Merge tag 'amd-drm-fixes-6.12-2024-10-16' of https://gitlab.freedesktop.org/agd5f/linux into drm-fixes

amd-drm-fixes-6.12-2024-10-16:

amdgpu:
- SR-IOV fix
- CS chunk handling fix
- MES fixes
- SMU13 fixes

amdkfd:
- VRAM usage reporting fix

radeon:
- Fix possible_clones handling

Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241016200514.3520286-1-alexander.deucher@amd.com

11 files changed, 61 insertions, 35 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 1e475eb01417..d891ab779ca7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -265,7 +265,7 @@ static int amdgpu_cs_pass1(struct amdgpu_cs_parser *p,
 
 			/* Only a single BO list is allowed to simplify handling. */
 			if (p->bo_list)
-				ret = -EINVAL;
+				goto free_partial_kdata;
 
 			ret = amdgpu_cs_p1_bo_handles(p, p->chunks[i].kdata);
 			if (ret)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
index 83e54697f0ee..f1ffab5a1eae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
@@ -1635,11 +1635,9 @@ int amdgpu_gfx_sysfs_isolation_shader_init(struct amdgpu_device *adev)
 {
 	int r;
 
-	if (!amdgpu_sriov_vf(adev)) {
-		r = device_create_file(adev->dev, &dev_attr_enforce_isolation);
-		if (r)
-			return r;
-	}
+	r = device_create_file(adev->dev, &dev_attr_enforce_isolation);
+	if (r)
+		return r;
 
 	r = device_create_file(adev->dev, &dev_attr_run_cleaner_shader);
 	if (r)
@@ -1650,8 +1648,7 @@ int amdgpu_gfx_sysfs_isolation_shader_init(struct amdgpu_device *adev)
 
 void amdgpu_gfx_sysfs_isolation_shader_fini(struct amdgpu_device *adev)
 {
-	if (!amdgpu_sriov_vf(adev))
-		device_remove_file(adev->dev, &dev_attr_enforce_isolation);
+	device_remove_file(adev->dev, &dev_attr_enforce_isolation);
 	device_remove_file(adev->dev, &dev_attr_run_cleaner_shader);
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
index 10b61ff63802..7d4b540340e0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
@@ -1203,8 +1203,10 @@ int amdgpu_mes_add_ring(struct amdgpu_device *adev, int gang_id,
 
 	r = amdgpu_ring_init(adev, ring, 1024, NULL, 0,
 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
-	if (r)
+	if (r) {
+		amdgpu_mes_unlock(&adev->mes);
 		goto clean_up_memory;
+	}
 
 	amdgpu_mes_ring_to_queue_props(adev, ring, &qprops);
 
@@ -1237,7 +1239,6 @@ clean_up_ring:
 	amdgpu_ring_fini(ring);
 clean_up_memory:
 	kfree(ring);
-	amdgpu_mes_unlock(&adev->mes);
 	return r;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
index 8d27421689c9..a37a6801c9ea 100644
--- a/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/mes_v12_0.c
@@ -621,7 +621,7 @@ static int mes_v12_0_set_hw_resources(struct amdgpu_mes *mes, int pipe)
 
 	if (amdgpu_mes_log_enable) {
 		mes_set_hw_res_pkt.enable_mes_event_int_logging = 1;
-		mes_set_hw_res_pkt.event_intr_history_gpu_mc_ptr = mes->event_log_gpu_addr;
+		mes_set_hw_res_pkt.event_intr_history_gpu_mc_ptr = mes->event_log_gpu_addr + pipe * AMDGPU_MES_LOG_BUFFER_SIZE;
 	}
 
 	return mes_v12_0_submit_pkt_and_poll_completion(mes, pipe,
@@ -1336,7 +1336,7 @@ static int mes_v12_0_sw_init(void *handle)
 	adev->mes.kiq_hw_fini = &mes_v12_0_kiq_hw_fini;
 	adev->mes.enable_legacy_queue_map = true;
 
-	adev->mes.event_log_size = AMDGPU_MES_LOG_BUFFER_SIZE;
+	adev->mes.event_log_size = adev->enable_uni_mes ? (AMDGPU_MAX_MES_PIPES * AMDGPU_MES_LOG_BUFFER_SIZE) : AMDGPU_MES_LOG_BUFFER_SIZE;
 
 	r = amdgpu_mes_init(adev);
 	if (r)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 9044bdb38cf4..3e6b4736a7fe 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -1148,7 +1148,7 @@ static int kfd_ioctl_alloc_memory_of_gpu(struct file *filep,
 
 		if (flags & KFD_IOC_ALLOC_MEM_FLAGS_AQL_QUEUE_MEM)
 			size >>= 1;
-		WRITE_ONCE(pdd->vram_usage, pdd->vram_usage + PAGE_ALIGN(size));
+		atomic64_add(PAGE_ALIGN(size), &pdd->vram_usage);
 	}
 
 	mutex_unlock(&p->mutex);
@@ -1219,7 +1219,7 @@ static int kfd_ioctl_free_memory_of_gpu(struct file *filep,
 		kfd_process_device_remove_obj_handle(
 			pdd, GET_IDR_HANDLE(args->handle));
 
