drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -1194,9 +1194,15 @@ struct amdgpu_device {
 	bool                            debug_exp_resets;
 	bool                            debug_disable_gpu_ring_reset;
 
-	bool				enforce_isolation[MAX_XCP];
-	/* Added this mutex for cleaner shader isolation between GFX and compute processes */
+	/* Protection for the following isolation structure */
 	struct mutex                    enforce_isolation_mutex;
+	bool				enforce_isolation[MAX_XCP];
+	struct amdgpu_isolation {
+		void			*owner;
+		struct dma_fence	*spearhead;
+		struct amdgpu_sync	active;
+		struct amdgpu_sync	prev;
+	} isolation[MAX_XCP];
 
 	struct amdgpu_init_level *init_lvl;
 
@@ -1482,6 +1488,9 @@ void amdgpu_device_pcie_port_wreg(struct amdgpu_device *adev,
 struct dma_fence *amdgpu_device_get_gang(struct amdgpu_device *adev);
 struct dma_fence *amdgpu_device_switch_gang(struct amdgpu_device *adev,
 					    struct dma_fence *gang);
+struct dma_fence *amdgpu_device_enforce_isolation(struct amdgpu_device *adev,
+						  struct amdgpu_ring *ring,
+						  struct amdgpu_job *job);
 bool amdgpu_device_has_display_hardware(struct amdgpu_device *adev);
 ssize_t amdgpu_get_soft_full_reset_mask(struct amdgpu_ring *ring);
 ssize_t amdgpu_show_reset_mask(char *buf, uint32_t supported_reset);

drivers/gpu/drm/amd/amdgpu/amdgpu_aca.c

@@ -391,6 +391,7 @@ static void aca_banks_generate_cper(struct amdgpu_device *adev,
 {
 	struct aca_bank_node *node;
 	struct aca_bank *bank;
+	int r;
 
 	if (!adev->cper.enabled)
 		return;
@@ -402,11 +403,27 @@ static void aca_banks_generate_cper(struct amdgpu_device *adev,
 
 	/* UEs must be encoded into separate CPER entries */
 	if (type == ACA_SMU_TYPE_UE) {
+		struct aca_banks de_banks;
+
+		aca_banks_init(&de_banks);
 		list_for_each_entry(node, &banks->list, node) {
 			bank = &node->bank;
-			if (amdgpu_cper_generate_ue_record(adev, bank))
-				dev_warn(adev->dev, "fail to generate ue cper records\n");
+			if (bank->aca_err_type == ACA_ERROR_TYPE_DEFERRED) {
+				r = aca_banks_add_bank(&de_banks, bank);
+				if (r)
+					dev_warn(adev->dev, "fail to add de banks, ret = %d\n", r);
+			} else {
+				if (amdgpu_cper_generate_ue_record(adev, bank))
+					dev_warn(adev->dev, "fail to generate ue cper records\n");
+			}
+		}
+
+		if (!list_empty(&de_banks.list)) {
+			if (amdgpu_cper_generate_ce_records(adev, &de_banks, de_banks.nr_banks))
+				dev_warn(adev->dev, "fail to generate de cper records\n");
 		}
+
+		aca_banks_release(&de_banks);
 	} else {
 		/*
 		 * SMU_TYPE_CE banks are combined into 1 CPER entries,
@@ -541,6 +558,10 @@ static int __aca_get_error_data(struct amdgpu_device *adev, struct aca_handle *h
 	if (ret)
 		return ret;
 
+	/* DEs may contain in CEs or UEs */
+	if (type != ACA_ERROR_TYPE_DEFERRED)
+		aca_log_aca_error(handle, ACA_ERROR_TYPE_DEFERRED, err_data);
+
 	return aca_log_aca_error(handle, type, err_data);
 }
 

drivers/gpu/drm/amd/amdgpu/amdgpu_aca.h

@@ -76,11 +76,17 @@ struct ras_query_context;
 #define mmSMNAID_XCD1_MCA_SMU		0x38430400	/* SMN AID XCD1 */
 #define mmSMNXCD_XCD0_MCA_SMU		0x40430400	/* SMN XCD XCD0 */
 
-#define ACA_BANK_ERR_CE_DE_DECODE(bank)                                  \
-	((ACA_REG__STATUS__POISON((bank)->regs[ACA_REG_IDX_STATUS]) ||   \
-	  ACA_REG__STATUS__DEFERRED((bank)->regs[ACA_REG_IDX_STATUS])) ? \
-		ACA_ERROR_TYPE_DEFERRED :                                \
-		ACA_ERROR_TYPE_CE)
+#define ACA_BANK_ERR_IS_DEFFERED(bank)                                \
+	(ACA_REG__STATUS__POISON((bank)->regs[ACA_REG_IDX_STATUS]) || \
+	 ACA_REG__STATUS__DEFERRED((bank)->regs[ACA_REG_IDX_STATUS]))
+
+#define ACA_BANK_ERR_CE_DE_DECODE(bank)                             \
+	(ACA_BANK_ERR_IS_DEFFERED(bank) ? ACA_ERROR_TYPE_DEFERRED : \
+					  ACA_ERROR_TYPE_CE)
+
+#define ACA_BANK_ERR_UE_DE_DECODE(bank)                             \
+	(ACA_BANK_ERR_IS_DEFFERED(bank) ? ACA_ERROR_TYPE_DEFERRED : \
+					  ACA_ERROR_TYPE_UE)
 
 enum aca_reg_idx {
 	ACA_REG_IDX_CTL			= 0,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c

@@ -189,7 +189,7 @@ const struct kfd2kgd_calls aldebaran_kfd2kgd = {
 	.set_address_watch = kgd_gfx_aldebaran_set_address_watch,
 	.clear_address_watch = kgd_gfx_v9_clear_address_watch,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
-	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
+	.build_dequeue_wait_counts_packet_info = kgd_gfx_v9_build_dequeue_wait_counts_packet_info,
 	.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings,
 	.hqd_get_pq_addr = kgd_gfx_v9_hqd_get_pq_addr,
 	.hqd_reset = kgd_gfx_v9_hqd_reset,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c

