[lkml]   [2010]   [Jul]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: Memory corruption during hibernation since 2.6.31
    On Thu, 29 Jul 2010 11:44:31 -0700 (PDT)
    Hugh Dickins <> wrote:

    > So, in between snapshotting the image and actually hibernating, we have
    > two parallel universes, one frozen in the image, the other writing that
    > out to swap: with the danger that the latter (which is about to die)
    > will introduce fatal inconsistencies in the former by placing pages in
    > swap locations innocently reallocated from it. (Excuse me while I go
    > write the movie script.)
    > I'd never got to think about that before.
    > Your fix is neat though hacky (it's somewhat of a coincidence that the
    > usage arg happens to distinguish the hibernation case), and should be
    > enough to fix "your" regression, but is it really enough?
    yes, it's hacky and this just hides a problem caused by the patch.

    > I'm worrying about the try_to_free_swap() calls in shrink_page_list():
    > can't those get called from direct reclaim (even if GFP_NOIO), and can't
    > direct reclaim get invoked from somewhere in the I/O path below
    > swap_writepage(), used for writing out the hibernation image?
    IIUC, before hibernation, shirnk_memory() is called and hibernation codes
    expect there are enough memory. But you're right, there are no guarantee.

    I think the best way is kexec(). But maybe rollback from hibernation failure
    will be difficult. Considering how crash-dump works well and under maintainance
    by many enterprise guys, hibernation-by-kexec is a choice. I think. It can make
    reuse of kdump code, ...or, hibernation-resume code can eat kdump image
    directly. Maybe the problem will be the speed of dump.

    Otherwise, we need to take care of status changes in already-saved-memory.
    Maybe anyone don't like to add COW method...


     \ /
      Last update: 2010-07-30 02:09    [W:0.023 / U:6.756 seconds]
    ©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site