[RFC][PATCH 00/14] Rewrite sched_domain/sched_group creation

[RFC][PATCH 00/14] Rewrite sched_domain/sched_group creation

[Peter Zijlstra]

So I got annoyed with the whole sched_domain/sched_group creation mess again
and decided to actuall do something about it. See here ;-)

Its not completely done yet, but much of the ground-work is there.

The final goal is to be able to have a simple way to dynamically specify the
architecture topoplogy and have the generic code in charge of building the data
structures. The architecture would need to provide a function that maps a cpu
to a cpumask and a function that initializes the sched_domain (flags etc.),
this would replace the now still hard-coded __build_*_sched_domain() calls in
__build_sched_domains().

With that we can also kill the current NODE and ALLNODES mess and generate
appropriate masks from say the ACPI SLIT table by grouping CPUs on their node
distance.

This would allow us to kill the horrid mess in x86's cpu_coregroup_mask() and
properly support the AMD magnycours stuff.

Anyway, not quite there yet.. patches build and boot on a 2*6*2 wsm box.

---
 include/linux/sched.h |   19 +-
 kernel/sched.c        |  817 +++++++++++++------------------------------------
 kernel/sched_fair.c   |   32 ++-
 3 files changed, 258 insertions(+), 610 deletions(-)

[RFC][PATCH 01/14] sched: Remove obsolete arch_ prefixes

[Peter Zijlstra]

[-- Attachment #1: sched-remove-arch-prefixes.patch --]
[-- Type: text/plain, Size: 2392 bytes --]

Non weak static functions clearly are not arch specific, so remove the
arch_ prefix.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -232,7 +232,7 @@ static void destroy_rt_bandwidth(struct
 #endif
 
 /*
- * sched_domains_mutex serializes calls to arch_init_sched_domains,
+ * sched_domains_mutex serializes calls to init_sched_domains,
  * detach_destroy_domains and partition_sched_domains.
  */
 static DEFINE_MUTEX(sched_domains_mutex);
@@ -7646,7 +7646,7 @@ void free_sched_domains(cpumask_var_t do
  * For now this just excludes isolated cpus, but could be used to
  * exclude other special cases in the future.
  */
-static int arch_init_sched_domains(const struct cpumask *cpu_map)
+static int init_sched_domains(const struct cpumask *cpu_map)
 {
 	int err;
 
@@ -7663,7 +7663,7 @@ static int arch_init_sched_domains(const
 	return err;
 }
 
-static void arch_destroy_sched_domains(const struct cpumask *cpu_map,
+static void destroy_sched_domains(const struct cpumask *cpu_map,
 				       struct cpumask *tmpmask)
 {
 	free_sched_groups(cpu_map, tmpmask);
@@ -7682,7 +7682,7 @@ static void detach_destroy_domains(const
 	for_each_cpu(i, cpu_map)
 		cpu_attach_domain(NULL, &def_root_domain, i);
 	synchronize_sched();
-	arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask));
+	destroy_sched_domains(cpu_map, to_cpumask(tmpmask));
 }
 
 /* handle null as "default" */
@@ -7791,7 +7791,7 @@ void partition_sched_domains(int ndoms_n
 }
 
 #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
-static void arch_reinit_sched_domains(void)
+static void reinit_sched_domains(void)
 {
 	get_online_cpus();
 
@@ -7824,7 +7824,7 @@ static ssize_t sched_power_savings_store
 	else
 		sched_mc_power_savings = level;
 
-	arch_reinit_sched_domains();
+	reinit_sched_domains();
 
 	return count;
 }
@@ -7950,7 +7950,7 @@ void __init sched_init_smp(void)
 #endif
 	get_online_cpus();
 	mutex_lock(&sched_domains_mutex);
-	arch_init_sched_domains(cpu_active_mask);
+	init_sched_domains(cpu_active_mask);
 	cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);
 	if (cpumask_empty(non_isolated_cpus))
 		cpumask_set_cpu(smp_processor_id(), non_isolated_cpus);

[RFC][PATCH 02/14] sched: Simplify cpu_power initialization

[Peter Zijlstra]

[-- Attachment #1: sched-simplify-cpu_power.patch --]
[-- Type: text/plain, Size: 2778 bytes --]

The code in update_group_power() does what init_sched_groups_power()
does and more, so remove the special init_ code and call the generic
code instead.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 kernel/sched.c |   44 +++++---------------------------------------
 1 file changed, 5 insertions(+), 39 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -6655,9 +6655,6 @@ cpu_attach_domain(struct sched_domain *s
 	struct rq *rq = cpu_rq(cpu);
 	struct sched_domain *tmp;
 
-	for (tmp = sd; tmp; tmp = tmp->parent)
-		tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
-
 	/* Remove the sched domains which do not contribute to scheduling. */
 	for (tmp = sd; tmp; ) {
 		struct sched_domain *parent = tmp->parent;
@@ -7135,11 +7132,6 @@ static void free_sched_groups(const stru
  */
 static void init_sched_groups_power(int cpu, struct sched_domain *sd)
 {
-	struct sched_domain *child;
-	struct sched_group *group;
-	long power;
-	int weight;
-
 	WARN_ON(!sd || !sd->groups);
 
 	if (cpu != group_first_cpu(sd->groups))
@@ -7147,36 +7139,7 @@ static void init_sched_groups_power(int
 
 	sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups));
 
-	child = sd->child;
-
-	sd->groups->cpu_power = 0;
-
-	if (!child) {
-		power = SCHED_LOAD_SCALE;
-		weight = cpumask_weight(sched_domain_span(sd));
-		/*
-		 * SMT siblings share the power of a single core.
-		 * Usually multiple threads get a better yield out of
-		 * that one core than a single thread would have,
-		 * reflect that in sd->smt_gain.
-		 */
-		if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) {
-			power *= sd->smt_gain;
-			power /= weight;
-			power >>= SCHED_LOAD_SHIFT;
-		}
-		sd->groups->cpu_power += power;
-		return;
-	}
-
-	/*
-	 * Add cpu_power of each child group to this groups cpu_power.
-	 */
-	group = child->groups;
-	do {
-		sd->groups->cpu_power += group->cpu_power;
-		group = group->next;
-	} while (group != child->groups);
+	update_group_power(sd, cpu);
 }
 
 /*
@@ -7483,7 +7446,7 @@ static int __build_sched_domains(const s
 {
 	enum s_alloc alloc_state = sa_none;
 	struct s_data d;
-	struct sched_domain *sd;
+	struct sched_domain *sd, *tmp;
 	int i;
 #ifdef CONFIG_NUMA
 	d.sd_allnodes = 0;
@@ -7506,6 +7469,9 @@ static int __build_sched_domains(const s
 		sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i);
 		sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
 		sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
+
+		for (tmp = sd; tmp; tmp = tmp->parent)
+			tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
 	}
 
 	for_each_cpu(i, cpu_map) {

Re: [RFC][PATCH 02/14] sched: Simplify cpu_power initialization

[Steven Rostedt]

On Mon, Mar 14, 2011 at 04:06:15PM +0100, Peter Zijlstra wrote:
> The code in update_group_power() does what init_sched_groups_power()
> does and more, so remove the special init_ code and call the generic
> code instead.
> 
> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
> LKML-Reference: <new-submission>
> ---
>  kernel/sched.c |   44 +++++---------------------------------------
>  1 file changed, 5 insertions(+), 39 deletions(-)
> 
> Index: linux-2.6/kernel/sched.c
> ===================================================================
> --- linux-2.6.orig/kernel/sched.c
> +++ linux-2.6/kernel/sched.c
> @@ -6655,9 +6655,6 @@ cpu_attach_domain(struct sched_domain *s
>  	struct rq *rq = cpu_rq(cpu);
>  	struct sched_domain *tmp;
>  
> -	for (tmp = sd; tmp; tmp = tmp->parent)
> -		tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
> -
>  	/* Remove the sched domains which do not contribute to scheduling. */
>  	for (tmp = sd; tmp; ) {
>  		struct sched_domain *parent = tmp->parent;

This and ...

[ snip what was explained in change log ]

>  
>  /*
> @@ -7483,7 +7446,7 @@ static int __build_sched_domains(const s
>  {
>  	enum s_alloc alloc_state = sa_none;
>  	struct s_data d;
> -	struct sched_domain *sd;
> +	struct sched_domain *sd, *tmp;
>  	int i;
>  #ifdef CONFIG_NUMA
>  	d.sd_allnodes = 0;
> @@ -7506,6 +7469,9 @@ static int __build_sched_domains(const s
>  		sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i);
>  		sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
>  		sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
> +
> +		for (tmp = sd; tmp; tmp = tmp->parent)
> +			tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
>  	}
>  
>  	for_each_cpu(i, cpu_map) {
> 

this, looks like a separate change than what was explained in the change
log. Did you forget a "quilt new" between these two changes?

-- Steve

Re: [RFC][PATCH 02/14] sched: Simplify cpu_power initialization

[Peter Zijlstra]

On Mon, 2011-03-14 at 17:35 -0400, Steven Rostedt wrote:
> On Mon, Mar 14, 2011 at 04:06:15PM +0100, Peter Zijlstra wrote:
> > The code in update_group_power() does what init_sched_groups_power()
> > does and more, so remove the special init_ code and call the generic
> > code instead.
> > 
> > Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
> > LKML-Reference: <new-submission>
> > ---
> >  kernel/sched.c |   44 +++++---------------------------------------
> >  1 file changed, 5 insertions(+), 39 deletions(-)
> > 
> > Index: linux-2.6/kernel/sched.c
> > ===================================================================
> > --- linux-2.6.orig/kernel/sched.c
> > +++ linux-2.6/kernel/sched.c
> > @@ -6655,9 +6655,6 @@ cpu_attach_domain(struct sched_domain *s
> >  	struct rq *rq = cpu_rq(cpu);
> >  	struct sched_domain *tmp;
> >  
> > -	for (tmp = sd; tmp; tmp = tmp->parent)
> > -		tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
> > -
> >  	/* Remove the sched domains which do not contribute to scheduling. */
> >  	for (tmp = sd; tmp; ) {
> >  		struct sched_domain *parent = tmp->parent;
> 
> This and ...
> 
> [ snip what was explained in change log ]
> 
> >  
> >  /*
> > @@ -7483,7 +7446,7 @@ static int __build_sched_domains(const s
> >  {
> >  	enum s_alloc alloc_state = sa_none;
> >  	struct s_data d;
> > -	struct sched_domain *sd;
> > +	struct sched_domain *sd, *tmp;
> >  	int i;
> >  #ifdef CONFIG_NUMA
> >  	d.sd_allnodes = 0;
> > @@ -7506,6 +7469,9 @@ static int __build_sched_domains(const s
> >  		sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i);
> >  		sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
> >  		sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
> > +
> > +		for (tmp = sd; tmp; tmp = tmp->parent)
> > +			tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
> >  	}
> >  
> >  	for_each_cpu(i, cpu_map) {
> > 
> 
> this, looks like a separate change than what was explained in the change
> log. Did you forget a "quilt new" between these two changes?

Ah, no, I simply wrote the changelog several days and a weekend after
the patch and completely forgot about it, its needed because
update_group_power() needs to have sched_domain::span_weight set up, so
it needs to be done sooner.

I guess I can split it into a separate patch before the other bits
though.

[RFC][PATCH 03/14] sched: Simplify build_sched_groups

[Peter Zijlstra]

[-- Attachment #1: sched-foo3.patch --]
[-- Type: text/plain, Size: 4741 bytes --]

Notice that the mask being computed is the same as the domain span we
just computed. By using the domain_span we can avoid some mask
allocations and computations.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 kernel/sched.c |   52 ++++++++++++++++------------------------------------
 1 file changed, 16 insertions(+), 36 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -6842,9 +6842,6 @@ struct s_data {
 	cpumask_var_t		notcovered;
 #endif
 	cpumask_var_t		nodemask;
-	cpumask_var_t		this_sibling_map;
-	cpumask_var_t		this_core_map;
-	cpumask_var_t		this_book_map;
 	cpumask_var_t		send_covered;
 	cpumask_var_t		tmpmask;
 	struct sched_group	**sched_group_nodes;
@@ -6856,9 +6853,6 @@ enum s_alloc {
 	sa_rootdomain,
 	sa_tmpmask,
 	sa_send_covered,
-	sa_this_book_map,
-	sa_this_core_map,
-	sa_this_sibling_map,
 	sa_nodemask,
 	sa_sched_group_nodes,
 #ifdef CONFIG_NUMA
@@ -7201,12 +7195,6 @@ static void __free_domain_allocs(struct
 		free_cpumask_var(d->tmpmask); /* fall through */
 	case sa_send_covered:
 		free_cpumask_var(d->send_covered); /* fall through */
-	case sa_this_book_map:
-		free_cpumask_var(d->this_book_map); /* fall through */
-	case sa_this_core_map:
-		free_cpumask_var(d->this_core_map); /* fall through */
-	case sa_this_sibling_map:
-		free_cpumask_var(d->this_sibling_map); /* fall through */
 	case sa_nodemask:
 		free_cpumask_var(d->nodemask); /* fall through */
 	case sa_sched_group_nodes:
@@ -7245,14 +7233,8 @@ static enum s_alloc __visit_domain_alloc
 #endif
 	if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL))
 		return sa_sched_group_nodes;
-	if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL))
-		return sa_nodemask;
-	if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL))
-		return sa_this_sibling_map;
-	if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL))
-		return sa_this_core_map;
 	if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
-		return sa_this_book_map;
+		return sa_nodemask;
 	if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
 		return sa_send_covered;
 	d->rd = alloc_rootdomain();
@@ -7364,39 +7346,40 @@ static struct sched_domain *__build_smt_
 static void build_sched_groups(struct s_data *d, enum sched_domain_level l,
 			       const struct cpumask *cpu_map, int cpu)
 {
+	struct sched_domain *sd;
+
 	switch (l) {
 #ifdef CONFIG_SCHED_SMT
 	case SD_LV_SIBLING: /* set up CPU (sibling) groups */
-		cpumask_and(d->this_sibling_map, cpu_map,
-			    topology_thread_cpumask(cpu));
-		if (cpu == cpumask_first(d->this_sibling_map))
-			init_sched_build_groups(d->this_sibling_map, cpu_map,
+		sd = &per_cpu(cpu_domains, cpu).sd;
+		if (cpu == cpumask_first(sched_domain_span(sd)))
+			init_sched_build_groups(sched_domain_span(sd), cpu_map,
 						&cpu_to_cpu_group,
 						d->send_covered, d->tmpmask);
 		break;
 #endif
 #ifdef CONFIG_SCHED_MC
 	case SD_LV_MC: /* set up multi-core groups */
-		cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu));
-		if (cpu == cpumask_first(d->this_core_map))
-			init_sched_build_groups(d->this_core_map, cpu_map,
+		sd = &per_cpu(core_domains, cpu).sd;
+		if (cpu == cpumask_first(sched_domain_span(sd)))
+			init_sched_build_groups(sched_domain_span(sd), cpu_map,
 						&cpu_to_core_group,
 						d->send_covered, d->tmpmask);
 		break;
 #endif
 #ifdef CONFIG_SCHED_BOOK
 	case SD_LV_BOOK: /* set up book groups */
-		cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu));
-		if (cpu == cpumask_first(d->this_book_map))
-			init_sched_build_groups(d->this_book_map, cpu_map,
+		sd = &per_cpu(book_domains, cpu).sd;
+		if (cpu == cpumask_first(sched_domain_span(sd)))
+			init_sched_build_groups(sched_domain_span(sd), cpu_map,
 						&cpu_to_book_group,
 						d->send_covered, d->tmpmask);
 		break;
 #endif
 	case SD_LV_CPU: /* set up physical groups */
-		cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map);
-		if (!cpumask_empty(d->nodemask))
-			init_sched_build_groups(d->nodemask, cpu_map,
+		sd = &per_cpu(phys_domains, cpu).sd;
+		if (cpu == cpumask_first(sched_domain_span(sd)))
+			init_sched_build_groups(sched_domain_span(sd), cpu_map,
 						&cpu_to_phys_group,
 						d->send_covered, d->tmpmask);
 		break;
@@ -7452,11 +7435,8 @@ static int __build_sched_domains(const s
 		build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i);
 		build_sched_groups(&d, SD_LV_BOOK, cpu_map, i);
 		build_sched_groups(&d, SD_LV_MC, cpu_map, i);
-	}
-
-	/* Set up physical groups */
-	for (i = 0; i < nr_node_ids; i++)
 		build_sched_groups(&d, SD_LV_CPU, cpu_map, i);
+	}
 
 #ifdef CONFIG_NUMA
 	/* Set up node groups */

[RFC][PATCH 04/14] sched: Change NODE sched_domain group creation

[Peter Zijlstra]

[-- Attachment #1: sched-foo1.patch --]
[-- Type: text/plain, Size: 11443 bytes --]

The NODE sched_domain is 'special' in that it allocates sched_groups
per CPU, instead of sharing the sched_groups between all CPUs.

While this might have some benefits on large NUMA and avoid remote
memory accesses when iterating the sched_groups, this does break
current code that assumes sched_groups are shared between all
sched_domains (since the dynamic cpu_power patches).

So refactor the NODE groups to behave like all other groups.

(The ALLNODES domain again shared its groups across the CPUs for some
reason).

If someone does measure a performance decrease due to this change we
need to revisit this and come up with another way to have both dynamic
cpu_power and NUMA work nice together.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 kernel/sched.c |  231 ++++++++-------------------------------------------------
 1 file changed, 33 insertions(+), 198 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -6837,29 +6837,18 @@ struct static_sched_domain {
 struct s_data {
 #ifdef CONFIG_NUMA
 	int			sd_allnodes;
-	cpumask_var_t		domainspan;
-	cpumask_var_t		covered;
-	cpumask_var_t		notcovered;
 #endif
 	cpumask_var_t		nodemask;
 	cpumask_var_t		send_covered;
 	cpumask_var_t		tmpmask;
-	struct sched_group	**sched_group_nodes;
 	struct root_domain	*rd;
 };
 
 enum s_alloc {
-	sa_sched_groups = 0,
 	sa_rootdomain,
 	sa_tmpmask,
 	sa_send_covered,
 	sa_nodemask,
-	sa_sched_group_nodes,
-#ifdef CONFIG_NUMA
-	sa_notcovered,
-	sa_covered,
-	sa_domainspan,
-#endif
 	sa_none,
 };
 
@@ -6955,18 +6944,10 @@ cpu_to_phys_group(int cpu, const struct
 }
 
 #ifdef CONFIG_NUMA
-/*
- * The init_sched_build_groups can't handle what we want to do with node
- * groups, so roll our own. Now each node has its own list of groups which
- * gets dynamically allocated.
- */
 static DEFINE_PER_CPU(struct static_sched_domain, node_domains);
-static struct sched_group ***sched_group_nodes_bycpu;
-
-static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains);
-static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes);
+static DEFINE_PER_CPU(struct static_sched_group, sched_group_node);
 
-static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map,
+static int cpu_to_node_group(int cpu, const struct cpumask *cpu_map,
 				 struct sched_group **sg,
 				 struct cpumask *nodemask)
 {
@@ -6976,142 +6957,27 @@ static int cpu_to_allnodes_group(int cpu
 	group = cpumask_first(nodemask);
 
 	if (sg)
-		*sg = &per_cpu(sched_group_allnodes, group).sg;
+		*sg = &per_cpu(sched_group_node, group).sg;
 	return group;
 }
 
-static void init_numa_sched_groups_power(struct sched_group *group_head)
-{
-	struct sched_group *sg = group_head;
-	int j;
-
-	if (!sg)
-		return;
-	do {
-		for_each_cpu(j, sched_group_cpus(sg)) {
-			struct sched_domain *sd;
-
-			sd = &per_cpu(phys_domains, j).sd;
-			if (j != group_first_cpu(sd->groups)) {
-				/*
-				 * Only add "power" once for each
-				 * physical package.
-				 */
-				continue;
-			}
-
-			sg->cpu_power += sd->groups->cpu_power;
-		}
-		sg = sg->next;
-	} while (sg != group_head);
-}
+static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains);
+static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes);
 
-static int build_numa_sched_groups(struct s_data *d,
-				   const struct cpumask *cpu_map, int num)
+static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map,
+				 struct sched_group **sg,
+				 struct cpumask *nodemask)
 {
-	struct sched_domain *sd;
-	struct sched_group *sg, *prev;
-	int n, j;
+	int group;
 
-	cpumask_clear(d->covered);
-	cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map);
-	if (cpumask_empty(d->nodemask)) {
-		d->sched_group_nodes[num] = NULL;
-		goto out;
-	}
-
-	sched_domain_node_span(num, d->domainspan);
-	cpumask_and(d->domainspan, d->domainspan, cpu_map);
-
-	sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
-			  GFP_KERNEL, num);
-	if (!sg) {
-		printk(KERN_WARNING "Can not alloc domain group for node %d\n",
-		       num);
-		return -ENOMEM;
-	}
-	d->sched_group_nodes[num] = sg;
-
-	for_each_cpu(j, d->nodemask) {
-		sd = &per_cpu(node_domains, j).sd;
-		sd->groups = sg;
-	}
-
-	sg->cpu_power = 0;
-	cpumask_copy(sched_group_cpus(sg), d->nodemask);
-	sg->next = sg;
-	cpumask_or(d->covered, d->covered, d->nodemask);
-
-	prev = sg;
-	for (j = 0; j < nr_node_ids; j++) {
-		n = (num + j) % nr_node_ids;
-		cpumask_complement(d->notcovered, d->covered);
-		cpumask_and(d->tmpmask, d->notcovered, cpu_map);
-		cpumask_and(d->tmpmask, d->tmpmask, d->domainspan);
-		if (cpumask_empty(d->tmpmask))
-			break;
-		cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n));
-		if (cpumask_empty(d->tmpmask))
-			continue;
-		sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(),
-				  GFP_KERNEL, num);
-		if (!sg) {
-			printk(KERN_WARNING
-			       "Can not alloc domain group for node %d\n", j);
-			return -ENOMEM;
-		}
-		sg->cpu_power = 0;
-		cpumask_copy(sched_group_cpus(sg), d->tmpmask);
-		sg->next = prev->next;
-		cpumask_or(d->covered, d->covered, d->tmpmask);
-		prev->next = sg;
-		prev = sg;
-	}
-out:
-	return 0;
-}
-#endif /* CONFIG_NUMA */
+	cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map);
+	group = cpumask_first(nodemask);
 