@@ -415,7 +415,7 @@ const struct kfd2kgd_calls arcturus_kfd2kgd = {
 	.set_address_watch = kgd_gfx_v9_set_address_watch,
 	.clear_address_watch = kgd_gfx_v9_clear_address_watch,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
-	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
+	.build_dequeue_wait_counts_packet_info = kgd_gfx_v9_build_dequeue_wait_counts_packet_info,
 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
 	.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings,
 	.hqd_get_pq_addr = kgd_gfx_v9_hqd_get_pq_addr,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gc_9_4_3.c

@@ -541,8 +541,8 @@ const struct kfd2kgd_calls gc_9_4_3_kfd2kgd = {
 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
 	.program_trap_handler_settings =
 				kgd_gfx_v9_program_trap_handler_settings,
-	.build_grace_period_packet_info =
-				kgd_gfx_v9_build_grace_period_packet_info,
+	.build_dequeue_wait_counts_packet_info =
+				kgd_gfx_v9_build_dequeue_wait_counts_packet_info,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
 	.enable_debug_trap = kgd_aldebaran_enable_debug_trap,
 	.disable_debug_trap = kgd_gfx_v9_4_3_disable_debug_trap,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c

@@ -1021,25 +1021,25 @@ void kgd_gfx_v10_get_iq_wait_times(struct amdgpu_device *adev,
 	*wait_times = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2));
 }
 
-void kgd_gfx_v10_build_grace_period_packet_info(struct amdgpu_device *adev,
+void kgd_gfx_v10_build_dequeue_wait_counts_packet_info(struct amdgpu_device *adev,
 						uint32_t wait_times,
-						uint32_t grace_period,
+						uint32_t sch_wave,
+						uint32_t que_sleep,
 						uint32_t *reg_offset,
 						uint32_t *reg_data)
 {
 	*reg_data = wait_times;
 
-	/*
-	 * The CP cannont handle a 0 grace period input and will result in
-	 * an infinite grace period being set so set to 1 to prevent this.
-	 */
-	if (grace_period == 0)
-		grace_period = 1;
-
-	*reg_data = REG_SET_FIELD(*reg_data,
-			CP_IQ_WAIT_TIME2,
-			SCH_WAVE,
-			grace_period);
+	if (sch_wave)
+		*reg_data = REG_SET_FIELD(*reg_data,
+				CP_IQ_WAIT_TIME2,
+				SCH_WAVE,
+				sch_wave);
+	if (que_sleep)
+		*reg_data = REG_SET_FIELD(*reg_data,
+				CP_IQ_WAIT_TIME2,
+				QUE_SLEEP,
+				que_sleep);
 
 	*reg_offset = SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2);
 }
@@ -1115,7 +1115,7 @@ const struct kfd2kgd_calls gfx_v10_kfd2kgd = {
 	.set_address_watch = kgd_gfx_v10_set_address_watch,
 	.clear_address_watch = kgd_gfx_v10_clear_address_watch,
 	.get_iq_wait_times = kgd_gfx_v10_get_iq_wait_times,
-	.build_grace_period_packet_info = kgd_gfx_v10_build_grace_period_packet_info,
+	.build_dequeue_wait_counts_packet_info = kgd_gfx_v10_build_dequeue_wait_counts_packet_info,
 	.program_trap_handler_settings = program_trap_handler_settings,
 	.hqd_get_pq_addr = kgd_gfx_v10_hqd_get_pq_addr,
 	.hqd_reset = kgd_gfx_v10_hqd_reset,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h

@@ -51,9 +51,10 @@ uint32_t kgd_gfx_v10_clear_address_watch(struct amdgpu_device *adev,
 void kgd_gfx_v10_get_iq_wait_times(struct amdgpu_device *adev,
 				uint32_t *wait_times,
 				uint32_t inst);
-void kgd_gfx_v10_build_grace_period_packet_info(struct amdgpu_device *adev,
+void kgd_gfx_v10_build_dequeue_wait_counts_packet_info(struct amdgpu_device *adev,
 					       uint32_t wait_times,
-					       uint32_t grace_period,
+					       uint32_t sch_wave,
+					       uint32_t que_sleep,
 					       uint32_t *reg_offset,
 					       uint32_t *reg_data);
 uint64_t kgd_gfx_v10_hqd_get_pq_addr(struct amdgpu_device *adev,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c

@@ -673,7 +673,7 @@ const struct kfd2kgd_calls gfx_v10_3_kfd2kgd = {
 	.set_vm_context_page_table_base = set_vm_context_page_table_base_v10_3,
 	.program_trap_handler_settings = program_trap_handler_settings_v10_3,
 	.get_iq_wait_times = kgd_gfx_v10_get_iq_wait_times,
-	.build_grace_period_packet_info = kgd_gfx_v10_build_grace_period_packet_info,
+	.build_dequeue_wait_counts_packet_info = kgd_gfx_v10_build_dequeue_wait_counts_packet_info,
 	.enable_debug_trap = kgd_gfx_v10_enable_debug_trap,
 	.disable_debug_trap = kgd_gfx_v10_disable_debug_trap,
 	.validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c

@@ -1077,25 +1077,25 @@ void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev,
 				adev->gfx.cu_info.max_waves_per_simd;
 }
 
-void kgd_gfx_v9_build_grace_period_packet_info(struct amdgpu_device *adev,
+void kgd_gfx_v9_build_dequeue_wait_counts_packet_info(struct amdgpu_device *adev,
 		uint32_t wait_times,
-		uint32_t grace_period,
+		uint32_t sch_wave,
+		uint32_t que_sleep,
 		uint32_t *reg_offset,
 		uint32_t *reg_data)
 {
 	*reg_data = wait_times;
 
-	/*
-	 * The CP cannot handle a 0 grace period input and will result in
-	 * an infinite grace period being set so set to 1 to prevent this.
-	 */
-	if (grace_period == 0)
-		grace_period = 1;
-
-	*reg_data = REG_SET_FIELD(*reg_data,
-			CP_IQ_WAIT_TIME2,
-			SCH_WAVE,
-			grace_period);
+	if (sch_wave)
+		*reg_data = REG_SET_FIELD(*reg_data,
+				CP_IQ_WAIT_TIME2,
+				SCH_WAVE,
+				sch_wave);
+	if (que_sleep)
+		*reg_data = REG_SET_FIELD(*reg_data,
+				CP_IQ_WAIT_TIME2,
+				QUE_SLEEP,
+				que_sleep);
 
 	*reg_offset = SOC15_REG_OFFSET(GC, 0, mmCP_IQ_WAIT_TIME2);
 }
@@ -1255,7 +1255,7 @@ const struct kfd2kgd_calls gfx_v9_kfd2kgd = {
 	.set_address_watch = kgd_gfx_v9_set_address_watch,
 	.clear_address_watch = kgd_gfx_v9_clear_address_watch,
 	.get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times,
-	.build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info,
+	.build_dequeue_wait_counts_packet_info = kgd_gfx_v9_build_dequeue_wait_counts_packet_info,
 	.get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy,
 	.program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings,
 	.hqd_get_pq_addr = kgd_gfx_v9_hqd_get_pq_addr,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h

@@ -97,9 +97,10 @@ uint32_t kgd_gfx_v9_clear_address_watch(struct amdgpu_device *adev,
 void kgd_gfx_v9_get_iq_wait_times(struct amdgpu_device *adev,
 				uint32_t *wait_times,
 				uint32_t inst);
-void kgd_gfx_v9_build_grace_period_packet_info(struct amdgpu_device *adev,
+void kgd_gfx_v9_build_dequeue_wait_counts_packet_info(struct amdgpu_device *adev,
 					       uint32_t wait_times,
-					       uint32_t grace_period,
+					       uint32_t sch_wave,
+					       uint32_t que_sleep,
 					       uint32_t *reg_offset,
 					       uint32_t *reg_data);
 uint64_t kgd_gfx_v9_hqd_get_pq_addr(struct amdgpu_device *adev,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c

@@ -491,7 +491,7 @@ static int vm_update_pds(struct amdgpu_vm *vm, struct amdgpu_sync *sync)
 	if (ret)
 		return ret;
 
-	return amdgpu_sync_fence(sync, vm->last_update);
+	return amdgpu_sync_fence(sync, vm->last_update, GFP_KERNEL);
 }
 
 static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem)
@@ -1249,7 +1249,7 @@ static int unmap_bo_from_gpuvm(struct kgd_mem *mem,
 
 	(void)amdgpu_vm_clear_freed(adev, vm, &bo_va->last_pt_update);
 
-	(void)amdgpu_sync_fence(sync, bo_va->last_pt_update);
+	(void)amdgpu_sync_fence(sync, bo_va->last_pt_update, GFP_KERNEL);
 
 	return 0;
 }
@@ -1273,7 +1273,7 @@ static int update_gpuvm_pte(struct kgd_mem *mem,
 		return ret;
 	}
 
-	return amdgpu_sync_fence(sync, bo_va->last_pt_update);
+	return amdgpu_sync_fence(sync, bo_va->last_pt_update, GFP_KERNEL);
 }
 
 static int map_bo_to_gpuvm(struct kgd_mem *mem,
@@ -2913,7 +2913,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence __rcu *
 		}
 		dma_resv_for_each_fence(&cursor, bo->tbo.base.resv,
 					DMA_RESV_USAGE_KERNEL, fence) {
-			ret = amdgpu_sync_fence(&sync_obj, fence);
+			ret = amdgpu_sync_fence(&sync_obj, fence, GFP_KERNEL);
 			if (ret) {
 				pr_debug("Memory eviction: Sync BO fence failed. Try again\n");
 				goto validate_map_fail;

drivers/gpu/drm/amd/amdgpu/amdgpu_cper.c

@@ -455,10 +455,10 @@ static u32 amdgpu_cper_ring_get_ent_sz(struct amdgpu_ring *ring, u64 pos)
 		return umin(rec_len, chunk);
 }
 
-void amdgpu_cper_ring_write(struct amdgpu_ring *ring,
-					      void *src, int count)
+void amdgpu_cper_ring_write(struct amdgpu_ring *ring, void *src, int count)
 {
 	u64 pos, wptr_old, rptr = *ring->rptr_cpu_addr & ring->ptr_mask;
+	int rec_cnt_dw = count >> 2;
 	u32 chunk, ent_sz;
 	u8 *s = (u8 *)src;
 
@@ -485,6 +485,9 @@ void amdgpu_cper_ring_write(struct amdgpu_ring *ring,
 		s += chunk;
 	}
 
+	if (ring->count_dw < rec_cnt_dw)
+		ring->count_dw = 0;
+
 	/* the buffer is overflow, adjust rptr */
 	if (((wptr_old < rptr) && (rptr <= ring->wptr)) ||
 	    ((ring->wptr < wptr_old) && (wptr_old < rptr)) ||
@@ -501,12 +504,10 @@ void amdgpu_cper_ring_write(struct amdgpu_ring *ring,
 			pos = rptr;
 		} while (!amdgpu_cper_is_hdr(ring, rptr));
 	}
-	mutex_unlock(&ring->adev->cper.ring_lock);
 
-	if (ring->count_dw >= (count >> 2))
-		ring->count_dw -= (count >> 2);
-	else
-		ring->count_dw = 0;
+	if (ring->count_dw >= rec_cnt_dw)
+		ring->count_dw -= rec_cnt_dw;
+	mutex_unlock(&ring->adev->cper.ring_lock);
 }
 
 static u64 amdgpu_cper_ring_get_rptr(struct amdgpu_ring *ring)

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c

@@ -428,7 +428,7 @@ static int amdgpu_cs_p2_dependencies(struct amdgpu_cs_parser *p,
 			dma_fence_put(old);
 		}
 
-		r = amdgpu_sync_fence(&p->sync, fence);
+		r = amdgpu_sync_fence(&p->sync, fence, GFP_KERNEL);
 		dma_fence_put(fence);
 		if (r)
 			return r;
@@ -450,7 +450,7 @@ static int amdgpu_syncobj_lookup_and_add(struct amdgpu_cs_parser *p,
 		return r;
 	}
 
