Red Hat Bugzilla – Bug 87629
(NET TG3)crash kernel tg3
Last modified: 2005-10-31 17:00:50 EST
From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.2.1) Gecko/20030128
Description of problem:
fter upgrading from kernel-smp-2.4.18-19.7.aragorn2.i686.rpm to
kernel-smp-2.4.18-27.8.0.i686.rpm my pe2650 has begun crashing again.
I don't have any debug info, but the server ran flawnes with the aragon driver,
for a mounth, but now it's crashed two times in 4 days using 2.4.18-27.8.0.i686.
I'm connecting a monitor, and hoping to se some debug info next time.
Version-Release number of selected component (if applicable):
Machine: Dell PE 2650
Compiler: gcc 2.96-110
NIC: Broadcom gigabit ethernet
This kernel does seem to be better than the previous ones, but doesnt
completely resolve the problem. Under bursty traffic, the system either
(1) hangs, with console disabled (nothing appears the monitor, and
PE 2650's power recyle button doesnt work
(2) system hangs, but console is not disabled and power recyle button still
(3) There is a kernel panic (see attachment "TG3 crash kernel panic"
Created attachment 90873 [details]
TG3 crash kernel panic
Here is one more manifestation of the tg3 bug.
The kernel does not crash, but the network interface does not work anymore. So
I brought down the network interace (ifconfig eth0 down), then tried to bring
it back up (ifconfig eth0 up), and I got this message.
tg3: tg3_stop_block time out, ofs=4800 enable_bits=2
try the official_broadcomt driver :
until this problem goes away in tg3
I just have found a possible solution with tg3 :
to boot with kernel parameter acpi=off or acpi=oldboot
maybe it works but if the force is with you:
or you get it from a recent 2.4.20-rc
Xose Vazquez: The RHL kernel does not include acpi, so acpi=off doesn't do anything
We have many dual Xeon 2.6 GHz machines equipped with Broadcom cards
01:03.0 Ethernet controller: Broadcom Corporation NetXtreme BCM5701 Gigabit
Ethernet (rev 15)
The machines routinely hung hard with the tg3 driver (compiled with 2.96-112)
under heavy traffic, only power cycle helped. We changed the driver to bcm5700
(compiled with 2.96-112) and the machines are happy.
No more 8.x updates, so this is unlikely to be fixed.