Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | /* |
| 3 | * Implement CPU time clocks for the POSIX clock interface. |
| 4 | */ |
| 5 | |
Ingo Molnar | 3f07c01 | 2017-02-08 18:51:30 +0100 | [diff] [blame] | 6 | #include <linux/sched/signal.h> |
Ingo Molnar | 32ef551 | 2017-02-05 11:48:36 +0100 | [diff] [blame] | 7 | #include <linux/sched/cputime.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 8 | #include <linux/posix-timers.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | #include <linux/errno.h> |
Roman Zippel | f8bd225 | 2008-05-01 04:34:31 -0700 | [diff] [blame] | 10 | #include <linux/math64.h> |
Linus Torvalds | 7c0f6ba | 2016-12-24 11:46:01 -0800 | [diff] [blame] | 11 | #include <linux/uaccess.h> |
Frank Mayhar | bb34d92 | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 12 | #include <linux/kernel_stat.h> |
Xiao Guangrong | 3f0a525 | 2009-08-10 10:52:30 +0800 | [diff] [blame] | 13 | #include <trace/events/timer.h> |
Frederic Weisbecker | a857216 | 2013-04-18 01:31:13 +0200 | [diff] [blame] | 14 | #include <linux/tick.h> |
| 15 | #include <linux/workqueue.h> |
Al Viro | edbeda4 | 2017-06-07 09:42:31 +0100 | [diff] [blame] | 16 | #include <linux/compat.h> |
Juri Lelli | 34be393 | 2017-12-12 12:10:24 +0100 | [diff] [blame] | 17 | #include <linux/sched/deadline.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | |
Thomas Gleixner | bab0aae9 | 2017-05-30 23:15:41 +0200 | [diff] [blame] | 19 | #include "posix-timers.h" |
| 20 | |
Thomas Gleixner | f37fb0a | 2017-05-30 23:15:47 +0200 | [diff] [blame] | 21 | static void posix_cpu_timer_rearm(struct k_itimer *timer); |
| 22 | |
Thomas Gleixner | 3a245c0 | 2019-08-21 21:09:06 +0200 | [diff] [blame] | 23 | void posix_cputimers_group_init(struct posix_cputimers *pct, u64 cpu_limit) |
| 24 | { |
| 25 | posix_cputimers_init(pct); |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 26 | if (cpu_limit != RLIM_INFINITY) { |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 27 | pct->bases[CPUCLOCK_PROF].nextevt = cpu_limit * NSEC_PER_SEC; |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 28 | pct->timers_active = true; |
| 29 | } |
Thomas Gleixner | 3a245c0 | 2019-08-21 21:09:06 +0200 | [diff] [blame] | 30 | } |
| 31 | |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 32 | /* |
Stanislaw Gruszka | f55db60 | 2010-03-11 14:04:37 -0800 | [diff] [blame] | 33 | * Called after updating RLIMIT_CPU to run cpu timer and update |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 34 | * tsk->signal->posix_cputimers.bases[clock].nextevt expiration cache if |
| 35 | * necessary. Needs siglock protection since other code may update the |
Thomas Gleixner | 3a245c0 | 2019-08-21 21:09:06 +0200 | [diff] [blame] | 36 | * expiration cache as well. |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 37 | */ |
Jiri Slaby | 5ab46b3 | 2009-08-28 14:05:12 +0200 | [diff] [blame] | 38 | void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new) |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 39 | { |
Frederic Weisbecker | 858cf3a | 2017-01-31 04:09:35 +0100 | [diff] [blame] | 40 | u64 nsecs = rlim_new * NSEC_PER_SEC; |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 41 | |
Jiri Slaby | 5ab46b3 | 2009-08-28 14:05:12 +0200 | [diff] [blame] | 42 | spin_lock_irq(&task->sighand->siglock); |
Frederic Weisbecker | 858cf3a | 2017-01-31 04:09:35 +0100 | [diff] [blame] | 43 | set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL); |
Jiri Slaby | 5ab46b3 | 2009-08-28 14:05:12 +0200 | [diff] [blame] | 44 | spin_unlock_irq(&task->sighand->siglock); |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 45 | } |
| 46 | |
Thomas Gleixner | 6ae40e3 | 2019-08-21 21:08:48 +0200 | [diff] [blame] | 47 | /* |
| 48 | * Functions for validating access to tasks. |
| 49 | */ |
Eric W. Biederman | 9649877 | 2020-04-27 07:55:07 -0500 | [diff] [blame] | 50 | static struct pid *pid_for_clock(const clockid_t clock, bool gettime) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 51 | { |
Eric W. Biederman | 9649877 | 2020-04-27 07:55:07 -0500 | [diff] [blame] | 52 | const bool thread = !!CPUCLOCK_PERTHREAD(clock); |
| 53 | const pid_t upid = CPUCLOCK_PID(clock); |
| 54 | struct pid *pid; |
| 55 | |
| 56 | if (CPUCLOCK_WHICH(clock) >= CPUCLOCK_MAX) |
| 57 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | |
Thomas Gleixner | 77b4b54 | 2019-09-05 23:15:08 +0200 | [diff] [blame] | 59 | /* |
| 60 | * If the encoded PID is 0, then the timer is targeted at current |
| 61 | * or the process to which current belongs. |
| 62 | */ |
Eric W. Biederman | 9649877 | 2020-04-27 07:55:07 -0500 | [diff] [blame] | 63 | if (upid == 0) |
| 64 | return thread ? task_pid(current) : task_tgid(current); |
| 65 | |
| 66 | pid = find_vpid(upid); |
Thomas Gleixner | 6ae40e3 | 2019-08-21 21:08:48 +0200 | [diff] [blame] | 67 | if (!pid) |
Thomas Gleixner | 6ae40e3 | 2019-08-21 21:08:48 +0200 | [diff] [blame] | 68 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | |
Eric W. Biederman | 9649877 | 2020-04-27 07:55:07 -0500 | [diff] [blame] | 70 | if (thread) { |
| 71 | struct task_struct *tsk = pid_task(pid, PIDTYPE_PID); |
| 72 | return (tsk && same_thread_group(tsk, current)) ? pid : NULL; |
| 73 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | |
Eric W. Biederman | 9649877 | 2020-04-27 07:55:07 -0500 | [diff] [blame] | 75 | /* |
| 76 | * For clock_gettime(PROCESS) allow finding the process by |
| 77 | * with the pid of the current task. The code needs the tgid |
| 78 | * of the process so that pid_task(pid, PIDTYPE_TGID) can be |
| 79 | * used to find the process. |
| 80 | */ |
| 81 | if (gettime && (pid == task_pid(current))) |
| 82 | return task_tgid(current); |
Thomas Gleixner | 77b4b54 | 2019-09-05 23:15:08 +0200 | [diff] [blame] | 83 | |
Eric W. Biederman | 9649877 | 2020-04-27 07:55:07 -0500 | [diff] [blame] | 84 | /* |
| 85 | * For processes require that pid identifies a process. |
| 86 | */ |
| 87 | return pid_has_task(pid, PIDTYPE_TGID) ? pid : NULL; |
Thomas Gleixner | 6ae40e3 | 2019-08-21 21:08:48 +0200 | [diff] [blame] | 88 | } |
| 89 | |
| 90 | static inline int validate_clock_permissions(const clockid_t clock) |
| 91 | { |
Eric W. Biederman | 9bf7c32 | 2020-04-25 18:38:54 -0500 | [diff] [blame] | 92 | int ret; |
| 93 | |
| 94 | rcu_read_lock(); |
Eric W. Biederman | 9649877 | 2020-04-27 07:55:07 -0500 | [diff] [blame] | 95 | ret = pid_for_clock(clock, false) ? 0 : -EINVAL; |
Eric W. Biederman | 9bf7c32 | 2020-04-25 18:38:54 -0500 | [diff] [blame] | 96 | rcu_read_unlock(); |
| 97 | |
| 98 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | } |
| 100 | |
Eric W. Biederman | fece9826 | 2020-04-27 09:38:29 -0500 | [diff] [blame] | 101 | static inline enum pid_type clock_pid_type(const clockid_t clock) |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 102 | { |
Eric W. Biederman | fece9826 | 2020-04-27 09:38:29 -0500 | [diff] [blame] | 103 | return CPUCLOCK_PERTHREAD(clock) ? PIDTYPE_PID : PIDTYPE_TGID; |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 104 | } |
| 105 | |
| 106 | static inline struct task_struct *cpu_timer_task_rcu(struct k_itimer *timer) |
| 107 | { |
Eric W. Biederman | fece9826 | 2020-04-27 09:38:29 -0500 | [diff] [blame] | 108 | return pid_task(timer->it.cpu.pid, clock_pid_type(timer->it_clock)); |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 109 | } |
| 110 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 111 | /* |
| 112 | * Update expiry time from increment, and increase overrun count, |
| 113 | * given the current clock sample. |
| 114 | */ |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 115 | static u64 bump_cpu_timer(struct k_itimer *timer, u64 now) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | { |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 117 | u64 delta, incr, expires = timer->it.cpu.node.expires; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 118 | int i; |
| 119 | |
Thomas Gleixner | 1611879 | 2019-01-11 14:33:17 +0100 | [diff] [blame] | 120 | if (!timer->it_interval) |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 121 | return expires; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 123 | if (now < expires) |
| 124 | return expires; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 125 | |
Thomas Gleixner | 1611879 | 2019-01-11 14:33:17 +0100 | [diff] [blame] | 126 | incr = timer->it_interval; |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 127 | delta = now + incr - expires; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | |
Frederic Weisbecker | 55ccb61 | 2013-06-28 00:06:42 +0000 | [diff] [blame] | 129 | /* Don't use (incr*2 < delta), incr*2 might overflow. */ |
| 130 | for (i = 0; incr < delta - incr; i++) |
| 131 | incr = incr << 1; |
| 132 | |
| 133 | for (; i >= 0; incr >>= 1, i--) { |
| 134 | if (delta < incr) |
| 135 | continue; |
| 136 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 137 | timer->it.cpu.node.expires += incr; |
Thomas Gleixner | 78c9c4d | 2018-06-26 15:21:32 +0200 | [diff] [blame] | 138 | timer->it_overrun += 1LL << i; |
Frederic Weisbecker | 55ccb61 | 2013-06-28 00:06:42 +0000 | [diff] [blame] | 139 | delta -= incr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | } |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 141 | return timer->it.cpu.node.expires; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | } |
| 143 | |
Thomas Gleixner | 2bbdbda | 2019-08-21 21:09:19 +0200 | [diff] [blame] | 144 | /* Check whether all cache entries contain U64_MAX, i.e. eternal expiry time */ |
| 145 | static inline bool expiry_cache_is_inactive(const struct posix_cputimers *pct) |
Frederic Weisbecker | 555347f | 2013-04-19 16:17:38 +0200 | [diff] [blame] | 146 | { |
Thomas Gleixner | 2bbdbda | 2019-08-21 21:09:19 +0200 | [diff] [blame] | 147 | return !(~pct->bases[CPUCLOCK_PROF].nextevt | |
| 148 | ~pct->bases[CPUCLOCK_VIRT].nextevt | |
| 149 | ~pct->bases[CPUCLOCK_SCHED].nextevt); |
Frederic Weisbecker | 555347f | 2013-04-19 16:17:38 +0200 | [diff] [blame] | 150 | } |
| 151 | |
Thomas Gleixner | bc2c8ea | 2011-02-01 13:52:12 +0000 | [diff] [blame] | 152 | static int |
Deepa Dinamani | d2e3e0c | 2017-03-26 12:04:15 -0700 | [diff] [blame] | 153 | posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | { |
Thomas Gleixner | 6ae40e3 | 2019-08-21 21:08:48 +0200 | [diff] [blame] | 155 | int error = validate_clock_permissions(which_clock); |
| 156 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | if (!error) { |
| 158 | tp->tv_sec = 0; |
| 159 | tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ); |
| 160 | if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) { |
| 161 | /* |
| 162 | * If sched_clock is using a cycle counter, we |
| 163 | * don't have any idea of its true resolution |
| 164 | * exported, but it is much more than 1s/HZ. |
| 165 | */ |
| 166 | tp->tv_nsec = 1; |
| 167 | } |
| 168 | } |
| 169 | return error; |
| 170 | } |
| 171 | |
Thomas Gleixner | bc2c8ea | 2011-02-01 13:52:12 +0000 | [diff] [blame] | 172 | static int |
Thomas Gleixner | 6ae40e3 | 2019-08-21 21:08:48 +0200 | [diff] [blame] | 173 | posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | { |
Thomas Gleixner | 6ae40e3 | 2019-08-21 21:08:48 +0200 | [diff] [blame] | 175 | int error = validate_clock_permissions(clock); |
| 176 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | /* |
| 178 | * You can never reset a CPU clock, but we check for other errors |
| 179 | * in the call before failing with EPERM. |
| 180 | */ |
Thomas Gleixner | 6ae40e3 | 2019-08-21 21:08:48 +0200 | [diff] [blame] | 181 | return error ? : -EPERM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | } |
| 183 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | /* |
Thomas Gleixner | 2092c1d4 | 2019-08-21 21:09:00 +0200 | [diff] [blame] | 185 | * Sample a per-thread clock for the given task. clkid is validated. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | */ |
Thomas Gleixner | 8c2d74f | 2019-08-21 21:09:01 +0200 | [diff] [blame] | 187 | static u64 cpu_clock_sample(const clockid_t clkid, struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | { |
Thomas Gleixner | ab693c5 | 2019-08-21 21:09:03 +0200 | [diff] [blame] | 189 | u64 utime, stime; |
| 190 | |
| 191 | if (clkid == CPUCLOCK_SCHED) |
| 192 | return task_sched_runtime(p); |
| 193 | |
| 194 | task_cputime(p, &utime, &stime); |
| 195 | |
