2001-07-06 22:32:57

by Rik van Riel

Subject: [PATCH #2] OOM kill trigger

Hi,

As Hugh pointed out, the info on how many pages we have in the
swap cache is (of course) present in the swapper_space structure.

Patch has been shrunk accordingly...

Rik
--
Executive summary of a recent Microsoft press release:
"we are concerned about the GNU General Public License (GPL)"


http://www.surriel.com/
http://www.conectiva.com/ http://distro.conectiva.com/



--- linux-2.4.6/mm/oom_kill.c.orig Fri Jul 6 17:32:58 2001
+++ linux-2.4.6/mm/oom_kill.c Fri Jul 6 19:19:25 2001
@@ -191,11 +191,28 @@
*/
int out_of_memory(void)
{
+ long cache_mem, limit;
+
/* Enough free memory? Not OOM. */
if (nr_free_pages() > freepages.min)
return 0;

if (nr_free_pages() + nr_inactive_clean_pages() > freepages.low)
+ return 0;
+
+ /*
+ * If the buffer and page cache (excluding swap cache) are over
+ * their (/proc tunable) minimum, we're still not OOM. We test
+ * this to make sure we don't return OOM when the system simply
+ * has a hard time with the cache.
+ */
+ cache_mem = atomic_read(&page_cache_size);
+ cache_mem += atomic_read(&buffermem_pages);
+ cache_mem -= swapper_space.nrpages;
+ limit = (page_cache.min_percent + buffer_mem.min_percent);
+ limit *= num_physpages / 100;
+
+ if (cache_mem > limit)
return 0;

/* Enough swap space left? Not OOM. */

2001-07-07 02:12:34

by Jeff Garzik

Subject: Re: [PATCH #2] OOM kill trigger

Rik van Riel wrote:
> + cache_mem = atomic_read(&page_cache_size);
> + cache_mem += atomic_read(&buffermem_pages);
> + cache_mem -= swapper_space.nrpages;
> + limit = (page_cache.min_percent + buffer_mem.min_percent);

Don't you need extra protection around swapper_space.nrpages? A barrier
right above it?

--
Jeff Garzik | A recent study has shown that too much soup
Building 1024 | can cause malaise in laboratory mice.
MandrakeSoft |

2001-07-07 18:06:01

by Rik van Riel

Subject: Re: [PATCH #2] OOM kill trigger

On Fri, 6 Jul 2001, Jeff Garzik wrote:
> Rik van Riel wrote:
> > + cache_mem = atomic_read(&page_cache_size);
> > + cache_mem += atomic_read(&buffermem_pages);
> > + cache_mem -= swapper_space.nrpages;
> > + limit = (page_cache.min_percent + buffer_mem.min_percent);
>
> Don't you need extra protection around swapper_space.nrpages?
> A barrier right above it?

No. It's just a heuristic.

Besides, all that could change is the VALUE of
swapper_space.nrpages and it won't change by all
that much...

Rik
--
Executive summary of a recent Microsoft press release:
"we are concerned about the GNU General Public License (GPL)"


http://www.surriel.com/
http://www.conectiva.com/ http://distro.conectiva.com/