Re: [BUG] e1000 fails on 2.4.26+bk with CONFIG_SMP=y

From: Willy TARREAU
Date: Mon Apr 19 2004 - 16:57:18 EST

On Mon, Apr 19, 2004 at 05:42:47PM -0400, Joe Korty wrote:
> On Mon, Apr 19, 2004 at 09:39:30PM +0200, Willy Tarreau wrote:
> > On Mon, Apr 19, 2004 at 12:54:25PM -0400, Joe Korty wrote:
> >>
> >> Uniprocessor 2.4.26 works fine.
> >> Uniprocessor 2.4.26 + local apic works fine.
> >> Uniprocessor 2.4.26 + local apic + io apic fails.
> >
> > interesting. Unfortunately, I didn't have time to try on the machine I told
> > you about last day. But right here, I have a dual athlon communicating
> > with an alpha, both with e1000 (544) in 2.4.26. Since there's a PCI
> > bridge on your quad, I wonder if the IOAPIC doesn't trigger an interrupt
> > routing problem with bridges. Are all the ports unusable or do some of
> > them work reliably in APIC mode ?
> I just verified that the Quad Ethernet board works with 2.6.5 SMP, so it
> is unlikely to be a bridge or other hardware problem.

Sorry that was not what I meant. I meant that the 2.4.26 irq routing or
I don't know what might experience difficulties due to the presence of
a PCI bridge on this board, and potentially this one particularly. I once
encountered such a problem in a desktop machine which would only see 3 out
of the 4 ports on an adaptec quad board, depending on the PCI slot it was
fit in ! It turned out to be a problem with region alignment or something
like this which was not correctly forwarded through the bridge (bios bug
or so at initialization).

> For 2.4.26, the Dell 650 has an Intel 82545EM Gigabit Ethernet Controller
> soldered in, which also uses the e1000 driver, and that works fine, so
> we know the 2.4.26 e1000 driver works with some of these Intel chips.

OK, same here.

> When the Quad board works, it negotiates down to 10 MBits/sec Half Duplex.
> Not sure yet if this is what it is supposed to be doing in our environment.

I remember something like this not long ago. It was what ethtool reported, but
the card was running either gigabit or a forced 100FDX without auto-neg on the
other side, I don't remember. Can you try a simple 4-pairs cable between two
ports on the quad to check what media is negociated ?

Also, can you cat /proc/interrupts with and without io/apic, and try without
ACPI as Zwane suggested ?


To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at