1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _KERNEL_STATS_H 3 #define _KERNEL_STATS_H 4 5 #ifdef CONFIG_SCHEDSTATS 6 7 extern struct static_key_false sched_schedstats; 8 9 /* 10 * Expects runqueue lock to be held for atomicity of update 11 */ 12 static inline void 13 rq_sched_info_arrive(struct rq *rq, unsigned long long delta) 14 { 15 if (rq) { 16 rq->rq_sched_info.run_delay += delta; 17 rq->rq_sched_info.pcount++; 18 } 19 } 20 21 /* 22 * Expects runqueue lock to be held for atomicity of update 23 */ 24 static inline void 25 rq_sched_info_depart(struct rq *rq, unsigned long long delta) 26 { 27 if (rq) 28 rq->rq_cpu_time += delta; 29 } 30 31 static inline void 32 rq_sched_info_dequeue(struct rq *rq, unsigned long long delta) 33 { 34 if (rq) 35 rq->rq_sched_info.run_delay += delta; 36 } 37 #define schedstat_enabled() static_branch_unlikely(&sched_schedstats) 38 #define __schedstat_inc(var) do { var++; } while (0) 39 #define schedstat_inc(var) do { if (schedstat_enabled()) { var++; } } while (0) 40 #define __schedstat_add(var, amt) do { var += (amt); } while (0) 41 #define schedstat_add(var, amt) do { if (schedstat_enabled()) { var += (amt); } } while (0) 42 #define __schedstat_set(var, val) do { var = (val); } while (0) 43 #define schedstat_set(var, val) do { if (schedstat_enabled()) { var = (val); } } while (0) 44 #define schedstat_val(var) (var) 45 #define schedstat_val_or_zero(var) ((schedstat_enabled()) ? (var) : 0) 46 47 void __update_stats_wait_start(struct rq *rq, struct task_struct *p, 48 struct sched_statistics *stats); 49 50 void __update_stats_wait_end(struct rq *rq, struct task_struct *p, 51 struct sched_statistics *stats); 52 void __update_stats_enqueue_sleeper(struct rq *rq, struct task_struct *p, 53 struct sched_statistics *stats); 54 55 static inline void 56 check_schedstat_required(void) 57 { 58 if (schedstat_enabled()) 59 return; 60 61 /* Force schedstat enabled if a dependent tracepoint is active */ 62 if (trace_sched_stat_wait_enabled() || 63 trace_sched_stat_sleep_enabled() || 64 trace_sched_stat_iowait_enabled() || 65 trace_sched_stat_blocked_enabled() || 66 trace_sched_stat_runtime_enabled()) 67 printk_deferred_once("Scheduler tracepoints stat_sleep, stat_iowait, stat_blocked and stat_runtime require the kernel parameter schedstats=enable or kernel.sched_schedstats=1\n"); 68 } 69 70 #else /* !CONFIG_SCHEDSTATS: */ 71 72 static inline void rq_sched_info_arrive (struct rq *rq, unsigned long long delta) { } 73 static inline void rq_sched_info_dequeue(struct rq *rq, unsigned long long delta) { } 74 static inline void rq_sched_info_depart (struct rq *rq, unsigned long long delta) { } 75 # define schedstat_enabled() 0 76 # define __schedstat_inc(var) do { } while (0) 77 # define schedstat_inc(var) do { } while (0) 78 # define __schedstat_add(var, amt) do { } while (0) 79 # define schedstat_add(var, amt) do { } while (0) 80 # define __schedstat_set(var, val) do { } while (0) 81 # define schedstat_set(var, val) do { } while (0) 82 # define schedstat_val(var) 0 83 # define schedstat_val_or_zero(var) 0 84 85 # define __update_stats_wait_start(rq, p, stats) do { } while (0) 86 # define __update_stats_wait_end(rq, p, stats) do { } while (0) 87 # define __update_stats_enqueue_sleeper(rq, p, stats) do { } while (0) 88 # define check_schedstat_required() do { } while (0) 89 90 #endif /* CONFIG_SCHEDSTATS */ 91 92 #ifdef