-#ifdef CONFIG_NUMA
-/* Free memory allocated for various sched_group structures */
-static void free_sched_groups(const struct cpumask *cpu_map,
-			      struct cpumask *nodemask)
-{
-	int cpu, i;
-
-	for_each_cpu(cpu, cpu_map) {
-		struct sched_group **sched_group_nodes
-			= sched_group_nodes_bycpu[cpu];
-
-		if (!sched_group_nodes)
-			continue;
-
-		for (i = 0; i < nr_node_ids; i++) {
-			struct sched_group *oldsg, *sg = sched_group_nodes[i];
-
-			cpumask_and(nodemask, cpumask_of_node(i), cpu_map);
-			if (cpumask_empty(nodemask))
-				continue;
-
-			if (sg == NULL)
-				continue;
-			sg = sg->next;
-next_sg:
-			oldsg = sg;
-			sg = sg->next;
-			kfree(oldsg);
-			if (oldsg != sched_group_nodes[i])
-				goto next_sg;
-		}
-		kfree(sched_group_nodes);
-		sched_group_nodes_bycpu[cpu] = NULL;
-	}
-}
-#else /* !CONFIG_NUMA */
-static void free_sched_groups(const struct cpumask *cpu_map,
-			      struct cpumask *nodemask)
-{
+	if (sg)
+		*sg = &per_cpu(sched_group_allnodes, group).sg;
+	return group;
 }
+
 #endif /* CONFIG_NUMA */
 
 /*
@@ -7212,9 +7078,6 @@ static void __free_domain_allocs(struct
 				 const struct cpumask *cpu_map)
 {
 	switch (what) {
-	case sa_sched_groups:
-		free_sched_groups(cpu_map, d->tmpmask); /* fall through */
-		d->sched_group_nodes = NULL;
 	case sa_rootdomain:
 		free_rootdomain(d->rd); /* fall through */
 	case sa_tmpmask:
@@ -7223,16 +7086,6 @@ static void __free_domain_allocs(struct
 		free_cpumask_var(d->send_covered); /* fall through */
 	case sa_nodemask:
 		free_cpumask_var(d->nodemask); /* fall through */
-	case sa_sched_group_nodes:
-#ifdef CONFIG_NUMA
-		kfree(d->sched_group_nodes); /* fall through */
-	case sa_notcovered:
-		free_cpumask_var(d->notcovered); /* fall through */
-	case sa_covered:
-		free_cpumask_var(d->covered); /* fall through */
-	case sa_domainspan:
-		free_cpumask_var(d->domainspan); /* fall through */
-#endif
 	case sa_none:
 		break;
 	}
@@ -7241,24 +7094,8 @@ static void __free_domain_allocs(struct
 static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
 						   const struct cpumask *cpu_map)
 {
-#ifdef CONFIG_NUMA
-	if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL))
-		return sa_none;
-	if (!alloc_cpumask_var(&d->covered, GFP_KERNEL))
-		return sa_domainspan;
-	if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL))
-		return sa_covered;
-	/* Allocate the per-node list of sched groups */
-	d->sched_group_nodes = kcalloc(nr_node_ids,
-				      sizeof(struct sched_group *), GFP_KERNEL);
-	if (!d->sched_group_nodes) {
-		printk(KERN_WARNING "Can not alloc sched group node list\n");
-		return sa_notcovered;
-	}
-	sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes;
-#endif
 	if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL))
-		return sa_sched_group_nodes;
+		return sa_none;
 	if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
 		return sa_nodemask;
 	if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
@@ -7298,6 +7135,7 @@ static struct sched_domain *__build_numa
 	if (parent)
 		parent->child = sd;
 	cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map);
+	cpu_to_node_group(i, cpu_map, &sd->groups, d->tmpmask);
 #endif
 	return sd;
 }
@@ -7410,6 +7248,13 @@ static void build_sched_groups(struct s_
 						d->send_covered, d->tmpmask);
 		break;
 #ifdef CONFIG_NUMA
+	case SD_LV_NODE:
+		sd = &per_cpu(node_domains, cpu).sd;
+		if (cpu == cpumask_first(sched_domain_span(sd)))
+			init_sched_build_groups(sched_domain_span(sd), cpu_map,
+						&cpu_to_node_group,
+						d->send_covered, d->tmpmask);
+
 	case SD_LV_ALLNODES:
 		init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group,
 					d->send_covered, d->tmpmask);
@@ -7438,7 +7283,6 @@ static int __build_sched_domains(const s
 	alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
 	if (alloc_state != sa_rootdomain)
 		goto error;
-	alloc_state = sa_sched_groups;
 
 	/*
 	 * Set up domains for cpus specified by the cpu_map.
@@ -7462,16 +7306,13 @@ static int __build_sched_domains(const s
 		build_sched_groups(&d, SD_LV_BOOK, cpu_map, i);
 		build_sched_groups(&d, SD_LV_MC, cpu_map, i);
 		build_sched_groups(&d, SD_LV_CPU, cpu_map, i);
+		build_sched_groups(&d, SD_LV_NODE, cpu_map, i);
 	}
 
 #ifdef CONFIG_NUMA
 	/* Set up node groups */
 	if (d.sd_allnodes)
 		build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0);
-
-	for (i = 0; i < nr_node_ids; i++)
-		if (build_numa_sched_groups(&d, cpu_map, i))
-			goto error;
 #endif
 
 	/* Calculate CPU power for physical packages and nodes */
@@ -7500,15 +7341,16 @@ static int __build_sched_domains(const s
 	}
 
 #ifdef CONFIG_NUMA
-	for (i = 0; i < nr_node_ids; i++)
-		init_numa_sched_groups_power(d.sched_group_nodes[i]);
+	for_each_cpu(i, cpu_map) {
+		sd = &per_cpu(node_domains, i).sd;
+		init_sched_groups_power(i, sd);
+	}
 
 	if (d.sd_allnodes) {
-		struct sched_group *sg;
-
-		cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg,
-								d.tmpmask);
-		init_numa_sched_groups_power(sg);
+		for_each_cpu(i, cpu_map) {
+			sd = &per_cpu(allnodes_domains, i).sd;
+			init_sched_groups_power(i, sd);
+		}
 	}
 #endif
 
@@ -7526,7 +7368,6 @@ static int __build_sched_domains(const s
 		cpu_attach_domain(sd, d.rd, i);
 	}
 
-	d.sched_group_nodes = NULL; /* don't free this we still need it */
 	__free_domain_allocs(&d, sa_tmpmask, cpu_map);
 	return 0;
 
@@ -7612,7 +7453,6 @@ static int init_sched_domains(const stru
 static void destroy_sched_domains(const struct cpumask *cpu_map,
 				       struct cpumask *tmpmask)
 {
-	free_sched_groups(cpu_map, tmpmask);
 }
 
 /*
@@ -7889,11 +7729,6 @@ void __init sched_init_smp(void)
 	alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL);
 	alloc_cpumask_var(&fallback_doms, GFP_KERNEL);
 
-#if defined(CONFIG_NUMA)
-	sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **),
-								GFP_KERNEL);
-	BUG_ON(sched_group_nodes_bycpu == NULL);
-#endif
 	get_online_cpus();
 	mutex_lock(&sched_domains_mutex);
 	init_sched_domains(cpu_active_mask);

[RFC][PATCH 05/14] sched: Clean up some ALLNODES code

[Peter Zijlstra]

[-- Attachment #1: sched-foo4.patch --]
[-- Type: text/plain, Size: 1227 bytes --]


Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 kernel/sched.c |   11 ++++-------
 1 file changed, 4 insertions(+), 7 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -7256,7 +7256,9 @@ static void build_sched_groups(struct s_
 						d->send_covered, d->tmpmask);
 
 	case SD_LV_ALLNODES:
-		init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group,
+		if (cpu == cpumask_first(cpu_map))
+			init_sched_build_groups(cpu_map, cpu_map,
+					&cpu_to_allnodes_group,
 					d->send_covered, d->tmpmask);
 		break;
 #endif
@@ -7307,14 +7309,9 @@ static int __build_sched_domains(const s
 		build_sched_groups(&d, SD_LV_MC, cpu_map, i);
 		build_sched_groups(&d, SD_LV_CPU, cpu_map, i);
 		build_sched_groups(&d, SD_LV_NODE, cpu_map, i);
+		build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, i);
 	}
 
-#ifdef CONFIG_NUMA
-	/* Set up node groups */
-	if (d.sd_allnodes)
-		build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0);
-#endif
-
 	/* Calculate CPU power for physical packages and nodes */
 #ifdef CONFIG_SCHED_SMT
 	for_each_cpu(i, cpu_map) {

[RFC][PATCH 06/14] Simplify group creation

[Peter Zijlstra]

[-- Attachment #1: sched-foo5.patch --]
[-- Type: text/plain, Size: 3198 bytes --]

Instead of calling build_sched_groups() for each possible sched_domain
we might have created, note that we can simply iterate the
sched_domain tree and call it for each sched_domain present.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 kernel/sched.c |   24 +++++-------------------
 1 file changed, 5 insertions(+), 19 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -7207,15 +7207,12 @@ static struct sched_domain *__build_smt_
 	return sd;
 }
 
-static void build_sched_groups(struct s_data *d, enum sched_domain_level l,
+static void build_sched_groups(struct s_data *d, struct sched_domain *sd,
 			       const struct cpumask *cpu_map, int cpu)
 {
-	struct sched_domain *sd;
-
-	switch (l) {
+	switch (sd->level) {
 #ifdef CONFIG_SCHED_SMT
 	case SD_LV_SIBLING: /* set up CPU (sibling) groups */
-		sd = &per_cpu(cpu_domains, cpu).sd;
 		if (cpu == cpumask_first(sched_domain_span(sd)))
 			init_sched_build_groups(sched_domain_span(sd), cpu_map,
 						&cpu_to_cpu_group,
@@ -7224,7 +7221,6 @@ static void build_sched_groups(struct s_
 #endif
 #ifdef CONFIG_SCHED_MC
 	case SD_LV_MC: /* set up multi-core groups */
-		sd = &per_cpu(core_domains, cpu).sd;
 		if (cpu == cpumask_first(sched_domain_span(sd)))
 			init_sched_build_groups(sched_domain_span(sd), cpu_map,
 						&cpu_to_core_group,
@@ -7233,7 +7229,6 @@ static void build_sched_groups(struct s_
 #endif
 #ifdef CONFIG_SCHED_BOOK
 	case SD_LV_BOOK: /* set up book groups */
-		sd = &per_cpu(book_domains, cpu).sd;
 		if (cpu == cpumask_first(sched_domain_span(sd)))
 			init_sched_build_groups(sched_domain_span(sd), cpu_map,
 						&cpu_to_book_group,
@@ -7241,7 +7236,6 @@ static void build_sched_groups(struct s_
 		break;
 #endif
 	case SD_LV_CPU: /* set up physical groups */
-		sd = &per_cpu(phys_domains, cpu).sd;
 		if (cpu == cpumask_first(sched_domain_span(sd)))
 			init_sched_build_groups(sched_domain_span(sd), cpu_map,
 						&cpu_to_phys_group,
@@ -7249,7 +7243,6 @@ static void build_sched_groups(struct s_
 		break;
 #ifdef CONFIG_NUMA
 	case SD_LV_NODE:
-		sd = &per_cpu(node_domains, cpu).sd;
 		if (cpu == cpumask_first(sched_domain_span(sd)))
 			init_sched_build_groups(sched_domain_span(sd), cpu_map,
 						&cpu_to_node_group,
@@ -7299,17 +7292,10 @@ static int __build_sched_domains(const s
 		sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
 		sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
 
