| 1 | /* |
| 2 | * Implement CPU time clocks for the POSIX clock interface. |
| 3 | */ |
| 4 | |
| 5 | #include <linux/sched.h> |
| 6 | #include <linux/posix-timers.h> |
| 7 | #include <linux/errno.h> |
| 8 | #include <linux/math64.h> |
| 9 | #include <asm/uaccess.h> |
| 10 | #include <linux/kernel_stat.h> |
| 11 | #include <trace/events/timer.h> |
| 12 | #include <linux/random.h> |
| 13 | #include <linux/tick.h> |
| 14 | #include <linux/workqueue.h> |
| 15 | |
| 16 | /* |
| 17 | * Called after updating RLIMIT_CPU to run cpu timer and update |
| 18 | * tsk->signal->cputime_expires expiration cache if necessary. Needs |
| 19 | * siglock protection since other code may update expiration cache as |
| 20 | * well. |
| 21 | */ |
| 22 | void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new) |
| 23 | { |
| 24 | cputime_t cputime = secs_to_cputime(rlim_new); |
| 25 | |
| 26 | spin_lock_irq(&task->sighand->siglock); |
| 27 | set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL); |
| 28 | spin_unlock_irq(&task->sighand->siglock); |
| 29 | } |
| 30 | |
| 31 | static int check_clock(const clockid_t which_clock) |
| 32 | { |
| 33 | int error = 0; |
| 34 | struct task_struct *p; |
| 35 | const pid_t pid = CPUCLOCK_PID(which_clock); |
| 36 | |
| 37 | if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX) |
| 38 | return -EINVAL; |
| 39 | |
| 40 | if (pid == 0) |
| 41 | return 0; |
| 42 | |
| 43 | rcu_read_lock(); |
| 44 | p = find_task_by_vpid(pid); |
| 45 | if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ? |
| 46 | same_thread_group(p, current) : has_group_leader_pid(p))) { |
| 47 | error = -EINVAL; |
| 48 | } |
| 49 | rcu_read_unlock(); |
| 50 | |
| 51 | return error; |
| 52 | } |
| 53 | |
| 54 | static inline unsigned long long |
| 55 | timespec_to_sample(const clockid_t which_clock, const struct timespec *tp) |
| 56 | { |
| 57 | unsigned long long ret; |
| 58 | |
| 59 | ret = 0; /* high half always zero when .cpu used */ |
| 60 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { |
| 61 | ret = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec; |
| 62 | } else { |
| 63 | ret = cputime_to_expires(timespec_to_cputime(tp)); |
| 64 | } |
| 65 | return ret; |
| 66 | } |
| 67 | |
| 68 | static void sample_to_timespec(const clockid_t which_clock, |
| 69 | unsigned long long expires, |
| 70 | struct timespec *tp) |
| 71 | { |
| 72 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) |
| 73 | *tp = ns_to_timespec(expires); |
| 74 | else |
| 75 | cputime_to_timespec((__force cputime_t)expires, tp); |
| 76 | } |
| 77 | |
| 78 | /* |
| 79 | * Update expiry time from increment, and increase overrun count, |
| 80 | * given the current clock sample. |
| 81 | */ |
| 82 | static void bump_cpu_timer(struct k_itimer *timer, |
| 83 | unsigned long long now) |
| 84 | { |
| 85 | int i; |
| 86 | unsigned long long delta, incr; |
| 87 | |
| 88 | if (timer->it.cpu.incr == 0) |
| 89 | return; |
| 90 | |
| 91 | if (now < timer->it.cpu.expires) |
| 92 | return; |
| 93 | |
| 94 | incr = timer->it.cpu.incr; |
| 95 | delta = now + incr - timer->it.cpu.expires; |
| 96 | |
| 97 | /* Don't use (incr*2 < delta), incr*2 might overflow. */ |
| 98 | for (i = 0; incr < delta - incr; i++) |
| 99 | incr = incr << 1; |
| 100 | |
| 101 | for (; i >= 0; incr >>= 1, i--) { |
| 102 | if (delta < incr) |
| 103 | continue; |
| 104 | |
| 105 | timer->it.cpu.expires += incr; |
| 106 | timer->it_overrun += 1 << i; |
| 107 | delta -= incr; |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | /** |
| 112 | * task_cputime_zero - Check a task_cputime struct for all zero fields. |
| 113 | * |
| 114 | * @cputime: The struct to compare. |
| 115 | * |
| 116 | * Checks @cputime to see if all fields are zero. Returns true if all fields |
| 117 | * are zero, false if any field is nonzero. |
| 118 | */ |
| 119 | static inline int task_cputime_zero(const struct task_cputime *cputime) |
| 120 | { |
| 121 | if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime) |
| 122 | return 1; |
| 123 | return 0; |
| 124 | } |
| 125 | |
| 126 | static inline unsigned long long prof_ticks(struct task_struct *p) |
| 127 | { |
| 128 | cputime_t utime, stime; |
| 129 | |
| 130 | task_cputime(p, &utime, &stime); |
| 131 | |
| 132 | return cputime_to_expires(utime + stime); |
| 133 | } |
| 134 | static inline unsigned long long virt_ticks(struct task_struct *p) |
| 135 | { |
| 136 | cputime_t utime; |
| 137 | |
| 138 | task_cputime(p, &utime, NULL); |
| 139 | |
| 140 | return cputime_to_expires(utime); |
| 141 | } |
| 142 | |
| 143 | static int |
| 144 | posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp) |
| 145 | { |
| 146 | int error = check_clock(which_clock); |
| 147 | if (!error) { |
| 148 | tp->tv_sec = 0; |
| 149 | tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ); |
| 150 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { |
| 151 | /* |
| 152 | * If sched_clock is using a cycle counter, we |
| 153 | * don't have any idea of its true resolution |
| 154 | * exported, but it is much more than 1s/HZ. |
| 155 | */ |
| 156 | tp->tv_nsec = 1; |
| 157 | } |
| 158 | } |
| 159 | return error; |
| 160 | } |
| 161 | |
| 162 | static int |
| 163 | posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp) |
| 164 | { |
| 165 | /* |
| 166 | * You can never reset a CPU clock, but we check for other errors |
| 167 | * in the call before failing with EPERM. |
| 168 | */ |
| 169 | int error = check_clock(which_clock); |
| 170 | if (error == 0) { |
| 171 | error = -EPERM; |
| 172 | } |
| 173 | return error; |
| 174 | } |
| 175 | |
| 176 | |
| 177 | /* |
| 178 | * Sample a per-thread clock for the given task. |
| 179 | */ |
| 180 | static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p, |
| 181 | unsigned long long *sample) |
| 182 | { |
| 183 | switch (CPUCLOCK_WHICH(which_clock)) { |
| 184 | default: |
| 185 | return -EINVAL; |
| 186 | case CPUCLOCK_PROF: |
| 187 | *sample = prof_ticks(p); |
| 188 | break; |
| 189 | case CPUCLOCK_VIRT: |
| 190 | *sample = virt_ticks(p); |
| 191 | break; |
| 192 | case CPUCLOCK_SCHED: |
| 193 | *sample = task_sched_runtime(p); |
| 194 | break; |
| 195 | } |
| 196 | return 0; |
| 197 | } |
| 198 | |
| 199 | static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b) |
| 200 | { |
| 201 | if (b->utime > a->utime) |
| 202 | a->utime = b->utime; |
| 203 | |
| 204 | if (b->stime > a->stime) |
| 205 | a->stime = b->stime; |
| 206 | |
| 207 | if (b->sum_exec_runtime > a->sum_exec_runtime) |
| 208 | a->sum_exec_runtime = b->sum_exec_runtime; |
| 209 | } |
| 210 | |
| 211 | void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times) |
| 212 | { |
