Eric Dumazet <da...@cosmosbay.com> wrote on 25/03/2009 15:04:26:
> Joakim Tjernlund a écrit :
> >>From 1c2f23b1f37f4818c0fd0217b93eb38ab6564840 Mon Sep 17 00:00:00 2001
> > From: Joakim Tjernlund <joakim.tjernl...@transmode.se>
> > Date: Tue, 24 Mar 2009 10:19:27 +0100
> > Subject: [PATCH] ucc_geth: Move freeing of TX packets to NAPI context.
> > Also increase NAPI weight somewhat.
> > This will make the system alot more responsive while
> > ping flooding the ucc_geth ethernet interaface.
> >
> >
> > Signed-off-by: Joakim Tjernlund <joakim.tjernl...@transmode.se>
> > ---
> > drivers/net/ucc_geth.c | 30 +++++++++++-------------------
> > drivers/net/ucc_geth.h | 2 +-
> > 2 files changed, 12 insertions(+), 20 deletions(-)
> >
> > diff --git a/drivers/net/ucc_geth.c b/drivers/net/ucc_geth.c
> > index 097aed8..7d5d110 100644
> > --- a/drivers/net/ucc_geth.c
> > +++ b/drivers/net/ucc_geth.c
> > @@ -3214,7 +3214,7 @@ static int ucc_geth_tx(struct net_device *dev,
u8 txQ)
> > dev->stats.tx_packets++;
> >
> > /* Free the sk buffer associated with this TxBD */
> > - dev_kfree_skb_irq(ugeth->
> > + dev_kfree_skb(ugeth->
> > tx_skbuff[txQ][ugeth->skb_dirtytx[txQ]]);
> > ugeth->tx_skbuff[txQ][ugeth->skb_dirtytx[txQ]] = NULL;
> > ugeth->skb_dirtytx[txQ] =
> > @@ -3248,9 +3248,16 @@ static int ucc_geth_poll(struct napi_struct
*napi, int budget)
> > for (i = 0; i < ug_info->numQueuesRx; i++)
> > howmany += ucc_geth_rx(ugeth, i, budget - howmany);
> >
>
> Cant you test (ucce & UCCE_TX_EVENTS) or something here to avoid
> taking lock and checking queues if not necessary ?
Probably, but I want this patch as simple as possible. There
are lots of optimizations left to do in this driver.
>
> > + /* Tx event processing */
> > + spin_lock(&ugeth->lock);
> > + for (i = 0; i < ug_info->numQueuesTx; i++) {
> > + ucc_geth_tx(ugeth->dev, i);
> > + }
> > + spin_unlock(&ugeth->lock);
> > +
>
> Why tx completions dont change "howmany" ?
> It seems strange you changed UCC_GETH_DEV_WEIGHT if not taking into
account tx event above...
This is unclear and last I checked not very common amongst other drivers
in the
tree.
UCC_GETH_DEV_WEIGHT needs to be a bit bigger than the number of RX HW
buffers avaliable,
otherwise one won't be able to drain the whole queue in one go. Changing
weight to something bigger made a big difference.
Jocke
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-dev