Thomas Gleixner | 2092c1d4 | 2019-08-21 21:09:00 +0200 | [diff] [blame] | 196 | switch (clkid) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | case CPUCLOCK_PROF: |
Thomas Gleixner | ab693c5 | 2019-08-21 21:09:03 +0200 | [diff] [blame] | 198 | return utime + stime; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | case CPUCLOCK_VIRT: |
Thomas Gleixner | ab693c5 | 2019-08-21 21:09:03 +0200 | [diff] [blame] | 200 | return utime; |
Thomas Gleixner | 2092c1d4 | 2019-08-21 21:09:00 +0200 | [diff] [blame] | 201 | default: |
| 202 | WARN_ON_ONCE(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 203 | } |
Thomas Gleixner | 8c2d74f | 2019-08-21 21:09:01 +0200 | [diff] [blame] | 204 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | } |
| 206 | |
Thomas Gleixner | b0d524f | 2019-08-21 21:09:12 +0200 | [diff] [blame] | 207 | static inline void store_samples(u64 *samples, u64 stime, u64 utime, u64 rtime) |
| 208 | { |
| 209 | samples[CPUCLOCK_PROF] = stime + utime; |
| 210 | samples[CPUCLOCK_VIRT] = utime; |
| 211 | samples[CPUCLOCK_SCHED] = rtime; |
| 212 | } |
| 213 | |
| 214 | static void task_sample_cputime(struct task_struct *p, u64 *samples) |
| 215 | { |
| 216 | u64 stime, utime; |
| 217 | |
| 218 | task_cputime(p, &utime, &stime); |
| 219 | store_samples(samples, stime, utime, p->se.sum_exec_runtime); |
| 220 | } |
| 221 | |
| 222 | static void proc_sample_cputime_atomic(struct task_cputime_atomic *at, |
| 223 | u64 *samples) |
| 224 | { |
| 225 | u64 stime, utime, rtime; |
| 226 | |
| 227 | utime = atomic64_read(&at->utime); |
| 228 | stime = atomic64_read(&at->stime); |
| 229 | rtime = atomic64_read(&at->sum_exec_runtime); |
| 230 | store_samples(samples, stime, utime, rtime); |
| 231 | } |
| 232 | |
Jason Low | 1018016 | 2015-04-28 13:00:22 -0700 | [diff] [blame] | 233 | /* |
| 234 | * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg |
| 235 | * to avoid race conditions with concurrent updates to cputime. |
| 236 | */ |
| 237 | static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime) |
Peter Zijlstra | 4da94d49 | 2009-02-11 11:30:27 +0100 | [diff] [blame] | 238 | { |
Jason Low | 1018016 | 2015-04-28 13:00:22 -0700 | [diff] [blame] | 239 | u64 curr_cputime; |
| 240 | retry: |
| 241 | curr_cputime = atomic64_read(cputime); |
| 242 | if (sum_cputime > curr_cputime) { |
| 243 | if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime) |
| 244 | goto retry; |
| 245 | } |
| 246 | } |
Peter Zijlstra | 4da94d49 | 2009-02-11 11:30:27 +0100 | [diff] [blame] | 247 | |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 248 | static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, |
| 249 | struct task_cputime *sum) |
Jason Low | 1018016 | 2015-04-28 13:00:22 -0700 | [diff] [blame] | 250 | { |
Jason Low | 7110744 | 2015-04-28 13:00:24 -0700 | [diff] [blame] | 251 | __update_gt_cputime(&cputime_atomic->utime, sum->utime); |
| 252 | __update_gt_cputime(&cputime_atomic->stime, sum->stime); |
| 253 | __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime); |
Jason Low | 1018016 | 2015-04-28 13:00:22 -0700 | [diff] [blame] | 254 | } |
Peter Zijlstra | 4da94d49 | 2009-02-11 11:30:27 +0100 | [diff] [blame] | 255 | |
Thomas Gleixner | 19298fb | 2019-08-21 21:08:51 +0200 | [diff] [blame] | 256 | /** |
| 257 | * thread_group_sample_cputime - Sample cputime for a given task |
| 258 | * @tsk: Task for which cputime needs to be started |
Yi Wang | 7f2cbcb | 2019-10-21 15:44:12 +0800 | [diff] [blame] | 259 | * @samples: Storage for time samples |
Thomas Gleixner | 19298fb | 2019-08-21 21:08:51 +0200 | [diff] [blame] | 260 | * |
| 261 | * Called from sys_getitimer() to calculate the expiry time of an active |
| 262 | * timer. That means group cputime accounting is already active. Called |
| 263 | * with task sighand lock held. |
| 264 | * |
| 265 | * Updates @times with an uptodate sample of the thread group cputimes. |
| 266 | */ |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 267 | void thread_group_sample_cputime(struct task_struct *tsk, u64 *samples) |
Thomas Gleixner | 19298fb | 2019-08-21 21:08:51 +0200 | [diff] [blame] | 268 | { |
| 269 | struct thread_group_cputimer *cputimer = &tsk->signal->cputimer; |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 270 | struct posix_cputimers *pct = &tsk->signal->posix_cputimers; |
Thomas Gleixner | 19298fb | 2019-08-21 21:08:51 +0200 | [diff] [blame] | 271 | |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 272 | WARN_ON_ONCE(!pct->timers_active); |
Thomas Gleixner | 19298fb | 2019-08-21 21:08:51 +0200 | [diff] [blame] | 273 | |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 274 | proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples); |
Thomas Gleixner | 19298fb | 2019-08-21 21:08:51 +0200 | [diff] [blame] | 275 | } |
| 276 | |
Thomas Gleixner | c506bef4 | 2019-08-21 21:08:54 +0200 | [diff] [blame] | 277 | /** |
| 278 | * thread_group_start_cputime - Start cputime and return a sample |
| 279 | * @tsk: Task for which cputime needs to be started |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 280 | * @samples: Storage for time samples |
Thomas Gleixner | c506bef4 | 2019-08-21 21:08:54 +0200 | [diff] [blame] | 281 | * |
| 282 | * The thread group cputime accouting is avoided when there are no posix |
| 283 | * CPU timers armed. Before starting a timer it's required to check whether |
| 284 | * the time accounting is active. If not, a full update of the atomic |
| 285 | * accounting store needs to be done and the accounting enabled. |
| 286 | * |
| 287 | * Updates @times with an uptodate sample of the thread group cputimes. |
| 288 | */ |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 289 | static void thread_group_start_cputime(struct task_struct *tsk, u64 *samples) |
Peter Zijlstra | 4da94d49 | 2009-02-11 11:30:27 +0100 | [diff] [blame] | 290 | { |
| 291 | struct thread_group_cputimer *cputimer = &tsk->signal->cputimer; |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 292 | struct posix_cputimers *pct = &tsk->signal->posix_cputimers; |
Peter Zijlstra | 4da94d49 | 2009-02-11 11:30:27 +0100 | [diff] [blame] | 293 | |
Jason Low | 1018016 | 2015-04-28 13:00:22 -0700 | [diff] [blame] | 294 | /* Check if cputimer isn't running. This is accessed without locking. */ |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 295 | if (!READ_ONCE(pct->timers_active)) { |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 296 | struct task_cputime sum; |
| 297 | |
Peter Zijlstra | 4da94d49 | 2009-02-11 11:30:27 +0100 | [diff] [blame] | 298 | /* |
| 299 | * The POSIX timer interface allows for absolute time expiry |
| 300 | * values through the TIMER_ABSTIME flag, therefore we have |
Jason Low | 1018016 | 2015-04-28 13:00:22 -0700 | [diff] [blame] | 301 | * to synchronize the timer to the clock every time we start it. |
Peter Zijlstra | 4da94d49 | 2009-02-11 11:30:27 +0100 | [diff] [blame] | 302 | */ |
Frederic Weisbecker | ebd7e7f | 2017-01-31 04:09:34 +0100 | [diff] [blame] | 303 | thread_group_cputime(tsk, &sum); |
Jason Low | 7110744 | 2015-04-28 13:00:24 -0700 | [diff] [blame] | 304 | update_gt_cputime(&cputimer->cputime_atomic, &sum); |
Jason Low | 1018016 | 2015-04-28 13:00:22 -0700 | [diff] [blame] | 305 | |
| 306 | /* |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 307 | * We're setting timers_active without a lock. Ensure this |
| 308 | * only gets written to in one operation. We set it after |
| 309 | * update_gt_cputime() as a small optimization, but |
| 310 | * barriers are not required because update_gt_cputime() |
Jason Low | 1018016 | 2015-04-28 13:00:22 -0700 | [diff] [blame] | 311 | * can handle concurrent updates. |
| 312 | */ |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 313 | WRITE_ONCE(pct->timers_active, true); |
Jason Low | 1018016 | 2015-04-28 13:00:22 -0700 | [diff] [blame] | 314 | } |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 315 | proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples); |
| 316 | } |
| 317 | |
| 318 | static void __thread_group_cputime(struct task_struct *tsk, u64 *samples) |
| 319 | { |
| 320 | struct task_cputime ct; |
| 321 | |
| 322 | thread_group_cputime(tsk, &ct); |
| 323 | store_samples(samples, ct.stime, ct.utime, ct.sum_exec_runtime); |
Peter Zijlstra | 4da94d49 | 2009-02-11 11:30:27 +0100 | [diff] [blame] | 324 | } |
| 325 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 326 | /* |
Thomas Gleixner | 24ab7f5 | 2019-08-21 21:08:55 +0200 | [diff] [blame] | 327 | * Sample a process (thread group) clock for the given task clkid. If the |
| 328 | * group's cputime accounting is already enabled, read the atomic |
Eric W. Biederman | a2efdbf | 2020-02-28 11:08:45 -0600 | [diff] [blame] | 329 | * store. Otherwise a full update is required. clkid is already validated. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 330 | */ |
Thomas Gleixner | 8c2d74f | 2019-08-21 21:09:01 +0200 | [diff] [blame] | 331 | static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p, |
| 332 | bool start) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 333 | { |
Thomas Gleixner | 24ab7f5 | 2019-08-21 21:08:55 +0200 | [diff] [blame] | 334 | struct thread_group_cputimer *cputimer = &p->signal->cputimer; |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 335 | struct posix_cputimers *pct = &p->signal->posix_cputimers; |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 336 | u64 samples[CPUCLOCK_MAX]; |
Frank Mayhar | bb34d92 | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 337 | |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 338 | if (!READ_ONCE(pct->timers_active)) { |
Thomas Gleixner | 24ab7f5 | 2019-08-21 21:08:55 +0200 | [diff] [blame] | 339 | if (start) |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 340 | thread_group_start_cputime(p, samples); |
Thomas Gleixner | 24ab7f5 | 2019-08-21 21:08:55 +0200 | [diff] [blame] | 341 | else |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 342 | __thread_group_cputime(p, samples); |
Thomas Gleixner | 24ab7f5 | 2019-08-21 21:08:55 +0200 | [diff] [blame] | 343 | } else { |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 344 | proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples); |
Thomas Gleixner | 24ab7f5 | 2019-08-21 21:08:55 +0200 | [diff] [blame] | 345 | } |
| 346 | |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 347 | return samples[clkid]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | } |
| 349 | |
Thomas Gleixner | bfcf3e9 | 2019-08-21 21:08:49 +0200 | [diff] [blame] | 350 | static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp) |
Frederic Weisbecker | 33ab0fe | 2013-10-11 17:41:11 +0200 | [diff] [blame] | 351 | { |
Thomas Gleixner | bfcf3e9 | 2019-08-21 21:08:49 +0200 | [diff] [blame] | 352 | const clockid_t clkid = CPUCLOCK_WHICH(clock); |
| 353 | struct task_struct *tsk; |
| 354 | u64 t; |
Frederic Weisbecker | 33ab0fe | 2013-10-11 17:41:11 +0200 | [diff] [blame] | 355 | |
Eric W. Biederman | 9bf7c32 | 2020-04-25 18:38:54 -0500 | [diff] [blame] | 356 | rcu_read_lock(); |
Eric W. Biederman | 9649877 | 2020-04-27 07:55:07 -0500 | [diff] [blame] | 357 | tsk = pid_task(pid_for_clock(clock, true), clock_pid_type(clock)); |
Eric W. Biederman | 9bf7c32 | 2020-04-25 18:38:54 -0500 | [diff] [blame] | 358 | if (!tsk) { |
| 359 | rcu_read_unlock(); |
Thomas Gleixner | bfcf3e9 | 2019-08-21 21:08:49 +0200 | [diff] [blame] | 360 | return -EINVAL; |
Eric W. Biederman | 9bf7c32 | 2020-04-25 18:38:54 -0500 | [diff] [blame] | 361 | } |
Frederic Weisbecker | 33ab0fe | 2013-10-11 17:41:11 +0200 | [diff] [blame] | 362 | |
Thomas Gleixner | bfcf3e9 | 2019-08-21 21:08:49 +0200 | [diff] [blame] | 363 | if (CPUCLOCK_PERTHREAD(clock)) |
Thomas Gleixner | 8c2d74f | 2019-08-21 21:09:01 +0200 | [diff] [blame] | 364 | t = cpu_clock_sample(clkid, tsk); |
Thomas Gleixner | bfcf3e9 | 2019-08-21 21:08:49 +0200 | [diff] [blame] | 365 | else |
Thomas Gleixner | 8c2d74f | 2019-08-21 21:09:01 +0200 | [diff] [blame] | 366 | t = cpu_clock_sample_group(clkid, tsk, false); |
Eric W. Biederman | 9bf7c32 | 2020-04-25 18:38:54 -0500 | [diff] [blame] | 367 | rcu_read_unlock(); |
Frederic Weisbecker | 33ab0fe | 2013-10-11 17:41:11 +0200 | [diff] [blame] | 368 | |
Thomas Gleixner | bfcf3e9 | 2019-08-21 21:08:49 +0200 | [diff] [blame] | 369 | *tp = ns_to_timespec64(t); |
| 370 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 371 | } |
