Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 24 Jan 2005 09:53:44 +0000 (GMT)
From:      Robert Watson <rwatson@freebsd.org>
To:        Ganbold <ganbold@micom.mng.net>
Cc:        freebsd-current@freebsd.org
Subject:   Re: fxp0: device timed out problem
Message-ID:  <Pine.NEB.3.96L.1050124093419.63183A-100000@fledge.watson.org>
In-Reply-To: <6.2.0.14.2.20050124113106.03402770@202.179.0.80>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 24 Jan 2005, Ganbold wrote:

> > > I turned off debug.mpsafenet to 0 and it seems like problem goes away.
> ><snip>
> > > Is this problem related to network stack? Or is it related to fxp driver?
> >
> >It's most likely a problem with the device driver or interrupt
> >configuration on your system.  There are a couple of other variables you
> >might try frobbing:
> >
> >- Use of ACPI to configure the hardware
> >- Use of "device apic" if the system is non-SMP
> 
> I see. One of the server is SMP system and device apic option is used in
> kernel config file. I didn't try device apic on non-SMP machine. 

Any luck with disabling ACPI?  In particular, are the interrupt
assignments substantially different between booting with ACPI and without?
You can probably just diff -u the old dmesg.boot and the new one...

> >Usually a device timed out error is related to interrupts from the device
> >not being delivered, being delivered improperly, etc.  Does your dmesg
> >contain any references to interrupt storms?  Once the above message has
> >printed, do you see any further interrupts on the fxp interrupt source
> >when checking intermittently with "systat -vmstat 1" or "vmstat -i"?
> 
> I couldn't check the system by issuing those commands.  Following is the
> dmesg output with debug.mpsafenet disabled: 

Couldn't as in, not possible for administrative reasons, because you
couldn't log in once the failure occurred so couldn't get the output, or
because they don't work, or...?  Just want to make sure I understand if
this is an administrative issue or symptomatic.

> I didn't do much investigation on those servers that time. However
> without debug.mpsafenet, servers are working fine for more than 3 weeks. 

That is certainly suggestive -- I wonder if we're looking at a locking bug
in fxp0 involving serialization with the hardware.  However, it's not
conclusive, I think -- when running MPSAFE, the timing is quite different
on UP as well as SMP hardware, which could trigger other existing bugs. 
The big open question, I think, is whether an interrupt delivery problem
is involved. 

Robert N M Watson




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.NEB.3.96L.1050124093419.63183A-100000>