1 files changed, 269 insertions, 37 deletions

diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index a2024247b..32cd0c978 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -13,11 +13,14 @@
 #include <drm/ttm/ttm_execbuf_util.h>
 #include <drm/ttm/ttm_tt.h>
 #include <drm/xe_drm.h>
+#include <linux/ascii85.h>
 #include <linux/delay.h>
 #include <linux/kthread.h>
 #include <linux/mm.h>
 #include <linux/swap.h>
 
+#include <generated/xe_wa_oob.h>
+
 #include "xe_assert.h"
 #include "xe_bo.h"
 #include "xe_device.h"
@@ -34,7 +37,6 @@
 #include "xe_res_cursor.h"
 #include "xe_sync.h"
 #include "xe_trace.h"
-#include "generated/xe_wa_oob.h"
 #include "xe_wa.h"
 
 static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm)
@@ -480,17 +482,53 @@ static int xe_gpuvm_validate(struct drm_gpuvm_bo *vm_bo, struct drm_exec *exec)
 	return 0;
 }
 
+/**
+ * xe_vm_validate_rebind() - Validate buffer objects and rebind vmas
+ * @vm: The vm for which we are rebinding.
+ * @exec: The struct drm_exec with the locked GEM objects.
+ * @num_fences: The number of fences to reserve for the operation, not
+ * including rebinds and validations.
+ *
+ * Validates all evicted gem objects and rebinds their vmas. Note that
+ * rebindings may cause evictions and hence the validation-rebind
+ * sequence is rerun until there are no more objects to validate.
+ *
+ * Return: 0 on success, negative error code on error. In particular,
+ * may return -EINTR or -ERESTARTSYS if interrupted, and -EDEADLK if
+ * the drm_exec transaction needs to be restarted.
+ */
+int xe_vm_validate_rebind(struct xe_vm *vm, struct drm_exec *exec,
+			  unsigned int num_fences)
+{
+	struct drm_gem_object *obj;
+	unsigned long index;
+	int ret;
+
+	do {
+		ret = drm_gpuvm_validate(&vm->gpuvm, exec);
+		if (ret)
+			return ret;
+
+		ret = xe_vm_rebind(vm, false);
+		if (ret)
+			return ret;
+	} while (!list_empty(&vm->gpuvm.evict.list));
+
+	drm_exec_for_each_locked_object(exec, index, obj) {
+		ret = dma_resv_reserve_fences(obj->resv, num_fences);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
 static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
 				 bool *done)
 {
 	int err;
 
-	/*
-	 * 1 fence for each preempt fence plus a fence for each tile from a
-	 * possible rebind
-	 */
-	err = drm_gpuvm_prepare_vm(&vm->gpuvm, exec, vm->preempt.num_exec_queues +
-				   vm->xe->info.tile_count);
+	err = drm_gpuvm_prepare_vm(&vm->gpuvm, exec, 0);
 	if (err)
 		return err;
 
@@ -505,7 +543,7 @@ static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
 		return 0;
 	}
 
-	err = drm_gpuvm_prepare_objects(&vm->gpuvm, exec, vm->preempt.num_exec_queues);
+	err = drm_gpuvm_prepare_objects(&vm->gpuvm, exec, 0);
 	if (err)
 		return err;
 
@@ -513,7 +551,13 @@ static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
 	if (err)
 		return err;
 
-	return drm_gpuvm_validate(&vm->gpuvm, exec);
+	/*
+	 * Add validation and rebinding to the locking loop since both can
+	 * cause evictions which may require blocing dma_resv locks.
+	 * The fence reservation here is intended for the new preempt fences
+	 * we attach at the end of the rebind work.
+	 */
+	return xe_vm_validate_rebind(vm, exec, vm->preempt.num_exec_queues);
 }
 
 static void preempt_rebind_work_func(struct work_struct *w)
@@ -801,6 +845,7 @@ static void xe_vma_free(struct xe_vma *vma)
 
