[PATCH] sched: reduce sched_domain_topology space

[PATCH] sched: reduce sched_domain_topology space

[liguang]

sched_domain_topology space was allocated by the size
ARRAY_SIZE(default_topology) + sched_domains_numa_levels,
suppose CONFIG_SCHED_SMT, CONFIG_SCHED_MC were y,
"static struct sched_domain_topology_level default_topology[] = {
         { sd_init_SIBLING, cpu_smt_mask, },
         { sd_init_MC, cpu_coregroup_mask, },
         { sd_init_CPU, cpu_cpu_mask, },
         { NULL, },
};"
then, ARRAY_SIZE(default_topology) equals to 4,
default_topology's NULL entry was accounted,
but, the following allocation for other sched_domains_numa_levels
entries was startd at 3(for default_topology[3].init is NULL),
so, 1 entry was forgot.

Signed-off-by: liguang <lig.fnst@cn.fujitsu.com>
---
 kernel/sched/core.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 257002c..83f947e 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -6241,7 +6241,7 @@ static void sched_init_numa(void)
 		}
 	}
 
-	tl = kzalloc((ARRAY_SIZE(default_topology) + level) *
+	tl = kzalloc((ARRAY_SIZE(default_topology) + level - 1 ) *
 			sizeof(struct sched_domain_topology_level), GFP_KERNEL);
 	if (!tl)
 		return;
-- 
1.7.2.5

[PATCH] sched: reduce sched_domain_topology space

[liguang]

sched_domain_topology space was allocated by the size
ARRAY_SIZE(default_topology) + sched_domains_numa_levels,
suppose CONFIG_SCHED_SMT, CONFIG_SCHED_MC were y,
"static struct sched_domain_topology_level default_topology[] = {
         { sd_init_SIBLING, cpu_smt_mask, },
         { sd_init_MC, cpu_coregroup_mask, },
         { sd_init_CPU, cpu_cpu_mask, },
         { NULL, },
};"
then, ARRAY_SIZE(default_topology) equals to 4,
default_topology's NULL entry was accounted,
but, the following allocation for other sched_domains_numa_levels
entries was startd at 3(for default_topology[3].init is NULL),
so, 1 entry was forgot.

Signed-off-by: liguang <lig.fnst@cn.fujitsu.com>
---
 kernel/sched/core.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 257002c..6ba5ecf 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -6241,7 +6241,7 @@ static void sched_init_numa(void)
 		}
 	}
 
-	tl = kzalloc((ARRAY_SIZE(default_topology) + level) *
+	tl = kzalloc((ARRAY_SIZE(default_topology) + level - 1) *
 			sizeof(struct sched_domain_topology_level), GFP_KERNEL);
 	if (!tl)
 		return;
-- 
1.7.2.5

Re: [PATCH] sched: reduce sched_domain_topology space

[Namhyung Kim]

Hi, liguang

On Tue, 15 Jan 2013 17:15:18 +0800, liguang wrote:
> sched_domain_topology space was allocated by the size
> ARRAY_SIZE(default_topology) + sched_domains_numa_levels,
> suppose CONFIG_SCHED_SMT, CONFIG_SCHED_MC were y,
> "static struct sched_domain_topology_level default_topology[] = {
>          { sd_init_SIBLING, cpu_smt_mask, },
>          { sd_init_MC, cpu_coregroup_mask, },
>          { sd_init_CPU, cpu_cpu_mask, },
>          { NULL, },
> };"
> then, ARRAY_SIZE(default_topology) equals to 4,
> default_topology's NULL entry was accounted,
> but, the following allocation for other sched_domains_numa_levels
> entries was startd at 3(for default_topology[3].init is NULL),
> so, 1 entry was forgot.

I don't think it's forgotten.

The sched_domain_topology still needs NULL entry at the end.  So what
the current code does is to set up following sched_domain_topology:

{ default_topology (without last NULL entry), numa topology, NULL }

Thanks,
Namhyung

Re: [PATCH] sched: reduce sched_domain_topology space

[li guang]

在 2013-01-16三的 10:51 +0900，Namhyung Kim写道：
> Hi, liguang
> 
> On Tue, 15 Jan 2013 17:15:18 +0800, liguang wrote:
> > sched_domain_topology space was allocated by the size
> > ARRAY_SIZE(default_topology) + sched_domains_numa_levels,
> > suppose CONFIG_SCHED_SMT, CONFIG_SCHED_MC were y,
> > "static struct sched_domain_topology_level default_topology[] = {
> >          { sd_init_SIBLING, cpu_smt_mask, },
> >          { sd_init_MC, cpu_coregroup_mask, },
> >          { sd_init_CPU, cpu_cpu_mask, },
> >          { NULL, },
> > };"
> > then, ARRAY_SIZE(default_topology) equals to 4,
> > default_topology's NULL entry was accounted,
> > but, the following allocation for other sched_domains_numa_levels
> > entries was startd at 3(for default_topology[3].init is NULL),
> > so, 1 entry was forgot.
> 
> I don't think it's forgotten.
> 
> The sched_domain_topology still needs NULL entry at the end.  So what
> the current code does is to set up following sched_domain_topology:
> 
> { default_topology (without last NULL entry), numa topology, NULL }
> 
> Thanks,
> Namhyung

Oh, that's a trick!
Thanks!

-- 
regards!
li guang