}
+
+static const struct dev_pm_ops vc4_v3d_pm_ops = {
+	SET_RUNTIME_PM_OPS(vc4_v3d_runtime_suspend, vc4_v3d_runtime_resume, NULL)
+};
+
+static const struct component_ops vc4_v3d_ops = {
+	.bind   = vc4_v3d_bind,
+	.unbind = vc4_v3d_unbind,
+};
+
+static int vc4_v3d_dev_probe(struct platform_device *pdev)
+{
+	return component_add(&pdev->dev, &vc4_v3d_ops);
+}
+
+static int vc4_v3d_dev_remove(struct platform_device *pdev)
+{
+	component_del(&pdev->dev, &vc4_v3d_ops);
+	return 0;
+}
+
+static const struct of_device_id vc4_v3d_dt_match[] = {
+	{ .compatible = "brcm,bcm2835-v3d" },
+	{ .compatible = "brcm,cygnus-v3d" },
+	{ .compatible = "brcm,vc4-v3d" },
+	{}
+};
+
+struct platform_driver vc4_v3d_driver = {
+	.probe = vc4_v3d_dev_probe,
+	.remove = vc4_v3d_dev_remove,
+	.driver = {
+		.name = "vc4_v3d",
+		.of_match_table = vc4_v3d_dt_match,
+		.pm = &vc4_v3d_pm_ops,
+	},
+};
diff --git a/drivers/gpu/drm/vc4/vc4_validate.c b/drivers/gpu/drm/vc4/vc4_validate.c
new file mode 100644
index 000000000..eec76af49
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_validate.c
@@ -0,0 +1,939 @@
+/*
+ * Copyright © 2014 Broadcom
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+/**
+ * DOC: Command list validator for VC4.
+ *
+ * Since the VC4 has no IOMMU between it and system memory, a user
+ * with access to execute command lists could escalate privilege by
+ * overwriting system memory (drawing to it as a framebuffer) or
+ * reading system memory it shouldn't (reading it as a vertex buffer
+ * or index buffer)
+ *
+ * We validate binner command lists to ensure that all accesses are
+ * within the bounds of the GEM objects referenced by the submitted
+ * job.  It explicitly whitelists packets, and looks at the offsets in
+ * any address fields to make sure they're contained within the BOs
+ * they reference.
+ *
+ * Note that because CL validation is already reading the
+ * user-submitted CL and writing the validated copy out to the memory
+ * that the GPU will actually read, this is also where GEM relocation
+ * processing (turning BO references into actual addresses for the GPU
+ * to use) happens.
+ */
+
+#include "uapi/drm/vc4_drm.h"
+#include "vc4_drv.h"
+#include "vc4_packet.h"
+
+#define VALIDATE_ARGS \
+	struct vc4_exec_info *exec,			\
+	void *validated,				\
+	void *untrusted
+
+/** Return the width in pixels of a 64-byte microtile. */
+static uint32_t
+utile_width(int cpp)
+{
+	switch (cpp) {
+	case 1:
+	case 2:
+		return 8;
+	case 4:
+		return 4;
+	case 8:
+		return 2;
+	default:
+		DRM_ERROR("unknown cpp: %d\n", cpp);
+		return 1;
+	}
+}
+
+/** Return the height in pixels of a 64-byte microtile. */
+static uint32_t
+utile_height(int cpp)
+{
+	switch (cpp) {
+	case 1:
+		return 8;
+	case 2:
+	case 4:
+	case 8:
+		return 4;
+	default:
+		DRM_ERROR("unknown cpp: %d\n", cpp);
+		return 1;
+	}
+}
+
+/**
+ * size_is_lt() - Returns whether a miplevel of the given size will
+ * use the lineartile (LT) tiling layout rather than the normal T
+ * tiling layout.
+ * @width: Width in pixels of the miplevel
+ * @height: Height in pixels of the miplevel
+ * @cpp: Bytes per pixel of the pixel format
+ */
+static bool
+size_is_lt(uint32_t width, uint32_t height, int cpp)
+{
+	return (width <= 4 * utile_width(cpp) ||
+		height <= 4 * utile_height(cpp));
+}
+
+struct drm_gem_cma_object *
+vc4_use_bo(struct vc4_exec_info *exec, uint32_t hindex)
+{
+	struct drm_gem_cma_object *obj;
+	struct vc4_bo *bo;
+
+	if (hindex >= exec->bo_count) {
+		DRM_DEBUG("BO index %d greater than BO count %d\n",
+			  hindex, exec->bo_count);
+		return NULL;
+	}
+	obj = exec->bo[hindex];
+	bo = to_vc4_bo(&obj->base);
+
+	if (bo->validated_shader) {
+		DRM_DEBUG("Trying to use shader BO as something other than "
+			  "a shader\n");
+		return NULL;
+	}
+
+	return obj;
+}
+
+static struct drm_gem_cma_object *
+vc4_use_handle(struct vc4_exec_info *exec, uint32_t gem_handles_packet_index)
+{
+	return vc4_use_bo(exec, exec->bo_index[gem_handles_packet_index]);
+}
+
+static bool
+validate_bin_pos(struct vc4_exec_info *exec, void *untrusted, uint32_t pos)
+{
+	/* Note that the untrusted pointer passed to these functions is
+	 * incremented past the packet byte.
+	 */
+	return (untrusted - 1 == exec->bin_u + pos);
+}
+
+static uint32_t
+gl_shader_rec_size(uint32_t pointer_bits)
+{
+	uint32_t attribute_count = pointer_bits & 7;
+	bool extended = pointer_bits & 8;
+
+	if (attribute_count == 0)
+		attribute_count = 8;
+
+	if (extended)
+		return 100 + attribute_count * 4;
+	else
+		return 36 + attribute_count * 8;
+}
+
+bool
+vc4_check_tex_size(struct vc4_exec_info *exec, struct drm_gem_cma_object *fbo,
+		   uint32_t offset, uint8_t tiling_format,
+		   uint32_t width, uint32_t height, uint8_t cpp)
+{
+	uint32_t aligned_width, aligned_height, stride, size;
+	uint32_t utile_w = utile_width(cpp);
+	uint32_t utile_h = utile_height(cpp);
+
+	/* The shaded vertex format stores signed 12.4 fixed point
+	 * (-2048,2047) offsets from the viewport center, so we should
+	 * never have a render target larger than 4096.  The texture
+	 * unit can only sample from 2048x2048, so it's even more
+	 * restricted.  This lets us avoid worrying about overflow in
+	 * our math.
+	 */
+	if (width > 4096 || height > 4096) {
+		DRM_DEBUG("Surface dimensions (%d,%d) too large",
+			  width, height);
+		return false;
+	}
+
+	switch (tiling_format) {
+	case VC4_TILING_FORMAT_LINEAR:
+		aligned_width = round_up(width, utile_w);
+		aligned_height = height;
+		break;
+	case VC4_TILING_FORMAT_T:
+		aligned_width = round_up(width, utile_w * 8);
+		aligned_height = round_up(height, utile_h * 8);
+		break;
+	case VC4_TILING_FORMAT_LT:
+		aligned_width = round_up(width, utile_w);
+		aligned_height = round_up(height, utile_h);
+		break;
+	default:
+		DRM_DEBUG("buffer tiling %d unsupported\n", tiling_format);
+		return false;
+	}
+
+	stride = aligned_width * cpp;
+	size = stride * aligned_height;
+
+	if (size + offset < size ||
+	    size + offset > fbo->base.size) {
+		DRM_DEBUG("Overflow in %dx%d (%dx%d) fbo size (%d + %d > %zd)\n",
+			  width, height,
+			  aligned_width, aligned_height,
+			  size, offset, fbo->base.size);
+		return false;
+	}
+
+	return true;
+}
+
+static int
+validate_flush(VALIDATE_ARGS)
+{
+	if (!validate_bin_pos(exec, untrusted, exec->args->bin_cl_size - 1)) {
+		DRM_DEBUG("Bin CL must end with VC4_PACKET_FLUSH\n");
+		return -EINVAL;
+	}
+	exec->found_flush = true;
+
+	return 0;
+}
+
+static int
+validate_start_tile_binning(VALIDATE_ARGS)
+{
+	if (exec->found_start_tile_binning_packet) {
+		DRM_DEBUG("Duplicate VC4_PACKET_START_TILE_BINNING\n");
+		return -EINVAL;
+	}
+	exec->found_start_tile_binning_packet = true;
+
+	if (!exec->found_tile_binning_mode_config_packet) {
+		DRM_DEBUG("missing VC4_PACKET_TILE_BINNING_MODE_CONFIG\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int
+validate_increment_semaphore(VALIDATE_ARGS)
+{
+	if (!validate_bin_pos(exec, untrusted, exec->args->bin_cl_size - 2)) {
+		DRM_DEBUG("Bin CL must end with "
+			  "VC4_PACKET_INCREMENT_SEMAPHORE\n");
+		return -EINVAL;
+	}
+	exec->found_increment_semaphore_packet = true;
+
+	return 0;
+}
+
+static int
+validate_indexed_prim_list(VALIDATE_ARGS)
+{
+	struct drm_gem_cma_object *ib;
+	uint32_t length = *(uint32_t *)(untrusted + 1);
+	uint32_t offset = *(uint32_t *)(untrusted + 5);
+	uint32_t max_index = *(uint32_t *)(untrusted + 9);
+	uint32_t index_size = (*(uint8_t *)(untrusted + 0) >> 4) ? 2 : 1;
+	struct vc4_shader_state *shader_state;
+
+	/* Check overflow condition */
+	if (exec->shader_state_count == 0) {
+		DRM_DEBUG("shader state must precede primitives\n");
+		return -EINVAL;
+	}
+	shader_state = &exec->shader_state[exec->shader_state_count - 1];
+
+	if (max_index > shader_state->max_index)
+		shader_state->max_index = max_index;
+
+	ib = vc4_use_handle(exec, 0);
+	if (!ib)
+		return -EINVAL;
+
+	exec->bin_dep_seqno = max(exec->bin_dep_seqno,
+				  to_vc4_bo(&ib->base)->write_seqno);
+
+	if (offset > ib->base.size ||
+	    (ib->base.size - offset) / index_size < length) {
+		DRM_DEBUG("IB access overflow (%d + %d*%d > %zd)\n",
+			  offset, length, index_size, ib->base.size);
+		return -EINVAL;
+	}
+
+	*(uint32_t *)(validated + 5) = ib->paddr + offset;
+
+	return 0;
+}
+
+static int
+validate_gl_array_primitive(VALIDATE_ARGS)
+{
+	uint32_t length = *(uint32_t *)(untrusted + 1);
+	uint32_t base_index = *(uint32_t *)(untrusted + 5);
+	uint32_t max_index;
+	struct vc4_shader_state *shader_state;
+
+	/* Check overflow condition */
+	if (exec->shader_state_count == 0) {
+		DRM_DEBUG("shader state must precede primitives\n");
+		return -EINVAL;
+	}
+	shader_state = &exec->shader_state[exec->shader_state_count - 1];
+
+	if (length + base_index < length) {
+		DRM_DEBUG("primitive vertex count overflow\n");
+		return -EINVAL;
+	}
+	max_index = length + base_index - 1;
+
+	if (max_index > shader_state->max_index)
+		shader_state->max_index = max_index;
+
+	return 0;
+}
+
+static int
+validate_gl_shader_state(VALIDATE_ARGS)
+{
+	uint32_t i = exec->shader_state_count++;
+
+	if (i >= exec->shader_state_size) {
+		DRM_DEBUG("More requests for shader states than declared\n");
+		return -EINVAL;
+	}
+
+	exec->shader_state[i].addr = *(uint32_t *)untrusted;
+	exec->shader_state[i].max_index = 0;
+
+	if (exec->shader_state[i].addr & ~0xf) {
+		DRM_DEBUG("high bits set in GL shader rec reference\n");
+		return -EINVAL;
+	}
+
+	*(uint32_t *)validated = (exec->shader_rec_p +
+				  exec->shader_state[i].addr);
+
+	exec->shader_rec_p +=
+		roundup(gl_shader_rec_size(exec->shader_state[i].addr), 16);
+
+	return 0;
+}
+
+static int
+validate_tile_binning_config(VALIDATE_ARGS)
+{
+	struct drm_device *dev = exec->exec_bo->base.dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	uint8_t flags;
+	uint32_t tile_state_size;
+	uint32_t tile_count, bin_addr;
+	int bin_slot;
+
+	if (exec->found_tile_binning_mode_config_packet) {
+		DRM_DEBUG("Duplicate VC4_PACKET_TILE_BINNING_MODE_CONFIG\n");
+		return -EINVAL;
+	}
+	exec->found_tile_binning_mode_config_packet = true;
+
+	exec->bin_tiles_x = *(uint8_t *)(untrusted + 12);
+	exec->bin_tiles_y = *(uint8_t *)(untrusted + 13);
+	tile_count = exec->bin_tiles_x * exec->bin_tiles_y;
+	flags = *(uint8_t *)(untrusted + 14);
+
+	if (exec->bin_tiles_x == 0 ||
+	    exec->bin_tiles_y == 0) {
+		DRM_DEBUG("Tile binning config of %dx%d too small\n",
+			  exec->bin_tiles_x, exec->bin_tiles_y);
+		return -EINVAL;
+	}
+
+	if (flags & (VC4_BIN_CONFIG_DB_NON_MS |
+		     VC4_BIN_CONFIG_TILE_BUFFER_64BIT)) {
+		DRM_DEBUG("unsupported binning config flags 0x%02x\n", flags);
+		return -EINVAL;
+	}
+
+	bin_slot = vc4_v3d_get_bin_slot(vc4);
+	if (bin_slot < 0) {
+		if (bin_slot != -EINTR && bin_slot != -ERESTARTSYS) {
+			DRM_ERROR("Failed to allocate binner memory: %d\n",
+				  bin_slot);
+		}
+		return bin_slot;
+	}
+
+	/* The slot we allocated will only be used by this job, and is
+	 * free when the job completes rendering.
