Bug 38582

Created attachment 64062 [details]
dmesg

Created attachment 64072 [details]
lspci -vv

Created attachment 64082 [details]
/proc/interrupts

Created attachment 64092 [details]
/proc/modules

As briefly described in 

20110630220050.GB30291@marc.osknowledge.org

I cannot resume the box after successful suspend with 3.0.0-rc5 with the
following configuration:

Hi there,

has anyone been testing this? I re-tested it with git HEAD some minutes ago and the issue still persists. As I assume 3.0.0 will be released very soon it would be great if anyone could advise me what to do or send me patches I could test.

I really consider this a show stopper...

Marc

Changed the affected kernel version to rc1.

did 2.6.39 work?
does 3.0 work or fail?
If Yes and No, respectively, can you bisect to find
what commit broke the feature?

you mention 32-bit in the description, does that mean 64-bit works
and 32-bit fails?

have you tried the debug steps here?
Documentation/power/basic-pm-debugging.txt

OK, here goes:

a) 2.6.39.3 works absolutely fine.
b) 3.0 final doesn't work
c) 3.0-rc1 didn't work either
d) I cannot bisect it as already stated on LKML due to having not much time currently
e) I use 32-bit instalation on the 64-bit hardware, but I did not try a 64-bit installation. Thus I cannot say if 64-bit is broken as well.
d) Does testing that make sense? The box hibernates as usual, but upon resume the screens stays black (no backlight). Testing proc freezing and such might not bring me any further, does it?

Regards,
Marc

(In reply to comment #9)
> Testing proc freezing and such might not bring me any further, does it?

In fact, it might give us a clue.  Without any more information we can't
really do anything here.

Do you use i915?

Hi Rafael,

so, I still have no time to bisect it right now. I guess when I have time for it at least 3.1 will be out thus that won't bring us any further. I'm really sorry I can't do that, but you know there's nothing more bad and time-consuming to bisect than resume issues.

I do not use i915, just nouveau or maybe the blob when I need CUDA.

Please come back to me in case you need any more info...

Regards,
Marc

I suspect your hibernate problem is related to the graphics driver.

Perhaps you can verify this somehow?

Hi Rafael,

I suspended the box without the blob and/or nouveau loaded. It didn't resume as before.

I attach the current .config of the box.

Regards,
Marc

Created attachment 68982 [details]
.config 3.1.0-rc2-dezzy-00324-g93ee7a9-dirty

(In reply to comment #9)
> The box hibernates as usual, but upon resume
> the screens stays black (no backlight).

Does capslock toggle its LED?  Does networking work --- can you log into the machine remotely?  When you switch between a graphical and a text console with ctrl-alt-f1 / f7, does that affect the backlist? Can you get a log from this state (for example using netconsole)?

s/backlist/backlight, sorry.

(In reply to comment #11)
> so, I still have no time to bisect it right now.

If you get a chance, please try reverting 983bbf1af066 (x86: Don't unmask disabled irqs when migrating them, 2011-05-06).

Hey all,

I reverted the commit on to of current HEAD but no luck - the box still hangs. The backlight doesn't turn on and NUMLOCK doesn't toggle. It hangs in such an early stage that not even disc IO happens (besides one short flashing of the disc LED which is most probably the disc spinning up or something).

Now that I read it - you meant CAPSLOCK, not NUMLOCK. Wait, I'll give it a second try...

OK, CAPSLOCK toggles... :)

I close this one (finally after the long downtime) as I cannot reproduce this anymore.

I was able to trigger this even with 2.6.39.4 and thus it's a .config option triggered issue that makes this happen.

However, I lost the .config which resulted in this issue to happen.