1b2441318SGreg Kroah-Hartman /* SPDX-License-Identifier: GPL-2.0 */ 297fb7a0aSIngo Molnar /* 397fb7a0aSIngo Molnar * Scheduler internal types and methods: 497fb7a0aSIngo Molnar */ 5391e43daSPeter Zijlstra #include <linux/sched.h> 6325ea10cSIngo Molnar 7dfc3401aSIngo Molnar #include <linux/sched/autogroup.h> 8e6017571SIngo Molnar #include <linux/sched/clock.h> 9325ea10cSIngo Molnar #include <linux/sched/coredump.h> 1055687da1SIngo Molnar #include <linux/sched/cpufreq.h> 11325ea10cSIngo Molnar #include <linux/sched/cputime.h> 12325ea10cSIngo Molnar #include <linux/sched/deadline.h> 13b17b0153SIngo Molnar #include <linux/sched/debug.h> 14ef8bd77fSIngo Molnar #include <linux/sched/hotplug.h> 15325ea10cSIngo Molnar #include <linux/sched/idle.h> 16325ea10cSIngo Molnar #include <linux/sched/init.h> 17325ea10cSIngo Molnar #include <linux/sched/isolation.h> 18325ea10cSIngo Molnar #include <linux/sched/jobctl.h> 19325ea10cSIngo Molnar #include <linux/sched/loadavg.h> 20325ea10cSIngo Molnar #include <linux/sched/mm.h> 21325ea10cSIngo Molnar #include <linux/sched/nohz.h> 22325ea10cSIngo Molnar #include <linux/sched/numa_balancing.h> 23325ea10cSIngo Molnar #include <linux/sched/prio.h> 24325ea10cSIngo Molnar #include <linux/sched/rt.h> 25325ea10cSIngo Molnar #include <linux/sched/signal.h> 26325ea10cSIngo Molnar #include <linux/sched/stat.h> 27325ea10cSIngo Molnar #include <linux/sched/sysctl.h> 2829930025SIngo Molnar #include <linux/sched/task.h> 2968db0cf1SIngo Molnar #include <linux/sched/task_stack.h> 30325ea10cSIngo Molnar #include <linux/sched/topology.h> 31325ea10cSIngo Molnar #include <linux/sched/user.h> 32325ea10cSIngo Molnar #include <linux/sched/wake_q.h> 33325ea10cSIngo Molnar #include <linux/sched/xacct.h> 34ef8bd77fSIngo Molnar 35325ea10cSIngo Molnar #include <uapi/linux/sched/types.h> 36325ea10cSIngo Molnar 373866e845SSteven Rostedt (Red Hat) #include <linux/binfmts.h> 38325ea10cSIngo Molnar #include <linux/blkdev.h> 39325ea10cSIngo Molnar #include <linux/compat.h> 40325ea10cSIngo Molnar #include <linux/context_tracking.h> 41325ea10cSIngo Molnar #include <linux/cpufreq.h> 42325ea10cSIngo Molnar #include <linux/cpuidle.h> 43325ea10cSIngo Molnar #include <linux/cpuset.h> 44325ea10cSIngo Molnar #include <linux/ctype.h> 45325ea10cSIngo Molnar #include <linux/debugfs.h> 46325ea10cSIngo Molnar #include <linux/delayacct.h> 47325ea10cSIngo Molnar #include <linux/init_task.h> 48325ea10cSIngo Molnar #include <linux/kprobes.h> 49325ea10cSIngo Molnar #include <linux/kthread.h> 50325ea10cSIngo Molnar #include <linux/membarrier.h> 51325ea10cSIngo Molnar #include <linux/migrate.h> 52325ea10cSIngo Molnar #include <linux/mmu_context.h> 53325ea10cSIngo Molnar #include <linux/nmi.h> 54325ea10cSIngo Molnar #include <linux/proc_fs.h> 55325ea10cSIngo Molnar #include <linux/prefetch.h> 56325ea10cSIngo Molnar #include <linux/profile.h> 57325ea10cSIngo Molnar #include <linux/rcupdate_wait.h> 58325ea10cSIngo Molnar #include <linux/security.h> 59325ea10cSIngo Molnar #include <linux/stackprotector.h> 60391e43daSPeter Zijlstra #include <linux/stop_machine.h> 61325ea10cSIngo Molnar #include <linux/suspend.h> 62325ea10cSIngo Molnar #include <linux/swait.h> 63325ea10cSIngo Molnar #include <linux/syscalls.h> 64325ea10cSIngo Molnar #include <linux/task_work.h> 65325ea10cSIngo Molnar #include <linux/tsacct_kern.h> 66325ea10cSIngo Molnar 67325ea10cSIngo Molnar #include <asm/tlb.h> 68391e43daSPeter Zijlstra 697fce777cSIngo Molnar #ifdef CONFIG_PARAVIRT 707fce777cSIngo Molnar # include <asm/paravirt.h> 717fce777cSIngo Molnar #endif 727fce777cSIngo Molnar 73391e43daSPeter Zijlstra #include "cpupri.h" 746bfd6d72SJuri Lelli #include "cpudeadline.h" 75391e43daSPeter Zijlstra 769148a3a1SPeter Zijlstra #ifdef CONFIG_SCHED_DEBUG 779148a3a1SPeter Zijlstra # define SCHED_WARN_ON(x) WARN_ONCE(x, #x) 789148a3a1SPeter Zijlstra #else 796d3aed3dSIngo Molnar # define SCHED_WARN_ON(x) ({ (void)(x), 0; }) 809148a3a1SPeter Zijlstra #endif 819148a3a1SPeter Zijlstra 8245ceebf7SPaul Gortmaker struct rq; 83442bf3aaSDaniel Lezcano struct cpuidle_state; 8445ceebf7SPaul Gortmaker 85da0c1e65SKirill Tkhai /* task_struct::on_rq states: */ 86da0c1e65SKirill Tkhai #define TASK_ON_RQ_QUEUED 1 87cca26e80SKirill Tkhai #define TASK_ON_RQ_MIGRATING 2 88da0c1e65SKirill Tkhai 89391e43daSPeter Zijlstra extern __read_mostly int scheduler_running; 90391e43daSPeter Zijlstra 9145ceebf7SPaul Gortmaker extern unsigned long calc_load_update; 9245ceebf7SPaul Gortmaker extern atomic_long_t calc_load_tasks; 9345ceebf7SPaul Gortmaker 943289bdb4SPeter Zijlstra extern void calc_global_load_tick(struct rq *this_rq); 95d60585c5SThomas Gleixner extern long calc_load_fold_active(struct rq *this_rq, long adjust); 963289bdb4SPeter Zijlstra 973289bdb4SPeter Zijlstra #ifdef CONFIG_SMP 98cee1afceSFrederic Weisbecker extern void cpu_load_update_active(struct rq *this_rq); 993289bdb4SPeter Zijlstra #else 100cee1afceSFrederic Weisbecker static inline void cpu_load_update_active(struct rq *this_rq) { } 1013289bdb4SPeter Zijlstra #endif 10245ceebf7SPaul Gortmaker 103391e43daSPeter Zijlstra /* 104391e43daSPeter Zijlstra * Helpers for converting nanosecond timing to jiffy resolution 105391e43daSPeter Zijlstra */ 106391e43daSPeter Zijlstra #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) 107391e43daSPeter Zijlstra 108cc1f4b1fSLi Zefan /* 109cc1f4b1fSLi Zefan * Increase resolution of nice-level calculations for 64-bit architectures. 110cc1f4b1fSLi Zefan * The extra resolution improves shares distribution and load balancing of 111cc1f4b1fSLi Zefan * low-weight task groups (eg. nice +19 on an autogroup), deeper taskgroup 112cc1f4b1fSLi Zefan * hierarchies, especially on larger systems. This is not a user-visible change 113cc1f4b1fSLi Zefan * and does not change the user-interface for setting shares/weights. 114cc1f4b1fSLi Zefan * 115cc1f4b1fSLi Zefan * We increase resolution only if we have enough bits to allow this increased 11697fb7a0aSIngo Molnar * resolution (i.e. 64-bit). The costs for increasing resolution when 32-bit 11797fb7a0aSIngo Molnar * are pretty high and the returns do not justify the increased costs. 1182159197dSPeter Zijlstra * 11997fb7a0aSIngo Molnar * Really only required when CONFIG_FAIR_GROUP_SCHED=y is also set, but to 12097fb7a0aSIngo Molnar * increase coverage and consistency always enable it on 64-bit platforms. 121cc1f4b1fSLi Zefan */ 1222159197dSPeter Zijlstra #ifdef CONFIG_64BIT 123172895e6SYuyang Du # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT + SCHED_FIXEDPOINT_SHIFT) 1246ecdd749SYuyang Du # define scale_load(w) ((w) << SCHED_FIXEDPOINT_SHIFT) 1256ecdd749SYuyang Du # define scale_load_down(w) ((w) >> SCHED_FIXEDPOINT_SHIFT) 126cc1f4b1fSLi Zefan #else 127172895e6SYuyang Du # define NICE_0_LOAD_SHIFT (SCHED_FIXEDPOINT_SHIFT) 128cc1f4b1fSLi Zefan # define scale_load(w) (w) 129cc1f4b1fSLi Zefan # define scale_load_down(w) (w) 130cc1f4b1fSLi Zefan #endif 131cc1f4b1fSLi Zefan 1326ecdd749SYuyang Du /* 133172895e6SYuyang Du * Task weight (visible to users) and its load (invisible to users) have 134172895e6SYuyang Du * independent resolution, but they should be well calibrated. We use 135172895e6SYuyang Du * scale_load() and scale_load_down(w) to convert between them. The 136172895e6SYuyang Du * following must be true: 137172895e6SYuyang Du * 138172895e6SYuyang Du * scale_load(sched_prio_to_weight[USER_PRIO(NICE_TO_PRIO(0))]) == NICE_0_LOAD 139172895e6SYuyang Du * 1406ecdd749SYuyang Du */ 141172895e6SYuyang Du #define NICE_0_LOAD (1L << NICE_0_LOAD_SHIFT) 142391e43daSPeter Zijlstra 143391e43daSPeter Zijlstra /* 144332ac17eSDario Faggioli * Single value that decides SCHED_DEADLINE internal math precision. 145332ac17eSDario Faggioli * 10 -> just above 1us 146332ac17eSDario Faggioli * 9 -> just above 0.5us 147332ac17eSDario Faggioli */ 14897fb7a0aSIngo Molnar #define DL_SCALE 10 149332ac17eSDario Faggioli 150332ac17eSDario Faggioli /* 15197fb7a0aSIngo Molnar * Single value that denotes runtime == period, ie unlimited time. 152391e43daSPeter Zijlstra */ 153391e43daSPeter Zijlstra #define RUNTIME_INF ((u64)~0ULL) 154391e43daSPeter Zijlstra 15520f9cd2aSHenrik Austad static inline int idle_policy(int policy) 15620f9cd2aSHenrik Austad { 15720f9cd2aSHenrik Austad return policy == SCHED_IDLE; 15820f9cd2aSHenrik Austad } 159d50dde5aSDario Faggioli static inline int fair_policy(int policy) 160d50dde5aSDario Faggioli { 161d50dde5aSDario Faggioli return policy == SCHED_NORMAL || policy == SCHED_BATCH; 162d50dde5aSDario Faggioli } 163d50dde5aSDario Faggioli 164391e43daSPeter Zijlstra static inline int rt_policy(int policy) 165391e43daSPeter Zijlstra { 166d50dde5aSDario Faggioli return policy == SCHED_FIFO || policy == SCHED_RR; 167391e43daSPeter Zijlstra } 168391e43daSPeter Zijlstra 169aab03e05SDario Faggioli static inline int dl_policy(int policy) 170aab03e05SDario Faggioli { 171aab03e05SDario Faggioli return policy == SCHED_DEADLINE; 172aab03e05SDario Faggioli } 17320f9cd2aSHenrik Austad static inline bool valid_policy(int policy) 17420f9cd2aSHenrik Austad { 17520f9cd2aSHenrik Austad return idle_policy(policy) || fair_policy(policy) || 17620f9cd2aSHenrik Austad rt_policy(policy) || dl_policy(policy); 17720f9cd2aSHenrik Austad } 178aab03e05SDario Faggioli 179391e43daSPeter Zijlstra static inline int task_has_rt_policy(struct task_struct *p) 180391e43daSPeter Zijlstra { 181391e43daSPeter Zijlstra return rt_policy(p->policy); 182391e43daSPeter Zijlstra } 183391e43daSPeter Zijlstra 184aab03e05SDario Faggioli static inline int task_has_dl_policy(struct task_struct *p) 185aab03e05SDario Faggioli { 186aab03e05SDario Faggioli return dl_policy(p->policy); 187aab03e05SDario Faggioli } 188aab03e05SDario Faggioli 18907881166SJuri Lelli #define cap_scale(v, s) ((v)*(s) >> SCHED_CAPACITY_SHIFT) 19007881166SJuri Lelli 1912d3d891dSDario Faggioli /* 192794a56ebSJuri Lelli * !! For sched_setattr_nocheck() (kernel) only !! 193794a56ebSJuri Lelli * 194794a56ebSJuri Lelli * This is actually gross. :( 195794a56ebSJuri Lelli * 196794a56ebSJuri Lelli * It is used to make schedutil kworker(s) higher priority than SCHED_DEADLINE 197794a56ebSJuri Lelli * tasks, but still be able to sleep. We need this on platforms that cannot 198794a56ebSJuri Lelli * atomically change clock frequency. Remove once fast switching will be 199794a56ebSJuri Lelli * available on such platforms. 200794a56ebSJuri Lelli * 201794a56ebSJuri Lelli * SUGOV stands for SchedUtil GOVernor. 202794a56ebSJuri Lelli */ 203794a56ebSJuri Lelli #define SCHED_FLAG_SUGOV 0x10000000 204794a56ebSJuri Lelli 205794a56ebSJuri Lelli static inline bool dl_entity_is_special(struct sched_dl_entity *dl_se) 206794a56ebSJuri Lelli { 207794a56ebSJuri Lelli #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 208794a56ebSJuri Lelli return unlikely(dl_se->flags & SCHED_FLAG_SUGOV); 209794a56ebSJuri Lelli #else 210794a56ebSJuri Lelli return false; 211794a56ebSJuri Lelli #endif 212794a56ebSJuri Lelli } 213794a56ebSJuri Lelli 214794a56ebSJuri Lelli /* 2152d3d891dSDario Faggioli * Tells if entity @a should preempt entity @b. 2162d3d891dSDario Faggioli */ 217332ac17eSDario Faggioli static inline bool 218332ac17eSDario Faggioli dl_entity_preempt(struct sched_dl_entity *a, struct sched_dl_entity *b) 2192d3d891dSDario Faggioli { 220794a56ebSJuri Lelli return dl_entity_is_special(a) || 221794a56ebSJuri Lelli dl_time_before(a->deadline, b->deadline); 2222d3d891dSDario Faggioli } 2232d3d891dSDario Faggioli 224391e43daSPeter Zijlstra /* 225391e43daSPeter Zijlstra * This is the priority-queue data structure of the RT scheduling class: 226391e43daSPeter Zijlstra */ 227391e43daSPeter Zijlstra struct rt_prio_array { 228391e43daSPeter Zijlstra DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ 229391e43daSPeter Zijlstra struct list_head queue[MAX_RT_PRIO]; 230391e43daSPeter Zijlstra }; 231391e43daSPeter Zijlstra 232391e43daSPeter Zijlstra struct rt_bandwidth { 233391e43daSPeter Zijlstra /* nests inside the rq lock: */ 234391e43daSPeter Zijlstra raw_spinlock_t rt_runtime_lock; 235391e43daSPeter Zijlstra ktime_t rt_period; 236391e43daSPeter Zijlstra u64 rt_runtime; 237391e43daSPeter Zijlstra struct hrtimer rt_period_timer; 2384cfafd30SPeter Zijlstra unsigned int rt_period_active; 239391e43daSPeter Zijlstra }; 240a5e7be3bSJuri Lelli 241a5e7be3bSJuri Lelli void __dl_clear_params(struct task_struct *p); 242a5e7be3bSJuri Lelli 243332ac17eSDario Faggioli /* 244332ac17eSDario Faggioli * To keep the bandwidth of -deadline tasks and groups under control 245332ac17eSDario Faggioli * we need some place where: 246332ac17eSDario Faggioli * - store the maximum -deadline bandwidth of the system (the group); 247332ac17eSDario Faggioli * - cache the fraction of that bandwidth that is currently allocated. 248332ac17eSDario Faggioli * 249332ac17eSDario Faggioli * This is all done in the data structure below. It is similar to the 250332ac17eSDario Faggioli * one used for RT-throttling (rt_bandwidth), with the main difference 251332ac17eSDario Faggioli * that, since here we are only interested in admission control, we 252332ac17eSDario Faggioli * do not decrease any runtime while the group "executes", neither we 253332ac17eSDario Faggioli * need a timer to replenish it. 254332ac17eSDario Faggioli * 255332ac17eSDario Faggioli * With respect to SMP, the bandwidth is given on a per-CPU basis, 256332ac17eSDario Faggioli * meaning that: 257332ac17eSDario Faggioli * - dl_bw (< 100%) is the bandwidth of the system (group) on each CPU; 258332ac17eSDario Faggioli * - dl_total_bw array contains, in the i-eth element, the currently 259332ac17eSDario Faggioli * allocated bandwidth on the i-eth CPU. 260332ac17eSDario Faggioli * Moreover, groups consume bandwidth on each CPU, while tasks only 261332ac17eSDario Faggioli * consume bandwidth on the CPU they're running on. 262332ac17eSDario Faggioli * Finally, dl_total_bw_cpu is used to cache the index of dl_total_bw 263332ac17eSDario Faggioli * that will be shown the next time the proc or cgroup controls will 264332ac17eSDario Faggioli * be red. It on its turn can be changed by writing on its own 265332ac17eSDario Faggioli * control. 