[PATCH] mm: fix execution order in alloc_contig_range_noprof

[PATCH] mm: fix execution order in alloc_contig_range_noprof

[Hua Su]

Fix the execution order issue in alloc_contig_range_noprof where
drain_all_pages was called after start_isolate_page_range, which
may be lead to race conditions.

Based on community patches commit ec6e8c7e0314 ("mm, page_alloc:
disable pcplists during memory offline") and commit d479960e44f27
("mm: disable LRU pagevec during the migration temporarily"), we
disable pcplists and LRU cache before page isolation to ensure no
pages are left in per-cpu lists during isolation.

Signed-off-by: Hua Su <huasu@xiaomi.com>
Signed-off-by: Hua Su <suhua.tanke@gmail.com>
---
 mm/page_alloc.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 6a47443c48ff..d08f929ca64c 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -6815,8 +6815,6 @@ static int __alloc_contig_migrate_range(struct compact_control *cc,
 		.reason = MR_CONTIG_RANGE,
 	};
 
-	lru_cache_disable();
-
 	while (pfn < end || !list_empty(&cc->migratepages)) {
 		if (fatal_signal_pending(current)) {
 			ret = -EINTR;
@@ -6850,7 +6848,6 @@ static int __alloc_contig_migrate_range(struct compact_control *cc,
 			break;
 	}
 
-	lru_cache_enable();
 	if (ret < 0) {
 		if (!(cc->gfp_mask & __GFP_NOWARN) && ret == -EBUSY)
 			alloc_contig_dump_pages(&cc->migratepages);
@@ -6973,6 +6970,9 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
 	if (__alloc_contig_verify_gfp_mask(gfp_mask, (gfp_t *)&cc.gfp_mask))
 		return -EINVAL;
 
+	zone_pcp_disable(cc.zone);
+	lru_cache_disable();
+
 	/*
 	 * What we do here is we mark all pageblocks in range as
 	 * MIGRATE_ISOLATE.  Because pageblock and max order pages may
@@ -6998,8 +6998,6 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
 	if (ret)
 		goto done;
 
-	drain_all_pages(cc.zone);
-
 	/*
 	 * In case of -EBUSY, we'd like to know which page causes problem.
 	 * So, just fall through. test_pages_isolated() has a tracepoint
@@ -7076,6 +7074,8 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
 		     start, end, outer_start, outer_end);
 	}
 done:
+	lru_cache_enable();
+	zone_pcp_enable(cc.zone);
 	undo_isolate_page_range(start, end);
 	return ret;
 }
-- 
2.34.1

Re: [PATCH] mm: fix execution order in alloc_contig_range_noprof

[Michal Hocko]

On Mon 05-01-26 21:32:50, Hua Su wrote:
> Fix the execution order issue in alloc_contig_range_noprof where
> drain_all_pages was called after start_isolate_page_range, which
> may be lead to race conditions.
> 
> Based on community patches commit ec6e8c7e0314 ("mm, page_alloc:
> disable pcplists during memory offline") and commit d479960e44f27
> ("mm: disable LRU pagevec during the migration temporarily"), we
> disable pcplists and LRU cache before page isolation to ensure no
> pages are left in per-cpu lists during isolation.

What exactly is the problem you are trying to fix here? Is this based on
code review or are you hitting any real problem. I find the changelog
rather hard to grasp.

> Signed-off-by: Hua Su <huasu@xiaomi.com>
> Signed-off-by: Hua Su <suhua.tanke@gmail.com>
> ---
>  mm/page_alloc.c | 10 +++++-----
>  1 file changed, 5 insertions(+), 5 deletions(-)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 6a47443c48ff..d08f929ca64c 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -6815,8 +6815,6 @@ static int __alloc_contig_migrate_range(struct compact_control *cc,
>  		.reason = MR_CONTIG_RANGE,
>  	};
>  
> -	lru_cache_disable();
> -
>  	while (pfn < end || !list_empty(&cc->migratepages)) {
>  		if (fatal_signal_pending(current)) {
>  			ret = -EINTR;
> @@ -6850,7 +6848,6 @@ static int __alloc_contig_migrate_range(struct compact_control *cc,
>  			break;
>  	}
>  
> -	lru_cache_enable();
>  	if (ret < 0) {
>  		if (!(cc->gfp_mask & __GFP_NOWARN) && ret == -EBUSY)
>  			alloc_contig_dump_pages(&cc->migratepages);
> @@ -6973,6 +6970,9 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
>  	if (__alloc_contig_verify_gfp_mask(gfp_mask, (gfp_t *)&cc.gfp_mask))
>  		return -EINVAL;
>  
> +	zone_pcp_disable(cc.zone);
> +	lru_cache_disable();
> +
>  	/*
>  	 * What we do here is we mark all pageblocks in range as
>  	 * MIGRATE_ISOLATE.  Because pageblock and max order pages may
> @@ -6998,8 +6998,6 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
>  	if (ret)
>  		goto done;
>  
> -	drain_all_pages(cc.zone);
> -
>  	/*
>  	 * In case of -EBUSY, we'd like to know which page causes problem.
>  	 * So, just fall through. test_pages_isolated() has a tracepoint
> @@ -7076,6 +7074,8 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
>  		     start, end, outer_start, outer_end);
>  	}
>  done:
> +	lru_cache_enable();
> +	zone_pcp_enable(cc.zone);
>  	undo_isolate_page_range(start, end);
>  	return ret;
>  }
> -- 
> 2.34.1

