On 4/18/19 3:54 PM, Mel Gorman wrote:
> On Wed, Apr 17, 2019 at 10:54:38AM +0200, Vlastimil Babka wrote:
>> On 4/17/19 10:35 AM, Li Wang wrote:
>>> Hi there,
>>>
>>> I catched this warning on v5.1-rc5(s390x). It was trggiered in fork & malloc & memset stress test, but the reproduced rate is very low. I'm working on find a stable reproducer for it.
>>>
>>> Anyone can have a look first?
>>>
>>> [ 1422.124060] WARNING: CPU: 0 PID: 9783 at mm/page_alloc.c:3777 __alloc_pages_irect_compact+0x182/0x190
>>
>> This means compaction was either skipped or deferred, yet it captured a
>> page. We have some registers with value 1 and 2, which is
>> COMPACT_SKIPPED and COMPACT_DEFERRED, so it could be one of those.
>> Probably COMPACT_SKIPPED. I think a race is possible:
>>
>> - compact_zone_order() sets up current->capture_control
>> - compact_zone() calls compaction_suitable() which returns
>> COMPACT_SKIPPED, so it also returns
>> - interrupt comes and its processing happens to free a page that forms
>> high-order page, since 'current' isn't changed during interrupt (IIRC?)
>> the capture_control is still active and the page is captured
>> - compact_zone_order() does *capture = capc.page
>>
>> What do you think, Mel, does it look plausible?
>
> It's plausible, just extremely unlikely. I think the most likely result
> was that a page filled the per-cpu lists and a bunch of pages got freed
> in a batch from interrupt context.
Sure, good point. Per-cpu lists make the scenario even more rare, but
once it's full, there's a higher change the batch free from the
interrupt will result in high-order page being formed.
>> Not sure whether we want
>> to try avoiding this scenario, or just remove the warning and be
>> grateful for the successful capture :)
>>
>
> Avoiding the scenario is pointless because it's not wrong. The check was
> initially meant to catch serious programming errors such as using a
> stale page pointer so I think the right patch is below. Li Wang, how
> reproducible is this and would you be willing to test it?
>
> ---8<---
> mm, page_alloc: Always use a captured page regardless of compaction result
>
> During the development of commit 5e1f0f098b46 ("mm, compaction: capture
> a page under direct compaction"), a paranoid check was added to ensure
> that if a captured page was available after compaction that it was
> consistent with the final state of compaction. The intent was to catch
> serious programming bugs such as using a stale page pointer and causing
> corruption problems.
>
> However, it is possible to get a captured page even if compaction was
> unsuccessful if an interrupt triggered and happened to free pages in
> interrupt context that got merged into a suitable high-order page. It's
> highly unlikely but Li Wang did report the following warning on s390
>
> [ 1422.124060] WARNING: CPU: 0 PID: 9783 at mm/page_alloc.c:3777 __alloc_pages_irect_compact+0x182/0x190
> [ 1422.124065] Modules linked in: rpcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver
> nfs lockd grace fscache sunrpc pkey ghash_s390 prng xts aes_s390 des_s390
> des_generic sha512_s390 zcrypt_cex4 zcrypt vmur binfmt_misc ip_tables xfs
> libcrc32c dasd_fba_mod qeth_l2 dasd_eckd_mod dasd_mod qeth qdio lcs ctcm
> ccwgroup fsm dm_mirror dm_region_hash dm_log dm_mod
> [ 1422.124086] CPU: 0 PID: 9783 Comm: copy.sh Kdump: loaded Not tainted 5.1.0-rc 5 #1
>
> This patch simply removes the check entirely instead of trying to be
> clever about pages freed from interrupt context. If a serious programming
> error was introduced, it is highly likely to be caught by prep_new_page()
> instead.
>
> Fixes: 5e1f0f098b46 ("mm, compaction: capture a page under direct compaction")
> Reported-by: Li Wang <liwang@xxxxxxxxxx>
> Signed-off-by: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
Yup, no need for a Cc: stable on a very rare WARN_ON_ONCE. So the AI
will pick it anyway...
Acked-by: Vlastimil Babka <vbabka@xxxxxxx>
> ---
> mm/page_alloc.c | 5 -----
> 1 file changed, 5 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index d96ca5bc555b..cfaba3889fa2 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3773,11 +3773,6 @@ __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
> memalloc_noreclaim_restore(noreclaim_flag);
> psi_memstall_leave(&pflags);
>
> - if (*compact_result <= COMPACT_INACTIVE) {
> - WARN_ON_ONCE(page);
> - return NULL;
> - }
> -
> /*
> * At least in one zone compaction wasn't deferred or skipped, so let's
> * count a compaction stall
>
