[Devel] [PATCH hci-8.1 0/3] mm/memcontrol: tidy up the memory.numa_migrate code

[Devel] [PATCH hci-8.1 0/3] mm/memcontrol: tidy up the memory.numa_migrate code

[Konstantin Khorenko]

Leftovers found while fixing VSTOR-151607, where the allocator of
memory.numa_migrate had fallen behind the upstream function it was
copied from. Nothing here changes behaviour. Applies on top of
"mm/memcontrol: numa_migrate destinations of the source folio order".

- 1: update_lru_sizes() lost its only caller when the isolation loop
  switched to lruvec_del_folio(); static __always_inline, so no
  warning ever came.
- 2: the comment in the isolation loop told the rebase history of the
  code. It now names the upstream functions the loop and the allocator
  copy (isolate_migratepages_block(), isolate_lru_folios(),
  alloc_migration_target()), which is what the next rebase has to
  compare against. The commented-out conditions and the pre-folio name
  of the allocator go with it.
- 3: the MAX_SCAN argument of memory.numa_migrate is documented in
  pages while the scan counts folios.

https://virtuozzo.atlassian.net/browse/VSTOR-152197

Assisted-by: LLM
Signed-off-by: Konstantin Khorenko <khorenko@virtuozzo.com>
---
Konstantin Khorenko (3):
      mm/memcontrol: drop the unused update_lru_sizes()
      mm/memcontrol: say where the numa_migrate code comes from
      mm/memcontrol: memory.numa_migrate MAX_SCAN counts folios

 mm/memcontrol.c | 54 +++++++++++++-----------------------------------------
 1 file changed, 13 insertions(+), 41 deletions(-)
---
base-commit: 8f4b2db111bee113348656f1b1f98c841d7c19e7
change-id: 20261028-b4-prep-khorenko-vstor-152197-numa-migrate-cleanup-dcc488d8d106

Best regards,
--  
Konstantin Khorenko,
Virtuozzo Linux Kernel Team

_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel

[Devel] [PATCH hci-8.1 1/3] mm/memcontrol: drop the unused update_lru_sizes()

[Konstantin Khorenko]

It has had no caller since commit 3c6e0c1ac1d6 ("mm/memcontrol: use
lruvec_del_folio() in memcg_numa_isolate_pages()"), and being static
__always_inline it never drew a warning.

https://virtuozzo.atlassian.net/browse/VSTOR-152197

Feature: mm: interface to migrate memory between NUMA nodes upon userspace request
Signed-off-by: Konstantin Khorenko <khorenko@virtuozzo.com>
---
 mm/memcontrol.c | 18 ------------------
 1 file changed, 18 deletions(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index d8a87913ace37..a82d5e7bc8694 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -3497,24 +3497,6 @@ static struct folio *memcg_numa_migrate_new_folio(struct folio *old_folio,
 			     ms->target_nodes);
 }
 
-/*
- * Update LRU sizes after isolating pages. The LRU size updates must
- * be complete before mem_cgroup_update_lru_size due to a santity check.
- */
-static __always_inline void update_lru_sizes(struct lruvec *lruvec,
-			enum lru_list lru, unsigned long *nr_zone_taken)
-{
-	int zid;
-
-	for (zid = 0; zid < MAX_NR_ZONES; zid++) {
-		if (!nr_zone_taken[zid])
-			continue;
-
-		update_lru_size(lruvec, lru, zid, -nr_zone_taken[zid]);
-	}
-
-}
-
 /*
  * Isolate at most @nr_to_scan pages from @lruvec for further migration and
  * store them in @dst. Returns the number of pages scanned. Return value of 0

-- 
2.43.0

_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel

[Devel] [PATCH hci-8.1 2/3] mm/memcontrol: say where the numa_migrate code comes from

[Konstantin Khorenko]

The comment in the isolation loop tells the rebase history of the code
instead of naming the upstream functions it copies, which is what a
rebase has to compare against. Replace it with those names, name the
original of the allocator too, and drop the leftovers of the copy: the
commented-out conditions and the pre-folio name of the allocator in
the comment to its private argument.

No functional change.

https://virtuozzo.atlassian.net/browse/VSTOR-152197

Feature: mm: interface to migrate memory between NUMA nodes upon userspace request
Signed-off-by: Konstantin Khorenko <khorenko@virtuozzo.com>
---
 mm/memcontrol.c | 24 +++++++-----------------
 1 file changed, 7 insertions(+), 17 deletions(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index a82d5e7bc8694..4c3c06cfcf9fc 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -3462,7 +3462,7 @@ int ve_enough_memory(long pages)
 #ifdef CONFIG_NUMA
 
 /*
- * memcg_numa_migrate_new_page() private argument. @target_nodes specifies the
+ * memcg_numa_migrate_new_folio() private argument. @target_nodes specifies the
  * set of nodes to allocate pages from. @current_node is the current preferable
  * node, it gets rotated after each allocation.
  */
@@ -3473,7 +3473,8 @@ struct memcg_numa_migrate_struct {
 
