path: root/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1279-drm-amdkfd-Move-doorbell_bitmap-to-per-process-struc.patch

From b453a28cec2c23798add20268437be07e437c81c Mon Sep 17 00:00:00 2001
From: Shaoyun Liu <Shaoyun.Liu@amd.com>
Date: Thu, 4 May 2017 17:39:58 -0400
Subject: [PATCH 1279/4131] drm/amdkfd: Move doorbell_bitmap to per process
 structure

The original DQM structure is not per process, so when process
terminated, the doorbell_bitmap will not be cleared. If user
didn't destroy the queue correctly, driver will run out of
the resource eventually.

Change-Id: Ifb5b0b5b9745f79730be9b730bf0be5b5c3c37d2
Signed-off-by: Shaoyun Liu <Shaoyun.Liu@amd.com>
---
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c  | 78 +++++---------------
 .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.h  |  1 -
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h              |  3 +
 drivers/gpu/drm/amd/amdkfd/kfd_process.c           | 85 ++++++++++++++++------
 4 files changed, 84 insertions(+), 83 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index 702fc4a..2a4a556 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -111,38 +111,11 @@ void program_sh_mem_settings(struct device_queue_manager *dqm,
 						qpd->sh_mem_bases);
 }
 
-static int init_doorbell_bitmap(struct device_queue_manager *dqm)
+static int allocate_doorbell(struct qcm_process_device *qpd, struct queue *q)
 {
-	unsigned int i;
-
-	if (!KFD_IS_SOC15(dqm->dev->device_info->asic_family))
-		return 0;
-
-	dqm->doorbell_bitmap =
-		kzalloc(DIV_ROUND_UP(KFD_MAX_NUM_OF_QUEUES_PER_PROCESS,
-				     BITS_PER_BYTE), GFP_KERNEL);
-	if (dqm->doorbell_bitmap == NULL)
-		return -ENOMEM;
-
-	/* Mask out any reserved doorbells */
-	for (i = 0; i < KFD_MAX_NUM_OF_QUEUES_PER_PROCESS; i++)
-		if ((dqm->dev->shared_resources.reserved_doorbell_mask & i) ==
-		    dqm->dev->shared_resources.reserved_doorbell_val) {
-			set_bit(i, dqm->doorbell_bitmap);
-			pr_debug("reserved doorbell 0x%03x\n", i);
-		}
-
-	return 0;
-}
+	struct kfd_dev *dev = qpd->dqm->dev;
 
-static void uninit_doorbell_bitmap(struct device_queue_manager *dqm)
-{
-	kfree(dqm->doorbell_bitmap);
-}
-
-static int allocate_doorbell(struct device_queue_manager *dqm, struct queue *q)
-{
-	if (!KFD_IS_SOC15(dqm->dev->device_info->asic_family)) {
+	if (!KFD_IS_SOC15(dev->device_info->asic_family)) {
 		/* On pre-SOC15 chips we need to use the queue ID to
 		 * preserve the user mode ABI.
 		 */
@@ -151,40 +124,41 @@ static int allocate_doorbell(struct device_queue_manager *dqm, struct queue *q)
 		/* For SDMA queues on SOC15, use static doorbell
 		 * assignments based on the engine and queue.
 		 */
-		q->doorbell_id = dqm->dev->shared_resources.sdma_doorbell
+		q->doorbell_id = dev->shared_resources.sdma_doorbell
 			[q->properties.sdma_engine_id]
 			[q->properties.sdma_queue_id];
 	} else {
 		/* For CP queues on SOC15 reserve a free doorbell ID */
 		unsigned int found;
 
