aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/sched.c
diff options
context:
space:
mode:
authorPeter Williams <pwil3058@bigpond.net.au>2006-06-27 05:54:34 -0400
committerLinus Torvalds <torvalds@g5.osdl.org>2006-06-27 20:32:44 -0400
commit2dd73a4f09beacadde827a032cf15fd8b1fa3d48 (patch)
treef81752d44e68240231518d6a3f05ac9ff6410a2d /kernel/sched.c
parentefc30814a88bdbe2bfe4ac94de2eb089ad80bee3 (diff)
[PATCH] sched: implement smpnice
Problem: The introduction of separate run queues per CPU has brought with it "nice" enforcement problems that are best described by a simple example. For the sake of argument suppose that on a single CPU machine with a nice==19 hard spinner and a nice==0 hard spinner running that the nice==0 task gets 95% of the CPU and the nice==19 task gets 5% of the CPU. Now suppose that there is a system with 2 CPUs and 2 nice==19 hard spinners and 2 nice==0 hard spinners running. The user of this system would be entitled to expect that the nice==0 tasks each get 95% of a CPU and the nice==19 tasks only get 5% each. However, whether this expectation is met is pretty much down to luck as there are four equally likely distributions of the tasks to the CPUs that the load balancing code will consider to be balanced with loads of 2.0 for each CPU. Two of these distributions involve one nice==0 and one nice==19 task per CPU and in these circumstances the users expectations will be met. The other two distributions both involve both nice==0 tasks being on one CPU and both nice==19 being on the other CPU and each task will get 50% of a CPU and the user's expectations will not be met. Solution: The solution to this problem that is implemented in the attached patch is to use weighted loads when determining if the system is balanced and, when an imbalance is detected, to move an amount of weighted load between run queues (as opposed to a number of tasks) to restore the balance. Once again, the easiest way to explain why both of these measures are necessary is to use a simple example. Suppose that (in a slight variation of the above example) that we have a two CPU system with 4 nice==0 and 4 nice=19 hard spinning tasks running and that the 4 nice==0 tasks are on one CPU and the 4 nice==19 tasks are on the other CPU. The weighted loads for the two CPUs would be 4.0 and 0.2 respectively and the load balancing code would move 2 tasks resulting in one CPU with a load of 2.0 and the other with load of 2.2. If this was considered to be a big enough imbalance to justify moving a task and that task was moved using the current move_tasks() then it would move the highest priority task that it found and this would result in one CPU with a load of 3.0 and the other with a load of 1.2 which would result in the movement of a task in the opposite direction and so on -- infinite loop. If, on the other hand, an amount of load to be moved is calculated from the imbalance (in this case 0.1) and move_tasks() skips tasks until it find ones whose contributions to the weighted load are less than this amount it would move two of the nice==19 tasks resulting in a system with 2 nice==0 and 2 nice=19 on each CPU with loads of 2.1 for each CPU. One of the advantages of this mechanism is that on a system where all tasks have nice==0 the load balancing calculations would be mathematically identical to the current load balancing code. Notes: struct task_struct: has a new field load_weight which (in a trade off of space for speed) stores the contribution that this task makes to a CPU's weighted load when it is runnable. struct runqueue: has a new field raw_weighted_load which is the sum of the load_weight values for the currently runnable tasks on this run queue. This field always needs to be updated when nr_running is updated so two new inline functions inc_nr_running() and dec_nr_running() have been created to make sure that this happens. This also offers a convenient way to optimize away this part of the smpnice mechanism when CONFIG_SMP is not defined. int try_to_wake_up(): in this function the value SCHED_LOAD_BALANCE is used to represent the load contribution of a single task in various calculations in the code that decides which CPU to put the waking task on. While this would be a valid on a system where the nice values for the runnable tasks were distributed evenly around zero it will lead to anomalous load balancing if the distribution is skewed in either direction. To overcome this problem SCHED_LOAD_SCALE has been replaced by the load_weight for the relevant task or by the average load_weight per task for the queue in question (as appropriate). int move_tasks(): The modifications to this function were complicated by the fact that active_load_balance() uses it to move exactly one task without checking whether an imbalance actually exists. This precluded the simple overloading of max_nr_move with max_load_move and necessitated the addition of the latter as an extra argument to the function. The internal implementation is then modified to move up to max_nr_move tasks and max_load_move of weighted load. This slightly complicates the code where move_tasks() is called and if ever active_load_balance() is changed to not use move_tasks() the implementation of move_tasks() should be simplified accordingly. struct sched_group *find_busiest_group(): Similar to try_to_wake_up(), there are places in this function where SCHED_LOAD_SCALE is used to represent the load contribution of a single task and the same issues are created. A similar solution is adopted except that it is now the average per task contribution to a group's load (as opposed to a run queue) that is required. As this value is not directly available from the group it is calculated on the fly as the queues in the groups are visited when determining the busiest group. A key change to this function is that it is no longer to scale down *imbalance on exit as move_tasks() uses the load in its scaled form. void set_user_nice(): has been modified to update the task's load_weight field when it's nice value and also to ensure that its run queue's raw_weighted_load field is updated if it was runnable. From: "Siddha, Suresh B" <suresh.b.siddha@intel.com> With smpnice, sched groups with highest priority tasks can mask the imbalance between the other sched groups with in the same domain. This patch fixes some of the listed down scenarios by not considering the sched groups which are lightly loaded. a) on a simple 4-way MP system, if we have one high priority and 4 normal priority tasks, with smpnice we would like to see the high priority task scheduled on one cpu, two other cpus getting one normal task each and the fourth cpu getting the remaining two normal tasks. but with current smpnice extra normal priority task keeps jumping from one cpu to another cpu having the normal priority task. This is because of the busiest_has_loaded_cpus, nr_loaded_cpus logic.. We are not including the cpu with high priority task in max_load calculations but including that in total and avg_load calcuations.. leading to max_load < avg_load and load balance between cpus running normal priority tasks(2 Vs 1) will always show imbalanace as one normal priority and the extra normal priority task will keep moving from one cpu to another cpu having normal priority task.. b) 4-way system with HT (8 logical processors). Package-P0 T0 has a highest priority task, T1 is idle. Package-P1 Both T0 and T1 have 1 normal priority task each.. P2 and P3 are idle. With this patch, one of the normal priority tasks on P1 will be moved to P2 or P3.. c) With the current weighted smp nice calculations, it doesn't always make sense to look at the highest weighted runqueue in the busy group.. Consider a load balance scenario on a DP with HT system, with Package-0 containing one high priority and one low priority, Package-1 containing one low priority(with other thread being idle).. Package-1 thinks that it need to take the low priority thread from Package-0. And find_busiest_queue() returns the cpu thread with highest priority task.. And ultimately(with help of active load balance) we move high priority task to Package-1. And same continues with Package-0 now, moving high priority task from package-1 to package-0.. Even without the presence of active load balance, load balance will fail to balance the above scenario.. Fix find_busiest_queue to use "imbalance" when it is lightly loaded. [kernel@kolivas.org: sched: store weighted load on up] [kernel@kolivas.org: sched: add discrete weighted cpu load function] [suresh.b.siddha@intel.com: sched: remove dead code] Signed-off-by: Peter Williams <pwil3058@bigpond.com.au> Cc: "Siddha, Suresh B" <suresh.b.siddha@intel.com> Cc: "Chen, Kenneth W" <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Signed-off-by: Con Kolivas <kernel@kolivas.org> Cc: John Hawkes <hawkes@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'kernel/sched.c')
-rw-r--r--kernel/sched.c284
1 files changed, 219 insertions, 65 deletions
diff --git a/kernel/sched.c b/kernel/sched.c
index 678335a8b390..1847a4456a2d 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -168,15 +168,21 @@
168 */ 168 */
169 169
170#define SCALE_PRIO(x, prio) \ 170#define SCALE_PRIO(x, prio) \
171 max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO/2), MIN_TIMESLICE) 171 max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO / 2), MIN_TIMESLICE)
172 172
173static unsigned int task_timeslice(task_t *p) 173static unsigned int static_prio_timeslice(int static_prio)
174{ 174{
175 if (p->static_prio < NICE_TO_PRIO(0)) 175 if (static_prio < NICE_TO_PRIO(0))
176 return SCALE_PRIO(DEF_TIMESLICE*4, p->static_prio); 176 return SCALE_PRIO(DEF_TIMESLICE * 4, static_prio);
177 else 177 else
178 return SCALE_PRIO(DEF_TIMESLICE, p->static_prio); 178 return SCALE_PRIO(DEF_TIMESLICE, static_prio);
179} 179}
180
181static inline unsigned int task_timeslice(task_t *p)
182{
183 return static_prio_timeslice(p->static_prio);
184}
185
180#define task_hot(p, now, sd) ((long long) ((now) - (p)->last_ran) \ 186#define task_hot(p, now, sd) ((long long) ((now) - (p)->last_ran) \
181 < (long long) (sd)->cache_hot_time) 187 < (long long) (sd)->cache_hot_time)
182 188
@@ -207,6 +213,7 @@ struct runqueue {
207 * remote CPUs use both these fields when doing load calculation. 213 * remote CPUs use both these fields when doing load calculation.
208 */ 214 */
209 unsigned long nr_running; 215 unsigned long nr_running;
216 unsigned long raw_weighted_load;
210#ifdef CONFIG_SMP 217#ifdef CONFIG_SMP
211 unsigned long cpu_load[3]; 218 unsigned long cpu_load[3];
212#endif 219#endif
@@ -662,6 +669,68 @@ static int effective_prio(task_t *p)
662} 669}
663 670
664/* 671/*
672 * To aid in avoiding the subversion of "niceness" due to uneven distribution
673 * of tasks with abnormal "nice" values across CPUs the contribution that
674 * each task makes to its run queue's load is weighted according to its
675 * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a
676 * scaled version of the new time slice allocation that they receive on time
677 * slice expiry etc.
678 */
679
680/*
681 * Assume: static_prio_timeslice(NICE_TO_PRIO(0)) == DEF_TIMESLICE
682 * If static_prio_timeslice() is ever changed to break this assumption then
683 * this code will need modification
684 */
685#define TIME_SLICE_NICE_ZERO DEF_TIMESLICE
686#define LOAD_WEIGHT(lp) \
687 (((lp) * SCHED_LOAD_SCALE) / TIME_SLICE_NICE_ZERO)
688#define PRIO_TO_LOAD_WEIGHT(prio) \
689 LOAD_WEIGHT(static_prio_timeslice(prio))
690#define RTPRIO_TO_LOAD_WEIGHT(rp) \
691 (PRIO_TO_LOAD_WEIGHT(MAX_RT_PRIO) + LOAD_WEIGHT(rp))
692
693static void set_load_weight(task_t *p)
694{
695 if (rt_task(p)) {
696#ifdef CONFIG_SMP
697 if (p == task_rq(p)->migration_thread)
698 /*
699 * The migration thread does the actual balancing.
