Re: [PATCH] Revert "usb: core: remove local_irq_save() around ->complete() handler"

[Soeren Moch]

On 12.06.19 16:47, Sebastian Andrzej Siewior wrote:
> On 2019-06-12 10:38:11 [-0400], Alan Stern wrote:
>>> When stopping hostapd after it hangs:
>>> [  903.504475] ieee80211 phy0: rt2x00queue_flush_queue: Warning - Queue
>>> 14 failed to flush
>> Instead of reverting the original commit, can you prevent the problem 
>> by adding local_irq_save() and local_irq_restore() to the URB 
>> completion routines in that wireless driver?
>>
>> Probably people who aren't already pretty familiar with the driver code
>> won't easily be able to locate the race.  Still, a little overkill may
>> be an acceptable solution.
> Not from RT point of view because you do
> 	local_irq_save() -> spin_lock()
>
> but it might be worth checking if the local_irq_save() within that
> wireless driver avoids the race or not.
I just sent a patch for the underlying problem in the rt2x00 driver, see
[1].
So we can drop this patch here and keep RT folks happy.
I really hope that no other usb drivers are affected by similar problems.

Zyxel support just sent me (some?) source code for the firmware of my
internet router (my real problem I mentioned before). Need to look into
this...

Thanks for your help,
Soeren


[1] https://lkml.org/lkml/2019/6/17/238
>> Alan Stern
> Sebastian