slab: change return type of kmem_getpages() to struct page
authorJoonsoo Kim <iamjoonsoo.kim@lge.com>
Thu, 24 Oct 2013 01:07:38 +0000 (10:07 +0900)
committerPekka Enberg <penberg@iki.fi>
Thu, 24 Oct 2013 17:17:23 +0000 (20:17 +0300)
It is more understandable that kmem_getpages() return struct page.
And, with this, we can reduce one translation from virt addr to page and
makes better code than before. Below is a change of this patch.

* Before
   text    data     bss     dec     hex filename
  22123   23434       4   45561    b1f9 mm/slab.o

* After
   text    data     bss     dec     hex filename
  22074   23434       4   45512    b1c8 mm/slab.o

And this help following patch to remove struct slab's colouroff.

Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Pekka Enberg <penberg@iki.fi>
mm/slab.c

index 0b4ddafd8a030fb2269af01134a4a5a9eb342b6d..7d79bd76600293f77757c3bc7035e2323c17d3e3 100644 (file)
--- a/mm/slab.c
+++ b/mm/slab.c
@@ -205,7 +205,7 @@ typedef unsigned int kmem_bufctl_t;
 struct slab_rcu {
        struct rcu_head head;
        struct kmem_cache *cachep;
-       void *addr;
+       struct page *page;
 };
 
 /*
@@ -1737,7 +1737,8 @@ slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid)
  * did not request dmaable memory, we might get it, but that
  * would be relatively rare and ignorable.
  */
-static void *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, int nodeid)
+static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags,
+                                                               int nodeid)
 {
        struct page *page;
        int nr_pages;
@@ -1790,16 +1791,15 @@ static void *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, int nodeid)
                        kmemcheck_mark_unallocated_pages(page, nr_pages);
        }
 
-       return page_address(page);
+       return page;
 }
 
 /*
  * Interface to system's page release.
  */
-static void kmem_freepages(struct kmem_cache *cachep, void *addr)
+static void kmem_freepages(struct kmem_cache *cachep, struct page *page)
 {
        unsigned long i = (1 << cachep->gfporder);
-       struct page *page = virt_to_page(addr);
        const unsigned long nr_freed = i;
 
        kmemcheck_free_shadow(page, cachep->gfporder);
@@ -1821,7 +1821,7 @@ static void kmem_freepages(struct kmem_cache *cachep, void *addr)
        memcg_release_pages(cachep, cachep->gfporder);
        if (current->reclaim_state)
                current->reclaim_state->reclaimed_slab += nr_freed;
-       free_memcg_kmem_pages((unsigned long)addr, cachep->gfporder);
+       __free_memcg_kmem_pages(page, cachep->gfporder);
 }
 
 static void kmem_rcu_free(struct rcu_head *head)
@@ -1829,7 +1829,7 @@ static void kmem_rcu_free(struct rcu_head *head)
        struct slab_rcu *slab_rcu = (struct slab_rcu *)head;
        struct kmem_cache *cachep = slab_rcu->cachep;
 
-       kmem_freepages(cachep, slab_rcu->addr);
+       kmem_freepages(cachep, slab_rcu->page);
        if (OFF_SLAB(cachep))
                kmem_cache_free(cachep->slabp_cache, slab_rcu);
 }
@@ -2048,7 +2048,7 @@ static void slab_destroy_debugcheck(struct kmem_cache *cachep, struct slab *slab
  */
 static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp)
 {
-       void *addr = slabp->s_mem - slabp->colouroff;
+       struct page *page = virt_to_head_page(slabp->s_mem);
 
        slab_destroy_debugcheck(cachep, slabp);
        if (unlikely(cachep->flags & SLAB_DESTROY_BY_RCU)) {
@@ -2056,10 +2056,10 @@ static void slab_destroy(struct kmem_cache *cachep, struct slab *slabp)
 
