> There is a race condition on accept_pend_cnt. Change it to atomic.
This is much too terse, so I don't know what the race is or how the
patch fixes it. But...
> + if (atomic_dec_and_test(&cm_node->accept_pend)) {
you do atomic_dec_and_test() but then the only other manipulations of
accept_pend that I see are:
> + atomic_set(&cm_node->accept_pend, 0);
> + atomic_set(&cm_node->accept_pend, 1);
and there's no particular ordering between atomic_set() and
atomic_dec_and_test() that I know of to protect against races.
So at least a better desription of the patch, please.
- R.
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