| 372 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 373 | /* |
| 374 | * Validate the clockid_t for a new CPU-clock timer, and initialize the timer. |
Stanislaw Gruszka | ba5ea95 | 2009-11-17 14:14:13 -0800 | [diff] [blame] | 375 | * This is called from sys_timer_create() and do_cpu_nanosleep() with the |
| 376 | * new timer already all-zeros initialized. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | */ |
Thomas Gleixner | bc2c8ea | 2011-02-01 13:52:12 +0000 | [diff] [blame] | 378 | static int posix_cpu_timer_create(struct k_itimer *new_timer) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | { |
Thomas Gleixner | 1fb497d | 2020-07-30 12:14:06 +0200 | [diff] [blame] | 380 | static struct lock_class_key posix_cpu_timers_key; |
Eric W. Biederman | 9649877 | 2020-04-27 07:55:07 -0500 | [diff] [blame] | 381 | struct pid *pid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | |
Eric W. Biederman | 9bf7c32 | 2020-04-25 18:38:54 -0500 | [diff] [blame] | 383 | rcu_read_lock(); |
Eric W. Biederman | 9649877 | 2020-04-27 07:55:07 -0500 | [diff] [blame] | 384 | pid = pid_for_clock(new_timer->it_clock, false); |
| 385 | if (!pid) { |
Eric W. Biederman | 9bf7c32 | 2020-04-25 18:38:54 -0500 | [diff] [blame] | 386 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 387 | return -EINVAL; |
Eric W. Biederman | 9bf7c32 | 2020-04-25 18:38:54 -0500 | [diff] [blame] | 388 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 389 | |
Thomas Gleixner | 1fb497d | 2020-07-30 12:14:06 +0200 | [diff] [blame] | 390 | /* |
| 391 | * If posix timer expiry is handled in task work context then |
| 392 | * timer::it_lock can be taken without disabling interrupts as all |
| 393 | * other locking happens in task context. This requires a seperate |
| 394 | * lock class key otherwise regular posix timer expiry would record |
| 395 | * the lock class being taken in interrupt context and generate a |
| 396 | * false positive warning. |
| 397 | */ |
| 398 | if (IS_ENABLED(CONFIG_POSIX_CPU_TIMERS_TASK_WORK)) |
| 399 | lockdep_set_class(&new_timer->it_lock, &posix_cpu_timers_key); |
| 400 | |
Thomas Gleixner | d97bb75 | 2017-05-30 23:15:44 +0200 | [diff] [blame] | 401 | new_timer->kclock = &clock_posix_cpu; |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 402 | timerqueue_init(&new_timer->it.cpu.node); |
Eric W. Biederman | 9649877 | 2020-04-27 07:55:07 -0500 | [diff] [blame] | 403 | new_timer->it.cpu.pid = get_pid(pid); |
Eric W. Biederman | 9bf7c32 | 2020-04-25 18:38:54 -0500 | [diff] [blame] | 404 | rcu_read_unlock(); |
Thomas Gleixner | e5a8b65 | 2019-08-21 21:08:50 +0200 | [diff] [blame] | 405 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 406 | } |
| 407 | |
| 408 | /* |
| 409 | * Clean up a CPU-clock timer that is about to be destroyed. |
| 410 | * This is called from timer deletion with the timer already locked. |
| 411 | * If we return TIMER_RETRY, it's necessary to release the timer's lock |
| 412 | * and try again. (This happens when the timer is in the middle of firing.) |
| 413 | */ |
Thomas Gleixner | bc2c8ea | 2011-02-01 13:52:12 +0000 | [diff] [blame] | 414 | static int posix_cpu_timer_del(struct k_itimer *timer) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 415 | { |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 416 | struct cpu_timer *ctmr = &timer->it.cpu; |
Frederic Weisbecker | 3d7a142 | 2013-10-11 17:41:11 +0200 | [diff] [blame] | 417 | struct sighand_struct *sighand; |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 418 | struct task_struct *p; |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 419 | unsigned long flags; |
| 420 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 422 | rcu_read_lock(); |
| 423 | p = cpu_timer_task_rcu(timer); |
| 424 | if (!p) |
| 425 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 426 | |
Frederic Weisbecker | 3d7a142 | 2013-10-11 17:41:11 +0200 | [diff] [blame] | 427 | /* |
| 428 | * Protect against sighand release/switch in exit/exec and process/ |
| 429 | * thread timer list entry concurrent read/writes. |
| 430 | */ |
| 431 | sighand = lock_task_sighand(p, &flags); |
| 432 | if (unlikely(sighand == NULL)) { |
Frederic Weisbecker | a3222f8 | 2013-10-11 00:37:39 +0200 | [diff] [blame] | 433 | /* |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 434 | * This raced with the reaping of the task. The exit cleanup |
| 435 | * should have removed this timer from the timer queue. |
Frederic Weisbecker | a3222f8 | 2013-10-11 00:37:39 +0200 | [diff] [blame] | 436 | */ |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 437 | WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node)); |
Frederic Weisbecker | a3222f8 | 2013-10-11 00:37:39 +0200 | [diff] [blame] | 438 | } else { |
Frederic Weisbecker | a3222f8 | 2013-10-11 00:37:39 +0200 | [diff] [blame] | 439 | if (timer->it.cpu.firing) |
| 440 | ret = TIMER_RETRY; |
| 441 | else |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 442 | cpu_timer_dequeue(ctmr); |
Frederic Weisbecker | 3d7a142 | 2013-10-11 17:41:11 +0200 | [diff] [blame] | 443 | |
| 444 | unlock_task_sighand(p, &flags); |
Oleg Nesterov | 108150e | 2005-10-23 20:25:39 +0400 | [diff] [blame] | 445 | } |
Frederic Weisbecker | a3222f8 | 2013-10-11 00:37:39 +0200 | [diff] [blame] | 446 | |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 447 | out: |
| 448 | rcu_read_unlock(); |
Frederic Weisbecker | a3222f8 | 2013-10-11 00:37:39 +0200 | [diff] [blame] | 449 | if (!ret) |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 450 | put_pid(ctmr->pid); |
Oleg Nesterov | 108150e | 2005-10-23 20:25:39 +0400 | [diff] [blame] | 451 | |
| 452 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | } |
| 454 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 455 | static void cleanup_timerqueue(struct timerqueue_head *head) |
Frederic Weisbecker | 1a7fa51 | 2013-06-28 00:06:42 +0000 | [diff] [blame] | 456 | { |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 457 | struct timerqueue_node *node; |
| 458 | struct cpu_timer *ctmr; |
Frederic Weisbecker | 1a7fa51 | 2013-06-28 00:06:42 +0000 | [diff] [blame] | 459 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 460 | while ((node = timerqueue_getnext(head))) { |
| 461 | timerqueue_del(head, node); |
| 462 | ctmr = container_of(node, struct cpu_timer, node); |
| 463 | ctmr->head = NULL; |
| 464 | } |
Frederic Weisbecker | 1a7fa51 | 2013-06-28 00:06:42 +0000 | [diff] [blame] | 465 | } |
| 466 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | /* |
Thomas Gleixner | 7cb9a94 | 2019-08-19 16:31:45 +0200 | [diff] [blame] | 468 | * Clean out CPU timers which are still armed when a thread exits. The |
| 469 | * timers are only removed from the list. No other updates are done. The |
| 470 | * corresponding posix timers are still accessible, but cannot be rearmed. |
| 471 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | * This must be called with the siglock held. |
| 473 | */ |
Thomas Gleixner | 2b69942 | 2019-08-21 21:09:04 +0200 | [diff] [blame] | 474 | static void cleanup_timers(struct posix_cputimers *pct) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 475 | { |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 476 | cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead); |
| 477 | cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead); |
| 478 | cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 479 | } |
| 480 | |
| 481 | /* |
| 482 | * These are both called with the siglock held, when the current thread |
| 483 | * is being reaped. When the final (leader) thread in the group is reaped, |
| 484 | * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit. |
| 485 | */ |
| 486 | void posix_cpu_timers_exit(struct task_struct *tsk) |
| 487 | { |
Thomas Gleixner | 2b69942 | 2019-08-21 21:09:04 +0200 | [diff] [blame] | 488 | cleanup_timers(&tsk->posix_cputimers); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | } |
| 490 | void posix_cpu_timers_exit_group(struct task_struct *tsk) |
| 491 | { |
Thomas Gleixner | 2b69942 | 2019-08-21 21:09:04 +0200 | [diff] [blame] | 492 | cleanup_timers(&tsk->signal->posix_cputimers); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | } |
| 494 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | /* |
| 496 | * Insert the timer on the appropriate list before any timers that |
Frederic Weisbecker | e73d84e | 2013-10-11 18:56:49 +0200 | [diff] [blame] | 497 | * expire later. This must be called with the sighand lock held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 498 | */ |
Eric W. Biederman | beb41d9 | 2020-02-28 11:09:46 -0600 | [diff] [blame] | 499 | static void arm_timer(struct k_itimer *timer, struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 500 | { |
Thomas Gleixner | 3b495b2 | 2019-08-21 21:09:08 +0200 | [diff] [blame] | 501 | int clkidx = CPUCLOCK_WHICH(timer->it_clock); |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 502 | struct cpu_timer *ctmr = &timer->it.cpu; |
| 503 | u64 newexp = cpu_timer_getexpires(ctmr); |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 504 | struct posix_cputimer_base *base; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 505 | |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 506 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
| 507 | base = p->posix_cputimers.bases + clkidx; |
| 508 | else |
| 509 | base = p->signal->posix_cputimers.bases + clkidx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 510 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 511 | if (!cpu_timer_enqueue(&base->tqhead, ctmr)) |
Thomas Gleixner | 3b495b2 | 2019-08-21 21:09:08 +0200 | [diff] [blame] | 512 | return; |
Stanislaw Gruszka | 5eb9aa6 | 2010-03-11 14:04:38 -0800 | [diff] [blame] | 513 | |
Thomas Gleixner | 3b495b2 | 2019-08-21 21:09:08 +0200 | [diff] [blame] | 514 | /* |
| 515 | * We are the new earliest-expiring POSIX 1.b timer, hence |
| 516 | * need to update expiration cache. Take into account that |
| 517 | * for process timers we share expiration cache with itimers |
| 518 | * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME. |
| 519 | */ |
Thomas Gleixner | 2bbdbda | 2019-08-21 21:09:19 +0200 | [diff] [blame] | 520 | if (newexp < base->nextevt) |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 521 | base->nextevt = newexp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 522 | |
Thomas Gleixner | 3b495b2 | 2019-08-21 21:09:08 +0200 | [diff] [blame] | 523 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
| 524 | tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER); |
| 525 | else |
| 526 | tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | } |
| 528 | |
| 529 | /* |
| 530 | * The timer is locked, fire it and arrange for its reload. |
| 531 | */ |
| 532 | static void cpu_timer_fire(struct k_itimer *timer) |
| 533 | { |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 534 | struct cpu_timer *ctmr = &timer->it.cpu; |
| 535 | |
Stanislaw Gruszka | 1f169f8 | 2010-03-11 14:04:41 -0800 | [diff] [blame] | 536 | if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) { |
| 537 | /* |
| 538 | * User don't want any signal. |
| 539 | */ |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 540 | cpu_timer_setexpires(ctmr, 0); |
Stanislaw Gruszka | 1f169f8 | 2010-03-11 14:04:41 -0800 | [diff] [blame] | 541 | } else if (unlikely(timer->sigq == NULL)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 542 | /* |
| 543 | * This a special case for clock_nanosleep, |
| 544 | * not a normal timer from sys_timer_create. |
| 545 | */ |
| 546 | wake_up_process(timer->it_process); |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 547 | cpu_timer_setexpires(ctmr, 0); |
Thomas Gleixner | 1611879 | 2019-01-11 14:33:17 +0100 | [diff] [blame] | 548 | } else if (!timer->it_interval) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 549 | /* |
| 550 | * One-shot timer. Clear it as soon as it's fired. |
| 551 | */ |
| 552 | posix_timer_event(timer, 0); |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 553 | cpu_timer_setexpires(ctmr, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 554 | } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) { |
| 555 | /* |
| 556 | * The signal did not get queued because the signal |
| 557 | * was ignored, so we won't get any callback to |
| 558 | * reload the timer. But we need to keep it |