266332ac17eSDario Faggioli */ 267332ac17eSDario Faggioli struct dl_bandwidth { 268332ac17eSDario Faggioli raw_spinlock_t dl_runtime_lock; 269332ac17eSDario Faggioli u64 dl_runtime; 270332ac17eSDario Faggioli u64 dl_period; 271332ac17eSDario Faggioli }; 272332ac17eSDario Faggioli 273332ac17eSDario Faggioli static inline int dl_bandwidth_enabled(void) 274332ac17eSDario Faggioli { 2751724813dSPeter Zijlstra return sysctl_sched_rt_runtime >= 0; 276332ac17eSDario Faggioli } 277332ac17eSDario Faggioli 278332ac17eSDario Faggioli struct dl_bw { 279332ac17eSDario Faggioli raw_spinlock_t lock; 28097fb7a0aSIngo Molnar u64 bw; 28197fb7a0aSIngo Molnar u64 total_bw; 282332ac17eSDario Faggioli }; 283332ac17eSDario Faggioli 284daec5798SLuca Abeni static inline void __dl_update(struct dl_bw *dl_b, s64 bw); 285daec5798SLuca Abeni 2867f51412aSJuri Lelli static inline 2878c0944ceSPeter Zijlstra void __dl_sub(struct dl_bw *dl_b, u64 tsk_bw, int cpus) 2887f51412aSJuri Lelli { 2897f51412aSJuri Lelli dl_b->total_bw -= tsk_bw; 290daec5798SLuca Abeni __dl_update(dl_b, (s32)tsk_bw / cpus); 2917f51412aSJuri Lelli } 2927f51412aSJuri Lelli 2937f51412aSJuri Lelli static inline 294daec5798SLuca Abeni void __dl_add(struct dl_bw *dl_b, u64 tsk_bw, int cpus) 2957f51412aSJuri Lelli { 2967f51412aSJuri Lelli dl_b->total_bw += tsk_bw; 297daec5798SLuca Abeni __dl_update(dl_b, -((s32)tsk_bw / cpus)); 2987f51412aSJuri Lelli } 2997f51412aSJuri Lelli 3007f51412aSJuri Lelli static inline 3017f51412aSJuri Lelli bool __dl_overflow(struct dl_bw *dl_b, int cpus, u64 old_bw, u64 new_bw) 3027f51412aSJuri Lelli { 3037f51412aSJuri Lelli return dl_b->bw != -1 && 3047f51412aSJuri Lelli dl_b->bw * cpus < dl_b->total_bw - old_bw + new_bw; 3057f51412aSJuri Lelli } 3067f51412aSJuri Lelli 30797fb7a0aSIngo Molnar extern void dl_change_utilization(struct task_struct *p, u64 new_bw); 308f2cb1360SIngo Molnar extern void init_dl_bw(struct dl_bw *dl_b); 30906a76fe0SNicolas Pitre extern int sched_dl_global_validate(void); 31006a76fe0SNicolas Pitre extern void sched_dl_do_global(void); 31197fb7a0aSIngo Molnar extern int sched_dl_overflow(struct task_struct *p, int policy, const struct sched_attr *attr); 31206a76fe0SNicolas Pitre extern void __setparam_dl(struct task_struct *p, const struct sched_attr *attr); 31306a76fe0SNicolas Pitre extern void __getparam_dl(struct task_struct *p, struct sched_attr *attr); 31406a76fe0SNicolas Pitre extern bool __checkparam_dl(const struct sched_attr *attr); 31506a76fe0SNicolas Pitre extern bool dl_param_changed(struct task_struct *p, const struct sched_attr *attr); 31697fb7a0aSIngo Molnar extern int dl_task_can_attach(struct task_struct *p, const struct cpumask *cs_cpus_allowed); 31797fb7a0aSIngo Molnar extern int dl_cpuset_cpumask_can_shrink(const struct cpumask *cur, const struct cpumask *trial); 31806a76fe0SNicolas Pitre extern bool dl_cpu_busy(unsigned int cpu); 319391e43daSPeter Zijlstra 320391e43daSPeter Zijlstra #ifdef CONFIG_CGROUP_SCHED 321391e43daSPeter Zijlstra 322391e43daSPeter Zijlstra #include <linux/cgroup.h> 323391e43daSPeter Zijlstra 324391e43daSPeter Zijlstra struct cfs_rq; 325391e43daSPeter Zijlstra struct rt_rq; 326391e43daSPeter Zijlstra 32735cf4e50SMike Galbraith extern struct list_head task_groups; 328391e43daSPeter Zijlstra 329391e43daSPeter Zijlstra struct cfs_bandwidth { 330391e43daSPeter Zijlstra #ifdef CONFIG_CFS_BANDWIDTH 331391e43daSPeter Zijlstra raw_spinlock_t lock; 332391e43daSPeter Zijlstra ktime_t period; 33397fb7a0aSIngo Molnar u64 quota; 33497fb7a0aSIngo Molnar u64 runtime; 3359c58c79aSZhihui Zhang s64 hierarchical_quota; 336391e43daSPeter Zijlstra u64 runtime_expires; 337512ac999SXunlei Pang int expires_seq; 338391e43daSPeter Zijlstra 339512ac999SXunlei Pang short idle; 340512ac999SXunlei Pang short period_active; 34197fb7a0aSIngo Molnar struct hrtimer period_timer; 34297fb7a0aSIngo Molnar struct hrtimer slack_timer; 343391e43daSPeter Zijlstra struct list_head throttled_cfs_rq; 344391e43daSPeter Zijlstra 34597fb7a0aSIngo Molnar /* Statistics: */ 34697fb7a0aSIngo Molnar int nr_periods; 34797fb7a0aSIngo Molnar int nr_throttled; 348391e43daSPeter Zijlstra u64 throttled_time; 349391e43daSPeter Zijlstra #endif 350391e43daSPeter Zijlstra }; 351391e43daSPeter Zijlstra 35297fb7a0aSIngo Molnar /* Task group related information */ 353391e43daSPeter Zijlstra struct task_group { 354391e43daSPeter Zijlstra struct cgroup_subsys_state css; 355391e43daSPeter Zijlstra 356391e43daSPeter Zijlstra #ifdef CONFIG_FAIR_GROUP_SCHED 35797fb7a0aSIngo Molnar /* schedulable entities of this group on each CPU */ 358391e43daSPeter Zijlstra struct sched_entity **se; 35997fb7a0aSIngo Molnar /* runqueue "owned" by this group on each CPU */ 360391e43daSPeter Zijlstra struct cfs_rq **cfs_rq; 361391e43daSPeter Zijlstra unsigned long shares; 362391e43daSPeter Zijlstra 363fa6bddebSAlex Shi #ifdef CONFIG_SMP 364b0367629SWaiman Long /* 365b0367629SWaiman Long * load_avg can be heavily contended at clock tick time, so put 366b0367629SWaiman Long * it in its own cacheline separated from the fields above which 367b0367629SWaiman Long * will also be accessed at each tick. 368b0367629SWaiman Long */ 369b0367629SWaiman Long atomic_long_t load_avg ____cacheline_aligned; 370391e43daSPeter Zijlstra #endif 371fa6bddebSAlex Shi #endif 372391e43daSPeter Zijlstra 373391e43daSPeter Zijlstra #ifdef CONFIG_RT_GROUP_SCHED 374391e43daSPeter Zijlstra struct sched_rt_entity **rt_se; 375391e43daSPeter Zijlstra struct rt_rq **rt_rq; 376391e43daSPeter Zijlstra 377391e43daSPeter Zijlstra struct rt_bandwidth rt_bandwidth; 378391e43daSPeter Zijlstra #endif 379391e43daSPeter Zijlstra 380391e43daSPeter Zijlstra struct rcu_head rcu; 381391e43daSPeter Zijlstra struct list_head list; 382391e43daSPeter Zijlstra 383391e43daSPeter Zijlstra struct task_group *parent; 384391e43daSPeter Zijlstra struct list_head siblings; 385391e43daSPeter Zijlstra struct list_head children; 386391e43daSPeter Zijlstra 387391e43daSPeter Zijlstra #ifdef CONFIG_SCHED_AUTOGROUP 388391e43daSPeter Zijlstra struct autogroup *autogroup; 389391e43daSPeter Zijlstra #endif 390391e43daSPeter Zijlstra 391391e43daSPeter Zijlstra struct cfs_bandwidth cfs_bandwidth; 392391e43daSPeter Zijlstra }; 393391e43daSPeter Zijlstra 394391e43daSPeter Zijlstra #ifdef CONFIG_FAIR_GROUP_SCHED 395391e43daSPeter Zijlstra #define ROOT_TASK_GROUP_LOAD NICE_0_LOAD 396391e43daSPeter Zijlstra 397391e43daSPeter Zijlstra /* 398391e43daSPeter Zijlstra * A weight of 0 or 1 can cause arithmetics problems. 399391e43daSPeter Zijlstra * A weight of a cfs_rq is the sum of weights of which entities 400391e43daSPeter Zijlstra * are queued on this cfs_rq, so a weight of a entity should not be 401391e43daSPeter Zijlstra * too large, so as the shares value of a task group. 402391e43daSPeter Zijlstra * (The default weight is 1024 - so there's no practical 403391e43daSPeter Zijlstra * limitation from this.) 404391e43daSPeter Zijlstra */ 405391e43daSPeter Zijlstra #define MIN_SHARES (1UL << 1) 406391e43daSPeter Zijlstra #define MAX_SHARES (1UL << 18) 407391e43daSPeter Zijlstra #endif 408391e43daSPeter Zijlstra 409391e43daSPeter Zijlstra typedef int (*tg_visitor)(struct task_group *, void *); 410391e43daSPeter Zijlstra 411391e43daSPeter Zijlstra extern int walk_tg_tree_from(struct task_group *from, 412391e43daSPeter Zijlstra tg_visitor down, tg_visitor up, void *data); 413391e43daSPeter Zijlstra 414391e43daSPeter Zijlstra /* 415391e43daSPeter Zijlstra * Iterate the full tree, calling @down when first entering a node and @up when 416391e43daSPeter Zijlstra * leaving it for the final time. 417391e43daSPeter Zijlstra * 418391e43daSPeter Zijlstra * Caller must hold rcu_lock or sufficient equivalent. 419391e43daSPeter Zijlstra */ 420391e43daSPeter Zijlstra static inline int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) 421391e43daSPeter Zijlstra { 422391e43daSPeter Zijlstra return walk_tg_tree_from(&root_task_group, down, up, data); 423391e43daSPeter Zijlstra } 424391e43daSPeter Zijlstra 425391e43daSPeter Zijlstra extern int tg_nop(struct task_group *tg, void *data); 426391e43daSPeter Zijlstra 427391e43daSPeter Zijlstra extern void free_fair_sched_group(struct task_group *tg); 428391e43daSPeter Zijlstra extern int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent); 4298663e24dSPeter Zijlstra extern void online_fair_sched_group(struct task_group *tg); 4306fe1f348SPeter Zijlstra extern void unregister_fair_sched_group(struct task_group *tg); 431391e43daSPeter Zijlstra extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, 432391e43daSPeter Zijlstra struct sched_entity *se, int cpu, 433391e43daSPeter Zijlstra struct sched_entity *parent); 434391e43daSPeter Zijlstra extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 435391e43daSPeter Zijlstra 436391e43daSPeter Zijlstra extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b); 43777a4d1a1SPeter Zijlstra extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b); 438391e43daSPeter Zijlstra extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq); 439391e43daSPeter Zijlstra 440391e43daSPeter Zijlstra extern void free_rt_sched_group(struct task_group *tg); 441391e43daSPeter Zijlstra extern int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent); 442391e43daSPeter Zijlstra extern void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, 443391e43daSPeter Zijlstra struct sched_rt_entity *rt_se, int cpu, 444391e43daSPeter Zijlstra struct sched_rt_entity *parent); 4458887cd99SNicolas Pitre extern int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us); 4468887cd99SNicolas Pitre extern int sched_group_set_rt_period(struct task_group *tg, u64 rt_period_us); 4478887cd99SNicolas Pitre extern long sched_group_rt_runtime(struct task_group *tg); 4488887cd99SNicolas Pitre extern long sched_group_rt_period(struct task_group *tg); 4498887cd99SNicolas Pitre extern int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk); 450391e43daSPeter Zijlstra 45125cc7da7SLi Zefan extern struct task_group *sched_create_group(struct task_group *parent); 45225cc7da7SLi Zefan extern void sched_online_group(struct task_group *tg, 45325cc7da7SLi Zefan struct task_group *parent); 45425cc7da7SLi Zefan extern void sched_destroy_group(struct task_group *tg); 45525cc7da7SLi Zefan extern void sched_offline_group(struct task_group *tg); 45625cc7da7SLi Zefan 45725cc7da7SLi Zefan extern void sched_move_task(struct task_struct *tsk); 45825cc7da7SLi Zefan 45925cc7da7SLi Zefan #ifdef CONFIG_FAIR_GROUP_SCHED 46025cc7da7SLi Zefan extern int sched_group_set_shares(struct task_group *tg, unsigned long shares); 461ad936d86SByungchul Park 462ad936d86SByungchul Park #ifdef CONFIG_SMP 463ad936d86SByungchul Park extern void set_task_rq_fair(struct sched_entity *se, 464ad936d86SByungchul Park struct cfs_rq *prev, struct cfs_rq *next); 465ad936d86SByungchul Park #else /* !CONFIG_SMP */ 466ad936d86SByungchul Park static inline void set_task_rq_fair(struct sched_entity *se, 467ad936d86SByungchul Park struct cfs_rq *prev, struct cfs_rq *next) { } 468ad936d86SByungchul Park #endif /* CONFIG_SMP */ 469ad936d86SByungchul Park #endif /* CONFIG_FAIR_GROUP_SCHED */ 47025cc7da7SLi Zefan 471391e43daSPeter Zijlstra #else /* CONFIG_CGROUP_SCHED */ 472391e43daSPeter Zijlstra 473391e43daSPeter Zijlstra struct cfs_bandwidth { }; 474391e43daSPeter Zijlstra 475391e43daSPeter Zijlstra #endif /* CONFIG_CGROUP_SCHED */ 476391e43daSPeter Zijlstra 477391e43daSPeter Zijlstra /* CFS-related fields in a runqueue */ 478391e43daSPeter Zijlstra struct cfs_rq { 479391e43daSPeter Zijlstra struct load_weight load; 4801ea6c46aSPeter Zijlstra unsigned long runnable_weight; 48197fb7a0aSIngo Molnar unsigned int nr_running; 48297fb7a0aSIngo Molnar unsigned int h_nr_running; 483391e43daSPeter Zijlstra 484391e43daSPeter Zijlstra u64 exec_clock; 485391e43daSPeter Zijlstra u64 min_vruntime; 486391e43daSPeter Zijlstra #ifndef CONFIG_64BIT 487391e43daSPeter Zijlstra u64 min_vruntime_copy; 488391e43daSPeter Zijlstra #endif 489391e43daSPeter Zijlstra 490bfb06889SDavidlohr Bueso struct rb_root_cached tasks_timeline; 491391e43daSPeter Zijlstra 492391e43daSPeter Zijlstra /* 493391e43daSPeter Zijlstra * 'curr' points to currently running entity on this cfs_rq. 494391e43daSPeter Zijlstra * It is set to NULL otherwise (i.e when none are currently running). 495391e43daSPeter Zijlstra */ 49697fb7a0aSIngo Molnar struct sched_entity *curr; 49797fb7a0aSIngo Molnar struct sched_entity *next; 49897fb7a0aSIngo Molnar struct sched_entity *last; 49997fb7a0aSIngo Molnar struct sched_entity *skip; 500391e43daSPeter Zijlstra 501391e43daSPeter Zijlstra #ifdef CONFIG_SCHED_DEBUG 502391e43daSPeter Zijlstra unsigned int nr_spread_over; 503391e43daSPeter Zijlstra #endif 504391e43daSPeter Zijlstra 5052dac754eSPaul Turner #ifdef CONFIG_SMP 5062dac754eSPaul Turner /* 5079d89c257SYuyang Du * CFS load tracking 5082dac754eSPaul Turner */ 5099d89c257SYuyang Du struct sched_avg avg; 5102a2f5d4eSPeter Zijlstra #ifndef CONFIG_64BIT 5112a2f5d4eSPeter Zijlstra u64 load_last_update_time_copy; 5122a2f5d4eSPeter Zijlstra #endif 5132a2f5d4eSPeter Zijlstra struct { 5142a2f5d4eSPeter Zijlstra raw_spinlock_t lock ____cacheline_aligned; 5152a2f5d4eSPeter Zijlstra int nr; 5162a2f5d4eSPeter Zijlstra unsigned long load_avg; 5172a2f5d4eSPeter Zijlstra unsigned long util_avg; 5180e2d2aaaSPeter Zijlstra unsigned long runnable_sum; 5192a2f5d4eSPeter Zijlstra } removed; 520141965c7SAlex Shi 521c566e8e9SPaul Turner #ifdef CONFIG_FAIR_GROUP_SCHED 5220e2d2aaaSPeter Zijlstra unsigned long tg_load_avg_contrib; 5230e2d2aaaSPeter Zijlstra long propagate; 5240e2d2aaaSPeter Zijlstra long prop_runnable_sum; 5250e2d2aaaSPeter Zijlstra 52682958366SPaul Turner /* 52782958366SPaul Turner * h_load = weight * f(tg) 52882958366SPaul Turner * 52982958366SPaul Turner * Where f(tg) is the recursive weight fraction assigned to 53082958366SPaul Turner * this group. 53182958366SPaul Turner */ 53282958366SPaul Turner unsigned long h_load; 53368520796SVladimir Davydov u64 last_h_load_update; 53468520796SVladimir Davydov struct sched_entity *h_load_next; 53568520796SVladimir Davydov #endif /* CONFIG_FAIR_GROUP_SCHED */ 53682958366SPaul Turner #endif /* CONFIG_SMP */ 53782958366SPaul Turner 538391e43daSPeter Zijlstra #ifdef CONFIG_FAIR_GROUP_SCHED 53997fb7a0aSIngo Molnar struct rq *rq; /* CPU runqueue to which this cfs_rq is attached */ 540391e43daSPeter Zijlstra 541391e43daSPeter Zijlstra /* 542391e43daSPeter Zijlstra * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in 543391e43daSPeter Zijlstra * a hierarchy). Non-leaf lrqs hold other higher schedulable entities 544391e43daSPeter Zijlstra * (like users, containers etc.) 545391e43daSPeter Zijlstra * 54697fb7a0aSIngo Molnar * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a CPU. 54797fb7a0aSIngo Molnar * This list is used during load balance. 