-- 
Michal Hocko
SUSE Labs

Re: [PATCH] mm: fix execution order in alloc_contig_range_noprof

[Su Hua]

Michal Hocko <mhocko@suse.com> 于2026年1月6日周二 22:55写道：
>
> On Mon 05-01-26 21:32:50, Hua Su wrote:
> > Fix the execution order issue in alloc_contig_range_noprof where
> > drain_all_pages was called after start_isolate_page_range, which
> > may be lead to race conditions.
> >
> > Based on community patches commit ec6e8c7e0314 ("mm, page_alloc:
> > disable pcplists during memory offline") and commit d479960e44f27
> > ("mm: disable LRU pagevec during the migration temporarily"), we
> > disable pcplists and LRU cache before page isolation to ensure no
> > pages are left in per-cpu lists during isolation.
>
> What exactly is the problem you are trying to fix here? Is this based on
> code review or are you hitting any real problem. I find the changelog
> rather hard to grasp.

This submission was discovered during code review. However, it seems
I made a mistake.

Disable LRU pagevec is for the problem where migrate page might fail
(origin code has already disabled). It seems not related to the isolated page
operation.

>
> > Signed-off-by: Hua Su <huasu@xiaomi.com>
> > Signed-off-by: Hua Su <suhua.tanke@gmail.com>
> > ---
> >  mm/page_alloc.c | 10 +++++-----
> >  1 file changed, 5 insertions(+), 5 deletions(-)
> >
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index 6a47443c48ff..d08f929ca64c 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -6815,8 +6815,6 @@ static int __alloc_contig_migrate_range(struct compact_control *cc,
> >               .reason = MR_CONTIG_RANGE,
> >       };
> >
> > -     lru_cache_disable();
> > -
> >       while (pfn < end || !list_empty(&cc->migratepages)) {
> >               if (fatal_signal_pending(current)) {
> >                       ret = -EINTR;
> > @@ -6850,7 +6848,6 @@ static int __alloc_contig_migrate_range(struct compact_control *cc,
> >                       break;
> >       }
> >
> > -     lru_cache_enable();
> >       if (ret < 0) {
> >               if (!(cc->gfp_mask & __GFP_NOWARN) && ret == -EBUSY)
> >                       alloc_contig_dump_pages(&cc->migratepages);
> > @@ -6973,6 +6970,9 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
> >       if (__alloc_contig_verify_gfp_mask(gfp_mask, (gfp_t *)&cc.gfp_mask))
> >               return -EINVAL;
> >
> > +     zone_pcp_disable(cc.zone);
> > +     lru_cache_disable();
> > +
> >       /*
> >        * What we do here is we mark all pageblocks in range as
> >        * MIGRATE_ISOLATE.  Because pageblock and max order pages may
> > @@ -6998,8 +6998,6 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
> >       if (ret)
> >               goto done;
> >
> > -     drain_all_pages(cc.zone);
> > -
> >       /*
> >        * In case of -EBUSY, we'd like to know which page causes problem.
> >        * So, just fall through. test_pages_isolated() has a tracepoint
> > @@ -7076,6 +7074,8 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
> >                    start, end, outer_start, outer_end);
> >       }
> >  done:
> > +     lru_cache_enable();
> > +     zone_pcp_enable(cc.zone);
> >       undo_isolate_page_range(start, end);
> >       return ret;
> >  }
> > --
> > 2.34.1
>
> --
> Michal Hocko
> SUSE Labs