Hi, Thanks for your replying. According to the OOM message, OOM killer is invoked by the function seq_read, I found two patches in the latest kernel which can be avoid or fixed this problem. https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/fs/seq_file.c?id=058504edd02667eef8fac9be27ab3ea74332e9b4 https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/commit/fs/seq_file.c?id=5cec38ac866bfb8775638e71a86e4d8cac30caae As the patches said, it changed the seq_file code fallback to vmalloc allocations if kmalloc failed, instead of OOM kill processes. -----邮件原件----- 发件人: David Rientjes [mailto:rientjes@xxxxxxxxxx] 发送时间: 2015年7月15日 8:10 收件人: Xuzhichuang 抄送: linux-mm@xxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Songjiangtao (mygirlsjt); Zhangwei (FF); Qiuxishi 主题: Re: [BUG REPORT] OOM Killer is invoked while the system still has much memory On Tue, 14 Jul 2015, Xuzhichuang wrote: > Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138764] iostat invoked > oom-killer: gfp_mask=0xd0, order=2, oom_adj=0, oom_score_adj=0 Jul 10 > 12:33:03 BMS_CNA04 kernel: [18136514.138769] iostat cpuset=/ > mems_allowed=0 Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138773] Pid: 18117, comm: iostat Tainted: P W NX 3.0.58-0.6.6-xen #1 Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138775] Call Trace: > Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138800] > [<ffffffff800088be>] dump_trace+0x6e/0x1a0 Jul 10 12:33:03 BMS_CNA04 > kernel: [18136514.138810] [<ffffffff803f773d>] dump_stack+0x69/0x6f > Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138821] > [<ffffffff800dbced>] dump_header+0x9d/0x120 Jul 10 12:33:03 BMS_CNA04 > kernel: [18136514.138826] [<ffffffff800dc505>] > oom_kill_process+0x95/0x1a0 Jul 10 12:33:03 BMS_CNA04 kernel: > [18136514.138830] [<ffffffff800dc746>] out_of_memory+0x136/0x220 Jul > 10 12:33:03 BMS_CNA04 kernel: [18136514.138834] [<ffffffff800e0fda>] > __alloc_pages_slowpath+0x7ba/0x810 > Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138838] > [<ffffffff800e1219>] __alloc_pages_nodemask+0x1e9/0x200 > Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138845] > [<ffffffff8011ae38>] cache_grow+0x348/0x450 Jul 10 12:33:03 BMS_CNA04 > kernel: [18136514.138850] [<ffffffff8011b243>] > cache_alloc_refill+0x303/0x4d0 Jul 10 12:33:03 BMS_CNA04 kernel: > [18136514.138854] [<ffffffff8011ba70>] __kmalloc+0x1b0/0x290 Jul 10 > 12:33:03 BMS_CNA04 kernel: [18136514.138862] [<ffffffff8014c1da>] > seq_read+0x13a/0x3b0 Jul 10 12:33:03 BMS_CNA04 kernel: > [18136514.138869] [<ffffffff8018a762>] proc_reg_read+0x92/0xe0 Jul 10 > 12:33:03 BMS_CNA04 kernel: [18136514.138877] [<ffffffff80129877>] > vfs_read+0xc7/0x130 Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138881] [<ffffffff801299e3>] sys_read+0x53/0xa0 Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138887] [<ffffffff80402d73>] system_call_fastpath+0x16/0x1b Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138922] [<00007f935f57f4c0>] 0x7f935f57f4bf Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138923] Mem-Info: > Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138925] DMA per-cpu: > Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138927] CPU 0: hi: > 0, btch: 1 usd: 0 Jul 10 12:33:03 BMS_CNA04 kernel: > [18136514.138929] CPU 1: hi: 0, btch: 1 usd: 0 Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138930] DMA32 