[openib-general] [PATCH] add cq error events

[Caitlin Bestler]

If the semantics were defined such that an overrun meant
that an event had been lost, but that the CQ was still
intact, then the user can definitely adjust and continue.

The QP for which the event was lost will have its connection
broken, which could allow many applications to determine
what the lost event was. If they were then able to recover
the lost resources the application could simple destroy
the QP, recreate it, and continue (or perhaps only reset
it).

On the other hand if the CQ is corrupt there is very little
that the application can do to recover. They'll have to tear
down every connection that uses the CQ and recreate it.

The distinction between "something went wrong and some
of *your* resources may be in a funny state" and "something
went wrong and some of *my* resources may be in a funny
state" can be very important. The user may be able to
repair their resources, but not those of the RDMA device.


  

> -----Original Message-----
> From: openib-general-bounces at openib.org 
> [mailto:openib-general-bounces at openib.org] On Behalf Of Sean Hefty
> Sent: Thursday, September 22, 2005 2:06 PM
> To: Michael S. Tsirkin
> Cc: Roland Dreier; openib-general at openib.org
> Subject: Re: [openib-general] [PATCH] add cq error events
> 
> Michael S. Tsirkin wrote:
> > As a side note, the spec says: "Two types of CQ errors can 
> occur: the 
> > CQ can overrun or it can become inaccessible": I wander 
> whether this 
> > should be interpreted in a sense that that there should be 
> two types 
> > of events: IB_EVENT_CQ_OVERRUN and IB_EVENT_CQ_ACCESS, rather than 
> > just a generic IB_EVENT_CQ_ERR
> > 
> > What do you think?
> 
> I'm fine with specifying the CQ errors, though I'm not sure 
> that the user can do anything differently.  Maybe use 
> CQ_ERROR or CQ_ACCESS_ERROR, rather than CQ_ACCESS.
> 
> I'll let Roland comment on the changes to mthca.
> 
> - Sean
> _______________________________________________
> openib-general mailing list
> openib-general at openib.org
> http://openib.org/mailman/listinfo/openib-general
> 
> To unsubscribe, please visit 
> http://openib.org/mailman/listinfo/openib-general
> 
>