From 25263da37693c764e198b31ab4c647934bf80e07 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Tue, 19 Jul 2022 12:13:08 -0400 Subject: drm/amdgpu: rework SR-IOV virtual display handling MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit virtual display is enabled unconditionally in SR-IOV, but without specifying the virtual_display module, the number of crtcs defaults to 0. Set a single display by default for SR-IOV if the virtual_display parameter is not set. Only enable virtual display by default on SR-IOV on asics which actually have display hardware. Acked-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 13 +++++++++++-- 1 file changed, 11 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 5b9f992e4607..72d803060730 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -1915,6 +1915,16 @@ static void amdgpu_device_enable_virtual_display(struct amdgpu_device *adev) } } +void amdgpu_device_set_sriov_virtual_display(struct amdgpu_device *adev) +{ + if (amdgpu_sriov_vf(adev) && !adev->enable_virtual_display) { + adev->mode_info.num_crtc = 1; + adev->enable_virtual_display = true; + DRM_INFO("virtual_display:%d, num_crtc:%d\n", + adev->enable_virtual_display, adev->mode_info.num_crtc); + } +} + /** * amdgpu_device_parse_gpu_info_fw - parse gpu info firmware * @@ -3347,8 +3357,7 @@ bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type) */ bool amdgpu_device_has_dc_support(struct amdgpu_device *adev) { - if (amdgpu_sriov_vf(adev) || - adev->enable_virtual_display || + if (adev->enable_virtual_display || (adev->harvest_ip_mask & AMD_HARVEST_IP_DMU_MASK)) return false; -- cgit From d09ef243035b75a6d403ebfeb7e87fa20d7e25c6 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Tue, 19 Jul 2022 14:18:41 -0400 Subject: drm/amdgpu: clarify DC checks MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit There are several places where we don't want to check if a particular asic could support DC, but rather, if DC is enabled. Set a flag if DC is enabled and check for that rather than if a device supports DC or not. Acked-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 32 ++++++++++++++++-------------- 1 file changed, 17 insertions(+), 15 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 72d803060730..ee69cc3a5efd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4224,25 +4224,27 @@ int amdgpu_device_resume(struct drm_device *dev, bool fbcon) amdgpu_ras_resume(adev); - /* - * Most of the connector probing functions try to acquire runtime pm - * refs to ensure that the GPU is powered on when connector polling is - * performed. Since we're calling this from a runtime PM callback, - * trying to acquire rpm refs will cause us to deadlock. - * - * Since we're guaranteed to be holding the rpm lock, it's safe to - * temporarily disable the rpm helpers so this doesn't deadlock us. - */ + if (adev->mode_info.num_crtc) { + /* + * Most of the connector probing functions try to acquire runtime pm + * refs to ensure that the GPU is powered on when connector polling is + * performed. Since we're calling this from a runtime PM callback, + * trying to acquire rpm refs will cause us to deadlock. + * + * Since we're guaranteed to be holding the rpm lock, it's safe to + * temporarily disable the rpm helpers so this doesn't deadlock us. + */ #ifdef CONFIG_PM - dev->dev->power.disable_depth++; + dev->dev->power.disable_depth++; #endif - if (!amdgpu_device_has_dc_support(adev)) - drm_helper_hpd_irq_event(dev); - else - drm_kms_helper_hotplug_event(dev); + if (!adev->dc_enabled) + drm_helper_hpd_irq_event(dev); + else + drm_kms_helper_hotplug_event(dev); #ifdef CONFIG_PM - dev->dev->power.disable_depth--; + dev->dev->power.disable_depth--; #endif + } adev->in_suspend = false; if (amdgpu_acpi_smart_shift_update(dev, AMDGPU_SS_DEV_D0)) -- cgit From 220c8cc855ce7075a29926533279a8dfefdd8788 