broken support for UC in rdma_cm

broken support for UC in rdma_cm

[Eli Cohen]

We got failing tests in our regression that use UC transport from
userspace. Looking at cma_connect_ib(), I see that it uses IB_QPT_RC
in the CM request instead of using whatever the transport type of the
requesting QP. There is no easy way to retrieve that information too
since the QP is created in userspace. We do have the QP number. One
way to address this problem is to have an API that returns a pointer
to struct ib_qp given the ib device and the QP number. Another way
would be to pass qp_type through the call to rdma_init_qp_attr().
Thoughts?

Re: broken support for UC in rdma_cm

[Eli Cohen]

I just noticed that Vlad already opened a bugzilla bug (1874) on this.
I quote Sean's response:

RDMA CM supports UD and RC QPs (port spaces UDP/TCP) only.  Support
for UC QPs should come from another port space.

This makes sense to me. Still we need to address the issues I raised
below. Sean, are you going to fix this?

On Mon, Mar 15, 2010 at 01:12:28PM +0200, Eli Cohen wrote:
> We got failing tests in our regression that use UC transport from
> userspace. Looking at cma_connect_ib(), I see that it uses IB_QPT_RC
> in the CM request instead of using whatever the transport type of the
> requesting QP. There is no easy way to retrieve that information too
> since the QP is created in userspace. We do have the QP number. One
> way to address this problem is to have an API that returns a pointer
> to struct ib_qp given the ib device and the QP number. Another way
> would be to pass qp_type through the call to rdma_init_qp_attr().
> Thoughts?
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: broken support for UC in rdma_cm

[Sean Hefty]

>I just noticed that Vlad already opened a bugzilla bug (1874) on this.
>I quote Sean's response:
>
>RDMA CM supports UD and RC QPs (port spaces UDP/TCP) only.  Support
>for UC QPs should come from another port space.
>
>This makes sense to me. Still we need to address the issues I raised
>below. Sean, are you going to fix this?

This is really a new feature request, which I may not have time to address.  It
depends on the scope of the changes that are needed.

If both sides try using UC QPs, do you know what specific setting causes the
connection to fail?
 
- Sean

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: broken support for UC in rdma_cm

[Eli Cohen]

On Mon, Mar 15, 2010 at 08:38:00AM -0700, Sean Hefty wrote:
> 
> This is really a new feature request, which I may not have time to address.  It
> depends on the scope of the changes that are needed.
> 
> If both sides try using UC QPs, do you know what specific setting causes the
> connection to fail?

The actual failure is to modify the QP from init to rtr since the
attribute mask sets IB_QP_MAX_DEST_RD_ATOMIC and IB_QP_MIN_RNR_TIMER
which are invalid attributes in UC. The flags are set in
cm_init_qp_rtr_attr() since the transport is IB_QPT_RC. 
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html