CONFIG_FAIR_GROUP_SCHED 93 struct sched_entity_stats { 94 struct sched_entity se; 95 struct sched_statistics stats; 96 } __no_randomize_layout; 97 #endif 98 99 static inline struct sched_statistics * 100 __schedstats_from_se(struct sched_entity *se) 101 { 102 #ifdef CONFIG_FAIR_GROUP_SCHED 103 if (!entity_is_task(se)) 104 return &container_of(se, struct sched_entity_stats, se)->stats; 105 #endif 106 return &task_of(se)->stats; 107 } 108 109 #ifdef CONFIG_PSI 110 void psi_task_change(struct task_struct *task, int clear, int set); 111 void psi_task_switch(struct task_struct *prev, struct task_struct *next, 112 bool sleep); 113 void psi_account_irqtime(struct task_struct *task, u32 delta); 114 115 /* 116 * PSI tracks state that persists across sleeps, such as iowaits and 117 * memory stalls. As a result, it has to distinguish between sleeps, 118 * where a task's runnable state changes, and requeues, where a task 119 * and its state are being moved between CPUs and runqueues. 120 */ 121 static inline void psi_enqueue(struct task_struct *p, bool wakeup) 122 { 123 int clear = 0, set = TSK_RUNNING; 124 125 if (static_branch_likely(&psi_disabled)) 126 return; 127 128 if (p->in_memstall) 129 set |= TSK_MEMSTALL_RUNNING; 130 131 if (!wakeup || p->sched_psi_wake_requeue) { 132 if (p->in_memstall) 133 set |= TSK_MEMSTALL; 134 if (p->sched_psi_wake_requeue) 135 p->sched_psi_wake_requeue = 0; 136 } else { 137 if (p->in_iowait) 138 clear |= TSK_IOWAIT; 139 } 140 141 psi_task_change(p, clear, set); 142 } 143 144 static inline void psi_dequeue(struct task_struct *p, bool sleep) 145 { 146 int clear = TSK_RUNNING; 147 148 if (static_branch_likely(&psi_disabled)) 149 return; 150 151 /* 152 * A voluntary sleep is a dequeue followed by a task switch. To 153 * avoid walking all ancestors twice, psi_task_switch() handles 154 * TSK_RUNNING and TSK_IOWAIT for us when it moves TSK_ONCPU. 155 * Do nothing here. 156 */ 157 if (sleep) 158 return; 159 160 if (p->in_memstall) 161 clear |= (TSK_MEMSTALL | TSK_MEMSTALL_RUNNING); 162 163 psi_task_change(p, clear, 0); 164 } 165 166 static inline void psi_ttwu_dequeue(struct task_struct *p) 167 { 168 if (static_branch_likely(&psi_disabled)) 169 return; 170 /* 171 * Is the task being migrated during a wakeup? Make sure to 172 * deregister its sleep-persistent psi states from the old 173 * queue, and let psi_enqueue() know it has to requeue. 174 */ 175 if (unlikely(p->in_iowait || p->in_memstall)) { 176 struct rq_flags rf; 177 struct rq *rq; 178 int clear = 0; 179 180 if (p->in_iowait) 181 clear |= TSK_IOWAIT; 182 if (p->in_memstall) 183 clear |= TSK_MEMSTALL; 184 185 rq = __task_rq_lock(p, &rf); 186 psi_task_change(p, clear, 0); 187 p->sched_psi_wake_requeue = 1; 188 __task_rq_unlock(rq, &rf); 189 } 190 } 191 192 static inline void psi_sched_switch(struct task_struct *prev, 193 struct task_struct *next, 194 bool sleep) 195 { 196 if (static_branch_likely(&psi_disabled)) 197 return; 198 199 psi_task_switch(prev, next, sleep); 200 } 201 202 #else /* CONFIG_PSI */ 203 static inline void psi_enqueue(struct task_struct *p, bool wakeup) {} 204 static inline void psi_dequeue(struct task_struct *p, bool sleep) {} 205 static inline void psi_ttwu_dequeue(struct task_struct *p) {} 206 static inline void psi_sched_switch(struct task_struct *prev, 207 struct task_struct *next, 208 bool sleep) {} 209 static inline void psi_account_irqtime(struct