blob: a71758e34e4564ad8db4d9081e0078287adab71d [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * Implement CPU time clocks for the POSIX clock interface.
4 */
5
Ingo Molnar3f07c012017-02-08 18:51:30 +01006#include <linux/sched/signal.h>
Ingo Molnar32ef5512017-02-05 11:48:36 +01007#include <linux/sched/cputime.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/posix-timers.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/errno.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070010#include <linux/math64.h>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080011#include <linux/uaccess.h>
Frank Mayharbb34d922008-09-12 09:54:39 -070012#include <linux/kernel_stat.h>
Xiao Guangrong3f0a5252009-08-10 10:52:30 +080013#include <trace/events/timer.h>
Frederic Weisbeckera8572162013-04-18 01:31:13 +020014#include <linux/tick.h>
15#include <linux/workqueue.h>
Al Viroedbeda42017-06-07 09:42:31 +010016#include <linux/compat.h>
Juri Lelli34be3932017-12-12 12:10:24 +010017#include <linux/sched/deadline.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018
Thomas Gleixnerbab0aae92017-05-30 23:15:41 +020019#include "posix-timers.h"
20
Thomas Gleixnerf37fb0a2017-05-30 23:15:47 +020021static void posix_cpu_timer_rearm(struct k_itimer *timer);
22
Thomas Gleixner3a245c02019-08-21 21:09:06 +020023void posix_cputimers_group_init(struct posix_cputimers *pct, u64 cpu_limit)
24{
25 posix_cputimers_init(pct);
Thomas Gleixner244d49e2019-08-21 21:09:24 +020026 if (cpu_limit != RLIM_INFINITY) {
Thomas Gleixner87dc6442019-08-26 20:22:24 +020027 pct->bases[CPUCLOCK_PROF].nextevt = cpu_limit * NSEC_PER_SEC;
Thomas Gleixner244d49e2019-08-21 21:09:24 +020028 pct->timers_active = true;
29 }
Thomas Gleixner3a245c02019-08-21 21:09:06 +020030}
31
Frank Mayharf06febc2008-09-12 09:54:39 -070032/*
Stanislaw Gruszkaf55db602010-03-11 14:04:37 -080033 * Called after updating RLIMIT_CPU to run cpu timer and update
Thomas Gleixner87dc6442019-08-26 20:22:24 +020034 * tsk->signal->posix_cputimers.bases[clock].nextevt expiration cache if
35 * necessary. Needs siglock protection since other code may update the
Thomas Gleixner3a245c02019-08-21 21:09:06 +020036 * expiration cache as well.
Frank Mayharf06febc2008-09-12 09:54:39 -070037 */
Jiri Slaby5ab46b32009-08-28 14:05:12 +020038void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
Frank Mayharf06febc2008-09-12 09:54:39 -070039{
Frederic Weisbecker858cf3a2017-01-31 04:09:35 +010040 u64 nsecs = rlim_new * NSEC_PER_SEC;
Frank Mayharf06febc2008-09-12 09:54:39 -070041
Jiri Slaby5ab46b32009-08-28 14:05:12 +020042 spin_lock_irq(&task->sighand->siglock);
Frederic Weisbecker858cf3a2017-01-31 04:09:35 +010043 set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL);
Jiri Slaby5ab46b32009-08-28 14:05:12 +020044 spin_unlock_irq(&task->sighand->siglock);
Frank Mayharf06febc2008-09-12 09:54:39 -070045}
46
Thomas Gleixner6ae40e32019-08-21 21:08:48 +020047/*
48 * Functions for validating access to tasks.
49 */
Eric W. Biederman96498772020-04-27 07:55:07 -050050static struct pid *pid_for_clock(const clockid_t clock, bool gettime)
Linus Torvalds1da177e2005-04-16 15:20:36 -070051{
Eric W. Biederman96498772020-04-27 07:55:07 -050052 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 Torvalds1da177e2005-04-16 15:20:36 -070058
Thomas Gleixner77b4b542019-09-05 23:15:08 +020059 /*
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. Biederman96498772020-04-27 07:55:07 -050063 if (upid == 0)
64 return thread ? task_pid(current) : task_tgid(current);
65
66 pid = find_vpid(upid);
Thomas Gleixner6ae40e32019-08-21 21:08:48 +020067 if (!pid)
Thomas Gleixner6ae40e32019-08-21 21:08:48 +020068 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
Eric W. Biederman96498772020-04-27 07:55:07 -050070 if (thread) {
71 struct task_struct *tsk = pid_task(pid, PIDTYPE_PID);
72 return (tsk && same_thread_group(tsk, current)) ? pid : NULL;
73 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Eric W. Biederman96498772020-04-27 07:55:07 -050075 /*
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 Gleixner77b4b542019-09-05 23:15:08 +020083
Eric W. Biederman96498772020-04-27 07:55:07 -050084 /*
85 * For processes require that pid identifies a process.
