From 71a8d61ebc38f0f1b96011518fefd0a1e07efa74 Mon Sep 17 00:00:00 2001 From: Tao Zhou Date: Mon, 19 Feb 2024 15:55:24 +0800 Subject: drm/amdgpu: retire gfx ras query_utcl2_poison_status Replace it with related interface in gfxhub functions. v2: replace node id with xcc id. get node id for query_utcl2_poison_status Signed-off-by: Tao Zhou Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index 35dd6effa9a3..d5fde8adf19b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -769,10 +769,11 @@ int amdgpu_amdkfd_send_close_event_drain_irq(struct amdgpu_device *adev, return 0; } -bool amdgpu_amdkfd_ras_query_utcl2_poison_status(struct amdgpu_device *adev) +bool amdgpu_amdkfd_ras_query_utcl2_poison_status(struct amdgpu_device *adev, + int xcc_id) { - if (adev->gfx.ras && adev->gfx.ras->query_utcl2_poison_status) - return adev->gfx.ras->query_utcl2_poison_status(adev); + if (adev->gfxhub.funcs->query_utcl2_poison_status) + return adev->gfxhub.funcs->query_utcl2_poison_status(adev, xcc_id); else return false; } -- cgit v1.2.3-59-g8ed1b From d8070c4241087d8c1ad3fc21632bac268dd7578a Mon Sep 17 00:00:00 2001 From: Tao Zhou Date: Mon, 11 Mar 2024 16:53:53 +0800 Subject: drm/amdgpu: support utcl2 RAS poison query for mmhub Support the query for both gfxhub and mmhub, also replace xcc_id with hub_inst. Signed-off-by: Tao Zhou Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 17 ++++++++++++----- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 5 ++--- drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c | 17 +++++++++++------ drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 17 +++++++++++------ 5 files changed, 37 insertions(+), 21 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index d5fde8adf19b..8ee18c2c082a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -770,12 +770,19 @@ int amdgpu_amdkfd_send_close_event_drain_irq(struct amdgpu_device *adev, } bool amdgpu_amdkfd_ras_query_utcl2_poison_status(struct amdgpu_device *adev, - int xcc_id) + int hub_inst, int hub_type) { - if (adev->gfxhub.funcs->query_utcl2_poison_status) - return adev->gfxhub.funcs->query_utcl2_poison_status(adev, xcc_id); - else - return false; + if (!hub_type) { + if (adev->gfxhub.funcs->query_utcl2_poison_status) + return adev->gfxhub.funcs->query_utcl2_poison_status(adev, hub_inst); + else + return false; + } else { + if (adev->mmhub.funcs->query_utcl2_poison_status) + return adev->mmhub.funcs->query_utcl2_poison_status(adev, hub_inst); + else + return false; + } } int amdgpu_amdkfd_check_and_lock_kfd(struct amdgpu_device *adev) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index caee36e52a09..6b67f0025966 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -342,7 +342,7 @@ bool amdgpu_amdkfd_bo_mapped_to_dev(struct amdgpu_device *adev, struct kgd_mem * void amdgpu_amdkfd_block_mmu_notifications(void *p); int amdgpu_amdkfd_criu_resume(void *p); bool amdgpu_amdkfd_ras_query_utcl2_poison_status(struct amdgpu_device *adev, - int xcc_id); + int hub_inst, int hub_type); int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev, uint64_t size, u32 alloc_flag, int8_t xcp_id); void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev, diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c index 05d52b9274a9..470a146f2f43 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c @@ -666,10 +666,9 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev, rw = REG_GET_FIELD(status, VM_L2_PROTECTION_FAULT_STATUS, RW); fed = REG_GET_FIELD(status, VM_L2_PROTECTION_FAULT_STATUS, FED); - /* for gfx fed error, kfd will handle it, return directly */ + /* for fed error, kfd will handle it, return directly */ if (fed && amdgpu_ras_is_poison_mode_supported(adev) && - (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(9, 4, 2)) && - (vmhub < AMDGPU_MMHUB0_START)) + (amdgpu_ip_version(adev, GC_HWIP, 0) >= IP_VERSION(9, 4, 2))) return 0; WREG32_P(hub->vm_l2_pro_fault_cntl, 1, ~1); 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 a8e76287dde0..650da18b0d87 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c @@ -369,18 +369,23 @@ static void event_interrupt_wq_v10(struct kfd_node *dev, 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 xcc_id = 0; + int hub_inst = 0; struct kfd_hsa_memory_exception_data exception_data; + /* gfxhub */ if (!vmid_type && dev->adev->gfx.funcs->ih_node_to_logical_xcc) { - xcc_id = dev->adev->gfx.funcs->ih_node_to_logical_xcc(dev->adev, + hub_inst = dev->adev->gfx.funcs->ih_node_to_logical_xcc(dev->adev, node_id); - if (xcc_id < 0) - xcc_id = 0; + if (hub_inst < 0) + hub_inst = 0; } - if (client_id == SOC15_IH_CLIENTID_UTCL2 && !vmid_type && - amdgpu_amdkfd_ras_query_utcl2_poison_status(dev->adev, xcc_id)) { + /* 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_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c index ff7392336795..11641f4645e6 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c @@ -415,18 +415,23 @@ static void event_interrupt_wq_v9(struct kfd_node *dev, 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 xcc_id = 0; + int hub_inst = 0; struct kfd_hsa_memory_exception_data exception_data; + /* gfxhub */ if (!vmid_type && dev->adev->gfx.funcs->ih_node_to_logical_xcc) { - xcc_id = dev->adev->gfx.funcs->ih_node_to_logical_xcc(dev->adev, + hub_inst = dev->adev->gfx.funcs->ih_node_to_logical_xcc(dev->adev, node_id); - if (xcc_id < 0) - xcc_id = 0; + if (hub_inst < 0) + hub_inst = 0; } - if (client_id == SOC15_IH_CLIENTID_UTCL2 && !vmid_type && - amdgpu_amdkfd_ras_query_utcl2_poison_status(dev->adev, xcc_id)) { + /* 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; } -- cgit v1.2.3-59-g8ed1b From 2fc46e0b2fe8802c98a68d5c903e0109e1d4a49c Mon Sep 17 00:00:00 2001 From: Tao Zhou Date: Tue, 12 Mar 2024 11:30:09 +0800 Subject: drm/amdgpu: make reset method configurable for RAS poison Each RAS block has different requirement for gpu reset in poison consumption handling. Add support for mmhub RAS poison consumption handling. v2: remove the mmhub poison support for kfd int v10. Signed-off-by: Tao Zhou Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 16 +++++++--------- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h | 4 ++-- drivers/gpu/drm/amd/amdkfd/kfd_int_process_v10.c | 13 ++++++++----- drivers/gpu/drm/amd/amdkfd/kfd_int_process_v11.c | 9 +++++---- drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 20 +++++++++++++++----- 8 files changed, 41 insertions(+), 29 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index 8ee18c2c082a..3b4591f554f1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -748,7 +748,7 @@ bool amdgpu_amdkfd_is_fed(struct amdgpu_device *adev) } void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev, - enum amdgpu_ras_block block, bool reset) + enum amdgpu_ras_block block, uint32_t reset) { amdgpu_umc_poison_handler(adev, block, reset); } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index 6b67f0025966..c51954c9052e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -336,7 +336,7 @@ void amdgpu_amdkfd_debug_mem_fence(struct amdgpu_device *adev); int amdgpu_amdkfd_get_tile_config(struct amdgpu_device *adev, struct tile_config *config); void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev, - enum amdgpu_ras_block block, bool reset); + enum amdgpu_ras_block block, uint32_t reset); bool amdgpu_amdkfd_is_fed(struct amdgpu_device *adev); bool amdgpu_amdkfd_bo_mapped_to_dev(struct amdgpu_device *adev, struct kgd_mem *mem); void amdgpu_amdkfd_block_mmu_notifications(void *p); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c index 26662c76b293..3c6d532824f6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c @@ -2051,7 +2051,7 @@ static void amdgpu_ras_interrupt_poison_consumption_handler(struct ras_manager * } } - amdgpu_umc_poison_handler(adev, obj->head.block, false); + amdgpu_umc_poison_handler(adev, obj->head.block, 0); if (block_obj->hw_ops && block_obj->hw_ops->handle_poison_consumption) poison_stat = block_obj->hw_ops->handle_poison_consumption(adev); @@ -2704,7 +2704,7 @@ static int amdgpu_ras_page_retirement_thread(void *param) atomic_dec(&con->page_retirement_req_cnt); amdgpu_umc_bad_page_polling_timeout(adev, - false, MAX_UMC_POISON_POLLING_TIME_ASYNC); + 0, MAX_UMC_POISON_POLLING_TIME_ASYNC); } return 0; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index 20436f81856a..f486510fc94c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -177,7 +177,7 @@ static void amdgpu_umc_handle_bad_pages(struct amdgpu_device *adev, static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev, void *ras_error_status, struct amdgpu_iv_entry *entry, - bool reset) + uint32_t reset) { struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; struct amdgpu_ras *con = amdgpu_ras_get_context(adev); @@ -186,9 +186,7 @@ static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev, amdgpu_umc_handle_bad_pages(adev, ras_error_status); if (err_data->ue_count && reset) { - /* use mode-2 reset for poison consumption */ - if (!entry) - con->gpu_reset_flags |= AMDGPU_RAS_GPU_RESET_MODE2_RESET; + con->gpu_reset_flags |= reset; amdgpu_ras_reset_gpu(adev); } @@ -196,7 +194,7 @@ static int amdgpu_umc_do_page_retirement(struct amdgpu_device *adev, } int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev, - bool reset, uint32_t timeout_ms) + uint32_t reset, uint32_t timeout_ms) { struct ras_err_data err_data; struct ras_common_if head = { @@ -238,8 +236,7 @@ int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev, if (reset) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); - /* use mode-2 reset for poison consumption */ - con->gpu_reset_flags |= AMDGPU_RAS_GPU_RESET_MODE2_RESET; + con->gpu_reset_flags |= reset; amdgpu_ras_reset_gpu(adev); } @@ -247,7 +244,7 @@ int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev, } int amdgpu_umc_poison_handler(struct amdgpu_device *adev, - enum amdgpu_ras_block block, bool reset) + enum amdgpu_ras_block block, uint32_t reset) { int ret = AMDGPU_RAS_SUCCESS; @@ -311,7 +308,8 @@ int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev, void *ras_error_status, struct amdgpu_iv_entry *entry) { - return amdgpu_umc_do_page_retirement(adev, ras_error_status, entry, true); + return amdgpu_umc_do_page_retirement(adev, ras_error_status, entry, + AMDGPU_RAS_GPU_RESET_MODE1_RESET); } int amdgpu_umc_ras_sw_init(struct amdgpu_device *adev) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h index 5954e839d580..563b0249247e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h @@ -101,7 +101,7 @@ struct amdgpu_umc { int amdgpu_umc_ras_sw_init(struct amdgpu_device *adev); int amdgpu_umc_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block); int amdgpu_umc_poison_handler(struct amdgpu_device *adev, - enum amdgpu_ras_block block, bool reset); + enum amdgpu_ras_block block, uint32_t reset); int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry); @@ -121,5 +121,5 @@ int amdgpu_umc_loop_channels(struct amdgpu_device *adev, umc_func func, void *data); int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev, - bool reset, uint32_t timeout_ms); + uint32_t reset, uint32_t timeout_ms); #endif 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 650da18b0d87..740f89aafbc0 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, 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 7e2859736a55..d3d6b5c180b3 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 11641f4645e6..2a37ab7a7150 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c @@ -145,6 +145,7 @@ static void event_interrupt_poison_consumption_v9(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) @@ -164,6 +165,15 @@ static void event_interrupt_poison_consumption_v9(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_VMC: + case SOC15_IH_CLIENTID_VMC1: + ret = kfd_dqm_evict_pasid(dev->dqm, pasid); + block = AMDGPU_RAS_BLOCK__MMHUB; + if (ret) + reset = AMDGPU_RAS_GPU_RESET_MODE1_RESET; break; case