-	r = amdgpu_sync_fence(&p->sync, fence);
+	r = amdgpu_sync_fence(&p->sync, fence, GFP_KERNEL);
 	dma_fence_put(fence);
 	return r;
 }
@@ -1111,7 +1111,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 			struct drm_gpu_scheduler *sched = entity->rq->sched;
 			struct amdgpu_ring *ring = to_amdgpu_ring(sched);
 
-			if (amdgpu_vmid_uses_reserved(adev, vm, ring->vm_hub))
+			if (amdgpu_vmid_uses_reserved(vm, ring->vm_hub))
 				return -EINVAL;
 		}
 	}
@@ -1124,7 +1124,8 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 	if (r)
 		return r;
 
-	r = amdgpu_sync_fence(&p->sync, fpriv->prt_va->last_pt_update);
+	r = amdgpu_sync_fence(&p->sync, fpriv->prt_va->last_pt_update,
+			      GFP_KERNEL);
 	if (r)
 		return r;
 
@@ -1135,7 +1136,8 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 		if (r)
 			return r;
 
-		r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update);
+		r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update,
+				      GFP_KERNEL);
 		if (r)
 			return r;
 	}
@@ -1154,7 +1156,8 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 		if (r)
 			return r;
 
-		r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update);
+		r = amdgpu_sync_fence(&p->sync, bo_va->last_pt_update,
+				      GFP_KERNEL);
 		if (r)
 			return r;
 	}
@@ -1167,7 +1170,7 @@ static int amdgpu_cs_vm_handling(struct amdgpu_cs_parser *p)
 	if (r)
 		return r;
 
-	r = amdgpu_sync_fence(&p->sync, vm->last_update);
+	r = amdgpu_sync_fence(&p->sync, vm->last_update, GFP_KERNEL);
 	if (r)
 		return r;
 
@@ -1248,7 +1251,8 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 			continue;
 		}
 
-		r = amdgpu_sync_fence(&p->gang_leader->explicit_sync, fence);
+		r = amdgpu_sync_fence(&p->gang_leader->explicit_sync, fence,
+				      GFP_KERNEL);
 		dma_fence_put(fence);
 		if (r)
 			return r;

drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c

@@ -1990,7 +1990,7 @@ static int amdgpu_debugfs_sclk_set(void *data, u64 val)
 	uint32_t max_freq, min_freq;
 	struct amdgpu_device *adev = (struct amdgpu_device *)data;
 
-	if (amdgpu_sriov_vf(adev) && !amdgpu_sriov_is_pp_one_vf(adev))
+	if (amdgpu_sriov_multi_vf_mode(adev))
 		return -EINVAL;
 
 	ret = pm_runtime_get_sync(adev_to_drm(adev)->dev);

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c

@@ -227,6 +227,24 @@ static ssize_t amdgpu_device_get_pcie_replay_count(struct device *dev,
 static DEVICE_ATTR(pcie_replay_count, 0444,
 		amdgpu_device_get_pcie_replay_count, NULL);
 
+static int amdgpu_device_attr_sysfs_init(struct amdgpu_device *adev)
+{
+	int ret = 0;
+
+	if (!amdgpu_sriov_vf(adev))
+		ret = sysfs_create_file(&adev->dev->kobj,
+					&dev_attr_pcie_replay_count.attr);
+
+	return ret;
+}
+
+static void amdgpu_device_attr_sysfs_fini(struct amdgpu_device *adev)
+{
+	if (!amdgpu_sriov_vf(adev))
+		sysfs_remove_file(&adev->dev->kobj,
+				  &dev_attr_pcie_replay_count.attr);
+}
+
 static ssize_t amdgpu_sysfs_reg_state_get(struct file *f, struct kobject *kobj,
 					  struct bin_attribute *attr, char *buf,
 					  loff_t ppos, size_t count)
@@ -2757,6 +2775,9 @@ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
 	if (!total)
 		return -ENODEV;
 
+	if (adev->gmc.xgmi.supported)
+		amdgpu_xgmi_early_init(adev);
+
 	ip_block = amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_GFX);
 	if (ip_block->status.valid != false)
 		amdgpu_amdkfd_device_probe(adev);
@@ -4169,11 +4190,6 @@ static bool amdgpu_device_check_iommu_remap(struct amdgpu_device *adev)
 }
 #endif
 
-static const struct attribute *amdgpu_dev_attributes[] = {
-	&dev_attr_pcie_replay_count.attr,
-	NULL
-};
-
 static void amdgpu_device_set_mcbp(struct amdgpu_device *adev)
 {
 	if (amdgpu_mcbp == 1)
@@ -4278,7 +4294,14 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	mutex_init(&adev->gfx.reset_sem_mutex);
 	/* Initialize the mutex for cleaner shader isolation between GFX and compute processes */
 	mutex_init(&adev->enforce_isolation_mutex);
+	for (i = 0; i < MAX_XCP; ++i) {
+		adev->isolation[i].spearhead = dma_fence_get_stub();
+		amdgpu_sync_create(&adev->isolation[i].active);
+		amdgpu_sync_create(&adev->isolation[i].prev);
+	}
 	mutex_init(&adev->gfx.kfd_sch_mutex);
+	mutex_init(&adev->gfx.workload_profile_mutex);
+	mutex_init(&adev->vcn.workload_profile_mutex);
 
 	amdgpu_device_init_apu_flags(adev);
 
@@ -4396,10 +4419,17 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	if (r)
 		return r;
 
-	/* Get rid of things like offb */
-	r = aperture_remove_conflicting_pci_devices(adev->pdev, amdgpu_kms_driver.name);
-	if (r)
-		return r;
+	/*
+	 * No need to remove conflicting FBs for non-display class devices.
+	 * This prevents the sysfb from being freed accidently.
+	 */
+	if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA ||
+	    (pdev->class >> 8) == PCI_CLASS_DISPLAY_OTHER) {
+		/* Get rid of things like offb */
+		r = aperture_remove_conflicting_pci_devices(adev->pdev, amdgpu_kms_driver.name);
+		if (r)
+			return r;
+	}
 