-		found = find_first_zero_bit(dqm->doorbell_bitmap,
+		found = find_first_zero_bit(qpd->doorbell_bitmap,
 					    KFD_MAX_NUM_OF_QUEUES_PER_PROCESS);
 		if (found >= KFD_MAX_NUM_OF_QUEUES_PER_PROCESS) {
 			pr_debug("amdkfd: No doorbells available");
 			return -EBUSY;
 		}
-		set_bit(found, dqm->doorbell_bitmap);
+		set_bit(found, qpd->doorbell_bitmap);
 		q->doorbell_id = found;
 	}
 
 	q->properties.doorbell_off =
-		kfd_doorbell_id_to_offset(dqm->dev, q->process,
+		kfd_doorbell_id_to_offset(dev, q->process,
 					  q->doorbell_id);
 
 	return 0;
 }
 
-static void deallocate_doorbell(struct device_queue_manager *dqm,
+static void deallocate_doorbell(struct qcm_process_device *qpd,
 				struct queue *q)
 {
 	unsigned int old;
+	struct kfd_dev *dev = qpd->dqm->dev;
 
-	if (!KFD_IS_SOC15(dqm->dev->device_info->asic_family) ||
+	if (!KFD_IS_SOC15(dev->device_info->asic_family) ||
 	    q->properties.type == KFD_QUEUE_TYPE_SDMA)
 		return;
 
-	old = test_and_clear_bit(q->doorbell_id, dqm->doorbell_bitmap);
+	old = test_and_clear_bit(q->doorbell_id, qpd->doorbell_bitmap);
 	WARN_ON(!old);
 }
 
@@ -386,7 +360,7 @@ static int create_compute_queue_nocpsch(struct device_queue_manager *dqm,
 	if (retval != 0)
 		return retval;
 
-	retval = allocate_doorbell(dqm, q);
+	retval = allocate_doorbell(qpd, q);
 	if (retval)
 		goto out_deallocate_hqd;
 
@@ -415,7 +389,7 @@ static int create_compute_queue_nocpsch(struct device_queue_manager *dqm,
 out_uninit_mqd:
 	mqd->uninit_mqd(mqd, q->mqd, q->mqd_mem_obj);
 out_deallocate_doorbell:
-	deallocate_doorbell(dqm, q);
+	deallocate_doorbell(qpd, q);
 out_deallocate_hqd:
 	deallocate_hqd(dqm, q);
 
@@ -441,7 +415,7 @@ static int destroy_queue_nocpsch_locked(struct device_queue_manager *dqm,
 		goto out;
 	}
 
-	deallocate_doorbell(dqm, q);
+	deallocate_doorbell(qpd, q);
 
 	if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE)
 		deallocate_hqd(dqm, q);
@@ -826,7 +800,7 @@ static int init_scheduler(struct device_queue_manager *dqm)
 
 static int initialize_nocpsch(struct device_queue_manager *dqm)
 {
-	int i, ret;
+	int i;
 
 	BUG_ON(!dqm);
 
@@ -838,12 +812,6 @@ static int initialize_nocpsch(struct device_queue_manager *dqm)
 	if (!dqm->allocated_queues)
 		return -ENOMEM;
 
-	ret = init_doorbell_bitmap(dqm);
-	if (ret) {
-		kfree(dqm->allocated_queues);
-		return ret;
-	}
-
 	mutex_init(&dqm->lock);
 	INIT_LIST_HEAD(&dqm->queues);
 	dqm->queue_count = dqm->next_pipe_to_allocate = 0;
@@ -867,7 +835,6 @@ static void uninitialize_nocpsch(struct device_queue_manager *dqm)
 
 	BUG_ON(dqm->queue_count > 0 || dqm->processes_count > 0);
 
-	uninit_doorbell_bitmap(dqm);
 	kfree(dqm->allocated_queues);
 	for (i = 0 ; i < KFD_MQD_TYPE_MAX ; i++)
 		kfree(dqm->mqds[i]);
@@ -928,7 +895,7 @@ static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm,
 	q->properties.sdma_queue_id = q->sdma_id / CIK_SDMA_QUEUES_PER_ENGINE;
 	q->properties.sdma_engine_id = q->sdma_id % CIK_SDMA_QUEUES_PER_ENGINE;
 