                slab_rcu = (struct slab_rcu *)slabp;
                slab_rcu->cachep = cachep;
-               slab_rcu->addr = addr;
+               slab_rcu->page = page;
                call_rcu(&slab_rcu->head, kmem_rcu_free);
        } else {
-               kmem_freepages(cachep, addr);
+               kmem_freepages(cachep, page);
                if (OFF_SLAB(cachep))
                        kmem_cache_free(cachep->slabp_cache, slabp);
        }
@@ -2604,11 +2604,12 @@ int __kmem_cache_shutdown(struct kmem_cache *cachep)
  * kmem_find_general_cachep till the initialization is complete.
  * Hence we cannot have slabp_cache same as the original cache.
  */
-static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, void *objp,
-                                  int colour_off, gfp_t local_flags,
-                                  int nodeid)
+static struct slab *alloc_slabmgmt(struct kmem_cache *cachep,
+                                  struct page *page, int colour_off,
+                                  gfp_t local_flags, int nodeid)
 {
        struct slab *slabp;
+       void *addr = page_address(page);
 
        if (OFF_SLAB(cachep)) {
                /* Slab management obj is off-slab. */
@@ -2625,12 +2626,12 @@ static struct slab *alloc_slabmgmt(struct kmem_cache *cachep, void *objp,
                if (!slabp)
                        return NULL;
        } else {
-               slabp = objp + colour_off;
+               slabp = addr + colour_off;
                colour_off += cachep->slab_size;
        }
        slabp->inuse = 0;
        slabp->colouroff = colour_off;
-       slabp->s_mem = objp + colour_off;
+       slabp->s_mem = addr + colour_off;
        slabp->nodeid = nodeid;
        slabp->free = 0;
        return slabp;
@@ -2741,12 +2742,9 @@ static void slab_put_obj(struct kmem_cache *cachep, struct slab *slabp,
  * virtual address for kfree, ksize, and slab debugging.
  */
 static void slab_map_pages(struct kmem_cache *cache, struct slab *slab,
-                          void *addr)
+                          struct page *page)
 {
        int nr_pages;
-       struct page *page;
-
-       page = virt_to_page(addr);
 
        nr_pages = 1;
        if (likely(!PageCompound(page)))
@@ -2764,7 +2762,7 @@ static void slab_map_pages(struct kmem_cache *cache, struct slab *slab,
  * kmem_cache_alloc() when there are no active objs left in a cache.
  */
 static int cache_grow(struct kmem_cache *cachep,
-               gfp_t flags, int nodeid, void *objp)
+               gfp_t flags, int nodeid, struct page *page)
 {
        struct slab *slabp;
        size_t offset;
@@ -2807,18 +2805,18 @@ static int cache_grow(struct kmem_cache *cachep,
         * Get mem for the objs.  Attempt to allocate a physical page from
         * 'nodeid'.
         */
-       if (!objp)
-               objp = kmem_getpages(cachep, local_flags, nodeid);
-       if (!objp)
+       if (!page)
+               page = kmem_getpages(cachep, local_flags, nodeid);
+       if (!page)
                goto failed;
 
        /* Get slab management. */
-       slabp = alloc_slabmgmt(cachep, objp, offset,
+       slabp = alloc_slabmgmt(cachep, page, offset,
                        local_flags & ~GFP_CONSTRAINT_MASK, nodeid);
        if (!slabp)
                goto opps1;
 
-       slab_map_pages(cachep, slabp, objp);
+       slab_map_pages(cachep, slabp, page);
 
        cache_init_objs(cachep, slabp);
 
@@ -2834,7 +2832,7 @@ static int cache_grow(struct kmem_cache *cachep,
        spin_unlock(&n->list_lock);
        return 1;
 opps1:
-       kmem_freepages(cachep, objp);
+       kmem_freepages(cachep, page);
 failed:
        if (local_flags & __GFP_WAIT)
                local_irq_disable();
@@ -3250,18 +3248,20 @@ retry:
                 * We may trigger various forms of reclaim on the allowed
                 * set and go into memory reserves if necessary.
                 */
+               struct page *page;
+
                if (local_flags & __GFP_WAIT)
                        local_irq_enable();
                kmem_flagcheck(cache, flags);
-               obj = kmem_getpages(cache, local_flags, numa_mem_id());
+               page = kmem_getpages(cache, local_flags, numa_mem_id());
                if (local_flags & __GFP_WAIT)
                        local_irq_disable();
-               if (obj) {
+               if (page) {
                        /*
                         * Insert into the appropriate per node queues
                         */
-                       nid = page_to_nid(virt_to_page(obj));
-                       if (cache_grow(cache, flags, nid, obj)) {
+                       nid = page_to_nid(page);
+                       if (cache_grow(cache, flags, nid, page)) {
                                obj = ____cache_alloc_node(cache,
                                        flags | GFP_THISNODE, nid);
                                if (!obj)