86 */
87 return pid_has_task(pid, PIDTYPE_TGID) ? pid : NULL;
Thomas Gleixner6ae40e32019-08-21 21:08:48 +020088}
89
90static inline int validate_clock_permissions(const clockid_t clock)
91{
Eric W. Biederman9bf7c322020-04-25 18:38:54 -050092 int ret;
93
94 rcu_read_lock();
Eric W. Biederman96498772020-04-27 07:55:07 -050095 ret = pid_for_clock(clock, false) ? 0 : -EINVAL;
Eric W. Biederman9bf7c322020-04-25 18:38:54 -050096 rcu_read_unlock();
97
98 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099}
100
Eric W. Biedermanfece98262020-04-27 09:38:29 -0500101static inline enum pid_type clock_pid_type(const clockid_t clock)
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600102{
Eric W. Biedermanfece98262020-04-27 09:38:29 -0500103 return CPUCLOCK_PERTHREAD(clock) ? PIDTYPE_PID : PIDTYPE_TGID;
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600104}
105
106static inline struct task_struct *cpu_timer_task_rcu(struct k_itimer *timer)
107{
Eric W. Biedermanfece98262020-04-27 09:38:29 -0500108 return pid_task(timer->it.cpu.pid, clock_pid_type(timer->it_clock));
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600109}
110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111/*
112 * Update expiry time from increment, and increase overrun count,
113 * given the current clock sample.
114 */
Thomas Gleixner60bda032019-08-27 21:31:02 +0200115static u64 bump_cpu_timer(struct k_itimer *timer, u64 now)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
Thomas Gleixner60bda032019-08-27 21:31:02 +0200117 u64 delta, incr, expires = timer->it.cpu.node.expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 int i;
119
Thomas Gleixner16118792019-01-11 14:33:17 +0100120 if (!timer->it_interval)
Thomas Gleixner60bda032019-08-27 21:31:02 +0200121 return expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Thomas Gleixner60bda032019-08-27 21:31:02 +0200123 if (now < expires)
124 return expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
Thomas Gleixner16118792019-01-11 14:33:17 +0100126 incr = timer->it_interval;
Thomas Gleixner60bda032019-08-27 21:31:02 +0200127 delta = now + incr - expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
Frederic Weisbecker55ccb612013-06-28 00:06:42 +0000129 /* 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 Gleixner60bda032019-08-27 21:31:02 +0200137 timer->it.cpu.node.expires += incr;
Thomas Gleixner78c9c4d2018-06-26 15:21:32 +0200138 timer->it_overrun += 1LL << i;
Frederic Weisbecker55ccb612013-06-28 00:06:42 +0000139 delta -= incr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 }
Thomas Gleixner60bda032019-08-27 21:31:02 +0200141 return timer->it.cpu.node.expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142}
143
Thomas Gleixner2bbdbda2019-08-21 21:09:19 +0200144/* Check whether all cache entries contain U64_MAX, i.e. eternal expiry time */
145static inline bool expiry_cache_is_inactive(const struct posix_cputimers *pct)
Frederic Weisbecker555347f2013-04-19 16:17:38 +0200146{
Thomas Gleixner2bbdbda2019-08-21 21:09:19 +0200147 return !(~pct->bases[CPUCLOCK_PROF].nextevt |
148 ~pct->bases[CPUCLOCK_VIRT].nextevt |
149 ~pct->bases[CPUCLOCK_SCHED].nextevt);
Frederic Weisbecker555347f2013-04-19 16:17:38 +0200150}
151
Thomas Gleixnerbc2c8ea2011-02-01 13:52:12 +0000152static int
Deepa Dinamanid2e3e0c2017-03-26 12:04:15 -0700153posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
Thomas Gleixner6ae40e32019-08-21 21:08:48 +0200155 int error = validate_clock_permissions(which_clock);
156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 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 Gleixnerbc2c8ea2011-02-01 13:52:12 +0000172static int
Thomas Gleixner6ae40e32019-08-21 21:08:48 +0200173posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174{
Thomas Gleixner6ae40e32019-08-21 21:08:48 +0200175 int error = validate_clock_permissions(clock);
176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 /*
178 * You can never reset a CPU clock, but we check for other errors
179 * in the call before failing with EPERM.