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Fri, 11 Nov 2022 12:50:38 -0500 Subject: drm/amdgpu: there is no vbios fb on devices with no display hw (v2) If we enable virtual display functionality on parts with no display hardware we can end up trying to check for and reserve the vbios FB area on devices where it doesn't exist. Check if display hardware is actually present on the hardware before trying to reserve the memory. v2: move the check into common code Acked-by: Evan Quan Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 41 ++++++++++++++++++++++++++++++ 1 file changed, 41 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 ee69cc3a5efd..31357101c704 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -6057,3 +6057,44 @@ struct dma_fence *amdgpu_device_switch_gang(struct amdgpu_device *adev, dma_fence_put(old); return NULL; } + +bool amdgpu_device_has_display_hardware(struct amdgpu_device *adev) +{ + switch (adev->asic_type) { +#ifdef CONFIG_DRM_AMDGPU_SI + case CHIP_HAINAN: +#endif + case CHIP_TOPAZ: + /* chips with no display hardware */ + return false; +#ifdef CONFIG_DRM_AMDGPU_SI + case CHIP_TAHITI: + case CHIP_PITCAIRN: + case CHIP_VERDE: + case CHIP_OLAND: +#endif +#ifdef CONFIG_DRM_AMDGPU_CIK + case CHIP_BONAIRE: + case CHIP_HAWAII: + case CHIP_KAVERI: + case CHIP_KABINI: + case CHIP_MULLINS: +#endif + case CHIP_TONGA: + case CHIP_FIJI: + case CHIP_POLARIS10: + case CHIP_POLARIS11: + case CHIP_POLARIS12: + case CHIP_VEGAM: + case CHIP_CARRIZO: + case CHIP_STONEY: + /* chips with display hardware */ + return true; + default: + /* IP discovery */ + if (!adev->ip_versions[DCE_HWIP][0] || + (adev->harvest_ip_mask & AMD_HARVEST_IP_DMU_MASK)) + return false; + return true; + } +} -- cgit From d293470e10d02ce1afb59cfe43fdf1a6cdb3d726 Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Fri, 4 Nov 2022 17:50:12 +0800 Subject: drm/amdgpu: Fixed the problem that ras error can't be queried after gpu recovery is completed Amdgpu_ras_set_error_query_ready is called at the start of amdgpu_device_gpu_recover to disable query ras error, but the code behind only enables query ras error in full reset path, but not in soft reset path, emergency restart path and skip the hardware reset path. Signed-off-by: YiPeng Chai Reviewed-by: Hawking Zhang 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 31357101c704..947ed5d97942 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5454,6 +5454,8 @@ skip_sched_resume: amdgpu_device_resume_display_audio(tmp_adev); amdgpu_device_unset_mp1_state(tmp_adev); + + amdgpu_ras_set_error_query_ready(tmp_adev, true); } recover_end: -- cgit From 06a2d7cc3f0476be4682ef90eb09a28fa3daed37 Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 26 Oct 2022 12:26:37 +0200 Subject: drm/amdgpu: revert "implement tdr advanced mode" MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This reverts commit e6c6338f393b74ac0b303d567bb918b44ae7ad75. This feature basically re-submits one job after another to figure out which one was the one causing a hang. This is obviously incompatible with gang-submit which requires that multiple jobs run at the same time. It's also absolutely not helpful to crash the hardware multiple times if a clean recovery is desired. For testing and debugging environments we should rather disable recovery alltogether to be able to inspect the state with a hw debugger. Additional to that the sw implementation is clearly buggy and causes reference count issues for the hardware fence. Signed-off-by: Christian König Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 103 ----------------------------- 1 file changed, 103 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 947ed5d97942..2a8166fb69f2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5088,94 +5088,6 @@ static int amdgpu_device_suspend_display_audio(struct amdgpu_device *adev) return 0; } -static