+	 */
+	exec->bin_slots |= BIT(bin_slot);
+	bin_addr = vc4->bin_bo->base.paddr + bin_slot * vc4->bin_alloc_size;
+
+	/* The tile state data array is 48 bytes per tile, and we put it at
+	 * the start of a BO containing both it and the tile alloc.
+	 */
+	tile_state_size = 48 * tile_count;
+
+	/* Since the tile alloc array will follow us, align. */
+	exec->tile_alloc_offset = bin_addr + roundup(tile_state_size, 4096);
+
+	*(uint8_t *)(validated + 14) =
+		((flags & ~(VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_MASK |
+			    VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_MASK)) |
+		 VC4_BIN_CONFIG_AUTO_INIT_TSDA |
+		 VC4_SET_FIELD(VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE_32,
+			       VC4_BIN_CONFIG_ALLOC_INIT_BLOCK_SIZE) |
+		 VC4_SET_FIELD(VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE_128,
+			       VC4_BIN_CONFIG_ALLOC_BLOCK_SIZE));
+
+	/* tile alloc address. */
+	*(uint32_t *)(validated + 0) = exec->tile_alloc_offset;
+	/* tile alloc size. */
+	*(uint32_t *)(validated + 4) = (bin_addr + vc4->bin_alloc_size -
+					exec->tile_alloc_offset);
+	/* tile state address. */
+	*(uint32_t *)(validated + 8) = bin_addr;
+
+	return 0;
+}
+
+static int
+validate_gem_handles(VALIDATE_ARGS)
+{
+	memcpy(exec->bo_index, untrusted, sizeof(exec->bo_index));
+	return 0;
+}
+
+#define VC4_DEFINE_PACKET(packet, func) \
+	[packet] = { packet ## _SIZE, #packet, func }
+
+static const struct cmd_info {
+	uint16_t len;
+	const char *name;
+	int (*func)(struct vc4_exec_info *exec, void *validated,
+		    void *untrusted);
+} cmd_info[] = {
+	VC4_DEFINE_PACKET(VC4_PACKET_HALT, NULL),
+	VC4_DEFINE_PACKET(VC4_PACKET_NOP, NULL),
+	VC4_DEFINE_PACKET(VC4_PACKET_FLUSH, validate_flush),
+	VC4_DEFINE_PACKET(VC4_PACKET_FLUSH_ALL, NULL),
+	VC4_DEFINE_PACKET(VC4_PACKET_START_TILE_BINNING,
+			  validate_start_tile_binning),
+	VC4_DEFINE_PACKET(VC4_PACKET_INCREMENT_SEMAPHORE,
+			  validate_increment_semaphore),
+
+	VC4_DEFINE_PACKET(VC4_PACKET_GL_INDEXED_PRIMITIVE,
+			  validate_indexed_prim_list),
+	VC4_DEFINE_PACKET(VC4_PACKET_GL_ARRAY_PRIMITIVE,
+			  validate_gl_array_primitive),
+
+	VC4_DEFINE_PACKET(VC4_PACKET_PRIMITIVE_LIST_FORMAT, NULL),
+
+	VC4_DEFINE_PACKET(VC4_PACKET_GL_SHADER_STATE, validate_gl_shader_state),
+
+	VC4_DEFINE_PACKET(VC4_PACKET_CONFIGURATION_BITS, NULL),
+	VC4_DEFINE_PACKET(VC4_PACKET_FLAT_SHADE_FLAGS, NULL),
+	VC4_DEFINE_PACKET(VC4_PACKET_POINT_SIZE, NULL),
+	VC4_DEFINE_PACKET(VC4_PACKET_LINE_WIDTH, NULL),
+	VC4_DEFINE_PACKET(VC4_PACKET_RHT_X_BOUNDARY, NULL),
+	VC4_DEFINE_PACKET(VC4_PACKET_DEPTH_OFFSET, NULL),
+	VC4_DEFINE_PACKET(VC4_PACKET_CLIP_WINDOW, NULL),
+	VC4_DEFINE_PACKET(VC4_PACKET_VIEWPORT_OFFSET, NULL),
+	VC4_DEFINE_PACKET(VC4_PACKET_CLIPPER_XY_SCALING, NULL),
+	/* Note: The docs say this was also 105, but it was 106 in the
+	 * initial userland code drop.
+	 */
+	VC4_DEFINE_PACKET(VC4_PACKET_CLIPPER_Z_SCALING, NULL),
+
+	VC4_DEFINE_PACKET(VC4_PACKET_TILE_BINNING_MODE_CONFIG,
+			  validate_tile_binning_config),
+
+	VC4_DEFINE_PACKET(VC4_PACKET_GEM_HANDLES, validate_gem_handles),
+};
+
+int
+vc4_validate_bin_cl(struct drm_device *dev,
+		    void *validated,
+		    void *unvalidated,
+		    struct vc4_exec_info *exec)
+{
+	uint32_t len = exec->args->bin_cl_size;
+	uint32_t dst_offset = 0;
+	uint32_t src_offset = 0;
+
+	while (src_offset < len) {
+		void *dst_pkt = validated + dst_offset;
+		void *src_pkt = unvalidated + src_offset;
+		u8 cmd = *(uint8_t *)src_pkt;
+		const struct cmd_info *info;
+
+		if (cmd >= ARRAY_SIZE(cmd_info)) {
+			DRM_DEBUG("0x%08x: packet %d out of bounds\n",
+				  src_offset, cmd);
+			return -EINVAL;
+		}
+
+		info = &cmd_info[cmd];
+		if (!info->name) {
+			DRM_DEBUG("0x%08x: packet %d invalid\n",
+				  src_offset, cmd);
+			return -EINVAL;
+		}
+
+		if (src_offset + info->len > len) {
+			DRM_DEBUG("0x%08x: packet %d (%s) length 0x%08x "
+				  "exceeds bounds (0x%08x)\n",
+				  src_offset, cmd, info->name, info->len,
+				  src_offset + len);
+			return -EINVAL;
+		}
+
+		if (cmd != VC4_PACKET_GEM_HANDLES)
+			memcpy(dst_pkt, src_pkt, info->len);
+
+		if (info->func && info->func(exec,
+					     dst_pkt + 1,
+					     src_pkt + 1)) {
+			DRM_DEBUG("0x%08x: packet %d (%s) failed to validate\n",
+				  src_offset, cmd, info->name);
+			return -EINVAL;
+		}
+
+		src_offset += info->len;
+		/* GEM handle loading doesn't produce HW packets. */
+		if (cmd != VC4_PACKET_GEM_HANDLES)
+			dst_offset += info->len;
+
+		/* When the CL hits halt, it'll stop reading anything else. */
+		if (cmd == VC4_PACKET_HALT)
+			break;
+	}
+
+	exec->ct0ea = exec->ct0ca + dst_offset;
+
+	if (!exec->found_start_tile_binning_packet) {
+		DRM_DEBUG("Bin CL missing VC4_PACKET_START_TILE_BINNING\n");
+		return -EINVAL;
+	}
+
+	/* The bin CL must be ended with INCREMENT_SEMAPHORE and FLUSH.  The
+	 * semaphore is used to trigger the render CL to start up, and the
+	 * FLUSH is what caps the bin lists with
+	 * VC4_PACKET_RETURN_FROM_SUB_LIST (so they jump back to the main
+	 * render CL when they get called to) and actually triggers the queued
+	 * semaphore increment.
+	 */
+	if (!exec->found_increment_semaphore_packet || !exec->found_flush) {
+		DRM_DEBUG("Bin CL missing VC4_PACKET_INCREMENT_SEMAPHORE + "
+			  "VC4_PACKET_FLUSH\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static bool
+reloc_tex(struct vc4_exec_info *exec,
+	  void *uniform_data_u,
+	  struct vc4_texture_sample_info *sample,
+	  uint32_t texture_handle_index, bool is_cs)
+{
+	struct drm_gem_cma_object *tex;
+	uint32_t p0 = *(uint32_t *)(uniform_data_u + sample->p_offset[0]);
+	uint32_t p1 = *(uint32_t *)(uniform_data_u + sample->p_offset[1]);
+	uint32_t p2 = (sample->p_offset[2] != ~0 ?
+		       *(uint32_t *)(uniform_data_u + sample->p_offset[2]) : 0);
+	uint32_t p3 = (sample->p_offset[3] != ~0 ?
+		       *(uint32_t *)(uniform_data_u + sample->p_offset[3]) : 0);
+	uint32_t *validated_p0 = exec->uniforms_v + sample->p_offset[0];
+	uint32_t offset = p0 & VC4_TEX_P0_OFFSET_MASK;
+	uint32_t miplevels = VC4_GET_FIELD(p0, VC4_TEX_P0_MIPLVLS);
+	uint32_t width = VC4_GET_FIELD(p1, VC4_TEX_P1_WIDTH);
+	uint32_t height = VC4_GET_FIELD(p1, VC4_TEX_P1_HEIGHT);
+	uint32_t cpp, tiling_format, utile_w, utile_h;
+	uint32_t i;
+	uint32_t cube_map_stride = 0;
+	enum vc4_texture_data_type type;
+
+	tex = vc4_use_bo(exec, texture_handle_index);
+	if (!tex)
+		return false;
+
+	if (sample->is_direct) {
+		uint32_t remaining_size = tex->base.size - p0;
+
+		if (p0 > tex->base.size - 4) {
+			DRM_DEBUG("UBO offset greater than UBO size\n");
+			goto fail;
+		}
+		if (p1 > remaining_size - 4) {
+			DRM_DEBUG("UBO clamp would allow reads "
+				  "outside of UBO\n");
+			goto fail;
+		}
+		*validated_p0 = tex->paddr + p0;
+		return true;
+	}
+
+	if (width == 0)
+		width = 2048;
+	if (height == 0)
+		height = 2048;
+
+	if (p0 & VC4_TEX_P0_CMMODE_MASK) {
+		if (VC4_GET_FIELD(p2, VC4_TEX_P2_PTYPE) ==
+		    VC4_TEX_P2_PTYPE_CUBE_MAP_STRIDE)
+			cube_map_stride = p2 & VC4_TEX_P2_CMST_MASK;
+		if (VC4_GET_FIELD(p3, VC4_TEX_P2_PTYPE) ==
+		    VC4_TEX_P2_PTYPE_CUBE_MAP_STRIDE) {
+			if (cube_map_stride) {
+				DRM_DEBUG("Cube map stride set twice\n");
+				goto fail;
+			}
+
+			cube_map_stride = p3 & VC4_TEX_P2_CMST_MASK;
+		}
+		if (!cube_map_stride) {
+			DRM_DEBUG("Cube map stride not set\n");
+			goto fail;
+		}
+	}
+
+	type = (VC4_GET_FIELD(p0, VC4_TEX_P0_TYPE) |
+		(VC4_GET_FIELD(p1, VC4_TEX_P1_TYPE4) << 4));
+
+	switch (type) {
+	case VC4_TEXTURE_TYPE_RGBA8888:
+	case VC4_TEXTURE_TYPE_RGBX8888:
+	case VC4_TEXTURE_TYPE_RGBA32R:
+		cpp = 4;
+		break;
+	case VC4_TEXTURE_TYPE_RGBA4444:
+	case VC4_TEXTURE_TYPE_RGBA5551:
+	case VC4_TEXTURE_TYPE_RGB565:
+	case VC4_TEXTURE_TYPE_LUMALPHA:
+	case VC4_TEXTURE_TYPE_S16F:
+	case VC4_TEXTURE_TYPE_S16:
+		cpp = 2;
+		break;
+	case VC4_TEXTURE_TYPE_LUMINANCE:
+	case VC4_TEXTURE_TYPE_ALPHA:
+	case VC4_TEXTURE_TYPE_S8:
+		cpp = 1;
+		break;
+	case VC4_TEXTURE_TYPE_ETC1:
+		/* ETC1 is arranged as 64-bit blocks, where each block is 4x4
+		 * pixels.
