drm/amdkfd: Use drm_priv to pass VM from KFD to amdgpu

amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu needs the drm_priv to allow mmap
to access the BO through the corresponding file descriptor. The VM can
also be extracted from drm_priv, so drm_priv can replace the vm parameter
in the kfd2kgd interface.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Philip Yang <philip.yang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

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

@@ -236,20 +236,20 @@ uint8_t amdgpu_amdkfd_get_xgmi_hops_count(struct kgd_dev *dst, struct kgd_dev *s
 /* GPUVM API */
 int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct kgd_dev *kgd,
 					struct file *filp, u32 pasid,
-					void **vm, void **process_info,
+					void **process_info,
 					struct dma_fence **ef);
-void amdgpu_amdkfd_gpuvm_release_process_vm(struct kgd_dev *kgd, void *vm);
-uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *vm);
+void amdgpu_amdkfd_gpuvm_release_process_vm(struct kgd_dev *kgd, void *drm_priv);
+uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *drm_priv);
 int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 		struct kgd_dev *kgd, uint64_t va, uint64_t size,
-		void *vm, struct kgd_mem **mem,
+		void *drm_priv, struct kgd_mem **mem,
 		uint64_t *offset, uint32_t flags);
 int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 		struct kgd_dev *kgd, struct kgd_mem *mem, uint64_t *size);
 int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
-		struct kgd_dev *kgd, struct kgd_mem *mem, void *vm);
+		struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv);
 int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
-		struct kgd_dev *kgd, struct kgd_mem *mem, void *vm);
+		struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv);
 int amdgpu_amdkfd_gpuvm_sync_memory(
 		struct kgd_dev *kgd, struct kgd_mem *mem, bool intr);
 int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct kgd_dev *kgd,
@@ -260,7 +260,7 @@ int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct kgd_dev *kgd,
 					      struct kfd_vm_fault_info *info);
 int amdgpu_amdkfd_gpuvm_import_dmabuf(struct kgd_dev *kgd,
 				      struct dma_buf *dmabuf,
-				      uint64_t va, void *vm,
+				      uint64_t va, void *drm_priv,
 				      struct kgd_mem **mem, uint64_t *size,
 				      uint64_t *mmap_offset);
 int amdgpu_amdkfd_get_tile_config(struct kgd_dev *kgd,

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

@@ -948,6 +948,13 @@ static int process_update_pds(struct amdkfd_process_info *process_info,
 	return 0;
 }
 
+static struct amdgpu_vm *drm_priv_to_vm(struct drm_file *drm_priv)
+{
+	struct amdgpu_fpriv *fpriv = drm_priv->driver_priv;
+
+	return &fpriv->vm;
+}
+
 static int init_kfd_vm(struct amdgpu_vm *vm, void **process_info,
 		       struct dma_fence **ef)
 {
@@ -1036,15 +1043,19 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void **process_info,
 
 int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct kgd_dev *kgd,
 					   struct file *filp, u32 pasid,
-					   void **vm, void **process_info,
+					   void **process_info,
 					   struct dma_fence **ef)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	struct drm_file *drm_priv = filp->private_data;
-	struct amdgpu_fpriv *drv_priv = drm_priv->driver_priv;
-	struct amdgpu_vm *avm = &drv_priv->vm;
+	struct amdgpu_fpriv *drv_priv;
+	struct amdgpu_vm *avm;
 	int ret;
 
+	ret = amdgpu_file_to_fpriv(filp, &drv_priv);
+	if (ret)
+		return ret;
+	avm = &drv_priv->vm;
+
 	/* Already a compute VM? */
 	if (avm->process_info)
 		return -EINVAL;
@@ -1059,7 +1070,7 @@ int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct kgd_dev *kgd,
 	if (ret)
 		return ret;
 
-	*vm = (void *)avm;
+	amdgpu_vm_set_task_info(avm);
 
 	return 0;
 }
@@ -1100,15 +1111,17 @@ void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev,
 	}
 }
 
-void amdgpu_amdkfd_gpuvm_release_process_vm(struct kgd_dev *kgd, void *vm)
+void amdgpu_amdkfd_gpuvm_release_process_vm(struct kgd_dev *kgd, void *drm_priv)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	struct amdgpu_vm *avm = (struct amdgpu_vm *)vm;
+	struct amdgpu_vm *avm;
 
