On 11/07/07, Stephen Hemminger <[EMAIL PROTECTED]> wrote:
On Wed, 11 Jul 2007 22:39:49 +0100
"Daniel J Blueman" <[EMAIL PROTECTED]> wrote:
> On 11/07/07, Daniel J Blueman <[EMAIL PROTECTED]> wrote:
> > > > On 05/07/07, Stephen Hemminger <[EMAIL PROTECTED]> wrote:
> > > > > Well, it didn't fix my test, but it made it better. The following
seemed
> > > > > to work longer...
> > > > >
> > > > > --- a/drivers/net/sky2.c 2007-07-05 09:09:45.000000000 -0700
> > > > > +++ b/drivers/net/sky2.c 2007-07-05 09:09:51.000000000 -0700
> > > > > @@ -2490,6 +2490,13 @@ static int sky2_poll(struct net_device *
> > > > >
> > > > > work_done = sky2_status_intr(hw, work_limit);
> > > > > if (work_done < work_limit) {
> > > > > + /* Bug/Errata workaround?
> > > > > + * Need to kick the TX irq moderation timer.
> > > > > + */
> > > > > + if (sky2_read8(hw, STAT_TX_TIMER_CTRL) == TIM_START) {
> > > > > + sky2_write8(hw, STAT_TX_TIMER_CTRL, TIM_STOP);
> > > > > + sky2_write8(hw, STAT_TX_TIMER_CTRL,
TIM_START);
> > > > > + }
> > > > > netif_rx_complete(dev0);
> > > > >
> > > > > /* end of interrupt, re-enables also acts as I/O
synchronization */
> > > >
> > > > I spoke too soon on this. With the above patch on 2.6.22-rc7, it
> > > > failed much sooner than the previous patch with the
> > > > read32(B0_Y2_SP_LISR); I'll try to reproduce with the older patch.
> > > >
> > > > Note the ifconfig error/dropped/frame count at the time of failure:
> [snip]
> > > The last message means some how frame was received with checksum for count
> > > wrong. I have only seen it when coalescing is messed up.
> > >
> > > I ran for 2+ days with the patch, and only 20min without. Usually my ISP
connection
> > > gives up after that because of crappy DSL box, and that makes DNS not
work.
> >
> > It wedged when I was copying a few GBs of data from my server to a
> > local disk at the time, and running rsync over ssh on a large file on
> > my server to my laptop's disk.
> >
> > This would be the typical load that would cause the NIC to lockup from
> > missing an IRQ or otherwise, however, it did feel like the new code
> > didn't un-wedge the Yukon-EC's bus master unit.
> >
> > What other tricks can be used to reset the Yukon-EC's bus master unit?
> >
> > I'll try the read32(B0_Y2_SP_LISR) trick, as before.
>
> Nope, this still locks up as you found.
>
> I have a reliable reproducer:
>
> 1. export directory over NFS TCP on server
> 2. mount directory on client
> 3. run 'iozone -a' in directory on client
>
> I'm reproducing this with NFSv4 (with callbacks working) with 1500
> octet MTU with one client, all gigabit. It would be good to hear if
> you can reproduce the problem there.
>
> Daniel
Please try again with post 2.6.22 git version (1.16)?
Reproduced with 2.6.22 w/ sky2 1.16 from git. We observe this
characteristic failure on the NFS server (always around 2-3GB of
transmit):
$ ifconfig lan0
lan0 Link encap:Ethernet HWaddr 00:03:2D:05:9C:27
inet addr:192.168.0.250 Bcast:192.168.0.255 Mask:255.255.255.0
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:24007220 errors:1 dropped:1 overruns:0 frame:1
TX packets:13886495 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:171026170 (163.1 MiB) TX bytes:2262910580 (2.1 GiB)
Interrupt:16
I'll rebuild with debugfs and grab the debug you've exported.
Daniel
--
Daniel J Blueman
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
