From ed098aa34c51d81a16de93f59c64e36a136fd201 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Thu, 18 Mar 2021 16:44:10 -0400 Subject: drm/amdgpu: Add additional Sienna Cichlid PCI ID Add new DID. Reviewed-by: Guchun Chen Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 1 + 1 file changed, 1 insertion(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 33991b4a5627..fcda7189d4ab 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -1161,6 +1161,7 @@ static const struct pci_device_id pciidlist[] = { {0x1002, 0x73A3, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID}, {0x1002, 0x73AB, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID}, {0x1002, 0x73AE, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID}, + {0x1002, 0x73AF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID}, {0x1002, 0x73BF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_SIENNA_CICHLID}, /* Van Gogh */ -- cgit v1.2.3 From e25443d2765f40a9b3d0056dc4d560a007dd850c Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Wed, 10 Mar 2021 00:41:49 -0500 Subject: drm/amdgpu: add a dev_pm_ops prepare callback (v2) as per: https://www.kernel.org/doc/html/latest/driver-api/pm/devices.html The prepare callback is required to support the DPM_FLAG_SMART_SUSPEND driver flag. This allows runtime pm to auto complete when the system goes into suspend avoiding a wake up on suspend and on resume. Apply this for hybrid gfx and BOCO systems where d3cold is provided by the ACPI platform. v2: check if device is runtime suspended in prepare. Reviewed-by: Evan Quan Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index fcda7189d4ab..1508da68cfbb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -36,6 +36,7 @@ #include #include #include +#include #include "amdgpu.h" #include "amdgpu_irq.h" @@ -1403,6 +1404,27 @@ static void amdgpu_drv_delayed_reset_work_handler(struct work_struct *work) return; } +static int amdgpu_pmops_prepare(struct device *dev) +{ + struct drm_device *drm_dev = dev_get_drvdata(dev); + + /* Return a positive number here so + * DPM_FLAG_SMART_SUSPEND works properly + */ + if ((amdgpu_device_supports_atpx(drm_dev) && + amdgpu_is_atpx_hybrid()) || + amdgpu_device_supports_boco(drm_dev)) + return pm_runtime_suspended(dev) && + pm_suspend_via_firmware(); + + return 0; +} + +static void amdgpu_pmops_complete(struct device *dev) +{ + /* nothing to do */ +} + static int amdgpu_pmops_suspend(struct device *dev) { struct drm_device *drm_dev = dev_get_drvdata(dev); @@ -1621,6 +1643,8 @@ out: } static const struct dev_pm_ops amdgpu_pm_ops = { + .prepare = amdgpu_pmops_prepare, + .complete = amdgpu_pmops_complete, .suspend = amdgpu_pmops_suspend, .resume = amdgpu_pmops_resume, .freeze = amdgpu_pmops_freeze, -- cgit v1.2.3 From b98c6299ef992660f5ca4392287a11ea2439c664 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Wed, 10 Mar 2021 00:43:35 -0500 Subject: drm/amdgpu: disentangle HG systems from vgaswitcheroo There's no need to keep vgaswitcheroo around for HG systems. They don't use muxes and their power control is handled via ACPI. Reviewed-by: Evan Quan Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 3 ++- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 38 +++++++++++++----------------- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 34 +++++++++++--------------- drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 9 +++---- 4 files changed, 35 insertions(+), 49 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 4c2fabe93d59..7da8d26f96b8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1275,8 +1275,9 @@ void amdgpu_device_program_register_sequence(struct amdgpu_device *adev, const u32 *registers, const u32 array_size); -bool amdgpu_device_supports_atpx(struct drm_device *dev); int amdgpu_device_mode1_reset(struct amdgpu_device *adev); +bool amdgpu_device_supports_atpx(struct drm_device *dev); +bool amdgpu_device_supports_px(struct drm_device *dev); bool amdgpu_device_supports_boco(struct drm_device *dev); bool amdgpu_device_supports_baco(struct drm_device *dev); bool amdgpu_device_is_peer_accessible(struct amdgpu_device *adev, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 7db60edfb5d1..1c3044f2e767 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -212,18 +212,18 @@ static DEVICE_ATTR(serial_number, S_IRUGO, amdgpu_device_get_serial_number, NULL); /** - * amdgpu_device_supports_atpx - Is the device a dGPU with HG/PX power control + * amdgpu_device_supports_px - Is the device a dGPU with ATPX power control * * @dev: drm_device pointer * - * Returns true if the device is a dGPU with HG/PX power control, + * Returns true if the device is a dGPU with ATPX power control, * otherwise return false. */ -bool amdgpu_device_supports_atpx(struct drm_device *dev) +bool amdgpu_device_supports_px(struct drm_device *dev) { struct amdgpu_device *adev = drm_to_adev(dev); - if (adev->flags & AMD_IS_PX) + if ((adev->flags & AMD_IS_PX) && !amdgpu_is_atpx_hybrid()) return true; return false; } @@ -233,14 +233,15 @@ bool amdgpu_device_supports_atpx(struct drm_device *dev) * * @dev: drm_device pointer * - * Returns true if the device is a dGPU with HG/PX power control, + * Returns true if the device is a dGPU with ACPI power control, * otherwise return false. */ bool amdgpu_device_supports_boco(struct drm_device *dev) { struct amdgpu_device *adev = drm_to_adev(dev); - if (adev->has_pr3) + if (adev->has_pr3 || + ((adev->flags & AMD_IS_PX) && amdgpu_is_atpx_hybrid())) return true; return false; } @@ -1420,7 +1421,7 @@ static void amdgpu_switcheroo_set_state(struct pci_dev *pdev, struct drm_device *dev = pci_get_drvdata(pdev); int r; - if (amdgpu_device_supports_atpx(dev) && state == VGA_SWITCHEROO_OFF) + if (amdgpu_device_supports_px(dev) && state == VGA_SWITCHEROO_OFF) return; if (state == VGA_SWITCHEROO_ON) { @@ -3226,7 +3227,7 @@ int amdgpu_device_init(struct amdgpu_device *adev, struct drm_device *ddev = adev_to_drm(adev); struct pci_dev *pdev = adev->pdev; int r, i; - bool atpx = false; + bool px = false; u32 max_MBps; adev->shutdown = false; @@ -3388,16 +3389,12 @@ int amdgpu_device_init(struct amdgpu_device *adev, if ((adev->pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA) vga_client_register(adev->pdev, adev, NULL, amdgpu_device_vga_set_decode); - if (amdgpu_device_supports_atpx(ddev)) - atpx = true; - if (amdgpu_has_atpx() && - (amdgpu_is_atpx_hybrid() || - amdgpu_has_atpx_dgpu_power_cntl()) && - !pci_is_thunderbolt_attached(adev->pdev)) + if (amdgpu_device_supports_px(ddev)) { + px = true; vga_switcheroo_register_client(adev->pdev, - &amdgpu_switcheroo_ops, atpx); - if (atpx) + &amdgpu_switcheroo_ops, px); vga_switcheroo_init_domain_pm_ops(adev->dev, &adev->vga_pm_domain); + } if (amdgpu_emu_mode == 1) { /* post the asic on emulation mode */ @@ -3604,7 +3601,7 @@ release_ras_con: failed: amdgpu_vf_error_trans_all(adev); - if (atpx) + if (px) vga_switcheroo_fini_domain_pm_ops(adev->dev); failed_unmap: @@ -3664,13 +3661,10 @@ void amdgpu_device_fini(struct amdgpu_device *adev) kfree(adev->bios); adev->bios = NULL; - if (amdgpu_has_atpx() && - (amdgpu_is_atpx_hybrid() || - amdgpu_has_atpx_dgpu_power_cntl()) && - !pci_is_thunderbolt_attached(adev->pdev)) + if (amdgpu_device_supports_px(adev_to_drm(adev))) { vga_switcheroo_unregister_client(adev->pdev); - if (amdgpu_device_supports_atpx(adev_to_drm(adev))) vga_switcheroo_fini_domain_pm_ops(adev->dev); + } if ((adev->pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA) vga_client_register(adev->pdev, NULL, NULL, NULL); iounmap(adev->rmmio); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 1508da68cfbb..daaacf9067b8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -1411,9 +1411,7 @@ static int amdgpu_pmops_prepare(struct device *dev) /* Return a positive number here so * DPM_FLAG_SMART_SUSPEND works properly */ - if ((amdgpu_device_supports_atpx(drm_dev) && - amdgpu_is_atpx_hybrid()) || - amdgpu_device_supports_boco(drm_dev)) + if (amdgpu_device_supports_boco(drm_dev)) return pm_runtime_suspended(dev) && pm_suspend_via_firmware(); @@ -1502,7 +1500,7 @@ static int amdgpu_pmops_runtime_suspend(struct device *dev) } adev->in_runpm = true; - if (amdgpu_device_supports_atpx(drm_dev)) + if (amdgpu_device_supports_px(drm_dev)) drm_dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; ret = amdgpu_device_suspend(drm_dev, false); @@ -1511,16 +1509,14 @@ static int amdgpu_pmops_runtime_suspend(struct device *dev) return ret; } - if (amdgpu_device_supports_atpx(drm_dev)) { + if (amdgpu_device_supports_px(drm_dev)) { /* Only need to handle PCI state in the driver for ATPX * PCI core handles it for _PR3. */ - if (!amdgpu_is_atpx_hybrid()) { - amdgpu_device_cache_pci_state(pdev); - pci_disable_device(pdev); - pci_ignore_hotplug(pdev); - pci_set_power_state(pdev, PCI_D3cold); - } + amdgpu_device_cache_pci_state(pdev); + pci_disable_device(pdev); + pci_ignore_hotplug(pdev); + pci_set_power_state(pdev, PCI_D3cold); drm_dev->switch_power_state = DRM_SWITCH_POWER_DYNAMIC_OFF; } else if (amdgpu_device_supports_baco(drm_dev)) { amdgpu_device_baco_enter(drm_dev); @@ -1539,19 +1535,17 @@ static int amdgpu_pmops_runtime_resume(struct device *dev) if (!adev->runpm) return -EINVAL; - if (amdgpu_device_supports_atpx(drm_dev)) { + if (amdgpu_device_supports_px(drm_dev)) { drm_dev->switch_power_state = DRM_SWITCH_POWER_CHANGING; /* Only need to handle PCI state in the driver for ATPX * PCI core handles it for _PR3. */ - if (!amdgpu_is_atpx_hybrid()) { - pci_set_power_state(pdev, PCI_D0); - amdgpu_device_load_pci_state(pdev); - ret = pci_enable_device(pdev); - if (ret) - return ret; - } + pci_set_power_state(pdev, PCI_D0); + amdgpu_device_load_pci_state(pdev); + ret = pci_enable_device(pdev); + if (ret) + return ret; pci_set_master(pdev); } else if (amdgpu_device_supports_boco(drm_dev)) { /* Only need to handle PCI state in the driver for ATPX @@ -1562,7 +1556,7 @@ static int amdgpu_pmops_runtime_resume(struct device *dev) amdgpu_device_baco_exit(drm_dev); } ret = amdgpu_device_resume(drm_dev, false); - if (amdgpu_device_supports_atpx(drm_dev)) + if (amdgpu_device_supports_px(drm_dev)) drm_dev->switch_power_state = DRM_SWITCH_POWER_ON; adev->in_runpm = false; return 0; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c index 23c8f4cb0ac9..8844f650b17f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c @@ -159,7 +159,7 @@ int amdgpu_driver_load_kms(struct amdgpu_device *adev, unsigned long flags) goto out; } - if (amdgpu_device_supports_atpx(dev) && + if (amdgpu_device_supports_px(dev) && (amdgpu_runtime_pm != 0)) { /* enable runpm by default for atpx */ adev->runpm = true; dev_info(adev->dev, "Using ATPX for runtime pm\n"); @@ -200,13 +200,10 @@ int amdgpu_driver_load_kms(struct amdgpu_device *adev, unsigned long flags) if (adev->runpm) { /* only need to skip on ATPX */ - if (amdgpu_device_supports_atpx(dev) && - !amdgpu_is_atpx_hybrid()) + if (amdgpu_device_supports_px(dev)) dev_pm_set_driver_flags(dev->dev, DPM_FLAG_NO_DIRECT_COMPLETE); /* we want direct complete for BOCO */ - if ((amdgpu_device_supports_atpx(dev) && - amdgpu_is_atpx_hybrid()) || - amdgpu_device_supports_boco(dev)) + if (amdgpu_device_supports_boco(dev)) dev_pm_set_driver_flags(dev->dev, DPM_FLAG_SMART_PREPARE | DPM_FLAG_SMART_SUSPEND | DPM_FLAG_MAY_SKIP_RESUME); -- cgit v1.2.3 From 62498733d4c4fde8bc15215c5502923ff8224f86 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Fri, 12 Mar 2021 15:22:36 -0500 Subject: drm/amdgpu: rework S3/S4/S0ix state handling Set flags at the top level pmops callbacks to track state. This cleans up the current set of flags and properly handles S4 on S0ix capable systems. Reviewed-by: Evan Quan Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 10 +++------- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 12 +++--------- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 30 ++++++++++++++++++------------ drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c | 2 +- 4 files changed, 25 insertions(+), 29 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 7da8d26f96b8..3316618f809d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1030,13 +1030,9 @@ struct amdgpu_device { /* s3/s4 mask */ bool in_suspend; - bool in_hibernate; - - /* - * The combination flag in_poweroff_reboot_com used to identify the poweroff - * and reboot opt in the s0i3 system-wide suspend. - */ - bool in_poweroff_reboot_com; + bool in_s3; + bool in_s4; + bool in_s0ix; atomic_t in_gpu_reset; enum pp_mp1_state mp1_state; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 87ebf67086e0..2359449567be 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -2685,8 +2685,7 @@ static int amdgpu_device_ip_suspend_phase1(struct amdgpu_device *adev) { int i, r; - if (adev->in_poweroff_reboot_com || adev->in_hibernate || - !amdgpu_acpi_is_s0ix_supported(adev) || amdgpu_in_reset(adev)) { + if (!adev->in_s0ix || amdgpu_in_reset(adev)) { amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE); amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE); } @@ -3766,12 +3765,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool fbcon) amdgpu_fence_driver_suspend(adev); - /* - * TODO: Need figure out the each GNB IP idle off dependency and then - * improve the AMDGPU suspend/resume sequence for system-wide Sx entry/exit. - */ - if (adev->in_poweroff_reboot_com || adev->in_hibernate || - !amdgpu_acpi_is_s0ix_supported(adev) || amdgpu_in_reset(adev)) + if (!adev->in_s0ix || amdgpu_in_reset(adev)) r = amdgpu_device_ip_suspend_phase2(adev); else amdgpu_gfx_state_change_set(adev, sGpuChangeState_D3Entry); @@ -3805,7 +3799,7 @@ int amdgpu_device_resume(struct drm_device *dev, bool fbcon) if (dev->switch_power_state == DRM_SWITCH_POWER_OFF) return 0; - if (amdgpu_acpi_is_s0ix_supported(adev)) + if (adev->in_s0ix) amdgpu_gfx_state_change_set(adev, sGpuChangeState_D0Entry); /* post card */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index daaacf9067b8..02b75fa64d7d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -1335,9 +1335,7 @@ amdgpu_pci_shutdown(struct pci_dev *pdev) */ if (!amdgpu_passthrough(adev)) adev->mp1_state = PP_MP1_STATE_UNLOAD; - adev->in_poweroff_reboot_com = true; amdgpu_device_ip_suspend(adev); - adev->in_poweroff_reboot_com = false; adev->mp1_state = PP_MP1_STATE_NONE; } @@ -1426,15 +1424,28 @@ static void amdgpu_pmops_complete(struct device *dev) static int amdgpu_pmops_suspend(struct device *dev) { struct drm_device *drm_dev = dev_get_drvdata(dev); + struct amdgpu_device *adev = drm_to_adev(drm_dev); + int r; - return amdgpu_device_suspend(drm_dev, true); + if (amdgpu_acpi_is_s0ix_supported(adev)) + adev->in_s0ix = true; + adev->in_s3 = true; + r = amdgpu_device_suspend(drm_dev, true); + adev->in_s3 = false; + + return r; } static int amdgpu_pmops_resume(struct device *dev) { struct drm_device *drm_dev = dev_get_drvdata(dev); + struct amdgpu_device *adev = drm_to_adev(drm_dev); + int r; - return amdgpu_device_resume(drm_dev, true); + r = amdgpu_device_resume(drm_dev, true); + if (amdgpu_acpi_is_s0ix_supported(adev)) + adev->in_s0ix = false; + return r; } static int amdgpu_pmops_freeze(struct device *dev) @@ -1443,9 +1454,9 @@ static int amdgpu_pmops_freeze(struct device *dev) struct amdgpu_device *adev = drm_to_adev(drm_dev); int r; - adev->in_hibernate = true; + adev->in_s4 = true; r = amdgpu_device_suspend(drm_dev, true); - adev->in_hibernate = false; + adev->in_s4 = false; if (r) return r; return amdgpu_asic_reset(adev); @@ -1461,13 +1472,8 @@ static int amdgpu_pmops_thaw(struct device *dev) static int amdgpu_pmops_poweroff(struct device *dev) { struct drm_device *drm_dev = dev_get_drvdata(dev); - struct amdgpu_device *adev = drm_to_adev(drm_dev); - int r; - adev->in_poweroff_reboot_com = true; - r = amdgpu_device_suspend(drm_dev, true); - adev->in_poweroff_reboot_com = false; - return r; + return amdgpu_device_suspend(drm_dev, true); } static int amdgpu_pmops_restore(struct device *dev) diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c index e722adcf2f53..cfcac110ed84 100644 --- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c +++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c @@ -1337,7 +1337,7 @@ static int smu_disable_dpms(struct smu_context *smu) bool use_baco = !smu->is_apu && ((amdgpu_in_reset(adev) && (amdgpu_asic_reset_method(adev) == AMD_RESET_METHOD_BACO)) || - ((adev->in_runpm || adev->in_hibernate) && amdgpu_asic_supports_baco(adev))); + ((adev->in_runpm || adev->in_s4) && amdgpu_asic_supports_baco(adev))); /* * For custom pptable uploading, skip the DPM features -- cgit v1.2.3 From e6c6338f393b74ac0b303d567bb918b44ae7ad75 Mon Sep 17 00:00:00 2001 From: Jack Zhang Date: Mon, 8 Mar 2021 12:41:27 +0800 Subject: drm/amd/amdgpu implement tdr advanced mode [Why] Previous tdr design treats the first job in job_timeout as the bad job. But sometimes a later bad compute job can block a good gfx job and cause an unexpected gfx job timeout because gfx and compute ring share internal GC HW mutually. [How] This patch implements an advanced tdr mode.It involves an additinal synchronous pre-resubmit step(Step0 Resubmit) before normal resubmit step in order to find the real bad job. 1. At Step0 Resubmit stage, it synchronously submits and pends for the first job being signaled. If it gets timeout, we identify it as guilty and do hw reset. After that, we would do the normal resubmit step to resubmit left jobs. 2. For whole gpu reset(vram lost), do resubmit as the old way. v2: squash in build fix (Alex) Signed-off-by: Jack Zhang Reviewed-by: Andrey Grodzovsky Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 81 +++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 2 +- drivers/gpu/drm/scheduler/sched_main.c | 103 ++++++++++++++++++++--------- include/drm/gpu_scheduler.h | 3 + 4 files changed, 156 insertions(+), 33 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 165192b2c336..3cbe8137a6af 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4575,6 +4575,73 @@ static int amdgpu_device_suspend_display_audio(struct amdgpu_device *adev) return 0; } +void amdgpu_device_recheck_guilty_jobs(struct amdgpu_device *adev, + struct amdgpu_hive_info *hive, + struct list_head *device_list_handle, + bool *need_full_reset) +{ + 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); + + 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); + + /* set guilty */ + drm_sched_increase_karma(s_job); +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 { + r = amdgpu_do_asic_reset(hive, device_list_handle, + need_full_reset, false); + 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); + } +} + /** * amdgpu_device_gpu_recover - reset the asic and recover scheduler * @@ -4597,6 +4664,7 @@ 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; /* * Special case: RAS triggered and full reset isn't supported @@ -4748,6 +4816,7 @@ retry: /* Rest of adevs pre asic reset from XGMI hive. */ } } + tmp_vram_lost_counter = atomic_read(&((adev)->vram_lost_counter)); /* Actual ASIC resets if needed.