-		for (tmp = sd; tmp; tmp = tmp->parent)
+		for (tmp = sd; tmp; tmp = tmp->parent) {
 			tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
-	}
-
-	for_each_cpu(i, cpu_map) {
-		build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i);
-		build_sched_groups(&d, SD_LV_BOOK, cpu_map, i);
-		build_sched_groups(&d, SD_LV_MC, cpu_map, i);
-		build_sched_groups(&d, SD_LV_CPU, cpu_map, i);
-		build_sched_groups(&d, SD_LV_NODE, cpu_map, i);
-		build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, i);
+			build_sched_groups(&d, tmp, cpu_map, i);
+		}
 	}
 
 	/* Calculate CPU power for physical packages and nodes */

[RFC][PATCH 07/14] sched: Simplify finding the lowest sched_domain

[Peter Zijlstra]

[-- Attachment #1: sched-foo6.patch --]
[-- Type: text/plain, Size: 2456 bytes --]

Instead of relying on knowing the build order and various CONFIG_
flags simply remember the bottom most sched_domain when we created the
domain hierarchy.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 include/asm-generic/percpu.h |    4 ++++
 kernel/sched.c               |   23 +++++++++++++----------
 2 files changed, 17 insertions(+), 10 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -6841,11 +6841,13 @@ struct s_data {
 	cpumask_var_t		nodemask;
 	cpumask_var_t		send_covered;
 	cpumask_var_t		tmpmask;
+	struct sched_domain ** __percpu sd;
 	struct root_domain	*rd;
 };
 
 enum s_alloc {
 	sa_rootdomain,
+	sa_sd,
 	sa_tmpmask,
 	sa_send_covered,
 	sa_nodemask,
@@ -7080,6 +7082,8 @@ static void __free_domain_allocs(struct
 	switch (what) {
 	case sa_rootdomain:
 		free_rootdomain(d->rd); /* fall through */
+	case sa_sd:
+		free_percpu(d->sd); /* fall through */
 	case sa_tmpmask:
 		free_cpumask_var(d->tmpmask); /* fall through */
 	case sa_send_covered:
@@ -7100,10 +7104,15 @@ static enum s_alloc __visit_domain_alloc
 		return sa_nodemask;
 	if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
 		return sa_send_covered;
+	d->sd = alloc_percpu(struct sched_domain *);
+	if (!d->sd) {
+		printk(KERN_WARNING "Cannot alloc per-cpu pointers\n");
+		return sa_tmpmask;
+	}
 	d->rd = alloc_rootdomain();
 	if (!d->rd) {
 		printk(KERN_WARNING "Cannot alloc root domain\n");
-		return sa_tmpmask;
+		return sa_sd;
 	}
 	return sa_rootdomain;
 }
@@ -7292,6 +7301,8 @@ static int __build_sched_domains(const s
 		sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
 		sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
 
+		*per_cpu_ptr(d.sd, i) = sd;
+
 		for (tmp = sd; tmp; tmp = tmp->parent) {
 			tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
 			build_sched_groups(&d, tmp, cpu_map, i);
@@ -7339,15 +7350,7 @@ static int __build_sched_domains(const s
 
 	/* Attach the domains */
 	for_each_cpu(i, cpu_map) {
-#ifdef CONFIG_SCHED_SMT
-		sd = &per_cpu(cpu_domains, i).sd;
-#elif defined(CONFIG_SCHED_MC)
-		sd = &per_cpu(core_domains, i).sd;
-#elif defined(CONFIG_SCHED_BOOK)
-		sd = &per_cpu(book_domains, i).sd;
-#else
-		sd = &per_cpu(phys_domains, i).sd;
-#endif
+		sd = *per_cpu_ptr(d.sd, i);
 		cpu_attach_domain(sd, d.rd, i);
 	}
 

[RFC][PATCH 08/14] sched: Simplify sched_groups_power initialization

[Peter Zijlstra]

[-- Attachment #1: sched-foo7.patch --]
[-- Type: text/plain, Size: 1997 bytes --]

Again, instead of relying on knowing the possible domains and their
order, simply rely on the sched_domain tree and whatever domains are
present in there to initialize the sched_group cpu_power.

Note: we need to iterate the CPU mask backwards because of the
cpumask_first() condition for iterating up the tree. By iterating the
mask backwards we ensure all groups of a domain are set-up before
starting on the parent groups that rely on its children to be
completely done.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 kernel/sched.c |   39 +++++----------------------------------
 1 file changed, 5 insertions(+), 34 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -7310,43 +7310,14 @@ static int __build_sched_domains(const s
 	}
 
 	/* Calculate CPU power for physical packages and nodes */
-#ifdef CONFIG_SCHED_SMT
-	for_each_cpu(i, cpu_map) {
-		sd = &per_cpu(cpu_domains, i).sd;
-		init_sched_groups_power(i, sd);
-	}
-#endif
-#ifdef CONFIG_SCHED_MC
-	for_each_cpu(i, cpu_map) {
-		sd = &per_cpu(core_domains, i).sd;
-		init_sched_groups_power(i, sd);
-	}
-#endif
-#ifdef CONFIG_SCHED_BOOK
-	for_each_cpu(i, cpu_map) {
-		sd = &per_cpu(book_domains, i).sd;
-		init_sched_groups_power(i, sd);
-	}
-#endif
-
-	for_each_cpu(i, cpu_map) {
-		sd = &per_cpu(phys_domains, i).sd;
-		init_sched_groups_power(i, sd);
-	}
-
-#ifdef CONFIG_NUMA
-	for_each_cpu(i, cpu_map) {
-		sd = &per_cpu(node_domains, i).sd;
-		init_sched_groups_power(i, sd);
-	}
+	for (i = nr_cpumask_bits-1; i >= 0; i--) {
+		if (!cpumask_test_cpu(i, cpu_map))
+			continue;
 
-	if (d.sd_allnodes) {
-		for_each_cpu(i, cpu_map) {
-			sd = &per_cpu(allnodes_domains, i).sd;
+		sd = *per_cpu_ptr(d.sd, i);
+		for (; sd; sd = sd->parent)
 			init_sched_groups_power(i, sd);
-		}
 	}
-#endif
 
 	/* Attach the domains */
 	for_each_cpu(i, cpu_map) {

[RFC][PATCH 09/14] sched: Dynamically allocate sched_domain/sched_group data-structures

[Peter Zijlstra]

[-- Attachment #1: sched-foo8.patch --]
[-- Type: text/plain, Size: 26574 bytes --]

Instead of relying on static allocations for the sched_domain and
sched_group trees, dynamically allocate and RCU free them.

Allocating this dynamically also allows for some build_sched_groups()
simplification since we can now (like with other simplifications) rely
on the sched_domain tree instead of hard-coded knowledge.

One tricky to note is that detach_destroy_domains() needs to hold
rcu_read_lock() over the entire tear-down, per-cpu is not sufficient
since that can lead to partial sched_group existance (could possibly
be solved by doing the tear-down backwards but this is much more
robust).

A concequence of the above is that we can no longer print the
sched_domain debug stuff from cpu_attach_domain() since that might now
run with preemption disabled (due to classic RCU etc.) and
sched_domain_debug() does some GFP_KERNEL allocations.

Another thing to note is that we now fully rely on normal RCU and not
RCU-sched, this is because with the new and exiting RCU flavours we
grew over the years BH doesn't necessarily hold off RCU-sched grace
periods (-rt is known to break this). This would in fact already cause
us grief since we do sched_domain/sched_group iterations from softirq
context.

This patch is somewhat larger than I would like it to be, but I didn't
find any means of shrinking/splitting this.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 include/linux/sched.h |    5 
 kernel/sched.c        |  464 +++++++++++++++++++-------------------------------
 kernel/sched_fair.c   |   32 ++-
 3 files changed, 214 insertions(+), 287 deletions(-)

Index: linux-2.6/include/linux/sched.h
===================================================================
--- linux-2.6.orig/include/linux/sched.h
+++ linux-2.6/include/linux/sched.h
@@ -867,6 +867,7 @@ static inline int sd_power_saving_flags(
 
 struct sched_group {
 	struct sched_group *next;	/* Must be a circular list */
+	atomic_t ref;
 
 	/*
 	 * CPU power of this group, SCHED_LOAD_SCALE being max power for a
@@ -972,6 +973,10 @@ struct sched_domain {
 #ifdef CONFIG_SCHED_DEBUG
 	char *name;
 #endif
+	union {
+		void *private;		/* used during construction */
+		struct rcu_head rcu;	/* used during destruction */
+	};
 
 	unsigned int span_weight;
 	/*
Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -418,6 +418,7 @@ struct rt_rq {
  */
 struct root_domain {
 	atomic_t refcount;
+	struct rcu_head rcu;
 	cpumask_var_t span;
 	cpumask_var_t online;
 
@@ -572,7 +573,7 @@ static inline int cpu_of(struct rq *rq)
 
 #define rcu_dereference_check_sched_domain(p) \
 	rcu_dereference_check((p), \
-			      rcu_read_lock_sched_held() || \
+			      rcu_read_lock_held() || \
 			      lockdep_is_held(&sched_domains_mutex))
 
 /*
@@ -6548,12 +6549,11 @@ sd_parent_degenerate(struct sched_domain
 	return 1;
 }
 
-static void free_rootdomain(struct root_domain *rd)
+static void free_rootdomain(struct rcu_head *rcu)
 {
-	synchronize_sched();
+	struct root_domain *rd = container_of(rcu, struct root_domain, rcu);
 
 	cpupri_cleanup(&rd->cpupri);
-
 	free_cpumask_var(rd->rto_mask);
 	free_cpumask_var(rd->online);
 	free_cpumask_var(rd->span);
@@ -6594,7 +6594,7 @@ static void rq_attach_root(struct rq *rq
 	raw_spin_unlock_irqrestore(&rq->lock, flags);
 
 	if (old_rd)
-		free_rootdomain(old_rd);
+		call_rcu_sched(&old_rd->rcu, free_rootdomain);
 }
 
 static int init_rootdomain(struct root_domain *rd)
@@ -6645,6 +6645,25 @@ static struct root_domain *alloc_rootdom
 	return rd;
 }
 