548391e43daSPeter Zijlstra */ 549391e43daSPeter Zijlstra int on_list; 550391e43daSPeter Zijlstra struct list_head leaf_cfs_rq_list; 551391e43daSPeter Zijlstra struct task_group *tg; /* group that "owns" this runqueue */ 552391e43daSPeter Zijlstra 553391e43daSPeter Zijlstra #ifdef CONFIG_CFS_BANDWIDTH 554391e43daSPeter Zijlstra int runtime_enabled; 555512ac999SXunlei Pang int expires_seq; 556391e43daSPeter Zijlstra u64 runtime_expires; 557391e43daSPeter Zijlstra s64 runtime_remaining; 558391e43daSPeter Zijlstra 55997fb7a0aSIngo Molnar u64 throttled_clock; 56097fb7a0aSIngo Molnar u64 throttled_clock_task; 561f1b17280SPaul Turner u64 throttled_clock_task_time; 56297fb7a0aSIngo Molnar int throttled; 56397fb7a0aSIngo Molnar int throttle_count; 564391e43daSPeter Zijlstra struct list_head throttled_list; 565391e43daSPeter Zijlstra #endif /* CONFIG_CFS_BANDWIDTH */ 566391e43daSPeter Zijlstra #endif /* CONFIG_FAIR_GROUP_SCHED */ 567391e43daSPeter Zijlstra }; 568391e43daSPeter Zijlstra 569391e43daSPeter Zijlstra static inline int rt_bandwidth_enabled(void) 570391e43daSPeter Zijlstra { 571391e43daSPeter Zijlstra return sysctl_sched_rt_runtime >= 0; 572391e43daSPeter Zijlstra } 573391e43daSPeter Zijlstra 574b6366f04SSteven Rostedt /* RT IPI pull logic requires IRQ_WORK */ 5754bdced5cSSteven Rostedt (Red Hat) #if defined(CONFIG_IRQ_WORK) && defined(CONFIG_SMP) 576b6366f04SSteven Rostedt # define HAVE_RT_PUSH_IPI 577b6366f04SSteven Rostedt #endif 578b6366f04SSteven Rostedt 579391e43daSPeter Zijlstra /* Real-Time classes' related field in a runqueue: */ 580391e43daSPeter Zijlstra struct rt_rq { 581391e43daSPeter Zijlstra struct rt_prio_array active; 582c82513e5SPeter Zijlstra unsigned int rt_nr_running; 58301d36d0aSFrederic Weisbecker unsigned int rr_nr_running; 584391e43daSPeter Zijlstra #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 585391e43daSPeter Zijlstra struct { 586391e43daSPeter Zijlstra int curr; /* highest queued rt task prio */ 587391e43daSPeter Zijlstra #ifdef CONFIG_SMP 588391e43daSPeter Zijlstra int next; /* next highest */ 589391e43daSPeter Zijlstra #endif 590391e43daSPeter Zijlstra } highest_prio; 591391e43daSPeter Zijlstra #endif 592391e43daSPeter Zijlstra #ifdef CONFIG_SMP 593391e43daSPeter Zijlstra unsigned long rt_nr_migratory; 594391e43daSPeter Zijlstra unsigned long rt_nr_total; 595391e43daSPeter Zijlstra int overloaded; 596391e43daSPeter Zijlstra struct plist_head pushable_tasks; 597371bf427SVincent Guittot 598b6366f04SSteven Rostedt #endif /* CONFIG_SMP */ 599f4ebcbc0SKirill Tkhai int rt_queued; 600f4ebcbc0SKirill Tkhai 601391e43daSPeter Zijlstra int rt_throttled; 602391e43daSPeter Zijlstra u64 rt_time; 603391e43daSPeter Zijlstra u64 rt_runtime; 604391e43daSPeter Zijlstra /* Nests inside the rq lock: */ 605391e43daSPeter Zijlstra raw_spinlock_t rt_runtime_lock; 606391e43daSPeter Zijlstra 607391e43daSPeter Zijlstra #ifdef CONFIG_RT_GROUP_SCHED 608391e43daSPeter Zijlstra unsigned long rt_nr_boosted; 609391e43daSPeter Zijlstra 610391e43daSPeter Zijlstra struct rq *rq; 611391e43daSPeter Zijlstra struct task_group *tg; 612391e43daSPeter Zijlstra #endif 613391e43daSPeter Zijlstra }; 614391e43daSPeter Zijlstra 615296b2ffeSVincent Guittot static inline bool rt_rq_is_runnable(struct rt_rq *rt_rq) 616296b2ffeSVincent Guittot { 617296b2ffeSVincent Guittot return rt_rq->rt_queued && rt_rq->rt_nr_running; 618296b2ffeSVincent Guittot } 619296b2ffeSVincent Guittot 620aab03e05SDario Faggioli /* Deadline class' related fields in a runqueue */ 621aab03e05SDario Faggioli struct dl_rq { 622aab03e05SDario Faggioli /* runqueue is an rbtree, ordered by deadline */ 6232161573eSDavidlohr Bueso struct rb_root_cached root; 624aab03e05SDario Faggioli 625aab03e05SDario Faggioli unsigned long dl_nr_running; 6261baca4ceSJuri Lelli 6271baca4ceSJuri Lelli #ifdef CONFIG_SMP 6281baca4ceSJuri Lelli /* 6291baca4ceSJuri Lelli * Deadline values of the currently executing and the 6301baca4ceSJuri Lelli * earliest ready task on this rq. Caching these facilitates 6311baca4ceSJuri Lelli * the decision wether or not a ready but not running task 6321baca4ceSJuri Lelli * should migrate somewhere else. 6331baca4ceSJuri Lelli */ 6341baca4ceSJuri Lelli struct { 6351baca4ceSJuri Lelli u64 curr; 6361baca4ceSJuri Lelli u64 next; 6371baca4ceSJuri Lelli } earliest_dl; 6381baca4ceSJuri Lelli 6391baca4ceSJuri Lelli unsigned long dl_nr_migratory; 6401baca4ceSJuri Lelli int overloaded; 6411baca4ceSJuri Lelli 6421baca4ceSJuri Lelli /* 6431baca4ceSJuri Lelli * Tasks on this rq that can be pushed away. They are kept in 6441baca4ceSJuri Lelli * an rb-tree, ordered by tasks' deadlines, with caching 6451baca4ceSJuri Lelli * of the leftmost (earliest deadline) element. 6461baca4ceSJuri Lelli */ 6472161573eSDavidlohr Bueso struct rb_root_cached pushable_dl_tasks_root; 648332ac17eSDario Faggioli #else 649332ac17eSDario Faggioli struct dl_bw dl_bw; 6501baca4ceSJuri Lelli #endif 651e36d8677SLuca Abeni /* 652e36d8677SLuca Abeni * "Active utilization" for this runqueue: increased when a 653e36d8677SLuca Abeni * task wakes up (becomes TASK_RUNNING) and decreased when a 654e36d8677SLuca Abeni * task blocks 655e36d8677SLuca Abeni */ 656e36d8677SLuca Abeni u64 running_bw; 6574da3abceSLuca Abeni 6584da3abceSLuca Abeni /* 6598fd27231SLuca Abeni * Utilization of the tasks "assigned" to this runqueue (including 6608fd27231SLuca Abeni * the tasks that are in runqueue and the tasks that executed on this 6618fd27231SLuca Abeni * CPU and blocked). Increased when a task moves to this runqueue, and 6628fd27231SLuca Abeni * decreased when the task moves away (migrates, changes scheduling 6638fd27231SLuca Abeni * policy, or terminates). 6648fd27231SLuca Abeni * This is needed to compute the "inactive utilization" for the 6658fd27231SLuca Abeni * runqueue (inactive utilization = this_bw - running_bw). 6668fd27231SLuca Abeni */ 6678fd27231SLuca Abeni u64 this_bw; 668daec5798SLuca Abeni u64 extra_bw; 6698fd27231SLuca Abeni 6708fd27231SLuca Abeni /* 6714da3abceSLuca Abeni * Inverse of the fraction of CPU utilization that can be reclaimed 6724da3abceSLuca Abeni * by the GRUB algorithm. 6734da3abceSLuca Abeni */ 6744da3abceSLuca Abeni u64 bw_ratio; 675aab03e05SDario Faggioli }; 676aab03e05SDario Faggioli 677c0796298SVincent Guittot #ifdef CONFIG_FAIR_GROUP_SCHED 678c0796298SVincent Guittot /* An entity is a task if it doesn't "own" a runqueue */ 679c0796298SVincent Guittot #define entity_is_task(se) (!se->my_q) 680c0796298SVincent Guittot #else 681c0796298SVincent Guittot #define entity_is_task(se) 1 682c0796298SVincent Guittot #endif 683c0796298SVincent Guittot 684391e43daSPeter Zijlstra #ifdef CONFIG_SMP 685c0796298SVincent Guittot /* 686c0796298SVincent Guittot * XXX we want to get rid of these helpers and use the full load resolution. 687c0796298SVincent Guittot */ 688c0796298SVincent Guittot static inline long se_weight(struct sched_entity *se) 689c0796298SVincent Guittot { 690c0796298SVincent Guittot return scale_load_down(se->load.weight); 691c0796298SVincent Guittot } 692c0796298SVincent Guittot 693c0796298SVincent Guittot static inline long se_runnable(struct sched_entity *se) 694c0796298SVincent Guittot { 695c0796298SVincent Guittot return scale_load_down(se->runnable_weight); 696c0796298SVincent Guittot } 697391e43daSPeter Zijlstra 698afe06efdSTim Chen static inline bool sched_asym_prefer(int a, int b) 699afe06efdSTim Chen { 700afe06efdSTim Chen return arch_asym_cpu_priority(a) > arch_asym_cpu_priority(b); 701afe06efdSTim Chen } 702afe06efdSTim Chen 703391e43daSPeter Zijlstra /* 704391e43daSPeter Zijlstra * We add the notion of a root-domain which will be used to define per-domain 705391e43daSPeter Zijlstra * variables. Each exclusive cpuset essentially defines an island domain by 70697fb7a0aSIngo Molnar * fully partitioning the member CPUs from any other cpuset. Whenever a new 707391e43daSPeter Zijlstra * exclusive cpuset is created, we also create and attach a new root-domain 708391e43daSPeter Zijlstra * object. 709391e43daSPeter Zijlstra * 710391e43daSPeter Zijlstra */ 711391e43daSPeter Zijlstra struct root_domain { 712391e43daSPeter Zijlstra atomic_t refcount; 713391e43daSPeter Zijlstra atomic_t rto_count; 714391e43daSPeter Zijlstra struct rcu_head rcu; 715391e43daSPeter Zijlstra cpumask_var_t span; 716391e43daSPeter Zijlstra cpumask_var_t online; 717391e43daSPeter Zijlstra 7184486edd1STim Chen /* Indicate more than one runnable task for any CPU */ 719*575638d1SValentin Schneider int overload; 7204486edd1STim Chen 721391e43daSPeter Zijlstra /* 7221baca4ceSJuri Lelli * The bit corresponding to a CPU gets set here if such CPU has more 7231baca4ceSJuri Lelli * than one runnable -deadline task (as it is below for RT tasks). 7241baca4ceSJuri Lelli */ 7251baca4ceSJuri Lelli cpumask_var_t dlo_mask; 7261baca4ceSJuri Lelli atomic_t dlo_count; 727332ac17eSDario Faggioli struct dl_bw dl_bw; 7286bfd6d72SJuri Lelli struct cpudl cpudl; 7291baca4ceSJuri Lelli 7304bdced5cSSteven Rostedt (Red Hat) #ifdef HAVE_RT_PUSH_IPI 7314bdced5cSSteven Rostedt (Red Hat) /* 7324bdced5cSSteven Rostedt (Red Hat) * For IPI pull requests, loop across the rto_mask. 7334bdced5cSSteven Rostedt (Red Hat) */ 7344bdced5cSSteven Rostedt (Red Hat) struct irq_work rto_push_work; 7354bdced5cSSteven Rostedt (Red Hat) raw_spinlock_t rto_lock; 7364bdced5cSSteven Rostedt (Red Hat) /* These are only updated and read within rto_lock */ 7374bdced5cSSteven Rostedt (Red Hat) int rto_loop; 7384bdced5cSSteven Rostedt (Red Hat) int rto_cpu; 7394bdced5cSSteven Rostedt (Red Hat) /* These atomics are updated outside of a lock */ 7404bdced5cSSteven Rostedt (Red Hat) atomic_t rto_loop_next; 7414bdced5cSSteven Rostedt (Red Hat) atomic_t rto_loop_start; 7424bdced5cSSteven Rostedt (Red Hat) #endif 7431baca4ceSJuri Lelli /* 744391e43daSPeter Zijlstra * The "RT overload" flag: it gets set if a CPU has more than 745391e43daSPeter Zijlstra * one runnable RT task. 746391e43daSPeter Zijlstra */ 747391e43daSPeter Zijlstra cpumask_var_t rto_mask; 748391e43daSPeter Zijlstra struct cpupri cpupri; 749cd92bfd3SDietmar Eggemann 750cd92bfd3SDietmar Eggemann unsigned long max_cpu_capacity; 751391e43daSPeter Zijlstra }; 752391e43daSPeter Zijlstra 753391e43daSPeter Zijlstra extern struct root_domain def_root_domain; 754f2cb1360SIngo Molnar extern struct mutex sched_domains_mutex; 755f2cb1360SIngo Molnar 756f2cb1360SIngo Molnar extern void init_defrootdomain(void); 7578d5dc512SPeter Zijlstra extern int sched_init_domains(const struct cpumask *cpu_map); 758f2cb1360SIngo Molnar extern void rq_attach_root(struct rq *rq, struct root_domain *rd); 759364f5665SSteven Rostedt (VMware) extern void sched_get_rd(struct root_domain *rd); 760364f5665SSteven Rostedt (VMware) extern void sched_put_rd(struct root_domain *rd); 761391e43daSPeter Zijlstra 7624bdced5cSSteven Rostedt (Red Hat) #ifdef HAVE_RT_PUSH_IPI 7634bdced5cSSteven Rostedt (Red Hat) extern void rto_push_irq_work_func(struct irq_work *work); 7644bdced5cSSteven Rostedt (Red Hat) #endif 765391e43daSPeter Zijlstra #endif /* CONFIG_SMP */ 766391e43daSPeter Zijlstra 767391e43daSPeter Zijlstra /* 768391e43daSPeter Zijlstra * This is the main, per-CPU runqueue data structure. 769391e43daSPeter Zijlstra * 770391e43daSPeter Zijlstra * Locking rule: those places that want to lock multiple runqueues 771391e43daSPeter Zijlstra * (such as the load balancing or the thread migration code), lock 772391e43daSPeter Zijlstra * acquire operations must be ordered by ascending &runqueue. 773391e43daSPeter Zijlstra */ 774391e43daSPeter Zijlstra struct rq { 775391e43daSPeter Zijlstra /* runqueue lock: */ 776391e43daSPeter Zijlstra raw_spinlock_t lock; 777391e43daSPeter Zijlstra 778391e43daSPeter Zijlstra /* 779391e43daSPeter Zijlstra * nr_running and cpu_load should be in the same cacheline because 780391e43daSPeter Zijlstra * remote CPUs use both these fields when doing load calculation. 781391e43daSPeter Zijlstra */ 782c82513e5SPeter Zijlstra unsigned int nr_running; 7830ec8aa00SPeter Zijlstra #ifdef CONFIG_NUMA_BALANCING 7840ec8aa00SPeter Zijlstra unsigned int nr_numa_running; 7850ec8aa00SPeter Zijlstra unsigned int nr_preferred_running; 7860ec8aa00SPeter Zijlstra #endif 787391e43daSPeter Zijlstra #define CPU_LOAD_IDX_MAX 5 788391e43daSPeter Zijlstra unsigned long cpu_load[CPU_LOAD_IDX_MAX]; 7893451d024SFrederic Weisbecker #ifdef CONFIG_NO_HZ_COMMON 7909fd81dd5SFrederic Weisbecker #ifdef CONFIG_SMP 7919fd81dd5SFrederic Weisbecker unsigned long last_load_update_tick; 792e022e0d3SPeter Zijlstra unsigned long last_blocked_load_update_tick; 793f643ea22SVincent Guittot unsigned int has_blocked_load; 7949fd81dd5SFrederic Weisbecker #endif /* CONFIG_SMP */ 79500357f5eSPeter Zijlstra unsigned int nohz_tick_stopped; 796a22e47a4SPeter Zijlstra atomic_t nohz_flags; 7979fd81dd5SFrederic Weisbecker #endif /* CONFIG_NO_HZ_COMMON */ 798dcdedb24SFrederic Weisbecker 79997fb7a0aSIngo Molnar /* capture load from *all* tasks on this CPU: */ 800391e43daSPeter Zijlstra struct load_weight load; 801391e43daSPeter Zijlstra unsigned long nr_load_updates; 802391e43daSPeter Zijlstra u64 nr_switches; 803391e43daSPeter Zijlstra 804391e43daSPeter Zijlstra struct cfs_rq cfs; 805391e43daSPeter Zijlstra struct rt_rq rt; 806aab03e05SDario Faggioli struct dl_rq dl; 807391e43daSPeter Zijlstra 808391e43daSPeter Zijlstra #ifdef CONFIG_FAIR_GROUP_SCHED 80997fb7a0aSIngo Molnar /* list of leaf cfs_rq on this CPU: */ 810391e43daSPeter Zijlstra struct list_head leaf_cfs_rq_list; 8119c2791f9SVincent Guittot struct list_head *tmp_alone_branch; 812a35b6466SPeter Zijlstra #endif /* CONFIG_FAIR_GROUP_SCHED */ 813a35b6466SPeter Zijlstra 814391e43daSPeter Zijlstra /* 815391e43daSPeter Zijlstra * This is part of a global counter where only the total sum 816391e43daSPeter Zijlstra * over all CPUs matters. A task can increase this counter on 817391e43daSPeter Zijlstra * one CPU and if it got migrated afterwards it may decrease 818391e43daSPeter Zijlstra * it on another CPU. Always updated under the runqueue lock: 819391e43daSPeter Zijlstra */ 820391e43daSPeter Zijlstra unsigned long nr_uninterruptible; 821391e43daSPeter Zijlstra 82297fb7a0aSIngo Molnar struct task_struct *curr; 82397fb7a0aSIngo Molnar struct task_struct *idle; 82497fb7a0aSIngo Molnar struct task_struct *stop; 825391e43daSPeter Zijlstra unsigned long next_balance; 826391e43daSPeter Zijlstra struct mm_struct *prev_mm; 827391e43daSPeter Zijlstra 828cb42c9a3SMatt Fleming unsigned int clock_update_flags; 829391e43daSPeter Zijlstra u64 clock; 830391e43daSPeter Zijlstra u64 clock_task; 831391e43daSPeter Zijlstra 832391e43daSPeter Zijlstra atomic_t nr_iowait; 833391e43daSPeter Zijlstra 834391e43daSPeter Zijlstra #ifdef CONFIG_SMP 835391e43daSPeter Zijlstra struct root_domain *rd; 836391e43daSPeter Zijlstra struct sched_domain *sd; 837391e43daSPeter Zijlstra 838ced549faSNicolas Pitre unsigned long cpu_capacity; 839ca6d75e6SVincent Guittot unsigned long cpu_capacity_orig; 840391e43daSPeter Zijlstra 841e3fca9e7SPeter Zijlstra struct callback_head *balance_callback; 842e3fca9e7SPeter Zijlstra 843391e43daSPeter Zijlstra unsigned char idle_balance; 84497fb7a0aSIngo Molnar 8453b1baa64SMorten Rasmussen unsigned long misfit_task_load; 8463b1baa64SMorten Rasmussen 847391e43daSPeter Zijlstra /* For active balancing */ 848391e43daSPeter Zijlstra int active_balance; 849391e43daSPeter Zijlstra int push_cpu; 850391e43daSPeter Zijlstra struct cpu_stop_work active_balance_work; 85197fb7a0aSIngo Molnar 85297fb7a0aSIngo Molnar /* CPU of this runqueue: */ 853391e43daSPeter Zijlstra int cpu; 854391e43daSPeter Zijlstra int online; 855391e43daSPeter Zijlstra 856367456c7SPeter Zijlstra struct list_head cfs_tasks; 857367456c7SPeter Zijlstra 858371bf427SVincent Guittot struct sched_avg avg_rt; 8593727e0e1SVincent Guittot struct sched_avg avg_dl; 86091c27493SVincent Guittot #if defined(CONFIG_IRQ_TIME_ACCOUNTING) || defined(CONFIG_PARAVIRT_TIME_ACCOUNTING) 8612e62c474SVincent Guittot #define HAVE_SCHED_AVG_IRQ 86291c27493SVincent Guittot struct sched_avg avg_irq; 86391c27493SVincent Guittot #endif 864391e43daSPeter Zijlstra u64 idle_stamp; 865391e43daSPeter Zijlstra u64 avg_idle; 8669bd721c5SJason Low 8679bd721c5SJason Low /* This is used to determine avg_idle's max value */ 8689bd721c5SJason Low u64 max_idle_balance_cost; 869391e43daSPeter Zijlstra #endif 870391e43daSPeter Zijlstra 871391e43daSPeter Zijlstra #ifdef CONFIG_IRQ_TIME_ACCOUNTING 872391e43daSPeter Zijlstra u64 prev_irq_time; 873391e43daSPeter Zijlstra #endif 874391e43daSPeter Zijlstra #ifdef CONFIG_PARAVIRT 875391e43daSPeter Zijlstra u64 prev_steal_time; 876391e43daSPeter Zijlstra #endif 877391e43daSPeter Zijlstra #ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING 878391e43daSPeter Zijlstra u64 prev_steal_time_rq; 879391e43daSPeter Zijlstra #endif 880391e43daSPeter Zijlstra 881391e43daSPeter Zijlstra /* calc_load related fields */ 882391e43daSPeter Zijlstra unsigned long calc_load_update; 883391e43daSPeter Zijlstra long calc_load_active; 884391e43daSPeter Zijlstra 885391e43daSPeter Zijlstra #ifdef CONFIG_SCHED_HRTICK 886391e43daSPeter Zijlstra #ifdef CONFIG_SMP 887391e43daSPeter Zijlstra int hrtick_csd_pending; 888966a9671SYing Huang call_single_data_t hrtick_csd; 889391e43daSPeter Zijlstra #endif 890391e43daSPeter Zijlstra struct hrtimer hrtick_timer; 891391e43daSPeter Zijlstra #endif 892391e43daSPeter Zijlstra 893391e43daSPeter Zijlstra #ifdef CONFIG_SCHEDSTATS 894391e43daSPeter Zijlstra /* latency stats */ 895391e43daSPeter Zijlstra struct sched_info rq_sched_info; 896391e43daSPeter Zijlstra unsigned long long rq_cpu_time; 897391e43daSPeter Zijlstra /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ 898391e43daSPeter Zijlstra 899391e43daSPeter Zijlstra /* sys_sched_yield() stats */ 900391e43daSPeter Zijlstra unsigned int yld_count; 901391e43daSPeter Zijlstra 902391e43daSPeter Zijlstra /* schedule() stats */ 903391e43daSPeter Zijlstra unsigned int sched_count; 904391e43daSPeter Zijlstra unsigned int sched_goidle; 905391e43daSPeter Zijlstra 906391e43daSPeter Zijlstra /* try_to_wake_up() stats */ 907391e43daSPeter Zijlstra unsigned int ttwu_count; 908391e43daSPeter Zijlstra unsigned int ttwu_local; 909391e43daSPeter Zijlstra #endif 910391e43daSPeter Zijlstra 911391e43daSPeter Zijlstra #ifdef CONFIG_SMP 912391e43daSPeter Zijlstra struct llist_head wake_list; 913391e43daSPeter Zijlstra #endif 914442bf3aaSDaniel Lezcano 915442bf3aaSDaniel Lezcano #ifdef CONFIG_CPU_IDLE 916442bf3aaSDaniel Lezcano /* Must be inspected within a rcu lock section */ 917442bf3aaSDaniel Lezcano struct cpuidle_state *idle_state; 918442bf3aaSDaniel Lezcano #endif 919391e43daSPeter Zijlstra }; 920391e43daSPeter Zijlstra 921391e43daSPeter Zijlstra static inline int cpu_of(struct rq *rq) 922391e43daSPeter Zijlstra { 923391e43daSPeter Zijlstra #ifdef CONFIG_SMP 924391e43daSPeter Zijlstra return rq->cpu; 925391e43daSPeter Zijlstra #else 926391e43daSPeter Zijlstra return 0; 927391e43daSPeter Zijlstra #endif 928391e43daSPeter Zijlstra } 929391e43daSPeter Zijlstra 9301b568f0aSPeter Zijlstra 9311b568f0aSPeter Zijlstra #ifdef CONFIG_SCHED_SMT 9321b568f0aSPeter Zijlstra 9331b568f0aSPeter Zijlstra extern struct static_key_false sched_smt_present; 9341b568f0aSPeter Zijlstra 9351b568f0aSPeter Zijlstra extern void __update_idle_core(struct rq *rq); 9361b568f0aSPeter Zijlstra 9371b568f0aSPeter Zijlstra static inline void update_idle_core(struct rq *rq) 9381b568f0aSPeter Zijlstra { 9391b568f0aSPeter Zijlstra if (static_branch_unlikely(&sched_smt_present)) 9401b568f0aSPeter Zijlstra __update_idle_core(rq); 9411b568f0aSPeter Zijlstra } 9421b568f0aSPeter Zijlstra 9431b568f0aSPeter Zijlstra #else 9441b568f0aSPeter Zijlstra static inline void update_idle_core(struct rq *rq) { } 9451b568f0aSPeter Zijlstra #endif 9461b568f0aSPeter Zijlstra 9478b06c55bSPranith Kumar DECLARE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); 948391e43daSPeter Zijlstra 949518cd623SPeter Zijlstra #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) 9504a32fea9SChristoph Lameter #define this_rq() this_cpu_ptr(&runqueues) 951518cd623SPeter Zijlstra #define task_rq(p) cpu_rq(task_cpu(p)) 952518cd623SPeter Zijlstra #define cpu_curr(cpu) (cpu_rq(cpu)->curr) 9534a32fea9SChristoph Lameter #define raw_rq() raw_cpu_ptr(&runqueues) 954518cd623SPeter Zijlstra 955cebde6d6SPeter Zijlstra static inline u64 __rq_clock_broken(struct rq *rq) 956cebde6d6SPeter Zijlstra { 957316c1608SJason Low return READ_ONCE(rq->clock); 958cebde6d6SPeter Zijlstra } 959cebde6d6SPeter Zijlstra 960cb42c9a3SMatt Fleming /* 961cb42c9a3SMatt Fleming * rq::clock_update_flags bits 962cb42c9a3SMatt Fleming * 963cb42c9a3SMatt Fleming * %RQCF_REQ_SKIP - will request skipping of clock update on the next 964cb42c9a3SMatt Fleming * call to __schedule(). This is an optimisation to avoid 965cb42c9a3SMatt Fleming * neighbouring rq clock updates. 966cb42c9a3SMatt Fleming * 967cb42c9a3SMatt Fleming * %RQCF_ACT_SKIP - is set from inside of __schedule() when skipping is 968cb42c9a3SMatt Fleming * in effect and calls to update_rq_clock() are being ignored. 969cb42c9a3SMatt Fleming * 970cb42c9a3SMatt Fleming * %RQCF_UPDATED - is a debug flag that indicates whether a call has been 971cb42c9a3SMatt Fleming * made to update_rq_clock() since the last time rq::lock was pinned. 972cb42c9a3SMatt Fleming * 973cb42c9a3SMatt Fleming * If inside of __schedule(), clock_update_flags will have been 974cb42c9a3SMatt Fleming * shifted left (a left shift is a cheap operation for the fast path 975cb42c9a3SMatt Fleming * to promote %RQCF_REQ_SKIP to %RQCF_ACT_SKIP), so you must use, 976cb42c9a3SMatt Fleming * 977cb42c9a3SMatt Fleming * if (rq-clock_update_flags >= RQCF_UPDATED) 978cb42c9a3SMatt Fleming * 979cb42c9a3SMatt Fleming * to check if %RQCF_UPADTED is set. It'll never be shifted more than 980cb42c9a3SMatt Fleming * one position though, because the next rq_unpin_lock() will shift it 981cb42c9a3SMatt Fleming * back. 982cb42c9a3SMatt Fleming */ 983cb42c9a3SMatt Fleming #define RQCF_REQ_SKIP 0x01 984cb42c9a3SMatt Fleming #define RQCF_ACT_SKIP 0x02 985cb42c9a3SMatt Fleming #define RQCF_UPDATED 0x04 986cb42c9a3SMatt Fleming 987cb42c9a3SMatt Fleming static inline void assert_clock_updated(struct rq *rq) 988cb42c9a3SMatt Fleming { 989cb42c9a3SMatt Fleming /* 990cb42c9a3SMatt Fleming * The only reason for not seeing a clock update since the 991cb42c9a3SMatt Fleming * last rq_pin_lock() is if we're currently skipping updates. 992cb42c9a3SMatt Fleming */ 993cb42c9a3SMatt Fleming SCHED_WARN_ON(rq->clock_update_flags < RQCF_ACT_SKIP); 994cb42c9a3SMatt Fleming } 995cb42c9a3SMatt Fleming 99678becc27SFrederic Weisbecker static inline u64 rq_clock(struct rq *rq) 99778becc27SFrederic Weisbecker { 998cebde6d6SPeter Zijlstra lockdep_assert_held(&rq->lock); 999cb42c9a3SMatt Fleming assert_clock_updated(rq); 1000cb42c9a3SMatt Fleming 100178becc27SFrederic Weisbecker return rq->clock; 100278becc27SFrederic Weisbecker } 100378becc27SFrederic Weisbecker 100478becc27SFrederic Weisbecker static inline u64 rq_clock_task(struct rq *rq) 100578becc27SFrederic Weisbecker { 1006cebde6d6SPeter Zijlstra lockdep_assert_held(&rq->lock); 1007cb42c9a3SMatt Fleming assert_clock_updated(rq); 1008cb42c9a3SMatt Fleming 100978becc27SFrederic Weisbecker return rq->clock_task; 101078becc27SFrederic Weisbecker } 101178becc27SFrederic Weisbecker 1012adcc8da8SDavidlohr Bueso static inline void rq_clock_skip_update(struct rq *rq) 10139edfbfedSPeter Zijlstra { 10149edfbfedSPeter Zijlstra lockdep_assert_held(&rq->lock); 1015cb42c9a3SMatt Fleming rq->clock_update_flags |= RQCF_REQ_SKIP; 1016adcc8da8SDavidlohr Bueso } 1017adcc8da8SDavidlohr Bueso 1018adcc8da8SDavidlohr Bueso /* 1019595058b6SDavidlohr Bueso * See rt task throttling, which is the only time a skip 1020adcc8da8SDavidlohr Bueso * request is cancelled. 1021adcc8da8SDavidlohr Bueso */ 1022adcc8da8SDavidlohr Bueso static inline void rq_clock_cancel_skipupdate(struct rq *rq) 1023adcc8da8SDavidlohr Bueso { 1024adcc8da8SDavidlohr Bueso lockdep_assert_held(&rq->lock); 1025cb42c9a3SMatt Fleming rq->clock_update_flags &= ~RQCF_REQ_SKIP; 10269edfbfedSPeter Zijlstra } 10279edfbfedSPeter Zijlstra 1028d8ac8971SMatt Fleming struct rq_flags { 1029d8ac8971SMatt Fleming unsigned long flags; 1030d8ac8971SMatt Fleming struct pin_cookie cookie; 1031cb42c9a3SMatt Fleming #ifdef CONFIG_SCHED_DEBUG 1032cb42c9a3SMatt Fleming /* 1033cb42c9a3SMatt Fleming * A copy of (rq::clock_update_flags & RQCF_UPDATED) for the 1034cb42c9a3SMatt Fleming * current pin context is stashed here in case it needs to be 1035cb42c9a3SMatt Fleming * restored in rq_repin_lock(). 1036cb42c9a3SMatt Fleming */ 1037cb42c9a3SMatt Fleming unsigned int clock_update_flags; 1038cb42c9a3SMatt Fleming #endif 1039d8ac8971SMatt Fleming }; 1040d8ac8971SMatt Fleming 1041d8ac8971SMatt Fleming static inline void rq_pin_lock(struct rq *rq, struct rq_flags *rf) 1042d8ac8971SMatt Fleming { 1043d8ac8971SMatt Fleming rf->cookie = lockdep_pin_lock(&rq->lock); 1044cb42c9a3SMatt Fleming 1045cb42c9a3SMatt Fleming #ifdef CONFIG_SCHED_DEBUG 1046cb42c9a3SMatt Fleming rq->clock_update_flags &= (RQCF_REQ_SKIP|RQCF_ACT_SKIP); 1047cb42c9a3SMatt Fleming rf->clock_update_flags = 0; 1048cb42c9a3SMatt Fleming #endif 1049d8ac8971SMatt Fleming } 1050d8ac8971SMatt Fleming 1051d8ac8971SMatt Fleming static inline void rq_unpin_lock(struct rq *rq, struct rq_flags *rf) 1052d8ac8971SMatt Fleming { 1053cb42c9a3SMatt Fleming #ifdef CONFIG_SCHED_DEBUG 1054cb42c9a3SMatt Fleming if (rq->clock_update_flags > RQCF_ACT_SKIP) 1055cb42c9a3SMatt Fleming rf->clock_update_flags = RQCF_UPDATED; 1056cb42c9a3SMatt Fleming #endif 1057cb42c9a3SMatt Fleming 1058d8ac8971SMatt Fleming lockdep_unpin_lock(&rq->lock, rf->cookie); 1059d8ac8971SMatt Fleming } 1060d8ac8971SMatt Fleming 1061d8ac8971SMatt Fleming static inline void rq_repin_lock(struct rq *rq, struct rq_flags *rf) 1062d8ac8971SMatt Fleming { 1063d8ac8971SMatt Fleming lockdep_repin_lock(&rq->lock, rf->cookie); 1064cb42c9a3SMatt Fleming 1065cb42c9a3SMatt Fleming #ifdef CONFIG_SCHED_DEBUG 1066cb42c9a3SMatt Fleming /* 1067cb42c9a3SMatt Fleming * Restore the value we stashed in @rf for this pin context. 1068cb42c9a3SMatt Fleming */ 1069cb42c9a3SMatt Fleming rq->clock_update_flags |= rf->clock_update_flags; 1070cb42c9a3SMatt Fleming #endif 1071d8ac8971SMatt Fleming } 1072d8ac8971SMatt Fleming 10739942f79bSRik van Riel #ifdef CONFIG_NUMA 1074e3fe70b1SRik van Riel enum numa_topology_type { 1075e3fe70b1SRik van Riel NUMA_DIRECT, 1076e3fe70b1SRik van Riel NUMA_GLUELESS_MESH, 1077e3fe70b1SRik van Riel NUMA_BACKPLANE, 1078e3fe70b1SRik van Riel }; 1079e3fe70b1SRik van Riel extern enum numa_topology_type sched_numa_topology_type; 10809942f79bSRik van Riel extern int sched_max_numa_distance; 10819942f79bSRik van Riel extern bool find_numa_distance(int distance); 10829942f79bSRik van Riel #endif 10839942f79bSRik van Riel 1084f2cb1360SIngo Molnar #ifdef CONFIG_NUMA 1085f2cb1360SIngo Molnar extern void sched_init_numa(void); 1086f2cb1360SIngo Molnar extern void sched_domains_numa_masks_set(unsigned int cpu); 1087f2cb1360SIngo Molnar extern void sched_domains_numa_masks_clear(unsigned int cpu); 1088f2cb1360SIngo Molnar #else 1089f2cb1360SIngo Molnar static inline void sched_init_numa(void) { } 1090f2cb1360SIngo Molnar static inline void sched_domains_numa_masks_set(unsigned int cpu) { } 1091f2cb1360SIngo Molnar static inline void sched_domains_numa_masks_clear(unsigned int cpu) { } 1092f2cb1360SIngo Molnar #endif 1093f2cb1360SIngo Molnar 1094f809ca9aSMel Gorman #ifdef CONFIG_NUMA_BALANCING 109544dba3d5SIulia Manda /* The regions in numa_faults array from task_struct */ 109644dba3d5SIulia Manda enum numa_faults_stats { 109744dba3d5SIulia Manda NUMA_MEM = 0, 109844dba3d5SIulia Manda NUMA_CPU, 109944dba3d5SIulia Manda NUMA_MEMBUF, 110044dba3d5SIulia Manda NUMA_CPUBUF 110144dba3d5SIulia Manda }; 11020ec8aa00SPeter Zijlstra extern void sched_setnuma(struct task_struct *p, int node); 1103e6628d5bSMel Gorman extern int migrate_task_to(struct task_struct *p, int cpu); 11040ad4e3dfSSrikar Dronamraju extern int migrate_swap(struct task_struct *p, struct task_struct *t, 11050ad4e3dfSSrikar Dronamraju int cpu, int scpu); 110613784475SMel Gorman extern void init_numa_balancing(unsigned long clone_flags, struct task_struct *p); 110713784475SMel Gorman #else 110813784475SMel Gorman static inline void 110913784475SMel Gorman init_numa_balancing(unsigned long clone_flags, struct task_struct *p) 111013784475SMel Gorman { 111113784475SMel Gorman } 1112f809ca9aSMel Gorman #endif /* CONFIG_NUMA_BALANCING */ 1113f809ca9aSMel Gorman 1114518cd623SPeter Zijlstra #ifdef CONFIG_SMP 1115518cd623SPeter Zijlstra 1116e3fca9e7SPeter Zijlstra static inline void 1117e3fca9e7SPeter Zijlstra queue_balance_callback(struct rq *rq, 1118e3fca9e7SPeter Zijlstra struct callback_head *head, 1119e3fca9e7SPeter Zijlstra void (*func)(struct rq *rq)) 1120e3fca9e7SPeter Zijlstra { 1121e3fca9e7SPeter Zijlstra lockdep_assert_held(&rq->lock); 1122e3fca9e7SPeter Zijlstra 1123e3fca9e7SPeter Zijlstra if (unlikely(head->next)) 1124e3fca9e7SPeter Zijlstra return; 1125e3fca9e7SPeter Zijlstra 1126e3fca9e7SPeter Zijlstra head->func = (void (*)(struct callback_head *))func; 1127e3fca9e7SPeter Zijlstra head->next = rq->balance_callback; 1128e3fca9e7SPeter Zijlstra rq->balance_callback = head; 1129e3fca9e7SPeter Zijlstra } 1130e3fca9e7SPeter Zijlstra 1131e3baac47SPeter Zijlstra extern void sched_ttwu_pending(void); 1132e3baac47SPeter Zijlstra 1133391e43daSPeter Zijlstra #define rcu_dereference_check_sched_domain(p) \ 1134391e43daSPeter Zijlstra rcu_dereference_check((p), \ 1135391e43daSPeter Zijlstra lockdep_is_held(&sched_domains_mutex)) 1136391e43daSPeter Zijlstra 1137391e43daSPeter Zijlstra /* 1138391e43daSPeter Zijlstra * The domain tree (rq->sd) is protected by RCU's quiescent state transition. 