| 213 | struct thread_group_cputimer *cputimer = &tsk->signal->cputimer; |
| 214 | struct task_cputime sum; |
| 215 | unsigned long flags; |
| 216 | |
| 217 | if (!cputimer->running) { |
| 218 | /* |
| 219 | * The POSIX timer interface allows for absolute time expiry |
| 220 | * values through the TIMER_ABSTIME flag, therefore we have |
| 221 | * to synchronize the timer to the clock every time we start |
| 222 | * it. |
| 223 | */ |
| 224 | thread_group_cputime(tsk, &sum); |
| 225 | raw_spin_lock_irqsave(&cputimer->lock, flags); |
| 226 | cputimer->running = 1; |
| 227 | update_gt_cputime(&cputimer->cputime, &sum); |
| 228 | } else |
| 229 | raw_spin_lock_irqsave(&cputimer->lock, flags); |
| 230 | *times = cputimer->cputime; |
| 231 | raw_spin_unlock_irqrestore(&cputimer->lock, flags); |
| 232 | } |
| 233 | |
| 234 | /* |
| 235 | * Sample a process (thread group) clock for the given group_leader task. |
| 236 | * Must be called with tasklist_lock held for reading. |
| 237 | */ |
| 238 | static int cpu_clock_sample_group(const clockid_t which_clock, |
| 239 | struct task_struct *p, |
| 240 | unsigned long long *sample) |
| 241 | { |
| 242 | struct task_cputime cputime; |
| 243 | |
| 244 | switch (CPUCLOCK_WHICH(which_clock)) { |
| 245 | default: |
| 246 | return -EINVAL; |
| 247 | case CPUCLOCK_PROF: |
| 248 | thread_group_cputime(p, &cputime); |
| 249 | *sample = cputime_to_expires(cputime.utime + cputime.stime); |
| 250 | break; |
| 251 | case CPUCLOCK_VIRT: |
| 252 | thread_group_cputime(p, &cputime); |
| 253 | *sample = cputime_to_expires(cputime.utime); |
| 254 | break; |
| 255 | case CPUCLOCK_SCHED: |
| 256 | thread_group_cputime(p, &cputime); |
| 257 | *sample = cputime.sum_exec_runtime; |
| 258 | break; |
| 259 | } |
| 260 | return 0; |
| 261 | } |
| 262 | |
| 263 | |
| 264 | static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp) |
| 265 | { |
| 266 | const pid_t pid = CPUCLOCK_PID(which_clock); |
| 267 | int error = -EINVAL; |
| 268 | unsigned long long rtn; |
| 269 | |
| 270 | if (pid == 0) { |
| 271 | /* |
| 272 | * Special case constant value for our own clocks. |
| 273 | * We don't have to do any lookup to find ourselves. |
| 274 | */ |
| 275 | if (CPUCLOCK_PERTHREAD(which_clock)) { |
| 276 | /* |
| 277 | * Sampling just ourselves we can do with no locking. |
| 278 | */ |
| 279 | error = cpu_clock_sample(which_clock, |
| 280 | current, &rtn); |
| 281 | } else { |
| 282 | read_lock(&tasklist_lock); |
| 283 | error = cpu_clock_sample_group(which_clock, |
| 284 | current, &rtn); |
| 285 | read_unlock(&tasklist_lock); |
| 286 | } |
| 287 | } else { |
| 288 | /* |
| 289 | * Find the given PID, and validate that the caller |
| 290 | * should be able to see it. |
| 291 | */ |
| 292 | struct task_struct *p; |
| 293 | rcu_read_lock(); |
| 294 | p = find_task_by_vpid(pid); |
| 295 | if (p) { |
| 296 | if (CPUCLOCK_PERTHREAD(which_clock)) { |
| 297 | if (same_thread_group(p, current)) { |
| 298 | error = cpu_clock_sample(which_clock, |
| 299 | p, &rtn); |
| 300 | } |
| 301 | } else { |
| 302 | read_lock(&tasklist_lock); |
| 303 | if (thread_group_leader(p) && p->sighand) { |
| 304 | error = |
| 305 | cpu_clock_sample_group(which_clock, |
| 306 | p, &rtn); |
| 307 | } |
| 308 | read_unlock(&tasklist_lock); |
| 309 | } |
| 310 | } |
| 311 | rcu_read_unlock(); |
| 312 | } |
| 313 | |
| 314 | if (error) |
| 315 | return error; |
| 316 | sample_to_timespec(which_clock, rtn, tp); |
| 317 | return 0; |
| 318 | } |
| 319 | |
| 320 | |
| 321 | /* |
| 322 | * Validate the clockid_t for a new CPU-clock timer, and initialize the timer. |
| 323 | * This is called from sys_timer_create() and do_cpu_nanosleep() with the |
| 324 | * new timer already all-zeros initialized. |
| 325 | */ |
| 326 | static int posix_cpu_timer_create(struct k_itimer *new_timer) |
| 327 | { |
| 328 | int ret = 0; |
| 329 | const pid_t pid = CPUCLOCK_PID(new_timer->it_clock); |
| 330 | struct task_struct *p; |
| 331 | |
| 332 | if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX) |
| 333 | return -EINVAL; |
| 334 | |
| 335 | INIT_LIST_HEAD(&new_timer->it.cpu.entry); |
| 336 | |
| 337 | rcu_read_lock(); |
| 338 | if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) { |
| 339 | if (pid == 0) { |
| 340 | p = current; |
| 341 | } else { |
| 342 | p = find_task_by_vpid(pid); |
| 343 | if (p && !same_thread_group(p, current)) |
| 344 | p = NULL; |
| 345 | } |
| 346 | } else { |
| 347 | if (pid == 0) { |
| 348 | p = current->group_leader; |
| 349 | } else { |
| 350 | p = find_task_by_vpid(pid); |
| 351 | if (p && !has_group_leader_pid(p)) |
| 352 | p = NULL; |
| 353 | } |
| 354 | } |
| 355 | new_timer->it.cpu.task = p; |
| 356 | if (p) { |
| 357 | get_task_struct(p); |
| 358 | } else { |
| 359 | ret = -EINVAL; |
| 360 | } |
| 361 | rcu_read_unlock(); |
| 362 | |
| 363 | return ret; |
| 364 | } |
| 365 | |
| 366 | /* |
| 367 | * Clean up a CPU-clock timer that is about to be destroyed. |
| 368 | * This is called from timer deletion with the timer already locked. |
| 369 | * If we return TIMER_RETRY, it's necessary to release the timer's lock |
| 370 | * and try again. (This happens when the timer is in the middle of firing.) |
| 371 | */ |
| 372 | static int posix_cpu_timer_del(struct k_itimer *timer) |
| 373 | { |
| 374 | struct task_struct *p = timer->it.cpu.task; |
| 375 | int ret = 0; |
| 376 | |
| 377 | if (likely(p != NULL)) { |
| 378 | read_lock(&tasklist_lock); |
| 379 | if (unlikely(p->sighand == NULL)) { |
| 380 | /* |
| 381 | * We raced with the reaping of the task. |
| 382 | * The deletion should have cleared us off the list. |
| 383 | */ |
| 384 | BUG_ON(!list_empty(&timer->it.cpu.entry)); |
| 385 | } else { |
| 386 | spin_lock(&p->sighand->siglock); |
| 387 | if (timer->it.cpu.firing) |
| 388 | ret = TIMER_RETRY; |
| 389 | else |
| 390 | list_del(&timer->it.cpu.entry); |
| 391 | spin_unlock(&p->sighand->siglock); |
| 392 | } |
| 393 | read_unlock(&tasklist_lock); |
| 394 | |
| 395 | if (!ret) |
| 396 | put_task_struct(p); |
| 397 | } |
| 398 | |
| 399 | return ret; |
| 400 | } |
| 401 | |
| 402 | static void cleanup_timers_list(struct list_head *head, |
| 403 | unsigned long long curr) |
| 404 | { |
| 405 | struct cpu_timer_list *timer, *next; |
| 406 | |
| 407 | list_for_each_entry_safe(timer, next, head, entry) |
| 408 | list_del_init(&timer->entry); |
| 409 | } |
| 410 | |
| 411 | /* |
| 412 | * Clean out CPU timers still ticking when a thread exited. The task |
| 413 | * pointer is cleared, and the expiry time is replaced with the residual |
| 414 | * time for later timer_gettime calls to return. |
| 415 | * This must be called with the siglock held. |
| 416 | */ |
| 417 | static void cleanup_timers(struct list_head *head, |
