diff mbox series

[for-rc] RDMA/mlx4: Fix return value when QP type isn't supported

Message ID 20200922134429.130255-1-kamalheib1@gmail.com (mailing list archive)
State Rejected
Headers show
Series [for-rc] RDMA/mlx4: Fix return value when QP type isn't supported | expand

Commit Message

Kamal Heib Sept. 22, 2020, 1:44 p.m. UTC
The proper return code is "-EOPNOTSUPP" when trying to modify a raw
packet QP over an IB port.

Fixes: 3987a2d3193c ("IB/mlx4: Add raw packet QP support")
Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
---
 drivers/infiniband/hw/mlx4/qp.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

Comments

Jason Gunthorpe Sept. 22, 2020, 11:30 p.m. UTC | #1
On Tue, Sep 22, 2020 at 04:44:29PM +0300, Kamal Heib wrote:
> The proper return code is "-EOPNOTSUPP" when trying to modify a raw
> packet QP over an IB port.
> 
> Fixes: 3987a2d3193c ("IB/mlx4: Add raw packet QP support")
> Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
> ---
>  drivers/infiniband/hw/mlx4/qp.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)

Is it? Why? EOPNOTSUPP should be used by the uverbs layer to indicate
that the operation is not implemented in the driver

Calling modify_qp against a RAW_QP when it is not supported by spec is
EINVAL?

Jason
Kamal Heib Sept. 27, 2020, 1:04 p.m. UTC | #2
On Tue, Sep 22, 2020 at 08:30:03PM -0300, Jason Gunthorpe wrote:
> On Tue, Sep 22, 2020 at 04:44:29PM +0300, Kamal Heib wrote:
> > The proper return code is "-EOPNOTSUPP" when trying to modify a raw
> > packet QP over an IB port.
> > 
> > Fixes: 3987a2d3193c ("IB/mlx4: Add raw packet QP support")
> > Signed-off-by: Kamal Heib <kamalheib1@gmail.com>
> > ---
> >  drivers/infiniband/hw/mlx4/qp.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> Is it? Why? EOPNOTSUPP should be used by the uverbs layer to indicate
> that the operation is not implemented in the driver
> 
> Calling modify_qp against a RAW_QP when it is not supported by spec is
> EINVAL?
> 
> Jason

Please drop this patch...

The idea behind it was to avoid the pyverbs test_qp.py failure when trying
to create a RAW QP over an mlx4 IB port, this was fixed using the
following change in the test_qp.py:

https://github.com/linux-rdma/rdma-core/pull/833

Thanks,
Kamal
diff mbox series

Patch

diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 2975f350b9fd..3c6ed7a7d407 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -2785,8 +2785,10 @@  static int _mlx4_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 
 	if ((attr_mask & IB_QP_PORT) && (ibqp->qp_type == IB_QPT_RAW_PACKET) &&
 	    (rdma_port_get_link_layer(&dev->ib_dev, attr->port_num) !=
-	     IB_LINK_LAYER_ETHERNET))
+	     IB_LINK_LAYER_ETHERNET)) {
+		err = -EOPNOTSUPP;
 		goto out;
+	}
 
 	if (attr_mask & IB_QP_PKEY_INDEX) {
 		int p = attr_mask & IB_QP_PORT ? attr->port_num : qp->port;