1139391e43daSPeter Zijlstra * See detach_destroy_domains: synchronize_sched for details. 1140391e43daSPeter Zijlstra * 1141391e43daSPeter Zijlstra * The domain tree of any CPU may only be accessed from within 1142391e43daSPeter Zijlstra * preempt-disabled sections. 1143391e43daSPeter Zijlstra */ 1144391e43daSPeter Zijlstra #define for_each_domain(cpu, __sd) \ 1145518cd623SPeter Zijlstra for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); \ 1146518cd623SPeter Zijlstra __sd; __sd = __sd->parent) 1147391e43daSPeter Zijlstra 114877e81365SSuresh Siddha #define for_each_lower_domain(sd) for (; sd; sd = sd->child) 114977e81365SSuresh Siddha 1150518cd623SPeter Zijlstra /** 1151518cd623SPeter Zijlstra * highest_flag_domain - Return highest sched_domain containing flag. 115297fb7a0aSIngo Molnar * @cpu: The CPU whose highest level of sched domain is to 1153518cd623SPeter Zijlstra * be returned. 1154518cd623SPeter Zijlstra * @flag: The flag to check for the highest sched_domain 115597fb7a0aSIngo Molnar * for the given CPU. 1156518cd623SPeter Zijlstra * 115797fb7a0aSIngo Molnar * Returns the highest sched_domain of a CPU which contains the given flag. 1158518cd623SPeter Zijlstra */ 1159518cd623SPeter Zijlstra static inline struct sched_domain *highest_flag_domain(int cpu, int flag) 1160518cd623SPeter Zijlstra { 1161518cd623SPeter Zijlstra struct sched_domain *sd, *hsd = NULL; 1162518cd623SPeter Zijlstra 1163518cd623SPeter Zijlstra for_each_domain(cpu, sd) { 1164518cd623SPeter Zijlstra if (!(sd->flags & flag)) 1165518cd623SPeter Zijlstra break; 1166518cd623SPeter Zijlstra hsd = sd; 1167518cd623SPeter Zijlstra } 1168518cd623SPeter Zijlstra 1169518cd623SPeter Zijlstra return hsd; 1170518cd623SPeter Zijlstra } 1171518cd623SPeter Zijlstra 1172fb13c7eeSMel Gorman static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) 1173fb13c7eeSMel Gorman { 1174fb13c7eeSMel Gorman struct sched_domain *sd; 1175fb13c7eeSMel Gorman 1176fb13c7eeSMel Gorman for_each_domain(cpu, sd) { 1177fb13c7eeSMel Gorman if (sd->flags & flag) 1178fb13c7eeSMel Gorman break; 1179fb13c7eeSMel Gorman } 1180fb13c7eeSMel Gorman 1181fb13c7eeSMel Gorman return sd; 1182fb13c7eeSMel Gorman } 1183fb13c7eeSMel Gorman 1184518cd623SPeter Zijlstra DECLARE_PER_CPU(struct sched_domain *, sd_llc); 11857d9ffa89SPeter Zijlstra DECLARE_PER_CPU(int, sd_llc_size); 1186518cd623SPeter Zijlstra DECLARE_PER_CPU(int, sd_llc_id); 11870e369d75SPeter Zijlstra DECLARE_PER_CPU(struct sched_domain_shared *, sd_llc_shared); 1188fb13c7eeSMel Gorman DECLARE_PER_CPU(struct sched_domain *, sd_numa); 118937dc6b50SPreeti U Murthy DECLARE_PER_CPU(struct sched_domain *, sd_asym); 1190df054e84SMorten Rasmussen extern struct static_key_false sched_asym_cpucapacity; 1191518cd623SPeter Zijlstra 119263b2ca30SNicolas Pitre struct sched_group_capacity { 11935e6521eaSLi Zefan atomic_t ref; 11945e6521eaSLi Zefan /* 1195172895e6SYuyang Du * CPU capacity of this group, SCHED_CAPACITY_SCALE being max capacity 119663b2ca30SNicolas Pitre * for a single CPU. 11975e6521eaSLi Zefan */ 1198bf475ce0SMorten Rasmussen unsigned long capacity; 1199bf475ce0SMorten Rasmussen unsigned long min_capacity; /* Min per-CPU capacity in group */ 1200e3d6d0cbSMorten Rasmussen unsigned long max_capacity; /* Max per-CPU capacity in group */ 12015e6521eaSLi Zefan unsigned long next_update; 120263b2ca30SNicolas Pitre int imbalance; /* XXX unrelated to capacity but shared group state */ 12035e6521eaSLi Zefan 1204005f874dSPeter Zijlstra #ifdef CONFIG_SCHED_DEBUG 1205005f874dSPeter Zijlstra int id; 1206005f874dSPeter Zijlstra #endif 1207005f874dSPeter Zijlstra 120897fb7a0aSIngo Molnar unsigned long cpumask[0]; /* Balance mask */ 12095e6521eaSLi Zefan }; 12105e6521eaSLi Zefan 12115e6521eaSLi Zefan struct sched_group { 12125e6521eaSLi Zefan struct sched_group *next; /* Must be a circular list */ 12135e6521eaSLi Zefan atomic_t ref; 12145e6521eaSLi Zefan 12155e6521eaSLi Zefan unsigned int group_weight; 121663b2ca30SNicolas Pitre struct sched_group_capacity *sgc; 121797fb7a0aSIngo Molnar int asym_prefer_cpu; /* CPU of highest priority in group */ 12185e6521eaSLi Zefan 12195e6521eaSLi Zefan /* 12205e6521eaSLi Zefan * The CPUs this group covers. 12215e6521eaSLi Zefan * 12225e6521eaSLi Zefan * NOTE: this field is variable length. (Allocated dynamically 12235e6521eaSLi Zefan * by attaching extra space to the end of the structure, 12245e6521eaSLi Zefan * depending on how many CPUs the kernel has booted up with) 12255e6521eaSLi Zefan */ 12265e6521eaSLi Zefan unsigned long cpumask[0]; 12275e6521eaSLi Zefan }; 12285e6521eaSLi Zefan 1229ae4df9d6SPeter Zijlstra static inline struct cpumask *sched_group_span(struct sched_group *sg) 12305e6521eaSLi Zefan { 12315e6521eaSLi Zefan return to_cpumask(sg->cpumask); 12325e6521eaSLi Zefan } 12335e6521eaSLi Zefan 12345e6521eaSLi Zefan /* 1235e5c14b1fSPeter Zijlstra * See build_balance_mask(). 12365e6521eaSLi Zefan */ 1237e5c14b1fSPeter Zijlstra static inline struct cpumask *group_balance_mask(struct sched_group *sg) 12385e6521eaSLi Zefan { 123963b2ca30SNicolas Pitre return to_cpumask(sg->sgc->cpumask); 12405e6521eaSLi Zefan } 12415e6521eaSLi Zefan 12425e6521eaSLi Zefan /** 124397fb7a0aSIngo Molnar * group_first_cpu - Returns the first CPU in the cpumask of a sched_group. 124497fb7a0aSIngo Molnar * @group: The group whose first CPU is to be returned. 12455e6521eaSLi Zefan */ 12465e6521eaSLi Zefan static inline unsigned int group_first_cpu(struct sched_group *group) 12475e6521eaSLi Zefan { 1248ae4df9d6SPeter Zijlstra return cpumask_first(sched_group_span(group)); 12495e6521eaSLi Zefan } 12505e6521eaSLi Zefan 1251c1174876SPeter Zijlstra extern int group_balance_cpu(struct sched_group *sg); 1252c1174876SPeter Zijlstra 12533866e845SSteven Rostedt (Red Hat) #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) 12543866e845SSteven Rostedt (Red Hat) void register_sched_domain_sysctl(void); 1255bbdacdfeSPeter Zijlstra void dirty_sched_domain_sysctl(int cpu); 12563866e845SSteven Rostedt (Red Hat) void unregister_sched_domain_sysctl(void); 12573866e845SSteven Rostedt (Red Hat) #else 12583866e845SSteven Rostedt (Red Hat) static inline void register_sched_domain_sysctl(void) 12593866e845SSteven Rostedt (Red Hat) { 12603866e845SSteven Rostedt (Red Hat) } 1261bbdacdfeSPeter Zijlstra static inline void dirty_sched_domain_sysctl(int cpu) 1262bbdacdfeSPeter Zijlstra { 1263bbdacdfeSPeter Zijlstra } 12643866e845SSteven Rostedt (Red Hat) static inline void unregister_sched_domain_sysctl(void) 12653866e845SSteven Rostedt (Red Hat) { 12663866e845SSteven Rostedt (Red Hat) } 12673866e845SSteven Rostedt (Red Hat) #endif 12683866e845SSteven Rostedt (Red Hat) 1269e3baac47SPeter Zijlstra #else 1270e3baac47SPeter Zijlstra 1271e3baac47SPeter Zijlstra static inline void sched_ttwu_pending(void) { } 1272e3baac47SPeter Zijlstra 1273518cd623SPeter Zijlstra #endif /* CONFIG_SMP */ 1274391e43daSPeter Zijlstra 1275391e43daSPeter Zijlstra #include "stats.h" 12761051408fSIngo Molnar #include "autogroup.h" 1277391e43daSPeter Zijlstra 1278391e43daSPeter Zijlstra #ifdef CONFIG_CGROUP_SCHED 1279391e43daSPeter Zijlstra 1280391e43daSPeter Zijlstra /* 1281391e43daSPeter Zijlstra * Return the group to which this tasks belongs. 1282391e43daSPeter Zijlstra * 12838af01f56STejun Heo * We cannot use task_css() and friends because the cgroup subsystem 12848af01f56STejun Heo * changes that value before the cgroup_subsys::attach() method is called, 12858af01f56STejun Heo * therefore we cannot pin it and might observe the wrong value. 12868323f26cSPeter Zijlstra * 12878323f26cSPeter Zijlstra * The same is true for autogroup's p->signal->autogroup->tg, the autogroup 12888323f26cSPeter Zijlstra * core changes this before calling sched_move_task(). 12898323f26cSPeter Zijlstra * 12908323f26cSPeter Zijlstra * Instead we use a 'copy' which is updated from sched_move_task() while 12918323f26cSPeter Zijlstra * holding both task_struct::pi_lock and rq::lock. 1292391e43daSPeter Zijlstra */ 1293391e43daSPeter Zijlstra static inline struct task_group *task_group(struct task_struct *p) 1294391e43daSPeter Zijlstra { 12958323f26cSPeter Zijlstra return p->sched_task_group; 1296391e43daSPeter Zijlstra } 1297391e43daSPeter Zijlstra 1298391e43daSPeter Zijlstra /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ 1299391e43daSPeter Zijlstra static inline void set_task_rq(struct task_struct *p, unsigned int cpu) 1300391e43daSPeter Zijlstra { 1301391e43daSPeter Zijlstra #if defined(CONFIG_FAIR_GROUP_SCHED) || defined(CONFIG_RT_GROUP_SCHED) 1302391e43daSPeter Zijlstra struct task_group *tg = task_group(p); 1303391e43daSPeter Zijlstra #endif 1304391e43daSPeter Zijlstra 1305391e43daSPeter Zijlstra #ifdef CONFIG_FAIR_GROUP_SCHED 1306ad936d86SByungchul Park set_task_rq_fair(&p->se, p->se.cfs_rq, tg->cfs_rq[cpu]); 1307391e43daSPeter Zijlstra p->se.cfs_rq = tg->cfs_rq[cpu]; 1308391e43daSPeter Zijlstra p->se.parent = tg->se[cpu]; 1309391e43daSPeter Zijlstra #endif 1310391e43daSPeter Zijlstra 1311391e43daSPeter Zijlstra #ifdef CONFIG_RT_GROUP_SCHED 1312391e43daSPeter Zijlstra p->rt.rt_rq = tg->rt_rq[cpu]; 1313391e43daSPeter Zijlstra p->rt.parent = tg->rt_se[cpu]; 1314391e43daSPeter Zijlstra #endif 1315391e43daSPeter Zijlstra } 1316391e43daSPeter Zijlstra 1317391e43daSPeter Zijlstra #else /* CONFIG_CGROUP_SCHED */ 1318391e43daSPeter Zijlstra 1319391e43daSPeter Zijlstra static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } 1320391e43daSPeter Zijlstra static inline struct task_group *task_group(struct task_struct *p) 1321391e43daSPeter Zijlstra { 1322391e43daSPeter Zijlstra return NULL; 1323391e43daSPeter Zijlstra } 1324391e43daSPeter Zijlstra 1325391e43daSPeter Zijlstra #endif /* CONFIG_CGROUP_SCHED */ 1326391e43daSPeter Zijlstra 1327391e43daSPeter Zijlstra static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) 1328391e43daSPeter Zijlstra { 1329391e43daSPeter Zijlstra set_task_rq(p, cpu); 1330391e43daSPeter Zijlstra #ifdef CONFIG_SMP 1331391e43daSPeter Zijlstra /* 1332391e43daSPeter Zijlstra * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be 1333391e43daSPeter Zijlstra * successfuly executed on another CPU. We must ensure that updates of 1334391e43daSPeter Zijlstra * per-task data have been completed by this moment. 1335391e43daSPeter Zijlstra */ 1336391e43daSPeter Zijlstra smp_wmb(); 1337c65eacbeSAndy Lutomirski #ifdef CONFIG_THREAD_INFO_IN_TASK 1338c65eacbeSAndy Lutomirski p->cpu = cpu; 1339c65eacbeSAndy Lutomirski #else 1340391e43daSPeter Zijlstra task_thread_info(p)->cpu = cpu; 1341c65eacbeSAndy Lutomirski #endif 1342ac66f547SPeter Zijlstra p->wake_cpu = cpu; 1343391e43daSPeter Zijlstra #endif 1344391e43daSPeter Zijlstra } 1345391e43daSPeter Zijlstra 1346391e43daSPeter Zijlstra /* 1347391e43daSPeter Zijlstra * Tunables that become constants when CONFIG_SCHED_DEBUG is off: 1348391e43daSPeter Zijlstra */ 1349391e43daSPeter Zijlstra #ifdef CONFIG_SCHED_DEBUG 1350c5905afbSIngo Molnar # include <linux/static_key.h> 1351391e43daSPeter Zijlstra # define const_debug __read_mostly 1352391e43daSPeter Zijlstra #else 1353391e43daSPeter Zijlstra # define const_debug const 1354391e43daSPeter Zijlstra #endif 1355391e43daSPeter Zijlstra 1356391e43daSPeter Zijlstra #define SCHED_FEAT(name, enabled) \ 1357391e43daSPeter Zijlstra __SCHED_FEAT_##name , 1358391e43daSPeter Zijlstra 1359391e43daSPeter Zijlstra enum { 1360391e43daSPeter Zijlstra #include "features.h" 1361f8b6d1ccSPeter Zijlstra __SCHED_FEAT_NR, 1362391e43daSPeter Zijlstra }; 1363391e43daSPeter Zijlstra 1364391e43daSPeter Zijlstra #undef SCHED_FEAT 1365391e43daSPeter Zijlstra 1366f8b6d1ccSPeter Zijlstra #if defined(CONFIG_SCHED_DEBUG) && defined(HAVE_JUMP_LABEL) 1367765cc3a4SPatrick Bellasi 1368765cc3a4SPatrick Bellasi /* 1369765cc3a4SPatrick Bellasi * To support run-time toggling of sched features, all the translation units 1370765cc3a4SPatrick Bellasi * (but core.c) reference the sysctl_sched_features defined in core.c. 1371765cc3a4SPatrick Bellasi */ 1372765cc3a4SPatrick Bellasi extern const_debug unsigned int sysctl_sched_features; 1373765cc3a4SPatrick Bellasi 1374f8b6d1ccSPeter Zijlstra #define SCHED_FEAT(name, enabled) \ 1375c5905afbSIngo Molnar static __always_inline bool static_branch_##name(struct static_key *key) \ 1376f8b6d1ccSPeter Zijlstra { \ 13776e76ea8aSJason Baron return static_key_##enabled(key); \ 1378f8b6d1ccSPeter Zijlstra } 1379f8b6d1ccSPeter Zijlstra 1380f8b6d1ccSPeter Zijlstra #include "features.h" 1381f8b6d1ccSPeter Zijlstra #undef SCHED_FEAT 1382f8b6d1ccSPeter Zijlstra 1383c5905afbSIngo Molnar extern struct static_key sched_feat_keys[__SCHED_FEAT_NR]; 1384f8b6d1ccSPeter Zijlstra #define sched_feat(x) (static_branch_##x(&sched_feat_keys[__SCHED_FEAT_##x])) 1385765cc3a4SPatrick Bellasi 1386f8b6d1ccSPeter Zijlstra #else /* !(SCHED_DEBUG && HAVE_JUMP_LABEL) */ 1387765cc3a4SPatrick Bellasi 1388765cc3a4SPatrick Bellasi /* 1389765cc3a4SPatrick Bellasi * Each translation unit has its own copy of sysctl_sched_features to allow 1390765cc3a4SPatrick Bellasi * constants propagation at compile time and compiler optimization based on 1391765cc3a4SPatrick Bellasi * features default. 