-	retval = allocate_doorbell(dqm, q);
+	retval = allocate_doorbell(qpd, q);
 	if (retval)
 		goto out_deallocate_sdma_queue;
 
@@ -951,7 +918,7 @@ static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm,
 out_uninit_mqd:
 	mqd->uninit_mqd(mqd, q->mqd, q->mqd_mem_obj);
 out_deallocate_doorbell:
-	deallocate_doorbell(dqm, q);
+	deallocate_doorbell(qpd, q);
 out_deallocate_sdma_queue:
 	deallocate_sdma_queue(dqm, q->sdma_id);
 
@@ -1015,10 +982,6 @@ static int initialize_cpsch(struct device_queue_manager *dqm)
 	pr_debug("kfd: In func %s num of pipes: %d\n",
 			__func__, get_pipes_per_mec(dqm));
 
-	retval = init_doorbell_bitmap(dqm);
-	if (retval)
-		return retval;
-
 	mutex_init(&dqm->lock);
 	INIT_LIST_HEAD(&dqm->queues);
 	dqm->queue_count = dqm->processes_count = 0;
@@ -1032,7 +995,6 @@ static int initialize_cpsch(struct device_queue_manager *dqm)
 	return 0;
 
 fail_init_pipelines:
-	uninit_doorbell_bitmap(dqm);
 	mutex_destroy(&dqm->lock);
 	return retval;
 }
@@ -1184,7 +1146,7 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q,
 			q->sdma_id % CIK_SDMA_QUEUES_PER_ENGINE;
 	}
 
-	retval = allocate_doorbell(dqm, q);
+	retval = allocate_doorbell(qpd, q);
 	if (retval)
 		goto out_deallocate_sdma_queue;
 
@@ -1234,7 +1196,7 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q,
 	return retval;
 
 out_deallocate_doorbell:
-	deallocate_doorbell(dqm, q);
+	deallocate_doorbell(qpd, q);
 out_deallocate_sdma_queue:
 	if (q->properties.type == KFD_QUEUE_TYPE_SDMA)
 		deallocate_sdma_queue(dqm, q->sdma_id);
@@ -1402,7 +1364,7 @@ static int destroy_queue_cpsch(struct device_queue_manager *dqm,
 		goto failed;
 	}
 
-	deallocate_doorbell(dqm, q);
+	deallocate_doorbell(qpd, q);
 
 	if (q->properties.type == KFD_QUEUE_TYPE_SDMA) {
 		dqm->sdma_queue_count--;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
index c269e5e..113ac1f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
@@ -182,7 +182,6 @@ struct device_queue_manager {
 	unsigned int		*allocated_queues;
 	unsigned int		sdma_bitmap;
 	unsigned int		vmid_bitmap;
-	unsigned long           *doorbell_bitmap;
 	uint64_t		pipelines_addr;
 	struct kfd_mem_obj	*pipeline_mem;
 	uint64_t		fence_gpu_addr;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 7bdddab..a81dfe7 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -577,6 +577,9 @@ struct qcm_process_device {
 	/* IB memory */
 	uint64_t ib_base; /* ib_base+ib_size must be below cwsr_base */
 	void *ib_kaddr;
+
+	/*doorbell resources per process per device*/
+	unsigned long           *doorbell_bitmap;
 };
 
 /* KFD Memory Eviction */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 5343dad..601e551 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -356,6 +356,9 @@ static void kfd_process_destroy_pdds(struct kfd_process *p)
 				get_order(pdd->dev->cwsr_size));
 		}
 
+		kfree(pdd->qpd.doorbell_bitmap);
+		idr_destroy(&pdd->alloc_idr);
+
 		kfree(pdd);
 	}
 }
@@ -633,6 +636,31 @@ static struct kfd_process *create_process(const struct task_struct *thread,
 	return ERR_PTR(err);
 }
 