void amdgpu_device_recheck_guilty_jobs( - struct amdgpu_device *adev, struct list_head *device_list_handle, - struct amdgpu_reset_context *reset_context) -{ - int i, r = 0; - - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - struct amdgpu_ring *ring = adev->rings[i]; - int ret = 0; - struct drm_sched_job *s_job; - - if (!ring || !ring->sched.thread) - continue; - - s_job = list_first_entry_or_null(&ring->sched.pending_list, - struct drm_sched_job, list); - if (s_job == NULL) - continue; - - /* clear job's guilty and depend the folowing step to decide the real one */ - drm_sched_reset_karma(s_job); - drm_sched_resubmit_jobs_ext(&ring->sched, 1); - - if (!s_job->s_fence->parent) { - DRM_WARN("Failed to get a HW fence for job!"); - continue; - } - - ret = dma_fence_wait_timeout(s_job->s_fence->parent, false, ring->sched.timeout); - if (ret == 0) { /* timeout */ - DRM_ERROR("Found the real bad job! ring:%s, job_id:%llx\n", - ring->sched.name, s_job->id); - - - amdgpu_fence_driver_isr_toggle(adev, true); - - /* Clear this failed job from fence array */ - amdgpu_fence_driver_clear_job_fences(ring); - - amdgpu_fence_driver_isr_toggle(adev, false); - - /* Since the job won't signal and we go for - * another resubmit drop this parent pointer - */ - dma_fence_put(s_job->s_fence->parent); - s_job->s_fence->parent = NULL; - - /* set guilty */ - drm_sched_increase_karma(s_job); - amdgpu_reset_prepare_hwcontext(adev, reset_context); -retry: - /* do hw reset */ - if (amdgpu_sriov_vf(adev)) { - amdgpu_virt_fini_data_exchange(adev); - r = amdgpu_device_reset_sriov(adev, false); - if (r) - adev->asic_reset_res = r; - } else { - clear_bit(AMDGPU_SKIP_HW_RESET, - &reset_context->flags); - r = amdgpu_do_asic_reset(device_list_handle, - reset_context); - if (r && r == -EAGAIN) - goto retry; - } - - /* - * add reset counter so that the following - * resubmitted job could flush vmid - */ - atomic_inc(&adev->gpu_reset_counter); - continue; - } - - /* got the hw fence, signal finished fence */ - atomic_dec(ring->sched.score); - dma_fence_get(&s_job->s_fence->finished); - dma_fence_signal(&s_job->s_fence->finished); - dma_fence_put(&s_job->s_fence->finished); - - /* remove node from list and free the job */ - spin_lock(&ring->sched.job_list_lock); - list_del_init(&s_job->list); - spin_unlock(&ring->sched.job_list_lock); - ring->sched.ops->free_job(s_job); - } -} - static inline void amdgpu_device_stop_pending_resets(struct amdgpu_device *adev) { struct amdgpu_ras *con = amdgpu_ras_get_context(adev); @@ -5196,7 +5108,6 @@ static inline void amdgpu_device_stop_pending_resets(struct amdgpu_device *adev) } - /** * amdgpu_device_gpu_recover - reset the asic and recover scheduler * @@ -5219,7 +5130,6 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, int i, r = 0; bool need_emergency_restart = false; bool audio_suspended = false; - int tmp_vram_lost_counter; bool gpu_reset_for_dev_remove = false; gpu_reset_for_dev_remove = @@ -5365,7 +5275,6 @@ retry: /* Rest of adevs pre asic reset from XGMI hive. */ amdgpu_device_stop_pending_resets(tmp_adev); } - tmp_vram_lost_counter = atomic_read(&((adev)->vram_lost_counter)); /* Actual ASIC resets if needed.*/ /* Host driver will handle XGMI hive reset for SRIOV */ if (amdgpu_sriov_vf(adev)) { @@ -5390,18 +5299,6 @@ skip_hw_reset: /* Post ASIC reset for all devs .*/ list_for_each_entry(tmp_adev, device_list_handle, reset_list) { - /* - * Sometimes a later bad compute job can block a good gfx job as gfx - * and compute ring share internal GC HW mutually. We add an additional - * guilty jobs recheck step to find the real guilty job, it synchronously - * submits and pends for the first job being signaled. If it gets timeout, - * we identify it as a real guilty job. - */ - if (amdgpu_gpu_recovery == 2 && - !