1392765cc3a4SPatrick Bellasi */ 1393765cc3a4SPatrick Bellasi #define SCHED_FEAT(name, enabled) \ 1394765cc3a4SPatrick Bellasi (1UL << __SCHED_FEAT_##name) * enabled | 1395765cc3a4SPatrick Bellasi static const_debug __maybe_unused unsigned int sysctl_sched_features = 1396765cc3a4SPatrick Bellasi #include "features.h" 1397765cc3a4SPatrick Bellasi 0; 1398765cc3a4SPatrick Bellasi #undef SCHED_FEAT 1399765cc3a4SPatrick Bellasi 1400391e43daSPeter Zijlstra #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) 1401765cc3a4SPatrick Bellasi 1402f8b6d1ccSPeter Zijlstra #endif /* SCHED_DEBUG && HAVE_JUMP_LABEL */ 1403391e43daSPeter Zijlstra 14042a595721SSrikar Dronamraju extern struct static_key_false sched_numa_balancing; 1405cb251765SMel Gorman extern struct static_key_false sched_schedstats; 1406cbee9f88SPeter Zijlstra 1407391e43daSPeter Zijlstra static inline u64 global_rt_period(void) 1408391e43daSPeter Zijlstra { 1409391e43daSPeter Zijlstra return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; 1410391e43daSPeter Zijlstra } 1411391e43daSPeter Zijlstra 1412391e43daSPeter Zijlstra static inline u64 global_rt_runtime(void) 1413391e43daSPeter Zijlstra { 1414391e43daSPeter Zijlstra if (sysctl_sched_rt_runtime < 0) 1415391e43daSPeter Zijlstra return RUNTIME_INF; 1416391e43daSPeter Zijlstra 1417391e43daSPeter Zijlstra return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; 1418391e43daSPeter Zijlstra } 1419391e43daSPeter Zijlstra 1420391e43daSPeter Zijlstra static inline int task_current(struct rq *rq, struct task_struct *p) 1421391e43daSPeter Zijlstra { 1422391e43daSPeter Zijlstra return rq->curr == p; 1423391e43daSPeter Zijlstra } 1424391e43daSPeter Zijlstra 1425391e43daSPeter Zijlstra static inline int task_running(struct rq *rq, struct task_struct *p) 1426391e43daSPeter Zijlstra { 1427391e43daSPeter Zijlstra #ifdef CONFIG_SMP 1428391e43daSPeter Zijlstra return p->on_cpu; 1429391e43daSPeter Zijlstra #else 1430391e43daSPeter Zijlstra return task_current(rq, p); 1431391e43daSPeter Zijlstra #endif 1432391e43daSPeter Zijlstra } 1433391e43daSPeter Zijlstra 1434da0c1e65SKirill Tkhai static inline int task_on_rq_queued(struct task_struct *p) 1435da0c1e65SKirill Tkhai { 1436da0c1e65SKirill Tkhai return p->on_rq == TASK_ON_RQ_QUEUED; 1437da0c1e65SKirill Tkhai } 1438391e43daSPeter Zijlstra 1439cca26e80SKirill Tkhai static inline int task_on_rq_migrating(struct task_struct *p) 1440cca26e80SKirill Tkhai { 1441cca26e80SKirill Tkhai return p->on_rq == TASK_ON_RQ_MIGRATING; 1442cca26e80SKirill Tkhai } 1443cca26e80SKirill Tkhai 1444b13095f0SLi Zefan /* 1445b13095f0SLi Zefan * wake flags 1446b13095f0SLi Zefan */ 144797fb7a0aSIngo Molnar #define WF_SYNC 0x01 /* Waker goes to sleep after wakeup */ 144897fb7a0aSIngo Molnar #define WF_FORK 0x02 /* Child wakeup after fork */ 144997fb7a0aSIngo Molnar #define WF_MIGRATED 0x4 /* Internal use, task got migrated */ 1450b13095f0SLi Zefan 1451391e43daSPeter Zijlstra /* 1452391e43daSPeter Zijlstra * To aid in avoiding the subversion of "niceness" due to uneven distribution 1453391e43daSPeter Zijlstra * of tasks with abnormal "nice" values across CPUs the contribution that 1454391e43daSPeter Zijlstra * each task makes to its run queue's load is weighted according to its 1455391e43daSPeter Zijlstra * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a 1456391e43daSPeter Zijlstra * scaled version of the new time slice allocation that they receive on time 1457391e43daSPeter Zijlstra * slice expiry etc. 1458391e43daSPeter Zijlstra */ 1459391e43daSPeter Zijlstra 1460391e43daSPeter Zijlstra #define WEIGHT_IDLEPRIO 3 1461391e43daSPeter Zijlstra #define WMULT_IDLEPRIO 1431655765 1462391e43daSPeter Zijlstra 1463ed82b8a1SAndi Kleen extern const int sched_prio_to_weight[40]; 1464ed82b8a1SAndi Kleen extern const u32 sched_prio_to_wmult[40]; 1465391e43daSPeter Zijlstra 1466ff77e468SPeter Zijlstra /* 1467ff77e468SPeter Zijlstra * {de,en}queue flags: 1468ff77e468SPeter Zijlstra * 1469ff77e468SPeter Zijlstra * DEQUEUE_SLEEP - task is no longer runnable 1470ff77e468SPeter Zijlstra * ENQUEUE_WAKEUP - task just became runnable 1471ff77e468SPeter Zijlstra * 1472ff77e468SPeter Zijlstra * SAVE/RESTORE - an otherwise spurious dequeue/enqueue, done to ensure tasks 1473ff77e468SPeter Zijlstra * are in a known state which allows modification. Such pairs 1474ff77e468SPeter Zijlstra * should preserve as much state as possible. 1475ff77e468SPeter Zijlstra * 1476ff77e468SPeter Zijlstra * MOVE - paired with SAVE/RESTORE, explicitly does not preserve the location 1477ff77e468SPeter Zijlstra * in the runqueue. 1478ff77e468SPeter Zijlstra * 1479ff77e468SPeter Zijlstra * ENQUEUE_HEAD - place at front of runqueue (tail if not specified) 1480ff77e468SPeter Zijlstra * ENQUEUE_REPLENISH - CBS (replenish runtime and postpone deadline) 148159efa0baSPeter Zijlstra * ENQUEUE_MIGRATED - the task was migrated during wakeup 1482ff77e468SPeter Zijlstra * 1483ff77e468SPeter Zijlstra */ 1484ff77e468SPeter Zijlstra 1485ff77e468SPeter Zijlstra #define DEQUEUE_SLEEP 0x01 148697fb7a0aSIngo Molnar #define DEQUEUE_SAVE 0x02 /* Matches ENQUEUE_RESTORE */ 148797fb7a0aSIngo Molnar #define DEQUEUE_MOVE 0x04 /* Matches ENQUEUE_MOVE */ 148897fb7a0aSIngo Molnar #define DEQUEUE_NOCLOCK 0x08 /* Matches ENQUEUE_NOCLOCK */ 1489ff77e468SPeter Zijlstra 14901de64443SPeter Zijlstra #define ENQUEUE_WAKEUP 0x01 1491ff77e468SPeter Zijlstra #define ENQUEUE_RESTORE 0x02 1492ff77e468SPeter Zijlstra #define ENQUEUE_MOVE 0x04 14930a67d1eeSPeter Zijlstra #define ENQUEUE_NOCLOCK 0x08 1494ff77e468SPeter Zijlstra 14950a67d1eeSPeter Zijlstra #define ENQUEUE_HEAD 0x10 14960a67d1eeSPeter Zijlstra #define ENQUEUE_REPLENISH 0x20 1497c82ba9faSLi Zefan #ifdef CONFIG_SMP 14980a67d1eeSPeter Zijlstra #define ENQUEUE_MIGRATED 0x40 1499c82ba9faSLi Zefan #else 150059efa0baSPeter Zijlstra #define ENQUEUE_MIGRATED 0x00 1501c82ba9faSLi Zefan #endif 1502c82ba9faSLi Zefan 150337e117c0SPeter Zijlstra #define RETRY_TASK ((void *)-1UL) 150437e117c0SPeter Zijlstra 1505c82ba9faSLi Zefan struct sched_class { 1506c82ba9faSLi Zefan const struct sched_class *next; 1507c82ba9faSLi Zefan 1508c82ba9faSLi Zefan void (*enqueue_task) (struct rq *rq, struct task_struct *p, int flags); 1509c82ba9faSLi Zefan void (*dequeue_task) (struct rq *rq, struct task_struct *p, int flags); 1510c82ba9faSLi Zefan void (*yield_task) (struct rq *rq); 1511c82ba9faSLi Zefan bool (*yield_to_task)(struct rq *rq, struct task_struct *p, bool preempt); 1512c82ba9faSLi Zefan 1513c82ba9faSLi Zefan void (*check_preempt_curr)(struct rq *rq, struct task_struct *p, int flags); 1514c82ba9faSLi Zefan 1515606dba2eSPeter Zijlstra /* 1516606dba2eSPeter Zijlstra * It is the responsibility of the pick_next_task() method that will 1517606dba2eSPeter Zijlstra * return the next task to call put_prev_task() on the @prev task or 1518606dba2eSPeter Zijlstra * something equivalent. 151937e117c0SPeter Zijlstra * 152037e117c0SPeter Zijlstra * May return RETRY_TASK when it finds a higher prio class has runnable 152137e117c0SPeter Zijlstra * tasks. 1522606dba2eSPeter Zijlstra */ 1523606dba2eSPeter Zijlstra struct task_struct * (*pick_next_task)(struct rq *rq, 1524e7904a28SPeter Zijlstra struct task_struct *prev, 1525d8ac8971SMatt Fleming struct rq_flags *rf); 1526c82ba9faSLi Zefan void (*put_prev_task)(struct rq *rq, struct task_struct *p); 1527c82ba9faSLi Zefan 1528c82ba9faSLi Zefan #ifdef CONFIG_SMP 1529ac66f547SPeter Zijlstra int (*select_task_rq)(struct task_struct *p, int task_cpu, int sd_flag, int flags); 15305a4fd036Sxiaofeng.yan void (*migrate_task_rq)(struct task_struct *p); 1531c82ba9faSLi Zefan 1532c82ba9faSLi Zefan void (*task_woken)(struct rq *this_rq, struct task_struct *task); 1533c82ba9faSLi Zefan 1534c82ba9faSLi Zefan void (*set_cpus_allowed)(struct task_struct *p, 1535c82ba9faSLi Zefan const struct cpumask *newmask); 1536c82ba9faSLi Zefan 1537c82ba9faSLi Zefan void (*rq_online)(struct rq *rq); 1538c82ba9faSLi Zefan void (*rq_offline)(struct rq *rq); 1539c82ba9faSLi Zefan #endif 1540c82ba9faSLi Zefan 1541c82ba9faSLi Zefan void (*set_curr_task)(struct rq *rq); 1542c82ba9faSLi Zefan void (*task_tick)(struct rq *rq, struct task_struct *p, int queued); 1543c82ba9faSLi Zefan void (*task_fork)(struct task_struct *p); 1544e6c390f2SDario Faggioli void (*task_dead)(struct task_struct *p); 1545c82ba9faSLi Zefan 154667dfa1b7SKirill Tkhai /* 154767dfa1b7SKirill Tkhai * The switched_from() call is allowed to drop rq->lock, therefore we 154867dfa1b7SKirill Tkhai * cannot assume the switched_from/switched_to pair is serliazed by 154967dfa1b7SKirill Tkhai * rq->lock. They are however serialized by p->pi_lock. 155067dfa1b7SKirill Tkhai */ 1551c82ba9faSLi Zefan void (*switched_from)(struct rq *this_rq, struct task_struct *task); 1552c82ba9faSLi Zefan void (*switched_to) (struct rq *this_rq, struct task_struct *task); 1553c82ba9faSLi Zefan void (*prio_changed) (struct rq *this_rq, struct task_struct *task, 1554c82ba9faSLi Zefan int oldprio); 1555c82ba9faSLi Zefan 1556c82ba9faSLi Zefan unsigned int (*get_rr_interval)(struct rq *rq, 1557c82ba9faSLi Zefan struct task_struct *task); 1558c82ba9faSLi Zefan 15596e998916SStanislaw Gruszka void (*update_curr)(struct rq *rq); 15606e998916SStanislaw Gruszka 1561ea86cb4bSVincent Guittot #define TASK_SET_GROUP 0 1562ea86cb4bSVincent Guittot #define TASK_MOVE_GROUP 1 1563ea86cb4bSVincent Guittot 1564c82ba9faSLi Zefan #ifdef CONFIG_FAIR_GROUP_SCHED 1565ea86cb4bSVincent Guittot void (*task_change_group)(struct task_struct *p, int type); 1566c82ba9faSLi Zefan #endif 1567c82ba9faSLi Zefan }; 1568391e43daSPeter Zijlstra 15693f1d2a31SPeter Zijlstra static inline void put_prev_task(struct rq *rq, struct task_struct *prev) 15703f1d2a31SPeter Zijlstra { 15713f1d2a31SPeter Zijlstra prev->sched_class->put_prev_task(rq, prev); 15723f1d2a31SPeter Zijlstra } 15733f1d2a31SPeter Zijlstra 1574b2bf6c31SPeter Zijlstra static inline void set_curr_task(struct rq *rq, struct task_struct *curr) 1575b2bf6c31SPeter Zijlstra { 1576b2bf6c31SPeter Zijlstra curr->sched_class->set_curr_task(rq); 1577b2bf6c31SPeter Zijlstra } 1578b2bf6c31SPeter Zijlstra 1579f5832c19SNicolas Pitre #ifdef CONFIG_SMP 1580391e43daSPeter Zijlstra #define sched_class_highest (&stop_sched_class) 1581f5832c19SNicolas Pitre #else 1582f5832c19SNicolas Pitre #define sched_class_highest (&dl_sched_class) 1583f5832c19SNicolas Pitre #endif 1584391e43daSPeter Zijlstra #define for_each_class(class) \ 1585391e43daSPeter Zijlstra for (class = sched_class_highest; class; class = class->next) 1586391e43daSPeter Zijlstra 1587391e43daSPeter Zijlstra extern const struct sched_class stop_sched_class; 1588aab03e05SDario Faggioli extern const struct sched_class dl_sched_class; 1589391e43daSPeter Zijlstra extern const struct sched_class rt_sched_class; 1590391e43daSPeter Zijlstra extern const struct sched_class fair_sched_class; 1591391e43daSPeter Zijlstra extern const struct sched_class idle_sched_class; 1592391e43daSPeter Zijlstra 1593391e43daSPeter Zijlstra 1594391e43daSPeter Zijlstra #ifdef CONFIG_SMP 1595391e43daSPeter Zijlstra 159663b2ca30SNicolas Pitre extern void update_group_capacity(struct sched_domain *sd, int cpu); 1597b719203bSLi Zefan 15987caff66fSDaniel Lezcano extern void trigger_load_balance(struct rq *rq); 1599391e43daSPeter Zijlstra 1600c5b28038SPeter Zijlstra extern void set_cpus_allowed_common(struct task_struct *p, const struct cpumask *new_mask); 1601c5b28038SPeter Zijlstra 1602391e43daSPeter Zijlstra #endif 1603391e43daSPeter Zijlstra 1604442bf3aaSDaniel Lezcano #ifdef CONFIG_CPU_IDLE 1605442bf3aaSDaniel Lezcano static inline void idle_set_state(struct rq *rq, 1606442bf3aaSDaniel Lezcano struct cpuidle_state *idle_state) 1607442bf3aaSDaniel Lezcano { 1608442bf3aaSDaniel Lezcano rq->idle_state = idle_state; 1609442bf3aaSDaniel Lezcano } 1610442bf3aaSDaniel Lezcano 1611442bf3aaSDaniel Lezcano static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1612442bf3aaSDaniel Lezcano { 16139148a3a1SPeter Zijlstra SCHED_WARN_ON(!rcu_read_lock_held()); 161497fb7a0aSIngo Molnar 1615442bf3aaSDaniel Lezcano return rq->idle_state; 1616442bf3aaSDaniel Lezcano } 1617442bf3aaSDaniel Lezcano #else 1618442bf3aaSDaniel Lezcano static inline void idle_set_state(struct rq *rq, 1619442bf3aaSDaniel Lezcano struct cpuidle_state *idle_state) 1620442bf3aaSDaniel Lezcano { 1621442bf3aaSDaniel Lezcano } 1622442bf3aaSDaniel Lezcano 1623442bf3aaSDaniel Lezcano static inline struct cpuidle_state *idle_get_state(struct rq *rq) 1624442bf3aaSDaniel Lezcano { 1625442bf3aaSDaniel Lezcano return NULL; 1626442bf3aaSDaniel Lezcano } 1627442bf3aaSDaniel Lezcano #endif 1628442bf3aaSDaniel Lezcano 16298663effbSSteven Rostedt (VMware) extern void schedule_idle(void); 16308663effbSSteven Rostedt (VMware) 1631391e43daSPeter Zijlstra extern void sysrq_sched_debug_show(void); 1632391e43daSPeter Zijlstra extern void sched_init_granularity(void); 1633391e43daSPeter Zijlstra extern void update_max_interval(void); 16341baca4ceSJuri Lelli 16351baca4ceSJuri Lelli extern void init_sched_dl_class(void); 1636391e43daSPeter Zijlstra extern void init_sched_rt_class(void); 1637391e43daSPeter Zijlstra extern void init_sched_fair_class(void); 1638391e43daSPeter Zijlstra 16399059393eSVincent Guittot extern void reweight_task(struct task_struct *p, int prio); 16409059393eSVincent Guittot 16418875125eSKirill Tkhai extern void resched_curr(struct rq *rq); 1642391e43daSPeter Zijlstra extern void resched_cpu(int cpu); 1643391e43daSPeter Zijlstra 1644391e43daSPeter Zijlstra extern struct rt_bandwidth def_rt_bandwidth; 1645391e43daSPeter Zijlstra extern void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime); 1646391e43daSPeter Zijlstra 1647332ac17eSDario Faggioli extern struct dl_bandwidth def_dl_bandwidth; 1648332ac17eSDario Faggioli extern void init_dl_bandwidth(struct dl_bandwidth *dl_b, u64 period, u64 runtime); 1649aab03e05SDario Faggioli extern void init_dl_task_timer(struct sched_dl_entity *dl_se); 1650209a0cbdSLuca Abeni extern void init_dl_inactive_task_timer(struct sched_dl_entity *dl_se); 16514da3abceSLuca Abeni extern void init_dl_rq_bw_ratio(struct dl_rq *dl_rq); 1652aab03e05SDario Faggioli 1653c52f14d3SLuca Abeni #define BW_SHIFT 20 1654c52f14d3SLuca Abeni #define BW_UNIT (1 << BW_SHIFT) 16554da3abceSLuca Abeni #define RATIO_SHIFT 8 1656332ac17eSDario Faggioli unsigned long to_ratio(u64 period, u64 runtime); 1657332ac17eSDario Faggioli 1658540247fbSYuyang Du extern void init_entity_runnable_average(struct sched_entity *se); 16592b8c41daSYuyang Du extern void post_init_entity_util_avg(struct sched_entity *se); 1660a75cdaa9SAlex Shi 166176d92ac3SFrederic Weisbecker #ifdef CONFIG_NO_HZ_FULL 166276d92ac3SFrederic Weisbecker extern bool sched_can_stop_tick(struct rq *rq); 1663d84b3131SFrederic Weisbecker extern int __init sched_tick_offload_init(void); 166476d92ac3SFrederic Weisbecker 166576d92ac3SFrederic Weisbecker /* 166676d92ac3SFrederic Weisbecker * Tick may be needed by tasks in the runqueue depending on their policy and 166776d92ac3SFrederic Weisbecker * requirements. If tick is needed, lets send the target an IPI to kick it out of 166876d92ac3SFrederic Weisbecker * nohz mode if necessary. 