| 559 | * ticking in case the signal is deliverable next time. |
| 560 | */ |
Thomas Gleixner | f37fb0a | 2017-05-30 23:15:47 +0200 | [diff] [blame] | 561 | posix_cpu_timer_rearm(timer); |
Thomas Gleixner | af888d6 | 2017-05-30 23:15:42 +0200 | [diff] [blame] | 562 | ++timer->it_requeue_pending; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 563 | } |
| 564 | } |
| 565 | |
| 566 | /* |
| 567 | * Guts of sys_timer_settime for CPU timers. |
| 568 | * This is called with the timer locked and interrupts disabled. |
| 569 | * If we return TIMER_RETRY, it's necessary to release the timer's lock |
| 570 | * and try again. (This happens when the timer is in the middle of firing.) |
| 571 | */ |
Frederic Weisbecker | e73d84e | 2013-10-11 18:56:49 +0200 | [diff] [blame] | 572 | static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags, |
Deepa Dinamani | 5f252b3 | 2017-03-26 12:04:17 -0700 | [diff] [blame] | 573 | struct itimerspec64 *new, struct itimerspec64 *old) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 574 | { |
Thomas Gleixner | c7a37c6 | 2019-08-21 21:08:56 +0200 | [diff] [blame] | 575 | clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock); |
Frederic Weisbecker | ebd7e7f | 2017-01-31 04:09:34 +0100 | [diff] [blame] | 576 | u64 old_expires, new_expires, old_incr, val; |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 577 | struct cpu_timer *ctmr = &timer->it.cpu; |
Thomas Gleixner | c7a37c6 | 2019-08-21 21:08:56 +0200 | [diff] [blame] | 578 | struct sighand_struct *sighand; |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 579 | struct task_struct *p; |
Thomas Gleixner | c7a37c6 | 2019-08-21 21:08:56 +0200 | [diff] [blame] | 580 | unsigned long flags; |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 581 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 582 | |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 583 | rcu_read_lock(); |
| 584 | p = cpu_timer_task_rcu(timer); |
| 585 | if (!p) { |
| 586 | /* |
| 587 | * If p has just been reaped, we can no |
| 588 | * longer get any information about it at all. |
| 589 | */ |
| 590 | rcu_read_unlock(); |
| 591 | return -ESRCH; |
| 592 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 593 | |
Thomas Gleixner | 098b0e0 | 2017-06-20 17:37:36 +0200 | [diff] [blame] | 594 | /* |
| 595 | * Use the to_ktime conversion because that clamps the maximum |
| 596 | * value to KTIME_MAX and avoid multiplication overflows. |
| 597 | */ |
| 598 | new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 599 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | /* |
Frederic Weisbecker | e73d84e | 2013-10-11 18:56:49 +0200 | [diff] [blame] | 601 | * Protect against sighand release/switch in exit/exec and p->cpu_timers |
| 602 | * and p->signal->cpu_timers read/write in arm_timer() |
| 603 | */ |
| 604 | sighand = lock_task_sighand(p, &flags); |
| 605 | /* |
| 606 | * If p has just been reaped, we can no |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | * longer get any information about it at all. |
| 608 | */ |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 609 | if (unlikely(sighand == NULL)) { |
| 610 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | return -ESRCH; |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 612 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | |
| 614 | /* |
| 615 | * Disarm any old timer after extracting its expiry time. |
| 616 | */ |
Thomas Gleixner | 1611879 | 2019-01-11 14:33:17 +0100 | [diff] [blame] | 617 | old_incr = timer->it_interval; |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 618 | old_expires = cpu_timer_getexpires(ctmr); |
| 619 | |
Oleg Nesterov | a69ac4a | 2005-10-24 18:29:58 +0400 | [diff] [blame] | 620 | if (unlikely(timer->it.cpu.firing)) { |
| 621 | timer->it.cpu.firing = -1; |
| 622 | ret = TIMER_RETRY; |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 623 | } else { |
| 624 | cpu_timer_dequeue(ctmr); |
| 625 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 626 | |
| 627 | /* |
| 628 | * We need to sample the current value to convert the new |
| 629 | * value from to relative and absolute, and to convert the |
| 630 | * old value from absolute to relative. To set a process |
| 631 | * timer, we need a sample to balance the thread expiry |
| 632 | * times (in arm_timer). With an absolute time, we must |
| 633 | * check if it's already passed. In short, we need a sample. |
| 634 | */ |
Thomas Gleixner | 8c2d74f | 2019-08-21 21:09:01 +0200 | [diff] [blame] | 635 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
| 636 | val = cpu_clock_sample(clkid, p); |
| 637 | else |
| 638 | val = cpu_clock_sample_group(clkid, p, true); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 639 | |
| 640 | if (old) { |
Frederic Weisbecker | 55ccb61 | 2013-06-28 00:06:42 +0000 | [diff] [blame] | 641 | if (old_expires == 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 642 | old->it_value.tv_sec = 0; |
| 643 | old->it_value.tv_nsec = 0; |
| 644 | } else { |
| 645 | /* |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 646 | * Update the timer in case it has overrun already. |
| 647 | * If it has, we'll report it as having overrun and |
| 648 | * with the next reloaded timer already ticking, |
| 649 | * though we are swallowing that pending |
| 650 | * notification here to install the new setting. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 651 | */ |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 652 | u64 exp = bump_cpu_timer(timer, val); |
| 653 | |
| 654 | if (val < exp) { |
| 655 | old_expires = exp - val; |
Deepa Dinamani | 5f252b3 | 2017-03-26 12:04:17 -0700 | [diff] [blame] | 656 | old->it_value = ns_to_timespec64(old_expires); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 657 | } else { |
| 658 | old->it_value.tv_nsec = 1; |
| 659 | old->it_value.tv_sec = 0; |
| 660 | } |
| 661 | } |
| 662 | } |
| 663 | |
Oleg Nesterov | a69ac4a | 2005-10-24 18:29:58 +0400 | [diff] [blame] | 664 | if (unlikely(ret)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 665 | /* |
| 666 | * We are colliding with the timer actually firing. |
| 667 | * Punt after filling in the timer's old value, and |
| 668 | * disable this firing since we are already reporting |
| 669 | * it as an overrun (thanks to bump_cpu_timer above). |
| 670 | */ |
Frederic Weisbecker | e73d84e | 2013-10-11 18:56:49 +0200 | [diff] [blame] | 671 | unlock_task_sighand(p, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 672 | goto out; |
| 673 | } |
| 674 | |
Frederic Weisbecker | e73d84e | 2013-10-11 18:56:49 +0200 | [diff] [blame] | 675 | if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) { |
Frederic Weisbecker | 55ccb61 | 2013-06-28 00:06:42 +0000 | [diff] [blame] | 676 | new_expires += val; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | } |
| 678 | |
| 679 | /* |
| 680 | * Install the new expiry time (or zero). |
| 681 | * For a timer with no notification action, we don't actually |
| 682 | * arm the timer (we'll just fake it for timer_gettime). |
| 683 | */ |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 684 | cpu_timer_setexpires(ctmr, new_expires); |
Frederic Weisbecker | 55ccb61 | 2013-06-28 00:06:42 +0000 | [diff] [blame] | 685 | if (new_expires != 0 && val < new_expires) { |
Eric W. Biederman | beb41d9 | 2020-02-28 11:09:46 -0600 | [diff] [blame] | 686 | arm_timer(timer, p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 687 | } |
| 688 | |
Frederic Weisbecker | e73d84e | 2013-10-11 18:56:49 +0200 | [diff] [blame] | 689 | unlock_task_sighand(p, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | /* |
| 691 | * Install the new reload setting, and |
| 692 | * set up the signal and overrun bookkeeping. |
| 693 | */ |
Thomas Gleixner | 1611879 | 2019-01-11 14:33:17 +0100 | [diff] [blame] | 694 | timer->it_interval = timespec64_to_ktime(new->it_interval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | |
| 696 | /* |
| 697 | * This acts as a modification timestamp for the timer, |
| 698 | * so any automatic reload attempt will punt on seeing |
| 699 | * that we have reset the timer manually. |
| 700 | */ |
| 701 | timer->it_requeue_pending = (timer->it_requeue_pending + 2) & |
| 702 | ~REQUEUE_PENDING; |
| 703 | timer->it_overrun_last = 0; |
| 704 | timer->it_overrun = -1; |
| 705 | |
Frederic Weisbecker | 55ccb61 | 2013-06-28 00:06:42 +0000 | [diff] [blame] | 706 | if (new_expires != 0 && !(val < new_expires)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 707 | /* |
| 708 | * The designated time already passed, so we notify |
| 709 | * immediately, even if the thread never runs to |
| 710 | * accumulate more time on this clock. |
| 711 | */ |
| 712 | cpu_timer_fire(timer); |
| 713 | } |
| 714 | |
| 715 | ret = 0; |
| 716 | out: |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 717 | rcu_read_unlock(); |
Frederic Weisbecker | ebd7e7f | 2017-01-31 04:09:34 +0100 | [diff] [blame] | 718 | if (old) |
Deepa Dinamani | 5f252b3 | 2017-03-26 12:04:17 -0700 | [diff] [blame] | 719 | old->it_interval = ns_to_timespec64(old_incr); |
Frederic Weisbecker | b787830 | 2015-07-17 22:25:49 +0200 | [diff] [blame] | 720 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 721 | return ret; |
| 722 | } |
| 723 | |
Deepa Dinamani | 5f252b3 | 2017-03-26 12:04:17 -0700 | [diff] [blame] | 724 | static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 725 | { |
Thomas Gleixner | 99093c5 | 2019-08-21 21:08:57 +0200 | [diff] [blame] | 726 | clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock); |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 727 | struct cpu_timer *ctmr = &timer->it.cpu; |
| 728 | u64 now, expires = cpu_timer_getexpires(ctmr); |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 729 | struct task_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 730 | |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 731 | rcu_read_lock(); |
| 732 | p = cpu_timer_task_rcu(timer); |
| 733 | if (!p) |
| 734 | goto out; |
Frederic Weisbecker | a3222f8 | 2013-10-11 00:37:39 +0200 | [diff] [blame] | 735 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 736 | /* |
| 737 | * Easy part: convert the reload time. |
| 738 | */ |
Thomas Gleixner | 1611879 | 2019-01-11 14:33:17 +0100 | [diff] [blame] | 739 | itp->it_interval = ktime_to_timespec64(timer->it_interval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 740 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 741 | if (!expires) |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 742 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 743 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 744 | /* |
| 745 | * Sample the clock to take the difference with the expiry time. |
| 746 | */ |
Eric W. Biederman | 60f2cea | 2020-02-28 11:09:19 -0600 | [diff] [blame] | 747 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
Thomas Gleixner | 8c2d74f | 2019-08-21 21:09:01 +0200 | [diff] [blame] | 748 | now = cpu_clock_sample(clkid, p); |
Eric W. Biederman | 60f2cea | 2020-02-28 11:09:19 -0600 | [diff] [blame] | 749 | else |
| 750 | now = cpu_clock_sample_group(clkid, p, false); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 751 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 752 | if (now < expires) { |
| 753 | itp->it_value = ns_to_timespec64(expires - now); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 754 | } else { |
| 755 | /* |
| 756 | * The timer should have expired already, but the firing |
| 757 | * hasn't taken place yet. Say it's just about to expire. |
| 758 | */ |
| 759 | itp->it_value.tv_nsec = 1; |
| 760 | itp->it_value.tv_sec = 0; |
| 761 | } |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 762 | out: |
| 763 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 764 | } |
| 765 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 766 | #define MAX_COLLECTED 20 |
| 767 | |
| 768 | static u64 collect_timerqueue(struct timerqueue_head *head, |
| 769 | struct list_head *firing, u64 now) |
Frederic Weisbecker | 2473f3e | 2013-06-28 00:06:43 +0000 | [diff] [blame] | 770 | { |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 771 | struct timerqueue_node *next; |
| 772 | int i = 0; |
Frederic Weisbecker | 2473f3e | 2013-06-28 00:06:43 +0000 | [diff] [blame] | 773 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 774 | while ((next = timerqueue_getnext(head))) { |
| 775 | struct cpu_timer *ctmr; |
