From 3afb00d3f86d3d924f88b56fa8285d4e9db85852 Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Wed, 7 Aug 2024 15:17:52 +0200 Subject: Merging upstream version 6.10.3. Signed-off-by: Daniel Baumann --- drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 5 ++ drivers/gpu/drm/amd/amdkfd/kfd_device.c | 4 +- .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 3 +- drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c | 32 ++++++++++--- drivers/gpu/drm/amd/amdkfd/kfd_int_process_v11.c | 9 ++-- drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 53 ++++++++++++++-------- drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c | 2 + drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 6 +-- drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c | 18 ++++++++ drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h | 4 +- drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c | 6 +-- drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c | 6 +-- drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c | 6 +-- drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c | 27 +++++++++-- drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c | 6 +-- drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 6 +-- drivers/gpu/drm/amd/amdkfd/kfd_svm.h | 1 - 17 files changed, 135 insertions(+), 59 deletions(-) (limited to 'drivers/gpu/drm/amd/amdkfd') diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index 1e6cc0bfc4..fdf171ad4a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -371,6 +371,11 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, err = -EINVAL; goto err_wptr_map_gart; } + if (dev->adev != amdgpu_ttm_adev(wptr_bo->tbo.bdev)) { + pr_err("Queue memory allocated to wrong device\n"); + err = -EINVAL; + goto err_wptr_map_gart; + } err = amdgpu_amdkfd_map_gtt_bo_to_gart(wptr_bo); if (err) { diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index ff01610fbc..afc57df421 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -428,12 +428,12 @@ struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf) if (!f2g) { if (amdgpu_ip_version(adev, GC_HWIP, 0)) - dev_err(kfd_device, + dev_info(kfd_device, "GC IP %06x %s not supported in kfd\n", amdgpu_ip_version(adev, GC_HWIP, 0), vf ? "VF" : ""); else - dev_err(kfd_device, "%s %s not supported in kfd\n", + dev_info(kfd_device, "%s %s not supported in kfd\n", amdgpu_asic_name[adev->asic_type], vf ? "VF" : ""); return NULL; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c index 0b655555e1..c08b6ee252 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -1997,8 +1997,7 @@ static int unmap_queues_cpsch(struct device_queue_manager *dqm, * check those fields */ mqd_mgr = dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]; - if (mqd_mgr->read_doorbell_id(dqm->packet_mgr.priv_queue->queue->mqd)) { - dev_err(dev, "HIQ MQD's queue_doorbell_id0 is not 0, Queue preemption time out\n"); + if (mqd_mgr->check_preemption_failed(mqd_mgr, dqm->packet_mgr.priv_queue->queue->mqd)) { while (halt_if_hws_hang) schedule(); kfd_hws_hang(dqm); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c index 40a21be6c0..8e0d0356e8 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c @@ -134,6 +134,7 @@ static void event_interrupt_poison_consumption(struct kfd_node *dev, { enum amdgpu_ras_block block = 0; int old_poison, ret = -EINVAL; + uint32_t reset = 0; struct kfd_process *p = kfd_lookup_process_by_pasid(pasid); if (!p) @@ -153,6 +154,8 @@ static void event_interrupt_poison_consumption(struct kfd_node *dev, case SOC15_IH_CLIENTID_UTCL2: ret = kfd_dqm_evict_pasid(dev->dqm, pasid); block = AMDGPU_RAS_BLOCK__GFX; + if (ret) + reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET; break; case SOC15_IH_CLIENTID_SDMA0: case SOC15_IH_CLIENTID_SDMA1: @@ -160,6 +163,7 @@ static void event_interrupt_poison_consumption(struct kfd_node *dev, case SOC15_IH_CLIENTID_SDMA3: case SOC15_IH_CLIENTID_SDMA4: block = AMDGPU_RAS_BLOCK__SDMA; + reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET; break; default: break; @@ -170,17 +174,16 @@ static void event_interrupt_poison_consumption(struct kfd_node *dev, /* resetting queue passes, do page retirement without gpu reset * resetting queue fails, fallback to gpu reset solution */ - if (!ret) { + if (!ret) dev_warn(dev->adev->dev, "RAS poison consumption, unmap queue flow succeeded: client id %d\n", client_id); - amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, false); - } else { + else dev_warn(dev->adev->dev, "RAS poison consumption, fall back to gpu reset flow: client id %d\n", client_id); - amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, true); - } + + amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, reset); } static bool event_interrupt_isr_v10(struct kfd_node *dev, @@ -368,10 +371,25 @@ static void event_interrupt_wq_v10(struct kfd_node *dev, client_id == SOC15_IH_CLIENTID_UTCL2) { struct kfd_vm_fault_info info = {0}; uint16_t ring_id = SOC15_RING_ID_FROM_IH_ENTRY(ih_ring_entry); + uint32_t node_id = SOC15_NODEID_FROM_IH_ENTRY(ih_ring_entry); + uint32_t vmid_type = SOC15_VMID_TYPE_FROM_IH_ENTRY(ih_ring_entry); + int hub_inst = 0; struct kfd_hsa_memory_exception_data exception_data; - if (client_id == SOC15_IH_CLIENTID_UTCL2 && - amdgpu_amdkfd_ras_query_utcl2_poison_status(dev->adev)) { + /* gfxhub */ + if (!vmid_type && dev->adev->gfx.funcs->ih_node_to_logical_xcc) { + hub_inst = dev->adev->gfx.funcs->ih_node_to_logical_xcc(dev->adev, + node_id); + if (hub_inst < 0) + hub_inst = 0; + } + + /* mmhub */ + if (vmid_type && client_id == SOC15_IH_CLIENTID_VMC) + hub_inst = node_id / 4; + + if (amdgpu_amdkfd_ras_query_utcl2_poison_status(dev->adev, + hub_inst, vmid_type)) { event_interrupt_poison_consumption(dev, pasid, client_id); return; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v11.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v11.c index fe2ad0c0de..f524a55eee 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v11.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v11.c @@ -193,6 +193,7 @@ static void event_interrupt_poison_consumption_v11(struct kfd_node *dev, { enum amdgpu_ras_block block = 0; int ret = -EINVAL; + uint32_t reset = 0; struct kfd_process *p = kfd_lookup_process_by_pasid(pasid); if (!p) @@ -212,10 +213,13 @@ static void event_interrupt_poison_consumption_v11(struct kfd_node *dev, if (dev->dqm->ops.reset_queues) ret = dev->dqm->ops.reset_queues(dev->dqm, pasid); block = AMDGPU_RAS_BLOCK__GFX; + if (ret) + reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET; break; case SOC21_INTSRC_SDMA_ECC: default: block = AMDGPU_RAS_BLOCK__GFX; + reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET; break; } @@ -223,10 +227,7 @@ static void event_interrupt_poison_consumption_v11(struct kfd_node *dev, /* resetting queue passes, do page retirement without gpu reset resetting queue fails, fallback to gpu reset solution */ - if (!ret) - amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, false); - else - amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, true); + amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, reset); } static bool event_interrupt_isr_v11(struct kfd_node *dev, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c index c4c6a29052..e1c21d2506 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c @@ -144,7 +144,8 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev, uint16_t pasid, uint16_t client_id) { enum amdgpu_ras_block block = 0; - int old_poison, ret = -EINVAL; + int old_poison; + uint32_t reset = 0; struct kfd_process *p = kfd_lookup_process_by_pasid(pasid); if (!p) @@ -162,8 +163,13 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev, case SOC15_IH_CLIENTID_SE2SH: case SOC15_IH_CLIENTID_SE3SH: case SOC15_IH_CLIENTID_UTCL2: - ret = kfd_dqm_evict_pasid(dev->dqm, pasid); block = AMDGPU_RAS_BLOCK__GFX; + reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET; + break; + case SOC15_IH_CLIENTID_VMC: + case SOC15_IH_CLIENTID_VMC1: + block = AMDGPU_RAS_BLOCK__MMHUB; + reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET; break; case