+static void free_sched_domain(struct rcu_head *rcu)
+{
+	struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu);
+	if (atomic_dec_and_test(&sd->groups->ref))
+		kfree(sd->groups);
+	kfree(sd);
+}
+
+static void destroy_sched_domain(struct sched_domain *sd, int cpu)
+{
+	call_rcu(&sd->rcu, free_sched_domain);
+}
+
+static void destroy_sched_domains(struct sched_domain *sd, int cpu)
+{
+	for (; sd; sd = sd->parent)
+		destroy_sched_domain(sd, cpu);
+}
+
 /*
  * Attach the domain 'sd' to 'cpu' as its base domain. Callers must
  * hold the hotplug lock.
@@ -6665,20 +6684,25 @@ cpu_attach_domain(struct sched_domain *s
 			tmp->parent = parent->parent;
 			if (parent->parent)
 				parent->parent->child = tmp;
+			destroy_sched_domain(parent, cpu);
 		} else
 			tmp = tmp->parent;
 	}
 
 	if (sd && sd_degenerate(sd)) {
+		tmp = sd;
 		sd = sd->parent;
+		destroy_sched_domain(tmp, cpu);
 		if (sd)
 			sd->child = NULL;
 	}
 
-	sched_domain_debug(sd, cpu);
+//	sched_domain_debug(sd, cpu);
 
 	rq_attach_root(rq, rd);
+	tmp = rq->sd;
 	rcu_assign_pointer(rq->sd, sd);
+	destroy_sched_domains(tmp, cpu);
 }
 
 /* cpus with isolated domains */
@@ -6694,56 +6718,6 @@ static int __init isolated_cpu_setup(cha
 
 __setup("isolcpus=", isolated_cpu_setup);
 
-/*
- * init_sched_build_groups takes the cpumask we wish to span, and a pointer
- * to a function which identifies what group(along with sched group) a CPU
- * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids
- * (due to the fact that we keep track of groups covered with a struct cpumask).
- *
- * init_sched_build_groups will build a circular linked list of the groups
- * covered by the given span, and will set each group's ->cpumask correctly,
- * and ->cpu_power to 0.
- */
-static void
-init_sched_build_groups(const struct cpumask *span,
-			const struct cpumask *cpu_map,
-			int (*group_fn)(int cpu, const struct cpumask *cpu_map,
-					struct sched_group **sg,
-					struct cpumask *tmpmask),
-			struct cpumask *covered, struct cpumask *tmpmask)
-{
-	struct sched_group *first = NULL, *last = NULL;
-	int i;
-
-	cpumask_clear(covered);
-
-	for_each_cpu(i, span) {
-		struct sched_group *sg;
-		int group = group_fn(i, cpu_map, &sg, tmpmask);
-		int j;
-
-		if (cpumask_test_cpu(i, covered))
-			continue;
-
-		cpumask_clear(sched_group_cpus(sg));
-		sg->cpu_power = 0;
-
-		for_each_cpu(j, span) {
-			if (group_fn(j, cpu_map, NULL, tmpmask) != group)
-				continue;
-
-			cpumask_set_cpu(j, covered);
-			cpumask_set_cpu(j, sched_group_cpus(sg));
-		}
-		if (!first)
-			first = sg;
-		if (last)
-			last->next = sg;
-		last = sg;
-	}
-	last->next = first;
-}
-
 #define SD_NODES_PER_DOMAIN 16
 
 #ifdef CONFIG_NUMA
@@ -6834,154 +6808,96 @@ struct static_sched_domain {
 	DECLARE_BITMAP(span, CONFIG_NR_CPUS);
 };
 
+struct sd_data {
+	struct sched_domain **__percpu sd;
+	struct sched_group **__percpu sg;
+};
+
 struct s_data {
 #ifdef CONFIG_NUMA
 	int			sd_allnodes;
 #endif
 	cpumask_var_t		nodemask;
 	cpumask_var_t		send_covered;
-	cpumask_var_t		tmpmask;
 	struct sched_domain ** __percpu sd;
+	struct sd_data 		sdd[SD_LV_MAX];
 	struct root_domain	*rd;
 };
 
 enum s_alloc {
 	sa_rootdomain,
 	sa_sd,
-	sa_tmpmask,
+	sa_sd_storage,
 	sa_send_covered,
 	sa_nodemask,
 	sa_none,
 };
 
 /*
- * SMT sched-domains:
+ * Assumes the sched_domain tree is fully constructed
  */
-#ifdef CONFIG_SCHED_SMT
-static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains);
-static DEFINE_PER_CPU(struct static_sched_group, sched_groups);
-
-static int
-cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map,
-		 struct sched_group **sg, struct cpumask *unused)
+static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg)
 {
-	if (sg)
-		*sg = &per_cpu(sched_groups, cpu).sg;
-	return cpu;
-}
-#endif /* CONFIG_SCHED_SMT */
+	struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu);
+	struct sched_domain *child = sd->child;
 
-/*
- * multi-core sched-domains:
- */
-#ifdef CONFIG_SCHED_MC
-static DEFINE_PER_CPU(struct static_sched_domain, core_domains);
-static DEFINE_PER_CPU(struct static_sched_group, sched_group_core);
+	if (child)
+		cpu = cpumask_first(sched_domain_span(child));
 
-static int
-cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
-		  struct sched_group **sg, struct cpumask *mask)
-{
-	int group;
-#ifdef CONFIG_SCHED_SMT
-	cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
-	group = cpumask_first(mask);
-#else
-	group = cpu;
-#endif
 	if (sg)
-		*sg = &per_cpu(sched_group_core, group).sg;
-	return group;
+		*sg = *per_cpu_ptr(sdd->sg, cpu);
+
+	return cpu;
 }
-#endif /* CONFIG_SCHED_MC */
 
 /*
- * book sched-domains:
+ * build_sched_groups takes the cpumask we wish to span, and a pointer
+ * to a function which identifies what group(along with sched group) a CPU
+ * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids
+ * (due to the fact that we keep track of groups covered with a struct cpumask).
+ *
+ * build_sched_groups will build a circular linked list of the groups
+ * covered by the given span, and will set each group's ->cpumask correctly,
+ * and ->cpu_power to 0.
  */
-#ifdef CONFIG_SCHED_BOOK
-static DEFINE_PER_CPU(struct static_sched_domain, book_domains);
-static DEFINE_PER_CPU(struct static_sched_group, sched_group_book);
-
-static int
-cpu_to_book_group(int cpu, const struct cpumask *cpu_map,
-		  struct sched_group **sg, struct cpumask *mask)
-{
-	int group = cpu;
-#ifdef CONFIG_SCHED_MC
-	cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
-	group = cpumask_first(mask);
-#elif defined(CONFIG_SCHED_SMT)
-	cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
-	group = cpumask_first(mask);
-#endif
-	if (sg)
-		*sg = &per_cpu(sched_group_book, group).sg;
-	return group;
-}
-#endif /* CONFIG_SCHED_BOOK */
-
-static DEFINE_PER_CPU(struct static_sched_domain, phys_domains);
-static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys);
-
-static int
-cpu_to_phys_group(int cpu, const struct cpumask *cpu_map,
-		  struct sched_group **sg, struct cpumask *mask)
+static void
+build_sched_groups(struct sched_domain *sd, struct cpumask *covered)
 {
-	int group;
-#ifdef CONFIG_SCHED_BOOK
-	cpumask_and(mask, cpu_book_mask(cpu), cpu_map);
-	group = cpumask_first(mask);
-#elif defined(CONFIG_SCHED_MC)
-	cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
-	group = cpumask_first(mask);
-#elif defined(CONFIG_SCHED_SMT)
-	cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
-	group = cpumask_first(mask);
-#else
-	group = cpu;
-#endif
-	if (sg)
-		*sg = &per_cpu(sched_group_phys, group).sg;
-	return group;
-}
-
-#ifdef CONFIG_NUMA
-static DEFINE_PER_CPU(struct static_sched_domain, node_domains);
-static DEFINE_PER_CPU(struct static_sched_group, sched_group_node);
+	struct sched_group *first = NULL, *last = NULL;
+	struct sd_data *sdd = sd->private;
+	const struct cpumask *span = sched_domain_span(sd);
+	int i;
 
-static int cpu_to_node_group(int cpu, const struct cpumask *cpu_map,
-				 struct sched_group **sg,
-				 struct cpumask *nodemask)
-{
-	int group;
+	cpumask_clear(covered);
 
-	cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map);
-	group = cpumask_first(nodemask);
+	for_each_cpu(i, span) {
+		struct sched_group *sg;
+		int group = get_group(i, sdd, &sg);
+		int j;
 
-	if (sg)
-		*sg = &per_cpu(sched_group_node, group).sg;
-	return group;
-}
+		if (cpumask_test_cpu(i, covered))
+			continue;
 
-static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains);
-static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes);
+		cpumask_clear(sched_group_cpus(sg));
+		sg->cpu_power = 0;
 
-static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map,
-				 struct sched_group **sg,
-				 struct cpumask *nodemask)
-{
-	int group;
+		for_each_cpu(j, span) {
+			if (get_group(j, sdd, NULL) != group)
+				continue;
 
-	cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map);
-	group = cpumask_first(nodemask);
+			cpumask_set_cpu(j, covered);
+			cpumask_set_cpu(j, sched_group_cpus(sg));
+		}
 
-	if (sg)
-		*sg = &per_cpu(sched_group_allnodes, group).sg;
-	return group;
+		if (!first)
+			first = sg;
+		if (last)
+			last->next = sg;
+		last = sg;
+	}
+	last->next = first;
 }
 
-#endif /* CONFIG_NUMA */
-
 /*
  * Initialize sched groups cpu_power.
  *
@@ -7015,15 +6931,15 @@ static void init_sched_groups_power(int
 # define SD_INIT_NAME(sd, type)		do { } while (0)
 #endif
 
-#define	SD_INIT(sd, type)	sd_init_##type(sd)
-
-#define SD_INIT_FUNC(type)	\
-static noinline void sd_init_##type(struct sched_domain *sd)	\
-{								\
-	memset(sd, 0, sizeof(*sd));				\
-	*sd = SD_##type##_INIT;					\
-	sd->level = SD_LV_##type;				\
-	SD_INIT_NAME(sd, type);					\
+#define SD_INIT_FUNC(type)						       \
+static noinline struct sched_domain *sd_init_##type(struct s_data *d, int cpu) \
+{									       \
+	struct sched_domain *sd = *per_cpu_ptr(d->sdd[SD_LV_##type].sd, cpu);  \
+	*sd = SD_##type##_INIT;						       \
+	sd->level = SD_LV_##type;					       \
+	SD_INIT_NAME(sd, type);						       \
+	sd->private = &d->sdd[SD_LV_##type];				       \
+	return sd;							       \
 }
 
