Dear RT Folks,
I'm pleased to announce the 3.6.3-rt7 release.
Changes since 3.6.3-rt6:
* Enable SLUB for RT
Last time I looked at SLUB for RT (some years ago) it was just
way more painful than dealing with SLAB, but Christoph Lameter
has done major surgery on the SLUB code since then and it turns
out that making SLUB usable for RT has become very simple. Thanks
Christoph!
slab.c: 172 insertions(+), 58 deletions(-)
slub.c: 17 insertions(+), 13 deletions(-)
I did some quick comparisons and even a simple hackbench run
shows a significant speedup with SLUB vs. SLAB on RT. I'm not too
surprised as SLUBs fastpath does not have the RT induced
contention problems which we can observe with SLAB.
As usual, give it a good testing and report whatever explodes :)
The delta patch against 3.6.3-rt6 is appended below and can be found
here:
http://www.kernel.org/pub/linux/kernel/projects/rt/3.6/incr/patch-3.6.3-rt6-rt7.patch.xz
The RT patch against 3.6.3 can be found here:
http://www.kernel.org/pub/linux/kernel/projects/rt/3.6/patch-3.6.3-rt7.patch.xz
The split quilt queue is available at:
http://www.kernel.org/pub/linux/kernel/projects/rt/3.6/patches-3.6.3-rt7.tar.xz
Enjoy,
tglx
---------->
Index: linux-stable/init/Kconfig
===================================================================
--- linux-stable.orig/init/Kconfig
+++ linux-stable/init/Kconfig
@@ -1442,7 +1442,6 @@ config SLAB
config SLUB
bool "SLUB (Unqueued Allocator)"
- depends on !PREEMPT_RT_FULL
help
SLUB is a slab allocator that minimizes cache line usage
instead of managing queues of cached objects (SLAB approach).
Index: linux-stable/localversion-rt
===================================================================
--- linux-stable.orig/localversion-rt
+++ linux-stable/localversion-rt
@@ -1 +1 @@
--rt6
+-rt7
Index: linux-stable/mm/slub.c
===================================================================
--- linux-stable.orig/mm/slub.c
+++ linux-stable/mm/slub.c
@@ -31,6 +31,7 @@
#include <linux/fault-inject.h>
#include <linux/stacktrace.h>
#include <linux/prefetch.h>
+#include <linux/locallock.h>
#include <trace/events/kmem.h>
@@ -225,6 +226,8 @@ static inline void stat(const struct kme
#endif
}
+static DEFINE_LOCAL_IRQ_LOCK(slub_lock);
+
/********************************************************************
* Core slab cache functions
*******************************************************************/
@@ -1278,7 +1281,7 @@ static struct page *allocate_slab(struct
flags &= gfp_allowed_mask;
if (flags & __GFP_WAIT)
- local_irq_enable();
+ local_unlock_irq(slub_lock);
flags |= s->allocflags;
@@ -1318,7 +1321,7 @@ static struct page *allocate_slab(struct
}
if (flags & __GFP_WAIT)
- local_irq_disable();
+ local_lock_irq(slub_lock);
if (!page)
return NULL;
@@ -1959,9 +1962,9 @@ int put_cpu_partial(struct kmem_cache *s
* partial array is full. Move the existing
* set to the per node partial list.
*/
- local_irq_save(flags);
+ local_lock_irqsave(slub_lock, flags);
unfreeze_partials(s);
- local_irq_restore(flags);
+ local_unlock_irqrestore(slub_lock, flags);
pobjects = 0;
pages = 0;
stat(s, CPU_PARTIAL_DRAIN);
@@ -2201,7 +2204,7 @@ static void *__slab_alloc(struct kmem_ca
struct page *page;
unsigned long flags;
- local_irq_save(flags);
+ local_lock_irqsave(slub_lock, flags);
#ifdef CONFIG_PREEMPT
/*
* We may have been preempted and rescheduled on a different
@@ -2262,7 +2265,7 @@ load_freelist:
VM_BUG_ON(!c->page->frozen);
c->freelist = get_freepointer(s, freelist);
c->tid = next_tid(c->tid);
- local_irq_restore(flags);
+ local_unlock_irqrestore(slub_lock, flags);
return freelist;
new_slab:
@@ -2281,7 +2284,7 @@ new_slab:
if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit())
slab_out_of_memory(s, gfpflags, node);
- local_irq_restore(flags);
+ local_unlock_irqrestore(slub_lock, flags);
return NULL;
}
@@ -2296,7 +2299,7 @@ new_slab:
deactivate_slab(s, page, get_freepointer(s, freelist));
c->page = NULL;
c->freelist = NULL;
- local_irq_restore(flags);
+ local_unlock_irqrestore(slub_lock, flags);
return freelist;
}
@@ -2488,7 +2491,8 @@ static void __slab_free(struct kmem_cach
* Otherwise the list_lock will synchronize with
* other processors updating the list of slabs.
*/
- spin_lock_irqsave(&n->list_lock, flags);
+ local_spin_lock_irqsave(slub_lock,
+ &n->list_lock, flags);
}
}
@@ -2538,7 +2542,7 @@ static void __slab_free(struct kmem_cach
stat(s, FREE_ADD_PARTIAL);
}
}
- spin_unlock_irqrestore(&n->list_lock, flags);
+ local_spin_unlock_irqrestore(slub_lock, &n->list_lock, flags);
return;
slab_empty:
@@ -2552,7 +2556,7 @@ slab_empty:
/* Slab must be on the full list */
remove_full(s, page);
- spin_unlock_irqrestore(&n->list_lock, flags);
+ local_spin_unlock_irqrestore(slub_lock, &n->list_lock, flags);
stat(s, FREE_SLAB);
discard_slab(s, page);
}
@@ -4002,9 +4006,9 @@ static int __cpuinit slab_cpuup_callback
case CPU_DEAD_FROZEN:
mutex_lock(&slab_mutex);
list_for_each_entry(s, &slab_caches, list) {
- local_irq_save(flags);
+ local_lock_irqsave(slub_lock, flags);
__flush_cpu_slab(s, cpu);
- local_irq_restore(flags);
+ local_unlock_irqrestore(slub_lock, flags);
}
mutex_unlock(&slab_mutex);
break;
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
