From cfbb6b0047448e2d986160d9f30d60f604d9ad0f Mon Sep 17 00:00:00 2001 From: Andrey Grodzovsky Date: Fri, 21 Jan 2022 17:23:32 -0500 Subject: drm/amdgpu: Rework reset domain to be refcounted. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The reset domain contains register access semaphor now and so needs to be present as long as each device in a hive needs it and so it cannot be binded to XGMI hive life cycle. Adress this by making reset domain refcounted and pointed by each member of the hive and the hive itself. v4: Fix crash on boot witrh XGMI hive by adding type to reset_domain. XGMI will only create a new reset_domain if prevoius was of single device type meaning it's first boot. Otherwsie it will take a refocunt to exsiting reset_domain from the amdgou device. Add a wrapper around reset_domain->refcount get/put and a wrapper around send to reset wq (Lijo) Signed-off-by: Andrey Grodzovsky Acked-by: Christian König Link: https://www.spinics.net/lists/amd-gfx/msg74121.html --- drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c | 40 +++++++++++++++++++++++++++++++ 1 file changed, 40 insertions(+) (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 index 02afd4115675..91864947063f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c @@ -96,3 +96,43 @@ int amdgpu_reset_perform_reset(struct amdgpu_device *adev, return reset_handler->restore_hwcontext(adev->reset_cntl, reset_context); } + + +void amdgpu_reset_destroy_reset_domain(struct kref *ref) +{ + struct amdgpu_reset_domain *reset_domain = container_of(ref, + struct amdgpu_reset_domain, + refcount); + if (reset_domain->wq) + destroy_workqueue(reset_domain->wq); + + kvfree(reset_domain); +} + +struct amdgpu_reset_domain *amdgpu_reset_create_reset_domain(enum amdgpu_reset_domain_type type, + char *wq_name) +{ + struct amdgpu_reset_domain *reset_domain; + + reset_domain = kvzalloc(sizeof(struct amdgpu_reset_domain), GFP_KERNEL); + if (!reset_domain) { + DRM_ERROR("Failed to allocate amdgpu_reset_domain!"); + return NULL; + } + + reset_domain->type = type; + kref_init(&reset_domain->refcount); + + reset_domain->wq = create_singlethread_workqueue(wq_name); + if (!reset_domain->wq) { + DRM_ERROR("Failed to allocate wq for amdgpu_reset_domain!"); + amdgpu_reset_put_reset_domain(reset_domain); + return NULL; + + } + + return reset_domain; +} + + + -- cgit From d0fb18b535679a28b1f55a312b7454563b9bb36e Mon Sep 17 00:00:00 2001 From: Andrey Grodzovsky Date: Wed, 19 Jan 2022 17:09:58 -0500 Subject: drm/amdgpu: Move reset sem into reset_domain MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit We want single instance of reset sem across all reset clients because in case of XGMI we should stop access cross device MMIO because any of them could be in a reset in the moment. Signed-off-by: Andrey Grodzovsky Reviewed-by: Christian König Link: https://www.spinics.net/lists/amd-gfx/msg74117.html --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 - drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 10 ++++++---- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 23 +++++++++++------------ drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c | 18 ++++++++++-------- drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h | 1 + drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 6 ++++-- drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 14 ++++++++------ drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c | 4 ++-- 10 files changed, 46 insertions(+), 37 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index cb9764513df8..ddfbcc8fd3d3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1058,7 +1058,6 @@ struct amdgpu_device { atomic_t in_gpu_reset; enum pp_mp1_state mp1_state; - struct rw_semaphore reset_sem; struct amdgpu_doorbell_index doorbell_index; struct mutex notifier_lock; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c index 25e2e5bf90eb..c3728061d65a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c @@ -37,6 +37,8 @@ #include "amdgpu_fw_attestation.h" #include "amdgpu_umr.h" +#include "amdgpu_reset.h" + #if defined(CONFIG_DEBUG_FS) /** @@ -1279,7 +1281,7 @@ static int amdgpu_debugfs_test_ib_show(struct seq_file *m, void *unused) } /* Avoid accidently unparking the sched thread during GPU reset */ - r = down_write_killable(&adev->reset_sem); + r = down_write_killable(&adev->reset_domain->sem); if (r) return r; @@ -1308,7 +1310,7 @@ static int amdgpu_debugfs_test_ib_show(struct seq_file *m, void *unused) kthread_unpark(ring->sched.thread); } - up_write(&adev->reset_sem); + up_write(&adev->reset_domain->sem); pm_runtime_mark_last_busy(dev->dev); pm_runtime_put_autosuspend(dev->dev); @@ -1517,7 +1519,7 @@ static int amdgpu_debugfs_ib_preempt(void *data, u64 val) return -ENOMEM; /* Avoid accidently unparking the sched thread during GPU reset */ - r = down_read_killable(&adev->reset_sem); + r = down_read_killable(&adev->reset_domain->sem); if (r) goto pro_end; @@ -1560,7 +1562,7 @@ failure: /* restart the scheduler */ kthread_unpark(ring->sched.thread); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); ttm_bo_unlock_delayed_workqueue(&adev->mman.bdev, resched); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index d61bc0a0457c..dcbb175d336f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -424,10 +424,10 @@ bool amdgpu_device_skip_hw_access(struct amdgpu_device *adev) * the lock. */ if (in_task()) { - if (down_read_trylock(&adev->reset_sem)) - up_read(&adev->reset_sem); + if (down_read_trylock(&adev->reset_domain->sem)) + up_read(&adev->reset_domain->sem); else - lockdep_assert_held(&adev->reset_sem); + lockdep_assert_held(&adev->reset_domain->sem); } #endif return false; @@ -453,9 +453,9 @@ uint32_t amdgpu_device_rreg(struct amdgpu_device *adev, if ((reg * 4) < adev->rmmio_size) { if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev) && - down_read_trylock(&adev->reset_sem)) { + down_read_trylock(&adev->reset_domain->sem)) { ret = amdgpu_kiq_rreg(adev, reg); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); } else { ret = readl(((void __iomem *)adev->rmmio) + (reg * 4)); } @@ -538,9 +538,9 @@ void amdgpu_device_wreg(struct amdgpu_device *adev, if ((reg * 4) < adev->rmmio_size) { if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev) && - down_read_trylock(&adev->reset_sem)) { + down_read_trylock(&adev->reset_domain->sem)) { amdgpu_kiq_wreg(adev, reg, v); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); } else { writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); } @@ -3555,7 +3555,6 @@ int amdgpu_device_init(struct amdgpu_device *adev, mutex_init(&adev->virt.vf_errors.lock); hash_init(adev->mn_hash); atomic_set(&adev->in_gpu_reset, 0); - init_rwsem(&adev->reset_sem); mutex_init(&adev->psp.mutex); mutex_init(&adev->notifier_lock); @@ -4833,9 +4832,9 @@ static void amdgpu_device_lock_adev(struct amdgpu_device *adev, atomic_set(&adev->in_gpu_reset, 1); if (hive) { - down_write_nest_lock(&adev->reset_sem, &hive->hive_lock); + down_write_nest_lock(&adev->reset_domain->sem, &hive->hive_lock); } else { - down_write(&adev->reset_sem); + down_write(&adev->reset_domain->sem); } switch (amdgpu_asic_reset_method(adev)) { @@ -4856,7 +4855,7 @@ static void amdgpu_device_unlock_adev(struct amdgpu_device *adev) amdgpu_vf_error_trans_all(adev); adev->mp1_state = PP_MP1_STATE_NONE; atomic_set(&adev->in_gpu_reset, 0); - up_write(&adev->reset_sem); + up_write(&adev->reset_domain->sem); } static void amdgpu_device_resume_display_audio(struct amdgpu_device *adev) @@ -5476,7 +5475,7 @@ pci_ers_result_t amdgpu_pci_error_detected(struct pci_dev *pdev, pci_channel_sta /* Fatal error, prepare for slot reset */ case pci_channel_io_frozen: /* - * Locking adev->reset_sem will prevent any external access + * Locking adev->reset_domain->sem will prevent any external access * to GPU during PCI error recovery */ amdgpu_device_lock_adev(adev, NULL); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c index 