 SD_INIT_FUNC(CPU)
@@ -7079,13 +6995,22 @@ static void set_domain_attribute(struct
 static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
 				 const struct cpumask *cpu_map)
 {
+	int i, j;
+
 	switch (what) {
 	case sa_rootdomain:
-		free_rootdomain(d->rd); /* fall through */
+		free_rootdomain(&d->rd->rcu); /* fall through */
 	case sa_sd:
 		free_percpu(d->sd); /* fall through */
-	case sa_tmpmask:
-		free_cpumask_var(d->tmpmask); /* fall through */
+	case sa_sd_storage:
+		for (i = 0; i < SD_LV_MAX; i++) {
+			for_each_cpu(j, cpu_map) {
+				kfree(*per_cpu_ptr(d->sdd[i].sd, j));
+				kfree(*per_cpu_ptr(d->sdd[i].sg, j));
+			}
+			free_percpu(d->sdd[i].sd);
+			free_percpu(d->sdd[i].sg);
+		} /* fall through */
 	case sa_send_covered:
 		free_cpumask_var(d->send_covered); /* fall through */
 	case sa_nodemask:
@@ -7098,25 +7023,70 @@ static void __free_domain_allocs(struct
 static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
 						   const struct cpumask *cpu_map)
 {
+	int i, j;
+
+	memset(d, 0, sizeof(*d));
+
 	if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL))
 		return sa_none;
 	if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
 		return sa_nodemask;
-	if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
-		return sa_send_covered;
-	d->sd = alloc_percpu(struct sched_domain *);
-	if (!d->sd) {
-		printk(KERN_WARNING "Cannot alloc per-cpu pointers\n");
-		return sa_tmpmask;
+	for (i = 0; i < SD_LV_MAX; i++) {
+		d->sdd[i].sd = alloc_percpu(struct sched_domain *);
+		if (!d->sdd[i].sd)
+			return sa_sd_storage;
+
+		d->sdd[i].sg = alloc_percpu(struct sched_group *);
+		if (!d->sdd[i].sg)
+			return sa_sd_storage;
+
+		for_each_cpu(j, cpu_map) {
+			struct sched_domain *sd;
+			struct sched_group *sg;
+
+		       	sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(),
+					GFP_KERNEL, cpu_to_node(j));
+			if (!sd)
+				return sa_sd_storage;
+
+			*per_cpu_ptr(d->sdd[i].sd, j) = sd;
+
+			sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(),
+					GFP_KERNEL, cpu_to_node(j));
+			if (!sg)
+				return sa_sd_storage;
+
+			*per_cpu_ptr(d->sdd[i].sg, j) = sg;
+		}
 	}
+	d->sd = alloc_percpu(struct sched_domain *);
+	if (!d->sd)
+		return sa_sd_storage;
 	d->rd = alloc_rootdomain();
-	if (!d->rd) {
-		printk(KERN_WARNING "Cannot alloc root domain\n");
+	if (!d->rd)
 		return sa_sd;
-	}
 	return sa_rootdomain;
 }
 
+/*
+ * NULL the sd_data elements we've used to build the sched_domain and
+ * sched_group structure so that the subsequent __free_domain_allocs()
+ * will not free the data we're using.
+ */
+static void claim_allocations(int cpu, struct sched_domain *sd)
+{
+	struct sd_data *sdd = sd->private;
+	struct sched_group *sg = sd->groups;
+
+	WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd);
+	*per_cpu_ptr(sdd->sd, cpu) = NULL;
+
+	if (cpu == cpumask_first(sched_group_cpus(sg))) {
+		WARN_ON_ONCE(*per_cpu_ptr(sdd->sg, cpu) != sg);
+		*per_cpu_ptr(sdd->sg, cpu) = NULL;
+	}
+}
+
 static struct sched_domain *__build_numa_sched_domains(struct s_data *d,
 	const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i)
 {
@@ -7127,24 +7097,20 @@ static struct sched_domain *__build_numa
 	d->sd_allnodes = 0;
 	if (cpumask_weight(cpu_map) >
 	    SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) {
-		sd = &per_cpu(allnodes_domains, i).sd;
-		SD_INIT(sd, ALLNODES);
+		sd = sd_init_ALLNODES(d, i);
 		set_domain_attribute(sd, attr);
 		cpumask_copy(sched_domain_span(sd), cpu_map);
-		cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask);
 		d->sd_allnodes = 1;
 	}
 	parent = sd;
 
-	sd = &per_cpu(node_domains, i).sd;
-	SD_INIT(sd, NODE);
+	sd = sd_init_NODE(d, i);
 	set_domain_attribute(sd, attr);
 	sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd));
 	sd->parent = parent;
 	if (parent)
 		parent->child = sd;
 	cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map);
-	cpu_to_node_group(i, cpu_map, &sd->groups, d->tmpmask);
 #endif
 	return sd;
 }
@@ -7154,14 +7120,12 @@ static struct sched_domain *__build_cpu_
 	struct sched_domain *parent, int i)
 {
 	struct sched_domain *sd;
-	sd = &per_cpu(phys_domains, i).sd;
-	SD_INIT(sd, CPU);
+	sd = sd_init_CPU(d, i);
 	set_domain_attribute(sd, attr);
 	cpumask_copy(sched_domain_span(sd), d->nodemask);
 	sd->parent = parent;
 	if (parent)
 		parent->child = sd;
-	cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask);
 	return sd;
 }
 
@@ -7171,13 +7135,11 @@ static struct sched_domain *__build_book
 {
 	struct sched_domain *sd = parent;
 #ifdef CONFIG_SCHED_BOOK
-	sd = &per_cpu(book_domains, i).sd;
-	SD_INIT(sd, BOOK);
+	sd = sd_init_BOOK(d, i);
 	set_domain_attribute(sd, attr);
 	cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i));
 	sd->parent = parent;
 	parent->child = sd;
-	cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask);
 #endif
 	return sd;
 }
@@ -7188,13 +7150,11 @@ static struct sched_domain *__build_mc_s
 {
 	struct sched_domain *sd = parent;
 #ifdef CONFIG_SCHED_MC
-	sd = &per_cpu(core_domains, i).sd;
-	SD_INIT(sd, MC);
+	sd = sd_init_MC(d, i);
 	set_domain_attribute(sd, attr);
 	cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i));
 	sd->parent = parent;
 	parent->child = sd;
-	cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask);
 #endif
 	return sd;
 }
@@ -7205,70 +7165,15 @@ static struct sched_domain *__build_smt_
 {
 	struct sched_domain *sd = parent;
 #ifdef CONFIG_SCHED_SMT
-	sd = &per_cpu(cpu_domains, i).sd;
-	SD_INIT(sd, SIBLING);
+	sd = sd_init_SIBLING(d, i);
 	set_domain_attribute(sd, attr);
 	cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i));
 	sd->parent = parent;
 	parent->child = sd;
-	cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask);
 #endif
 	return sd;
 }
 
-static void build_sched_groups(struct s_data *d, struct sched_domain *sd,
-			       const struct cpumask *cpu_map, int cpu)
-{
-	switch (sd->level) {
-#ifdef CONFIG_SCHED_SMT
-	case SD_LV_SIBLING: /* set up CPU (sibling) groups */
-		if (cpu == cpumask_first(sched_domain_span(sd)))
-			init_sched_build_groups(sched_domain_span(sd), cpu_map,
-						&cpu_to_cpu_group,
-						d->send_covered, d->tmpmask);
-		break;
-#endif
-#ifdef CONFIG_SCHED_MC
-	case SD_LV_MC: /* set up multi-core groups */
-		if (cpu == cpumask_first(sched_domain_span(sd)))
-			init_sched_build_groups(sched_domain_span(sd), cpu_map,
-						&cpu_to_core_group,
-						d->send_covered, d->tmpmask);
-		break;
-#endif
-#ifdef CONFIG_SCHED_BOOK
-	case SD_LV_BOOK: /* set up book groups */
-		if (cpu == cpumask_first(sched_domain_span(sd)))
-			init_sched_build_groups(sched_domain_span(sd), cpu_map,
-						&cpu_to_book_group,
-						d->send_covered, d->tmpmask);
-		break;
-#endif
-	case SD_LV_CPU: /* set up physical groups */
-		if (cpu == cpumask_first(sched_domain_span(sd)))
-			init_sched_build_groups(sched_domain_span(sd), cpu_map,
-						&cpu_to_phys_group,
-						d->send_covered, d->tmpmask);
-		break;
-#ifdef CONFIG_NUMA
-	case SD_LV_NODE:
-		if (cpu == cpumask_first(sched_domain_span(sd)))
-			init_sched_build_groups(sched_domain_span(sd), cpu_map,
-						&cpu_to_node_group,
-						d->send_covered, d->tmpmask);
-
-	case SD_LV_ALLNODES:
-		if (cpu == cpumask_first(cpu_map))
-			init_sched_build_groups(cpu_map, cpu_map,
-					&cpu_to_allnodes_group,
-					d->send_covered, d->tmpmask);
-		break;
-#endif
-	default:
-		break;
-	}
-}
-
 /*
  * Build sched domains for a given set of cpus and attach the sched domains
  * to the individual cpus
@@ -7277,20 +7182,15 @@ static int __build_sched_domains(const s
 				 struct sched_domain_attr *attr)
 {
 	enum s_alloc alloc_state = sa_none;
+	struct sched_domain *sd;
 	struct s_data d;
-	struct sched_domain *sd, *tmp;
 	int i;
-#ifdef CONFIG_NUMA
-	d.sd_allnodes = 0;
-#endif
 
 	alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
 	if (alloc_state != sa_rootdomain)
 		goto error;
 
-	/*
-	 * Set up domains for cpus specified by the cpu_map.
-	 */
+	/* Set up domains for cpus specified by the cpu_map. */
 	for_each_cpu(i, cpu_map) {
 		cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)),
 			    cpu_map);
@@ -7302,10 +7202,19 @@ static int __build_sched_domains(const s
 		sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
 
 		*per_cpu_ptr(d.sd, i) = sd;
+	}
 
-		for (tmp = sd; tmp; tmp = tmp->parent) {
-			tmp->span_weight = cpumask_weight(sched_domain_span(tmp));
-			build_sched_groups(&d, tmp, cpu_map, i);
+	/* Build the groups for the domains */
+	for_each_cpu(i, cpu_map) {
+		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
+			sd->span_weight = cpumask_weight(sched_domain_span(sd));
+			get_group(i, sd->private, &sd->groups);
+			atomic_inc(&sd->groups->ref);
+
+			if (i != cpumask_first(sched_domain_span(sd)))
+				continue;
+
+			build_sched_groups(sd, d.send_covered);
 		}
 	}
 
@@ -7314,18 +7223,20 @@ static int __build_sched_domains(const s
 		if (!cpumask_test_cpu(i, cpu_map))
 			continue;
 
-		sd = *per_cpu_ptr(d.sd, i);
-		for (; sd; sd = sd->parent)
+		for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
+			claim_allocations(i, sd);
 			init_sched_groups_power(i, sd);
+		}
 	}
 
 	/* Attach the domains */
 	for_each_cpu(i, cpu_map) {
 		sd = *per_cpu_ptr(d.sd, i);
 		cpu_attach_domain(sd, d.rd, i);
+		sched_domain_debug(sd, i);
 	}
 
-	__free_domain_allocs(&d, sa_tmpmask, cpu_map);
+	__free_domain_allocs(&d, sa_sd, cpu_map);
 	return 0;
 
 error:
@@ -7407,25 +7318,18 @@ static int init_sched_domains(const stru
 	return err;
 }
 
-static void destroy_sched_domains(const struct cpumask *cpu_map,
-				       struct cpumask *tmpmask)
-{
-}
-
 /*
  * Detach sched domains from a group of cpus specified in cpu_map
  * These cpus will now be attached to the NULL domain
  */
 static void detach_destroy_domains(const struct cpumask *cpu_map)
 {
-	/* Save because hotplug lock held. */
-	static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS);
 	int i;
 
