[ofa-general] Re: IPOIB CM (NOSRQ)[PATCH V4] patch for review

[Michael S. Tsirkin]

> Quoting Pradeep Satyanarayana <pradeeps at linux.vnet.ibm.com>:
> Subject: Re: IPOIB CM (NOSRQ)[PATCH V4] patch for review
> 
> Michael S. Tsirkin wrote:
> >>+void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
> >>+{
> >>+	struct ipoib_dev_priv *priv = netdev_priv(dev);
> >>+
> >>+	if (priv->cm.srq)
> >>+		handle_rx_wc_srq(dev, wc);
> >>+	else
> >>+		handle_rx_wc_nosrq(dev, wc);
> >> }
> >
> >I still think this conditional branch on datapath should be avoided
> >by using separate RX handlers for SRQ/non SRQ cases.
> >And same for the one on alloc_rx_skb.
> >
> 
> I attempted implementing this. With NAPI now included,
> the code looked real ugly and so decided not do so.

Why?
The only difference with NAPI is that instead of a separate
completion handler, you should have a separate poll routine.

-- 
MST