05117eda105b..d3e055314804 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c @@ -31,6 +31,8 @@ #include #include +#include "amdgpu_reset.h" + #define EEPROM_I2C_MADDR_VEGA20 0x0 #define EEPROM_I2C_MADDR_ARCTURUS 0x40000 #define EEPROM_I2C_MADDR_ARCTURUS_D342 0x0 @@ -193,12 +195,12 @@ static int __write_table_header(struct amdgpu_ras_eeprom_control *control) __encode_table_header_to_buf(&control->tbl_hdr, buf); /* i2c may be unstable in gpu reset */ - down_read(&adev->reset_sem); + down_read(&adev->reset_domain->sem); res = amdgpu_eeprom_write(&adev->pm.smu_i2c, control->i2c_address + control->ras_header_offset, buf, RAS_TABLE_HEADER_SIZE); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); if (res < 0) { DRM_ERROR("Failed to write EEPROM table header:%d", res); @@ -387,13 +389,13 @@ static int __amdgpu_ras_eeprom_write(struct amdgpu_ras_eeprom_control *control, int res; /* i2c may be unstable in gpu reset */ - down_read(&adev->reset_sem); + down_read(&adev->reset_domain->sem); buf_size = num * RAS_TABLE_RECORD_SIZE; res = amdgpu_eeprom_write(&adev->pm.smu_i2c, control->i2c_address + RAS_INDEX_TO_OFFSET(control, fri), buf, buf_size); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); if (res < 0) { DRM_ERROR("Writing %d EEPROM table records error:%d", num, res); @@ -547,12 +549,12 @@ amdgpu_ras_eeprom_update_header(struct amdgpu_ras_eeprom_control *control) goto Out; } - down_read(&adev->reset_sem); + down_read(&adev->reset_domain->sem); res = amdgpu_eeprom_read(&adev->pm.smu_i2c, control->i2c_address + control->ras_record_offset, buf, buf_size); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); if (res < 0) { DRM_ERROR("EEPROM failed reading records:%d\n", res); @@ -642,13 +644,13 @@ static int __amdgpu_ras_eeprom_read(struct amdgpu_ras_eeprom_control *control, int res; /* i2c may be unstable in gpu reset */ - down_read(&adev->reset_sem); + down_read(&adev->reset_domain->sem); buf_size = num * RAS_TABLE_RECORD_SIZE; res = amdgpu_eeprom_read(&adev->pm.smu_i2c, control->i2c_address + RAS_INDEX_TO_OFFSET(control, fri), buf, buf_size); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); if (res < 0) { DRM_ERROR("Reading %d EEPROM table records error:%d", num, res); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c index 91864947063f..c0988c804459 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c @@ -131,6 +131,8 @@ struct amdgpu_reset_domain *amdgpu_reset_create_reset_domain(enum amdgpu_reset_d } + init_rwsem(&reset_domain->sem); + return reset_domain; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h index cc625e441fa0..80f918e87d4f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h @@ -80,6 +80,7 @@ struct amdgpu_reset_domain { struct kref refcount; struct workqueue_struct *wq; enum amdgpu_reset_domain_type type; + struct rw_semaphore sem; }; diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c index 38bb42727715..222b1da9d601 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c @@ -48,6 +48,8 @@ #include "athub_v2_0.h" #include "athub_v2_1.h" +#include "amdgpu_reset.h" + #if 0 static const struct soc15_reg_golden golden_settings_navi10_hdp[] = { @@ -328,7 +330,7 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid, */ if (adev->gfx.kiq.ring.sched.ready && (amdgpu_sriov_runtime(adev) || !amdgpu_sriov_vf(adev)) && - down_read_trylock(&adev->reset_sem)) { + down_read_trylock(&adev->reset_domain->sem)) { struct amdgpu_vmhub *hub = &adev->vmhub[vmhub]; const unsigned eng = 17; u32 inv_req = hub->vmhub_funcs->get_invalidate_req(vmid, flush_type); @@ -338,7 +340,7 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid, amdgpu_virt_kiq_reg_write_reg_wait(adev, req, ack, inv_req, 1 << vmid); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); return; } diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c index 88c1eb9ad068..3a5efe969735 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c @@ -62,6 +62,8 @@ #include "amdgpu_ras.h" #include "amdgpu_xgmi.h" +#include "amdgpu_reset.h" + /* add these here since we already include dce12 headers and these are for DCN */ #define mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION 0x055d #define mmHUBP0_DCSURF_PRI_VIEWPORT_DIMENSION_BASE_IDX 2 @@ -787,13 +789,13 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid, */ if (adev->gfx.kiq.ring.sched.ready && (amdgpu_sriov_runtime(adev) || !amdgpu_sriov_vf(adev)) && - down_read_trylock(&adev->reset_sem)) { + down_read_trylock(&adev->reset_domain->sem)) { uint32_t req = hub->vm_inv_eng0_req + hub->eng_distance * eng; uint32_t ack = hub->vm_inv_eng0_ack + hub->eng_distance * eng; amdgpu_virt_kiq_reg_write_reg_wait(adev, req, ack, inv_req, 1 << vmid); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); return; } @@ -900,7 +902,7 @@ static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev, if (amdgpu_in_reset(adev)) return -EIO; - if (ring->sched.ready && down_read_trylock(&adev->reset_sem)) { + if (ring->sched.ready && down_read_trylock(&adev->reset_domain->sem)) { /* Vega20+XGMI caches PTEs in TC and TLB. Add a * heavy-weight TLB flush (type 2), which flushes * both. Due to a race condition with concurrent @@ -927,7 +929,7 @@ static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev, if (r) { amdgpu_ring_undo(ring); spin_unlock(&adev->gfx.kiq.ring_lock); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); return -ETIME; } @@ -936,10 +938,10 @@ static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev, r = amdgpu_fence_wait_polling(ring, seq, adev->usec_timeout); if (r < 1) { dev_err(adev->dev, "wait for kiq fence error: %ld.\n", r); - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); return -ETIME; } - up_read(&adev->reset_sem); + up_read(&adev->reset_domain->sem); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c index 6740eef84ee1..4e23c29e665c 100644 --- a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c +++ b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c @@ -262,7 +262,7 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work) if (atomic_cmpxchg(&adev->in_gpu_reset, 0, 1) != 0) return; - down_write(&adev->reset_sem); + down_write(&adev->reset_domain->sem); amdgpu_virt_fini_data_exchange(adev); @@ -278,7 +278,7 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work) flr_done: atomic_set(&adev->in_gpu_reset, 0); - up_write(&adev->reset_sem); + up_write(&adev->reset_domain->sem); /* Trigger recovery for world switch failure if no TDR */ if (amdgpu_device_should_recover_gpu(adev) diff --git a/drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c b/drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c index e967d61c7134..f715780f7d20 100644 --- a/drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c +++ b/drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c @@ -286,7 +286,7 @@ static void xgpu_nv_mailbox_flr_work(struct work_struct *work) if (atomic_cmpxchg(&adev->in_gpu_reset, 0, 1) != 0) return; - down_write(&adev->reset_sem); + down_write(&adev->reset_domain->sem); amdgpu_virt_fini_data_exchange(adev); @@ -302,7 +302,7 @@ static void xgpu_nv_mailbox_flr_work(struct work_struct *work) flr_done: atomic_set(&adev->in_gpu_reset, 0); - up_write(&adev->reset_sem); + up_write(&adev->reset_domain->sem); /* Trigger recovery for world switch failure if no TDR */ if (amdgpu_device_should_recover_gpu(adev) -- cgit From 89a7a87093d67e2c633e1ed400ba00ffd15bdae5 Mon Sep 17 00:00:00 2001 From: Andrey Grodzovsky Date: Wed, 19 Jan 2022 17:20:00 -0500 Subject: drm/amdgpu: Move in_gpu_reset into reset_domain MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit We should have a single instance per entrire reset domain. Signed-off-by: Andrey Grodzovsky Suggested-by: Lijo Lazar Reviewed-by: Christian König Link: https://www.spinics.net/lists/amd-gfx/msg74116.html --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 7 ++----- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 10 +++++++--- drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h | 1 + drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c | 4 ++-- 6 files changed, 15 insertions(+), 12 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index ddfbcc8fd3d3..b89406b01694 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1056,7 +1056,6 @@ struct amdgpu_device { bool in_s4; bool in_s0ix; - atomic_t in_gpu_reset; enum pp_mp1_state mp1_state; struct amdgpu_doorbell_index doorbell_index; @@ -1463,8 +1462,6 @@ static inline bool amdgpu_is_tmz(struct amdgpu_device *adev) return adev->gmc.tmz_enabled; } -static inline int amdgpu_in_reset(struct amdgpu_device *adev) -{ - return atomic_read(&adev->in_gpu_reset); -} +int amdgpu_in_reset(struct amdgpu_device *adev); + #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index dcbb175d336f..e05d7cbefd2c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -3554,7 +3554,6 @@ int amdgpu_device_init(struct amdgpu_device *adev, mutex_init(&adev->mn_lock); mutex_init(&adev->virt.vf_errors.lock); hash_init(adev->mn_hash); - atomic_set(&adev->in_gpu_reset, 0); mutex_init(&adev->psp.mutex); mutex_init(&adev->notifier_lock); @@ -4829,7 +4828,7 @@ end: static void amdgpu_device_lock_adev(struct amdgpu_device *adev, struct amdgpu_hive_info *hive) { - atomic_set(&adev->in_gpu_reset, 1); + atomic_set(&adev->reset_domain->in_gpu_reset, 1); if (hive) { down_write_nest_lock(&adev->reset_domain->sem, &hive->hive_lock); @@ -4854,7 +4853,7 @@ static void amdgpu_device_unlock_adev(struct amdgpu_device *adev) { amdgpu_vf_error_trans_all(adev); adev->mp1_state = PP_MP1_STATE_NONE; - atomic_set(&adev->in_gpu_reset, 0); + atomic_set(&adev->reset_domain->in_gpu_reset, 0); up_write(&adev->reset_domain->sem); } @@ -5699,6 +5698,11 @@ void amdgpu_device_invalidate_hdp(struct amdgpu_device *adev, amdgpu_asic_invalidate_hdp(adev, ring); } +int amdgpu_in_reset(struct amdgpu_device *adev) +{ + return atomic_read(&adev->reset_domain->in_gpu_reset); + } + /** * amdgpu_device_halt() - bring hardware to some kind of halt state * diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c index c0988c804459..5ab72c3bfbda 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c @@ -131,6 +131,7 @@ struct amdgpu_reset_domain *amdgpu_reset_create_reset_domain(enum amdgpu_reset_d } + atomic_set(&reset_domain->in_gpu_reset, 0); init_rwsem(&reset_domain->sem); return reset_domain; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h index 80f918e87d4f..ea6fc98ea927 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h @@ -81,6 +81,7 @@ struct amdgpu_reset_domain { struct workqueue_struct *wq; enum amdgpu_reset_domain_type type; struct rw_semaphore sem; + atomic_t in_gpu_reset; }; diff --git a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c index 4e23c29e665c..b81acf59870c 100644 --- a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c +++ b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c @@ -259,7 +259,7 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work) * otherwise the mailbox msg will be ruined/reseted by * the VF FLR. */ - if (atomic_cmpxchg(&adev->in_gpu_reset, 0, 1) != 0) + if (atomic_cmpxchg(&adev->reset_domain->in_gpu_reset, 0, 1) != 0) return; down_write(&adev->reset_domain->sem); @@ -277,7 +277,7 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work) } while (timeout > 1); flr_done: - atomic_set(&adev->in_gpu_reset, 0); + atomic_set(&adev->reset_domain->in_gpu_reset, 0); up_write(&adev->reset_domain->sem); /* Trigger recovery for world switch failure if no TDR */ diff --git a/drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c b/drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c index f715780f7d20..22c10b97ea81 100644 --- a/drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c +++ b/drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c @@ -283,7 +283,7 @@ static void xgpu_nv_mailbox_flr_work(struct work_struct *work) * otherwise the mailbox msg will be ruined/reseted by * the VF FLR. */ - if (atomic_cmpxchg(&adev->in_gpu_reset, 