 /*
  * Used as an argument for migrate_pages(). Allocated pages are spread evenly
- * among destination nodes.
+ * among destination nodes. The order and gfp choice follow
+ * alloc_migration_target().
  */
 static struct folio *memcg_numa_migrate_new_folio(struct folio *old_folio,
 						  unsigned long private)
@@ -3520,19 +3521,10 @@ static long memcg_numa_isolate_pages(struct lruvec *lruvec,
 		scanned++;
 
 		/*
-		 * Previously here we had a call to
-		 * __isolate_lru_page_prepare(page, ISOLATE_ASYNC_MIGRATE)
-		 * => need to get same parts from __isolate_lru_page_prepare()
-		 * like it's done in isolate_migratepages_block() in commit
-		 * 89f6c88a6ab4 ("mm: __isolate_lru_page_prepare() in
-		 * isolate_migratepages_block()")
-		 *
-		 * The code block below is taken from
-		 * isolate_migratepages_block(), mode variable is defined and
-		 * set to ISOLATE_ASYNC_MIGRATE in order to keep the original
-		 * code unchanged where possible.
-		 *
-		 * Also it is switched to folio, based on isolate_lru_folios().
+		 * The checks below are a copy of the per-folio checks in
+		 * isolate_migratepages_block() with mode == ISOLATE_ASYNC_MIGRATE;
+		 * the isolation itself follows isolate_lru_folios(). Keep both
+		 * in sync with their originals on a rebase.
 		 */
 		folio = lru_to_folio(src);
 
@@ -3552,7 +3544,6 @@ static long memcg_numa_isolate_pages(struct lruvec *lruvec,
 		is_unevictable = folio_test_unevictable(folio);
 
 		/* Compaction might skip unevictable pages but CMA takes them */
-		/* if (is_unevictable) */
 		if (!(mode & ISOLATE_UNEVICTABLE) && is_unevictable)
 			goto isolate_fail;
 
@@ -3567,7 +3558,6 @@ static long memcg_numa_isolate_pages(struct lruvec *lruvec,
 
 		is_dirty = folio_test_dirty(folio);
 
-		/* if (is_dirty) */
 		if (((mode & ISOLATE_ASYNC_MIGRATE) && is_dirty) ||
 		    (mapping && is_unevictable)) {
 			bool migrate_dirty = true;

-- 
2.43.0

_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel

[Devel] [PATCH hci-8.1 3/3] mm/memcontrol: memory.numa_migrate MAX_SCAN counts folios

[Konstantin Khorenko]

memcg_numa_isolate_pages() counts a scanned folio as one whatever its
size, so the MAX_SCAN limit of memory.numa_migrate and the @nr_to_scan
arguments are in folios, not in pages as the comments say.

No functional change.

https://virtuozzo.atlassian.net/browse/VSTOR-152197

Feature: mm: interface to migrate memory between NUMA nodes upon userspace request
Signed-off-by: Konstantin Khorenko <khorenko@virtuozzo.com>
---
 mm/memcontrol.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 4c3c06cfcf9fc..f93b0400c786d 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -3499,8 +3499,8 @@ static struct folio *memcg_numa_migrate_new_folio(struct folio *old_folio,
 }
 
 /*
- * Isolate at most @nr_to_scan pages from @lruvec for further migration and
- * store them in @dst. Returns the number of pages scanned. Return value of 0
+ * Isolate at most @nr_to_scan folios from @lruvec for further migration and
+ * store them in @dst. Returns the number of folios scanned. Return value of 0
  * means that @lruvec is empty.
  */
 static long memcg_numa_isolate_pages(struct lruvec *lruvec,
@@ -3661,9 +3661,9 @@ static long __memcg_numa_migrate_pages(struct lruvec *lruvec, struct list_head *
 }
 
 /*
- * Migrate at most @nr_to_scan pages accounted to @memcg to @target_nodes.
- * Pages are spreaded evenly among destination nodes. If @nr_to_scan is <= 0,
- * then the function will attempt to migrate all pages accounted to @memcg.
+ * Scan at most @nr_to_scan folios accounted to @memcg and migrate them to
+ * @target_nodes, spread evenly among the nodes. If @nr_to_scan is <= 0, the
+ * function attempts to migrate all pages accounted to @memcg.
  */
 static int memcg_numa_migrate_pages(struct mem_cgroup *memcg,
 				    nodemask_t *target_nodes, long nr_to_scan)
@@ -3750,7 +3750,7 @@ static int memcg_numa_migrate_pages(struct mem_cgroup *memcg,
  *
  * where NODELIST is a comma-separated list of ranges N1-N2 specifying the set
  * of nodes to migrate pages of this cgroup to, and the optional MAX_SCAN
- * imposes a limit on the number of pages that can be migrated in one go.
+ * imposes a limit on the number of folios scanned in one go.
  *
  * The call may be interrupted by a signal, in which case -EINTR is returned.
  */

-- 
2.43.0

_______________________________________________
Devel mailing list
Devel@openvz.org
https://lists.openvz.org/mailman/listinfo/devel