lkml.org 
[lkml]   [2020]   [Jun]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
SubjectRE:(2) [PATCH v2] page_alloc: consider highatomic reserve in wmartermark fast
From
Date
>On 6/14/20 1:17 AM, Baoquan He wrote:
>> On 06/13/20 at 10:08pm, Jaewon Kim wrote:
>> ...
>>> > > This is an example of ALLOC_HARDER allocation failure.
>>> > >
>>> > > <4>[ 6207.637280] [3: Binder:9343_3:22875] Binder:9343_3: page allocation failure: order:0, mode:0x480020(GFP_ATOMIC), nodemask=(null)
>>> > > <4>[ 6207.637311] [3: Binder:9343_3:22875] Call trace:
>>> > > <4>[ 6207.637346] [3: Binder:9343_3:22875] [<ffffff8008f40f8c>] dump_stack+0xb8/0xf0
>>> > > <4>[ 6207.637356] [3: Binder:9343_3:22875] [<ffffff8008223320>] warn_alloc+0xd8/0x12c
>>> > > <4>[ 6207.637365] [3: Binder:9343_3:22875] [<ffffff80082245e4>] __alloc_pages_nodemask+0x120c/0x1250
>>> > > <4>[ 6207.637374] [3: Binder:9343_3:22875] [<ffffff800827f6e8>] new_slab+0x128/0x604
>>> > > <4>[ 6207.637381] [3: Binder:9343_3:22875] [<ffffff800827b0cc>] ___slab_alloc+0x508/0x670
>>> > > <4>[ 6207.637387] [3: Binder:9343_3:22875] [<ffffff800827ba00>] __kmalloc+0x2f8/0x310
>>> > > <4>[ 6207.637396] [3: Binder:9343_3:22875] [<ffffff80084ac3e0>] context_struct_to_string+0x104/0x1cc
>>> > > <4>[ 6207.637404] [3: Binder:9343_3:22875] [<ffffff80084ad8fc>] security_sid_to_context_core+0x74/0x144
>>> > > <4>[ 6207.637412] [3: Binder:9343_3:22875] [<ffffff80084ad880>] security_sid_to_context+0x10/0x18
>>> > > <4>[ 6207.637421] [3: Binder:9343_3:22875] [<ffffff800849bd80>] selinux_secid_to_secctx+0x20/0x28
>>> > > <4>[ 6207.637430] [3: Binder:9343_3:22875] [<ffffff800849109c>] security_secid_to_secctx+0x3c/0x70
>>> > > <4>[ 6207.637442] [3: Binder:9343_3:22875] [<ffffff8008bfe118>] binder_transaction+0xe68/0x454c
>>> > > <4>[ 6207.637569] [3: Binder:9343_3:22875] Mem-Info:
>>> > > <4>[ 6207.637595] [3: Binder:9343_3:22875] active_anon:102061 inactive_anon:81551 isolated_anon:0
>>> > > <4>[ 6207.637595] [3: Binder:9343_3:22875] active_file:59102 inactive_file:68924 isolated_file:64
>>> > > <4>[ 6207.637595] [3: Binder:9343_3:22875] unevictable:611 dirty:63 writeback:0 unstable:0
>>> > > <4>[ 6207.637595] [3: Binder:9343_3:22875] slab_reclaimable:13324 slab_unreclaimable:44354
>>> > > <4>[ 6207.637595] [3: Binder:9343_3:22875] mapped:83015 shmem:4858 pagetables:26316 bounce:0
>>> > > <4>[ 6207.637595] [3: Binder:9343_3:22875] free:2727 free_pcp:1035 free_cma:178
>>> > > <4>[ 6207.637616] [3: Binder:9343_3:22875] Node 0 active_anon:408244kB inactive_anon:326204kB active_file:236408kB inactive_file:275696kB unevictable:2444kB isolated(anon):0kB isolated(file):256kB mapped:332060kB dirty:252kB writeback:0kB shmem:19432kB writeback_tmp:0kB unstable:0kB all_unreclaimable? no
>>> > > <4>[ 6207.637627] [3: Binder:9343_3:22875] Normal free:10908kB min:6192kB low:44388kB high:47060kB active_anon:409160kB inactive_anon:325924kB active_file:235820kB inactive_file:276628kB unevictable:2444kB writepending:252kB present:3076096kB managed:2673676kB mlocked:2444kB kernel_stack:62512kB pagetables:105264kB bounce:0kB free_pcp:4140kB local_pcp:40kB free_cma:712kB
>>
>> Checked this mem info, wondering why there's no 'reserved_highatomic'
>> printing in all these examples.
>
>Yeah, it better be printed, especially after it's included in watermark
>calculation, so we're less confused by reports of allocation failure where
>watermarks are seemingly ok.
>