180 */
Thomas Gleixner6ae40e32019-08-21 21:08:48 +0200181 return error ? : -EPERM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184/*
Thomas Gleixner2092c1d42019-08-21 21:09:00 +0200185 * Sample a per-thread clock for the given task. clkid is validated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 */
Thomas Gleixner8c2d74f2019-08-21 21:09:01 +0200187static u64 cpu_clock_sample(const clockid_t clkid, struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
Thomas Gleixnerab693c52019-08-21 21:09:03 +0200189 u64 utime, stime;
190
191 if (clkid == CPUCLOCK_SCHED)
192 return task_sched_runtime(p);
193
194 task_cputime(p, &utime, &stime);
195
Thomas Gleixner2092c1d42019-08-21 21:09:00 +0200196 switch (clkid) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 case CPUCLOCK_PROF:
Thomas Gleixnerab693c52019-08-21 21:09:03 +0200198 return utime + stime;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 case CPUCLOCK_VIRT:
Thomas Gleixnerab693c52019-08-21 21:09:03 +0200200 return utime;
Thomas Gleixner2092c1d42019-08-21 21:09:00 +0200201 default:
202 WARN_ON_ONCE(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 }
Thomas Gleixner8c2d74f2019-08-21 21:09:01 +0200204 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205}
206
Thomas Gleixnerb0d524f2019-08-21 21:09:12 +0200207static 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
214static 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
222static 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 Low10180162015-04-28 13:00:22 -0700233/*
234 * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
235 * to avoid race conditions with concurrent updates to cputime.
236 */
237static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
Peter Zijlstra4da94d492009-02-11 11:30:27 +0100238{
Jason Low10180162015-04-28 13:00:22 -0700239 u64 curr_cputime;
240retry:
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 Zijlstra4da94d492009-02-11 11:30:27 +0100247
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200248static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic,
249 struct task_cputime *sum)
Jason Low10180162015-04-28 13:00:22 -0700250{
Jason Low71107442015-04-28 13:00:24 -0700251 __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 Low10180162015-04-28 13:00:22 -0700254}
Peter Zijlstra4da94d492009-02-11 11:30:27 +0100255
Thomas Gleixner19298fb2019-08-21 21:08:51 +0200256/**
257 * thread_group_sample_cputime - Sample cputime for a given task
258 * @tsk: Task for which cputime needs to be started
Yi Wang7f2cbcb2019-10-21 15:44:12 +0800259 * @samples: Storage for time samples
Thomas Gleixner19298fb2019-08-21 21:08:51 +0200260 *
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 Gleixnerb7be4ef2019-08-21 21:09:16 +0200267void thread_group_sample_cputime(struct task_struct *tsk, u64 *samples)
Thomas Gleixner19298fb2019-08-21 21:08:51 +0200268{
269 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200270 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
Thomas Gleixner19298fb2019-08-21 21:08:51 +0200271
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200272 WARN_ON_ONCE(!pct->timers_active);
Thomas Gleixner19298fb2019-08-21 21:08:51 +0200273
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200274 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
Thomas Gleixner19298fb2019-08-21 21:08:51 +0200275}
276
Thomas Gleixnerc506bef42019-08-21 21:08:54 +0200277/**
278 * thread_group_start_cputime - Start cputime and return a sample
279 * @tsk: Task for which cputime needs to be started
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200280 * @samples: Storage for time samples
Thomas Gleixnerc506bef42019-08-21 21:08:54 +0200281 *
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 Gleixnerb7be4ef2019-08-21 21:09:16 +0200289static void thread_group_start_cputime(struct task_struct *tsk, u64 *samples)
Peter Zijlstra4da94d492009-02-11 11:30:27 +0100290{
291 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200292 struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
Peter Zijlstra4da94d492009-02-11 11:30:27 +0100293
Jason Low10180162015-04-28 13:00:22 -0700294 /* Check if cputimer isn't running. This is accessed without locking. */
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200295 if (!READ_ONCE(pct->timers_active)) {
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200296 struct task_cputime sum;
297
Peter Zijlstra4da94d492009-02-11 11:30:27 +0100298 /*
299 * The POSIX timer interface allows for absolute time expiry
300 * values through the TIMER_ABSTIME flag, therefore we have
Jason Low10180162015-04-28 13:00:22 -0700301 * to synchronize the timer to the clock every time we start it.
