zone->present_pages shoule be: spanned pages - absent pages - bootmem pages(including memmap pages), but now it's: spanned pages - absent pages - memmap pages. And it didn't consider whether the memmap pages is actully allocated from the zone or not which may cause problem when memory hotplug is improved recently. For example: numa node 1 has ZONE_NORMAL and ZONE_MOVABLE, it's memmap and other bootmem allocated from ZONE_MOVABLE. So ZONE_NORMAL's present_pages should be spanned pages - absent pages, but now it also minus memmap pages, which are actually allocated from ZONE_MOVABLE. This is wrong and when offlining all memory of this zone: (zone->present_pages -= offline_pages) will less than 0. Since present_pages is unsigned long type, that is actually a very large integer which will cause zone->watermark[WMARK_MIN] becomes a large integer too(see setup_per_zone_wmarks()). As a result, totalreserve_pages become a large integer also and finally memory allocating will fail in __vm_enough_memory(). Related discuss: http://lkml.org/lkml/2012/11/5/866 https://patchwork.kernel.org/patch/1346751/ Related patches in mmotm: mm: fix-up zone present pages(7f1290f2f2a4d2c) (sometimes cause egression) mm: fix a regression with HIGHMEM(fe2cebd5a259eec) (Andrew have some feedback) Jiang Liu have sent a series patches to fix this issue by adding a managed_pages area to zone struct: [RFT PATCH v1 0/5] fix up inaccurate zone->present_pages But i think it's too complicated. Mine is based on the two related patches already in mmotm(need to revert them first) It fix the calculation of zone->present_pages by: 1. Reset the zone->present_pages to zero before free_all_bootmem(),free_all_bootmem_node() and free_low_memory_core_early(). I think these should already included all path in all arch. 2. If there is a page freed to buddy system in __free_pages_bootmem(), add zone->present_pages accrodingly. Note this patch assumes that bootmem won't use memory above ZONE_HIGHMEM, so only zones below ZONE_HIGHMEM are reset/fixed. If not, some update is needed. For ZONE_HIGHMEM, only fix it's init value to: panned_pages - absent_pages in free_area_init_core(). Only did some simple test currently. Signed-off-by: Jianguo Wu <wujianguo@xxxxxxxxxx> Signed-off-by: Jiang Liu <jiang.liu@xxxxxxxxxx> Signed-off-by: Bob Liu <lliubbo@xxxxxxxxx> --- include/linux/mm.h | 3 +++ mm/bootmem.c | 2 ++ mm/nobootmem.c | 1 + mm/page_alloc.c | 49 +++++++++++++++++++++++++++++++++++++------------ 4 files changed, 43 insertions(+), 12 deletions(-) diff --git a/include/linux/mm.h b/include/linux/mm.h index 7b03cab..3b40eb6 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -1763,5 +1763,8 @@ static inline unsigned int debug_guardpage_minorder(void) { return 0; } static inline bool page_is_guard(struct page *page) { return false; } #endif /* CONFIG_DEBUG_PAGEALLOC */ +extern void reset_lowmem_zone_present_pages_pernode(pg_data_t *pgdat); +extern void reset_lowmem_zone_present_pages(void); + #endif /* __KERNEL__ */ #endif /* _LINUX_MM_H */ diff --git a/mm/bootmem.c b/mm/bootmem.c index 26d057a..661775b 100644 --- a/mm/bootmem.c +++ b/mm/bootmem.c @@ -238,6 +238,7 @@ static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) { register_page_bootmem_info_node(pgdat); + reset_lowmem_zone_present_pages_pernode(pgdat); return free_all_bootmem_core(pgdat->bdata); } @@ -251,6 +252,7 @@ unsigned long __init free_all_bootmem(void) unsigned long total_pages = 0; bootmem_data_t *bdata; + reset_lowmem_zone_present_pages(); list_for_each_entry(bdata, &bdata_list, list) total_pages += free_all_bootmem_core(bdata); diff --git a/mm/nobootmem.c b/mm/nobootmem.c index bd82f6b..378d50a 100644 --- a/mm/nobootmem.c +++ b/mm/nobootmem.c @@ -126,6 +126,7 @@ unsigned long __init free_low_memory_core_early(int nodeid) phys_addr_t start, end, size; u64 i; + reset_lowmem_zone_present_pages(); for_each_free_mem_range(i, MAX_NUMNODES, &start, &end, NULL) count += __free_memory_core(start, end); diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 07425a7..76d37f0 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -735,6 +735,7 @@ void __meminit __free_pages_bootmem(struct page *page, unsigned int order) { unsigned int nr_pages = 1 << order; unsigned int loop; + struct zone *zone; prefetchw(page); for (loop = 0; loop < nr_pages; loop++) { @@ -748,6 +749,9 @@ void __meminit __free_pages_bootmem(struct page *page, unsigned int order) set_page_refcounted(page); __free_pages(page, order); + zone = page_zone(page); + WARN_ON(!(is_normal(zone) || is_dma(zone) || is_dma32(zone))); + zone->present_pages += nr_pages; } #ifdef CONFIG_CMA @@ -4547,18 +4551,20 @@ static void __paginginit free_area_init_core(struct pglist_data *pgdat, * is used by this zone for memmap. This affects the watermark * and per-cpu initialisations */ - memmap_pages = - PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; - if (realsize >= memmap_pages) { - realsize -= memmap_pages; - if (memmap_pages) - printk(KERN_DEBUG - " %s zone: %lu pages used for memmap\n", - zone_names[j], memmap_pages); - } else - printk(KERN_WARNING - " %s zone: %lu pages exceeds realsize %lu\n", - zone_names[j], memmap_pages, realsize); + if (j < ZONE_HIGHMEM) { + memmap_pages = + PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; + if (realsize >= memmap_pages) { + realsize -= memmap_pages; + if (memmap_pages) + printk(KERN_DEBUG + " %s zone: %lu pages used for memmap\n", + zone_names[j], memmap_pages); + } else + printk(KERN_WARNING + " %s zone: %lu pages exceeds realsize %lu\n", + zone_names[j], memmap_pages, realsize); + } /* Account for reserved pages */ if (j == 0 && realsize > dma_reserve) { @@ -6143,3 +6149,22 @@ void dump_page(struct page *page) dump_page_flags(page->flags); mem_cgroup_print_bad_page(page); } + +/* reset zone->present_pages to 0 for zones below ZONE_HIGHMEM */ +void reset_lowmem_zone_present_pages_pernode(pg_data_t *pgdat) +{ + int i; + struct zone *z; + for (i = 0; i < ZONE_HIGHMEM; i++) { + z = pgdat->node_zones + i; + z->present_pages = 0; + } +} + +void reset_lowmem_zone_present_pages(void) +{ + int nid; + + for_each_node_state(nid, N_HIGH_MEMORY) + reset_lowmem_zone_present_pages_pernode(NODE_DATA(nid)); +} -- 1.7.9.5 -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>