From e071dce38f0a3acb2f6d019c6bb7933fd20d3357 Mon Sep 17 00:00:00 2001 From: Lijo Lazar Date: Tue, 16 Mar 2021 20:19:06 +0800 Subject: drm/amdgpu: Add reset control to amdgpu_device v1: Add generic amdgpu_reset_control to handle different types of resets. It may be added at device, hive or ip level. Each reset control has a list of handlers associated with it to handle different types of reset. Reset control is responsible for choosing the right handler given a particular reset context. Handler objects may implement a set of functions on how to handle a particular type of reset. prepare_env = Prepare environment/software context (not used currently). prepare_hwcontext = Prepare hardware context for the reset. perform_reset = Perform the type of reset. restore_hwcontext = Restore the hw context after reset. restore_env = Restore the environment after reset (not used currently). Reset context carries the context of reset, as of now this is based on the parameters used for current set of resets. v2: Fix coding style Signed-off-by: Lijo Lazar Reviewed-by: Feifei Xu Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c | 82 +++++++++++++++++++++++++++++++ 1 file changed, 82 insertions(+) create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c new file mode 100644 index 000000000000..5bc94b8320ba --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c @@ -0,0 +1,82 @@ +/* + * Copyright 2021 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#include "amdgpu_reset.h" +#include "aldebaran.h" + +int amdgpu_reset_add_handler(struct amdgpu_reset_control *reset_ctl, + struct amdgpu_reset_handler *handler) +{ + /* TODO: Check if handler exists? */ + list_add_tail(&handler->handler_list, &reset_ctl->reset_handlers); + return 0; +} + +int amdgpu_reset_init(struct amdgpu_device *adev) +{ + int ret = 0; + + return ret; +} + +int amdgpu_reset_fini(struct amdgpu_device *adev) +{ + int ret = 0; + + return ret; +} + +int amdgpu_reset_prepare_hwcontext(struct amdgpu_device *adev, + struct amdgpu_reset_context *reset_context) +{ + struct amdgpu_reset_handler *reset_handler = NULL; + + if (adev->reset_cntl && adev->reset_cntl->get_reset_handler) + reset_handler = adev->reset_cntl->get_reset_handler( + adev->reset_cntl, reset_context); + if (!reset_handler) + return -ENOSYS; + + return reset_handler->prepare_hwcontext(adev->reset_cntl, + reset_context); +} + +int amdgpu_reset_perform_reset(struct amdgpu_device *adev, + struct amdgpu_reset_context *reset_context) +{ + int ret; + struct amdgpu_reset_handler *reset_handler = NULL; + + if (adev->reset_cntl) + reset_handler = adev->reset_cntl->get_reset_handler( + adev->reset_cntl, reset_context); + if (!reset_handler) + return -ENOSYS; + + ret = reset_handler->perform_reset(adev->reset_cntl, reset_context); + if (ret) + return ret; + + return reset_handler->restore_hwcontext(adev->reset_cntl, + reset_context); +} -- cgit v1.2.3 From 142600e854b17f7dac25c8f971c030bb1700d2b6 Mon Sep 17 00:00:00 2001 From: Lijo Lazar Date: Tue, 16 Mar 2021 21:36:28 +0800 Subject: drm/amdgpu: Add mode2 reset support for aldebaran v1: Aldebaran uses reset control to support mode2 reset. The sequences to reset and restore hardware context are specific to a particular configuration. v2: Clear bus mastering before reset. Fix coding style issues, drop unwanted variables and info log. Signed-off-by: Lijo Lazar Reviewed-by: Feifei Xu Reviewed-by: Hawking Zhang Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/Makefile | 2 +- drivers/gpu/drm/amd/amdgpu/aldebaran.c | 407 ++++++++++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/aldebaran.h | 32 +++ drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c | 16 ++ drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 8 +- 5 files changed, 463 insertions(+), 2 deletions(-) create mode 100644 drivers/gpu/drm/amd/amdgpu/aldebaran.c create mode 100644 drivers/gpu/drm/amd/amdgpu/aldebaran.h (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c') diff --git a/drivers/gpu/drm/amd/amdgpu/Makefile b/drivers/gpu/drm/amd/amdgpu/Makefile index a0a5fd1788b5..42153638a55c 100644 --- a/drivers/gpu/drm/amd/amdgpu/Makefile +++ b/drivers/gpu/drm/amd/amdgpu/Makefile @@ -71,7 +71,7 @@ amdgpu-y += \ vi.o mxgpu_vi.o nbio_v6_1.o soc15.o emu_soc.o mxgpu_ai.o nbio_v7_0.o vega10_reg_init.o \ vega20_reg_init.o nbio_v7_4.o nbio_v2_3.o nv.o navi10_reg_init.o navi14_reg_init.o \ arct_reg_init.o navi12_reg_init.o mxgpu_nv.o sienna_cichlid_reg_init.o vangogh_reg_init.o \ - nbio_v7_2.o dimgrey_cavefish_reg_init.o hdp_v4_0.o hdp_v5_0.o aldebaran_reg_init.o + nbio_v7_2.o dimgrey_cavefish_reg_init.o hdp_v4_0.o hdp_v5_0.o aldebaran_reg_init.o aldebaran.o # add DF block amdgpu-y += \ diff --git a/drivers/gpu/drm/amd/amdgpu/aldebaran.c b/drivers/gpu/drm/amd/amdgpu/aldebaran.c new file mode 100644 index 000000000000..39604a461bf5 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/aldebaran.c @@ -0,0 +1,407 @@ +/* + * Copyright 2021 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#include "aldebaran.h" +#include "amdgpu_reset.h" +#include "amdgpu_amdkfd.h" +#include "amdgpu_dpm.h" +#include "amdgpu_job.h" +#include "amdgpu_ring.h" +#include "amdgpu_ras.h" +#include "amdgpu_psp.h" +#include "amdgpu_xgmi.h" + +static struct amdgpu_reset_handler * +aldebaran_get_reset_handler(struct amdgpu_reset_control *reset_ctl, + struct amdgpu_reset_context *reset_context) +{ + struct amdgpu_reset_handler *handler; + struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle; + + if (reset_context->method != AMD_RESET_METHOD_NONE) { + dev_dbg(adev->dev, "Getting reset handler for method %d\n", + reset_context->method); + list_for_each_entry(handler, &reset_ctl->reset_handlers, + handler_list) { + if (handler->reset_method == reset_context->method) + return handler; + } + } + + if (adev->gmc.xgmi.connected_to_cpu) { + list_for_each_entry(handler, &reset_ctl->reset_handlers, + handler_list) { + if (handler->reset_method == AMD_RESET_METHOD_MODE2) { + reset_context->method = AMD_RESET_METHOD_MODE2; + return handler; + } + } + } + + dev_dbg(adev->dev, "Reset handler not found!\n"); + + return NULL; +} + +static int aldebaran_mode2_suspend_ip(struct amdgpu_device *adev) +{ + int r, i; + + amdgpu_device_set_pg_state(adev, AMD_PG_STATE_UNGATE); + amdgpu_device_set_cg_state(adev, AMD_CG_STATE_UNGATE); + + for (i = adev->num_ip_blocks - 1; i >= 0; i--) { + if (!(adev->ip_blocks[i].version->type == + AMD_IP_BLOCK_TYPE_GFX || + adev->ip_blocks[i].version->type == + AMD_IP_BLOCK_TYPE_SDMA)) + continue; + + r = adev->ip_blocks[i].version->funcs->suspend(adev); + + if (r) { + dev_err(adev->dev, + "suspend of IP block <%s> failed %d\n", + adev->ip_blocks[i].version->funcs->name, r); + return r; + } + + adev->ip_blocks[i].status.hw = false; + } + + return r; +} + +static int +aldebaran_mode2_prepare_hwcontext(struct amdgpu_reset_control *reset_ctl, + struct amdgpu_reset_context *reset_context) +{ + int r = 0; + struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle; + + dev_dbg(adev->dev, "Aldebaran prepare hw context\n"); + /* Don't suspend on bare metal if we are not going to HW reset the ASIC */ + if (!amdgpu_sriov_vf(adev)) + r = aldebaran_mode2_suspend_ip(adev); + + return r; +} + +void aldebaran_async_reset(struct work_struct *work) +{ + struct amdgpu_reset_handler *handler; + struct amdgpu_reset_control *reset_ctl = + container_of(work, struct amdgpu_reset_control, reset_work); + struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle; + + list_for_each_entry(handler, &reset_ctl->reset_handlers, + handler_list) { + if (handler->reset_method == reset_ctl->active_reset) { + dev_dbg(adev->dev, "Resetting device\n"); + handler->do_reset(adev); + break; + } + } +} + +static int aldebaran_mode2_reset(struct amdgpu_device *adev) +{ + /* disable BM */ + pci_clear_master(adev->pdev); + adev->asic_reset_res = amdgpu_dpm_mode2_reset(adev); + return adev->asic_reset_res; +} + +static int +aldebaran_mode2_perform_reset(struct amdgpu_reset_control *reset_ctl, + struct amdgpu_reset_context *reset_context) +{ + struct amdgpu_device *tmp_adev = NULL; + struct amdgpu_device *adev = (struct amdgpu_device *)reset_ctl->handle; + int r = 0; + + dev_dbg(adev->dev, "aldebaran perform hw reset\n"); + if (reset_context->hive == NULL) { + /* Wrong context, return error */ + return -EINVAL; + } + + list_for_each_entry(tmp_adev, &reset_context->hive->device_list, + gmc.xgmi.head) { + mutex_lock(&tmp_adev->reset_cntl->reset_lock); + tmp_adev->reset_cntl->active_reset = AMD_RESET_METHOD_MODE2; + } + /* + * Mode2 reset doesn't need any sync between nodes in XGMI hive, instead launch + * them together so that they can be completed asynchronously on multiple nodes + */ + list_for_each_entry(tmp_adev, &reset_context->hive->device_list, + gmc.xgmi.head) { + /* For XGMI run all resets in parallel to speed up the process */ + if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { + if (!queue_work(system_unbound_wq, + &tmp_adev->reset_cntl->reset_work)) + r = -EALREADY; + } else + r = aldebaran_mode2_reset(tmp_adev); + if (r) { + dev_err(tmp_adev->dev, + "ASIC reset failed with error, %d for drm dev, %s", + r, adev_to_drm(tmp_adev)->unique); + break; + } + } + + /* For XGMI wait for all resets to complete before proceed */ + if (!r) { + list_for_each_entry(tmp_adev, + &reset_context->hive->device_list, + gmc.xgmi.head) { + if (tmp_adev->gmc.xgmi.num_physical_nodes > 1) { + flush_work(&tmp_adev->reset_cntl->reset_work); + r = tmp_adev->asic_reset_res; + if (r) + break; + } + } + } + + list_for_each_entry(tmp_adev, &reset_context->hive->device_list, + gmc.xgmi.head) { + mutex_unlock(&tmp_adev->reset_cntl->reset_lock); + tmp_adev->reset_cntl->active_reset = AMD_RESET_METHOD_NONE; + } + + return r; +} + +static int aldebaran_mode2_restore_ip(struct amdgpu_device *adev) +{ + struct amdgpu_firmware_info *ucode_list[AMDGPU_UCODE_ID_MAXIMUM]; + struct amdgpu_firmware_info *ucode; + struct amdgpu_ip_block *cmn_block; + int ucode_count = 0; + int i, r; + + dev_dbg(adev->dev, "Reloading ucodes after reset\n"); + for (i = 0; i < adev->firmware.max_ucodes; i++) { + ucode = &adev->firmware.ucode[i]; + if (!ucode->fw) + continue; + switch (ucode->ucode_id) { + case AMDGPU_UCODE_ID_SDMA0: + case AMDGPU_UCODE_ID_SDMA1: + case AMDGPU_UCODE_ID_SDMA2: + case AMDGPU_UCODE_ID_SDMA3: + case AMDGPU_UCODE_ID_SDMA4: + case AMDGPU_UCODE_ID_SDMA5: + case AMDGPU_UCODE_ID_SDMA6: + case AMDGPU_UCODE_ID_SDMA7: + case AMDGPU_UCODE_ID_CP_MEC1: + case AMDGPU_UCODE_ID_CP_MEC1_JT: + case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL: + case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM: + case AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM: + case AMDGPU_UCODE_ID_RLC_G: + ucode_list[ucode_count++] = ucode; + break; + default: + break; + }; + } + + /* Reinit NBIF block */ + cmn_block = + amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_COMMON); + if (unlikely(!cmn_block)) { + dev_err(adev->dev, "Failed to get BIF handle\n"); + return -EINVAL; + } + r = cmn_block->version->funcs->resume(adev); + if (r) + return r; + + /* Reinit GFXHUB */ + adev->gfxhub.funcs->init(adev); + r = adev->gfxhub.funcs->gart_enable(adev); + if (r) { + dev_err(adev->dev, "GFXHUB gart reenable failed after reset\n"); + return r; + } + + /* Reload GFX firmware */ + r = psp_load_fw_list(&adev->psp, ucode_list, ucode_count); + if (r) { + dev_err(adev->dev, "GFX ucode load failed after reset\n"); + return r; + } + + /* Resume RLC, FW needs RLC alive to complete reset process */ + adev->gfx.rlc.funcs->resume(adev); + + /* Wait for FW reset event complete */ + r = smu_wait_for_event(adev, SMU_EVENT_RESET_COMPLETE, 0); + if (r) { + dev_err(adev->dev, + "Failed to get response from firmware after reset\n"); + return r; + } + + for (i = 0; i < adev->num_ip_blocks; i++) { + if (!(adev->ip_blocks[i].version->type == + AMD_IP_BLOCK_TYPE_GFX || + adev->ip_blocks[i].version->type == + AMD_IP_BLOCK_TYPE_SDMA)) + continue; + r = adev->ip_blocks[i].version->funcs->resume(adev); + if (r) { + dev_err(adev->dev, + "resume of IP block <%s> failed %d\n", + adev->ip_blocks[i].version->funcs->name, r); + return r; + } + + adev->ip_blocks[i].status.hw = true; + } + + for (i = 0; i < adev->num_ip_blocks; i++) { + if (!(adev->ip_blocks[i].version->type == + AMD_IP_BLOCK_TYPE_GFX || + adev->ip_blocks[i].version->type == + AMD_IP_BLOCK_TYPE_SDMA || + adev->ip_blocks[i].version->type == + AMD_IP_BLOCK_TYPE_COMMON)) + continue; + + if (adev->ip_blocks[i].version->funcs->late_init) { + r = adev->ip_blocks[i].version->funcs->late_init( + (void *)adev); + if (r) { + dev_err(adev->dev, + "late_init of IP block <%s> failed %d after reset\n", + adev->ip_blocks[i].version->funcs->name, + r); + return r; + } + } + adev->ip_blocks[i].status.late_initialized = true; + } + + amdgpu_device_set_cg_state(adev, AMD_CG_STATE_GATE); + amdgpu_device_set_pg_state(adev, AMD_PG_STATE_GATE); + + return r; +} + +static int +aldebaran_mode2_restore_hwcontext(struct amdgpu_reset_control *reset_ctl, + struct amdgpu_reset_context *reset_context) +{ + int r; + struct amdgpu_device *tmp_adev = NULL; + + if (reset_context->hive == NULL) { + /* Wrong context, return error */ + return -EINVAL; + } + + list_for_each_entry(tmp_adev, &reset_context->hive->device_list, + gmc.xgmi.head) { + dev_info(tmp_adev->dev, + "GPU reset succeeded, trying to resume\n"); + r = aldebaran_mode2_restore_ip(tmp_adev); + if (r) + goto end; + + /* + * Add this ASIC as tracked as reset was already + * complete successfully. + */ + amdgpu_register_gpu_instance(tmp_adev); + + /* Resume RAS */ + amdgpu_ras_resume(tmp_adev); + + /* Update PSP FW topology after reset */ + if (reset_context->hive && + tmp_adev->gmc.xgmi.num_physical_nodes > 1) + r = amdgpu_xgmi_update_topology(reset_context->hive, + tmp_adev); + + if (!r) { + amdgpu_irq_gpu_reset_resume_helper(tmp_adev); + + r = amdgpu_ib_ring_tests(tmp_adev); + if (r) { + dev_err(tmp_adev->dev, + "ib ring test failed (%d).