Skip to contentSkip to conversation

mm: remove VM_BUG_ON(PageSlab()) from page_mapcount()

PATCH In review Konstantin Khlebnikov 4 participants 4 messages May 24, 2020 – May 25, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 mm: remove VM_BUG_ON(PageSlab()) from page_mapcount()

Replace superfluous VM_BUG_ON() with comment about correct usage.

Technically reverts commit 1d148e218a0d0566b1c06f2f45f1436d53b049b2 ("mm: add VM_BUG_ON_PAGE() to page_mapcount()"), but context have changed.

Function isolate_migratepages_block() runs some checks out of lru_lock when choose pages for migration. After checking PageLRU() it checks extra page references by comparing page_count() and page_mapcount(). Between these two checks page could be removed from lru, freed and taken by slab.

As a result this race triggers VM_BUG_ON(PageSlab()) in page_mapcount(). Race window is tiny. For certain workload this happens around once a year.

page:ffffea0105ca9380 count:1 mapcount:0 mapping:ffff88ff7712c180 index:0x0 compound_mapcount: 0
flags: 0x500000000008100(slab|head)

 raw: 0500000000008100 dead000000000100 dead000000000200 ffff88ff7712c180
 raw: 0000000000000000 0000000080200020 00000001ffffffff 0000000000000000
 page dumped because: VM_BUG_ON_PAGE(PageSlab(page))
 ------------[ cut here ]------------
 kernel BUG at ./include/linux/mm.h:628!
 invalid opcode: 0000 [#1] SMP NOPTI
 CPU: 77 PID: 504 Comm: kcompactd1 Tainted: G        W         4.19.109-27 #1
 Hardware name: Yandex T175-N41-Y3N/MY81-EX0-Y3N, BIOS R05 06/20/2019
 RIP: 0010:isolate_migratepages_block+0x986/0x9b0

Code in isolate_migratepages_block() was added in commit 119d6d59dcc0 ("mm, compaction: avoid isolating pinned pages") before adding VM_BUG_ON into page_mapcount().

This race has been predicted in 2015 by Vlastimil Babka (see link below).

---
 include/linux/mm.h |   14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
include/linux/mm.h +12 -2
index 5a323422d783..95f777f482ac 100644
@@ -782,6 +782,11 @@ static inline void *kvcalloc(size_t n, size_t size, gfp_t flags)
 
 extern void kvfree(const void *addr);
 
+/*
+ * Mapcount of compound page as a whole, not includes mapped sub-pages.
+ *
+ * Must be called only for compound pages or any their tail sub-pages.
+ */
 static inline int compound_mapcount(struct page *page)
 {
 	VM_BUG_ON_PAGE(!PageCompound(page), page);
@@ -801,10 +806,15 @@ static inline void page_mapcount_reset(struct page *page)
 
 int __page_mapcount(struct page *page);
 
+/*
+ * Mapcount of 0-order page, for sub-page includes compound_mapcount().
+ *
+ * Result is undefined for pages which cannot be mapped into userspace.
+ * For example SLAB or special types of pages. See function page_has_type().
+ * They use this place in struct page differently.
+ */
 static inline int page_mapcount(struct page *page)
 {
-	VM_BUG_ON_PAGE(PageSlab(page), page);
-
 	if (unlikely(PageCompound(page)))
 		return __page_mapcount(page);
 	return atomic_read(&page->_mapcount) + 1;
Acked-by on v2 1/1
#1 lore raw
Quoted 27 lines · Konstantin Khlebnikov

Replace superfluous VM_BUG_ON() with comment about correct usage.

Technically reverts commit 1d148e218a0d0566b1c06f2f45f1436d53b049b2 ("mm: add VM_BUG_ON_PAGE() to page_mapcount()"), but context have changed.

Function isolate_migratepages_block() runs some checks out of lru_lock when choose pages for migration. After checking PageLRU() it checks extra page references by comparing page_count() and page_mapcount(). Between these two checks page could be removed from lru, freed and taken by slab.

