Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * kernel/sched.c |
| 3 | * |
| 4 | * Kernel scheduler and related syscalls |
| 5 | * |
| 6 | * Copyright (C) 1991-2002 Linus Torvalds |
| 7 | * |
| 8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and |
| 9 | * make semaphores SMP safe |
| 10 | * 1998-11-19 Implemented schedule_timeout() and related stuff |
| 11 | * by Andrea Arcangeli |
| 12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: |
| 13 | * hybrid priority-list and round-robin design with |
| 14 | * an array-switch method of distributing timeslices |
| 15 | * and per-CPU runqueues. Cleanups and useful suggestions |
| 16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. |
| 17 | * 2003-09-03 Interactivity tuning by Con Kolivas. |
| 18 | * 2004-04-02 Scheduler domains code by Nick Piggin |
Ingo Molnar | c31f2e8 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
| 20 | * fair scheduling design by Con Kolivas. |
| 21 | * 2007-05-05 Load balancing (smp-nice) and other improvements |
| 22 | * by Peter Williams |
| 23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith |
| 24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | */ |
| 26 | |
| 27 | #include <linux/mm.h> |
| 28 | #include <linux/module.h> |
| 29 | #include <linux/nmi.h> |
| 30 | #include <linux/init.h> |
Ingo Molnar | dff06c1 | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 31 | #include <linux/uaccess.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | #include <linux/highmem.h> |
| 33 | #include <linux/smp_lock.h> |
| 34 | #include <asm/mmu_context.h> |
| 35 | #include <linux/interrupt.h> |
Randy.Dunlap | c59ede7 | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 36 | #include <linux/capability.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | #include <linux/completion.h> |
| 38 | #include <linux/kernel_stat.h> |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 39 | #include <linux/debug_locks.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | #include <linux/security.h> |
| 41 | #include <linux/notifier.h> |
| 42 | #include <linux/profile.h> |
Nigel Cunningham | 7dfb710 | 2006-12-06 20:34:23 -0800 | [diff] [blame] | 43 | #include <linux/freezer.h> |
akpm@osdl.org | 198e2f1 | 2006-01-12 01:05:30 -0800 | [diff] [blame] | 44 | #include <linux/vmalloc.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 45 | #include <linux/blkdev.h> |
| 46 | #include <linux/delay.h> |
| 47 | #include <linux/smp.h> |
| 48 | #include <linux/threads.h> |
| 49 | #include <linux/timer.h> |
| 50 | #include <linux/rcupdate.h> |
| 51 | #include <linux/cpu.h> |
| 52 | #include <linux/cpuset.h> |
| 53 | #include <linux/percpu.h> |
| 54 | #include <linux/kthread.h> |
| 55 | #include <linux/seq_file.h> |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 56 | #include <linux/sysctl.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | #include <linux/syscalls.h> |
| 58 | #include <linux/times.h> |
Jay Lan | 8f0ab51 | 2006-09-30 23:28:59 -0700 | [diff] [blame] | 59 | #include <linux/tsacct_kern.h> |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 60 | #include <linux/kprobes.h> |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 61 | #include <linux/delayacct.h> |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 62 | #include <linux/reciprocal_div.h> |
Ingo Molnar | dff06c1 | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 63 | #include <linux/unistd.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 64 | |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 65 | #include <asm/tlb.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 66 | |
| 67 | /* |
Alexey Dobriyan | b035b6d | 2007-02-10 01:45:10 -0800 | [diff] [blame] | 68 | * Scheduler clock - returns current time in nanosec units. |
| 69 | * This is default implementation. |
| 70 | * Architectures and sub-architectures can override this. |
| 71 | */ |
| 72 | unsigned long long __attribute__((weak)) sched_clock(void) |
| 73 | { |
| 74 | return (unsigned long long)jiffies * (1000000000 / HZ); |
| 75 | } |
| 76 | |
| 77 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | * Convert user-nice values [ -20 ... 0 ... 19 ] |
| 79 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], |
| 80 | * and back. |
| 81 | */ |
| 82 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) |
| 83 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) |
| 84 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) |
| 85 | |
| 86 | /* |
| 87 | * 'User priority' is the nice value converted to something we |
| 88 | * can work with better when scaling various scheduler parameters, |
| 89 | * it's a [ 0 ... 39 ] range. |
| 90 | */ |
| 91 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) |
| 92 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) |
| 93 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) |
| 94 | |
| 95 | /* |
| 96 | * Some helpers for converting nanosecond timing to jiffy resolution |
| 97 | */ |
| 98 | #define NS_TO_JIFFIES(TIME) ((TIME) / (1000000000 / HZ)) |
| 99 | #define JIFFIES_TO_NS(TIME) ((TIME) * (1000000000 / HZ)) |
| 100 | |
Ingo Molnar | 6aa645e | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 101 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
| 102 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT |
| 103 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | /* |
| 105 | * These are the 'tuning knobs' of the scheduler: |
| 106 | * |
| 107 | * Minimum timeslice is 5 msecs (or 1 jiffy, whichever is larger), |
| 108 | * default timeslice is 100 msecs, maximum timeslice is 800 msecs. |
| 109 | * Timeslices get refilled after they expire. |
| 110 | */ |
| 111 | #define MIN_TIMESLICE max(5 * HZ / 1000, 1) |
| 112 | #define DEF_TIMESLICE (100 * HZ / 1000) |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 113 | |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 114 | #ifdef CONFIG_SMP |
| 115 | /* |
| 116 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) |
| 117 | * Since cpu_power is a 'constant', we can use a reciprocal divide. |
| 118 | */ |
| 119 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) |
| 120 | { |
| 121 | return reciprocal_divide(load, sg->reciprocal_cpu_power); |
| 122 | } |
| 123 | |
| 124 | /* |
| 125 | * Each time a sched group cpu_power is changed, |
| 126 | * we must compute its reciprocal value |
| 127 | */ |
| 128 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) |
| 129 | { |
| 130 | sg->__cpu_power += val; |
| 131 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); |
| 132 | } |
| 133 | #endif |
| 134 | |
Ingo Molnar | 634fa8c | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 135 | #define SCALE_PRIO(x, prio) \ |
| 136 | max(x * (MAX_PRIO - prio) / (MAX_USER_PRIO / 2), MIN_TIMESLICE) |
Borislav Petkov | 91fcdd4 | 2006-10-19 23:28:29 -0700 | [diff] [blame] | 137 | |
Ingo Molnar | 634fa8c | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 138 | /* |
| 139 | * static_prio_timeslice() scales user-nice values [ -20 ... 0 ... 19 ] |
| 140 | * to time slice values: [800ms ... 100ms ... 5ms] |
| 141 | */ |
| 142 | static unsigned int static_prio_timeslice(int static_prio) |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 143 | { |
Ingo Molnar | 634fa8c | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 144 | if (static_prio == NICE_TO_PRIO(19)) |
| 145 | return 1; |
| 146 | |
| 147 | if (static_prio < NICE_TO_PRIO(0)) |
| 148 | return SCALE_PRIO(DEF_TIMESLICE * 4, static_prio); |
| 149 | else |
| 150 | return SCALE_PRIO(DEF_TIMESLICE, static_prio); |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 151 | } |
| 152 | |
Ingo Molnar | e05606d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 153 | static inline int rt_policy(int policy) |
| 154 | { |
| 155 | if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) |
| 156 | return 1; |
| 157 | return 0; |
| 158 | } |
| 159 | |
| 160 | static inline int task_has_rt_policy(struct task_struct *p) |
| 161 | { |
| 162 | return rt_policy(p->policy); |
| 163 | } |
| 164 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 165 | /* |
Ingo Molnar | 6aa645e | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 166 | * This is the priority-queue data structure of the RT scheduling class: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | */ |
Ingo Molnar | 6aa645e | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 168 | struct rt_prio_array { |
| 169 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ |
| 170 | struct list_head queue[MAX_RT_PRIO]; |
| 171 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | |
Ingo Molnar | 6aa645e | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 173 | struct load_stat { |
| 174 | struct load_weight load; |
| 175 | u64 load_update_start, load_update_last; |
| 176 | unsigned long delta_fair, delta_exec, delta_stat; |
| 177 | }; |
| 178 | |
| 179 | /* CFS-related fields in a runqueue */ |
| 180 | struct cfs_rq { |
| 181 | struct load_weight load; |
| 182 | unsigned long nr_running; |
| 183 | |
| 184 | s64 fair_clock; |
| 185 | u64 exec_clock; |
| 186 | s64 wait_runtime; |
| 187 | u64 sleeper_bonus; |
| 188 | unsigned long wait_runtime_overruns, wait_runtime_underruns; |
| 189 | |
| 190 | struct rb_root tasks_timeline; |
| 191 | struct rb_node *rb_leftmost; |
| 192 | struct rb_node *rb_load_balance_curr; |
| 193 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 194 | /* 'curr' points to currently running entity on this cfs_rq. |
| 195 | * It is set to NULL otherwise (i.e when none are currently running). |
| 196 | */ |
| 197 | struct sched_entity *curr; |
| 198 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
| 199 | |
| 200 | /* leaf cfs_rqs are those that hold tasks (lowest schedulable entity in |
| 201 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
| 202 | * (like users, containers etc.) |
| 203 | * |
| 204 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This |
| 205 | * list is used during load balance. |
| 206 | */ |
| 207 | struct list_head leaf_cfs_rq_list; /* Better name : task_cfs_rq_list? */ |
| 208 | #endif |
| 209 | }; |
| 210 | |
| 211 | /* Real-Time classes' related field in a runqueue: */ |
| 212 | struct rt_rq { |
| 213 | struct rt_prio_array active; |
| 214 | int rt_load_balance_idx; |
| 215 | struct list_head *rt_load_balance_head, *rt_load_balance_curr; |
| 216 | }; |
| 217 | |
| 218 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | * This is the main, per-CPU runqueue data structure. |
| 220 | * |
| 221 | * Locking rule: those places that want to lock multiple runqueues |
| 222 | * (such as the load balancing or the thread migration code), lock |
| 223 | * acquire operations must be ordered by ascending &runqueue. |
| 224 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 225 | struct rq { |
Ingo Molnar | 6aa645e | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 226 | spinlock_t lock; /* runqueue lock */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | |
| 228 | /* |
| 229 | * nr_running and cpu_load should be in the same cacheline because |
| 230 | * remote CPUs use both these fields when doing load calculation. |
| 231 | */ |
| 232 | unsigned long nr_running; |
Ingo Molnar | 6aa645e | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 233 | #define CPU_LOAD_IDX_MAX 5 |
| 234 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; |
Siddha, Suresh B | bdecea3 | 2007-05-08 00:32:48 -0700 | [diff] [blame] | 235 | unsigned char idle_at_tick; |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 236 | #ifdef CONFIG_NO_HZ |
| 237 | unsigned char in_nohz_recently; |
| 238 | #endif |
Ingo Molnar | 6aa645e | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 239 | struct load_stat ls; /* capture load from *all* tasks on this cpu */ |
| 240 | unsigned long nr_load_updates; |
| 241 | u64 nr_switches; |
| 242 | |
| 243 | struct cfs_rq cfs; |
| 244 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 245 | struct list_head leaf_cfs_rq_list; /* list of leaf cfs_rq on this cpu */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 246 | #endif |
Ingo Molnar | 6aa645e | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 247 | struct rt_rq rt; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 248 | |
| 249 | /* |
| 250 | * This is part of a global counter where only the total sum |
| 251 | * over all CPUs matters. A task can increase this counter on |
| 252 | * one CPU and if it got migrated afterwards it may decrease |
| 253 | * it on another CPU. Always updated under the runqueue lock: |
| 254 | */ |
| 255 | unsigned long nr_uninterruptible; |
| 256 | |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 257 | struct task_struct *curr, *idle; |
Christoph Lameter | c9819f4 | 2006-12-10 02:20:25 -0800 | [diff] [blame] | 258 | unsigned long next_balance; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 259 | struct mm_struct *prev_mm; |
Ingo Molnar | 6aa645e | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 260 | |
Ingo Molnar | 6aa645e | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 261 | u64 clock, prev_clock_raw; |
| 262 | s64 clock_max_delta; |
| 263 | |
| 264 | unsigned int clock_warps, clock_overflows; |
Ingo Molnar | 2aa44d0 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 265 | u64 idle_clock; |
| 266 | unsigned int clock_deep_idle_events; |
Ingo Molnar | 529c772 | 2007-08-10 23:05:11 +0200 | [diff] [blame] | 267 | u64 tick_timestamp; |
Ingo Molnar | 6aa645e | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 268 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 269 | atomic_t nr_iowait; |
| 270 | |
| 271 | #ifdef CONFIG_SMP |
| 272 | struct sched_domain *sd; |
| 273 | |
| 274 | /* For active balancing */ |
| 275 | int active_balance; |
| 276 | int push_cpu; |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 277 | int cpu; /* cpu of this runqueue */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 278 | |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 279 | struct task_struct *migration_thread; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | struct list_head migration_queue; |
| 281 | #endif |
| 282 | |
| 283 | #ifdef CONFIG_SCHEDSTATS |
| 284 | /* latency stats */ |
| 285 | struct sched_info rq_sched_info; |
| 286 | |
| 287 | /* sys_sched_yield() stats */ |
| 288 | unsigned long yld_exp_empty; |
| 289 | unsigned long yld_act_empty; |
| 290 | unsigned long yld_both_empty; |
| 291 | unsigned long yld_cnt; |
| 292 | |
| 293 | /* schedule() stats */ |
| 294 | unsigned long sched_switch; |
| 295 | unsigned long sched_cnt; |
| 296 | unsigned long sched_goidle; |
| 297 | |
| 298 | /* try_to_wake_up() stats */ |
| 299 | unsigned long ttwu_cnt; |
| 300 | unsigned long ttwu_local; |
| 301 | #endif |
Ingo Molnar | fcb9937 | 2006-07-03 00:25:10 -0700 | [diff] [blame] | 302 | struct lock_class_key rq_lock_key; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 303 | }; |
| 304 | |
Fenghua Yu | f34e3b6 | 2007-07-19 01:48:13 -0700 | [diff] [blame] | 305 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 306 | static DEFINE_MUTEX(sched_hotcpu_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 308 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
| 309 | { |
| 310 | rq->curr->sched_class->check_preempt_curr(rq, p); |
| 311 | } |
| 312 | |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 313 | static inline int cpu_of(struct rq *rq) |
| 314 | { |
| 315 | #ifdef CONFIG_SMP |
| 316 | return rq->cpu; |
| 317 | #else |
| 318 | return 0; |
| 319 | #endif |
| 320 | } |
| 321 | |
Nick Piggin | 674311d | 2005-06-25 14:57:27 -0700 | [diff] [blame] | 322 | /* |
Ingo Molnar | b04a0f4 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 323 | * Update the per-runqueue clock, as finegrained as the platform can give |
| 324 | * us, but without assuming monotonicity, etc.: |
Ingo Molnar | 20d315d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 325 | */ |
Ingo Molnar | b04a0f4 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 326 | static void __update_rq_clock(struct rq *rq) |
Ingo Molnar | 20d315d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 327 | { |
| 328 | u64 prev_raw = rq->prev_clock_raw; |
| 329 | u64 now = sched_clock(); |
| 330 | s64 delta = now - prev_raw; |
| 331 | u64 clock = rq->clock; |
| 332 | |
Ingo Molnar | b04a0f4 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 333 | #ifdef CONFIG_SCHED_DEBUG |
| 334 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); |
| 335 | #endif |
Ingo Molnar | 20d315d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 336 | /* |
| 337 | * Protect against sched_clock() occasionally going backwards: |
| 338 | */ |
| 339 | if (unlikely(delta < 0)) { |
| 340 | clock++; |
| 341 | rq->clock_warps++; |
| 342 | } else { |
| 343 | /* |
| 344 | * Catch too large forward jumps too: |
| 345 | */ |
Ingo Molnar | 529c772 | 2007-08-10 23:05:11 +0200 | [diff] [blame] | 346 | if (unlikely(clock + delta > rq->tick_timestamp + TICK_NSEC)) { |
| 347 | if (clock < rq->tick_timestamp + TICK_NSEC) |
| 348 | clock = rq->tick_timestamp + TICK_NSEC; |
| 349 | else |
| 350 | clock++; |
Ingo Molnar | 20d315d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 351 | rq->clock_overflows++; |
| 352 | } else { |
| 353 | if (unlikely(delta > rq->clock_max_delta)) |
| 354 | rq->clock_max_delta = delta; |
| 355 | clock += delta; |
| 356 | } |
| 357 | } |
| 358 | |
| 359 | rq->prev_clock_raw = now; |
| 360 | rq->clock = clock; |
Ingo Molnar | 20d315d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 361 | } |
| 362 | |
Ingo Molnar | b04a0f4 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 363 | static void update_rq_clock(struct rq *rq) |
Ingo Molnar | 20d315d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 364 | { |
Ingo Molnar | b04a0f4 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 365 | if (likely(smp_processor_id() == cpu_of(rq))) |
| 366 | __update_rq_clock(rq); |
| 367 | } |
Ingo Molnar | 20d315d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 368 | |
Ingo Molnar | 20d315d | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 369 | /* |
Nick Piggin | 674311d | 2005-06-25 14:57:27 -0700 | [diff] [blame] | 370 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 371 | * See detach_destroy_domains: synchronize_sched for details. |
Nick Piggin | 674311d | 2005-06-25 14:57:27 -0700 | [diff] [blame] | 372 | * |
| 373 | * The domain tree of any CPU may only be accessed from within |
| 374 | * preempt-disabled sections. |
| 375 | */ |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 376 | #define for_each_domain(cpu, __sd) \ |
| 377 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 378 | |
| 379 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) |
| 380 | #define this_rq() (&__get_cpu_var(runqueues)) |
| 381 | #define task_rq(p) cpu_rq(task_cpu(p)) |
| 382 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) |
| 383 | |
Ingo Molnar | e436d80 | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 384 | /* |
| 385 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu |
| 386 | * clock constructed from sched_clock(): |
| 387 | */ |
| 388 | unsigned long long cpu_clock(int cpu) |
| 389 | { |
Ingo Molnar | e436d80 | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 390 | unsigned long long now; |
| 391 | unsigned long flags; |
Ingo Molnar | b04a0f4 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 392 | struct rq *rq; |
Ingo Molnar | e436d80 | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 393 | |
Ingo Molnar | 2cd4d0e | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 394 | local_irq_save(flags); |
Ingo Molnar | b04a0f4 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 395 | rq = cpu_rq(cpu); |
| 396 | update_rq_clock(rq); |
| 397 | now = rq->clock; |
Ingo Molnar | 2cd4d0e | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 398 | local_irq_restore(flags); |
Ingo Molnar | e436d80 | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 399 | |
| 400 | return now; |
| 401 | } |
| 402 | |
Ingo Molnar | 138a8ae | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 403 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 404 | /* Change a task's ->cfs_rq if it moves across CPUs */ |
| 405 | static inline void set_task_cfs_rq(struct task_struct *p) |
| 406 | { |
| 407 | p->se.cfs_rq = &task_rq(p)->cfs; |
| 408 | } |
| 409 | #else |
| 410 | static inline void set_task_cfs_rq(struct task_struct *p) |
| 411 | { |
| 412 | } |
| 413 | #endif |
| 414 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 415 | #ifndef prepare_arch_switch |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 416 | # define prepare_arch_switch(next) do { } while (0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 417 | #endif |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 418 | #ifndef finish_arch_switch |
| 419 | # define finish_arch_switch(prev) do { } while (0) |
| 420 | #endif |
| 421 | |
| 422 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 423 | static inline int task_running(struct rq *rq, struct task_struct *p) |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 424 | { |
| 425 | return rq->curr == p; |
| 426 | } |
| 427 | |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 428 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 429 | { |
| 430 | } |
| 431 | |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 432 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 433 | { |
Ingo Molnar | da04c03 | 2005-09-13 11:17:59 +0200 | [diff] [blame] | 434 | #ifdef CONFIG_DEBUG_SPINLOCK |
| 435 | /* this is a valid case when another task releases the spinlock */ |
| 436 | rq->lock.owner = current; |
| 437 | #endif |
Ingo Molnar | 8a25d5d | 2006-07-03 00:24:54 -0700 | [diff] [blame] | 438 | /* |
| 439 | * If we are tracking spinlock dependencies then we have to |
| 440 | * fix up the runqueue lock - which gets 'carried over' from |
| 441 | * prev into current: |
| 442 | */ |
| 443 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); |
| 444 | |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 445 | spin_unlock_irq(&rq->lock); |
| 446 | } |
| 447 | |
| 448 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 449 | static inline int task_running(struct rq *rq, struct task_struct *p) |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 450 | { |
| 451 | #ifdef CONFIG_SMP |
| 452 | return p->oncpu; |
| 453 | #else |
| 454 | return rq->curr == p; |
| 455 | #endif |
| 456 | } |
| 457 | |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 458 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 459 | { |
| 460 | #ifdef CONFIG_SMP |
| 461 | /* |
| 462 | * We can optimise this out completely for !SMP, because the |
| 463 | * SMP rebalancing from interrupt is the only thing that cares |
| 464 | * here. |
| 465 | */ |
| 466 | next->oncpu = 1; |
| 467 | #endif |
| 468 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
| 469 | spin_unlock_irq(&rq->lock); |
| 470 | #else |
| 471 | spin_unlock(&rq->lock); |
| 472 | #endif |
| 473 | } |
| 474 | |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 475 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 476 | { |
| 477 | #ifdef CONFIG_SMP |
| 478 | /* |
| 479 | * After ->oncpu is cleared, the task can be moved to a different CPU. |
| 480 | * We must ensure this doesn't happen until the switch is completely |
| 481 | * finished. |
| 482 | */ |
| 483 | smp_wmb(); |
| 484 | prev->oncpu = 0; |
| 485 | #endif |
| 486 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
| 487 | local_irq_enable(); |
| 488 | #endif |
| 489 | } |
| 490 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 491 | |
| 492 | /* |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 493 | * __task_rq_lock - lock the runqueue a given task resides on. |
| 494 | * Must be called interrupts disabled. |
| 495 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 496 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 497 | __acquires(rq->lock) |
| 498 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 499 | struct rq *rq; |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 500 | |
| 501 | repeat_lock_task: |
| 502 | rq = task_rq(p); |
| 503 | spin_lock(&rq->lock); |
| 504 | if (unlikely(rq != task_rq(p))) { |
| 505 | spin_unlock(&rq->lock); |
| 506 | goto repeat_lock_task; |
| 507 | } |
| 508 | return rq; |
| 509 | } |
| 510 | |
| 511 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 512 | * task_rq_lock - lock the runqueue a given task resides on and disable |
| 513 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
| 514 | * explicitly disabling preemption. |
| 515 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 516 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 517 | __acquires(rq->lock) |
| 518 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 519 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 520 | |
| 521 | repeat_lock_task: |
| 522 | local_irq_save(*flags); |
| 523 | rq = task_rq(p); |
| 524 | spin_lock(&rq->lock); |
| 525 | if (unlikely(rq != task_rq(p))) { |
| 526 | spin_unlock_irqrestore(&rq->lock, *flags); |
| 527 | goto repeat_lock_task; |
| 528 | } |
| 529 | return rq; |
| 530 | } |
| 531 | |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 532 | static inline void __task_rq_unlock(struct rq *rq) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 533 | __releases(rq->lock) |
| 534 | { |
| 535 | spin_unlock(&rq->lock); |
| 536 | } |
| 537 | |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 538 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 539 | __releases(rq->lock) |
| 540 | { |
| 541 | spin_unlock_irqrestore(&rq->lock, *flags); |
| 542 | } |
| 543 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | /* |
Robert P. J. Day | cc2a73b | 2006-12-10 02:20:00 -0800 | [diff] [blame] | 545 | * this_rq_lock - lock this runqueue and disable interrupts. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 546 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 547 | static inline struct rq *this_rq_lock(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | __acquires(rq->lock) |
| 549 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 550 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 551 | |
| 552 | local_irq_disable(); |
| 553 | rq = this_rq(); |
| 554 | spin_lock(&rq->lock); |
| 555 | |
| 556 | return rq; |
| 557 | } |
| 558 | |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 559 | /* |
Ingo Molnar | 2aa44d0 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 560 | * We are going deep-idle (irqs are disabled): |
Ingo Molnar | 1b9f19c | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 561 | */ |
Ingo Molnar | 2aa44d0 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 562 | void sched_clock_idle_sleep_event(void) |
Ingo Molnar | 1b9f19c | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 563 | { |
Ingo Molnar | 2aa44d0 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 564 | struct rq *rq = cpu_rq(smp_processor_id()); |
Ingo Molnar | 1b9f19c | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 565 | |
Ingo Molnar | 2aa44d0 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 566 | spin_lock(&rq->lock); |
| 567 | __update_rq_clock(rq); |
| 568 | spin_unlock(&rq->lock); |
| 569 | rq->clock_deep_idle_events++; |
Ingo Molnar | 1b9f19c | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 570 | } |
Ingo Molnar | 2aa44d0 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 571 | EXPORT_SYMBOL_GPL(sched_clock_idle_sleep_event); |
| 572 | |
| 573 | /* |
| 574 | * We just idled delta nanoseconds (called with irqs disabled): |
| 575 | */ |
| 576 | void sched_clock_idle_wakeup_event(u64 delta_ns) |
| 577 | { |
| 578 | struct rq *rq = cpu_rq(smp_processor_id()); |
| 579 | u64 now = sched_clock(); |
| 580 | |
| 581 | rq->idle_clock += delta_ns; |
| 582 | /* |
| 583 | * Override the previous timestamp and ignore all |
| 584 | * sched_clock() deltas that occured while we idled, |
| 585 | * and use the PM-provided delta_ns to advance the |
| 586 | * rq clock: |
| 587 | */ |
| 588 | spin_lock(&rq->lock); |
| 589 | rq->prev_clock_raw = now; |
| 590 | rq->clock += delta_ns; |
| 591 | spin_unlock(&rq->lock); |
| 592 | } |
| 593 | EXPORT_SYMBOL_GPL(sched_clock_idle_wakeup_event); |
Ingo Molnar | 1b9f19c | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 594 | |
| 595 | /* |
Ingo Molnar | c24d20d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 596 | * resched_task - mark a task 'to be rescheduled now'. |
| 597 | * |
| 598 | * On UP this means the setting of the need_resched flag, on SMP it |
| 599 | * might also involve a cross-CPU call to trigger the scheduler on |
| 600 | * the target CPU. |
| 601 | */ |
| 602 | #ifdef CONFIG_SMP |
| 603 | |
| 604 | #ifndef tsk_is_polling |
| 605 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) |
| 606 | #endif |
| 607 | |
| 608 | static void resched_task(struct task_struct *p) |
| 609 | { |
| 610 | int cpu; |
| 611 | |
| 612 | assert_spin_locked(&task_rq(p)->lock); |
| 613 | |
| 614 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
| 615 | return; |
| 616 | |
| 617 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
| 618 | |
| 619 | cpu = task_cpu(p); |
| 620 | if (cpu == smp_processor_id()) |
| 621 | return; |
| 622 | |
| 623 | /* NEED_RESCHED must be visible before we test polling */ |
| 624 | smp_mb(); |
| 625 | if (!tsk_is_polling(p)) |
| 626 | smp_send_reschedule(cpu); |
| 627 | } |
| 628 | |
| 629 | static void resched_cpu(int cpu) |
| 630 | { |
| 631 | struct rq *rq = cpu_rq(cpu); |
| 632 | unsigned long flags; |
| 633 | |
| 634 | if (!spin_trylock_irqsave(&rq->lock, flags)) |
| 635 | return; |
| 636 | resched_task(cpu_curr(cpu)); |
| 637 | spin_unlock_irqrestore(&rq->lock, flags); |
| 638 | } |
| 639 | #else |
| 640 | static inline void resched_task(struct task_struct *p) |
| 641 | { |
| 642 | assert_spin_locked(&task_rq(p)->lock); |
| 643 | set_tsk_need_resched(p); |
| 644 | } |
| 645 | #endif |
| 646 | |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 647 | static u64 div64_likely32(u64 divident, unsigned long divisor) |
| 648 | { |
| 649 | #if BITS_PER_LONG == 32 |
| 650 | if (likely(divident <= 0xffffffffULL)) |
| 651 | return (u32)divident / divisor; |
| 652 | do_div(divident, divisor); |
| 653 | |
| 654 | return divident; |
| 655 | #else |
| 656 | return divident / divisor; |
| 657 | #endif |
| 658 | } |
| 659 | |
| 660 | #if BITS_PER_LONG == 32 |
| 661 | # define WMULT_CONST (~0UL) |
| 662 | #else |
| 663 | # define WMULT_CONST (1UL << 32) |
| 664 | #endif |
| 665 | |
| 666 | #define WMULT_SHIFT 32 |
| 667 | |
Ingo Molnar | 194081e | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 668 | /* |
| 669 | * Shift right and round: |
| 670 | */ |
Ingo Molnar | cf2ab46 | 2007-09-05 14:32:49 +0200 | [diff] [blame] | 671 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
Ingo Molnar | 194081e | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 672 | |
Ingo Molnar | cb1c4fc | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 673 | static unsigned long |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 674 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
| 675 | struct load_weight *lw) |
| 676 | { |
| 677 | u64 tmp; |
| 678 | |
| 679 | if (unlikely(!lw->inv_weight)) |
Ingo Molnar | 194081e | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 680 | lw->inv_weight = (WMULT_CONST - lw->weight/2) / lw->weight + 1; |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 681 | |
| 682 | tmp = (u64)delta_exec * weight; |
| 683 | /* |
| 684 | * Check whether we'd overflow the 64-bit multiplication: |
| 685 | */ |
Ingo Molnar | 194081e | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 686 | if (unlikely(tmp > WMULT_CONST)) |
Ingo Molnar | cf2ab46 | 2007-09-05 14:32:49 +0200 | [diff] [blame] | 687 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
Ingo Molnar | 194081e | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 688 | WMULT_SHIFT/2); |
| 689 | else |
Ingo Molnar | cf2ab46 | 2007-09-05 14:32:49 +0200 | [diff] [blame] | 690 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 691 | |
Ingo Molnar | ecf691d | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 692 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 693 | } |
| 694 | |
| 695 | static inline unsigned long |
| 696 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) |
| 697 | { |
| 698 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); |
| 699 | } |
| 700 | |
| 701 | static void update_load_add(struct load_weight *lw, unsigned long inc) |
| 702 | { |
| 703 | lw->weight += inc; |
| 704 | lw->inv_weight = 0; |
| 705 | } |
| 706 | |
| 707 | static void update_load_sub(struct load_weight *lw, unsigned long dec) |
| 708 | { |
| 709 | lw->weight -= dec; |
| 710 | lw->inv_weight = 0; |
| 711 | } |
| 712 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 713 | /* |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 714 | * To aid in avoiding the subversion of "niceness" due to uneven distribution |
| 715 | * of tasks with abnormal "nice" values across CPUs the contribution that |
| 716 | * each task makes to its run queue's load is weighted according to its |
| 717 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
| 718 | * scaled version of the new time slice allocation that they receive on time |
| 719 | * slice expiry etc. |
| 720 | */ |
| 721 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 722 | #define WEIGHT_IDLEPRIO 2 |
| 723 | #define WMULT_IDLEPRIO (1 << 31) |
| 724 | |
| 725 | /* |
| 726 | * Nice levels are multiplicative, with a gentle 10% change for every |
| 727 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to |
| 728 | * nice 1, it will get ~10% less CPU time than another CPU-bound task |
| 729 | * that remained on nice 0. |
| 730 | * |
| 731 | * The "10% effect" is relative and cumulative: from _any_ nice level, |
| 732 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level |
Ingo Molnar | f9153ee | 2007-07-16 09:46:30 +0200 | [diff] [blame] | 733 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
| 734 | * If a task goes up by ~10% and another task goes down by ~10% then |
| 735 | * the relative distance between them is ~25%.) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 736 | */ |
| 737 | static const int prio_to_weight[40] = { |
Ingo Molnar | 254753d | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 738 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
| 739 | /* -15 */ 29154, 23254, 18705, 14949, 11916, |
| 740 | /* -10 */ 9548, 7620, 6100, 4904, 3906, |
| 741 | /* -5 */ 3121, 2501, 1991, 1586, 1277, |
| 742 | /* 0 */ 1024, 820, 655, 526, 423, |
| 743 | /* 5 */ 335, 272, 215, 172, 137, |
| 744 | /* 10 */ 110, 87, 70, 56, 45, |
| 745 | /* 15 */ 36, 29, 23, 18, 15, |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 746 | }; |
| 747 | |
Ingo Molnar | 5714d2d | 2007-07-16 09:46:31 +0200 | [diff] [blame] | 748 | /* |
| 749 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. |
| 750 | * |
| 751 | * In cases where the weight does not change often, we can use the |
| 752 | * precalculated inverse to speed up arithmetics by turning divisions |
| 753 | * into multiplications: |
| 754 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 755 | static const u32 prio_to_wmult[40] = { |
Ingo Molnar | 254753d | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 756 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
| 757 | /* -15 */ 147320, 184698, 229616, 287308, 360437, |
| 758 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, |
| 759 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, |
| 760 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, |
| 761 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, |
| 762 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, |
| 763 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 764 | }; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 765 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 766 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
| 767 | |
| 768 | /* |
| 769 | * runqueue iterator, to support SMP load-balancing between different |
| 770 | * scheduling classes, without having to expose their internal data |
| 771 | * structures to the load-balancing proper: |
| 772 | */ |
| 773 | struct rq_iterator { |
| 774 | void *arg; |
| 775 | struct task_struct *(*start)(void *); |
| 776 | struct task_struct *(*next)(void *); |
| 777 | }; |
| 778 | |
| 779 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
| 780 | unsigned long max_nr_move, unsigned long max_load_move, |
| 781 | struct sched_domain *sd, enum cpu_idle_type idle, |
| 782 | int *all_pinned, unsigned long *load_moved, |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 783 | int *this_best_prio, struct rq_iterator *iterator); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 784 | |
| 785 | #include "sched_stats.h" |
| 786 | #include "sched_rt.c" |
| 787 | #include "sched_fair.c" |
| 788 | #include "sched_idletask.c" |
| 789 | #ifdef CONFIG_SCHED_DEBUG |
| 790 | # include "sched_debug.c" |
| 791 | #endif |
| 792 | |
| 793 | #define sched_class_highest (&rt_sched_class) |
| 794 | |
Ingo Molnar | 9c21724 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 795 | static void __update_curr_load(struct rq *rq, struct load_stat *ls) |
| 796 | { |
| 797 | if (rq->curr != rq->idle && ls->load.weight) { |
| 798 | ls->delta_exec += ls->delta_stat; |
| 799 | ls->delta_fair += calc_delta_fair(ls->delta_stat, &ls->load); |
| 800 | ls->delta_stat = 0; |
| 801 | } |
| 802 | } |
| 803 | |
| 804 | /* |
| 805 | * Update delta_exec, delta_fair fields for rq. |
| 806 | * |
| 807 | * delta_fair clock advances at a rate inversely proportional to |
| 808 | * total load (rq->ls.load.weight) on the runqueue, while |
| 809 | * delta_exec advances at the same rate as wall-clock (provided |
| 810 | * cpu is not idle). |
| 811 | * |
| 812 | * delta_exec / delta_fair is a measure of the (smoothened) load on this |
| 813 | * runqueue over any given interval. This (smoothened) load is used |
| 814 | * during load balance. |
| 815 | * |
| 816 | * This function is called /before/ updating rq->ls.load |
| 817 | * and when switching tasks. |
| 818 | */ |
Ingo Molnar | 84a1d7a | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 819 | static void update_curr_load(struct rq *rq) |
Ingo Molnar | 9c21724 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 820 | { |
| 821 | struct load_stat *ls = &rq->ls; |
| 822 | u64 start; |
| 823 | |
| 824 | start = ls->load_update_start; |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 825 | ls->load_update_start = rq->clock; |
| 826 | ls->delta_stat += rq->clock - start; |
Ingo Molnar | 9c21724 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 827 | /* |
| 828 | * Stagger updates to ls->delta_fair. Very frequent updates |
| 829 | * can be expensive. |
| 830 | */ |
| 831 | if (ls->delta_stat >= sysctl_sched_stat_granularity) |
| 832 | __update_curr_load(rq, ls); |
| 833 | } |
| 834 | |
Ingo Molnar | 29b4b62 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 835 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
Ingo Molnar | 9c21724 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 836 | { |
Ingo Molnar | 84a1d7a | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 837 | update_curr_load(rq); |
Ingo Molnar | 9c21724 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 838 | update_load_add(&rq->ls.load, p->se.load.weight); |
| 839 | } |
| 840 | |
Ingo Molnar | 79b5ddd | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 841 | static inline void dec_load(struct rq *rq, const struct task_struct *p) |
Ingo Molnar | 9c21724 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 842 | { |
Ingo Molnar | 84a1d7a | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 843 | update_curr_load(rq); |
Ingo Molnar | 9c21724 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 844 | update_load_sub(&rq->ls.load, p->se.load.weight); |
| 845 | } |
| 846 | |
Ingo Molnar | e5fa223 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 847 | static void inc_nr_running(struct task_struct *p, struct rq *rq) |
Ingo Molnar | 9c21724 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 848 | { |
| 849 | rq->nr_running++; |
Ingo Molnar | 29b4b62 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 850 | inc_load(rq, p); |
Ingo Molnar | 9c21724 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 851 | } |
| 852 | |
Ingo Molnar | db53181 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 853 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
Ingo Molnar | 9c21724 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 854 | { |
| 855 | rq->nr_running--; |
Ingo Molnar | 79b5ddd | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 856 | dec_load(rq, p); |
Ingo Molnar | 9c21724 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 857 | } |
| 858 | |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 859 | static void set_load_weight(struct task_struct *p) |
| 860 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 861 | p->se.wait_runtime = 0; |
| 862 | |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 863 | if (task_has_rt_policy(p)) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 864 | p->se.load.weight = prio_to_weight[0] * 2; |
| 865 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; |
| 866 | return; |
| 867 | } |
| 868 | |
| 869 | /* |
| 870 | * SCHED_IDLE tasks get minimal weight: |
| 871 | */ |
| 872 | if (p->policy == SCHED_IDLE) { |
| 873 | p->se.load.weight = WEIGHT_IDLEPRIO; |
| 874 | p->se.load.inv_weight = WMULT_IDLEPRIO; |
| 875 | return; |
| 876 | } |
| 877 | |
| 878 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
| 879 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; |
Ingo Molnar | 45bf76d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 880 | } |
| 881 | |
Ingo Molnar | 8159f87 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 882 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
Ingo Molnar | 71f8bd4 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 883 | { |
| 884 | sched_info_queued(p); |