| 418 | cputime_t utime, cputime_t stime, |
| 419 | unsigned long long sum_exec_runtime) |
| 420 | { |
| 421 | |
| 422 | cputime_t ptime = utime + stime; |
| 423 | |
| 424 | cleanup_timers_list(head, cputime_to_expires(ptime)); |
| 425 | cleanup_timers_list(++head, cputime_to_expires(utime)); |
| 426 | cleanup_timers_list(++head, sum_exec_runtime); |
| 427 | } |
| 428 | |
| 429 | /* |
| 430 | * These are both called with the siglock held, when the current thread |
| 431 | * is being reaped. When the final (leader) thread in the group is reaped, |
| 432 | * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit. |
| 433 | */ |
| 434 | void posix_cpu_timers_exit(struct task_struct *tsk) |
| 435 | { |
| 436 | cputime_t utime, stime; |
| 437 | |
| 438 | add_device_randomness((const void*) &tsk->se.sum_exec_runtime, |
| 439 | sizeof(unsigned long long)); |
| 440 | task_cputime(tsk, &utime, &stime); |
| 441 | cleanup_timers(tsk->cpu_timers, |
| 442 | utime, stime, tsk->se.sum_exec_runtime); |
| 443 | |
| 444 | } |
| 445 | void posix_cpu_timers_exit_group(struct task_struct *tsk) |
| 446 | { |
| 447 | struct signal_struct *const sig = tsk->signal; |
| 448 | cputime_t utime, stime; |
| 449 | |
| 450 | task_cputime(tsk, &utime, &stime); |
| 451 | cleanup_timers(tsk->signal->cpu_timers, |
| 452 | utime + sig->utime, stime + sig->stime, |
| 453 | tsk->se.sum_exec_runtime + sig->sum_sched_runtime); |
| 454 | } |
| 455 | |
| 456 | static void clear_dead_task(struct k_itimer *itimer, unsigned long long now) |
| 457 | { |
| 458 | struct cpu_timer_list *timer = &itimer->it.cpu; |
| 459 | |
| 460 | /* |
| 461 | * That's all for this thread or process. |
| 462 | * We leave our residual in expires to be reported. |
| 463 | */ |
| 464 | put_task_struct(timer->task); |
| 465 | timer->task = NULL; |
| 466 | if (timer->expires < now) { |
| 467 | timer->expires = 0; |
| 468 | } else { |
| 469 | timer->expires -= now; |
| 470 | } |
| 471 | } |
| 472 | |
| 473 | static inline int expires_gt(cputime_t expires, cputime_t new_exp) |
| 474 | { |
| 475 | return expires == 0 || expires > new_exp; |
| 476 | } |
| 477 | |
| 478 | /* |
| 479 | * Insert the timer on the appropriate list before any timers that |
| 480 | * expire later. This must be called with the tasklist_lock held |
| 481 | * for reading, interrupts disabled and p->sighand->siglock taken. |
| 482 | */ |
| 483 | static void arm_timer(struct k_itimer *timer) |
| 484 | { |
| 485 | struct task_struct *p = timer->it.cpu.task; |
| 486 | struct list_head *head, *listpos; |
| 487 | struct task_cputime *cputime_expires; |
| 488 | struct cpu_timer_list *const nt = &timer->it.cpu; |
| 489 | struct cpu_timer_list *next; |
| 490 | |
| 491 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { |
| 492 | head = p->cpu_timers; |
| 493 | cputime_expires = &p->cputime_expires; |
| 494 | } else { |
| 495 | head = p->signal->cpu_timers; |
| 496 | cputime_expires = &p->signal->cputime_expires; |
| 497 | } |
| 498 | head += CPUCLOCK_WHICH(timer->it_clock); |
| 499 | |
| 500 | listpos = head; |
| 501 | list_for_each_entry(next, head, entry) { |
| 502 | if (nt->expires < next->expires) |
| 503 | break; |
| 504 | listpos = &next->entry; |
| 505 | } |
| 506 | list_add(&nt->entry, listpos); |
| 507 | |
| 508 | if (listpos == head) { |
| 509 | unsigned long long exp = nt->expires; |
| 510 | |
| 511 | /* |
| 512 | * We are the new earliest-expiring POSIX 1.b timer, hence |
| 513 | * need to update expiration cache. Take into account that |
| 514 | * for process timers we share expiration cache with itimers |
| 515 | * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME. |
| 516 | */ |
| 517 | |
| 518 | switch (CPUCLOCK_WHICH(timer->it_clock)) { |
| 519 | case CPUCLOCK_PROF: |
| 520 | if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp))) |
| 521 | cputime_expires->prof_exp = expires_to_cputime(exp); |
| 522 | break; |
| 523 | case CPUCLOCK_VIRT: |
| 524 | if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp))) |
| 525 | cputime_expires->virt_exp = expires_to_cputime(exp); |
| 526 | break; |
| 527 | case CPUCLOCK_SCHED: |
| 528 | if (cputime_expires->sched_exp == 0 || |
| 529 | cputime_expires->sched_exp > exp) |
| 530 | cputime_expires->sched_exp = exp; |
| 531 | break; |
| 532 | } |
| 533 | } |
| 534 | } |
| 535 | |
| 536 | /* |
| 537 | * The timer is locked, fire it and arrange for its reload. |
| 538 | */ |
| 539 | static void cpu_timer_fire(struct k_itimer *timer) |
| 540 | { |
| 541 | if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) { |
| 542 | /* |
| 543 | * User don't want any signal. |
| 544 | */ |
| 545 | timer->it.cpu.expires = 0; |
| 546 | } else if (unlikely(timer->sigq == NULL)) { |
| 547 | /* |
| 548 | * This a special case for clock_nanosleep, |
| 549 | * not a normal timer from sys_timer_create. |
| 550 | */ |
| 551 | wake_up_process(timer->it_process); |
| 552 | timer->it.cpu.expires = 0; |
| 553 | } else if (timer->it.cpu.incr == 0) { |
| 554 | /* |
| 555 | * One-shot timer. Clear it as soon as it's fired. |
| 556 | */ |
| 557 | posix_timer_event(timer, 0); |
| 558 | timer->it.cpu.expires = 0; |
| 559 | } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) { |
| 560 | /* |
| 561 | * The signal did not get queued because the signal |
| 562 | * was ignored, so we won't get any callback to |
| 563 | * reload the timer. But we need to keep it |
| 564 | * ticking in case the signal is deliverable next time. |
| 565 | */ |
| 566 | posix_cpu_timer_schedule(timer); |
| 567 | } |
| 568 | } |
| 569 | |
| 570 | /* |
| 571 | * Sample a process (thread group) timer for the given group_leader task. |
| 572 | * Must be called with tasklist_lock held for reading. |
| 573 | */ |
| 574 | static int cpu_timer_sample_group(const clockid_t which_clock, |
| 575 | struct task_struct *p, |
| 576 | unsigned long long *sample) |
| 577 | { |
| 578 | struct task_cputime cputime; |
| 579 | |
| 580 | thread_group_cputimer(p, &cputime); |
| 581 | switch (CPUCLOCK_WHICH(which_clock)) { |
| 582 | default: |
| 583 | return -EINVAL; |
| 584 | case CPUCLOCK_PROF: |
| 585 | *sample = cputime_to_expires(cputime.utime + cputime.stime); |
| 586 | break; |
| 587 | case CPUCLOCK_VIRT: |
| 588 | *sample = cputime_to_expires(cputime.utime); |
| 589 | break; |
| 590 | case CPUCLOCK_SCHED: |
| 591 | *sample = cputime.sum_exec_runtime + task_delta_exec(p); |
| 592 | break; |
| 593 | } |
| 594 | return 0; |
| 595 | } |
| 596 | |
| 597 | #ifdef CONFIG_NO_HZ_FULL |
| 598 | static void nohz_kick_work_fn(struct work_struct *work) |
| 599 | { |
| 600 | tick_nohz_full_kick_all(); |
| 601 | } |
| 602 | |
| 603 | static DECLARE_WORK(nohz_kick_work, nohz_kick_work_fn); |