0, 1) != 0) + if (atomic_cmpxchg(&adev->reset_domain->in_gpu_reset, 0, 1) != 0) return; down_write(&adev->reset_domain->sem); @@ -301,7 +301,7 @@ static void xgpu_nv_mailbox_flr_work(struct work_struct *work) } while (timeout > 1); flr_done: - atomic_set(&adev->in_gpu_reset, 0); + atomic_set(&adev->reset_domain->in_gpu_reset, 0); up_write(&adev->reset_domain->sem); /* Trigger recovery for world switch failure if no TDR */ -- cgit From e923be9934a9c54a94e443f9e77bda5b9fbd1ce5 Mon Sep 17 00:00:00 2001 From: Andrey Grodzovsky Date: Tue, 25 Jan 2022 11:32:47 -0500 Subject: drm/amdgpu: Rework amdgpu_device_lock_adev MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This functions needs to be split into 2 parts where one is called only once for locking single instance of reset_domain's sem and reset flag and the other part which handles MP1 states should still be called for each device in XGMI hive. Signed-off-by: Andrey Grodzovsky Reviewed-by: Christian König Link: https://www.spinics.net/lists/amd-gfx/msg74118.html --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 37 ++++++++++++++++-------------- drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c | 19 +++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h | 4 ++++ 3 files changed, 43 insertions(+), 17 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index e05d7cbefd2c..f69ab22ff096 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -4825,16 +4825,8 @@ end: return r; } -static void amdgpu_device_lock_adev(struct amdgpu_device *adev, - struct amdgpu_hive_info *hive) +static void amdgpu_device_set_mp1_state(struct amdgpu_device *adev) { - atomic_set(&adev->reset_domain->in_gpu_reset, 1); - - if (hive) { - down_write_nest_lock(&adev->reset_domain->sem, &hive->hive_lock); - } else { - down_write(&adev->reset_domain->sem); - } switch (amdgpu_asic_reset_method(adev)) { case AMD_RESET_METHOD_MODE1: @@ -4849,12 +4841,10 @@ static void amdgpu_device_lock_adev(struct amdgpu_device *adev, } } -static void amdgpu_device_unlock_adev(struct amdgpu_device *adev) +static void amdgpu_device_unset_mp1_state(struct amdgpu_device *adev) { amdgpu_vf_error_trans_all(adev); adev->mp1_state = PP_MP1_STATE_NONE; - atomic_set(&adev->reset_domain->in_gpu_reset, 0); - up_write(&adev->reset_domain->sem); } static void amdgpu_device_resume_display_audio(struct amdgpu_device *adev) @@ -5060,10 +5050,15 @@ int amdgpu_device_gpu_recover_imp(struct amdgpu_device *adev, device_list_handle = &device_list; } + /* We need to lock reset domain only once both for XGMI and single device */ + tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device, + reset_list); + amdgpu_device_lock_reset_domain(tmp_adev->reset_domain, hive); + /* block all schedulers and reset given job's ring */ list_for_each_entry(tmp_adev, device_list_handle, reset_list) { - amdgpu_device_lock_adev(tmp_adev, hive); + amdgpu_device_set_mp1_state(tmp_adev); /* * Try to put the audio codec into suspend state @@ -5213,9 +5208,14 @@ skip_sched_resume: if (audio_suspended) amdgpu_device_resume_display_audio(tmp_adev); - amdgpu_device_unlock_adev(tmp_adev); + + amdgpu_device_unset_mp1_state(tmp_adev); } + tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device, + reset_list); + amdgpu_device_unlock_reset_domain(tmp_adev->reset_domain); + if (hive) { mutex_unlock(&hive->hive_lock); amdgpu_put_xgmi_hive(hive); @@ -5477,7 +5477,8 @@ pci_ers_result_t amdgpu_pci_error_detected(struct pci_dev *pdev, pci_channel_sta * Locking adev->reset_domain->sem will prevent any external access * to GPU during PCI error recovery */ - amdgpu_device_lock_adev(adev, NULL); + amdgpu_device_lock_reset_domain(adev->reset_domain, NULL); + amdgpu_device_set_mp1_state(adev); /* * Block any work scheduling as we do for regular GPU reset @@ -5584,7 +5585,8 @@ out: DRM_INFO("PCIe error recovery succeeded\n"); } else { DRM_ERROR("PCIe error recovery