vdpa/mlx5: Add support for running with virtio_vdpa

In order to support running vdpa using vritio_vdpa driver, we need  to
create a different kind of MR, one that has 1:1 mapping, since the
addresses referring to virtqueues are dma addresses.

We create the 1:1 MR in mlx5_vdpa_dev_add() only in case firmware
supports the general capability umem_uid_0. The reason for that is that
1:1 MRs must be created with uid == 0 while virtqueue objects can be
created with uid == 0 only when the firmware capability is on.

If the set_map() callback is called with new translations provided
through iotlb, the driver will destroy the 1:1 MR and create a regular
one.

Signed-off-by: Eli Cohen <elic@nvidia.com>
Link: https://lore.kernel.org/r/20210602085854.62690-1-elic@nvidia.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>

drivers/vdpa/mlx5/core/mlx5_vdpa.h

@@ -35,6 +35,7 @@ struct mlx5_vdpa_mr {
 
 	/* serialize mkey creation and destruction */
 	struct mutex mkey_mtx;
+	bool user_mr;
 };
 
 struct mlx5_vdpa_resources {

drivers/vdpa/mlx5/core/mr.c

@@ -355,7 +355,7 @@ err_alloc:
  * indirect memory key that provides access to the enitre address space given
  * by iotlb.
  */
-static int _mlx5_vdpa_create_mr(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
+static int create_user_mr(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
 {
 	struct mlx5_vdpa_mr *mr = &mvdev->mr;
 	struct mlx5_vdpa_direct_mr *dmr;
@@ -369,9 +369,6 @@ static int _mlx5_vdpa_create_mr(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb
 	int err = 0;
 	int nnuls;
 
-	if (mr->initialized)
-		return 0;
-
 	INIT_LIST_HEAD(&mr->head);
 	for (map = vhost_iotlb_itree_first(iotlb, start, last); map;
 	     map = vhost_iotlb_itree_next(map, start, last)) {
@@ -409,7 +406,7 @@ static int _mlx5_vdpa_create_mr(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb
 	if (err)
 		goto err_chain;
 
-	mr->initialized = true;
+	mr->user_mr = true;
 	return 0;
 
 err_chain:
@@ -421,33 +418,94 @@ err_chain:
 	return err;
 }
 
-int mlx5_vdpa_create_mr(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
+static int create_dma_mr(struct mlx5_vdpa_dev *mvdev, struct mlx5_vdpa_mr *mr)
+{
+	int inlen = MLX5_ST_SZ_BYTES(create_mkey_in);
+	void *mkc;
+	u32 *in;
+	int err;
+
+	in = kzalloc(inlen, GFP_KERNEL);
+	if (!in)
+		return -ENOMEM;
+
+	mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
+
+	MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_PA);
+	MLX5_SET(mkc, mkc, length64, 1);
+	MLX5_SET(mkc, mkc, lw, 1);
+	MLX5_SET(mkc, mkc, lr, 1);
+	MLX5_SET(mkc, mkc, pd, mvdev->res.pdn);
+	MLX5_SET(mkc, mkc, qpn, 0xffffff);
+
+	err = mlx5_vdpa_create_mkey(mvdev, &mr->mkey, in, inlen);
+	if (!err)
+		mr->user_mr = false;
+
+	kfree(in);
+	return err;
+}
+
+static void destroy_dma_mr(struct mlx5_vdpa_dev *mvdev, struct mlx5_vdpa_mr *mr)
+{
+	mlx5_vdpa_destroy_mkey(mvdev, &mr->mkey);
+}
+
+static int _mlx5_vdpa_create_mr(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
 {
 	struct mlx5_vdpa_mr *mr = &mvdev->mr;
 	int err;
 
-	mutex_lock(&mr->mkey_mtx);
+	if (mr->initialized)
-	err = _mlx5_vdpa_create_mr(mvdev, iotlb);
+		return 0;
-	mutex_unlock(&mr->mkey_mtx);
+
+	if (iotlb)
+		err = create_user_mr(mvdev, iotlb);
+	else
+		err = create_dma_mr(mvdev, mr);
+
+	if (!err)
+		mr->initialized = true;
+
 	return err;
 }
 
