From 3b37e2725ab32c9055bec00ef41caa63839efd37 Mon Sep 17 00:00:00 2001 From: Yifan Zhang Date: Thu, 18 Jul 2024 13:18:53 +0800 Subject: drm/amdgpu: skip kfd init if GFX is not ready. avoid kfd init crash in that case. Signed-off-by: Yifan Zhang Acked-by: Alex Deucher Tested-by: Jesse Zhang Reviewed-by: Jesse Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index bcacf2e35eba..730dae77570c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -2471,6 +2471,7 @@ out: */ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev) { + struct amdgpu_ip_block *ip_block; struct pci_dev *parent; int i, r; bool total; @@ -2608,7 +2609,10 @@ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev) if (!total) return -ENODEV; - amdgpu_amdkfd_device_probe(adev); + ip_block = amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_GFX); + if (ip_block->status.valid != false) + amdgpu_amdkfd_device_probe(adev); + adev->cg_flags &= amdgpu_cg_mask; adev->pg_flags &= amdgpu_pg_mask; -- cgit From 83b501c1799a96a41e163973e88826253ffadfb3 Mon Sep 17 00:00:00 2001 From: Christian König Date: Fri, 19 Jul 2024 17:24:14 +0200 Subject: drm/scheduler: remove full_recover from drm_sched_start MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This was basically just another one of amdgpus hacks. The parameter allowed to restart the scheduler without turning fence signaling on again. That this is absolutely not a good idea should be obvious by now since the fences will then just sit there and never signal. While at it cleanup the code a bit. Signed-off-by: Christian König Reviewed-by: Matthew Brost Link: https://patchwork.freedesktop.org/patch/msgid/20240722083816.99685-1-christian.koenig@amd.com --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 861ccff78af9..c186fdb198ad 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5862,7 +5862,7 @@ skip_hw_reset: if (!amdgpu_ring_sched_ready(ring)) continue; - drm_sched_start(&ring->sched, true); + drm_sched_start(&ring->sched); } if (!drm_drv_uses_atomic_modeset(adev_to_drm(tmp_adev)) && !job_signaled) @@ -6360,7 +6360,7 @@ void amdgpu_pci_resume(struct pci_dev *pdev) if (!amdgpu_ring_sched_ready(ring)) continue; - drm_sched_start(&ring->sched, true); + drm_sched_start(&ring->sched); } amdgpu_device_unset_mp1_state(adev); -- cgit From 13d8850a3387635c051c5ed1c8a8b6c1e9bd1341 Mon Sep 17 00:00:00 2001 From: Sunil Khatri Date: Fri, 26 Jul 2024 18:07:41 +0530 Subject: drm/amdgpu: trigger ip dump before suspend of IP's Problem: IP dump right now is done post suspend of all IP's which for some IP's could change power state and software state too which we do not want to reflect in the dump as it might not be same at the time of hang. Solution: IP should be dumped as close to the HW state when the GPU was in hung state without trying to reinitialize any resource. Acked-by: Alex Deucher Signed-off-by: Sunil Khatri Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 60 +++++++++++++++--------------- 1 file changed, 30 insertions(+), 30 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 730dae77570c..a2a1a3da17e3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5277,11 +5277,29 @@ mode1_reset_failed: return ret; } +static int amdgpu_reset_reg_dumps(struct amdgpu_device *adev) +{ + int i; + + lockdep_assert_held(&adev->reset_domain->sem); + + for (i = 0; i < adev->reset_info.num_regs; i++) { + adev->reset_info.reset_dump_reg_value[i] = + RREG32(adev->reset_info.reset_dump_reg_list[i]); + + trace_amdgpu_reset_reg_dumps(adev->reset_info.reset_dump_reg_list[i], + adev->reset_info.reset_dump_reg_value[i]); + } + + return 0; +} + int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, struct amdgpu_reset_context *reset_context) { int i, r = 0; struct amdgpu_job *job = NULL; + struct amdgpu_device *tmp_adev = reset_context->reset_req_dev; bool need_full_reset = test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); @@ -5340,6 +5358,18 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, } } + if (!test_bit(AMDGPU_SKIP_COREDUMP, &reset_context->flags)) { + amdgpu_reset_reg_dumps(tmp_adev); + + dev_info(tmp_adev->dev, "Dumping IP State\n"); + /* Trigger ip dump before we reset the asic */ + for (i = 0; i < tmp_adev->num_ip_blocks; i++) + if (tmp_adev->ip_blocks[i].version->funcs->dump_ip_state) + tmp_adev->ip_blocks[i].version->funcs + ->dump_ip_state((void *)tmp_adev); + dev_info(tmp_adev->dev, "Dumping IP State Completed\n"); + } + if (need_full_reset) r = amdgpu_device_ip_suspend(adev); if (need_full_reset) @@ -5352,47 +5382,17 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, return r; } -static int amdgpu_reset_reg_dumps(struct amdgpu_device *adev) -{ - int i; - - lockdep_assert_held(&adev->reset_domain->sem); - - for (i = 0; i < adev->reset_info.num_regs; i++) { - adev->reset_info.reset_dump_reg_value[i] = - RREG32(adev->reset_info.reset_dump_reg_list[i]); - - trace_amdgpu_reset_reg_dumps(adev->reset_info.reset_dump_reg_list[i], - adev->reset_info.reset_dump_reg_value[i]); - } - - return 0; -} - int amdgpu_do_asic_reset(struct list_head *device_list_handle, struct amdgpu_reset_context *reset_context) { struct amdgpu_device *tmp_adev = NULL; bool need_full_reset, skip_hw_reset, vram_lost = false; int r = 0; - uint32_t i; /* Try reset handler method first */ tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device, reset_list); - if (!test_bit(AMDGPU_SKIP_COREDUMP, &reset_context->flags)) { - amdgpu_reset_reg_dumps(tmp_adev); - - dev_info(tmp_adev->dev, "Dumping IP State\n"); - /* Trigger ip dump before we reset the asic */ - for (i = 0; i < tmp_adev->num_ip_blocks; i++) - if (tmp_adev->ip_blocks[i].version->funcs->dump_ip_state) - tmp_adev->ip_blocks[i].version->funcs - ->dump_ip_state((void *)tmp_adev); - dev_info(tmp_adev->dev, "Dumping IP State Completed\n"); - } - reset_context->reset_device_list = device_list_handle; r = amdgpu_reset_perform_reset(tmp_adev, reset_context); /* If reset handler not implemented, continue; otherwise return */ -- cgit From 836af5be1b6d8e93d736c252e711a20db7dbde9d Mon Sep 17 00:00:00 2001 From: Sunil Khatri Date: Tue, 30 Jul 2024 11:19:53 +0530 Subject: drm/amdgpu: Clean up the register dump via debugfs list MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit debugfs register list for dump is cleaned as it have some issues related to proper power state of the IP before register read. Since the above mentioned is removed we no longer want this to be dumped part of the devcoredump and hence removed. Reviewed-by: Christian König Signed-off-by: Sunil Khatri Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 19 ------------------- 1 file changed, 19 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index a2a1a3da17e3..3a43754e7f10 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5277,23 +5277,6 @@ mode1_reset_failed: return ret; } -static int amdgpu_reset_reg_dumps(struct amdgpu_device *adev) -{ - int i; - - lockdep_assert_held(&adev->reset_domain->sem); - - for (i = 0; i < adev->reset_info.num_regs; i++) { - adev->reset_info.reset_dump_reg_value[i] = - RREG32(adev->reset_info.reset_dump_reg_list[i]); - - trace_amdgpu_reset_reg_dumps(adev->reset_info.reset_dump_reg_list[i], - adev->reset_info.reset_dump_reg_value[i]); - } - - return 0; -} - int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, struct amdgpu_reset_context *reset_context) { @@ -5359,8 +5342,6 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, } if (!test_bit(AMDGPU_SKIP_COREDUMP, &reset_context->flags)) { - amdgpu_reset_reg_dumps(tmp_adev); - dev_info(tmp_adev->dev, "Dumping IP State\n"); /* Trigger ip dump before we reset the asic */ for (i = 0; i < tmp_adev->num_ip_blocks; i++) -- cgit From dd3e296289346109d41c6317124f51aee0269c25 Mon Sep 17 00:00:00 2001 From: Tao Zhou Date: Thu, 1 Aug 2024 16:26:27 +0800 Subject: drm/amdgpu: update bad state check in GPU recovery Return RMA status without message print. Signed-off-by: Tao Zhou Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 3a43754e7f10..f595ba6ba7c7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5498,7 +5498,7 @@ int amdgpu_do_asic_reset(struct list_head *device_list_handle, * bad_page_threshold value to fix this once * probing driver again. */ - if (!amdgpu_ras_eeprom_check_err_threshold(tmp_adev)) { + if (!amdgpu_ras_is_rma(tmp_adev)) { /* must succeed. */ amdgpu_ras_resume(tmp_adev); } else { -- cgit From b3a3c9a6b27b68310f1d4d486f47556808c7c855 Mon Sep 17 00:00:00 2001 From: Tao Zhou Date: Wed, 31 Jul 2024 15:54:27 +0800 Subject: drm/amdgpu: report bad status in GPU recovery Instead of printing GPU reset failed. v2: add check for reset_context->src. Signed-off-by: Tao Zhou Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index f595ba6ba7c7..29a4adee9286 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5876,8 +5876,14 @@ skip_hw_reset: tmp_adev->asic_reset_res = 0; if (r) { - /* bad news, how to tell it to userspace ? */ - dev_info(tmp_adev->dev, "GPU reset(%d) failed\n", atomic_read(&tmp_adev->gpu_reset_counter)); + /* bad news, how to tell it to userspace ? + * for ras error, we should report GPU bad status instead of + * reset failure + */ + if (reset_context->src != AMDGPU_RESET_SRC_RAS || + !amdgpu_ras_eeprom_check_err_threshold(tmp_adev)) + dev_info(tmp_adev->dev, "GPU reset(%d) failed\n", + atomic_read(&tmp_adev->gpu_reset_counter)); amdgpu_vf_error_put(tmp_adev, AMDGIM_ERROR_VF_GPU_RESET_FAIL, 0, r); } else { dev_info(tmp_adev->dev, "GPU reset(%d) succeeded!\n", atomic_read(&tmp_adev->gpu_reset_counter)); -- cgit From f83cec3b3a7c968bbceb810b7acd1baf3fe8cd87 Mon Sep 17 00:00:00 2001 From: Victor Skvortsov Date: Thu, 8 Aug 2024 13:22:34 -0400 Subject: drm/amdgpu: Disable dpm_enabled flag while VF is in reset VFs do not perform HW fini/suspend in FLR, so the dpm_enabled is incorrectly kept enabled. Add interface to disable it in virt_pre_reset call. v2: Made implementation generic for all asics v3: Re-order conditionals so PP_MP1_STATE_FLR is only evaluated on VF Signed-off-by: Victor Skvortsov Reviewed-by: Lijo Lazar Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 29a4adee9286..a6b8d0ba4758 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5289,10 +5289,8 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, if (reset_context->reset_req_dev == adev) job = reset_context->job; - if (amdgpu_sriov_vf(adev)) { - /* stop the data exchange thread */ - amdgpu_virt_fini_data_exchange(adev); - } + if (amdgpu_sriov_vf(adev)) + amdgpu_virt_pre_reset(adev); amdgpu_fence_driver_isr_toggle(adev, true); -- cgit From 76acba7b7f12517990f326fabfecb6f55e334233 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Fri, 12 Jul 2024 16:37:33 -0400 Subject: drm/amdgpu/gfx11: add a mutex for the gfx semaphore This will be used in more places in the future so add a mutex. Acked-by: Vitaly Prosyak Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 1 + 1 file changed, 1 insertion(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index a6b8d0ba4758..482db4ebcc4b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4059,6 +4059,7 @@ int amdgpu_device_init(struct amdgpu_device *adev, mutex_init(&adev->notifier_lock); mutex_init(&adev->pm.stable_pstate_ctx_lock); mutex_init(&adev->benchmark_mutex); + mutex_init(&adev->gfx.reset_sem_mutex); amdgpu_device_init_apu_flags(adev); -- cgit From 96595204195d7e13736a84295e217316610d4cdb Mon Sep 17 00:00:00 2001 From: Srinivasan Shanmugam Date: Mon, 29 Jul 2024 21:35:26 +0530 Subject: drm/amdgpu: Make enforce_isolation setting per GPU MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This commit makes enforce_isolation setting to be per GPU and per partition by adding the enforce_isolation array to the adev structure. The adev variable is set based on the global enforce_isolation module parameter during device initialization. In amdgpu_ids.c, the adev->enforce_isolation value for the current GPU is used to determine whether to enforce isolation between graphics and compute processes on that GPU. In amdgpu_ids.c, the adev->enforce_isolation value for the current GPU and partition is used to determine whether to enforce