| 776 | u64 expires; |
Frederic Weisbecker | 2473f3e | 2013-06-28 00:06:43 +0000 | [diff] [blame] | 777 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 778 | ctmr = container_of(next, struct cpu_timer, node); |
| 779 | expires = cpu_timer_getexpires(ctmr); |
| 780 | /* Limit the number of timers to expire at once */ |
| 781 | if (++i == MAX_COLLECTED || now < expires) |
| 782 | return expires; |
Frederic Weisbecker | 2473f3e | 2013-06-28 00:06:43 +0000 | [diff] [blame] | 783 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 784 | ctmr->firing = 1; |
| 785 | cpu_timer_dequeue(ctmr); |
| 786 | list_add_tail(&ctmr->elist, firing); |
Frederic Weisbecker | 2473f3e | 2013-06-28 00:06:43 +0000 | [diff] [blame] | 787 | } |
| 788 | |
Thomas Gleixner | 2bbdbda | 2019-08-21 21:09:19 +0200 | [diff] [blame] | 789 | return U64_MAX; |
Frederic Weisbecker | 2473f3e | 2013-06-28 00:06:43 +0000 | [diff] [blame] | 790 | } |
| 791 | |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 792 | static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples, |
| 793 | struct list_head *firing) |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 794 | { |
| 795 | struct posix_cputimer_base *base = pct->bases; |
| 796 | int i; |
| 797 | |
| 798 | for (i = 0; i < CPUCLOCK_MAX; i++, base++) { |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 799 | base->nextevt = collect_timerqueue(&base->tqhead, firing, |
| 800 | samples[i]); |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 801 | } |
| 802 | } |
| 803 | |
Juri Lelli | 34be393 | 2017-12-12 12:10:24 +0100 | [diff] [blame] | 804 | static inline void check_dl_overrun(struct task_struct *tsk) |
| 805 | { |
| 806 | if (tsk->dl.dl_overrun) { |
| 807 | tsk->dl.dl_overrun = 0; |
| 808 | __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk); |
| 809 | } |
| 810 | } |
| 811 | |
Thomas Gleixner | 8991afe | 2019-08-21 21:09:23 +0200 | [diff] [blame] | 812 | static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard) |
| 813 | { |
| 814 | if (time < limit) |
| 815 | return false; |
| 816 | |
| 817 | if (print_fatal_signals) { |
| 818 | pr_info("%s Watchdog Timeout (%s): %s[%d]\n", |
| 819 | rt ? "RT" : "CPU", hard ? "hard" : "soft", |
| 820 | current->comm, task_pid_nr(current)); |
| 821 | } |
| 822 | __group_send_sig_info(signo, SEND_SIG_PRIV, current); |
| 823 | return true; |
| 824 | } |
| 825 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 826 | /* |
| 827 | * Check for any per-thread CPU timers that have fired and move them off |
| 828 | * the tsk->cpu_timers[N] list onto the firing list. Here we update the |
| 829 | * tsk->it_*_expires values to reflect the remaining thread CPU timers. |
| 830 | */ |
| 831 | static void check_thread_timers(struct task_struct *tsk, |
| 832 | struct list_head *firing) |
| 833 | { |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 834 | struct posix_cputimers *pct = &tsk->posix_cputimers; |
| 835 | u64 samples[CPUCLOCK_MAX]; |
Jiri Slaby | d4bb5274 | 2010-03-05 13:42:53 -0800 | [diff] [blame] | 836 | unsigned long soft; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 837 | |
Juri Lelli | 34be393 | 2017-12-12 12:10:24 +0100 | [diff] [blame] | 838 | if (dl_task(tsk)) |
| 839 | check_dl_overrun(tsk); |
| 840 | |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 841 | if (expiry_cache_is_inactive(pct)) |
Jason Low | 934715a | 2015-10-14 12:07:54 -0700 | [diff] [blame] | 842 | return; |
| 843 | |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 844 | task_sample_cputime(tsk, samples); |
| 845 | collect_posix_cputimers(pct, samples, firing); |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 846 | |
| 847 | /* |
| 848 | * Check for the special case thread timers. |
| 849 | */ |
Krzysztof Opasiak | 3cf2949 | 2017-07-05 19:25:48 +0200 | [diff] [blame] | 850 | soft = task_rlimit(tsk, RLIMIT_RTTIME); |
Jiri Slaby | d4bb5274 | 2010-03-05 13:42:53 -0800 | [diff] [blame] | 851 | if (soft != RLIM_INFINITY) { |
Thomas Gleixner | 8ea1de9 | 2019-08-21 21:09:21 +0200 | [diff] [blame] | 852 | /* Task RT timeout is accounted in jiffies. RTTIME is usec */ |
Thomas Gleixner | 8991afe | 2019-08-21 21:09:23 +0200 | [diff] [blame] | 853 | unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ); |
Krzysztof Opasiak | 3cf2949 | 2017-07-05 19:25:48 +0200 | [diff] [blame] | 854 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME); |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 855 | |
Thomas Gleixner | 8991afe | 2019-08-21 21:09:23 +0200 | [diff] [blame] | 856 | /* At the hard limit, send SIGKILL. No further action. */ |
| 857 | if (hard != RLIM_INFINITY && |
| 858 | check_rlimit(rttime, hard, SIGKILL, true, true)) |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 859 | return; |
Thomas Gleixner | dd67022 | 2019-08-21 21:09:22 +0200 | [diff] [blame] | 860 | |
Thomas Gleixner | 8991afe | 2019-08-21 21:09:23 +0200 | [diff] [blame] | 861 | /* At the soft limit, send a SIGXCPU every second */ |
| 862 | if (check_rlimit(rttime, soft, SIGXCPU, true, false)) { |
Thomas Gleixner | dd67022 | 2019-08-21 21:09:22 +0200 | [diff] [blame] | 863 | soft += USEC_PER_SEC; |
| 864 | tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft; |
Peter Zijlstra | 78f2c7d | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 865 | } |
| 866 | } |
Thomas Gleixner | c02b078 | 2019-08-21 21:09:10 +0200 | [diff] [blame] | 867 | |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 868 | if (expiry_cache_is_inactive(pct)) |
Frederic Weisbecker | b787830 | 2015-07-17 22:25:49 +0200 | [diff] [blame] | 869 | tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 870 | } |
| 871 | |
Jason Low | 1018016 | 2015-04-28 13:00:22 -0700 | [diff] [blame] | 872 | static inline void stop_process_timers(struct signal_struct *sig) |
Peter Zijlstra | 3fccfd6 | 2009-02-10 16:37:31 +0100 | [diff] [blame] | 873 | { |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 874 | struct posix_cputimers *pct = &sig->posix_cputimers; |
Peter Zijlstra | 3fccfd6 | 2009-02-10 16:37:31 +0100 | [diff] [blame] | 875 | |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 876 | /* Turn off the active flag. This is done without locking. */ |
| 877 | WRITE_ONCE(pct->timers_active, false); |
Frederic Weisbecker | b787830 | 2015-07-17 22:25:49 +0200 | [diff] [blame] | 878 | tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER); |
Peter Zijlstra | 3fccfd6 | 2009-02-10 16:37:31 +0100 | [diff] [blame] | 879 | } |
| 880 | |
Stanislaw Gruszka | 42c4ab4 | 2009-07-29 12:15:26 +0200 | [diff] [blame] | 881 | static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it, |
Frederic Weisbecker | ebd7e7f | 2017-01-31 04:09:34 +0100 | [diff] [blame] | 882 | u64 *expires, u64 cur_time, int signo) |
Stanislaw Gruszka | 42c4ab4 | 2009-07-29 12:15:26 +0200 | [diff] [blame] | 883 | { |
Martin Schwidefsky | 6486163 | 2011-12-15 14:56:09 +0100 | [diff] [blame] | 884 | if (!it->expires) |
Stanislaw Gruszka | 42c4ab4 | 2009-07-29 12:15:26 +0200 | [diff] [blame] | 885 | return; |
| 886 | |
Frederic Weisbecker | 858cf3a | 2017-01-31 04:09:35 +0100 | [diff] [blame] | 887 | if (cur_time >= it->expires) { |
| 888 | if (it->incr) |
Martin Schwidefsky | 6486163 | 2011-12-15 14:56:09 +0100 | [diff] [blame] | 889 | it->expires += it->incr; |
Frederic Weisbecker | 858cf3a | 2017-01-31 04:09:35 +0100 | [diff] [blame] | 890 | else |
Martin Schwidefsky | 6486163 | 2011-12-15 14:56:09 +0100 | [diff] [blame] | 891 | it->expires = 0; |
Stanislaw Gruszka | 42c4ab4 | 2009-07-29 12:15:26 +0200 | [diff] [blame] | 892 | |
Xiao Guangrong | 3f0a525 | 2009-08-10 10:52:30 +0800 | [diff] [blame] | 893 | trace_itimer_expire(signo == SIGPROF ? |
| 894 | ITIMER_PROF : ITIMER_VIRTUAL, |
Eric W. Biederman | 6883f81 | 2017-06-04 04:32:13 -0500 | [diff] [blame] | 895 | task_tgid(tsk), cur_time); |
Stanislaw Gruszka | 42c4ab4 | 2009-07-29 12:15:26 +0200 | [diff] [blame] | 896 | __group_send_sig_info(signo, SEND_SIG_PRIV, tsk); |
| 897 | } |
| 898 | |
Thomas Gleixner | 2bbdbda | 2019-08-21 21:09:19 +0200 | [diff] [blame] | 899 | if (it->expires && it->expires < *expires) |
Frederic Weisbecker | 858cf3a | 2017-01-31 04:09:35 +0100 | [diff] [blame] | 900 | *expires = it->expires; |
Stanislaw Gruszka | 42c4ab4 | 2009-07-29 12:15:26 +0200 | [diff] [blame] | 901 | } |
| 902 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 903 | /* |
| 904 | * Check for any per-thread CPU timers that have fired and move them |
| 905 | * off the tsk->*_timers list onto the firing list. Per-thread timers |
| 906 | * have already been taken off. |
| 907 | */ |
| 908 | static void check_process_timers(struct task_struct *tsk, |
| 909 | struct list_head *firing) |
| 910 | { |
| 911 | struct signal_struct *const sig = tsk->signal; |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 912 | struct posix_cputimers *pct = &sig->posix_cputimers; |
| 913 | u64 samples[CPUCLOCK_MAX]; |
Jiri Slaby | d4bb5274 | 2010-03-05 13:42:53 -0800 | [diff] [blame] | 914 | unsigned long soft; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 915 | |
| 916 | /* |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 917 | * If there are no active process wide timers (POSIX 1.b, itimers, |
Thomas Gleixner | a2ed4fd | 2019-08-29 12:52:28 +0200 | [diff] [blame] | 918 | * RLIMIT_CPU) nothing to check. Also skip the process wide timer |
| 919 | * processing when there is already another task handling them. |
Jason Low | 934715a | 2015-10-14 12:07:54 -0700 | [diff] [blame] | 920 | */ |
Thomas Gleixner | a2ed4fd | 2019-08-29 12:52:28 +0200 | [diff] [blame] | 921 | if (!READ_ONCE(pct->timers_active) || pct->expiry_active) |
Jason Low | 934715a | 2015-10-14 12:07:54 -0700 | [diff] [blame] | 922 | return; |
| 923 | |
Thomas Gleixner | a2ed4fd | 2019-08-29 12:52:28 +0200 | [diff] [blame] | 924 | /* |
Jason Low | c8d75aa | 2015-10-14 12:07:56 -0700 | [diff] [blame] | 925 | * Signify that a thread is checking for process timers. |
| 926 | * Write access to this field is protected by the sighand lock. |
| 927 | */ |
Thomas Gleixner | a2ed4fd | 2019-08-29 12:52:28 +0200 | [diff] [blame] | 928 | pct->expiry_active = true; |
Jason Low | c8d75aa | 2015-10-14 12:07:56 -0700 | [diff] [blame] | 929 | |
Jason Low | 934715a | 2015-10-14 12:07:54 -0700 | [diff] [blame] | 930 | /* |
Thomas Gleixner | a324956 | 2019-08-21 21:08:53 +0200 | [diff] [blame] | 931 | * Collect the current process totals. Group accounting is active |
| 932 | * so the sample can be taken directly. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 933 | */ |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 934 | proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples); |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 935 | collect_posix_cputimers(pct, samples, firing); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 936 | |
| 937 | /* |
| 938 | * Check for the special case process timers. |
| 939 | */ |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 940 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], |
| 941 | &pct->bases[CPUCLOCK_PROF].nextevt, |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 942 | samples[CPUCLOCK_PROF], SIGPROF); |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 943 | check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], |
| 944 | &pct->bases[CPUCLOCK_VIRT].nextevt, |
| 945 | samples[CPUCLOCK_VIRT], SIGVTALRM); |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 946 | |
Krzysztof Opasiak | 3cf2949 | 2017-07-05 19:25:48 +0200 | [diff] [blame] | 947 | soft = task_rlimit(tsk, RLIMIT_CPU); |
Jiri Slaby | d4bb5274 | 2010-03-05 13:42:53 -0800 | [diff] [blame] | 948 | if (soft != RLIM_INFINITY) { |
Thomas Gleixner | 8ea1de9 | 2019-08-21 21:09:21 +0200 | [diff] [blame] | 949 | /* RLIMIT_CPU is in seconds. Samples are nanoseconds */ |
Krzysztof Opasiak | 3cf2949 | 2017-07-05 19:25:48 +0200 | [diff] [blame] | 950 | unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU); |
Thomas Gleixner | 8ea1de9 | 2019-08-21 21:09:21 +0200 | [diff] [blame] | 951 | u64 ptime = samples[CPUCLOCK_PROF]; |
| 952 | u64 softns = (u64)soft * NSEC_PER_SEC; |
| 953 | u64 hardns = (u64)hard * NSEC_PER_SEC; |
Thomas Gleixner | b7be4ef | 2019-08-21 21:09:16 +0200 | [diff] [blame] | 954 | |
Thomas Gleixner | 8991afe | 2019-08-21 21:09:23 +0200 | [diff] [blame] | 955 | /* At the hard limit, send SIGKILL. No further action. */ |