per-cpu: > Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138932] CPU 0: hi: > 155, btch: 38 usd: 11 Jul 10 12:33:03 BMS_CNA04 kernel: > [18136514.138933] CPU 1: hi: 155, btch: 38 usd: 0 Jul 10 > 12:33:03 BMS_CNA04 kernel: [18136514.138936] active_anon:227111 > inactive_anon:10382 isolated_anon:0 Jul 10 12:33:03 BMS_CNA04 kernel: > [18136514.138937] active_file:203 inactive_file:189 isolated_file:47 > Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138938] unevictable:95395 > dirty:0 writeback:0 unstable:0 Jul 10 12:33:03 BMS_CNA04 kernel: > [18136514.138939] free:247834 slab_reclaimable:18187 > slab_unreclaimable:53853 Jul 10 12:33:03 BMS_CNA04 kernel: > [18136514.138940] mapped:11485 shmem:11167 pagetables:0 bounce:0 Jul > 10 12:33:03 BMS_CNA04 kernel: [18136514.138945] DMA free:984kB > min:36kB low:44kB high:52kB active_anon:0kB inactive_anon:0kB > active_file:0kB inactive_file:0kB unevictable:0kB isolated(anon):0kB > isolated(file):0kB present:16160kB mlocked:0kB dirty:0kB writeback:0kB > mapped:0kB shmem:0kB slab_reclaimable:0kB slab_unreclaimable:0kB > kernel_stack:0kB pagetables:0kB unstable:0kB bounce:0kB > writeback_tmp:0kB pages_scanned:0 all_unreclaimable? yes Jul 10 > 12:33:03 BMS_CNA04 kernel: [18136514.138949] lowmem_reserve[]: 0 3014 > 3014 3014 Jul 10 12:33:03 BMS_CNA04 kernel: [18136514.138955] DMA32 > free:990352kB min:7004kB low:8752kB high:10504kB active_anon:908444kB > inactive_anon:41528kB active_file:812kB inactive_file:756kB > unevictable:381580kB isolated(anon):0kB isolated(file):188kB > present:3025264kB mlocked:381580kB dirty:0kB writeback:0kB > mapped:45940kB shmem:44668kB slab_reclaimable:72748kB > slab_unreclaimable:215412kB kernel_stack:12456kB pagetables:0kB > unstable:0kB bounce:0kB writeback_tmp:0kB pages_scanned:192 > all_unreclaimable? no Jul 10 12:33:03 BMS_CNA04 kernel: > [18136514.138960] lowmem_reserve[]: 0 0 0 0 Jul 10 12:33:03 BMS_CNA04 > kernel: [18136514.138962] DMA: 2*4kB 4*8kB 3*16kB 4*32kB 2*64kB > 1*128kB 2*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 984kB Jul 10 > 12:33:03 BMS_CNA04 kernel: [18136514.138968] DMA32: 188513*4kB > 29459*8kB 2*16kB 2*32kB 1*64kB 0*128kB 0*256kB 1*512kB 0*1024kB > 0*2048kB 0*4096kB = 990396kB The problem is most of your memory for ZONE_DMA32 is available only in sizes of order-0 and order-1 and the slab allocator is trying to allocate order-2 memory with no possibility of fallback to a smaller order. You're running on a 3.0.58 kernel, but the watermark calculation should be the same in recent kernels. If you follow the logic of __zone_watermark_ok(), which uses the same watermarks as printed above, the min watermark for this zone is 1751 pages and the total zone free pages is 247588. Discounting order-0 memory, there are only 59075 pages free with a min watermark of 875 pages. Discounting order-1 memory, there are 157 pages free with a min watermark of 437 pages. This is where your allocation fails. Even though the zone has 672KB of memory available, the per-order watermark fails. The only option you have to avoid this other than changing your workload is to alter lowmem_reserve_ratio, see Documentation/sysctl/vm.txt. You have 916KB of memory in ZONE_DMA that could be used for this allocation if it wasn't reserved for DMA allocations. ��.n������g��� �a��� �&ޖ)���)��h���&������梷���� �Ǟ�m������)������^�����������v���O��zf������