summaryrefslogtreecommitdiffstats
path: root/mm/memory.c
diff options
context:
space:
mode:
authorKirill A. Shutemov <kirill.shutemov@linux.intel.com>2014-01-21 15:49:07 -0800
committerLinus Torvalds <torvalds@linux-foundation.org>2014-01-21 16:19:44 -0800
commitb35f1819acd9243a3ff7ad25b1fa8bd6bfe80fb2 (patch)
tree140f810d09226caa12531e0a02126af64e71e314 /mm/memory.c
parent943dca1a1fcbccb58de944669b833fd38a6c809b (diff)
downloadop-kernel-dev-b35f1819acd9243a3ff7ad25b1fa8bd6bfe80fb2.zip
op-kernel-dev-b35f1819acd9243a3ff7ad25b1fa8bd6bfe80fb2.tar.gz
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>
Diffstat (limited to 'mm/memory.c')
-rw-r--r--mm/memory.c13
1 files changed, 11 insertions, 2 deletions
diff --git a/mm/memory.c b/mm/memory.c
index e9c5504..86487df 100644
--- a/mm/memory.c
+++ b/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
OpenPOWER on IntegriCloud