1 files changed, 65 insertions, 50 deletions

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
index 01b960b15274..c643e0ccec52 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
@@ -69,8 +69,8 @@ static int find_available_queue_slot(struct process_queue_manager *pqm,
 	pr_debug("The new slot id %lu\n", found);
 
 	if (found >= KFD_MAX_NUM_OF_QUEUES_PER_PROCESS) {
-		pr_info("Cannot open more queues for process with pasid 0x%x\n",
-				pqm->process->pasid);
+		pr_info("Cannot open more queues for process with pid %d\n",
+			pqm->process->lead_thread->pid);
 		return -ENOMEM;
 	}
 
@@ -86,9 +86,12 @@ void kfd_process_dequeue_from_device(struct kfd_process_device *pdd)
 
 	if (pdd->already_dequeued)
 		return;
-
+	/* The MES context flush needs to filter out the case which the
+	 * KFD process is created without setting up the MES context and
+	 * queue for creating a compute queue.
+	 */
 	dev->dqm->ops.process_termination(dev->dqm, &pdd->qpd);
-	if (dev->kfd->shared_resources.enable_mes &&
+	if (dev->kfd->shared_resources.enable_mes && !!pdd->proc_ctx_gpu_addr &&
 	    down_read_trylock(&dev->adev->reset_domain->sem)) {
 		amdgpu_mes_flush_shader_debugger(dev->adev,
 						 pdd->proc_ctx_gpu_addr);
@@ -131,8 +134,9 @@ int pqm_set_gws(struct process_queue_manager *pqm, unsigned int qid,
 	if (!gws && pdd->qpd.num_gws == 0)
 		return -EINVAL;
 
-	if (KFD_GC_VERSION(dev) != IP_VERSION(9, 4, 3) &&
-	    KFD_GC_VERSION(dev) != IP_VERSION(9, 4, 4) &&
+	if ((KFD_GC_VERSION(dev) != IP_VERSION(9, 4, 3) &&
+	     KFD_GC_VERSION(dev) != IP_VERSION(9, 4, 4) &&
+	     KFD_GC_VERSION(dev) != IP_VERSION(9, 5, 0)) &&
 	    !dev->kfd->shared_resources.enable_mes) {
 		if (gws)
 			ret = amdgpu_amdkfd_add_gws_to_process(pdd->process->kgd_process_info,
@@ -197,6 +201,7 @@ static void pqm_clean_queue_resource(struct process_queue_manager *pqm,
 	if (pqn->q->gws) {
 		if (KFD_GC_VERSION(pqn->q->device) != IP_VERSION(9, 4, 3) &&
 		    KFD_GC_VERSION(pqn->q->device) != IP_VERSION(9, 4, 4) &&
+		    KFD_GC_VERSION(pqn->q->device) != IP_VERSION(9, 5, 0) &&
 		    !dev->kfd->shared_resources.enable_mes)
 			amdgpu_amdkfd_remove_gws_from_process(
 				pqm->process->kgd_process_info, pqn->q->gws);
@@ -212,13 +217,17 @@ static void pqm_clean_queue_resource(struct process_queue_manager *pqm,
 void pqm_uninit(struct process_queue_manager *pqm)
 {
 	struct process_queue_node *pqn, *next;
-	struct kfd_process_device *pdd;
 
 	list_for_each_entry_safe(pqn, next, &pqm->queues, process_queue_list) {
 		if (pqn->q) {
-			pdd = kfd_get_process_device_data(pqn->q->device, pqm->process);
-			kfd_queue_unref_bo_vas(pdd, &pqn->q->properties);
-			kfd_queue_release_buffers(pdd, &pqn->q->properties);
+			struct kfd_process_device *pdd = kfd_get_process_device_data(pqn->q->device,
+										     pqm->process);
+			if (pdd) {
+				kfd_queue_unref_bo_vas(pdd, &pqn->q->properties);
+				kfd_queue_release_buffers(pdd, &pqn->q->properties);
+			} else {
+				WARN_ON(!pdd);
+			}
 			pqm_clean_queue_resource(pqm, pqn);
 		}
 
@@ -235,7 +244,7 @@ void pqm_uninit(struct process_queue_manager *pqm)
 static int init_user_queue(struct process_queue_manager *pqm,
 				struct kfd_node *dev, struct queue **q,
 				struct queue_properties *q_properties,
-				struct file *f, unsigned int qid)
+				unsigned int qid)
 {
 	int retval;
 
@@ -270,20 +279,17 @@ static int init_user_queue(struct process_queue_manager *pqm,
 		/* Starting with GFX11, wptr BOs must be mapped to GART for MES to determine work
 		 * on unmapped queues for usermode queue oversubscription (no aggregated doorbell)
 		 */
-		if (((dev->adev->mes.sched_version & AMDGPU_MES_API_VERSION_MASK)
-		    >> AMDGPU_MES_API_VERSION_SHIFT) >= 2) {
-			if (dev->adev != amdgpu_ttm_adev(q_properties->wptr_bo->tbo.bdev)) {
-				pr_err("Queue memory allocated to wrong device\n");
-				retval = -EINVAL;
-				goto free_gang_ctx_bo;
-			}
+		if (dev->adev != amdgpu_ttm_adev(q_properties->wptr_bo->tbo.bdev)) {
+			pr_err("Queue memory allocated to wrong device\n");
+			retval = -EINVAL;
+			goto free_gang_ctx_bo;
+		}
 
-			retval = amdgpu_amdkfd_map_gtt_bo_to_gart(q_properties->wptr_bo,
-								  &(*q)->wptr_bo_gart);
-			if (retval) {
-				pr_err("Failed to map wptr bo to GART\n");
-				goto free_gang_ctx_bo;
-			}
+		retval = amdgpu_amdkfd_map_gtt_bo_to_gart(q_properties->wptr_bo,
+							  &(*q)->wptr_bo_gart);
+		if (retval) {
+			pr_err("Failed to map wptr bo to GART\n");
+			goto free_gang_ctx_bo;
 		}
 	}
 
@@ -291,7 +297,7 @@ static int init_user_queue(struct process_queue_manager *pqm,
 	return 0;
 
 free_gang_ctx_bo:
-	amdgpu_amdkfd_free_gtt_mem(dev->adev, (*q)->gang_ctx_bo);
+	amdgpu_amdkfd_free_gtt_mem(dev->adev, &(*q)->gang_ctx_bo);
 cleanup:
 	uninit_queue(*q);
 	*q = NULL;
@@ -300,7 +306,6 @@ cleanup:
 
 int pqm_create_queue(struct process_queue_manager *pqm,
 			    struct kfd_node *dev,
-			    struct file *f,
 			    struct queue_properties *properties,
 			    unsigned int *qid,
 			    const struct kfd_criu_queue_priv_data *q_data,
@@ -317,11 +322,12 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 	unsigned int max_queues = 127; /* HWS limit */
 
 	/*
-	 * On GFX 9.4.3, increase the number of queues that
-	 * can be created to 255. No HWS limit on GFX 9.4.3.
+	 * On GFX 9.4.3/9.5.0, increase the number of queues that
+	 * can be created to 255. No HWS limit on GFX 9.4.3/9.5.0.
 	 */
 	if (KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 3) ||
-	    KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 4))
+	    KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 4) ||
+	    KFD_GC_VERSION(dev) == IP_VERSION(9, 5, 0))
 		max_queues = 255;
 
 	q = NULL;
@@ -354,10 +360,26 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 	if (retval != 0)
 		return retval;
 
+	/* Register process if this is the first queue */
 	if (list_empty(&pdd->qpd.queues_list) &&
 	    list_empty(&pdd->qpd.priv_queue_list))
 		dev->dqm->ops.register_process(dev->dqm, &pdd->qpd);
 