-	if (WARN_ON(!kgd || !vm))
+	if (WARN_ON(!kgd || !drm_priv))
 		return;
 
-	pr_debug("Releasing process vm %p\n", vm);
+	avm = drm_priv_to_vm(drm_priv);
+
+	pr_debug("Releasing process vm %p\n", avm);
 
 	/* The original pasid of amdgpu vm has already been
 	 * released during making a amdgpu vm to a compute vm
@@ -1119,9 +1132,9 @@ void amdgpu_amdkfd_gpuvm_release_process_vm(struct kgd_dev *kgd, void *vm)
 	amdgpu_vm_release_compute(adev, avm);
 }
 
-uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *vm)
+uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *drm_priv)
 {
-	struct amdgpu_vm *avm = (struct amdgpu_vm *)vm;
+	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
 	struct amdgpu_bo *pd = avm->root.base.bo;
 	struct amdgpu_device *adev = amdgpu_ttm_adev(pd->tbo.bdev);
 
@@ -1132,11 +1145,11 @@ uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *vm)
 
 int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 		struct kgd_dev *kgd, uint64_t va, uint64_t size,
-		void *vm, struct kgd_mem **mem,
+		void *drm_priv, struct kgd_mem **mem,
 		uint64_t *offset, uint32_t flags)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	struct amdgpu_vm *avm = (struct amdgpu_vm *)vm;
+	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
 	enum ttm_bo_type bo_type = ttm_bo_type_device;
 	struct sg_table *sg = NULL;
 	uint64_t user_addr = 0;
@@ -1347,10 +1360,10 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 }
 
 int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
-		struct kgd_dev *kgd, struct kgd_mem *mem, void *vm)
+		struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	struct amdgpu_vm *avm = (struct amdgpu_vm *)vm;
+	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
 	int ret;
 	struct amdgpu_bo *bo;
 	uint32_t domain;
@@ -1391,9 +1404,9 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 	pr_debug("Map VA 0x%llx - 0x%llx to vm %p domain %s\n",
 			mem->va,
 			mem->va + bo_size * (1 + mem->aql_queue),
-			vm, domain_string(domain));
+			avm, domain_string(domain));
 
-	ret = reserve_bo_and_vm(mem, vm, &ctx);
+	ret = reserve_bo_and_vm(mem, avm, &ctx);
 	if (unlikely(ret))
 		goto out;
 
@@ -1437,7 +1450,7 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 	}
 
 	list_for_each_entry(entry, &mem->bo_va_list, bo_list) {
-		if (entry->bo_va->base.vm == vm && !entry->is_mapped) {
+		if (entry->bo_va->base.vm == avm && !entry->is_mapped) {
 			pr_debug("\t map VA 0x%llx - 0x%llx in entry %p\n",
 					entry->va, entry->va + bo_size,
 					entry);
@@ -1449,7 +1462,7 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 				goto map_bo_to_gpuvm_failed;
 			}
 
-			ret = vm_update_pds(vm, ctx.sync);
+			ret = vm_update_pds(avm, ctx.sync);
 			if (ret) {
 				pr_err("Failed to update page directories\n");
 				goto map_bo_to_gpuvm_failed;
@@ -1485,11 +1498,11 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 }
 
 int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
-		struct kgd_dev *kgd, struct kgd_mem *mem, void *vm)
+		struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	struct amdkfd_process_info *process_info =
-		((struct amdgpu_vm *)vm)->process_info;
+	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
+	struct amdkfd_process_info *process_info = avm->process_info;
 	unsigned long bo_size = mem->bo->tbo.base.size;
 	struct kfd_bo_va_list *entry;
 	struct bo_vm_reservation_context ctx;
@@ -1497,7 +1510,7 @@ int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
 
 	mutex_lock(&mem->lock);
 
-	ret = reserve_bo_and_cond_vms(mem, vm, BO_VM_MAPPED, &ctx);
+	ret = reserve_bo_and_cond_vms(mem, avm, BO_VM_MAPPED, &ctx);
 	if (unlikely(ret))
 		goto out;
 	/* If no VMs were reserved, it means the BO wasn't actually mapped */
@@ -1506,17 +1519,17 @@ int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
 		goto unreserve_out;
 	}
 
