Re: [BUG] 4.11.0-rc1 panic on shutdown X61s

From: Borislav Petkov
Date: Sun Mar 12 2017 - 18:27:48 EST


On Sun, Mar 12, 2017 at 03:55:08PM +0200, Andy Shevchenko wrote:
> On Sun, Mar 12, 2017 at 2:26 PM, Borislav Petkov <bp@xxxxxxxxx> wrote:
> > On Sun, Mar 12, 2017 at 12:57:03PM +0100, Borislav Petkov wrote:
> >> On Sat, Mar 11, 2017 at 09:37:23PM -0800, lkml@xxxxxxxxxxx wrote:
> >> > Hello list,
> >> >
> >> > Here's a photo of the panic, on imgur to be kind to vger:
> >> > http://imgur.com/a/wZI32
> >> >
> >> > I'm out on a sailboat so can't really do much, but had a chance with internet
> >>
> >> So you didn't bring another box with you on the sailboat to connect it to the
> >> laptop over netconsole to catch full dmesg, did you?
> >
> > Hahah, you're so in luck: I just sent this mail and hibernated my laptop
> > and got the same BUG. What's the chance of that happening?! Apparently
> > big enough.
> >
> > But I was able to catch the warning before it too. So the question is,
> > do you have an e1000e eth controller in that machine too?
> >
> > Because the symptoms below are consistent with the observed behavior:
> > e1000e fails to initialize MSI interrupts for whatever reason and falls
> > back to legacy interrupts.
> >
> > Then, PCI core shuts down and BUGs because the msi_list is empty.
> >
> > Anyway, lemme add e1000e people too to the fun thread.
> >
>
> The only change that IMHO matters happened between v4.10 and v4.11-rc1 is this:
>
> @@ -6276,8 +6274,8 @@ static int e1000e_pm_freeze(struct device *dev)
> /* Quiesce the device without resetting the hardware */
> e1000e_down(adapter, false);
> e1000_free_irq(adapter);
> + e1000e_reset_interrupt_capability(adapter);
> }
> - e1000e_reset_interrupt_capability(adapter);
>
> So, it apparently misses something for the other case, like
> pci_disable_msi() call or so.

Well, lemme add the people from

7e54d9d063fa ("e1000e: driver trying to free already-free irq")

to CC then. :-)

Thanks.

--
Regards/Gruss,
Boris.

Good mailing practices for 400: avoid top-posting and trim the reply.