*/ /* TODO Implement XGMI hive reset logic for SRIOV */ if (amdgpu_sriov_vf(adev)) { @@ -4765,6 +4834,18 @@ 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, hive, + device_list_handle, &need_full_reset); + for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { struct amdgpu_ring *ring = tmp_adev->rings[i]; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 02b75fa64d7d..0e10c3958f94 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -516,7 +516,7 @@ module_param_named(compute_multipipe, amdgpu_compute_multipipe, int, 0444); * DOC: gpu_recovery (int) * Set to enable GPU recovery mechanism (1 = enable, 0 = disable). The default is -1 (auto, disabled except SRIOV). */ -MODULE_PARM_DESC(gpu_recovery, "Enable GPU recovery mechanism, (1 = enable, 0 = disable, -1 = auto)"); +MODULE_PARM_DESC(gpu_recovery, "Enable GPU recovery mechanism, (2 = advanced tdr mode, 1 = enable, 0 = disable, -1 = auto)"); module_param_named(gpu_recovery, amdgpu_gpu_recovery, int, 0444); /** diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c index d82a7ebf6099..92d8de24d0a1 100644 --- a/drivers/gpu/drm/scheduler/sched_main.c +++ b/drivers/gpu/drm/scheduler/sched_main.c @@ -361,40 +361,16 @@ static void drm_sched_job_timedout(struct work_struct *work) */ void drm_sched_increase_karma(struct drm_sched_job *bad) { - int i; - struct drm_sched_entity *tmp; - struct drm_sched_entity *entity; - struct drm_gpu_scheduler *sched = bad->sched; - - /* don't increase @bad's karma if it's from KERNEL RQ, - * because sometimes GPU hang would cause kernel jobs (like VM updating jobs) - * corrupt but keep in mind that kernel jobs always considered good. - */ - if (bad->s_priority != DRM_SCHED_PRIORITY_KERNEL) { - atomic_inc(&bad->karma); - for (i = DRM_SCHED_PRIORITY_MIN; i < DRM_SCHED_PRIORITY_KERNEL; - i++) { - struct drm_sched_rq *rq = &sched->sched_rq[i]; - - spin_lock(&rq->lock); - list_for_each_entry_safe(entity, tmp, &rq->entities, list) { - if (bad->s_fence->scheduled.context == - entity->fence_context) { - if (atomic_read(&bad->karma) > - bad->sched->hang_limit) - if (entity->guilty) - atomic_set(entity->guilty, 1); - break; - } - } - spin_unlock(&rq->lock); - if (&entity->list != &rq->entities) - break; - } - } + drm_sched_increase_karma_ext(bad, 1); } EXPORT_SYMBOL(drm_sched_increase_karma); +void drm_sched_reset_karma(struct drm_sched_job *bad) +{ + drm_sched_increase_karma_ext(bad, 0); +} +EXPORT_SYMBOL(drm_sched_reset_karma); + /** * drm_sched_stop - stop the scheduler * @@ -533,15 +509,32 @@ EXPORT_SYMBOL(drm_sched_start); * */ void drm_sched_resubmit_jobs(struct drm_gpu_scheduler *sched) +{ + drm_sched_resubmit_jobs_ext(sched, INT_MAX); +} +EXPORT_SYMBOL(drm_sched_resubmit_jobs); + +/** + * drm_sched_resubmit_jobs_ext - helper to relunch certain number of jobs from mirror ring list + * + * @sched: scheduler instance + * @max: job numbers to relaunch + * + */ +void drm_sched_resubmit_jobs_ext(struct drm_gpu_scheduler *sched, int max) { struct drm_sched_job *s_job, *tmp; uint64_t guilty_context; bool found_guilty = false; struct dma_fence *fence; + int i = 0; list_for_each_entry_safe(s_job, tmp, &sched->pending_list, list) { struct drm_sched_fence *s_fence = s_job->s_fence; + if (i >= max) + break; + if (!found_guilty && atomic_read(&s_job->karma) > sched->hang_limit) { found_guilty = true; guilty_context = s_job->s_fence->scheduled.context; @@ -552,6 +545,7 @@ void drm_sched_resubmit_jobs(struct drm_gpu_scheduler *sched) dma_fence_put(s_job->s_fence->parent); fence = sched->ops->run_job(s_job); + i++; if (IS_ERR_OR_NULL(fence)) { if (IS_ERR(fence)) @@ -563,7 +557,7 @@ void drm_sched_resubmit_jobs(struct drm_gpu_scheduler *sched) } } } -EXPORT_SYMBOL(drm_sched_resubmit_jobs); +EXPORT_SYMBOL(drm_sched_resubmit_jobs_ext); /** * drm_sched_job_init - init a scheduler job @@ -903,3 +897,48 @@ void drm_sched_fini(struct drm_gpu_scheduler *sched) sched->ready = false; } EXPORT_SYMBOL(drm_sched_fini); + +/** + * drm_sched_increase_karma_ext - Update sched_entity guilty flag + * + * @bad: The job guilty of time out + * @type: type for increase/reset karma + * + */ +void drm_sched_increase_karma_ext(struct drm_sched_job *bad, int type) +{ + int i; + struct drm_sched_entity *tmp; + struct drm_sched_entity *entity; + struct drm_gpu_scheduler *sched = bad->sched; + + /* don't change @bad's karma if it's from KERNEL RQ, + * because sometimes GPU hang would cause kernel jobs (like VM updating jobs) + * corrupt but keep in mind that kernel jobs always considered good. + */ + if (bad->s_priority != DRM_SCHED_PRIORITY_KERNEL) { + if (type == 0) + atomic_set(&bad->karma, 0); + else if (type == 1) + atomic_inc(&bad->karma); + + for (i = DRM_SCHED_PRIORITY_MIN; i < DRM_SCHED_PRIORITY_KERNEL; + i++) { + struct drm_sched_rq *rq = &sched->sched_rq[i]; + + spin_lock(&rq->lock); + list_for_each_entry_safe(entity, tmp, &rq->entities, list) { + if (bad->s_fence->scheduled.context == + entity->fence_context) { + if (entity->guilty) + atomic_set(entity->guilty, type); + break; + } + } + spin_unlock(&rq->lock); + if (&entity->list != &rq->entities) + break; + } + } +} +EXPORT_SYMBOL(drm_sched_increase_karma_ext); diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h