lkml.org 
[lkml]   [2003]   [Apr]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Fix SWSUSP & !SWAP
>> > I don't believer I've ever seen things get OOM killed. Instead, page
>> > cache is discarded until things do fit.
>>
>> What happens if user allocated pages are filling up all the space,
>> not page cache? Trust me, it happens ;-)
>
> Yep, just because I haven't seen it, doesn't mean a thing. :>. In that
> case, there are two issues: memory to work in to start with and how to
> save the image without corrupting it.
>
> Regarding #1, there must still be some memory available, mustn't there?
> swsusp only approx (nrpages in use/256) pages to do its work. Surely
> we'd always be able to get .4% of the number of pages? Even if we can't
> get that many, we should be able to adjust the algorithm to be able to
> suspend a machine with only 10 or so pages available to start with (no,
> I'm not volunteering to do it! I want to merge with 2.5 and get on to
> other projects!).

Well, if you have the pre-reserved area, you just start swapping pages out
into it. If we need a few pages (like 10) to manage data structures, etc,
we can just allocate those at boot, and keep them ready.

> Regarding #2, my algorithm (ie not the version in 2.5 at the mo)
> separates pages to be saved into 2 types. Pageset1 are pages we expect
> to be needed during suspend. Pageset2 is those that will definitely not
> be needed. My algorithm for saving the data goes: Save pageset2 pages to
> disk then (as per the original/current method) make a copy of pageset1
> pages (using the pageset2 locations + extra allocated memory if needsbe)
> and save the copy. Loading the image is the reverse process. Pageset 2
> currently only consists of all highmem pages + active and inactive list
> pages.

Not sure that quite works ... don't you need PTE's to know what to swap
out? Those can be in highmem. However, all user pages would fit in pageset
2, I think.

> If we refined the algorithm, perhaps that would address your
> issue. The other point here is that since we have to be able to make a
> copy of pageset1 pages, and since I haven't inlined kmap/unmap in the
> routine to copy pageset1 pages back on resume (Pavel will say whew to
> that, I'm sure!), pageset1 has a miximum size of half normal memory. I
> reckon refining the algoritm so that pageset1 can be [nearly] guaranteed
> to always be smaller is the better area to focus on, and I'm perfectly
> happy to try suggestions, particularly when they come in the form of a
> code fragment that include a call to SetPagePageset2(struct page * page)
> for the relevant targets :>

The more I think about this, the more it seems so much simpler to just
require a reserved swap area the size of your RAM to suspend into. Would
make the code so much simpler ... forget the option bit I suggested earlier
;-)

M.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:34    [from the cache]
©2003-2014 Jasper Spaans. Advertise on this site