Suspend regression with 2.6.29-rc

Suspend regression with 2.6.29-rc

[Tino Keitel]

Hi,

with 2.6.28, I have sporadic suspend failures with suspend to RAM on my
Mac mini core duo (Intel Core Duo, i945 chipset and graphics).  I
traced it down to the DRM driver: without X and DRM unloaded, I had
more than 200 suspens in a row without any failure.

With 2.6.29-rc6-00029-gf7e603a, even with DRM unloaded the computer
hangs at suspend time after some attempts.  I also tried other
2.6.29-rc kernels, and all had unstable suspend behaviour.  I can't
give any further details, as the console is dark and I have no serial
console.  Any hints what I might try except for bisecting?

Regards,
Tino

Re: Suspend regression with 2.6.29-rc

[Andreas Mohr]

What, no_console_suspend of course ;)

Needless to say I'm also stuck as always, waiting for -rc resume regressions
(due to i945 I think, -rc4 currently) to vanish
(I believe -rc6 wouldn't fix it yet, may try -rc7 then).
How many suspend/resume regression reports did we have in 2.6.29-rc? 5? 6? 7?
When did I first have 30+ working suspend/resumes on a desktop PC? 2001?
2000? 1999?
I'm not entirely sure whether I can achieve similar reliability
on a convincing number of desktops 10 years later...

Andreas Mohr

Re: Suspend regression with 2.6.29-rc

[Tino Keitel]

On Tue, Feb 24, 2009 at 21:52:27 +0100, Andreas Mohr wrote:

[...]

> I'm not entirely sure whether I can achieve similar reliability
> on a convincing number of desktops 10 years later...

Then I guess I should be happy to have _rock_ _solid_ suspend to RAM
and to disk (using TuxOnIce) on my notebook (X61s) with kernel 2.6.27. 
:-) It also comes with Intel graphics, but with i965 instead of i945.

Regards,
Tino

Re: Suspend regression with 2.6.29-rc

[Rafael J. Wysocki]

On Tuesday 24 February 2009, Tino Keitel wrote:
> Hi,
> 
> with 2.6.28, I have sporadic suspend failures with suspend to RAM on my
> Mac mini core duo (Intel Core Duo, i945 chipset and graphics).  I
> traced it down to the DRM driver: without X and DRM unloaded, I had
> more than 200 suspens in a row without any failure.
> 
> With 2.6.29-rc6-00029-gf7e603a, even with DRM unloaded the computer
> hangs at suspend time after some attempts.  I also tried other
> 2.6.29-rc kernels, and all had unstable suspend behaviour.  I can't
> give any further details, as the console is dark and I have no serial
> console.  Any hints what I might try except for bisecting?

Well, this most likely is a result of the recent PCI PM core changes, but
it's hard to say which of them exactly is responsible in this particular case.

Please open a bug at bugzilla.kernel.org with a CC to me and put your system
information in there (the output of 'lspci -vv', /proc/iomem, /proc/interrupts,
boot log).

Also please try to reproduce the problem in the minimal configuration
(init=/bin/bash, /proc and /sys mounted manually).

Thanks,
Rafael

Re: Suspend regression with 2.6.29-rc

[Rafael J. Wysocki]

On Tuesday 24 February 2009, Andreas Mohr wrote:
> What, no_console_suspend of course ;)
> 
> Needless to say I'm also stuck as always, waiting for -rc resume regressions
> (due to i945 I think, -rc4 currently) to vanish
> (I believe -rc6 wouldn't fix it yet, may try -rc7 then).
> How many suspend/resume regression reports did we have in 2.6.29-rc? 5? 6? 7?

Probably more. :-(

Unfortunately, we had too many changes affecting suspend-resume during this
cycle, but that only turned out after the fact.  Moreover, all of them were
done in response to problem reports, but our fixes happened to break things on
systems that were not affected by the original issues being fixed.

> When did I first have 30+ working suspend/resumes on a desktop PC? 2001?
> 2000? 1999?
> I'm not entirely sure whether I can achieve similar reliability
> on a convincing number of desktops 10 years later...

Actually, I can.

Thanks,
Rafael