+		 */
+		cpp = 8;
+		width = (width + 3) >> 2;
+		height = (height + 3) >> 2;
+		break;
+	case VC4_TEXTURE_TYPE_BW1:
+	case VC4_TEXTURE_TYPE_A4:
+	case VC4_TEXTURE_TYPE_A1:
+	case VC4_TEXTURE_TYPE_RGBA64:
+	case VC4_TEXTURE_TYPE_YUV422R:
+	default:
+		DRM_DEBUG("Texture format %d unsupported\n", type);
+		goto fail;
+	}
+	utile_w = utile_width(cpp);
+	utile_h = utile_height(cpp);
+
+	if (type == VC4_TEXTURE_TYPE_RGBA32R) {
+		tiling_format = VC4_TILING_FORMAT_LINEAR;
+	} else {
+		if (size_is_lt(width, height, cpp))
+			tiling_format = VC4_TILING_FORMAT_LT;
+		else
+			tiling_format = VC4_TILING_FORMAT_T;
+	}
+
+	if (!vc4_check_tex_size(exec, tex, offset + cube_map_stride * 5,
+				tiling_format, width, height, cpp)) {
+		goto fail;
+	}
+
+	/* The mipmap levels are stored before the base of the texture.  Make
+	 * sure there is actually space in the BO.
+	 */
+	for (i = 1; i <= miplevels; i++) {
+		uint32_t level_width = max(width >> i, 1u);
+		uint32_t level_height = max(height >> i, 1u);
+		uint32_t aligned_width, aligned_height;
+		uint32_t level_size;
+
+		/* Once the levels get small enough, they drop from T to LT. */
+		if (tiling_format == VC4_TILING_FORMAT_T &&
+		    size_is_lt(level_width, level_height, cpp)) {
+			tiling_format = VC4_TILING_FORMAT_LT;
+		}
+
+		switch (tiling_format) {
+		case VC4_TILING_FORMAT_T:
+			aligned_width = round_up(level_width, utile_w * 8);
+			aligned_height = round_up(level_height, utile_h * 8);
+			break;
+		case VC4_TILING_FORMAT_LT:
+			aligned_width = round_up(level_width, utile_w);
+			aligned_height = round_up(level_height, utile_h);
+			break;
+		default:
+			aligned_width = round_up(level_width, utile_w);
+			aligned_height = level_height;
+			break;
+		}
+
+		level_size = aligned_width * cpp * aligned_height;
+
+		if (offset < level_size) {
+			DRM_DEBUG("Level %d (%dx%d -> %dx%d) size %db "
+				  "overflowed buffer bounds (offset %d)\n",
+				  i, level_width, level_height,
+				  aligned_width, aligned_height,
+				  level_size, offset);
+			goto fail;
+		}
+
+		offset -= level_size;
+	}
+
+	*validated_p0 = tex->paddr + p0;
+
+	if (is_cs) {
+		exec->bin_dep_seqno = max(exec->bin_dep_seqno,
+					  to_vc4_bo(&tex->base)->write_seqno);
+	}
+
+	return true;
+ fail:
+	DRM_INFO("Texture p0 at %d: 0x%08x\n", sample->p_offset[0], p0);
+	DRM_INFO("Texture p1 at %d: 0x%08x\n", sample->p_offset[1], p1);
+	DRM_INFO("Texture p2 at %d: 0x%08x\n", sample->p_offset[2], p2);
+	DRM_INFO("Texture p3 at %d: 0x%08x\n", sample->p_offset[3], p3);
+	return false;
+}
+
+static int
+validate_gl_shader_rec(struct drm_device *dev,
+		       struct vc4_exec_info *exec,
+		       struct vc4_shader_state *state)
+{
+	uint32_t *src_handles;
+	void *pkt_u, *pkt_v;
+	static const uint32_t shader_reloc_offsets[] = {
+		4, /* fs */
+		16, /* vs */
+		28, /* cs */
+	};
+	uint32_t shader_reloc_count = ARRAY_SIZE(shader_reloc_offsets);
+	struct drm_gem_cma_object *bo[ARRAY_SIZE(shader_reloc_offsets) + 8];
+	uint32_t nr_attributes, nr_relocs, packet_size;
+	int i;
+
+	nr_attributes = state->addr & 0x7;
+	if (nr_attributes == 0)
+		nr_attributes = 8;
+	packet_size = gl_shader_rec_size(state->addr);
+
+	nr_relocs = ARRAY_SIZE(shader_reloc_offsets) + nr_attributes;
+	if (nr_relocs * 4 > exec->shader_rec_size) {
+		DRM_DEBUG("overflowed shader recs reading %d handles "
+			  "from %d bytes left\n",
+			  nr_relocs, exec->shader_rec_size);
+		return -EINVAL;
+	}
+	src_handles = exec->shader_rec_u;
+	exec->shader_rec_u += nr_relocs * 4;
+	exec->shader_rec_size -= nr_relocs * 4;
+
+	if (packet_size > exec->shader_rec_size) {
+		DRM_DEBUG("overflowed shader recs copying %db packet "
+			  "from %d bytes left\n",
+			  packet_size, exec->shader_rec_size);
+		return -EINVAL;
+	}
+	pkt_u = exec->shader_rec_u;
+	pkt_v = exec->shader_rec_v;
+	memcpy(pkt_v, pkt_u, packet_size);
+	exec->shader_rec_u += packet_size;
+	/* Shader recs have to be aligned to 16 bytes (due to the attribute
+	 * flags being in the low bytes), so round the next validated shader
+	 * rec address up.  This should be safe, since we've got so many
+	 * relocations in a shader rec packet.
+	 */
+	BUG_ON(roundup(packet_size, 16) - packet_size > nr_relocs * 4);
+	exec->shader_rec_v += roundup(packet_size, 16);
+	exec->shader_rec_size -= packet_size;
+
+	for (i = 0; i < shader_reloc_count; i++) {
+		if (src_handles[i] > exec->bo_count) {
+			DRM_DEBUG("Shader handle %d too big\n", src_handles[i]);
+			return -EINVAL;
+		}
+
+		bo[i] = exec->bo[src_handles[i]];
+		if (!bo[i])
+			return -EINVAL;
+	}
+	for (i = shader_reloc_count; i < nr_relocs; i++) {
+		bo[i] = vc4_use_bo(exec, src_handles[i]);
+		if (!bo[i])
+			return -EINVAL;
+	}
+
+	if (((*(uint16_t *)pkt_u & VC4_SHADER_FLAG_FS_SINGLE_THREAD) == 0) !=
+	    to_vc4_bo(&bo[0]->base)->validated_shader->is_threaded) {
+		DRM_DEBUG("Thread mode of CL and FS do not match\n");
+		return -EINVAL;
+	}
+
+	if (to_vc4_bo(&bo[1]->base)->validated_shader->is_threaded ||
+	    to_vc4_bo(&bo[2]->base)->validated_shader->is_threaded) {
+		DRM_DEBUG("cs and vs cannot be threaded\n");
+		return -EINVAL;
+	}
+
+	for (i = 0; i < shader_reloc_count; i++) {
+		struct vc4_validated_shader_info *validated_shader;
+		uint32_t o = shader_reloc_offsets[i];
+		uint32_t src_offset = *(uint32_t *)(pkt_u + o);
+		uint32_t *texture_handles_u;
+		void *uniform_data_u;
+		uint32_t tex, uni;
+
+		*(uint32_t *)(pkt_v + o) = bo[i]->paddr + src_offset;
+
+		if (src_offset != 0) {
+			DRM_DEBUG("Shaders must be at offset 0 of "
+				  "the BO.\n");
+			return -EINVAL;
+		}
+
+		validated_shader = to_vc4_bo(&bo[i]->base)->validated_shader;
+		if (!validated_shader)
+			return -EINVAL;
+
+		if (validated_shader->uniforms_src_size >
+		    exec->uniforms_size) {
+			DRM_DEBUG("Uniforms src buffer overflow\n");
+			return -EINVAL;
+		}
+
+		texture_handles_u = exec->uniforms_u;
+		uniform_data_u = (texture_handles_u +
+				  validated_shader->num_texture_samples);
+
+		memcpy(exec->uniforms_v, uniform_data_u,
+		       validated_shader->uniforms_size);
+
+		for (tex = 0;
+		     tex < validated_shader->num_texture_samples;
+		     tex++) {
+			if (!reloc_tex(exec,
+				       uniform_data_u,
+				       &validated_shader->texture_samples[tex],
+				       texture_handles_u[tex],
+				       i == 2)) {
+				return -EINVAL;
+			}
+		}
+
+		/* Fill in the uniform slots that need this shader's
+		 * start-of-uniforms address (used for resetting the uniform
+		 * stream in the presence of control flow).
+		 */
+		for (uni = 0;
+		     uni < validated_shader->num_uniform_addr_offsets;
+		     uni++) {
+			uint32_t o = validated_shader->uniform_addr_offsets[uni];
+			((uint32_t *)exec->uniforms_v)[o] = exec->uniforms_p;
+		}
+
+		*(uint32_t *)(pkt_v + o + 4) = exec->uniforms_p;
+
+		exec->uniforms_u += validated_shader->uniforms_src_size;
+		exec->uniforms_v += validated_shader->uniforms_size;
+		exec->uniforms_p += validated_shader->uniforms_size;
+	}
+
+	for (i = 0; i < nr_attributes; i++) {
+		struct drm_gem_cma_object *vbo =
+			bo[ARRAY_SIZE(shader_reloc_offsets) + i];
+		uint32_t o = 36 + i * 8;
+		uint32_t offset = *(uint32_t *)(pkt_u + o + 0);
+		uint32_t attr_size = *(uint8_t *)(pkt_u + o + 4) + 1;
+		uint32_t stride = *(uint8_t *)(pkt_u + o + 5);
+		uint32_t max_index;
+
+		exec->bin_dep_seqno = max(exec->bin_dep_seqno,
+					  to_vc4_bo(&vbo->base)->write_seqno);
+
+		if (state->addr & 0x8)
+			stride |= (*(uint32_t *)(pkt_u + 100 + i * 4)) & ~0xff;
+
+		if (vbo->base.size < offset ||
+		    vbo->base.size - offset < attr_size) {
+			DRM_DEBUG("BO offset overflow (%d + %d > %zu)\n",
+				  offset, attr_size, vbo->base.size);
+			return -EINVAL;
+		}
+
+		if (stride != 0) {
+			max_index = ((vbo->base.size - offset - attr_size) /
+				     stride);
+			if (state->max_index > max_index) {
+				DRM_DEBUG("primitives use index %d out of "
+					  "supplied %d\n",
+					  state->max_index, max_index);
+				return -EINVAL;
+			}
+		}
+
+		*(uint32_t *)(pkt_v + o) = vbo->paddr + offset;
+	}
+
+	return 0;
+}
+
+int
+vc4_validate_shader_recs(struct drm_device *dev,
+			 struct vc4_exec_info *exec)
+{
+	uint32_t i;
+	int ret = 0;
+
+	for (i = 0; i < exec->shader_state_count; i++) {
+		ret = validate_gl_shader_rec(dev, exec, &exec->shader_state[i]);
+		if (ret)
+			return ret;
+	}
+
+	return ret;
+}
diff --git a/drivers/gpu/drm/vc4/vc4_validate_shaders.c b/drivers/gpu/drm/vc4/vc4_validate_shaders.c
new file mode 100644
index 000000000..7cf82b071
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_validate_shaders.c
@@ -0,0 +1,950 @@
+/*
+ * Copyright © 2014 Broadcom
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+/**
+ * DOC: Shader validator for VC4.
+ *
+ * Since the VC4 has no IOMMU between it and system memory, a user
+ * with access to execute shaders could escalate privilege by
+ * overwriting system memory (using the VPM write address register in
+ * the general-purpose DMA mode) or reading system memory it shouldn't
+ * (reading it as a texture, uniform data, or direct-addressed TMU
+ * lookup).