(tmp_vram_lost_counter < atomic_read(&adev->vram_lost_counter))) - amdgpu_device_recheck_guilty_jobs( - tmp_adev, device_list_handle, reset_context); - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { struct amdgpu_ring *ring = tmp_adev->rings[i]; -- cgit From 6868a2c46560670efc0d1f2b446cc57edcaf960d Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 26 Oct 2022 12:57:44 +0200 Subject: drm/amdgpu: stop resubmitting jobs for GPU reset v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Re-submitting IBs by the kernel has many problems because pre- requisite state is not automatically re-created as well. In other words neither binary semaphores nor things like ring buffer pointers are in the state they should be when the hardware starts to work on the IBs again. Additional to that even after more than 5 years of developing this feature it is still not stable and we have massively problems getting the reference counts right. As discussed with user space developers this behavior is not helpful in the first place. For graphics and multimedia workloads it makes much more sense to either completely re-create the context or at least re-submitting the IBs from userspace. For compute use cases re-submitting is also not very helpful since userspace must rely on the accuracy of the result. Because of this we stop this practice and instead just properly note that the fence submission was canceled. The only use case we keep the re-submission for now is SRIOV and function level resets. v2: as suggested by Sshaoyun stop resubmitting jobs even for SRIOV Signed-off-by: Christian König Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 6 +----- 1 file changed, 1 insertion(+), 5 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 2a8166fb69f2..62b508053eb1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5305,11 +5305,7 @@ skip_hw_reset: if (!ring || !ring->sched.thread) continue; - /* No point to resubmit jobs if we didn't HW reset*/ - if (!tmp_adev->asic_reset_res && !job_signaled) - drm_sched_resubmit_jobs(&ring->sched); - - drm_sched_start(&ring->sched, !tmp_adev->asic_reset_res); + drm_sched_start(&ring->sched, true); } if (adev->enable_mes && adev->ip_versions[GC_HWIP][0] != IP_VERSION(11, 0, 3)) -- cgit From 0788a47e7cec7ebdcb1ad8912754b8b8b06ee915 Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 26 Oct 2022 12:48:24 +0200 Subject: drm/amdgpu: stop resubmittting jobs in amdgpu_pci_resume MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The state of VRAM is unreliable due to a PCI event like AER, link reset or DPC. Signed-off-by: Christian König Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 -- 1 file changed, 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 62b508053eb1..0f5cbc70682a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5760,8 +5760,6 @@ void amdgpu_pci_resume(struct pci_dev *pdev) if (!ring || !ring->sched.thread) continue; - - drm_sched_resubmit_jobs(&ring->sched); drm_sched_start(&ring->sched, true); } -- cgit From 1a11a65d5395ccdcd07f19a75da82a3d74c368dd Mon Sep 17 00:00:00 2001 From: YiPeng Chai Date: Tue, 8 Nov 2022 17:11:18 +0800 Subject: drm/amdgpu: Enable mode-1 reset for RAS recovery in fatal error mode The patch is enabling mode-1 reset for RAS recovery in fatal error mode. Signed-off-by: YiPeng Chai Reviewed-by: Hawking Zhang Reviewed-by: Tao Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ++++ 1 file changed, 4 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 0f5cbc70682a..3d5d5d49cfab 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4593,6 +4593,10 @@ bool amdgpu_device_should_recover_gpu(struct amdgpu_device *adev) if (amdgpu_gpu_recovery == 0) goto disabled; + /* Skip soft reset check in fatal error mode */ + if (!amdgpu_ras_is_poison_mode_supported(adev)) + return true; + if (!amdgpu_device_ip_check_soft_reset(adev)) { dev_info(adev->dev,"Timeout, but no hardware hang detected.\n"); return false; -- cgit