On 7/13/19 8:53 PM, Hillf Danton wrote:
On Wed, 10 Jul 2019 14:43:28 -0700 (PDT) Qian Cai wrote:
Running LTP oom01 test case with swap triggers a crash below. Revert the series
"Make deferred split shrinker memcg aware" [1] seems fix the issue.
aefde94195ca mm: thp: make deferred split shrinker memcg aware
cf402211cacc mm-shrinker-make-shrinker-not-depend-on-memcg-kmem-fix-2-fix
ca37e9e5f18d mm-shrinker-make-shrinker-not-depend-on-memcg-kmem-fix-2
5f419d89cab4 mm-shrinker-make-shrinker-not-depend-on-memcg-kmem-fix
c9d49e69e887 mm: shrinker: make shrinker not depend on memcg kmem
1c0af4b86bcf mm: move mem_cgroup_uncharge out of __page_cache_release()
4e050f2df876 mm: thp: extract split_queue_* into a struct
[1] https://lore.kernel.org/linux-mm/1561507361-59349-1-git-send-email-yang.shi@xxxxxxxxxxxxxxxxx/
[ 1145.730682][ T5764] list_del corruption, ffffea00251c8098->next is LIST_POISON1 (dead000000000100)
[ 1145.739763][ T5764] ------------[ cut here ]------------
[ 1145.745126][ T5764] kernel BUG at lib/list_debug.c:47!
[ 1145.750320][ T5764] invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN NOPTI
[ 1145.757513][ T5764] CPU: 1 PID: 5764 Comm: oom01 Tainted: G W 5.2.0-next-20190710+ #7
[ 1145.766709][ T5764] Hardware name: HPE ProLiant DL385 Gen10/ProLiant DL385 Gen10, BIOS A40 01/25/2019
[ 1145.776000][ T5764] RIP: 0010:__list_del_entry_valid.cold.0+0x12/0x4a
[ 1145.782491][ T5764] Code: c7 40 5a 33 af e8 ac fe bc ff 0f 0b 48 c7 c7 80 9e
a1 af e8 f6 4c 01 00 4c 89 ea 48 89 de 48 c7 c7 20 59 33 af e8 8c fe bc ff <0f>
0b 48 c7 c7 40 9f a1 af e8 d6 4c 01 00 4c 89 e2 48 89 de 48 c7
[ 1145.802078][ T5764] RSP: 0018:ffff888514d773c0 EFLAGS: 00010082
[ 1145.808042][ T5764] RAX: 000000000000004e RBX: ffffea00251c8098 RCX: ffffffffae95d318
[ 1145.815923][ T5764] RDX: 0000000000000000 RSI: 0000000000000008 RDI: ffff8888440bd380
[ 1145.823806][ T5764] RBP: ffff888514d773d8 R08: ffffed1108817a71 R09: ffffed1108817a70
[ 1145.831689][ T5764] R10: ffffed1108817a70 R11: ffff8888440bd387 R12: dead000000000122
[ 1145.839571][ T5764] R13: dead000000000100 R14: ffffea00251c8034 R15: dead000000000100
[ 1145.847455][ T5764] FS: 00007f765ad4d700(0000) GS:ffff888844080000(0000) knlGS:0000000000000000
[ 1145.856299][ T5764] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 1145.862784][ T5764] CR2: 00007f8cebec7000 CR3: 0000000459338000 CR4: 00000000001406a0
[ 1145.870664][ T5764] Call Trace:
[ 1145.873835][ T5764] deferred_split_scan+0x337/0x740
[ 1145.878835][ T5764] ? split_huge_page_to_list+0xe30/0xe30
[ 1145.884364][ T5764] ? __radix_tree_lookup+0x12d/0x1e0
[ 1145.889539][ T5764] ? node_tag_get.part.0.constprop.6+0x40/0x40
[ 1145.895592][ T5764] do_shrink_slab+0x244/0x5a0
[ 1145.900159][ T5764] shrink_slab+0x253/0x440
[ 1145.904462][ T5764] ? unregister_shrinker+0x110/0x110
[ 1145.909641][ T5764] ? kasan_check_read+0x11/0x20
[ 1145.914383][ T5764] ? mem_cgroup_protected+0x20f/0x260
[ 1145.919645][ T5764] shrink_node+0x31e/0xa30
[ 1145.923949][ T5764] ? shrink_node_memcg+0x1560/0x1560
[ 1145.929126][ T5764] ? ktime_get+0x93/0x110
