[ofa-general] Re: [PATCH] - [resend] Corrects a race in ipoib_cm_post_receive_nonsrq()

[Roland Dreier]

 > - enhance ipoib_cm_post_receive_nonsrq() with wr and sge passed to by
 >   the caller, ie ipoib_poll() -> ipoib_cm_handle_rx_wc() resp.
 >   ipoib_cm_nonsrq_init_rx()
 > - ipoib_cm_handle_rx_wc() passes rx_wr and rx_sge as already defined
 >   in struct ipoib_cm_dev_priv to ipoib_cm_post_receive_nonsrq()
 >   since we use one cq, there should not be a race in this path
 > - ipoib_cm_nonsrq_init_rx() allocates wr and sge on the heap or slabcache,
 >   passes them to ipoib_cm_post_receive_nonsrq() and frees them thereafter

I just wrote a patch before reading your email, and this is exactly what
I did.

Thanks,
  Roland