Ingo Molnar | fd390f6 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 885 | p->sched_class->enqueue_task(rq, p, wakeup); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 886 | p->se.on_rq = 1; |
| 887 | } |
| 888 | |
Ingo Molnar | 69be72c | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 889 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 890 | { |
Ingo Molnar | f02231e | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 891 | p->sched_class->dequeue_task(rq, p, sleep); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 892 | p->se.on_rq = 0; |
Ingo Molnar | 71f8bd4 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 893 | } |
| 894 | |
| 895 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 896 | * __normal_prio - return the priority that is based on the static prio |
Ingo Molnar | 71f8bd4 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 897 | */ |
Ingo Molnar | 1453118 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 898 | static inline int __normal_prio(struct task_struct *p) |
| 899 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 900 | return p->static_prio; |
Ingo Molnar | 1453118 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 901 | } |
| 902 | |
| 903 | /* |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 904 | * Calculate the expected normal priority: i.e. priority |
| 905 | * without taking RT-inheritance into account. Might be |
| 906 | * boosted by interactivity modifiers. Changes upon fork, |
| 907 | * setprio syscalls, and whenever the interactivity |
| 908 | * estimator recalculates. |
| 909 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 910 | static inline int normal_prio(struct task_struct *p) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 911 | { |
| 912 | int prio; |
| 913 | |
Ingo Molnar | e05606d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 914 | if (task_has_rt_policy(p)) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 915 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
| 916 | else |
| 917 | prio = __normal_prio(p); |
| 918 | return prio; |
| 919 | } |
| 920 | |
| 921 | /* |
| 922 | * Calculate the current priority, i.e. the priority |
| 923 | * taken into account by the scheduler. This value might |
| 924 | * be boosted by RT tasks, or might be boosted by |
| 925 | * interactivity modifiers. Will be RT if the task got |
| 926 | * RT-boosted. If not then it returns p->normal_prio. |
| 927 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 928 | static int effective_prio(struct task_struct *p) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 929 | { |
| 930 | p->normal_prio = normal_prio(p); |
| 931 | /* |
| 932 | * If we are RT tasks or we were boosted to RT priority, |
| 933 | * keep the priority unchanged. Otherwise, update priority |
| 934 | * to the normal priority: |
| 935 | */ |
| 936 | if (!rt_prio(p->prio)) |
| 937 | return p->normal_prio; |
| 938 | return p->prio; |
| 939 | } |
| 940 | |
| 941 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 942 | * activate_task - move a task to the runqueue. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 943 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 944 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 945 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 946 | if (p->state == TASK_UNINTERRUPTIBLE) |
| 947 | rq->nr_uninterruptible--; |
| 948 | |
Ingo Molnar | 8159f87 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 949 | enqueue_task(rq, p, wakeup); |
Ingo Molnar | e5fa223 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 950 | inc_nr_running(p, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 951 | } |
| 952 | |
| 953 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 954 | * activate_idle_task - move idle task to the _front_ of runqueue. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 955 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 956 | static inline void activate_idle_task(struct task_struct *p, struct rq *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 957 | { |
Ingo Molnar | a8e504d | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 958 | update_rq_clock(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 959 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 960 | if (p->state == TASK_UNINTERRUPTIBLE) |
| 961 | rq->nr_uninterruptible--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 962 | |
Ingo Molnar | 8159f87 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 963 | enqueue_task(rq, p, 0); |
Ingo Molnar | e5fa223 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 964 | inc_nr_running(p, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 965 | } |
| 966 | |
| 967 | /* |
| 968 | * deactivate_task - remove a task from the runqueue. |
| 969 | */ |
Ingo Molnar | 2e1cb74 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 970 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 971 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 972 | if (p->state == TASK_UNINTERRUPTIBLE) |
| 973 | rq->nr_uninterruptible++; |
| 974 | |
Ingo Molnar | 69be72c | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 975 | dequeue_task(rq, p, sleep); |
Ingo Molnar | db53181 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 976 | dec_nr_running(p, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 977 | } |
| 978 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 979 | /** |
| 980 | * task_curr - is this task currently executing on a CPU? |
| 981 | * @p: the task in question. |
| 982 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 983 | inline int task_curr(const struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 984 | { |
| 985 | return cpu_curr(task_cpu(p)) == p; |
| 986 | } |
| 987 | |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 988 | /* Used instead of source_load when we know the type == 0 */ |
| 989 | unsigned long weighted_cpuload(const int cpu) |
| 990 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 991 | return cpu_rq(cpu)->ls.load.weight; |
| 992 | } |
| 993 | |
| 994 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
| 995 | { |
| 996 | #ifdef CONFIG_SMP |
| 997 | task_thread_info(p)->cpu = cpu; |
| 998 | set_task_cfs_rq(p); |
| 999 | #endif |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1000 | } |
| 1001 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1002 | #ifdef CONFIG_SMP |
Ingo Molnar | c65cc87 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1003 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1004 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
Ingo Molnar | c65cc87 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1005 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1006 | int old_cpu = task_cpu(p); |
| 1007 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); |
| 1008 | u64 clock_offset, fair_clock_offset; |
| 1009 | |
| 1010 | clock_offset = old_rq->clock - new_rq->clock; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 1011 | fair_clock_offset = old_rq->cfs.fair_clock - new_rq->cfs.fair_clock; |
| 1012 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1013 | if (p->se.wait_start_fair) |
| 1014 | p->se.wait_start_fair -= fair_clock_offset; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 1015 | if (p->se.sleep_start_fair) |
| 1016 | p->se.sleep_start_fair -= fair_clock_offset; |
| 1017 | |
| 1018 | #ifdef CONFIG_SCHEDSTATS |
| 1019 | if (p->se.wait_start) |
| 1020 | p->se.wait_start -= clock_offset; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1021 | if (p->se.sleep_start) |
| 1022 | p->se.sleep_start -= clock_offset; |
| 1023 | if (p->se.block_start) |
| 1024 | p->se.block_start -= clock_offset; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 1025 | #endif |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1026 | |
| 1027 | __set_task_cpu(p, new_cpu); |
Ingo Molnar | c65cc87 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 1028 | } |
| 1029 | |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1030 | struct migration_req { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1031 | struct list_head list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1032 | |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1033 | struct task_struct *task; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | int dest_cpu; |
| 1035 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | struct completion done; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1037 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1038 | |
| 1039 | /* |
| 1040 | * The task's runqueue lock must be held. |
| 1041 | * Returns true if you have to wait for migration thread. |
| 1042 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1043 | static int |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1044 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1045 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1046 | struct rq *rq = task_rq(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1047 | |
| 1048 | /* |
| 1049 | * If the task is not on a runqueue (and not running), then |
| 1050 | * it is sufficient to simply update the task's cpu field. |
| 1051 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1052 | if (!p->se.on_rq && !task_running(rq, p)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1053 | set_task_cpu(p, dest_cpu); |
| 1054 | return 0; |
| 1055 | } |
| 1056 | |
| 1057 | init_completion(&req->done); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1058 | req->task = p; |
| 1059 | req->dest_cpu = dest_cpu; |
| 1060 | list_add(&req->list, &rq->migration_queue); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 1061 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1062 | return 1; |
| 1063 | } |
| 1064 | |
| 1065 | /* |
| 1066 | * wait_task_inactive - wait for a thread to unschedule. |
| 1067 | * |
| 1068 | * The caller must ensure that the task *will* unschedule sometime soon, |
| 1069 | * else this function might spin for a *long* time. This function can't |
| 1070 | * be called with interrupts off, or it may introduce deadlock with |
| 1071 | * smp_call_function() if an IPI is sent by the same process we are |
| 1072 | * waiting to become inactive. |
| 1073 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1074 | void wait_task_inactive(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1075 | { |
| 1076 | unsigned long flags; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1077 | int running, on_rq; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1078 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1079 | |
| 1080 | repeat: |
Linus Torvalds | fa490cf | 2007-06-18 09:34:40 -0700 | [diff] [blame] | 1081 | /* |
| 1082 | * We do the initial early heuristics without holding |
| 1083 | * any task-queue locks at all. We'll only try to get |
| 1084 | * the runqueue lock when things look like they will |
| 1085 | * work out! |
| 1086 | */ |
| 1087 | rq = task_rq(p); |
| 1088 | |
| 1089 | /* |
| 1090 | * If the task is actively running on another CPU |
| 1091 | * still, just relax and busy-wait without holding |
| 1092 | * any locks. |
| 1093 | * |
| 1094 | * NOTE! Since we don't hold any locks, it's not |
| 1095 | * even sure that "rq" stays as the right runqueue! |
| 1096 | * But we don't care, since "task_running()" will |
| 1097 | * return false if the runqueue has changed and p |
| 1098 | * is actually now running somewhere else! |
| 1099 | */ |
| 1100 | while (task_running(rq, p)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1101 | cpu_relax(); |
Linus Torvalds | fa490cf | 2007-06-18 09:34:40 -0700 | [diff] [blame] | 1102 | |
| 1103 | /* |
| 1104 | * Ok, time to look more closely! We need the rq |
| 1105 | * lock now, to be *sure*. If we're wrong, we'll |
| 1106 | * just go back and repeat. |
| 1107 | */ |
| 1108 | rq = task_rq_lock(p, &flags); |
| 1109 | running = task_running(rq, p); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1110 | on_rq = p->se.on_rq; |
Linus Torvalds | fa490cf | 2007-06-18 09:34:40 -0700 | [diff] [blame] | 1111 | task_rq_unlock(rq, &flags); |
| 1112 | |
| 1113 | /* |
| 1114 | * Was it really running after all now that we |
| 1115 | * checked with the proper locks actually held? |
| 1116 | * |
| 1117 | * Oops. Go back and try again.. |
| 1118 | */ |
| 1119 | if (unlikely(running)) { |
| 1120 | cpu_relax(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1121 | goto repeat; |
| 1122 | } |
Linus Torvalds | fa490cf | 2007-06-18 09:34:40 -0700 | [diff] [blame] | 1123 | |
| 1124 | /* |
| 1125 | * It's not enough that it's not actively running, |
| 1126 | * it must be off the runqueue _entirely_, and not |
| 1127 | * preempted! |
| 1128 | * |
| 1129 | * So if it wa still runnable (but just not actively |
| 1130 | * running right now), it's preempted, and we should |
| 1131 | * yield - it could be a while. |
| 1132 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1133 | if (unlikely(on_rq)) { |
Linus Torvalds | fa490cf | 2007-06-18 09:34:40 -0700 | [diff] [blame] | 1134 | yield(); |
| 1135 | goto repeat; |
| 1136 | } |
| 1137 | |
| 1138 | /* |
| 1139 | * Ahh, all good. It wasn't running, and it wasn't |
| 1140 | * runnable, which means that it will never become |
| 1141 | * running in the future either. We're all done! |
| 1142 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | } |
| 1144 | |
| 1145 | /*** |
| 1146 | * kick_process - kick a running thread to enter/exit the kernel |
| 1147 | * @p: the to-be-kicked thread |
| 1148 | * |
| 1149 | * Cause a process which is running on another CPU to enter |
| 1150 | * kernel-mode, without any delay. (to get signals handled.) |
| 1151 | * |
| 1152 | * NOTE: this function doesnt have to take the runqueue lock, |
| 1153 | * because all it wants to ensure is that the remote task enters |
| 1154 | * the kernel. If the IPI races and the task has been migrated |
| 1155 | * to another CPU then no harm is done and the purpose has been |
| 1156 | * achieved as well. |
| 1157 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1158 | void kick_process(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1159 | { |
| 1160 | int cpu; |
| 1161 | |
| 1162 | preempt_disable(); |
| 1163 | cpu = task_cpu(p); |
| 1164 | if ((cpu != smp_processor_id()) && task_curr(p)) |
| 1165 | smp_send_reschedule(cpu); |
| 1166 | preempt_enable(); |
| 1167 | } |
| 1168 | |
| 1169 | /* |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1170 | * Return a low guess at the load of a migration-source cpu weighted |
| 1171 | * according to the scheduling class and "nice" value. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1172 | * |
| 1173 | * We want to under-estimate the load of migration sources, to |
| 1174 | * balance conservatively. |
| 1175 | */ |
Con Kolivas | b910472 | 2005-11-08 21:38:55 -0800 | [diff] [blame] | 1176 | static inline unsigned long source_load(int cpu, int type) |
| 1177 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1178 | struct rq *rq = cpu_rq(cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1179 | unsigned long total = weighted_cpuload(cpu); |
Nick Piggin | a200057 | 2006-02-10 01:51:02 -0800 | [diff] [blame] | 1180 | |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1181 | if (type == 0) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1182 | return total; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1183 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1184 | return min(rq->cpu_load[type-1], total); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1185 | } |
| 1186 | |
| 1187 | /* |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1188 | * Return a high guess at the load of a migration-target cpu weighted |
| 1189 | * according to the scheduling class and "nice" value. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1190 | */ |
Con Kolivas | b910472 | 2005-11-08 21:38:55 -0800 | [diff] [blame] | 1191 | static inline unsigned long target_load(int cpu, int type) |
| 1192 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1193 | struct rq *rq = cpu_rq(cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1194 | unsigned long total = weighted_cpuload(cpu); |
Nick Piggin | a200057 | 2006-02-10 01:51:02 -0800 | [diff] [blame] | 1195 | |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1196 | if (type == 0) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1197 | return total; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1198 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1199 | return max(rq->cpu_load[type-1], total); |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1200 | } |
| 1201 | |
| 1202 | /* |
| 1203 | * Return the average load per task on the cpu's run queue |
| 1204 | */ |
| 1205 | static inline unsigned long cpu_avg_load_per_task(int cpu) |
| 1206 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1207 | struct rq *rq = cpu_rq(cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1208 | unsigned long total = weighted_cpuload(cpu); |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1209 | unsigned long n = rq->nr_running; |
| 1210 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1211 | return n ? total / n : SCHED_LOAD_SCALE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1212 | } |
| 1213 | |
Nick Piggin | 147cbb4 | 2005-06-25 14:57:19 -0700 | [diff] [blame] | 1214 | /* |
| 1215 | * find_idlest_group finds and returns the least busy CPU group within the |
| 1216 | * domain. |
| 1217 | */ |
| 1218 | static struct sched_group * |
| 1219 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) |
| 1220 | { |
| 1221 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; |
| 1222 | unsigned long min_load = ULONG_MAX, this_load = 0; |
| 1223 | int load_idx = sd->forkexec_idx; |
| 1224 | int imbalance = 100 + (sd->imbalance_pct-100)/2; |
| 1225 | |
| 1226 | do { |
| 1227 | unsigned long load, avg_load; |
| 1228 | int local_group; |
| 1229 | int i; |
| 1230 | |
M.Baris Demiray | da5a552 | 2005-09-10 00:26:09 -0700 | [diff] [blame] | 1231 | /* Skip over this group if it has no CPUs allowed */ |
| 1232 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) |
| 1233 | goto nextgroup; |
| 1234 | |
Nick Piggin | 147cbb4 | 2005-06-25 14:57:19 -0700 | [diff] [blame] | 1235 | local_group = cpu_isset(this_cpu, group->cpumask); |
Nick Piggin | 147cbb4 | 2005-06-25 14:57:19 -0700 | [diff] [blame] | 1236 | |
| 1237 | /* Tally up the load of all CPUs in the group */ |
| 1238 | avg_load = 0; |
| 1239 | |
| 1240 | for_each_cpu_mask(i, group->cpumask) { |
| 1241 | /* Bias balancing toward cpus of our domain */ |
| 1242 | if (local_group) |
| 1243 | load = source_load(i, load_idx); |
| 1244 | else |
| 1245 | load = target_load(i, load_idx); |
| 1246 | |
| 1247 | avg_load += load; |
| 1248 | } |
| 1249 | |
| 1250 | /* Adjust by relative CPU power of the group */ |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 1251 | avg_load = sg_div_cpu_power(group, |
| 1252 | avg_load * SCHED_LOAD_SCALE); |
Nick Piggin | 147cbb4 | 2005-06-25 14:57:19 -0700 | [diff] [blame] | 1253 | |
| 1254 | if (local_group) { |
| 1255 | this_load = avg_load; |
| 1256 | this = group; |
| 1257 | } else if (avg_load < min_load) { |
| 1258 | min_load = avg_load; |
| 1259 | idlest = group; |
| 1260 | } |
M.Baris Demiray | da5a552 | 2005-09-10 00:26:09 -0700 | [diff] [blame] | 1261 | nextgroup: |
Nick Piggin | 147cbb4 | 2005-06-25 14:57:19 -0700 | [diff] [blame] | 1262 | group = group->next; |
| 1263 | } while (group != sd->groups); |
| 1264 | |
| 1265 | if (!idlest || 100*this_load < imbalance*min_load) |
| 1266 | return NULL; |
| 1267 | return idlest; |
| 1268 | } |
| 1269 | |
| 1270 | /* |
Satoru Takeuchi | 0feaece | 2006-10-03 01:14:10 -0700 | [diff] [blame] | 1271 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
Nick Piggin | 147cbb4 | 2005-06-25 14:57:19 -0700 | [diff] [blame] | 1272 | */ |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 1273 | static int |
| 1274 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
Nick Piggin | 147cbb4 | 2005-06-25 14:57:19 -0700 | [diff] [blame] | 1275 | { |
M.Baris Demiray | da5a552 | 2005-09-10 00:26:09 -0700 | [diff] [blame] | 1276 | cpumask_t tmp; |
Nick Piggin | 147cbb4 | 2005-06-25 14:57:19 -0700 | [diff] [blame] | 1277 | unsigned long load, min_load = ULONG_MAX; |
| 1278 | int idlest = -1; |
| 1279 | int i; |
| 1280 | |
M.Baris Demiray | da5a552 | 2005-09-10 00:26:09 -0700 | [diff] [blame] | 1281 | /* Traverse only the allowed CPUs */ |
| 1282 | cpus_and(tmp, group->cpumask, p->cpus_allowed); |
| 1283 | |
| 1284 | for_each_cpu_mask(i, tmp) { |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1285 | load = weighted_cpuload(i); |
Nick Piggin | 147cbb4 | 2005-06-25 14:57:19 -0700 | [diff] [blame] | 1286 | |
| 1287 | if (load < min_load || (load == min_load && i == this_cpu)) { |
| 1288 | min_load = load; |
| 1289 | idlest = i; |
| 1290 | } |
| 1291 | } |
| 1292 | |
| 1293 | return idlest; |
| 1294 | } |
| 1295 | |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1296 | /* |
| 1297 | * sched_balance_self: balance the current task (running on cpu) in domains |
| 1298 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and |
| 1299 | * SD_BALANCE_EXEC. |
| 1300 | * |
| 1301 | * Balance, ie. select the least loaded group. |
| 1302 | * |
| 1303 | * Returns the target CPU number, or the same CPU if no balancing is needed. |
| 1304 | * |
| 1305 | * preempt must be disabled. |
| 1306 | */ |
| 1307 | static int sched_balance_self(int cpu, int flag) |
| 1308 | { |
| 1309 | struct task_struct *t = current; |
| 1310 | struct sched_domain *tmp, *sd = NULL; |
Nick Piggin | 147cbb4 | 2005-06-25 14:57:19 -0700 | [diff] [blame] | 1311 | |
Chen, Kenneth W | c96d145 | 2006-06-27 02:54:28 -0700 | [diff] [blame] | 1312 | for_each_domain(cpu, tmp) { |
Ingo Molnar | 9761eea | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 1313 | /* |
| 1314 | * If power savings logic is enabled for a domain, stop there. |
| 1315 | */ |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 1316 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
| 1317 | break; |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1318 | if (tmp->flags & flag) |
| 1319 | sd = tmp; |
Chen, Kenneth W | c96d145 | 2006-06-27 02:54:28 -0700 | [diff] [blame] | 1320 | } |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1321 | |
| 1322 | while (sd) { |
| 1323 | cpumask_t span; |
| 1324 | struct sched_group *group; |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 1325 | int new_cpu, weight; |
| 1326 | |
| 1327 | if (!(sd->flags & flag)) { |
| 1328 | sd = sd->child; |
| 1329 | continue; |
| 1330 | } |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1331 | |
| 1332 | span = sd->span; |
| 1333 | group = find_idlest_group(sd, t, cpu); |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 1334 | if (!group) { |
| 1335 | sd = sd->child; |
| 1336 | continue; |
| 1337 | } |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1338 | |
M.Baris Demiray | da5a552 | 2005-09-10 00:26:09 -0700 | [diff] [blame] | 1339 | new_cpu = find_idlest_cpu(group, t, cpu); |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 1340 | if (new_cpu == -1 || new_cpu == cpu) { |
| 1341 | /* Now try balancing at a lower domain level of cpu */ |
| 1342 | sd = sd->child; |
| 1343 | continue; |
| 1344 | } |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1345 | |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 1346 | /* Now try balancing at a lower domain level of new_cpu */ |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1347 | cpu = new_cpu; |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1348 | sd = NULL; |
| 1349 | weight = cpus_weight(span); |
| 1350 | for_each_domain(cpu, tmp) { |
| 1351 | if (weight <= cpus_weight(tmp->span)) |
| 1352 | break; |
| 1353 | if (tmp->flags & flag) |
| 1354 | sd = tmp; |
| 1355 | } |
| 1356 | /* while loop will break here if sd == NULL */ |
| 1357 | } |
| 1358 | |
| 1359 | return cpu; |
| 1360 | } |
| 1361 | |
| 1362 | #endif /* CONFIG_SMP */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1363 | |
| 1364 | /* |
| 1365 | * wake_idle() will wake a task on an idle cpu if task->cpu is |
| 1366 | * not idle and an idle cpu is available. The span of cpus to |
| 1367 | * search starts with cpus closest then further out as needed, |
| 1368 | * so we always favor a closer, idle cpu. |
| 1369 | * |
| 1370 | * Returns the CPU we should wake onto. |
| 1371 | */ |
| 1372 | #if defined(ARCH_HAS_SCHED_WAKE_IDLE) |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1373 | static int wake_idle(int cpu, struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1374 | { |
| 1375 | cpumask_t tmp; |
| 1376 | struct sched_domain *sd; |
| 1377 | int i; |
| 1378 | |
Siddha, Suresh B | 4953198 | 2007-05-08 00:33:01 -0700 | [diff] [blame] | 1379 | /* |
| 1380 | * If it is idle, then it is the best cpu to run this task. |
| 1381 | * |
| 1382 | * This cpu is also the best, if it has more than one task already. |
| 1383 | * Siblings must be also busy(in most cases) as they didn't already |
| 1384 | * pickup the extra load from this cpu and hence we need not check |
| 1385 | * sibling runqueue info. This will avoid the checks and cache miss |
| 1386 | * penalities associated with that. |
| 1387 | */ |
| 1388 | if (idle_cpu(cpu) || cpu_rq(cpu)->nr_running > 1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1389 | return cpu; |
| 1390 | |
| 1391 | for_each_domain(cpu, sd) { |
| 1392 | if (sd->flags & SD_WAKE_IDLE) { |
Nick Piggin | e0f364f | 2005-06-25 14:57:06 -0700 | [diff] [blame] | 1393 | cpus_and(tmp, sd->span, p->cpus_allowed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1394 | for_each_cpu_mask(i, tmp) { |
| 1395 | if (idle_cpu(i)) |
| 1396 | return i; |
| 1397 | } |
Ingo Molnar | 9761eea | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 1398 | } else { |
Nick Piggin | e0f364f | 2005-06-25 14:57:06 -0700 | [diff] [blame] | 1399 | break; |
Ingo Molnar | 9761eea | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 1400 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1401 | } |
| 1402 | return cpu; |
| 1403 | } |
| 1404 | #else |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1405 | static inline int wake_idle(int cpu, struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1406 | { |
| 1407 | return cpu; |
| 1408 | } |
| 1409 | #endif |
| 1410 | |
| 1411 | /*** |
| 1412 | * try_to_wake_up - wake up a thread |
| 1413 | * @p: the to-be-woken-up thread |
| 1414 | * @state: the mask of task states that can be woken |
| 1415 | * @sync: do a synchronous wakeup? |
| 1416 | * |
| 1417 | * Put it on the run-queue if it's not already there. The "current" |
| 1418 | * thread is always on the run-queue (except when the actual |
| 1419 | * re-schedule is in progress), and as such you're allowed to do |
| 1420 | * the simpler "current->state = TASK_RUNNING" to mark yourself |
| 1421 | * runnable without the overhead of this. |
| 1422 | * |
| 1423 | * returns failure only if the task is already active. |
| 1424 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1425 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1426 | { |
| 1427 | int cpu, this_cpu, success = 0; |
| 1428 | unsigned long flags; |
| 1429 | long old_state; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1430 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1431 | #ifdef CONFIG_SMP |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 1432 | struct sched_domain *sd, *this_sd = NULL; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1433 | unsigned long load, this_load; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1434 | int new_cpu; |
| 1435 | #endif |
| 1436 | |
| 1437 | rq = task_rq_lock(p, &flags); |
| 1438 | old_state = p->state; |
| 1439 | if (!(old_state & state)) |
| 1440 | goto out; |
| 1441 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1442 | if (p->se.on_rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1443 | goto out_running; |
| 1444 | |
| 1445 | cpu = task_cpu(p); |
| 1446 | this_cpu = smp_processor_id(); |
| 1447 | |
| 1448 | #ifdef CONFIG_SMP |
| 1449 | if (unlikely(task_running(rq, p))) |
| 1450 | goto out_activate; |
| 1451 | |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 1452 | new_cpu = cpu; |
| 1453 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1454 | schedstat_inc(rq, ttwu_cnt); |
| 1455 | if (cpu == this_cpu) { |
| 1456 | schedstat_inc(rq, ttwu_local); |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 1457 | goto out_set_cpu; |
| 1458 | } |
| 1459 | |
| 1460 | for_each_domain(this_cpu, sd) { |
| 1461 | if (cpu_isset(cpu, sd->span)) { |
| 1462 | schedstat_inc(sd, ttwu_wake_remote); |
| 1463 | this_sd = sd; |
| 1464 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1465 | } |
| 1466 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1467 | |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 1468 | if (unlikely(!cpu_isset(this_cpu, p->cpus_allowed))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1469 | goto out_set_cpu; |
| 1470 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1471 | /* |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 1472 | * Check for affine wakeup and passive balancing possibilities. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1473 | */ |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 1474 | if (this_sd) { |
| 1475 | int idx = this_sd->wake_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1476 | unsigned int imbalance; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1477 | |
Nick Piggin | a3f21bc | 2005-06-25 14:57:15 -0700 | [diff] [blame] | 1478 | imbalance = 100 + (this_sd->imbalance_pct - 100) / 2; |
| 1479 | |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 1480 | load = source_load(cpu, idx); |
| 1481 | this_load = target_load(this_cpu, idx); |
| 1482 | |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 1483 | new_cpu = this_cpu; /* Wake to this CPU if we can */ |
| 1484 | |
Nick Piggin | a3f21bc | 2005-06-25 14:57:15 -0700 | [diff] [blame] | 1485 | if (this_sd->flags & SD_WAKE_AFFINE) { |
| 1486 | unsigned long tl = this_load; |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 1487 | unsigned long tl_per_task; |
| 1488 | |
| 1489 | tl_per_task = cpu_avg_load_per_task(this_cpu); |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1490 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1491 | /* |
Nick Piggin | a3f21bc | 2005-06-25 14:57:15 -0700 | [diff] [blame] | 1492 | * If sync wakeup then subtract the (maximum possible) |
| 1493 | * effect of the currently running task from the load |
| 1494 | * of the current CPU: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1495 | */ |
Nick Piggin | a3f21bc | 2005-06-25 14:57:15 -0700 | [diff] [blame] | 1496 | if (sync) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1497 | tl -= current->se.load.weight; |
Nick Piggin | a3f21bc | 2005-06-25 14:57:15 -0700 | [diff] [blame] | 1498 | |
| 1499 | if ((tl <= load && |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 1500 | tl + target_load(cpu, idx) <= tl_per_task) || |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1501 | 100*(tl + p->se.load.weight) <= imbalance*load) { |
Nick Piggin | a3f21bc | 2005-06-25 14:57:15 -0700 | [diff] [blame] | 1502 | /* |
| 1503 | * This domain has SD_WAKE_AFFINE and |
| 1504 | * p is cache cold in this domain, and |
| 1505 | * there is no bad imbalance. |
| 1506 | */ |
| 1507 | schedstat_inc(this_sd, ttwu_move_affine); |
| 1508 | goto out_set_cpu; |
| 1509 | } |
| 1510 | } |
| 1511 | |
| 1512 | /* |
| 1513 | * Start passive balancing when half the imbalance_pct |
| 1514 | * limit is reached. |
| 1515 | */ |
| 1516 | if (this_sd->flags & SD_WAKE_BALANCE) { |
| 1517 | if (imbalance*this_load <= 100*load) { |
| 1518 | schedstat_inc(this_sd, ttwu_move_balance); |
| 1519 | goto out_set_cpu; |
| 1520 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1521 | } |
| 1522 | } |
| 1523 | |
| 1524 | new_cpu = cpu; /* Could not wake to this_cpu. Wake to cpu instead */ |
| 1525 | out_set_cpu: |
| 1526 | new_cpu = wake_idle(new_cpu, p); |
| 1527 | if (new_cpu != cpu) { |
| 1528 | set_task_cpu(p, new_cpu); |
| 1529 | task_rq_unlock(rq, &flags); |
| 1530 | /* might preempt at this point */ |
| 1531 | rq = task_rq_lock(p, &flags); |
| 1532 | old_state = p->state; |
| 1533 | if (!(old_state & state)) |
| 1534 | goto out; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1535 | if (p->se.on_rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1536 | goto out_running; |
| 1537 | |
| 1538 | this_cpu = smp_processor_id(); |
| 1539 | cpu = task_cpu(p); |
| 1540 | } |
| 1541 | |
| 1542 | out_activate: |
| 1543 | #endif /* CONFIG_SMP */ |
Ingo Molnar | 2daa357 | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 1544 | update_rq_clock(rq); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1545 | activate_task(rq, p, 1); |
Ingo Molnar | d79fc0f | 2005-09-10 00:26:12 -0700 | [diff] [blame] | 1546 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1547 | * Sync wakeups (i.e. those types of wakeups where the waker |
| 1548 | * has indicated that it will leave the CPU in short order) |
| 1549 | * don't trigger a preemption, if the woken up task will run on |
| 1550 | * this cpu. (in this case the 'I will reschedule' promise of |
| 1551 | * the waker guarantees that the freshly woken up task is going |
| 1552 | * to be considered on this CPU.) |
| 1553 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1554 | if (!sync || cpu != this_cpu) |
| 1555 | check_preempt_curr(rq, p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1556 | success = 1; |
| 1557 | |
| 1558 | out_running: |
| 1559 | p->state = TASK_RUNNING; |
| 1560 | out: |
| 1561 | task_rq_unlock(rq, &flags); |
| 1562 | |
| 1563 | return success; |
| 1564 | } |
| 1565 | |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1566 | int fastcall wake_up_process(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1567 | { |
| 1568 | return try_to_wake_up(p, TASK_STOPPED | TASK_TRACED | |
| 1569 | TASK_INTERRUPTIBLE | TASK_UNINTERRUPTIBLE, 0); |
| 1570 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1571 | EXPORT_SYMBOL(wake_up_process); |
| 1572 | |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1573 | int fastcall wake_up_state(struct task_struct *p, unsigned int state) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1574 | { |
| 1575 | return try_to_wake_up(p, state, 0); |
| 1576 | } |
| 1577 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1578 | /* |
| 1579 | * Perform scheduler related setup for a newly forked process p. |
| 1580 | * p is forked by current. |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1581 | * |
| 1582 | * __sched_fork() is basic setup used by init_idle() too: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1583 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1584 | static void __sched_fork(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1585 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1586 | p->se.wait_start_fair = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1587 | p->se.exec_start = 0; |
| 1588 | p->se.sum_exec_runtime = 0; |
Ingo Molnar | f6cf891 | 2007-08-28 12:53:24 +0200 | [diff] [blame] | 1589 | p->se.prev_sum_exec_runtime = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1590 | p->se.delta_exec = 0; |
| 1591 | p->se.delta_fair_run = 0; |
| 1592 | p->se.delta_fair_sleep = 0; |
| 1593 | p->se.wait_runtime = 0; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 1594 | p->se.sleep_start_fair = 0; |
| 1595 | |
| 1596 | #ifdef CONFIG_SCHEDSTATS |
| 1597 | p->se.wait_start = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1598 | p->se.sum_wait_runtime = 0; |
| 1599 | p->se.sum_sleep_runtime = 0; |
| 1600 | p->se.sleep_start = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1601 | p->se.block_start = 0; |
| 1602 | p->se.sleep_max = 0; |
| 1603 | p->se.block_max = 0; |
| 1604 | p->se.exec_max = 0; |
| 1605 | p->se.wait_max = 0; |
| 1606 | p->se.wait_runtime_overruns = 0; |
| 1607 | p->se.wait_runtime_underruns = 0; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 1608 | #endif |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1609 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1610 | INIT_LIST_HEAD(&p->run_list); |
| 1611 | p->se.on_rq = 0; |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1612 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1613 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
| 1614 | INIT_HLIST_HEAD(&p->preempt_notifiers); |
| 1615 | #endif |
| 1616 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1617 | /* |
| 1618 | * We mark the process as running here, but have not actually |
| 1619 | * inserted it onto the runqueue yet. This guarantees that |
| 1620 | * nobody will actually run it, and a signal or other external |
| 1621 | * event cannot wake it up and insert it on the runqueue either. |
| 1622 | */ |
| 1623 | p->state = TASK_RUNNING; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1624 | } |
| 1625 | |
| 1626 | /* |
| 1627 | * fork()/clone()-time setup: |
| 1628 | */ |
| 1629 | void sched_fork(struct task_struct *p, int clone_flags) |
| 1630 | { |
| 1631 | int cpu = get_cpu(); |
| 1632 | |
| 1633 | __sched_fork(p); |
| 1634 | |
| 1635 | #ifdef CONFIG_SMP |
| 1636 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); |
| 1637 | #endif |
| 1638 | __set_task_cpu(p, cpu); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 1639 | |
| 1640 | /* |
| 1641 | * Make sure we do not leak PI boosting priority to the child: |
| 1642 | */ |
| 1643 | p->prio = current->normal_prio; |
| 1644 | |
Chandra Seetharaman | 52f17b6 | 2006-07-14 00:24:38 -0700 | [diff] [blame] | 1645 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1646 | if (likely(sched_info_on())) |
Chandra Seetharaman | 52f17b6 | 2006-07-14 00:24:38 -0700 | [diff] [blame] | 1647 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1648 | #endif |
Chen, Kenneth W | d6077cb | 2006-02-14 13:53:10 -0800 | [diff] [blame] | 1649 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1650 | p->oncpu = 0; |
| 1651 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1652 | #ifdef CONFIG_PREEMPT |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1653 | /* Want to start with kernel preemption disabled. */ |
Al Viro | a1261f5 | 2005-11-13 16:06:55 -0800 | [diff] [blame] | 1654 | task_thread_info(p)->preempt_count = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1655 | #endif |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 1656 | put_cpu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1657 | } |
| 1658 | |
| 1659 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1660 | * After fork, child runs first. (default) If set to 0 then |
| 1661 | * parent will (try to) run first. |
| 1662 | */ |
| 1663 | unsigned int __read_mostly sysctl_sched_child_runs_first = 1; |
| 1664 | |
| 1665 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1666 | * wake_up_new_task - wake up a newly created task for the first time. |
| 1667 | * |
| 1668 | * This function will do some initial scheduler statistics housekeeping |
| 1669 | * that must be done for every newly created context, then puts the task |
| 1670 | * on the runqueue and wakes it. |
| 1671 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1672 | void fastcall wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1673 | { |
| 1674 | unsigned long flags; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1675 | struct rq *rq; |
| 1676 | int this_cpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1677 | |
| 1678 | rq = task_rq_lock(p, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1679 | BUG_ON(p->state != TASK_RUNNING); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1680 | this_cpu = smp_processor_id(); /* parent's CPU */ |
Ingo Molnar | a8e504d | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 1681 | update_rq_clock(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1682 | |
| 1683 | p->prio = effective_prio(p); |
| 1684 | |
Hiroshi Shimamoto | 9c95e73 | 2007-09-19 23:34:46 +0200 | [diff] [blame] | 1685 | if (rt_prio(p->prio)) |
| 1686 | p->sched_class = &rt_sched_class; |
| 1687 | else |
| 1688 | p->sched_class = &fair_sched_class; |
| 1689 | |
Ingo Molnar | cad60d9 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 1690 | if (!p->sched_class->task_new || !sysctl_sched_child_runs_first || |
| 1691 | (clone_flags & CLONE_VM) || task_cpu(p) != this_cpu || |
| 1692 | !current->se.on_rq) { |
| 1693 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1694 | activate_task(rq, p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1695 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1696 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1697 | * Let the scheduling class do new task startup |
| 1698 | * management (if any): |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1699 | */ |
Ingo Molnar | ee0827d | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 1700 | p->sched_class->task_new(rq, p); |
Ingo Molnar | e5fa223 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 1701 | inc_nr_running(p, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1702 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1703 | check_preempt_curr(rq, p); |
| 1704 | task_rq_unlock(rq, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1705 | } |
| 1706 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1707 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
| 1708 | |
| 1709 | /** |
Randy Dunlap | 421cee2 | 2007-07-31 00:37:50 -0700 | [diff] [blame] | 1710 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
| 1711 | * @notifier: notifier struct to register |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1712 | */ |
| 1713 | void preempt_notifier_register(struct preempt_notifier *notifier) |
| 1714 | { |
| 1715 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); |
| 1716 | } |
| 1717 | EXPORT_SYMBOL_GPL(preempt_notifier_register); |
| 1718 | |
| 1719 | /** |
| 1720 | * preempt_notifier_unregister - no longer interested in preemption notifications |
Randy Dunlap | 421cee2 | 2007-07-31 00:37:50 -0700 | [diff] [blame] | 1721 | * @notifier: notifier struct to unregister |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1722 | * |
| 1723 | * This is safe to call from within a preemption notifier. |
| 1724 | */ |
| 1725 | void preempt_notifier_unregister(struct preempt_notifier *notifier) |
| 1726 | { |
| 1727 | hlist_del(¬ifier->link); |
| 1728 | } |
| 1729 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); |
| 1730 | |
| 1731 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
| 1732 | { |
| 1733 | struct preempt_notifier *notifier; |
| 1734 | struct hlist_node *node; |
| 1735 | |
| 1736 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) |
| 1737 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); |
| 1738 | } |
| 1739 | |
| 1740 | static void |
| 1741 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
| 1742 | struct task_struct *next) |
| 1743 | { |
| 1744 | struct preempt_notifier *notifier; |
| 1745 | struct hlist_node *node; |
| 1746 | |
| 1747 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) |
| 1748 | notifier->ops->sched_out(notifier, next); |
| 1749 | } |
| 1750 | |
| 1751 | #else |
| 1752 | |
| 1753 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) |
| 1754 | { |
| 1755 | } |
| 1756 | |
| 1757 | static void |
| 1758 | fire_sched_out_preempt_notifiers(struct task_struct *curr, |
| 1759 | struct task_struct *next) |
| 1760 | { |
| 1761 | } |
| 1762 | |
| 1763 | #endif |
| 1764 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1765 | /** |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1766 | * prepare_task_switch - prepare to switch tasks |
| 1767 | * @rq: the runqueue preparing to switch |
Randy Dunlap | 421cee2 | 2007-07-31 00:37:50 -0700 | [diff] [blame] | 1768 | * @prev: the current task that is being switched out |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1769 | * @next: the task we are going to switch to. |
| 1770 | * |
| 1771 | * This is called with the rq lock held and interrupts off. It must |
| 1772 | * be paired with a subsequent finish_task_switch after the context |
| 1773 | * switch. |
| 1774 | * |
| 1775 | * prepare_task_switch sets up locking and calls architecture specific |
| 1776 | * hooks. |
| 1777 | */ |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1778 | static inline void |
| 1779 | prepare_task_switch(struct rq *rq, struct task_struct *prev, |
| 1780 | struct task_struct *next) |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1781 | { |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1782 | fire_sched_out_preempt_notifiers(prev, next); |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1783 | prepare_lock_switch(rq, next); |
| 1784 | prepare_arch_switch(next); |
| 1785 | } |
| 1786 | |
| 1787 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1788 | * finish_task_switch - clean up after a task-switch |
Jeff Garzik | 344baba | 2005-09-07 01:15:17 -0400 | [diff] [blame] | 1789 | * @rq: runqueue associated with task-switch |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1790 | * @prev: the thread we just switched away from. |
| 1791 | * |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1792 | * finish_task_switch must be called after the context switch, paired |
| 1793 | * with a prepare_task_switch call before the context switch. |
| 1794 | * finish_task_switch will reconcile locking set up by prepare_task_switch, |
| 1795 | * and do any other architecture-specific cleanup actions. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1796 | * |
| 1797 | * Note that we may have delayed dropping an mm in context_switch(). If |
| 1798 | * so, we finish that here outside of the runqueue lock. (Doing it |
| 1799 | * with the lock held can cause deadlocks; see schedule() for |
| 1800 | * details.) |
| 1801 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1802 | static inline void finish_task_switch(struct rq *rq, struct task_struct *prev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1803 | __releases(rq->lock) |
| 1804 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1805 | struct mm_struct *mm = rq->prev_mm; |
Oleg Nesterov | 55a101f | 2006-09-29 02:01:10 -0700 | [diff] [blame] | 1806 | long prev_state; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1807 | |
| 1808 | rq->prev_mm = NULL; |
| 1809 | |
| 1810 | /* |
| 1811 | * A task struct has one reference for the use as "current". |
Oleg Nesterov | c394cc9 | 2006-09-29 02:01:11 -0700 | [diff] [blame] | 1812 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
Oleg Nesterov | 55a101f | 2006-09-29 02:01:10 -0700 | [diff] [blame] | 1813 | * schedule one last time. The schedule call will never return, and |
| 1814 | * the scheduled task must drop that reference. |
Oleg Nesterov | c394cc9 | 2006-09-29 02:01:11 -0700 | [diff] [blame] | 1815 | * The test for TASK_DEAD must occur while the runqueue locks are |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1816 | * still held, otherwise prev could be scheduled on another cpu, die |
| 1817 | * there before we look at prev->state, and then the reference would |
| 1818 | * be dropped twice. |
| 1819 | * Manfred Spraul <manfred@colorfullife.com> |
| 1820 | */ |
Oleg Nesterov | 55a101f | 2006-09-29 02:01:10 -0700 | [diff] [blame] | 1821 | prev_state = prev->state; |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1822 | finish_arch_switch(prev); |
| 1823 | finish_lock_switch(rq, prev); |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1824 | fire_sched_in_preempt_notifiers(current); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1825 | if (mm) |
| 1826 | mmdrop(mm); |
Oleg Nesterov | c394cc9 | 2006-09-29 02:01:11 -0700 | [diff] [blame] | 1827 | if (unlikely(prev_state == TASK_DEAD)) { |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 1828 | /* |
| 1829 | * Remove function-return probe instances associated with this |
| 1830 | * task and put them back on the free list. |
Ingo Molnar | 9761eea | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 1831 | */ |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 1832 | kprobe_flush_task(prev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1833 | put_task_struct(prev); |
bibo mao | c6fd91f | 2006-03-26 01:38:20 -0800 | [diff] [blame] | 1834 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1835 | } |
| 1836 | |
| 1837 | /** |
| 1838 | * schedule_tail - first thing a freshly forked thread must call. |
| 1839 | * @prev: the thread we just switched away from. |
| 1840 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1841 | asmlinkage void schedule_tail(struct task_struct *prev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1842 | __releases(rq->lock) |
| 1843 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1844 | struct rq *rq = this_rq(); |
| 1845 | |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 1846 | finish_task_switch(rq, prev); |
| 1847 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
| 1848 | /* In this case, finish_task_switch does not reenable preemption */ |
| 1849 | preempt_enable(); |
| 1850 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1851 | if (current->set_child_tid) |
| 1852 | put_user(current->pid, current->set_child_tid); |
| 1853 | } |
| 1854 | |
| 1855 | /* |
| 1856 | * context_switch - switch to the new MM and the new |
| 1857 | * thread's register state. |
| 1858 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1859 | static inline void |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 1860 | context_switch(struct rq *rq, struct task_struct *prev, |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 1861 | struct task_struct *next) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1862 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1863 | struct mm_struct *mm, *oldmm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1864 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 1865 | prepare_task_switch(rq, prev, next); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1866 | mm = next->mm; |
| 1867 | oldmm = prev->active_mm; |
Zachary Amsden | 9226d12 | 2007-02-13 13:26:21 +0100 | [diff] [blame] | 1868 | /* |
| 1869 | * For paravirt, this is coupled with an exit in switch_to to |
| 1870 | * combine the page table reload and the switch backend into |
| 1871 | * one hypercall. |
| 1872 | */ |
| 1873 | arch_enter_lazy_cpu_mode(); |
| 1874 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1875 | if (unlikely(!mm)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1876 | next->active_mm = oldmm; |
| 1877 | atomic_inc(&oldmm->mm_count); |
| 1878 | enter_lazy_tlb(oldmm, next); |
| 1879 | } else |
| 1880 | switch_mm(oldmm, mm, next); |
| 1881 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1882 | if (unlikely(!prev->mm)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1883 | prev->active_mm = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1884 | rq->prev_mm = oldmm; |
| 1885 | } |
Ingo Molnar | 3a5f5e4 | 2006-07-14 00:24:27 -0700 | [diff] [blame] | 1886 | /* |
| 1887 | * Since the runqueue lock will be released by the next |
| 1888 | * task (which is an invalid locking op but in the case |
| 1889 | * of the scheduler it's an obvious special-case), so we |
| 1890 | * do an early lockdep release here: |
| 1891 | */ |
| 1892 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
Ingo Molnar | 8a25d5d | 2006-07-03 00:24:54 -0700 | [diff] [blame] | 1893 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
Ingo Molnar | 3a5f5e4 | 2006-07-14 00:24:27 -0700 | [diff] [blame] | 1894 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1895 | |
| 1896 | /* Here we just switch the register state and the stack. */ |
| 1897 | switch_to(prev, next, prev); |
| 1898 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1899 | barrier(); |
| 1900 | /* |
| 1901 | * this_rq must be evaluated again because prev may have moved |
| 1902 | * CPUs since it called schedule(), thus the 'rq' on its stack |
| 1903 | * frame will be invalid. |
| 1904 | */ |
| 1905 | finish_task_switch(this_rq(), prev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1906 | } |
| 1907 | |
| 1908 | /* |
| 1909 | * nr_running, nr_uninterruptible and nr_context_switches: |
| 1910 | * |
| 1911 | * externally visible scheduler statistics: current number of runnable |
| 1912 | * threads, current number of uninterruptible-sleeping threads, total |
| 1913 | * number of context switches performed since bootup. |
| 1914 | */ |
| 1915 | unsigned long nr_running(void) |
| 1916 | { |
| 1917 | unsigned long i, sum = 0; |
| 1918 | |
| 1919 | for_each_online_cpu(i) |
| 1920 | sum += cpu_rq(i)->nr_running; |
| 1921 | |
| 1922 | return sum; |
| 1923 | } |
| 1924 | |
| 1925 | unsigned long nr_uninterruptible(void) |
| 1926 | { |
| 1927 | unsigned long i, sum = 0; |
| 1928 | |
KAMEZAWA Hiroyuki | 0a94502 | 2006-03-28 01:56:37 -0800 | [diff] [blame] | 1929 | for_each_possible_cpu(i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1930 | sum += cpu_rq(i)->nr_uninterruptible; |
| 1931 | |
| 1932 | /* |
| 1933 | * Since we read the counters lockless, it might be slightly |
| 1934 | * inaccurate. Do not allow it to go below zero though: |
| 1935 | */ |
| 1936 | if (unlikely((long)sum < 0)) |
| 1937 | sum = 0; |
| 1938 | |
| 1939 | return sum; |
| 1940 | } |
| 1941 | |
| 1942 | unsigned long long nr_context_switches(void) |
| 1943 | { |
Steven Rostedt | cc94abf | 2006-06-27 02:54:31 -0700 | [diff] [blame] | 1944 | int i; |
| 1945 | unsigned long long sum = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1946 | |
KAMEZAWA Hiroyuki | 0a94502 | 2006-03-28 01:56:37 -0800 | [diff] [blame] | 1947 | for_each_possible_cpu(i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1948 | sum += cpu_rq(i)->nr_switches; |
| 1949 | |
| 1950 | return sum; |
| 1951 | } |
| 1952 | |
| 1953 | unsigned long nr_iowait(void) |
| 1954 | { |
| 1955 | unsigned long i, sum = 0; |
| 1956 | |
KAMEZAWA Hiroyuki | 0a94502 | 2006-03-28 01:56:37 -0800 | [diff] [blame] | 1957 | for_each_possible_cpu(i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1958 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
| 1959 | |
| 1960 | return sum; |
| 1961 | } |
| 1962 | |
Jack Steiner | db1b1fe | 2006-03-31 02:31:21 -0800 | [diff] [blame] | 1963 | unsigned long nr_active(void) |
| 1964 | { |
| 1965 | unsigned long i, running = 0, uninterruptible = 0; |
| 1966 | |
| 1967 | for_each_online_cpu(i) { |
| 1968 | running += cpu_rq(i)->nr_running; |
| 1969 | uninterruptible += cpu_rq(i)->nr_uninterruptible; |
| 1970 | } |
| 1971 | |
| 1972 | if (unlikely((long)uninterruptible < 0)) |
| 1973 | uninterruptible = 0; |
| 1974 | |
| 1975 | return running + uninterruptible; |
| 1976 | } |
| 1977 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1978 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1979 | * Update rq->cpu_load[] statistics. This function is usually called every |
| 1980 | * scheduler tick (TICK_NSEC). |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 1981 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1982 | static void update_cpu_load(struct rq *this_rq) |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 1983 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1984 | u64 fair_delta64, exec_delta64, idle_delta64, sample_interval64, tmp64; |
| 1985 | unsigned long total_load = this_rq->ls.load.weight; |
| 1986 | unsigned long this_load = total_load; |
| 1987 | struct load_stat *ls = &this_rq->ls; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1988 | int i, scale; |
| 1989 | |
| 1990 | this_rq->nr_load_updates++; |
| 1991 | if (unlikely(!(sysctl_sched_features & SCHED_FEAT_PRECISE_CPU_LOAD))) |
| 1992 | goto do_avg; |
| 1993 | |
| 1994 | /* Update delta_fair/delta_exec fields first */ |
Ingo Molnar | 84a1d7a | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 1995 | update_curr_load(this_rq); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 1996 | |
| 1997 | fair_delta64 = ls->delta_fair + 1; |
| 1998 | ls->delta_fair = 0; |
| 1999 | |
| 2000 | exec_delta64 = ls->delta_exec + 1; |
| 2001 | ls->delta_exec = 0; |
| 2002 | |
Ingo Molnar | d281918 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 2003 | sample_interval64 = this_rq->clock - ls->load_update_last; |
| 2004 | ls->load_update_last = this_rq->clock; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2005 | |
| 2006 | if ((s64)sample_interval64 < (s64)TICK_NSEC) |
| 2007 | sample_interval64 = TICK_NSEC; |
| 2008 | |
| 2009 | if (exec_delta64 > sample_interval64) |
| 2010 | exec_delta64 = sample_interval64; |
| 2011 | |
| 2012 | idle_delta64 = sample_interval64 - exec_delta64; |
| 2013 | |
| 2014 | tmp64 = div64_64(SCHED_LOAD_SCALE * exec_delta64, fair_delta64); |
| 2015 | tmp64 = div64_64(tmp64 * exec_delta64, sample_interval64); |
| 2016 | |
| 2017 | this_load = (unsigned long)tmp64; |
| 2018 | |
| 2019 | do_avg: |
| 2020 | |
| 2021 | /* Update our load: */ |
| 2022 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { |
| 2023 | unsigned long old_load, new_load; |
| 2024 | |
| 2025 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
| 2026 | |
| 2027 | old_load = this_rq->cpu_load[i]; |
| 2028 | new_load = this_load; |
| 2029 | |
| 2030 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
| 2031 | } |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2032 | } |
| 2033 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2034 | #ifdef CONFIG_SMP |
| 2035 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2036 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2037 | * double_rq_lock - safely lock two runqueues |
| 2038 | * |
| 2039 | * Note this does not disable interrupts like task_rq_lock, |
| 2040 | * you need to do so manually before calling. |
| 2041 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2042 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2043 | __acquires(rq1->lock) |
| 2044 | __acquires(rq2->lock) |
| 2045 | { |
Kirill Korotaev | 054b910 | 2006-12-10 02:20:11 -0800 | [diff] [blame] | 2046 | BUG_ON(!irqs_disabled()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2047 | if (rq1 == rq2) { |
| 2048 | spin_lock(&rq1->lock); |
| 2049 | __acquire(rq2->lock); /* Fake it out ;) */ |
| 2050 | } else { |
Chen, Kenneth W | c96d145 | 2006-06-27 02:54:28 -0700 | [diff] [blame] | 2051 | if (rq1 < rq2) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2052 | spin_lock(&rq1->lock); |
| 2053 | spin_lock(&rq2->lock); |
| 2054 | } else { |
| 2055 | spin_lock(&rq2->lock); |
| 2056 | spin_lock(&rq1->lock); |
| 2057 | } |
| 2058 | } |
Ingo Molnar | 6e82a3b | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 2059 | update_rq_clock(rq1); |
| 2060 | update_rq_clock(rq2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2061 | } |
| 2062 | |
| 2063 | /* |
| 2064 | * double_rq_unlock - safely unlock two runqueues |
| 2065 | * |
| 2066 | * Note this does not restore interrupts like task_rq_unlock, |
| 2067 | * you need to do so manually after calling. |
| 2068 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2069 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2070 | __releases(rq1->lock) |
| 2071 | __releases(rq2->lock) |
| 2072 | { |
| 2073 | spin_unlock(&rq1->lock); |
| 2074 | if (rq1 != rq2) |
| 2075 | spin_unlock(&rq2->lock); |
| 2076 | else |
| 2077 | __release(rq2->lock); |
| 2078 | } |
| 2079 | |
| 2080 | /* |
| 2081 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. |
| 2082 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2083 | static void double_lock_balance(struct rq *this_rq, struct rq *busiest) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2084 | __releases(this_rq->lock) |
| 2085 | __acquires(busiest->lock) |
| 2086 | __acquires(this_rq->lock) |
| 2087 | { |
Kirill Korotaev | 054b910 | 2006-12-10 02:20:11 -0800 | [diff] [blame] | 2088 | if (unlikely(!irqs_disabled())) { |
| 2089 | /* printk() doesn't work good under rq->lock */ |
| 2090 | spin_unlock(&this_rq->lock); |
| 2091 | BUG_ON(1); |
| 2092 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2093 | if (unlikely(!spin_trylock(&busiest->lock))) { |
Chen, Kenneth W | c96d145 | 2006-06-27 02:54:28 -0700 | [diff] [blame] | 2094 | if (busiest < this_rq) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2095 | spin_unlock(&this_rq->lock); |
| 2096 | spin_lock(&busiest->lock); |
| 2097 | spin_lock(&this_rq->lock); |
| 2098 | } else |
| 2099 | spin_lock(&busiest->lock); |
| 2100 | } |
| 2101 | } |
| 2102 | |
| 2103 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2104 | * If dest_cpu is allowed for this process, migrate the task to it. |
| 2105 | * This is accomplished by forcing the cpu_allowed mask to only |
| 2106 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
| 2107 | * the cpu_allowed mask is restored. |
| 2108 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 2109 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2110 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2111 | struct migration_req req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2112 | unsigned long flags; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2113 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2114 | |
| 2115 | rq = task_rq_lock(p, &flags); |
| 2116 | if (!cpu_isset(dest_cpu, p->cpus_allowed) |
| 2117 | || unlikely(cpu_is_offline(dest_cpu))) |
| 2118 | goto out; |
| 2119 | |
| 2120 | /* force the process onto the specified CPU */ |
| 2121 | if (migrate_task(p, dest_cpu, &req)) { |
| 2122 | /* Need to wait for migration thread (might exit: take ref). */ |
| 2123 | struct task_struct *mt = rq->migration_thread; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 2124 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2125 | get_task_struct(mt); |
| 2126 | task_rq_unlock(rq, &flags); |
| 2127 | wake_up_process(mt); |
| 2128 | put_task_struct(mt); |
| 2129 | wait_for_completion(&req.done); |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 2130 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2131 | return; |
| 2132 | } |
| 2133 | out: |
| 2134 | task_rq_unlock(rq, &flags); |
| 2135 | } |
| 2136 | |
| 2137 | /* |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 2138 | * sched_exec - execve() is a valuable balancing opportunity, because at |
| 2139 | * this point the task has the smallest effective memory and cache footprint. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2140 | */ |
| 2141 | void sched_exec(void) |
| 2142 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2143 | int new_cpu, this_cpu = get_cpu(); |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 2144 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2145 | put_cpu(); |
Nick Piggin | 476d139 | 2005-06-25 14:57:29 -0700 | [diff] [blame] | 2146 | if (new_cpu != this_cpu) |
| 2147 | sched_migrate_task(current, new_cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2148 | } |
| 2149 | |
| 2150 | /* |
| 2151 | * pull_task - move a task from a remote runqueue to the local runqueue. |
| 2152 | * Both runqueues must be locked. |
| 2153 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2154 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
| 2155 | struct rq *this_rq, int this_cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2156 | { |
Ingo Molnar | 2e1cb74 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 2157 | deactivate_task(src_rq, p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2158 | set_task_cpu(p, this_cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2159 | activate_task(this_rq, p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2160 | /* |
| 2161 | * Note that idle threads have a prio of MAX_PRIO, for this test |
| 2162 | * to be always true for them. |
| 2163 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2164 | check_preempt_curr(this_rq, p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2165 | } |
| 2166 | |
| 2167 | /* |
| 2168 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? |
| 2169 | */ |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 2170 | static |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2171 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2172 | struct sched_domain *sd, enum cpu_idle_type idle, |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 2173 | int *all_pinned) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2174 | { |
| 2175 | /* |
| 2176 | * We do not migrate tasks that are: |
| 2177 | * 1) running (obviously), or |
| 2178 | * 2) cannot be migrated to this CPU due to cpus_allowed, or |
| 2179 | * 3) are cache-hot on their current CPU. |
| 2180 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2181 | if (!cpu_isset(this_cpu, p->cpus_allowed)) |
| 2182 | return 0; |
Nick Piggin | 8102679 | 2005-06-25 14:57:07 -0700 | [diff] [blame] | 2183 | *all_pinned = 0; |
| 2184 | |
| 2185 | if (task_running(rq, p)) |
| 2186 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2187 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2188 | return 1; |
| 2189 | } |
| 2190 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2191 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
| 2192 | unsigned long max_nr_move, unsigned long max_load_move, |
| 2193 | struct sched_domain *sd, enum cpu_idle_type idle, |
| 2194 | int *all_pinned, unsigned long *load_moved, |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2195 | int *this_best_prio, struct rq_iterator *iterator) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2196 | { |
| 2197 | int pulled = 0, pinned = 0, skip_for_load; |
| 2198 | struct task_struct *p; |
| 2199 | long rem_load_move = max_load_move; |
| 2200 | |
| 2201 | if (max_nr_move == 0 || max_load_move == 0) |
| 2202 | goto out; |
| 2203 | |
| 2204 | pinned = 1; |
| 2205 | |
| 2206 | /* |
| 2207 | * Start the load-balancing iterator: |
| 2208 | */ |
| 2209 | p = iterator->start(iterator->arg); |
| 2210 | next: |
| 2211 | if (!p) |
| 2212 | goto out; |
| 2213 | /* |
| 2214 | * To help distribute high priority tasks accross CPUs we don't |
| 2215 | * skip a task if it will be the highest priority task (i.e. smallest |
| 2216 | * prio value) on its new queue regardless of its load weight |
| 2217 | */ |
| 2218 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
| 2219 | SCHED_LOAD_SCALE_FUZZ; |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2220 | if ((skip_for_load && p->prio >= *this_best_prio) || |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2221 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2222 | p = iterator->next(iterator->arg); |
| 2223 | goto next; |
| 2224 | } |
| 2225 | |
| 2226 | pull_task(busiest, p, this_rq, this_cpu); |
| 2227 | pulled++; |
| 2228 | rem_load_move -= p->se.load.weight; |
| 2229 | |
| 2230 | /* |
| 2231 | * We only want to steal up to the prescribed number of tasks |
| 2232 | * and the prescribed amount of weighted load. |
| 2233 | */ |
| 2234 | if (pulled < max_nr_move && rem_load_move > 0) { |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2235 | if (p->prio < *this_best_prio) |
| 2236 | *this_best_prio = p->prio; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2237 | p = iterator->next(iterator->arg); |
| 2238 | goto next; |
| 2239 | } |
| 2240 | out: |
| 2241 | /* |
| 2242 | * Right now, this is the only place pull_task() is called, |
| 2243 | * so we can safely collect pull_task() stats here rather than |
| 2244 | * inside pull_task(). |
| 2245 | */ |
| 2246 | schedstat_add(sd, lb_gained[idle], pulled); |
| 2247 | |
| 2248 | if (all_pinned) |
| 2249 | *all_pinned = pinned; |
| 2250 | *load_moved = max_load_move - rem_load_move; |
| 2251 | return pulled; |
| 2252 | } |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2253 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2254 | /* |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2255 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
| 2256 | * this_rq, as part of a balancing operation within domain "sd". |
| 2257 | * Returns 1 if successful and 0 otherwise. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2258 | * |
| 2259 | * Called with both runqueues locked. |
| 2260 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2261 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2262 | unsigned long max_load_move, |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2263 | struct sched_domain *sd, enum cpu_idle_type idle, |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2264 | int *all_pinned) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2265 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2266 | struct sched_class *class = sched_class_highest; |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2267 | unsigned long total_load_moved = 0; |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2268 | int this_best_prio = this_rq->curr->prio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2269 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2270 | do { |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2271 | total_load_moved += |
| 2272 | class->load_balance(this_rq, this_cpu, busiest, |
| 2273 | ULONG_MAX, max_load_move - total_load_moved, |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2274 | sd, idle, all_pinned, &this_best_prio); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2275 | class = class->next; |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2276 | } while (class && max_load_move > total_load_moved); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2277 | |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2278 | return total_load_moved > 0; |
| 2279 | } |
| 2280 | |
| 2281 | /* |
| 2282 | * move_one_task tries to move exactly one task from busiest to this_rq, as |
| 2283 | * part of active balancing operations within "domain". |
| 2284 | * Returns 1 if successful and 0 otherwise. |
| 2285 | * |
| 2286 | * Called with both runqueues locked. |
| 2287 | */ |
| 2288 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, |
| 2289 | struct sched_domain *sd, enum cpu_idle_type idle) |
| 2290 | { |
| 2291 | struct sched_class *class; |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2292 | int this_best_prio = MAX_PRIO; |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2293 | |
| 2294 | for (class = sched_class_highest; class; class = class->next) |
| 2295 | if (class->load_balance(this_rq, this_cpu, busiest, |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2296 | 1, ULONG_MAX, sd, idle, NULL, |
| 2297 | &this_best_prio)) |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2298 | return 1; |
| 2299 | |
| 2300 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2301 | } |
| 2302 | |
| 2303 | /* |
| 2304 | * find_busiest_group finds and returns the busiest CPU group within the |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2305 | * domain. It calculates and returns the amount of weighted load which |
| 2306 | * should be moved to restore balance via the imbalance parameter. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2307 | */ |
| 2308 | static struct sched_group * |
| 2309 | find_busiest_group(struct sched_domain *sd, int this_cpu, |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2310 | unsigned long *imbalance, enum cpu_idle_type idle, |
| 2311 | int *sd_idle, cpumask_t *cpus, int *balance) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2312 | { |
| 2313 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; |
| 2314 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; |
Siddha, Suresh B | 0c117f1 | 2005-09-10 00:26:21 -0700 | [diff] [blame] | 2315 | unsigned long max_pull; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2316 | unsigned long busiest_load_per_task, busiest_nr_running; |
| 2317 | unsigned long this_load_per_task, this_nr_running; |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 2318 | int load_idx; |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2319 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
| 2320 | int power_savings_balance = 1; |
| 2321 | unsigned long leader_nr_running = 0, min_load_per_task = 0; |
| 2322 | unsigned long min_nr_running = ULONG_MAX; |
| 2323 | struct sched_group *group_min = NULL, *group_leader = NULL; |
| 2324 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2325 | |
| 2326 | max_load = this_load = total_load = total_pwr = 0; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2327 | busiest_load_per_task = busiest_nr_running = 0; |
| 2328 | this_load_per_task = this_nr_running = 0; |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2329 | if (idle == CPU_NOT_IDLE) |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 2330 | load_idx = sd->busy_idx; |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2331 | else if (idle == CPU_NEWLY_IDLE) |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 2332 | load_idx = sd->newidle_idx; |
| 2333 | else |
| 2334 | load_idx = sd->idle_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2335 | |
| 2336 | do { |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2337 | unsigned long load, group_capacity; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2338 | int local_group; |
| 2339 | int i; |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2340 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2341 | unsigned long sum_nr_running, sum_weighted_load; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2342 | |
| 2343 | local_group = cpu_isset(this_cpu, group->cpumask); |
| 2344 | |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2345 | if (local_group) |
| 2346 | balance_cpu = first_cpu(group->cpumask); |
| 2347 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2348 | /* Tally up the load of all CPUs in the group */ |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2349 | sum_weighted_load = sum_nr_running = avg_load = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2350 | |
| 2351 | for_each_cpu_mask(i, group->cpumask) { |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2352 | struct rq *rq; |
| 2353 | |
| 2354 | if (!cpu_isset(i, *cpus)) |
| 2355 | continue; |
| 2356 | |
| 2357 | rq = cpu_rq(i); |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2358 | |
Suresh Siddha | 9439aab | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 2359 | if (*sd_idle && rq->nr_running) |
Nick Piggin | 5969fe0 | 2005-09-10 00:26:19 -0700 | [diff] [blame] | 2360 | *sd_idle = 0; |
| 2361 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2362 | /* Bias balancing toward cpus of our domain */ |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2363 | if (local_group) { |
| 2364 | if (idle_cpu(i) && !first_idle_cpu) { |
| 2365 | first_idle_cpu = 1; |
| 2366 | balance_cpu = i; |
| 2367 | } |
| 2368 | |
Nick Piggin | a200057 | 2006-02-10 01:51:02 -0800 | [diff] [blame] | 2369 | load = target_load(i, load_idx); |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2370 | } else |
Nick Piggin | a200057 | 2006-02-10 01:51:02 -0800 | [diff] [blame] | 2371 | load = source_load(i, load_idx); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2372 | |
| 2373 | avg_load += load; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2374 | sum_nr_running += rq->nr_running; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2375 | sum_weighted_load += weighted_cpuload(i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2376 | } |
| 2377 | |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2378 | /* |
| 2379 | * First idle cpu or the first cpu(busiest) in this sched group |
| 2380 | * is eligible for doing load balancing at this and above |
Suresh Siddha | 9439aab | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 2381 | * domains. In the newly idle case, we will allow all the cpu's |
| 2382 | * to do the newly idle load balance. |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2383 | */ |
Suresh Siddha | 9439aab | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 2384 | if (idle != CPU_NEWLY_IDLE && local_group && |
| 2385 | balance_cpu != this_cpu && balance) { |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2386 | *balance = 0; |
| 2387 | goto ret; |
| 2388 | } |
| 2389 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2390 | total_load += avg_load; |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 2391 | total_pwr += group->__cpu_power; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2392 | |
| 2393 | /* Adjust by relative CPU power of the group */ |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 2394 | avg_load = sg_div_cpu_power(group, |
| 2395 | avg_load * SCHED_LOAD_SCALE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2396 | |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 2397 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2398 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2399 | if (local_group) { |
| 2400 | this_load = avg_load; |
| 2401 | this = group; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2402 | this_nr_running = sum_nr_running; |
| 2403 | this_load_per_task = sum_weighted_load; |
| 2404 | } else if (avg_load > max_load && |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2405 | sum_nr_running > group_capacity) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2406 | max_load = avg_load; |
| 2407 | busiest = group; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2408 | busiest_nr_running = sum_nr_running; |
| 2409 | busiest_load_per_task = sum_weighted_load; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2410 | } |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2411 | |
| 2412 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
| 2413 | /* |
| 2414 | * Busy processors will not participate in power savings |
| 2415 | * balance. |
| 2416 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2417 | if (idle == CPU_NOT_IDLE || |
| 2418 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
| 2419 | goto group_next; |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2420 | |
| 2421 | /* |
| 2422 | * If the local group is idle or completely loaded |
| 2423 | * no need to do power savings balance at this domain |
| 2424 | */ |
| 2425 | if (local_group && (this_nr_running >= group_capacity || |
| 2426 | !this_nr_running)) |
| 2427 | power_savings_balance = 0; |
| 2428 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2429 | /* |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2430 | * If a group is already running at full capacity or idle, |
| 2431 | * don't include that group in power savings calculations |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2432 | */ |
| 2433 | if (!power_savings_balance || sum_nr_running >= group_capacity |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2434 | || !sum_nr_running) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2435 | goto group_next; |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2436 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2437 | /* |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2438 | * Calculate the group which has the least non-idle load. |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2439 | * This is the group from where we need to pick up the load |
| 2440 | * for saving power |
| 2441 | */ |
| 2442 | if ((sum_nr_running < min_nr_running) || |
| 2443 | (sum_nr_running == min_nr_running && |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2444 | first_cpu(group->cpumask) < |
| 2445 | first_cpu(group_min->cpumask))) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2446 | group_min = group; |
| 2447 | min_nr_running = sum_nr_running; |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2448 | min_load_per_task = sum_weighted_load / |
| 2449 | sum_nr_running; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2450 | } |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2451 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2452 | /* |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2453 | * Calculate the group which is almost near its |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2454 | * capacity but still has some space to pick up some load |
| 2455 | * from other group and save more power |
| 2456 | */ |
| 2457 | if (sum_nr_running <= group_capacity - 1) { |
| 2458 | if (sum_nr_running > leader_nr_running || |
| 2459 | (sum_nr_running == leader_nr_running && |
| 2460 | first_cpu(group->cpumask) > |
| 2461 | first_cpu(group_leader->cpumask))) { |
| 2462 | group_leader = group; |
| 2463 | leader_nr_running = sum_nr_running; |
| 2464 | } |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2465 | } |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2466 | group_next: |
| 2467 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2468 | group = group->next; |
| 2469 | } while (group != sd->groups); |
| 2470 | |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2471 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2472 | goto out_balanced; |
| 2473 | |
| 2474 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; |
| 2475 | |
| 2476 | if (this_load >= avg_load || |
| 2477 | 100*max_load <= sd->imbalance_pct*this_load) |
| 2478 | goto out_balanced; |
| 2479 | |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2480 | busiest_load_per_task /= busiest_nr_running; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2481 | /* |
| 2482 | * We're trying to get all the cpus to the average_load, so we don't |
| 2483 | * want to push ourselves above the average load, nor do we wish to |
| 2484 | * reduce the max loaded cpu below the average load, as either of these |
| 2485 | * actions would just result in more rebalancing later, and ping-pong |
| 2486 | * tasks around. Thus we look for the minimum possible imbalance. |
| 2487 | * Negative imbalances (*we* are more loaded than anyone else) will |
| 2488 | * be counted as no imbalance for these purposes -- we can't fix that |
| 2489 | * by pulling tasks to us. Be careful of negative numbers as they'll |
| 2490 | * appear as very large values with unsigned longs. |
| 2491 | */ |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2492 | if (max_load <= busiest_load_per_task) |
| 2493 | goto out_balanced; |
| 2494 | |
| 2495 | /* |
| 2496 | * In the presence of smp nice balancing, certain scenarios can have |
| 2497 | * max load less than avg load(as we skip the groups at or below |
| 2498 | * its cpu_power, while calculating max_load..) |
| 2499 | */ |
| 2500 | if (max_load < avg_load) { |
| 2501 | *imbalance = 0; |
| 2502 | goto small_imbalance; |
| 2503 | } |
Siddha, Suresh B | 0c117f1 | 2005-09-10 00:26:21 -0700 | [diff] [blame] | 2504 | |
| 2505 | /* Don't want to pull so many tasks that a group would go idle */ |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2506 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
Siddha, Suresh B | 0c117f1 | 2005-09-10 00:26:21 -0700 | [diff] [blame] | 2507 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2508 | /* How much load to actually move to equalise the imbalance */ |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 2509 | *imbalance = min(max_pull * busiest->__cpu_power, |
| 2510 | (avg_load - this_load) * this->__cpu_power) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2511 | / SCHED_LOAD_SCALE; |
| 2512 | |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2513 | /* |
| 2514 | * if *imbalance is less than the average load per runnable task |
| 2515 | * there is no gaurantee that any tasks will be moved so we'll have |
| 2516 | * a think about bumping its value to force at least one task to be |
| 2517 | * moved |
| 2518 | */ |
Suresh Siddha | 7fd0d2d | 2007-09-05 14:32:48 +0200 | [diff] [blame] | 2519 | if (*imbalance < busiest_load_per_task) { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2520 | unsigned long tmp, pwr_now, pwr_move; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2521 | unsigned int imbn; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2522 | |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2523 | small_imbalance: |
| 2524 | pwr_move = pwr_now = 0; |
| 2525 | imbn = 2; |
| 2526 | if (this_nr_running) { |
| 2527 | this_load_per_task /= this_nr_running; |
| 2528 | if (busiest_load_per_task > this_load_per_task) |
| 2529 | imbn = 1; |
| 2530 | } else |
| 2531 | this_load_per_task = SCHED_LOAD_SCALE; |
| 2532 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2533 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
