Re: [Update][PATCH] MM / PM: Force GFP_NOIO during suspend/hibernation and resume

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Saturday 30 January 2010, Maxim Levitsky wrote:
> On Sat, 2010-01-30 at 19:56 +0100, Rafael J. Wysocki wrote: 
> > On Monday 25 January 2010, Alexey Starikovskiy wrote:
> > > Rafael J. Wysocki пишет:
> > > > On Saturday 23 January 2010, Maxim Levitsky wrote:
> > > >> On Fri, 2010-01-22 at 22:19 +0100, Rafael J. Wysocki wrote: 
> > > >>> On Friday 22 January 2010, Maxim Levitsky wrote:
> > > >>>> On Fri, 2010-01-22 at 10:42 +0900, KOSAKI Motohiro wrote: 
> > > >>>>>>>> Probably we have multiple option. but I don't think GFP_NOIO is good
> > > >>>>>>>> option. It assume the system have lots non-dirty cache memory and it isn't
> > > >>>>>>>> guranteed.
> > > >>>>>>> Basically nothing is guaranteed in this case.  However, does it actually make
> > > >>>>>>> things _worse_?  
> > > >>>>>> Hmm..
> > > >>>>>> Do you mean we don't need to prevent accidental suspend failure?
> > > >>>>>> Perhaps, I did misunderstand your intention. If you think your patch solve
> > > >>>>>> this this issue, I still disagree. but If you think your patch mitigate
> > > >>>>>> the pain of this issue, I agree it. I don't have any reason to oppose your
> > > >>>>>> first patch.
> > > >>>>> One question. Have anyone tested Rafael's $subject patch? 
> > > >>>>> Please post test result. if the issue disapper by the patch, we can
> > > >>>>> suppose the slowness is caused by i/o layer.
> > > >>>> I did.
> > > >>>>
> > > >>>> As far as I could see, patch does solve the problem I described.
> > > >>>>
> > > >>>> Does it affect speed of suspend? I can't say for sure. It seems to be
> > > >>>> the same.
> > > >>> Thanks for testing.
> > > >> I'll test that too, soon.
> > > >> Just to note that I left my hibernate loop run overnight, and now I am
> > > >> posting from my notebook after it did 590 hibernate cycles.
> > > > 
> > > > Did you have a chance to test it?
> > > > 
> > > >> Offtopic, but Note that to achieve that I had to stop using global acpi
> > > >> hardware lock. I tried all kinds of things, but for now it just hands
> > > >> from time to time.
> > > >> See http://bugzilla.kernel.org/show_bug.cgi?id=14668
> > > > 
> > > > I'm going to look at that later this week, although I'm not sure I can do more
> > > > than Alex about that.
> > > > 
> > > > Rafael
> > > Rafael,
> > > If you can point to where one may insert callback to be called just before handing control to resume kernel,
> > > it may help...
> > 
> > Generally speaking, I'd do that in a .suspend() callback of one of devices.
> > 
> > If that's inconvenient, you can also place it in the .pre_restore() platform
> > hibernate callback (drivers/acpi/sleep.c).  It only disables GPEs right now,
> > it might release the global lock as well.
> > 
> > The .pre_restore() callback is executed after all devices have been suspended,
> > so there's no danger any driver would re-acquire the global lock after that.
> 
> 
> Well, I did that very late, very close to image restore.
> Still, it didn't work (It hung after the resume, in the kernel that was
> just restored, on access to the hardware lock, or in other words in same
> way)
> 
> Here is what I did:

I saw the patch in the bug entry
(http://bugzilla.kernel.org/show_bug.cgi?id=14668).
Please see the comments in there.

Please also test the patch I attached and let's use the bug entry for the
tracking of this issue from now on.

Rafael