Skip to contentSkip to conversation

mm: disable LRU pagevec during the migration temporarily

PATCH Applied Minchan Kim 4 participants 8 messages Mar 19, 2021 – Mar 22, 2021

Subsystems (7)
lore
Share & tools
lore
#0 lore raw

PATCH v4 1/3 mm: disable LRU pagevec during the migration temporarily

LRU pagevec holds refcount of pages until the pagevec are drained. It could prevent migration since the refcount of the page is greater than the expection in migration logic. To mitigate the issue, callers of migrate_pages drains LRU pagevec via migrate_prep or lru_add_drain_all before migrate_pages call.

However, it's not enough because pages coming into pagevec after the draining call still could stay at the pagevec so it could keep preventing page migration. Since some callers of migrate_pages have retrial logic with LRU draining, the page would migrate at next trail but it is still fragile in that it doesn't close the fundamental race between upcoming LRU pages into pagvec and migration so the migration failure could cause contiguous memory allocation failure in the end.

To close the race, this patch disables lru caches(i.e, pagevec) during ongoing migration until migrate is done.

Since it's really hard to reproduce, I measured how many times migrate_pages retried with force mode(it is about a fallback to a sync migration) with below debug code.

int migrate_pages(struct list_head *from, new_page_t get_new_page,

			..
			..
if (rc && reason == MR_CONTIG_RANGE && pass > 2) {
       printk(KERN_ERR, "pfn 0x%lx reason %d\n", page_to_pfn(page), rc);
       dump_page(page, "fail to migrate");
}

The test was repeating android apps launching with cma allocation in background every five seconds. Total cma allocation count was about 500 during the testing. With this patch, the dump_page count was reduced from 400 to 30.

The new interface is also useful for memory hotplug which currently drains lru pcp caches after each migration failure. This is rather suboptimal as it has to disrupt others running during the operation. With the new interface the operation happens only once. This is also in line with pcp allocator cache which are disabled for the offlining as well.

---
 include/linux/migrate.h |  2 ++
 include/linux/swap.h    | 14 +++++++++
 mm/memory_hotplug.c     |  3 +-
 mm/mempolicy.c          |  4 +++
 mm/migrate.c            | 11 ++++---
 mm/page_alloc.c         |  2 ++
 mm/swap.c               | 64 +++++++++++++++++++++++++++++++++++------
 7 files changed, 86 insertions(+), 14 deletions(-)
include/linux/migrate.h +2 -0
index 3a389633b68f..9e4a2dc8622c 100644
@@ -46,6 +46,7 @@ extern int isolate_movable_page(struct page *page, isolate_mode_t mode);
 extern void putback_movable_page(struct page *page);
 
 extern void migrate_prep(void);
+extern void migrate_finish(void);
 extern void migrate_prep_local(void);
 extern void migrate_page_states(struct page *newpage, struct page *page);
 extern void migrate_page_copy(struct page *newpage, struct page *page);
@@ -67,6 +68,7 @@ static inline int isolate_movable_page(struct page *page, isolate_mode_t mode)
 	{ return -EBUSY; }
 
 static inline int migrate_prep(void) { return -ENOSYS; }
+static inline int migrate_finish(void) { return -ENOSYS; }
 static inline int migrate_prep_local(void) { return -ENOSYS; }
 
 static inline void migrate_page_states(struct page *newpage, struct page *page)
include/linux/swap.h +14 -0
index 32f665b1ee85..85cf022072a0 100644
@@ -339,6 +339,20 @@ extern void lru_note_cost(struct lruvec *lruvec, bool file,
 extern void lru_note_cost_page(struct page *);
 extern void lru_cache_add(struct page *);
 extern void mark_page_accessed(struct page *);
+
+extern atomic_t lru_disable_count;
+
+static inline bool lru_cache_disabled(void)
+{
+	return atomic_read(&lru_disable_count);
+}
+
+static inline void lru_cache_enable(void)
+{
+	atomic_dec(&lru_disable_count);
+}
+
+extern void lru_cache_disable(void);
 extern void lru_add_drain(void);
 extern void lru_add_drain_cpu(int cpu);
 extern void lru_add_drain_cpu_zone(struct zone *zone);
mm/memory_hotplug.c +2 -1
index 5ba51a8bdaeb..959f659ef085 100644
@@ -1611,6 +1611,7 @@ int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages)
 	 * in a way that pages from isolated pageblock are left on pcplists.
 	 */
 	zone_pcp_disable(zone);
+	lru_cache_disable();
 