As a result this race triggers VM_BUG_ON(PageSlab()) in page_mapcount(). Race window is tiny. For certain workload this happens around once a year.

page:ffffea0105ca9380 count:1 mapcount:0 mapping:ffff88ff7712c180 index:0x0 compound_mapcount: 0
flags: 0x500000000008100(slab|head)

 raw: 0500000000008100 dead000000000100 dead000000000200 ffff88ff7712c180
 raw: 0000000000000000 0000000080200020 00000001ffffffff 0000000000000000
 page dumped because: VM_BUG_ON_PAGE(PageSlab(page))
 ------------[ cut here ]------------
 kernel BUG at ./include/linux/mm.h:628!
 invalid opcode: 0000 [#1] SMP NOPTI
 CPU: 77 PID: 504 Comm: kcompactd1 Tainted: G        W         4.19.109-27 #1
 Hardware name: Yandex T175-N41-Y3N/MY81-EX0-Y3N, BIOS R05 06/20/2019
 RIP: 0010:isolate_migratepages_block+0x986/0x9b0
Quoted 5 lines

Code in isolate_migratepages_block() was added in commit 119d6d59dcc0 ("mm, compaction: avoid isolating pinned pages") before adding VM_BUG_ON into page_mapcount().

This race has been predicted in 2015 by Vlastimil Babka (see link below).

Ah, kudos to Vlastimil.

Quoted 2 lines

Thanks,

For the BUG deletion. I got a bit confused by the comments at first:
slight rewording suggested below.

I had also intended a comment in page_mapcount(), on being used in that racy exceptional way by isolate_migratepages_block(): but now that the BUG is gone, I think you're right not to add that - it's not for the function to document its various callers, nor what's not in it. That kind of info belongs to the commit message, as you have done.

Quoted 12 lines
---
 include/linux/mm.h |   14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)
include/linux/mm.h +0 -0
index 5a323422d783..95f777f482ac 100644

include/linux/mm.h static inline void *kvcalloc(size_t n, size_t size, gfp_t flags) · line 786

Quoted 5 lines
include/linux/mm.h +2 -0
@@ -782,6 +782,11 @@ static inline void *kvcalloc(size_t n, size_t size, gfp_t flags)
 
 extern void kvfree(const void *addr);
 
+/*
+ * Mapcount of compound page as a whole, not includes mapped sub-pages.

s/not includes/does not include/

include/linux/mm.h static inline void page_mapcount_reset(struct page *page) · line 810

Quoted 12 lines
include/linux/mm.h +5 -0
+ *
+ * Must be called only for compound pages or any their tail sub-pages.
+ */
 static inline int compound_mapcount(struct page *page)
 {
 	VM_BUG_ON_PAGE(!PageCompound(page), page);
@@ -801,10 +806,15 @@ static inline void page_mapcount_reset(struct page *page)
 
 int __page_mapcount(struct page *page);
 
+/*
+ * Mapcount of 0-order page, for sub-page includes compound_mapcount().

s/, for sub-page/; when compound sub-page,/

Quoted 12 lines
include/linux/mm.h +5 -2
+ *
+ * Result is undefined for pages which cannot be mapped into userspace.
+ * For example SLAB or special types of pages. See function page_has_type().
+ * They use this place in struct page differently.
+ */
 static inline int page_mapcount(struct page *page)
 {
-	VM_BUG_ON_PAGE(PageSlab(page), page);
-
 	if (unlikely(PageCompound(page)))
 		return __page_mapcount(page);
 	return atomic_read(&page->_mapcount) + 1;
Acked-by on v2 1/1
#2 lore raw
In reply to Konstantin Khlebnikov
Quoted 31 lines · Konstantin Khlebnikov

Replace superfluous VM_BUG_ON() with comment about correct usage.

Technically reverts commit 1d148e218a0d0566b1c06f2f45f1436d53b049b2 ("mm: add VM_BUG_ON_PAGE() to page_mapcount()"), but context have changed.

