On Tuesday 07 March 2006 04:08, PaNiC wrote:
> Thanks for your reply.
> I have an idea what you're talking about, but no more i'm afraid.
> I'm no programmer and i don't know how to try what you suggested.
> What i did try was applying some patches manually that i found in the
> mailing list archives.
>
> This is one of them:
I'll try to elaborate a little, but since my experience in dealing with the
NETDEV watchdog timeout issue is confined to x86 architectures,
some translation may be necessary...
I've seen two causes for this: 1) driver bug, 2) firmware bug. I have yet to
come across one where the kernel itself was at fault.
1) if its a driver bug, updating to the latest driver should have an effect,
hopefully a positive one. Generally, the driver maintainer is at least aware
of the problem, if this is the cause.
2) If its a firmware bug, you'll have to upgrade to the latest version of
the BIOS, uboot, rommon, or whatever your platform uses to boot the
OS.
Case 2) is the type of problem where many people can be using the exact
same kernel and exact same driver, and only a few experience any problems.
This is because its not the kernel or the driver, but the firmware that is at fault.
In particular, on the x86 platform where I've root-caused this issue, a different
version of the BIOS was present (I did an upgrade, and the upgrade failed to
implement the proper register programming, and that's when the problem showed
up). The manifestation is that heavy PCI traffic (not just ethernet traffic) would
overload the host-PCI bridge, and bad things started happening. In some cases,
just ethernet traffic alone was enough to overpower the bridge, but that dependended
on the link speed. 100 Mbs took a lot longer to NETDEV watchdog timeout than
1000 Mbs. Another fun aspect of case 2) is that since its PCI traffic related, its very hard
to recreate exactly, since there's usually other stuff on the PCI bus.
So if nobody seems to have a solution for your problem, it is possible that its
firmware-related, so try getting a different firmware. In particular, if you have any
experience at this, or know someone in your organization who does, try focusing on
the MTT (multi-transaction timeout) register, and the equivalent (if any) of the ICH
configuration register in the host-PCI bridge. These are both proprietary registers, and
are not present in the standard 64 byte PCI header, so its pretty hard to "guess" where they
are what they should be set to. If you have a similar platform that does not have any
NETDEV watchodog timeout issues, it may be possible to dump the PCI config. space
of the host-PCI bridge on that platform, and compare it with the failing platform, and start
"guessing" from there, but the firmware is the responsible entity that should have these
registers programmed correctly in the first place.
Sorry for being so verbose, I'm hoping that others googling in the future may save themselves
some time in isolating this pretty obscure cause...
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