 	/* Enable TMZ based on IP_VERSION */
 	amdgpu_gmc_tmz_set(adev);
@@ -4610,7 +4640,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	} else
 		adev->ucode_sysfs_en = true;
 
-	r = sysfs_create_files(&adev->dev->kobj, amdgpu_dev_attributes);
+	r = amdgpu_device_attr_sysfs_init(adev);
 	if (r)
 		dev_err(adev->dev, "Could not create amdgpu device attr\n");
 
@@ -4747,7 +4777,7 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev)
 		amdgpu_pm_sysfs_fini(adev);
 	if (adev->ucode_sysfs_en)
 		amdgpu_ucode_sysfs_fini(adev);
-	sysfs_remove_files(&adev->dev->kobj, amdgpu_dev_attributes);
+	amdgpu_device_attr_sysfs_fini(adev);
 	amdgpu_fru_sysfs_fini(adev);
 
 	amdgpu_reg_state_sysfs_fini(adev);
@@ -4774,7 +4804,7 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev)
 
 void amdgpu_device_fini_sw(struct amdgpu_device *adev)
 {
-	int idx;
+	int i, idx;
 	bool px;
 
 	amdgpu_device_ip_fini(adev);
@@ -4782,6 +4812,11 @@ void amdgpu_device_fini_sw(struct amdgpu_device *adev)
 	amdgpu_ucode_release(&adev->firmware.gpu_info_fw);
 	adev->accel_working = false;
 	dma_fence_put(rcu_dereference_protected(adev->gang_submit, true));
+	for (i = 0; i < MAX_XCP; ++i) {
+		dma_fence_put(adev->isolation[i].spearhead);
+		amdgpu_sync_free(&adev->isolation[i].active);
+		amdgpu_sync_free(&adev->isolation[i].prev);
+	}
 
 	amdgpu_reset_fini(adev);
 
@@ -4797,6 +4832,9 @@ void amdgpu_device_fini_sw(struct amdgpu_device *adev)
 	kfree(adev->fru_info);
 	adev->fru_info = NULL;
 
+	kfree(adev->xcp_mgr);
+	adev->xcp_mgr = NULL;
+
 	px = amdgpu_device_supports_px(adev_to_drm(adev));
 
 	if (px || (!dev_is_removable(&adev->pdev->dev) &&
@@ -5328,6 +5366,7 @@ static int amdgpu_device_reset_sriov(struct amdgpu_device *adev,
 	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 2) ||
 	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
 	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4) ||
+	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 5, 0) ||
 	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(11, 0, 3))
 		amdgpu_ras_resume(adev);
 
@@ -6900,22 +6939,117 @@ struct dma_fence *amdgpu_device_switch_gang(struct amdgpu_device *adev,
 {
 	struct dma_fence *old = NULL;
 
+	dma_fence_get(gang);
 	do {
 		dma_fence_put(old);
 		old = amdgpu_device_get_gang(adev);
 		if (old == gang)
 			break;
 
-		if (!dma_fence_is_signaled(old))
+		if (!dma_fence_is_signaled(old)) {
+			dma_fence_put(gang);
 			return old;
+		}
 
 	} while (cmpxchg((struct dma_fence __force **)&adev->gang_submit,
 			 old, gang) != old);
 
+	/*
+	 * Drop it once for the exchanged reference in adev and once for the
+	 * thread local reference acquired in amdgpu_device_get_gang().
+	 */
+	dma_fence_put(old);
 	dma_fence_put(old);
 	return NULL;
 }
 
+/**
+ * amdgpu_device_enforce_isolation - enforce HW isolation
+ * @adev: the amdgpu device pointer
+ * @ring: the HW ring the job is supposed to run on
+ * @job: the job which is about to be pushed to the HW ring
+ *
+ * Makes sure that only one client at a time can use the GFX block.
+ * Returns: The dependency to wait on before the job can be pushed to the HW.
+ * The function is called multiple times until NULL is returned.
+ */
+struct dma_fence *amdgpu_device_enforce_isolation(struct amdgpu_device *adev,
+						  struct amdgpu_ring *ring,
+						  struct amdgpu_job *job)
+{
+	struct amdgpu_isolation *isolation = &adev->isolation[ring->xcp_id];
+	struct drm_sched_fence *f = job->base.s_fence;
+	struct dma_fence *dep;
+	void *owner;
+	int r;
+
+	/*
+	 * For now enforce isolation only for the GFX block since we only need
+	 * the cleaner shader on those rings.
+	 */
+	if (ring->funcs->type != AMDGPU_RING_TYPE_GFX &&
+	    ring->funcs->type != AMDGPU_RING_TYPE_COMPUTE)
+		return NULL;
+
+	/*
+	 * All submissions where enforce isolation is false are handled as if
+	 * they come from a single client. Use ~0l as the owner to distinct it
+	 * from kernel submissions where the owner is NULL.
+	 */
+	owner = job->enforce_isolation ? f->owner : (void *)~0l;
+
+	mutex_lock(&adev->enforce_isolation_mutex);
+
+	/*
+	 * The "spearhead" submission is the first one which changes the
+	 * ownership to its client. We always need to wait for it to be
+	 * pushed to the HW before proceeding with anything.
+	 */
+	if (&f->scheduled != isolation->spearhead &&
+	    !dma_fence_is_signaled(isolation->spearhead)) {
+		dep = isolation->spearhead;
+		goto out_grab_ref;
+	}
+
+	if (isolation->owner != owner) {
+
+		/*
+		 * Wait for any gang to be assembled before switching to a
+		 * different owner or otherwise we could deadlock the
+		 * submissions.
+		 */
+		if (!job->gang_submit) {
+			dep = amdgpu_device_get_gang(adev);
+			if (!dma_fence_is_signaled(dep))
+				goto out_return_dep;
+			dma_fence_put(dep);
+		}
+
+		dma_fence_put(isolation->spearhead);
+		isolation->spearhead = dma_fence_get(&f->scheduled);
+		amdgpu_sync_move(&isolation->active, &isolation->prev);
+		trace_amdgpu_isolation(isolation->owner, owner);
+		isolation->owner = owner;
+	}
+
+	/*
+	 * Specifying the ring here helps to pipeline submissions even when
+	 * isolation is enabled. If that is not desired for testing NULL can be
+	 * used instead of the ring to enforce a CPU round trip while switching
+	 * between clients.
+	 */
+	dep = amdgpu_sync_peek_fence(&isolation->prev, ring);
+	r = amdgpu_sync_fence(&isolation->active, &f->finished, GFP_NOWAIT);
+	if (r)
+		DRM_WARN("OOM tracking isolation\n");
+
+out_grab_ref:
+	dma_fence_get(dep);
+out_return_dep:
+	mutex_unlock(&adev->enforce_isolation_mutex);
+	return dep;
+}
+
 bool amdgpu_device_has_display_hardware(struct amdgpu_device *adev)
 {
 	switch (adev->asic_type) {

drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c

@@ -113,8 +113,13 @@
 #include "amdgpu_isp.h"
 #endif
 
-#define FIRMWARE_IP_DISCOVERY "amdgpu/ip_discovery.bin"
-MODULE_FIRMWARE(FIRMWARE_IP_DISCOVERY);
+MODULE_FIRMWARE("amdgpu/ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/vega10_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/vega12_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/vega20_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/raven_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/raven2_ip_discovery.bin");
+MODULE_FIRMWARE("amdgpu/picasso_ip_discovery.bin");
 
 #define mmIP_DISCOVERY_VERSION  0x16A00
 #define mmRCC_CONFIG_MEMSIZE	0xde3
@@ -297,21 +302,13 @@ static int amdgpu_discovery_read_binary_from_mem(struct amdgpu_device *adev,
 	return ret;
 }
 
-static int amdgpu_discovery_read_binary_from_file(struct amdgpu_device *adev, uint8_t *binary)
+static int amdgpu_discovery_read_binary_from_file(struct amdgpu_device *adev,
+							uint8_t *binary,
+							const char *fw_name)
 {
 	const struct firmware *fw;
-	const char *fw_name;
 	int r;
 
-	switch (amdgpu_discovery) {
-	case 2:
-		fw_name = FIRMWARE_IP_DISCOVERY;
-		break;
-	default:
-		dev_warn(adev->dev, "amdgpu_discovery is not set properly\n");
-		return -EINVAL;
-	}
-
 	r = request_firmware(&fw, fw_name, adev->dev);
 	if (r) {
 		dev_err(adev->dev, "can't load firmware \"%s\"\n",
@@ -404,10 +401,39 @@ static int amdgpu_discovery_verify_npsinfo(struct amdgpu_device *adev,
 	return 0;
 }
 
+static const char *amdgpu_discovery_get_fw_name(struct amdgpu_device *adev)
+{
+	if (amdgpu_discovery == 2)
+		return "amdgpu/ip_discovery.bin";
+
+	switch (adev->asic_type) {
+	case CHIP_VEGA10:
+		return "amdgpu/vega10_ip_discovery.bin";
+	case CHIP_VEGA12:
+		return "amdgpu/vega12_ip_discovery.bin";
+	case CHIP_RAVEN:
+		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
+			return "amdgpu/raven2_ip_discovery.bin";
+		else if (adev->apu_flags & AMD_APU_IS_PICASSO)
+			return "amdgpu/picasso_ip_discovery.bin";
+		else
+			return "amdgpu/raven_ip_discovery.bin";
+	case CHIP_VEGA20:
+		return "amdgpu/vega20_ip_discovery.bin";
+	case CHIP_ARCTURUS:
+		return "amdgpu/arcturus_ip_discovery.bin";
+	case CHIP_ALDEBARAN:
+		return "amdgpu/aldebaran_ip_discovery.bin";
+	default:
+		return NULL;
+	}
+}
+
 static int amdgpu_discovery_init(struct amdgpu_device *adev)
 {
 	struct table_info *info;
 	struct binary_header *bhdr;
+	const char *fw_name;
 	uint16_t offset;
 	uint16_t size;
 	uint16_t checksum;
@@ -419,9 +445,10 @@ static int amdgpu_discovery_init(struct amdgpu_device *adev)
 		return -ENOMEM;
 
 	/* Read from file if it is the preferred option */
-	if (amdgpu_discovery == 2) {
+	fw_name = amdgpu_discovery_get_fw_name(adev);
+	if (fw_name != NULL) {
 		dev_info(adev->dev, "use ip discovery information from file");
-		r = amdgpu_discovery_read_binary_from_file(adev, adev->mman.discovery_bin);
+		r = amdgpu_discovery_read_binary_from_file(adev, adev->mman.discovery_bin, fw_name);
 
 		if (r) {
 			dev_err(adev->dev, "failed to read ip discovery binary from file\n");
@@ -1290,6 +1317,7 @@ static int amdgpu_discovery_reg_base_init(struct amdgpu_device *adev)
 	uint16_t die_offset;
 	uint16_t ip_offset;
 	uint16_t num_dies;
+	uint32_t wafl_ver;
 	uint16_t num_ips;
 	uint16_t hw_id;
 	uint8_t inst;
@@ -1303,6 +1331,7 @@ static int amdgpu_discovery_reg_base_init(struct amdgpu_device *adev)
 		return r;
 	}
 
+	wafl_ver = 0;
 	adev->gfx.xcc_mask = 0;
 	adev->sdma.sdma_mask = 0;
 	adev->vcn.inst_mask = 0;
@@ -1403,6 +1432,10 @@ static int amdgpu_discovery_reg_base_init(struct amdgpu_device *adev)
 				adev->gfx.xcc_mask |=
 					(1U << ip->instance_number);
 
+			if (!wafl_ver && le16_to_cpu(ip->hw_id) == WAFLC_HWID)
+				wafl_ver = IP_VERSION_FULL(ip->major, ip->minor,
+							   ip->revision, 0, 0);
+
 			for (k = 0; k < num_base_address; k++) {
 				/*
 				 * convert the endianness of base addresses in place,
@@ -1468,6 +1501,9 @@ static int amdgpu_discovery_reg_base_init(struct amdgpu_device *adev)
 		}
 	}
 