 	/* set above range as isolated */
 	ret = start_isolate_page_range(start_pfn, end_pfn,
@@ -1642,7 +1643,6 @@ int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages)
 			}
 
 			cond_resched();
-			lru_add_drain_all();
 
 			ret = scan_movable_pages(pfn, end_pfn, &pfn);
 			if (!ret) {
@@ -1687,6 +1687,7 @@ int __ref offline_pages(unsigned long start_pfn, unsigned long nr_pages)
 	zone->nr_isolate_pageblock -= nr_pages / pageblock_nr_pages;
 	spin_unlock_irqrestore(&zone->lock, flags);
 
+	lru_cache_enable();
 	zone_pcp_enable(zone);
 
 	/* removal success */
mm/mempolicy.c +4 -0
index ab51132547b8..495b43a4b0f8 100644
@@ -1208,6 +1208,8 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 			break;
 	}
 	mmap_read_unlock(mm);
+
+	migrate_finish();
 	if (err < 0)
 		return err;
 	return busy;
@@ -1371,6 +1373,8 @@ static long do_mbind(unsigned long start, unsigned long len,
 	mmap_write_unlock(mm);
 mpol_out:
 	mpol_put(new);
+	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
+		migrate_finish();
 	return err;
 }
 
mm/migrate.c +7 -4
index 62b81d5257aa..4d6c306d41c6 100644
@@ -66,11 +66,13 @@ void migrate_prep(void)
 {
 	/*
 	 * Clear the LRU lists so pages can be isolated.
-	 * Note that pages may be moved off the LRU after we have
-	 * drained them. Those pages will fail to migrate like other
-	 * pages that may be busy.
 	 */
-	lru_add_drain_all();
+	lru_cache_disable();
+}
+
+void migrate_finish(void)
+{
+	lru_cache_enable();
 }
 
 /* Do the necessary work of migrate_prep but not if it involves other CPUs */
@@ -1838,6 +1840,7 @@ static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
 	if (err >= 0)
 		err = err1;
 out:
+	migrate_finish();
 	return err;
 }
 
mm/page_alloc.c +2 -0
index 2e8348936df8..af5d4eeb2999 100644
@@ -8495,6 +8495,8 @@ static int __alloc_contig_migrate_range(struct compact_control *cc,
 		ret = migrate_pages(&cc->migratepages, alloc_migration_target,
 				NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE);
 	}
+
+	migrate_finish();
 	if (ret < 0) {
 		putback_movable_pages(&cc->migratepages);
 		return ret;
mm/swap.c +55 -9

136 lines not on this page · raw message

Signature · 1 line
2.31.0.rc2.261.g7f71774620-goog
#2 lore raw

PATCH v4 3/3 mm: fs: Invalidate BH LRU during page migration

Pages containing buffer_heads that are in one of the per-CPU buffer_head LRU caches will be pinned and thus cannot be migrated. This can prevent CMA allocations from succeeding, which are often used on platforms with co-processors (such as a DSP) that can only use physically contiguous memory. It can also prevent memory hot-unplugging from succeeding, which involves migrating at least MIN_MEMORY_BLOCK_SIZE bytes of memory, which ranges from 8 MiB to 1 GiB based on the architecture in use.

Correspondingly, invalidate the BH LRU caches before a migration starts and stop any buffer_head from being cached in the LRU caches, until migration has finished.

