Revert "SLUB: Alternate fast paths using cmpxchg_local"
This reverts commit 1f84260c8ce3b1ce26d4c1d6dedc2f33a3a29c0c, which is
suspected to be the reason for some very occasional and hard-to-trigger
crashes that usually look related to memory allocation (mostly reported
in networking, but since that's generally the most common source of
shortlived allocations - and allocations in interrupt contexts - that in
itself is not a big clue).
See for example
http://bugzilla.kernel.org/show_bug.cgi?id=9973
http://lkml.org/lkml/2008/2/19/278
etc.
One promising suspicion for what the root cause of bug is (which also
explains why it's so hard to trigger in practice) came from Eric
Dumazet:
"I wonder how SLUB_FASTPATH is supposed to work, since it is affected
by a classical ABA problem of lockless algo.
cmpxchg_local(&c->freelist, object, object[c->offset]) can succeed,
while an interrupt came (on this cpu), and several allocations were
done, and one free was performed at the end of this interruption, so
'object' was recycled.
c->freelist can then contain the previous value (object), but
object[c->offset] was changed by IRQ.
We then put back in freelist an already allocated object."
but another reason for the revert is simply that everybody agrees that
this code was the main suspect just by virtue of the pattern of oopses.
Cc: Torsten Kaiser <just.for.lkml@googlemail.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
diff --git a/mm/slub.c b/mm/slub.c
index 4b3895c..74c65af 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -149,13 +149,6 @@
/* Enable to test recovery from slab corruption on boot */
#undef SLUB_RESILIENCY_TEST
-/*
- * Currently fastpath is not supported if preemption is enabled.
- */
-#if defined(CONFIG_FAST_CMPXCHG_LOCAL) && !defined(CONFIG_PREEMPT)
-#define SLUB_FASTPATH
-#endif
-
#if PAGE_SHIFT <= 12
/*
@@ -1514,11 +1507,7 @@
{
void **object;
struct page *new;
-#ifdef SLUB_FASTPATH
- unsigned long flags;
- local_irq_save(flags);
-#endif
if (!c->page)
goto new_slab;
@@ -1541,9 +1530,6 @@
unlock_out:
slab_unlock(c->page);
stat(c, ALLOC_SLOWPATH);
-#ifdef SLUB_FASTPATH
- local_irq_restore(flags);
-#endif
return object;
another_slab:
@@ -1575,9 +1561,7 @@
c->page = new;
goto load_freelist;
}
-#ifdef SLUB_FASTPATH
- local_irq_restore(flags);
-#endif
+
/*
* No memory available.
*
@@ -1619,34 +1603,6 @@
{
void **object;
struct kmem_cache_cpu *c;
-
-/*
- * The SLUB_FASTPATH path is provisional and is currently disabled if the
- * kernel is compiled with preemption or if the arch does not support
- * fast cmpxchg operations. There are a couple of coming changes that will
- * simplify matters and allow preemption. Ultimately we may end up making
- * SLUB_FASTPATH the default.
- *
- * 1. The introduction of the per cpu allocator will avoid array lookups
- * through get_cpu_slab(). A special register can be used instead.
- *
- * 2. The introduction of per cpu atomic operations (cpu_ops) means that
- * we can realize the logic here entirely with per cpu atomics. The
- * per cpu atomic ops will take care of the preemption issues.
- */
-
-#ifdef SLUB_FASTPATH
- c = get_cpu_slab(s, raw_smp_processor_id());
- do {
- object = c->freelist;
- if (unlikely(is_end(object) || !node_match(c, node))) {
- object = __slab_alloc(s, gfpflags, node, addr, c);
- break;
- }
- stat(c, ALLOC_FASTPATH);
- } while (cmpxchg_local(&c->freelist, object, object[c->offset])
- != object);
-#else
unsigned long flags;
local_irq_save(flags);
@@ -1661,7 +1617,6 @@
stat(c, ALLOC_FASTPATH);
}
local_irq_restore(flags);
-#endif
if (unlikely((gfpflags & __GFP_ZERO) && object))
memset(object, 0, c->objsize);
@@ -1698,11 +1653,6 @@
void **object = (void *)x;
struct kmem_cache_cpu *c;
-#ifdef SLUB_FASTPATH
- unsigned long flags;
-
- local_irq_save(flags);
-#endif
c = get_cpu_slab(s, raw_smp_processor_id());
stat(c, FREE_SLOWPATH);
slab_lock(page);
@@ -1734,9 +1684,6 @@
out_unlock:
slab_unlock(page);
-#ifdef SLUB_FASTPATH
- local_irq_restore(flags);
-#endif
return;
slab_empty:
@@ -1749,9 +1696,6 @@
}
slab_unlock(page);
stat(c, FREE_SLAB);
-#ifdef SLUB_FASTPATH
- local_irq_restore(flags);
-#endif
discard_slab(s, page);
return;
@@ -1777,34 +1721,6 @@
{
void **object = (void *)x;
struct kmem_cache_cpu *c;
-
-#ifdef SLUB_FASTPATH
- void **freelist;
-
- c = get_cpu_slab(s, raw_smp_processor_id());
- debug_check_no_locks_freed(object, s->objsize);
- do {
- freelist = c->freelist;
- barrier();
- /*
- * If the compiler would reorder the retrieval of c->page to
- * come before c->freelist then an interrupt could
- * change the cpu slab before we retrieve c->freelist. We
- * could be matching on a page no longer active and put the
- * object onto the freelist of the wrong slab.
- *
- * On the other hand: If we already have the freelist pointer
- * then any change of cpu_slab will cause the cmpxchg to fail
- * since the freelist pointers are unique per slab.
- */
- if (unlikely(page != c->page || c->node < 0)) {
- __slab_free(s, page, x, addr, c->offset);
- break;
- }
- object[c->offset] = freelist;
- stat(c, FREE_FASTPATH);
- } while (cmpxchg_local(&c->freelist, freelist, object) != freelist);
-#else
unsigned long flags;
local_irq_save(flags);
@@ -1818,7 +1734,6 @@
__slab_free(s, page, x, addr, c->offset);
local_irq_restore(flags);
-#endif
}
void kmem_cache_free(struct kmem_cache *s, void *x)