+ *
+ * The shader validator walks over a shader's BO, ensuring that its
+ * accesses are appropriately bounded, and recording where texture
+ * accesses are made so that we can do relocations for them in the
+ * uniform stream.
+ *
+ * Shader BO are immutable for their lifetimes (enforced by not
+ * allowing mmaps, GEM prime export, or rendering to from a CL), so
+ * this validation is only performed at BO creation time.
+ */
+
+#include "vc4_drv.h"
+#include "vc4_qpu_defines.h"
+
+#define LIVE_REG_COUNT (32 + 32 + 4)
+
+struct vc4_shader_validation_state {
+	/* Current IP being validated. */
+	uint32_t ip;
+
+	/* IP at the end of the BO, do not read shader[max_ip] */
+	uint32_t max_ip;
+
+	uint64_t *shader;
+
+	struct vc4_texture_sample_info tmu_setup[2];
+	int tmu_write_count[2];
+
+	/* For registers that were last written to by a MIN instruction with
+	 * one argument being a uniform, the address of the uniform.
+	 * Otherwise, ~0.
+	 *
+	 * This is used for the validation of direct address memory reads.
+	 */
+	uint32_t live_min_clamp_offsets[LIVE_REG_COUNT];
+	bool live_max_clamp_regs[LIVE_REG_COUNT];
+	uint32_t live_immediates[LIVE_REG_COUNT];
+
+	/* Bitfield of which IPs are used as branch targets.
+	 *
+	 * Used for validation that the uniform stream is updated at the right
+	 * points and clearing the texturing/clamping state.
+	 */
+	unsigned long *branch_targets;
+
+	/* Set when entering a basic block, and cleared when the uniform
+	 * address update is found.  This is used to make sure that we don't
+	 * read uniforms when the address is undefined.
+	 */
+	bool needs_uniform_address_update;
+
+	/* Set when we find a backwards branch.  If the branch is backwards,
+	 * the taraget is probably doing an address reset to read uniforms,
+	 * and so we need to be sure that a uniforms address is present in the
+	 * stream, even if the shader didn't need to read uniforms in later
+	 * basic blocks.
+	 */
+	bool needs_uniform_address_for_loop;
+
+	/* Set when we find an instruction writing the top half of the
+	 * register files.  If we allowed writing the unusable regs in
+	 * a threaded shader, then the other shader running on our
+	 * QPU's clamp validation would be invalid.
+	 */
+	bool all_registers_used;
+};
+
+static uint32_t
+waddr_to_live_reg_index(uint32_t waddr, bool is_b)
+{
+	if (waddr < 32) {
+		if (is_b)
+			return 32 + waddr;
+		else
+			return waddr;
+	} else if (waddr <= QPU_W_ACC3) {
+		return 64 + waddr - QPU_W_ACC0;
+	} else {
+		return ~0;
+	}
+}
+
+static uint32_t
+raddr_add_a_to_live_reg_index(uint64_t inst)
+{
+	uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
+	uint32_t add_a = QPU_GET_FIELD(inst, QPU_ADD_A);
+	uint32_t raddr_a = QPU_GET_FIELD(inst, QPU_RADDR_A);
+	uint32_t raddr_b = QPU_GET_FIELD(inst, QPU_RADDR_B);
+
+	if (add_a == QPU_MUX_A)
+		return raddr_a;
+	else if (add_a == QPU_MUX_B && sig != QPU_SIG_SMALL_IMM)
+		return 32 + raddr_b;
+	else if (add_a <= QPU_MUX_R3)
+		return 64 + add_a;
+	else
+		return ~0;
+}
+
+static bool
+live_reg_is_upper_half(uint32_t lri)
+{
+	return	(lri >= 16 && lri < 32) ||
+		(lri >= 32 + 16 && lri < 32 + 32);
+}
+
+static bool
+is_tmu_submit(uint32_t waddr)
+{
+	return (waddr == QPU_W_TMU0_S ||
+		waddr == QPU_W_TMU1_S);
+}
+
+static bool
+is_tmu_write(uint32_t waddr)
+{
+	return (waddr >= QPU_W_TMU0_S &&
+		waddr <= QPU_W_TMU1_B);
+}
+
+static bool
+record_texture_sample(struct vc4_validated_shader_info *validated_shader,
+		      struct vc4_shader_validation_state *validation_state,
+		      int tmu)
+{
+	uint32_t s = validated_shader->num_texture_samples;
+	int i;
+	struct vc4_texture_sample_info *temp_samples;
+
+	temp_samples = krealloc(validated_shader->texture_samples,
+				(s + 1) * sizeof(*temp_samples),
+				GFP_KERNEL);
+	if (!temp_samples)
+		return false;
+
+	memcpy(&temp_samples[s],
+	       &validation_state->tmu_setup[tmu],
+	       sizeof(*temp_samples));
+
+	validated_shader->num_texture_samples = s + 1;
+	validated_shader->texture_samples = temp_samples;
+
+	for (i = 0; i < 4; i++)
+		validation_state->tmu_setup[tmu].p_offset[i] = ~0;
+
+	return true;
+}
+
+static bool
+check_tmu_write(struct vc4_validated_shader_info *validated_shader,
+		struct vc4_shader_validation_state *validation_state,
+		bool is_mul)
+{
+	uint64_t inst = validation_state->shader[validation_state->ip];
+	uint32_t waddr = (is_mul ?
+			  QPU_GET_FIELD(inst, QPU_WADDR_MUL) :
+			  QPU_GET_FIELD(inst, QPU_WADDR_ADD));
+	uint32_t raddr_a = QPU_GET_FIELD(inst, QPU_RADDR_A);
+	uint32_t raddr_b = QPU_GET_FIELD(inst, QPU_RADDR_B);
+	int tmu = waddr > QPU_W_TMU0_B;
+	bool submit = is_tmu_submit(waddr);
+	bool is_direct = submit && validation_state->tmu_write_count[tmu] == 0;
+	uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
+
+	if (is_direct) {
+		uint32_t add_b = QPU_GET_FIELD(inst, QPU_ADD_B);
+		uint32_t clamp_reg, clamp_offset;
+
+		if (sig == QPU_SIG_SMALL_IMM) {
+			DRM_DEBUG("direct TMU read used small immediate\n");
+			return false;
+		}
+
+		/* Make sure that this texture load is an add of the base
+		 * address of the UBO to a clamped offset within the UBO.
+		 */
+		if (is_mul ||
+		    QPU_GET_FIELD(inst, QPU_OP_ADD) != QPU_A_ADD) {
+			DRM_DEBUG("direct TMU load wasn't an add\n");
+			return false;
+		}
+
+		/* We assert that the clamped address is the first
+		 * argument, and the UBO base address is the second argument.
+		 * This is arbitrary, but simpler than supporting flipping the
+		 * two either way.
+		 */
+		clamp_reg = raddr_add_a_to_live_reg_index(inst);
+		if (clamp_reg == ~0) {
+			DRM_DEBUG("direct TMU load wasn't clamped\n");
+			return false;
+		}
+
+		clamp_offset = validation_state->live_min_clamp_offsets[clamp_reg];
+		if (clamp_offset == ~0) {
+			DRM_DEBUG("direct TMU load wasn't clamped\n");
+			return false;
+		}
+
+		/* Store the clamp value's offset in p1 (see reloc_tex() in
+		 * vc4_validate.c).
+		 */
+		validation_state->tmu_setup[tmu].p_offset[1] =
+			clamp_offset;
+
+		if (!(add_b == QPU_MUX_A && raddr_a == QPU_R_UNIF) &&
+		    !(add_b == QPU_MUX_B && raddr_b == QPU_R_UNIF)) {
+			DRM_DEBUG("direct TMU load didn't add to a uniform\n");
+			return false;
+		}
+
+		validation_state->tmu_setup[tmu].is_direct = true;
+	} else {
+		if (raddr_a == QPU_R_UNIF || (sig != QPU_SIG_SMALL_IMM &&
+					      raddr_b == QPU_R_UNIF)) {
+			DRM_DEBUG("uniform read in the same instruction as "
+				  "texture setup.\n");
+			return false;
+		}
+	}
+
+	if (validation_state->tmu_write_count[tmu] >= 4) {
+		DRM_DEBUG("TMU%d got too many parameters before dispatch\n",
+			  tmu);
+		return false;
+	}
+	validation_state->tmu_setup[tmu].p_offset[validation_state->tmu_write_count[tmu]] =
+		validated_shader->uniforms_size;
+	validation_state->tmu_write_count[tmu]++;
+	/* Since direct uses a RADDR uniform reference, it will get counted in
+	 * check_instruction_reads()
+	 */
+	if (!is_direct) {
+		if (validation_state->needs_uniform_address_update) {
+			DRM_DEBUG("Texturing with undefined uniform address\n");
+			return false;
+		}
+
+		validated_shader->uniforms_size += 4;
+	}
+
+	if (submit) {
+		if (!record_texture_sample(validated_shader,
+					   validation_state, tmu)) {
+			return false;
+		}
+
+		validation_state->tmu_write_count[tmu] = 0;
+	}
+
+	return true;
+}
+
+static bool require_uniform_address_uniform(struct vc4_validated_shader_info *validated_shader)
+{
+	uint32_t o = validated_shader->num_uniform_addr_offsets;
+	uint32_t num_uniforms = validated_shader->uniforms_size / 4;
+
+	validated_shader->uniform_addr_offsets =
+		krealloc(validated_shader->uniform_addr_offsets,
+			 (o + 1) *
+			 sizeof(*validated_shader->uniform_addr_offsets),
+			 GFP_KERNEL);
+	if (!validated_shader->uniform_addr_offsets)
+		return false;
+
+	validated_shader->uniform_addr_offsets[o] = num_uniforms;
+	validated_shader->num_uniform_addr_offsets++;
+
+	return true;
+}
+
+static bool
+validate_uniform_address_write(struct vc4_validated_shader_info *validated_shader,
+			       struct vc4_shader_validation_state *validation_state,
+			       bool is_mul)
+{
+	uint64_t inst = validation_state->shader[validation_state->ip];
+	u32 add_b = QPU_GET_FIELD(inst, QPU_ADD_B);
+	u32 raddr_a = QPU_GET_FIELD(inst, QPU_RADDR_A);
+	u32 raddr_b = QPU_GET_FIELD(inst, QPU_RADDR_B);
+	u32 add_lri = raddr_add_a_to_live_reg_index(inst);
+	/* We want our reset to be pointing at whatever uniform follows the
+	 * uniforms base address.
+	 */
+	u32 expected_offset = validated_shader->uniforms_size + 4;
+
+	/* We only support absolute uniform address changes, and we
+	 * require that they be in the current basic block before any
+	 * of its uniform reads.
+	 *
+	 * One could potentially emit more efficient QPU code, by
+	 * noticing that (say) an if statement does uniform control
+	 * flow for all threads and that the if reads the same number
+	 * of uniforms on each side.  However, this scheme is easy to
+	 * validate so it's all we allow for now.
+	 */
+	switch (QPU_GET_FIELD(inst, QPU_SIG)) {
+	case QPU_SIG_NONE:
+	case QPU_SIG_SCOREBOARD_UNLOCK:
+	case QPU_SIG_COLOR_LOAD:
+	case QPU_SIG_LOAD_TMU0:
+	case QPU_SIG_LOAD_TMU1:
+		break;
+	default:
+		DRM_DEBUG("uniforms address change must be "
+			  "normal math\n");
+		return false;
+	}
+
+	if (is_mul || QPU_GET_FIELD(inst, QPU_OP_ADD) != QPU_A_ADD) {
+		DRM_DEBUG("Uniform address reset must be an ADD.\n");
+		return false;
+	}
+
+	if (QPU_GET_FIELD(inst, QPU_COND_ADD) != QPU_COND_ALWAYS) {
+		DRM_DEBUG("Uniform address reset must be unconditional.\n");
+		return false;
+	}
+
+	if (QPU_GET_FIELD(inst, QPU_PACK) != QPU_PACK_A_NOP &&
+	    !(inst & QPU_PM)) {
+		DRM_DEBUG("No packing allowed on uniforms reset\n");
+		return false;
+	}
+
+	if (add_lri == -1) {
+		DRM_DEBUG("First argument of uniform address write must be "
+			  "an immediate value.\n");
+		return false;
+	}
+
+	if (validation_state->live_immediates[add_lri] != expected_offset) {
+		DRM_DEBUG("Resetting uniforms with offset %db instead of %db\n",
+			  validation_state->live_immediates[add_lri],
+			  expected_offset);
+		return false;
+	}
+
+	if (!(add_b == QPU_MUX_A && raddr_a == QPU_R_UNIF) &&
+	    !(add_b == QPU_MUX_B && raddr_b == QPU_R_UNIF)) {
+		DRM_DEBUG("Second argument of uniform address write must be "
+			  "a uniform.\n");
+		return false;
+	}
+
+	validation_state->needs_uniform_address_update = false;
+	validation_state->needs_uniform_address_for_loop = false;
+	return require_uniform_address_uniform(validated_shader);
+}
+
+static bool
+check_reg_write(struct vc4_validated_shader_info *validated_shader,
+		struct vc4_shader_validation_state *validation_state,
+		bool is_mul)
+{
+	uint64_t inst = validation_state->shader[validation_state->ip];
+	uint32_t waddr = (is_mul ?