| 604 | |
| 605 | /* |
| 606 | * We need the IPIs to be sent from sane process context. |
| 607 | * The posix cpu timers are always set with irqs disabled. |
| 608 | */ |
| 609 | static void posix_cpu_timer_kick_nohz(void) |
| 610 | { |
| 611 | schedule_work(&nohz_kick_work); |
| 612 | } |
| 613 | |
| 614 | bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk) |
| 615 | { |
| 616 | if (!task_cputime_zero(&tsk->cputime_expires)) |
| 617 | return false; |
| 618 | |
| 619 | if (tsk->signal->cputimer.running) |
| 620 | return false; |
| 621 | |
| 622 | return true; |
| 623 | } |
| 624 | #else |
| 625 | static inline void posix_cpu_timer_kick_nohz(void) { } |
| 626 | #endif |
| 627 | |
| 628 | /* |
| 629 | * Guts of sys_timer_settime for CPU timers. |
| 630 | * This is called with the timer locked and interrupts disabled. |
| 631 | * If we return TIMER_RETRY, it's necessary to release the timer's lock |
| 632 | * and try again. (This happens when the timer is in the middle of firing.) |
| 633 | */ |
| 634 | static int posix_cpu_timer_set(struct k_itimer *timer, int flags, |
| 635 | struct itimerspec *new, struct itimerspec *old) |
| 636 | { |
| 637 | struct task_struct *p = timer->it.cpu.task; |
| 638 | unsigned long long old_expires, new_expires, old_incr, val; |
| 639 | int ret; |
| 640 | |
| 641 | if (unlikely(p == NULL)) { |
| 642 | /* |
| 643 | * Timer refers to a dead task's clock. |
| 644 | */ |
| 645 | return -ESRCH; |
| 646 | } |
| 647 | |
| 648 | new_expires = timespec_to_sample(timer->it_clock, &new->it_value); |
| 649 | |
| 650 | read_lock(&tasklist_lock); |
| 651 | /* |
| 652 | * We need the tasklist_lock to protect against reaping that |
| 653 | * clears p->sighand. If p has just been reaped, we can no |
| 654 | * longer get any information about it at all. |
| 655 | */ |
| 656 | if (unlikely(p->sighand == NULL)) { |
| 657 | read_unlock(&tasklist_lock); |
| 658 | put_task_struct(p); |
| 659 | timer->it.cpu.task = NULL; |
| 660 | return -ESRCH; |
| 661 | } |
| 662 | |
| 663 | /* |
| 664 | * Disarm any old timer after extracting its expiry time. |
| 665 | */ |
| 666 | BUG_ON(!irqs_disabled()); |
| 667 | |
| 668 | ret = 0; |
| 669 | old_incr = timer->it.cpu.incr; |
| 670 | spin_lock(&p->sighand->siglock); |
| 671 | old_expires = timer->it.cpu.expires; |
| 672 | if (unlikely(timer->it.cpu.firing)) { |
| 673 | timer->it.cpu.firing = -1; |
| 674 | ret = TIMER_RETRY; |
| 675 | } else |
| 676 | list_del_init(&timer->it.cpu.entry); |
| 677 | |
| 678 | /* |
| 679 | * We need to sample the current value to convert the new |
| 680 | * value from to relative and absolute, and to convert the |
| 681 | * old value from absolute to relative. To set a process |
| 682 | * timer, we need a sample to balance the thread expiry |
| 683 | * times (in arm_timer). With an absolute time, we must |
| 684 | * check if it's already passed. In short, we need a sample. |
| 685 | */ |
| 686 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { |
| 687 | cpu_clock_sample(timer->it_clock, p, &val); |
| 688 | } else { |
| 689 | cpu_timer_sample_group(timer->it_clock, p, &val); |
| 690 | } |
| 691 | |
| 692 | if (old) { |
| 693 | if (old_expires == 0) { |
| 694 | old->it_value.tv_sec = 0; |
| 695 | old->it_value.tv_nsec = 0; |
| 696 | } else { |
| 697 | /* |
| 698 | * Update the timer in case it has |
| 699 | * overrun already. If it has, |
| 700 | * we'll report it as having overrun |
| 701 | * and with the next reloaded timer |
| 702 | * already ticking, though we are |
| 703 | * swallowing that pending |
| 704 | * notification here to install the |
| 705 | * new setting. |
| 706 | */ |
| 707 | bump_cpu_timer(timer, val); |
| 708 | if (val < timer->it.cpu.expires) { |
| 709 | old_expires = timer->it.cpu.expires - val; |
| 710 | sample_to_timespec(timer->it_clock, |
| 711 | old_expires, |
| 712 | &old->it_value); |
| 713 | } else { |
| 714 | old->it_value.tv_nsec = 1; |
| 715 | old->it_value.tv_sec = 0; |
| 716 | } |
| 717 | } |
| 718 | } |
| 719 | |
| 720 | if (unlikely(ret)) { |
| 721 | /* |
| 722 | * We are colliding with the timer actually firing. |
| 723 | * Punt after filling in the timer's old value, and |
| 724 | * disable this firing since we are already reporting |
| 725 | * it as an overrun (thanks to bump_cpu_timer above). |
| 726 | */ |
| 727 | spin_unlock(&p->sighand->siglock); |
| 728 | read_unlock(&tasklist_lock); |
| 729 | goto out; |
| 730 | } |
| 731 | |
| 732 | if (new_expires != 0 && !(flags & TIMER_ABSTIME)) { |
| 733 | new_expires += val; |
| 734 | } |
| 735 | |
| 736 | /* |
| 737 | * Install the new expiry time (or zero). |
| 738 | * For a timer with no notification action, we don't actually |
| 739 | * arm the timer (we'll just fake it for timer_gettime). |
| 740 | */ |
| 741 | timer->it.cpu.expires = new_expires; |
| 742 | if (new_expires != 0 && val < new_expires) { |
| 743 | arm_timer(timer); |
| 744 | } |
| 745 | |
| 746 | spin_unlock(&p->sighand->siglock); |
| 747 | read_unlock(&tasklist_lock); |
| 748 | |
| 749 | /* |
| 750 | * Install the new reload setting, and |
| 751 | * set up the signal and overrun bookkeeping. |
| 752 | */ |
| 753 | timer->it.cpu.incr = timespec_to_sample(timer->it_clock, |
| 754 | &new->it_interval); |
| 755 | |
| 756 | /* |
| 757 | * This acts as a modification timestamp for the timer, |
| 758 | * so any automatic reload attempt will punt on seeing |
| 759 | * that we have reset the timer manually. |
| 760 | */ |
| 761 | timer->it_requeue_pending = (timer->it_requeue_pending + 2) & |
| 762 | ~REQUEUE_PENDING; |
| 763 | timer->it_overrun_last = 0; |
| 764 | timer->it_overrun = -1; |
| 765 | |
| 766 | if (new_expires != 0 && !(val < new_expires)) { |
| 767 | /* |
| 768 | * The designated time already passed, so we notify |
| 769 | * immediately, even if the thread never runs to |
| 770 | * accumulate more time on this clock. |
| 771 | */ |
| 772 | cpu_timer_fire(timer); |
| 773 | } |
| 774 | |
| 775 | ret = 0; |
| 776 | out: |
| 777 | if (old) { |
| 778 | sample_to_timespec(timer->it_clock, |
| 779 | old_incr, &old->it_interval); |
| 780 | } |
| 781 | if (!ret) |
| 782 | posix_cpu_timer_kick_nohz(); |
| 783 | return ret; |
| 784 | } |
| 785 | |
| 786 | static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp) |
| 787 | { |
| 788 | unsigned long long now; |
| 789 | struct task_struct *p = timer->it.cpu.task; |
| 790 | int clear_dead; |
| 791 | |
| 792 | /* |
| 793 | * Easy part: convert the reload time. |
| 794 | */ |
| 795 | sample_to_timespec(timer->it_clock, |
| 796 | timer->it.cpu.incr, &itp->it_interval); |
| 797 | |
| 798 | if (timer->it.cpu.expires == 0) { /* Timer not armed at all. */ |
| 799 | itp->it_value.tv_sec = itp->it_value.tv_nsec = 0; |
| 800 | return; |
| 801 | } |
| 802 | |
| 803 | if (unlikely(p == NULL)) { |
| 804 | /* |
| 805 | * This task already died and the timer will never fire. |
| 806 | * In this case, expires is actually the dead value. |
| 807 | */ |
| 808 | dead: |
| 809 | sample_to_timespec(timer->it_clock, timer->it.cpu.expires, |
| 810 | &itp->it_value); |
| 811 | return; |
| 812 | } |
| 813 | |
| 814 | /* |
| 815 | * Sample the clock to take the difference with the expiry time. |
| 816 | */ |
| 817 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { |
| 818 | cpu_clock_sample(timer->it_clock, p, &now); |
| 819 | clear_dead = p->exit_state; |
| 820 | } else { |
| 821 | read_lock(&tasklist_lock); |
| 822 | if (unlikely(p->sighand == NULL)) { |
| 823 | /* |
| 824 | * The process has been reaped. |
| 825 | * We can't even collect a sample any more. |
| 826 | * Call the timer disarmed, nothing else to do. |
| 827 | */ |
| 828 | put_task_struct(p); |
| 829 | timer->it.cpu.task = NULL; |
| 830 | timer->it.cpu.expires = 0; |
| 831 | read_unlock(&tasklist_lock); |
| 832 | goto dead; |
| 833 | } else { |
| 834 | cpu_timer_sample_group(timer->it_clock, p, &now); |
| 835 | clear_dead = (unlikely(p->exit_state) && |
| 836 | thread_group_empty(p)); |
| 837 | } |
| 838 | read_unlock(&tasklist_lock); |
| 839 | } |
| 840 | |
| 841 | if (unlikely(clear_dead)) { |
| 842 | /* |
| 843 | * We've noticed that the thread is dead, but |
| 844 | * not yet reaped. Take this opportunity to |
| 845 | * drop our task ref. |
| 846 | */ |
| 847 | clear_dead_task(timer, now); |
| 848 | goto dead; |
| 849 | } |
| 850 | |
| 851 | if (now < timer->it.cpu.expires) { |
| 852 | sample_to_timespec(timer->it_clock, |
| 853 | timer->it.cpu.expires - now, |
| 854 | &itp->it_value); |
| 855 | } else { |
| 856 | /* |
| 857 | * The timer should have expired already, but the firing |
| 858 | * hasn't taken place yet. Say it's just about to expire. |
| 859 | */ |
| 860 | itp->it_value.tv_nsec = 1; |
| 861 | itp->it_value.tv_sec = 0; |
| 862 | } |
| 863 | } |
| 864 | |
| 865 | static unsigned long long |
| 866 | check_timers_list(struct list_head *timers, |
| 867 | struct list_head *firing, |
| 868 | unsigned long long curr) |
| 869 | { |
| 870 | int maxfire = 20; |
| 871 | |
| 872 | while (!list_empty(timers)) { |
| 873 | struct cpu_timer_list *t; |
| 874 | |
| 875 | t = list_first_entry(timers, struct cpu_timer_list, entry); |
| 876 | |
| 877 | if (!--maxfire || curr < t->expires) |
| 878 | return t->expires; |
| 879 | |
| 880 | t->firing = 1; |
| 881 | list_move_tail(&t->entry, firing); |
| 882 | } |
| 883 | |
| 884 | return 0; |
| 885 | } |
| 886 | |
| 887 | /* |
| 888 | * Check for any per-thread CPU timers that have fired and move them off |
| 889 | * the tsk->cpu_timers[N] list onto the firing list. Here we update the |
| 890 | * tsk->it_*_expires values to reflect the remaining thread CPU timers. |
| 891 | */ |
| 892 | static void check_thread_timers(struct task_struct *tsk, |
| 893 | struct list_head *firing) |
| 894 | { |
| 895 | struct list_head *timers = tsk->cpu_timers; |
| 896 | struct signal_struct *const sig = tsk->signal; |
| 897 | struct task_cputime *tsk_expires = &tsk->cputime_expires; |
| 898 | unsigned long long expires; |
| 899 | unsigned long soft; |
| 900 | |
| 901 | expires = check_timers_list(timers, firing, prof_ticks(tsk)); |
| 902 | tsk_expires->prof_exp = expires_to_cputime(expires); |
| 903 | |
| 904 | expires = check_timers_list(++timers, firing, virt_ticks(tsk)); |
| 905 | tsk_expires->virt_exp = expires_to_cputime(expires); |
| 906 | |
| 907 | tsk_expires->sched_exp = check_timers_list(++timers, firing, |
| 908 | tsk->se.sum_exec_runtime); |
| 909 | |
| 910 | /* |
| 911 | * Check for the special case thread timers. |
| 912 | */ |
| 913 | soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur); |
| 914 | if (soft != RLIM_INFINITY) { |
| 915 | unsigned long hard = |
| 916 | ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max); |
| 917 | |
| 918 | if (hard != RLIM_INFINITY && |
| 919 | tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) { |
| 920 | /* |
| 921 | * At the hard limit, we just die. |
| 922 | * No need to calculate anything else now. |
| 923 | */ |
| 924 | __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk); |
| 925 | return; |
| 926 | } |
| 927 | if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) { |
| 928 | /* |
| 929 | * At the soft limit, send a SIGXCPU every second. |
| 930 | */ |
| 931 | if (soft < hard) { |
| 932 | soft += USEC_PER_SEC; |
| 933 | sig->rlim[RLIMIT_RTTIME].rlim_cur = soft; |
| 934 | } |
| 935 | printk(KERN_INFO |
| 936 | "RT Watchdog Timeout: %s[%d]\n", |
| 937 | tsk->comm, task_pid_nr(tsk)); |
| 938 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); |
| 939 | } |
| 940 | } |
| 941 | } |
| 942 | |
| 943 | static void stop_process_timers(struct signal_struct *sig) |
| 944 | { |
| 945 | struct thread_group_cputimer *cputimer = &sig->cputimer; |
| 946 | unsigned long flags; |
| 947 | |
| 948 | raw_spin_lock_irqsave(&cputimer->lock, flags); |
| 949 | cputimer->running = 0; |
| 950 | raw_spin_unlock_irqrestore(&cputimer->lock, flags); |
| 951 | } |
| 952 | |
| 953 | static u32 onecputick; |
| 954 | |
| 955 | static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it, |
| 956 | unsigned long long *expires, |
| 957 | unsigned long long cur_time, int signo) |
| 958 | { |
| 959 | if (!it->expires) |
| 960 | return; |
| 961 | |
| 962 | if (cur_time >= it->expires) { |
| 963 | if (it->incr) { |
| 964 | it->expires += it->incr; |
| 965 | it->error += it->incr_error; |
| 966 | if (it->error >= onecputick) { |
| 967 | it->expires -= cputime_one_jiffy; |
| 968 | it->error -= onecputick; |
| 969 | } |
| 970 | } else { |
| 971 | it->expires = 0; |
| 972 | } |
| 973 | |
| 974 | trace_itimer_expire(signo == SIGPROF ? |
| 975 | ITIMER_PROF : ITIMER_VIRTUAL, |
| 976 | tsk->signal->leader_pid, cur_time); |
| 977 | __group_send_sig_info(signo, SEND_SIG_PRIV, tsk); |
| 978 | } |
| 979 | |
| 980 | if (it->expires && (!*expires || it->expires < *expires)) { |
| 981 | *expires = it->expires; |
| 982 | } |
| 983 | } |
| 984 | |
| 985 | /* |
| 986 | * Check for any per-thread CPU timers that have fired and move them |
| 987 | * off the tsk->*_timers list onto the firing list. Per-thread timers |
| 988 | * have already been taken off. |
| 989 | */ |
| 990 | static void check_process_timers(struct task_struct *tsk, |
| 991 | struct list_head *firing) |
| 992 | { |
| 993 | struct signal_struct *const sig = tsk->signal; |
| 994 | unsigned long long utime, ptime, virt_expires, prof_expires; |
| 995 | unsigned long long sum_sched_runtime, sched_expires; |