+static int init_doorbell_bitmap(struct qcm_process_device *qpd,
+			struct kfd_dev *dev)
+{
+	unsigned int i;
+
+	if (!KFD_IS_SOC15(dev->device_info->asic_family))
+		return 0;
+
+	qpd->doorbell_bitmap =
+		kzalloc(DIV_ROUND_UP(KFD_MAX_NUM_OF_QUEUES_PER_PROCESS,
+				     BITS_PER_BYTE), GFP_KERNEL);
+	if (qpd->doorbell_bitmap == NULL)
+		return -ENOMEM;
+
+	/* Mask out any reserved doorbells */
+	for (i = 0; i < KFD_MAX_NUM_OF_QUEUES_PER_PROCESS; i++)
+		if ((dev->shared_resources.reserved_doorbell_mask & i) ==
+		    dev->shared_resources.reserved_doorbell_val) {
+			set_bit(i, qpd->doorbell_bitmap);
+			pr_debug("reserved doorbell 0x%03x\n", i);
+		}
+
+	return 0;
+}
+
 struct kfd_process_device *kfd_get_process_device_data(struct kfd_dev *dev,
 							struct kfd_process *p)
 {
@@ -651,33 +679,42 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev,
 	struct kfd_process_device *pdd = NULL;
 
 	pdd = kzalloc(sizeof(*pdd), GFP_KERNEL);
-	if (pdd != NULL) {
-		pdd->dev = dev;
-		INIT_LIST_HEAD(&pdd->qpd.queues_list);
-		INIT_LIST_HEAD(&pdd->qpd.priv_queue_list);
-		pdd->qpd.dqm = dev->dqm;
-		pdd->qpd.pqm = &p->pqm;
-		pdd->qpd.evicted = 0;
-		pdd->reset_wavefronts = false;
-		pdd->process = p;
-		pdd->bound = PDD_UNBOUND;
-		pdd->already_dequeued = false;
-		list_add(&pdd->per_device_list, &p->per_device_data);
-
-		/* Init idr used for memory handle translation */
-		idr_init(&pdd->alloc_idr);
-
-		/* Create the GPUVM context for this specific device */
-		if (dev->kfd2kgd->create_process_vm(dev->kgd, &pdd->vm,
-						&p->process_info)) {
-			pr_err("Failed to create process VM object\n");
-			list_del(&pdd->per_device_list);
-			kfree(pdd);
-			pdd = NULL;
-		}
+	if (!pdd)
+		return NULL;
+
+	pdd->dev = dev;
+	INIT_LIST_HEAD(&pdd->qpd.queues_list);
+	INIT_LIST_HEAD(&pdd->qpd.priv_queue_list);
+	pdd->qpd.dqm = dev->dqm;
+	pdd->qpd.pqm = &p->pqm;
+	pdd->qpd.evicted = 0;
+	pdd->reset_wavefronts = false;
+	pdd->process = p;
+	pdd->bound = PDD_UNBOUND;
+	pdd->already_dequeued = false;
+	list_add(&pdd->per_device_list, &p->per_device_data);
+
+	/* Init idr used for memory handle translation */
+	idr_init(&pdd->alloc_idr);
+	if (init_doorbell_bitmap(&pdd->qpd, dev)) {
+		pr_err("Failed to init doorbell for process\n");
+		goto err_create_pdd;
 	}
 
+	/* Create the GPUVM context for this specific device */
+	if (dev->kfd2kgd->create_process_vm(dev->kgd, &pdd->vm,
+					&p->process_info)) {
+		pr_err("Failed to create process VM object\n");
+		goto err_create_pdd;
+	}
 	return pdd;
+
+err_create_pdd:
+	kfree(pdd->qpd.doorbell_bitmap);
+	idr_destroy(&pdd->alloc_idr);
+	list_del(&pdd->per_device_list);
+	kfree(pdd);
+	return NULL;
 }
 
 /*
-- 
2.7.4