Peter Zijlstra4da94d492009-02-11 11:30:27 +0100302 */
Frederic Weisbeckerebd7e7f2017-01-31 04:09:34 +0100303 thread_group_cputime(tsk, &sum);
Jason Low71107442015-04-28 13:00:24 -0700304 update_gt_cputime(&cputimer->cputime_atomic, &sum);
Jason Low10180162015-04-28 13:00:22 -0700305
306 /*
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200307 * 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 Low10180162015-04-28 13:00:22 -0700311 * can handle concurrent updates.
312 */
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200313 WRITE_ONCE(pct->timers_active, true);
Jason Low10180162015-04-28 13:00:22 -0700314 }
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200315 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
316}
317
318static 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 Zijlstra4da94d492009-02-11 11:30:27 +0100324}
325
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326/*
Thomas Gleixner24ab7f52019-08-21 21:08:55 +0200327 * 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. Biedermana2efdbf2020-02-28 11:08:45 -0600329 * store. Otherwise a full update is required. clkid is already validated.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 */
Thomas Gleixner8c2d74f2019-08-21 21:09:01 +0200331static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p,
332 bool start)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
Thomas Gleixner24ab7f52019-08-21 21:08:55 +0200334 struct thread_group_cputimer *cputimer = &p->signal->cputimer;
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200335 struct posix_cputimers *pct = &p->signal->posix_cputimers;
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200336 u64 samples[CPUCLOCK_MAX];
Frank Mayharbb34d922008-09-12 09:54:39 -0700337
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200338 if (!READ_ONCE(pct->timers_active)) {
Thomas Gleixner24ab7f52019-08-21 21:08:55 +0200339 if (start)
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200340 thread_group_start_cputime(p, samples);
Thomas Gleixner24ab7f52019-08-21 21:08:55 +0200341 else
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200342 __thread_group_cputime(p, samples);
Thomas Gleixner24ab7f52019-08-21 21:08:55 +0200343 } else {
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200344 proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
Thomas Gleixner24ab7f52019-08-21 21:08:55 +0200345 }
346
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200347 return samples[clkid];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348}
349
Thomas Gleixnerbfcf3e92019-08-21 21:08:49 +0200350static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp)
Frederic Weisbecker33ab0fe2013-10-11 17:41:11 +0200351{
Thomas Gleixnerbfcf3e92019-08-21 21:08:49 +0200352 const clockid_t clkid = CPUCLOCK_WHICH(clock);
353 struct task_struct *tsk;
354 u64 t;
Frederic Weisbecker33ab0fe2013-10-11 17:41:11 +0200355
Eric W. Biederman9bf7c322020-04-25 18:38:54 -0500356 rcu_read_lock();
Eric W. Biederman96498772020-04-27 07:55:07 -0500357 tsk = pid_task(pid_for_clock(clock, true), clock_pid_type(clock));
Eric W. Biederman9bf7c322020-04-25 18:38:54 -0500358 if (!tsk) {
359 rcu_read_unlock();
Thomas Gleixnerbfcf3e92019-08-21 21:08:49 +0200360 return -EINVAL;
Eric W. Biederman9bf7c322020-04-25 18:38:54 -0500361 }
Frederic Weisbecker33ab0fe2013-10-11 17:41:11 +0200362
Thomas Gleixnerbfcf3e92019-08-21 21:08:49 +0200363 if (CPUCLOCK_PERTHREAD(clock))
Thomas Gleixner8c2d74f2019-08-21 21:09:01 +0200364 t = cpu_clock_sample(clkid, tsk);
Thomas Gleixnerbfcf3e92019-08-21 21:08:49 +0200365 else
Thomas Gleixner8c2d74f2019-08-21 21:09:01 +0200366 t = cpu_clock_sample_group(clkid, tsk, false);
Eric W. Biederman9bf7c322020-04-25 18:38:54 -0500367 rcu_read_unlock();
Frederic Weisbecker33ab0fe2013-10-11 17:41:11 +0200368
Thomas Gleixnerbfcf3e92019-08-21 21:08:49 +0200369 *tp = ns_to_timespec64(t);
370 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371}
372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373/*
374 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
Stanislaw Gruszkaba5ea952009-11-17 14:14:13 -0800375 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