isolation between graphics and compute processes on that GPU and partition. This allows the enforce_isolation setting to be controlled individually for each GPU and each partition, which is useful in a system with multiple GPUs and partitions where different isolation settings might be desired for different GPUs and partitions. v2: fix loop in amdgpu_vmid_mgr_init() (Alex) Cc: Christian König Cc: Alex Deucher Signed-off-by: Srinivasan Shanmugam Signed-off-by: Alex Deucher Suggested-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 482db4ebcc4b..e623af740aa3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -1916,6 +1916,8 @@ static int amdgpu_device_init_apu_flags(struct amdgpu_device *adev) */ static int amdgpu_device_check_arguments(struct amdgpu_device *adev) { + int i; + if (amdgpu_sched_jobs < 4) { dev_warn(adev->dev, "sched jobs (%d) must be at least 4\n", amdgpu_sched_jobs); @@ -1970,6 +1972,9 @@ static int amdgpu_device_check_arguments(struct amdgpu_device *adev) adev->firmware.load_type = amdgpu_ucode_get_load_type(adev, amdgpu_fw_load_type); + for (i = 0; i < MAX_XCP; i++) + adev->enforce_isolation[i] = !!enforce_isolation; + return 0; } -- cgit From e189be9b2e3820c88164d95090f1fd6343cd77fc Mon Sep 17 00:00:00 2001 From: Srinivasan Shanmugam Date: Mon, 27 May 2024 07:30:47 +0530 Subject: drm/amdgpu: Add enforce_isolation sysfs attribute MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This commit adds a new sysfs attribute 'enforce_isolation' to control the 'enforce_isolation' setting per GPU. The attribute can be read and written, and accepts values 0 (disabled) and 1 (enabled). When 'enforce_isolation' is enabled, reserved VMIDs are allocated for each ring. When it's disabled, the reserved VMIDs are freed. The set function locks a mutex before changing the 'enforce_isolation' flag and the VMIDs, and unlocks it afterwards. This ensures that these operations are atomic and prevents race conditions and other concurrency issues. Cc: Christian König Cc: Alex Deucher Signed-off-by: Srinivasan Shanmugam Suggested-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index e623af740aa3..2f1bc02309fe 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4065,6 +4065,8 @@ int amdgpu_device_init(struct amdgpu_device *adev, mutex_init(&adev->pm.stable_pstate_ctx_lock); mutex_init(&adev->benchmark_mutex); mutex_init(&adev->gfx.reset_sem_mutex); + /* Initialize the mutex for cleaner shader isolation between GFX and compute processes */ + mutex_init(&adev->enforce_isolation_mutex); amdgpu_device_init_apu_flags(adev); -- cgit From afefd6f245024684fff75100052065d6a9e8f75f Mon Sep 17 00:00:00 2001 From: Srinivasan Shanmugam Date: Thu, 6 Jun 2024 13:28:02 +0530 Subject: drm/amdgpu: Implement Enforce Isolation Handler for KGD/KFD serialization MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This commit introduces the Enforce Isolation Handler designed to enforce shader isolation on AMD GPUs, which helps to prevent data leakage between different processes. The handler counts the number of emitted fences for each GFX and compute ring. If there are any fences, it schedules the `enforce_isolation_work` to be run after a delay of `GFX_SLICE_PERIOD`. If there are no fences, it signals the Kernel Fusion Driver (KFD) to resume the runqueue. The function is synchronized using the `enforce_isolation_mutex`. This commit also introduces a reference count mechanism (kfd_sch_req_count) to keep track of the number of requests to enable the KFD scheduler. When a request to enable the KFD scheduler is made, the reference count is decremented. When the reference count reaches zero, a delayed work is scheduled to enforce isolation after a delay of GFX_SLICE_PERIOD. When a request to disable the KFD scheduler is made, the function first checks if the reference count is zero. If it is, it cancels the delayed work for enforcing isolation and checks if the KFD scheduler is active. If the KFD scheduler is active, it sends a request to stop the KFD scheduler and sets the KFD