SOC15_IH_CLIENTID_SDMA0: case SOC15_IH_CLIENTID_SDMA1: @@ -171,6 +181,7 @@ 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; @@ -181,17 +192,16 @@ static void event_interrupt_poison_consumption_v9(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 context_id_expected(struct kfd_dev *dev) -- cgit v1.2.3-59-g8ed1b From bfa579b38b865879223f61f6ae295c939d4f7d11 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Mon, 22 Apr 2024 17:40:03 +0800 Subject: drm/amdgpu: prepare to handle pasid poison consumption Prepare to handle pasid poison consumption. Signed-off-by: YiPeng Chai Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 9 ++++++++- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 5 +++++ drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c | 20 +++++++++++++------- drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h | 3 +++ drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 3 ++- 5 files changed, 31 insertions(+), 9 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index 3b4591f554f1..7ba05f030dd1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -747,10 +747,17 @@ bool amdgpu_amdkfd_is_fed(struct amdgpu_device *adev) return amdgpu_ras_get_fed_status(adev); } +void amdgpu_amdkfd_ras_pasid_poison_consumption_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset) +{ + amdgpu_umc_pasid_poison_handler(adev, block, pasid, pasid_fn, data, reset); +} + void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev, enum amdgpu_ras_block block, uint32_t reset) { - amdgpu_umc_poison_handler(adev, block, reset); + amdgpu_umc_pasid_poison_handler(adev, block, 0, NULL, NULL, reset); } int amdgpu_amdkfd_send_close_event_drain_irq(struct amdgpu_device *adev, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index c51954c9052e..1de021ebdd46 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -337,6 +337,11 @@ int amdgpu_amdkfd_get_tile_config(struct amdgpu_device *adev, struct tile_config *config); void amdgpu_amdkfd_ras_poison_consumption_handler(struct amdgpu_device *adev, enum amdgpu_ras_block block, uint32_t reset); + +void amdgpu_amdkfd_ras_pasid_poison_consumption_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset); + bool amdgpu_amdkfd_is_fed(struct amdgpu_device *adev); bool amdgpu_amdkfd_bo_mapped_to_dev(struct amdgpu_device *adev, struct kgd_mem *mem); void amdgpu_amdkfd_block_mmu_notifications(void *p); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c index dcda3d24bee3..8ebbca9e2e22 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.c @@ -252,8 +252,9 @@ int amdgpu_umc_bad_page_polling_timeout(struct amdgpu_device *adev, return 0; } -int amdgpu_umc_poison_handler(struct amdgpu_device *adev, - enum amdgpu_ras_block block, uint32_t reset) +int amdgpu_umc_pasid_poison_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset) { int ret = AMDGPU_RAS_SUCCESS; @@ -291,16 +292,14 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev, amdgpu_ras_error_data_fini(&err_data); } else { - if (reset) { - amdgpu_umc_bad_page_polling_timeout(adev, - reset, MAX_UMC_POISON_POLLING_TIME_SYNC); - } else { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); + amdgpu_ras_put_poison_req(adev, + block, pasid, pasid_fn, data, reset); + atomic_inc(&con->page_retirement_req_cnt); wake_up(&con->page_retirement_wq); - } } } else { if (adev->virt.ops && adev->virt.ops->ras_poison_handler) @@ -313,6 +312,13 @@ int amdgpu_umc_poison_handler(struct amdgpu_device *adev, return ret; } +int amdgpu_umc_poison_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint32_t reset) +{ + return amdgpu_umc_pasid_poison_handler(adev, + block, 0, NULL, NULL, reset); +} + int amdgpu_umc_process_ras_data_cb(struct amdgpu_device *adev, void *ras_error_status, struct amdgpu_iv_entry *entry) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h index 9e77e6d48e3b..5f50c69c3cec 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_umc.h @@ -106,6 +106,9 @@ int amdgpu_umc_ras_sw_init(struct amdgpu_device *adev); int amdgpu_umc_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *ras_block); int amdgpu_umc_poison_handler(struct amdgpu_device *adev, enum amdgpu_ras_block block, uint32_t reset); +int amdgpu_umc_pasid_poison_handler(struct amdgpu_device *adev, + enum amdgpu_ras_block block, uint16_t pasid, + pasid_notify pasid_fn, void *data, uint32_t reset); int amdgpu_umc_process_ecc_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry); 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 c3beb872adf8..e1c21d250611 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c @@ -190,7 +190,8 @@ static void event_interrupt_poison_consumption_v9(struct kfd_node *dev, dev_warn(dev->adev->dev, "poison is consumed by client %d, kick off gpu reset flow\n", client_id); - amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, block, reset); + amdgpu_amdkfd_ras_pasid_poison_consumption_handler(dev->adev, + block, pasid, NULL, NULL, reset); } static bool context_id_expected(struct kfd_dev *dev) -- cgit v1.2.3-59-g8ed1b From eb853413d02c8d9b27942429b261a9eef228f005 Mon Sep 17 00:00:00 2001 From: Lang Yu Date: Fri, 26 Apr 2024 14:56:35 +0800 Subject: drm/amdkfd: Let VRAM allocations go to GTT domain on small APUs Small APUs(i.e., consumer, embedded products) usually have a small carveout device memory which can't satisfy most compute workloads memory allocation requirements. We can't even run a Basic MNIST Example with a default 512MB carveout. https://github.com/pytorch/examples/tree/main/mnist. Error Log: "torch.cuda.OutOfMemoryError: HIP out of memory. Tried to allocate 84.00 MiB. GPU 0 has a total capacity of 512.00 MiB of which 0 bytes is free. Of the allocated memory 103.83 MiB is allocated by PyTorch, and 22.17 MiB is reserved by PyTorch but unallocated" Though we can change BIOS settings to enlarge carveout size, which is inflexible and may bring complaint. On the other hand, the memory resource can't be effectively used between host and device. The solution is MI300A approach, i.e., let VRAM allocations go to GTT. Then device and host can flexibly and effectively share memory resource. v2: Report local_mem_size_private as 0. (Felix) Signed-off-by: Lang Yu Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 5 +++++ drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 20 +++++++++++--------- drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 2 +- drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 6 ++++-- drivers/gpu/drm/amd/amdkfd/kfd_svm.h | 3 ++- 5 files changed, 23 insertions(+), 13 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index 7ba05f030dd1..e3738d417245 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -455,6 +455,9 @@ void amdgpu_amdkfd_get_local_mem_info(struct amdgpu_device *adev, else mem_info->local_mem_size_private = KFD_XCP_MEMORY_SIZE(adev, xcp->id); + } else if (adev->flags & AMD_IS_APU) { + mem_info->local_mem_size_public = (ttm_tt_pages_limit() << PAGE_SHIFT); + mem_info->local_mem_size_private = 0; } else { mem_info->local_mem_size_public = adev->gmc.visible_vram_size; mem_info->local_mem_size_private = adev->gmc.real_vram_size - @@ -824,6 +827,8 @@ u64 amdgpu_amdkfd_xcp_memory_size(struct amdgpu_device *adev, int xcp_id) } do_div(tmp, adev->xcp_mgr->num_xcp_per_mem_partition); return ALIGN_DOWN(tmp, PAGE_SIZE); + } else if (adev->flags & AMD_IS_APU) { + return (ttm_tt_pages_limit() << PAGE_SHIFT); } else { return adev->gmc.real_vram_size; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index 0535b07987d9..8975cf41a91a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -196,7 +196,7 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev, return -EINVAL; vram_size = KFD_XCP_MEMORY_SIZE(adev, xcp_id); - if (adev->gmc.is_app_apu) { + if (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) { system_mem_needed = size; ttm_mem_needed = size; } @@ -232,7 +232,8 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev, "adev reference can't be null when vram