\n", r); + r = -EAGAIN; + tmp_adev->asic_reset_res = r; + goto end; + } + } + } + +end: + return r; +} + +static struct amdgpu_reset_handler aldebaran_mode2_handler = { + .reset_method = AMD_RESET_METHOD_MODE2, + .prepare_env = NULL, + .prepare_hwcontext = aldebaran_mode2_prepare_hwcontext, + .perform_reset = aldebaran_mode2_perform_reset, + .restore_hwcontext = aldebaran_mode2_restore_hwcontext, + .restore_env = NULL, + .do_reset = aldebaran_mode2_reset, +}; + +int aldebaran_reset_init(struct amdgpu_device *adev) +{ + struct amdgpu_reset_control *reset_ctl; + + reset_ctl = kzalloc(sizeof(*reset_ctl), GFP_KERNEL); + if (!reset_ctl) + return -ENOMEM; + + reset_ctl->handle = adev; + reset_ctl->async_reset = aldebaran_async_reset; + reset_ctl->active_reset = AMD_RESET_METHOD_NONE; + reset_ctl->get_reset_handler = aldebaran_get_reset_handler; + + INIT_LIST_HEAD(&reset_ctl->reset_handlers); + INIT_WORK(&reset_ctl->reset_work, reset_ctl->async_reset); + /* Only mode2 is handled through reset control now */ + amdgpu_reset_add_handler(reset_ctl, &aldebaran_mode2_handler); + + adev->reset_cntl = reset_ctl; + + return 0; +} + +int aldebaran_reset_fini(struct amdgpu_device *adev) +{ + kfree(adev->reset_cntl); + adev->reset_cntl = NULL; + return 0; +} diff --git a/drivers/gpu/drm/amd/amdgpu/aldebaran.h b/drivers/gpu/drm/amd/amdgpu/aldebaran.h new file mode 100644 index 000000000000..a07db5454d49 --- /dev/null +++ b/drivers/gpu/drm/amd/amdgpu/aldebaran.h @@ -0,0 +1,32 @@ +/* + * Copyright 2021 Advanced Micro Devices, Inc. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + * OTHER DEALINGS IN THE SOFTWARE. + * + */ + +#ifndef __ALDEBARAN_H__ +#define __ALDEBARAN_H__ + +#include "amdgpu.h" + +int aldebaran_reset_init(struct amdgpu_device *adev); +int aldebaran_reset_fini(struct amdgpu_device *adev); + +#endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c index 5bc94b8320ba..02afd4115675 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c @@ -36,6 +36,14 @@ int amdgpu_reset_init(struct amdgpu_device *adev) { int ret = 0; + switch (adev->asic_type) { + case CHIP_ALDEBARAN: + ret = aldebaran_reset_init(adev); + break; + default: + break; + } + return ret; } @@ -43,6 +51,14 @@ int amdgpu_reset_fini(struct amdgpu_device *adev) { int ret = 0; + switch (adev->asic_type) { + case CHIP_ALDEBARAN: + ret = aldebaran_reset_fini(adev); + break; + default: + break; + } + return ret; } diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c index 63512370a51b..99f58439f3d5 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c @@ -4024,8 +4024,14 @@ static int gfx_v9_0_hw_fini(void *handle) } gfx_v9_0_cp_enable(adev, false); - adev->gfx.rlc.funcs->stop(adev); + /* Skip suspend with A+A reset */ + if (adev->gmc.xgmi.connected_to_cpu && amdgpu_in_reset(adev)) { + dev_dbg(adev->dev, "Device in reset. Skipping RLC halt\n"); + return 0; + } + + adev->gfx.rlc.funcs->stop(adev); return 0; } -- cgit v1.2.3