---
 fs/buffer.c                 | 36 ++++++++++++++++++++++++++++++------
 include/linux/buffer_head.h |  4 ++++
 mm/swap.c                   |  5 ++++-
 3 files changed, 38 insertions(+), 7 deletions(-)
fs/buffer.c +30 -6
index 0cb7ffd4977c..e9872d0dcbf1 100644
@@ -1264,6 +1264,15 @@ static void bh_lru_install(struct buffer_head *bh)
 	int i;
 
 	check_irqs_on();
+	/*
+	 * the refcount of buffer_head in bh_lru prevents dropping the
+	 * attached page(i.e., try_to_free_buffers) so it could cause
+	 * failing page migration.
+	 * Skip putting upcoming bh into bh_lru until migration is done.
+	 */
+	if (lru_cache_disabled())
+		return;
+
 	bh_lru_lock();
 
 	b = this_cpu_ptr(&bh_lrus);
@@ -1404,6 +1413,15 @@ __bread_gfp(struct block_device *bdev, sector_t block,
 }
 EXPORT_SYMBOL(__bread_gfp);
 
+static void __invalidate_bh_lrus(struct bh_lru *b)
+{
+	int i;
+
+	for (i = 0; i < BH_LRU_SIZE; i++) {
+		brelse(b->bhs[i]);
+		b->bhs[i] = NULL;
+	}
+}
 /*
  * invalidate_bh_lrus() is called rarely - but not only at unmount.
  * This doesn't race because it runs in each cpu either in irq
@@ -1412,16 +1430,12 @@ EXPORT_SYMBOL(__bread_gfp);
 static void invalidate_bh_lru(void *arg)
 {
 	struct bh_lru *b = &get_cpu_var(bh_lrus);
-	int i;
 
-	for (i = 0; i < BH_LRU_SIZE; i++) {
-		brelse(b->bhs[i]);
-		b->bhs[i] = NULL;
-	}
+	__invalidate_bh_lrus(b);
 	put_cpu_var(bh_lrus);
 }
 
-static bool has_bh_in_lru(int cpu, void *dummy)
+bool has_bh_in_lru(int cpu, void *dummy)
 {
 	struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
 	int i;
@@ -1440,6 +1454,16 @@ void invalidate_bh_lrus(void)
 }
 EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
 
+void invalidate_bh_lrus_cpu(int cpu)
+{
+	struct bh_lru *b;
+
+	bh_lru_lock();
+	b = per_cpu_ptr(&bh_lrus, cpu);
+	__invalidate_bh_lrus(b);
+	bh_lru_unlock();
+}
+
 void set_bh_page(struct buffer_head *bh,
 		struct page *page, unsigned long offset)
 {
include/linux/buffer_head.h +4 -0
index 6b47f94378c5..e7e99da31349 100644
@@ -194,6 +194,8 @@ void __breadahead_gfp(struct block_device *, sector_t block, unsigned int size,
 struct buffer_head *__bread_gfp(struct block_device *,
 				sector_t block, unsigned size, gfp_t gfp);
 void invalidate_bh_lrus(void);
+void invalidate_bh_lrus_cpu(int cpu);
+bool has_bh_in_lru(int cpu, void *dummy);
 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags);
 void free_buffer_head(struct buffer_head * bh);
 void unlock_buffer(struct buffer_head *bh);
@@ -406,6 +408,8 @@ static inline int inode_has_buffers(struct inode *inode) { return 0; }
 static inline void invalidate_inode_buffers(struct inode *inode) {}
 static inline int remove_inode_buffers(struct inode *inode) { return 1; }
 static inline int sync_mapping_buffers(struct address_space *mapping) { return 0; }
+static inline void invalidate_bh_lrus_cpu(int cpu) {}
+static inline bool has_bh_in_lru(int cpu, void *dummy) { return 0; }
 #define buffer_heads_over_limit 0
 
 #endif /* CONFIG_BLOCK */
mm/swap.c +4 -1
index c94f55e7b649..a75a8265302b 100644
@@ -36,6 +36,7 @@
 #include <linux/hugetlb.h>
 #include <linux/page_idle.h>
 #include <linux/local_lock.h>
+#include <linux/buffer_head.h>
 
