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>

include/linux/mm.h

@@ -1350,6 +1350,7 @@ static inline pmd_t *pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long a
 
 #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 @@ static inline spinlock_t *ptlock_ptr(struct page *page)
 	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 @@ static inline spinlock_t *pte_lockptr(struct mm_struct *mm, pmd_t *pmd)
 {
 	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);

init/main.c

@@ -476,7 +476,7 @@ static void __init mm_init(void)
 	mem_init();
 	kmem_cache_init();
 	percpu_init_late();
-	pgtable_cache_init();
+	pgtable_init();
 	vmalloc_init();
 }
 

mm/memory.c

@@ -4275,11 +4275,20 @@ void copy_user_huge_page(struct page *dst, struct page *src,
 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
 
 #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
+
+static struct kmem_cache *page_ptl_cachep;
+
+void __init ptlock_cache_init(void)
+{
+	page_ptl_cachep = kmem_cache_create("page->ptl", sizeof(spinlock_t), 0,
+			SLAB_PANIC, NULL);
+}
+
 bool ptlock_alloc(struct page *page)
 {
 	spinlock_t *ptl;
 
-	ptl = kmalloc(sizeof(spinlock_t), GFP_KERNEL);
+	ptl = kmem_cache_alloc(page_ptl_cachep, GFP_KERNEL);
 	if (!ptl)
 		return false;
 	page->ptl = ptl;
@@ -4288,6 +4297,6 @@ bool ptlock_alloc(struct page *page)
 
 void ptlock_free(struct page *page)
 {
-	kfree(page->ptl);
+	kmem_cache_free(page_ptl_cachep, page->ptl);
 }
 #endif