lkml.org 
[lkml]   [2002]   [Mar]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: Sun GEM card looses TX on x86 32bit PCI
From
Date
Hi David,

Sorry about the mindless waffle earlier on - I'd just come in from work
and was suffering from brain death ;) I must remember to wait an hour or
so before touching a computer after I come in from work.

Ok, I've been fiddling around with the driver tonight and have managed
to get a little further by forcing the driver to do a full reset of the
chip when the RX buffer over flows. I achieved this by sticking a return
1; at the top of gem_rxmac_reset().

I'm guessing this isn't an "optimal" reset for the situation but so far
it's having /reasonable/ results (i.e. I don't have to bring the
interface up and down every 30 seconds!).

Here's the output of the ping;

monkey:/home/andy# ping -f -s 1472 shroom
PING shroom.beezly.org.uk (10.0.0.15) from 10.0.0.12 : 1472(1500) bytes
of
data.............................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
--- shroom.beezly.org.uk ping statistics ---
366892 packets transmitted, 366288 received, 0% loss, time 343836ms
rtt min/avg/max/mdev = 0.445/0.471/5.377/0.038 ms, pipe 2, ipg/ewma
0.937/0.465 ms
Hope this helps,

Beezly

On Mon, 2002-03-11 at 21:23, Beezly wrote:
> David,
>
> I've been looking some more at what my changes did.... it might be best
> to completely ignore them ;) I had no clue what I was doing!
>
> Cheers,
>
> Beezly
>
> On Mon, 2002-03-11 at 21:11, Beezly wrote:
> > Hi David,
> >
> > It seems I fubar'd. I recompiled the module and run it through the test
> > again... no hang. It looks like I forgot to copy the new module into my
> > /lib/modules/<blah>. Apologies for messing up there.
> >
> > Anyway... the new driver still drops packets after the initial RX
> > overflow, so I had a poke around with it and I've seen some definate
> > improvement by forcing the whole chip to reset when the RX overflows.
> >
> > My modifications to the driver are evil and I only intend them to be a
> > test, but it helps to shed some extra light on what's going on.
> >
> > When the chip does a full reset I loose a whole load of packets, but I'm
> > guessing this is normal :(
> >
> > Also, I can't remember where I read it, but the Extreme Summit 48 is
> > supposed to support *receiving* the xon/xoff Pause stuff (I'm no expert
> > in this area, so I could be talking complete twaddle!), no transmit
> > capability though.
> >
> > Here's what I get out of the module when it resets (with my limit=5000
> > mod);
> >
> > eth0: RX buffer overflowed - running rxmac_reset
> > eth0: RX MAC resetting
> > eth0: RX MAC *ONLY* reset
> > eth0: RX MAC reset ok?
> > eth0: RX MAC will not disable, resetting whole chip.
> > eth0: PCS AutoNEG complete.
> > eth0: PCS link is now up.
> >
> > Without the limit=5000, it appears that the module detects the RX
> > section is "un-hung" when it isn't.
> >
> > Cheers,
> >
> > Beezly
> >
> > On Mon, 2002-03-11 at 19:02, David S. Miller wrote:
> > > From: Beezly <beezly@beezly.org.uk>
> > > Date: 11 Mar 2002 18:35:01 +0000
> > >
> > > Sorry it took so long for me to get back to you. Sadly it also hung with
> > > this patch ;) I was unable to get an oops out of it (machine was
> > > completely hosed and in X so I couldn't even note the oops on paper :(
> > > ).
> > >
> > > So rerun the test not under X please?
> >
> > ----
> >
>
> > --- sungem.c Mon Mar 11 20:37:57 2002
> > +++ sungem.c.testing Mon Mar 11 20:31:12 2002
> > @@ -302,14 +302,23 @@
> > u64 desc_dma;
> > u32 val;
> >
> > + printk(KERN_ERR "%s: RX MAC resetting\n", dev->name);
> > /* First, reset MAC RX. */
> > writel(gp->mac_rx_cfg & ~MAC_RXCFG_ENAB,
> > gp->regs + MAC_RXCFG);
> > + printk(KERN_ERR "%s: RX MAC *ONLY* reset\n", dev->name);
> > +
> > for (limit = 0; limit < 5000; limit++) {
> > - if (!(readl(gp->regs + MAC_RXCFG) & MAC_RXCFG_ENAB))
> > + if (!(readl(gp->regs + MAC_RXCFG) & MAC_RXCFG_ENAB)) {
> > + printk(KERN_ERR "%s: RX MAC reset ok?\n", dev->name);
> > break;
> > + }
> > udelay(10);
> > }
> > +
> > + /* RX MAC reset doesn't appear to work so I force a whole reset */
> > + limit = 5000;
> > +
> > if (limit == 5000) {
> > printk(KERN_ERR "%s: RX MAC will not disable, resetting whole "
> > "chip.\n", dev->name);
> > @@ -323,6 +332,9 @@
> > break;
> > udelay(10);
> > }
> > +
> > + limit=5000;
> > +
> > if (limit == 5000) {
> > printk(KERN_ERR "%s: RX DMA will not disable, resetting whole "
> > "chip.\n", dev->name);
> > @@ -399,6 +411,8 @@
> > if (rxmac_stat & MAC_RXSTAT_OFLW) {
> > gp->net_stats.rx_over_errors++;
> > gp->net_stats.rx_fifo_errors++;
> > + printk(KERN_DEBUG "%s: RX buffer overflowed - running rxmac_reset\n",
> > + gp->dev->name);
> >
> > ret = gem_rxmac_reset(gp);
> > }
>

[unhandled content-type:application/pgp-signature]
\
 
 \ /
  Last update: 2005-03-22 13:21    [from the cache]
©2003-2011 Jasper Spaans