 #include "internal.h"
 
@@ -641,6 +642,7 @@ void lru_add_drain_cpu(int cpu)
 		pagevec_lru_move_fn(pvec, lru_lazyfree_fn);
 
 	activate_page_drain(cpu);
+	invalidate_bh_lrus_cpu(cpu);
 }
 
 /**
@@ -828,7 +830,8 @@ inline void __lru_add_drain_all(bool force_all_cpus)
 		    pagevec_count(&per_cpu(lru_pvecs.lru_deactivate_file, cpu)) ||
 		    pagevec_count(&per_cpu(lru_pvecs.lru_deactivate, cpu)) ||
 		    pagevec_count(&per_cpu(lru_pvecs.lru_lazyfree, cpu)) ||
-		    need_activate_page_drain(cpu)) {
+		    need_activate_page_drain(cpu) ||
+		    has_bh_in_lru(cpu, NULL)) {
 			INIT_WORK(work, lru_add_drain_per_cpu);
 			queue_work_on(cpu, mm_percpu_wq, work);
 			__cpumask_set_cpu(cpu, &has_work);
Signature · 1 line
2.31.0.rc2.261.g7f71774620-goog
#1 lore raw

PATCH v4 2/3 mm: replace migrate_[prep|finish] with lru_cache_[disable|enable]

In reply to Minchan Kim

Currently, migrate_[prep|finish] is merely a wrapper of lru_cache_[disable|enable]. There is not much to gain from having additional abstraction.

Use lru_cache_[disable|enable] instead of migrate_[prep|finish], which would be more descriptive.

note: migrate_prep_local in compaction.c changed into lru_add_drain to avoid CPU schedule cost with involving many other CPUs to keep keep old behavior.

---
 include/linux/migrate.h |  7 -------
 mm/compaction.c         |  3 ++-
 mm/mempolicy.c          |  8 ++++----
 mm/migrate.c            | 28 ++--------------------------
 mm/page_alloc.c         |  4 ++--
 5 files changed, 10 insertions(+), 40 deletions(-)
include/linux/migrate.h +0 -7
index 9e4a2dc8622c..6155d97ec76c 100644
@@ -45,9 +45,6 @@ extern struct page *alloc_migration_target(struct page *page, unsigned long priv
 extern int isolate_movable_page(struct page *page, isolate_mode_t mode);
 extern void putback_movable_page(struct page *page);
 
-extern void migrate_prep(void);
-extern void migrate_finish(void);
-extern void migrate_prep_local(void);
 extern void migrate_page_states(struct page *newpage, struct page *page);
 extern void migrate_page_copy(struct page *newpage, struct page *page);
 extern int migrate_huge_page_move_mapping(struct address_space *mapping,
@@ -67,10 +64,6 @@ static inline struct page *alloc_migration_target(struct page *page,
 static inline int isolate_movable_page(struct page *page, isolate_mode_t mode)
 	{ return -EBUSY; }
 
-static inline int migrate_prep(void) { return -ENOSYS; }
-static inline int migrate_finish(void) { return -ENOSYS; }
-static inline int migrate_prep_local(void) { return -ENOSYS; }
-
 static inline void migrate_page_states(struct page *newpage, struct page *page)
 {
 }
mm/compaction.c +2 -1
index e04f4476e68e..3be017ececc0 100644
@@ -2319,7 +2319,8 @@ compact_zone(struct compact_control *cc, struct capture_control *capc)
 	trace_mm_compaction_begin(start_pfn, cc->migrate_pfn,
 				cc->free_pfn, end_pfn, sync);
 
-	migrate_prep_local();
+	/* lru_add_drain_all could be expensive with involving other CPUs */
+	lru_add_drain();
 
 	while ((ret = compact_finished(cc)) == COMPACT_CONTINUE) {
 		int err;
mm/mempolicy.c +4 -4
index 495b43a4b0f8..6daf9cc4c843 100644
@@ -1124,7 +1124,7 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 	int err = 0;
 	nodemask_t tmp;
 
