Re: known oom issues on numa in -mm tree?

----- Original Message -----

BTW, the latest linux-next also had the similar issue.

- kswapd was running for a long time.

runnable tasks:
task PID tree-key switches prio exec-runtime sum-exec sum-sleep
----------------------------------------------------------------------------------------------------------
R kswapd1 82 564957.671501 32922 120 564986.657154 1188257.259742
3555627.915255

- it was looping for drain_local_pages.

oom02 R running task 0 2023 1969 0x00000088
0000000000000282 ffff88041d219df0 ffff88041fbf8ef0 ffffffff81100800
ffff880418ab5b18 0000000000000282 ffffffff8100c9ee ffff880418ab5ba8
0000000087654321 0000000000000000 ffff880000000000 0000000000000001
Call Trace:
[<ffffffff81100800>] ? drain_local_pages+0x0/0x20
[<ffffffff8100c9ee>] ? apic_timer_interrupt+0xe/0x20
[<ffffffff81097ea6>] ? smp_call_function_many+0x1b6/0x210
[<ffffffff81097e82>] ? smp_call_function_many+0x192/0x210
[<ffffffff81100800>] ? drain_local_pages+0x0/0x20
[<ffffffff81097f22>] ? smp_call_function+0x22/0x30
[<ffffffff81068184>] ? on_each_cpu+0x24/0x50
[<ffffffff810fe68c>] ? drain_all_pages+0x1c/0x20
[<ffffffff81100d04>] ? __alloc_pages_nodemask+0x4e4/0x840
[<ffffffff81138e09>] ? alloc_page_vma+0x89/0x140
[<ffffffff8111c481>] ? handle_mm_fault+0x871/0xd80
[<ffffffff814a4ecd>] ? schedule+0x3fd/0x980
[<ffffffff8100c9ee>] ? apic_timer_interrupt+0xe/0x20
[<ffffffff8100c9ee>] ? apic_timer_interrupt+0xe/0x20
[<ffffffff814aadd3>] ? do_page_fault+0x143/0x4b0
[<ffffffff8100a7b4>] ? __switch_to+0x194/0x320
[<ffffffff814a4ecd>] ? schedule+0x3fd/0x980
[<ffffffff814a7ad5>] ? page_fault+0x25/0x30

- although the local pages were low.

Node 1 Normal free:8052kB min:8136kB low:10168kB high:12204kB
active_anon:8026400kB inactive_anon:0kB active_file:0kB
inactive_file:20kB unevictable:0kB isolated(anon):0kB
isolated(file):0kB present:8273920kB mlocked:0kB dirty:0kB
writeback:0kB mapped:0kB shmem:116kB slab_reclaimable:23728kB
slab_unreclaimable:175932kB kernel_stack:176kB pagetables:17136kB
unstable:0kB bounce:0kB writeback_tmp:0kB pages_scanned:10
all_unreclaimable? no
lowmem_reserve[]: 0 0 0 0
Node 0 DMA: 0*4kB 0*8kB 1*16kB 0*32kB 2*64kB 1*128kB 1*256kB 0*512kB
1*1024kB 1*2048kB 3*4096kB = 15888kB
Node 0 DMA32: 10*4kB 10*8kB 7*16kB 8*32kB 4*64kB 8*128kB 10*256kB
2*512kB 4*1024kB 2*2048kB 534*4096kB = 2200808kB
Node 0 Normal: 519*4kB 824*8kB 440*16kB 202*32kB 87*64kB 29*128kB
15*256kB 11*512kB 3*1024kB 8*2048kB 1065*4096kB = 4422620kB
Node 1 Normal: 579*4kB 205*8kB 0*16kB 0*32kB 0*64kB 0*128kB 0*256kB
0*512kB 0*1024kB 0*2048kB 1*4096kB = 8052kB

It could also be reproduced by using mempolicy alone. The test just
hung at memset() while allocating memory.

set_mempolicy(MPOL_BIND, &nmask, MAXNODES);
while(1) {
s = mmap(NULL, length, PROT_READ|PROT_WRITE,
MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
memset(s, '\a', length);
}

Is it possible that,

drain_all_pages()
    on_each_cpu()
        drain_local_pages()

Thought that there were still free pages in other node, so it was still
trying to call direct reclaim path?

CAI Qian

--
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/ .
Fight unfair telecom policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>