[rdma-rc] RDMA/mlx5: Fix a race for DMABUF MR which can lead to CQE with error

Commit Message

Leon Romanovsky Feb. 3, 2025, 12:50 p.m. UTC

From: Yishai Hadas <[e-mail]>

This patch addresses a potential race condition for a DMABUF MR that can
result in a CQE with an error on the UMR QP.

During the __mlx5_ib_dereg_mr() flow, the following sequence of calls
occurs:
mlx5_revoke_mr()
mlx5r_umr_revoke_mr()
mlx5r_umr_post_send_wait()
At this point, the lkey is freed from the hardware's perspective.

However, concurrently, mlx5_ib_dmabuf_invalidate_cb() might be triggered
by another task attempting to invalidate the MR having that freed lkey.

Since the lkey has already been freed, this can lead to a CQE error,
causing the UMR QP to enter an error state.

To resolve this race condition, the dma_resv_lock() which was hold as
part of the mlx5_ib_dmabuf_invalidate_cb() is now also acquired as part
of the mlx5_revoke_mr() scope.

Upon a successful revoke, we set umem_dmabuf->private which points to
that MR to NULL, preventing any further invalidation attempts on its
lkey.

Fixes: e6fb246ccafb ("RDMA/mlx5: Consolidate MR destruction to mlx5_ib_dereg_mr()")
Signed-off-by: Yishai Hadas <[e-mail]>
Reviewed-by: Artemy Kovalyov <[e-mail]>
Signed-off-by: Leon Romanovsky <[e-mail]>
---
 drivers/infiniband/hw/mlx5/mr.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

Comments

Leon Romanovsky Feb. 6, 2025, 8:44 a.m. UTC | #1

On Mon, 03 Feb 2025 14:50:59 +0200, Leon Romanovsky wrote:
> This patch addresses a potential race condition for a DMABUF MR that can
> result in a CQE with an error on the UMR QP.
> 
> During the __mlx5_ib_dereg_mr() flow, the following sequence of calls
> occurs:
> mlx5_revoke_mr()
> mlx5r_umr_revoke_mr()
> mlx5r_umr_post_send_wait()
> At this point, the lkey is freed from the hardware's perspective.
> 
> [...]

Applied, thanks!

[1/1] RDMA/mlx5: Fix a race for DMABUF MR which can lead to CQE with error
      https://git.kernel.org/rdma/rdma/c/cc668a11e6ac8a

Best regards,

Patch

diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c
index 9e8993ea68e7..32ab489517f6 100644
--- a/drivers/infiniband/hw/mlx5/mr.c
+++ b/drivers/infiniband/hw/mlx5/mr.c
@@ -1605,7 +1605,7 @@  static void mlx5_ib_dmabuf_invalidate_cb(struct dma_buf_attachment *attach)
 
 	dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv);
 
-	if (!umem_dmabuf->sgt)
+	if (!umem_dmabuf->sgt || !mr)
 		return;
 
 	mlx5r_umr_update_mr_pas(mr, MLX5_IB_UPD_XLT_ZAP);
@@ -2088,11 +2088,16 @@  static int mlx5_revoke_mr(struct mlx5_ib_mr *mr)
 	struct mlx5_ib_dev *dev = to_mdev(mr->ibmr.device);
 	struct mlx5_cache_ent *ent = mr->mmkey.cache_ent;
 	bool is_odp = is_odp_mr(mr);
+	bool is_odp_dma_buf = is_dmabuf_mr(mr) &&
+			!to_ib_umem_dmabuf(mr->umem)->pinned;
 	int ret = 0;
 
 	if (is_odp)
 		mutex_lock(&to_ib_umem_odp(mr->umem)->umem_mutex);
 
+	if (is_odp_dma_buf)
+		dma_resv_lock(to_ib_umem_dmabuf(mr->umem)->attach->dmabuf->resv, NULL);
+
 	if (mr->mmkey.cacheable && !mlx5r_umr_revoke_mr(mr) && !cache_ent_find_and_store(dev, mr)) {
 		ent = mr->mmkey.cache_ent;
 		/* upon storing to a clean temp entry - schedule its cleanup */
@@ -2128,6 +2133,12 @@  static int mlx5_revoke_mr(struct mlx5_ib_mr *mr)
 		mutex_unlock(&to_ib_umem_odp(mr->umem)->umem_mutex);
 	}
 
+	if (is_odp_dma_buf) {
+		if (!ret)
+			to_ib_umem_dmabuf(mr->umem)->private = NULL;
+		dma_resv_unlock(to_ib_umem_dmabuf(mr->umem)->attach->dmabuf->resv);
+	}
+
 	return ret;
 }