index f888b5e9583a..10225a0a35d0 100644 --- a/include/drm/gpu_scheduler.h +++ b/include/drm/gpu_scheduler.h @@ -322,7 +322,10 @@ void drm_sched_wakeup(struct drm_gpu_scheduler *sched); void drm_sched_stop(struct drm_gpu_scheduler *sched, struct drm_sched_job *bad); void drm_sched_start(struct drm_gpu_scheduler *sched, bool full_recovery); void drm_sched_resubmit_jobs(struct drm_gpu_scheduler *sched); +void drm_sched_resubmit_jobs_ext(struct drm_gpu_scheduler *sched, int max); void drm_sched_increase_karma(struct drm_sched_job *bad); +void drm_sched_reset_karma(struct drm_sched_job *bad); +void drm_sched_increase_karma_ext(struct drm_sched_job *bad, int type); bool drm_sched_dependency_optimized(struct dma_fence* fence, struct drm_sched_entity *entity); void drm_sched_fault(struct drm_gpu_scheduler *sched); -- cgit v1.2.3 From 04442bf70debb197d4ed4e850aa77213e685b352 Mon Sep 17 00:00:00 2001 From: Lijo Lazar Date: Tue, 16 Mar 2021 20:31:51 +0800 Subject: drm/amdgpu: Add reset control handling to reset workflow This prefers reset control based handling if it's implemented for a particular ASIC. If not, it takes the legacy path. It uses the legacy method of preparing environment (job, scheduler tasks) and restoring environment. v2: remove unused variable (Alex) Signed-off-by: Lijo Lazar Reviewed-by: Feifei Xu Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 11 ++- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 108 +++++++++++++++++++++-------- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 17 ++++- 3 files changed, 97 insertions(+), 39 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 4b72fb32f22c..ec21a71c74eb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -270,6 +270,7 @@ struct amdgpu_bo_va_mapping; struct amdgpu_atif; struct kfd_vm_fault_info; struct amdgpu_hive_info; +struct amdgpu_reset_context; struct amdgpu_reset_control; enum amdgpu_cp_irq { @@ -1075,6 +1076,7 @@ struct amdgpu_device { bool in_pci_err_recovery; struct pci_saved_state *pci_state; + struct amdgpu_reset_control *reset_cntl; }; @@ -1127,13 +1129,10 @@ bool amdgpu_device_asic_has_dc_support(enum amd_asic_type asic_type); bool amdgpu_device_has_dc_support(struct amdgpu_device *adev); int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, - struct amdgpu_job *job, - bool *need_full_reset_arg); + struct amdgpu_reset_context *reset_context); -int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive, - struct list_head *device_list_handle, - bool *need_full_reset_arg, - bool skip_hw_reset); +int amdgpu_do_asic_reset(struct list_head *device_list_handle, + struct amdgpu_reset_context *reset_context); int emu_soc_asic_init(struct amdgpu_device *adev); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 3cbe8137a6af..a51f470631d4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -65,6 +65,7 @@ #include "amdgpu_ras.h" #include "amdgpu_pmu.h" #include "amdgpu_fru_eeprom.h" +#include "amdgpu_reset.h" #include #include @@ -3421,6 +3422,8 @@ int amdgpu_device_init(struct amdgpu_device *adev, goto fence_driver_init; } + amdgpu_reset_init(adev); + /* detect if we are with an SRIOV vbios */ amdgpu_device_detect_sriov_bios(adev); @@ -3671,6 +3674,9 @@ void amdgpu_device_fini(struct amdgpu_device *adev) release_firmware(adev->firmware.gpu_info_fw); adev->firmware.gpu_info_fw = NULL; adev->accel_working = false; + + amdgpu_reset_fini(adev); + /* free i2c buses */ if (!amdgpu_device_has_dc_support(adev)) amdgpu_i2c_fini(adev); @@ -4239,11 +4245,15 @@ int amdgpu_device_mode1_reset(struct amdgpu_device *adev) } int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, - struct amdgpu_job *job, - bool *need_full_reset_arg) + struct amdgpu_reset_context *reset_context) { int i, r = 0; - bool need_full_reset = *need_full_reset_arg; + struct amdgpu_job *job = NULL; + bool need_full_reset = + test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); + + if (reset_context->reset_req_dev == adev) + job = reset_context->job; /* no need to dump if device is not in good state during probe period */ if (!adev->gmc.xgmi.pending_reset) @@ -4268,6 +4278,10 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, if(job) drm_sched_increase_karma(&job->base); + r = amdgpu_reset_prepare_hwcontext(adev, reset_context); + if (r != -ENOSYS) + return r; + /* Don't suspend on bare metal if we are not going to HW reset the ASIC */ if (!amdgpu_sriov_vf(adev)) { @@ -4286,22 +4300,36 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev, if (need_full_reset) r = amdgpu_device_ip_suspend(adev); - - *need_full_reset_arg = need_full_reset; + if (need_full_reset) + set_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); + else + clear_bit(AMDGPU_NEED_FULL_RESET, + &reset_context->flags); } return