[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Wed, 5 May 2010 08:01:11 -0700
From: "Michael Chan" <mchan@...adcom.com>
To: "'Eric Dumazet'" <eric.dumazet@...il.com>
cc: "davem@...emloft.net" <davem@...emloft.net>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: [PATCH -next 2/3] bnx2: Add prefetches to rx path.
Eric Dumazet wrote:
> > @@ -3097,7 +3099,11 @@ bnx2_rx_int(struct bnx2 *bp, struct
> bnx2_napi *bnapi, int budget)
> >
> > rx_buf = &rxr->rx_buf_ring[sw_ring_cons];
> > skb = rx_buf->skb;
> > + prefetch(skb);
>
> why not a prefetchw() ?
Yes, didn't know there was a prefetchw() before you pointed it
out.
>
> >
> > + next_rx_buf =
> > +
> &rxr->rx_buf_ring[RX_RING_IDX(NEXT_RX_BD(sw_cons))];
> > + prefetch(next_rx_buf->desc);
>
> So cpu is allowed to start a memory transaction on
> next_skb->data, while
> not yes DMA unmapped ?
Very good point. The prefetch() will not work and will be
wasted on systems that have pci_dma_sync_...() defined. I
think we can skip the prefetch if pci_dma_sync_...() is
defined. The logic to determine if the next descriptor is
ready, dma_sync it, and prefetch it will be complicated and
we may end up not gaining any performance in the end.
What do you think? Thanks.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists