]> www.infradead.org Git - users/jedix/linux-maple.git/commitdiff
memcg: slub: fix SUnreclaim for post charged objects
authorShakeel Butt <shakeel.butt@linux.dev>
Tue, 10 Dec 2024 04:06:57 +0000 (20:06 -0800)
committerVlastimil Babka <vbabka@suse.cz>
Tue, 10 Dec 2024 08:25:39 +0000 (09:25 +0100)
Large kmalloc directly allocates from the page allocator and then use
lruvec_stat_mod_folio() to increment the unreclaimable slab stats for
global and memcg. However when post memcg charging of slab objects was
added in commit 9028cdeb38e1 ("memcg: add charging of already allocated
slab objects"), it missed to correctly handle the unreclaimable slab
stats for memcg.

One user visisble effect of that bug is that the node level
unreclaimable slab stat will work correctly but the memcg level stat can
underflow as kernel correctly handles the free path but the charge path
missed to increment the memcg level unreclaimable slab stat. Let's fix
by correctly handle in the post charge code path.

Fixes: 9028cdeb38e1 ("memcg: add charging of already allocated slab objects")
Signed-off-by: Shakeel Butt <shakeel.butt@linux.dev>
Cc: <stable@vger.kernel.org>
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
mm/slub.c

index 19980419b176c28d2e6247f8b076c0a83b65a436..c2151c9fee228d121a9cbcc220c3ae054769dacf 100644 (file)
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2189,9 +2189,24 @@ bool memcg_slab_post_charge(void *p, gfp_t flags)
 
        folio = virt_to_folio(p);
        if (!folio_test_slab(folio)) {
-               return folio_memcg_kmem(folio) ||
-                       (__memcg_kmem_charge_page(folio_page(folio, 0), flags,
-                                                 folio_order(folio)) == 0);
+               int size;
+
+               if (folio_memcg_kmem(folio))
+                       return true;
+
+               if (__memcg_kmem_charge_page(folio_page(folio, 0), flags,
+                                            folio_order(folio)))
+                       return false;
+
+               /*
+                * This folio has already been accounted in the global stats but
+                * not in the memcg stats. So, subtract from the global and use
+                * the interface which adds to both global and memcg stats.
+                */
+               size = folio_size(folio);
+               node_stat_mod_folio(folio, NR_SLAB_UNRECLAIMABLE_B, -size);
+               lruvec_stat_mod_folio(folio, NR_SLAB_UNRECLAIMABLE_B, size);
+               return true;
        }
 
        slab = folio_slab(folio);