Messages in this thread | | | Date | Mon, 4 Jan 2010 09:47:11 +0900 | Subject | Re: [PATCH 2/3 -mmotm-2009-12-10-17-19] Count zero page as file_rss | From | Minchan Kim <> |
| |
On Mon, Jan 4, 2010 at 8:43 AM, KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote: > On Wed, 30 Dec 2009 16:49:52 +0000 (GMT) > Hugh Dickins <hugh.dickins@tiscali.co.uk> wrote: > >> > > >> > > Kame reported following as >> > > "Before starting zero-page works, I checked "questions" in lkml and >> > > found some reports that some applications start to go OOM after zero-page >> > > removal. >> > > >> > > For me, I know one of my customer's application depends on behavior of >> > > zero page (on RHEL5). So, I tried to add again it before RHEL6 because >> > > I think removal of zero-page corrupts compatibility." >> > > >> > > So how about adding zero page as file_rss again for compatibility? >> >> I think not. >> >> KAMEZAWA-san can correct me (when he returns in the New Year) if I'm >> wrong, but I don't think his customer's OOMs had anything to do with >> whether the ZERO_PAGE was counted in file_rss or not: the OOMs came >> from the fact that many pages were being used up where just the one >> ZERO_PAGE had been good before. Wouldn't he have complained if the >> zero_pfn patches hadn't solved that problem? >> >> You are right that I completely overlooked the issue of whether to >> include the ZERO_PAGE in rss counts (now being a !vm_normal_page, >> it was just natural to leave it out); and I overlooked the fact that >> it used to be counted into file_rss in the old days (being !PageAnon). >> >> So I'm certainly at fault for that, and thank you for bringing the >> issue to attention; but once considered, I can't actually see a good >> reason why we should add code to count ZERO_PAGEs into file_rss now. >> And if this patch falls, then 1/3 and 3/3 would fall also. >> >> And the patch below would be incomplete anyway, wouldn't it? >> There would need to be a matching change to zap_pte_range(), >> but I don't see that. >> >> We really don't want to be adding more and more ZERO_PAGE/zero_pfn >> tests around the place if we can avoid them: KOSAKI-san has a strong >> argument for adding such a test in kernel/futex.c, but I don't the >> argument here. >> > > I agree that ZERO_PAGE shouldn't be counted as rss. Now, I feel that old > counting method(in old zero-page implementation) was bad. > > Minchan-san, I'm sorry for noise.
That's all right. It was my mistake.
I will drop this and repost Matt and Hugh's ACK version. Thanks for all. :)
> > Thanks, > -Kame > > > >
-- Kind regards, Minchan Kim -- 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/
| |