1 files changed, 80 insertions, 5 deletions

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 1c7016958d6d..dc774ddf3445 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -65,6 +65,25 @@ static int kfd_char_dev_major = -1;
 static struct class *kfd_class;
 struct device *kfd_device;
 
+static inline struct kfd_process_device *kfd_lock_pdd_by_id(struct kfd_process *p, __u32 gpu_id)
+{
+	struct kfd_process_device *pdd;
+
+	mutex_lock(&p->mutex);
+	pdd = kfd_process_device_data_by_id(p, gpu_id);
+
+	if (pdd)
+		return pdd;
+
+	mutex_unlock(&p->mutex);
+	return NULL;
+}
+
+static inline void kfd_unlock_pdd(struct kfd_process_device *pdd)
+{
+	mutex_unlock(&pdd->process->mutex);
+}
+
 int kfd_chardev_init(void)
 {
 	int err = 0;
@@ -280,6 +299,7 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,
 	struct kfd_process_device *pdd;
 	struct queue_properties q_properties;
 	uint32_t doorbell_offset_in_process = 0;
+	struct amdgpu_bo *wptr_bo = NULL;
 
 	memset(&q_properties, 0, sizeof(struct queue_properties));
 
@@ -307,12 +327,49 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,
 		goto err_bind_process;
 	}
 
+	/* 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->shared_resources.enable_mes &&
+			((dev->adev->mes.sched_version & AMDGPU_MES_API_VERSION_MASK)
+			>> AMDGPU_MES_API_VERSION_SHIFT) >= 2) {
+		struct amdgpu_bo_va_mapping *wptr_mapping;
+		struct amdgpu_vm *wptr_vm;
+
+		wptr_vm = drm_priv_to_vm(pdd->drm_priv);
+		err = amdgpu_bo_reserve(wptr_vm->root.bo, false);
+		if (err)
+			goto err_wptr_map_gart;
+
+		wptr_mapping = amdgpu_vm_bo_lookup_mapping(
+				wptr_vm, args->write_pointer_address >> PAGE_SHIFT);
+		amdgpu_bo_unreserve(wptr_vm->root.bo);
+		if (!wptr_mapping) {
+			pr_err("Failed to lookup wptr bo\n");
+			err = -EINVAL;
+			goto err_wptr_map_gart;
+		}
+
+		wptr_bo = wptr_mapping->bo_va->base.bo;
+		if (wptr_bo->tbo.base.size > PAGE_SIZE) {
+			pr_err("Requested GART mapping for wptr bo larger than one page\n");
+			err = -EINVAL;
+			goto err_wptr_map_gart;
+		}
+
+		err = amdgpu_amdkfd_map_gtt_bo_to_gart(dev->adev, wptr_bo);
+		if (err) {
+			pr_err("Failed to map wptr bo to GART\n");
+			goto err_wptr_map_gart;
+		}
+	}
+
 	pr_debug("Creating queue for PASID 0x%x on gpu 0x%x\n",
 			p->pasid,
 			dev->id);
 
-	err = pqm_create_queue(&p->pqm, dev, filep, &q_properties, &queue_id, NULL, NULL, NULL,
-			&doorbell_offset_in_process);
+	err = pqm_create_queue(&p->pqm, dev, filep, &q_properties, &queue_id, wptr_bo,
+			NULL, NULL, NULL, &doorbell_offset_in_process);
 	if (err != 0)
 		goto err_create_queue;
 
@@ -344,6 +401,9 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,
 	return 0;
 
 err_create_queue:
+	if (wptr_bo)
+		amdgpu_amdkfd_free_gtt_mem(dev->adev, wptr_bo);
+err_wptr_map_gart:
 err_bind_process:
 err_pdd:
 	mutex_unlock(&p->mutex);
@@ -814,7 +874,7 @@ static int kfd_ioctl_wait_events(struct file *filp, struct kfd_process *p,
 	err = kfd_wait_on_events(p, args->num_events,
 			(void __user *)args->events_ptr,
 			(args->wait_for_all != 0),
-			args->timeout, &args->wait_result);
+			&args->timeout, &args->wait_result);
 
 	return err;
 }
@@ -958,6 +1018,19 @@ bool kfd_dev_is_large_bar(struct kfd_dev *dev)
 	return false;
 }
 
+static int kfd_ioctl_get_available_memory(struct file *filep,
+					  struct kfd_process *p, void *data)
+{
+	struct kfd_ioctl_get_available_memory_args *args = data;
+	struct kfd_process_device *pdd = kfd_lock_pdd_by_id(p, args->gpu_id);
+
+	if (!pdd)
+		return -EINVAL;
+	args->available = amdgpu_amdkfd_get_available_memory(pdd->dev->adev);
+	kfd_unlock_pdd(pdd);
+	return 0;
+}
+
 static int kfd_ioctl_alloc_memory_of_gpu(struct file *filep,
 					struct kfd_process *p, void *data)
 {
@@ -2361,7 +2434,7 @@ static int criu_restore(struct file *filep,
 	 * Set the process to evicted state to avoid running any new queues before all the memory
 	 * mappings are ready.
 	 */
-	ret = kfd_process_evict_queues(p);
+	ret = kfd_process_evict_queues(p, KFD_QUEUE_EVICTION_CRIU_RESTORE);
 	if (ret)
 		goto exit_unlock;
 
@@ -2480,7 +2553,7 @@ static int criu_process_info(struct file *filep,
 		goto err_unlock;
 	}
 
-	ret = kfd_process_evict_queues(p);
+	ret = kfd_process_evict_queues(p, KFD_QUEUE_EVICTION_CRIU_CHECKPOINT);
 	if (ret)
 		goto err_unlock;
 
@@ -2648,6 +2721,8 @@ static const struct amdkfd_ioctl_desc amdkfd_ioctls[] = {
 	AMDKFD_IOCTL_DEF(AMDKFD_IOC_CRIU_OP,
 			kfd_ioctl_criu, KFD_IOC_FLAG_CHECKPOINT_RESTORE),
 
+	AMDKFD_IOCTL_DEF(AMDKFD_IOC_AVAILABLE_MEMORY,
+			kfd_ioctl_get_available_memory, 0),
 };
 
 #define AMDKFD_CORE_IOCTL_COUNT	ARRAY_SIZE(amdkfd_ioctls)