diff mbox series

[for-rc] RDMA/core: Update CMA destination address on rdma_resolve_addr

Message ID 20230712234133.1343-1-shiraz.saleem@intel.com (mailing list archive)
State Accepted
Headers show
Series [for-rc] RDMA/core: Update CMA destination address on rdma_resolve_addr | expand

Commit Message

Shiraz Saleem July 12, 2023, 11:41 p.m. UTC
8d037973d48c ("RDMA/core: Refactor rdma_bind_addr") intoduces as regression
on irdma devices on certain tests which uses rdma CM, such as cmtime.

No connections can be established with the MAD QP experiences a fatal
error on the active side.

The cma destination address is not updated with the dst_addr when ULP
on active side calls rdma_bind_addr followed by rdma_resolve_addr.
The id_priv state is 'bound' in resolve_prepare_src and update is skipped.

This leaves the dgid passed into irdma driver to create an Address Handle
(AH) for the MAD QP at 0. The create AH descriptor as well as the ARP cache
entry is invalid and HW throws an asynchronous events as result.

[ 1207.656888] resolve_prepare_src caller: ucma_resolve_addr+0xff/0x170 [rdma_ucm] daddr=200.0.4.28 id_priv->state=7
[....]
[ 1207.680362] ice 0000:07:00.1 rocep7s0f1: caller: irdma_create_ah+0x3e/0x70 [irdma] ah_id=0 arp_idx=0 dest_ip=0.0.0.0
destMAC=00:00:64:ca:b7:52 ipvalid=1 raw=0000:0000:0000:0000:0000:ffff:0000:0000
[ 1207.682077] ice 0000:07:00.1 rocep7s0f1: abnormal ae_id = 0x401 bool qp=1 qp_id = 1, ae_src=5
[ 1207.691657] infiniband rocep7s0f1: Fatal error (1) on MAD QP (1)

Fix this by updating the CMA destination address when the ULP calls
a resolve address with the CM state already bound.

Fixes: 8d037973d48c ("RDMA/core: Refactor rdma_bind_addr")
Signed-off-by: Shiraz Saleem <shiraz.saleem@intel.com>
---
 drivers/infiniband/core/cma.c | 2 ++
 1 file changed, 2 insertions(+)

Comments

Leon Romanovsky July 16, 2023, 11:41 a.m. UTC | #1
On Wed, 12 Jul 2023 18:41:33 -0500, Shiraz Saleem wrote:
> 8d037973d48c ("RDMA/core: Refactor rdma_bind_addr") intoduces as regression
> on irdma devices on certain tests which uses rdma CM, such as cmtime.
> 
> No connections can be established with the MAD QP experiences a fatal
> error on the active side.
> 
> The cma destination address is not updated with the dst_addr when ULP
> on active side calls rdma_bind_addr followed by rdma_resolve_addr.
> The id_priv state is 'bound' in resolve_prepare_src and update is skipped.
> 
> [...]

Applied, thanks!

[1/1] RDMA/core: Update CMA destination address on rdma_resolve_addr
      https://git.kernel.org/rdma/rdma/c/21b09f979229d9

Best regards,
diff mbox series

Patch

diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index 1ee87c3..9891c7d 100644
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -4062,6 +4062,8 @@  static int resolve_prepare_src(struct rdma_id_private *id_priv,
 					   RDMA_CM_ADDR_QUERY)))
 			return -EINVAL;
 
+	} else {
+		memcpy(cma_dst_addr(id_priv), dst_addr, rdma_addr_size(dst_addr));
 	}
 
 	if (cma_family(id_priv) != dst_addr->sa_family) {