700 * Giving its load any weight will skew balancing
701 * adversely.
702 */
703 p->load_weight = 0;
704 else
705#endif
706 p->load_weight = RTPRIO_TO_LOAD_WEIGHT(p->rt_priority);
707 } else
708 p->load_weight = PRIO_TO_LOAD_WEIGHT(p->static_prio);
709}
710
711static inline void inc_raw_weighted_load(runqueue_t *rq, const task_t *p)
712{
713 rq->raw_weighted_load += p->load_weight;
714}
715
716static inline void dec_raw_weighted_load(runqueue_t *rq, const task_t *p)
717{
718 rq->raw_weighted_load -= p->load_weight;
719}
720
721static inline void inc_nr_running(task_t *p, runqueue_t *rq)
722{
723 rq->nr_running++;
724 inc_raw_weighted_load(rq, p);
725}
726
727static inline void dec_nr_running(task_t *p, runqueue_t *rq)
728{
729 rq->nr_running--;
730 dec_raw_weighted_load(rq, p);
731}
732
733/*
665 * __activate_task - move a task to the runqueue. 734 * __activate_task - move a task to the runqueue.
666 */ 735 */
667static void __activate_task(task_t *p, runqueue_t *rq) 736static void __activate_task(task_t *p, runqueue_t *rq)
@@ -671,7 +740,7 @@ static void __activate_task(task_t *p, runqueue_t *rq)
671 if (batch_task(p)) 740 if (batch_task(p))
672 target = rq->expired; 741 target = rq->expired;
673 enqueue_task(p, target); 742 enqueue_task(p, target);
674 rq->nr_running++; 743 inc_nr_running(p, rq);
675} 744}
676 745
677/* 746/*
@@ -680,7 +749,7 @@ static void __activate_task(task_t *p, runqueue_t *rq)
680static inline void __activate_idle_task(task_t *p, runqueue_t *rq) 749static inline void __activate_idle_task(task_t *p, runqueue_t *rq)
681{ 750{
682 enqueue_task_head(p, rq->active); 751 enqueue_task_head(p, rq->active);
683 rq->nr_running++; 752 inc_nr_running(p, rq);
684} 753}
685 754
686static int recalc_task_prio(task_t *p, unsigned long long now) 755static int recalc_task_prio(task_t *p, unsigned long long now)
@@ -804,7 +873,7 @@ static void activate_task(task_t *p, runqueue_t *rq, int local)
804 */ 873 */
805static void deactivate_task(struct task_struct *p, runqueue_t *rq) 874static void deactivate_task(struct task_struct *p, runqueue_t *rq)
806{ 875{
807 rq->nr_running--; 876 dec_nr_running(p, rq);
808 dequeue_task(p, p->array); 877 dequeue_task(p, p->array);
809 p->array = NULL; 878 p->array = NULL;
810} 879}
@@ -859,6 +928,12 @@ inline int task_curr(const task_t *p)
859 return cpu_curr(task_cpu(p)) == p; 928 return cpu_curr(task_cpu(p)) == p;
860} 929}
861 930
931/* Used instead of source_load when we know the type == 0 */
932unsigned long weighted_cpuload(const int cpu)
933{
934 return cpu_rq(cpu)->raw_weighted_load;
935}
936
862#ifdef CONFIG_SMP 937#ifdef CONFIG_SMP
863typedef struct { 938typedef struct {
864 struct list_head list; 939 struct list_head list;
@@ -948,7 +1023,8 @@ void kick_process(task_t *p)
948} 1023}
949 1024
950/* 1025/*
951 * Return a low guess at the load of a migration-source cpu. 1026 * Return a low guess at the load of a migration-source cpu weighted
1027 * according to the scheduling class and "nice" value.
952 * 1028 *
953 * We want to under-estimate the load of migration sources, to 1029 * We want to under-estimate the load of migration sources, to
954 * balance conservatively. 1030 * balance conservatively.
@@ -956,24 +1032,36 @@ void kick_process(task_t *p)
956static inline unsigned long source_load(int cpu, int type) 1032static inline unsigned long source_load(int cpu, int type)
957{ 1033{
958 runqueue_t *rq = cpu_rq(cpu); 1034 runqueue_t *rq = cpu_rq(cpu);
959 unsigned long load_now = rq->nr_running * SCHED_LOAD_SCALE; 1035
960 if (type == 0) 1036 if (type == 0)
961 return load_now; 1037 return rq->raw_weighted_load;
962 1038
963 return min(rq->cpu_load[type-1], load_now); 1039 return min(rq->cpu_load[type-1], rq->raw_weighted_load);
964} 1040}
965 1041
966/* 1042/*
967 * Return a high guess at the load of a migration-target cpu 1043 * Return a high guess at the load of a migration-target cpu weighted
1044 * according to the scheduling class and "nice" value.
968 */ 1045 */
969static inline unsigned long target_load(int cpu, int type) 1046static inline unsigned long target_load(int cpu, int type)
970{ 1047{
971 runqueue_t *rq = cpu_rq(cpu); 1048 runqueue_t *rq = cpu_rq(cpu);
972 unsigned long load_now = rq->nr_running * SCHED_LOAD_SCALE; 1049
973 if (type == 0) 1050 if (type == 0)
974 return load_now; 1051 return rq->raw_weighted_load;
975 1052
976 return max(rq->cpu_load[type-1], load_now); 1053 return max(rq->cpu_load[type-1], rq->raw_weighted_load);
1054}
1055
1056/*
1057 * Return the average load per task on the cpu's run queue
1058 */
1059static inline unsigned long cpu_avg_load_per_task(int cpu)
1060{
1061 runqueue_t *rq = cpu_rq(cpu);
1062 unsigned long n = rq->nr_running;
1063
1064 return n ? rq->raw_weighted_load / n : SCHED_LOAD_SCALE;
977} 1065}
978 1066
979/* 1067/*
@@ -1046,7 +1134,7 @@ find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
1046 cpus_and(tmp, group->cpumask, p->cpus_allowed); 1134 cpus_and(tmp, group->cpumask, p->cpus_allowed);
1047 1135
1048 for_each_cpu_mask(i, tmp) { 1136 for_each_cpu_mask(i, tmp) {
1049 load = source_load(i, 0); 1137 load = weighted_cpuload(i);
1050 1138
1051 if (load < min_load || (load == min_load && i == this_cpu)) { 1139 if (load < min_load || (load == min_load && i == this_cpu)) {
1052 min_load = load; 1140 min_load = load;
@@ -1226,17 +1314,19 @@ static int try_to_wake_up(task_t *p, unsigned int state, int sync)
1226 1314
1227 if (this_sd->flags & SD_WAKE_AFFINE) { 1315 if (this_sd->flags & SD_WAKE_AFFINE) {
1228 unsigned long tl = this_load; 1316 unsigned long tl = this_load;
1317 unsigned long tl_per_task = cpu_avg_load_per_task(this_cpu);
1318
1229 /* 1319 /*
1230 * If sync wakeup then subtract the (maximum possible) 1320 * If sync wakeup then subtract the (maximum possible)
1231 * effect of the currently running task from the load 1321 * effect of the currently running task from the load
1232 * of the current CPU: 1322 * of the current CPU:
1233 */ 1323 */
1234 if (sync) 1324 if (sync)
1235 tl -= SCHED_LOAD_SCALE; 1325 tl -= current->load_weight;
1236 1326
1237 if ((tl <= load && 1327 if ((tl <= load &&
1238 tl + target_load(cpu, idx) <= SCHED_LOAD_SCALE) || 1328 tl + target_load(cpu, idx) <= tl_per_task) ||
1239 100*(tl + SCHED_LOAD_SCALE) <= imbalance*load) { 1329 100*(tl + p->load_weight) <= imbalance*load) {
1240 /* 1330 /*
1241 * This domain has SD_WAKE_AFFINE and 1331 * This domain has SD_WAKE_AFFINE and
1242 * p is cache cold in this domain, and 1332 * p is cache cold in this domain, and
@@ -1435,7 +1525,7 @@ void fastcall wake_up_new_task(task_t *p, unsigned long clone_flags)
1435 list_add_tail(&p->run_list, &current->run_list); 1525 list_add_tail(&p->run_list, &current->run_list);
1436 p->array = current->array; 1526 p->array = current->array;
1437 p->array->nr_active++; 1527 p->array->nr_active++;
1438 rq->nr_running++; 1528 inc_nr_running(p, rq);
1439 } 1529 }
1440 set_need_resched(); 1530 set_need_resched();
1441 } else 1531 } else
@@ -1802,9 +1892,9 @@ void pull_task(runqueue_t *src_rq, prio_array_t *src_array, task_t *p,
1802 runqueue_t *this_rq, prio_array_t *this_array, int this_cpu) 1892 runqueue_t *this_rq, prio_array_t *this_array, int this_cpu)
1803{ 1893{
1804 dequeue_task(p, src_array); 1894 dequeue_task(p, src_array);
1805 src_rq->nr_running--; 1895 dec_nr_running(p, src_rq);
1806 set_task_cpu(p, this_cpu); 1896 set_task_cpu(p, this_cpu);
1807 this_rq->nr_running++; 1897 inc_nr_running(p, this_rq);
1808 enqueue_task(p, this_array); 1898 enqueue_task(p, this_array);
1809 p->timestamp = (p->timestamp - src_rq->timestamp_last_tick) 1899 p->timestamp = (p->timestamp - src_rq->timestamp_last_tick)
1810 + this_rq->timestamp_last_tick; 1900 + this_rq->timestamp_last_tick;
@@ -1852,24 +1942,27 @@ int can_migrate_task(task_t *p, runqueue_t *rq, int this_cpu,
1852} 1942}
1853 1943
1854/* 1944/*
1855 * move_tasks tries to move up to max_nr_move tasks from busiest to this_rq, 1945 * move_tasks tries to move up to max_nr_move tasks and max_load_move weighted
1856 * as part of a balancing operation within "domain". Returns the number of 1946 * load from busiest to this_rq, as part of a balancing operation within
1857 * tasks moved. 1947 * "domain". Returns the number of tasks moved.
1858 * 1948 *
1859 * Called with both runqueues locked. 1949 * Called with both runqueues locked.
1860 */ 1950 */
1861static int move_tasks(runqueue_t *this_rq, int this_cpu, runqueue_t *busiest, 1951static int move_tasks(runqueue_t *this_rq, int this_cpu, runqueue_t *busiest,
1862 unsigned long max_nr_move, struct sched_domain *sd, 1952 unsigned long max_nr_move, unsigned long max_load_move,
1863 enum idle_type idle, int *all_pinned) 1953 struct sched_domain *sd, enum idle_type idle,
1954 int *all_pinned)
1864{ 1955{
1865 prio_array_t *array, *dst_array; 1956 prio_array_t *array, *dst_array;
1866 struct list_head *head, *curr; 1957 struct list_head *head, *curr;
1867 int idx, pulled = 0, pinned = 0; 1958 int idx, pulled = 0, pinned = 0;
1959 long rem_load_move;
1868 task_t *tmp; 1960 task_t *tmp;
1869 1961
1870 if (max_nr_move == 0) 1962 if (max_nr_move == 0 || max_load_move == 0)
1871 goto out; 1963 goto out;
1872 1964
1965 rem_load_move = max_load_move;
1873 pinned = 1; 1966 pinned = 1;
1874 1967
1875 /* 1968 /*
@@ -1910,7 +2003,8 @@ skip_queue:
1910 2003
1911 curr = curr->prev; 2004 curr = curr->prev;
1912 2005
1913 if (!can_migrate_task(tmp, busiest, this_cpu, sd, idle, &pinned)) { 2006 if (tmp->load_weight > rem_load_move ||
2007 !can_migrate_task(tmp, busiest, this_cpu, sd, idle, &pinned)) {
1914 if (curr != head) 2008 if (curr != head)
1915 goto skip_queue; 2009 goto skip_queue;
1916 idx++; 2010 idx++;
@@ -1924,9 +2018,13 @@ skip_queue:
1924 2018
1925 pull_task(busiest, array, tmp, this_rq, dst_array, this_cpu); 2019 pull_task(busiest, array, tmp, this_rq, dst_array, this_cpu);
1926 pulled++; 2020 pulled++;
2021 rem_load_move -= tmp->load_weight;
1927 2022
1928 /* We only want to steal up to the prescribed number of tasks. */ 2023 /*
1929 if (pulled < max_nr_move) { 2024 * We only want to steal up to the prescribed number of tasks
2025 * and the prescribed amount of weighted load.
2026 */
2027 if (pulled < max_nr_move && rem_load_move > 0) {
1930 if (curr != head) 2028 if (curr != head)
1931 goto skip_queue; 2029 goto skip_queue;
1932 idx++; 2030 idx++;
@@ -1947,7 +2045,7 @@ out:
1947 2045
1948/* 2046/*
1949 * find_busiest_group finds and returns the busiest CPU group within the 2047 * find_busiest_group finds and returns the busiest CPU group within the
1950 * domain. It calculates and returns the number of tasks which should be 2048 * domain. It calculates and returns the amount of weighted load which should be
1951 * moved to restore balance via the imbalance parameter. 2049 * moved to restore balance via the imbalance parameter.
1952 */ 2050 */
1953static struct sched_group * 2051static struct sched_group *
@@ -1957,9 +2055,13 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
1957 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; 2055 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups;
1958 unsigned long max_load, avg_load, total_load, this_load, total_pwr; 2056 unsigned long max_load, avg_load, total_load, this_load, total_pwr;
1959 unsigned long max_pull; 2057 unsigned long max_pull;
2058 unsigned long busiest_load_per_task, busiest_nr_running;
2059 unsigned long this_load_per_task, this_nr_running;
1960 int load_idx; 2060 int load_idx;
1961 2061
1962 max_load = this_load = total_load = total_pwr = 0; 2062 max_load = this_load = total_load = total_pwr = 0;
2063 busiest_load_per_task = busiest_nr_running = 0;
2064 this_load_per_task = this_nr_running = 0;
1963 if (idle == NOT_IDLE) 2065 if (idle == NOT_IDLE)
1964 load_idx = sd->busy_idx; 2066 load_idx = sd->busy_idx;
1965 else if (idle == NEWLY_IDLE) 2067 else if (idle == NEWLY_IDLE)
@@ -1971,13 +2073,16 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
1971 unsigned long load; 2073 unsigned long load;
1972 int local_group; 2074 int local_group;
1973 int i; 2075 int i;
2076 unsigned long sum_nr_running, sum_weighted_load;
1974 2077
1975 local_group = cpu_isset(this_cpu, group->cpumask); 2078 local_group = cpu_isset(this_cpu, group->cpumask);
1976 2079
1977 /* Tally up the load of all CPUs in the group */ 2080 /* Tally up the load of all CPUs in the group */
1978 avg_load = 0; 2081 sum_weighted_load = sum_nr_running = avg_load = 0;
1979 2082
1980 for_each_cpu_mask(i, group->cpumask) { 2083 for_each_cpu_mask(i, group->cpumask) {
2084 runqueue_t *rq = cpu_rq(i);
2085
1981 if (*sd_idle && !idle_cpu(i)) 2086 if (*sd_idle && !idle_cpu(i))
1982 *sd_idle = 0; 2087 *sd_idle = 0;
1983 2088
@@ -1988,6 +2093,8 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
1988 load = source_load(i, load_idx); 2093 load = source_load(i, load_idx);
1989 2094
1990 avg_load += load; 2095 avg_load += load;
2096 sum_nr_running += rq->nr_running;
2097 sum_weighted_load += rq->raw_weighted_load;
1991 } 2098 }
1992 2099
1993 total_load += avg_load; 2100 total_load += avg_load;
@@ -1999,14 +2106,19 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
1999 if (local_group) { 2106 if (local_group) {
2000 this_load = avg_load; 2107 this_load = avg_load;
2001 this = group; 2108 this = group;
2002 } else if (avg_load > max_load) { 2109 this_nr_running = sum_nr_running;
2110 this_load_per_task = sum_weighted_load;
2111 } else if (avg_load > max_load &&
2112 sum_nr_running > group->cpu_power / SCHED_LOAD_SCALE) {
2003 max_load = avg_load; 2113 max_load = avg_load;
2004 busiest = group; 2114 busiest = group;
2115 busiest_nr_running = sum_nr_running;
2116 busiest_load_per_task = sum_weighted_load;
2005 } 2117 }
2006 group = group->next; 2118 group = group->next;
2007 } while (group != sd->groups); 2119 } while (group != sd->groups);
2008 2120
2009 if (!busiest || this_load >= max_load || max_load <= SCHED_LOAD_SCALE) 2121 if (!busiest || this_load >= max_load || busiest_nr_running == 0)
2010 goto out_balanced; 2122 goto out_balanced;
2011 2123
2012 avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; 2124 avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr;
@@ -2015,6 +2127,7 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
2015 100*max_load <= sd->imbalance_pct*this_load) 2127 100*max_load <= sd->imbalance_pct*this_load)
2016 goto out_balanced; 2128 goto out_balanced;
2017 2129
2130 busiest_load_per_task /= busiest_nr_running;
2018 /* 2131 /*
2019 * We're trying to get all the cpus to the average_load, so we don't 2132 * We're trying to get all the cpus to the average_load, so we don't
2020 * want to push ourselves above the average load, nor do we wish to 2133 * want to push ourselves above the average load, nor do we wish to
@@ -2026,21 +2139,50 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
2026 * by pulling tasks to us. Be careful of negative numbers as they'll 2139 * by pulling tasks to us. Be careful of negative numbers as they'll
2027 * appear as very large values with unsigned longs. 2140 * appear as very large values with unsigned longs.
2028 */ 2141 */
2142 if (max_load <= busiest_load_per_task)
2143 goto out_balanced;
2144
2145 /*
2146 * In the presence of smp nice balancing, certain scenarios can have
2147 * max load less than avg load(as we skip the groups at or below
2148 * its cpu_power, while calculating max_load..)
2149 */
2150 if (max_load < avg_load) {
2151 *imbalance = 0;
2152 goto small_imbalance;
2153 }
2029 2154
2030 /* Don't want to pull so many tasks that a group would go idle */ 2155 /* Don't want to pull so many tasks that a group would go idle */
2031 max_pull = min(max_load - avg_load, max_load - SCHED_LOAD_SCALE); 2156 max_pull = min(max_load - avg_load, max_load - busiest_load_per_task);
2032 2157
2033 /* How much load to actually move to equalise the imbalance */ 2158 /* How much load to actually move to equalise the imbalance */
2034 *imbalance = min(max_pull * busiest->cpu_power, 2159 *imbalance = min(max_pull * busiest->cpu_power,
2035 (avg_load - this_load) * this->cpu_power) 2160 (avg_load - this_load) * this->cpu_power)
2036 / SCHED_LOAD_SCALE; 2161 / SCHED_LOAD_SCALE;
2037 2162
2038 if (*imbalance < SCHED_LOAD_SCALE) { 2163 /*
2039 unsigned long pwr_now = 0, pwr_move = 0; 2164 * if *imbalance is less than the average load per runnable task
2165 * there is no gaurantee that any tasks will be moved so we'll have
2166 * a think about bumping its value to force at least one task to be
2167 * moved
2168 */
2169 if (*imbalance < busiest_load_per_task) {
2170 unsigned long pwr_now, pwr_move;
2040 unsigned long tmp; 2171 unsigned long tmp;
2172 unsigned int imbn;
2173
2174small_imbalance:
2175 pwr_move = pwr_now = 0;
2176 imbn = 2;
2177 if (this_nr_running) {
2178 this_load_per_task /= this_nr_running;
2179 if (busiest_load_per_task > this_load_per_task)
2180 imbn = 1;
2181 } else
2182 this_load_per_task = SCHED_LOAD_SCALE;
2041 2183
2042 if (max_load - this_load >= SCHED_LOAD_SCALE*2) { 2184 if (max_load - this_load >= busiest_load_per_task * imbn) {
2043 *imbalance = 1; 2185 *imbalance = busiest_load_per_task;
2044 return busiest; 2186 return busiest;
2045 } 2187 }
2046 2188
@@ -2050,35 +2192,34 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
2050 * moving them. 2192 * moving them.