166976d92ac3SFrederic Weisbecker */ 167076d92ac3SFrederic Weisbecker static inline void sched_update_tick_dependency(struct rq *rq) 167176d92ac3SFrederic Weisbecker { 167276d92ac3SFrederic Weisbecker int cpu; 167376d92ac3SFrederic Weisbecker 167476d92ac3SFrederic Weisbecker if (!tick_nohz_full_enabled()) 167576d92ac3SFrederic Weisbecker return; 167676d92ac3SFrederic Weisbecker 167776d92ac3SFrederic Weisbecker cpu = cpu_of(rq); 167876d92ac3SFrederic Weisbecker 167976d92ac3SFrederic Weisbecker if (!tick_nohz_full_cpu(cpu)) 168076d92ac3SFrederic Weisbecker return; 168176d92ac3SFrederic Weisbecker 168276d92ac3SFrederic Weisbecker if (sched_can_stop_tick(rq)) 168376d92ac3SFrederic Weisbecker tick_nohz_dep_clear_cpu(cpu, TICK_DEP_BIT_SCHED); 168476d92ac3SFrederic Weisbecker else 168576d92ac3SFrederic Weisbecker tick_nohz_dep_set_cpu(cpu, TICK_DEP_BIT_SCHED); 168676d92ac3SFrederic Weisbecker } 168776d92ac3SFrederic Weisbecker #else 1688d84b3131SFrederic Weisbecker static inline int sched_tick_offload_init(void) { return 0; } 168976d92ac3SFrederic Weisbecker static inline void sched_update_tick_dependency(struct rq *rq) { } 169076d92ac3SFrederic Weisbecker #endif 169176d92ac3SFrederic Weisbecker 169272465447SKirill Tkhai static inline void add_nr_running(struct rq *rq, unsigned count) 1693391e43daSPeter Zijlstra { 169472465447SKirill Tkhai unsigned prev_nr = rq->nr_running; 169572465447SKirill Tkhai 169672465447SKirill Tkhai rq->nr_running = prev_nr + count; 16979f3660c2SFrederic Weisbecker 169872465447SKirill Tkhai if (prev_nr < 2 && rq->nr_running >= 2) { 16994486edd1STim Chen #ifdef CONFIG_SMP 17004486edd1STim Chen if (!rq->rd->overload) 1701*575638d1SValentin Schneider rq->rd->overload = 1; 17024486edd1STim Chen #endif 170376d92ac3SFrederic Weisbecker } 17044486edd1STim Chen 170576d92ac3SFrederic Weisbecker sched_update_tick_dependency(rq); 17064486edd1STim Chen } 1707391e43daSPeter Zijlstra 170872465447SKirill Tkhai static inline void sub_nr_running(struct rq *rq, unsigned count) 1709391e43daSPeter Zijlstra { 171072465447SKirill Tkhai rq->nr_running -= count; 171176d92ac3SFrederic Weisbecker /* Check if we still need preemption */ 171276d92ac3SFrederic Weisbecker sched_update_tick_dependency(rq); 1713391e43daSPeter Zijlstra } 1714391e43daSPeter Zijlstra 1715391e43daSPeter Zijlstra extern void update_rq_clock(struct rq *rq); 1716391e43daSPeter Zijlstra 1717391e43daSPeter Zijlstra extern void activate_task(struct rq *rq, struct task_struct *p, int flags); 1718391e43daSPeter Zijlstra extern void deactivate_task(struct rq *rq, struct task_struct *p, int flags); 1719391e43daSPeter Zijlstra 1720391e43daSPeter Zijlstra extern void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); 1721391e43daSPeter Zijlstra 1722391e43daSPeter Zijlstra extern const_debug unsigned int sysctl_sched_nr_migrate; 1723391e43daSPeter Zijlstra extern const_debug unsigned int sysctl_sched_migration_cost; 1724391e43daSPeter Zijlstra 1725391e43daSPeter Zijlstra #ifdef CONFIG_SCHED_HRTICK 1726391e43daSPeter Zijlstra 1727391e43daSPeter Zijlstra /* 1728391e43daSPeter Zijlstra * Use hrtick when: 1729391e43daSPeter Zijlstra * - enabled by features 1730391e43daSPeter Zijlstra * - hrtimer is actually high res 1731391e43daSPeter Zijlstra */ 1732391e43daSPeter Zijlstra static inline int hrtick_enabled(struct rq *rq) 1733391e43daSPeter Zijlstra { 1734391e43daSPeter Zijlstra if (!sched_feat(HRTICK)) 1735391e43daSPeter Zijlstra return 0; 1736391e43daSPeter Zijlstra if (!cpu_active(cpu_of(rq))) 1737391e43daSPeter Zijlstra return 0; 1738391e43daSPeter Zijlstra return hrtimer_is_hres_active(&rq->hrtick_timer); 1739391e43daSPeter Zijlstra } 1740391e43daSPeter Zijlstra 1741391e43daSPeter Zijlstra void hrtick_start(struct rq *rq, u64 delay); 1742391e43daSPeter Zijlstra 1743b39e66eaSMike Galbraith #else 1744b39e66eaSMike Galbraith 1745b39e66eaSMike Galbraith static inline int hrtick_enabled(struct rq *rq) 1746b39e66eaSMike Galbraith { 1747b39e66eaSMike Galbraith return 0; 1748b39e66eaSMike Galbraith } 1749b39e66eaSMike Galbraith 1750391e43daSPeter Zijlstra #endif /* CONFIG_SCHED_HRTICK */ 1751391e43daSPeter Zijlstra 1752dfbca41fSPeter Zijlstra #ifndef arch_scale_freq_capacity 1753dfbca41fSPeter Zijlstra static __always_inline 17547673c8a4SJuri Lelli unsigned long arch_scale_freq_capacity(int cpu) 1755dfbca41fSPeter Zijlstra { 1756dfbca41fSPeter Zijlstra return SCHED_CAPACITY_SCALE; 1757dfbca41fSPeter Zijlstra } 1758dfbca41fSPeter Zijlstra #endif 1759b5b4860dSVincent Guittot 17607e1a9208SJuri Lelli #ifdef CONFIG_SMP 17618cd5601cSMorten Rasmussen #ifndef arch_scale_cpu_capacity 17628cd5601cSMorten Rasmussen static __always_inline 17638cd5601cSMorten Rasmussen unsigned long arch_scale_cpu_capacity(struct sched_domain *sd, int cpu) 17648cd5601cSMorten Rasmussen { 1765e3279a2eSDietmar Eggemann if (sd && (sd->flags & SD_SHARE_CPUCAPACITY) && (sd->span_weight > 1)) 17668cd5601cSMorten Rasmussen return sd->smt_gain / sd->span_weight; 17678cd5601cSMorten Rasmussen 17688cd5601cSMorten Rasmussen return SCHED_CAPACITY_SCALE; 17698cd5601cSMorten Rasmussen } 17708cd5601cSMorten Rasmussen #endif 1771391e43daSPeter Zijlstra #else 17727e1a9208SJuri Lelli #ifndef arch_scale_cpu_capacity 17737e1a9208SJuri Lelli static __always_inline 17747e1a9208SJuri Lelli unsigned long arch_scale_cpu_capacity(void __always_unused *sd, int cpu) 17757e1a9208SJuri Lelli { 17767e1a9208SJuri Lelli return SCHED_CAPACITY_SCALE; 17777e1a9208SJuri Lelli } 17787e1a9208SJuri Lelli #endif 1779391e43daSPeter Zijlstra #endif 1780391e43daSPeter Zijlstra 1781eb580751SPeter Zijlstra struct rq *__task_rq_lock(struct task_struct *p, struct rq_flags *rf) 17823e71a462SPeter Zijlstra __acquires(rq->lock); 17838a8c69c3SPeter Zijlstra 1784eb580751SPeter Zijlstra struct rq *task_rq_lock(struct task_struct *p, struct rq_flags *rf) 17853960c8c0SPeter Zijlstra __acquires(p->pi_lock) 17863e71a462SPeter Zijlstra __acquires(rq->lock); 17873960c8c0SPeter Zijlstra 1788eb580751SPeter Zijlstra static inline void __task_rq_unlock(struct rq *rq, struct rq_flags *rf) 17893960c8c0SPeter Zijlstra __releases(rq->lock) 17903960c8c0SPeter Zijlstra { 1791d8ac8971SMatt Fleming rq_unpin_lock(rq, rf); 17923960c8c0SPeter Zijlstra raw_spin_unlock(&rq->lock); 17933960c8c0SPeter Zijlstra } 17943960c8c0SPeter Zijlstra 17953960c8c0SPeter Zijlstra static inline void 1796eb580751SPeter Zijlstra task_rq_unlock(struct rq *rq, struct task_struct *p, struct rq_flags *rf) 17973960c8c0SPeter Zijlstra __releases(rq->lock) 17983960c8c0SPeter Zijlstra __releases(p->pi_lock) 17993960c8c0SPeter Zijlstra { 1800d8ac8971SMatt Fleming rq_unpin_lock(rq, rf); 18013960c8c0SPeter Zijlstra raw_spin_unlock(&rq->lock); 1802eb580751SPeter Zijlstra raw_spin_unlock_irqrestore(&p->pi_lock, rf->flags); 18033960c8c0SPeter Zijlstra } 18043960c8c0SPeter Zijlstra 18058a8c69c3SPeter Zijlstra static inline void 18068a8c69c3SPeter Zijlstra rq_lock_irqsave(struct rq *rq, struct rq_flags *rf) 18078a8c69c3SPeter Zijlstra __acquires(rq->lock) 18088a8c69c3SPeter Zijlstra { 18098a8c69c3SPeter Zijlstra raw_spin_lock_irqsave(&rq->lock, rf->flags); 18108a8c69c3SPeter Zijlstra rq_pin_lock(rq, rf); 18118a8c69c3SPeter Zijlstra } 18128a8c69c3SPeter Zijlstra 18138a8c69c3SPeter Zijlstra static inline void 18148a8c69c3SPeter Zijlstra rq_lock_irq(struct rq *rq, struct rq_flags *rf) 18158a8c69c3SPeter Zijlstra __acquires(rq->lock) 18168a8c69c3SPeter Zijlstra { 18178a8c69c3SPeter Zijlstra raw_spin_lock_irq(&rq->lock); 18188a8c69c3SPeter Zijlstra rq_pin_lock(rq, rf); 18198a8c69c3SPeter Zijlstra } 18208a8c69c3SPeter Zijlstra 18218a8c69c3SPeter Zijlstra static inline void 18228a8c69c3SPeter Zijlstra rq_lock(struct rq *rq, struct rq_flags *rf) 18238a8c69c3SPeter Zijlstra __acquires(rq->lock) 18248a8c69c3SPeter Zijlstra { 18258a8c69c3SPeter Zijlstra raw_spin_lock(&rq->lock); 18268a8c69c3SPeter Zijlstra rq_pin_lock(rq, rf); 18278a8c69c3SPeter Zijlstra } 18288a8c69c3SPeter Zijlstra 18298a8c69c3SPeter Zijlstra static inline void 18308a8c69c3SPeter Zijlstra rq_relock(struct rq *rq, struct rq_flags *rf) 18318a8c69c3SPeter Zijlstra __acquires(rq->lock) 18328a8c69c3SPeter Zijlstra { 18338a8c69c3SPeter Zijlstra raw_spin_lock(&rq->lock); 18348a8c69c3SPeter Zijlstra rq_repin_lock(rq, rf); 18358a8c69c3SPeter Zijlstra } 18368a8c69c3SPeter Zijlstra 18378a8c69c3SPeter Zijlstra static inline void 18388a8c69c3SPeter Zijlstra rq_unlock_irqrestore(struct rq *rq, struct rq_flags *rf) 18398a8c69c3SPeter Zijlstra __releases(rq->lock) 18408a8c69c3SPeter Zijlstra { 18418a8c69c3SPeter Zijlstra rq_unpin_lock(rq, rf); 18428a8c69c3SPeter Zijlstra raw_spin_unlock_irqrestore(&rq->lock, rf->flags); 18438a8c69c3SPeter Zijlstra } 18448a8c69c3SPeter Zijlstra 18458a8c69c3SPeter Zijlstra static inline void 18468a8c69c3SPeter Zijlstra rq_unlock_irq(struct rq *rq, struct rq_flags *rf) 18478a8c69c3SPeter Zijlstra __releases(rq->lock) 18488a8c69c3SPeter Zijlstra { 18498a8c69c3SPeter Zijlstra rq_unpin_lock(rq, rf); 18508a8c69c3SPeter Zijlstra raw_spin_unlock_irq(&rq->lock); 18518a8c69c3SPeter Zijlstra } 18528a8c69c3SPeter Zijlstra 18538a8c69c3SPeter Zijlstra static inline void 18548a8c69c3SPeter Zijlstra rq_unlock(struct rq *rq, struct rq_flags *rf) 18558a8c69c3SPeter Zijlstra __releases(rq->lock) 18568a8c69c3SPeter Zijlstra { 18578a8c69c3SPeter Zijlstra rq_unpin_lock(rq, rf); 18588a8c69c3SPeter Zijlstra raw_spin_unlock(&rq->lock); 18598a8c69c3SPeter Zijlstra } 18608a8c69c3SPeter Zijlstra 1861391e43daSPeter Zijlstra #ifdef CONFIG_SMP 1862391e43daSPeter Zijlstra #ifdef CONFIG_PREEMPT 1863391e43daSPeter Zijlstra 1864391e43daSPeter Zijlstra static inline void double_rq_lock(struct rq *rq1, struct rq *rq2); 1865391e43daSPeter Zijlstra 1866391e43daSPeter Zijlstra /* 1867391e43daSPeter Zijlstra * fair double_lock_balance: Safely acquires both rq->locks in a fair 1868391e43daSPeter Zijlstra * way at the expense of forcing extra atomic operations in all 1869391e43daSPeter Zijlstra * invocations. This assures that the double_lock is acquired using the 1870391e43daSPeter Zijlstra * same underlying policy as the spinlock_t on this architecture, which 1871391e43daSPeter Zijlstra * reduces latency compared to the unfair variant below. However, it 1872391e43daSPeter Zijlstra * also adds more overhead and therefore may reduce throughput. 1873391e43daSPeter Zijlstra */ 1874391e43daSPeter Zijlstra static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 1875391e43daSPeter Zijlstra __releases(this_rq->lock) 1876391e43daSPeter Zijlstra __acquires(busiest->lock) 1877391e43daSPeter Zijlstra __acquires(this_rq->lock) 1878391e43daSPeter Zijlstra { 1879391e43daSPeter Zijlstra raw_spin_unlock(&this_rq->lock); 1880391e43daSPeter Zijlstra double_rq_lock(this_rq, busiest); 1881391e43daSPeter Zijlstra 1882391e43daSPeter Zijlstra return 1; 1883391e43daSPeter Zijlstra } 1884391e43daSPeter Zijlstra 1885391e43daSPeter Zijlstra #else 1886391e43daSPeter Zijlstra /* 1887391e43daSPeter Zijlstra * Unfair double_lock_balance: Optimizes throughput at the expense of 1888391e43daSPeter Zijlstra * latency by eliminating extra atomic operations when the locks are 188997fb7a0aSIngo Molnar * already in proper order on entry. This favors lower CPU-ids and will 189097fb7a0aSIngo Molnar * grant the double lock to lower CPUs over higher ids under contention, 1891391e43daSPeter Zijlstra * regardless of entry order into the function. 1892391e43daSPeter Zijlstra */ 1893391e43daSPeter Zijlstra static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) 1894391e43daSPeter Zijlstra __releases(this_rq->lock) 1895391e43daSPeter Zijlstra __acquires(busiest->lock) 1896391e43daSPeter Zijlstra __acquires(this_rq->lock) 1897391e43daSPeter Zijlstra { 1898391e43daSPeter Zijlstra int ret = 0; 1899391e43daSPeter Zijlstra 1900391e43daSPeter Zijlstra if (unlikely(!raw_spin_trylock(&busiest->lock))) { 1901391e43daSPeter Zijlstra if (busiest < this_rq) { 1902391e43daSPeter Zijlstra raw_spin_unlock(&this_rq->lock); 1903391e43daSPeter Zijlstra raw_spin_lock(&busiest->lock); 1904391e43daSPeter Zijlstra raw_spin_lock_nested(&this_rq->lock, 1905391e43daSPeter Zijlstra SINGLE_DEPTH_NESTING); 1906391e43daSPeter Zijlstra ret = 1; 1907391e43daSPeter Zijlstra } else 1908391e43daSPeter Zijlstra raw_spin_lock_nested(&busiest->lock, 1909391e43daSPeter Zijlstra SINGLE_DEPTH_NESTING); 1910391e43daSPeter Zijlstra } 1911391e43daSPeter Zijlstra return ret; 1912391e43daSPeter Zijlstra } 1913391e43daSPeter Zijlstra 1914391e43daSPeter Zijlstra #endif /* CONFIG_PREEMPT */ 1915391e43daSPeter Zijlstra 1916391e43daSPeter Zijlstra /* 1917391e43daSPeter Zijlstra * double_lock_balance - lock the busiest runqueue, this_rq is locked already. 1918391e43daSPeter Zijlstra */ 1919391e43daSPeter Zijlstra static inline int double_lock_balance(struct rq *this_rq, struct rq *busiest) 1920391e43daSPeter Zijlstra { 1921391e43daSPeter Zijlstra if (unlikely(!irqs_disabled())) { 192297fb7a0aSIngo Molnar /* printk() doesn't work well under rq->lock */ 1923391e43daSPeter Zijlstra raw_spin_unlock(&this_rq->lock); 1924391e43daSPeter Zijlstra BUG_ON(1); 1925391e43daSPeter Zijlstra } 1926391e43daSPeter Zijlstra 1927391e43daSPeter Zijlstra return _double_lock_balance(this_rq, busiest); 1928391e43daSPeter Zijlstra } 1929391e43daSPeter Zijlstra 1930391e43daSPeter Zijlstra static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) 1931391e43daSPeter Zijlstra __releases(busiest->lock) 1932391e43daSPeter Zijlstra { 1933391e43daSPeter Zijlstra raw_spin_unlock(&busiest->lock); 1934391e43daSPeter Zijlstra lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); 1935391e43daSPeter Zijlstra } 1936391e43daSPeter Zijlstra 193774602315SPeter Zijlstra static inline void double_lock(spinlock_t *l1, spinlock_t *l2) 193874602315SPeter Zijlstra { 193974602315SPeter Zijlstra if (l1 > l2) 194074602315SPeter Zijlstra swap(l1, l2); 194174602315SPeter Zijlstra 194274602315SPeter Zijlstra spin_lock(l1); 194374602315SPeter Zijlstra spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 194474602315SPeter Zijlstra } 194574602315SPeter Zijlstra 194660e69eedSMike Galbraith static inline void double_lock_irq(spinlock_t *l1, spinlock_t *l2) 194760e69eedSMike Galbraith { 194860e69eedSMike Galbraith if (l1 > l2) 194960e69eedSMike Galbraith swap(l1, l2); 195060e69eedSMike Galbraith 195160e69eedSMike Galbraith spin_lock_irq(l1); 195260e69eedSMike Galbraith spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 195360e69eedSMike Galbraith } 195460e69eedSMike Galbraith 195574602315SPeter Zijlstra static inline void double_raw_lock(raw_spinlock_t *l1, raw_spinlock_t *l2) 195674602315SPeter Zijlstra { 195774602315SPeter Zijlstra if (l1 > l2) 195874602315SPeter Zijlstra swap(l1, l2); 195974602315SPeter Zijlstra 196074602315SPeter Zijlstra raw_spin_lock(l1); 196174602315SPeter Zijlstra raw_spin_lock_nested(l2, SINGLE_DEPTH_NESTING); 196274602315SPeter Zijlstra } 196374602315SPeter Zijlstra 1964391e43daSPeter Zijlstra /* 1965391e43daSPeter Zijlstra * double_rq_lock - safely lock two runqueues 1966391e43daSPeter Zijlstra * 1967391e43daSPeter Zijlstra * Note this does not disable interrupts like task_rq_lock, 1968391e43daSPeter Zijlstra * you need to do so manually before calling. 