Hi!
> > When the machine doesn't well handle the e820 persistent when hibernate
> > resuming, then it may causes page fault when writing image to snapshot
> > buffer:
> >
> > [ 17.929495] BUG: unable to handle kernel paging request at
> > ffff880069d4f000
> > [ 17.933469] IP: [<ffffffff810a1cf0>] load_image_lzo+0x810/0xe40
> > [ 17.933469] PGD 2194067 PUD 77ffff067 PMD 2197067 PTE 0
> > [ 17.933469] Oops: 0002 [#1] SMP
> > ...
> >
> > The ffff880069d4f000 page is in e820 reserved region of resume boot
> > kernel:
> >
> > [ 0.000000] BIOS-e820: [mem 0x0000000069d4f000-0x0000000069e12fff]
> > reserved
> > ...
> > [ 0.000000] PM: Registered nosave memory: [mem 0x69d4f000-0x69e12fff]
> >
> > So snapshot.c mark the pfn to forbidden pages map. But, this
> > page is also in the memory bitmap in snapshot image because it's an
> > original page used by image kernel, so it will also mark as an
> > unsafe(free) page in prepare_image().
> >
> > That means the page in e820 when resuming mark as "forbidden" and
> > "free", it causes get_buffer() treat it as an allocated unsafe page.
> > Then snapshot_write_next() return this page to load_image, load_image
> > writing content to this address, but this page didn't really allocated
> > . So, we got page fault.
> >
> > Although the root cause is from BIOS, I think aggressive check and
> > significant message in kernel will better then a page fault for
> > issue tracking, especially when serial console unavailable.
> >
> > This patch adds code in mark_unsafe_pages() for check does free pages in
> > nosave region. If so, then it print message and return fault to stop whole
> > S4 resume process:
> >
> > [ 8.166004] PM: Image loading progress: 0%
> > [ 8.658717] PM: 0x6796c000 in e820 nosave region: [mem
> > 0x6796c000-0x6796cfff]
> > [ 8.918737] PM: Read 2511940 kbytes in 1.04 seconds (2415.32 MB/s)
> > [ 8.926633] PM: Error -14 resuming
> > [ 8.933534] PM: Failed to load hibernation image, recovering.
> >
> > v2:
> > + removed empty check of nosave_regions list.
> > + fixed the typo of "region" in code for error message and patch comment.
> >
> > Cc: "Rafael J. Wysocki" <r...@rjwysocki.net>
> > Cc: Len Brown <len.br...@intel.com>
> > Cc: Takashi Iwai <ti...@suse.de>
> > Acked-by: Pavel Machek <pa...@ucw.cz>
> > Signed-off-by: Lee, Chun-Yi <j...@suse.com>
>
> I discussed with Vojtech Pavlik for this patch, he raised a situation is:
>
> Maybe e820 changed but image kernel original pages do not fall into new e820
> region.
> Then the hibernate will recovery success, but later kernel drivers may got
> problem
> when accessing memory.
>
> My idea is hashing the start/end pfn of each nosave region sequentially, put
> this
> nosave region digest to hibernate header then compare e820 digest in
> check_header()
> when hibernate resuming.
>
> I am developing patch, then we don't need check unsafe page should not in
> unsave(e820)
> regions.
Actually, if you are doing such a check... it makes sense to check for
_all_ the regions, nosave or not. If e820 map changed at all, it is
not safe to resume.
Pavel
--
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures)
http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