failed, err:%d", r); - amdgpu_device_unlock_adev(adev); + amdgpu_device_unset_mp1_state(adev); + amdgpu_device_unlock_reset_domain(adev->reset_domain); } return r ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED; @@ -5621,7 +5623,8 @@ void amdgpu_pci_resume(struct pci_dev *pdev) drm_sched_start(&ring->sched, true); } - amdgpu_device_unlock_adev(adev); + amdgpu_device_unset_mp1_state(adev); + amdgpu_device_unlock_reset_domain(adev->reset_domain); } bool amdgpu_device_cache_pci_state(struct pci_dev *pdev) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c index 5ab72c3bfbda..9b18ad0b77ab 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c @@ -137,5 +137,24 @@ struct amdgpu_reset_domain *amdgpu_reset_create_reset_domain(enum amdgpu_reset_d return reset_domain; } +void amdgpu_device_lock_reset_domain(struct amdgpu_reset_domain *reset_domain, + struct amdgpu_hive_info *hive) +{ + atomic_set(&reset_domain->in_gpu_reset, 1); + + if (hive) { + down_write_nest_lock(&reset_domain->sem, &hive->hive_lock); + } else { + down_write(&reset_domain->sem); + } +} + + +void amdgpu_device_unlock_reset_domain(struct amdgpu_reset_domain *reset_domain) +{ + atomic_set(&reset_domain->in_gpu_reset, 0); + up_write(&reset_domain->sem); +} + diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h index ea6fc98ea927..92de3b7965a1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h @@ -118,5 +118,9 @@ static inline bool amdgpu_reset_domain_schedule(struct amdgpu_reset_domain *doma return queue_work(domain->wq, work); } +void amdgpu_device_lock_reset_domain(struct amdgpu_reset_domain *reset_domain, + struct amdgpu_hive_info *hive); + +void amdgpu_device_unlock_reset_domain(struct amdgpu_reset_domain *reset_domain); #endif -- cgit From 3675c2f26f33ab4928859fb8950a4697a16be5c9 Mon Sep 17 00:00:00 2001 From: Andrey Grodzovsky Date: Tue, 25 Jan 2022 11:36:18 -0500 Subject: drm/amdgpu: Revert 'drm/amdgpu: annotate a false positive recursive locking' MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Since we have a single instance of reset semaphore which we lock only once even for XGMI hive we don't need the nested locking hint anymore. Signed-off-by: Andrey Grodzovsky Reviewed-by: Christian König Link: https://www.spinics.net/lists/amd-gfx/msg74120.html --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c | 10 ++-------- 2 files changed, 4 insertions(+), 10 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index f69ab22ff096..54f8e1fa4cae 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -5053,7 +5053,7 @@ int amdgpu_device_gpu_recover_imp(struct amdgpu_device *adev, /* We need to lock reset domain only once both for XGMI and single device */ tmp_adev = list_first_entry(device_list_handle, struct amdgpu_device, reset_list); - amdgpu_device_lock_reset_domain(tmp_adev->reset_domain, hive); + amdgpu_device_lock_reset_domain(tmp_adev->reset_domain); /* block all schedulers and reset given job's ring */ list_for_each_entry(tmp_adev, device_list_handle, reset_list) { @@ -5477,7 +5477,7 @@ pci_ers_result_t amdgpu_pci_error_detected(struct pci_dev *pdev, pci_channel_sta * Locking adev->reset_domain->sem will prevent any external access * to GPU during PCI error recovery */ - amdgpu_device_lock_reset_domain(adev->reset_domain, NULL); + amdgpu_device_lock_reset_domain(adev->reset_domain); amdgpu_device_set_mp1_state(adev); /* diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c index 9b18ad0b77ab..248d64158721 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c @@ -137,16 +137,10 @@ struct amdgpu_reset_domain *amdgpu_reset_create_reset_domain(enum amdgpu_reset_d return reset_domain; } -void amdgpu_device_lock_reset_domain(struct amdgpu_reset_domain *reset_domain, - struct amdgpu_hive_info *hive) +void amdgpu_device_lock_reset_domain(struct amdgpu_reset_domain *reset_domain) { atomic_set(&reset_domain->in_gpu_reset, 1); - - if (hive) { - down_write_nest_lock(&reset_domain->sem, &hive->hive_lock); - } else { - down_write(&reset_domain->sem); - } + down_write(&reset_domain->sem); } -- cgit