v4.9 to v4.10 regression: oops when USB cable is plugged in.

[Tony Lindgren]

* Pali Rohár <pali.rohar at gmail.com> [170124 02:02]:
> On Tuesday 24 January 2017 10:18:17 Pavel Machek wrote:
> > Hi!
> > On Mon 2017-01-23 14:44:54, Tony Lindgren wrote:
> > > * Pavel Machek <pavel at ucw.cz> [170123 14:26]:
> > > > [25392.239837] Unhandled fault: external abort on non-linefetch (0x1028) at 0xfa0ab060
> > > > [25392.239868] pgd = c0004000
> > > > [25392.239898] [fa0ab060] *pgd=48011452(bad)
> > > > [25392.239929] Internal error: : 1028 [#1] ARM
> > > > [25392.239929] Modules linked in:
> > > > [25392.239959] CPU: 0 PID: 24322 Comm: kworker/0:1 Not tainted 4.10.0-rc5-142127-g41f2839-dirty #222
> > > > [25392.239990] Hardware name: Nokia RX-51 board
> > > > [25392.240020] Workqueue: events musb_irq_work
> > > > [25392.240051] task: cd44d5c0 task.stack: cd308000
> > > > [25392.240051] PC is at musb_default_readb+0x0/0xc
> > > > [25392.240081] LR is at musb_irq_work+0x1c/0x1b0
> > > 
> > > OK I'm pretty sure the patch I posted few days ago fixes
> > > this. Can you please test patch "[PATCH] usb: musb: Fix
> > > external abort on non-linefetch for musb_irq_work()"?
> > 
> > Can I get the copy of the patch?
> > 
> > http://www.spinics.net/lists/linux-usb/msg152542.html
> > 
> > ...but it is html mangled with no obvious way to unmangle it.

Bounced it to you. FYI, patchwork.kernel.org should have it too, the
"mbox" option there works the best.

> Another place when caller of pm_runtime_get_sync forgot to check return
> value? This is not first time I see this problem related to Nokia N900!

No that's a completely missing pm_runtime_get on this one that's the
most likely cause.

> In past I already suggested to use gcc attribute for pm_runtime_get_sync
> to issue warning when caller does not check return value.

Yeah that would be nice, needs all the missing use cases fixed first
though..

> > > I was able to hit that only once so far, do you hit it
> > > every time with your built-in g_ether .config?
> > 
> > I get it "way too often", like once a day. I don't yet know how to hit
> > it reliably :-(.

OK, well let's hope the patch linked above fixes it. At some point the
number of musb fixes should just start going down if I'm predicting right :)

Anyways, hitting these issues during late -rc cycle is too late. We
really should have some n900 usability testing for core features with
Linux next on at least weekly basis.

I've noticed that testing with Linux next is way less effort than chasing
bugs every -rc cycle when it's too late. For about past four months or so
next has been usable for me with only occasional minor issues that get
fixed within a day or two.

Regards,

Tony