[ 1145.933340][ T5764] do_try_to_free_pages+0x22f/0x820
[ 1145.938429][ T5764] ? shrink_node+0xa30/0xa30
[ 1145.942906][ T5764] ? kasan_check_read+0x11/0x20
[ 1145.947647][ T5764] ? check_chain_key+0x1df/0x2e0
[ 1145.952474][ T5764] try_to_free_pages+0x242/0x4d0
[ 1145.957299][ T5764] ? do_try_to_free_pages+0x820/0x820
[ 1145.962566][ T5764] __alloc_pages_nodemask+0x9ce/0x1bc0
[ 1145.967917][ T5764] ? kasan_check_read+0x11/0x20
[ 1145.972657][ T5764] ? gfp_pfmemalloc_allowed+0xc0/0xc0
[ 1145.977920][ T5764] ? kasan_check_read+0x11/0x20
[ 1145.982659][ T5764] ? check_chain_key+0x1df/0x2e0
[ 1145.987487][ T5764] ? do_anonymous_page+0x343/0xe30
[ 1145.992489][ T5764] ? lock_downgrade+0x390/0x390
[ 1145.997230][ T5764] ? __count_memcg_events+0x8b/0x1c0
[ 1146.002404][ T5764] ? kasan_check_read+0x11/0x20
[ 1146.007145][ T5764] ? __lru_cache_add+0x122/0x160
[ 1146.011974][ T5764] alloc_pages_vma+0x89/0x2c0
[ 1146.016538][ T5764] do_anonymous_page+0x3e1/0xe30
[ 1146.021367][ T5764] ? __update_load_avg_cfs_rq+0x2c/0x490
[ 1146.026893][ T5764] ? finish_fault+0x120/0x120
[ 1146.031461][ T5764] ? call_function_interrupt+0xa/0x20
[ 1146.036724][ T5764] handle_pte_fault+0x457/0x12c0
[ 1146.041552][ T5764] __handle_mm_fault+0x79a/0xa50
[ 1146.046378][ T5764] ? vmf_insert_mixed_mkwrite+0x20/0x20
[ 1146.051817][ T5764] ? kasan_check_read+0x11/0x20
[ 1146.056557][ T5764] ? __count_memcg_events+0x8b/0x1c0
[ 1146.061732][ T5764] handle_mm_fault+0x17f/0x370
[ 1146.066386][ T5764] __do_page_fault+0x25b/0x5d0
[ 1146.071037][ T5764] do_page_fault+0x4c/0x2cf
[ 1146.075426][ T5764] ? page_fault+0x5/0x20
[ 1146.079553][ T5764] page_fault+0x1b/0x20
[ 1146.083594][ T5764] RIP: 0033:0x410be0
[ 1146.087373][ T5764] Code: 89 de e8 e3 23 ff ff 48 83 f8 ff 0f 84 86 00 00 00
48 89 c5 41 83 fc 02 74 28 41 83 fc 03 74 62 e8 95 29 ff ff 31 d2 48 98 90 <c6>
44 15 00 07 48 01 c2 48 39 d3 7f f3 31 c0 5b 5d 41 5c c3 0f 1f
[ 1146.106959][ T5764] RSP: 002b:00007f765ad4cec0 EFLAGS: 00010206
[ 1146.112921][ T5764] RAX: 0000000000001000 RBX: 00000000c0000000 RCX: 00007f98f2674497
[ 1146.120804][ T5764] RDX: 0000000001d95000 RSI: 00000000c0000000 RDI: 0000000000000000
[ 1146.128687][ T5764] RBP: 00007f74d9d4c000 R08: 00000000ffffffff R09: 0000000000000000
[ 1146.136569][ T5764] R10: 0000000000000022 R11: 000000000
[ 1147.588181][ T5764] Shutting down cpus with NMI
[ 1147.592756][ T5764] Kernel Offset: 0x2d400000 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0xffffffffbfffffff)
[ 1147.604414][ T5764] ---[ end Kernel panic - not syncing: Fatal exception ]---
Ignore the noise if there is no chance you think to corrupt the local list walk
in some way like:
CPU0 CPU1
---- ----
take no lock spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
list_for_each_safe(pos, next,
&list)
list_del(page_deferred_list(page));
page = list_entry((void *)pos,
struct page, mapping);
spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags);
IMHO, I didn't see the race could happen really.
list_del() is called at 3 places:
1. Parallel free_transhuge_page(): The refcount bump should prevent from
the race.