r; } -int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive, - struct list_head *device_list_handle, - bool *need_full_reset_arg, - bool skip_hw_reset) +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 = *need_full_reset_arg, vram_lost = false; + bool need_full_reset, skip_hw_reset, vram_lost = false; int r = 0; + /* Try reset handler method first */ + tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device, + reset_list); + r = amdgpu_reset_perform_reset(tmp_adev, reset_context); + + if (r != -ENOSYS) + return r; + + /* Reset handler not implemented, use the default method */ + need_full_reset = + test_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); + skip_hw_reset = test_bit(AMDGPU_SKIP_HW_RESET, &reset_context->flags); + /* * ASIC reset has to be done on all XGMI hive nodes ASAP * to allow proper links negotiation in FW (within 1 sec) @@ -4385,7 +4413,8 @@ int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive, */ amdgpu_register_gpu_instance(tmp_adev); - if (!hive && tmp_adev->gmc.xgmi.num_physical_nodes > 1) + if (!reset_context->hive && + tmp_adev->gmc.xgmi.num_physical_nodes > 1) amdgpu_xgmi_add_device(tmp_adev); r = amdgpu_device_ip_late_init(tmp_adev); @@ -4413,8 +4442,10 @@ int amdgpu_do_asic_reset(struct amdgpu_hive_info *hive, } /* Update PSP FW topology after reset */ - if (hive && tmp_adev->gmc.xgmi.num_physical_nodes > 1) - r = amdgpu_xgmi_update_topology(hive, tmp_adev); + if (reset_context->hive && + tmp_adev->gmc.xgmi.num_physical_nodes > 1) + r = amdgpu_xgmi_update_topology( + reset_context->hive, tmp_adev); } } @@ -4438,7 +4469,10 @@ out: } end: - *need_full_reset_arg = need_full_reset; + if (need_full_reset) + set_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); + else + clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context->flags); return r; } @@ -4575,10 +4609,9 @@ static int amdgpu_device_suspend_display_audio(struct amdgpu_device *adev) return 0; } -void amdgpu_device_recheck_guilty_jobs(struct amdgpu_device *adev, - struct amdgpu_hive_info *hive, - struct list_head *device_list_handle, - bool *need_full_reset) +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; @@ -4614,8 +4647,10 @@ retry: if (r) adev->asic_reset_res = r; } else { - r = amdgpu_do_asic_reset(hive, device_list_handle, - need_full_reset, false); + 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; } @@ -4657,7 +4692,6 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, struct amdgpu_job *job) { struct list_head device_list, *device_list_handle = NULL; - bool need_full_reset = false; bool job_signaled = false; struct amdgpu_hive_info *hive = NULL; struct amdgpu_device *tmp_adev = NULL; @@ -4665,6 +4699,9 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, bool need_emergency_restart = false; bool audio_suspended = false; int tmp_vram_lost_counter; + struct amdgpu_reset_context reset_context; + + memset(&reset_context, 0, sizeof(reset_context)); /* * Special case: RAS triggered and full reset isn't supported @@ -4705,6 +4742,12 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, mutex_lock(&hive->hive_lock); } + reset_context.method = AMD_RESET_METHOD_NONE; + reset_context.reset_req_dev = adev; + reset_context.job = job; + reset_context.hive = hive; + clear_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags); + /* * lock the device before we try to operate the linked list * if didn't get the device lock, don't touch the linked list since @@ -4805,9 +4848,7 @@ int amdgpu_device_gpu_recover(struct amdgpu_device *adev, retry: /* Rest of adevs pre asic reset from XGMI hive. */ list_for_each_entry(tmp_adev, device_list_handle, reset_list) { - r = amdgpu_device_pre_asic_reset(tmp_adev, - (tmp_adev == adev) ? job : NULL, - &need_full_reset); + r = amdgpu_device_pre_asic_reset(tmp_adev, &reset_context); /*TODO Should we stop ?*/ if (r) { dev_err(tmp_adev->dev, "GPU pre asic reset failed with err, %d for drm dev, %s ", @@ -4824,7 +4865,7 @@ retry: /* Rest of adevs pre asic reset from XGMI hive. */ if (r) adev->asic_reset_res = r; } else { - r = amdgpu_do_asic_reset(hive, device_list_handle, &need_full_reset, false); + r = amdgpu_do_asic_reset(device_list_handle, &reset_context); if (r && r == -EAGAIN) goto retry; } @@ -4843,8 +4884,8 @@ skip_hw_reset: */ if (amdgpu_gpu_recovery == 2 && !(tmp_vram_lost_counter < atomic_read(&adev->vram_lost_counter))) - amdgpu_device_recheck_guilty_jobs(tmp_adev, hive, - device_list_handle, &need_full_reset); + 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]; @@ -5189,12 +5230,14 @@ pci_ers_result_t amdgpu_pci_slot_reset(struct pci_dev *pdev) struct drm_device *dev = pci_get_drvdata(pdev); struct amdgpu_device *adev = drm_to_adev(dev); int r, i; - bool need_full_reset = true; + struct amdgpu_reset_context reset_context; u32 memsize; struct list_head device_list; DRM_INFO("PCI error: slot reset callback!!