[openib-general] Re: [PATCH] core: fix user_mad memory leaks on timeout

[Michael S. Tsirkin]

Quoting Hal Rosenstock <halr at voltaire.com>:
> Couldn't the callback be rescheduled for some time later where the
> allocation might succeed ?

Arent we allocating with GFP_KERNEL already?
I think this means "try as hard as you can" already. No?

Maybe we could allocate the timeout packet together with allocating the
send mad.

> > lets at least be careful to avoid a crash or memory leak.
> 
> Agreed.

-- 
MST