376 * new timer already all-zeros initialized.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 */
Thomas Gleixnerbc2c8ea2011-02-01 13:52:12 +0000378static int posix_cpu_timer_create(struct k_itimer *new_timer)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
Thomas Gleixner1fb497d2020-07-30 12:14:06 +0200380 static struct lock_class_key posix_cpu_timers_key;
Eric W. Biederman96498772020-04-27 07:55:07 -0500381 struct pid *pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
Eric W. Biederman9bf7c322020-04-25 18:38:54 -0500383 rcu_read_lock();
Eric W. Biederman96498772020-04-27 07:55:07 -0500384 pid = pid_for_clock(new_timer->it_clock, false);
385 if (!pid) {
Eric W. Biederman9bf7c322020-04-25 18:38:54 -0500386 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 return -EINVAL;
Eric W. Biederman9bf7c322020-04-25 18:38:54 -0500388 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Thomas Gleixner1fb497d2020-07-30 12:14:06 +0200390 /*
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 Gleixnerd97bb752017-05-30 23:15:44 +0200401 new_timer->kclock = &clock_posix_cpu;
Thomas Gleixner60bda032019-08-27 21:31:02 +0200402 timerqueue_init(&new_timer->it.cpu.node);
Eric W. Biederman96498772020-04-27 07:55:07 -0500403 new_timer->it.cpu.pid = get_pid(pid);
Eric W. Biederman9bf7c322020-04-25 18:38:54 -0500404 rcu_read_unlock();
Thomas Gleixnere5a8b652019-08-21 21:08:50 +0200405 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406}
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 Gleixnerbc2c8ea2011-02-01 13:52:12 +0000414static int posix_cpu_timer_del(struct k_itimer *timer)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
Thomas Gleixner60bda032019-08-27 21:31:02 +0200416 struct cpu_timer *ctmr = &timer->it.cpu;
Frederic Weisbecker3d7a1422013-10-11 17:41:11 +0200417 struct sighand_struct *sighand;
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600418 struct task_struct *p;
Thomas Gleixner60bda032019-08-27 21:31:02 +0200419 unsigned long flags;
420 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600422 rcu_read_lock();
423 p = cpu_timer_task_rcu(timer);
424 if (!p)
425 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
Frederic Weisbecker3d7a1422013-10-11 17:41:11 +0200427 /*
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 Weisbeckera3222f82013-10-11 00:37:39 +0200433 /*
Thomas Gleixner60bda032019-08-27 21:31:02 +0200434 * This raced with the reaping of the task. The exit cleanup
435 * should have removed this timer from the timer queue.
Frederic Weisbeckera3222f82013-10-11 00:37:39 +0200436 */
Thomas Gleixner60bda032019-08-27 21:31:02 +0200437 WARN_ON_ONCE(ctmr->head || timerqueue_node_queued(&ctmr->node));
Frederic Weisbeckera3222f82013-10-11 00:37:39 +0200438 } else {
Frederic Weisbeckera3222f82013-10-11 00:37:39 +0200439 if (timer->it.cpu.firing)
440 ret = TIMER_RETRY;
441 else
Thomas Gleixner60bda032019-08-27 21:31:02 +0200442 cpu_timer_dequeue(ctmr);
Frederic Weisbecker3d7a1422013-10-11 17:41:11 +0200443
444 unlock_task_sighand(p, &flags);
Oleg Nesterov108150e2005-10-23 20:25:39 +0400445 }
Frederic Weisbeckera3222f82013-10-11 00:37:39 +0200446
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600447out:
448 rcu_read_unlock();
Frederic Weisbeckera3222f82013-10-11 00:37:39 +0200449 if (!ret)
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600450 put_pid(ctmr->pid);
Oleg Nesterov108150e2005-10-23 20:25:39 +0400451
452 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453}
454
Thomas Gleixner60bda032019-08-27 21:31:02 +0200455static void cleanup_timerqueue(struct timerqueue_head *head)
Frederic Weisbecker1a7fa512013-06-28 00:06:42 +0000456{
Thomas Gleixner60bda032019-08-27 21:31:02 +0200457 struct timerqueue_node *node;
458 struct cpu_timer *ctmr;
Frederic Weisbecker1a7fa512013-06-28 00:06:42 +0000459
Thomas Gleixner60bda032019-08-27 21:31:02 +0200460 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 Weisbecker1a7fa512013-06-28 00:06:42 +0000465}
466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467/*
Thomas Gleixner7cb9a942019-08-19 16:31:45 +0200468 * 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 Torvalds1da177e2005-04-16 15:20:36 -0700472 * This must be called with the siglock held.
473 */
Thomas Gleixner2b699422019-08-21 21:09:04 +0200474static void cleanup_timers(struct posix_cputimers *pct)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475{
Thomas Gleixner60bda032019-08-27 21:31:02 +0200476 cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead);
477 cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead);
478 cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479}
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 */
486void posix_cpu_timers_exit(struct task_struct *tsk)
487{
Thomas Gleixner2b699422019-08-21 21:09:04 +0200488 cleanup_timers(&tsk->posix_cputimers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489}
490void posix_cpu_timers_exit_group(struct task_struct *tsk)
491{
Thomas Gleixner2b699422019-08-21 21:09:04 +0200492 cleanup_timers(&tsk->signal->posix_cputimers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493}
494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495/*
496 * Insert the timer on the appropriate list before any timers that
Frederic Weisbeckere73d84e2013-10-11 18:56:49 +0200497 * expire later. This must be called with the sighand lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 */
Eric W. Biedermanbeb41d92020-02-28 11:09:46 -0600499static void arm_timer(struct k_itimer *timer, struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
Thomas Gleixner3b495b22019-08-21 21:09:08 +0200501 int clkidx = CPUCLOCK_WHICH(timer->it_clock);
Thomas Gleixner60bda032019-08-27 21:31:02 +0200502 struct cpu_timer *ctmr = &timer->it.cpu;
503 u64 newexp = cpu_timer_getexpires(ctmr);
Thomas Gleixner87dc6442019-08-26 20:22:24 +0200504 struct posix_cputimer_base *base;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
Thomas Gleixner87dc6442019-08-26 20:22:24 +0200506 if (CPUCLOCK_PERTHREAD(timer->it_clock))
507 base = p->posix_cputimers.bases + clkidx;
508 else
509 base = p->signal->posix_cputimers.bases + clkidx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Thomas Gleixner60bda032019-08-27 21:31:02 +0200511 if (!cpu_timer_enqueue(&base->tqhead, ctmr))
Thomas Gleixner3b495b22019-08-21 21:09:08 +0200512 return;
Stanislaw Gruszka5eb9aa62010-03-11 14:04:38 -0800513
Thomas Gleixner3b495b22019-08-21 21:09:08 +0200514 /*
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 Gleixner2bbdbda2019-08-21 21:09:19 +0200520 if (newexp < base->nextevt)
Thomas Gleixner87dc6442019-08-26 20:22:24 +0200521 base->nextevt = newexp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522
Thomas Gleixner3b495b22019-08-21 21:09:08 +0200523 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 Torvalds1da177e2005-04-16 15:20:36 -0700527}
528
529/*
530 * The timer is locked, fire it and arrange for its reload.
531 */
532static void cpu_timer_fire(struct k_itimer *timer)
533{
Thomas Gleixner60bda032019-08-27 21:31:02 +0200534 struct cpu_timer *ctmr = &timer->it.cpu;
535
Stanislaw Gruszka1f169f82010-03-11 14:04:41 -0800536 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
537 /*
538 * User don't want any signal.
539 */
Thomas Gleixner60bda032019-08-27 21:31:02 +0200540 cpu_timer_setexpires(ctmr, 0);
Stanislaw Gruszka1f169f82010-03-11 14:04:41 -0800541 } else if (unlikely(timer->sigq == NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 /*
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 Gleixner60bda032019-08-27 21:31:02 +0200547 cpu_timer_setexpires(ctmr, 0);
Thomas Gleixner16118792019-01-11 14:33:17 +0100548 } else if (!timer->it_interval) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 /*
550 * One-shot timer. Clear it as soon as it's fired.