Hello Vlastimil and Baoquan

The log in previous mail was captured from kernel based on v4.14.
After adding the reserved_highatomic log, I finally got a new log below
Let me change description in next patch.

There seems be reserved_highatomic:32768KB and actually 14232kB free.

[ 4738.329298] kswapd0: page allocation failure: order:0, mode:0x140000a(GFP_NOIO|__GFP_HIGHMEM|__GFP_MOVABLE), nodemask=(null)
[ 4738.329325] kswapd0 cpuset=/ mems_allowed=0
[ 4738.329339] CPU: 4 PID: 1221 Comm: kswapd0 Not tainted 4.14.113-18770262-userdebug #1
[ 4738.329350] Call trace:
[ 4738.329366] [<0000000000000000>] dump_backtrace+0x0/0x248
[ 4738.329377] [<0000000000000000>] show_stack+0x18/0x20
[ 4738.329390] [<0000000000000000>] __dump_stack+0x20/0x28
[ 4738.329398] [<0000000000000000>] dump_stack+0x68/0x90
[ 4738.329409] [<0000000000000000>] warn_alloc+0x104/0x198
[ 4738.329417] [<0000000000000000>] __alloc_pages_nodemask+0xdc0/0xdf0
[ 4738.329427] [<0000000000000000>] zs_malloc+0x148/0x3d0
[ 4738.329438] [<0000000000000000>] zram_bvec_rw+0x410/0x798
[ 4738.329446] [<0000000000000000>] zram_rw_page+0x88/0xdc
[ 4738.329455] [<0000000000000000>] bdev_write_page+0x70/0xbc
[ 4738.329463] [<0000000000000000>] __swap_writepage+0x58/0x37c
[ 4738.329469] [<0000000000000000>] swap_writepage+0x40/0x4c
[ 4738.329478] [<0000000000000000>] shrink_page_list+0xc30/0xf48
[ 4738.329486] [<0000000000000000>] shrink_inactive_list+0x2b0/0x61c
[ 4738.329494] [<0000000000000000>] shrink_node_memcg+0x23c/0x618
[ 4738.329501] [<0000000000000000>] shrink_node+0x1c8/0x304
[ 4738.329509] [<0000000000000000>] kswapd+0x680/0x7c4
[ 4738.329518] [<0000000000000000>] kthread+0x110/0x120
[ 4738.329527] [<0000000000000000>] ret_from_fork+0x10/0x18
[ 4738.329538] Mem-Info:
[ 4738.329574] active_anon:111826 inactive_anon:65557 isolated_anon:0\x0a active_file:44260 inactive_file:83422 isolated_file:0\x0a unevictable:4158 dirty:117 writeback:0 unstable:0\x0a slab_reclaimable:13943 slab_unreclaimable:43315\x0a mapped:102511 shmem:3299 pagetables:19566 bounce:0\x0a free:3510 free_pcp:553 free_cma:0
[ 4738.329593] Node 0 active_anon:447304kB inactive_anon:262228kB active_file:177040kB inactive_file:333688kB unevictable:16632kB isolated(anon):0kB isolated(file):0kB mapped:410044kB d irty:468kB writeback:0kB shmem:13196kB writeback_tmp:0kB unstable:0kB all_unreclaimable? no
[ 4738.329603] Normal free:14040kB min:7440kB low:94500kB high:98136kB reserved_highatomic:32768KB active_anon:447336kB inactive_anon:261668kB active_file:177572kB inactive_file:333768k B unevictable:16632kB writepending:480kB present:4081664kB managed:3637088kB mlocked:16632kB kernel_stack:47072kB pagetables:78264kB bounce:0kB free_pcp:2280kB local_pcp:720kB free_cma:0kB [ 4738.329607] lowmem_reserve[]: 0 0
[ 4738.329615] Normal: 860*4kB (H) 453*8kB (H) 180*16kB (H) 26*32kB (H) 34*64kB (H) 6*128kB (H) 2*256kB (H) 0*512kB 0*1024kB 0*2048kB 0*4096kB = 14232kB