 #define VMA_CREATE_FLAG_READ_ONLY	BIT(0)
 #define VMA_CREATE_FLAG_IS_NULL		BIT(1)
+#define VMA_CREATE_FLAG_DUMPABLE	BIT(2)
 
 static struct xe_vma *xe_vma_create(struct xe_vm *vm,
 				    struct xe_bo *bo,
@@ -813,6 +858,7 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
 	u8 id;
 	bool read_only = (flags & VMA_CREATE_FLAG_READ_ONLY);
 	bool is_null = (flags & VMA_CREATE_FLAG_IS_NULL);
+	bool dumpable = (flags & VMA_CREATE_FLAG_DUMPABLE);
 
 	xe_assert(vm->xe, start < end);
 	xe_assert(vm->xe, end < vm->size);
@@ -847,6 +893,8 @@ static struct xe_vma *xe_vma_create(struct xe_vm *vm,
 	vma->gpuva.va.range = end - start + 1;
 	if (read_only)
 		vma->gpuva.flags |= XE_VMA_READ_ONLY;
+	if (dumpable)
+		vma->gpuva.flags |= XE_VMA_DUMPABLE;
 
 	for_each_tile(tile, vm->xe, id)
 		vma->tile_mask |= 0x1 << id;
@@ -990,35 +1038,26 @@ static void xe_vma_destroy(struct xe_vma *vma, struct dma_fence *fence)
 }
 
 /**
- * xe_vm_prepare_vma() - drm_exec utility to lock a vma
+ * xe_vm_lock_vma() - drm_exec utility to lock a vma
  * @exec: The drm_exec object we're currently locking for.
  * @vma: The vma for witch we want to lock the vm resv and any attached
  * object's resv.
- * @num_shared: The number of dma-fence slots to pre-allocate in the
- * objects' reservation objects.
  *
  * Return: 0 on success, negative error code on error. In particular
  * may return -EDEADLK on WW transaction contention and -EINTR if
  * an interruptible wait is terminated by a signal.
  */
-int xe_vm_prepare_vma(struct drm_exec *exec, struct xe_vma *vma,
-		      unsigned int num_shared)
+int xe_vm_lock_vma(struct drm_exec *exec, struct xe_vma *vma)
 {
 	struct xe_vm *vm = xe_vma_vm(vma);
 	struct xe_bo *bo = xe_vma_bo(vma);
 	int err;
 
 	XE_WARN_ON(!vm);
-	if (num_shared)
-		err = drm_exec_prepare_obj(exec, xe_vm_obj(vm), num_shared);
-	else
-		err = drm_exec_lock_obj(exec, xe_vm_obj(vm));
-	if (!err && bo && !bo->vm) {
-		if (num_shared)
-			err = drm_exec_prepare_obj(exec, &bo->ttm.base, num_shared);
-		else
-			err = drm_exec_lock_obj(exec, &bo->ttm.base);
-	}
+
+	err = drm_exec_lock_obj(exec, xe_vm_obj(vm));
+	if (!err && bo && !bo->vm)
+		err = drm_exec_lock_obj(exec, &bo->ttm.base);
 
 	return err;
 }
@@ -1030,7 +1069,7 @@ static void xe_vma_destroy_unlocked(struct xe_vma *vma)
 
 	drm_exec_init(&exec, 0, 0);
 	drm_exec_until_all_locked(&exec) {
-		err = xe_vm_prepare_vma(&exec, vma, 0);
+		err = xe_vm_lock_vma(&exec, vma);
 		drm_exec_retry_on_contention(&exec);
 		if (XE_WARN_ON(err))
 			break;
@@ -1065,7 +1104,9 @@ static int xe_vm_insert_vma(struct xe_vm *vm, struct xe_vma *vma)
 	xe_assert(vm->xe, xe_vma_vm(vma) == vm);
 	lockdep_assert_held(&vm->lock);
 
+	mutex_lock(&vm->snap_mutex);
 	err = drm_gpuva_insert(&vm->gpuvm, &vma->gpuva);
+	mutex_unlock(&vm->snap_mutex);
 	XE_WARN_ON(err);	/* Shouldn't be possible */
 
 	return err;
@@ -1076,7 +1117,9 @@ static void xe_vm_remove_vma(struct xe_vm *vm, struct xe_vma *vma)
 	xe_assert(vm->xe, xe_vma_vm(vma) == vm);
 	lockdep_assert_held(&vm->lock);
 
