TG3 driver dies with "irq 12: nobody cared" on 2.6.10 (x86)

[Scott Laird]

I'm having a huge pile of problems with my home file server.  Since 
upgrading from 2.6.2 to 2.6.10, the box now falls off the network at 
the drop of a hat.  Here's one round of logs:

Jan 12 17:16:49 nfs kernel: irq 12: nobody cared!
Jan 12 17:16:49 nfs kernel:  [<c012a2ea>] __report_bad_irq+0x2a/0x90
Jan 12 17:16:49 nfs kernel:  [<c0129dd0>] handle_IRQ_event+0x30/0x70
Jan 12 17:16:49 nfs kernel:  [<c012a3dc>] note_interrupt+0x6c/0xd0
Jan 12 17:16:49 nfs kernel:  [<c0129eea>] __do_IRQ+0xda/0xf0
Jan 12 17:16:49 nfs kernel:  [<c0103ff9>] do_IRQ+0x19/0x30
Jan 12 17:16:49 nfs kernel:  [<c01026be>] common_interrupt+0x1a/0x20
Jan 12 17:16:49 nfs kernel:  [<c0116130>] __do_softirq+0x30/0x90
Jan 12 17:16:49 nfs kernel:  [<c01161b6>] do_softirq+0x26/0x30
Jan 12 17:16:49 nfs kernel:  [<c0103ffe>] do_IRQ+0x1e/0x30
Jan 12 17:16:49 nfs kernel:  [<c01026be>] common_interrupt+0x1a/0x20
Jan 12 17:16:49 nfs kernel:  [<c012e299>] mempool_free+0x19/0x80
Jan 12 17:16:49 nfs kernel:  [<c014b1b8>] bio_destructor+0x38/0x60
Jan 12 17:16:49 nfs kernel:  [<c014b3b9>] bio_put+0x29/0x40
Jan 12 17:16:49 nfs kernel:  [<c01667e0>] mpage_end_io_read+0x60/0x70
Jan 12 17:16:49 nfs kernel:  [<c0337a9e>] handle_stripe+0x73e/0xc70
Jan 12 17:16:49 nfs kernel:  [<c033d5ae>] md_update_sb+0xae/0xe0
Jan 12 17:16:49 nfs kernel:  [<c03384c0>] raid5d+0x60/0xf0
Jan 12 17:16:49 nfs kernel:  [<c033f913>] md_thread+0x123/0x170
Jan 12 17:16:49 nfs kernel:  [<c0124060>] 
autoremove_wake_function+0x0/0x60
Jan 12 17:16:49 nfs kernel:  [<c010245e>] ret_from_fork+0x6/0x14
Jan 12 17:16:49 nfs kernel:  [<c0124060>] 
autoremove_wake_function+0x0/0x60
Jan 12 17:16:49 nfs kernel:  [<c033f7f0>] md_thread+0x0/0x170
Jan 12 17:16:49 nfs kernel:  [<c0100851>] kernel_thread_helper+0x5/0x14
Jan 12 17:16:49 nfs kernel: handlers:
Jan 12 17:16:49 nfs kernel: [<c0308830>] (usb_hcd_irq+0x0/0x70)
Jan 12 17:16:49 nfs kernel: [<c02aae30>] (tg3_interrupt+0x0/0x130)
Jan 12 17:16:49 nfs kernel: Disabling IRQ #12

I moved cards around to get the TG3 onto its own IRQ, and that didn't 
make any difference.  This box is an Athlon 700.  It's running a 
single-processor non-preempt kernel with 8k stacks.  If anyone cares, 
the full .config is at http://scottstuff.net/misc/config-2.6.10.txt

The last three reboots have only lasted for a minute or two before the 
TG3 interrupt problem pops up.  Once this happens, nothing that I can 
do seems to get the network working again--'ip link set eth0 down ; ip 
link set eth0 up' doesn't help.  I'm not using modules, so I can't try 
unloading and reloading.

The single weirdest part of this is the numbers from /proc/interrupts: 
the last two hangs happened with *exactly* 200,000 TG3 interrupts.  The 
time before that, it was exactly 400,000 interrupts.

I have a second TG3 card that I can swap in, but it's sitting on my 
desk at work, so I won't be able to try until tomorrow.

Does anyone have any suggestions?


Scott

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