+			  QPU_GET_FIELD(inst, QPU_WADDR_MUL) :
+			  QPU_GET_FIELD(inst, QPU_WADDR_ADD));
+	uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
+	bool ws = inst & QPU_WS;
+	bool is_b = is_mul ^ ws;
+	u32 lri = waddr_to_live_reg_index(waddr, is_b);
+
+	if (lri != -1) {
+		uint32_t cond_add = QPU_GET_FIELD(inst, QPU_COND_ADD);
+		uint32_t cond_mul = QPU_GET_FIELD(inst, QPU_COND_MUL);
+
+		if (sig == QPU_SIG_LOAD_IMM &&
+		    QPU_GET_FIELD(inst, QPU_PACK) == QPU_PACK_A_NOP &&
+		    ((is_mul && cond_mul == QPU_COND_ALWAYS) ||
+		     (!is_mul && cond_add == QPU_COND_ALWAYS))) {
+			validation_state->live_immediates[lri] =
+				QPU_GET_FIELD(inst, QPU_LOAD_IMM);
+		} else {
+			validation_state->live_immediates[lri] = ~0;
+		}
+
+		if (live_reg_is_upper_half(lri))
+			validation_state->all_registers_used = true;
+	}
+
+	switch (waddr) {
+	case QPU_W_UNIFORMS_ADDRESS:
+		if (is_b) {
+			DRM_DEBUG("relative uniforms address change "
+				  "unsupported\n");
+			return false;
+		}
+
+		return validate_uniform_address_write(validated_shader,
+						      validation_state,
+						      is_mul);
+
+	case QPU_W_TLB_COLOR_MS:
+	case QPU_W_TLB_COLOR_ALL:
+	case QPU_W_TLB_Z:
+		/* These only interact with the tile buffer, not main memory,
+		 * so they're safe.
+		 */
+		return true;
+
+	case QPU_W_TMU0_S:
+	case QPU_W_TMU0_T:
+	case QPU_W_TMU0_R:
+	case QPU_W_TMU0_B:
+	case QPU_W_TMU1_S:
+	case QPU_W_TMU1_T:
+	case QPU_W_TMU1_R:
+	case QPU_W_TMU1_B:
+		return check_tmu_write(validated_shader, validation_state,
+				       is_mul);
+
+	case QPU_W_HOST_INT:
+	case QPU_W_TMU_NOSWAP:
+	case QPU_W_TLB_ALPHA_MASK:
+	case QPU_W_MUTEX_RELEASE:
+		/* XXX: I haven't thought about these, so don't support them
+		 * for now.
+		 */
+		DRM_DEBUG("Unsupported waddr %d\n", waddr);
+		return false;
+
+	case QPU_W_VPM_ADDR:
+		DRM_DEBUG("General VPM DMA unsupported\n");
+		return false;
+
+	case QPU_W_VPM:
+	case QPU_W_VPMVCD_SETUP:
+		/* We allow VPM setup in general, even including VPM DMA
+		 * configuration setup, because the (unsafe) DMA can only be
+		 * triggered by QPU_W_VPM_ADDR writes.
+		 */
+		return true;
+
+	case QPU_W_TLB_STENCIL_SETUP:
+		return true;
+	}
+
+	return true;
+}
+
+static void
+track_live_clamps(struct vc4_validated_shader_info *validated_shader,
+		  struct vc4_shader_validation_state *validation_state)
+{
+	uint64_t inst = validation_state->shader[validation_state->ip];
+	uint32_t op_add = QPU_GET_FIELD(inst, QPU_OP_ADD);
+	uint32_t waddr_add = QPU_GET_FIELD(inst, QPU_WADDR_ADD);
+	uint32_t waddr_mul = QPU_GET_FIELD(inst, QPU_WADDR_MUL);
+	uint32_t cond_add = QPU_GET_FIELD(inst, QPU_COND_ADD);
+	uint32_t add_a = QPU_GET_FIELD(inst, QPU_ADD_A);
+	uint32_t add_b = QPU_GET_FIELD(inst, QPU_ADD_B);
+	uint32_t raddr_a = QPU_GET_FIELD(inst, QPU_RADDR_A);
+	uint32_t raddr_b = QPU_GET_FIELD(inst, QPU_RADDR_B);
+	uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
+	bool ws = inst & QPU_WS;
+	uint32_t lri_add_a, lri_add, lri_mul;
+	bool add_a_is_min_0;
+
+	/* Check whether OP_ADD's A argumennt comes from a live MAX(x, 0),
+	 * before we clear previous live state.
+	 */
+	lri_add_a = raddr_add_a_to_live_reg_index(inst);
+	add_a_is_min_0 = (lri_add_a != ~0 &&
+			  validation_state->live_max_clamp_regs[lri_add_a]);
+
+	/* Clear live state for registers written by our instruction. */
+	lri_add = waddr_to_live_reg_index(waddr_add, ws);
+	lri_mul = waddr_to_live_reg_index(waddr_mul, !ws);
+	if (lri_mul != ~0) {
+		validation_state->live_max_clamp_regs[lri_mul] = false;
+		validation_state->live_min_clamp_offsets[lri_mul] = ~0;
+	}
+	if (lri_add != ~0) {
+		validation_state->live_max_clamp_regs[lri_add] = false;
+		validation_state->live_min_clamp_offsets[lri_add] = ~0;
+	} else {
+		/* Nothing further to do for live tracking, since only ADDs
+		 * generate new live clamp registers.
+		 */
+		return;
+	}
+
+	/* Now, handle remaining live clamp tracking for the ADD operation. */
+
+	if (cond_add != QPU_COND_ALWAYS)
+		return;
+
+	if (op_add == QPU_A_MAX) {
+		/* Track live clamps of a value to a minimum of 0 (in either
+		 * arg).
+		 */
+		if (sig != QPU_SIG_SMALL_IMM || raddr_b != 0 ||
+		    (add_a != QPU_MUX_B && add_b != QPU_MUX_B)) {
+			return;
+		}
+
+		validation_state->live_max_clamp_regs[lri_add] = true;
+	} else if (op_add == QPU_A_MIN) {
+		/* Track live clamps of a value clamped to a minimum of 0 and
+		 * a maximum of some uniform's offset.
+		 */
+		if (!add_a_is_min_0)
+			return;
+
+		if (!(add_b == QPU_MUX_A && raddr_a == QPU_R_UNIF) &&
+		    !(add_b == QPU_MUX_B && raddr_b == QPU_R_UNIF &&
+		      sig != QPU_SIG_SMALL_IMM)) {
+			return;
+		}
+
+		validation_state->live_min_clamp_offsets[lri_add] =
+			validated_shader->uniforms_size;
+	}
+}
+
+static bool
+check_instruction_writes(struct vc4_validated_shader_info *validated_shader,
+			 struct vc4_shader_validation_state *validation_state)
+{
+	uint64_t inst = validation_state->shader[validation_state->ip];
+	uint32_t waddr_add = QPU_GET_FIELD(inst, QPU_WADDR_ADD);
+	uint32_t waddr_mul = QPU_GET_FIELD(inst, QPU_WADDR_MUL);
+	bool ok;
+
+	if (is_tmu_write(waddr_add) && is_tmu_write(waddr_mul)) {
+		DRM_DEBUG("ADD and MUL both set up textures\n");
+		return false;
+	}
+
+	ok = (check_reg_write(validated_shader, validation_state, false) &&
+	      check_reg_write(validated_shader, validation_state, true));
+
+	track_live_clamps(validated_shader, validation_state);
+
+	return ok;
+}
+
+static bool
+check_branch(uint64_t inst,
+	     struct vc4_validated_shader_info *validated_shader,
+	     struct vc4_shader_validation_state *validation_state,
+	     int ip)
+{
+	int32_t branch_imm = QPU_GET_FIELD(inst, QPU_BRANCH_TARGET);
+	uint32_t waddr_add = QPU_GET_FIELD(inst, QPU_WADDR_ADD);
+	uint32_t waddr_mul = QPU_GET_FIELD(inst, QPU_WADDR_MUL);
+
+	if ((int)branch_imm < 0)
+		validation_state->needs_uniform_address_for_loop = true;
+
+	/* We don't want to have to worry about validation of this, and
+	 * there's no need for it.
+	 */
+	if (waddr_add != QPU_W_NOP || waddr_mul != QPU_W_NOP) {
+		DRM_DEBUG("branch instruction at %d wrote a register.\n",
+			  validation_state->ip);
+		return false;
+	}
+
+	return true;
+}
+
+static bool
+check_instruction_reads(struct vc4_validated_shader_info *validated_shader,
+			struct vc4_shader_validation_state *validation_state)
+{
+	uint64_t inst = validation_state->shader[validation_state->ip];
+	uint32_t raddr_a = QPU_GET_FIELD(inst, QPU_RADDR_A);
+	uint32_t raddr_b = QPU_GET_FIELD(inst, QPU_RADDR_B);
+	uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
+
+	if (raddr_a == QPU_R_UNIF ||
+	    (raddr_b == QPU_R_UNIF && sig != QPU_SIG_SMALL_IMM)) {
+		/* This can't overflow the uint32_t, because we're reading 8
+		 * bytes of instruction to increment by 4 here, so we'd
+		 * already be OOM.
+		 */
+		validated_shader->uniforms_size += 4;
+
+		if (validation_state->needs_uniform_address_update) {
+			DRM_DEBUG("Uniform read with undefined uniform "
+				  "address\n");
+			return false;
+		}
+	}
+
+	if ((raddr_a >= 16 && raddr_a < 32) ||
+	    (raddr_b >= 16 && raddr_b < 32 && sig != QPU_SIG_SMALL_IMM)) {
+		validation_state->all_registers_used = true;
+	}
+
+	return true;
+}
+
+/* Make sure that all branches are absolute and point within the shader, and
+ * note their targets for later.
+ */
+static bool
+vc4_validate_branches(struct vc4_shader_validation_state *validation_state)
+{
+	uint32_t max_branch_target = 0;
+	int ip;
+	int last_branch = -2;
+
+	for (ip = 0; ip < validation_state->max_ip; ip++) {
+		uint64_t inst = validation_state->shader[ip];
+		int32_t branch_imm = QPU_GET_FIELD(inst, QPU_BRANCH_TARGET);
+		uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
+		uint32_t after_delay_ip = ip + 4;
+		uint32_t branch_target_ip;
+
+		if (sig == QPU_SIG_PROG_END) {
+			/* There are two delay slots after program end is
+			 * signaled that are still executed, then we're
+			 * finished.  validation_state->max_ip is the
+			 * instruction after the last valid instruction in the
+			 * program.
+			 */
+			validation_state->max_ip = ip + 3;
+			continue;
+		}
+
+		if (sig != QPU_SIG_BRANCH)
+			continue;
+
+		if (ip - last_branch < 4) {
+			DRM_DEBUG("Branch at %d during delay slots\n", ip);
+			return false;
+		}
+		last_branch = ip;
+
+		if (inst & QPU_BRANCH_REG) {
+			DRM_DEBUG("branching from register relative "
+				  "not supported\n");
+			return false;
+		}
+
+		if (!(inst & QPU_BRANCH_REL)) {
+			DRM_DEBUG("relative branching required\n");
+			return false;
+		}
+
+		/* The actual branch target is the instruction after the delay
+		 * slots, plus whatever byte offset is in the low 32 bits of
+		 * the instruction.  Make sure we're not branching beyond the
+		 * end of the shader object.