+	mutex_lock(&vm->snap_mutex);
 	drm_gpuva_remove(&vma->gpuva);
+	mutex_unlock(&vm->snap_mutex);
 	if (vm->usm.last_fault_vma == vma)
 		vm->usm.last_fault_vma = NULL;
 }
@@ -1095,7 +1138,7 @@ static struct drm_gpuva_op *xe_vm_op_alloc(void)
 
 static void xe_vm_free(struct drm_gpuvm *gpuvm);
 
-static struct drm_gpuvm_ops gpuvm_ops = {
+static const struct drm_gpuvm_ops gpuvm_ops = {
 	.op_alloc = xe_vm_op_alloc,
 	.vm_bo_validate = xe_gpuvm_validate,
 	.vm_free = xe_vm_free,
@@ -1303,6 +1346,7 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
 	vm->flags = flags;
 
 	init_rwsem(&vm->lock);
+	mutex_init(&vm->snap_mutex);
 
 	INIT_LIST_HEAD(&vm->rebind_list);
 
@@ -1428,6 +1472,7 @@ err_close:
 	return ERR_PTR(err);
 
 err_no_resv:
+	mutex_destroy(&vm->snap_mutex);
 	for_each_tile(tile, xe, id)
 		xe_range_fence_tree_fini(&vm->rftree[id]);
 	kfree(vm);
@@ -1532,6 +1577,16 @@ void xe_vm_close_and_put(struct xe_vm *vm)
 		xe->usm.num_vm_in_fault_mode--;
 	else if (!(vm->flags & XE_VM_FLAG_MIGRATION))
 		xe->usm.num_vm_in_non_fault_mode--;
+
+	if (vm->usm.asid) {
+		void *lookup;
+
+		xe_assert(xe, xe->info.has_asid);
+		xe_assert(xe, !(vm->flags & XE_VM_FLAG_MIGRATION));
+
+		lookup = xa_erase(&xe->usm.asid_to_vm, vm->usm.asid);
+		xe_assert(xe, lookup == vm);
+	}
 	mutex_unlock(&xe->usm.lock);
 
 	for_each_tile(tile, xe, id)
@@ -1547,21 +1602,17 @@ static void vm_destroy_work_func(struct work_struct *w)
 	struct xe_device *xe = vm->xe;
 	struct xe_tile *tile;
 	u8 id;
-	void *lookup;
 
 	/* xe_vm_close_and_put was not called? */
 	xe_assert(xe, !vm->size);
 
-	if (!(vm->flags & XE_VM_FLAG_MIGRATION)) {
-		xe_device_mem_access_put(xe);
+	if (xe_vm_in_preempt_fence_mode(vm))
+		flush_work(&vm->preempt.rebind_work);
 
-		if (xe->info.has_asid && vm->usm.asid) {
-			mutex_lock(&xe->usm.lock);
-			lookup = xa_erase(&xe->usm.asid_to_vm, vm->usm.asid);
-			xe_assert(xe, lookup == vm);
-			mutex_unlock(&xe->usm.lock);
-		}
-	}
+	mutex_destroy(&vm->snap_mutex);
+
+	if (!(vm->flags & XE_VM_FLAG_MIGRATION))
+		xe_device_mem_access_put(xe);
 
 	for_each_tile(tile, xe, id)
 		XE_WARN_ON(vm->pt_root[id]);
@@ -2163,6 +2214,7 @@ vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo,
 
 		if (__op->op == DRM_GPUVA_OP_MAP) {
 			op->map.is_null = flags & DRM_XE_VM_BIND_FLAG_NULL;
+			op->map.dumpable = flags & DRM_XE_VM_BIND_FLAG_DUMPABLE;
 			op->map.pat_index = pat_index;
 		} else if (__op->op == DRM_GPUVA_OP_PREFETCH) {
 			op->prefetch.region = prefetch_region;
@@ -2328,6 +2380,7 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q,
 				   struct xe_sync_entry *syncs, u32 num_syncs,
 				   struct list_head *ops_list, bool last)
 {
+	struct xe_device *xe = vm->xe;
 	struct xe_vma_op *last_op = NULL;
 	struct drm_gpuva_op *__op;
 	int err = 0;
@@ -2356,6 +2409,8 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q,
 		{
 			flags |= op->map.is_null ?
 				VMA_CREATE_FLAG_IS_NULL : 0;
+			flags |= op->map.dumpable ?
+				VMA_CREATE_FLAG_DUMPABLE : 0;
 
