sched: arch_reinit_sched_domains() must destroy domains to force rebuild
PATCH 1/1 sched: arch_reinit_sched_domains() must destroy domains to force rebuild
What I realized recently is that calling rebuild_sched_domains() in arch_reinit_sched_domains() by itself is not enough when cpusets are enabled. partition_sched_domains() code is trying to avoid unnecessary domain rebuilds and will not actually rebuild anything if new domain masks match the old ones.
What this means is that doing
echo 1 > /sys/devices/system/cpu/sched_mc_power_savings
on a system with cpusets enabled will not take affect untill something changes in the cpuset setup (ie new sets created or deleted).
This patch fixes restore correct behaviour where domains must be rebuilt in order to enable MC powersaving flags.
Test on quad-core Core2 box with both CONFIG_CPUSETS and !CONFIG_CPUSETS. Also tested on dual-core Core2 laptop. Lockdep is happy and things are working as expected.
Ingo, please apply.
btw We also need to push my other cpuset patch into mainline. We currently calling rebuild_sched_domains() without cgroup lock which is bad. When I made original sched: changes the assumption was that cpuset patch will also go in. I'm talking about
"cpuset: Rework sched domains and CPU hotplug handling"
It's been ACKed by Paul has been in the -tip for awhile now.
--- include/linux/cpuset.h | 2 +- kernel/sched.c | 19 +++++++++++++------ 2 files changed, 14 insertions(+), 7 deletions(-)
include/linux/cpuset.h +1 -1
@@ -160,7 +160,7 @@ static inline int current_cpuset_is_being_rebound(void)static inline void rebuild_sched_domains (void ) { -partition_sched_domains (0 ,NULL ,NULL ); +partition_sched_domains (1 ,NULL ,NULL ); }#endif /* !CONFIG_CPUSETS */
kernel/sched.c +13 -6
@@ -7637,24 +7637,27 @@ static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,* and partition_sched_domains() will fallback to the single partition * 'fallback_doms', it also forces the domains to be rebuilt. * +* If doms_new==NULL it will be replaced with cpu_online_map. +* ndoms_new==0 is a special case for destroying existing domains. +* It will not create the default domain. +* * Call with hotplug lock held */ void partition_sched_domains (int ndoms_new,cpumask_t *doms_new,struct sched_domain_attr *dattr_new) { -int i, j; +int i, j, n;mutex_lock (&sched_domains_mutex);/* always unregister in case we don't destroy any domains */ unregister_sched_domain_sysctl (); -if (doms_new ==NULL ) - ndoms_new =0 ; + n = doms_new ? ndoms_new :0 ;/* Destroy deleted domains */ for (i =0 ; i < ndoms_cur; i++) { -for (j =0 ; j < ndoms_new; j++) { +for (j =0 ; j < n; j++) {if (cpus_equal (doms_cur[i], doms_new[j]) &&dattrs_equal (dattr_cur, i, dattr_new, j)) goto match1;
@@ -7667,7 +7670,6 @@ match1:if (doms_new ==NULL ) { ndoms_cur =0 ; - ndoms_new =1 ; doms_new = &fallback_doms;cpus_andnot (doms_new[0 ], cpu_online_map, cpu_isolated_map); dattr_new =NULL ;
@@ -7704,8 +7706,13 @@ match2:int arch_reinit_sched_domains (void ) {get_online_cpus (); + +/* Destroy domains first to force the rebuild */ +partition_sched_domains (0 ,NULL ,NULL ); +rebuild_sched_domains ();put_online_cpus (); +return 0 ; }
@@ -7789,7 +7796,7 @@ static int update_sched_domains(struct notifier_block *nfb,case CPU_ONLINE_FROZEN :case CPU_DEAD :case CPU_DEAD_FROZEN : -partition_sched_domains (0 ,NULL ,NULL ); +partition_sched_domains (1 ,NULL ,NULL );return NOTIFY_OK ;default :
Signature · 1 line
1.5.5.1
Quoted 25 lines · Max Krasnyansky
What I realized recently is that calling
rebuild_sched_domains()inarch_reinit_sched_domains()by itself is not enough when cpusets are enabled.partition_sched_domains()code is trying to avoid unnecessary domain rebuilds and will not actually rebuild anything if new domain masks match the old ones.What this means is that doing
echo 1 >/sys/devices/system/cpu/sched_mc_power_savings
on a system with cpusets enabled will not take affect untill something changes in the cpuset setup (ie new sets created or deleted).This patch fixes restore correct behaviour where domains must be rebuilt in order to enable MC powersaving flags.