2. Parallel reclaimer: split_queue_lock should prevent this, so the
other reclaimer should not see the same page.
3. Parallel split_huge_page(): I'm not sure about this one. But, page
lock should be acquired before calling split_huge_page() in other call
paths too.
I'm not sure if I miss anything, please feel free to correct me.
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -2765,7 +2765,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
if (!mapcount && page_ref_freeze(head, 1 + extra_pins)) {
if (!list_empty(page_deferred_list(head))) {
ds_queue->split_queue_len--;
- list_del(page_deferred_list(head));
+ list_del_init(page_deferred_list(head));
}
if (mapping)
__dec_node_page_state(page, NR_SHMEM_THPS);
@@ -2814,7 +2814,7 @@ void free_transhuge_page(struct page *page)
spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
if (!list_empty(page_deferred_list(page))) {
ds_queue->split_queue_len--;
- list_del(page_deferred_list(page));
+ list_del_init(page_deferred_list(page));
}
spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags);
free_compound_page(page);
--
I proposed the similar thing.
The major important is listed above; the minor trivial part below.
Both are only for thought collectings.
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -2869,9 +2869,8 @@ static unsigned long deferred_split_scan(struct shrinker *shrink,
struct pglist_data *pgdata = NODE_DATA(sc->nid);
struct deferred_split *ds_queue;
unsigned long flags;
- LIST_HEAD(list), *pos, *next;
struct page *page;
- int split = 0;
+ unsigned long nr_split = 0;
#ifdef CONFIG_MEMCG
if (sc->memcg)
@@ -2884,44 +2883,44 @@ static unsigned long deferred_split_scan(struct shrinker *shrink,
spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
/* Take pin on all head pages to avoid freeing them under us */
- list_for_each_safe(pos, next, &ds_queue->split_queue) {
- page = list_entry((void *)pos, struct page, mapping);
+ while (sc->nr_to_scan && !list_empty(&ds_queue->split_queue)) {
+ bool locked, pinned;
+
+ page = list_first_entry(&ds_queue->split_queue, struct page,
+ mapping);
page = compound_head(page);
+
if (get_page_unless_zero(page)) {
- list_move(page_deferred_list(page), &list);
+ pinned = true;
+ locked = trylock_page(page);
} else {
/* We lost race with put_compound_page() */
- list_del_init(page_deferred_list(page));
- ds_queue->split_queue_len--;
+ pinned = false;
+ locked = false;
+ }
+ list_del_init(page_deferred_list(page));
+ ds_queue->split_queue_len--;
+ --sc->nr_to_scan;
+ if (!pinned)
+ continue;
+ spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags);
+ if (locked) {
+ if (!split_huge_page(page))
+ nr_split++;
+ unlock_page(page);
}
- if (!--sc->nr_to_scan)
- break;
- }
- spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags);
-
- list_for_each_safe(pos, next, &list) {
- page = list_entry((void *)pos, struct page, mapping);
- if (!trylock_page(page))
- goto next;
- /* split_huge_page() removes page from list on success */
- if (!split_huge_page(page))
- split++;
- unlock_page(page);
-next:
put_page(page);
+ spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
}
-
- spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
- list_splice_tail(&list, &ds_queue->split_queue);
spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags);
/*
* Stop shrinker if we didn't split any page, but the queue is empty.
* This can happen if pages were freed under us.
*/
- if (!split && list_empty(&ds_queue->split_queue))
+ if (!nr_split && list_empty(&ds_queue->split_queue))
return SHRINK_STOP;
- return split;
+ return nr_split;
}
static struct shrinker deferred_split_shrinker = {
--
