--- Documentation/sysctl/kernel.txt | 2 - kernel/sched.c | 45 ++++++++++++++++++++-------------------- kernel/softirq.c | 2 - kernel/sysctl.c | 12 ++++++---- 4 files changed, 32 insertions(+), 29 deletions(-) Index: linux-2.6.20.7-sd/kernel/sched.c =================================================================== --- linux-2.6.20.7-sd.orig/kernel/sched.c 2007-04-19 12:17:46.000000000 +1000 +++ linux-2.6.20.7-sd/kernel/sched.c 2007-04-20 00:45:05.000000000 +1000 @@ -53,6 +53,7 @@ #include #include #include +#include #include #include @@ -79,7 +80,7 @@ /* Some helpers for converting to/from various scales.*/ #define NS_TO_JIFFIES(TIME) ((TIME) / (1000000000 / HZ)) #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ)) -#define MS_TO_NS(TIME) ((TIME) * 1000000) +#define MS_TO_US(TIME) ((TIME) * 1000) /* Can return 0 */ #define MS_TO_JIFFIES(TIME) ((TIME) * HZ / 1000) #define JIFFIES_TO_MS(TIME) ((TIME) * 1000 / HZ) @@ -91,9 +92,8 @@ * Value is in ms and set to a minimum of 8ms. Scales with number of cpus. * Tunable via /proc interface. */ -int rr_interval __read_mostly; +int rr_interval __read_mostly = 8; -#define RR_INTERVAL 8 #define DEF_TIMESLICE (rr_interval * 20) /* @@ -975,23 +975,20 @@ static int effective_prio(struct task_st * tick still. Below nice 0 they get progressively larger. * ie nice -6..0 = rr_interval. nice -10 = 2.5 * rr_interval * nice -20 = 10 * rr_interval. nice 1-19 = rr_interval / 2. - * Value returned is in nanoseconds. + * Value returned is in microseconds. */ static unsigned int rr_quota(struct task_struct *p) { int nice = TASK_NICE(p), rr = rr_interval; - /* Ensure that rr_interval is at least 1 tick */ - if (unlikely(!MS_TO_JIFFIES(rr))) - rr = rr_interval = JIFFIES_TO_MS(1) ? : 1; if (!rt_task(p)) { if (nice < -6) { rr *= nice * nice; rr /= 40; - } else if (nice > 0 && (rr * HZ / 1000 / 2) > 0) - rr /= 2; + } else if (nice > 0) + rr = rr / 2 ? : 1; } - return MS_TO_NS(rr); + return MS_TO_US(rr); } /* @@ -3010,16 +3007,17 @@ EXPORT_PER_CPU_SYMBOL(kstat); /* * This is called on clock ticks and on context switches. * Bank in p->sched_time the ns elapsed since the last tick or switch. - * CPU scheduler quota accounting is also performed here. + * CPU scheduler quota accounting is also performed here in microseconds. * The value returned from sched_clock() occasionally gives bogus values so * some sanity checking is required. */ -static inline void +static void update_cpu_clock(struct task_struct *p, struct rq *rq, unsigned long long now, int tick) { cputime64_t time_diff = now - p->last_ran; - unsigned int min_diff = 1000; + const unsigned int min_diff = 1000; + int us_time_diff; if (tick) { /* @@ -3038,8 +3036,11 @@ update_cpu_clock(struct task_struct *p, if (time_diff > JIFFIES_TO_NS(1) || time_diff < min_diff) time_diff = min_diff; } + /* time_slice accounting is done in usecs to avoid overflow on 32bit */ + us_time_diff = time_diff; + us_time_diff /= 1000; if (p != rq->idle && p->policy != SCHED_FIFO) - p->time_slice -= time_diff; + p->time_slice -= us_time_diff; p->sched_time += time_diff; p->last_ran = rq->most_recent_timestamp = now; } @@ -3140,8 +3141,7 @@ void account_steal_time(struct task_stru static void task_expired_entitlement(struct rq *rq, struct task_struct *p) { struct prio_array *old_array; - int overrun; - int old_prio; + int overrun, old_prio; if (unlikely(p->first_time_slice)) p->first_time_slice = 0; @@ -6826,6 +6826,13 @@ void __init sched_init_smp(void) /* Move init over to a non-isolated CPU */ if (set_cpus_allowed(current, non_isolated_cpus) < 0) BUG(); + + /* + * Assume that every added cpu gives us slightly less overall latency + * allowing us to increase the base rr_interval, but in a non linear + * fashion. + */ + rr_interval *= 1 + ilog2(num_online_cpus()); } #else void __init sched_init_smp(void) @@ -6846,7 +6853,6 @@ int in_sched_functions(unsigned long add void __init sched_init(void) { int i, j, k; - unsigned int rr_us = 0, rr_inc = RR_INTERVAL * 1000; /* Generate the priority matrix */ for (i = 0; i < PRIO_RANGE; i++) { @@ -6897,12 +6903,7 @@ void __init sched_init(void) __set_bit(MAX_PRIO, array->prio_bitmap); } - /* Every added cpu increases the rr_interval */ - rr_us += rr_inc; - rr_inc /= 2; } - rr_interval = rr_us / 1000; - set_load_weight(&init_task); #ifdef CONFIG_SMP Index: linux-2.6.20.7-sd/kernel/sysctl.c =================================================================== --- linux-2.6.20.7-sd.orig/kernel/sysctl.c 2007-04-19 09:51:30.000000000 +1000 +++ linux-2.6.20.7-sd/kernel/sysctl.c 2007-04-20 00:45:59.000000000 +1000 @@ -229,11 +229,13 @@ static void register_proc_table(ctl_tabl static void unregister_proc_table(ctl_table *, struct proc_dir_entry *); #endif -/* Constants for minimum and maximum testing in vm_table. + +/* Constants for minimum and maximum testing. We use these as one-element integer vectors. */ -static int __read_mostly zero; -static int __read_mostly one = 1; -static int __read_mostly one_hundred = 100; +static int __read_mostly zero; +static int __read_mostly one = 1; +static int __read_mostly one_hundred = 100; +static int __read_mostly five_thousand = 5000; /* The default sysctl tables: */ @@ -693,7 +695,7 @@ static ctl_table kern_table[] = { .proc_handler = &proc_dointvec_minmax, .strategy = &sysctl_intvec, .extra1 = &one, - .extra2 = &one_hundred, + .extra2 = &five_thousand, }, #if defined(CONFIG_X86_LOCAL_APIC) && defined(CONFIG_X86) { Index: linux-2.6.20.7-sd/Documentation/sysctl/kernel.txt =================================================================== --- linux-2.6.20.7-sd.orig/Documentation/sysctl/kernel.txt 2007-04-19 09:51:30.000000000 +1000 +++ linux-2.6.20.7-sd/Documentation/sysctl/kernel.txt 2007-04-20 00:45:05.000000000 +1000 @@ -298,7 +298,7 @@ overall. This value is in milliseconds a depends on the number of cpus available at scheduler initialisation with a minimum of 8. -Valid values are from 1-100. +Valid values are from 1-5000. ============================================================== Index: linux-2.6.20.7-sd/kernel/softirq.c =================================================================== --- linux-2.6.20.7-sd.orig/kernel/softirq.c 2007-02-05 22:52:04.000000000 +1100 +++ linux-2.6.20.7-sd/kernel/softirq.c 2007-04-20 00:45:05.000000000 +1000 @@ -469,7 +469,7 @@ void __init softirq_init(void) static int ksoftirqd(void * __bind_cpu) { - set_user_nice(current, 19); + set_user_nice(current, 15); current->flags |= PF_NOFREEZE; set_current_state(TASK_INTERRUPTIBLE);