-	ret = vm_validate_pt_pd_bos((struct amdgpu_vm *)vm);
+	ret = vm_validate_pt_pd_bos(avm);
 	if (unlikely(ret))
 		goto unreserve_out;
 
 	pr_debug("Unmap VA 0x%llx - 0x%llx from vm %p\n",
 		mem->va,
 		mem->va + bo_size * (1 + mem->aql_queue),
-		vm);
+		avm);
 
 	list_for_each_entry(entry, &mem->bo_va_list, bo_list) {
-		if (entry->bo_va->base.vm == vm && entry->is_mapped) {
+		if (entry->bo_va->base.vm == avm && entry->is_mapped) {
 			pr_debug("\t unmap VA 0x%llx - 0x%llx from entry %p\n",
 					entry->va,
 					entry->va + bo_size,
@@ -1642,14 +1655,14 @@ int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct kgd_dev *kgd,
 
 int amdgpu_amdkfd_gpuvm_import_dmabuf(struct kgd_dev *kgd,
 				      struct dma_buf *dma_buf,
-				      uint64_t va, void *vm,
+				      uint64_t va, void *drm_priv,
 				      struct kgd_mem **mem, uint64_t *size,
 				      uint64_t *mmap_offset)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
+	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
 	struct drm_gem_object *obj;
 	struct amdgpu_bo *bo;
-	struct amdgpu_vm *avm = (struct amdgpu_vm *)vm;
 
 	if (dma_buf->ops != &amdgpu_dmabuf_ops)
 		/* Can't handle non-graphics buffers */

drivers/gpu/drm/amd/amdkfd/kfd_chardev.c

@@ -1297,7 +1297,7 @@ static int kfd_ioctl_alloc_memory_of_gpu(struct file *filep,
 
 	err = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 		dev->kgd, args->va_addr, args->size,
-		pdd->vm, (struct kgd_mem **) &mem, &offset,
+		pdd->drm_priv, (struct kgd_mem **) &mem, &offset,
 		flags);
 
 	if (err)
@@ -1448,7 +1448,7 @@ static int kfd_ioctl_map_memory_to_gpu(struct file *filep,
 			goto get_mem_obj_from_handle_failed;
 		}
 		err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
-			peer->kgd, (struct kgd_mem *)mem, peer_pdd->vm);
+			peer->kgd, (struct kgd_mem *)mem, peer_pdd->drm_priv);
 		if (err) {
 			pr_err("Failed to map to gpu %d/%d\n",
 			       i, args->n_devices);
@@ -1555,7 +1555,7 @@ static int kfd_ioctl_unmap_memory_from_gpu(struct file *filep,
 			goto get_mem_obj_from_handle_failed;
 		}
 		err = amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
-			peer->kgd, (struct kgd_mem *)mem, peer_pdd->vm);
+			peer->kgd, (struct kgd_mem *)mem, peer_pdd->drm_priv);
 		if (err) {
 			pr_err("Failed to unmap from gpu %d/%d\n",
 			       i, args->n_devices);
@@ -1701,7 +1701,7 @@ static int kfd_ioctl_import_dmabuf(struct file *filep,
 	}
 
 	r = amdgpu_amdkfd_gpuvm_import_dmabuf(dev->kgd, dmabuf,
-					      args->va_addr, pdd->vm,
+					      args->va_addr, pdd->drm_priv,
 					      (struct kgd_mem **)&mem, &size,
 					      NULL);
 	if (r)

drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c

@@ -738,7 +738,7 @@ static int restore_process_queues_nocpsch(struct device_queue_manager *dqm,
 
 	pdd = qpd_to_pdd(qpd);
 	/* Retrieve PD base */
-	pd_base = amdgpu_amdkfd_gpuvm_get_process_page_dir(pdd->vm);
+	pd_base = amdgpu_amdkfd_gpuvm_get_process_page_dir(pdd->drm_priv);
 
 	dqm_lock(dqm);
 	if (WARN_ON_ONCE(!qpd->evicted)) /* already restored, do nothing */
@@ -821,7 +821,7 @@ static int restore_process_queues_cpsch(struct device_queue_manager *dqm,
 
 	pdd = qpd_to_pdd(qpd);
 	/* Retrieve PD base */
-	pd_base = amdgpu_amdkfd_gpuvm_get_process_page_dir(pdd->vm);
+	pd_base = amdgpu_amdkfd_gpuvm_get_process_page_dir(pdd->drm_priv);
 
 	dqm_lock(dqm);
 	if (WARN_ON_ONCE(!qpd->evicted)) /* already restored, do nothing */
@@ -873,7 +873,7 @@ static int register_process(struct device_queue_manager *dqm,
 