2051 */ 2193 */
2052 2194
2053 pwr_now += busiest->cpu_power*min(SCHED_LOAD_SCALE, max_load); 2195 pwr_now += busiest->cpu_power *
2054 pwr_now += this->cpu_power*min(SCHED_LOAD_SCALE, this_load); 2196 min(busiest_load_per_task, max_load);
2197 pwr_now += this->cpu_power *
2198 min(this_load_per_task, this_load);
2055 pwr_now /= SCHED_LOAD_SCALE; 2199 pwr_now /= SCHED_LOAD_SCALE;
2056 2200
2057 /* Amount of load we'd subtract */ 2201 /* Amount of load we'd subtract */
2058 tmp = SCHED_LOAD_SCALE*SCHED_LOAD_SCALE/busiest->cpu_power; 2202 tmp = busiest_load_per_task*SCHED_LOAD_SCALE/busiest->cpu_power;
2059 if (max_load > tmp) 2203 if (max_load > tmp)
2060 pwr_move += busiest->cpu_power*min(SCHED_LOAD_SCALE, 2204 pwr_move += busiest->cpu_power *
2061 max_load - tmp); 2205 min(busiest_load_per_task, max_load - tmp);
2062 2206
2063 /* Amount of load we'd add */ 2207 /* Amount of load we'd add */
2064 if (max_load*busiest->cpu_power < 2208 if (max_load*busiest->cpu_power <
2065 SCHED_LOAD_SCALE*SCHED_LOAD_SCALE) 2209 busiest_load_per_task*SCHED_LOAD_SCALE)
2066 tmp = max_load*busiest->cpu_power/this->cpu_power; 2210 tmp = max_load*busiest->cpu_power/this->cpu_power;
2067 else 2211 else
2068 tmp = SCHED_LOAD_SCALE*SCHED_LOAD_SCALE/this->cpu_power; 2212 tmp = busiest_load_per_task*SCHED_LOAD_SCALE/this->cpu_power;
2069 pwr_move += this->cpu_power*min(SCHED_LOAD_SCALE, this_load + tmp); 2213 pwr_move += this->cpu_power*min(this_load_per_task, this_load + tmp);
2070 pwr_move /= SCHED_LOAD_SCALE; 2214 pwr_move /= SCHED_LOAD_SCALE;
2071 2215
2072 /* Move if we gain throughput */ 2216 /* Move if we gain throughput */
2073 if (pwr_move <= pwr_now) 2217 if (pwr_move <= pwr_now)
2074 goto out_balanced; 2218 goto out_balanced;
2075 2219
2076 *imbalance = 1; 2220 *imbalance = busiest_load_per_task;
2077 return busiest;
2078 } 2221 }
2079 2222
2080 /* Get rid of the scaling factor, rounding down as we divide */
2081 *imbalance = *imbalance / SCHED_LOAD_SCALE;
2082 return busiest; 2223 return busiest;
2083 2224
2084out_balanced: 2225out_balanced:
@@ -2091,18 +2232,21 @@ out_balanced:
2091 * find_busiest_queue - find the busiest runqueue among the cpus in group. 2232 * find_busiest_queue - find the busiest runqueue among the cpus in group.
2092 */ 2233 */
2093static runqueue_t *find_busiest_queue(struct sched_group *group, 2234static runqueue_t *find_busiest_queue(struct sched_group *group,
2094 enum idle_type idle) 2235 enum idle_type idle, unsigned long imbalance)
2095{ 2236{
2096 unsigned long load, max_load = 0; 2237 unsigned long max_load = 0;
2097 runqueue_t *busiest = NULL; 2238 runqueue_t *busiest = NULL, *rqi;
2098 int i; 2239 int i;
2099 2240
2100 for_each_cpu_mask(i, group->cpumask) { 2241 for_each_cpu_mask(i, group->cpumask) {
2101 load = source_load(i, 0); 2242 rqi = cpu_rq(i);
2243
2244 if (rqi->nr_running == 1 && rqi->raw_weighted_load > imbalance)
2245 continue;
2102 2246
2103 if (load > max_load) { 2247 if (rqi->raw_weighted_load > max_load) {
2104 max_load = load; 2248 max_load = rqi->raw_weighted_load;
2105 busiest = cpu_rq(i); 2249 busiest = rqi;
2106 } 2250 }
2107 } 2251 }
2108 2252
@@ -2115,6 +2259,7 @@ static runqueue_t *find_busiest_queue(struct sched_group *group,
2115 */ 2259 */
2116#define MAX_PINNED_INTERVAL 512 2260#define MAX_PINNED_INTERVAL 512
2117 2261
2262#define minus_1_or_zero(n) ((n) > 0 ? (n) - 1 : 0)
2118/* 2263/*
2119 * Check this_cpu to ensure it is balanced within domain. Attempt to move 2264 * Check this_cpu to ensure it is balanced within domain. Attempt to move
2120 * tasks if there is an imbalance. 2265 * tasks if there is an imbalance.
