When testing page flags with PF_NO_TAIL (enforce=0), tail pages are
legit and they are converted by compound_head(). When modifying page
flags (enforce=1), tail pages are not legit and they either trigger
VM_BUG_ON_PGFLAGS() or are "corrected" by compound_head().
There is no evidence such "correction" is beneficial in terms of
preventing a buggy kernel from crashing. But there is clear evidence
it's detrimental to the size of vmlinux because compound_head() is
redundantly inlined for all modifications of small and head page flags
using PF_NO_TAIL.
This patch makes PF_NO_TAIL skip compound_head() when modifying page
flags. There won't be any behavior change unless a piece of buggy code
tries to modify tail page flags using PF_NO_TAIL. Such code won't be
"corrected", if VM_BUG_ON_PGFLAGS() isn't there to catch it. Again,
there is no evidence such "correction" is beneficial.
bloat-o-meter result:
add/remove: 0/0 grow/shrink: 4/62 up/down: 309/-2851 (-2542)
Signed-off-by: Yu Zhao <yuzhao@xxxxxxxxxx>
---
include/linux/page-flags.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
index db914477057b..1995208a3763 100644
--- a/include/linux/page-flags.h
+++ b/include/linux/page-flags.h
@@ -248,7 +248,7 @@ static inline void page_init_poison(struct page *page, size_t size)
PF_POISONED_CHECK(page); })
#define PF_NO_TAIL(page, enforce) ({ \
VM_BUG_ON_PGFLAGS(enforce && PageTail(page), page); \
- PF_POISONED_CHECK(compound_head(page)); })
+ PF_POISONED_CHECK((enforce) ? (page) : compound_head(page)); })
#define PF_NO_COMPOUND(page, enforce) ({ \
VM_BUG_ON_PGFLAGS(enforce && PageCompound(page), page); \
PF_POISONED_CHECK(page); })
--
2.30.1.766.gb4fecdf3b7-goog