| 996 | struct list_head *timers = sig->cpu_timers; |
| 997 | struct task_cputime cputime; |
| 998 | unsigned long soft; |
| 999 | |
| 1000 | /* |
| 1001 | * Collect the current process totals. |
| 1002 | */ |
| 1003 | thread_group_cputimer(tsk, &cputime); |
| 1004 | utime = cputime_to_expires(cputime.utime); |
| 1005 | ptime = utime + cputime_to_expires(cputime.stime); |
| 1006 | sum_sched_runtime = cputime.sum_exec_runtime; |
| 1007 | |
| 1008 | prof_expires = check_timers_list(timers, firing, ptime); |
| 1009 | virt_expires = check_timers_list(++timers, firing, utime); |
| 1010 | sched_expires = check_timers_list(++timers, firing, sum_sched_runtime); |
| 1011 | |
| 1012 | /* |
| 1013 | * Check for the special case process timers. |
| 1014 | */ |
| 1015 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime, |
| 1016 | SIGPROF); |
| 1017 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime, |
| 1018 | SIGVTALRM); |
| 1019 | soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur); |
| 1020 | if (soft != RLIM_INFINITY) { |
| 1021 | unsigned long psecs = cputime_to_secs(ptime); |
| 1022 | unsigned long hard = |
| 1023 | ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max); |
| 1024 | cputime_t x; |
| 1025 | if (psecs >= hard) { |
| 1026 | /* |
| 1027 | * At the hard limit, we just die. |
| 1028 | * No need to calculate anything else now. |
| 1029 | */ |
| 1030 | __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk); |
| 1031 | return; |
| 1032 | } |
| 1033 | if (psecs >= soft) { |
| 1034 | /* |
| 1035 | * At the soft limit, send a SIGXCPU every second. |
| 1036 | */ |
| 1037 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); |
| 1038 | if (soft < hard) { |
| 1039 | soft++; |
| 1040 | sig->rlim[RLIMIT_CPU].rlim_cur = soft; |
| 1041 | } |
| 1042 | } |
| 1043 | x = secs_to_cputime(soft); |
| 1044 | if (!prof_expires || x < prof_expires) { |
| 1045 | prof_expires = x; |
| 1046 | } |
| 1047 | } |
| 1048 | |
| 1049 | sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires); |
| 1050 | sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires); |
| 1051 | sig->cputime_expires.sched_exp = sched_expires; |
| 1052 | if (task_cputime_zero(&sig->cputime_expires)) |
| 1053 | stop_process_timers(sig); |
| 1054 | } |
| 1055 | |
| 1056 | /* |
| 1057 | * This is called from the signal code (via do_schedule_next_timer) |
| 1058 | * when the last timer signal was delivered and we have to reload the timer. |
| 1059 | */ |
| 1060 | void posix_cpu_timer_schedule(struct k_itimer *timer) |
| 1061 | { |
| 1062 | struct task_struct *p = timer->it.cpu.task; |
| 1063 | unsigned long long now; |
| 1064 | |
| 1065 | if (unlikely(p == NULL)) |
| 1066 | /* |
| 1067 | * The task was cleaned up already, no future firings. |
| 1068 | */ |
| 1069 | goto out; |
| 1070 | |
| 1071 | /* |
| 1072 | * Fetch the current sample and update the timer's expiry time. |
| 1073 | */ |
| 1074 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) { |
| 1075 | cpu_clock_sample(timer->it_clock, p, &now); |
| 1076 | bump_cpu_timer(timer, now); |
| 1077 | if (unlikely(p->exit_state)) { |
| 1078 | clear_dead_task(timer, now); |
| 1079 | goto out; |
| 1080 | } |
| 1081 | read_lock(&tasklist_lock); /* arm_timer needs it. */ |
| 1082 | spin_lock(&p->sighand->siglock); |
| 1083 | } else { |
| 1084 | read_lock(&tasklist_lock); |
| 1085 | if (unlikely(p->sighand == NULL)) { |
| 1086 | /* |
| 1087 | * The process has been reaped. |
| 1088 | * We can't even collect a sample any more. |
| 1089 | */ |
| 1090 | put_task_struct(p); |
| 1091 | timer->it.cpu.task = p = NULL; |
| 1092 | timer->it.cpu.expires = 0; |
| 1093 | goto out_unlock; |
| 1094 | } else if (unlikely(p->exit_state) && thread_group_empty(p)) { |
| 1095 | /* |
| 1096 | * We've noticed that the thread is dead, but |
| 1097 | * not yet reaped. Take this opportunity to |
| 1098 | * drop our task ref. |
| 1099 | */ |
| 1100 | cpu_timer_sample_group(timer->it_clock, p, &now); |
| 1101 | clear_dead_task(timer, now); |
| 1102 | goto out_unlock; |
| 1103 | } |
| 1104 | spin_lock(&p->sighand->siglock); |
| 1105 | cpu_timer_sample_group(timer->it_clock, p, &now); |
| 1106 | bump_cpu_timer(timer, now); |
| 1107 | /* Leave the tasklist_lock locked for the call below. */ |
| 1108 | } |
| 1109 | |
| 1110 | /* |
| 1111 | * Now re-arm for the new expiry time. |
| 1112 | */ |
| 1113 | BUG_ON(!irqs_disabled()); |
| 1114 | arm_timer(timer); |
| 1115 | spin_unlock(&p->sighand->siglock); |
| 1116 | |
| 1117 | out_unlock: |
| 1118 | read_unlock(&tasklist_lock); |
| 1119 | |
| 1120 | out: |
| 1121 | timer->it_overrun_last = timer->it_overrun; |
| 1122 | timer->it_overrun = -1; |
| 1123 | ++timer->it_requeue_pending; |
| 1124 | } |
| 1125 | |
| 1126 | /** |
| 1127 | * task_cputime_expired - Compare two task_cputime entities. |
| 1128 | * |
| 1129 | * @sample: The task_cputime structure to be checked for expiration. |
| 1130 | * @expires: Expiration times, against which @sample will be checked. |
| 1131 | * |
| 1132 | * Checks @sample against @expires to see if any field of @sample has expired. |
| 1133 | * Returns true if any field of the former is greater than the corresponding |
| 1134 | * field of the latter if the latter field is set. Otherwise returns false. |
| 1135 | */ |
| 1136 | static inline int task_cputime_expired(const struct task_cputime *sample, |
| 1137 | const struct task_cputime *expires) |
| 1138 | { |
| 1139 | if (expires->utime && sample->utime >= expires->utime) |
| 1140 | return 1; |
| 1141 | if (expires->stime && sample->utime + sample->stime >= expires->stime) |
| 1142 | return 1; |
| 1143 | if (expires->sum_exec_runtime != 0 && |
| 1144 | sample->sum_exec_runtime >= expires->sum_exec_runtime) |
| 1145 | return 1; |
| 1146 | return 0; |
| 1147 | } |
| 1148 | |
| 1149 | /** |
| 1150 | * fastpath_timer_check - POSIX CPU timers fast path. |
| 1151 | * |
| 1152 | * @tsk: The task (thread) being checked. |
| 1153 | * |
| 1154 | * Check the task and thread group timers. If both are zero (there are no |
| 1155 | * timers set) return false. Otherwise snapshot the task and thread group |
| 1156 | * timers and compare them with the corresponding expiration times. Return |
| 1157 | * true if a timer has expired, else return false. |
| 1158 | */ |
| 1159 | static inline int fastpath_timer_check(struct task_struct *tsk) |
| 1160 | { |
| 1161 | struct signal_struct *sig; |
| 1162 | cputime_t utime, stime; |
| 1163 | |
| 1164 | task_cputime(tsk, &utime, &stime); |
| 1165 | |
| 1166 | if (!task_cputime_zero(&tsk->cputime_expires)) { |
| 1167 | struct task_cputime task_sample = { |
| 1168 | .utime = utime, |
| 1169 | .stime = stime, |