| 2534 | busiest_load_per_task * imbn) { |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2535 | *imbalance = busiest_load_per_task; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2536 | return busiest; |
| 2537 | } |
| 2538 | |
| 2539 | /* |
| 2540 | * OK, we don't have enough imbalance to justify moving tasks, |
| 2541 | * however we may be able to increase total CPU power used by |
| 2542 | * moving them. |
| 2543 | */ |
| 2544 | |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 2545 | pwr_now += busiest->__cpu_power * |
| 2546 | min(busiest_load_per_task, max_load); |
| 2547 | pwr_now += this->__cpu_power * |
| 2548 | min(this_load_per_task, this_load); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2549 | pwr_now /= SCHED_LOAD_SCALE; |
| 2550 | |
| 2551 | /* Amount of load we'd subtract */ |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 2552 | tmp = sg_div_cpu_power(busiest, |
| 2553 | busiest_load_per_task * SCHED_LOAD_SCALE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2554 | if (max_load > tmp) |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 2555 | pwr_move += busiest->__cpu_power * |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2556 | min(busiest_load_per_task, max_load - tmp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2557 | |
| 2558 | /* Amount of load we'd add */ |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 2559 | if (max_load * busiest->__cpu_power < |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 2560 | busiest_load_per_task * SCHED_LOAD_SCALE) |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 2561 | tmp = sg_div_cpu_power(this, |
| 2562 | max_load * busiest->__cpu_power); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2563 | else |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 2564 | tmp = sg_div_cpu_power(this, |
| 2565 | busiest_load_per_task * SCHED_LOAD_SCALE); |
| 2566 | pwr_move += this->__cpu_power * |
| 2567 | min(this_load_per_task, this_load + tmp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2568 | pwr_move /= SCHED_LOAD_SCALE; |
| 2569 | |
| 2570 | /* Move if we gain throughput */ |
Suresh Siddha | 7fd0d2d | 2007-09-05 14:32:48 +0200 | [diff] [blame] | 2571 | if (pwr_move > pwr_now) |
| 2572 | *imbalance = busiest_load_per_task; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2573 | } |
| 2574 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2575 | return busiest; |
| 2576 | |
| 2577 | out_balanced: |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2578 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2579 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2580 | goto ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2581 | |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2582 | if (this == group_leader && group_leader != group_min) { |
| 2583 | *imbalance = min_load_per_task; |
| 2584 | return group_min; |
| 2585 | } |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 2586 | #endif |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2587 | ret: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2588 | *imbalance = 0; |
| 2589 | return NULL; |
| 2590 | } |
| 2591 | |
| 2592 | /* |
| 2593 | * find_busiest_queue - find the busiest runqueue among the cpus in group. |
| 2594 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2595 | static struct rq * |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2596 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2597 | unsigned long imbalance, cpumask_t *cpus) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2598 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2599 | struct rq *busiest = NULL, *rq; |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2600 | unsigned long max_load = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2601 | int i; |
| 2602 | |
| 2603 | for_each_cpu_mask(i, group->cpumask) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2604 | unsigned long wl; |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2605 | |
| 2606 | if (!cpu_isset(i, *cpus)) |
| 2607 | continue; |
| 2608 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2609 | rq = cpu_rq(i); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2610 | wl = weighted_cpuload(i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2611 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2612 | if (rq->nr_running == 1 && wl > imbalance) |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 2613 | continue; |
| 2614 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2615 | if (wl > max_load) { |
| 2616 | max_load = wl; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2617 | busiest = rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2618 | } |
| 2619 | } |
| 2620 | |
| 2621 | return busiest; |
| 2622 | } |
| 2623 | |
| 2624 | /* |
Nick Piggin | 77391d7 | 2005-06-25 14:57:30 -0700 | [diff] [blame] | 2625 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but |
| 2626 | * so long as it is large enough. |
| 2627 | */ |
| 2628 | #define MAX_PINNED_INTERVAL 512 |
| 2629 | |
| 2630 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2631 | * Check this_cpu to ensure it is balanced within domain. Attempt to move |
| 2632 | * tasks if there is an imbalance. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2633 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2634 | static int load_balance(int this_cpu, struct rq *this_rq, |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2635 | struct sched_domain *sd, enum cpu_idle_type idle, |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2636 | int *balance) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2637 | { |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2638 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2639 | struct sched_group *group; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2640 | unsigned long imbalance; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2641 | struct rq *busiest; |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2642 | cpumask_t cpus = CPU_MASK_ALL; |
Christoph Lameter | fe2eea3 | 2006-12-10 02:20:21 -0800 | [diff] [blame] | 2643 | unsigned long flags; |
Nick Piggin | 5969fe0 | 2005-09-10 00:26:19 -0700 | [diff] [blame] | 2644 | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 2645 | /* |
| 2646 | * When power savings policy is enabled for the parent domain, idle |
| 2647 | * sibling can pick up load irrespective of busy siblings. In this case, |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2648 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2649 | * portraying it as CPU_NOT_IDLE. |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 2650 | */ |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2651 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 2652 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
Nick Piggin | 5969fe0 | 2005-09-10 00:26:19 -0700 | [diff] [blame] | 2653 | sd_idle = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2654 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2655 | schedstat_inc(sd, lb_cnt[idle]); |
| 2656 | |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2657 | redo: |
| 2658 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2659 | &cpus, balance); |
| 2660 | |
Chen, Kenneth W | 0606671 | 2006-12-10 02:20:35 -0800 | [diff] [blame] | 2661 | if (*balance == 0) |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2662 | goto out_balanced; |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2663 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2664 | if (!group) { |
| 2665 | schedstat_inc(sd, lb_nobusyg[idle]); |
| 2666 | goto out_balanced; |
| 2667 | } |
| 2668 | |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2669 | busiest = find_busiest_queue(group, idle, imbalance, &cpus); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2670 | if (!busiest) { |
| 2671 | schedstat_inc(sd, lb_nobusyq[idle]); |
| 2672 | goto out_balanced; |
| 2673 | } |
| 2674 | |
Nick Piggin | db935db | 2005-06-25 14:57:11 -0700 | [diff] [blame] | 2675 | BUG_ON(busiest == this_rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2676 | |
| 2677 | schedstat_add(sd, lb_imbalance[idle], imbalance); |
| 2678 | |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2679 | ld_moved = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2680 | if (busiest->nr_running > 1) { |
| 2681 | /* |
| 2682 | * Attempt to move tasks. If find_busiest_group has found |
| 2683 | * an imbalance but busiest->nr_running <= 1, the group is |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2684 | * still unbalanced. ld_moved simply stays zero, so it is |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2685 | * correctly treated as an imbalance. |
| 2686 | */ |
Christoph Lameter | fe2eea3 | 2006-12-10 02:20:21 -0800 | [diff] [blame] | 2687 | local_irq_save(flags); |
Nick Piggin | e17224b | 2005-09-10 00:26:18 -0700 | [diff] [blame] | 2688 | double_rq_lock(this_rq, busiest); |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2689 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2690 | imbalance, sd, idle, &all_pinned); |
Nick Piggin | e17224b | 2005-09-10 00:26:18 -0700 | [diff] [blame] | 2691 | double_rq_unlock(this_rq, busiest); |
Christoph Lameter | fe2eea3 | 2006-12-10 02:20:21 -0800 | [diff] [blame] | 2692 | local_irq_restore(flags); |
Nick Piggin | 8102679 | 2005-06-25 14:57:07 -0700 | [diff] [blame] | 2693 | |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 2694 | /* |
| 2695 | * some other cpu did the load balance for us. |
| 2696 | */ |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2697 | if (ld_moved && this_cpu != smp_processor_id()) |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 2698 | resched_cpu(this_cpu); |
| 2699 | |
Nick Piggin | 8102679 | 2005-06-25 14:57:07 -0700 | [diff] [blame] | 2700 | /* All tasks on this runqueue were pinned by CPU affinity */ |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2701 | if (unlikely(all_pinned)) { |
| 2702 | cpu_clear(cpu_of(busiest), cpus); |
| 2703 | if (!cpus_empty(cpus)) |
| 2704 | goto redo; |
Nick Piggin | 8102679 | 2005-06-25 14:57:07 -0700 | [diff] [blame] | 2705 | goto out_balanced; |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2706 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2707 | } |
Nick Piggin | 8102679 | 2005-06-25 14:57:07 -0700 | [diff] [blame] | 2708 | |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2709 | if (!ld_moved) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2710 | schedstat_inc(sd, lb_failed[idle]); |
| 2711 | sd->nr_balance_failed++; |
| 2712 | |
| 2713 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2714 | |
Christoph Lameter | fe2eea3 | 2006-12-10 02:20:21 -0800 | [diff] [blame] | 2715 | spin_lock_irqsave(&busiest->lock, flags); |
Siddha, Suresh B | fa3b6dd | 2005-09-10 00:26:21 -0700 | [diff] [blame] | 2716 | |
| 2717 | /* don't kick the migration_thread, if the curr |
| 2718 | * task on busiest cpu can't be moved to this_cpu |
| 2719 | */ |
| 2720 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { |
Christoph Lameter | fe2eea3 | 2006-12-10 02:20:21 -0800 | [diff] [blame] | 2721 | spin_unlock_irqrestore(&busiest->lock, flags); |
Siddha, Suresh B | fa3b6dd | 2005-09-10 00:26:21 -0700 | [diff] [blame] | 2722 | all_pinned = 1; |
| 2723 | goto out_one_pinned; |
| 2724 | } |
| 2725 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2726 | if (!busiest->active_balance) { |
| 2727 | busiest->active_balance = 1; |
| 2728 | busiest->push_cpu = this_cpu; |
Nick Piggin | 8102679 | 2005-06-25 14:57:07 -0700 | [diff] [blame] | 2729 | active_balance = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2730 | } |
Christoph Lameter | fe2eea3 | 2006-12-10 02:20:21 -0800 | [diff] [blame] | 2731 | spin_unlock_irqrestore(&busiest->lock, flags); |
Nick Piggin | 8102679 | 2005-06-25 14:57:07 -0700 | [diff] [blame] | 2732 | if (active_balance) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2733 | wake_up_process(busiest->migration_thread); |
| 2734 | |
| 2735 | /* |
| 2736 | * We've kicked active balancing, reset the failure |
| 2737 | * counter. |
| 2738 | */ |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2739 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2740 | } |
Nick Piggin | 8102679 | 2005-06-25 14:57:07 -0700 | [diff] [blame] | 2741 | } else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2742 | sd->nr_balance_failed = 0; |
| 2743 | |
Nick Piggin | 8102679 | 2005-06-25 14:57:07 -0700 | [diff] [blame] | 2744 | if (likely(!active_balance)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2745 | /* We were unbalanced, so reset the balancing interval */ |
| 2746 | sd->balance_interval = sd->min_interval; |
Nick Piggin | 8102679 | 2005-06-25 14:57:07 -0700 | [diff] [blame] | 2747 | } else { |
| 2748 | /* |
| 2749 | * If we've begun active balancing, start to back off. This |
| 2750 | * case may not be covered by the all_pinned logic if there |
| 2751 | * is only 1 task on the busy runqueue (because we don't call |
| 2752 | * move_tasks). |
| 2753 | */ |
| 2754 | if (sd->balance_interval < sd->max_interval) |
| 2755 | sd->balance_interval *= 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2756 | } |
| 2757 | |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2758 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 2759 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
Nick Piggin | 5969fe0 | 2005-09-10 00:26:19 -0700 | [diff] [blame] | 2760 | return -1; |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2761 | return ld_moved; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2762 | |
| 2763 | out_balanced: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2764 | schedstat_inc(sd, lb_balanced[idle]); |
| 2765 | |
Nick Piggin | 16cfb1c | 2005-06-25 14:57:08 -0700 | [diff] [blame] | 2766 | sd->nr_balance_failed = 0; |
Siddha, Suresh B | fa3b6dd | 2005-09-10 00:26:21 -0700 | [diff] [blame] | 2767 | |
| 2768 | out_one_pinned: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2769 | /* tune up the balancing interval */ |
Nick Piggin | 77391d7 | 2005-06-25 14:57:30 -0700 | [diff] [blame] | 2770 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
| 2771 | (sd->balance_interval < sd->max_interval)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2772 | sd->balance_interval *= 2; |
| 2773 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2774 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 2775 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
Nick Piggin | 5969fe0 | 2005-09-10 00:26:19 -0700 | [diff] [blame] | 2776 | return -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2777 | return 0; |
| 2778 | } |
| 2779 | |
| 2780 | /* |
| 2781 | * Check this_cpu to ensure it is balanced within domain. Attempt to move |
| 2782 | * tasks if there is an imbalance. |
| 2783 | * |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2784 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2785 | * this_rq is locked. |
| 2786 | */ |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2787 | static int |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2788 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2789 | { |
| 2790 | struct sched_group *group; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2791 | struct rq *busiest = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2792 | unsigned long imbalance; |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2793 | int ld_moved = 0; |
Nick Piggin | 5969fe0 | 2005-09-10 00:26:19 -0700 | [diff] [blame] | 2794 | int sd_idle = 0; |
Suresh Siddha | 969bb4e | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 2795 | int all_pinned = 0; |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2796 | cpumask_t cpus = CPU_MASK_ALL; |
Nick Piggin | 5969fe0 | 2005-09-10 00:26:19 -0700 | [diff] [blame] | 2797 | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 2798 | /* |
| 2799 | * When power savings policy is enabled for the parent domain, idle |
| 2800 | * sibling can pick up load irrespective of busy siblings. In this case, |
| 2801 | * let the state of idle sibling percolate up as IDLE, instead of |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2802 | * portraying it as CPU_NOT_IDLE. |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 2803 | */ |
| 2804 | if (sd->flags & SD_SHARE_CPUPOWER && |
| 2805 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
Nick Piggin | 5969fe0 | 2005-09-10 00:26:19 -0700 | [diff] [blame] | 2806 | sd_idle = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2807 | |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2808 | schedstat_inc(sd, lb_cnt[CPU_NEWLY_IDLE]); |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2809 | redo: |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2810 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 2811 | &sd_idle, &cpus, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2812 | if (!group) { |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2813 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
Nick Piggin | 16cfb1c | 2005-06-25 14:57:08 -0700 | [diff] [blame] | 2814 | goto out_balanced; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2815 | } |
| 2816 | |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2817 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2818 | &cpus); |
Nick Piggin | db935db | 2005-06-25 14:57:11 -0700 | [diff] [blame] | 2819 | if (!busiest) { |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2820 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
Nick Piggin | 16cfb1c | 2005-06-25 14:57:08 -0700 | [diff] [blame] | 2821 | goto out_balanced; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2822 | } |
| 2823 | |
Nick Piggin | db935db | 2005-06-25 14:57:11 -0700 | [diff] [blame] | 2824 | BUG_ON(busiest == this_rq); |
| 2825 | |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2826 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
Nick Piggin | d6d5cfa | 2005-09-10 00:26:16 -0700 | [diff] [blame] | 2827 | |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2828 | ld_moved = 0; |
Nick Piggin | d6d5cfa | 2005-09-10 00:26:16 -0700 | [diff] [blame] | 2829 | if (busiest->nr_running > 1) { |
| 2830 | /* Attempt to move tasks */ |
| 2831 | double_lock_balance(this_rq, busiest); |
Ingo Molnar | 6e82a3b | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 2832 | /* this_rq->clock is already updated */ |
| 2833 | update_rq_clock(busiest); |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2834 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
Suresh Siddha | 969bb4e | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 2835 | imbalance, sd, CPU_NEWLY_IDLE, |
| 2836 | &all_pinned); |
Nick Piggin | d6d5cfa | 2005-09-10 00:26:16 -0700 | [diff] [blame] | 2837 | spin_unlock(&busiest->lock); |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2838 | |
Suresh Siddha | 969bb4e | 2007-07-19 21:28:35 +0200 | [diff] [blame] | 2839 | if (unlikely(all_pinned)) { |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 2840 | cpu_clear(cpu_of(busiest), cpus); |
| 2841 | if (!cpus_empty(cpus)) |
| 2842 | goto redo; |
| 2843 | } |
Nick Piggin | d6d5cfa | 2005-09-10 00:26:16 -0700 | [diff] [blame] | 2844 | } |
| 2845 | |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2846 | if (!ld_moved) { |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2847 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 2848 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
| 2849 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
Nick Piggin | 5969fe0 | 2005-09-10 00:26:19 -0700 | [diff] [blame] | 2850 | return -1; |
| 2851 | } else |
Nick Piggin | 16cfb1c | 2005-06-25 14:57:08 -0700 | [diff] [blame] | 2852 | sd->nr_balance_failed = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2853 | |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2854 | return ld_moved; |
Nick Piggin | 16cfb1c | 2005-06-25 14:57:08 -0700 | [diff] [blame] | 2855 | |
| 2856 | out_balanced: |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 2857 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2858 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 2859 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
Nick Piggin | 5969fe0 | 2005-09-10 00:26:19 -0700 | [diff] [blame] | 2860 | return -1; |
Nick Piggin | 16cfb1c | 2005-06-25 14:57:08 -0700 | [diff] [blame] | 2861 | sd->nr_balance_failed = 0; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2862 | |
Nick Piggin | 16cfb1c | 2005-06-25 14:57:08 -0700 | [diff] [blame] | 2863 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2864 | } |
| 2865 | |
| 2866 | /* |
| 2867 | * idle_balance is called by schedule() if this_cpu is about to become |
| 2868 | * idle. Attempts to pull tasks from other CPUs. |
| 2869 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2870 | static void idle_balance(int this_cpu, struct rq *this_rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2871 | { |
| 2872 | struct sched_domain *sd; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2873 | int pulled_task = -1; |
| 2874 | unsigned long next_balance = jiffies + HZ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2875 | |
| 2876 | for_each_domain(this_cpu, sd) { |
Christoph Lameter | 92c4ca5 | 2007-06-23 17:16:33 -0700 | [diff] [blame] | 2877 | unsigned long interval; |
| 2878 | |
| 2879 | if (!(sd->flags & SD_LOAD_BALANCE)) |
| 2880 | continue; |
| 2881 | |
| 2882 | if (sd->flags & SD_BALANCE_NEWIDLE) |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2883 | /* If we've pulled tasks over stop searching: */ |
Christoph Lameter | 1bd77f2 | 2006-12-10 02:20:27 -0800 | [diff] [blame] | 2884 | pulled_task = load_balance_newidle(this_cpu, |
Christoph Lameter | 92c4ca5 | 2007-06-23 17:16:33 -0700 | [diff] [blame] | 2885 | this_rq, sd); |
| 2886 | |
| 2887 | interval = msecs_to_jiffies(sd->balance_interval); |
| 2888 | if (time_after(next_balance, sd->last_balance + interval)) |
| 2889 | next_balance = sd->last_balance + interval; |
| 2890 | if (pulled_task) |
| 2891 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2892 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2893 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
Christoph Lameter | 1bd77f2 | 2006-12-10 02:20:27 -0800 | [diff] [blame] | 2894 | /* |
| 2895 | * We are going idle. next_balance may be set based on |
| 2896 | * a busy processor. So reset next_balance. |
| 2897 | */ |
| 2898 | this_rq->next_balance = next_balance; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 2899 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2900 | } |
| 2901 | |
| 2902 | /* |
| 2903 | * active_load_balance is run by migration threads. It pushes running tasks |
| 2904 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be |
| 2905 | * running on each physical CPU where possible, and avoids physical / |
| 2906 | * logical imbalances. |
| 2907 | * |
| 2908 | * Called with busiest_rq locked. |
| 2909 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2910 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2911 | { |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2912 | int target_cpu = busiest_rq->push_cpu; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 2913 | struct sched_domain *sd; |
| 2914 | struct rq *target_rq; |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2915 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2916 | /* Is there any task to move? */ |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2917 | if (busiest_rq->nr_running <= 1) |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2918 | return; |
| 2919 | |
| 2920 | target_rq = cpu_rq(target_cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2921 | |
| 2922 | /* |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2923 | * This condition is "impossible", if it occurs |
| 2924 | * we need to fix it. Originally reported by |
| 2925 | * Bjorn Helgaas on a 128-cpu setup. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2926 | */ |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2927 | BUG_ON(busiest_rq == target_rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2928 | |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2929 | /* move a task from busiest_rq to target_rq */ |
| 2930 | double_lock_balance(busiest_rq, target_rq); |
Ingo Molnar | 6e82a3b | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 2931 | update_rq_clock(busiest_rq); |
| 2932 | update_rq_clock(target_rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2933 | |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2934 | /* Search for an sd spanning us and the target CPU. */ |
Chen, Kenneth W | c96d145 | 2006-06-27 02:54:28 -0700 | [diff] [blame] | 2935 | for_each_domain(target_cpu, sd) { |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2936 | if ((sd->flags & SD_LOAD_BALANCE) && |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2937 | cpu_isset(busiest_cpu, sd->span)) |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2938 | break; |
Chen, Kenneth W | c96d145 | 2006-06-27 02:54:28 -0700 | [diff] [blame] | 2939 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2940 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2941 | if (likely(sd)) { |
| 2942 | schedstat_inc(sd, alb_cnt); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2943 | |
Peter Williams | 4301065 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 2944 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
| 2945 | sd, CPU_IDLE)) |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 2946 | schedstat_inc(sd, alb_pushed); |
| 2947 | else |
| 2948 | schedstat_inc(sd, alb_failed); |
| 2949 | } |
Nick Piggin | 3950745 | 2005-06-25 14:57:09 -0700 | [diff] [blame] | 2950 | spin_unlock(&target_rq->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2951 | } |
| 2952 | |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 2953 | #ifdef CONFIG_NO_HZ |
| 2954 | static struct { |
| 2955 | atomic_t load_balancer; |
| 2956 | cpumask_t cpu_mask; |
| 2957 | } nohz ____cacheline_aligned = { |
| 2958 | .load_balancer = ATOMIC_INIT(-1), |
| 2959 | .cpu_mask = CPU_MASK_NONE, |
| 2960 | }; |
| 2961 | |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 2962 | /* |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 2963 | * This routine will try to nominate the ilb (idle load balancing) |
| 2964 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle |
| 2965 | * load balancing on behalf of all those cpus. If all the cpus in the system |
| 2966 | * go into this tickless mode, then there will be no ilb owner (as there is |
| 2967 | * no need for one) and all the cpus will sleep till the next wakeup event |
| 2968 | * arrives... |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 2969 | * |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 2970 | * For the ilb owner, tick is not stopped. And this tick will be used |
| 2971 | * for idle load balancing. ilb owner will still be part of |
| 2972 | * nohz.cpu_mask.. |
| 2973 | * |
| 2974 | * While stopping the tick, this cpu will become the ilb owner if there |
| 2975 | * is no other owner. And will be the owner till that cpu becomes busy |
| 2976 | * or if all cpus in the system stop their ticks at which point |
| 2977 | * there is no need for ilb owner. |
| 2978 | * |
| 2979 | * When the ilb owner becomes busy, it nominates another owner, during the |
| 2980 | * next busy scheduler_tick() |
| 2981 | */ |
| 2982 | int select_nohz_load_balancer(int stop_tick) |
| 2983 | { |
| 2984 | int cpu = smp_processor_id(); |
| 2985 | |
| 2986 | if (stop_tick) { |
| 2987 | cpu_set(cpu, nohz.cpu_mask); |
| 2988 | cpu_rq(cpu)->in_nohz_recently = 1; |
| 2989 | |
| 2990 | /* |
| 2991 | * If we are going offline and still the leader, give up! |
| 2992 | */ |
| 2993 | if (cpu_is_offline(cpu) && |
| 2994 | atomic_read(&nohz.load_balancer) == cpu) { |
| 2995 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
| 2996 | BUG(); |
| 2997 | return 0; |
| 2998 | } |
| 2999 | |
| 3000 | /* time for ilb owner also to sleep */ |
| 3001 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { |
| 3002 | if (atomic_read(&nohz.load_balancer) == cpu) |
| 3003 | atomic_set(&nohz.load_balancer, -1); |
| 3004 | return 0; |
| 3005 | } |
| 3006 | |
| 3007 | if (atomic_read(&nohz.load_balancer) == -1) { |
| 3008 | /* make me the ilb owner */ |
| 3009 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) |
| 3010 | return 1; |
| 3011 | } else if (atomic_read(&nohz.load_balancer) == cpu) |
| 3012 | return 1; |
| 3013 | } else { |
| 3014 | if (!cpu_isset(cpu, nohz.cpu_mask)) |
| 3015 | return 0; |
| 3016 | |
| 3017 | cpu_clear(cpu, nohz.cpu_mask); |
| 3018 | |
| 3019 | if (atomic_read(&nohz.load_balancer) == cpu) |
| 3020 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
| 3021 | BUG(); |
| 3022 | } |
| 3023 | return 0; |
| 3024 | } |
| 3025 | #endif |
| 3026 | |
| 3027 | static DEFINE_SPINLOCK(balancing); |
| 3028 | |
| 3029 | /* |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 3030 | * It checks each scheduling domain to see if it is due to be balanced, |
| 3031 | * and initiates a balancing operation if so. |
| 3032 | * |
| 3033 | * Balancing parameters are set up in arch_init_sched_domains. |
| 3034 | */ |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 3035 | static inline void rebalance_domains(int cpu, enum cpu_idle_type idle) |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 3036 | { |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3037 | int balance = 1; |
| 3038 | struct rq *rq = cpu_rq(cpu); |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 3039 | unsigned long interval; |
| 3040 | struct sched_domain *sd; |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3041 | /* Earliest time when we have to do rebalance again */ |
Christoph Lameter | c9819f4 | 2006-12-10 02:20:25 -0800 | [diff] [blame] | 3042 | unsigned long next_balance = jiffies + 60*HZ; |
Suresh Siddha | f549da8 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 3043 | int update_next_balance = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3044 | |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3045 | for_each_domain(cpu, sd) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3046 | if (!(sd->flags & SD_LOAD_BALANCE)) |
| 3047 | continue; |
| 3048 | |
| 3049 | interval = sd->balance_interval; |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 3050 | if (idle != CPU_IDLE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3051 | interval *= sd->busy_factor; |
| 3052 | |
| 3053 | /* scale ms to jiffies */ |
| 3054 | interval = msecs_to_jiffies(interval); |
| 3055 | if (unlikely(!interval)) |
| 3056 | interval = 1; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3057 | if (interval > HZ*NR_CPUS/10) |
| 3058 | interval = HZ*NR_CPUS/10; |
| 3059 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3060 | |
Christoph Lameter | 08c183f | 2006-12-10 02:20:29 -0800 | [diff] [blame] | 3061 | if (sd->flags & SD_SERIALIZE) { |
| 3062 | if (!spin_trylock(&balancing)) |
| 3063 | goto out; |
| 3064 | } |
| 3065 | |
Christoph Lameter | c9819f4 | 2006-12-10 02:20:25 -0800 | [diff] [blame] | 3066 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3067 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
Siddha, Suresh B | fa3b6dd | 2005-09-10 00:26:21 -0700 | [diff] [blame] | 3068 | /* |
| 3069 | * We've pulled tasks over so either we're no |
Nick Piggin | 5969fe0 | 2005-09-10 00:26:19 -0700 | [diff] [blame] | 3070 | * longer idle, or one of our SMT siblings is |
| 3071 | * not idle. |
| 3072 | */ |
Ingo Molnar | d15bcfd | 2007-07-09 18:51:57 +0200 | [diff] [blame] | 3073 | idle = CPU_NOT_IDLE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3074 | } |
Christoph Lameter | 1bd77f2 | 2006-12-10 02:20:27 -0800 | [diff] [blame] | 3075 | sd->last_balance = jiffies; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3076 | } |
Christoph Lameter | 08c183f | 2006-12-10 02:20:29 -0800 | [diff] [blame] | 3077 | if (sd->flags & SD_SERIALIZE) |
| 3078 | spin_unlock(&balancing); |
| 3079 | out: |
Suresh Siddha | f549da8 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 3080 | if (time_after(next_balance, sd->last_balance + interval)) { |
Christoph Lameter | c9819f4 | 2006-12-10 02:20:25 -0800 | [diff] [blame] | 3081 | next_balance = sd->last_balance + interval; |
Suresh Siddha | f549da8 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 3082 | update_next_balance = 1; |
| 3083 | } |
Siddha, Suresh B | 783609c | 2006-12-10 02:20:33 -0800 | [diff] [blame] | 3084 | |
| 3085 | /* |
| 3086 | * Stop the load balance at this level. There is another |
| 3087 | * CPU in our sched group which is doing load balancing more |
| 3088 | * actively. |
| 3089 | */ |
| 3090 | if (!balance) |
| 3091 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3092 | } |
Suresh Siddha | f549da8 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 3093 | |
| 3094 | /* |
| 3095 | * next_balance will be updated only when there is a need. |
| 3096 | * When the cpu is attached to null domain for ex, it will not be |
| 3097 | * updated. |
| 3098 | */ |
| 3099 | if (likely(update_next_balance)) |
| 3100 | rq->next_balance = next_balance; |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3101 | } |
| 3102 | |
| 3103 | /* |
| 3104 | * run_rebalance_domains is triggered when needed from the scheduler tick. |
| 3105 | * In CONFIG_NO_HZ case, the idle load balance owner will do the |
| 3106 | * rebalancing for all the cpus for whom scheduler ticks are stopped. |
| 3107 | */ |
| 3108 | static void run_rebalance_domains(struct softirq_action *h) |
| 3109 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3110 | int this_cpu = smp_processor_id(); |
| 3111 | struct rq *this_rq = cpu_rq(this_cpu); |
| 3112 | enum cpu_idle_type idle = this_rq->idle_at_tick ? |
| 3113 | CPU_IDLE : CPU_NOT_IDLE; |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3114 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3115 | rebalance_domains(this_cpu, idle); |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3116 | |
| 3117 | #ifdef CONFIG_NO_HZ |
| 3118 | /* |
| 3119 | * If this cpu is the owner for idle load balancing, then do the |
| 3120 | * balancing on behalf of the other idle cpus whose ticks are |
| 3121 | * stopped. |
| 3122 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3123 | if (this_rq->idle_at_tick && |
| 3124 | atomic_read(&nohz.load_balancer) == this_cpu) { |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3125 | cpumask_t cpus = nohz.cpu_mask; |
| 3126 | struct rq *rq; |
| 3127 | int balance_cpu; |
| 3128 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3129 | cpu_clear(this_cpu, cpus); |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3130 | for_each_cpu_mask(balance_cpu, cpus) { |
| 3131 | /* |
| 3132 | * If this cpu gets work to do, stop the load balancing |
| 3133 | * work being done for other cpus. Next load |
| 3134 | * balancing owner will pick it up. |
| 3135 | */ |
| 3136 | if (need_resched()) |
| 3137 | break; |
| 3138 | |
Oleg Nesterov | de0cf89 | 2007-08-12 18:08:19 +0200 | [diff] [blame] | 3139 | rebalance_domains(balance_cpu, CPU_IDLE); |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3140 | |
| 3141 | rq = cpu_rq(balance_cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3142 | if (time_after(this_rq->next_balance, rq->next_balance)) |
| 3143 | this_rq->next_balance = rq->next_balance; |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3144 | } |
| 3145 | } |
| 3146 | #endif |
| 3147 | } |
| 3148 | |
| 3149 | /* |
| 3150 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. |
| 3151 | * |
| 3152 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new |
| 3153 | * idle load balancing owner or decide to stop the periodic load balancing, |
| 3154 | * if the whole system is idle. |
| 3155 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3156 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3157 | { |
Siddha, Suresh B | 46cb4b7 | 2007-05-08 00:32:51 -0700 | [diff] [blame] | 3158 | #ifdef CONFIG_NO_HZ |
| 3159 | /* |
| 3160 | * If we were in the nohz mode recently and busy at the current |
| 3161 | * scheduler tick, then check if we need to nominate new idle |
| 3162 | * load balancer. |
| 3163 | */ |
| 3164 | if (rq->in_nohz_recently && !rq->idle_at_tick) { |
| 3165 | rq->in_nohz_recently = 0; |
| 3166 | |
| 3167 | if (atomic_read(&nohz.load_balancer) == cpu) { |
| 3168 | cpu_clear(cpu, nohz.cpu_mask); |
| 3169 | atomic_set(&nohz.load_balancer, -1); |
| 3170 | } |
| 3171 | |
| 3172 | if (atomic_read(&nohz.load_balancer) == -1) { |
| 3173 | /* |
| 3174 | * simple selection for now: Nominate the |
| 3175 | * first cpu in the nohz list to be the next |
| 3176 | * ilb owner. |
| 3177 | * |
| 3178 | * TBD: Traverse the sched domains and nominate |
| 3179 | * the nearest cpu in the nohz.cpu_mask. |
| 3180 | */ |
| 3181 | int ilb = first_cpu(nohz.cpu_mask); |
| 3182 | |
| 3183 | if (ilb != NR_CPUS) |
| 3184 | resched_cpu(ilb); |
| 3185 | } |
| 3186 | } |
| 3187 | |
| 3188 | /* |
| 3189 | * If this cpu is idle and doing idle load balancing for all the |
| 3190 | * cpus with ticks stopped, is it time for that to stop? |
| 3191 | */ |
| 3192 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && |
| 3193 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { |
| 3194 | resched_cpu(cpu); |
| 3195 | return; |
| 3196 | } |
| 3197 | |
| 3198 | /* |
| 3199 | * If this cpu is idle and the idle load balancing is done by |
| 3200 | * someone else, then no need raise the SCHED_SOFTIRQ |
| 3201 | */ |
| 3202 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && |
| 3203 | cpu_isset(cpu, nohz.cpu_mask)) |
| 3204 | return; |
| 3205 | #endif |
| 3206 | if (time_after_eq(jiffies, rq->next_balance)) |
| 3207 | raise_softirq(SCHED_SOFTIRQ); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3208 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3209 | |
| 3210 | #else /* CONFIG_SMP */ |
| 3211 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3212 | /* |
| 3213 | * on UP we do not need to balance between CPUs: |
| 3214 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 3215 | static inline void idle_balance(int cpu, struct rq *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3216 | { |
| 3217 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3218 | |
| 3219 | /* Avoid "used but not defined" warning on UP */ |
| 3220 | static int balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, |
| 3221 | unsigned long max_nr_move, unsigned long max_load_move, |
| 3222 | struct sched_domain *sd, enum cpu_idle_type idle, |
| 3223 | int *all_pinned, unsigned long *load_moved, |
Peter Williams | a4ac01c | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 3224 | int *this_best_prio, struct rq_iterator *iterator) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3225 | { |
| 3226 | *load_moved = 0; |
| 3227 | |
| 3228 | return 0; |
| 3229 | } |
| 3230 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3231 | #endif |
| 3232 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3233 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
| 3234 | |
| 3235 | EXPORT_PER_CPU_SYMBOL(kstat); |
| 3236 | |
| 3237 | /* |
Ingo Molnar | 41b86e9 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 3238 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
| 3239 | * that have not yet been banked in case the task is currently running. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3240 | */ |
Ingo Molnar | 41b86e9 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 3241 | unsigned long long task_sched_runtime(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3242 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3243 | unsigned long flags; |
Ingo Molnar | 41b86e9 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 3244 | u64 ns, delta_exec; |
| 3245 | struct rq *rq; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3246 | |
Ingo Molnar | 41b86e9 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 3247 | rq = task_rq_lock(p, &flags); |
| 3248 | ns = p->se.sum_exec_runtime; |
| 3249 | if (rq->curr == p) { |
Ingo Molnar | a8e504d | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 3250 | update_rq_clock(rq); |
| 3251 | delta_exec = rq->clock - p->se.exec_start; |
Ingo Molnar | 41b86e9 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 3252 | if ((s64)delta_exec > 0) |
| 3253 | ns += delta_exec; |
| 3254 | } |
| 3255 | task_rq_unlock(rq, &flags); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3256 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3257 | return ns; |
| 3258 | } |
| 3259 | |
| 3260 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3261 | * Account user cpu time to a process. |
| 3262 | * @p: the process that the cpu time gets accounted to |
| 3263 | * @hardirq_offset: the offset to subtract from hardirq_count() |
| 3264 | * @cputime: the cpu time spent in user space since the last update |
| 3265 | */ |
| 3266 | void account_user_time(struct task_struct *p, cputime_t cputime) |
| 3267 | { |
| 3268 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
| 3269 | cputime64_t tmp; |
| 3270 | |
| 3271 | p->utime = cputime_add(p->utime, cputime); |
| 3272 | |
| 3273 | /* Add user time to cpustat. */ |
| 3274 | tmp = cputime_to_cputime64(cputime); |
| 3275 | if (TASK_NICE(p) > 0) |
| 3276 | cpustat->nice = cputime64_add(cpustat->nice, tmp); |
| 3277 | else |
| 3278 | cpustat->user = cputime64_add(cpustat->user, tmp); |
| 3279 | } |
| 3280 | |
| 3281 | /* |
| 3282 | * Account system cpu time to a process. |
| 3283 | * @p: the process that the cpu time gets accounted to |
| 3284 | * @hardirq_offset: the offset to subtract from hardirq_count() |
| 3285 | * @cputime: the cpu time spent in kernel space since the last update |
| 3286 | */ |
| 3287 | void account_system_time(struct task_struct *p, int hardirq_offset, |
| 3288 | cputime_t cputime) |
| 3289 | { |
| 3290 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 3291 | struct rq *rq = this_rq(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3292 | cputime64_t tmp; |
| 3293 | |
| 3294 | p->stime = cputime_add(p->stime, cputime); |
| 3295 | |
| 3296 | /* Add system time to cpustat. */ |