scheduler state to inactive. Then, it increments the reference count. The function is synchronized using the kfd_sch_mutex to ensure that the KFD scheduler state and reference count are updated atomically. Cc: Christian König Cc: Alex Deucher Signed-off-by: Alex Deucher Signed-off-by: Srinivasan Shanmugam Suggested-by: Christian König Suggested-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 2f1bc02309fe..ad97f03f1358 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4067,6 +4067,7 @@ int amdgpu_device_init(struct amdgpu_device *adev, mutex_init(&adev->gfx.reset_sem_mutex); /* Initialize the mutex for cleaner shader isolation between GFX and compute processes */ mutex_init(&adev->enforce_isolation_mutex); + mutex_init(&adev->gfx.kfd_sch_mutex); amdgpu_device_init_apu_flags(adev); @@ -4098,6 +4099,21 @@ int amdgpu_device_init(struct amdgpu_device *adev, amdgpu_device_delayed_init_work_handler); INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work, amdgpu_device_delay_enable_gfx_off); + /* + * Initialize the enforce_isolation work structures for each XCP + * partition. This work handler is responsible for enforcing shader + * isolation on AMD GPUs. It counts the number of emitted fences for + * each GFX and compute ring. If there are any fences, it schedules + * the `enforce_isolation_work` to be run after a delay. If there are + * no fences, it signals the Kernel Fusion Driver (KFD) to resume the + * runqueue. + */ + for (i = 0; i < MAX_XCP; i++) { + INIT_DELAYED_WORK(&adev->gfx.enforce_isolation[i].work, + amdgpu_gfx_enforce_isolation_handler); + adev->gfx.enforce_isolation[i].adev = adev; + adev->gfx.enforce_isolation[i].xcp_id = i; + } INIT_WORK(&adev->xgmi_reset_work, amdgpu_device_xgmi_reset_func); -- cgit From 75f0efbc4b3b088cca20864d055b3854a51b5af0 Mon Sep 17 00:00:00 2001 From: Rahul Jain Date: Tue, 13 Aug 2024 13:41:11 +0530 Subject: drm/amdgpu: Take IOMMU remapping into account for p2p checks when trying to enable p2p the amdgpu_device_is_peer_accessible() checks the condition where address_mask overlaps the aper_base and hence returns 0, due to which the p2p disables for this platform IOMMU should remap the BAR addresses so the device can access them. Hence check if peer_adev is remapping DMA v5: (Felix, Alex) - fixing comment as per Alex feedback - refactor code as per Felix v4: (Alex) - fix the comment and description v3: - remove iommu_remap variable v2: (Alex) - Fix as per review comments - add new function amdgpu_device_check_iommu_remap to check if iommu remap Signed-off-by: Rahul Jain Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 43 ++++++++++++++++++++++++------ 1 file changed, 35 insertions(+), 8 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index ad97f03f1358..da06705f0026 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -3957,6 +3957,27 @@ static void amdgpu_device_check_iommu_direct_map(struct amdgpu_device *adev) adev->ram_is_direct_mapped = true; } +#if defined(CONFIG_HSA_AMD_P2P) +/** + * amdgpu_device_check_iommu_remap - Check if DMA remapping is enabled. + * + * @adev: amdgpu_device pointer + * + * return if IOMMU remapping bar address + */ +static bool amdgpu_device_check_iommu_remap(struct amdgpu_device *adev) +{ + struct iommu_domain *domain; + + domain = iommu_get_domain_for_dev(adev->dev); + if (domain && (domain->type == IOMMU_DOMAIN_DMA || + domain->type == IOMMU_DOMAIN_DMA_FQ)) + return true; + + return false; +} +#endif + static const struct attribute *amdgpu_dev_attributes[] = { &dev_attr_pcie_replay_count.attr, NULL @@ -6151,18 +6172,24 @@ bool amdgpu_device_is_peer_accessible(struct amdgpu_device *adev, struct amdgpu_device *peer_adev) { #ifdef CONFIG_HSA_AMD_P2P - uint64_t address_mask = peer_adev->dev->dma_mask ? - ~*peer_adev->dev->dma_mask : ~((1ULL << 32) - 1); - resource_size_t aper_limit = - adev->gmc.aper_base + adev->gmc.aper_size - 1; bool p2p_access = !adev->gmc.xgmi.connected_to_cpu && !