On Mon, 3 Mar 2008, Christian Kujau wrote:

quoted text
> On Mon, 3 Mar 2008, Christian Kujau wrote:
> > after upgrading to 2.6.25-rc3, kern.log shows:
> >
> > [ 1535.884848] ------------[ cut here ]------------
> > [ 1535.884855] WARNING: at drivers/usb/host/ehci-hcd.c:287
> 
> Hm, after ~24h of uptime, this message appeared 25 times already.
> The WLAN USB device was used the whole time, the usb-storage module
> was hardly used over the day:
> 
> # grep usb /proc/interrupts
>   10:   10693403    XT-PIC-XT        ohci_hcd:usb3, eth0
>   11:   10270770    XT-PIC-XT        sym53c8xx, ehci_hcd:usb1
>   12:         91    XT-PIC-XT        ohci_hcd:usb2
> 
> I even tried to trigger the system freeze[0] by using usb-storage
> and reading a lot from it, but no freeze, and the message could
> not be triggered either - they pop up every now and then, but too often, 
> IMHO.
> 
> Can anybody shed some light on this?

Dave, it seems to me that this must be an example of a race between the 
iaa watchdog timer expiring and end_unlink_async() running.  It's not 
good enough for end_unlink_async() to call iaa_watchdog_done(), because 
on an SMP system the timer may already have fired and the watchdog 
routine may be spinning on ehci->lock.

It's even worse, because end_unlink_async() can call 
start_unlink_async().  If that happens, the watchdog routine will think 
the timer has expired when in fact it has just been restarted.

How about replacing the

	WARN_ON(!ehci->reclaim);

line in ehci_iaa_watchdog() with

	if (unlikely(!ehci->reclaim || 
			!HC_IS_RUNNING(ehci_to_hcd(ehci)->state) ||
			timer_pending(&ehci->iaa_watchdog))
		goto done;

where "done:" is added just before the spin_unlock_irqrestore?

Alan Stern

--
unsubscribe notice
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/