+	rcu_read_lock();
 	for_each_cpu(i, cpu_map)
 		cpu_attach_domain(NULL, &def_root_domain, i);
-	synchronize_sched();
-	destroy_sched_domains(cpu_map, to_cpumask(tmpmask));
+	rcu_read_unlock();
 }
 
 /* handle null as "default" */
Index: linux-2.6/kernel/sched_fair.c
===================================================================
--- linux-2.6.orig/kernel/sched_fair.c
+++ linux-2.6/kernel/sched_fair.c
@@ -1621,6 +1621,7 @@ static int select_idle_sibling(struct ta
 	/*
 	 * Otherwise, iterate the domains and find an elegible idle cpu.
 	 */
+	rcu_read_lock();
 	for_each_domain(target, sd) {
 		if (!(sd->flags & SD_SHARE_PKG_RESOURCES))
 			break;
@@ -1640,6 +1641,7 @@ static int select_idle_sibling(struct ta
 		    cpumask_test_cpu(prev_cpu, sched_domain_span(sd)))
 			break;
 	}
+	rcu_read_unlock();
 
 	return target;
 }
@@ -1672,6 +1674,7 @@ select_task_rq_fair(struct rq *rq, struc
 		new_cpu = prev_cpu;
 	}
 
+	rcu_read_lock();
 	for_each_domain(cpu, tmp) {
 		if (!(tmp->flags & SD_LOAD_BALANCE))
 			continue;
@@ -1721,10 +1724,11 @@ select_task_rq_fair(struct rq *rq, struc
 	}
 
 	if (affine_sd) {
-		if (cpu == prev_cpu || wake_affine(affine_sd, p, sync))
-			return select_idle_sibling(p, cpu);
-		else
-			return select_idle_sibling(p, prev_cpu);
+		if (wake_affine(affine_sd, p, sync))
+			prev_cpu = cpu;
+
+		new_cpu = select_idle_sibling(p, prev_cpu);
+		goto unlock;
 	}
 
 	while (sd) {
@@ -1765,6 +1769,8 @@ select_task_rq_fair(struct rq *rq, struc
 		}
 		/* while loop will break here if sd == NULL */
 	}
+unlock:
+	rcu_read_unlock();
 
 	return new_cpu;
 }
@@ -3466,6 +3472,7 @@ static void idle_balance(int this_cpu, s
 	raw_spin_unlock(&this_rq->lock);
 
 	update_shares(this_cpu);
+	rcu_read_lock();
 	for_each_domain(this_cpu, sd) {
 		unsigned long interval;
 		int balance = 1;
@@ -3487,6 +3494,7 @@ static void idle_balance(int this_cpu, s
 			break;
 		}
 	}
+	rcu_read_unlock();
 
 	raw_spin_lock(&this_rq->lock);
 
@@ -3535,6 +3543,7 @@ static int active_load_balance_cpu_stop(
 	double_lock_balance(busiest_rq, target_rq);
 
 	/* Search for an sd spanning us and the target CPU. */
+	rcu_read_lock();
 	for_each_domain(target_cpu, sd) {
 		if ((sd->flags & SD_LOAD_BALANCE) &&
 		    cpumask_test_cpu(busiest_cpu, sched_domain_span(sd)))
@@ -3550,6 +3559,7 @@ static int active_load_balance_cpu_stop(
 		else
 			schedstat_inc(sd, alb_failed);
 	}
+	rcu_read_unlock();
 	double_unlock_balance(busiest_rq, target_rq);
 out_unlock:
 	busiest_rq->active_balance = 0;
@@ -3676,6 +3686,7 @@ static int find_new_ilb(int cpu)
 {
 	struct sched_domain *sd;
 	struct sched_group *ilb_group;
+	int ilb = nr_cpu_ids;
 
 	/*
 	 * Have idle load balancer selection from semi-idle packages only
@@ -3691,20 +3702,25 @@ static int find_new_ilb(int cpu)
 	if (cpumask_weight(nohz.idle_cpus_mask) < 2)
 		goto out_done;
 
+	rcu_read_lock();
 	for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) {
 		ilb_group = sd->groups;
 
 		do {
-			if (is_semi_idle_group(ilb_group))
-				return cpumask_first(nohz.grp_idle_mask);
+			if (is_semi_idle_group(ilb_group)) {
+				ilb = cpumask_first(nohz.grp_idle_mask);
+				goto unlock;
+			}
 
 			ilb_group = ilb_group->next;
 
 		} while (ilb_group != sd->groups);
 	}
+unlock:
+	rcu_read_unlock();
 
 out_done:
-	return nr_cpu_ids;
+	return ilb;
 }
 #else /*  (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */
 static inline int find_new_ilb(int call_cpu)
@@ -3838,6 +3854,7 @@ static void rebalance_domains(int cpu, e
 
 	update_shares(cpu);
 
+	rcu_read_lock();
 	for_each_domain(cpu, sd) {
 		if (!(sd->flags & SD_LOAD_BALANCE))
 			continue;
@@ -3886,6 +3903,7 @@ static void rebalance_domains(int cpu, e
 		if (!balance)
 			break;
 	}
+	rcu_read_unlock();
 
 	/*
 	 * next_balance will be updated only when there is a need.

Re: [RFC][PATCH 09/14] sched: Dynamically allocate sched_domain/sched_group data-structures

[Bharata B Rao]

On Mon, Mar 14, 2011 at 8:36 PM, Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:
>
>  enum s_alloc {
>        sa_rootdomain,
>        sa_sd,
> -       sa_tmpmask,
> +       sa_sd_storage,
>        sa_send_covered,
>        sa_nodemask,
>        sa_none,
>  };
>
>
>
>  SD_INIT_FUNC(CPU)
> @@ -7079,13 +6995,22 @@ static void set_domain_attribute(struct
>  static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
>                                 const struct cpumask *cpu_map)
>  {
> +       int i, j;
> +
>        switch (what) {
>        case sa_rootdomain:
> -               free_rootdomain(d->rd); /* fall through */
> +               free_rootdomain(&d->rd->rcu); /* fall through */
>        case sa_sd:
>                free_percpu(d->sd); /* fall through */
> -       case sa_tmpmask:
> -               free_cpumask_var(d->tmpmask); /* fall through */
> +       case sa_sd_storage:
> +               for (i = 0; i < SD_LV_MAX; i++) {
> +                       for_each_cpu(j, cpu_map) {
> +                               kfree(*per_cpu_ptr(d->sdd[i].sd, j));
> +                               kfree(*per_cpu_ptr(d->sdd[i].sg, j));
> +                       }
> +                       free_percpu(d->sdd[i].sd);
> +                       free_percpu(d->sdd[i].sg);
> +               } /* fall through */
>        case sa_send_covered:
>                free_cpumask_var(d->send_covered); /* fall through */
>        case sa_nodemask:
> @@ -7098,25 +7023,70 @@ static void __free_domain_allocs(struct
>  static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
>                                                   const struct cpumask *cpu_map)
>  {
> +       int i, j;
> +
> +       memset(d, 0, sizeof(*d));
> +
>        if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL))
>                return sa_none;
>        if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
>                return sa_nodemask;
> -       if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
> -               return sa_send_covered;

sa_send_covered enum member can be removed, since you no longer seem
to be using it.

Regards,
Bharata.

Re: [RFC][PATCH 09/14] sched: Dynamically allocate sched_domain/sched_group data-structures

[Peter Zijlstra]

On Fri, 2011-03-18 at 14:38 +0530, Bharata B Rao wrote:
> sa_send_covered enum member can be removed, since you no longer seem
> to be using it. 

Indeed, thanks!

Re: [RFC][PATCH 09/14] sched: Dynamically allocate sched_domain/sched_group data-structures

[Venkatesh Pallipadi]

On Mon, Mar 14, 2011 at 8:06 AM, Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:
> @@ -1721,10 +1724,11 @@ select_task_rq_fair(struct rq *rq, struc
>        }
>
>        if (affine_sd) {
> -               if (cpu == prev_cpu || wake_affine(affine_sd, p, sync))
> -                       return select_idle_sibling(p, cpu);
> -               else
> -                       return select_idle_sibling(p, prev_cpu);
> +               if (wake_affine(affine_sd, p, sync))
> +                       prev_cpu = cpu;
> +
> +               new_cpu = select_idle_sibling(p, prev_cpu);
> +               goto unlock;
>        }
>
>        while (sd) {

This would result in going through wake_affine() doing all
effective_load stuff even with cpu == prev_cpu. No?
So, we need either if (cpu != prev_cpu && wake_affine(affine_sd, p,
sync)) or an check at the start to have want_affine=0 for this case.

Overall patchset looks great!

Thanks,
Venki

Re: [RFC][PATCH 09/14] sched: Dynamically allocate sched_domain/sched_group data-structures

[Peter Zijlstra]

On Fri, 2011-03-18 at 18:23 -0700, Venkatesh Pallipadi wrote:
> >        if (affine_sd) {
> > -               if (cpu == prev_cpu || wake_affine(affine_sd, p, sync))
> > -                       return select_idle_sibling(p, cpu);
> > -               else
> > -                       return select_idle_sibling(p, prev_cpu);
> > +               if (wake_affine(affine_sd, p, sync))
> > +                       prev_cpu = cpu;
> > +
> > +               new_cpu = select_idle_sibling(p, prev_cpu);
> > +               goto unlock;
> >        }
> >
> >        while (sd) {
> 
> This would result in going through wake_affine() doing all
> effective_load stuff even with cpu == prev_cpu. No?
> So, we need either if (cpu != prev_cpu && wake_affine(affine_sd, p,
> sync)) or an check at the start to have want_affine=0 for this case.

D'0h yeah, I missed the conditional execution of wake_affine there, silly me.

[RFC][PATCH 10/14] sched: Simplify the free path some

[Peter Zijlstra]

[-- Attachment #1: sched-foo9.patch --]
[-- Type: text/plain, Size: 1416 bytes --]

If we check the root_domain reference count we can see if its been
used or not, use this observation to simplify some of the return
paths.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 kernel/sched.c |   11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -6999,7 +6999,8 @@ static void __free_domain_allocs(struct 
 
 	switch (what) {
 	case sa_rootdomain:
-		free_rootdomain(&d->rd->rcu); /* fall through */
+		if (!atomic_read(&d->rd->refcount))
+			free_rootdomain(&d->rd->rcu); /* fall through */
 	case sa_sd:
 		free_percpu(d->sd); /* fall through */
 	case sa_sd_storage:
@@ -7184,7 +7185,7 @@ static int __build_sched_domains(const s
 	enum s_alloc alloc_state = sa_none;
 	struct sched_domain *sd;
 	struct s_data d;
-	int i;
+	int i, ret = -ENOMEM;
 
 	alloc_state = __visit_domain_allocation_hell(&d, cpu_map);
 	if (alloc_state != sa_rootdomain)
@@ -7236,12 +7237,10 @@ static int __build_sched_domains(const s
 		sched_domain_debug(sd, i);
 	}
 
-	__free_domain_allocs(&d, sa_sd, cpu_map);
-	return 0;
-
+	ret = 0;
 error:
 	__free_domain_allocs(&d, alloc_state, cpu_map);
-	return -ENOMEM;
+	return ret;
 }
 
 static int build_sched_domains(const struct cpumask *cpu_map)

[RFC][PATCH 11/14] sched: Reduce some allocation pressure

[Peter Zijlstra]

[-- Attachment #1: sched-foo10.patch --]
[-- Type: text/plain, Size: 991 bytes --]

Since we now allocate SD_LV_MAX * nr_cpu_ids sched_domain/sched_group
structures when rebuilding the scheduler toplogy it might make sense
to shrink that depending on the CONFIG_ options.