+		 */
+		if (branch_imm % sizeof(inst) != 0) {
+			DRM_DEBUG("branch target not aligned\n");
+			return false;
+		}
+
+		branch_target_ip = after_delay_ip + (branch_imm >> 3);
+		if (branch_target_ip >= validation_state->max_ip) {
+			DRM_DEBUG("Branch at %d outside of shader (ip %d/%d)\n",
+				  ip, branch_target_ip,
+				  validation_state->max_ip);
+			return false;
+		}
+		set_bit(branch_target_ip, validation_state->branch_targets);
+
+		/* Make sure that the non-branching path is also not outside
+		 * the shader.
+		 */
+		if (after_delay_ip >= validation_state->max_ip) {
+			DRM_DEBUG("Branch at %d continues past shader end "
+				  "(%d/%d)\n",
+				  ip, after_delay_ip, validation_state->max_ip);
+			return false;
+		}
+		set_bit(after_delay_ip, validation_state->branch_targets);
+		max_branch_target = max(max_branch_target, after_delay_ip);
+	}
+
+	if (max_branch_target > validation_state->max_ip - 3) {
+		DRM_DEBUG("Branch landed after QPU_SIG_PROG_END");
+		return false;
+	}
+
+	return true;
+}
+
+/* Resets any known state for the shader, used when we may be branched to from
+ * multiple locations in the program (or at shader start).
+ */
+static void
+reset_validation_state(struct vc4_shader_validation_state *validation_state)
+{
+	int i;
+
+	for (i = 0; i < 8; i++)
+		validation_state->tmu_setup[i / 4].p_offset[i % 4] = ~0;
+
+	for (i = 0; i < LIVE_REG_COUNT; i++) {
+		validation_state->live_min_clamp_offsets[i] = ~0;
+		validation_state->live_max_clamp_regs[i] = false;
+		validation_state->live_immediates[i] = ~0;
+	}
+}
+
+static bool
+texturing_in_progress(struct vc4_shader_validation_state *validation_state)
+{
+	return (validation_state->tmu_write_count[0] != 0 ||
+		validation_state->tmu_write_count[1] != 0);
+}
+
+static bool
+vc4_handle_branch_target(struct vc4_shader_validation_state *validation_state)
+{
+	uint32_t ip = validation_state->ip;
+
+	if (!test_bit(ip, validation_state->branch_targets))
+		return true;
+
+	if (texturing_in_progress(validation_state)) {
+		DRM_DEBUG("Branch target landed during TMU setup\n");
+		return false;
+	}
+
+	/* Reset our live values tracking, since this instruction may have
+	 * multiple predecessors.
+	 *
+	 * One could potentially do analysis to determine that, for
+	 * example, all predecessors have a live max clamp in the same
+	 * register, but we don't bother with that.
+	 */
+	reset_validation_state(validation_state);
+
+	/* Since we've entered a basic block from potentially multiple
+	 * predecessors, we need the uniforms address to be updated before any
+	 * unforms are read.  We require that after any branch point, the next
+	 * uniform to be loaded is a uniform address offset.  That uniform's
+	 * offset will be marked by the uniform address register write
+	 * validation, or a one-off the end-of-program check.
+	 */
+	validation_state->needs_uniform_address_update = true;
+
+	return true;
+}
+
+struct vc4_validated_shader_info *
+vc4_validate_shader(struct drm_gem_cma_object *shader_obj)
+{
+	bool found_shader_end = false;
+	int shader_end_ip = 0;
+	uint32_t last_thread_switch_ip = -3;
+	uint32_t ip;
+	struct vc4_validated_shader_info *validated_shader = NULL;
+	struct vc4_shader_validation_state validation_state;
+
+	memset(&validation_state, 0, sizeof(validation_state));
+	validation_state.shader = shader_obj->vaddr;
+	validation_state.max_ip = shader_obj->base.size / sizeof(uint64_t);
+
+	reset_validation_state(&validation_state);
+
+	validation_state.branch_targets =
+		kcalloc(BITS_TO_LONGS(validation_state.max_ip),
+			sizeof(unsigned long), GFP_KERNEL);
+	if (!validation_state.branch_targets)
+		goto fail;
+
+	validated_shader = kcalloc(1, sizeof(*validated_shader), GFP_KERNEL);
+	if (!validated_shader)
+		goto fail;
+
+	if (!vc4_validate_branches(&validation_state))
+		goto fail;
+
+	for (ip = 0; ip < validation_state.max_ip; ip++) {
+		uint64_t inst = validation_state.shader[ip];
+		uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
+
+		validation_state.ip = ip;
+
+		if (!vc4_handle_branch_target(&validation_state))
+			goto fail;
+
+		if (ip == last_thread_switch_ip + 3) {
+			/* Reset r0-r3 live clamp data */
+			int i;
+
+			for (i = 64; i < LIVE_REG_COUNT; i++) {
+				validation_state.live_min_clamp_offsets[i] = ~0;
+				validation_state.live_max_clamp_regs[i] = false;
+				validation_state.live_immediates[i] = ~0;
+			}
+		}
+
+		switch (sig) {
+		case QPU_SIG_NONE:
+		case QPU_SIG_WAIT_FOR_SCOREBOARD:
+		case QPU_SIG_SCOREBOARD_UNLOCK:
+		case QPU_SIG_COLOR_LOAD:
+		case QPU_SIG_LOAD_TMU0:
+		case QPU_SIG_LOAD_TMU1:
+		case QPU_SIG_PROG_END:
+		case QPU_SIG_SMALL_IMM:
+		case QPU_SIG_THREAD_SWITCH:
+		case QPU_SIG_LAST_THREAD_SWITCH:
+			if (!check_instruction_writes(validated_shader,
+						      &validation_state)) {
+				DRM_DEBUG("Bad write at ip %d\n", ip);
+				goto fail;
+			}
+
+			if (!check_instruction_reads(validated_shader,
+						     &validation_state))
+				goto fail;
+
+			if (sig == QPU_SIG_PROG_END) {
+				found_shader_end = true;
+				shader_end_ip = ip;
+			}
+
+			if (sig == QPU_SIG_THREAD_SWITCH ||
+			    sig == QPU_SIG_LAST_THREAD_SWITCH) {
+				validated_shader->is_threaded = true;
+
+				if (ip < last_thread_switch_ip + 3) {
+					DRM_DEBUG("Thread switch too soon after "
+						  "last switch at ip %d\n", ip);
+					goto fail;
+				}
+				last_thread_switch_ip = ip;
+			}
+
+			break;
+
+		case QPU_SIG_LOAD_IMM:
+			if (!check_instruction_writes(validated_shader,
+						      &validation_state)) {
+				DRM_DEBUG("Bad LOAD_IMM write at ip %d\n", ip);
+				goto fail;
+			}
+			break;
+
+		case QPU_SIG_BRANCH:
+			if (!check_branch(inst, validated_shader,
+					  &validation_state, ip))
+				goto fail;
+
+			if (ip < last_thread_switch_ip + 3) {
+				DRM_DEBUG("Branch in thread switch at ip %d",
+					  ip);
+				goto fail;
+			}
+
+			break;
+		default:
+			DRM_DEBUG("Unsupported QPU signal %d at "
+				  "instruction %d\n", sig, ip);
+			goto fail;
+		}
+
+		/* There are two delay slots after program end is signaled
+		 * that are still executed, then we're finished.
+		 */
+		if (found_shader_end && ip == shader_end_ip + 2)
+			break;
+	}
+
+	if (ip == validation_state.max_ip) {
+		DRM_DEBUG("shader failed to terminate before "
+			  "shader BO end at %zd\n",
+			  shader_obj->base.size);
+		goto fail;
+	}
+
+	/* Might corrupt other thread */
+	if (validated_shader->is_threaded &&
+	    validation_state.all_registers_used) {
+		DRM_DEBUG("Shader uses threading, but uses the upper "
+			  "half of the registers, too\n");
+		goto fail;
+	}
+
+	/* If we did a backwards branch and we haven't emitted a uniforms
+	 * reset since then, we still need the uniforms stream to have the
+	 * uniforms address available so that the backwards branch can do its
+	 * uniforms reset.
+	 *
+	 * We could potentially prove that the backwards branch doesn't
+	 * contain any uses of uniforms until program exit, but that doesn't
+	 * seem to be worth the trouble.
+	 */
+	if (validation_state.needs_uniform_address_for_loop) {
+		if (!require_uniform_address_uniform(validated_shader))
+			goto fail;
+		validated_shader->uniforms_size += 4;
+	}
+
+	/* Again, no chance of integer overflow here because the worst case
+	 * scenario is 8 bytes of uniforms plus handles per 8-byte
+	 * instruction.
+	 */
+	validated_shader->uniforms_src_size =
+		(validated_shader->uniforms_size +
+		 4 * validated_shader->num_texture_samples);
+
+	kfree(validation_state.branch_targets);
+
+	return validated_shader;
+
+fail:
+	kfree(validation_state.branch_targets);
+	if (validated_shader) {
+		kfree(validated_shader->uniform_addr_offsets);
+		kfree(validated_shader->texture_samples);
+		kfree(validated_shader);
+	}
+	return NULL;
+}
diff --git a/drivers/gpu/drm/vc4/vc4_vec.c b/drivers/gpu/drm/vc4/vc4_vec.c
new file mode 100644
index 000000000..8e7facb65
--- /dev/null
+++ b/drivers/gpu/drm/vc4/vc4_vec.c
@@ -0,0 +1,661 @@
+/*
+ * Copyright (C) 2016 Broadcom
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published by
+ * the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+/**
+ * DOC: VC4 SDTV module
+ *
+ * The VEC encoder generates PAL or NTSC composite video output.
+ *
+ * TV mode selection is done by an atomic property on the encoder,
+ * because a drm_mode_modeinfo is insufficient to distinguish between
+ * PAL and PAL-M or NTSC and NTSC-J.
