[openib-general] Re: uverbs comp events

[Sean Hefty]

Roland Dreier wrote:
>     Sean> I'm leaning more towards Arlin's thinking on this, but
>     Sean> whatever is decided, I think that uCM and uAT should match.
> 
> Doesn't this seem like a good argument against adding support for
> synthetic events?  We already have a perfectly good mechanism for
> generating fd events (namely pipe() + write()), so why do we want to
> add three more mechanisms for generating synthetic events?

I think that the issue that Arlin is hitting is that once he calls 
blah_get_event() he doesn't have an easy way to release the thread. 
This may turn out to be a uCM / uAT issue, and not a verbs issue.

Verbs has calls to open and close the device that in turn open and close 
the fd.  It sounds like when he closes the device, his thread doesn't 
return, but he'll need to confirm this.  For uCM and uAT, the fd's are 
created during initialization, and I don't see where they're ever 
explicitly closed.

- Sean