task_struct *task, u32 delta) {} 210 #endif /* CONFIG_PSI */ 211 212 #ifdef CONFIG_SCHED_INFO 213 /* 214 * We are interested in knowing how long it was from the *first* time a 215 * task was queued to the time that it finally hit a CPU, we call this routine 216 * from dequeue_task() to account for possible rq->clock skew across CPUs. The 217 * delta taken on each CPU would annul the skew. 218 */ 219 static inline void sched_info_dequeue(struct rq *rq, struct task_struct *t) 220 { 221 unsigned long long delta = 0; 222 223 if (!t->sched_info.last_queued) 224 return; 225 226 delta = rq_clock(rq) - t->sched_info.last_queued; 227 t->sched_info.last_queued = 0; 228 t->sched_info.run_delay += delta; 229 230 rq_sched_info_dequeue(rq, delta); 231 } 232 233 /* 234 * Called when a task finally hits the CPU. We can now calculate how 235 * long it was waiting to run. We also note when it began so that we 236 * can keep stats on how long its timeslice is. 237 */ 238 static void sched_info_arrive(struct rq *rq, struct task_struct *t) 239 { 240 unsigned long long now, delta = 0; 241 242 if (!t->sched_info.last_queued) 243 return; 244 245 now = rq_clock(rq); 246 delta = now - t->sched_info.last_queued; 247 t->sched_info.last_queued = 0; 248 t->sched_info.run_delay += delta; 249 t->sched_info.last_arrival = now; 250 t->sched_info.pcount++; 251 252 rq_sched_info_arrive(rq, delta); 253 } 254 255 /* 256 * This function is only called from enqueue_task(), but also only updates 257 * the timestamp if it is already not set. It's assumed that 258 * sched_info_dequeue() will clear that stamp when appropriate. 259 */ 260 static inline void sched_info_enqueue(struct rq *rq, struct task_struct *t) 261 { 262 if (!t->sched_info.last_queued) 263 t->sched_info.last_queued = rq_clock(rq); 264 } 265 266 /* 267 * Called when a process ceases being the active-running process involuntarily 268 * due, typically, to expiring its time slice (this may also be called when 269 * switching to the idle task). Now we can calculate how long we ran. 270 * Also, if the process is still in the TASK_RUNNING state, call 271 * sched_info_enqueue() to mark that it has now again started waiting on 272 * the runqueue. 273 */ 274 static inline void sched_info_depart(struct rq *rq, struct task_struct *t) 275 { 276 unsigned long long delta = rq_clock(rq) - t->sched_info.last_arrival; 277 278 rq_sched_info_depart(rq, delta); 279 280 if (task_is_running(t)) 281 sched_info_enqueue(rq, t); 282 } 283 284 /* 285 * Called when tasks are switched involuntarily due, typically, to expiring 286 * their time slice. (This may also be called when switching to or from 287 * the idle task.) We are only called when prev != next. 288 */ 289 static inline void 290 sched_info_switch(struct rq *rq, struct task_struct *prev, struct task_struct *next) 291 { 292 /* 293 * prev now departs the CPU. It's not interesting to record 294 * stats about how efficient we were at scheduling the idle 295 * process, however. 296 */ 297 if (prev != rq->idle) 298 sched_info_depart(rq, prev); 299 300 if (next != rq->idle) 301 sched_info_arrive(rq, next); 302 } 303 304 #else /* !CONFIG_SCHED_INFO: */ 305 # define sched_info_enqueue(rq, t) do { } while (0) 306 # define sched_info_dequeue(rq, t) do { } while (0) 307 # define sched_info_switch(rq, t, next) do { } while (0) 308 #endif /* CONFIG_SCHED_INFO */ 309 310 #endif /* _KERNEL_STATS_H */ 311