-	WRITE_ONCE(pdd->vram_usage, pdd->vram_usage - size);
+	atomic64_sub(size, &pdd->vram_usage);
 
 err_unlock:
 err_pdd:
@@ -2347,7 +2347,7 @@ static int criu_restore_memory_of_gpu(struct kfd_process_device *pdd,
 	} else if (bo_bucket->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM) {
 		bo_bucket->restored_offset = offset;
 		/* Update the VRAM usage count */
-		WRITE_ONCE(pdd->vram_usage, pdd->vram_usage + bo_bucket->size);
+		atomic64_add(bo_bucket->size, &pdd->vram_usage);
 	}
 	return 0;
 }
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index d6530febabad..26e48fdc8728 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -775,7 +775,7 @@ struct kfd_process_device {
 	enum kfd_pdd_bound bound;
 
 	/* VRAM usage */
-	uint64_t vram_usage;
+	atomic64_t vram_usage;
 	struct attribute attr_vram;
 	char vram_filename[MAX_SYSFS_FILENAME_LEN];
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index d665ecdcd12f..d4aa843aacfd 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -332,7 +332,7 @@ static ssize_t kfd_procfs_show(struct kobject *kobj, struct attribute *attr,
 	} else if (strncmp(attr->name, "vram_", 5) == 0) {
 		struct kfd_process_device *pdd = container_of(attr, struct kfd_process_device,
 							      attr_vram);
-		return snprintf(buffer, PAGE_SIZE, "%llu\n", READ_ONCE(pdd->vram_usage));
+		return snprintf(buffer, PAGE_SIZE, "%llu\n", atomic64_read(&pdd->vram_usage));
 	} else if (strncmp(attr->name, "sdma_", 5) == 0) {
 		struct kfd_process_device *pdd = container_of(attr, struct kfd_process_device,
 							      attr_sdma);
@@ -1625,7 +1625,7 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_node *dev,
 	pdd->bound = PDD_UNBOUND;
 	pdd->already_dequeued = false;
 	pdd->runtime_inuse = false;
-	pdd->vram_usage = 0;
+	atomic64_set(&pdd->vram_usage, 0);
 	pdd->sdma_past_activity_counter = 0;
 	pdd->user_gpu_id = dev->id;
 	atomic64_set(&pdd->evict_duration_counter, 0);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 04e746923697..1893c27746a5 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -405,6 +405,27 @@ static void svm_range_bo_release(struct kref *kref)
 		spin_lock(&svm_bo->list_lock);
 	}
 	spin_unlock(&svm_bo->list_lock);
+
+	if (mmget_not_zero(svm_bo->eviction_fence->mm)) {
+		struct kfd_process_device *pdd;
+		struct kfd_process *p;
+		struct mm_struct *mm;
+
+		mm = svm_bo->eviction_fence->mm;
+		/*
+		 * The forked child process takes svm_bo device pages ref, svm_bo could be
+		 * released after parent process is gone.
+		 */
+		p = kfd_lookup_process_by_mm(mm);
+		if (p) {
+			pdd = kfd_get_process_device_data(svm_bo->node, p);
+			if (pdd)
+				atomic64_sub(amdgpu_bo_size(svm_bo->bo), &pdd->vram_usage);
+			kfd_unref_process(p);
+		}
+		mmput(mm);
+	}
+
 	if (!dma_fence_is_signaled(&svm_bo->eviction_fence->base))
 		/* We're not in the eviction worker. Signal the fence. */
 		dma_fence_signal(&svm_bo->eviction_fence->base);
@@ -532,6 +553,7 @@ int
 svm_range_vram_node_new(struct kfd_node *node, struct svm_range *prange,
 			bool clear)
 {
+	struct kfd_process_device *pdd;
 	struct amdgpu_bo_param bp;
 	struct svm_range_bo *svm_bo;
 	struct amdgpu_bo_user *ubo;
@@ -623,6 +645,10 @@ svm_range_vram_node_new(struct kfd_node *node, struct svm_range *prange,
 	list_add(&prange->svm_bo_list, &svm_bo->range_list);
 	spin_unlock(&svm_bo->list_lock);
 