Function isolate_migratepages_block() runs some checks out of lru_lock when choose pages for migration. After checking PageLRU() it checks extra page references by comparing page_count() and page_mapcount(). Between these two checks page could be removed from lru, freed and taken by slab.

As a result this race triggers VM_BUG_ON(PageSlab()) in page_mapcount(). Race window is tiny. For certain workload this happens around once a year.

page:ffffea0105ca9380 count:1 mapcount:0 mapping:ffff88ff7712c180 index:0x0 compound_mapcount: 0
flags: 0x500000000008100(slab|head)

 raw: 0500000000008100 dead000000000100 dead000000000200 ffff88ff7712c180
 raw: 0000000000000000 0000000080200020 00000001ffffffff 0000000000000000
 page dumped because: VM_BUG_ON_PAGE(PageSlab(page))
 ------------[ cut here ]------------
 kernel BUG at ./include/linux/mm.h:628!
 invalid opcode: 0000 [#1] SMP NOPTI
 CPU: 77 PID: 504 Comm: kcompactd1 Tainted: G        W         4.19.109-27 #1
 Hardware name: Yandex T175-N41-Y3N/MY81-EX0-Y3N, BIOS R05 06/20/2019
 RIP: 0010:isolate_migratepages_block+0x986/0x9b0

Code in isolate_migratepages_block() was added in commit 119d6d59dcc0 ("mm, compaction: avoid isolating pinned pages") before adding VM_BUG_ON into page_mapcount().

Quoted 6 lines

This race has been predicted in 2015 by Vlastimil Babka (see link below).

Signature · 1 line
 Kirill A. Shutemov
Acked-by on v2 1/1
#3 lore raw
In reply to Konstantin Khlebnikov
Quoted 27 lines · Konstantin Khlebnikov

Replace superfluous VM_BUG_ON() with comment about correct usage.

Technically reverts commit 1d148e218a0d0566b1c06f2f45f1436d53b049b2 ("mm: add VM_BUG_ON_PAGE() to page_mapcount()"), but context have changed.

Function isolate_migratepages_block() runs some checks out of lru_lock when choose pages for migration. After checking PageLRU() it checks extra page references by comparing page_count() and page_mapcount(). Between these two checks page could be removed from lru, freed and taken by slab.

As a result this race triggers VM_BUG_ON(PageSlab()) in page_mapcount(). Race window is tiny. For certain workload this happens around once a year.

page:ffffea0105ca9380 count:1 mapcount:0 mapping:ffff88ff7712c180 index:0x0 compound_mapcount: 0
flags: 0x500000000008100(slab|head)

 raw: 0500000000008100 dead000000000100 dead000000000200 ffff88ff7712c180
 raw: 0000000000000000 0000000080200020 00000001ffffffff 0000000000000000
 page dumped because: VM_BUG_ON_PAGE(PageSlab(page))
 ------------[ cut here ]------------
 kernel BUG at ./include/linux/mm.h:628!
 invalid opcode: 0000 [#1] SMP NOPTI
 CPU: 77 PID: 504 Comm: kcompactd1 Tainted: G        W         4.19.109-27 #1
 Hardware name: Yandex T175-N41-Y3N/MY81-EX0-Y3N, BIOS R05 06/20/2019
 RIP: 0010:isolate_migratepages_block+0x986/0x9b0
Quoted 5 lines

Code in isolate_migratepages_block() was added in commit 119d6d59dcc0 ("mm, compaction: avoid isolating pinned pages") before adding VM_BUG_ON into page_mapcount().

This race has been predicted in 2015 by Vlastimil Babka (see link below).

Huh, looks like I made that prediction only half year after that patch has been posted. Now if only I remembered why... I hope it was just a code inspection while chasing something else. I most likely didn't actually see the bug happen, as we don't compile with DEBUG_VM.

Quoted 4 lines

With Hugh's wording tweaks,

Thanks.