-void mlx5_vdpa_destroy_mr(struct mlx5_vdpa_dev *mvdev)
+int mlx5_vdpa_create_mr(struct mlx5_vdpa_dev *mvdev, struct vhost_iotlb *iotlb)
+{
+	int err;
+
+	mutex_lock(&mvdev->mr.mkey_mtx);
+	err = _mlx5_vdpa_create_mr(mvdev, iotlb);
+	mutex_unlock(&mvdev->mr.mkey_mtx);
+	return err;
+}
+
+static void destroy_user_mr(struct mlx5_vdpa_dev *mvdev, struct mlx5_vdpa_mr *mr)
 {
-	struct mlx5_vdpa_mr *mr = &mvdev->mr;
 	struct mlx5_vdpa_direct_mr *dmr;
 	struct mlx5_vdpa_direct_mr *n;
 
-	mutex_lock(&mr->mkey_mtx);
-	if (!mr->initialized)
-		goto out;
-
 	destroy_indirect_key(mvdev, mr);
 	list_for_each_entry_safe_reverse(dmr, n, &mr->head, list) {
 		list_del_init(&dmr->list);
 		unmap_direct_mr(mvdev, dmr);
 		kfree(dmr);
 	}
+}
+
+void mlx5_vdpa_destroy_mr(struct mlx5_vdpa_dev *mvdev)
+{
+	struct mlx5_vdpa_mr *mr = &mvdev->mr;
+
+	mutex_lock(&mr->mkey_mtx);
+	if (!mr->initialized)
+		goto out;
+
+	if (mr->user_mr)
+		destroy_user_mr(mvdev, mr);
+	else
+		destroy_dma_mr(mvdev, mr);
+
 	memset(mr, 0, sizeof(*mr));
 	mr->initialized = false;
 out:

drivers/vdpa/mlx5/net/mlx5_vnet.c

@@ -1781,6 +1781,10 @@ static void mlx5_vdpa_set_status(struct vdpa_device *vdev, u8 status)
 		ndev->mvdev.status = 0;
 		ndev->mvdev.mlx_features = 0;
 		++mvdev->generation;
+		if (MLX5_CAP_GEN(mvdev->mdev, umem_uid_0)) {
+			if (mlx5_vdpa_create_mr(mvdev, NULL))
+				mlx5_vdpa_warn(mvdev, "create MR failed\n");
+		}
 		return;
 	}
 
@@ -1861,6 +1865,7 @@ static void mlx5_vdpa_free(struct vdpa_device *vdev)
 	ndev = to_mlx5_vdpa_ndev(mvdev);
 
 	free_resources(ndev);
+	mlx5_vdpa_destroy_mr(mvdev);
 	if (!is_zero_ether_addr(ndev->config.mac)) {
 		pfmdev = pci_get_drvdata(pci_physfn(mvdev->mdev->pdev));
 		mlx5_mpfs_del_mac(pfmdev, ndev->config.mac);
@@ -2037,9 +2042,15 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name)
 	if (err)
 		goto err_mpfs;
 
+	if (MLX5_CAP_GEN(mvdev->mdev, umem_uid_0)) {
+		err = mlx5_vdpa_create_mr(mvdev, NULL);
+		if (err)
+			goto err_res;
+	}
+
 	err = alloc_resources(ndev);
 	if (err)
-		goto err_res;
+		goto err_mr;
 
 	mvdev->vdev.mdev = &mgtdev->mgtdev;
 	err = _vdpa_register_device(&mvdev->vdev, 2 * mlx5_vdpa_max_qps(max_vqs));
@@ -2051,6 +2062,8 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name)
 
 err_reg:
 	free_resources(ndev);
+err_mr:
+	mlx5_vdpa_destroy_mr(mvdev);
 err_res:
 	mlx5_vdpa_free_resources(&ndev->mvdev);
 err_mpfs: