sched: mc/smt power savings sched policy
Subsystems (2)
PATCH 1/1 sched: mc/smt power savings sched policy
Appended the patch. Can someone please test compile the powerpc change?
thanks,
suresh
--
sysfs entries 'sched_mc_power_savings' and 'sched_smt_power_savings' in /sys/devices/system/cpu/ control the MC/SMT power savings policy for the scheduler.
Based on the values (1-enable, 0-disable) for these controls, sched groups cpu power will be determined for different domains. When power savings policy is enabled and under light load conditions, scheduler will minimize the physical packages/cpu cores carrying the load and thus conserving power(with a perf impact based on the workload characteristics... see OLS 2005 CMP kernel scheduler paper for more details..)
arch/i386/kernel/smpboot.c +5 -3
@@ -446,10 +446,12 @@ cpumask_t cpu_coregroup_map(int cpu)struct cpuinfo_x86 *c = cpu_data + cpu;/* * For perf, we return last level cache shared map. -* TBD: when power saving sched policy is added, we will return -* cpu_core_map when power saving policy is enabled +* And for power savings, we return cpu_core_map */ -return c->llc_shared_map; +if (sched_mc_power_savings || sched_smt_power_savings) +return cpu_core_map[cpu]; +else +return c->llc_shared_map; }/* representing cpus for which sibling maps can be computed */
arch/x86_64/kernel/smpboot.c +5 -3
@@ -450,10 +450,12 @@ cpumask_t cpu_coregroup_map(int cpu)struct cpuinfo_x86 *c = cpu_data + cpu;/* * For perf, we return last level cache shared map. -* TBD: when power saving sched policy is added, we will return -* cpu_core_map when power saving policy is enabled +* And for power savings, we return cpu_core_map */ -return c->llc_shared_map; +if (sched_mc_power_savings || sched_smt_power_savings) +return cpu_core_map[cpu]; +else +return c->llc_shared_map; }/* representing cpus for which sibling maps can be computed */
drivers/base/cpu.c +9 -1
@@ -143,5 +143,13 @@ EXPORT_SYMBOL_GPL(get_cpu_sysdev);int __init cpu_dev_init (void ) { -return sysdev_class_register (&cpu_sysdev_class); +int err; + + err =sysdev_class_register (&cpu_sysdev_class); +#if defined(CONFIG_SCHED_MC ) || defined(CONFIG_SCHED_SMT ) +if (!err) + err =sched_create_sysfs_power_savings_entries (&cpu_sysdev_class); +#endif + +return err; }
include/asm-i386/topology.h +5 -0
@@ -112,4 +112,9 @@ extern unsigned long node_remap_size[];extern cpumask_t cpu_coregroup_map (int cpu); +#ifdef CONFIG_SMP +#define mc_capable () (boot_cpu_data.x86_max_cores > 1) +#define smt_capable () (smp_num_siblings > 1) +#endif +#endif /* _ASM_I386_TOPOLOGY_H */
include/asm-ia64/topology.h +1 -0
@@ -112,6 +112,7 @@ void build_cpu_to_node_map(void);#define topology_core_id (cpu) (cpu_data(cpu)->core_id)#define topology_core_siblings (cpu) (cpu_core_map[cpu])#define topology_thread_siblings (cpu) (cpu_sibling_map[cpu]) +#define smt_capable () (smp_num_siblings > 1)#endif #include <asm-generic/topology.h>
include/asm-powerpc/topology.h +5 -0
@@ -88,5 +88,10 @@ static inline void sysfs_remove_device_f#endif /* CONFIG_NUMA */ +#ifdef CONFIG_SMP +#include <asm/cputable.h> +#define smt_capable () (cpu_has_feature(CPU_FTR_SMT)) +#endif +#endif /* __KERNEL__ */ #endif /* _ASM_POWERPC_TOPOLOGY_H */
include/asm-x86_64/topology.h +2 -0
@@ -59,6 +59,8 @@ extern int __node_distance(int, int);#define topology_core_id (cpu) (cpu_data[cpu].cpu_core_id)#define topology_core_siblings (cpu) (cpu_core_map[cpu])#define topology_thread_siblings (cpu) (cpu_sibling_map[cpu]) +#define mc_capable () (boot_cpu_data.x86_max_cores > 1) +#define smt_capable () (smp_num_siblings > 1)#endif #include <asm-generic/topology.h>
include/linux/sched.h +10 -0
@@ -610,6 +610,11 @@ enum idle_type#define SD_WAKE_AFFINE 32/* Wake task to waking CPU */ #define SD_WAKE_BALANCE 64/* Perform balancing at task wakeup */ #define SD_SHARE_CPUPOWER 128/* Domain members share cpu power */ +#define SD_POWERSAVINGS_BALANCE 256/* Balance for power savings */ + +#define BALANCE_FOR_POWER ((sched_mc_power_savings || sched_smt_power_savings) \ + ? SD_POWERSAVINGS_BALANCE : 0) +struct sched_group {struct sched_group *next;/* Must be a circular list */
@@ -1527,6 +1532,11 @@ static inline void arch_pick_mmap_layoutextern long sched_setaffinity (pid_t pid,cpumask_t new_mask);extern long sched_getaffinity (pid_t pid,cpumask_t *mask); +#include <linux/sysdev.h> +extern int sched_mc_power_savings, sched_smt_power_savings; +extern struct sysdev_attribute attr_sched_mc_power_savings, attr_sched_smt_power_savings; +extern int sched_create_sysfs_power_savings_entries (struct sysdev_class *cls); +extern void normalize_rt_tasks (void );#ifdef CONFIG_PM
include/linux/topology.h +2 -1
@@ -134,7 +134,8 @@ .flags = SD_LOAD_BALANCE \ |SD_BALANCE_NEWIDLE \ |SD_BALANCE_EXEC \ - | SD_WAKE_AFFINE, \ + |SD_WAKE_AFFINE \ + | BALANCE_FOR_POWER, \ .last_balance = jiffies, \ .balance_interval = 1, \ .nr_balance_failed = 0, \
kernel/sched.c +216 -26
344 lines not on this page · raw message
Quoted 10 lines · Suresh B
Appended the patch. Can someone please test compile the powerpc change?
thanks,
suresh
--sysfs entries '
sched_mc_power_savings' and 'sched_smt_power_savings' in/sys/devices/system/cpu/control the MC/SMT power savings policy for the scheduler.
Quoted 6 lines
Based on the values (1-enable, 0-disable) for these controls, sched groups cpu power will be determined for different domains. When power savings policy is enabled and under light load conditions, scheduler will minimize the physical packages/cpu cores carrying the load and thus conserving power(with a perf impact based on the workload characteristics... see OLS 2005 CMP kernel scheduler paper for more details..)
This MC code is a maze of #ifdefs within functions and getting harder to follow with each subsequent patch. Can we not deviate so much from kernel style?
Signature · 1 line
-ck
Quoted 3 lines · Con Kolivas
This MC code is a maze of #ifdefs within functions and getting harder to follow with each subsequent patch. Can we not deviate so much from kernel style?
build_sched_domains() originally was not designed with this many sched domains in mind.. I can take a look at cleaning up of the sched domains setup..
thanks,
suresh
Quoted 1 line · Suresh B
Appended the patch. Can someone please test compile the powerpc change?
powerpc compiles and boots OK, but sparc64 is not so good.
kernel/built-in.o(.text+0x6ec0): In function `sched_create_sysfs_power_savings_entries':
: undefined reference to `smt_capable'
Quoted 9 lines · Andrew Morton
Quoted 1 line · Suresh B
Appended the patch. Can someone please test compile the powerpc change?
powerpc compiles and boots OK, but sparc64 is not so good.
kernel/built-in.o(.text+0x6ec0): In function `sched_create_sysfs_power_savings_entries':
: undefined reference to `smt_capable'
Dave, I am not sure if the appended patch is the correct(perhaps temporary?) mechanism(currently smp_prepare_cpus() also uses this mechanism) in identifying a SMT capable sparc64 processor. Can you confirm?
thanks,
suresh
--
include/asm-sparc64/topology.h +3 -0
@@ -1,6 +1,9 @@#ifndef _ASM_SPARC64_TOPOLOGY_H#define _ASM_SPARC64_TOPOLOGY_H +#include <asm/spitfire.h> +#define smt_capable () (tlb_type == hypervisor) +#include <asm-generic/topology.h> #endif /* _ASM_SPARC64_TOPOLOGY_H */