(pci_p2pdma_distance(adev->pdev, peer_adev->dev, false) < 0); - return pcie_p2p && p2p_access && (adev->gmc.visible_vram_size && - adev->gmc.real_vram_size == adev->gmc.visible_vram_size && - !(adev->gmc.aper_base & address_mask || - aper_limit & address_mask)); + bool is_large_bar = adev->gmc.visible_vram_size && + adev->gmc.real_vram_size == adev->gmc.visible_vram_size; + bool p2p_addressable = amdgpu_device_check_iommu_remap(peer_adev); + + if (!p2p_addressable) { + uint64_t address_mask = peer_adev->dev->dma_mask ? + ~*peer_adev->dev->dma_mask : ~((1ULL << 32) - 1); + resource_size_t aper_limit = + adev->gmc.aper_base + adev->gmc.aper_size - 1; + + p2p_addressable = !(adev->gmc.aper_base & address_mask || + aper_limit & address_mask); + } + return is_large_bar && p2p_access && p2p_addressable; #else return false; #endif -- cgit From 6122f5c72e38a88eda13c7168e2ebbd3bd80b681 Mon Sep 17 00:00:00 2001 From: Trigger Huang Date: Mon, 19 Aug 2024 15:53:22 +0800 Subject: drm/amdgpu: skip printing vram_lost if needed The vm lost status can only be obtained after a GPU reset occurs, but sometimes a dev core dump can be happened before GPU reset. So a new argument is added to tell the dev core dump implementation whether to skip printing the vram_lost status in the dump. And this patch is also trying to decouple the core dump function from the GPU reset function, by replacing the argument amdgpu_reset_context with amdgpu_job to specify the context for core dump. V2: Inform user if VRAM lost check is skipped so users don't assume VRAM wasn't lost (Alex) Signed-off-by: Trigger Huang Suggested-by: Alex Deucher Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 49ef22dcf7fb..45edf99ae7ec 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5489,7 +5489,7 @@ int amdgpu_do_asic_reset(struct list_head *device_list_handle, vram_lost = amdgpu_device_check_vram_lost(tmp_adev); if (!test_bit(AMDGPU_SKIP_COREDUMP, &reset_context->flags)) - amdgpu_coredump(tmp_adev, vram_lost, reset_context); + amdgpu_coredump(tmp_adev, false, vram_lost, reset_context->job); if (vram_lost) { DRM_INFO("VRAM is lost due to GPU reset!\n"); -- cgit From af76ca8e180f38a7d874c18cf810707762766627 Mon Sep 17 00:00:00 2001 From: Victor Zhao Date: Mon, 26 Aug 2024 00:14:26 +0800 Subject: drm/amd/amdgpu: move drain_workqueue before shutdown is set [background] when unloading amdgpu driver right after running a workload, drain_workqueue is causing "Fence fallback timer expired on ring sdma0.0". Under sriov, this issue will cause sriov full access timeout and a reset happening. move drain_workqueue before shutdown is set to allow ih process and before enter full access under sriov to avoid full access time cost. Signed-off-by: Victor Zhao Reviewed-by: Feifei Xu Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 45edf99ae7ec..f4628412dac4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4531,6 +4531,9 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev) { dev_info(adev->dev, "amdgpu: finishing device.\n"); flush_delayed_work(&adev->delayed_init_work); + + if (adev->mman.initialized) + drain_workqueue(adev->mman.bdev.wq); adev->shutdown = true; /* make sure IB test finished before entering exclusive mode @@ -4551,9 +4554,6 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev) } amdgpu_fence_driver_hw_fini(adev); - if (adev->mman.initialized) - drain_workqueue(adev->mman.bdev.wq); - if (adev->pm.sysfs_initialized) amdgpu_pm_sysfs_fini(adev); if (adev->ucode_sysfs_en) -- cgit From 28b0ef922738b74335c20b8ed4bf8e259353a3a3 Mon Sep 17 00:00:00 2001 From: Bob Zhou Date: Fri, 6 Sep 2024 17:48:20 +0800 Subject: drm/amdgpu: Fix missing check pcie_p2p module param The module param pcie_p2p should be checked for kfd p2p feature, so add it. Fixes: 75f0efbc4b3b ("drm/amdgpu: Take IOMMU remapping into account for p2p checks") Signed-off-by: Bob Zhou Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index f4628412dac4..afb643e03d3c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -6189,7 +6189,7 @@ bool amdgpu_device_is_peer_accessible(struct amdgpu_device *adev, p2p_addressable = !