 			vma = new_vma(vm, &op->base.map, op->map.pat_index,
 				      flags);
@@ -2380,6 +2435,9 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q,
 				flags |= op->base.remap.unmap->va->flags &
 					DRM_GPUVA_SPARSE ?
 					VMA_CREATE_FLAG_IS_NULL : 0;
+				flags |= op->base.remap.unmap->va->flags &
+					XE_VMA_DUMPABLE ?
+					VMA_CREATE_FLAG_DUMPABLE : 0;
 
 				vma = new_vma(vm, op->base.remap.prev,
 					      old->pat_index, flags);
@@ -2401,6 +2459,9 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q,
 						xe_vma_end(vma) -
 						xe_vma_start(old);
 					op->remap.start = xe_vma_end(vma);
+					vm_dbg(&xe->drm, "REMAP:SKIP_PREV: addr=0x%016llx, range=0x%016llx",
+					       (ULL)op->remap.start,
+					       (ULL)op->remap.range);
 				}
 			}
 
@@ -2411,6 +2472,9 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q,
 				flags |= op->base.remap.unmap->va->flags &
 					DRM_GPUVA_SPARSE ?
 					VMA_CREATE_FLAG_IS_NULL : 0;
+				flags |= op->base.remap.unmap->va->flags &
+					XE_VMA_DUMPABLE ?
+					VMA_CREATE_FLAG_DUMPABLE : 0;
 
 				vma = new_vma(vm, op->base.remap.next,
 					      old->pat_index, flags);
@@ -2431,6 +2495,9 @@ static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q,
 					op->remap.range -=
 						xe_vma_end(old) -
 						xe_vma_start(vma);
+					vm_dbg(&xe->drm, "REMAP:SKIP_NEXT: addr=0x%016llx, range=0x%016llx",
+					       (ULL)op->remap.start,
+					       (ULL)op->remap.range);
 				}
 			}
 			break;
@@ -2473,7 +2540,7 @@ static int op_execute(struct drm_exec *exec, struct xe_vm *vm,
 
 	lockdep_assert_held_write(&vm->lock);
 
-	err = xe_vm_prepare_vma(exec, vma, 1);
+	err = xe_vm_lock_vma(exec, vma);
 	if (err)
 		return err;
 
@@ -3284,3 +3351,168 @@ int xe_analyze_vm(struct drm_printer *p, struct xe_vm *vm, int gt_id)
 
