* MIGRATE_RESERVE pages in show_mem function problems @ 2014-05-28 2:24 Wang, Yalin 2014-05-30 14:59 ` Vlastimil Babka 0 siblings, 1 reply; 3+ messages in thread From: Wang, Yalin @ 2014-05-28 2:24 UTC (permalink / raw) To: 'akpm@linux-foundation.org', 'linux-mm@kvack.org', 'linux-kernel@vger.kernel.org', 'cody@linux.vnet.ibm.com', 'linux-arch-owner@vger.kernel.org', 'Will Deacon', 'hannes@cmpxchg.org' Hi I find the show_mem function show page MIGRATE types result is not correct for MIGRATE_RESERVE pages : Normal: 1582*4kB (UEMC) 1317*8kB (UEMC) 1020*16kB (UEMC) 450*32kB (UEMC) 206*64kB (UEMC) 40*128kB (UM) 10*256kB (UM) 10*512kB (UM) 1*1024kB (M) 0*2048kB 0*4096kB = 74592kB Some pages should be marked (R) , while it is changed into MIGRATE_MOVEABLE or UNMOVEABLE in free_area list , It's not correct for debug . I make a patch for this: diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 5dba293..6ef8ebe 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -1198,7 +1198,8 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order, list_add_tail(&page->lru, list); if (IS_ENABLED(CONFIG_CMA)) { mt = get_pageblock_migratetype(page); - if (!is_migrate_cma(mt) && !is_migrate_isolate(mt)) + if (!is_migrate_cma(mt) && !is_migrate_isolate(mt) + && mt != MIGRATE_RESERVE) mt = migratetype; } set_freepage_migratetype(page, mt); seems work ok , I am curious is it a BUG ? or designed like this for some reason ? Thanks <6>[ 250.751554] lowmem_reserve[]: 0 0 0 <6>[ 250.751606] Normal: 1582*4kB (UEMC) 1317*8kB (UEMC) 1020*16kB (UEMC) 450*32kB (UEMC) 206*64kB (UEMC) 40*128kB (UM) 10*256kB (UM) 10*512kB (UM) 1*1024kB (M) 0*2048kB 0*4096kB = 74592kB <6>[ 250.751848] HighMem: 167*4kB (UC) 3*8kB (U) 0*16kB 0*32kB 0*64kB 0*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 692kB <6>[ 250.752020] 62596 total pagecache pages <6>[ 250.752046] 0 pages in swap cache <6>[ 250.752074] Swap cache stats: add 0, delete 0, find 0/0 Sony Mobile Communications Tel: My Number +18610323092 yalin.wang@sonymobile.com sonymobile.com -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a> ^ permalink raw reply related [flat|nested] 3+ messages in thread
* Re: MIGRATE_RESERVE pages in show_mem function problems 2014-05-28 2:24 MIGRATE_RESERVE pages in show_mem function problems Wang, Yalin @ 2014-05-30 14:59 ` Vlastimil Babka 2014-06-03 6:43 ` Wang, Yalin 0 siblings, 1 reply; 3+ messages in thread From: Vlastimil Babka @ 2014-05-30 14:59 UTC (permalink / raw) To: Wang, Yalin Cc: 'akpm@linux-foundation.org', 'linux-mm@kvack.org', 'linux-kernel@vger.kernel.org', 'cody@linux.vnet.ibm.com', 'linux-arch-owner@vger.kernel.org', 'Will Deacon', 'hannes@cmpxchg.org', Joonsoo Kim On 05/28/2014 04:24 AM, Wang, Yalin wrote: > Hi > > I find the show_mem function show page MIGRATE types result is not correct for > MIGRATE_RESERVE pages : > > Normal: 1582*4kB (UEMC) 1317*8kB (UEMC) 1020*16kB (UEMC) 450*32kB (UEMC) 206*64kB (UEMC) 40*128kB (UM) 10*256kB (UM) 10*512kB (UM) 1*1024kB (M) 0*2048kB 0*4096kB = 74592kB > > Some pages should be marked (R) , while it is changed into MIGRATE_MOVEABLE or UNMOVEABLE in free_area list , > It's not correct for debug . > I make a patch for this: > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c > index 5dba293..6ef8ebe 100644 > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -1198,7 +1198,8 