| 956 | if (hard != RLIM_INFINITY && |
| 957 | check_rlimit(ptime, hardns, SIGKILL, false, true)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 958 | return; |
Thomas Gleixner | dd67022 | 2019-08-21 21:09:22 +0200 | [diff] [blame] | 959 | |
Thomas Gleixner | 8991afe | 2019-08-21 21:09:23 +0200 | [diff] [blame] | 960 | /* At the soft limit, send a SIGXCPU every second */ |
| 961 | if (check_rlimit(ptime, softns, SIGXCPU, false, false)) { |
Thomas Gleixner | dd67022 | 2019-08-21 21:09:22 +0200 | [diff] [blame] | 962 | sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1; |
| 963 | softns += NSEC_PER_SEC; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 964 | } |
Thomas Gleixner | 8ea1de9 | 2019-08-21 21:09:21 +0200 | [diff] [blame] | 965 | |
| 966 | /* Update the expiry cache */ |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 967 | if (softns < pct->bases[CPUCLOCK_PROF].nextevt) |
| 968 | pct->bases[CPUCLOCK_PROF].nextevt = softns; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 969 | } |
| 970 | |
Thomas Gleixner | 1cd07c0b | 2019-08-21 21:09:20 +0200 | [diff] [blame] | 971 | if (expiry_cache_is_inactive(pct)) |
Stanislaw Gruszka | 29f87b7 | 2010-04-27 14:12:15 -0700 | [diff] [blame] | 972 | stop_process_timers(sig); |
Jason Low | c8d75aa | 2015-10-14 12:07:56 -0700 | [diff] [blame] | 973 | |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 974 | pct->expiry_active = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 975 | } |
| 976 | |
| 977 | /* |
Thomas Gleixner | 96fe3b0 | 2017-05-30 23:15:46 +0200 | [diff] [blame] | 978 | * This is called from the signal code (via posixtimer_rearm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 979 | * when the last timer signal was delivered and we have to reload the timer. |
| 980 | */ |
Thomas Gleixner | f37fb0a | 2017-05-30 23:15:47 +0200 | [diff] [blame] | 981 | static void posix_cpu_timer_rearm(struct k_itimer *timer) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 982 | { |
Thomas Gleixner | da020ce | 2019-08-21 21:08:58 +0200 | [diff] [blame] | 983 | clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock); |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 984 | struct task_struct *p; |
Frederic Weisbecker | e73d84e | 2013-10-11 18:56:49 +0200 | [diff] [blame] | 985 | struct sighand_struct *sighand; |
| 986 | unsigned long flags; |
Frederic Weisbecker | ebd7e7f | 2017-01-31 04:09:34 +0100 | [diff] [blame] | 987 | u64 now; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 988 | |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 989 | rcu_read_lock(); |
| 990 | p = cpu_timer_task_rcu(timer); |
| 991 | if (!p) |
| 992 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 993 | |
| 994 | /* |
| 995 | * Fetch the current sample and update the timer's expiry time. |
| 996 | */ |
Eric W. Biederman | 60f2cea | 2020-02-28 11:09:19 -0600 | [diff] [blame] | 997 | if (CPUCLOCK_PERTHREAD(timer->it_clock)) |
Thomas Gleixner | 8c2d74f | 2019-08-21 21:09:01 +0200 | [diff] [blame] | 998 | now = cpu_clock_sample(clkid, p); |
Eric W. Biederman | 60f2cea | 2020-02-28 11:09:19 -0600 | [diff] [blame] | 999 | else |
Thomas Gleixner | 8c2d74f | 2019-08-21 21:09:01 +0200 | [diff] [blame] | 1000 | now = cpu_clock_sample_group(clkid, p, true); |
Eric W. Biederman | 60f2cea | 2020-02-28 11:09:19 -0600 | [diff] [blame] | 1001 | |
| 1002 | bump_cpu_timer(timer, now); |
| 1003 | |
| 1004 | /* Protect timer list r/w in arm_timer() */ |
| 1005 | sighand = lock_task_sighand(p, &flags); |
| 1006 | if (unlikely(sighand == NULL)) |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 1007 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1008 | |
| 1009 | /* |
| 1010 | * Now re-arm for the new expiry time. |
| 1011 | */ |
Eric W. Biederman | beb41d9 | 2020-02-28 11:09:46 -0600 | [diff] [blame] | 1012 | arm_timer(timer, p); |
Frederic Weisbecker | e73d84e | 2013-10-11 18:56:49 +0200 | [diff] [blame] | 1013 | unlock_task_sighand(p, &flags); |
Eric W. Biederman | 55e8c8e | 2020-02-28 11:11:06 -0600 | [diff] [blame] | 1014 | out: |
| 1015 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1016 | } |
| 1017 | |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1018 | /** |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 1019 | * task_cputimers_expired - Check whether posix CPU timers are expired |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1020 | * |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1021 | * @samples: Array of current samples for the CPUCLOCK clocks |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 1022 | * @pct: Pointer to a posix_cputimers container |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1023 | * |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 1024 | * Returns true if any member of @samples is greater than the corresponding |
| 1025 | * member of @pct->bases[CLK].nextevt. False otherwise |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1026 | */ |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 1027 | static inline bool |
Yi Wang | 7f2cbcb | 2019-10-21 15:44:12 +0800 | [diff] [blame] | 1028 | task_cputimers_expired(const u64 *samples, struct posix_cputimers *pct) |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1029 | { |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1030 | int i; |
| 1031 | |
| 1032 | for (i = 0; i < CPUCLOCK_MAX; i++) { |
Yi Wang | 7f2cbcb | 2019-10-21 15:44:12 +0800 | [diff] [blame] | 1033 | if (samples[i] >= pct->bases[i].nextevt) |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1034 | return true; |
| 1035 | } |
| 1036 | return false; |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1037 | } |
| 1038 | |
| 1039 | /** |
| 1040 | * fastpath_timer_check - POSIX CPU timers fast path. |
| 1041 | * |
| 1042 | * @tsk: The task (thread) being checked. |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1043 | * |
Frank Mayhar | bb34d92 | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1044 | * Check the task and thread group timers. If both are zero (there are no |
| 1045 | * timers set) return false. Otherwise snapshot the task and thread group |
| 1046 | * timers and compare them with the corresponding expiration times. Return |
| 1047 | * true if a timer has expired, else return false. |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1048 | */ |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1049 | static inline bool fastpath_timer_check(struct task_struct *tsk) |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1050 | { |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 1051 | struct posix_cputimers *pct = &tsk->posix_cputimers; |
Oleg Nesterov | ad133ba | 2008-11-17 15:39:47 +0100 | [diff] [blame] | 1052 | struct signal_struct *sig; |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1053 | |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 1054 | if (!expiry_cache_is_inactive(pct)) { |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1055 | u64 samples[CPUCLOCK_MAX]; |
Frank Mayhar | bb34d92 | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1056 | |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1057 | task_sample_cputime(tsk, samples); |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 1058 | if (task_cputimers_expired(samples, pct)) |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1059 | return true; |
Frank Mayhar | bb34d92 | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1060 | } |
Oleg Nesterov | ad133ba | 2008-11-17 15:39:47 +0100 | [diff] [blame] | 1061 | |
| 1062 | sig = tsk->signal; |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 1063 | pct = &sig->posix_cputimers; |
Jason Low | c8d75aa | 2015-10-14 12:07:56 -0700 | [diff] [blame] | 1064 | /* |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 1065 | * Check if thread group timers expired when timers are active and |
| 1066 | * no other thread in the group is already handling expiry for |
| 1067 | * thread group cputimers. These fields are read without the |
| 1068 | * sighand lock. However, this is fine because this is meant to be |
| 1069 | * a fastpath heuristic to determine whether we should try to |
| 1070 | * acquire the sighand lock to handle timer expiry. |
Jason Low | c8d75aa | 2015-10-14 12:07:56 -0700 | [diff] [blame] | 1071 | * |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 1072 | * In the worst case scenario, if concurrently timers_active is set |
| 1073 | * or expiry_active is cleared, but the current thread doesn't see |
| 1074 | * the change yet, the timer checks are delayed until the next |
| 1075 | * thread in the group gets a scheduler interrupt to handle the |
| 1076 | * timer. This isn't an issue in practice because these types of |
| 1077 | * delays with signals actually getting sent are expected. |
Jason Low | c8d75aa | 2015-10-14 12:07:56 -0700 | [diff] [blame] | 1078 | */ |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 1079 | if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) { |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1080 | u64 samples[CPUCLOCK_MAX]; |
Frank Mayhar | bb34d92 | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1081 | |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1082 | proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, |
| 1083 | samples); |
Oleg Nesterov | 8d1f431 | 2010-06-11 20:04:46 +0200 | [diff] [blame] | 1084 | |
Thomas Gleixner | 244d49e | 2019-08-21 21:09:24 +0200 | [diff] [blame] | 1085 | if (task_cputimers_expired(samples, pct)) |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1086 | return true; |
Frank Mayhar | bb34d92 | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1087 | } |
Oleg Nesterov | 37bebc7 | 2009-03-23 20:34:11 +0100 | [diff] [blame] | 1088 | |
Juri Lelli | 34be393 | 2017-12-12 12:10:24 +0100 | [diff] [blame] | 1089 | if (dl_task(tsk) && tsk->dl.dl_overrun) |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1090 | return true; |
Juri Lelli | 34be393 | 2017-12-12 12:10:24 +0100 | [diff] [blame] | 1091 | |
Thomas Gleixner | 001f797 | 2019-08-21 21:09:13 +0200 | [diff] [blame] | 1092 | return false; |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1093 | } |
| 1094 | |
Thomas Gleixner | 1fb497d | 2020-07-30 12:14:06 +0200 | [diff] [blame] | 1095 | static void handle_posix_cpu_timers(struct task_struct *tsk); |
| 1096 | |
| 1097 | #ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK |
| 1098 | static void posix_cpu_timers_work(struct callback_head *work) |
| 1099 | { |
| 1100 | handle_posix_cpu_timers(current); |
| 1101 | } |
| 1102 | |
| 1103 | /* |
| 1104 | * Initialize posix CPU timers task work in init task. Out of line to |
| 1105 | * keep the callback static and to avoid header recursion hell. |
| 1106 | */ |
| 1107 | void __init posix_cputimers_init_work(void) |
| 1108 | { |
| 1109 | init_task_work(¤t->posix_cputimers_work.work, |
| 1110 | posix_cpu_timers_work); |
| 1111 | } |
| 1112 | |
| 1113 | /* |
| 1114 | * Note: All operations on tsk->posix_cputimer_work.scheduled happen either |
| 1115 | * in hard interrupt context or in task context with interrupts |
| 1116 | * disabled. Aside of that the writer/reader interaction is always in the |
| 1117 | * context of the current task, which means they are strict per CPU. |
| 1118 | */ |
| 1119 | static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk) |
| 1120 | { |
| 1121 | return tsk->posix_cputimers_work.scheduled; |
| 1122 | } |
| 1123 | |
| 1124 | static inline void __run_posix_cpu_timers(struct task_struct *tsk) |
| 1125 | { |
| 1126 | if (WARN_ON_ONCE(tsk->posix_cputimers_work.scheduled)) |
| 1127 | return; |
| 1128 | |
| 1129 | /* Schedule task work to actually expire the timers */ |
| 1130 | tsk->posix_cputimers_work.scheduled = true; |
| 1131 | task_work_add(tsk, &tsk->posix_cputimers_work.work, TWA_RESUME); |
| 1132 | } |
| 1133 | |
| 1134 | static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk, |
| 1135 | unsigned long start) |
| 1136 | { |
| 1137 | bool ret = true; |
| 1138 | |
| 1139 | /* |
| 1140 | * On !RT kernels interrupts are disabled while collecting expired |
| 1141 | * timers, so no tick can happen and the fast path check can be |
| 1142 | * reenabled without further checks. |
| 1143 | */ |
| 1144 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { |
| 1145 | tsk->posix_cputimers_work.scheduled = false; |
| 1146 | return true; |
| 1147 | } |
| 1148 | |
| 1149 | /* |