 	return 0;
 }
+
+struct xe_vm_snapshot {
+	unsigned long num_snaps;
+	struct {
+		u64 ofs, bo_ofs;
+		unsigned long len;
+		struct xe_bo *bo;
+		void *data;
+		struct mm_struct *mm;
+	} snap[];
+};
+
+struct xe_vm_snapshot *xe_vm_snapshot_capture(struct xe_vm *vm)
+{
+	unsigned long num_snaps = 0, i;
+	struct xe_vm_snapshot *snap = NULL;
+	struct drm_gpuva *gpuva;
+
+	if (!vm)
+		return NULL;
+
+	mutex_lock(&vm->snap_mutex);
+	drm_gpuvm_for_each_va(gpuva, &vm->gpuvm) {
+		if (gpuva->flags & XE_VMA_DUMPABLE)
+			num_snaps++;
+	}
+
+	if (num_snaps)
+		snap = kvzalloc(offsetof(struct xe_vm_snapshot, snap[num_snaps]), GFP_NOWAIT);
+	if (!snap)
+		goto out_unlock;
+
+	snap->num_snaps = num_snaps;
+	i = 0;
+	drm_gpuvm_for_each_va(gpuva, &vm->gpuvm) {
+		struct xe_vma *vma = gpuva_to_vma(gpuva);
+		struct xe_bo *bo = vma->gpuva.gem.obj ?
+			gem_to_xe_bo(vma->gpuva.gem.obj) : NULL;
+
+		if (!(gpuva->flags & XE_VMA_DUMPABLE))
+			continue;
+
+		snap->snap[i].ofs = xe_vma_start(vma);
+		snap->snap[i].len = xe_vma_size(vma);
+		if (bo) {
+			snap->snap[i].bo = xe_bo_get(bo);
+			snap->snap[i].bo_ofs = xe_vma_bo_offset(vma);
+		} else if (xe_vma_is_userptr(vma)) {
+			struct mm_struct *mm =
+				to_userptr_vma(vma)->userptr.notifier.mm;
+
+			if (mmget_not_zero(mm))
+				snap->snap[i].mm = mm;
+			else
+				snap->snap[i].data = ERR_PTR(-EFAULT);
+
+			snap->snap[i].bo_ofs = xe_vma_userptr(vma);
+		} else {
+			snap->snap[i].data = ERR_PTR(-ENOENT);
+		}
+		i++;
+	}
+
+out_unlock:
+	mutex_unlock(&vm->snap_mutex);
+	return snap;
+}
+
+void xe_vm_snapshot_capture_delayed(struct xe_vm_snapshot *snap)
+{
+	for (int i = 0; i < snap->num_snaps; i++) {
+		struct xe_bo *bo = snap->snap[i].bo;
+		struct iosys_map src;
+		int err;
+
+		if (IS_ERR(snap->snap[i].data))
+			continue;
+
+		snap->snap[i].data = kvmalloc(snap->snap[i].len, GFP_USER);
+		if (!snap->snap[i].data) {
+			snap->snap[i].data = ERR_PTR(-ENOMEM);
+			goto cleanup_bo;
+		}
+
+		if (bo) {
+			dma_resv_lock(bo->ttm.base.resv, NULL);
+			err = ttm_bo_vmap(&bo->ttm, &src);
+			if (!err) {
+				xe_map_memcpy_from(xe_bo_device(bo),
+						   snap->snap[i].data,
+						   &src, snap->snap[i].bo_ofs,
+						   snap->snap[i].len);
+				ttm_bo_vunmap(&bo->ttm, &src);
+			}
+			dma_resv_unlock(bo->ttm.base.resv);
+		} else {
+			void __user *userptr = (void __user *)(size_t)snap->snap[i].bo_ofs;
+
+			kthread_use_mm(snap->snap[i].mm);
+			if (!copy_from_user(snap->snap[i].data, userptr, snap->snap[i].len))
+				err = 0;
+			else
+				err = -EFAULT;
+			kthread_unuse_mm(snap->snap[i].mm);
+
+			mmput(snap->snap[i].mm);
+			snap->snap[i].mm = NULL;
+		}
+
+		if (err) {
+			kvfree(snap->snap[i].data);
+			snap->snap[i].data = ERR_PTR(err);
+		}
+
+cleanup_bo:
+		xe_bo_put(bo);
+		snap->snap[i].bo = NULL;
+	}
+}
+
+void xe_vm_snapshot_print(struct xe_vm_snapshot *snap, struct drm_printer *p)
+{
+	unsigned long i, j;
+
+	for (i = 0; i < snap->num_snaps; i++) {
+		if (IS_ERR(snap->snap[i].data))
+			goto uncaptured;
+
+		drm_printf(p, "[%llx].length: 0x%lx\n", snap->snap[i].ofs, snap->snap[i].len);
+		drm_printf(p, "[%llx].data: ",
+			   snap->snap[i].ofs);
+
+		for (j = 0; j < snap->snap[i].len; j += sizeof(u32)) {
+			u32 *val = snap->snap[i].data + j;
+			char dumped[ASCII85_BUFSZ];
+
+			drm_puts(p, ascii85_encode(*val, dumped));
+		}
+
+		drm_puts(p, "\n");
+		continue;
+
+uncaptured:
+		drm_printf(p, "Unable to capture range [%llx-%llx]: %li\n",
+			   snap->snap[i].ofs, snap->snap[i].ofs + snap->snap[i].len - 1,
+			   PTR_ERR(snap->snap[i].data));
+	}
+}
+
+void xe_vm_snapshot_free(struct xe_vm_snapshot *snap)
+{
+	unsigned long i;
+
+	if (!snap)
+		return;
+
+	for (i = 0; i < snap->num_snaps; i++) {
+		if (!IS_ERR(snap->snap[i].data))
+			kvfree(snap->snap[i].data);
+		xe_bo_put(snap->snap[i].bo);
+		if (snap->snap[i].mm)
+			mmput(snap->snap[i].mm);
+	}
+	kvfree(snap);
+}