-	migrate_prep();
+	lru_cache_disable();
 
 	mmap_read_lock(mm);
 
@@ -1209,7 +1209,7 @@ int do_migrate_pages(struct mm_struct *mm, const nodemask_t *from,
 	}
 	mmap_read_unlock(mm);
 
-	migrate_finish();
+	lru_cache_enable();
 	if (err < 0)
 		return err;
 	return busy;
@@ -1325,7 +1325,7 @@ static long do_mbind(unsigned long start, unsigned long len,
 
 	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL)) {
 
-		migrate_prep();
+		lru_cache_disable();
 	}
 	{
 		NODEMASK_SCRATCH(scratch);
@@ -1374,7 +1374,7 @@ static long do_mbind(unsigned long start, unsigned long len,
 mpol_out:
 	mpol_put(new);
 	if (flags & (MPOL_MF_MOVE | MPOL_MF_MOVE_ALL))
-		migrate_finish();
+		lru_cache_enable();
 	return err;
 }
 
mm/migrate.c +2 -26
index 4d6c306d41c6..acc9913e4303 100644
@@ -57,30 +57,6 @@
 
 #include "internal.h"
 
-/*
- * migrate_prep() needs to be called before we start compiling a list of pages
- * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
- * undesirable, use migrate_prep_local()
- */
-void migrate_prep(void)
-{
-	/*
-	 * Clear the LRU lists so pages can be isolated.
-	 */
-	lru_cache_disable();
-}
-
-void migrate_finish(void)
-{
-	lru_cache_enable();
-}
-
-/* Do the necessary work of migrate_prep but not if it involves other CPUs */
-void migrate_prep_local(void)
-{
-	lru_add_drain();
-}
-
 int isolate_movable_page(struct page *page, isolate_mode_t mode)
 {
 	struct address_space *mapping;
@@ -1771,7 +1747,7 @@ static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
 	int start, i;
 	int err = 0, err1;
 
-	migrate_prep();
+	lru_cache_disable();
 
 	for (i = start = 0; i < nr_pages; i++) {
 		const void __user *p;
@@ -1840,7 +1816,7 @@ static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
 	if (err >= 0)
 		err = err1;
 out:
-	migrate_finish();
+	lru_cache_enable();
 	return err;
 }
 
mm/page_alloc.c +2 -2
index af5d4eeb2999..bf1606c7965a 100644
@@ -8467,7 +8467,7 @@ static int __alloc_contig_migrate_range(struct compact_control *cc,
 		.gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
 	};
 
-	migrate_prep();
+	lru_cache_disable();
 
 	while (pfn < end || !list_empty(&cc->migratepages)) {
 		if (fatal_signal_pending(current)) {
@@ -8496,7 +8496,7 @@ static int __alloc_contig_migrate_range(struct compact_control *cc,
 				NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE);
 	}
 
-	migrate_finish();
+	lru_cache_enable();
 	if (ret < 0) {
 		putback_movable_pages(&cc->migratepages);
 		return ret;
Signature · 1 line
2.31.0.rc2.261.g7f71774620-goog
maintainer
#3 lore raw

Re: [PATCH v4 3/3] mm: fs: Invalidate BH LRU during page migration

In reply to Minchan Kim
Quoted 9 lines · On Fri, 19 Mar 2021 10:51:27 -0700 Minchan Kim <[email protected]> wrote:

Pages containing buffer_heads that are in one of the per-CPU buffer_head LRU caches will be pinned and thus cannot be migrated. This can prevent CMA allocations from succeeding, which are often used on platforms with co-processors (such as a DSP) that can only use physically contiguous memory. It can also prevent memory hot-unplugging from succeeding, which involves migrating at least MIN_MEMORY_BLOCK_SIZE bytes of memory, which ranges from 8 MiB to 1 GiB based on the architecture in use.

Quoted 8 lines

Correspondingly, invalidate the BH LRU caches before a migration starts and stop any buffer_head from being cached in the LRU caches, until migration has finished.