\n"); + memset(&reset_context, 0, sizeof(reset_context)); + INIT_LIST_HEAD(&device_list); list_add_tail(&adev->reset_list, &device_list); @@ -5217,13 +5260,18 @@ pci_ers_result_t amdgpu_pci_slot_reset(struct pci_dev *pdev) goto out; } + reset_context.method = AMD_RESET_METHOD_NONE; + reset_context.reset_req_dev = adev; + set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags); + set_bit(AMDGPU_SKIP_HW_RESET, &reset_context.flags); + adev->in_pci_err_recovery = true; - r = amdgpu_device_pre_asic_reset(adev, NULL, &need_full_reset); + r = amdgpu_device_pre_asic_reset(adev, &reset_context); adev->in_pci_err_recovery = false; if (r) goto out; - r = amdgpu_do_asic_reset(NULL, &device_list, &need_full_reset, true); + r = amdgpu_do_asic_reset(&device_list, &reset_context); out: if (!r) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 0e10c3958f94..d8f131ed10cb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -47,6 +47,7 @@ #include "amdgpu_ras.h" #include "amdgpu_xgmi.h" +#include "amdgpu_reset.h" /* * KMS wrapper. @@ -1349,7 +1350,9 @@ static void amdgpu_drv_delayed_reset_work_handler(struct work_struct *work) struct list_head device_list; struct amdgpu_device *adev; int i, r; - bool need_full_reset = true; + struct amdgpu_reset_context reset_context; + + memset(&reset_context, 0, sizeof(reset_context)); mutex_lock(&mgpu_info.mutex); if (mgpu_info.pending_reset == true) { @@ -1359,9 +1362,14 @@ static void amdgpu_drv_delayed_reset_work_handler(struct work_struct *work) mgpu_info.pending_reset = true; mutex_unlock(&mgpu_info.mutex); + /* Use a common context, just need to make sure full reset is done */ + reset_context.method = AMD_RESET_METHOD_NONE; + set_bit(AMDGPU_NEED_FULL_RESET, &reset_context.flags); + for (i = 0; i < mgpu_info.num_dgpu; i++) { adev = mgpu_info.gpu_ins[i].adev; - r = amdgpu_device_pre_asic_reset(adev, NULL, &need_full_reset); + reset_context.reset_req_dev = adev; + r = amdgpu_device_pre_asic_reset(adev, &reset_context); if (r) { dev_err(adev->dev, "GPU pre asic reset failed with err, %d for drm dev, %s ", r, adev_to_drm(adev)->unique); @@ -1388,7 +1396,10 @@ static void amdgpu_drv_delayed_reset_work_handler(struct work_struct *work) list_for_each_entry(adev, &device_list, reset_list) amdgpu_unregister_gpu_instance(adev); - r = amdgpu_do_asic_reset(NULL, &device_list, &need_full_reset, true); + /* Use a common context, just need to make sure full reset is done */ + set_bit(AMDGPU_SKIP_HW_RESET, &reset_context.flags); + r = amdgpu_do_asic_reset(&device_list, &reset_context); + if (r) { DRM_ERROR("reinit gpus failure"); return; -- cgit v1.2.3 From 28a5d7a58949aede16f8572ba501aa2ee2a60657 Mon Sep 17 00:00:00 2001 From: Hawking Zhang Date: Fri, 16 Apr 2021 14:44:27 +0800 Subject: drm/amdgpu: correct default gfx wdt timeout setting When gfx wdt was configured to fatal_disable, the timeout period should be configured to 0x0 (timeout disabled) Signed-off-by: Hawking Zhang Reviewed-by: Dennis Li Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index d8f131ed10cb..922938931e1a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -185,7 +185,7 @@ uint amdgpu_ras_mask = 0xffffffff; int amdgpu_bad_page_threshold = -1; struct amdgpu_watchdog_timer amdgpu_watchdog_timer = { .timeout_fatal_disable = false, - .period = 0x23, /* default to max. timeout = 1 << 0x23 cycles */ + .period = 0x0, /* default to 0x0 (timeout disable) */ }; /** @@ -553,7 +553,7 @@ module_param_named(timeout_fatal_disable, amdgpu_watchdog_timer.timeout_fatal_di * DOC: timeout_period (uint) * Modify the watchdog timeout max_cycles as (1 << period) */ -MODULE_PARM_DESC(timeout_period, "watchdog timeout period (1 to 0x23(default), timeout maxCycles = (1 << period)"); +MODULE_PARM_DESC(timeout_period, "watchdog timeout period (0 = timeout disabled, 1 ~ 0x23 = timeout maxcycles = (1 << period)"); module_param_named(timeout_period, amdgpu_watchdog_timer.period, uint, 0644); /** -- cgit v1.2.3 From b45aeb2dea9142d4d32fa3a117ba381d84f27065 Mon Sep 17 00:00:00 2001 From: Pavan Kumar Ramayanam Date: Tue, 27 Apr 2021 10:21:18 +0530 Subject: drm/amdgpu: Handling of amdgpu_device_resume return value for graceful teardown The runtime resume PM op disregards the return value from amdgpu_device_resume(), masking errors for failed resumes at the PM layer. Reviewed-by: Alex Deucher Signed-off-by: Pavan Kumar Ramayanam Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index d8f131ed10cb..b0378cfe2218 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -1573,6 +1573,9 @@ static int amdgpu_pmops_runtime_resume(struct device *dev) amdgpu_device_baco_exit(drm_dev); } ret = amdgpu_device_resume(drm_dev, false); + if (ret) + return ret; + if (amdgpu_device_supports_px(drm_dev)) drm_dev->switch_power_state = DRM_SWITCH_POWER_ON; adev->in_runpm = false; -- cgit v1.2.3