(adev->gmc.aper_base & address_mask || aper_limit & address_mask); } - return is_large_bar && p2p_access && p2p_addressable; + return pcie_p2p && is_large_bar && p2p_access && p2p_addressable; #else return false; #endif -- cgit From 7181faaa4703705939580abffaf9cb5d6b50dbb7 Mon Sep 17 00:00:00 2001 From: Christian König Date: Tue, 27 Aug 2024 16:12:11 +0200 Subject: drm/amdgpu: nuke the VM PD/PT shadow handling MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This was only used as workaround for recovering the page tables after VRAM was lost and is no longer necessary after the function amdgpu_vm_bo_reset_state_machine() started to do the same. Compute never used shadows either, so the only proplematic case left is SVM and that is most likely not recoverable in any way when VRAM is lost. Signed-off-by: Christian König Acked-by: Lijo Lazar Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 87 +----------------------------- 1 file changed, 2 insertions(+), 85 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_device.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index afb643e03d3c..c2394c8b4d6b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4107,9 +4107,6 @@ int amdgpu_device_init(struct amdgpu_device *adev, spin_lock_init(&adev->mm_stats.lock); spin_lock_init(&adev->wb.lock); - INIT_LIST_HEAD(&adev->shadow_list); - mutex_init(&adev->shadow_list_lock); - INIT_LIST_HEAD(&adev->reset_list); INIT_LIST_HEAD(&adev->ras_list); @@ -5029,80 +5026,6 @@ static int amdgpu_device_ip_post_soft_reset(struct amdgpu_device *adev) return 0; } -/** - * amdgpu_device_recover_vram - Recover some VRAM contents - * - * @adev: amdgpu_device pointer - * - * Restores the contents of VRAM buffers from the shadows in GTT. Used to - * restore things like GPUVM page tables after a GPU reset where - * the contents of VRAM might be lost. - * - * Returns: - * 0 on success, negative error code on failure. - */ -static int amdgpu_device_recover_vram(struct amdgpu_device *adev) -{ - struct dma_fence *fence = NULL, *next = NULL; - struct amdgpu_bo *shadow; - struct amdgpu_bo_vm *vmbo; - long r = 1, tmo; - - if (amdgpu_sriov_runtime(adev)) - tmo = msecs_to_jiffies(8000); - else - tmo = msecs_to_jiffies(100); - - dev_info(adev->dev, "recover vram bo from shadow start\n"); - mutex_lock(&adev->shadow_list_lock); - list_for_each_entry(vmbo, &adev->shadow_list, shadow_list) { - /* If vm is compute context or adev is APU, shadow will be NULL */ - if (!vmbo->shadow) - continue; - shadow = vmbo->shadow; - - /* No need to recover an evicted BO */ - if (!shadow->tbo.resource || - shadow->tbo.resource->mem_type != TTM_PL_TT || - shadow->tbo.resource->start == AMDGPU_BO_INVALID_OFFSET || - shadow->parent->tbo.resource->mem_type != TTM_PL_VRAM) - continue; - - r = amdgpu_bo_restore_shadow(shadow, &next); - if (r) - break; - - if (fence) { - tmo = dma_fence_wait_timeout(fence, false, tmo); - dma_fence_put(fence); - fence = next; - if (tmo == 0) { - r = -ETIMEDOUT; - break; - } else if (tmo < 0) { - r = tmo; - break; - } - } else { - fence = next; - } - } - mutex_unlock(&adev->shadow_list_lock); - - if (fence) - tmo = dma_fence_wait_timeout(fence, false, tmo); - dma_fence_put(fence); - - if (r < 0 || tmo <= 0) { - dev_err(adev->dev, "recover vram bo from shadow failed, r is %ld, tmo is %ld\n", r, tmo); - return -EIO; - } - - dev_info(adev->dev, "recover vram bo from shadow done\n"); - return 0; -} - - /** * amdgpu_device_reset_sriov - reset ASIC for SR-IOV vf * @@ -5165,12 +5088,8 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device *adev, if (r) return r; - if (adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) { + if (adev->virt.gim_feature & AMDGIM_FEATURE_GIM_FLR_VRAMLOST) amdgpu_inc_vram_lost(adev); - r = amdgpu_device_recover_vram(adev); - } - if (r) - return r; /* need to be called during full access so we can't do it later like * bare-metal does. @@ -5569,9 +5488,7 @@ out: } } - if (!r) - r = amdgpu_device_recover_vram(tmp_adev); - else + if (r) tmp_adev->asic_reset_res = r; } -- cgit