+ */
+
+#include <drm/drm_atomic_helper.h>
+#include <drm/drm_crtc_helper.h>
+#include <drm/drm_edid.h>
+#include <drm/drm_panel.h>
+#include <linux/clk.h>
+#include <linux/component.h>
+#include <linux/of_graph.h>
+#include <linux/of_platform.h>
+#include <linux/pm_runtime.h>
+
+#include "vc4_drv.h"
+#include "vc4_regs.h"
+
+/* WSE Registers */
+#define VEC_WSE_RESET			0xc0
+
+#define VEC_WSE_CONTROL			0xc4
+#define VEC_WSE_WSS_ENABLE		BIT(7)
+
+#define VEC_WSE_WSS_DATA		0xc8
+#define VEC_WSE_VPS_DATA1		0xcc
+#define VEC_WSE_VPS_CONTROL		0xd0
+
+/* VEC Registers */
+#define VEC_REVID			0x100
+
+#define VEC_CONFIG0			0x104
+#define VEC_CONFIG0_YDEL_MASK		GENMASK(28, 26)
+#define VEC_CONFIG0_YDEL(x)		((x) << 26)
+#define VEC_CONFIG0_CDEL_MASK		GENMASK(25, 24)
+#define VEC_CONFIG0_CDEL(x)		((x) << 24)
+#define VEC_CONFIG0_PBPR_FIL		BIT(18)
+#define VEC_CONFIG0_CHROMA_GAIN_MASK	GENMASK(17, 16)
+#define VEC_CONFIG0_CHROMA_GAIN_UNITY	(0 << 16)
+#define VEC_CONFIG0_CHROMA_GAIN_1_32	(1 << 16)
+#define VEC_CONFIG0_CHROMA_GAIN_1_16	(2 << 16)
+#define VEC_CONFIG0_CHROMA_GAIN_1_8	(3 << 16)
+#define VEC_CONFIG0_CBURST_GAIN_MASK	GENMASK(14, 13)
+#define VEC_CONFIG0_CBURST_GAIN_UNITY	(0 << 13)
+#define VEC_CONFIG0_CBURST_GAIN_1_128	(1 << 13)
+#define VEC_CONFIG0_CBURST_GAIN_1_64	(2 << 13)
+#define VEC_CONFIG0_CBURST_GAIN_1_32	(3 << 13)
+#define VEC_CONFIG0_CHRBW1		BIT(11)
+#define VEC_CONFIG0_CHRBW0		BIT(10)
+#define VEC_CONFIG0_SYNCDIS		BIT(9)
+#define VEC_CONFIG0_BURDIS		BIT(8)
+#define VEC_CONFIG0_CHRDIS		BIT(7)
+#define VEC_CONFIG0_PDEN		BIT(6)
+#define VEC_CONFIG0_YCDELAY		BIT(4)
+#define VEC_CONFIG0_RAMPEN		BIT(2)
+#define VEC_CONFIG0_YCDIS		BIT(2)
+#define VEC_CONFIG0_STD_MASK		GENMASK(1, 0)
+#define VEC_CONFIG0_NTSC_STD		0
+#define VEC_CONFIG0_PAL_BDGHI_STD	1
+#define VEC_CONFIG0_PAL_N_STD		3
+
+#define VEC_SCHPH			0x108
+#define VEC_SOFT_RESET			0x10c
+#define VEC_CLMP0_START			0x144
+#define VEC_CLMP0_END			0x148
+#define VEC_FREQ3_2			0x180
+#define VEC_FREQ1_0			0x184
+
+#define VEC_CONFIG1			0x188
+#define VEC_CONFIG_VEC_RESYNC_OFF	BIT(18)
+#define VEC_CONFIG_RGB219		BIT(17)
+#define VEC_CONFIG_CBAR_EN		BIT(16)
+#define VEC_CONFIG_TC_OBB		BIT(15)
+#define VEC_CONFIG1_OUTPUT_MODE_MASK	GENMASK(12, 10)
+#define VEC_CONFIG1_C_Y_CVBS		(0 << 10)
+#define VEC_CONFIG1_CVBS_Y_C		(1 << 10)
+#define VEC_CONFIG1_PR_Y_PB		(2 << 10)
+#define VEC_CONFIG1_RGB			(4 << 10)
+#define VEC_CONFIG1_Y_C_CVBS		(5 << 10)
+#define VEC_CONFIG1_C_CVBS_Y		(6 << 10)
+#define VEC_CONFIG1_C_CVBS_CVBS		(7 << 10)
+#define VEC_CONFIG1_DIS_CHR		BIT(9)
+#define VEC_CONFIG1_DIS_LUMA		BIT(8)
+#define VEC_CONFIG1_YCBCR_IN		BIT(6)
+#define VEC_CONFIG1_DITHER_TYPE_LFSR	0
+#define VEC_CONFIG1_DITHER_TYPE_COUNTER	BIT(5)
+#define VEC_CONFIG1_DITHER_EN		BIT(4)
+#define VEC_CONFIG1_CYDELAY		BIT(3)
+#define VEC_CONFIG1_LUMADIS		BIT(2)
+#define VEC_CONFIG1_COMPDIS		BIT(1)
+#define VEC_CONFIG1_CUSTOM_FREQ		BIT(0)
+
+#define VEC_CONFIG2			0x18c
+#define VEC_CONFIG2_PROG_SCAN		BIT(15)
+#define VEC_CONFIG2_SYNC_ADJ_MASK	GENMASK(14, 12)
+#define VEC_CONFIG2_SYNC_ADJ(x)		(((x) / 2) << 12)
+#define VEC_CONFIG2_PBPR_EN		BIT(10)
+#define VEC_CONFIG2_UV_DIG_DIS		BIT(6)
+#define VEC_CONFIG2_RGB_DIG_DIS		BIT(5)
+#define VEC_CONFIG2_TMUX_MASK		GENMASK(3, 2)
+#define VEC_CONFIG2_TMUX_DRIVE0		(0 << 2)
+#define VEC_CONFIG2_TMUX_RG_COMP	(1 << 2)
+#define VEC_CONFIG2_TMUX_UV_YC		(2 << 2)
+#define VEC_CONFIG2_TMUX_SYNC_YC	(3 << 2)
+
+#define VEC_INTERRUPT_CONTROL		0x190
+#define VEC_INTERRUPT_STATUS		0x194
+#define VEC_FCW_SECAM_B			0x198
+#define VEC_SECAM_GAIN_VAL		0x19c
+
+#define VEC_CONFIG3			0x1a0
+#define VEC_CONFIG3_HORIZ_LEN_STD	(0 << 0)
+#define VEC_CONFIG3_HORIZ_LEN_MPEG1_SIF	(1 << 0)
+#define VEC_CONFIG3_SHAPE_NON_LINEAR	BIT(1)
+
+#define VEC_STATUS0			0x200
+#define VEC_MASK0			0x204
+
+#define VEC_CFG				0x208
+#define VEC_CFG_SG_MODE_MASK		GENMASK(6, 5)
+#define VEC_CFG_SG_MODE(x)		((x) << 5)
+#define VEC_CFG_SG_EN			BIT(4)
+#define VEC_CFG_VEC_EN			BIT(3)
+#define VEC_CFG_MB_EN			BIT(2)
+#define VEC_CFG_ENABLE			BIT(1)
+#define VEC_CFG_TB_EN			BIT(0)
+
+#define VEC_DAC_TEST			0x20c
+
+#define VEC_DAC_CONFIG			0x210
+#define VEC_DAC_CONFIG_LDO_BIAS_CTRL(x)	((x) << 24)
+#define VEC_DAC_CONFIG_DRIVER_CTRL(x)	((x) << 16)
+#define VEC_DAC_CONFIG_DAC_CTRL(x)	(x)
+
+#define VEC_DAC_MISC			0x214
+#define VEC_DAC_MISC_VCD_CTRL_MASK	GENMASK(31, 16)
+#define VEC_DAC_MISC_VCD_CTRL(x)	((x) << 16)
+#define VEC_DAC_MISC_VID_ACT		BIT(8)
+#define VEC_DAC_MISC_VCD_PWRDN		BIT(6)
+#define VEC_DAC_MISC_BIAS_PWRDN		BIT(5)
+#define VEC_DAC_MISC_DAC_PWRDN		BIT(2)
+#define VEC_DAC_MISC_LDO_PWRDN		BIT(1)
+#define VEC_DAC_MISC_DAC_RST_N		BIT(0)
+
+
+/* General VEC hardware state. */
+struct vc4_vec {
+	struct platform_device *pdev;
+
+	struct drm_encoder *encoder;
+	struct drm_connector *connector;
+
+	void __iomem *regs;
+
+	struct clk *clock;
+
+	const struct vc4_vec_tv_mode *tv_mode;
+};
+
+#define VEC_READ(offset) readl(vec->regs + (offset))
+#define VEC_WRITE(offset, val) writel(val, vec->regs + (offset))
+
+/* VC4 VEC encoder KMS struct */
+struct vc4_vec_encoder {
+	struct vc4_encoder base;
+	struct vc4_vec *vec;
+};
+
+static inline struct vc4_vec_encoder *
+to_vc4_vec_encoder(struct drm_encoder *encoder)
+{
+	return container_of(encoder, struct vc4_vec_encoder, base.base);
+}
+
+/* VC4 VEC connector KMS struct */
+struct vc4_vec_connector {
+	struct drm_connector base;
+	struct vc4_vec *vec;
+
+	/* Since the connector is attached to just the one encoder,
+	 * this is the reference to it so we can do the best_encoder()
+	 * hook.
+	 */
+	struct drm_encoder *encoder;
+};
+
+static inline struct vc4_vec_connector *
+to_vc4_vec_connector(struct drm_connector *connector)
+{
+	return container_of(connector, struct vc4_vec_connector, base);
+}
+
+enum vc4_vec_tv_mode_id {
+	VC4_VEC_TV_MODE_NTSC,
+	VC4_VEC_TV_MODE_NTSC_J,
+	VC4_VEC_TV_MODE_PAL,
+	VC4_VEC_TV_MODE_PAL_M,
+};
+
+struct vc4_vec_tv_mode {
+	const struct drm_display_mode *mode;
+	void (*mode_set)(struct vc4_vec *vec);
+};
+
+#define VEC_REG(reg) { reg, #reg }
+static const struct {
+	u32 reg;
+	const char *name;
+} vec_regs[] = {
+	VEC_REG(VEC_WSE_CONTROL),
+	VEC_REG(VEC_WSE_WSS_DATA),
+	VEC_REG(VEC_WSE_VPS_DATA1),
+	VEC_REG(VEC_WSE_VPS_CONTROL),
+	VEC_REG(VEC_REVID),
+	VEC_REG(VEC_CONFIG0),
+	VEC_REG(VEC_SCHPH),
+	VEC_REG(VEC_CLMP0_START),
+	VEC_REG(VEC_CLMP0_END),
+	VEC_REG(VEC_FREQ3_2),
+	VEC_REG(VEC_FREQ1_0),
+	VEC_REG(VEC_CONFIG1),
+	VEC_REG(VEC_CONFIG2),
+	VEC_REG(VEC_INTERRUPT_CONTROL),
+	VEC_REG(VEC_INTERRUPT_STATUS),
+	VEC_REG(VEC_FCW_SECAM_B),
+	VEC_REG(VEC_SECAM_GAIN_VAL),
+	VEC_REG(VEC_CONFIG3),
+	VEC_REG(VEC_STATUS0),
+	VEC_REG(VEC_MASK0),
+	VEC_REG(VEC_CFG),
+	VEC_REG(VEC_DAC_TEST),
+	VEC_REG(VEC_DAC_CONFIG),
+	VEC_REG(VEC_DAC_MISC),
+};
+
+#ifdef CONFIG_DEBUG_FS
+int vc4_vec_debugfs_regs(struct seq_file *m, void *unused)
+{
+	struct drm_info_node *node = (struct drm_info_node *)m->private;
+	struct drm_device *dev = node->minor->dev;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct vc4_vec *vec = vc4->vec;
+	int i;
+
+	if (!vec)
+		return 0;
+
+	for (i = 0; i < ARRAY_SIZE(vec_regs); i++) {
+		seq_printf(m, "%s (0x%04x): 0x%08x\n",
+			   vec_regs[i].name, vec_regs[i].reg,
+			   VEC_READ(vec_regs[i].reg));
+	}
+
+	return 0;
+}
+#endif
+
+static void vc4_vec_ntsc_mode_set(struct vc4_vec *vec)
+{
+	VEC_WRITE(VEC_CONFIG0, VEC_CONFIG0_NTSC_STD | VEC_CONFIG0_PDEN);
+	VEC_WRITE(VEC_CONFIG1, VEC_CONFIG1_C_CVBS_CVBS);
+}
+
+static void vc4_vec_ntsc_j_mode_set(struct vc4_vec *vec)
+{
+	VEC_WRITE(VEC_CONFIG0, VEC_CONFIG0_NTSC_STD);
+	VEC_WRITE(VEC_CONFIG1, VEC_CONFIG1_C_CVBS_CVBS);
+}
+
+static const struct drm_display_mode ntsc_mode = {
+	DRM_MODE("720x480", DRM_MODE_TYPE_DRIVER, 13500,
+		 720, 720 + 14, 720 + 14 + 64, 720 + 14 + 64 + 60, 0,
+		 480, 480 + 3, 480 + 3 + 3, 480 + 3 + 3 + 16, 0,
+		 DRM_MODE_FLAG_INTERLACE)
+};
+
+static void vc4_vec_pal_mode_set(struct vc4_vec *vec)
+{
+	VEC_WRITE(VEC_CONFIG0, VEC_CONFIG0_PAL_BDGHI_STD);
+	VEC_WRITE(VEC_CONFIG1, VEC_CONFIG1_C_CVBS_CVBS);
+}
+
+static void vc4_vec_pal_m_mode_set(struct vc4_vec *vec)
+{
+	VEC_WRITE(VEC_CONFIG0, VEC_CONFIG0_PAL_BDGHI_STD);
+	VEC_WRITE(VEC_CONFIG1,
+		  VEC_CONFIG1_C_CVBS_CVBS | VEC_CONFIG1_CUSTOM_FREQ);
+	VEC_WRITE(VEC_FREQ3_2, 0x223b);
+	VEC_WRITE(VEC_FREQ1_0, 0x61d1);
+}
+
+static const struct drm_display_mode pal_mode = {
+	DRM_MODE("720x576", DRM_MODE_TYPE_DRIVER, 13500,
+		 720, 720 + 20, 720 + 20 + 64, 720 + 20 + 64 + 60, 0,
+		 576, 576 + 2, 576 + 2 + 3, 576 + 2 + 3 + 20, 0,
+		 DRM_MODE_FLAG_INTERLACE)
+};
+
+static const struct vc4_vec_tv_mode vc4_vec_tv_modes[] = {
+	[VC4_VEC_TV_MODE_NTSC] = {
+		.mode = &ntsc_mode,
+		.mode_set = vc4_vec_ntsc_mode_set,
+	},
+	[VC4_VEC_TV_MODE_NTSC_J] = {
+		.mode = &ntsc_mode,
+		.mode_set = vc4_vec_ntsc_j_mode_set,
+	},
+	[VC4_VEC_TV_MODE_PAL] = {
+		.mode = &pal_mode,
+		.mode_set = vc4_vec_pal_mode_set,