| 3297 | tmp = cputime_to_cputime64(cputime); |
| 3298 | if (hardirq_count() - hardirq_offset) |
| 3299 | cpustat->irq = cputime64_add(cpustat->irq, tmp); |
| 3300 | else if (softirq_count()) |
| 3301 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); |
| 3302 | else if (p != rq->idle) |
| 3303 | cpustat->system = cputime64_add(cpustat->system, tmp); |
| 3304 | else if (atomic_read(&rq->nr_iowait) > 0) |
| 3305 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
| 3306 | else |
| 3307 | cpustat->idle = cputime64_add(cpustat->idle, tmp); |
| 3308 | /* Account for system time used */ |
| 3309 | acct_update_integrals(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3310 | } |
| 3311 | |
| 3312 | /* |
| 3313 | * Account for involuntary wait time. |
| 3314 | * @p: the process from which the cpu time has been stolen |
| 3315 | * @steal: the cpu time spent in involuntary wait |
| 3316 | */ |
| 3317 | void account_steal_time(struct task_struct *p, cputime_t steal) |
| 3318 | { |
| 3319 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
| 3320 | cputime64_t tmp = cputime_to_cputime64(steal); |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 3321 | struct rq *rq = this_rq(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3322 | |
| 3323 | if (p == rq->idle) { |
| 3324 | p->stime = cputime_add(p->stime, steal); |
| 3325 | if (atomic_read(&rq->nr_iowait) > 0) |
| 3326 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
| 3327 | else |
| 3328 | cpustat->idle = cputime64_add(cpustat->idle, tmp); |
| 3329 | } else |
| 3330 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
| 3331 | } |
| 3332 | |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 3333 | /* |
| 3334 | * This function gets called by the timer code, with HZ frequency. |
| 3335 | * We call it with interrupts disabled. |
| 3336 | * |
| 3337 | * It also gets called by the fork code, when changing the parent's |
| 3338 | * timeslices. |
| 3339 | */ |
| 3340 | void scheduler_tick(void) |
| 3341 | { |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 3342 | int cpu = smp_processor_id(); |
| 3343 | struct rq *rq = cpu_rq(cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3344 | struct task_struct *curr = rq->curr; |
Ingo Molnar | 529c772 | 2007-08-10 23:05:11 +0200 | [diff] [blame] | 3345 | u64 next_tick = rq->tick_timestamp + TICK_NSEC; |
Christoph Lameter | 7835b98 | 2006-12-10 02:20:22 -0800 | [diff] [blame] | 3346 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3347 | spin_lock(&rq->lock); |
Ingo Molnar | 546fe3c | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 3348 | __update_rq_clock(rq); |
Ingo Molnar | 529c772 | 2007-08-10 23:05:11 +0200 | [diff] [blame] | 3349 | /* |
| 3350 | * Let rq->clock advance by at least TICK_NSEC: |
| 3351 | */ |
| 3352 | if (unlikely(rq->clock < next_tick)) |
| 3353 | rq->clock = next_tick; |
| 3354 | rq->tick_timestamp = rq->clock; |
Ingo Molnar | f1a438d | 2007-08-09 11:16:45 +0200 | [diff] [blame] | 3355 | update_cpu_load(rq); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3356 | if (curr != rq->idle) /* FIXME: needed? */ |
| 3357 | curr->sched_class->task_tick(rq, curr); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3358 | spin_unlock(&rq->lock); |
| 3359 | |
Christoph Lameter | e418e1c | 2006-12-10 02:20:23 -0800 | [diff] [blame] | 3360 | #ifdef CONFIG_SMP |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3361 | rq->idle_at_tick = idle_cpu(cpu); |
| 3362 | trigger_load_balance(rq, cpu); |
Christoph Lameter | e418e1c | 2006-12-10 02:20:23 -0800 | [diff] [blame] | 3363 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3364 | } |
| 3365 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3366 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
| 3367 | |
| 3368 | void fastcall add_preempt_count(int val) |
| 3369 | { |
| 3370 | /* |
| 3371 | * Underflow? |
| 3372 | */ |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 3373 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
| 3374 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3375 | preempt_count() += val; |
| 3376 | /* |
| 3377 | * Spinlock count overflowing soon? |
| 3378 | */ |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 3379 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
| 3380 | PREEMPT_MASK - 10); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3381 | } |
| 3382 | EXPORT_SYMBOL(add_preempt_count); |
| 3383 | |
| 3384 | void fastcall sub_preempt_count(int val) |
| 3385 | { |
| 3386 | /* |
| 3387 | * Underflow? |
| 3388 | */ |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 3389 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
| 3390 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3391 | /* |
| 3392 | * Is the spinlock portion underflowing? |
| 3393 | */ |
Ingo Molnar | 9a11b49a | 2006-07-03 00:24:33 -0700 | [diff] [blame] | 3394 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
| 3395 | !(preempt_count() & PREEMPT_MASK))) |
| 3396 | return; |
| 3397 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3398 | preempt_count() -= val; |
| 3399 | } |
| 3400 | EXPORT_SYMBOL(sub_preempt_count); |
| 3401 | |
| 3402 | #endif |
| 3403 | |
| 3404 | /* |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3405 | * Print scheduling while atomic bug: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3406 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3407 | static noinline void __schedule_bug(struct task_struct *prev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3408 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3409 | printk(KERN_ERR "BUG: scheduling while atomic: %s/0x%08x/%d\n", |
| 3410 | prev->comm, preempt_count(), prev->pid); |
| 3411 | debug_show_held_locks(prev); |
| 3412 | if (irqs_disabled()) |
| 3413 | print_irqtrace_events(prev); |
| 3414 | dump_stack(); |
| 3415 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3416 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3417 | /* |
| 3418 | * Various schedule()-time debugging checks and statistics: |
| 3419 | */ |
| 3420 | static inline void schedule_debug(struct task_struct *prev) |
| 3421 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3422 | /* |
| 3423 | * Test if we are atomic. Since do_exit() needs to call into |
| 3424 | * schedule() atomically, we ignore that path for now. |
| 3425 | * Otherwise, whine if we are scheduling when we should not be. |
| 3426 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3427 | if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) |
| 3428 | __schedule_bug(prev); |
| 3429 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3430 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
| 3431 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3432 | schedstat_inc(this_rq(), sched_cnt); |
| 3433 | } |
| 3434 | |
| 3435 | /* |
| 3436 | * Pick up the highest-prio task: |
| 3437 | */ |
| 3438 | static inline struct task_struct * |
Ingo Molnar | ff95f3d | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 3439 | pick_next_task(struct rq *rq, struct task_struct *prev) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3440 | { |
| 3441 | struct sched_class *class; |
| 3442 | struct task_struct *p; |
| 3443 | |
| 3444 | /* |
| 3445 | * Optimization: we know that if all tasks are in |
| 3446 | * the fair class we can call that function directly: |
| 3447 | */ |
| 3448 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
Ingo Molnar | fb8d472 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 3449 | p = fair_sched_class.pick_next_task(rq); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3450 | if (likely(p)) |
| 3451 | return p; |
| 3452 | } |
| 3453 | |
| 3454 | class = sched_class_highest; |
| 3455 | for ( ; ; ) { |
Ingo Molnar | fb8d472 | 2007-08-09 11:16:48 +0200 | [diff] [blame] | 3456 | p = class->pick_next_task(rq); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3457 | if (p) |
| 3458 | return p; |
| 3459 | /* |
| 3460 | * Will never be NULL as the idle class always |
| 3461 | * returns a non-NULL p: |
| 3462 | */ |
| 3463 | class = class->next; |
| 3464 | } |
| 3465 | } |
| 3466 | |
| 3467 | /* |
| 3468 | * schedule() is the main scheduler function. |
| 3469 | */ |
| 3470 | asmlinkage void __sched schedule(void) |
| 3471 | { |
| 3472 | struct task_struct *prev, *next; |
| 3473 | long *switch_count; |
| 3474 | struct rq *rq; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3475 | int cpu; |
| 3476 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3477 | need_resched: |
| 3478 | preempt_disable(); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3479 | cpu = smp_processor_id(); |
| 3480 | rq = cpu_rq(cpu); |
| 3481 | rcu_qsctr_inc(cpu); |
| 3482 | prev = rq->curr; |
| 3483 | switch_count = &prev->nivcsw; |
| 3484 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3485 | release_kernel_lock(prev); |
| 3486 | need_resched_nonpreemptible: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3487 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3488 | schedule_debug(prev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3489 | |
| 3490 | spin_lock_irq(&rq->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3491 | clear_tsk_need_resched(prev); |
Ingo Molnar | c1b3da3 | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 3492 | __update_rq_clock(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3493 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3494 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
| 3495 | if (unlikely((prev->state & TASK_INTERRUPTIBLE) && |
| 3496 | unlikely(signal_pending(prev)))) { |
| 3497 | prev->state = TASK_RUNNING; |
| 3498 | } else { |
Ingo Molnar | 2e1cb74 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 3499 | deactivate_task(rq, prev, 1); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3500 | } |
| 3501 | switch_count = &prev->nvcsw; |
| 3502 | } |
| 3503 | |
| 3504 | if (unlikely(!rq->nr_running)) |
| 3505 | idle_balance(cpu, rq); |
| 3506 | |
Ingo Molnar | 31ee529 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 3507 | prev->sched_class->put_prev_task(rq, prev); |
Ingo Molnar | ff95f3d | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 3508 | next = pick_next_task(rq, prev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3509 | |
| 3510 | sched_info_switch(prev, next); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3511 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3512 | if (likely(prev != next)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3513 | rq->nr_switches++; |
| 3514 | rq->curr = next; |
| 3515 | ++*switch_count; |
| 3516 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3517 | context_switch(rq, prev, next); /* unlocks the rq */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3518 | } else |
| 3519 | spin_unlock_irq(&rq->lock); |
| 3520 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3521 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
| 3522 | cpu = smp_processor_id(); |
| 3523 | rq = cpu_rq(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3524 | goto need_resched_nonpreemptible; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3525 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3526 | preempt_enable_no_resched(); |
| 3527 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) |
| 3528 | goto need_resched; |
| 3529 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3530 | EXPORT_SYMBOL(schedule); |
| 3531 | |
| 3532 | #ifdef CONFIG_PREEMPT |
| 3533 | /* |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 3534 | * this is the entry point to schedule() from in-kernel preemption |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3535 | * off of preempt_enable. Kernel preemptions off return from interrupt |
| 3536 | * occur there and call schedule directly. |
| 3537 | */ |
| 3538 | asmlinkage void __sched preempt_schedule(void) |
| 3539 | { |
| 3540 | struct thread_info *ti = current_thread_info(); |
| 3541 | #ifdef CONFIG_PREEMPT_BKL |
| 3542 | struct task_struct *task = current; |
| 3543 | int saved_lock_depth; |
| 3544 | #endif |
| 3545 | /* |
| 3546 | * If there is a non-zero preempt_count or interrupts are disabled, |
| 3547 | * we do not want to preempt the current task. Just return.. |
| 3548 | */ |
Nick Piggin | beed33a | 2006-10-11 01:21:52 -0700 | [diff] [blame] | 3549 | if (likely(ti->preempt_count || irqs_disabled())) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3550 | return; |
| 3551 | |
| 3552 | need_resched: |
| 3553 | add_preempt_count(PREEMPT_ACTIVE); |
| 3554 | /* |
| 3555 | * We keep the big kernel semaphore locked, but we |
| 3556 | * clear ->lock_depth so that schedule() doesnt |
| 3557 | * auto-release the semaphore: |
| 3558 | */ |
| 3559 | #ifdef CONFIG_PREEMPT_BKL |
| 3560 | saved_lock_depth = task->lock_depth; |
| 3561 | task->lock_depth = -1; |
| 3562 | #endif |
| 3563 | schedule(); |
| 3564 | #ifdef CONFIG_PREEMPT_BKL |
| 3565 | task->lock_depth = saved_lock_depth; |
| 3566 | #endif |
| 3567 | sub_preempt_count(PREEMPT_ACTIVE); |
| 3568 | |
| 3569 | /* we could miss a preemption opportunity between schedule and now */ |
| 3570 | barrier(); |
| 3571 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) |
| 3572 | goto need_resched; |
| 3573 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3574 | EXPORT_SYMBOL(preempt_schedule); |
| 3575 | |
| 3576 | /* |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 3577 | * this is the entry point to schedule() from kernel preemption |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3578 | * off of irq context. |
| 3579 | * Note, that this is called and return with irqs disabled. This will |
| 3580 | * protect us against recursive calling from irq. |
| 3581 | */ |
| 3582 | asmlinkage void __sched preempt_schedule_irq(void) |
| 3583 | { |
| 3584 | struct thread_info *ti = current_thread_info(); |
| 3585 | #ifdef CONFIG_PREEMPT_BKL |
| 3586 | struct task_struct *task = current; |
| 3587 | int saved_lock_depth; |
| 3588 | #endif |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 3589 | /* Catch callers which need to be fixed */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3590 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
| 3591 | |
| 3592 | need_resched: |
| 3593 | add_preempt_count(PREEMPT_ACTIVE); |
| 3594 | /* |
| 3595 | * We keep the big kernel semaphore locked, but we |
| 3596 | * clear ->lock_depth so that schedule() doesnt |
| 3597 | * auto-release the semaphore: |
| 3598 | */ |
| 3599 | #ifdef CONFIG_PREEMPT_BKL |
| 3600 | saved_lock_depth = task->lock_depth; |
| 3601 | task->lock_depth = -1; |
| 3602 | #endif |
| 3603 | local_irq_enable(); |
| 3604 | schedule(); |
| 3605 | local_irq_disable(); |
| 3606 | #ifdef CONFIG_PREEMPT_BKL |
| 3607 | task->lock_depth = saved_lock_depth; |
| 3608 | #endif |
| 3609 | sub_preempt_count(PREEMPT_ACTIVE); |
| 3610 | |
| 3611 | /* we could miss a preemption opportunity between schedule and now */ |
| 3612 | barrier(); |
| 3613 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) |
| 3614 | goto need_resched; |
| 3615 | } |
| 3616 | |
| 3617 | #endif /* CONFIG_PREEMPT */ |
| 3618 | |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 3619 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
| 3620 | void *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3621 | { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3622 | return try_to_wake_up(curr->private, mode, sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3623 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3624 | EXPORT_SYMBOL(default_wake_function); |
| 3625 | |
| 3626 | /* |
| 3627 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
| 3628 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve |
| 3629 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
| 3630 | * |
| 3631 | * There are circumstances in which we can try to wake a task which has already |
| 3632 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
| 3633 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
| 3634 | */ |
| 3635 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
| 3636 | int nr_exclusive, int sync, void *key) |
| 3637 | { |
| 3638 | struct list_head *tmp, *next; |
| 3639 | |
| 3640 | list_for_each_safe(tmp, next, &q->task_list) { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3641 | wait_queue_t *curr = list_entry(tmp, wait_queue_t, task_list); |
| 3642 | unsigned flags = curr->flags; |
| 3643 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3644 | if (curr->func(curr, mode, sync, key) && |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3645 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3646 | break; |
| 3647 | } |
| 3648 | } |
| 3649 | |
| 3650 | /** |
| 3651 | * __wake_up - wake up threads blocked on a waitqueue. |
| 3652 | * @q: the waitqueue |
| 3653 | * @mode: which threads |
| 3654 | * @nr_exclusive: how many wake-one or wake-many threads to wake up |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 3655 | * @key: is directly passed to the wakeup function |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3656 | */ |
| 3657 | void fastcall __wake_up(wait_queue_head_t *q, unsigned int mode, |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 3658 | int nr_exclusive, void *key) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3659 | { |
| 3660 | unsigned long flags; |
| 3661 | |
| 3662 | spin_lock_irqsave(&q->lock, flags); |
| 3663 | __wake_up_common(q, mode, nr_exclusive, 0, key); |
| 3664 | spin_unlock_irqrestore(&q->lock, flags); |
| 3665 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3666 | EXPORT_SYMBOL(__wake_up); |
| 3667 | |
| 3668 | /* |
| 3669 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. |
| 3670 | */ |
| 3671 | void fastcall __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
| 3672 | { |
| 3673 | __wake_up_common(q, mode, 1, 0, NULL); |
| 3674 | } |
| 3675 | |
| 3676 | /** |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 3677 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3678 | * @q: the waitqueue |
| 3679 | * @mode: which threads |
| 3680 | * @nr_exclusive: how many wake-one or wake-many threads to wake up |
| 3681 | * |
| 3682 | * The sync wakeup differs that the waker knows that it will schedule |
| 3683 | * away soon, so while the target thread will be woken up, it will not |
| 3684 | * be migrated to another CPU - ie. the two threads are 'synchronized' |
| 3685 | * with each other. This can prevent needless bouncing between CPUs. |
| 3686 | * |
| 3687 | * On UP it can prevent extra preemption. |
| 3688 | */ |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 3689 | void fastcall |
| 3690 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3691 | { |
| 3692 | unsigned long flags; |
| 3693 | int sync = 1; |
| 3694 | |
| 3695 | if (unlikely(!q)) |
| 3696 | return; |
| 3697 | |
| 3698 | if (unlikely(!nr_exclusive)) |
| 3699 | sync = 0; |
| 3700 | |
| 3701 | spin_lock_irqsave(&q->lock, flags); |
| 3702 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); |
| 3703 | spin_unlock_irqrestore(&q->lock, flags); |
| 3704 | } |
| 3705 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
| 3706 | |
| 3707 | void fastcall complete(struct completion *x) |
| 3708 | { |
| 3709 | unsigned long flags; |
| 3710 | |
| 3711 | spin_lock_irqsave(&x->wait.lock, flags); |
| 3712 | x->done++; |
| 3713 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, |
| 3714 | 1, 0, NULL); |
| 3715 | spin_unlock_irqrestore(&x->wait.lock, flags); |
| 3716 | } |
| 3717 | EXPORT_SYMBOL(complete); |
| 3718 | |
| 3719 | void fastcall complete_all(struct completion *x) |
| 3720 | { |
| 3721 | unsigned long flags; |
| 3722 | |
| 3723 | spin_lock_irqsave(&x->wait.lock, flags); |
| 3724 | x->done += UINT_MAX/2; |
| 3725 | __wake_up_common(&x->wait, TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE, |
| 3726 | 0, 0, NULL); |
| 3727 | spin_unlock_irqrestore(&x->wait.lock, flags); |
| 3728 | } |
| 3729 | EXPORT_SYMBOL(complete_all); |
| 3730 | |
| 3731 | void fastcall __sched wait_for_completion(struct completion *x) |
| 3732 | { |
| 3733 | might_sleep(); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 3734 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3735 | spin_lock_irq(&x->wait.lock); |
| 3736 | if (!x->done) { |
| 3737 | DECLARE_WAITQUEUE(wait, current); |
| 3738 | |
| 3739 | wait.flags |= WQ_FLAG_EXCLUSIVE; |
| 3740 | __add_wait_queue_tail(&x->wait, &wait); |
| 3741 | do { |
| 3742 | __set_current_state(TASK_UNINTERRUPTIBLE); |
| 3743 | spin_unlock_irq(&x->wait.lock); |
| 3744 | schedule(); |
| 3745 | spin_lock_irq(&x->wait.lock); |
| 3746 | } while (!x->done); |
| 3747 | __remove_wait_queue(&x->wait, &wait); |
| 3748 | } |
| 3749 | x->done--; |
| 3750 | spin_unlock_irq(&x->wait.lock); |
| 3751 | } |
| 3752 | EXPORT_SYMBOL(wait_for_completion); |
| 3753 | |
| 3754 | unsigned long fastcall __sched |
| 3755 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
| 3756 | { |
| 3757 | might_sleep(); |
| 3758 | |
| 3759 | spin_lock_irq(&x->wait.lock); |
| 3760 | if (!x->done) { |
| 3761 | DECLARE_WAITQUEUE(wait, current); |
| 3762 | |
| 3763 | wait.flags |= WQ_FLAG_EXCLUSIVE; |
| 3764 | __add_wait_queue_tail(&x->wait, &wait); |
| 3765 | do { |
| 3766 | __set_current_state(TASK_UNINTERRUPTIBLE); |
| 3767 | spin_unlock_irq(&x->wait.lock); |
| 3768 | timeout = schedule_timeout(timeout); |
| 3769 | spin_lock_irq(&x->wait.lock); |
| 3770 | if (!timeout) { |
| 3771 | __remove_wait_queue(&x->wait, &wait); |
| 3772 | goto out; |
| 3773 | } |
| 3774 | } while (!x->done); |
| 3775 | __remove_wait_queue(&x->wait, &wait); |
| 3776 | } |
| 3777 | x->done--; |
| 3778 | out: |
| 3779 | spin_unlock_irq(&x->wait.lock); |
| 3780 | return timeout; |
| 3781 | } |
| 3782 | EXPORT_SYMBOL(wait_for_completion_timeout); |
| 3783 | |
| 3784 | int fastcall __sched wait_for_completion_interruptible(struct completion *x) |
| 3785 | { |
| 3786 | int ret = 0; |
| 3787 | |
| 3788 | might_sleep(); |
| 3789 | |
| 3790 | spin_lock_irq(&x->wait.lock); |
| 3791 | if (!x->done) { |
| 3792 | DECLARE_WAITQUEUE(wait, current); |
| 3793 | |
| 3794 | wait.flags |= WQ_FLAG_EXCLUSIVE; |
| 3795 | __add_wait_queue_tail(&x->wait, &wait); |
| 3796 | do { |
| 3797 | if (signal_pending(current)) { |
| 3798 | ret = -ERESTARTSYS; |
| 3799 | __remove_wait_queue(&x->wait, &wait); |
| 3800 | goto out; |
| 3801 | } |
| 3802 | __set_current_state(TASK_INTERRUPTIBLE); |
| 3803 | spin_unlock_irq(&x->wait.lock); |
| 3804 | schedule(); |
| 3805 | spin_lock_irq(&x->wait.lock); |
| 3806 | } while (!x->done); |
| 3807 | __remove_wait_queue(&x->wait, &wait); |
| 3808 | } |
| 3809 | x->done--; |
| 3810 | out: |
| 3811 | spin_unlock_irq(&x->wait.lock); |
| 3812 | |
| 3813 | return ret; |
| 3814 | } |
| 3815 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
| 3816 | |
| 3817 | unsigned long fastcall __sched |
| 3818 | wait_for_completion_interruptible_timeout(struct completion *x, |
| 3819 | unsigned long timeout) |
| 3820 | { |
| 3821 | might_sleep(); |
| 3822 | |
| 3823 | spin_lock_irq(&x->wait.lock); |
| 3824 | if (!x->done) { |
| 3825 | DECLARE_WAITQUEUE(wait, current); |
| 3826 | |
| 3827 | wait.flags |= WQ_FLAG_EXCLUSIVE; |
| 3828 | __add_wait_queue_tail(&x->wait, &wait); |
| 3829 | do { |
| 3830 | if (signal_pending(current)) { |
| 3831 | timeout = -ERESTARTSYS; |
| 3832 | __remove_wait_queue(&x->wait, &wait); |
| 3833 | goto out; |
| 3834 | } |
| 3835 | __set_current_state(TASK_INTERRUPTIBLE); |
| 3836 | spin_unlock_irq(&x->wait.lock); |
| 3837 | timeout = schedule_timeout(timeout); |
| 3838 | spin_lock_irq(&x->wait.lock); |
| 3839 | if (!timeout) { |
| 3840 | __remove_wait_queue(&x->wait, &wait); |
| 3841 | goto out; |
| 3842 | } |
| 3843 | } while (!x->done); |
| 3844 | __remove_wait_queue(&x->wait, &wait); |
| 3845 | } |
| 3846 | x->done--; |
| 3847 | out: |
| 3848 | spin_unlock_irq(&x->wait.lock); |
| 3849 | return timeout; |
| 3850 | } |
| 3851 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
| 3852 | |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3853 | static inline void |
| 3854 | sleep_on_head(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3855 | { |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3856 | spin_lock_irqsave(&q->lock, *flags); |
| 3857 | __add_wait_queue(q, wait); |
| 3858 | spin_unlock(&q->lock); |
| 3859 | } |
| 3860 | |
| 3861 | static inline void |
| 3862 | sleep_on_tail(wait_queue_head_t *q, wait_queue_t *wait, unsigned long *flags) |
| 3863 | { |
| 3864 | spin_lock_irq(&q->lock); |
| 3865 | __remove_wait_queue(q, wait); |
| 3866 | spin_unlock_irqrestore(&q->lock, *flags); |
| 3867 | } |
| 3868 | |
| 3869 | void __sched interruptible_sleep_on(wait_queue_head_t *q) |
| 3870 | { |
| 3871 | unsigned long flags; |
| 3872 | wait_queue_t wait; |
| 3873 | |
| 3874 | init_waitqueue_entry(&wait, current); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3875 | |
| 3876 | current->state = TASK_INTERRUPTIBLE; |
| 3877 | |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3878 | sleep_on_head(q, &wait, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3879 | schedule(); |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3880 | sleep_on_tail(q, &wait, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3881 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3882 | EXPORT_SYMBOL(interruptible_sleep_on); |
| 3883 | |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3884 | long __sched |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 3885 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3886 | { |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3887 | unsigned long flags; |
| 3888 | wait_queue_t wait; |
| 3889 | |
| 3890 | init_waitqueue_entry(&wait, current); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3891 | |
| 3892 | current->state = TASK_INTERRUPTIBLE; |
| 3893 | |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3894 | sleep_on_head(q, &wait, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3895 | timeout = schedule_timeout(timeout); |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3896 | sleep_on_tail(q, &wait, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3897 | |
| 3898 | return timeout; |
| 3899 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3900 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
| 3901 | |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3902 | void __sched sleep_on(wait_queue_head_t *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3903 | { |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3904 | unsigned long flags; |
| 3905 | wait_queue_t wait; |
| 3906 | |
| 3907 | init_waitqueue_entry(&wait, current); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3908 | |
| 3909 | current->state = TASK_UNINTERRUPTIBLE; |
| 3910 | |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3911 | sleep_on_head(q, &wait, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3912 | schedule(); |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3913 | sleep_on_tail(q, &wait, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3914 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3915 | EXPORT_SYMBOL(sleep_on); |
| 3916 | |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3917 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3918 | { |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3919 | unsigned long flags; |
| 3920 | wait_queue_t wait; |
| 3921 | |
| 3922 | init_waitqueue_entry(&wait, current); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3923 | |
| 3924 | current->state = TASK_UNINTERRUPTIBLE; |
| 3925 | |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3926 | sleep_on_head(q, &wait, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3927 | timeout = schedule_timeout(timeout); |
Ingo Molnar | 0fec171 | 2007-07-09 18:52:01 +0200 | [diff] [blame] | 3928 | sleep_on_tail(q, &wait, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3929 | |
| 3930 | return timeout; |
| 3931 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3932 | EXPORT_SYMBOL(sleep_on_timeout); |
| 3933 | |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3934 | #ifdef CONFIG_RT_MUTEXES |
| 3935 | |
| 3936 | /* |
| 3937 | * rt_mutex_setprio - set the current priority of a task |
| 3938 | * @p: task |
| 3939 | * @prio: prio value (kernel-internal form) |
| 3940 | * |
| 3941 | * This function changes the 'effective' priority of a task. It does |
| 3942 | * not touch ->normal_prio like __setscheduler(). |
| 3943 | * |
| 3944 | * Used by the rt_mutex code to implement priority inheritance logic. |
| 3945 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3946 | void rt_mutex_setprio(struct task_struct *p, int prio) |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3947 | { |
| 3948 | unsigned long flags; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3949 | int oldprio, on_rq; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 3950 | struct rq *rq; |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3951 | |
| 3952 | BUG_ON(prio < 0 || prio > MAX_PRIO); |
| 3953 | |
| 3954 | rq = task_rq_lock(p, &flags); |
Ingo Molnar | a8e504d | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 3955 | update_rq_clock(rq); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3956 | |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 3957 | oldprio = p->prio; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3958 | on_rq = p->se.on_rq; |
| 3959 | if (on_rq) |
Ingo Molnar | 69be72c | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 3960 | dequeue_task(rq, p, 0); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3961 | |
| 3962 | if (rt_prio(prio)) |
| 3963 | p->sched_class = &rt_sched_class; |
| 3964 | else |
| 3965 | p->sched_class = &fair_sched_class; |
| 3966 | |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3967 | p->prio = prio; |
| 3968 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3969 | if (on_rq) { |
Ingo Molnar | 8159f87 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 3970 | enqueue_task(rq, p, 0); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3971 | /* |
| 3972 | * Reschedule if we are currently running on this runqueue and |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 3973 | * our priority decreased, or if we are not currently running on |
| 3974 | * this runqueue and our priority is higher than the current's |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3975 | */ |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 3976 | if (task_running(rq, p)) { |
| 3977 | if (p->prio > oldprio) |
| 3978 | resched_task(rq->curr); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3979 | } else { |
| 3980 | check_preempt_curr(rq, p); |
| 3981 | } |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 3982 | } |
| 3983 | task_rq_unlock(rq, &flags); |
| 3984 | } |
| 3985 | |
| 3986 | #endif |
| 3987 | |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 3988 | void set_user_nice(struct task_struct *p, long nice) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3989 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 3990 | int old_prio, delta, on_rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3991 | unsigned long flags; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 3992 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3993 | |
| 3994 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) |
| 3995 | return; |
| 3996 | /* |
| 3997 | * We have to be careful, if called from sys_setpriority(), |
| 3998 | * the task might be in the middle of scheduling on another CPU. |
| 3999 | */ |
| 4000 | rq = task_rq_lock(p, &flags); |
Ingo Molnar | a8e504d | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 4001 | update_rq_clock(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4002 | /* |
| 4003 | * The RT priorities are set via sched_setscheduler(), but we still |
| 4004 | * allow the 'normal' nice value to be set - but as expected |
| 4005 | * it wont have any effect on scheduling until the task is |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4006 | * SCHED_FIFO/SCHED_RR: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4007 | */ |
Ingo Molnar | e05606d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4008 | if (task_has_rt_policy(p)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4009 | p->static_prio = NICE_TO_PRIO(nice); |
| 4010 | goto out_unlock; |
| 4011 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4012 | on_rq = p->se.on_rq; |
| 4013 | if (on_rq) { |
Ingo Molnar | 69be72c | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 4014 | dequeue_task(rq, p, 0); |
Ingo Molnar | 79b5ddd | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 4015 | dec_load(rq, p); |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 4016 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4017 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4018 | p->static_prio = NICE_TO_PRIO(nice); |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 4019 | set_load_weight(p); |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 4020 | old_prio = p->prio; |
| 4021 | p->prio = effective_prio(p); |
| 4022 | delta = p->prio - old_prio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4023 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4024 | if (on_rq) { |
Ingo Molnar | 8159f87 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 4025 | enqueue_task(rq, p, 0); |
Ingo Molnar | 29b4b62 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 4026 | inc_load(rq, p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4027 | /* |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 4028 | * If the task increased its priority or is running and |
| 4029 | * lowered its priority, then reschedule its CPU: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4030 | */ |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 4031 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4032 | resched_task(rq->curr); |
| 4033 | } |
| 4034 | out_unlock: |
| 4035 | task_rq_unlock(rq, &flags); |
| 4036 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4037 | EXPORT_SYMBOL(set_user_nice); |
| 4038 | |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 4039 | /* |
| 4040 | * can_nice - check if a task can reduce its nice value |
| 4041 | * @p: task |
| 4042 | * @nice: nice value |
| 4043 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4044 | int can_nice(const struct task_struct *p, const int nice) |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 4045 | { |
Matt Mackall | 024f474 | 2005-08-18 11:24:19 -0700 | [diff] [blame] | 4046 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
| 4047 | int nice_rlim = 20 - nice; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 4048 | |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 4049 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
| 4050 | capable(CAP_SYS_NICE)); |
| 4051 | } |
| 4052 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4053 | #ifdef __ARCH_WANT_SYS_NICE |
| 4054 | |
| 4055 | /* |
| 4056 | * sys_nice - change the priority of the current process. |
| 4057 | * @increment: priority increment |
| 4058 | * |
| 4059 | * sys_setpriority is a more generic, but much slower function that |
| 4060 | * does similar things. |
| 4061 | */ |
| 4062 | asmlinkage long sys_nice(int increment) |
| 4063 | { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 4064 | long nice, retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4065 | |
| 4066 | /* |
| 4067 | * Setpriority might change our priority at the same moment. |
| 4068 | * We don't have to worry. Conceptually one call occurs first |
| 4069 | * and we have a single winner. |
| 4070 | */ |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 4071 | if (increment < -40) |
| 4072 | increment = -40; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4073 | if (increment > 40) |
| 4074 | increment = 40; |
| 4075 | |
| 4076 | nice = PRIO_TO_NICE(current->static_prio) + increment; |
| 4077 | if (nice < -20) |
| 4078 | nice = -20; |
| 4079 | if (nice > 19) |
| 4080 | nice = 19; |
| 4081 | |
Matt Mackall | e43379f | 2005-05-01 08:59:00 -0700 | [diff] [blame] | 4082 | if (increment < 0 && !can_nice(current, nice)) |
| 4083 | return -EPERM; |
| 4084 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4085 | retval = security_task_setnice(current, nice); |
| 4086 | if (retval) |
| 4087 | return retval; |
| 4088 | |
| 4089 | set_user_nice(current, nice); |
| 4090 | return 0; |
| 4091 | } |
| 4092 | |
| 4093 | #endif |
| 4094 | |
| 4095 | /** |
| 4096 | * task_prio - return the priority value of a given task. |
| 4097 | * @p: the task in question. |
| 4098 | * |
| 4099 | * This is the priority value as seen by users in /proc. |
| 4100 | * RT tasks are offset by -200. Normal tasks are centered |
| 4101 | * around 0, value goes from -16 to +15. |
| 4102 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4103 | int task_prio(const struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4104 | { |
| 4105 | return p->prio - MAX_RT_PRIO; |
| 4106 | } |
| 4107 | |
| 4108 | /** |
| 4109 | * task_nice - return the nice value of a given task. |
| 4110 | * @p: the task in question. |
| 4111 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4112 | int task_nice(const struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4113 | { |
| 4114 | return TASK_NICE(p); |
| 4115 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4116 | EXPORT_SYMBOL_GPL(task_nice); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4117 | |
| 4118 | /** |
| 4119 | * idle_cpu - is a given cpu idle currently? |
| 4120 | * @cpu: the processor in question. |
| 4121 | */ |
| 4122 | int idle_cpu(int cpu) |
| 4123 | { |
| 4124 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; |
| 4125 | } |
| 4126 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4127 | /** |
| 4128 | * idle_task - return the idle task for a given cpu. |
| 4129 | * @cpu: the processor in question. |
| 4130 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4131 | struct task_struct *idle_task(int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4132 | { |
| 4133 | return cpu_rq(cpu)->idle; |
| 4134 | } |
| 4135 | |
| 4136 | /** |
| 4137 | * find_process_by_pid - find a process with a matching PID value. |
| 4138 | * @pid: the pid in question. |
| 4139 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4140 | static inline struct task_struct *find_process_by_pid(pid_t pid) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4141 | { |
| 4142 | return pid ? find_task_by_pid(pid) : current; |
| 4143 | } |
| 4144 | |
| 4145 | /* Actually do priority change: must hold rq lock. */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4146 | static void |
| 4147 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4148 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4149 | BUG_ON(p->se.on_rq); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 4150 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4151 | p->policy = policy; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4152 | switch (p->policy) { |
| 4153 | case SCHED_NORMAL: |
| 4154 | case SCHED_BATCH: |
| 4155 | case SCHED_IDLE: |
| 4156 | p->sched_class = &fair_sched_class; |
| 4157 | break; |
| 4158 | case SCHED_FIFO: |
| 4159 | case SCHED_RR: |
| 4160 | p->sched_class = &rt_sched_class; |
| 4161 | break; |
| 4162 | } |
| 4163 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4164 | p->rt_priority = prio; |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 4165 | p->normal_prio = normal_prio(p); |
| 4166 | /* we are holding p->pi_lock already */ |
| 4167 | p->prio = rt_mutex_getprio(p); |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 4168 | set_load_weight(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4169 | } |
| 4170 | |
| 4171 | /** |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 4172 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4173 | * @p: the task in question. |
| 4174 | * @policy: new policy. |
| 4175 | * @param: structure containing the new RT priority. |
Oleg Nesterov | 5fe1d75 | 2006-09-29 02:00:48 -0700 | [diff] [blame] | 4176 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 4177 | * NOTE that the task may be already dead. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4178 | */ |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 4179 | int sched_setscheduler(struct task_struct *p, int policy, |
| 4180 | struct sched_param *param) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4181 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4182 | int retval, oldprio, oldpolicy = -1, on_rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4183 | unsigned long flags; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4184 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4185 | |
Steven Rostedt | 66e5393 | 2006-06-27 02:54:44 -0700 | [diff] [blame] | 4186 | /* may grab non-irq protected spin_locks */ |
| 4187 | BUG_ON(in_interrupt()); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4188 | recheck: |
| 4189 | /* double check policy once rq lock held */ |
| 4190 | if (policy < 0) |
| 4191 | policy = oldpolicy = p->policy; |
| 4192 | else if (policy != SCHED_FIFO && policy != SCHED_RR && |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4193 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
| 4194 | policy != SCHED_IDLE) |
Ingo Molnar | b0a9499 | 2006-01-14 13:20:41 -0800 | [diff] [blame] | 4195 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4196 | /* |
| 4197 | * Valid priorities for SCHED_FIFO and SCHED_RR are |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4198 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
| 4199 | * SCHED_BATCH and SCHED_IDLE is 0. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4200 | */ |
| 4201 | if (param->sched_priority < 0 || |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 4202 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
Steven Rostedt | d46523e | 2005-07-25 16:28:39 -0400 | [diff] [blame] | 4203 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4204 | return -EINVAL; |
