While making slub available for RT I noticed, that during CPU offline
for each kmem_cache __flush_cpu_slab() is called on a live CPU. This
correctly flushs the cpu_slab of the dead CPU via flush_slab. Though
unfreeze_partials which is called from __flush_cpu_slab() after that
looks at the cpu_slab of the cpu on which this is called. So we fail
to look at the partials of the dead cpu.
Correct this by extending the arguments of unfreeze_partials with the
target cpu number and use per_cpu_ptr instead of this_cpu_ptr.
Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
---
mm/slub.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
Index: linux-2.6/mm/slub.c
===================================================================
--- linux-2.6.orig/mm/slub.c
+++ linux-2.6/mm/slub.c
@@ -1874,10 +1874,10 @@ redo:
*
* This function must be called with interrupt disabled.
*/
-static void unfreeze_partials(struct kmem_cache *s)
+static void unfreeze_partials(struct kmem_cache *s, unsigned int cpu)
{
struct kmem_cache_node *n = NULL, *n2 = NULL;
- struct kmem_cache_cpu *c = this_cpu_ptr(s->cpu_slab);
+ struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
struct page *page, *discard_page = NULL;
while ((page = c->partial)) {
@@ -1963,7 +1963,7 @@ static int put_cpu_partial(struct kmem_c
* set to the per node partial list.
*/
local_irq_save(flags);
- unfreeze_partials(s);
+ unfreeze_partials(s, smp_processor_id());
local_irq_restore(flags);
oldpage = NULL;
pobjects = 0;
@@ -2006,7 +2006,7 @@ static inline void __flush_cpu_slab(stru
if (c->page)
flush_slab(s, c);
- unfreeze_partials(s);
+ unfreeze_partials(s, cpu);
}
}
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>