+	pdd = svm_range_get_pdd_by_node(prange, node);
+	if (pdd)
+		atomic64_add(amdgpu_bo_size(bo), &pdd->vram_usage);
+
 	return 0;
 
 reserve_bo_failed:
diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
index bb3bc68dfc39..9ad9cf7a9c98 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
@@ -1264,7 +1264,11 @@ static int smu_sw_init(void *handle)
 	smu->workload_prority[PP_SMC_POWER_PROFILE_VR] = 4;
 	smu->workload_prority[PP_SMC_POWER_PROFILE_COMPUTE] = 5;
 	smu->workload_prority[PP_SMC_POWER_PROFILE_CUSTOM] = 6;
-	smu->workload_mask = 1 << smu->workload_prority[PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT];
+
+	if (smu->is_apu)
+		smu->workload_mask = 1 << smu->workload_prority[PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT];
+	else
+		smu->workload_mask = 1 << smu->workload_prority[PP_SMC_POWER_PROFILE_FULLSCREEN3D];
 
 	smu->workload_setting[0] = PP_SMC_POWER_PROFILE_BOOTUP_DEFAULT;
 	smu->workload_setting[1] = PP_SMC_POWER_PROFILE_FULLSCREEN3D;
@@ -2226,7 +2230,7 @@ static int smu_bump_power_profile_mode(struct smu_context *smu,
 static int smu_adjust_power_state_dynamic(struct smu_context *smu,
 					  enum amd_dpm_forced_level level,
 					  bool skip_display_settings,
-					  bool force_update)
+					  bool init)
 {
 	int ret = 0;
 	int index = 0;
@@ -2255,7 +2259,7 @@ static int smu_adjust_power_state_dynamic(struct smu_context *smu,
 		}
 	}
 
-	if (force_update || smu_dpm_ctx->dpm_level != level) {
+	if (smu_dpm_ctx->dpm_level != level) {
 		ret = smu_asic_set_performance_level(smu, level);
 		if (ret) {
 			dev_err(smu->adev->dev, "Failed to set performance level!");
@@ -2272,7 +2276,7 @@ static int smu_adjust_power_state_dynamic(struct smu_context *smu,
 		index = index > 0 && index <= WORKLOAD_POLICY_MAX ? index - 1 : 0;
 		workload[0] = smu->workload_setting[index];
 
-		if (force_update || smu->power_profile_mode != workload[0])
+		if (init || smu->power_profile_mode != workload[0])
 			smu_bump_power_profile_mode(smu, workload, 0);
 	}
 
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
index 1d024b122b0c..cb923e33fd6f 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_0_ppt.c
@@ -2555,18 +2555,16 @@ static int smu_v13_0_0_set_power_profile_mode(struct smu_context *smu,
 	workload_mask = 1 << workload_type;
 
 	/* Add optimizations for SMU13.0.0/10.  Reuse the power saving profile */
-	if (smu->power_profile_mode == PP_SMC_POWER_PROFILE_COMPUTE) {
-		if ((amdgpu_ip_version(smu->adev, MP1_HWIP, 0) == IP_VERSION(13, 0, 0) &&
-			((smu->adev->pm.fw_version == 0x004e6601) ||
-			(smu->adev->pm.fw_version >= 0x004e7300))) ||
-			(amdgpu_ip_version(smu->adev, MP1_HWIP, 0) == IP_VERSION(13, 0, 10) &&
-			 smu->adev->pm.fw_version >= 0x00504500)) {
-			workload_type = smu_cmn_to_asic_specific_index(smu,
-								CMN2ASIC_MAPPING_WORKLOAD,
-								PP_SMC_POWER_PROFILE_POWERSAVING);
-			if (workload_type >= 0)
-				workload_mask |= 1 << workload_type;
-		}
+	if ((amdgpu_ip_version(smu->adev, MP1_HWIP, 0) == IP_VERSION(13, 0, 0) &&
+	     ((smu->adev->pm.fw_version == 0x004e6601) ||
+	      (smu->adev->pm.fw_version >= 0x004e7300))) ||
+	    (amdgpu_ip_version(smu->adev, MP1_HWIP, 0) == IP_VERSION(13, 0, 10) &&
+	     smu->adev->pm.fw_version >= 0x00504500)) {
+		workload_type = smu_cmn_to_asic_specific_index(smu,
+							       CMN2ASIC_MAPPING_WORKLOAD,
+							       PP_SMC_POWER_PROFILE_POWERSAVING);
+		if (workload_type >= 0)
+			workload_mask |= 1 << workload_type;
 	}
 
 	ret = smu_cmn_send_smc_msg_with_param(smu,
diff --git a/drivers/gpu/drm/radeon/radeon_encoders.c b/drivers/gpu/drm/radeon/radeon_encoders.c
index 0f723292409e..fafed331e0a0 100644
--- a/drivers/gpu/drm/radeon/radeon_encoders.c
+++ b/drivers/gpu/drm/radeon/radeon_encoders.c
@@ -43,7 +43,7 @@ static uint32_t radeon_encoder_clones(struct drm_encoder *encoder)
 	struct radeon_device *rdev = dev->dev_private;
 	struct radeon_encoder *radeon_encoder = to_radeon_encoder(encoder);
 	struct drm_encoder *clone_encoder;
-	uint32_t index_mask = 0;
+	uint32_t index_mask = drm_encoder_mask(encoder);
 	int count;
 
 	/* DIG routing gets problematic */