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order, > list_add_tail(&page->lru, list); > if (IS_ENABLED(CONFIG_CMA)) { > mt = get_pageblock_migratetype(page); > - if (!is_migrate_cma(mt) && !is_migrate_isolate(mt)) > + if (!is_migrate_cma(mt) && !is_migrate_isolate(mt) > + && mt != MIGRATE_RESERVE) > mt = migratetype; > } > set_freepage_migratetype(page, mt); > > > seems work ok , I am curious is it a BUG ? or designed like this for some reason ? Hi, this is a known problem that should be fixed for the rmqueue_bulk() part by this patch: http://www.ozlabs.org/~akpm/mmotm/broken-out/mm-page_alloc-prevent-migrate_reserve-pages-from-being-misplaced.patch Testing is welcome if you can reproduce it easily enough. Note that even with the patch, MIGRATE_RESERVE pageblocks can still disappear for two reasons: - when MAX_ORDER-1 > pageblock_order (such as x86_64), there can be a single MIGRATE_RESERVE pageblock created for a smaller zone, and get merged with !MIGRATE_RESERVE buddy pageblock - when min_free_kbytes sysctl is used, creation of new MIGRATE_RESERVE pageblocks can race with other CPUs putting pages in their pcplists, and then freeing then on a wronge free_list. If try_to_steal_freepages happens to find such misplaced page, it might remark the pageblock. I think the second problem is extremely rare, but Joonsoo Kim confirmed the first one to happen. You can check the -mm archives for threads around the patch above. Vlastimil > Thanks > > > <6>[ 250.751554] lowmem_reserve[]: 0 0 0 > <6>[ 250.751606] Normal: 1582*4kB (UEMC) 1317*8kB (UEMC) 1020*16kB (UEMC) 450*32kB (UEMC) 206*64kB (UEMC) 40*128kB (UM) 10*256kB (UM) 10*512kB (UM) 1*1024kB (M) 0*2048kB 0*4096kB = 74592kB > <6>[ 250.751848] HighMem: 167*4kB (UC) 3*8kB (U) 0*16kB 0*32kB 0*64kB 0*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 692kB > <6>[ 250.752020] 62596 total pagecache pages > <6>[ 250.752046] 0 pages in swap cache > <6>[ 250.752074] Swap cache stats: add 0, delete 0, find 0/0 > > > > > Sony Mobile Communications > Tel: My Number +18610323092 > yalin.wang@sonymobile.com > sonymobile.com > > > > -- > To unsubscribe, send a message with 'unsubscribe linux-mm' in > the body to majordomo@kvack.org. For more info on Linux MM, > see: http://www.linux-mm.org/ . > Don't email: <a href=ilto:"dont@kvack.org"> email@kvack.org </a> > -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a> ^ permalink raw reply [flat|nested] 3+ messages in thread
* RE: MIGRATE_RESERVE pages in show_mem function problems 2014-05-30 14:59 ` Vlastimil Babka @ 2014-06-03 6:43 ` Wang, Yalin 0 siblings, 0 replies; 3+ messages in thread From: Wang, Yalin @ 2014-06-03 6:43 UTC (permalink / raw) To: 'Vlastimil Babka' Cc: 'akpm@linux-foundation.org', 'linux-mm@kvack.org', 'linux-kernel@vger.kernel.org', 'cody@linux.vnet.ibm.com', 'linux-arch-owner@vger.kernel.org', 'Will Deacon', 'hannes@cmpxchg.org', Joonsoo Kim Hi I see, Your patch should be ok to fix this problem, Could I know if this patch will be merged into kernel mainline branch? Thanks -----Original Message----- From: Vlastimil Babka [mailto:vbabka@suse.cz] Sent: Friday, May 30, 2014 10:59 PM To: Wang, Yalin Cc: 'akpm@linux-foundation.org'; 'linux-mm@kvack.org'; 'linux-kernel@vger.kernel.org'; 'cody@linux.vnet.ibm.com'; 'linux-arch-owner@vger.kernel.org'; 'Will Deacon'; 'hannes@cmpxchg.org'; Joonsoo Kim Subject: Re: MIGRATE_RESERVE pages in show_mem function problems On 05/28/2014 04:24 