slob: remove bigblock tracking
authorNick Piggin <npiggin@suse.de>
Mon, 16 Jul 2007 06:38:08 +0000 (23:38 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Mon, 16 Jul 2007 16:05:35 +0000 (09:05 -0700)
Remove the bigblock lists in favour of using compound pages and going directly
to the page allocator.  Allocation size is stored in page->private, which also
makes ksize more accurate than it previously was.

Saves ~.5K of code, and 12-24 bytes overhead per >= PAGE_SIZE allocation.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/slob.c

index 8ee64fed2bb56db97792b7fa4a7d8dc665123979..11a87586d4b75db623c523bea8339114d6d1901c 100644 (file)
--- a/mm/slob.c
+++ b/mm/slob.c
  * Above this is an implementation of kmalloc/kfree. Blocks returned
  * from kmalloc are 4-byte aligned and prepended with a 4-byte header.
  * If kmalloc is asked for objects of PAGE_SIZE or larger, it calls
- * __get_free_pages directly so that it can return page-aligned blocks
- * and keeps a linked list of such pages and their orders. These
- * objects are detected in kfree() by their page alignment.
+ * __get_free_pages directly, allocating compound pages so the page order
+ * does not have to be separately tracked, and also stores the exact
+ * allocation size in page->private so that it can be used to accurately
+ * provide ksize(). These objects are detected in kfree() because slob_page()
+ * is false for them.
  *
  * SLAB is emulated on top of SLOB by simply calling constructors and
  * destructors for every SLAB allocation. Objects are returned with the
@@ -29,7 +31,8 @@
  * alignment. Again, objects of page-size or greater are allocated by
  * calling __get_free_pages. As SLAB objects know their size, no separate
  * size bookkeeping is necessary and there is essentially no allocation
- * space overhead.
+ * space overhead, and compound pages aren't needed for multi-page
+ * allocations.
  */
 
 #include <linux/kernel.h>
@@ -381,48 +384,26 @@ out:
  * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend.
  */
 
-struct bigblock {
-       int order;
-       void *pages;
-       struct bigblock *next;
-};
-typedef struct bigblock bigblock_t;
-
-static bigblock_t *bigblocks;
-
-static DEFINE_SPINLOCK(block_lock);
-
-
 void *__kmalloc(size_t size, gfp_t gfp)
 {
-       slob_t *m;
-       bigblock_t *bb;
-       unsigned long flags;
-
        if (size < PAGE_SIZE - SLOB_UNIT) {
+               slob_t *m;
                m = slob_alloc(size + SLOB_UNIT, gfp, 0);
                if (m)
                        m->units = size;
                return m+1;
+       } else {
+               void *ret;
+
+               ret = (void *) __get_free_pages(gfp | __GFP_COMP,
+                                               get_order(size));
+               if (ret) {
+                       struct page *page;
+                       page = virt_to_page(ret);
+                       page->private = size;
+               }
+               return ret;
        }
-
-       bb = slob_alloc(sizeof(bigblock_t), gfp, 0);
-       if (!bb)
-               return 0;
-
-       bb->order = get_order(size);
-       bb->pages = (void *)__get_free_pages(gfp, bb->order);
-
-       if (bb->pages) {
-               spin_lock_irqsave(&block_lock, flags);
-               bb->next = bigblocks;
-               bigblocks = bb;
-               spin_unlock_irqrestore(&block_lock, flags);
-               return bb->pages;
-       }
-
-       slob_free(bb, sizeof(bigblock_t));
-       return 0;
 }
 EXPORT_SYMBOL(__kmalloc);
 
@@ -462,59 +443,33 @@ EXPORT_SYMBOL(krealloc);
 void kfree(const void *block)
 {
        struct slob_page *sp;
-       slob_t *m;
-       bigblock_t *bb, **last = &bigblocks;
-       unsigned long flags;
 
        if (!block)
                return;
 
        sp = (struct slob_page *)virt_to_page(block);
-       if (!slob_page(sp)) {
-               /* on the big block list */
-               spin_lock_irqsave(&block_lock, flags);
-               for (bb = bigblocks; bb; last = &bb->next, bb = bb->next) {
-                       if (bb->pages == block) {
-                               *last = bb->next;
-                               spin_unlock_irqrestore(&block_lock, flags);
-                               free_pages((unsigned long)block, bb->order);
-                               slob_free(bb, sizeof(bigblock_t));
-                               return;
-                       }
-               }
-               spin_unlock_irqrestore(&block_lock, flags);
-               WARN_ON(1);
-               return;
-       }
-
-       m = (slob_t *)block - 1;
-       slob_free(m, m->units + SLOB_UNIT);
-       return;
+       if (slob_page(sp)) {
+               slob_t *m = (slob_t *)block - 1;
+               slob_free(m, m->units + SLOB_UNIT);
+       } else
+               put_page(&sp->page);
 }
 
 EXPORT_SYMBOL(kfree);
 
+/* can't use ksize for kmem_cache_alloc memory, only kmalloc */
 size_t ksize(const void *block)
 {
        struct slob_page *sp;
-       bigblock_t *bb;
-       unsigned long flags;
 
        if (!block)
                return 0;
 
        sp = (struct slob_page *)virt_to_page(block);
-       if (!slob_page(sp)) {
-               spin_lock_irqsave(&block_lock, flags);
-               for (bb = bigblocks; bb; bb = bb->next)
-                       if (bb->pages == block) {
-                               spin_unlock_irqrestore(&slob_lock, flags);
-                               return PAGE_SIZE << bb->order;
-                       }
-               spin_unlock_irqrestore(&block_lock, flags);
-       }
-
-       return ((slob_t *)block - 1)->units + SLOB_UNIT;
+       if (slob_page(sp))
+               return ((slob_t *)block - 1)->units + SLOB_UNIT;
+       else
+               return sp->page.private;
 }
 
 struct kmem_cache {