| 1170 | .sum_exec_runtime = tsk->se.sum_exec_runtime |
| 1171 | }; |
| 1172 | |
| 1173 | if (task_cputime_expired(&task_sample, &tsk->cputime_expires)) |
| 1174 | return 1; |
| 1175 | } |
| 1176 | |
| 1177 | sig = tsk->signal; |
| 1178 | if (sig->cputimer.running) { |
| 1179 | struct task_cputime group_sample; |
| 1180 | |
| 1181 | raw_spin_lock(&sig->cputimer.lock); |
| 1182 | group_sample = sig->cputimer.cputime; |
| 1183 | raw_spin_unlock(&sig->cputimer.lock); |
| 1184 | |
| 1185 | if (task_cputime_expired(&group_sample, &sig->cputime_expires)) |
| 1186 | return 1; |
| 1187 | } |
| 1188 | |
| 1189 | return 0; |
| 1190 | } |
| 1191 | |
| 1192 | /* |
| 1193 | * This is called from the timer interrupt handler. The irq handler has |
| 1194 | * already updated our counts. We need to check if any timers fire now. |
| 1195 | * Interrupts are disabled. |
| 1196 | */ |
| 1197 | void run_posix_cpu_timers(struct task_struct *tsk) |
| 1198 | { |
| 1199 | LIST_HEAD(firing); |
| 1200 | struct k_itimer *timer, *next; |
| 1201 | unsigned long flags; |
| 1202 | |
| 1203 | BUG_ON(!irqs_disabled()); |
| 1204 | |
| 1205 | /* |
| 1206 | * The fast path checks that there are no expired thread or thread |
| 1207 | * group timers. If that's so, just return. |
| 1208 | */ |
| 1209 | if (!fastpath_timer_check(tsk)) |
| 1210 | return; |
| 1211 | |
| 1212 | if (!lock_task_sighand(tsk, &flags)) |
| 1213 | return; |
| 1214 | /* |
| 1215 | * Here we take off tsk->signal->cpu_timers[N] and |
| 1216 | * tsk->cpu_timers[N] all the timers that are firing, and |
| 1217 | * put them on the firing list. |
| 1218 | */ |
| 1219 | check_thread_timers(tsk, &firing); |
| 1220 | /* |
| 1221 | * If there are any active process wide timers (POSIX 1.b, itimers, |
| 1222 | * RLIMIT_CPU) cputimer must be running. |
| 1223 | */ |
| 1224 | if (tsk->signal->cputimer.running) |
| 1225 | check_process_timers(tsk, &firing); |
| 1226 | |
| 1227 | /* |
| 1228 | * We must release these locks before taking any timer's lock. |
| 1229 | * There is a potential race with timer deletion here, as the |
| 1230 | * siglock now protects our private firing list. We have set |
| 1231 | * the firing flag in each timer, so that a deletion attempt |
| 1232 | * that gets the timer lock before we do will give it up and |
| 1233 | * spin until we've taken care of that timer below. |
| 1234 | */ |
| 1235 | unlock_task_sighand(tsk, &flags); |
| 1236 | |
| 1237 | /* |
| 1238 | * Now that all the timers on our list have the firing flag, |
| 1239 | * no one will touch their list entries but us. We'll take |
| 1240 | * each timer's lock before clearing its firing flag, so no |
| 1241 | * timer call will interfere. |
| 1242 | */ |
| 1243 | list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) { |
| 1244 | int cpu_firing; |
| 1245 | |
| 1246 | spin_lock(&timer->it_lock); |
| 1247 | list_del_init(&timer->it.cpu.entry); |
| 1248 | cpu_firing = timer->it.cpu.firing; |
| 1249 | timer->it.cpu.firing = 0; |
| 1250 | /* |
| 1251 | * The firing flag is -1 if we collided with a reset |
| 1252 | * of the timer, which already reported this |
| 1253 | * almost-firing as an overrun. So don't generate an event. |
| 1254 | */ |
| 1255 | if (likely(cpu_firing >= 0)) |
| 1256 | cpu_timer_fire(timer); |
| 1257 | spin_unlock(&timer->it_lock); |
| 1258 | } |
| 1259 | |
| 1260 | /* |
| 1261 | * In case some timers were rescheduled after the queue got emptied, |
| 1262 | * wake up full dynticks CPUs. |
| 1263 | */ |
| 1264 | if (tsk->signal->cputimer.running) |
| 1265 | posix_cpu_timer_kick_nohz(); |
| 1266 | } |
| 1267 | |
| 1268 | /* |
| 1269 | * Set one of the process-wide special case CPU timers or RLIMIT_CPU. |
| 1270 | * The tsk->sighand->siglock must be held by the caller. |
| 1271 | */ |
| 1272 | void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx, |
| 1273 | cputime_t *newval, cputime_t *oldval) |
| 1274 | { |
| 1275 | unsigned long long now; |
| 1276 | |
| 1277 | BUG_ON(clock_idx == CPUCLOCK_SCHED); |
| 1278 | cpu_timer_sample_group(clock_idx, tsk, &now); |
| 1279 | |
| 1280 | if (oldval) { |
| 1281 | /* |
| 1282 | * We are setting itimer. The *oldval is absolute and we update |
| 1283 | * it to be relative, *newval argument is relative and we update |
| 1284 | * it to be absolute. |
| 1285 | */ |
| 1286 | if (*oldval) { |
| 1287 | if (*oldval <= now) { |
| 1288 | /* Just about to fire. */ |
| 1289 | *oldval = cputime_one_jiffy; |
| 1290 | } else { |
| 1291 | *oldval -= now; |
| 1292 | } |
| 1293 | } |
| 1294 | |
| 1295 | if (!*newval) |
| 1296 | goto out; |
| 1297 | *newval += now; |
| 1298 | } |
| 1299 | |
| 1300 | /* |
| 1301 | * Update expiration cache if we are the earliest timer, or eventually |
| 1302 | * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire. |
| 1303 | */ |
| 1304 | switch (clock_idx) { |
| 1305 | case CPUCLOCK_PROF: |
| 1306 | if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval)) |
| 1307 | tsk->signal->cputime_expires.prof_exp = *newval; |
| 1308 | break; |
| 1309 | case CPUCLOCK_VIRT: |
| 1310 | if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval)) |
| 1311 | tsk->signal->cputime_expires.virt_exp = *newval; |
| 1312 | break; |
| 1313 | } |
| 1314 | out: |
| 1315 | posix_cpu_timer_kick_nohz(); |
| 1316 | } |
| 1317 | |
| 1318 | static int do_cpu_nanosleep(const clockid_t which_clock, int flags, |
| 1319 | struct timespec *rqtp, struct itimerspec *it) |
| 1320 | { |
| 1321 | struct k_itimer timer; |
| 1322 | int error; |
| 1323 | |
| 1324 | /* |
| 1325 | * Set up a temporary timer and then wait for it to go off. |
| 1326 | */ |
| 1327 | memset(&timer, 0, sizeof timer); |
| 1328 | spin_lock_init(&timer.it_lock); |
| 1329 | timer.it_clock = which_clock; |
| 1330 | timer.it_overrun = -1; |
| 1331 | error = posix_cpu_timer_create(&timer); |
| 1332 | timer.it_process = current; |
| 1333 | if (!error) { |
| 1334 | static struct itimerspec zero_it; |
| 1335 | |
| 1336 | memset(it, 0, sizeof *it); |
| 1337 | it->it_value = *rqtp; |
| 1338 | |
| 1339 | spin_lock_irq(&timer.it_lock); |
| 1340 | error = posix_cpu_timer_set(&timer, flags, it, NULL); |
| 1341 | if (error) { |
| 1342 | spin_unlock_irq(&timer.it_lock); |
| 1343 | return error; |
| 1344 | } |
| 1345 | |
| 1346 | while (!signal_pending(current)) { |
| 1347 | if (timer.it.cpu.expires == 0) { |
| 1348 | /* |
| 1349 | * Our timer fired and was reset, below |
| 1350 | * deletion can not fail. |
| 1351 | */ |
| 1352 | posix_cpu_timer_del(&timer); |
| 1353 | spin_unlock_irq(&timer.it_lock); |
| 1354 | return 0; |
| 1355 | } |
| 1356 | |
| 1357 | /* |
| 1358 | * Block until cpu_timer_fire (or a signal) wakes us. |
| 1359 | */ |
| 1360 | __set_current_state(TASK_INTERRUPTIBLE); |
