[ofa-general] Non-Signaled RDMA Read (cxgb3)

[Steve Wise]

Roland Dreier wrote:
>  > Note that I _could_ add code in the driver/lib poll routine to "eat"
>  > these cqes.  But there is a fundamental problem in that if the CQ is
>  > armed, then the HW _will_ generate an interrupt in this case.  This
>  > leads to a "false notification" which breaks the semantics of the
>  > notification mechanism.  I'm not sure how bad that is?  To the
>  > application, it would receive a notification, then go poll and get
>  > nothing.
>
> There's also the issue that these completions take CQ entries... so if
> an app provisions a CQ for some amount of completions and then assumes
> that posting 100 RDMA reads doesn't use any CQ entries, things break.
>
>   
Note such an app could overflow the cq if the reads fail.  Granted that 
causes the connection to come crashing down anyway so maybe its ok.

> Not sure if there is a good way to deal with this quirk really.
>
>   
I tend to agree.  That's why I haven't done it. :) 

Stevo