This is only needed until we get rid of SD_LV_* alltogether and
provide a full dynamic topology interface.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 include/linux/sched.h |    8 ++++++++
 1 file changed, 8 insertions(+)

Index: linux-2.6/include/linux/sched.h
===================================================================
--- linux-2.6.orig/include/linux/sched.h
+++ linux-2.6/include/linux/sched.h
@@ -895,12 +895,20 @@ static inline struct cpumask *sched_grou
 
 enum sched_domain_level {
 	SD_LV_NONE = 0,
+#ifdef CONFIG_SCHED_SMT
 	SD_LV_SIBLING,
+#endif
+#ifdef CONFIG_SCHED_MC
 	SD_LV_MC,
+#endif
+#ifdef CONFIG_SCHED_BOOK
 	SD_LV_BOOK,
+#endif
 	SD_LV_CPU,
+#ifdef CONFIG_NUMA
 	SD_LV_NODE,
 	SD_LV_ALLNODES,
+#endif
 	SD_LV_MAX
 };
 

[RFC][PATCH 12/14] sched: Simplify NODE/ALLNODES domain creation

[Peter Zijlstra]

[-- Attachment #1: sched-foo11.patch --]
[-- Type: text/plain, Size: 2839 bytes --]

Don't treat ALLNODES/NODE different for difference's sake. Simply
always create the ALLNODES domain and let the sd_degenerate() checks
kill it when its redundant. This simplifies the code flow.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 kernel/sched.c |   40 ++++++++++++++++++++++------------------
 1 file changed, 22 insertions(+), 18 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -6814,9 +6814,6 @@ struct sd_data {
 };
 
 struct s_data {
-#ifdef CONFIG_NUMA
-	int			sd_allnodes;
-#endif
 	cpumask_var_t		nodemask;
 	cpumask_var_t		send_covered;
 	struct sched_domain ** __percpu sd;
@@ -7088,30 +7085,35 @@ static void claim_allocations(int cpu, s
 	}
 }
 
-static struct sched_domain *__build_numa_sched_domains(struct s_data *d,
-	const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i)
+static struct sched_domain *__build_allnodes_sched_domain(struct s_data *d,
+	const struct cpumask *cpu_map, struct sched_domain_attr *attr,
+	struct sched_domain *parent, int i)
 {
 	struct sched_domain *sd = NULL;
 #ifdef CONFIG_NUMA
-	struct sched_domain *parent;
-
-	d->sd_allnodes = 0;
-	if (cpumask_weight(cpu_map) >
-	    SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) {
-		sd = sd_init_ALLNODES(d, i);
-		set_domain_attribute(sd, attr);
-		cpumask_copy(sched_domain_span(sd), cpu_map);
-		d->sd_allnodes = 1;
-	}
-	parent = sd;
+	sd = sd_init_ALLNODES(d, i);
+	set_domain_attribute(sd, attr);
+	cpumask_copy(sched_domain_span(sd), cpu_map);
+	sd->parent = parent;
+	if (parent)
+		parent->child = sd;
+#endif
+	return sd;
+}
 
+static struct sched_domain *__build_node_sched_domain(struct s_data *d,
+	const struct cpumask *cpu_map, struct sched_domain_attr *attr,
+	struct sched_domain *parent, int i)
+{
+	struct sched_domain *sd = NULL;
+#ifdef CONFIG_NUMA
 	sd = sd_init_NODE(d, i);
 	set_domain_attribute(sd, attr);
 	sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd));
+	cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map);
 	sd->parent = parent;
 	if (parent)
 		parent->child = sd;
-	cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map);
 #endif
 	return sd;
 }
@@ -7196,7 +7198,9 @@ static int __build_sched_domains(const s
 		cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)),
 			    cpu_map);
 
-		sd = __build_numa_sched_domains(&d, cpu_map, attr, i);
+		sd = NULL;
+		sd = __build_allnodes_sched_domain(&d, cpu_map, attr, sd, i);
+		sd = __build_node_sched_domain(&d, cpu_map, attr, sd, i);
 		sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i);
 		sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i);
 		sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);

[RFC][PATCH 13/14] sched: Remove nodemask allocation

[Peter Zijlstra]

[-- Attachment #1: sched-foo12.patch --]
[-- Type: text/plain, Size: 2155 bytes --]

There's only one nodemask user left so remove it with a direct
computation and save some memory and reduce some code-flow
complexity.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 kernel/sched.c |   14 +++-----------
 1 file changed, 3 insertions(+), 11 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -6814,7 +6814,6 @@ struct sd_data {
 };
 
 struct s_data {
-	cpumask_var_t		nodemask;
 	cpumask_var_t		send_covered;
 	struct sched_domain ** __percpu sd;
 	struct sd_data 		sdd[SD_LV_MAX];
@@ -6826,7 +6825,6 @@ enum s_alloc {
 	sa_sd,
 	sa_sd_storage,
 	sa_send_covered,
-	sa_nodemask,
 	sa_none,
 };
 
@@ -7011,8 +7009,6 @@ static void __free_domain_allocs(struct 
 		} /* fall through */
 	case sa_send_covered:
 		free_cpumask_var(d->send_covered); /* fall through */
-	case sa_nodemask:
-		free_cpumask_var(d->nodemask); /* fall through */
 	case sa_none:
 		break;
 	}
@@ -7025,10 +7021,8 @@ static enum s_alloc __visit_domain_alloc
 
 	memset(d, 0, sizeof(*d));
 
-	if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL))
-		return sa_none;
 	if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
-		return sa_nodemask;
+		return sa_none;
 	for (i = 0; i < SD_LV_MAX; i++) {
 		d->sdd[i].sd = alloc_percpu(struct sched_domain *);
 		if (!d->sdd[i].sd)
@@ -7125,7 +7119,8 @@ static struct sched_domain *__build_cpu_
 	struct sched_domain *sd;
 	sd = sd_init_CPU(d, i);
 	set_domain_attribute(sd, attr);
-	cpumask_copy(sched_domain_span(sd), d->nodemask);
+	cpumask_and(sched_domain_span(sd),
+			cpumask_of_node(cpu_to_node(i)), cpu_map);
 	sd->parent = parent;
 	if (parent)
 		parent->child = sd;
@@ -7195,9 +7190,6 @@ static int __build_sched_domains(const s
 
 	/* Set up domains for cpus specified by the cpu_map. */
 	for_each_cpu(i, cpu_map) {
-		cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)),
-			    cpu_map);
-
 		sd = NULL;
 		sd = __build_allnodes_sched_domain(&d, cpu_map, attr, sd, i);
 		sd = __build_node_sched_domain(&d, cpu_map, attr, sd, i);

[RFC][PATCH 14/14] sched: Remove some dead code

[Peter Zijlstra]

[-- Attachment #1: sched-foo13.patch --]
[-- Type: text/plain, Size: 1967 bytes --]


Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
---
 include/linux/sched.h |    6 ------
 kernel/sched.c        |   16 ----------------
 2 files changed, 22 deletions(-)

Index: linux-2.6/include/linux/sched.h
===================================================================
--- linux-2.6.orig/include/linux/sched.h
+++ linux-2.6/include/linux/sched.h
@@ -881,9 +881,6 @@ struct sched_group {
 	 * NOTE: this field is variable length. (Allocated dynamically
 	 * by attaching extra space to the end of the structure,
 	 * depending on how many CPUs the kernel has booted up with)
-	 *
-	 * It is also be embedded into static data structures at build
-	 * time. (See 'struct static_sched_group' in kernel/sched.c)
 	 */
 	unsigned long cpumask[0];
 };
@@ -992,9 +989,6 @@ struct sched_domain {
 	 * NOTE: this field is variable length. (Allocated dynamically
 	 * by attaching extra space to the end of the structure,
 	 * depending on how many CPUs the kernel has booted up with)
-	 *
-	 * It is also be embedded into static data structures at build
-	 * time. (See 'struct static_sched_domain' in kernel/sched.c)
 	 */
 	unsigned long span[0];
 };
Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -6793,22 +6793,6 @@ static void sched_domain_node_span(int n
 
 int sched_smt_power_savings = 0, sched_mc_power_savings = 0;
 
-/*
- * The cpus mask in sched_group and sched_domain hangs off the end.
- *
- * ( See the the comments in include/linux/sched.h:struct sched_group
- *   and struct sched_domain. )
- */
-struct static_sched_group {
-	struct sched_group sg;
-	DECLARE_BITMAP(cpus, CONFIG_NR_CPUS);
-};
-
-struct static_sched_domain {
-	struct sched_domain sd;
-	DECLARE_BITMAP(span, CONFIG_NR_CPUS);
-};
-
 struct sd_data {
 	struct sched_domain **__percpu sd;
 	struct sched_group **__percpu sg;

Re: [RFC][PATCH 00/14] Rewrite sched_domain/sched_group creation

[Peter Zijlstra]

On Mon, 2011-03-14 at 16:06 +0100, Peter Zijlstra wrote:
> So I got annoyed with the whole sched_domain/sched_group creation mess again
> and decided to actuall do something about it. See here ;-)
> 
> Its not completely done yet, but much of the ground-work is there.
> 
> The final goal is to be able to have a simple way to dynamically specify the
> architecture topoplogy and have the generic code in charge of building the data
> structures. The architecture would need to provide a function that maps a cpu
> to a cpumask and a function that initializes the sched_domain (flags etc.),
> this would replace the now still hard-coded __build_*_sched_domain() calls in
> __build_sched_domains().

Got to here last friday a week ago, and started on:

> With that we can also kill the current NODE and ALLNODES mess and generate
> appropriate masks from say the ACPI SLIT table by grouping CPUs on their node
> distance.

which I just finished.

> This would allow us to kill the horrid mess in x86's cpu_coregroup_mask() and
> properly support the AMD magnycours stuff.

Still todo.

> Anyway, not quite there yet.. patches build and boot on a 2*6*2 wsm box.

Still boots, although a 2 node system isn't very interesting.

I pushed out an updated version to:

  git://git.kernel.org/pub/scm/linux/kernel/git/peterz/linux-2.6-sched.git sched_domain

we're now at:

 include/linux/sched.h    |   26 +-
 include/linux/topology.h |   25 -
 kernel/cpuset.c          |    2 +-
 kernel/sched.c           | 1094 ++++++++++++++++------------------------------
 kernel/sched_fair.c      |   32 +-
 5 files changed, 407 insertions(+), 772 deletions(-)

Re: [RFC][PATCH 00/14] Rewrite sched_domain/sched_group creation

[Peter Zijlstra]

On Fri, 2011-03-25 at 22:46 +0100, Peter Zijlstra wrote:
> > Anyway, not quite there yet.. patches build and boot on a 2*6*2 wsm
> box.
> 
> Still boots, although a 2 node system isn't very interesting.
> 
OK, halt made the domain stuff explode (also didn't test hotplug), so
there's something fishy.