Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 12 Jul 2010 18:08:54 +0200
From:      Markus Gebert <markus.gebert@hostpoint.ch>
To:        Adam McDougall <mcdouga9@egr.msu.edu>
Cc:        freebsd-stable <freebsd-stable@freebsd.org>
Subject:   Re: 8.1-RC2 - PCI fatal error or MCE triggered by USB/ehci on Sun X4100M2?
Message-ID:  <2CA1E184-D9FF-4093-A6D9-18AF6DDC7407@hostpoint.ch>
In-Reply-To: <20100712154316.GD15784@egr.msu.edu>
References:  <6B57591F-9FA2-45EB-825F-1DB025C0635D@hostpoint.ch> <20100712154316.GD15784@egr.msu.edu>

next in thread | previous in thread | raw e-mail | index | archive | help

On 12.07.2010, at 17:43, Adam McDougall wrote:

> I also get MCE on x4100m2 when causing significant disk activity in =
mpt
> while also downloading through em0 or em1.

Could you reproduce this on 6.x or 7.x? Because whatever we try here, we =
simply couldn't so far. A short test with Ubuntu also didn't show any =
sing of problems.


>  I was not able to trigger it
> while using nfe, however nfe locked up on me during normal DNS server
> traffic so that was a wash.

We had issues with nfe pre-8.x, that's why we have been using the em =
nics, which seem to be part of the problem now in 8.x.


>  What seemed to work for me was to add an
> Intel PCIE nic to the server and use it instead of the onboard NICS.

Thanks for the hint.


> For whatever reason I never experienced this problem until using ZFS.

We were able to reproduce it with UFS on 8.x. with just one disks (no =
gmirror), but I guess it's easier to trigger with ZFS especially in an =
mirror setup.


> I triggered it by downloading a 200m tgz file via http repeatedly
> over gigabit and it would reliably crash within a minute or two.

Our test case is basically:

1. fetch a large file using wget over em0 (100mbit link seems enough)
2. cp a large file locally to stress mpt
3. wait for MCE


> I ordered a dozen nics for probably around $20 each and was satisfied
> with this workaround given the age of the servers.  I'm pinched on =
time
> for work so I often don't get around to reporting issues where I've
> found a workaround, I'm glad you can get that started.=20


"Glad" we're not the only ones :-)


Markus





Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?2CA1E184-D9FF-4093-A6D9-18AF6DDC7407>