lkml.org 
[lkml]   [2011]   [Nov]   [22]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [PATCH] PM / Hibernation: Fix *massive* memory leak at early exits in hibernation
Date
On Tuesday, November 22, 2011, Srivatsa S. Bhat wrote:
> On 11/22/2011 05:15 PM, Pavel Machek wrote:
> > On Mon 2011-11-21 23:25:39, Rafael J. Wysocki wrote:
> >> On Monday, November 21, 2011, Srivatsa S. Bhat wrote:
> >>> At some of the early exit points during hibernation (exiting either due
> >>> to failure or after a successful hibernation test, the memory pre-allocated
> >>> for hibernation is not freed up. And this is *very* serious, because, during
> >>> pre-allocation, it could have allocated upto a few *gigabytes* of memory!
> >>> And hence, if a hibernation fails or even if we run some hibernation tests
> >>> using the 'pm_test' framework, the system is rendered unstable due to memory
> >>> becoming signifantly lower. Fix this bug.
> >>
> >> While the observation is valid, I'd prefer to do something like the patch
> >> below.
> >
> > The code slowly becomes goto maze :-(.
> >
>
> I agree.. It is already quite a mess.
>
> >> @@ -357,12 +357,14 @@ int hibernation_snapshot(int platform_mo
> >> * successful freezer test.
> >> */
> >> freezer_test_done = true;
> >> - goto Close;
> >> + goto Cleanup;
> >> }
> >>
> >> error = dpm_prepare(PMSG_FREEZE);
> >> - if (error)
> >> - goto Complete_devices;
> >> + if (error) {
> >> + dpm_complete(msg);
> >> + goto Cleanup;
> >> + }
> >
> > Perhaps dpm_prepare should be changed to clean after itself in the
> > error case? That is the normal convention AFAICT....
> >
>
> If the intention here is to merely clean up hibernation_snapshot() code,
> I would not prefer to change the behaviour of dpm_prepare(), considering
> things like, what parameter should we pass to dpm_complete(); is the
> resultant behaviour change in dpm_suspend_start() correct or not; what
> happens to all the code that uses the nice pair: dpm_suspend_start() and
> dpm_resume_end() and so on.
>
> Perhaps there are bigger issues involved there, since I observed on a brief
> look that the current code doesn't seem to strictly follow the above
> convention that whoever called dpm_prepare() should call dpm_complete()
> upon failure. Or may be its doing the right thing.. I don't know.
>
> But anyway, the good news is, even without changing dpm_prepare()'s
> behaviour, we can clean up quite a bit of code in hibernation_snapshot(),
> as it is.
>
> The first patch below does the cleanup, the second patch fixes the memory
> leak and applies on top of the first patch.

Wait, wait. These changes can be made in the 3.3 merge window, while I'd
like the fix the bug _now_.

Does anyone have any _technical_ problem with my patch posted previously
in this thread?

Rafael


\
 
 \ /
  Last update: 2011-11-22 21:31    [W:0.075 / U:0.144 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site