| 1361 | spin_unlock_irq(&timer.it_lock); |
| 1362 | schedule(); |
| 1363 | spin_lock_irq(&timer.it_lock); |
| 1364 | } |
| 1365 | |
| 1366 | /* |
| 1367 | * We were interrupted by a signal. |
| 1368 | */ |
| 1369 | sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp); |
| 1370 | error = posix_cpu_timer_set(&timer, 0, &zero_it, it); |
| 1371 | if (!error) { |
| 1372 | /* |
| 1373 | * Timer is now unarmed, deletion can not fail. |
| 1374 | */ |
| 1375 | posix_cpu_timer_del(&timer); |
| 1376 | } |
| 1377 | spin_unlock_irq(&timer.it_lock); |
| 1378 | |
| 1379 | while (error == TIMER_RETRY) { |
| 1380 | /* |
| 1381 | * We need to handle case when timer was or is in the |
| 1382 | * middle of firing. In other cases we already freed |
| 1383 | * resources. |
| 1384 | */ |
| 1385 | spin_lock_irq(&timer.it_lock); |
| 1386 | error = posix_cpu_timer_del(&timer); |
| 1387 | spin_unlock_irq(&timer.it_lock); |
| 1388 | } |
| 1389 | |
| 1390 | if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) { |
| 1391 | /* |
| 1392 | * It actually did fire already. |
| 1393 | */ |
| 1394 | return 0; |
| 1395 | } |
| 1396 | |
| 1397 | error = -ERESTART_RESTARTBLOCK; |
| 1398 | } |
| 1399 | |
| 1400 | return error; |
| 1401 | } |
| 1402 | |
| 1403 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block); |
| 1404 | |
| 1405 | static int posix_cpu_nsleep(const clockid_t which_clock, int flags, |
| 1406 | struct timespec *rqtp, struct timespec __user *rmtp) |
| 1407 | { |
| 1408 | struct restart_block *restart_block = |
| 1409 | ¤t_thread_info()->restart_block; |
| 1410 | struct itimerspec it; |
| 1411 | int error; |
| 1412 | |
| 1413 | /* |
| 1414 | * Diagnose required errors first. |
| 1415 | */ |
| 1416 | if (CPUCLOCK_PERTHREAD(which_clock) && |
| 1417 | (CPUCLOCK_PID(which_clock) == 0 || |
| 1418 | CPUCLOCK_PID(which_clock) == current->pid)) |
| 1419 | return -EINVAL; |
| 1420 | |
| 1421 | error = do_cpu_nanosleep(which_clock, flags, rqtp, &it); |
| 1422 | |
| 1423 | if (error == -ERESTART_RESTARTBLOCK) { |
| 1424 | |
| 1425 | if (flags & TIMER_ABSTIME) |
| 1426 | return -ERESTARTNOHAND; |
| 1427 | /* |
| 1428 | * Report back to the user the time still remaining. |
| 1429 | */ |
| 1430 | if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp)) |
| 1431 | return -EFAULT; |
| 1432 | |
| 1433 | restart_block->fn = posix_cpu_nsleep_restart; |
| 1434 | restart_block->nanosleep.clockid = which_clock; |
| 1435 | restart_block->nanosleep.rmtp = rmtp; |
| 1436 | restart_block->nanosleep.expires = timespec_to_ns(rqtp); |
| 1437 | } |
| 1438 | return error; |
| 1439 | } |
| 1440 | |
| 1441 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block) |
| 1442 | { |
| 1443 | clockid_t which_clock = restart_block->nanosleep.clockid; |
| 1444 | struct timespec t; |
| 1445 | struct itimerspec it; |
| 1446 | int error; |
| 1447 | |
| 1448 | t = ns_to_timespec(restart_block->nanosleep.expires); |
| 1449 | |
| 1450 | error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it); |
| 1451 | |
| 1452 | if (error == -ERESTART_RESTARTBLOCK) { |
| 1453 | struct timespec __user *rmtp = restart_block->nanosleep.rmtp; |
| 1454 | /* |
| 1455 | * Report back to the user the time still remaining. |
| 1456 | */ |
| 1457 | if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp)) |
| 1458 | return -EFAULT; |
| 1459 | |
| 1460 | restart_block->nanosleep.expires = timespec_to_ns(&t); |
| 1461 | } |
| 1462 | return error; |
| 1463 | |
| 1464 | } |
| 1465 | |
| 1466 | #define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED) |
| 1467 | #define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED) |
| 1468 | |
| 1469 | static int process_cpu_clock_getres(const clockid_t which_clock, |
| 1470 | struct timespec *tp) |
| 1471 | { |
| 1472 | return posix_cpu_clock_getres(PROCESS_CLOCK, tp); |
| 1473 | } |
| 1474 | static int process_cpu_clock_get(const clockid_t which_clock, |
| 1475 | struct timespec *tp) |
| 1476 | { |
| 1477 | return posix_cpu_clock_get(PROCESS_CLOCK, tp); |
| 1478 | } |
| 1479 | static int process_cpu_timer_create(struct k_itimer *timer) |
| 1480 | { |
| 1481 | timer->it_clock = PROCESS_CLOCK; |
| 1482 | return posix_cpu_timer_create(timer); |
| 1483 | } |
| 1484 | static int process_cpu_nsleep(const clockid_t which_clock, int flags, |
| 1485 | struct timespec *rqtp, |
| 1486 | struct timespec __user *rmtp) |
| 1487 | { |
| 1488 | return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp); |
| 1489 | } |
| 1490 | static long process_cpu_nsleep_restart(struct restart_block *restart_block) |
| 1491 | { |
| 1492 | return -EINVAL; |
| 1493 | } |
| 1494 | static int thread_cpu_clock_getres(const clockid_t which_clock, |
| 1495 | struct timespec *tp) |
| 1496 | { |
| 1497 | return posix_cpu_clock_getres(THREAD_CLOCK, tp); |
| 1498 | } |
| 1499 | static int thread_cpu_clock_get(const clockid_t which_clock, |
| 1500 | struct timespec *tp) |
| 1501 | { |
| 1502 | return posix_cpu_clock_get(THREAD_CLOCK, tp); |
| 1503 | } |
| 1504 | static int thread_cpu_timer_create(struct k_itimer *timer) |
| 1505 | { |
| 1506 | timer->it_clock = THREAD_CLOCK; |
| 1507 | return posix_cpu_timer_create(timer); |
| 1508 | } |
| 1509 | |
| 1510 | struct k_clock clock_posix_cpu = { |
| 1511 | .clock_getres = posix_cpu_clock_getres, |
| 1512 | .clock_set = posix_cpu_clock_set, |
| 1513 | .clock_get = posix_cpu_clock_get, |
| 1514 | .timer_create = posix_cpu_timer_create, |
| 1515 | .nsleep = posix_cpu_nsleep, |
| 1516 | .nsleep_restart = posix_cpu_nsleep_restart, |
| 1517 | .timer_set = posix_cpu_timer_set, |
| 1518 | .timer_del = posix_cpu_timer_del, |
| 1519 | .timer_get = posix_cpu_timer_get, |
| 1520 | }; |
| 1521 | |
| 1522 | static __init int init_posix_cpu_timers(void) |
| 1523 | { |
| 1524 | struct k_clock process = { |
| 1525 | .clock_getres = process_cpu_clock_getres, |
| 1526 | .clock_get = process_cpu_clock_get, |
| 1527 | .timer_create = process_cpu_timer_create, |
| 1528 | .nsleep = process_cpu_nsleep, |
| 1529 | .nsleep_restart = process_cpu_nsleep_restart, |
| 1530 | }; |
| 1531 | struct k_clock thread = { |
| 1532 | .clock_getres = thread_cpu_clock_getres, |
| 1533 | .clock_get = thread_cpu_clock_get, |
| 1534 | .timer_create = thread_cpu_timer_create, |
| 1535 | }; |
| 1536 | struct timespec ts; |
| 1537 | |
| 1538 | posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process); |
| 1539 | posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread); |
| 1540 | |
| 1541 | cputime_to_timespec(cputime_one_jiffy, &ts); |
| 1542 | onecputick = ts.tv_nsec; |
| 1543 | WARN_ON(ts.tv_sec != 0); |
| 1544 | |
| 1545 | return 0; |
| 1546 | } |
| 1547 | __initcall(init_posix_cpu_timers); |