Bug#611750: Bug#613790: 2nd update (Re: Bug#613790: pm-utils: eeepc 1005 PE recently stopped resuming from hibernate)

[Frédéric Boiteux <fboiteux@calistel.com>]

Le Mon, 21 Feb 2011 17:08:58 +0000,
Ben Hutchings <ben@decadent.org.uk> a écrit :

> On Mon, Feb 21, 2011 at 05:58:52PM +0100, Frédéric Boiteux wrote:
> > 	Hello,
> > 
> >   I've tried a self-compiled Debian 2.6.32-30 kernel without
> > kprobes, as suggested in bug #611551, but it didn't work for me,
> > after hibernation, my eeepc (1002HA) didn't resume, but returned to
> > Grub menu after few seconds.
> >   I've then tried a self-compiled Debian 2.6.32-30 with the 3
> > drm-i915 related patches reverted :
> > drm-i915-free-hardware-status-page-on-unload-when-physically-mapped.patch.bz2
> > drm-i915-overlay-ensure-that-the-reg_bo-is-in-the-gtt-prior-to-writing.patch.bz2
> > drm-i915-sdvo-always-add-a-30ms-delay-to-make-sdvo-tv-detection-reliable.patch.bz2
> > 
> >   But I had the same result, the hibernation is still broken :-(
> > 
> >   At now, I've returned to 2.6.32-29 which is OK.
> > 
> >   I've tried one time to do a git-bisect to find a bug, but I don't
> > know if it's usable with a Debian-patched kernel ?
>  
> If you clone git://git.debian.org/kernel/linux-2.6.git you can
> then use 'git bisect debian/2.6.32-30 debian/2.6.32-29'.
> 
> At least, you will after I update it later today.
> 
> Ben.
> 

	Hello Ben,

  I've actually found the culprit's commit. I has been harder than I
thought, as at first git bisect process, I was always finding 'good'
systems, ending with latest commit of 2.6.30 (which is unrelated, about
powerpc) :-( I doubted of my rebuild system (I wasn't do a 'make clean'
between kernel recompilations, I'm not sure if it's mandatory) and
finally found that the bug (resumimg from hibernation fails), I had to
wait some time, and remove the current plug for some time, or at least
do a cold boot.

I've found following commit as problematic :

eb28b7f273d768a62f58baacb7e8473481b05dc5 is first bad commit
commit eb28b7f273d768a62f58baacb7e8473481b05dc5
Author: Ben Hutchings <ben@decadent.org.uk>
Date:   Thu Feb 17 22:22:26 2011 +0000

    Revert most of "block: Deprecate QUEUE_FLAG_CLUSTER ..."
    
    This reverts the API- and ABI-breaking changes in commit
    e692cb668fdd5a712c6ed2a2d6f2a36ee83997b4 upstream, included in
    stable update 2.6.32.28.
    
    Restore both blk_queue_limits::no_cluster and QUEUE_FLAG_CLUSTER.
    Modify blk_queue_cluster() to check both of them, to cope with any
    drivers that fail to update them both.
    
    Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

:040000 040000 21b8932d10749394aacbb2f07c8e6b7e8f3256f9
2998f524ab5f85f9b894c3421aee2391b33d509c M      block :040000 040000
af298ebe2fac399537275d8515516964c1ffeb5c
46916a3bcd8b182d6f2b314b1b4b3125d7ac3563 M      drivers :040000 040000
53352a9d2a102c31a98a8e55370c4841e7b3b3ea
193a630d1fdb9a107f3e04f80bd580e1671a2294 M      include

I've tried to build then a kernel 2.6.32-30 with this specific patch,
and my tests were successful, hibertation worked here.

  I'm not enough experienced in kernel dev to understand why this
commit has a problem, I'm relying on you ;-)


   Hope this can be helpful for others too, and perhaps in time for
next Squeeze release...

     with regards,
	Fred.