mm: create a separate slab for page->ptl allocation
If DEBUG_SPINLOCK and DEBUG_LOCK_ALLOC are enabled spinlock_t on x86_64
is 72 bytes. For page->ptl they will be allocated from kmalloc-96 slab,
so we loose 24 on each. An average system can easily allocate few tens
thousands of page->ptl and overhead is significant.
Let's create a separate slab for page->ptl allocation to solve this.
To make sure that it really works this time, some numbers from my test
machine (just booted, no load):
Before:
# grep '^\(kmalloc-96\|page->ptl\)' /proc/slabinfo
kmalloc-96 31987 32190 128 30 1 : tunables 120 60 8 : slabdata 1073 1073 92
After:
# grep '^\(kmalloc-96\|page->ptl\)' /proc/slabinfo
page->ptl 27516 28143 72 53 1 : tunables 120 60 8 : slabdata 531 531 9
kmalloc-96 3853 5280 128 30 1 : tunables 120 60 8 : slabdata 176 176 0
Note that the patch is useful not only for debug case, but also for
PREEMPT_RT, where spinlock_t is always bloated.
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
diff --git a/include/linux/mm.h b/include/linux/mm.h
index 58202c2..fc44152 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1350,6 +1350,7 @@
#if USE_SPLIT_PTE_PTLOCKS
#if ALLOC_SPLIT_PTLOCKS
+void __init ptlock_cache_init(void);
extern bool ptlock_alloc(struct page *page);
extern void ptlock_free(struct page *page);
@@ -1358,6 +1359,10 @@
return page->ptl;
}
#else /* ALLOC_SPLIT_PTLOCKS */
+static inline void ptlock_cache_init(void)
+{
+}
+
static inline bool ptlock_alloc(struct page *page)
{
return true;
@@ -1410,10 +1415,17 @@
{
return &mm->page_table_lock;
}
+static inline void ptlock_cache_init(void) {}
static inline bool ptlock_init(struct page *page) { return true; }
static inline void pte_lock_deinit(struct page *page) {}
#endif /* USE_SPLIT_PTE_PTLOCKS */
+static inline void pgtable_init(void)
+{
+ ptlock_cache_init();
+ pgtable_cache_init();
+}
+
static inline bool pgtable_page_ctor(struct page *page)
{
inc_zone_page_state(page, NR_PAGETABLE);