Re: [PATCH v1 03/18] xprtrdma: Remove completion polling budgets

[Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>]

On Thu, Oct 01, 2015 at 12:37:36PM -0400, Chuck Lever wrote:

> A missed WC will result in an RPC/RDMA transport deadlock. In
> fact that is the reason for this particular patch (although
> it addresses only one source of missed WCs). So I would like
> to see that there are no windows here.

WCs are never missed.

The issue is a race where re-arming the CQ might not work, meaning you
don't get an event.

You can certainly use arrays with poll_cq. There is no race in the API
here.

But you have to use the IB_CQ_REPORT_MISSED_EVENTS scheme to guarantee
the CQ is actually armed or continue to loop again.

Basically you have to loop until ib_req_notify_cq succeeds.

Any driver that doesn't support this is broken, do we know of any?

while (1) {
  struct ib_wc wcs[100];
  int rc = ib_poll_cq(cw, NELEMS(wcs), wcs);

  .. process rc wcs ..

  if (rc != NELEMS(wcs))
    if (ib_req_notify_cq(cq, IB_CQ_NEXT_COMP |
        IB_CQ_REPORT_MISSED_EVENTS) == 0)
     break;
}

API wise, we should probably look at forcing
IB_CQ_REPORT_MISSED_EVENTS on and dropping the flag.

Jason
--
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