Red Hat Bugzilla – Bug 103705
tg3.o driver in SMP kernel does not enable network interface
Last modified: 2013-08-05 21:02:08 EDT
From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; SunOS sun4u; en-US; rv:1.1) Gecko/20020827
Description of problem:
In SMP mode the broadcom network cards initialize ok according to dmesg, but
fail to bring a network up. The same cards work in UP mode. The tg3 module is
The driver is tg3.o , the source is v1.6, SMP binary is 64240bytes,
the UP 64384
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1.boot SMP kernel
Actual Results: cannot assign ip address through dhcp
Expected Results: should bring up a card with ifup eth0
Broadcom NetXtreme BCM5704 Gigabit
beta 1 worked
Have needed to add acpi=off to even get SMP beta 2 to boot
this sounds like an irq routing problem; without ACPI you don't get irq's routed
to the tg3 card
Can you tell me what model system you are using, and describe how the SMP boot
fails (if you can produce and attach a serial console dump that would be ideal!)
Created attachment 94212 [details]
serial console output with SMP without acpi=off
Could you try again? The crucial part of that serial capture (crash location
and beginning of register dump) is garbled.
any more info on reproducing this?
In email with Jim I believe I am seeing an SMP bug in the serial console too.
I have manually typed in the last lines of the console output, the output is
being overwritten by other output lines. I'll see if I can decipher the rest
Code: Bad RIP value.
<<0>4>KeCormneplle ptiannicg : ReAtgiteonm/ptFeied ltd/o Bukiffller t/hPaec
lizInat iidoln:e <ta4>sk. -. no.t. syn.ci.ng
<this has corrupted the serial output reads>
<0> Kernel panic Attempted to kill the idle task!
<4>Completing Region/Field/Buffer/Package initialization <4> not syncing
Please run the /usr/sbin/dmidecode and attach the output. Thanks!
Created attachment 94771 [details]
On deciphering the crash myself I noticed we never get out of acpi
initialization as such, so the tg3 driver could be a red herring in this case.
Conversations with customer and AMD show that this is happening on a
prototype platform that will not go into production and that we do not
have access to. We can only support such configurations on a "best
effort" basis. Since we couldn't resolve the problem I'm changing
this to WONTFIX.
Good News I've been meaning to post. With a BIOS update and AS 3.0
Final release (this was originally reported on beta) the network card
now works in SMP mode