+	},
+	[VC4_VEC_TV_MODE_PAL_M] = {
+		.mode = &pal_mode,
+		.mode_set = vc4_vec_pal_m_mode_set,
+	},
+};
+
+static enum drm_connector_status
+vc4_vec_connector_detect(struct drm_connector *connector, bool force)
+{
+	return connector_status_unknown;
+}
+
+static void vc4_vec_connector_destroy(struct drm_connector *connector)
+{
+	drm_connector_unregister(connector);
+	drm_connector_cleanup(connector);
+}
+
+static int vc4_vec_connector_get_modes(struct drm_connector *connector)
+{
+	struct drm_connector_state *state = connector->state;
+	struct drm_display_mode *mode;
+
+	mode = drm_mode_duplicate(connector->dev,
+				  vc4_vec_tv_modes[state->tv.mode].mode);
+	if (!mode) {
+		DRM_ERROR("Failed to create a new display mode\n");
+		return -ENOMEM;
+	}
+
+	drm_mode_probed_add(connector, mode);
+
+	return 1;
+}
+
+static const struct drm_connector_funcs vc4_vec_connector_funcs = {
+	.detect = vc4_vec_connector_detect,
+	.fill_modes = drm_helper_probe_single_connector_modes,
+	.destroy = vc4_vec_connector_destroy,
+	.reset = drm_atomic_helper_connector_reset,
+	.atomic_duplicate_state = drm_atomic_helper_connector_duplicate_state,
+	.atomic_destroy_state = drm_atomic_helper_connector_destroy_state,
+};
+
+static const struct drm_connector_helper_funcs vc4_vec_connector_helper_funcs = {
+	.get_modes = vc4_vec_connector_get_modes,
+};
+
+static struct drm_connector *vc4_vec_connector_init(struct drm_device *dev,
+						    struct vc4_vec *vec)
+{
+	struct drm_connector *connector = NULL;
+	struct vc4_vec_connector *vec_connector;
+
+	vec_connector = devm_kzalloc(dev->dev, sizeof(*vec_connector),
+				     GFP_KERNEL);
+	if (!vec_connector)
+		return ERR_PTR(-ENOMEM);
+
+	connector = &vec_connector->base;
+	connector->interlace_allowed = true;
+
+	vec_connector->encoder = vec->encoder;
+	vec_connector->vec = vec;
+
+	drm_connector_init(dev, connector, &vc4_vec_connector_funcs,
+			   DRM_MODE_CONNECTOR_Composite);
+	drm_connector_helper_add(connector, &vc4_vec_connector_helper_funcs);
+
+	drm_object_attach_property(&connector->base,
+				   dev->mode_config.tv_mode_property,
+				   VC4_VEC_TV_MODE_NTSC);
+	vec->tv_mode = &vc4_vec_tv_modes[VC4_VEC_TV_MODE_NTSC];
+
+	drm_connector_attach_encoder(connector, vec->encoder);
+
+	return connector;
+}
+
+static const struct drm_encoder_funcs vc4_vec_encoder_funcs = {
+	.destroy = drm_encoder_cleanup,
+};
+
+static void vc4_vec_encoder_disable(struct drm_encoder *encoder)
+{
+	struct vc4_vec_encoder *vc4_vec_encoder = to_vc4_vec_encoder(encoder);
+	struct vc4_vec *vec = vc4_vec_encoder->vec;
+	int ret;
+
+	VEC_WRITE(VEC_CFG, 0);
+	VEC_WRITE(VEC_DAC_MISC,
+		  VEC_DAC_MISC_VCD_PWRDN |
+		  VEC_DAC_MISC_BIAS_PWRDN |
+		  VEC_DAC_MISC_DAC_PWRDN |
+		  VEC_DAC_MISC_LDO_PWRDN);
+
+	clk_disable_unprepare(vec->clock);
+
+	ret = pm_runtime_put(&vec->pdev->dev);
+	if (ret < 0) {
+		DRM_ERROR("Failed to release power domain: %d\n", ret);
+		return;
+	}
+}
+
+static void vc4_vec_encoder_enable(struct drm_encoder *encoder)
+{
+	struct vc4_vec_encoder *vc4_vec_encoder = to_vc4_vec_encoder(encoder);
+	struct vc4_vec *vec = vc4_vec_encoder->vec;
+	int ret;
+
+	ret = pm_runtime_get_sync(&vec->pdev->dev);
+	if (ret < 0) {
+		DRM_ERROR("Failed to retain power domain: %d\n", ret);
+		return;
+	}
+
+	/*
+	 * We need to set the clock rate each time we enable the encoder
+	 * because there's a chance we share the same parent with the HDMI
+	 * clock, and both drivers are requesting different rates.
+	 * The good news is, these 2 encoders cannot be enabled at the same
+	 * time, thus preventing incompatible rate requests.
+	 */
+	ret = clk_set_rate(vec->clock, 108000000);
+	if (ret) {
+		DRM_ERROR("Failed to set clock rate: %d\n", ret);
+		return;
+	}
+
+	ret = clk_prepare_enable(vec->clock);
+	if (ret) {
+		DRM_ERROR("Failed to turn on core clock: %d\n", ret);
+		return;
+	}
+
+	/* Reset the different blocks */
+	VEC_WRITE(VEC_WSE_RESET, 1);
+	VEC_WRITE(VEC_SOFT_RESET, 1);
+
+	/* Disable the CGSM-A and WSE blocks */
+	VEC_WRITE(VEC_WSE_CONTROL, 0);
+
+	/* Write config common to all modes. */
+
+	/*
+	 * Color subcarrier phase: phase = 360 * SCHPH / 256.
+	 * 0x28 <=> 39.375 deg.
+	 */
+	VEC_WRITE(VEC_SCHPH, 0x28);
+
+	/*
+	 * Reset to default values.
+	 */
+	VEC_WRITE(VEC_CLMP0_START, 0xac);
+	VEC_WRITE(VEC_CLMP0_END, 0xec);
+	VEC_WRITE(VEC_CONFIG2,
+		  VEC_CONFIG2_UV_DIG_DIS | VEC_CONFIG2_RGB_DIG_DIS);
+	VEC_WRITE(VEC_CONFIG3, VEC_CONFIG3_HORIZ_LEN_STD);
+	VEC_WRITE(VEC_DAC_CONFIG,
+		  VEC_DAC_CONFIG_DAC_CTRL(0xc) |
+		  VEC_DAC_CONFIG_DRIVER_CTRL(0xc) |
+		  VEC_DAC_CONFIG_LDO_BIAS_CTRL(0x46));
+
+	/* Mask all interrupts. */
+	VEC_WRITE(VEC_MASK0, 0);
+
+	vec->tv_mode->mode_set(vec);
+
+	VEC_WRITE(VEC_DAC_MISC,
+		  VEC_DAC_MISC_VID_ACT | VEC_DAC_MISC_DAC_RST_N);
+	VEC_WRITE(VEC_CFG, VEC_CFG_VEC_EN);
+}
+
+
+static bool vc4_vec_encoder_mode_fixup(struct drm_encoder *encoder,
+				       const struct drm_display_mode *mode,
+				       struct drm_display_mode *adjusted_mode)
+{
+	return true;
+}
+
+static void vc4_vec_encoder_atomic_mode_set(struct drm_encoder *encoder,
+					struct drm_crtc_state *crtc_state,
+					struct drm_connector_state *conn_state)
+{
+	struct vc4_vec_encoder *vc4_vec_encoder = to_vc4_vec_encoder(encoder);
+	struct vc4_vec *vec = vc4_vec_encoder->vec;
+
+	vec->tv_mode = &vc4_vec_tv_modes[conn_state->tv.mode];
+}
+
+static int vc4_vec_encoder_atomic_check(struct drm_encoder *encoder,
+					struct drm_crtc_state *crtc_state,
+					struct drm_connector_state *conn_state)
+{
+	const struct vc4_vec_tv_mode *vec_mode;
+
+	vec_mode = &vc4_vec_tv_modes[conn_state->tv.mode];
+
+	if (conn_state->crtc &&
+	    !drm_mode_equal(vec_mode->mode, &crtc_state->adjusted_mode))
+		return -EINVAL;
+
+	return 0;
+}
+
+static const struct drm_encoder_helper_funcs vc4_vec_encoder_helper_funcs = {
+	.disable = vc4_vec_encoder_disable,
+	.enable = vc4_vec_encoder_enable,
+	.mode_fixup = vc4_vec_encoder_mode_fixup,
+	.atomic_check = vc4_vec_encoder_atomic_check,
+	.atomic_mode_set = vc4_vec_encoder_atomic_mode_set,
+};
+
+static const struct of_device_id vc4_vec_dt_match[] = {
+	{ .compatible = "brcm,bcm2835-vec", .data = NULL },
+	{ /* sentinel */ },
+};
+
+static const char * const tv_mode_names[] = {
+	[VC4_VEC_TV_MODE_NTSC] = "NTSC",
+	[VC4_VEC_TV_MODE_NTSC_J] = "NTSC-J",
+	[VC4_VEC_TV_MODE_PAL] = "PAL",
+	[VC4_VEC_TV_MODE_PAL_M] = "PAL-M",
+};
+
+static int vc4_vec_bind(struct device *dev, struct device *master, void *data)
+{
+	struct platform_device *pdev = to_platform_device(dev);
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	struct vc4_vec *vec;
+	struct vc4_vec_encoder *vc4_vec_encoder;
+	int ret;
+
+	ret = drm_mode_create_tv_properties(drm, ARRAY_SIZE(tv_mode_names),
+					    tv_mode_names);
+	if (ret)
+		return ret;
+
+	vec = devm_kzalloc(dev, sizeof(*vec), GFP_KERNEL);
+	if (!vec)
+		return -ENOMEM;
+
+	vc4_vec_encoder = devm_kzalloc(dev, sizeof(*vc4_vec_encoder),
+				       GFP_KERNEL);
+	if (!vc4_vec_encoder)
+		return -ENOMEM;
+	vc4_vec_encoder->base.type = VC4_ENCODER_TYPE_VEC;
+	vc4_vec_encoder->vec = vec;
+	vec->encoder = &vc4_vec_encoder->base.base;
+
+	vec->pdev = pdev;
+	vec->regs = vc4_ioremap_regs(pdev, 0);
+	if (IS_ERR(vec->regs))
+		return PTR_ERR(vec->regs);
+
+	vec->clock = devm_clk_get(dev, NULL);
+	if (IS_ERR(vec->clock)) {
+		ret = PTR_ERR(vec->clock);
+		if (ret != -EPROBE_DEFER)
+			DRM_ERROR("Failed to get clock: %d\n", ret);
+		return ret;
+	}
+
+	pm_runtime_enable(dev);
+
+	drm_encoder_init(drm, vec->encoder, &vc4_vec_encoder_funcs,
+			 DRM_MODE_ENCODER_TVDAC, NULL);
+	drm_encoder_helper_add(vec->encoder, &vc4_vec_encoder_helper_funcs);
+
+	vec->connector = vc4_vec_connector_init(drm, vec);
+	if (IS_ERR(vec->connector)) {
+		ret = PTR_ERR(vec->connector);
+		goto err_destroy_encoder;
+	}
+
+	dev_set_drvdata(dev, vec);
+
+	vc4->vec = vec;
+
+	return 0;
+
+err_destroy_encoder:
+	drm_encoder_cleanup(vec->encoder);
+	pm_runtime_disable(dev);
+
+	return ret;
+}
+
+static void vc4_vec_unbind(struct device *dev, struct device *master,
+			   void *data)
+{
+	struct drm_device *drm = dev_get_drvdata(master);
+	struct vc4_dev *vc4 = to_vc4_dev(drm);
+	struct vc4_vec *vec = dev_get_drvdata(dev);
+
+	vc4_vec_connector_destroy(vec->connector);
+	drm_encoder_cleanup(vec->encoder);
+	pm_runtime_disable(dev);
+
+	vc4->vec = NULL;
+}
+
+static const struct component_ops vc4_vec_ops = {
+	.bind   = vc4_vec_bind,
+	.unbind = vc4_vec_unbind,
+};
+
+static int vc4_vec_dev_probe(struct platform_device *pdev)
+{
+	return component_add(&pdev->dev, &vc4_vec_ops);
+}
+
+static int vc4_vec_dev_remove(struct platform_device *pdev)
+{
+	component_del(&pdev->dev, &vc4_vec_ops);
+	return 0;
+}
+
+struct platform_driver vc4_vec_driver = {
+	.probe = vc4_vec_dev_probe,
+	.remove = vc4_vec_dev_remove,
+	.driver = {
+		.name = "vc4_vec",
+		.of_match_table = vc4_vec_dt_match,
+	},
+};