The signoff chain ordering might mean that Chris was the primary author, but there is no From:him. Please clarify?

#4 lore raw

Re: [PATCH v4 3/3] mm: fs: Invalidate BH LRU during page migration

Quoted 15 lines · Andrew Morton
Quoted 13 lines · On Fri, 19 Mar 2021 10:51:27 -0700 Minchan Kim <[email protected]> wrote:

Pages containing buffer_heads that are in one of the per-CPU buffer_head LRU caches will be pinned and thus cannot be migrated. This can prevent CMA allocations from succeeding, which are often used on platforms with co-processors (such as a DSP) that can only use physically contiguous memory. It can also prevent memory hot-unplugging from succeeding, which involves migrating at least MIN_MEMORY_BLOCK_SIZE bytes of memory, which ranges from 8 MiB to 1 GiB based on the architecture in use.

Correspondingly, invalidate the BH LRU caches before a migration starts and stop any buffer_head from being cached in the LRU caches, until migration has finished.

Quoted 7 lines
Quoted 4 lines

The signoff chain ordering might mean that Chris was the primary author, but there is no From:him. Please clarify?

He tried first version but was diffrent implementation since I changed a lot. That's why I added his SoB even though current implementaion is much different. So, maybe I am primary author?

#5 lore raw

Re: [PATCH v4 3/3] mm: fs: Invalidate BH LRU during page migration

Quoted 11 lines · Minchan Kim
Quoted 7 lines
Quoted 4 lines

The signoff chain ordering might mean that Chris was the primary author, but there is no From:him. Please clarify?

He tried first version but was diffrent implementation since I changed a lot. That's why I added his SoB even though current implementaion is much different. So, maybe I am primary author?

Maybe Chris is Reported-by: ? And don't forget Laura Abbott as original author of the patch Chris submitted. I think she should be Reported-by:
as well, since there is no code from either of them in this version of the patch.

#6 lore raw

Re: [PATCH v4 3/3] mm: fs: Invalidate BH LRU during page migration

Quoted 6 lines · Matthew Wilcox
Quoted 5 lines · Minchan Kim
Quoted 5 lines
Quoted 4 lines
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

The signoff chain ordering might mean that Chris was the primary author, but there is no From:him. Please clarify?

He tried first version but was diffrent implementation since I changed a lot. That's why I added his SoB even though current implementaion is much different. So, maybe I am primary author?

Hey Minchan, let's have you as the primary author.

Quoted 4 lines

Maybe Chris is Reported-by: ? And don't forget Laura Abbott as original author of the patch Chris submitted. I think she should be Reported-by:
as well, since there is no code from either of them in this version of the patch.

Yes, let's have a Reported-by: from Laura. We can change my Signed-off-by to Reported-by: as well.

Signature · 3 lines
The Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
author Tested-by Oliver Sang
#7 lore raw

Re: [PATCH v4 3/3] mm: fs: Invalidate BH LRU during page migration

Quoted 16 lines · Chris Goldsworthy
Quoted 6 lines · Matthew Wilcox
Quoted 5 lines · Minchan Kim
Quoted 5 lines
Quoted 4 lines
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines

The signoff chain ordering might mean that Chris was the primary author, but there is no From:him. Please clarify?

He tried first version but was diffrent implementation since I changed a lot. That's why I added his SoB even though current implementaion is much different. So, maybe I am primary author?

Hey Minchan, let's have you as the primary author.

Quoted 7 lines
Quoted 4 lines

Maybe Chris is Reported-by: ? And don't forget Laura Abbott as original author of the patch Chris submitted. I think she should be Reported-by:
as well, since there is no code from either of them in this version of the patch.

Yes, let's have a Reported-by: from Laura. We can change my Signed-off-by to
Reported-by: as well.

Thanks Matthew and Chris.

I just wanted to give more credit(something like Co-authored-by if we have such kinds of tag) to Chris since I felt he had more than. :)

Okay, let's have something like this.

From: Minchan Kim <[email protected]>

..

Andrew, if you want me resend the patch, let me know.

Thank you.