lkml.org 
[lkml]   [2003]   [Feb]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRE: lockups with 2.4.20 (tg3? net/core/dev.c|deliver_to_old_ones)
Date
From
> >> Since sometime in December two systems we have on site using P4 HT
> (one
> >> Dell 2650 and one Dell 4600, both dual CPU, both ht/mce capable)
have
> been
> >> locking up without any kernel output and without sysrq keys working
> (the
> >> keyboard is locked solid).
> >>[...]
> >> Using nmi_watchdog I've managed to get a stack track and ran
ksymoops
> over
> >> it (attached).
>
>
> > Good report. To tell the truth, I know that this lockup exists,
> > there's an RH issue-tracker item against me on this.
>
> Several of us at HP have been chasing this problem as well. Here is
why
> there is a deadlock: deliver_to_old_ones()attempts to stop all timers
> from running and then blocks until all the timers are no longer
running.
> This code is called from netif_receive_skb which is called from
tg3_poll
> while it is holding a lock in the tg3 driver. On another CPU, the
> tg3_timer routine is run but is blocked by the lock held in the
tg3_poll
> routine. The tg3_timer routine never finishes because it can't acquire
> the lock being held by tg3_poll on another CPU. That prevents
> deliver_to_old_ones from executing because there is still a timer
> routine executing.
>
> Here is the call stack of the deadlocked CPUs on a RH8.0 system with a
> 2.4.18-24.8.0 smp kernel:
> CPU 2:
> deliver_to_old_ones+45
> netif_receive_skb
> tg3_rx+27b
> tg3_poll+81
> net_rx_action
> do_softirq
> do_IRQ
> call_do_IRQ
>
> CPU 6:
> tg3_timer (tg3+9fc4)
> run_timer_list+0x112
> bh_action+55
> tasklet_hi_action+67
> do_softirq+d9
> do_IRQ
> call_do_IRQ+5
>
Until this is fixed, there are a couple work-arounds:
- Put the system behind a NAT firewall. This problem is caused by
another protocol on the wire, probably IPX (Netware).
- Switch to a non-NAPI driver. Since you're using the BroadComm NIC, try
the BCM5700 driver or switch to a different NIC.

While troubleshooting this problem here at HP, I have a couple of
systems that have run for over 2 weeks by using the above work-arounds.

Tom Rhodes
tom.rhodes@hp.com
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:33    [W:0.048 / U:4.736 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site