Skip to contentSkip to conversation

PROBLEM: kernel BUG in usb-ohci.c:902!

DISCUSSION David Brownell 2 participants 4 messages Sep 29, 2002

lore
Share & tools
lore
#0 lore raw

Hmm, if I look at that line in the 2.4.20-pre7 kernel, I see:

	/* likely some interface's driver has a refcount bug */
	err ("bus %s devnum %d deletion in interrupt",
			ohci->ohci_dev->slot_name, usb_dev->devnum);
	BUG ();

That BUG() has been an "oops now predictably, or oops later randomly" situation, and EVERY (!!!) time it's been an issue, the problem has been traced to exactly what that comment says: some interface's driver has bogus disconnect() processing, and is not cleaning up. (*)

It's likely the problem is in the driver for one of your USB devices. Given what I saw, it sure looks like "prism2_usb" is the problem.

And from what I saw last time I looked at that driver, I can VERY easily believe that ... in fact I think I've seen similar reports show up before, that's why I took a look at it. The prism2_usb folk need to make sure their disconnect() routine (a) prevents all further driver requests to that device, (b) unlinks all urbs it submitted, and (c) waits for them to finish unlinking.

- Dave

(*) Maybe a better response would be to just leak memory, but
that's how the ohci driver evolved ... it took a long time to establish that the oopsing was really caused by bogus disconnect() processing in the layered usb device handling.

#1 lore raw

Actually there is a file of that name 'uhci-hcd' in the 2.5.39 source, which I was talking about. I was also talking about the uhci-hcd driver in the 2.5.39 source which doesn't necessarily have to have a BUG in that code. Here is my call trace:

uhci-irq [uchi-hcd]
usb_hcd_irq_Rfba60562 [usbcore]
handle_IRQ_event
do_IRQ
common_interrupt
page_remove_rmap
zap_pke_range
zap_pmd_range
unmap_page_range
exit_mmap
mmput
exec_mmap
flush_old_exec
load_elf_binary
load_elf_binary
search_binary_handler
do_execve
sys_execve
syscall_call

Code: 83 7b 18 8d 73 04 74 38 c7 04 24 80 f8 88 d0 89 5c 24 04
  <0> Kernel panic:  Aiee, killing interrupt handler!

In interrupt handler - not syncing...

I am apparently hitting a different bug, but it inevitably comes from the uhci-hcd driver (according to the panic).

Regards,

Stephen Marz

Quoted 12 lines
Quoted 2 lines

I have noticed this problem in 2.5.39 except it occurs with the module uhci-hcd.

No you haven't. It doesn't have a file of that name, so you didn't see such a BUG(). And I don't know about you, but my copy of 2.5.39 has no BUG() anywhere in the ohci-hcd driver, so it'd be hard seeing _any_ BUG() coming from there.

You might be hitting a different BUG(), but in that case you would need to get your bug reports straight.

- Dave

#2 lore raw
Quoted 10 lines

code. Here is my call trace:

uhci-irq [uchi-hcd]
usb_hcd_irq_Rfba60562 [usbcore]
handle_IRQ_event
do_IRQ
...

I am apparently hitting a different bug, but it inevitably comes from the uhci-hcd driver (according to the panic).

As I said: you're not seeing "this problem". And it's not a BUG(). So now we agree ... ;-)

You might try sending the full oops report to the maintainer of that driver, or at least to the linux-usb-devel list. (The 2.5.39 code gives more info than you snipped...) So far as I know, this problem has not been reported there.

- Dave

Quoted 27 lines

Regards,

Stephen Marz

Quoted 15 lines
Quoted 2 lines

I have noticed this problem in 2.5.39 except it occurs with the module uhci-hcd.

No you haven't. It doesn't have a file of that name, so you didn't see such a BUG(). And I don't know about you, but my copy of 2.5.39 has no BUG() anywhere in the ohci-hcd driver, so it'd be hard seeing _any_ BUG() coming from there.

You might be hitting a different BUG(), but in that case you would need to get your bug reports straight.

- Dave

#3 lore raw

Its so nice how these things work out :).

Thanks for your time and help,

Stephen Marz

Quoted 59 lines · David Brownell
Quoted 10 lines

code. Here is my call trace:

uhci-irq [uchi-hcd]
usb_hcd_irq_Rfba60562 [usbcore]
handle_IRQ_event
do_IRQ
...

I am apparently hitting a different bug, but it inevitably comes from the uhci-hcd driver (according to the panic).

As I said: you're not seeing "this problem". And it's not a BUG(). So now we agree ... ;-)

You might try sending the full oops report to the maintainer of that driver, or at least to the linux-usb-devel list. (The 2.5.39 code gives more info than you snipped...) So far as I know, this problem has not been reported there.

- Dave

Quoted 31 lines

Regards,

Stephen Marz

Quoted 18 lines
Quoted 2 lines

I have noticed this problem in 2.5.39 except it occurs with the module uhci-hcd.

No you haven't. It doesn't have a file of that name, so you didn't see such a BUG(). And I don't know about you, but my copy of 2.5.39 has no BUG() anywhere in the ohci-hcd driver, so it'd be hard seeing _any_ BUG() coming from there.

You might be hitting a different BUG(), but in that case you would need to get your bug reports straight.

- Dave