Re: 2.6.31 regression: programs crashing after a couple of days of uptime with hibernation

[Ondrej Zary <linux@rainbow-software.org>]

On Saturday 10 April 2010 21:58:17 Rafael J. Wysocki wrote:
> On Saturday 10 April 2010, Ondrej Zary wrote:
> > On Friday 02 April 2010 18:58:28 Rafael J. Wysocki wrote:
> > > On Friday 02 April 2010, Ondrej Zary wrote:
> > > > On Friday 02 April 2010 00:33:55 Rafael J. Wysocki wrote:
> > > > > On Friday 02 April 2010, Ondrej Zary wrote:
> > > > > > On Thursday 01 April 2010 23:45:33 you wrote:
> > > > > > > On Thursday 01 April 2010, Ondrej Zary wrote:
> > > > > > > > Hello,
> > > > > > > > with kernel 2.6.30, I can have uptime of more than a month on
> > > > > > > > my desktop PC (with hibernation). It's impossible with
> > > > > > > > 2.6.31. After 1 to 3 days, processes that were running during
> > > > > > > > hibernation (e.g. konsole, kwin, kicker, xorg) start to crash
> > > > > > > > randomly in very weird ways. The kernel itself does not seem
> > > > > > > > to crash. When I run the crashed program again, it seems to
> > > > > > > > work. Looks like some memory corruption.
> > > > > > > >
> > > > > > > > This bug is also present in 2.6.32 and 2.6.33.
> > > > > > >
> > > > > > > Is the kernel 32-bit or 64-bit?  What kind of CPU is there in
> > > > > > > the box?
> > > > > >
> > > > > > It's old 32-bit i686 CPU - Cyrix MII.
> > > > >
> > > > > Please try with this patch applied:
> > > > > http://git.kernel.org/?p=linux/kernel/git/mingo/linux-2.6-x86.git;a
> > > > >=pat ch;h =8ae06d223f8203c72104e5c0c4ee49a000aedb42
> > > >
> > > > Thanks, I'll try it. But I doubt that it will fix the problem. There
> > > > were no changes in hibernate_asm_32.S between 2.6.30 and 2.6.31.
> > >
> > > The change that exposed this issue was elsewhere in the x86 arch code. 
> > > I don't know where exactly, but we surely didn't need the above patch
> > > before.
> >
> > It did not help. It started crashing with 2.6.32 and this patch applied
> > after two days.
>
> Well, no idea, then.

So I'll try to bisect it again.

Here's a bug report so it won't get lost: 
https://bugzilla.kernel.org/show_bug.cgi?id=15753

-- 
Ondrej Zary