+	if (wafl_ver && !adev->ip_versions[XGMI_HWIP][0])
+		adev->ip_versions[XGMI_HWIP][0] = wafl_ver;
+
 	return 0;
 }
 
@@ -2509,6 +2545,38 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev)
 {
 	int r;
 
+	switch (adev->asic_type) {
+	case CHIP_VEGA10:
+	case CHIP_VEGA12:
+	case CHIP_RAVEN:
+	case CHIP_VEGA20:
+	case CHIP_ARCTURUS:
+	case CHIP_ALDEBARAN:
+		/* this is not fatal.  We have a fallback below
+		 * if the new firmwares are not present. some of
+		 * this will be overridden below to keep things
+		 * consistent with the current behavior.
+		 */
+		r = amdgpu_discovery_reg_base_init(adev);
+		if (!r) {
+			amdgpu_discovery_harvest_ip(adev);
+			amdgpu_discovery_get_gfx_info(adev);
+			amdgpu_discovery_get_mall_info(adev);
+			amdgpu_discovery_get_vcn_info(adev);
+		}
+		break;
+	default:
+		r = amdgpu_discovery_reg_base_init(adev);
+		if (r)
+			return -EINVAL;
+
+		amdgpu_discovery_harvest_ip(adev);
+		amdgpu_discovery_get_gfx_info(adev);
+		amdgpu_discovery_get_mall_info(adev);
+		amdgpu_discovery_get_vcn_info(adev);
+		break;
+	}
+
 	switch (adev->asic_type) {
 	case CHIP_VEGA10:
 		vega10_reg_base_init(adev);
@@ -2673,14 +2741,6 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev)
 		adev->ip_versions[XGMI_HWIP][0] = IP_VERSION(6, 1, 0);
 		break;
 	default:
-		r = amdgpu_discovery_reg_base_init(adev);
-		if (r)
-			return -EINVAL;
-
-		amdgpu_discovery_harvest_ip(adev);
-		amdgpu_discovery_get_gfx_info(adev);
-		amdgpu_discovery_get_mall_info(adev);
-		amdgpu_discovery_get_vcn_info(adev);
 		break;
 	}
 
@@ -2772,13 +2832,6 @@ int amdgpu_discovery_set_ip_blocks(struct amdgpu_device *adev)
 		break;
 	}
 
-	if (amdgpu_ip_version(adev, XGMI_HWIP, 0) == IP_VERSION(4, 8, 0))
-		adev->gmc.xgmi.supported = true;
-
-	if (amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) ||
-	    amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4))
-		adev->ip_versions[XGMI_HWIP][0] = IP_VERSION(6, 4, 0);
-
 	/* set NBIO version */
 	switch (amdgpu_ip_version(adev, NBIO_HWIP, 0)) {
 	case IP_VERSION(6, 1, 0):

drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c

@@ -122,9 +122,10 @@
  * - 3.60.0 - Add AMDGPU_TILING_GFX12_DCC_WRITE_COMPRESS_DISABLE (Vulkan requirement)
  * - 3.61.0 - Contains fix for RV/PCO compute queues
  * - 3.62.0 - Add AMDGPU_IDS_FLAGS_MODE_PF, AMDGPU_IDS_FLAGS_MODE_VF & AMDGPU_IDS_FLAGS_MODE_PT
+ * - 3.63.0 - GFX12 display DCC supports 256B max compressed block size
  */
 #define KMS_DRIVER_MAJOR	3
-#define KMS_DRIVER_MINOR	62
+#define KMS_DRIVER_MINOR	63
 #define KMS_DRIVER_PATCHLEVEL	0
 
 /*
@@ -138,6 +139,7 @@ enum AMDGPU_DEBUG_MASK {
 	AMDGPU_DEBUG_ENABLE_RAS_ACA = BIT(4),
 	AMDGPU_DEBUG_ENABLE_EXP_RESETS = BIT(5),
 	AMDGPU_DEBUG_DISABLE_GPU_RING_RESET = BIT(6),
+	AMDGPU_DEBUG_SMU_POOL = BIT(7),
 };
 
 unsigned int amdgpu_vram_limit = UINT_MAX;
@@ -175,6 +177,7 @@ uint amdgpu_sdma_phase_quantum = 32;
 char *amdgpu_disable_cu;
 char *amdgpu_virtual_display;
 bool enforce_isolation;
+int amdgpu_modeset = -1;
 
 /* Specifies the default granularity for SVM, used in buffer
  * migration and restoration of backing memory when handling
@@ -1036,6 +1039,13 @@ module_param_named(user_partt_mode, amdgpu_user_partt_mode, uint, 0444);
 module_param(enforce_isolation, bool, 0444);
 MODULE_PARM_DESC(enforce_isolation, "enforce process isolation between graphics and compute . enforce_isolation = on");
 
+/**
+ * DOC: modeset (int)
+ * Override nomodeset (1 = override, -1 = auto). The default is -1 (auto).
+ */
+MODULE_PARM_DESC(modeset, "Override nomodeset (1 = enable, -1 = auto)");
+module_param_named(modeset, amdgpu_modeset, int, 0444);
+
 /**
  * DOC: seamless (int)
  * Seamless boot will keep the image on the screen during the boot process.
@@ -1052,6 +1062,11 @@ module_param_named(seamless, amdgpu_seamless, int, 0444);
  *   limits the VRAM size reported to ROCm applications to the visible
  *   size, usually 256MB.
  * - 0x4: Disable GPU soft recovery, always do a full reset
+ * - 0x8: Use VRAM for firmware loading
+ * - 0x10: Enable ACA based RAS logging
+ * - 0x20: Enable experimental resets
+ * - 0x40: Disable ring resets
+ * - 0x80: Use VRAM for SMU pool
  */
 MODULE_PARM_DESC(debug_mask, "debug options for amdgpu, disabled by default");
 module_param_named_unsafe(debug_mask, amdgpu_debug_mask, uint, 0444);
@@ -2229,6 +2244,10 @@ static void amdgpu_init_debug_options(struct amdgpu_device *adev)
 		pr_info("debug: ring reset disabled\n");
 		adev->debug_disable_gpu_ring_reset = true;
 	}
+	if (amdgpu_debug_mask & AMDGPU_DEBUG_SMU_POOL) {
+		pr_info("debug: use vram for smu pool\n");
+		adev->pm.smu_debug_mask |= SMU_DEBUG_POOL_USE_VRAM;
+	}
 }
 
 static unsigned long amdgpu_fix_asic_type(struct pci_dev *pdev, unsigned long flags)
@@ -2256,6 +2275,12 @@ static int amdgpu_pci_probe(struct pci_dev *pdev,
 	int ret, retry = 0, i;
 	bool supports_atomic = false;
 