Ingo Molnar | e05606d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4205 | if (rt_policy(policy) != (param->sched_priority != 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4206 | return -EINVAL; |
| 4207 | |
Olivier Croquette | 37e4ab3 | 2005-06-25 14:57:32 -0700 | [diff] [blame] | 4208 | /* |
| 4209 | * Allow unprivileged RT tasks to decrease priority: |
| 4210 | */ |
| 4211 | if (!capable(CAP_SYS_NICE)) { |
Ingo Molnar | e05606d | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4212 | if (rt_policy(policy)) { |
Oleg Nesterov | 8dc3e90 | 2006-09-29 02:00:50 -0700 | [diff] [blame] | 4213 | unsigned long rlim_rtprio; |
Oleg Nesterov | 5fe1d75 | 2006-09-29 02:00:48 -0700 | [diff] [blame] | 4214 | |
Oleg Nesterov | 8dc3e90 | 2006-09-29 02:00:50 -0700 | [diff] [blame] | 4215 | if (!lock_task_sighand(p, &flags)) |
| 4216 | return -ESRCH; |
| 4217 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; |
| 4218 | unlock_task_sighand(p, &flags); |
Oleg Nesterov | 5fe1d75 | 2006-09-29 02:00:48 -0700 | [diff] [blame] | 4219 | |
Oleg Nesterov | 8dc3e90 | 2006-09-29 02:00:50 -0700 | [diff] [blame] | 4220 | /* can't set/change the rt policy */ |
| 4221 | if (policy != p->policy && !rlim_rtprio) |
| 4222 | return -EPERM; |
| 4223 | |
| 4224 | /* can't increase priority */ |
| 4225 | if (param->sched_priority > p->rt_priority && |
| 4226 | param->sched_priority > rlim_rtprio) |
| 4227 | return -EPERM; |
| 4228 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4229 | /* |
| 4230 | * Like positive nice levels, dont allow tasks to |
| 4231 | * move out of SCHED_IDLE either: |
| 4232 | */ |
| 4233 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) |
| 4234 | return -EPERM; |
Oleg Nesterov | 8dc3e90 | 2006-09-29 02:00:50 -0700 | [diff] [blame] | 4235 | |
Olivier Croquette | 37e4ab3 | 2005-06-25 14:57:32 -0700 | [diff] [blame] | 4236 | /* can't change other user's priorities */ |
| 4237 | if ((current->euid != p->euid) && |
| 4238 | (current->euid != p->uid)) |
| 4239 | return -EPERM; |
| 4240 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4241 | |
| 4242 | retval = security_task_setscheduler(p, policy, param); |
| 4243 | if (retval) |
| 4244 | return retval; |
| 4245 | /* |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 4246 | * make sure no PI-waiters arrive (or leave) while we are |
| 4247 | * changing the priority of the task: |
| 4248 | */ |
| 4249 | spin_lock_irqsave(&p->pi_lock, flags); |
| 4250 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4251 | * To be able to change p->policy safely, the apropriate |
| 4252 | * runqueue lock must be held. |
| 4253 | */ |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 4254 | rq = __task_rq_lock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4255 | /* recheck policy now with rq lock held */ |
| 4256 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { |
| 4257 | policy = oldpolicy = -1; |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 4258 | __task_rq_unlock(rq); |
| 4259 | spin_unlock_irqrestore(&p->pi_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4260 | goto recheck; |
| 4261 | } |
Ingo Molnar | 2daa357 | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 4262 | update_rq_clock(rq); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4263 | on_rq = p->se.on_rq; |
Ingo Molnar | 2daa357 | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 4264 | if (on_rq) |
Ingo Molnar | 2e1cb74 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 4265 | deactivate_task(rq, p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4266 | oldprio = p->prio; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4267 | __setscheduler(rq, p, policy, param->sched_priority); |
| 4268 | if (on_rq) { |
| 4269 | activate_task(rq, p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4270 | /* |
| 4271 | * Reschedule if we are currently running on this runqueue and |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 4272 | * our priority decreased, or if we are not currently running on |
| 4273 | * this runqueue and our priority is higher than the current's |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4274 | */ |
Andrew Morton | d5f9f94 | 2007-05-08 20:27:06 -0700 | [diff] [blame] | 4275 | if (task_running(rq, p)) { |
| 4276 | if (p->prio > oldprio) |
| 4277 | resched_task(rq->curr); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4278 | } else { |
| 4279 | check_preempt_curr(rq, p); |
| 4280 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4281 | } |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 4282 | __task_rq_unlock(rq); |
| 4283 | spin_unlock_irqrestore(&p->pi_lock, flags); |
| 4284 | |
Thomas Gleixner | 95e02ca | 2006-06-27 02:55:02 -0700 | [diff] [blame] | 4285 | rt_mutex_adjust_pi(p); |
| 4286 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4287 | return 0; |
| 4288 | } |
| 4289 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
| 4290 | |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 4291 | static int |
| 4292 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4293 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4294 | struct sched_param lparam; |
| 4295 | struct task_struct *p; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4296 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4297 | |
| 4298 | if (!param || pid < 0) |
| 4299 | return -EINVAL; |
| 4300 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) |
| 4301 | return -EFAULT; |
Oleg Nesterov | 5fe1d75 | 2006-09-29 02:00:48 -0700 | [diff] [blame] | 4302 | |
| 4303 | rcu_read_lock(); |
| 4304 | retval = -ESRCH; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4305 | p = find_process_by_pid(pid); |
Oleg Nesterov | 5fe1d75 | 2006-09-29 02:00:48 -0700 | [diff] [blame] | 4306 | if (p != NULL) |
| 4307 | retval = sched_setscheduler(p, policy, &lparam); |
| 4308 | rcu_read_unlock(); |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4309 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4310 | return retval; |
| 4311 | } |
| 4312 | |
| 4313 | /** |
| 4314 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority |
| 4315 | * @pid: the pid in question. |
| 4316 | * @policy: new policy. |
| 4317 | * @param: structure containing the new RT priority. |
| 4318 | */ |
| 4319 | asmlinkage long sys_sched_setscheduler(pid_t pid, int policy, |
| 4320 | struct sched_param __user *param) |
| 4321 | { |
Jason Baron | c21761f | 2006-01-18 17:43:03 -0800 | [diff] [blame] | 4322 | /* negative values for policy are not valid */ |
| 4323 | if (policy < 0) |
| 4324 | return -EINVAL; |
| 4325 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4326 | return do_sched_setscheduler(pid, policy, param); |
| 4327 | } |
| 4328 | |
| 4329 | /** |
| 4330 | * sys_sched_setparam - set/change the RT priority of a thread |
| 4331 | * @pid: the pid in question. |
| 4332 | * @param: structure containing the new RT priority. |
| 4333 | */ |
| 4334 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) |
| 4335 | { |
| 4336 | return do_sched_setscheduler(pid, -1, param); |
| 4337 | } |
| 4338 | |
| 4339 | /** |
| 4340 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread |
| 4341 | * @pid: the pid in question. |
| 4342 | */ |
| 4343 | asmlinkage long sys_sched_getscheduler(pid_t pid) |
| 4344 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4345 | struct task_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4346 | int retval = -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4347 | |
| 4348 | if (pid < 0) |
| 4349 | goto out_nounlock; |
| 4350 | |
| 4351 | retval = -ESRCH; |
| 4352 | read_lock(&tasklist_lock); |
| 4353 | p = find_process_by_pid(pid); |
| 4354 | if (p) { |
| 4355 | retval = security_task_getscheduler(p); |
| 4356 | if (!retval) |
| 4357 | retval = p->policy; |
| 4358 | } |
| 4359 | read_unlock(&tasklist_lock); |
| 4360 | |
| 4361 | out_nounlock: |
| 4362 | return retval; |
| 4363 | } |
| 4364 | |
| 4365 | /** |
| 4366 | * sys_sched_getscheduler - get the RT priority of a thread |
| 4367 | * @pid: the pid in question. |
| 4368 | * @param: structure containing the RT priority. |
| 4369 | */ |
| 4370 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) |
| 4371 | { |
| 4372 | struct sched_param lp; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4373 | struct task_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4374 | int retval = -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4375 | |
| 4376 | if (!param || pid < 0) |
| 4377 | goto out_nounlock; |
| 4378 | |
| 4379 | read_lock(&tasklist_lock); |
| 4380 | p = find_process_by_pid(pid); |
| 4381 | retval = -ESRCH; |
| 4382 | if (!p) |
| 4383 | goto out_unlock; |
| 4384 | |
| 4385 | retval = security_task_getscheduler(p); |
| 4386 | if (retval) |
| 4387 | goto out_unlock; |
| 4388 | |
| 4389 | lp.sched_priority = p->rt_priority; |
| 4390 | read_unlock(&tasklist_lock); |
| 4391 | |
| 4392 | /* |
| 4393 | * This one might sleep, we cannot do it with a spinlock held ... |
| 4394 | */ |
| 4395 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; |
| 4396 | |
| 4397 | out_nounlock: |
| 4398 | return retval; |
| 4399 | |
| 4400 | out_unlock: |
| 4401 | read_unlock(&tasklist_lock); |
| 4402 | return retval; |
| 4403 | } |
| 4404 | |
| 4405 | long sched_setaffinity(pid_t pid, cpumask_t new_mask) |
| 4406 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4407 | cpumask_t cpus_allowed; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4408 | struct task_struct *p; |
| 4409 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4410 | |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 4411 | mutex_lock(&sched_hotcpu_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4412 | read_lock(&tasklist_lock); |
| 4413 | |
| 4414 | p = find_process_by_pid(pid); |
| 4415 | if (!p) { |
| 4416 | read_unlock(&tasklist_lock); |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 4417 | mutex_unlock(&sched_hotcpu_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4418 | return -ESRCH; |
| 4419 | } |
| 4420 | |
| 4421 | /* |
| 4422 | * It is not safe to call set_cpus_allowed with the |
| 4423 | * tasklist_lock held. We will bump the task_struct's |
| 4424 | * usage count and then drop tasklist_lock. |
| 4425 | */ |
| 4426 | get_task_struct(p); |
| 4427 | read_unlock(&tasklist_lock); |
| 4428 | |
| 4429 | retval = -EPERM; |
| 4430 | if ((current->euid != p->euid) && (current->euid != p->uid) && |
| 4431 | !capable(CAP_SYS_NICE)) |
| 4432 | goto out_unlock; |
| 4433 | |
David Quigley | e7834f8 | 2006-06-23 02:03:59 -0700 | [diff] [blame] | 4434 | retval = security_task_setscheduler(p, 0, NULL); |
| 4435 | if (retval) |
| 4436 | goto out_unlock; |
| 4437 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4438 | cpus_allowed = cpuset_cpus_allowed(p); |
| 4439 | cpus_and(new_mask, new_mask, cpus_allowed); |
| 4440 | retval = set_cpus_allowed(p, new_mask); |
| 4441 | |
| 4442 | out_unlock: |
| 4443 | put_task_struct(p); |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 4444 | mutex_unlock(&sched_hotcpu_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4445 | return retval; |
| 4446 | } |
| 4447 | |
| 4448 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, |
| 4449 | cpumask_t *new_mask) |
| 4450 | { |
| 4451 | if (len < sizeof(cpumask_t)) { |
| 4452 | memset(new_mask, 0, sizeof(cpumask_t)); |
| 4453 | } else if (len > sizeof(cpumask_t)) { |
| 4454 | len = sizeof(cpumask_t); |
| 4455 | } |
| 4456 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
| 4457 | } |
| 4458 | |
| 4459 | /** |
| 4460 | * sys_sched_setaffinity - set the cpu affinity of a process |
| 4461 | * @pid: pid of the process |
| 4462 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr |
| 4463 | * @user_mask_ptr: user-space pointer to the new cpu mask |
| 4464 | */ |
| 4465 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, |
| 4466 | unsigned long __user *user_mask_ptr) |
| 4467 | { |
| 4468 | cpumask_t new_mask; |
| 4469 | int retval; |
| 4470 | |
| 4471 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); |
| 4472 | if (retval) |
| 4473 | return retval; |
| 4474 | |
| 4475 | return sched_setaffinity(pid, new_mask); |
| 4476 | } |
| 4477 | |
| 4478 | /* |
| 4479 | * Represents all cpu's present in the system |
| 4480 | * In systems capable of hotplug, this map could dynamically grow |
| 4481 | * as new cpu's are detected in the system via any platform specific |
| 4482 | * method, such as ACPI for e.g. |
| 4483 | */ |
| 4484 | |
Andi Kleen | 4cef0c6 | 2006-01-11 22:44:57 +0100 | [diff] [blame] | 4485 | cpumask_t cpu_present_map __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4486 | EXPORT_SYMBOL(cpu_present_map); |
| 4487 | |
| 4488 | #ifndef CONFIG_SMP |
Andi Kleen | 4cef0c6 | 2006-01-11 22:44:57 +0100 | [diff] [blame] | 4489 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
Greg Banks | e16b38f | 2006-10-02 02:17:40 -0700 | [diff] [blame] | 4490 | EXPORT_SYMBOL(cpu_online_map); |
| 4491 | |
Andi Kleen | 4cef0c6 | 2006-01-11 22:44:57 +0100 | [diff] [blame] | 4492 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
Greg Banks | e16b38f | 2006-10-02 02:17:40 -0700 | [diff] [blame] | 4493 | EXPORT_SYMBOL(cpu_possible_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4494 | #endif |
| 4495 | |
| 4496 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
| 4497 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4498 | struct task_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4499 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4500 | |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 4501 | mutex_lock(&sched_hotcpu_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4502 | read_lock(&tasklist_lock); |
| 4503 | |
| 4504 | retval = -ESRCH; |
| 4505 | p = find_process_by_pid(pid); |
| 4506 | if (!p) |
| 4507 | goto out_unlock; |
| 4508 | |
David Quigley | e7834f8 | 2006-06-23 02:03:59 -0700 | [diff] [blame] | 4509 | retval = security_task_getscheduler(p); |
| 4510 | if (retval) |
| 4511 | goto out_unlock; |
| 4512 | |
Jack Steiner | 2f7016d | 2006-02-01 03:05:18 -0800 | [diff] [blame] | 4513 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4514 | |
| 4515 | out_unlock: |
| 4516 | read_unlock(&tasklist_lock); |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 4517 | mutex_unlock(&sched_hotcpu_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4518 | |
Ulrich Drepper | 9531b62 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 4519 | return retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4520 | } |
| 4521 | |
| 4522 | /** |
| 4523 | * sys_sched_getaffinity - get the cpu affinity of a process |
| 4524 | * @pid: pid of the process |
| 4525 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr |
| 4526 | * @user_mask_ptr: user-space pointer to hold the current cpu mask |
| 4527 | */ |
| 4528 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, |
| 4529 | unsigned long __user *user_mask_ptr) |
| 4530 | { |
| 4531 | int ret; |
| 4532 | cpumask_t mask; |
| 4533 | |
| 4534 | if (len < sizeof(cpumask_t)) |
| 4535 | return -EINVAL; |
| 4536 | |
| 4537 | ret = sched_getaffinity(pid, &mask); |
| 4538 | if (ret < 0) |
| 4539 | return ret; |
| 4540 | |
| 4541 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) |
| 4542 | return -EFAULT; |
| 4543 | |
| 4544 | return sizeof(cpumask_t); |
| 4545 | } |
| 4546 | |
| 4547 | /** |
| 4548 | * sys_sched_yield - yield the current processor to other threads. |
| 4549 | * |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4550 | * This function yields the current CPU to other tasks. If there are no |
| 4551 | * other threads running on this CPU then this function will return. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4552 | */ |
| 4553 | asmlinkage long sys_sched_yield(void) |
| 4554 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4555 | struct rq *rq = this_rq_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4556 | |
| 4557 | schedstat_inc(rq, yld_cnt); |
Ingo Molnar | 1799e35 | 2007-09-19 23:34:46 +0200 | [diff] [blame] | 4558 | current->sched_class->yield_task(rq, current); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4559 | |
| 4560 | /* |
| 4561 | * Since we are going to call schedule() anyway, there's |
| 4562 | * no need to preempt or enable interrupts: |
| 4563 | */ |
| 4564 | __release(rq->lock); |
Ingo Molnar | 8a25d5d | 2006-07-03 00:24:54 -0700 | [diff] [blame] | 4565 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4566 | _raw_spin_unlock(&rq->lock); |
| 4567 | preempt_enable_no_resched(); |
| 4568 | |
| 4569 | schedule(); |
| 4570 | |
| 4571 | return 0; |
| 4572 | } |
| 4573 | |
Andrew Morton | e7b3840 | 2006-06-30 01:56:00 -0700 | [diff] [blame] | 4574 | static void __cond_resched(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4575 | { |
Ingo Molnar | 8e0a43d | 2006-06-23 02:05:23 -0700 | [diff] [blame] | 4576 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
| 4577 | __might_sleep(__FILE__, __LINE__); |
| 4578 | #endif |
Ingo Molnar | 5bbcfd9 | 2005-07-07 17:57:04 -0700 | [diff] [blame] | 4579 | /* |
| 4580 | * The BKS might be reacquired before we have dropped |
| 4581 | * PREEMPT_ACTIVE, which could trigger a second |
| 4582 | * cond_resched() call. |
| 4583 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4584 | do { |
| 4585 | add_preempt_count(PREEMPT_ACTIVE); |
| 4586 | schedule(); |
| 4587 | sub_preempt_count(PREEMPT_ACTIVE); |
| 4588 | } while (need_resched()); |
| 4589 | } |
| 4590 | |
| 4591 | int __sched cond_resched(void) |
| 4592 | { |
Ingo Molnar | 9414232 | 2006-12-29 16:48:13 -0800 | [diff] [blame] | 4593 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
| 4594 | system_state == SYSTEM_RUNNING) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4595 | __cond_resched(); |
| 4596 | return 1; |
| 4597 | } |
| 4598 | return 0; |
| 4599 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4600 | EXPORT_SYMBOL(cond_resched); |
| 4601 | |
| 4602 | /* |
| 4603 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, |
| 4604 | * call schedule, and on return reacquire the lock. |
| 4605 | * |
| 4606 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
| 4607 | * operations here to prevent schedule() from being called twice (once via |
| 4608 | * spin_unlock(), once by hand). |
| 4609 | */ |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 4610 | int cond_resched_lock(spinlock_t *lock) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4611 | { |
Jan Kara | 6df3cec | 2005-06-13 15:52:32 -0700 | [diff] [blame] | 4612 | int ret = 0; |
| 4613 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4614 | if (need_lockbreak(lock)) { |
| 4615 | spin_unlock(lock); |
| 4616 | cpu_relax(); |
Jan Kara | 6df3cec | 2005-06-13 15:52:32 -0700 | [diff] [blame] | 4617 | ret = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4618 | spin_lock(lock); |
| 4619 | } |
Ingo Molnar | 9414232 | 2006-12-29 16:48:13 -0800 | [diff] [blame] | 4620 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
Ingo Molnar | 8a25d5d | 2006-07-03 00:24:54 -0700 | [diff] [blame] | 4621 | spin_release(&lock->dep_map, 1, _THIS_IP_); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4622 | _raw_spin_unlock(lock); |
| 4623 | preempt_enable_no_resched(); |
| 4624 | __cond_resched(); |
Jan Kara | 6df3cec | 2005-06-13 15:52:32 -0700 | [diff] [blame] | 4625 | ret = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4626 | spin_lock(lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4627 | } |
Jan Kara | 6df3cec | 2005-06-13 15:52:32 -0700 | [diff] [blame] | 4628 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4629 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4630 | EXPORT_SYMBOL(cond_resched_lock); |
| 4631 | |
| 4632 | int __sched cond_resched_softirq(void) |
| 4633 | { |
| 4634 | BUG_ON(!in_softirq()); |
| 4635 | |
Ingo Molnar | 9414232 | 2006-12-29 16:48:13 -0800 | [diff] [blame] | 4636 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
Thomas Gleixner | 98d82567 | 2007-05-23 13:58:18 -0700 | [diff] [blame] | 4637 | local_bh_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4638 | __cond_resched(); |
| 4639 | local_bh_disable(); |
| 4640 | return 1; |
| 4641 | } |
| 4642 | return 0; |
| 4643 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4644 | EXPORT_SYMBOL(cond_resched_softirq); |
| 4645 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4646 | /** |
| 4647 | * yield - yield the current processor to other threads. |
| 4648 | * |
Robert P. J. Day | 72fd4a3 | 2007-02-10 01:45:59 -0800 | [diff] [blame] | 4649 | * This is a shortcut for kernel-space yielding - it marks the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4650 | * thread runnable and calls sys_sched_yield(). |
| 4651 | */ |
| 4652 | void __sched yield(void) |
| 4653 | { |
| 4654 | set_current_state(TASK_RUNNING); |
| 4655 | sys_sched_yield(); |
| 4656 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4657 | EXPORT_SYMBOL(yield); |
| 4658 | |
| 4659 | /* |
| 4660 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
| 4661 | * that process accounting knows that this is a task in IO wait state. |
| 4662 | * |
| 4663 | * But don't do that if it is a deliberate, throttling IO wait (this task |
| 4664 | * has set its backing_dev_info: the queue against which it should throttle) |
| 4665 | */ |
| 4666 | void __sched io_schedule(void) |
| 4667 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4668 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4669 | |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 4670 | delayacct_blkio_start(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4671 | atomic_inc(&rq->nr_iowait); |
| 4672 | schedule(); |
| 4673 | atomic_dec(&rq->nr_iowait); |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 4674 | delayacct_blkio_end(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4675 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4676 | EXPORT_SYMBOL(io_schedule); |
| 4677 | |
| 4678 | long __sched io_schedule_timeout(long timeout) |
| 4679 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4680 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4681 | long ret; |
| 4682 | |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 4683 | delayacct_blkio_start(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4684 | atomic_inc(&rq->nr_iowait); |
| 4685 | ret = schedule_timeout(timeout); |
| 4686 | atomic_dec(&rq->nr_iowait); |
Shailabh Nagar | 0ff9224 | 2006-07-14 00:24:37 -0700 | [diff] [blame] | 4687 | delayacct_blkio_end(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4688 | return ret; |
| 4689 | } |
| 4690 | |
| 4691 | /** |
| 4692 | * sys_sched_get_priority_max - return maximum RT priority. |
| 4693 | * @policy: scheduling class. |
| 4694 | * |
| 4695 | * this syscall returns the maximum rt_priority that can be used |
| 4696 | * by a given scheduling class. |
| 4697 | */ |
| 4698 | asmlinkage long sys_sched_get_priority_max(int policy) |
| 4699 | { |
| 4700 | int ret = -EINVAL; |
| 4701 | |
| 4702 | switch (policy) { |
| 4703 | case SCHED_FIFO: |
| 4704 | case SCHED_RR: |
| 4705 | ret = MAX_USER_RT_PRIO-1; |
| 4706 | break; |
| 4707 | case SCHED_NORMAL: |
Ingo Molnar | b0a9499 | 2006-01-14 13:20:41 -0800 | [diff] [blame] | 4708 | case SCHED_BATCH: |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4709 | case SCHED_IDLE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4710 | ret = 0; |
| 4711 | break; |
| 4712 | } |
| 4713 | return ret; |
| 4714 | } |
| 4715 | |
| 4716 | /** |
| 4717 | * sys_sched_get_priority_min - return minimum RT priority. |
| 4718 | * @policy: scheduling class. |
| 4719 | * |
| 4720 | * this syscall returns the minimum rt_priority that can be used |
| 4721 | * by a given scheduling class. |
| 4722 | */ |
| 4723 | asmlinkage long sys_sched_get_priority_min(int policy) |
| 4724 | { |
| 4725 | int ret = -EINVAL; |
| 4726 | |
| 4727 | switch (policy) { |
| 4728 | case SCHED_FIFO: |
| 4729 | case SCHED_RR: |
| 4730 | ret = 1; |
| 4731 | break; |
| 4732 | case SCHED_NORMAL: |
Ingo Molnar | b0a9499 | 2006-01-14 13:20:41 -0800 | [diff] [blame] | 4733 | case SCHED_BATCH: |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4734 | case SCHED_IDLE: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4735 | ret = 0; |
| 4736 | } |
| 4737 | return ret; |
| 4738 | } |
| 4739 | |
| 4740 | /** |
| 4741 | * sys_sched_rr_get_interval - return the default timeslice of a process. |
| 4742 | * @pid: pid of the process. |
| 4743 | * @interval: userspace pointer to the timeslice value. |
| 4744 | * |
| 4745 | * this syscall writes the default timeslice value of a given process |
| 4746 | * into the user-space timespec buffer. A value of '0' means infinity. |
| 4747 | */ |
| 4748 | asmlinkage |
| 4749 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) |
| 4750 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4751 | struct task_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4752 | int retval = -EINVAL; |
| 4753 | struct timespec t; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4754 | |
| 4755 | if (pid < 0) |
| 4756 | goto out_nounlock; |
| 4757 | |
| 4758 | retval = -ESRCH; |
| 4759 | read_lock(&tasklist_lock); |
| 4760 | p = find_process_by_pid(pid); |
| 4761 | if (!p) |
| 4762 | goto out_unlock; |
| 4763 | |
| 4764 | retval = security_task_getscheduler(p); |
| 4765 | if (retval) |
| 4766 | goto out_unlock; |
| 4767 | |
Peter Williams | b78709c | 2006-06-26 16:58:00 +1000 | [diff] [blame] | 4768 | jiffies_to_timespec(p->policy == SCHED_FIFO ? |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4769 | 0 : static_prio_timeslice(p->static_prio), &t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4770 | read_unlock(&tasklist_lock); |
| 4771 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
| 4772 | out_nounlock: |
| 4773 | return retval; |
| 4774 | out_unlock: |
| 4775 | read_unlock(&tasklist_lock); |
| 4776 | return retval; |
| 4777 | } |
| 4778 | |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 4779 | static const char stat_nam[] = "RSDTtZX"; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4780 | |
| 4781 | static void show_task(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4782 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4783 | unsigned long free = 0; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4784 | unsigned state; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4785 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4786 | state = p->state ? __ffs(p->state) + 1 : 0; |
Andreas Mohr | 2ed6e34 | 2006-07-10 04:43:52 -0700 | [diff] [blame] | 4787 | printk("%-13.13s %c", p->comm, |
| 4788 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
Ingo Molnar | 4bd7732 | 2007-07-11 21:21:47 +0200 | [diff] [blame] | 4789 | #if BITS_PER_LONG == 32 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4790 | if (state == TASK_RUNNING) |
Ingo Molnar | 4bd7732 | 2007-07-11 21:21:47 +0200 | [diff] [blame] | 4791 | printk(" running "); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4792 | else |
Ingo Molnar | 4bd7732 | 2007-07-11 21:21:47 +0200 | [diff] [blame] | 4793 | printk(" %08lx ", thread_saved_pc(p)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4794 | #else |
| 4795 | if (state == TASK_RUNNING) |
Ingo Molnar | 4bd7732 | 2007-07-11 21:21:47 +0200 | [diff] [blame] | 4796 | printk(" running task "); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4797 | else |
| 4798 | printk(" %016lx ", thread_saved_pc(p)); |
| 4799 | #endif |
| 4800 | #ifdef CONFIG_DEBUG_STACK_USAGE |
| 4801 | { |
Al Viro | 10ebffd | 2005-11-13 16:06:56 -0800 | [diff] [blame] | 4802 | unsigned long *n = end_of_stack(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4803 | while (!*n) |
| 4804 | n++; |
Al Viro | 10ebffd | 2005-11-13 16:06:56 -0800 | [diff] [blame] | 4805 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4806 | } |
| 4807 | #endif |
Ingo Molnar | 4bd7732 | 2007-07-11 21:21:47 +0200 | [diff] [blame] | 4808 | printk("%5lu %5d %6d\n", free, p->pid, p->parent->pid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4809 | |
| 4810 | if (state != TASK_RUNNING) |
| 4811 | show_stack(p, NULL); |
| 4812 | } |
| 4813 | |
Ingo Molnar | e59e2ae | 2006-12-06 20:35:59 -0800 | [diff] [blame] | 4814 | void show_state_filter(unsigned long state_filter) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4815 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4816 | struct task_struct *g, *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4817 | |
Ingo Molnar | 4bd7732 | 2007-07-11 21:21:47 +0200 | [diff] [blame] | 4818 | #if BITS_PER_LONG == 32 |
| 4819 | printk(KERN_INFO |
| 4820 | " task PC stack pid father\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4821 | #else |
Ingo Molnar | 4bd7732 | 2007-07-11 21:21:47 +0200 | [diff] [blame] | 4822 | printk(KERN_INFO |
| 4823 | " task PC stack pid father\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4824 | #endif |
| 4825 | read_lock(&tasklist_lock); |
| 4826 | do_each_thread(g, p) { |
| 4827 | /* |
| 4828 | * reset the NMI-timeout, listing all files on a slow |
| 4829 | * console might take alot of time: |
| 4830 | */ |
| 4831 | touch_nmi_watchdog(); |
Ingo Molnar | 39bc89f | 2007-04-25 20:50:03 -0700 | [diff] [blame] | 4832 | if (!state_filter || (p->state & state_filter)) |
Ingo Molnar | e59e2ae | 2006-12-06 20:35:59 -0800 | [diff] [blame] | 4833 | show_task(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4834 | } while_each_thread(g, p); |
| 4835 | |
Jeremy Fitzhardinge | 04c9167 | 2007-05-08 00:28:05 -0700 | [diff] [blame] | 4836 | touch_all_softlockup_watchdogs(); |
| 4837 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4838 | #ifdef CONFIG_SCHED_DEBUG |
| 4839 | sysrq_sched_debug_show(); |
| 4840 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4841 | read_unlock(&tasklist_lock); |
Ingo Molnar | e59e2ae | 2006-12-06 20:35:59 -0800 | [diff] [blame] | 4842 | /* |
| 4843 | * Only show locks if all tasks are dumped: |
| 4844 | */ |
| 4845 | if (state_filter == -1) |
| 4846 | debug_show_all_locks(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4847 | } |
| 4848 | |
Ingo Molnar | 1df2105 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 4849 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
| 4850 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4851 | idle->sched_class = &idle_sched_class; |
Ingo Molnar | 1df2105 | 2007-07-09 18:51:58 +0200 | [diff] [blame] | 4852 | } |
| 4853 | |
Ingo Molnar | f340c0d | 2005-06-28 16:40:42 +0200 | [diff] [blame] | 4854 | /** |
| 4855 | * init_idle - set up an idle thread for a given CPU |
| 4856 | * @idle: task in question |
| 4857 | * @cpu: cpu the idle task belongs to |
| 4858 | * |
| 4859 | * NOTE: this function does not set the idle thread's NEED_RESCHED |
| 4860 | * flag, to make booting more robust. |
| 4861 | */ |
Nick Piggin | 5c1e176 | 2006-10-03 01:14:04 -0700 | [diff] [blame] | 4862 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4863 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4864 | struct rq *rq = cpu_rq(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4865 | unsigned long flags; |
| 4866 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4867 | __sched_fork(idle); |
| 4868 | idle->se.exec_start = sched_clock(); |
| 4869 | |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 4870 | idle->prio = idle->normal_prio = MAX_PRIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4871 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4872 | __set_task_cpu(idle, cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4873 | |
| 4874 | spin_lock_irqsave(&rq->lock, flags); |
| 4875 | rq->curr = rq->idle = idle; |
Nick Piggin | 4866cde | 2005-06-25 14:57:23 -0700 | [diff] [blame] | 4876 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
| 4877 | idle->oncpu = 1; |
| 4878 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4879 | spin_unlock_irqrestore(&rq->lock, flags); |
| 4880 | |
| 4881 | /* Set the preempt count _outside_ the spinlocks! */ |
| 4882 | #if defined(CONFIG_PREEMPT) && !defined(CONFIG_PREEMPT_BKL) |
Al Viro | a1261f5 | 2005-11-13 16:06:55 -0800 | [diff] [blame] | 4883 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4884 | #else |
Al Viro | a1261f5 | 2005-11-13 16:06:55 -0800 | [diff] [blame] | 4885 | task_thread_info(idle)->preempt_count = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4886 | #endif |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4887 | /* |
| 4888 | * The idle tasks have their own, simple scheduling class: |
| 4889 | */ |
| 4890 | idle->sched_class = &idle_sched_class; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4891 | } |
| 4892 | |
| 4893 | /* |
| 4894 | * In a system that switches off the HZ timer nohz_cpu_mask |
| 4895 | * indicates which cpus entered this state. This is used |
| 4896 | * in the rcu update to wait only for active cpus. For system |
| 4897 | * which do not switch off the HZ timer nohz_cpu_mask should |
| 4898 | * always be CPU_MASK_NONE. |
| 4899 | */ |
| 4900 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; |
| 4901 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4902 | /* |
| 4903 | * Increase the granularity value when there are more CPUs, |
| 4904 | * because with more CPUs the 'effective latency' as visible |
| 4905 | * to users decreases. But the relationship is not linear, |
| 4906 | * so pick a second-best guess by going with the log2 of the |
| 4907 | * number of CPUs. |
| 4908 | * |
| 4909 | * This idea comes from the SD scheduler of Con Kolivas: |
| 4910 | */ |
| 4911 | static inline void sched_init_granularity(void) |
| 4912 | { |
| 4913 | unsigned int factor = 1 + ilog2(num_online_cpus()); |
Peter Zijlstra | 2180508 | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 4914 | const unsigned long limit = 100000000; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4915 | |
Ingo Molnar | 172ac3d | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 4916 | sysctl_sched_min_granularity *= factor; |
| 4917 | if (sysctl_sched_min_granularity > limit) |
| 4918 | sysctl_sched_min_granularity = limit; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4919 | |
Peter Zijlstra | 2180508 | 2007-08-25 18:41:53 +0200 | [diff] [blame] | 4920 | sysctl_sched_latency *= factor; |
| 4921 | if (sysctl_sched_latency > limit) |
| 4922 | sysctl_sched_latency = limit; |
| 4923 | |
Ingo Molnar | 50c4663 | 2007-08-25 22:17:19 +0200 | [diff] [blame] | 4924 | sysctl_sched_runtime_limit = sysctl_sched_latency; |
| 4925 | sysctl_sched_wakeup_granularity = sysctl_sched_min_granularity / 2; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 4926 | } |
| 4927 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4928 | #ifdef CONFIG_SMP |
| 4929 | /* |
| 4930 | * This is how migration works: |
| 4931 | * |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4932 | * 1) we queue a struct migration_req structure in the source CPU's |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4933 | * runqueue and wake up that CPU's migration thread. |
| 4934 | * 2) we down() the locked semaphore => thread blocks. |
| 4935 | * 3) migration thread wakes up (implicitly it forces the migrated |
| 4936 | * thread off the CPU) |
| 4937 | * 4) it gets the migration request and checks whether the migrated |
| 4938 | * task is still in the wrong runqueue. |
| 4939 | * 5) if it's in the wrong runqueue then the migration thread removes |
| 4940 | * it and puts it into the right queue. |
| 4941 | * 6) migration thread up()s the semaphore. |
| 4942 | * 7) we wake up and the migration is done. |
| 4943 | */ |
| 4944 | |
| 4945 | /* |
| 4946 | * Change a given task's CPU affinity. Migrate the thread to a |
| 4947 | * proper CPU and schedule it away if the CPU it's executing on |
| 4948 | * is removed from the allowed bitmask. |
| 4949 | * |
| 4950 | * NOTE: the caller must have a valid reference to the task, the |
| 4951 | * task must not exit() & deallocate itself prematurely. The |
| 4952 | * call is not atomic; no spinlocks may be held. |
| 4953 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 4954 | int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4955 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4956 | struct migration_req req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4957 | unsigned long flags; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 4958 | struct rq *rq; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 4959 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4960 | |
| 4961 | rq = task_rq_lock(p, &flags); |
| 4962 | if (!cpus_intersects(new_mask, cpu_online_map)) { |
| 4963 | ret = -EINVAL; |
| 4964 | goto out; |
| 4965 | } |
| 4966 | |
| 4967 | p->cpus_allowed = new_mask; |
| 4968 | /* Can the task run on the task's current CPU? If so, we're done */ |
| 4969 | if (cpu_isset(task_cpu(p), new_mask)) |
| 4970 | goto out; |
| 4971 | |
| 4972 | if (migrate_task(p, any_online_cpu(new_mask), &req)) { |
| 4973 | /* Need help from migration thread: drop lock and wait. */ |
| 4974 | task_rq_unlock(rq, &flags); |
| 4975 | wake_up_process(rq->migration_thread); |
| 4976 | wait_for_completion(&req.done); |
| 4977 | tlb_migrate_finish(p->mm); |
| 4978 | return 0; |
| 4979 | } |
| 4980 | out: |
| 4981 | task_rq_unlock(rq, &flags); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 4982 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4983 | return ret; |
| 4984 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4985 | EXPORT_SYMBOL_GPL(set_cpus_allowed); |
| 4986 | |
| 4987 | /* |
| 4988 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
| 4989 | * this because either it can't run here any more (set_cpus_allowed() |
| 4990 | * away from this CPU, or CPU going down), or because we're |
| 4991 | * attempting to rebalance this task on exec (sched_exec). |
| 4992 | * |
| 4993 | * So we race with normal scheduler movements, but that's OK, as long |
| 4994 | * as the task is no longer on this CPU. |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 4995 | * |
| 4996 | * Returns non-zero if task was successfully migrated. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4997 | */ |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 4998 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4999 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5000 | struct rq *rq_dest, *rq_src; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5001 | int ret = 0, on_rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5002 | |
| 5003 | if (unlikely(cpu_is_offline(dest_cpu))) |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 5004 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5005 | |
| 5006 | rq_src = cpu_rq(src_cpu); |
| 5007 | rq_dest = cpu_rq(dest_cpu); |
| 5008 | |
| 5009 | double_rq_lock(rq_src, rq_dest); |
| 5010 | /* Already moved. */ |
| 5011 | if (task_cpu(p) != src_cpu) |
| 5012 | goto out; |
| 5013 | /* Affinity changed (again). */ |
| 5014 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) |
| 5015 | goto out; |
| 5016 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5017 | on_rq = p->se.on_rq; |
Ingo Molnar | 6e82a3b | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 5018 | if (on_rq) |
Ingo Molnar | 2e1cb74 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 5019 | deactivate_task(rq_src, p, 0); |
Ingo Molnar | 6e82a3b | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 5020 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5021 | set_task_cpu(p, dest_cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5022 | if (on_rq) { |
| 5023 | activate_task(rq_dest, p, 0); |
| 5024 | check_preempt_curr(rq_dest, p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5025 | } |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 5026 | ret = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5027 | out: |
| 5028 | double_rq_unlock(rq_src, rq_dest); |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 5029 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5030 | } |
| 5031 | |
| 5032 | /* |
| 5033 | * migration_thread - this is a highprio system thread that performs |
| 5034 | * thread migration by bumping thread off CPU then 'pushing' onto |
| 5035 | * another runqueue. |
| 5036 | */ |
Ingo Molnar | 95cdf3b | 2005-09-10 00:26:11 -0700 | [diff] [blame] | 5037 | static int migration_thread(void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5038 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5039 | int cpu = (long)data; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5040 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5041 | |
| 5042 | rq = cpu_rq(cpu); |
| 5043 | BUG_ON(rq->migration_thread != current); |
| 5044 | |
| 5045 | set_current_state(TASK_INTERRUPTIBLE); |
| 5046 | while (!kthread_should_stop()) { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5047 | struct migration_req *req; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5048 | struct list_head *head; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5049 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5050 | spin_lock_irq(&rq->lock); |
| 5051 | |
| 5052 | if (cpu_is_offline(cpu)) { |
| 5053 | spin_unlock_irq(&rq->lock); |
| 5054 | goto wait_to_die; |
| 5055 | } |
| 5056 | |
| 5057 | if (rq->active_balance) { |
| 5058 | active_load_balance(rq, cpu); |
| 5059 | rq->active_balance = 0; |
| 5060 | } |
| 5061 | |
| 5062 | head = &rq->migration_queue; |
| 5063 | |
| 5064 | if (list_empty(head)) { |
| 5065 | spin_unlock_irq(&rq->lock); |
| 5066 | schedule(); |
| 5067 | set_current_state(TASK_INTERRUPTIBLE); |
| 5068 | continue; |
| 5069 | } |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5070 | req = list_entry(head->next, struct migration_req, list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5071 | list_del_init(head->next); |
| 5072 | |
Nick Piggin | 674311d | 2005-06-25 14:57:27 -0700 | [diff] [blame] | 5073 | spin_unlock(&rq->lock); |
| 5074 | __migrate_task(req->task, cpu, req->dest_cpu); |
| 5075 | local_irq_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5076 | |
| 5077 | complete(&req->done); |
| 5078 | } |
| 5079 | __set_current_state(TASK_RUNNING); |
| 5080 | return 0; |
| 5081 | |
| 5082 | wait_to_die: |
| 5083 | /* Wait for kthread_stop */ |
| 5084 | set_current_state(TASK_INTERRUPTIBLE); |
| 5085 | while (!kthread_should_stop()) { |
| 5086 | schedule(); |
| 5087 | set_current_state(TASK_INTERRUPTIBLE); |
| 5088 | } |
| 5089 | __set_current_state(TASK_RUNNING); |
| 5090 | return 0; |
| 5091 | } |
| 5092 | |
| 5093 | #ifdef CONFIG_HOTPLUG_CPU |
Kirill Korotaev | 054b910 | 2006-12-10 02:20:11 -0800 | [diff] [blame] | 5094 | /* |
| 5095 | * Figure out where task on dead CPU should go, use force if neccessary. |
| 5096 | * NOTE: interrupts should be disabled by the caller |
| 5097 | */ |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5098 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5099 | { |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 5100 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5101 | cpumask_t mask; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5102 | struct rq *rq; |
| 5103 | int dest_cpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5104 | |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 5105 | restart: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5106 | /* On same node? */ |
| 5107 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5108 | cpus_and(mask, mask, p->cpus_allowed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5109 | dest_cpu = any_online_cpu(mask); |
| 5110 | |
| 5111 | /* On any allowed CPU? */ |
| 5112 | if (dest_cpu == NR_CPUS) |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5113 | dest_cpu = any_online_cpu(p->cpus_allowed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5114 | |
| 5115 | /* No more Mr. Nice Guy. */ |
| 5116 | if (dest_cpu == NR_CPUS) { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5117 | rq = task_rq_lock(p, &flags); |
| 5118 | cpus_setall(p->cpus_allowed); |
| 5119 | dest_cpu = any_online_cpu(p->cpus_allowed); |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 5120 | task_rq_unlock(rq, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5121 | |
| 5122 | /* |
| 5123 | * Don't tell them about moving exiting tasks or |
| 5124 | * kernel threads (both mm NULL), since they never |
| 5125 | * leave kernel. |
| 5126 | */ |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5127 | if (p->mm && printk_ratelimit()) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5128 | printk(KERN_INFO "process %d (%s) no " |
| 5129 | "longer affine to cpu%d\n", |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5130 | p->pid, p->comm, dead_cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5131 | } |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5132 | if (!__migrate_task(p, dead_cpu, dest_cpu)) |
Kirill Korotaev | efc3081 | 2006-06-27 02:54:32 -0700 | [diff] [blame] | 5133 | goto restart; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5134 | } |
| 5135 | |
| 5136 | /* |
| 5137 | * While a dead CPU has no uninterruptible tasks queued at this point, |
| 5138 | * it might still have a nonzero ->nr_uninterruptible counter, because |
| 5139 | * for performance reasons the counter is not stricly tracking tasks to |
| 5140 | * their home CPUs. So we just add the counter to another CPU's counter, |
| 5141 | * to keep the global sum constant after CPU-down: |
| 5142 | */ |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5143 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5144 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5145 | struct rq *rq_dest = cpu_rq(any_online_cpu(CPU_MASK_ALL)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5146 | unsigned long flags; |
| 5147 | |
| 5148 | local_irq_save(flags); |
| 5149 | double_rq_lock(rq_src, rq_dest); |
| 5150 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
| 5151 | rq_src->nr_uninterruptible = 0; |
| 5152 | double_rq_unlock(rq_src, rq_dest); |
| 5153 | local_irq_restore(flags); |
| 5154 | } |
| 5155 | |
| 5156 | /* Run through task list and migrate tasks from the dead cpu. */ |
| 5157 | static void migrate_live_tasks(int src_cpu) |
| 5158 | { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5159 | struct task_struct *p, *t; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5160 | |
| 5161 | write_lock_irq(&tasklist_lock); |
| 5162 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5163 | do_each_thread(t, p) { |
| 5164 | if (p == current) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5165 | continue; |
| 5166 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5167 | if (task_cpu(p) == src_cpu) |
| 5168 | move_task_off_dead_cpu(src_cpu, p); |
| 5169 | } while_each_thread(t, p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5170 | |
| 5171 | write_unlock_irq(&tasklist_lock); |
| 5172 | } |
| 5173 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5174 | /* |
| 5175 | * Schedules idle task to be the next runnable task on current CPU. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5176 | * It does so by boosting its priority to highest possible and adding it to |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5177 | * the _front_ of the runqueue. Used by CPU offline code. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5178 | */ |
| 5179 | void sched_idle_next(void) |
| 5180 | { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5181 | int this_cpu = smp_processor_id(); |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5182 | struct rq *rq = cpu_rq(this_cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5183 | struct task_struct *p = rq->idle; |
| 5184 | unsigned long flags; |
| 5185 | |
| 5186 | /* cpu has to be offline */ |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5187 | BUG_ON(cpu_online(this_cpu)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5188 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5189 | /* |
| 5190 | * Strictly not necessary since rest of the CPUs are stopped by now |
| 5191 | * and interrupts disabled on the current cpu. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5192 | */ |
| 5193 | spin_lock_irqsave(&rq->lock, flags); |
| 5194 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5195 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5196 | |
| 5197 | /* Add idle task to the _front_ of its priority queue: */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5198 | activate_idle_task(p, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5199 | |
| 5200 | spin_unlock_irqrestore(&rq->lock, flags); |
| 5201 | } |
| 5202 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5203 | /* |
| 5204 | * Ensures that the idle task is using init_mm right before its cpu goes |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5205 | * offline. |
| 5206 | */ |
| 5207 | void idle_task_exit(void) |
| 5208 | { |
| 5209 | struct mm_struct *mm = current->active_mm; |
| 5210 | |
| 5211 | BUG_ON(cpu_online(smp_processor_id())); |
| 5212 | |
| 5213 | if (mm != &init_mm) |
| 5214 | switch_mm(mm, &init_mm, current); |
| 5215 | mmdrop(mm); |
| 5216 | } |
| 5217 | |
Kirill Korotaev | 054b910 | 2006-12-10 02:20:11 -0800 | [diff] [blame] | 5218 | /* called under rq->lock with disabled interrupts */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 5219 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5220 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5221 | struct rq *rq = cpu_rq(dead_cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5222 | |
| 5223 | /* Must be exiting, otherwise would be on tasklist. */ |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5224 | BUG_ON(p->exit_state != EXIT_ZOMBIE && p->exit_state != EXIT_DEAD); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5225 | |
| 5226 | /* Cannot have done final schedule yet: would have vanished. */ |
Oleg Nesterov | c394cc9 | 2006-09-29 02:01:11 -0700 | [diff] [blame] | 5227 | BUG_ON(p->state == TASK_DEAD); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5228 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5229 | get_task_struct(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5230 | |
| 5231 | /* |
| 5232 | * Drop lock around migration; if someone else moves it, |
| 5233 | * that's OK. No task can be added to this CPU, so iteration is |
| 5234 | * fine. |
Kirill Korotaev | 054b910 | 2006-12-10 02:20:11 -0800 | [diff] [blame] | 5235 | * NOTE: interrupts should be left disabled --dev@ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5236 | */ |
Kirill Korotaev | 054b910 | 2006-12-10 02:20:11 -0800 | [diff] [blame] | 5237 | spin_unlock(&rq->lock); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5238 | move_task_off_dead_cpu(dead_cpu, p); |
Kirill Korotaev | 054b910 | 2006-12-10 02:20:11 -0800 | [diff] [blame] | 5239 | spin_lock(&rq->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5240 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5241 | put_task_struct(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5242 | } |
| 5243 | |
| 5244 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ |
| 5245 | static void migrate_dead_tasks(unsigned int dead_cpu) |
| 5246 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5247 | struct rq *rq = cpu_rq(dead_cpu); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5248 | struct task_struct *next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5249 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5250 | for ( ; ; ) { |
| 5251 | if (!rq->nr_running) |
| 5252 | break; |
Ingo Molnar | a8e504d | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 5253 | update_rq_clock(rq); |
Ingo Molnar | ff95f3d | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 5254 | next = pick_next_task(rq, rq->curr); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5255 | if (!next) |
| 5256 | break; |
| 5257 | migrate_dead(dead_cpu, next); |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5258 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5259 | } |
| 5260 | } |
| 5261 | #endif /* CONFIG_HOTPLUG_CPU */ |
| 5262 | |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5263 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
| 5264 | |
| 5265 | static struct ctl_table sd_ctl_dir[] = { |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5266 | { |
| 5267 | .procname = "sched_domain", |
Eric W. Biederman | c57baf1 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 5268 | .mode = 0555, |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5269 | }, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5270 | {0,}, |
| 5271 | }; |
| 5272 | |
| 5273 | static struct ctl_table sd_ctl_root[] = { |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5274 | { |
Eric W. Biederman | c57baf1 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 5275 | .ctl_name = CTL_KERN, |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5276 | .procname = "kernel", |
Eric W. Biederman | c57baf1 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 5277 | .mode = 0555, |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5278 | .child = sd_ctl_dir, |
| 5279 | }, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5280 | {0,}, |
| 5281 | }; |
| 5282 | |
| 5283 | static struct ctl_table *sd_alloc_ctl_entry(int n) |
| 5284 | { |
| 5285 | struct ctl_table *entry = |
| 5286 | kmalloc(n * sizeof(struct ctl_table), GFP_KERNEL); |
| 5287 | |
| 5288 | BUG_ON(!entry); |
| 5289 | memset(entry, 0, n * sizeof(struct ctl_table)); |
| 5290 | |
| 5291 | return entry; |
| 5292 | } |
| 5293 | |
| 5294 | static void |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5295 | set_table_entry(struct ctl_table *entry, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5296 | const char *procname, void *data, int maxlen, |
| 5297 | mode_t mode, proc_handler *proc_handler) |
| 5298 | { |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5299 | entry->procname = procname; |
| 5300 | entry->data = data; |
| 5301 | entry->maxlen = maxlen; |
| 5302 | entry->mode = mode; |
| 5303 | entry->proc_handler = proc_handler; |
| 5304 | } |
| 5305 | |
| 5306 | static struct ctl_table * |
| 5307 | sd_alloc_ctl_domain_table(struct sched_domain *sd) |
| 5308 | { |
| 5309 | struct ctl_table *table = sd_alloc_ctl_entry(14); |
| 5310 | |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5311 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5312 | sizeof(long), 0644, proc_doulongvec_minmax); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5313 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5314 | sizeof(long), 0644, proc_doulongvec_minmax); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5315 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5316 | sizeof(int), 0644, proc_dointvec_minmax); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5317 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5318 | sizeof(int), 0644, proc_dointvec_minmax); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5319 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5320 | sizeof(int), 0644, proc_dointvec_minmax); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5321 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5322 | sizeof(int), 0644, proc_dointvec_minmax); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5323 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5324 | sizeof(int), 0644, proc_dointvec_minmax); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5325 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5326 | sizeof(int), 0644, proc_dointvec_minmax); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5327 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5328 | sizeof(int), 0644, proc_dointvec_minmax); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5329 | set_table_entry(&table[10], "cache_nice_tries", |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5330 | &sd->cache_nice_tries, |
| 5331 | sizeof(int), 0644, proc_dointvec_minmax); |
Alexey Dobriyan | e036185 | 2007-08-09 11:16:46 +0200 | [diff] [blame] | 5332 | set_table_entry(&table[12], "flags", &sd->flags, |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5333 | sizeof(int), 0644, proc_dointvec_minmax); |
| 5334 | |
| 5335 | return table; |
| 5336 | } |
| 5337 | |
| 5338 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
| 5339 | { |
| 5340 | struct ctl_table *entry, *table; |
| 5341 | struct sched_domain *sd; |
| 5342 | int domain_num = 0, i; |
| 5343 | char buf[32]; |
| 5344 | |
| 5345 | for_each_domain(cpu, sd) |
| 5346 | domain_num++; |
| 5347 | entry = table = sd_alloc_ctl_entry(domain_num + 1); |
| 5348 | |
| 5349 | i = 0; |
| 5350 | for_each_domain(cpu, sd) { |
| 5351 | snprintf(buf, 32, "domain%d", i); |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5352 | entry->procname = kstrdup(buf, GFP_KERNEL); |
Eric W. Biederman | c57baf1 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 5353 | entry->mode = 0555; |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5354 | entry->child = sd_alloc_ctl_domain_table(sd); |
| 5355 | entry++; |
| 5356 | i++; |
| 5357 | } |
| 5358 | return table; |
| 5359 | } |
| 5360 | |
| 5361 | static struct ctl_table_header *sd_sysctl_header; |
| 5362 | static void init_sched_domain_sysctl(void) |
| 5363 | { |
| 5364 | int i, cpu_num = num_online_cpus(); |
| 5365 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
| 5366 | char buf[32]; |
| 5367 | |
| 5368 | sd_ctl_dir[0].child = entry; |
| 5369 | |
| 5370 | for (i = 0; i < cpu_num; i++, entry++) { |
| 5371 | snprintf(buf, 32, "cpu%d", i); |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5372 | entry->procname = kstrdup(buf, GFP_KERNEL); |
Eric W. Biederman | c57baf1 | 2007-08-23 15:18:02 +0200 | [diff] [blame] | 5373 | entry->mode = 0555; |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 5374 | entry->child = sd_alloc_ctl_cpu_table(i); |
| 5375 | } |
| 5376 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
| 5377 | } |
| 5378 | #else |
| 5379 | static void init_sched_domain_sysctl(void) |
| 5380 | { |
| 5381 | } |
| 5382 | #endif |
| 5383 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5384 | /* |
| 5385 | * migration_call - callback that gets triggered when a CPU is added. |
| 5386 | * Here we can start up the necessary migration thread for the new CPU. |
| 5387 | */ |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5388 | static int __cpuinit |
| 5389 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5390 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5391 | struct task_struct *p; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5392 | int cpu = (long)hcpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5393 | unsigned long flags; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5394 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5395 | |
| 5396 | switch (action) { |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 5397 | case CPU_LOCK_ACQUIRE: |
| 5398 | mutex_lock(&sched_hotcpu_mutex); |
| 5399 | break; |
| 5400 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5401 | case CPU_UP_PREPARE: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 5402 | case CPU_UP_PREPARE_FROZEN: |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5403 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5404 | if (IS_ERR(p)) |
| 5405 | return NOTIFY_BAD; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5406 | kthread_bind(p, cpu); |
| 5407 | /* Must be high prio: stop_machine expects to yield to it. */ |
| 5408 | rq = task_rq_lock(p, &flags); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5409 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5410 | task_rq_unlock(rq, &flags); |
| 5411 | cpu_rq(cpu)->migration_thread = p; |
| 5412 | break; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5413 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5414 | case CPU_ONLINE: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 5415 | case CPU_ONLINE_FROZEN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5416 | /* Strictly unneccessary, as first user will wake it. */ |
| 5417 | wake_up_process(cpu_rq(cpu)->migration_thread); |
| 5418 | break; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5419 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5420 | #ifdef CONFIG_HOTPLUG_CPU |
| 5421 | case CPU_UP_CANCELED: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 5422 | case CPU_UP_CANCELED_FROZEN: |
Heiko Carstens | fc75cdf | 2006-06-25 05:49:10 -0700 | [diff] [blame] | 5423 | if (!cpu_rq(cpu)->migration_thread) |
| 5424 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5425 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
Heiko Carstens | a4c4af7 | 2005-11-07 00:58:38 -0800 | [diff] [blame] | 5426 | kthread_bind(cpu_rq(cpu)->migration_thread, |
| 5427 | any_online_cpu(cpu_online_map)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5428 | kthread_stop(cpu_rq(cpu)->migration_thread); |
| 5429 | cpu_rq(cpu)->migration_thread = NULL; |
| 5430 | break; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5431 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5432 | case CPU_DEAD: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 5433 | case CPU_DEAD_FROZEN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5434 | migrate_live_tasks(cpu); |
| 5435 | rq = cpu_rq(cpu); |
| 5436 | kthread_stop(rq->migration_thread); |
| 5437 | rq->migration_thread = NULL; |
| 5438 | /* Idle task back to normal (off runqueue, low prio) */ |
| 5439 | rq = task_rq_lock(rq->idle, &flags); |
Ingo Molnar | a8e504d | 2007-08-09 11:16:47 +0200 | [diff] [blame] | 5440 | update_rq_clock(rq); |
Ingo Molnar | 2e1cb74 | 2007-08-09 11:16:49 +0200 | [diff] [blame] | 5441 | deactivate_task(rq, rq->idle, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5442 | rq->idle->static_prio = MAX_PRIO; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 5443 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
| 5444 | rq->idle->sched_class = &idle_sched_class; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5445 | migrate_dead_tasks(cpu); |
| 5446 | task_rq_unlock(rq, &flags); |
| 5447 | migrate_nr_uninterruptible(rq); |
| 5448 | BUG_ON(rq->nr_running != 0); |
| 5449 | |
| 5450 | /* No need to migrate the tasks: it was best-effort if |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 5451 | * they didn't take sched_hotcpu_mutex. Just wake up |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5452 | * the requestors. */ |
| 5453 | spin_lock_irq(&rq->lock); |
| 5454 | while (!list_empty(&rq->migration_queue)) { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5455 | struct migration_req *req; |
| 5456 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5457 | req = list_entry(rq->migration_queue.next, |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5458 | struct migration_req, list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5459 | list_del_init(&req->list); |
| 5460 | complete(&req->done); |
| 5461 | } |
| 5462 | spin_unlock_irq(&rq->lock); |
| 5463 | break; |
| 5464 | #endif |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 5465 | case CPU_LOCK_RELEASE: |
| 5466 | mutex_unlock(&sched_hotcpu_mutex); |
| 5467 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5468 | } |
| 5469 | return NOTIFY_OK; |
| 5470 | } |
| 5471 | |
| 5472 | /* Register at highest priority so that task migration (migrate_all_tasks) |
| 5473 | * happens before everything else. |
| 5474 | */ |
Chandra Seetharaman | 26c2143 | 2006-06-27 02:54:10 -0700 | [diff] [blame] | 5475 | static struct notifier_block __cpuinitdata migration_notifier = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5476 | .notifier_call = migration_call, |
| 5477 | .priority = 10 |
| 5478 | }; |
| 5479 | |
| 5480 | int __init migration_init(void) |
| 5481 | { |
| 5482 | void *cpu = (void *)(long)smp_processor_id(); |
Akinobu Mita | 07dccf3 | 2006-09-29 02:00:22 -0700 | [diff] [blame] | 5483 | int err; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5484 | |
| 5485 | /* Start one for the boot CPU: */ |
Akinobu Mita | 07dccf3 | 2006-09-29 02:00:22 -0700 | [diff] [blame] | 5486 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
| 5487 | BUG_ON(err == NOTIFY_BAD); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5488 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
| 5489 | register_cpu_notifier(&migration_notifier); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5490 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5491 | return 0; |
| 5492 | } |
| 5493 | #endif |
| 5494 | |
| 5495 | #ifdef CONFIG_SMP |
Christoph Lameter | 476f353 | 2007-05-06 14:48:58 -0700 | [diff] [blame] | 5496 | |
| 5497 | /* Number of possible processor ids */ |
| 5498 | int nr_cpu_ids __read_mostly = NR_CPUS; |
| 5499 | EXPORT_SYMBOL(nr_cpu_ids); |
| 5500 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 5501 | #undef SCHED_DOMAIN_DEBUG |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5502 | #ifdef SCHED_DOMAIN_DEBUG |
| 5503 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
| 5504 | { |
| 5505 | int level = 0; |
| 5506 | |
Nick Piggin | 41c7ce9 | 2005-06-25 14:57:24 -0700 | [diff] [blame] | 5507 | if (!sd) { |
| 5508 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); |
| 5509 | return; |
| 5510 | } |
| 5511 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5512 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
| 5513 | |
| 5514 | do { |
| 5515 | int i; |
| 5516 | char str[NR_CPUS]; |
| 5517 | struct sched_group *group = sd->groups; |
| 5518 | cpumask_t groupmask; |
| 5519 | |
| 5520 | cpumask_scnprintf(str, NR_CPUS, sd->span); |
| 5521 | cpus_clear(groupmask); |
| 5522 | |
| 5523 | printk(KERN_DEBUG); |
| 5524 | for (i = 0; i < level + 1; i++) |
| 5525 | printk(" "); |
| 5526 | printk("domain %d: ", level); |
| 5527 | |
| 5528 | if (!(sd->flags & SD_LOAD_BALANCE)) { |
| 5529 | printk("does not load-balance\n"); |
| 5530 | if (sd->parent) |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 5531 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
| 5532 | " has parent"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5533 | break; |
| 5534 | } |
| 5535 | |
| 5536 | printk("span %s\n", str); |
| 5537 | |
| 5538 | if (!cpu_isset(cpu, sd->span)) |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 5539 | printk(KERN_ERR "ERROR: domain->span does not contain " |
| 5540 | "CPU%d\n", cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5541 | if (!cpu_isset(cpu, group->cpumask)) |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 5542 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
| 5543 | " CPU%d\n", cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5544 | |
| 5545 | printk(KERN_DEBUG); |
| 5546 | for (i = 0; i < level + 2; i++) |
| 5547 | printk(" "); |
| 5548 | printk("groups:"); |
| 5549 | do { |
| 5550 | if (!group) { |
| 5551 | printk("\n"); |
| 5552 | printk(KERN_ERR "ERROR: group is NULL\n"); |
| 5553 | break; |
| 5554 | } |
| 5555 | |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 5556 | if (!group->__cpu_power) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5557 | printk("\n"); |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 5558 | printk(KERN_ERR "ERROR: domain->cpu_power not " |
| 5559 | "set\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5560 | } |
| 5561 | |
| 5562 | if (!cpus_weight(group->cpumask)) { |
| 5563 | printk("\n"); |
| 5564 | printk(KERN_ERR "ERROR: empty group\n"); |
| 5565 | } |
| 5566 | |
| 5567 | if (cpus_intersects(groupmask, group->cpumask)) { |
| 5568 | printk("\n"); |
| 5569 | printk(KERN_ERR "ERROR: repeated CPUs\n"); |
| 5570 | } |
| 5571 | |
| 5572 | cpus_or(groupmask, groupmask, group->cpumask); |
| 5573 | |
| 5574 | cpumask_scnprintf(str, NR_CPUS, group->cpumask); |
| 5575 | printk(" %s", str); |
| 5576 | |
| 5577 | group = group->next; |
| 5578 | } while (group != sd->groups); |
| 5579 | printk("\n"); |
| 5580 | |
| 5581 | if (!cpus_equal(sd->span, groupmask)) |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 5582 | printk(KERN_ERR "ERROR: groups don't span " |
| 5583 | "domain->span\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5584 | |
| 5585 | level++; |
| 5586 | sd = sd->parent; |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 5587 | if (!sd) |
| 5588 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5589 | |
Miguel Ojeda Sandonis | 33859f7 | 2006-12-10 02:20:38 -0800 | [diff] [blame] | 5590 | if (!cpus_subset(groupmask, sd->span)) |
| 5591 | printk(KERN_ERR "ERROR: parent span is not a superset " |
| 5592 | "of domain->span\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5593 | |
| 5594 | } while (sd); |
| 5595 | } |
| 5596 | #else |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5597 | # define sched_domain_debug(sd, cpu) do { } while (0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5598 | #endif |
| 5599 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 5600 | static int sd_degenerate(struct sched_domain *sd) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5601 | { |
| 5602 | if (cpus_weight(sd->span) == 1) |
| 5603 | return 1; |
| 5604 | |
| 5605 | /* Following flags need at least 2 groups */ |
| 5606 | if (sd->flags & (SD_LOAD_BALANCE | |
| 5607 | SD_BALANCE_NEWIDLE | |
| 5608 | SD_BALANCE_FORK | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5609 | SD_BALANCE_EXEC | |
| 5610 | SD_SHARE_CPUPOWER | |
| 5611 | SD_SHARE_PKG_RESOURCES)) { |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5612 | if (sd->groups != sd->groups->next) |
| 5613 | return 0; |
| 5614 | } |
| 5615 | |
| 5616 | /* Following flags don't use groups */ |
| 5617 | if (sd->flags & (SD_WAKE_IDLE | |
| 5618 | SD_WAKE_AFFINE | |
| 5619 | SD_WAKE_BALANCE)) |
| 5620 | return 0; |
| 5621 | |
| 5622 | return 1; |
| 5623 | } |
| 5624 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5625 | static int |
| 5626 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5627 | { |
| 5628 | unsigned long cflags = sd->flags, pflags = parent->flags; |
| 5629 | |
| 5630 | if (sd_degenerate(parent)) |
| 5631 | return 1; |
| 5632 | |
| 5633 | if (!cpus_equal(sd->span, parent->span)) |
| 5634 | return 0; |
| 5635 | |
| 5636 | /* Does parent contain flags not in child? */ |
| 5637 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ |
| 5638 | if (cflags & SD_WAKE_AFFINE) |
| 5639 | pflags &= ~SD_WAKE_BALANCE; |
| 5640 | /* Flags needing groups don't count if only 1 group in parent */ |
| 5641 | if (parent->groups == parent->groups->next) { |
| 5642 | pflags &= ~(SD_LOAD_BALANCE | |
| 5643 | SD_BALANCE_NEWIDLE | |
| 5644 | SD_BALANCE_FORK | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5645 | SD_BALANCE_EXEC | |
| 5646 | SD_SHARE_CPUPOWER | |
| 5647 | SD_SHARE_PKG_RESOURCES); |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5648 | } |
| 5649 | if (~cflags & pflags) |
| 5650 | return 0; |
| 5651 | |
| 5652 | return 1; |
| 5653 | } |
| 5654 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5655 | /* |
| 5656 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
| 5657 | * hold the hotplug lock. |
| 5658 | */ |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5659 | static void cpu_attach_domain(struct sched_domain *sd, int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5660 | { |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 5661 | struct rq *rq = cpu_rq(cpu); |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5662 | struct sched_domain *tmp; |
| 5663 | |
| 5664 | /* Remove the sched domains which do not contribute to scheduling. */ |
| 5665 | for (tmp = sd; tmp; tmp = tmp->parent) { |
| 5666 | struct sched_domain *parent = tmp->parent; |
| 5667 | if (!parent) |
| 5668 | break; |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 5669 | if (sd_parent_degenerate(tmp, parent)) { |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5670 | tmp->parent = parent->parent; |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 5671 | if (parent->parent) |
| 5672 | parent->parent->child = tmp; |
| 5673 | } |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5674 | } |
| 5675 | |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 5676 | if (sd && sd_degenerate(sd)) { |
Suresh Siddha | 245af2c | 2005-06-25 14:57:25 -0700 | [diff] [blame] | 5677 | sd = sd->parent; |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 5678 | if (sd) |
| 5679 | sd->child = NULL; |
| 5680 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5681 | |
| 5682 | sched_domain_debug(sd, cpu); |
| 5683 | |
Nick Piggin | 674311d | 2005-06-25 14:57:27 -0700 | [diff] [blame] | 5684 | rcu_assign_pointer(rq->sd, sd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5685 | } |
| 5686 | |
| 5687 | /* cpus with isolated domains */ |
Tim Chen | 67af63a | 2006-12-22 01:07:50 -0800 | [diff] [blame] | 5688 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5689 | |
| 5690 | /* Setup the mask of cpus configured for isolated domains */ |
| 5691 | static int __init isolated_cpu_setup(char *str) |
| 5692 | { |
| 5693 | int ints[NR_CPUS], i; |
| 5694 | |
| 5695 | str = get_options(str, ARRAY_SIZE(ints), ints); |
| 5696 | cpus_clear(cpu_isolated_map); |
| 5697 | for (i = 1; i <= ints[0]; i++) |
| 5698 | if (ints[i] < NR_CPUS) |
| 5699 | cpu_set(ints[i], cpu_isolated_map); |
| 5700 | return 1; |
| 5701 | } |
| 5702 | |
| 5703 | __setup ("isolcpus=", isolated_cpu_setup); |
| 5704 | |
| 5705 | /* |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5706 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
| 5707 | * to a function which identifies what group(along with sched group) a CPU |
| 5708 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS |
| 5709 | * (due to the fact that we keep track of groups covered with a cpumask_t). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5710 | * |
| 5711 | * init_sched_build_groups will build a circular linked list of the groups |
| 5712 | * covered by the given span, and will set each group's ->cpumask correctly, |
| 5713 | * and ->cpu_power to 0. |
| 5714 | */ |
Siddha, Suresh B | a616058 | 2006-10-03 01:14:06 -0700 | [diff] [blame] | 5715 | static void |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5716 | init_sched_build_groups(cpumask_t span, const cpumask_t *cpu_map, |
| 5717 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
| 5718 | struct sched_group **sg)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5719 | { |
| 5720 | struct sched_group *first = NULL, *last = NULL; |
| 5721 | cpumask_t covered = CPU_MASK_NONE; |
| 5722 | int i; |
| 5723 | |
| 5724 | for_each_cpu_mask(i, span) { |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5725 | struct sched_group *sg; |
| 5726 | int group = group_fn(i, cpu_map, &sg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5727 | int j; |
| 5728 | |
| 5729 | if (cpu_isset(i, covered)) |
| 5730 | continue; |
| 5731 | |
| 5732 | sg->cpumask = CPU_MASK_NONE; |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 5733 | sg->__cpu_power = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5734 | |
| 5735 | for_each_cpu_mask(j, span) { |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5736 | if (group_fn(j, cpu_map, NULL) != group) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5737 | continue; |
| 5738 | |
| 5739 | cpu_set(j, covered); |
| 5740 | cpu_set(j, sg->cpumask); |
| 5741 | } |
| 5742 | if (!first) |
| 5743 | first = sg; |
| 5744 | if (last) |
| 5745 | last->next = sg; |
| 5746 | last = sg; |
| 5747 | } |
| 5748 | last->next = first; |
| 5749 | } |
| 5750 | |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5751 | #define SD_NODES_PER_DOMAIN 16 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5752 | |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5753 | #ifdef CONFIG_NUMA |
akpm@osdl.org | 198e2f1 | 2006-01-12 01:05:30 -0800 | [diff] [blame] | 5754 | |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5755 | /** |
| 5756 | * find_next_best_node - find the next node to include in a sched_domain |
| 5757 | * @node: node whose sched_domain we're building |
| 5758 | * @used_nodes: nodes already in the sched_domain |
| 5759 | * |
| 5760 | * Find the next node to include in a given scheduling domain. Simply |
| 5761 | * finds the closest node not already in the @used_nodes map. |
| 5762 | * |
| 5763 | * Should use nodemask_t. |
| 5764 | */ |
| 5765 | static int find_next_best_node(int node, unsigned long *used_nodes) |
| 5766 | { |
| 5767 | int i, n, val, min_val, best_node = 0; |
| 5768 | |
| 5769 | min_val = INT_MAX; |
| 5770 | |
| 5771 | for (i = 0; i < MAX_NUMNODES; i++) { |
| 5772 | /* Start at @node */ |
| 5773 | n = (node + i) % MAX_NUMNODES; |
| 5774 | |
| 5775 | if (!nr_cpus_node(n)) |
| 5776 | continue; |
| 5777 | |
| 5778 | /* Skip already used nodes */ |
| 5779 | if (test_bit(n, used_nodes)) |
| 5780 | continue; |
| 5781 | |
| 5782 | /* Simple min distance search */ |
| 5783 | val = node_distance(node, n); |
| 5784 | |
| 5785 | if (val < min_val) { |
| 5786 | min_val = val; |
| 5787 | best_node = n; |
| 5788 | } |
| 5789 | } |
| 5790 | |
| 5791 | set_bit(best_node, used_nodes); |
| 5792 | return best_node; |
| 5793 | } |
| 5794 | |
| 5795 | /** |
| 5796 | * sched_domain_node_span - get a cpumask for a node's sched_domain |
| 5797 | * @node: node whose cpumask we're constructing |
| 5798 | * @size: number of nodes to include in this span |
| 5799 | * |
| 5800 | * Given a node, construct a good cpumask for its sched_domain to span. It |
| 5801 | * should be one that prevents unnecessary balancing, but also spreads tasks |
| 5802 | * out optimally. |
| 5803 | */ |
| 5804 | static cpumask_t sched_domain_node_span(int node) |
| 5805 | { |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5806 | DECLARE_BITMAP(used_nodes, MAX_NUMNODES); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5807 | cpumask_t span, nodemask; |
| 5808 | int i; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5809 | |
| 5810 | cpus_clear(span); |
| 5811 | bitmap_zero(used_nodes, MAX_NUMNODES); |
| 5812 | |
| 5813 | nodemask = node_to_cpumask(node); |
| 5814 | cpus_or(span, span, nodemask); |
| 5815 | set_bit(node, used_nodes); |
| 5816 | |
| 5817 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { |
| 5818 | int next_node = find_next_best_node(node, used_nodes); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5819 | |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5820 | nodemask = node_to_cpumask(next_node); |
| 5821 | cpus_or(span, span, nodemask); |
| 5822 | } |
| 5823 | |
| 5824 | return span; |
| 5825 | } |
| 5826 | #endif |
| 5827 | |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 5828 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5829 | |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5830 | /* |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5831 | * SMT sched-domains: |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5832 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5833 | #ifdef CONFIG_SCHED_SMT |
| 5834 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5835 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5836 | |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5837 | static int cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, |
| 5838 | struct sched_group **sg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5839 | { |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5840 | if (sg) |
| 5841 | *sg = &per_cpu(sched_group_cpus, cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5842 | return cpu; |
| 5843 | } |
| 5844 | #endif |
| 5845 | |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5846 | /* |
| 5847 | * multi-core sched-domains: |
| 5848 | */ |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 5849 | #ifdef CONFIG_SCHED_MC |
| 5850 | static DEFINE_PER_CPU(struct sched_domain, core_domains); |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5851 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 5852 | #endif |
| 5853 | |
| 5854 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5855 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
| 5856 | struct sched_group **sg) |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 5857 | { |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5858 | int group; |
Siddha, Suresh B | a616058 | 2006-10-03 01:14:06 -0700 | [diff] [blame] | 5859 | cpumask_t mask = cpu_sibling_map[cpu]; |
| 5860 | cpus_and(mask, mask, *cpu_map); |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5861 | group = first_cpu(mask); |
| 5862 | if (sg) |
| 5863 | *sg = &per_cpu(sched_group_core, group); |
| 5864 | return group; |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 5865 | } |
| 5866 | #elif defined(CONFIG_SCHED_MC) |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5867 | static int cpu_to_core_group(int cpu, const cpumask_t *cpu_map, |
| 5868 | struct sched_group **sg) |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 5869 | { |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5870 | if (sg) |
| 5871 | *sg = &per_cpu(sched_group_core, cpu); |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 5872 | return cpu; |
| 5873 | } |
| 5874 | #endif |
| 5875 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5876 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5877 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5878 | |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5879 | static int cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, |
| 5880 | struct sched_group **sg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5881 | { |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5882 | int group; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 5883 | #ifdef CONFIG_SCHED_MC |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 5884 | cpumask_t mask = cpu_coregroup_map(cpu); |
Siddha, Suresh B | a616058 | 2006-10-03 01:14:06 -0700 | [diff] [blame] | 5885 | cpus_and(mask, mask, *cpu_map); |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5886 | group = first_cpu(mask); |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 5887 | #elif defined(CONFIG_SCHED_SMT) |
Siddha, Suresh B | a616058 | 2006-10-03 01:14:06 -0700 | [diff] [blame] | 5888 | cpumask_t mask = cpu_sibling_map[cpu]; |
| 5889 | cpus_and(mask, mask, *cpu_map); |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5890 | group = first_cpu(mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5891 | #else |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5892 | group = cpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5893 | #endif |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5894 | if (sg) |
| 5895 | *sg = &per_cpu(sched_group_phys, group); |
| 5896 | return group; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5897 | } |
| 5898 | |
| 5899 | #ifdef CONFIG_NUMA |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5900 | /* |
| 5901 | * The init_sched_build_groups can't handle what we want to do with node |
| 5902 | * groups, so roll our own. Now each node has its own list of groups which |
| 5903 | * gets dynamically allocated. |
| 5904 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5905 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
John Hawkes | d1b5513 | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5906 | static struct sched_group **sched_group_nodes_bycpu[NR_CPUS]; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5907 | |
| 5908 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5909 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 5910 | |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5911 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
| 5912 | struct sched_group **sg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5913 | { |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5914 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(cpu)); |
| 5915 | int group; |
| 5916 | |
| 5917 | cpus_and(nodemask, nodemask, *cpu_map); |
| 5918 | group = first_cpu(nodemask); |
| 5919 | |
| 5920 | if (sg) |
| 5921 | *sg = &per_cpu(sched_group_allnodes, group); |
| 5922 | return group; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5923 | } |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 5924 | |
Siddha, Suresh B | 0806903 | 2006-03-27 01:15:23 -0800 | [diff] [blame] | 5925 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
| 5926 | { |
| 5927 | struct sched_group *sg = group_head; |
| 5928 | int j; |
| 5929 | |
| 5930 | if (!sg) |
| 5931 | return; |
| 5932 | next_sg: |
| 5933 | for_each_cpu_mask(j, sg->cpumask) { |
