Messages in this thread Patch in this message | | | Subject | Re: [PATCH] [10/23] SCHED: Only allocate per cpu cpu mask buffer with offstack cpumasks | From | Andi Kleen <> | Date | Thu, 10 Jun 2010 17:06:18 +0200 |
| |
Peter Zijlstra <peterz@infradead.org> writes:
> On Thu, 2010-06-10 at 16:52 +0200, Andi Kleen wrote: >> > This patch will actually break things.. please read the code. >> > >> > I guess we could move the unsigned long into that block, but I really >> > don't see the point. >> >> How does it break things? >> >> ptr is not used for anything unless that define is set. gcc doesn't >> lie on this.
Ok.
For the !FAIR_GROUP_SCHED || !CPUMASK_OFFSTACK case it's still wasted memory because it is not used for anything. If you enable the kmemleak tracer you should also get a warning about this
I updated the patch with a better ifdef to have CONFIG_UNFAIR_GROUP_SCHED too.
-Andi
---
SCHED: Only allocate per cpu cpu mask buffer with offstack cpumasks v2
This will save a few bytes in the non offstack cpumask case.
Found by gcc 4.6's new warnings.
Cc: peterz@infradead.org Cc: mingo@elte.hu Signed-off-by: Andi Kleen <ak@linux.intel.com>
--- kernel/sched.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
Index: linux-2.6.35-rc2-gcc/kernel/sched.c =================================================================== --- linux-2.6.35-rc2-gcc.orig/kernel/sched.c +++ linux-2.6.35-rc2-gcc/kernel/sched.c @@ -7482,7 +7482,7 @@ static void init_tg_rt_entry(struct task void __init sched_init(void) { int i, j; - unsigned long alloc_size = 0, ptr; + unsigned long alloc_size = 0; #ifdef CONFIG_FAIR_GROUP_SCHED alloc_size += 2 * nr_cpu_ids * sizeof(void **); @@ -7494,7 +7494,10 @@ void __init sched_init(void) alloc_size += num_possible_cpus() * cpumask_size(); #endif if (alloc_size) { +#if defined(CONFIG_CPUMASK_OFFSTACK) || defined(CONFIG_FAIR_GROUP_SCHED) + unsigned long ptr; ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); +#endif #ifdef CONFIG_FAIR_GROUP_SCHED init_task_group.se = (struct sched_entity **)ptr;
-- ak@linux.intel.com -- Speaking for myself only.
| |