AM, Wang, Yalin wrote: > Hi > > I find the show_mem function show page MIGRATE types result is not > correct for MIGRATE_RESERVE pages : > > Normal: 1582*4kB (UEMC) 1317*8kB (UEMC) 1020*16kB (UEMC) 450*32kB > (UEMC) 206*64kB (UEMC) 40*128kB (UM) 10*256kB (UM) 10*512kB (UM) > 1*1024kB (M) 0*2048kB 0*4096kB = 74592kB > > Some pages should be marked (R) , while it is changed into > MIGRATE_MOVEABLE or UNMOVEABLE in free_area list , It's not correct for debug . > I make a patch for this: > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 5dba293..6ef8ebe > 100644 > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -1198,7 +1198,8 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order, > list_add_tail(&page->lru, list); > if (IS_ENABLED(CONFIG_CMA)) { > mt = get_pageblock_migratetype(page); > - if (!is_migrate_cma(mt) && !is_migrate_isolate(mt)) > + if (!is_migrate_cma(mt) && !is_migrate_isolate(mt) > + && mt != MIGRATE_RESERVE) > mt = migratetype; > } > set_freepage_migratetype(page, mt); > > > seems work ok , I am curious is it a BUG ? or designed like this for some reason ? Hi, this is a known problem that should be fixed for the rmqueue_bulk() part by this patch: http://www.ozlabs.org/~akpm/mmotm/broken-out/mm-page_alloc-prevent-migrate_reserve-pages-from-being-misplaced.patch Testing is welcome if you can reproduce it easily enough. Note that even with the patch, MIGRATE_RESERVE pageblocks can still disappear for two reasons: - when MAX_ORDER-1 > pageblock_order (such as x86_64), there can be a single MIGRATE_RESERVE pageblock created for a smaller zone, and get merged with !MIGRATE_RESERVE buddy pageblock - when min_free_kbytes sysctl is used, creation of new MIGRATE_RESERVE pageblocks can race with other CPUs putting pages in their pcplists, and then freeing then on a wronge free_list. If try_to_steal_freepages happens to find such misplaced page, it might remark the pageblock. I think the second problem is extremely rare, but Joonsoo Kim confirmed the first one to happen. You can check the -mm archives for threads around the patch above. Vlastimil > Thanks > > > <6>[ 250.751554] lowmem_reserve[]: 0 0 0 <6>[ 250.751606] Normal: > 1582*4kB (UEMC) 1317*8kB (UEMC) 1020*16kB (UEMC) 450*32kB (UEMC) > 206*64kB (UEMC) 40*128kB (UM) 10*256kB (UM) 10*512kB (UM) 1*1024kB (M) > 0*2048kB 0*4096kB = 74592kB <6>[ 250.751848] HighMem: 167*4kB (UC) > 3*8kB (U) 0*16kB 0*32kB 0*64kB 0*128kB 0*256kB 0*512kB 0*1024kB > 0*2048kB 0*4096kB = 692kB <6>[ 250.752020] 62596 total pagecache > pages <6>[ 250.752046] 0 pages in swap cache <6>[ 250.752074] Swap > cache stats: add 0, delete 0, find 0/0 > > > > > Sony Mobile Communications > Tel: My Number +18610323092 > yalin.wang@sonymobile.com > sonymobile.com > > > > -- > To unsubscribe, send a message with 'unsubscribe linux-mm' in the body > to majordomo@kvack.org. For more info on Linux MM, > see: http://www.linux-mm.org/ . > Don't email: <a href=ilto:"dont@kvack.org"> email@kvack.org </a> > -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a> ^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2014-06-03 6:43 UTC | newest] Thread overview: 3+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2014-05-28 2:24 MIGRATE_RESERVE pages in show_mem function problems Wang, Yalin 2014-05-30 14:59 ` Vlastimil Babka 2014-06-03 6:43 ` Wang, Yalin
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).