summaryrefslogtreecommitdiff
path: root/init/main.c
diff options
context:
space:
mode:
authorKirill A. Shutemov <kirill.shutemov@linux.intel.com>2014-01-22 03:49:07 +0400
committerLinus Torvalds <torvalds@linux-foundation.org>2014-01-22 04:19:44 +0400
commitb35f1819acd9243a3ff7ad25b1fa8bd6bfe80fb2 (patch)
tree140f810d09226caa12531e0a02126af64e71e314 /init/main.c
parent943dca1a1fcbccb58de944669b833fd38a6c809b (diff)
downloadlinux-b35f1819acd9243a3ff7ad25b1fa8bd6bfe80fb2.tar.xz
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 'init/main.c')
-rw-r--r--init/main.c2
1 files changed, 1 insertions, 1 deletions
diff --git a/init/main.c b/init/main.c
index febc511e078a..01573fdfa186 100644
--- a/init/main.c
+++ b/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();
}