SLUB: simplify IRQ off handling
authorChristoph Lameter <clameter@sgi.com>
Wed, 17 Oct 2007 06:25:51 +0000 (23:25 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Wed, 17 Oct 2007 15:42:45 +0000 (08:42 -0700)
Move irq handling out of new slab into __slab_alloc.  That is useful for
Mathieu's cmpxchg_local patchset and also allows us to remove the crude
local_irq_off in early_kmem_cache_alloc().

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/slub.c

index f426f9b..0f862fb 100644 (file)
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -1085,9 +1085,6 @@ static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
 
        BUG_ON(flags & GFP_SLAB_BUG_MASK);
 
-       if (flags & __GFP_WAIT)
-               local_irq_enable();
-
        page = allocate_slab(s,
                flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
        if (!page)
@@ -1120,8 +1117,6 @@ static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
        page->freelist = start;
        page->inuse = 0;
 out:
-       if (flags & __GFP_WAIT)
-               local_irq_disable();
        return page;
 }
 
@@ -1505,7 +1500,14 @@ new_slab:
                goto load_freelist;
        }
 
+       if (gfpflags & __GFP_WAIT)
+               local_irq_enable();
+
        new = new_slab(s, gfpflags, node);
+
+       if (gfpflags & __GFP_WAIT)
+               local_irq_disable();
+
        if (new) {
                c = get_cpu_slab(s, smp_processor_id());
                if (c->page) {
@@ -2039,12 +2041,6 @@ static struct kmem_cache_node *early_kmem_cache_node_alloc(gfp_t gfpflags,
        init_kmem_cache_node(n);
        atomic_long_inc(&n->nr_slabs);
        add_partial(n, page);
-
-       /*
-        * new_slab() disables interupts. If we do not reenable interrupts here
-        * then bootup would continue with interrupts disabled.
-        */
-       local_irq_enable();
        return n;
 }