| 5934 | struct sched_domain *sd; |
| 5935 | |
| 5936 | sd = &per_cpu(phys_domains, j); |
| 5937 | if (j != first_cpu(sd->groups->cpumask)) { |
| 5938 | /* |
| 5939 | * Only add "power" once for each |
| 5940 | * physical package. |
| 5941 | */ |
| 5942 | continue; |
| 5943 | } |
| 5944 | |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 5945 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
Siddha, Suresh B | 0806903 | 2006-03-27 01:15:23 -0800 | [diff] [blame] | 5946 | } |
| 5947 | sg = sg->next; |
| 5948 | if (sg != group_head) |
| 5949 | goto next_sg; |
| 5950 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5951 | #endif |
| 5952 | |
Siddha, Suresh B | a616058 | 2006-10-03 01:14:06 -0700 | [diff] [blame] | 5953 | #ifdef CONFIG_NUMA |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 5954 | /* Free memory allocated for various sched_group structures */ |
| 5955 | static void free_sched_groups(const cpumask_t *cpu_map) |
| 5956 | { |
Siddha, Suresh B | a616058 | 2006-10-03 01:14:06 -0700 | [diff] [blame] | 5957 | int cpu, i; |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 5958 | |
| 5959 | for_each_cpu_mask(cpu, *cpu_map) { |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 5960 | struct sched_group **sched_group_nodes |
| 5961 | = sched_group_nodes_bycpu[cpu]; |
| 5962 | |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 5963 | if (!sched_group_nodes) |
| 5964 | continue; |
| 5965 | |
| 5966 | for (i = 0; i < MAX_NUMNODES; i++) { |
| 5967 | cpumask_t nodemask = node_to_cpumask(i); |
| 5968 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
| 5969 | |
| 5970 | cpus_and(nodemask, nodemask, *cpu_map); |
| 5971 | if (cpus_empty(nodemask)) |
| 5972 | continue; |
| 5973 | |
| 5974 | if (sg == NULL) |
| 5975 | continue; |
| 5976 | sg = sg->next; |
| 5977 | next_sg: |
| 5978 | oldsg = sg; |
| 5979 | sg = sg->next; |
| 5980 | kfree(oldsg); |
| 5981 | if (oldsg != sched_group_nodes[i]) |
| 5982 | goto next_sg; |
| 5983 | } |
| 5984 | kfree(sched_group_nodes); |
| 5985 | sched_group_nodes_bycpu[cpu] = NULL; |
| 5986 | } |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 5987 | } |
Siddha, Suresh B | a616058 | 2006-10-03 01:14:06 -0700 | [diff] [blame] | 5988 | #else |
| 5989 | static void free_sched_groups(const cpumask_t *cpu_map) |
| 5990 | { |
| 5991 | } |
| 5992 | #endif |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 5993 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5994 | /* |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 5995 | * Initialize sched groups cpu_power. |
| 5996 | * |
| 5997 | * cpu_power indicates the capacity of sched group, which is used while |
| 5998 | * distributing the load between different sched groups in a sched domain. |
| 5999 | * Typically cpu_power for all the groups in a sched domain will be same unless |
| 6000 | * there are asymmetries in the topology. If there are asymmetries, group |
| 6001 | * having more cpu_power will pickup more load compared to the group having |
| 6002 | * less cpu_power. |
| 6003 | * |
| 6004 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents |
| 6005 | * the maximum number of tasks a group can handle in the presence of other idle |
| 6006 | * or lightly loaded groups in the same sched domain. |
| 6007 | */ |
| 6008 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) |
| 6009 | { |
| 6010 | struct sched_domain *child; |
| 6011 | struct sched_group *group; |
| 6012 | |
| 6013 | WARN_ON(!sd || !sd->groups); |
| 6014 | |
| 6015 | if (cpu != first_cpu(sd->groups->cpumask)) |
| 6016 | return; |
| 6017 | |
| 6018 | child = sd->child; |
| 6019 | |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 6020 | sd->groups->__cpu_power = 0; |
| 6021 | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6022 | /* |
| 6023 | * For perf policy, if the groups in child domain share resources |
| 6024 | * (for example cores sharing some portions of the cache hierarchy |
| 6025 | * or SMT), then set this domain groups cpu_power such that each group |
| 6026 | * can handle only one task, when there are other idle groups in the |
| 6027 | * same sched domain. |
| 6028 | */ |
| 6029 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && |
| 6030 | (child->flags & |
| 6031 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 6032 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6033 | return; |
| 6034 | } |
| 6035 | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6036 | /* |
| 6037 | * add cpu_power of each child group to this groups cpu_power |
| 6038 | */ |
| 6039 | group = child->groups; |
| 6040 | do { |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 6041 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6042 | group = group->next; |
| 6043 | } while (group != child->groups); |
| 6044 | } |
| 6045 | |
| 6046 | /* |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6047 | * Build sched domains for a given set of cpus and attach the sched domains |
| 6048 | * to the individual cpus |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6049 | */ |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6050 | static int build_sched_domains(const cpumask_t *cpu_map) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6051 | { |
| 6052 | int i; |
John Hawkes | d1b5513 | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6053 | #ifdef CONFIG_NUMA |
| 6054 | struct sched_group **sched_group_nodes = NULL; |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 6055 | int sd_allnodes = 0; |
John Hawkes | d1b5513 | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6056 | |
| 6057 | /* |
| 6058 | * Allocate the per-node list of sched groups |
| 6059 | */ |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6060 | sched_group_nodes = kzalloc(sizeof(struct sched_group *)*MAX_NUMNODES, |
Srivatsa Vaddagiri | d3a5aa9 | 2006-06-27 02:54:39 -0700 | [diff] [blame] | 6061 | GFP_KERNEL); |
John Hawkes | d1b5513 | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6062 | if (!sched_group_nodes) { |
| 6063 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6064 | return -ENOMEM; |
John Hawkes | d1b5513 | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6065 | } |
| 6066 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; |
| 6067 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6068 | |
| 6069 | /* |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6070 | * Set up domains for cpus specified by the cpu_map. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6071 | */ |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6072 | for_each_cpu_mask(i, *cpu_map) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6073 | struct sched_domain *sd = NULL, *p; |
| 6074 | cpumask_t nodemask = node_to_cpumask(cpu_to_node(i)); |
| 6075 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6076 | cpus_and(nodemask, nodemask, *cpu_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6077 | |
| 6078 | #ifdef CONFIG_NUMA |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6079 | if (cpus_weight(*cpu_map) > |
| 6080 | SD_NODES_PER_DOMAIN*cpus_weight(nodemask)) { |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6081 | sd = &per_cpu(allnodes_domains, i); |
| 6082 | *sd = SD_ALLNODES_INIT; |
| 6083 | sd->span = *cpu_map; |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 6084 | cpu_to_allnodes_group(i, cpu_map, &sd->groups); |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6085 | p = sd; |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 6086 | sd_allnodes = 1; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6087 | } else |
| 6088 | p = NULL; |
| 6089 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6090 | sd = &per_cpu(node_domains, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6091 | *sd = SD_NODE_INIT; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6092 | sd->span = sched_domain_node_span(cpu_to_node(i)); |
| 6093 | sd->parent = p; |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 6094 | if (p) |
| 6095 | p->child = sd; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6096 | cpus_and(sd->span, sd->span, *cpu_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6097 | #endif |
| 6098 | |
| 6099 | p = sd; |
| 6100 | sd = &per_cpu(phys_domains, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6101 | *sd = SD_CPU_INIT; |
| 6102 | sd->span = nodemask; |
| 6103 | sd->parent = p; |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 6104 | if (p) |
| 6105 | p->child = sd; |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 6106 | cpu_to_phys_group(i, cpu_map, &sd->groups); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6107 | |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 6108 | #ifdef CONFIG_SCHED_MC |
| 6109 | p = sd; |
| 6110 | sd = &per_cpu(core_domains, i); |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 6111 | *sd = SD_MC_INIT; |
| 6112 | sd->span = cpu_coregroup_map(i); |
| 6113 | cpus_and(sd->span, sd->span, *cpu_map); |
| 6114 | sd->parent = p; |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 6115 | p->child = sd; |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 6116 | cpu_to_core_group(i, cpu_map, &sd->groups); |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 6117 | #endif |
| 6118 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6119 | #ifdef CONFIG_SCHED_SMT |
| 6120 | p = sd; |
| 6121 | sd = &per_cpu(cpu_domains, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6122 | *sd = SD_SIBLING_INIT; |
| 6123 | sd->span = cpu_sibling_map[i]; |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6124 | cpus_and(sd->span, sd->span, *cpu_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6125 | sd->parent = p; |
Siddha, Suresh B | 1a84887 | 2006-10-03 01:14:08 -0700 | [diff] [blame] | 6126 | p->child = sd; |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 6127 | cpu_to_cpu_group(i, cpu_map, &sd->groups); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6128 | #endif |
| 6129 | } |
| 6130 | |
| 6131 | #ifdef CONFIG_SCHED_SMT |
| 6132 | /* Set up CPU (sibling) groups */ |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6133 | for_each_cpu_mask(i, *cpu_map) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6134 | cpumask_t this_sibling_map = cpu_sibling_map[i]; |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6135 | cpus_and(this_sibling_map, this_sibling_map, *cpu_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6136 | if (i != first_cpu(this_sibling_map)) |
| 6137 | continue; |
| 6138 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6139 | init_sched_build_groups(this_sibling_map, cpu_map, |
| 6140 | &cpu_to_cpu_group); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6141 | } |
| 6142 | #endif |
| 6143 | |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 6144 | #ifdef CONFIG_SCHED_MC |
| 6145 | /* Set up multi-core groups */ |
| 6146 | for_each_cpu_mask(i, *cpu_map) { |
| 6147 | cpumask_t this_core_map = cpu_coregroup_map(i); |
| 6148 | cpus_and(this_core_map, this_core_map, *cpu_map); |
| 6149 | if (i != first_cpu(this_core_map)) |
| 6150 | continue; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6151 | init_sched_build_groups(this_core_map, cpu_map, |
| 6152 | &cpu_to_core_group); |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 6153 | } |
| 6154 | #endif |
| 6155 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6156 | /* Set up physical groups */ |
| 6157 | for (i = 0; i < MAX_NUMNODES; i++) { |
| 6158 | cpumask_t nodemask = node_to_cpumask(i); |
| 6159 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6160 | cpus_and(nodemask, nodemask, *cpu_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6161 | if (cpus_empty(nodemask)) |
| 6162 | continue; |
| 6163 | |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 6164 | init_sched_build_groups(nodemask, cpu_map, &cpu_to_phys_group); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6165 | } |
| 6166 | |
| 6167 | #ifdef CONFIG_NUMA |
| 6168 | /* Set up node groups */ |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 6169 | if (sd_allnodes) |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6170 | init_sched_build_groups(*cpu_map, cpu_map, |
| 6171 | &cpu_to_allnodes_group); |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6172 | |
| 6173 | for (i = 0; i < MAX_NUMNODES; i++) { |
| 6174 | /* Set up node groups */ |
| 6175 | struct sched_group *sg, *prev; |
| 6176 | cpumask_t nodemask = node_to_cpumask(i); |
| 6177 | cpumask_t domainspan; |
| 6178 | cpumask_t covered = CPU_MASK_NONE; |
| 6179 | int j; |
| 6180 | |
| 6181 | cpus_and(nodemask, nodemask, *cpu_map); |
John Hawkes | d1b5513 | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6182 | if (cpus_empty(nodemask)) { |
| 6183 | sched_group_nodes[i] = NULL; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6184 | continue; |
John Hawkes | d1b5513 | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6185 | } |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6186 | |
| 6187 | domainspan = sched_domain_node_span(i); |
| 6188 | cpus_and(domainspan, domainspan, *cpu_map); |
| 6189 | |
Srivatsa Vaddagiri | 15f0b67 | 2006-06-27 02:54:40 -0700 | [diff] [blame] | 6190 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6191 | if (!sg) { |
| 6192 | printk(KERN_WARNING "Can not alloc domain group for " |
| 6193 | "node %d\n", i); |
| 6194 | goto error; |
| 6195 | } |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6196 | sched_group_nodes[i] = sg; |
| 6197 | for_each_cpu_mask(j, nodemask) { |
| 6198 | struct sched_domain *sd; |
Ingo Molnar | 9761eea | 2007-07-09 18:52:00 +0200 | [diff] [blame] | 6199 | |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6200 | sd = &per_cpu(node_domains, j); |
| 6201 | sd->groups = sg; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6202 | } |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 6203 | sg->__cpu_power = 0; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6204 | sg->cpumask = nodemask; |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6205 | sg->next = sg; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6206 | cpus_or(covered, covered, nodemask); |
| 6207 | prev = sg; |
| 6208 | |
| 6209 | for (j = 0; j < MAX_NUMNODES; j++) { |
| 6210 | cpumask_t tmp, notcovered; |
| 6211 | int n = (i + j) % MAX_NUMNODES; |
| 6212 | |
| 6213 | cpus_complement(notcovered, covered); |
| 6214 | cpus_and(tmp, notcovered, *cpu_map); |
| 6215 | cpus_and(tmp, tmp, domainspan); |
| 6216 | if (cpus_empty(tmp)) |
| 6217 | break; |
| 6218 | |
| 6219 | nodemask = node_to_cpumask(n); |
| 6220 | cpus_and(tmp, tmp, nodemask); |
| 6221 | if (cpus_empty(tmp)) |
| 6222 | continue; |
| 6223 | |
Srivatsa Vaddagiri | 15f0b67 | 2006-06-27 02:54:40 -0700 | [diff] [blame] | 6224 | sg = kmalloc_node(sizeof(struct sched_group), |
| 6225 | GFP_KERNEL, i); |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6226 | if (!sg) { |
| 6227 | printk(KERN_WARNING |
| 6228 | "Can not alloc domain group for node %d\n", j); |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6229 | goto error; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6230 | } |
Eric Dumazet | 5517d86 | 2007-05-08 00:32:57 -0700 | [diff] [blame] | 6231 | sg->__cpu_power = 0; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6232 | sg->cpumask = tmp; |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6233 | sg->next = prev->next; |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6234 | cpus_or(covered, covered, tmp); |
| 6235 | prev->next = sg; |
| 6236 | prev = sg; |
| 6237 | } |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6238 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6239 | #endif |
| 6240 | |
| 6241 | /* Calculate CPU power for physical packages and nodes */ |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 6242 | #ifdef CONFIG_SCHED_SMT |
| 6243 | for_each_cpu_mask(i, *cpu_map) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6244 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
| 6245 | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6246 | init_sched_groups_power(i, sd); |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 6247 | } |
| 6248 | #endif |
| 6249 | #ifdef CONFIG_SCHED_MC |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6250 | for_each_cpu_mask(i, *cpu_map) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6251 | struct sched_domain *sd = &per_cpu(core_domains, i); |
| 6252 | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6253 | init_sched_groups_power(i, sd); |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 6254 | } |
| 6255 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6256 | |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 6257 | for_each_cpu_mask(i, *cpu_map) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6258 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
| 6259 | |
Siddha, Suresh B | 89c4710 | 2006-10-03 01:14:09 -0700 | [diff] [blame] | 6260 | init_sched_groups_power(i, sd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6261 | } |
| 6262 | |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6263 | #ifdef CONFIG_NUMA |
Siddha, Suresh B | 0806903 | 2006-03-27 01:15:23 -0800 | [diff] [blame] | 6264 | for (i = 0; i < MAX_NUMNODES; i++) |
| 6265 | init_numa_sched_groups_power(sched_group_nodes[i]); |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6266 | |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 6267 | if (sd_allnodes) { |
| 6268 | struct sched_group *sg; |
Siddha, Suresh B | f712c0c | 2006-07-30 03:02:59 -0700 | [diff] [blame] | 6269 | |
Siddha, Suresh B | 6711cab | 2006-12-10 02:20:07 -0800 | [diff] [blame] | 6270 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg); |
Siddha, Suresh B | f712c0c | 2006-07-30 03:02:59 -0700 | [diff] [blame] | 6271 | init_numa_sched_groups_power(sg); |
| 6272 | } |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6273 | #endif |
| 6274 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6275 | /* Attach the domains */ |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6276 | for_each_cpu_mask(i, *cpu_map) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6277 | struct sched_domain *sd; |
| 6278 | #ifdef CONFIG_SCHED_SMT |
| 6279 | sd = &per_cpu(cpu_domains, i); |
Siddha, Suresh B | 1e9f28f | 2006-03-27 01:15:22 -0800 | [diff] [blame] | 6280 | #elif defined(CONFIG_SCHED_MC) |
| 6281 | sd = &per_cpu(core_domains, i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6282 | #else |
| 6283 | sd = &per_cpu(phys_domains, i); |
| 6284 | #endif |
| 6285 | cpu_attach_domain(sd, i); |
| 6286 | } |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6287 | |
| 6288 | return 0; |
| 6289 | |
Siddha, Suresh B | a616058 | 2006-10-03 01:14:06 -0700 | [diff] [blame] | 6290 | #ifdef CONFIG_NUMA |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6291 | error: |
| 6292 | free_sched_groups(cpu_map); |
| 6293 | return -ENOMEM; |
Siddha, Suresh B | a616058 | 2006-10-03 01:14:06 -0700 | [diff] [blame] | 6294 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6295 | } |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6296 | /* |
| 6297 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
| 6298 | */ |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6299 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6300 | { |
| 6301 | cpumask_t cpu_default_map; |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6302 | int err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6303 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6304 | /* |
| 6305 | * Setup mask for cpus without special case scheduling requirements. |
| 6306 | * For now this just excludes isolated cpus, but could be used to |
| 6307 | * exclude other special cases in the future. |
| 6308 | */ |
| 6309 | cpus_andnot(cpu_default_map, *cpu_map, cpu_isolated_map); |
| 6310 | |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6311 | err = build_sched_domains(&cpu_default_map); |
| 6312 | |
| 6313 | return err; |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6314 | } |
| 6315 | |
| 6316 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6317 | { |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6318 | free_sched_groups(cpu_map); |
John Hawkes | 9c1cfda | 2005-09-06 15:18:14 -0700 | [diff] [blame] | 6319 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6320 | |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6321 | /* |
| 6322 | * Detach sched domains from a group of cpus specified in cpu_map |
| 6323 | * These cpus will now be attached to the NULL domain |
| 6324 | */ |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 6325 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6326 | { |
| 6327 | int i; |
| 6328 | |
| 6329 | for_each_cpu_mask(i, *cpu_map) |
| 6330 | cpu_attach_domain(NULL, i); |
| 6331 | synchronize_sched(); |
| 6332 | arch_destroy_sched_domains(cpu_map); |
| 6333 | } |
| 6334 | |
| 6335 | /* |
| 6336 | * Partition sched domains as specified by the cpumasks below. |
| 6337 | * This attaches all cpus from the cpumasks to the NULL domain, |
| 6338 | * waits for a RCU quiescent period, recalculates sched |
| 6339 | * domain information and then attaches them back to the |
| 6340 | * correct sched domains |
| 6341 | * Call with hotplug lock held |
| 6342 | */ |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6343 | int partition_sched_domains(cpumask_t *partition1, cpumask_t *partition2) |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6344 | { |
| 6345 | cpumask_t change_map; |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6346 | int err = 0; |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6347 | |
| 6348 | cpus_and(*partition1, *partition1, cpu_online_map); |
| 6349 | cpus_and(*partition2, *partition2, cpu_online_map); |
| 6350 | cpus_or(change_map, *partition1, *partition2); |
| 6351 | |
| 6352 | /* Detach sched domains from all of the affected cpus */ |
| 6353 | detach_destroy_domains(&change_map); |
| 6354 | if (!cpus_empty(*partition1)) |
Srivatsa Vaddagiri | 51888ca | 2006-06-27 02:54:38 -0700 | [diff] [blame] | 6355 | err = build_sched_domains(partition1); |
| 6356 | if (!err && !cpus_empty(*partition2)) |
| 6357 | err = build_sched_domains(partition2); |
| 6358 | |
| 6359 | return err; |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6360 | } |
| 6361 | |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 6362 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
Adrian Bunk | 6707de00 | 2007-08-12 18:08:19 +0200 | [diff] [blame] | 6363 | static int arch_reinit_sched_domains(void) |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 6364 | { |
| 6365 | int err; |
| 6366 | |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 6367 | mutex_lock(&sched_hotcpu_mutex); |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 6368 | detach_destroy_domains(&cpu_online_map); |
| 6369 | err = arch_init_sched_domains(&cpu_online_map); |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 6370 | mutex_unlock(&sched_hotcpu_mutex); |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 6371 | |
| 6372 | return err; |
| 6373 | } |
| 6374 | |
| 6375 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) |
| 6376 | { |
| 6377 | int ret; |
| 6378 | |
| 6379 | if (buf[0] != '0' && buf[0] != '1') |
| 6380 | return -EINVAL; |
| 6381 | |
| 6382 | if (smt) |
| 6383 | sched_smt_power_savings = (buf[0] == '1'); |
| 6384 | else |
| 6385 | sched_mc_power_savings = (buf[0] == '1'); |
| 6386 | |
| 6387 | ret = arch_reinit_sched_domains(); |
| 6388 | |
| 6389 | return ret ? ret : count; |
| 6390 | } |
| 6391 | |
Adrian Bunk | 6707de00 | 2007-08-12 18:08:19 +0200 | [diff] [blame] | 6392 | #ifdef CONFIG_SCHED_MC |
| 6393 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) |
| 6394 | { |
| 6395 | return sprintf(page, "%u\n", sched_mc_power_savings); |
| 6396 | } |
| 6397 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
| 6398 | const char *buf, size_t count) |
| 6399 | { |
| 6400 | return sched_power_savings_store(buf, count, 0); |
| 6401 | } |
| 6402 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
| 6403 | sched_mc_power_savings_store); |
| 6404 | #endif |
| 6405 | |
| 6406 | #ifdef CONFIG_SCHED_SMT |
| 6407 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) |
| 6408 | { |
| 6409 | return sprintf(page, "%u\n", sched_smt_power_savings); |
| 6410 | } |
| 6411 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
| 6412 | const char *buf, size_t count) |
| 6413 | { |
| 6414 | return sched_power_savings_store(buf, count, 1); |
| 6415 | } |
| 6416 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
| 6417 | sched_smt_power_savings_store); |
| 6418 | #endif |
| 6419 | |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 6420 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
| 6421 | { |
| 6422 | int err = 0; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 6423 | |
Siddha, Suresh B | 5c45bf2 | 2006-06-27 02:54:42 -0700 | [diff] [blame] | 6424 | #ifdef CONFIG_SCHED_SMT |
| 6425 | if (smt_capable()) |
| 6426 | err = sysfs_create_file(&cls->kset.kobj, |
| 6427 | &attr_sched_smt_power_savings.attr); |
| 6428 | #endif |
| 6429 | #ifdef CONFIG_SCHED_MC |
| 6430 | if (!err && mc_capable()) |
| 6431 | err = sysfs_create_file(&cls->kset.kobj, |
| 6432 | &attr_sched_mc_power_savings.attr); |
| 6433 | #endif |
| 6434 | return err; |
| 6435 | } |
| 6436 | #endif |
| 6437 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6438 | /* |
| 6439 | * Force a reinitialization of the sched domains hierarchy. The domains |
| 6440 | * and groups cannot be updated in place without racing with the balancing |
Nick Piggin | 41c7ce9 | 2005-06-25 14:57:24 -0700 | [diff] [blame] | 6441 | * code, so we temporarily attach all running cpus to the NULL domain |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6442 | * which will prevent rebalancing while the sched domains are recalculated. |
| 6443 | */ |
| 6444 | static int update_sched_domains(struct notifier_block *nfb, |
| 6445 | unsigned long action, void *hcpu) |
| 6446 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6447 | switch (action) { |
| 6448 | case CPU_UP_PREPARE: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 6449 | case CPU_UP_PREPARE_FROZEN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6450 | case CPU_DOWN_PREPARE: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 6451 | case CPU_DOWN_PREPARE_FROZEN: |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6452 | detach_destroy_domains(&cpu_online_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6453 | return NOTIFY_OK; |
| 6454 | |
| 6455 | case CPU_UP_CANCELED: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 6456 | case CPU_UP_CANCELED_FROZEN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6457 | case CPU_DOWN_FAILED: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 6458 | case CPU_DOWN_FAILED_FROZEN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6459 | case CPU_ONLINE: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 6460 | case CPU_ONLINE_FROZEN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6461 | case CPU_DEAD: |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 6462 | case CPU_DEAD_FROZEN: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6463 | /* |
| 6464 | * Fall through and re-initialise the domains. |
| 6465 | */ |
| 6466 | break; |
| 6467 | default: |
| 6468 | return NOTIFY_DONE; |
| 6469 | } |
| 6470 | |
| 6471 | /* The hotplug lock is already held by cpu_up/cpu_down */ |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6472 | arch_init_sched_domains(&cpu_online_map); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6473 | |
| 6474 | return NOTIFY_OK; |
| 6475 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6476 | |
| 6477 | void __init sched_init_smp(void) |
| 6478 | { |
Nick Piggin | 5c1e176 | 2006-10-03 01:14:04 -0700 | [diff] [blame] | 6479 | cpumask_t non_isolated_cpus; |
| 6480 | |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 6481 | mutex_lock(&sched_hotcpu_mutex); |
Dinakar Guniguntala | 1a20ff2 | 2005-06-25 14:57:33 -0700 | [diff] [blame] | 6482 | arch_init_sched_domains(&cpu_online_map); |
Nathan Lynch | e5e5673 | 2007-01-10 23:15:28 -0800 | [diff] [blame] | 6483 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
Nick Piggin | 5c1e176 | 2006-10-03 01:14:04 -0700 | [diff] [blame] | 6484 | if (cpus_empty(non_isolated_cpus)) |
| 6485 | cpu_set(smp_processor_id(), non_isolated_cpus); |
Gautham R Shenoy | 5be9361 | 2007-05-09 02:34:04 -0700 | [diff] [blame] | 6486 | mutex_unlock(&sched_hotcpu_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6487 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
| 6488 | hotcpu_notifier(update_sched_domains, 0); |
Nick Piggin | 5c1e176 | 2006-10-03 01:14:04 -0700 | [diff] [blame] | 6489 | |
Nick Piggin | e692ab5 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 6490 | init_sched_domain_sysctl(); |
| 6491 | |
Nick Piggin | 5c1e176 | 2006-10-03 01:14:04 -0700 | [diff] [blame] | 6492 | /* Move init over to a non-isolated CPU */ |
| 6493 | if (set_cpus_allowed(current, non_isolated_cpus) < 0) |
| 6494 | BUG(); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6495 | sched_init_granularity(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6496 | } |
| 6497 | #else |
| 6498 | void __init sched_init_smp(void) |
| 6499 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6500 | sched_init_granularity(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6501 | } |
| 6502 | #endif /* CONFIG_SMP */ |
| 6503 | |
| 6504 | int in_sched_functions(unsigned long addr) |
| 6505 | { |
| 6506 | /* Linker adds these: start and end of __sched functions */ |
| 6507 | extern char __sched_text_start[], __sched_text_end[]; |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 6508 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6509 | return in_lock_functions(addr) || |
| 6510 | (addr >= (unsigned long)__sched_text_start |
| 6511 | && addr < (unsigned long)__sched_text_end); |
| 6512 | } |
| 6513 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6514 | static inline void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
| 6515 | { |
| 6516 | cfs_rq->tasks_timeline = RB_ROOT; |
| 6517 | cfs_rq->fair_clock = 1; |
| 6518 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 6519 | cfs_rq->rq = rq; |
| 6520 | #endif |
| 6521 | } |
| 6522 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6523 | void __init sched_init(void) |
| 6524 | { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6525 | u64 now = sched_clock(); |
Christoph Lameter | 476f353 | 2007-05-06 14:48:58 -0700 | [diff] [blame] | 6526 | int highest_cpu = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6527 | int i, j; |
| 6528 | |
| 6529 | /* |
| 6530 | * Link up the scheduling class hierarchy: |
| 6531 | */ |
| 6532 | rt_sched_class.next = &fair_sched_class; |
| 6533 | fair_sched_class.next = &idle_sched_class; |
| 6534 | idle_sched_class.next = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6535 | |
KAMEZAWA Hiroyuki | 0a94502 | 2006-03-28 01:56:37 -0800 | [diff] [blame] | 6536 | for_each_possible_cpu(i) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6537 | struct rt_prio_array *array; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 6538 | struct rq *rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6539 | |
| 6540 | rq = cpu_rq(i); |
| 6541 | spin_lock_init(&rq->lock); |
Ingo Molnar | fcb9937 | 2006-07-03 00:25:10 -0700 | [diff] [blame] | 6542 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
Nick Piggin | 7897986 | 2005-06-25 14:57:13 -0700 | [diff] [blame] | 6543 | rq->nr_running = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6544 | rq->clock = 1; |
| 6545 | init_cfs_rq(&rq->cfs, rq); |
| 6546 | #ifdef CONFIG_FAIR_GROUP_SCHED |
| 6547 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
| 6548 | list_add(&rq->cfs.leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); |
| 6549 | #endif |
| 6550 | rq->ls.load_update_last = now; |
| 6551 | rq->ls.load_update_start = now; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6552 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6553 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
| 6554 | rq->cpu_load[j] = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6555 | #ifdef CONFIG_SMP |
Nick Piggin | 41c7ce9 | 2005-06-25 14:57:24 -0700 | [diff] [blame] | 6556 | rq->sd = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6557 | rq->active_balance = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6558 | rq->next_balance = jiffies; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6559 | rq->push_cpu = 0; |
Christoph Lameter | 0a2966b | 2006-09-25 23:30:51 -0700 | [diff] [blame] | 6560 | rq->cpu = i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6561 | rq->migration_thread = NULL; |
| 6562 | INIT_LIST_HEAD(&rq->migration_queue); |
| 6563 | #endif |
| 6564 | atomic_set(&rq->nr_iowait, 0); |
| 6565 | |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6566 | array = &rq->rt.active; |
| 6567 | for (j = 0; j < MAX_RT_PRIO; j++) { |
| 6568 | INIT_LIST_HEAD(array->queue + j); |
| 6569 | __clear_bit(j, array->bitmap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6570 | } |
Christoph Lameter | 476f353 | 2007-05-06 14:48:58 -0700 | [diff] [blame] | 6571 | highest_cpu = i; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6572 | /* delimiter for bitsearch: */ |
| 6573 | __set_bit(MAX_RT_PRIO, array->bitmap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6574 | } |
| 6575 | |
Peter Williams | 2dd73a4 | 2006-06-27 02:54:34 -0700 | [diff] [blame] | 6576 | set_load_weight(&init_task); |
Heiko Carstens | b50f60c | 2006-07-30 03:03:52 -0700 | [diff] [blame] | 6577 | |
Avi Kivity | e107be3 | 2007-07-26 13:40:43 +0200 | [diff] [blame] | 6578 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
| 6579 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); |
| 6580 | #endif |
| 6581 | |
Christoph Lameter | c9819f4 | 2006-12-10 02:20:25 -0800 | [diff] [blame] | 6582 | #ifdef CONFIG_SMP |
Christoph Lameter | 476f353 | 2007-05-06 14:48:58 -0700 | [diff] [blame] | 6583 | nr_cpu_ids = highest_cpu + 1; |
Christoph Lameter | c9819f4 | 2006-12-10 02:20:25 -0800 | [diff] [blame] | 6584 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); |
| 6585 | #endif |
| 6586 | |
Heiko Carstens | b50f60c | 2006-07-30 03:03:52 -0700 | [diff] [blame] | 6587 | #ifdef CONFIG_RT_MUTEXES |
| 6588 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); |
| 6589 | #endif |
| 6590 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6591 | /* |
| 6592 | * The boot idle thread does lazy MMU switching as well: |
| 6593 | */ |
| 6594 | atomic_inc(&init_mm.mm_count); |
| 6595 | enter_lazy_tlb(&init_mm, current); |
| 6596 | |
| 6597 | /* |
| 6598 | * Make us the idle thread. Technically, schedule() should not be |
| 6599 | * called from this thread, however somewhere below it might be, |
| 6600 | * but because we are the idle thread, we just pick up running again |
| 6601 | * when this runqueue becomes "idle". |
| 6602 | */ |
| 6603 | init_idle(current, smp_processor_id()); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6604 | /* |
| 6605 | * During early bootup we pretend to be a normal task: |
| 6606 | */ |
| 6607 | current->sched_class = &fair_sched_class; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6608 | } |
| 6609 | |
| 6610 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
| 6611 | void __might_sleep(char *file, int line) |
| 6612 | { |
Ingo Molnar | 48f24c4 | 2006-07-03 00:25:40 -0700 | [diff] [blame] | 6613 | #ifdef in_atomic |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6614 | static unsigned long prev_jiffy; /* ratelimiting */ |
| 6615 | |
| 6616 | if ((in_atomic() || irqs_disabled()) && |
| 6617 | system_state == SYSTEM_RUNNING && !oops_in_progress) { |
| 6618 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) |
| 6619 | return; |
| 6620 | prev_jiffy = jiffies; |
Ingo Molnar | 91368d7 | 2006-03-23 03:00:54 -0800 | [diff] [blame] | 6621 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6622 | " context at %s:%d\n", file, line); |
| 6623 | printk("in_atomic():%d, irqs_disabled():%d\n", |
| 6624 | in_atomic(), irqs_disabled()); |
Peter Zijlstra | a4c410f | 2006-12-06 20:37:21 -0800 | [diff] [blame] | 6625 | debug_show_held_locks(current); |
Ingo Molnar | 3117df0 | 2006-12-13 00:34:43 -0800 | [diff] [blame] | 6626 | if (irqs_disabled()) |
| 6627 | print_irqtrace_events(current); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6628 | dump_stack(); |
| 6629 | } |
| 6630 | #endif |
| 6631 | } |
| 6632 | EXPORT_SYMBOL(__might_sleep); |
| 6633 | #endif |
| 6634 | |
| 6635 | #ifdef CONFIG_MAGIC_SYSRQ |
| 6636 | void normalize_rt_tasks(void) |
| 6637 | { |
Ingo Molnar | a0f98a1 | 2007-06-17 18:37:45 +0200 | [diff] [blame] | 6638 | struct task_struct *g, *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6639 | unsigned long flags; |
Ingo Molnar | 70b97a7 | 2006-07-03 00:25:42 -0700 | [diff] [blame] | 6640 | struct rq *rq; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6641 | int on_rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6642 | |
| 6643 | read_lock_irq(&tasklist_lock); |
Ingo Molnar | a0f98a1 | 2007-06-17 18:37:45 +0200 | [diff] [blame] | 6644 | do_each_thread(g, p) { |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6645 | p->se.fair_key = 0; |
| 6646 | p->se.wait_runtime = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6647 | p->se.exec_start = 0; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 6648 | p->se.wait_start_fair = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6649 | p->se.sleep_start_fair = 0; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 6650 | #ifdef CONFIG_SCHEDSTATS |
| 6651 | p->se.wait_start = 0; |
| 6652 | p->se.sleep_start = 0; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6653 | p->se.block_start = 0; |
Ingo Molnar | 6cfb0d5 | 2007-08-02 17:41:40 +0200 | [diff] [blame] | 6654 | #endif |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6655 | task_rq(p)->cfs.fair_clock = 0; |
| 6656 | task_rq(p)->clock = 0; |
| 6657 | |
| 6658 | if (!rt_task(p)) { |
| 6659 | /* |
| 6660 | * Renice negative nice level userspace |
| 6661 | * tasks back to 0: |
| 6662 | */ |
| 6663 | if (TASK_NICE(p) < 0 && p->mm) |
| 6664 | set_user_nice(p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6665 | continue; |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6666 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6667 | |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 6668 | spin_lock_irqsave(&p->pi_lock, flags); |
| 6669 | rq = __task_rq_lock(p); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6670 | #ifdef CONFIG_SMP |
| 6671 | /* |
| 6672 | * Do not touch the migration thread: |
| 6673 | */ |
| 6674 | if (p == rq->migration_thread) |
| 6675 | goto out_unlock; |
| 6676 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6677 | |
Ingo Molnar | 2daa357 | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 6678 | update_rq_clock(rq); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6679 | on_rq = p->se.on_rq; |
Ingo Molnar | 2daa357 | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 6680 | if (on_rq) |
| 6681 | deactivate_task(rq, p, 0); |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6682 | __setscheduler(rq, p, SCHED_NORMAL, 0); |
| 6683 | if (on_rq) { |
Ingo Molnar | 2daa357 | 2007-08-09 11:16:51 +0200 | [diff] [blame] | 6684 | activate_task(rq, p, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6685 | resched_task(rq->curr); |
| 6686 | } |
Ingo Molnar | dd41f59 | 2007-07-09 18:51:59 +0200 | [diff] [blame] | 6687 | #ifdef CONFIG_SMP |
| 6688 | out_unlock: |
| 6689 | #endif |
Ingo Molnar | b29739f | 2006-06-27 02:54:51 -0700 | [diff] [blame] | 6690 | __task_rq_unlock(rq); |
| 6691 | spin_unlock_irqrestore(&p->pi_lock, flags); |
Ingo Molnar | a0f98a1 | 2007-06-17 18:37:45 +0200 | [diff] [blame] | 6692 | } while_each_thread(g, p); |
| 6693 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 6694 | read_unlock_irq(&tasklist_lock); |
| 6695 | } |
| 6696 | |
| 6697 | #endif /* CONFIG_MAGIC_SYSRQ */ |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 6698 | |
| 6699 | #ifdef CONFIG_IA64 |
| 6700 | /* |
| 6701 | * These functions are only useful for the IA64 MCA handling. |
| 6702 | * |
| 6703 | * They can only be called when the whole system has been |
| 6704 | * stopped - every CPU needs to be quiescent, and no scheduling |
| 6705 | * activity can take place. Using them for anything else would |
| 6706 | * be a serious bug, and as a result, they aren't even visible |
| 6707 | * under any other configuration. |
| 6708 | */ |
| 6709 | |
| 6710 | /** |
| 6711 | * curr_task - return the current task for a given cpu. |
| 6712 | * @cpu: the processor in question. |
| 6713 | * |
| 6714 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! |
| 6715 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 6716 | struct task_struct *curr_task(int cpu) |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 6717 | { |
| 6718 | return cpu_curr(cpu); |
| 6719 | } |
| 6720 | |
| 6721 | /** |
| 6722 | * set_curr_task - set the current task for a given cpu. |
| 6723 | * @cpu: the processor in question. |
| 6724 | * @p: the task pointer to set. |
| 6725 | * |
| 6726 | * Description: This function must only be used when non-maskable interrupts |
| 6727 | * are serviced on a separate stack. It allows the architecture to switch the |
| 6728 | * notion of the current task on a cpu in a non-blocking manner. This function |
| 6729 | * must be called with all CPU's synchronized, and interrupts disabled, the |
| 6730 | * and caller must save the original value of the current task (see |
| 6731 | * curr_task() above) and restore that value before reenabling interrupts and |
| 6732 | * re-starting the system. |
| 6733 | * |
| 6734 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! |
| 6735 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 6736 | void set_curr_task(int cpu, struct task_struct *p) |
Linus Torvalds | 1df5c10 | 2005-09-12 07:59:21 -0700 | [diff] [blame] | 6737 | { |
| 6738 | cpu_curr(cpu) = p; |
| 6739 | } |
| 6740 | |
| 6741 | #endif |