1969391e43daSPeter Zijlstra */ 1970391e43daSPeter Zijlstra static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 1971391e43daSPeter Zijlstra __acquires(rq1->lock) 1972391e43daSPeter Zijlstra __acquires(rq2->lock) 1973391e43daSPeter Zijlstra { 1974391e43daSPeter Zijlstra BUG_ON(!irqs_disabled()); 1975391e43daSPeter Zijlstra if (rq1 == rq2) { 1976391e43daSPeter Zijlstra raw_spin_lock(&rq1->lock); 1977391e43daSPeter Zijlstra __acquire(rq2->lock); /* Fake it out ;) */ 1978391e43daSPeter Zijlstra } else { 1979391e43daSPeter Zijlstra if (rq1 < rq2) { 1980391e43daSPeter Zijlstra raw_spin_lock(&rq1->lock); 1981391e43daSPeter Zijlstra raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); 1982391e43daSPeter Zijlstra } else { 1983391e43daSPeter Zijlstra raw_spin_lock(&rq2->lock); 1984391e43daSPeter Zijlstra raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); 1985391e43daSPeter Zijlstra } 1986391e43daSPeter Zijlstra } 1987391e43daSPeter Zijlstra } 1988391e43daSPeter Zijlstra 1989391e43daSPeter Zijlstra /* 1990391e43daSPeter Zijlstra * double_rq_unlock - safely unlock two runqueues 1991391e43daSPeter Zijlstra * 1992391e43daSPeter Zijlstra * Note this does not restore interrupts like task_rq_unlock, 1993391e43daSPeter Zijlstra * you need to do so manually after calling. 1994391e43daSPeter Zijlstra */ 1995391e43daSPeter Zijlstra static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 1996391e43daSPeter Zijlstra __releases(rq1->lock) 1997391e43daSPeter Zijlstra __releases(rq2->lock) 1998391e43daSPeter Zijlstra { 1999391e43daSPeter Zijlstra raw_spin_unlock(&rq1->lock); 2000391e43daSPeter Zijlstra if (rq1 != rq2) 2001391e43daSPeter Zijlstra raw_spin_unlock(&rq2->lock); 2002391e43daSPeter Zijlstra else 2003391e43daSPeter Zijlstra __release(rq2->lock); 2004391e43daSPeter Zijlstra } 2005391e43daSPeter Zijlstra 2006f2cb1360SIngo Molnar extern void set_rq_online (struct rq *rq); 2007f2cb1360SIngo Molnar extern void set_rq_offline(struct rq *rq); 2008f2cb1360SIngo Molnar extern bool sched_smp_initialized; 2009f2cb1360SIngo Molnar 2010391e43daSPeter Zijlstra #else /* CONFIG_SMP */ 2011391e43daSPeter Zijlstra 2012391e43daSPeter Zijlstra /* 2013391e43daSPeter Zijlstra * double_rq_lock - safely lock two runqueues 2014391e43daSPeter Zijlstra * 2015391e43daSPeter Zijlstra * Note this does not disable interrupts like task_rq_lock, 2016391e43daSPeter Zijlstra * you need to do so manually before calling. 2017391e43daSPeter Zijlstra */ 2018391e43daSPeter Zijlstra static inline void double_rq_lock(struct rq *rq1, struct rq *rq2) 2019391e43daSPeter Zijlstra __acquires(rq1->lock) 2020391e43daSPeter Zijlstra __acquires(rq2->lock) 2021391e43daSPeter Zijlstra { 2022391e43daSPeter Zijlstra BUG_ON(!irqs_disabled()); 2023391e43daSPeter Zijlstra BUG_ON(rq1 != rq2); 2024391e43daSPeter Zijlstra raw_spin_lock(&rq1->lock); 2025391e43daSPeter Zijlstra __acquire(rq2->lock); /* Fake it out ;) */ 2026391e43daSPeter Zijlstra } 2027391e43daSPeter Zijlstra 2028391e43daSPeter Zijlstra /* 2029391e43daSPeter Zijlstra * double_rq_unlock - safely unlock two runqueues 2030391e43daSPeter Zijlstra * 2031391e43daSPeter Zijlstra * Note this does not restore interrupts like task_rq_unlock, 2032391e43daSPeter Zijlstra * you need to do so manually after calling. 2033391e43daSPeter Zijlstra */ 2034391e43daSPeter Zijlstra static inline void double_rq_unlock(struct rq *rq1, struct rq *rq2) 2035391e43daSPeter Zijlstra __releases(rq1->lock) 2036391e43daSPeter Zijlstra __releases(rq2->lock) 2037391e43daSPeter Zijlstra { 2038391e43daSPeter Zijlstra BUG_ON(rq1 != rq2); 2039391e43daSPeter Zijlstra raw_spin_unlock(&rq1->lock); 2040391e43daSPeter Zijlstra __release(rq2->lock); 2041391e43daSPeter Zijlstra } 2042391e43daSPeter Zijlstra 2043391e43daSPeter Zijlstra #endif 2044391e43daSPeter Zijlstra 2045391e43daSPeter Zijlstra extern struct sched_entity *__pick_first_entity(struct cfs_rq *cfs_rq); 2046391e43daSPeter Zijlstra extern struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq); 20476b55c965SSrikar Dronamraju 20486b55c965SSrikar Dronamraju #ifdef CONFIG_SCHED_DEBUG 20499469eb01SPeter Zijlstra extern bool sched_debug_enabled; 20509469eb01SPeter Zijlstra 2051391e43daSPeter Zijlstra extern void print_cfs_stats(struct seq_file *m, int cpu); 2052391e43daSPeter Zijlstra extern void print_rt_stats(struct seq_file *m, int cpu); 2053acb32132SWanpeng Li extern void print_dl_stats(struct seq_file *m, int cpu); 2054f6a34630SMathieu Malaterre extern void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq); 2055f6a34630SMathieu Malaterre extern void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq); 2056f6a34630SMathieu Malaterre extern void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq); 2057397f2378SSrikar Dronamraju #ifdef CONFIG_NUMA_BALANCING 2058397f2378SSrikar Dronamraju extern void 2059397f2378SSrikar Dronamraju show_numa_stats(struct task_struct *p, struct seq_file *m); 2060397f2378SSrikar Dronamraju extern void 2061397f2378SSrikar Dronamraju print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 2062397f2378SSrikar Dronamraju unsigned long tpf, unsigned long gsf, unsigned long gpf); 2063397f2378SSrikar Dronamraju #endif /* CONFIG_NUMA_BALANCING */ 2064397f2378SSrikar Dronamraju #endif /* CONFIG_SCHED_DEBUG */ 2065391e43daSPeter Zijlstra 2066391e43daSPeter Zijlstra extern void init_cfs_rq(struct cfs_rq *cfs_rq); 206707c54f7aSAbel Vesa extern void init_rt_rq(struct rt_rq *rt_rq); 206807c54f7aSAbel Vesa extern void init_dl_rq(struct dl_rq *dl_rq); 2069391e43daSPeter Zijlstra 20701ee14e6cSBen Segall extern void cfs_bandwidth_usage_inc(void); 20711ee14e6cSBen Segall extern void cfs_bandwidth_usage_dec(void); 20721c792db7SSuresh Siddha 20733451d024SFrederic Weisbecker #ifdef CONFIG_NO_HZ_COMMON 207400357f5eSPeter Zijlstra #define NOHZ_BALANCE_KICK_BIT 0 207500357f5eSPeter Zijlstra #define NOHZ_STATS_KICK_BIT 1 2076a22e47a4SPeter Zijlstra 2077a22e47a4SPeter Zijlstra #define NOHZ_BALANCE_KICK BIT(NOHZ_BALANCE_KICK_BIT) 2078b7031a02SPeter Zijlstra #define NOHZ_STATS_KICK BIT(NOHZ_STATS_KICK_BIT) 2079b7031a02SPeter Zijlstra 2080b7031a02SPeter Zijlstra #define NOHZ_KICK_MASK (NOHZ_BALANCE_KICK | NOHZ_STATS_KICK) 20811c792db7SSuresh Siddha 20821c792db7SSuresh Siddha #define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) 208320a5c8ccSThomas Gleixner 208400357f5eSPeter Zijlstra extern void nohz_balance_exit_idle(struct rq *rq); 208520a5c8ccSThomas Gleixner #else 208600357f5eSPeter Zijlstra static inline void nohz_balance_exit_idle(struct rq *rq) { } 20871c792db7SSuresh Siddha #endif 208873fbec60SFrederic Weisbecker 2089daec5798SLuca Abeni 2090daec5798SLuca Abeni #ifdef CONFIG_SMP 2091daec5798SLuca Abeni static inline 2092daec5798SLuca Abeni void __dl_update(struct dl_bw *dl_b, s64 bw) 2093daec5798SLuca Abeni { 2094daec5798SLuca Abeni struct root_domain *rd = container_of(dl_b, struct root_domain, dl_bw); 2095daec5798SLuca Abeni int i; 2096daec5798SLuca Abeni 2097daec5798SLuca Abeni RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(), 2098daec5798SLuca Abeni "sched RCU must be held"); 2099daec5798SLuca Abeni for_each_cpu_and(i, rd->span, cpu_active_mask) { 2100daec5798SLuca Abeni struct rq *rq = cpu_rq(i); 2101daec5798SLuca Abeni 2102daec5798SLuca Abeni rq->dl.extra_bw += bw; 2103daec5798SLuca Abeni } 2104daec5798SLuca Abeni } 2105daec5798SLuca Abeni #else 2106daec5798SLuca Abeni static inline 2107daec5798SLuca Abeni void __dl_update(struct dl_bw *dl_b, s64 bw) 2108daec5798SLuca Abeni { 2109daec5798SLuca Abeni struct dl_rq *dl = container_of(dl_b, struct dl_rq, dl_bw); 2110daec5798SLuca Abeni 2111daec5798SLuca Abeni dl->extra_bw += bw; 2112daec5798SLuca Abeni } 2113daec5798SLuca Abeni #endif 2114daec5798SLuca Abeni 2115daec5798SLuca Abeni 211673fbec60SFrederic Weisbecker #ifdef CONFIG_IRQ_TIME_ACCOUNTING 211719d23dbfSFrederic Weisbecker struct irqtime { 211825e2d8c1SFrederic Weisbecker u64 total; 2119a499a5a1SFrederic Weisbecker u64 tick_delta; 212019d23dbfSFrederic Weisbecker u64 irq_start_time; 212119d23dbfSFrederic Weisbecker struct u64_stats_sync sync; 212219d23dbfSFrederic Weisbecker }; 212373fbec60SFrederic Weisbecker 212419d23dbfSFrederic Weisbecker DECLARE_PER_CPU(struct irqtime, cpu_irqtime); 212573fbec60SFrederic Weisbecker 212625e2d8c1SFrederic Weisbecker /* 212725e2d8c1SFrederic Weisbecker * Returns the irqtime minus the softirq time computed by ksoftirqd. 212825e2d8c1SFrederic Weisbecker * Otherwise ksoftirqd's sum_exec_runtime is substracted its own runtime 212925e2d8c1SFrederic Weisbecker * and never move forward. 213025e2d8c1SFrederic Weisbecker */ 213173fbec60SFrederic Weisbecker static inline u64 irq_time_read(int cpu) 213273fbec60SFrederic Weisbecker { 213319d23dbfSFrederic Weisbecker struct irqtime *irqtime = &per_cpu(cpu_irqtime, cpu); 213419d23dbfSFrederic Weisbecker unsigned int seq; 213519d23dbfSFrederic Weisbecker u64 total; 213673fbec60SFrederic Weisbecker 213773fbec60SFrederic Weisbecker do { 213819d23dbfSFrederic Weisbecker seq = __u64_stats_fetch_begin(&irqtime->sync); 213925e2d8c1SFrederic Weisbecker total = irqtime->total; 214019d23dbfSFrederic Weisbecker } while (__u64_stats_fetch_retry(&irqtime->sync, seq)); 214173fbec60SFrederic Weisbecker 214219d23dbfSFrederic Weisbecker return total; 214373fbec60SFrederic Weisbecker } 214473fbec60SFrederic Weisbecker #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ 2145adaf9fcdSRafael J. Wysocki 2146adaf9fcdSRafael J. Wysocki #ifdef CONFIG_CPU_FREQ 2147adaf9fcdSRafael J. Wysocki DECLARE_PER_CPU(struct update_util_data *, cpufreq_update_util_data); 2148adaf9fcdSRafael J. Wysocki 2149adaf9fcdSRafael J. Wysocki /** 2150adaf9fcdSRafael J. Wysocki * cpufreq_update_util - Take a note about CPU utilization changes. 215112bde33dSRafael J. Wysocki * @rq: Runqueue to carry out the update for. 215258919e83SRafael J. Wysocki * @flags: Update reason flags. 2153adaf9fcdSRafael J. Wysocki * 215458919e83SRafael J. Wysocki * This function is called by the scheduler on the CPU whose utilization is 215558919e83SRafael J. Wysocki * being updated. 2156adaf9fcdSRafael J. Wysocki * 2157adaf9fcdSRafael J. Wysocki * It can only be called from RCU-sched read-side critical sections. 2158adaf9fcdSRafael J. Wysocki * 2159adaf9fcdSRafael J. Wysocki * The way cpufreq is currently arranged requires it to evaluate the CPU 2160adaf9fcdSRafael J. Wysocki * performance state (frequency/voltage) on a regular basis to prevent it from 2161adaf9fcdSRafael J. Wysocki * being stuck in a completely inadequate performance level for too long. 2162e0367b12SJuri Lelli * That is not guaranteed to happen if the updates are only triggered from CFS 2163e0367b12SJuri Lelli * and DL, though, because they may not be coming in if only RT tasks are 2164e0367b12SJuri Lelli * active all the time (or there are RT tasks only). 2165adaf9fcdSRafael J. Wysocki * 2166e0367b12SJuri Lelli * As a workaround for that issue, this function is called periodically by the 2167e0367b12SJuri Lelli * RT sched class to trigger extra cpufreq updates to prevent it from stalling, 2168adaf9fcdSRafael J. Wysocki * but that really is a band-aid. Going forward it should be replaced with 2169e0367b12SJuri Lelli * solutions targeted more specifically at RT tasks. 2170adaf9fcdSRafael J. Wysocki */ 217112bde33dSRafael J. Wysocki static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) 2172adaf9fcdSRafael J. Wysocki { 217358919e83SRafael J. Wysocki struct update_util_data *data; 217458919e83SRafael J. Wysocki 2175674e7541SViresh Kumar data = rcu_dereference_sched(*per_cpu_ptr(&cpufreq_update_util_data, 2176674e7541SViresh Kumar cpu_of(rq))); 217758919e83SRafael J. Wysocki if (data) 217812bde33dSRafael J. Wysocki data->func(data, rq_clock(rq), flags); 217912bde33dSRafael J. Wysocki } 2180adaf9fcdSRafael J. Wysocki #else 218112bde33dSRafael J. Wysocki static inline void cpufreq_update_util(struct rq *rq, unsigned int flags) {} 2182adaf9fcdSRafael J. Wysocki #endif /* CONFIG_CPU_FREQ */ 2183be53f58fSLinus Torvalds 21849bdcb44eSRafael J. Wysocki #ifdef arch_scale_freq_capacity 21859bdcb44eSRafael J. Wysocki # ifndef arch_scale_freq_invariant 218697fb7a0aSIngo Molnar # define arch_scale_freq_invariant() true 21879bdcb44eSRafael J. Wysocki # endif 218897fb7a0aSIngo Molnar #else 218997fb7a0aSIngo Molnar # define arch_scale_freq_invariant() false 21909bdcb44eSRafael J. Wysocki #endif 2191d4edd662SJuri Lelli 2192794a56ebSJuri Lelli #ifdef CONFIG_CPU_FREQ_GOV_SCHEDUTIL 21938cc90515SVincent Guittot static inline unsigned long cpu_bw_dl(struct rq *rq) 2194d4edd662SJuri Lelli { 2195d4edd662SJuri Lelli return (rq->dl.running_bw * SCHED_CAPACITY_SCALE) >> BW_SHIFT; 2196d4edd662SJuri Lelli } 2197d4edd662SJuri Lelli 21988cc90515SVincent Guittot static inline unsigned long cpu_util_dl(struct rq *rq) 21998cc90515SVincent Guittot { 22008cc90515SVincent Guittot return READ_ONCE(rq->avg_dl.util_avg); 22018cc90515SVincent Guittot } 22028cc90515SVincent Guittot 2203d4edd662SJuri Lelli static inline unsigned long cpu_util_cfs(struct rq *rq) 2204d4edd662SJuri Lelli { 2205a07630b8SPatrick Bellasi unsigned long util = READ_ONCE(rq->cfs.avg.util_avg); 2206a07630b8SPatrick Bellasi 2207a07630b8SPatrick Bellasi if (sched_feat(UTIL_EST)) { 2208a07630b8SPatrick Bellasi util = max_t(unsigned long, util, 2209a07630b8SPatrick Bellasi READ_ONCE(rq->cfs.avg.util_est.enqueued)); 2210a07630b8SPatrick Bellasi } 2211a07630b8SPatrick Bellasi 2212a07630b8SPatrick Bellasi return util; 2213d4edd662SJuri Lelli } 2214371bf427SVincent Guittot 2215371bf427SVincent Guittot static inline unsigned long cpu_util_rt(struct rq *rq) 2216371bf427SVincent Guittot { 2217dfa444dcSVincent Guittot return READ_ONCE(rq->avg_rt.util_avg); 2218371bf427SVincent Guittot } 22192e62c474SVincent Guittot #endif 22209033ea11SVincent Guittot 22212e62c474SVincent Guittot #ifdef HAVE_SCHED_AVG_IRQ 22229033ea11SVincent Guittot static inline unsigned long cpu_util_irq(struct rq *rq) 22239033ea11SVincent Guittot { 22249033ea11SVincent Guittot return rq->avg_irq.util_avg; 22259033ea11SVincent Guittot } 22262e62c474SVincent Guittot 22272e62c474SVincent Guittot static inline 22282e62c474SVincent Guittot unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 22292e62c474SVincent Guittot { 22302e62c474SVincent Guittot util *= (max - irq); 22312e62c474SVincent Guittot util /= max; 22322e62c474SVincent Guittot 22332e62c474SVincent Guittot return util; 22342e62c474SVincent Guittot 22352e62c474SVincent Guittot } 22369033ea11SVincent Guittot #else 22379033ea11SVincent Guittot static inline unsigned long cpu_util_irq(struct rq *rq) 22389033ea11SVincent Guittot { 22399033ea11SVincent Guittot return 0; 22409033ea11SVincent Guittot } 22419033ea11SVincent Guittot 22422e62c474SVincent Guittot static inline 22432e62c474SVincent Guittot unsigned long scale_irq_capacity(unsigned long util, unsigned long irq, unsigned long max) 22442e62c474SVincent Guittot { 22452e62c474SVincent Guittot return util; 22462e62c474SVincent Guittot } 2247794a56ebSJuri Lelli #endif 2248