 	pdd = qpd_to_pdd(qpd);
 	/* Retrieve PD base */
-	pd_base = amdgpu_amdkfd_gpuvm_get_process_page_dir(pdd->vm);
+	pd_base = amdgpu_amdkfd_gpuvm_get_process_page_dir(pdd->drm_priv);
 
 	dqm_lock(dqm);
 	list_add(&n->list, &dqm->queues);

drivers/gpu/drm/amd/amdkfd/kfd_priv.h

@@ -669,7 +669,7 @@ struct kfd_process_device {
 
 	/* VM context for GPUVM allocations */
 	struct file *drm_file;
-	void *vm;
+	void *drm_priv;
 
 	/* GPUVM allocations storage */
 	struct idr alloc_idr;

drivers/gpu/drm/amd/amdkfd/kfd_process.c

@@ -647,7 +647,7 @@ static void kfd_process_free_gpuvm(struct kgd_mem *mem,
 {
 	struct kfd_dev *dev = pdd->dev;
 
-	amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(dev->kgd, mem, pdd->vm);
+	amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(dev->kgd, mem, pdd->drm_priv);
 	amdgpu_amdkfd_gpuvm_free_memory_of_gpu(dev->kgd, mem, NULL);
 }
 
@@ -667,11 +667,11 @@ static int kfd_process_alloc_gpuvm(struct kfd_process_device *pdd,
 	int err;
 
 	err = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(kdev->kgd, gpu_va, size,
-						 pdd->vm, &mem, NULL, flags);
+						 pdd->drm_priv, &mem, NULL, flags);
 	if (err)
 		goto err_alloc_mem;
 
-	err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu(kdev->kgd, mem, pdd->vm);
+	err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu(kdev->kgd, mem, pdd->drm_priv);
 	if (err)
 		goto err_map_mem;
 
@@ -901,10 +901,10 @@ static void kfd_process_device_free_bos(struct kfd_process_device *pdd)
 		for (i = 0; i < p->n_pdds; i++) {
 			struct kfd_process_device *peer_pdd = p->pdds[i];
 
-			if (!peer_pdd->vm)
+			if (!peer_pdd->drm_priv)
 				continue;
 			amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
-				peer_pdd->dev->kgd, mem, peer_pdd->vm);
+				peer_pdd->dev->kgd, mem, peer_pdd->drm_priv);
 		}
 
 		amdgpu_amdkfd_gpuvm_free_memory_of_gpu(pdd->dev->kgd, mem, NULL);
@@ -932,7 +932,7 @@ static void kfd_process_destroy_pdds(struct kfd_process *p)
 
 		if (pdd->drm_file) {
 			amdgpu_amdkfd_gpuvm_release_process_vm(
-					pdd->dev->kgd, pdd->vm);
+					pdd->dev->kgd, pdd->drm_priv);
 			fput(pdd->drm_file);
 		}
 
@@ -1375,7 +1375,7 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd,
 	if (!drm_file)
 		return -EINVAL;
 
-	if (pdd->vm)
+	if (pdd->drm_priv)
 		return -EBUSY;
 
 	p = pdd->process;
@@ -1383,13 +1383,12 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd,
 
 	ret = amdgpu_amdkfd_gpuvm_acquire_process_vm(
 		dev->kgd, drm_file, p->pasid,
-		&pdd->vm, &p->kgd_process_info, &p->ef);
+		&p->kgd_process_info, &p->ef);
 	if (ret) {
 		pr_err("Failed to create process VM object\n");
 		return ret;
 	}
-
-	amdgpu_vm_set_task_info(pdd->vm);
+	pdd->drm_priv = drm_file->private_data;
 
 	ret = kfd_process_device_reserve_ib_mem(pdd);
 	if (ret)
@@ -1405,7 +1404,7 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd,
 err_init_cwsr:
 err_reserve_ib_mem:
 	kfd_process_device_free_bos(pdd);
-	pdd->vm = NULL;
+	pdd->drm_priv = NULL;
 
 	return ret;
 }
@@ -1429,7 +1428,7 @@ struct kfd_process_device *kfd_bind_process_to_device(struct kfd_dev *dev,
 		return ERR_PTR(-ENOMEM);
 	}
 
-	if (!pdd->vm)
+	if (!pdd->drm_priv)
 		return ERR_PTR(-ENODEV);
 
 	/*