[openib-general] ucma into kernel.org

[Michael S. Tsirkin]

Quoting r. Sean Hefty <sean.hefty at intel.com>:
> Subject: RE: ucma into kernel.org
> 
> >Hmm, thanks.
> >So what you are saying is that we must destroy/recreate the QP as well,
> >and this implies this must be done at ULP level. Right?
> 
> I believe so.
> 
> >So I guess if CMA rounds the retry count down to the max supported
> >value, this will be enough for most ULPs, and if not, ULP
> >can get the etry count, see that it is smaller than what user set,
> >and destroy/retry.
> >
> >Sounds good?
> 
> I would rather the set failed if the value is too large.  We could expose the
> maximum supported number of retries through a #define in the header file.

OK, but wouldn't it be different for IB/TCP? Maybe a function then?

> So, is the proposal to provide a generic set_option call for connection
> parameters?  Is there any other parameter than retries?  Does a generic
> connection timeout value work for TCP as well?

Not that I know: I see MAX_TCP_SYNCNT but nothing to do with timers.
Ideally CMA would do something adaptive ...

-- 
MST