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

["Sören Moch" <smoch@web.de>]

On 01.06.19 12:50, Sebastian Andrzej Siewior wrote:
> On 2019-06-01 01:02:37 [+0200], Soeren Moch wrote:
>>> Why not just fix that driver?  Wouldn't that be easier?
>>>
>> I suspect there are more drivers to fix. I only tested WIFI sticks so
>> far, RTL8188 drivers also seem to suffer from this. I'm not sure how to
>> fix all this properly, maybe Sebastian as original patch author can help
>> here.
> Suspecting isn't helping here.
Yes, you are right.
When I encountered this problem half a year ago, I tried some other type
of wifi stick
and immediately ran into trouble with this, too.

Now I did a short test with a RTL8188CUS stick, I could not reproduce
this bug with
this stick so far.

>
>> This patch is mostly for -stable, to get an acceptable solution quickly.
>> It was really annoying to get such unstable WIFI connection over the
>> last three kernel releases to my development board.  Since my internet
>> service provider forcefully updated my router box 3 weeks ago, I
>> unfortunately see the same symptoms on my primary internet access.
>> That's even worse, I need to reset this router box every few days. I'm
>> not sure, however, that this is caused by the same problem, but it feels
>> like this.
>> So can we please fix this regression quickly and workout a proper fix
>> later? In the original patch there is no reason given, why this patch is
>> necessary. With this revert I at least see a stable connection.
> I will look into this. This patch got in in v4.20-rc1 and the final
> kernel was released by the end of 2018. This is the first report I am
> aware of over half year later…
It is not easy to reproduce this bug reliably within a reasonable period
of time. It
took days to bisect this. And usb core code was for sure not my first
candidate
to look at.

Reverting this patch solves the problem, but of course disabling interrupts
also can hide a bug elsewhere.

Soeren