>...
>
>>> > > /*
>>> > > * Fast check for order-0 only. If this fails then the reserves
>>> > > @@ -3598,9 +3604,12 @@ static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
>>> > > * the caller is !atomic then it'll uselessly search the free
>>> > > * list. That corner case is then slower but it is harmless.
>>> >
>>> > Do we need remove or adjust the code comment at this place? So Mel have
>>> > foreseen the corner case, just reclaiming to unreserve the highatomic
>>> > might be ignored.
>>> >
>>>
>>> Hello thank you for your comment.
>>>
>>> My previous mail to Vlastimil Babka seems to have html.
>>> Let me put again here because I also think the comment should be changed.
>>> I'd like to hear opinions from the open source community.
>>
>> Yeah, your replying mail to Vlastimil looks a little messy on format, I
>> didn't scroll down to check.
>>
>>>
>>> Additionally actually I think we need accurate counting of highatomic
>>> free after this patch.
>>>
>>> ----------------------------------------------------------------------------------------
>>>
>>> As Mel also agreed with me in v1 mail thread, this highatomic reserved should
>>> be considered even in watermark fast.
>>>
>>> The comment, I think, may need to be changed. Prior to this patch, non
>>> highatomic
>>> allocation may do useless search, but it also can take ALL non highatomic free.
>>>
>>> With this patch, non highatomic allocation will NOT do useless search. Rather,
>>> it may be required direct reclamation even when there are some non
>>> high atomic free.
>>>
>>> i.e)
>>> In following situation, watermark check fails (9MB - 8MB < 4MB) though there are
>>> enough free (9MB - 4MB > 4MB). If this is really matter, we need to
>>> count highatomic
>>> free accurately.
>>
>> Seems to make sense. We only use zone->nr_reserved_highatomic to account
>> the reserved highatomic, don't track how much have been used for
>> highatomic allocation. But not sure if this will happen often, or just a
>> very rare case, whether taking that into account will impact anything.
>
>Unfortunately there's a problem when trying to account free pages of a migrate
>type exactly, as e.g. during reserve_highatomic_pageblock(), some pages might be
>in pcplist of other cpu with other migratetype, and once they are freed, the
>buddy merging will merge the different migratetypes and distort the accounting.
>Fixing this for all migratetypes would have performance overhead, so we only do
>that for MIGRATE_ISOLATE which is not that frequent (and it took a while to
>eliminate all corner cases), and CMA which doesn't change pageblocks dynamically.

AFAIK we do not account free cma in pcp either. But yes accurate check could be
overhead. For example, __mod_zone_freepage_state should account highatomic free
as cma free. And we may see some incorrect accounting issue.

>
>So either we live with the possible overreclaim due to inaccurate counting per
>your example above, or we instead let order-0 atomic allocations use highatomic
>reserves.
>

Additionally regarding existing Mel's comment, let me remove some of it if you
don't mind.

/*
* Fast check for order-0 only. If this fails then the reserves
- * need to be calculated. There is a corner case where the check
- * passes but only the high-order atomic reserve are free. If
- * the caller is !atomic then it'll uselessly search the free
- * list. That corner case is then slower but it is harmless.
+ * need to be calculated.
*/
I will prepare v3 patch.
Thank you again.

\
 
 \ /
  Last update: 2020-06-16 09:31    [W:0.074 / U:18.608 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site