+	/* Allocate proc_ctx_bo only if MES is enabled and this is the first queue */
+	if (!pdd->proc_ctx_cpu_ptr && dev->kfd->shared_resources.enable_mes) {
+		retval = amdgpu_amdkfd_alloc_gtt_mem(dev->adev,
+						     AMDGPU_MES_PROC_CTX_SIZE,
+						     &pdd->proc_ctx_bo,
+						     &pdd->proc_ctx_gpu_addr,
+						     &pdd->proc_ctx_cpu_ptr,
+						     false);
+		if (retval) {
+			dev_err(dev->adev->dev, "failed to allocate process context bo\n");
+			return retval;
+		}
+		memset(pdd->proc_ctx_cpu_ptr, 0, AMDGPU_MES_PROC_CTX_SIZE);
+	}
+
 	pqn = kzalloc(sizeof(*pqn), GFP_KERNEL);
 	if (!pqn) {
 		retval = -ENOMEM;
@@ -374,7 +396,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 		 * allocate_sdma_queue() in create_queue() has the
 		 * corresponding check logic.
 		 */
-		retval = init_user_queue(pqm, dev, &q, properties, f, *qid);
+		retval = init_user_queue(pqm, dev, &q, properties, *qid);
 		if (retval != 0)
 			goto err_create_queue;
 		pqn->q = q;
@@ -395,7 +417,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 			goto err_create_queue;
 		}
 
-		retval = init_user_queue(pqm, dev, &q, properties, f, *qid);
+		retval = init_user_queue(pqm, dev, &q, properties, *qid);
 		if (retval != 0)
 			goto err_create_queue;
 		pqn->q = q;
@@ -426,8 +448,15 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 	}
 
 	if (retval != 0) {
-		pr_err("Pasid 0x%x DQM create queue type %d failed. ret %d\n",
-			pqm->process->pasid, type, retval);
+		if ((type == KFD_QUEUE_TYPE_SDMA ||
+		    type == KFD_QUEUE_TYPE_SDMA_XGMI ||
+		    type == KFD_QUEUE_TYPE_SDMA_BY_ENG_ID) &&
+		    retval == -ENOMEM)
+			pr_warn("process pid %d DQM create queue type %d failed. ret %d\n",
+				pqm->process->lead_thread->pid, type, retval);
+		else
+			pr_err("process pid %d DQM create queue type %d failed. ret %d\n",
+				pqm->process->lead_thread->pid, type, retval);
 		goto err_create_queue;
 	}
 
@@ -521,9 +550,9 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid)
 		retval = dqm->ops.destroy_queue(dqm, &pdd->qpd, pqn->q);
 		if (retval) {
 			pr_err("Pasid 0x%x destroy queue %d failed, ret %d\n",
-				pqm->process->pasid,
+				pdd->pasid,
 				pqn->q->properties.queue_id, retval);
-			if (retval != -ETIME)
+			if (retval != -ETIME && retval != -EIO)
 				goto err_destroy_queue;
 		}
 		kfd_procfs_del_queue(pqn->q);
@@ -643,19 +672,6 @@ int pqm_update_mqd(struct process_queue_manager *pqm,
 	return 0;
 }
 
-struct kernel_queue *pqm_get_kernel_queue(
-					struct process_queue_manager *pqm,
-					unsigned int qid)
-{
-	struct process_queue_node *pqn;
-
-	pqn = get_queue_by_qid(pqm, qid);
-	if (pqn && pqn->kq)
-		return pqn->kq;
-
-	return NULL;
-}
-
 struct queue *pqm_get_user_queue(struct process_queue_manager *pqm,
 					unsigned int qid)
 {
@@ -1029,8 +1045,7 @@ int kfd_criu_restore_queue(struct kfd_process *p,
 
 	print_queue_properties(&qp);
 
-	ret = pqm_create_queue(&p->pqm, pdd->dev, NULL, &qp, &queue_id, q_data, mqd, ctl_stack,
-				NULL);
+	ret = pqm_create_queue(&p->pqm, pdd->dev, &qp, &queue_id, q_data, mqd, ctl_stack, NULL);
 	if (ret) {
 		pr_err("Failed to create new queue err:%d\n", ret);
 		goto exit;