is used"); if (adev && xcp_id >= 0) { adev->kfd.vram_used[xcp_id] += vram_needed; - adev->kfd.vram_used_aligned[xcp_id] += adev->gmc.is_app_apu ? + adev->kfd.vram_used_aligned[xcp_id] += + (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) ? vram_needed : ALIGN(vram_needed, VRAM_AVAILABLITY_ALIGN); } @@ -260,7 +261,7 @@ void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev, if (adev) { adev->kfd.vram_used[xcp_id] -= size; - if (adev->gmc.is_app_apu) { + if (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) { adev->kfd.vram_used_aligned[xcp_id] -= size; kfd_mem_limit.system_mem_used -= size; kfd_mem_limit.ttm_mem_used -= size; @@ -889,7 +890,7 @@ static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem, * if peer device has large BAR. In contrast, access over xGMI is * allowed for both small and large BAR configurations of peer device */ - if ((adev != bo_adev && !adev->gmc.is_app_apu) && + if ((adev != bo_adev && !(adev->gmc.is_app_apu || adev->flags & AMD_IS_APU)) && ((mem->domain == AMDGPU_GEM_DOMAIN_VRAM) || (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_DOORBELL) || (mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP))) { @@ -1657,7 +1658,7 @@ size_t amdgpu_amdkfd_get_available_memory(struct amdgpu_device *adev, - atomic64_read(&adev->vram_pin_size) - reserved_for_pt; - if (adev->gmc.is_app_apu) { + if (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) { system_mem_available = no_system_mem_limit ? kfd_mem_limit.max_system_mem_limit : kfd_mem_limit.max_system_mem_limit - @@ -1705,7 +1706,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( if (flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) { domain = alloc_domain = AMDGPU_GEM_DOMAIN_VRAM; - if (adev->gmc.is_app_apu) { + if (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) { domain = AMDGPU_GEM_DOMAIN_GTT; alloc_domain = AMDGPU_GEM_DOMAIN_GTT; alloc_flags = 0; @@ -1952,7 +1953,7 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu( if (size) { if (!is_imported && (mem->bo->preferred_domains == AMDGPU_GEM_DOMAIN_VRAM || - (adev->gmc.is_app_apu && + ((adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) && mem->bo->preferred_domains == AMDGPU_GEM_DOMAIN_GTT))) *size = bo_size; else @@ -2374,8 +2375,9 @@ static int import_obj_create(struct amdgpu_device *adev, (*mem)->dmabuf = dma_buf; (*mem)->bo = bo; (*mem)->va = va; - (*mem)->domain = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) && !adev->gmc.is_app_apu ? - AMDGPU_GEM_DOMAIN_VRAM : AMDGPU_GEM_DOMAIN_GTT; + (*mem)->domain = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) && + !(adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) ? + AMDGPU_GEM_DOMAIN_VRAM : AMDGPU_GEM_DOMAIN_GTT; (*mem)->mapped_to_gpu_memory = 0; (*mem)->process_info = avm->process_info; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index 4bcfbeac48fb..4816fcb9803a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -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) + if (adev->gmc.is_app_apu || adev->flags & AMD_IS_APU) return 0; pgmap = &kfddev->pgmap; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index 386875e6eb96..069b81eeea03 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -2619,7 +2619,8 @@ svm_range_best_restore_location(struct svm_range *prange, return -1; } - if (node->adev->gmc.is_app_apu) + if (node->adev->gmc.is_app_apu || + node->adev->flags & AMD_IS_APU) return 0; if (prange->preferred_loc == gpuid || @@ -3337,7 +3338,8 @@ svm_range_best_prefetch_location(struct svm_range *prange) goto out; } - if (bo_node->adev->gmc.is_app_apu) { + if (bo_node->adev->gmc.is_app_apu || + 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 026863a0abcd..9c37bd0567ef 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h @@ -201,7 +201,8 @@ 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)->gmc.is_app_apu ||\ + ((adev)->flags & AMD_IS_APU)) void svm_range_bo_unref_async(struct svm_range_bo *svm_bo); -- cgit v1.2.3-59-g8ed1b