SOC15_IH_CLIENTID_SDMA0: case SOC15_IH_CLIENTID_SDMA1: @@ -171,27 +177,21 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev, case SOC15_IH_CLIENTID_SDMA3: case SOC15_IH_CLIENTID_SDMA4: block = AMDGPU_RAS_BLOCK__SDMA; + reset = AMDGPU_RAS_GPU_RESET_MODE2_RESET; break; default: - break; + dev_warn(dev->adev->dev, + "client %d does not support poison consumption\n", client_id); + return; } kfd_signal_poison_consumed_event(dev, pasid); - /* resetting queue passes, do page retirement without gpu reset - * resetting queue fails, fallback to gpu reset solution - */ - if (!ret) { - dev_warn(dev->adev->dev, - "RAS poison consumption, unmap queue flow succeeded: client id %d\n", - client_id); - amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, false); - } else { - dev_warn(dev->adev->dev, - "RAS poison consumption, fall back to gpu reset flow: client id %d\n", - client_id); - amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, true); - } + dev_warn(dev->adev->dev, + "poison is consumed by client %d, kick off gpu reset flow\n", client_id); + + amdgpu_amdkfd_ras_pasid_poison_consumption_handler(dev->adev, + block, pasid, NULL, NULL, reset); } static bool context_id_expected(struct kfd_dev *dev) @@ -414,10 +414,25 @@ static void event_interrupt_wq_v9(struct kfd_node *dev, client_id == SOC15_IH_CLIENTID_UTCL2) { struct kfd_vm_fault_info info = {0}; uint16_t ring_id = SOC15_RING_ID_FROM_IH_ENTRY(ih_ring_entry); + uint32_t node_id = SOC15_NODEID_FROM_IH_ENTRY(ih_ring_entry); + uint32_t vmid_type = SOC15_VMID_TYPE_FROM_IH_ENTRY(ih_ring_entry); + int hub_inst = 0; struct kfd_hsa_memory_exception_data exception_data; - if (client_id == SOC15_IH_CLIENTID_UTCL2 && - amdgpu_amdkfd_ras_query_utcl2_poison_status(dev->adev)) { + /* gfxhub */ + if (!vmid_type && dev->adev->gfx.funcs->ih_node_to_logical_xcc) { + hub_inst = dev->adev->gfx.funcs->ih_node_to_logical_xcc(dev->adev, + node_id); + if (hub_inst < 0) + hub_inst = 0; + } + + /* mmhub */ + if (vmid_type && client_id == SOC15_IH_CLIENTID_VMC) + hub_inst = node_id / 4; + + if (amdgpu_amdkfd_ras_query_utcl2_poison_status(dev->adev, + hub_inst, vmid_type)) { event_interrupt_poison_consumption_v9(dev, pasid, client_id); return; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c index dd3c43c1ad..9b6b6e8825 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c @@ -104,6 +104,8 @@ void kfd_interrupt_exit(struct kfd_node *node) */ flush_workqueue(node->ih_wq); + destroy_workqueue(node->ih_wq); + kfifo_free(&node->ih_fifo); } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index ba651d12f1..8ee3d07ffb 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -77,7 +77,7 @@ svm_migrate_gart_map(struct amdgpu_ring *ring, uint64_t npages, dst_addr = amdgpu_bo_gpu_offset(adev->gart.bo); amdgpu_emit_copy_buffer(adev, &job->ibs[0], src_addr, - dst_addr, num_bytes, false); + dst_addr, num_bytes, 0); amdgpu_ring_pad_ib(ring, &job->ibs[0]); WARN_ON(job->ibs[0].length_dw > num_dw); @@ -153,7 +153,7 @@ svm_migrate_copy_memory_gart(struct amdgpu_device *adev, dma_addr_t *sys, } r = amdgpu_copy_buffer(ring, gart_s, gart_d, size * PAGE_SIZE, - NULL, &next, false, true, false); + NULL, &next, false, true, 0); if (r) { dev_err(adev->dev, "fail %d to copy memory\n", r); goto out_unlock; @@ -1023,7 +1023,7 @@ int kgd2kfd_init_zone_device(struct amdgpu_device *adev) if (amdgpu_ip_version(adev, GC_HWIP, 0) < IP_VERSION(9, 0, 1)) return -EINVAL; - if (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) + if (adev->flags & AMD_IS_APU) return 0; pgmap = &kfddev->pgmap; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c index 050a6936ff..8746a61a85 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c @@ -290,3 +290,21 @@ uint64_t kfd_mqd_stride(struct mqd_manager *mm, { return mm->mqd_size; } + +bool kfd_check_hiq_mqd_doorbell_id(struct kfd_node *node, uint32_t doorbell_id, + uint32_t inst) +{ + if (doorbell_id) { + struct device *dev = node->adev->dev; + + if (node->adev->xcp_mgr && node->adev->xcp_mgr->num_xcps > 0) + dev_err(dev, "XCC %d: Queue preemption failed for queue with doorbell_id: %x\n", + inst, doorbell_id); + else + dev_err(dev, "Queue preemption failed for queue with doorbell_id: %x\n", + doorbell_id); + return true; + } + + return false; +} diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h index e5cc697a3c..17cc1f25c8 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h @@ -119,7 +119,7 @@ struct mqd_manager { #if defined(CONFIG_DEBUG_FS) int (*debugfs_show_mqd)(struct seq_file *m, void *data); #endif - uint32_t (*read_doorbell_id)(void *mqd); + bool (*check_preemption_failed)(struct mqd_manager *mm, void *mqd); uint64_t (*mqd_stride)(struct mqd_manager *mm, struct queue_properties *p); @@ -198,4 +198,6 @@ void kfd_get_hiq_xcc_mqd(struct kfd_node *dev, uint64_t kfd_hiq_mqd_stride(struct kfd_node *dev); uint64_t kfd_mqd_stride(struct mqd_manager *mm, struct queue_properties *q); +bool kfd_check_hiq_mqd_doorbell_id(struct kfd_node *node, uint32_t doorbell_id, + uint32_t inst); #endif /* KFD_MQD_MANAGER_H_ */ diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c index 1a4a69943c..05f3ac2eae 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c @@ -206,11 +206,11 @@ static void __update_mqd(struct mqd_manager *mm, void *mqd, q->is_active = QUEUE_IS_ACTIVE(*q); } -static uint32_t read_doorbell_id(void *mqd) +static bool check_preemption_failed(struct mqd_manager *mm, void *mqd) { struct cik_mqd *m = (struct cik_mqd *)mqd; - return m->queue_doorbell_id0; + return kfd_check_hiq_mqd_doorbell_id(mm->dev, m->queue_doorbell_id0, 0); } static void update_mqd(struct mqd_manager *mm, void *mqd, @@ -423,7 +423,7 @@ struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type, #if defined(CONFIG_DEBUG_FS) mqd->debugfs_show_mqd = debugfs_show_mqd; #endif - mqd->read_doorbell_id = read_doorbell_id; + mqd->check_preemption_failed = check_preemption_failed; break; case KFD_MQD_TYPE_DIQ: mqd->allocate_mqd = allocate_mqd; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c index 22cbfa1bda..2eff37aaf8 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c @@ -224,11 +224,11 @@ static void update_mqd(struct mqd_manager *mm, void *mqd, q->is_active = QUEUE_IS_ACTIVE(*q); } -static uint32_t read_doorbell_id(void *mqd) +static bool check_preemption_failed(struct mqd_manager *mm, void *mqd) { struct v10_compute_mqd *m = (struct v10_compute_mqd *)mqd; - return m->queue_doorbell_id0; + return kfd_check_hiq_mqd_doorbell_id(mm->dev, m->queue_doorbell_id0, 0); } static int get_wave_state(struct mqd_manager *mm, void *mqd, @@ -488,7 +488,7 @@ struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type, #if defined(CONFIG_DEBUG_FS) mqd->debugfs_show_mqd = debugfs_show_mqd; #endif - mqd->read_doorbell_id = read_doorbell_id; + mqd->check_preemption_failed = check_preemption_failed; pr_debug("%s@%i\n", __func__, __LINE__); break; case KFD_MQD_TYPE_DIQ: diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c index 826bc4f6c8..68dbc0399c 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c @@ -278,11 +278,11 @@ static void update_mqd(struct mqd_manager *mm, void *mqd, q->is_active = QUEUE_IS_ACTIVE(*q); } -static uint32_t read_doorbell_id(void *mqd) +static bool check_preemption_failed(struct mqd_manager *mm, void *mqd) { struct v11_compute_mqd *m = (struct v11_compute_mqd *)mqd; - return m->queue_doorbell_id0; + return kfd_check_hiq_mqd_doorbell_id(mm->dev, m->queue_doorbell_id0, 0); } static int get_wave_state(struct mqd_manager *mm, void *mqd, @@ -517,7 +517,7 @@ struct mqd_manager *mqd_manager_init_v11(enum KFD_MQD_TYPE type, #if defined(CONFIG_DEBUG_FS) mqd->debugfs_show_mqd = debugfs_show_mqd; #endif - mqd->read_doorbell_id = read_doorbell_id; + mqd->check_preemption_failed = check_preemption_failed; pr_debug("%s@%i\n", __func__, __LINE__); break; case KFD_MQD_TYPE_DIQ: diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c index 697b6d530d..8ec136eba5 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c @@ -316,11 +316,11 @@ static void update_mqd(struct mqd_manager *mm, void *mqd, } -static uint32_t read_doorbell_id(void *mqd) +static bool check_preemption_failed(struct mqd_manager *mm, void *mqd) { struct v9_mqd *m = (struct v9_mqd *)mqd; - return m->queue_doorbell_id0; + return kfd_check_hiq_mqd_doorbell_id(mm->dev, m->queue_doorbell_id0, 0); } static int get_wave_state(struct mqd_manager *mm, void *mqd, @@ -607,6 +607,24 @@ static int destroy_hiq_mqd_v9_4_3(struct mqd_manager *mm, void *mqd, return err; } +static bool check_preemption_failed_v9_4_3(struct mqd_manager *mm, void *mqd) +{ + uint64_t hiq_mqd_size = kfd_hiq_mqd_stride(mm->dev); + uint32_t xcc_mask = mm->dev->xcc_mask; + int inst = 0, xcc_id; + struct v9_mqd *m; + bool ret = false; + + for_each_inst(xcc_id, xcc_mask) { + m = get_mqd(mqd + hiq_mqd_size * inst); + ret |= kfd_check_hiq_mqd_doorbell_id(mm->dev, + m->queue_doorbell_id0, inst); + ++inst; + } + + return ret; +} + static void get_xcc_mqd(struct kfd_mem_obj *mqd_mem_obj, struct kfd_mem_obj *xcc_mqd_mem_obj, uint64_t offset) @@ -695,7 +713,7 @@ static void update_mqd_v9_4_3(struct mqd_manager *mm, void *mqd, m = get_mqd(mqd + size * xcc); update_mqd(mm, m, q, minfo); - update_cu_mask(mm, mqd, minfo, xcc); + update_cu_mask(mm, m, minfo, xcc); if (q->format == KFD_QUEUE_FORMAT_AQL) { switch (xcc) { @@ -881,15 +899,16 @@ struct mqd_manager *mqd_manager_init_v9(enum KFD_MQD_TYPE type, #if defined(CONFIG_DEBUG_FS) mqd->debugfs_show_mqd = debugfs_show_mqd; #endif - mqd->read_doorbell_id = read_doorbell_id; if (KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 3)) { mqd->init_mqd = init_mqd_hiq_v9_4_3; mqd->load_mqd = hiq_load_mqd_kiq_v9_4_3; mqd->destroy_mqd = destroy_hiq_mqd_v9_4_3; + mqd->check_preemption_failed = check_preemption_failed_v9_4_3; } else { mqd->init_mqd = init_mqd_hiq; mqd->load_mqd = kfd_hiq_load_mqd_kiq; mqd->destroy_mqd = destroy_hiq_mqd; + mqd->check_preemption_failed = check_preemption_failed; } break; case KFD_MQD_TYPE_DIQ: diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c index 3e1a574d4e..c1fafc5025 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c @@ -237,11 +237,11 @@ static void __update_mqd(struct mqd_manager *mm, void *mqd, q->is_active = QUEUE_IS_ACTIVE(*q); } -static uint32_t read_doorbell_id(void *mqd) +static bool check_preemption_failed(struct mqd_manager *mm, void *mqd) { struct vi_mqd *m = (struct vi_mqd *)mqd; - return m->queue_doorbell_id0; + return kfd_check_hiq_mqd_doorbell_id(mm->dev, m->queue_doorbell_id0, 0); } static void update_mqd(struct mqd_manager *mm, void *mqd, @@ -482,7 +482,7 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type, #if defined(CONFIG_DEBUG_FS) mqd->debugfs_show_mqd = debugfs_show_mqd; #endif - mqd->read_doorbell_id = read_doorbell_id; + mqd->check_preemption_failed = check_preemption_failed; break; case KFD_MQD_TYPE_DIQ: mqd->allocate_mqd = allocate_mqd; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index 069b81eeea..31e500859a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -2619,8 +2619,7 @@ svm_range_best_restore_location(struct svm_range *prange, return -1; } - if (node->adev->gmc.is_app_apu || - node->adev->flags & AMD_IS_APU) + if (node->adev->flags & AMD_IS_APU) return 0; if (prange->preferred_loc == gpuid || @@ -3338,8 +3337,7 @@ svm_range_best_prefetch_location(struct svm_range *prange) goto out; } - if (bo_node->adev->gmc.is_app_apu || - bo_node->adev->flags & AMD_IS_APU) { + if (bo_node->adev->flags & AMD_IS_APU) { best_loc = 0; goto out; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h index 9c37bd0567..70c1776611 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h @@ -201,7 +201,6 @@ void svm_range_list_lock_and_flush_work(struct svm_range_list *svms, struct mm_s * is initialized to not 0 when page migration register device memory. */ #define KFD_IS_SVM_API_SUPPORTED(adev) ((adev)->kfd.pgmap.type != 0 ||\ - (adev)->gmc.is_app_apu ||\ ((adev)->flags & AMD_IS_APU)) void svm_range_bo_unref_async(struct svm_range_bo *svm_bo); -- cgit v1.2.3