Test on quad-core Core2 box with both
CONFIG_CPUSETSand !CONFIG_CPUSETS. Also tested on dual-core Core2 laptop. Lockdep is happy and things are working as expected.Ingo, please apply.
btw We also need to push my other cpuset patch into mainline. We currently callingrebuild_sched_domains()without cgroup lock which is bad. When I made original sched: changes the assumption was that cpuset patch will also go in. I'm talking about
"cpuset: Rework sched domains and CPU hotplug handling"
It's been ACKed by Paul has been in the -tip for awhile now.
Quoted 4 lines
Signed-off-by Max Krasnyansky <[email protected]>
This fixes the problem on my 32-bit x86 quad-core, dual socket system. The sched domains are being rebuilt accurately as expected on changing sched_mc_power_savings.
However, the fix as such should be further reviewed for correctness. I verified only the rebuilding of sched domains in my system.
Thanks,
Vaidy
Quoted 88 lines
--- include/linux/cpuset.h | 2 +- kernel/sched.c | 19 +++++++++++++------ 2 files changed, 14 insertions(+), 7 deletions(-)include/linux/cpuset.h +1
-1@@ -160,7 +160,7 @@ static inline int current_cpuset_is_being_rebound(void)static inline void rebuild_sched_domains (void ) { -partition_sched_domains (0 ,NULL ,NULL ); +partition_sched_domains (1 ,NULL ,NULL ); }#endif /* !CONFIG_CPUSETS */ kernel/sched.c +13
-6@@ -7637,24 +7637,27 @@ static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur,* and partition_sched_domains() will fallback to the single partition * 'fallback_doms', it also forces the domains to be rebuilt. * +* If doms_new==NULL it will be replaced with cpu_online_map. +* ndoms_new==0 is a special case for destroying existing domains. +* It will not create the default domain. +* * Call with hotplug lock held */ void partition_sched_domains (int ndoms_new,cpumask_t *doms_new,struct sched_domain_attr *dattr_new) { -int i, j; +int i, j, n;mutex_lock (&sched_domains_mutex);/* always unregister in case we don't destroy any domains */ unregister_sched_domain_sysctl (); -if (doms_new ==NULL ) - ndoms_new =0 ; + n = doms_new ? ndoms_new :0 ;/* Destroy deleted domains */ for (i =0 ; i < ndoms_cur; i++) { -for (j =0 ; j < ndoms_new; j++) { +for (j =0 ; j < n; j++) {if (cpus_equal (doms_cur[i], doms_new[j]) &&dattrs_equal (dattr_cur, i, dattr_new, j)) goto match1;@@ -7667,7 +7670,6 @@ match1:if (doms_new ==NULL ) { ndoms_cur =0 ; - ndoms_new =1 ; doms_new = &fallback_doms;cpus_andnot (doms_new[0 ], cpu_online_map, cpu_isolated_map); dattr_new =NULL ;@@ -7704,8 +7706,13 @@ match2:int arch_reinit_sched_domains (void ) {get_online_cpus (); + +/* Destroy domains first to force the rebuild */ +partition_sched_domains (0 ,NULL ,NULL ); +rebuild_sched_domains ();put_online_cpus (); +return 0 ; }@@ -7789,7 +7796,7 @@ static int update_sched_domains(struct notifier_block *nfb,case CPU_ONLINE_FROZEN :case CPU_DEAD :case CPU_DEAD_FROZEN : -partition_sched_domains (0 ,NULL ,NULL ); +partition_sched_domains (1 ,NULL ,NULL );return NOTIFY_OK ;default :Signature · 1 line
1.5.5.1
Ingo Molnar
AppliedQuoted 10 lines · Max Krasnyansky
What I realized recently is that calling
rebuild_sched_domains()inarch_reinit_sched_domains()by itself is not enough when cpusets are enabled.partition_sched_domains()code is trying to avoid unnecessary domain rebuilds and will not actually rebuild anything if new domain masks match the old ones.What this means is that doing
echo 1 >/sys/devices/system/cpu/sched_mc_power_savings
on a system with cpusets enabled will not take affect untill something changes in the cpuset setup (ie new sets created or deleted).
Quoted 8 lines
This patch fixes restore correct behaviour where domains must be rebuilt in order to enable MC powersaving flags.
Test on quad-core Core2 box with both
CONFIG_CPUSETSand !CONFIG_CPUSETS. Also tested on dual-core Core2 laptop. Lockdep is happy and things are working as expected.Ingo, please apply.
applied to tip/sched/cpuset, thanks Max.
Quoted 1 line
btw We also need to push my other cpuset patch into mainline. [...]
will push it once this has been tested some more.
Ingo