lkml.org 
[lkml]   [2014]   [Mar]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] ARM: machine_power_off should not return
On Wed, Mar 26, 2014 at 07:12:27AM -0300, Ezequiel Garcia wrote:
> On Mar 26, Russell King - ARM Linux wrote:
> > On Tue, Mar 25, 2014 at 07:45:55PM -0300, Ezequiel Garcia wrote:
> > >
> > > Without this patch we got the heartbeat's reboot_notifier called twice while
> > > testing the recent hibernation patches, which was unexpected and produced a
> > > kernel panic: https://lkml.org/lkml/2014/3/19/363
> >
> > I don't see why we should make this change. kernel/reboot.c handles
> > this function returning, so other places should do too.
> >
> > Even on x86, this function can return:
> >
> [..]
> >
> > Therefore, I'd say... it's a bug in the hibernation code - or we probably
> > have many buggy architectures. I'd suggest fixing the hibernation code
> > rather than stuffing some workaround like an endless loop into every
> > architecture.
> >
>
> Which is exactly what Sebastian did first:
>
> https://lkml.org/lkml/2014/3/20/605
>
> But Pavel asked to fix ARM's machine_power_off instead.
>
> Also, looking at the other architectures, it seems this API is not well
> defined: some of them have an infinite loop, some don't. So it's hard to
> say the function is supposed to return or not.

I'm going by x86 (which I regard as definitive) and the generic power-off
kernel code (which I've looked at all the way back to 2.6.12-rc2).

The hibernation code path should really be fixed - the paths in
kernel/reboot.c have coped for 9+ with all of these platform hooks
returning, and it's only the silly switch() in the hibernation code
that doesn't use a "default" case to handle the kernel_halt() case
which is the real cause of the problem.

As you've found, calling kernel_power_off() followed by kernel_halt()
leads to bugs in drivers: this is not an architecture thing, it's partly
a hibernation code failure for doing that, and partly a driver bug for
trying to unregister something that it's already unregistered.

--
FTTC broadband for 0.8mile line: now at 9.7Mbps down 460kbps up... slowly
improving, and getting towards what was expected from it.


\
 
 \ /
  Last update: 2014-03-26 12:41    [W:0.046 / U:1.352 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site