@@ -2142,7 +2287,7 @@ static int load_balance(int this_cpu, runqueue_t *this_rq,
2142 goto out_balanced; 2287 goto out_balanced;
2143 } 2288 }
2144 2289
2145 busiest = find_busiest_queue(group, idle); 2290 busiest = find_busiest_queue(group, idle, imbalance);
2146 if (!busiest) { 2291 if (!busiest) {
2147 schedstat_inc(sd, lb_nobusyq[idle]); 2292 schedstat_inc(sd, lb_nobusyq[idle]);
2148 goto out_balanced; 2293 goto out_balanced;
@@ -2162,6 +2307,7 @@ static int load_balance(int this_cpu, runqueue_t *this_rq,
2162 */ 2307 */
2163 double_rq_lock(this_rq, busiest); 2308 double_rq_lock(this_rq, busiest);
2164 nr_moved = move_tasks(this_rq, this_cpu, busiest, 2309 nr_moved = move_tasks(this_rq, this_cpu, busiest,
2310 minus_1_or_zero(busiest->nr_running),
2165 imbalance, sd, idle, &all_pinned); 2311 imbalance, sd, idle, &all_pinned);
2166 double_rq_unlock(this_rq, busiest); 2312 double_rq_unlock(this_rq, busiest);
2167 2313
@@ -2265,7 +2411,7 @@ static int load_balance_newidle(int this_cpu, runqueue_t *this_rq,
2265 goto out_balanced; 2411 goto out_balanced;
2266 } 2412 }
2267 2413
2268 busiest = find_busiest_queue(group, NEWLY_IDLE); 2414 busiest = find_busiest_queue(group, NEWLY_IDLE, imbalance);
2269 if (!busiest) { 2415 if (!busiest) {
2270 schedstat_inc(sd, lb_nobusyq[NEWLY_IDLE]); 2416 schedstat_inc(sd, lb_nobusyq[NEWLY_IDLE]);
2271 goto out_balanced; 2417 goto out_balanced;
@@ -2280,6 +2426,7 @@ static int load_balance_newidle(int this_cpu, runqueue_t *this_rq,
2280 /* Attempt to move tasks */ 2426 /* Attempt to move tasks */
2281 double_lock_balance(this_rq, busiest); 2427 double_lock_balance(this_rq, busiest);
2282 nr_moved = move_tasks(this_rq, this_cpu, busiest, 2428 nr_moved = move_tasks(this_rq, this_cpu, busiest,
2429 minus_1_or_zero(busiest->nr_running),
2283 imbalance, sd, NEWLY_IDLE, NULL); 2430 imbalance, sd, NEWLY_IDLE, NULL);
2284 spin_unlock(&busiest->lock); 2431 spin_unlock(&busiest->lock);
2285 } 2432 }
@@ -2361,7 +2508,8 @@ static void active_load_balance(runqueue_t *busiest_rq, int busiest_cpu)
2361 2508
2362 schedstat_inc(sd, alb_cnt); 2509 schedstat_inc(sd, alb_cnt);
2363 2510
2364 if (move_tasks(target_rq, target_cpu, busiest_rq, 1, sd, SCHED_IDLE, NULL)) 2511 if (move_tasks(target_rq, target_cpu, busiest_rq, 1,
2512 RTPRIO_TO_LOAD_WEIGHT(100), sd, SCHED_IDLE, NULL))
2365 schedstat_inc(sd, alb_pushed); 2513 schedstat_inc(sd, alb_pushed);
2366 else 2514 else
2367 schedstat_inc(sd, alb_failed); 2515 schedstat_inc(sd, alb_failed);
@@ -2389,7 +2537,7 @@ static void rebalance_tick(int this_cpu, runqueue_t *this_rq,
2389 struct sched_domain *sd; 2537 struct sched_domain *sd;
2390 int i; 2538 int i;
2391 2539
2392 this_load = this_rq->nr_running * SCHED_LOAD_SCALE; 2540 this_load = this_rq->raw_weighted_load;
2393 /* Update our load */ 2541 /* Update our load */
2394 for (i = 0; i < 3; i++) { 2542 for (i = 0; i < 3; i++) {
2395 unsigned long new_load = this_load; 2543 unsigned long new_load = this_load;
@@ -3441,17 +3589,21 @@ void set_user_nice(task_t *p, long nice)
3441 goto out_unlock; 3589 goto out_unlock;
3442 } 3590 }
3443 array = p->array; 3591 array = p->array;
3444 if (array) 3592 if (array) {
3445 dequeue_task(p, array); 3593 dequeue_task(p, array);
3594 dec_raw_weighted_load(rq, p);
3595 }
3446 3596
3447 old_prio = p->prio; 3597 old_prio = p->prio;
3448 new_prio = NICE_TO_PRIO(nice); 3598 new_prio = NICE_TO_PRIO(nice);
3449 delta = new_prio - old_prio; 3599 delta = new_prio - old_prio;
3450 p->static_prio = NICE_TO_PRIO(nice); 3600 p->static_prio = NICE_TO_PRIO(nice);
3601 set_load_weight(p);
3451 p->prio += delta; 3602 p->prio += delta;
3452 3603
3453 if (array) { 3604 if (array) {
3454 enqueue_task(p, array); 3605 enqueue_task(p, array);
3606 inc_raw_weighted_load(rq, p);
3455 /* 3607 /*
3456 * If the task increased its priority or is running and 3608 * If the task increased its priority or is running and
3457 * lowered its priority, then reschedule its CPU: 3609 * lowered its priority, then reschedule its CPU:
@@ -3587,6 +3739,7 @@ static void __setscheduler(struct task_struct *p, int policy, int prio)
3587 if (policy == SCHED_BATCH) 3739 if (policy == SCHED_BATCH)
3588 p->sleep_avg = 0; 3740 p->sleep_avg = 0;
3589 } 3741 }
3742 set_load_weight(p);
3590} 3743}
3591 3744
3592/** 3745/**
@@ -6106,6 +6259,7 @@ void __init sched_init(void)
6106 } 6259 }
6107 } 6260 }
6108 6261
6262 set_load_weight(&init_task);
6109 /* 6263 /*
6110 * The boot idle thread does lazy MMU switching as well: 6264 * The boot idle thread does lazy MMU switching as well:
6111 */ 6265 */