+	if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA ||
+	    (pdev->class >> 8) == PCI_CLASS_DISPLAY_OTHER) {
+		if (drm_firmware_drivers_only() && amdgpu_modeset == -1)
+			return -EINVAL;
+	}
+
 	/* skip devices which are owned by radeon */
 	for (i = 0; i < ARRAY_SIZE(amdgpu_unsupported_pciidlist); i++) {
 		if (amdgpu_unsupported_pciidlist[i] == pdev->device)
@@ -2989,9 +3014,6 @@ static int __init amdgpu_init(void)
 {
 	int r;
 
-	if (drm_firmware_drivers_only())
-		return -EINVAL;
-
 	r = amdgpu_sync_init();
 	if (r)
 		goto error_sync;

drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c

@@ -1665,15 +1665,8 @@ static ssize_t amdgpu_gfx_set_enforce_isolation(struct device *dev,
 	}
 
 	mutex_lock(&adev->enforce_isolation_mutex);
-	for (i = 0; i < num_partitions; i++) {
-		if (adev->enforce_isolation[i] && !partition_values[i])
-			/* Going from enabled to disabled */
-			amdgpu_vmid_free_reserved(adev, AMDGPU_GFXHUB(i));
-		else if (!adev->enforce_isolation[i] && partition_values[i])
-			/* Going from disabled to enabled */
-			amdgpu_vmid_alloc_reserved(adev, AMDGPU_GFXHUB(i));
+	for (i = 0; i < num_partitions; i++)
 		adev->enforce_isolation[i] = partition_values[i];
-	}
 	mutex_unlock(&adev->enforce_isolation_mutex);
 
 	amdgpu_mes_update_enforce_isolation(adev);
@@ -2002,8 +1995,8 @@ void amdgpu_gfx_enforce_isolation_handler(struct work_struct *work)
 		if (adev->kfd.init_complete) {
 			WARN_ON_ONCE(!adev->gfx.kfd_sch_inactive[idx]);
 			WARN_ON_ONCE(adev->gfx.kfd_sch_req_count[idx]);
-				amdgpu_amdkfd_start_sched(adev, idx);
-				adev->gfx.kfd_sch_inactive[idx] = false;
+			amdgpu_amdkfd_start_sched(adev, idx);
+			adev->gfx.kfd_sch_inactive[idx] = false;
 		}
 	}
 	mutex_unlock(&adev->enforce_isolation_mutex);
@@ -2160,11 +2153,16 @@ void amdgpu_gfx_profile_idle_work_handler(struct work_struct *work)
 	for (i = 0; i < (AMDGPU_MAX_COMPUTE_RINGS * AMDGPU_MAX_GC_INSTANCES); ++i)
 		fences += amdgpu_fence_count_emitted(&adev->gfx.compute_ring[i]);
 	if (!fences && !atomic_read(&adev->gfx.total_submission_cnt)) {
-		r = amdgpu_dpm_switch_power_profile(adev, profile, false);
-		if (r)
-			dev_warn(adev->dev, "(%d) failed to disable %s power profile mode\n", r,
-				 profile == PP_SMC_POWER_PROFILE_FULLSCREEN3D ?
-				 "fullscreen 3D" : "compute");
+		mutex_lock(&adev->gfx.workload_profile_mutex);
+		if (adev->gfx.workload_profile_active) {
+			r = amdgpu_dpm_switch_power_profile(adev, profile, false);
+			if (r)
+				dev_warn(adev->dev, "(%d) failed to disable %s power profile mode\n", r,
+					 profile == PP_SMC_POWER_PROFILE_FULLSCREEN3D ?
+					 "fullscreen 3D" : "compute");
+			adev->gfx.workload_profile_active = false;
+		}
+		mutex_unlock(&adev->gfx.workload_profile_mutex);
 	} else {
 		schedule_delayed_work(&adev->gfx.idle_work, GFX_PROFILE_IDLE_TIMEOUT);
 	}
@@ -2183,13 +2181,25 @@ void amdgpu_gfx_profile_ring_begin_use(struct amdgpu_ring *ring)
 
 	atomic_inc(&adev->gfx.total_submission_cnt);
 
-	if (!cancel_delayed_work_sync(&adev->gfx.idle_work)) {
+	cancel_delayed_work_sync(&adev->gfx.idle_work);
+
+	/* We can safely return early here because we've cancelled the
+	 * the delayed work so there is no one else to set it to false
+	 * and we don't care if someone else sets it to true.
+	 */
+	if (adev->gfx.workload_profile_active)
+		return;
+
+	mutex_lock(&adev->gfx.workload_profile_mutex);
+	if (!adev->gfx.workload_profile_active) {
 		r = amdgpu_dpm_switch_power_profile(adev, profile, true);
 		if (r)
 			dev_warn(adev->dev, "(%d) failed to disable %s power profile mode\n", r,
 				 profile == PP_SMC_POWER_PROFILE_FULLSCREEN3D ?
 				 "fullscreen 3D" : "compute");
+		adev->gfx.workload_profile_active = true;
 	}
+	mutex_unlock(&adev->gfx.workload_profile_mutex);
 }
 
 void amdgpu_gfx_profile_ring_end_use(struct amdgpu_ring *ring)

drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h

@@ -482,6 +482,8 @@ struct amdgpu_gfx {
 
 	atomic_t			total_submission_cnt;
 	struct delayed_work		idle_work;
+	bool				workload_profile_active;
+	struct mutex                    workload_profile_mutex;
 };
 
 struct amdgpu_gfx_ras_reg_entry {