| 1150 | * On RT enabled kernels ticks can happen while the expired timers |
| 1151 | * are collected under sighand lock. But any tick which observes |
| 1152 | * the CPUTIMERS_WORK_SCHEDULED bit set, does not run the fastpath |
| 1153 | * checks. So reenabling the tick work has do be done carefully: |
| 1154 | * |
| 1155 | * Disable interrupts and run the fast path check if jiffies have |
| 1156 | * advanced since the collecting of expired timers started. If |
| 1157 | * jiffies have not advanced or the fast path check did not find |
| 1158 | * newly expired timers, reenable the fast path check in the timer |
| 1159 | * interrupt. If there are newly expired timers, return false and |
| 1160 | * let the collection loop repeat. |
| 1161 | */ |
| 1162 | local_irq_disable(); |
| 1163 | if (start != jiffies && fastpath_timer_check(tsk)) |
| 1164 | ret = false; |
| 1165 | else |
| 1166 | tsk->posix_cputimers_work.scheduled = false; |
| 1167 | local_irq_enable(); |
| 1168 | |
| 1169 | return ret; |
| 1170 | } |
| 1171 | #else /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */ |
| 1172 | static inline void __run_posix_cpu_timers(struct task_struct *tsk) |
| 1173 | { |
| 1174 | lockdep_posixtimer_enter(); |
| 1175 | handle_posix_cpu_timers(tsk); |
| 1176 | lockdep_posixtimer_exit(); |
| 1177 | } |
| 1178 | |
| 1179 | static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk) |
| 1180 | { |
| 1181 | return false; |
| 1182 | } |
| 1183 | |
| 1184 | static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk, |
| 1185 | unsigned long start) |
| 1186 | { |
| 1187 | return true; |
| 1188 | } |
| 1189 | #endif /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */ |
| 1190 | |
| 1191 | static void handle_posix_cpu_timers(struct task_struct *tsk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1192 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1193 | struct k_itimer *timer, *next; |
Thomas Gleixner | 1fb497d | 2020-07-30 12:14:06 +0200 | [diff] [blame] | 1194 | unsigned long flags, start; |
Thomas Gleixner | dce3e8f | 2019-08-19 16:31:47 +0200 | [diff] [blame] | 1195 | LIST_HEAD(firing); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1196 | |
Thomas Gleixner | 820903c7 | 2020-07-30 12:14:05 +0200 | [diff] [blame] | 1197 | if (!lock_task_sighand(tsk, &flags)) |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1198 | return; |
Ingo Molnar | 5ce73a4 | 2008-09-14 17:11:46 +0200 | [diff] [blame] | 1199 | |
Thomas Gleixner | 1fb497d | 2020-07-30 12:14:06 +0200 | [diff] [blame] | 1200 | do { |
| 1201 | /* |
| 1202 | * On RT locking sighand lock does not disable interrupts, |
| 1203 | * so this needs to be careful vs. ticks. Store the current |
| 1204 | * jiffies value. |
| 1205 | */ |
| 1206 | start = READ_ONCE(jiffies); |
| 1207 | barrier(); |
Jason Low | 934715a | 2015-10-14 12:07:54 -0700 | [diff] [blame] | 1208 | |
Thomas Gleixner | 1fb497d | 2020-07-30 12:14:06 +0200 | [diff] [blame] | 1209 | /* |
| 1210 | * Here we take off tsk->signal->cpu_timers[N] and |
| 1211 | * tsk->cpu_timers[N] all the timers that are firing, and |
| 1212 | * put them on the firing list. |
| 1213 | */ |
| 1214 | check_thread_timers(tsk, &firing); |
| 1215 | |
| 1216 | check_process_timers(tsk, &firing); |
| 1217 | |
| 1218 | /* |
| 1219 | * The above timer checks have updated the exipry cache and |
| 1220 | * because nothing can have queued or modified timers after |
| 1221 | * sighand lock was taken above it is guaranteed to be |
| 1222 | * consistent. So the next timer interrupt fastpath check |
| 1223 | * will find valid data. |
| 1224 | * |
| 1225 | * If timer expiry runs in the timer interrupt context then |
| 1226 | * the loop is not relevant as timers will be directly |
| 1227 | * expired in interrupt context. The stub function below |
| 1228 | * returns always true which allows the compiler to |
| 1229 | * optimize the loop out. |
| 1230 | * |
| 1231 | * If timer expiry is deferred to task work context then |
| 1232 | * the following rules apply: |
| 1233 | * |
| 1234 | * - On !RT kernels no tick can have happened on this CPU |
| 1235 | * after sighand lock was acquired because interrupts are |
| 1236 | * disabled. So reenabling task work before dropping |
| 1237 | * sighand lock and reenabling interrupts is race free. |
| 1238 | * |
| 1239 | * - On RT kernels ticks might have happened but the tick |
| 1240 | * work ignored posix CPU timer handling because the |
| 1241 | * CPUTIMERS_WORK_SCHEDULED bit is set. Reenabling work |
| 1242 | * must be done very carefully including a check whether |
| 1243 | * ticks have happened since the start of the timer |
| 1244 | * expiry checks. posix_cpu_timers_enable_work() takes |
| 1245 | * care of that and eventually lets the expiry checks |
| 1246 | * run again. |
| 1247 | */ |
| 1248 | } while (!posix_cpu_timers_enable_work(tsk, start)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1249 | |
Frank Mayhar | bb34d92 | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1250 | /* |
Thomas Gleixner | 1fb497d | 2020-07-30 12:14:06 +0200 | [diff] [blame] | 1251 | * We must release sighand lock before taking any timer's lock. |
Frank Mayhar | bb34d92 | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1252 | * There is a potential race with timer deletion here, as the |
| 1253 | * siglock now protects our private firing list. We have set |
| 1254 | * the firing flag in each timer, so that a deletion attempt |
| 1255 | * that gets the timer lock before we do will give it up and |
| 1256 | * spin until we've taken care of that timer below. |
| 1257 | */ |
Oleg Nesterov | 0bdd2ed | 2010-06-11 01:10:18 +0200 | [diff] [blame] | 1258 | unlock_task_sighand(tsk, &flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1259 | |
| 1260 | /* |
| 1261 | * Now that all the timers on our list have the firing flag, |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1262 | * no one will touch their list entries but us. We'll take |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1263 | * each timer's lock before clearing its firing flag, so no |
| 1264 | * timer call will interfere. |
| 1265 | */ |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 1266 | list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) { |
H Hartley Sweeten | 6e85c5b | 2009-04-29 19:14:32 -0400 | [diff] [blame] | 1267 | int cpu_firing; |
| 1268 | |
Thomas Gleixner | 1fb497d | 2020-07-30 12:14:06 +0200 | [diff] [blame] | 1269 | /* |
| 1270 | * spin_lock() is sufficient here even independent of the |
| 1271 | * expiry context. If expiry happens in hard interrupt |
| 1272 | * context it's obvious. For task work context it's safe |
| 1273 | * because all other operations on timer::it_lock happen in |
| 1274 | * task context (syscall or exit). |
| 1275 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1276 | spin_lock(&timer->it_lock); |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 1277 | list_del_init(&timer->it.cpu.elist); |
H Hartley Sweeten | 6e85c5b | 2009-04-29 19:14:32 -0400 | [diff] [blame] | 1278 | cpu_firing = timer->it.cpu.firing; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1279 | timer->it.cpu.firing = 0; |
| 1280 | /* |
| 1281 | * The firing flag is -1 if we collided with a reset |
| 1282 | * of the timer, which already reported this |
| 1283 | * almost-firing as an overrun. So don't generate an event. |
| 1284 | */ |
H Hartley Sweeten | 6e85c5b | 2009-04-29 19:14:32 -0400 | [diff] [blame] | 1285 | if (likely(cpu_firing >= 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1286 | cpu_timer_fire(timer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1287 | spin_unlock(&timer->it_lock); |
| 1288 | } |
Thomas Gleixner | 820903c7 | 2020-07-30 12:14:05 +0200 | [diff] [blame] | 1289 | } |
| 1290 | |
| 1291 | /* |
| 1292 | * This is called from the timer interrupt handler. The irq handler has |
| 1293 | * already updated our counts. We need to check if any timers fire now. |
| 1294 | * Interrupts are disabled. |
| 1295 | */ |
| 1296 | void run_posix_cpu_timers(void) |
| 1297 | { |
| 1298 | struct task_struct *tsk = current; |
| 1299 | |
| 1300 | lockdep_assert_irqs_disabled(); |
| 1301 | |
| 1302 | /* |
Thomas Gleixner | 1fb497d | 2020-07-30 12:14:06 +0200 | [diff] [blame] | 1303 | * If the actual expiry is deferred to task work context and the |
| 1304 | * work is already scheduled there is no point to do anything here. |
| 1305 | */ |
| 1306 | if (posix_cpu_timers_work_scheduled(tsk)) |
| 1307 | return; |
| 1308 | |
| 1309 | /* |
Thomas Gleixner | 820903c7 | 2020-07-30 12:14:05 +0200 | [diff] [blame] | 1310 | * The fast path checks that there are no expired thread or thread |
| 1311 | * group timers. If that's so, just return. |
| 1312 | */ |
| 1313 | if (!fastpath_timer_check(tsk)) |
| 1314 | return; |
| 1315 | |
Thomas Gleixner | 820903c7 | 2020-07-30 12:14:05 +0200 | [diff] [blame] | 1316 | __run_posix_cpu_timers(tsk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1317 | } |
| 1318 | |
| 1319 | /* |
Stanislaw Gruszka | f55db60 | 2010-03-11 14:04:37 -0800 | [diff] [blame] | 1320 | * Set one of the process-wide special case CPU timers or RLIMIT_CPU. |
Frank Mayhar | f06febc | 2008-09-12 09:54:39 -0700 | [diff] [blame] | 1321 | * The tsk->sighand->siglock must be held by the caller. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1322 | */ |
Thomas Gleixner | 1b0dd96 | 2019-08-21 21:09:09 +0200 | [diff] [blame] | 1323 | void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid, |
Frederic Weisbecker | 858cf3a | 2017-01-31 04:09:35 +0100 | [diff] [blame] | 1324 | u64 *newval, u64 *oldval) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1325 | { |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 1326 | u64 now, *nextevt; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1327 | |
Thomas Gleixner | 1b0dd96 | 2019-08-21 21:09:09 +0200 | [diff] [blame] | 1328 | if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED)) |
Thomas Gleixner | 692117c | 2019-08-19 16:31:46 +0200 | [diff] [blame] | 1329 | return; |
| 1330 | |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 1331 | nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt; |
Thomas Gleixner | 1b0dd96 | 2019-08-21 21:09:09 +0200 | [diff] [blame] | 1332 | now = cpu_clock_sample_group(clkid, tsk, true); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1333 | |
Thomas Gleixner | 5405d00 | 2019-08-21 21:08:59 +0200 | [diff] [blame] | 1334 | if (oldval) { |
Stanislaw Gruszka | f55db60 | 2010-03-11 14:04:37 -0800 | [diff] [blame] | 1335 | /* |
| 1336 | * We are setting itimer. The *oldval is absolute and we update |
| 1337 | * it to be relative, *newval argument is relative and we update |
| 1338 | * it to be absolute. |
| 1339 | */ |
Martin Schwidefsky | 6486163 | 2011-12-15 14:56:09 +0100 | [diff] [blame] | 1340 | if (*oldval) { |
Frederic Weisbecker | 858cf3a | 2017-01-31 04:09:35 +0100 | [diff] [blame] | 1341 | if (*oldval <= now) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1342 | /* Just about to fire. */ |
Frederic Weisbecker | 858cf3a | 2017-01-31 04:09:35 +0100 | [diff] [blame] | 1343 | *oldval = TICK_NSEC; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1344 | } else { |
Frederic Weisbecker | 858cf3a | 2017-01-31 04:09:35 +0100 | [diff] [blame] | 1345 | *oldval -= now; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1346 | } |
| 1347 | } |
| 1348 | |
Martin Schwidefsky | 6486163 | 2011-12-15 14:56:09 +0100 | [diff] [blame] | 1349 | if (!*newval) |
Frederic Weisbecker | b787830 | 2015-07-17 22:25:49 +0200 | [diff] [blame] | 1350 | return; |
Frederic Weisbecker | 858cf3a | 2017-01-31 04:09:35 +0100 | [diff] [blame] | 1351 | *newval += now; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1352 | } |
| 1353 | |
| 1354 | /* |
Thomas Gleixner | 1b0dd96 | 2019-08-21 21:09:09 +0200 | [diff] [blame] | 1355 | * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF |
| 1356 | * expiry cache is also used by RLIMIT_CPU!. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | */ |
Thomas Gleixner | 2bbdbda | 2019-08-21 21:09:19 +0200 | [diff] [blame] | 1358 | if (*newval < *nextevt) |
Thomas Gleixner | 87dc644 | 2019-08-26 20:22:24 +0200 | [diff] [blame] | 1359 | *nextevt = *newval; |
Frederic Weisbecker | b787830 | 2015-07-17 22:25:49 +0200 | [diff] [blame] | 1360 | |
| 1361 | tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1362 | } |
| 1363 | |
Toyo Abe | e4b7655 | 2006-09-29 02:00:29 -0700 | [diff] [blame] | 1364 | static int do_cpu_nanosleep(const clockid_t which_clock, int flags, |
Thomas Gleixner | 343d8fc | 2017-06-13 23:29:14 +0200 | [diff] [blame] | 1365 | const struct timespec64 *rqtp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1366 | { |
Al Viro | 86a9c44 | 2017-06-07 09:42:26 +0100 | [diff] [blame] | 1367 | struct itimerspec64 it; |
Thomas Gleixner | 343d8fc | 2017-06-13 23:29:14 +0200 | [diff] [blame] | 1368 | struct k_itimer timer; |
| 1369 | u64 expires; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1370 | int error; |
| 1371 | |
| 1372 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1373 | * Set up a temporary timer and then wait for it to go off. |
| 1374 | */ |
| 1375 | memset(&timer, 0, sizeof timer); |
| 1376 | spin_lock_init(&timer.it_lock); |
| 1377 | timer.it_clock = which_clock; |
| 1378 | timer.it_overrun = -1; |
| 1379 | error = posix_cpu_timer_create(&timer); |
| 1380 | timer.it_process = current; |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 1381 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1382 | if (!error) { |
Deepa Dinamani | 5f252b3 | 2017-03-26 12:04:17 -0700 | [diff] [blame] | 1383 | static struct itimerspec64 zero_it; |
Al Viro | edbeda4 | 2017-06-07 09:42:31 +0100 | [diff] [blame] | 1384 | struct restart_block *restart; |
Toyo Abe | e4b7655 | 2006-09-29 02:00:29 -0700 | [diff] [blame] | 1385 | |
Al Viro | edbeda4 | 2017-06-07 09:42:31 +0100 | [diff] [blame] | 1386 | memset(&it, 0, sizeof(it)); |
Al Viro | 86a9c44 | 2017-06-07 09:42:26 +0100 | [diff] [blame] | 1387 | it.it_value = *rqtp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1388 | |
| 1389 | spin_lock_irq(&timer.it_lock); |
Al Viro | 86a9c44 | 2017-06-07 09:42:26 +0100 | [diff] [blame] | 1390 | error = posix_cpu_timer_set(&timer, flags, &it, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1391 | if (error) { |
| 1392 | spin_unlock_irq(&timer.it_lock); |
| 1393 | return error; |
| 1394 | } |
| 1395 | |
| 1396 | while (!signal_pending(current)) { |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 1397 | if (!cpu_timer_getexpires(&timer.it.cpu)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1398 | /* |
Stanislaw Gruszka | e6c42c2 | 2013-02-15 11:08:11 +0100 | [diff] [blame] | 1399 | * Our timer fired and was reset, below |
| 1400 | * deletion can not fail. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1401 | */ |
Stanislaw Gruszka | e6c42c2 | 2013-02-15 11:08:11 +0100 | [diff] [blame] | 1402 | posix_cpu_timer_del(&timer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1403 | spin_unlock_irq(&timer.it_lock); |
| 1404 | return 0; |
| 1405 | } |
| 1406 | |
| 1407 | /* |
| 1408 | * Block until cpu_timer_fire (or a signal) wakes us. |
| 1409 | */ |
| 1410 | __set_current_state(TASK_INTERRUPTIBLE); |
| 1411 | spin_unlock_irq(&timer.it_lock); |
| 1412 | schedule(); |
| 1413 | spin_lock_irq(&timer.it_lock); |
| 1414 | } |
| 1415 | |
| 1416 | /* |
| 1417 | * We were interrupted by a signal. |
| 1418 | */ |
Thomas Gleixner | 60bda03 | 2019-08-27 21:31:02 +0200 | [diff] [blame] | 1419 | expires = cpu_timer_getexpires(&timer.it.cpu); |
Al Viro | 86a9c44 | 2017-06-07 09:42:26 +0100 | [diff] [blame] | 1420 | error = posix_cpu_timer_set(&timer, 0, &zero_it, &it); |
Stanislaw Gruszka | e6c42c2 | 2013-02-15 11:08:11 +0100 | [diff] [blame] | 1421 | if (!error) { |
| 1422 | /* |
| 1423 | * Timer is now unarmed, deletion can not fail. |
| 1424 | */ |
| 1425 | posix_cpu_timer_del(&timer); |
| 1426 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1427 | spin_unlock_irq(&timer.it_lock); |
| 1428 | |
Stanislaw Gruszka | e6c42c2 | 2013-02-15 11:08:11 +0100 | [diff] [blame] | 1429 | while (error == TIMER_RETRY) { |
| 1430 | /* |
| 1431 | * We need to handle case when timer was or is in the |
| 1432 | * middle of firing. In other cases we already freed |
| 1433 | * resources. |
| 1434 | */ |
| 1435 | spin_lock_irq(&timer.it_lock); |
| 1436 | error = posix_cpu_timer_del(&timer); |
| 1437 | spin_unlock_irq(&timer.it_lock); |
| 1438 | } |
| 1439 | |
Al Viro | 86a9c44 | 2017-06-07 09:42:26 +0100 | [diff] [blame] | 1440 | if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1441 | /* |
| 1442 | * It actually did fire already. |
| 1443 | */ |
| 1444 | return 0; |
| 1445 | } |
| 1446 | |
Toyo Abe | e4b7655 | 2006-09-29 02:00:29 -0700 | [diff] [blame] | 1447 | error = -ERESTART_RESTARTBLOCK; |
Al Viro | 86a9c44 | 2017-06-07 09:42:26 +0100 | [diff] [blame] | 1448 | /* |
| 1449 | * Report back to the user the time still remaining. |
| 1450 | */ |
Al Viro | edbeda4 | 2017-06-07 09:42:31 +0100 | [diff] [blame] | 1451 | restart = ¤t->restart_block; |
Thomas Gleixner | 343d8fc | 2017-06-13 23:29:14 +0200 | [diff] [blame] | 1452 | restart->nanosleep.expires = expires; |
Deepa Dinamani | c0edd7c | 2017-06-24 11:45:06 -0700 | [diff] [blame] | 1453 | if (restart->nanosleep.type != TT_NONE) |
| 1454 | error = nanosleep_copyout(restart, &it.it_value); |
Toyo Abe | e4b7655 | 2006-09-29 02:00:29 -0700 | [diff] [blame] | 1455 | } |
| 1456 | |
| 1457 | return error; |
| 1458 | } |
| 1459 | |
Thomas Gleixner | bc2c8ea | 2011-02-01 13:52:12 +0000 | [diff] [blame] | 1460 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block); |
| 1461 | |
| 1462 | static int posix_cpu_nsleep(const clockid_t which_clock, int flags, |
Thomas Gleixner | 938e7cf | 2017-06-13 23:34:33 +0200 | [diff] [blame] | 1463 | const struct timespec64 *rqtp) |
Toyo Abe | e4b7655 | 2006-09-29 02:00:29 -0700 | [diff] [blame] | 1464 | { |
Andy Lutomirski | f56141e | 2015-02-12 15:01:14 -0800 | [diff] [blame] | 1465 | struct restart_block *restart_block = ¤t->restart_block; |
Toyo Abe | e4b7655 | 2006-09-29 02:00:29 -0700 | [diff] [blame] | 1466 | int error; |
| 1467 | |
| 1468 | /* |
| 1469 | * Diagnose required errors first. |
| 1470 | */ |
| 1471 | if (CPUCLOCK_PERTHREAD(which_clock) && |
| 1472 | (CPUCLOCK_PID(which_clock) == 0 || |
Eric W. Biederman | 01a2197 | 2017-04-13 10:32:16 -0500 | [diff] [blame] | 1473 | CPUCLOCK_PID(which_clock) == task_pid_vnr(current))) |
Toyo Abe | e4b7655 | 2006-09-29 02:00:29 -0700 | [diff] [blame] | 1474 | return -EINVAL; |
| 1475 | |
Al Viro | 86a9c44 | 2017-06-07 09:42:26 +0100 | [diff] [blame] | 1476 | error = do_cpu_nanosleep(which_clock, flags, rqtp); |
Toyo Abe | e4b7655 | 2006-09-29 02:00:29 -0700 | [diff] [blame] | 1477 | |
| 1478 | if (error == -ERESTART_RESTARTBLOCK) { |
| 1479 | |
Thomas Gleixner | 3751f9f | 2011-02-01 13:51:20 +0000 | [diff] [blame] | 1480 | if (flags & TIMER_ABSTIME) |
Toyo Abe | e4b7655 | 2006-09-29 02:00:29 -0700 | [diff] [blame] | 1481 | return -ERESTARTNOHAND; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1482 | |
Toyo Abe | 1711ef3 | 2006-09-29 02:00:28 -0700 | [diff] [blame] | 1483 | restart_block->fn = posix_cpu_nsleep_restart; |
Thomas Gleixner | ab8177b | 2011-05-20 13:05:15 +0200 | [diff] [blame] | 1484 | restart_block->nanosleep.clockid = which_clock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1485 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1486 | return error; |
| 1487 | } |
| 1488 | |
Thomas Gleixner | bc2c8ea | 2011-02-01 13:52:12 +0000 | [diff] [blame] | 1489 | static long posix_cpu_nsleep_restart(struct restart_block *restart_block) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1490 | { |
Thomas Gleixner | ab8177b | 2011-05-20 13:05:15 +0200 | [diff] [blame] | 1491 | clockid_t which_clock = restart_block->nanosleep.clockid; |
Deepa Dinamani | ad19638 | 2017-03-26 12:04:18 -0700 | [diff] [blame] | 1492 | struct timespec64 t; |
Thomas Gleixner | 97735f2 | 2006-01-09 20:52:37 -0800 | [diff] [blame] | 1493 | |
Deepa Dinamani | ad19638 | 2017-03-26 12:04:18 -0700 | [diff] [blame] | 1494 | t = ns_to_timespec64(restart_block->nanosleep.expires); |
Thomas Gleixner | 97735f2 | 2006-01-09 20:52:37 -0800 | [diff] [blame] | 1495 | |
Al Viro | 86a9c44 | 2017-06-07 09:42:26 +0100 | [diff] [blame] | 1496 | return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1497 | } |
| 1498 | |
Nick Desaulniers | 29f1b2b | 2017-12-28 22:11:36 -0500 | [diff] [blame] | 1499 | #define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED) |
| 1500 | #define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1501 | |
Thomas Gleixner | a924b04 | 2006-01-09 20:52:27 -0800 | [diff] [blame] | 1502 | static int process_cpu_clock_getres(const clockid_t which_clock, |
Deepa Dinamani | d2e3e0c | 2017-03-26 12:04:15 -0700 | [diff] [blame] | 1503 | struct timespec64 *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1504 | { |
| 1505 | return posix_cpu_clock_getres(PROCESS_CLOCK, tp); |
| 1506 | } |
Thomas Gleixner | a924b04 | 2006-01-09 20:52:27 -0800 | [diff] [blame] | 1507 | static int process_cpu_clock_get(const clockid_t which_clock, |
Deepa Dinamani | 3c9c12f | 2017-03-26 12:04:14 -0700 | [diff] [blame] | 1508 | struct timespec64 *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1509 | { |
| 1510 | return posix_cpu_clock_get(PROCESS_CLOCK, tp); |
| 1511 | } |
| 1512 | static int process_cpu_timer_create(struct k_itimer *timer) |
| 1513 | { |
| 1514 | timer->it_clock = PROCESS_CLOCK; |
| 1515 | return posix_cpu_timer_create(timer); |
| 1516 | } |
Thomas Gleixner | a924b04 | 2006-01-09 20:52:27 -0800 | [diff] [blame] | 1517 | static int process_cpu_nsleep(const clockid_t which_clock, int flags, |
Thomas Gleixner | 938e7cf | 2017-06-13 23:34:33 +0200 | [diff] [blame] | 1518 | const struct timespec64 *rqtp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1519 | { |
Al Viro | 99e6c0e | 2017-06-07 09:42:30 +0100 | [diff] [blame] | 1520 | return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1521 | } |
Thomas Gleixner | a924b04 | 2006-01-09 20:52:27 -0800 | [diff] [blame] | 1522 | static int thread_cpu_clock_getres(const clockid_t which_clock, |
Deepa Dinamani | d2e3e0c | 2017-03-26 12:04:15 -0700 | [diff] [blame] | 1523 | struct timespec64 *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1524 | { |
| 1525 | return posix_cpu_clock_getres(THREAD_CLOCK, tp); |
| 1526 | } |
Thomas Gleixner | a924b04 | 2006-01-09 20:52:27 -0800 | [diff] [blame] | 1527 | static int thread_cpu_clock_get(const clockid_t which_clock, |
Deepa Dinamani | 3c9c12f | 2017-03-26 12:04:14 -0700 | [diff] [blame] | 1528 | struct timespec64 *tp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1529 | { |
| 1530 | return posix_cpu_clock_get(THREAD_CLOCK, tp); |
| 1531 | } |
| 1532 | static int thread_cpu_timer_create(struct k_itimer *timer) |
| 1533 | { |
| 1534 | timer->it_clock = THREAD_CLOCK; |
| 1535 | return posix_cpu_timer_create(timer); |
| 1536 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1537 | |
Christoph Hellwig | d3ba5a9 | 2017-05-26 12:03:11 +0300 | [diff] [blame] | 1538 | const struct k_clock clock_posix_cpu = { |
Andrei Vagin | 819a95f | 2019-11-12 01:26:54 +0000 | [diff] [blame] | 1539 | .clock_getres = posix_cpu_clock_getres, |
| 1540 | .clock_set = posix_cpu_clock_set, |
| 1541 | .clock_get_timespec = posix_cpu_clock_get, |
| 1542 | .timer_create = posix_cpu_timer_create, |
| 1543 | .nsleep = posix_cpu_nsleep, |
| 1544 | .timer_set = posix_cpu_timer_set, |
| 1545 | .timer_del = posix_cpu_timer_del, |
| 1546 | .timer_get = posix_cpu_timer_get, |
| 1547 | .timer_rearm = posix_cpu_timer_rearm, |
Thomas Gleixner | 1976945 | 2011-02-01 13:51:06 +0000 | [diff] [blame] | 1548 | }; |
| 1549 | |
Christoph Hellwig | d3ba5a9 | 2017-05-26 12:03:11 +0300 | [diff] [blame] | 1550 | const struct k_clock clock_process = { |
Andrei Vagin | 819a95f | 2019-11-12 01:26:54 +0000 | [diff] [blame] | 1551 | .clock_getres = process_cpu_clock_getres, |
| 1552 | .clock_get_timespec = process_cpu_clock_get, |
| 1553 | .timer_create = process_cpu_timer_create, |
| 1554 | .nsleep = process_cpu_nsleep, |
Christoph Hellwig | d3ba5a9 | 2017-05-26 12:03:11 +0300 | [diff] [blame] | 1555 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1556 | |
Christoph Hellwig | d3ba5a9 | 2017-05-26 12:03:11 +0300 | [diff] [blame] | 1557 | const struct k_clock clock_thread = { |
Andrei Vagin | 819a95f | 2019-11-12 01:26:54 +0000 | [diff] [blame] | 1558 | .clock_getres = thread_cpu_clock_getres, |
| 1559 | .clock_get_timespec = thread_cpu_clock_get, |
| 1560 | .timer_create = thread_cpu_timer_create, |
Christoph Hellwig | d3ba5a9 | 2017-05-26 12:03:11 +0300 | [diff] [blame] | 1561 | }; |