551 */
552 posix_timer_event(timer, 0);
Thomas Gleixner60bda032019-08-27 21:31:02 +0200553 cpu_timer_setexpires(ctmr, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 } 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 Gleixnerf37fb0a2017-05-30 23:15:47 +0200561 posix_cpu_timer_rearm(timer);
Thomas Gleixneraf888d62017-05-30 23:15:42 +0200562 ++timer->it_requeue_pending;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 }
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 Weisbeckere73d84e2013-10-11 18:56:49 +0200572static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
Deepa Dinamani5f252b32017-03-26 12:04:17 -0700573 struct itimerspec64 *new, struct itimerspec64 *old)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574{
Thomas Gleixnerc7a37c62019-08-21 21:08:56 +0200575 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
Frederic Weisbeckerebd7e7f2017-01-31 04:09:34 +0100576 u64 old_expires, new_expires, old_incr, val;
Thomas Gleixner60bda032019-08-27 21:31:02 +0200577 struct cpu_timer *ctmr = &timer->it.cpu;
Thomas Gleixnerc7a37c62019-08-21 21:08:56 +0200578 struct sighand_struct *sighand;
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600579 struct task_struct *p;
Thomas Gleixnerc7a37c62019-08-21 21:08:56 +0200580 unsigned long flags;
Thomas Gleixner60bda032019-08-27 21:31:02 +0200581 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600583 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 Torvalds1da177e2005-04-16 15:20:36 -0700593
Thomas Gleixner098b0e02017-06-20 17:37:36 +0200594 /*
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 Torvalds1da177e2005-04-16 15:20:36 -0700599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 /*
Frederic Weisbeckere73d84e2013-10-11 18:56:49 +0200601 * 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 Torvalds1da177e2005-04-16 15:20:36 -0700607 * longer get any information about it at all.
608 */
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600609 if (unlikely(sighand == NULL)) {
610 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 return -ESRCH;
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600612 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
614 /*
615 * Disarm any old timer after extracting its expiry time.
616 */
Thomas Gleixner16118792019-01-11 14:33:17 +0100617 old_incr = timer->it_interval;
Thomas Gleixner60bda032019-08-27 21:31:02 +0200618 old_expires = cpu_timer_getexpires(ctmr);
619
Oleg Nesterova69ac4a2005-10-24 18:29:58 +0400620 if (unlikely(timer->it.cpu.firing)) {
621 timer->it.cpu.firing = -1;
622 ret = TIMER_RETRY;
Thomas Gleixner60bda032019-08-27 21:31:02 +0200623 } else {
624 cpu_timer_dequeue(ctmr);
625 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
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 Gleixner8c2d74f2019-08-21 21:09:01 +0200635 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 Torvalds1da177e2005-04-16 15:20:36 -0700639
640 if (old) {
Frederic Weisbecker55ccb612013-06-28 00:06:42 +0000641 if (old_expires == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 old->it_value.tv_sec = 0;
643 old->it_value.tv_nsec = 0;
644 } else {
645 /*
Thomas Gleixner60bda032019-08-27 21:31:02 +0200646 * 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 Torvalds1da177e2005-04-16 15:20:36 -0700651 */
Thomas Gleixner60bda032019-08-27 21:31:02 +0200652 u64 exp = bump_cpu_timer(timer, val);
653
654 if (val < exp) {
655 old_expires = exp - val;
Deepa Dinamani5f252b32017-03-26 12:04:17 -0700656 old->it_value = ns_to_timespec64(old_expires);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 } else {
658 old->it_value.tv_nsec = 1;
659 old->it_value.tv_sec = 0;
660 }
661 }
662 }
663
Oleg Nesterova69ac4a2005-10-24 18:29:58 +0400664 if (unlikely(ret)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 /*
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 Weisbeckere73d84e2013-10-11 18:56:49 +0200671 unlock_task_sighand(p, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 goto out;
673 }
674
Frederic Weisbeckere73d84e2013-10-11 18:56:49 +0200675 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
Frederic Weisbecker55ccb612013-06-28 00:06:42 +0000676 new_expires += val;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 }
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 Gleixner60bda032019-08-27 21:31:02 +0200684 cpu_timer_setexpires(ctmr, new_expires);
Frederic Weisbecker55ccb612013-06-28 00:06:42 +0000685 if (new_expires != 0 && val < new_expires) {
Eric W. Biedermanbeb41d92020-02-28 11:09:46 -0600686 arm_timer(timer, p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 }
688
Frederic Weisbeckere73d84e2013-10-11 18:56:49 +0200689 unlock_task_sighand(p, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 /*
691 * Install the new reload setting, and
692 * set up the signal and overrun bookkeeping.
693 */
Thomas Gleixner16118792019-01-11 14:33:17 +0100694 timer->it_interval = timespec64_to_ktime(new->it_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
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 Weisbecker55ccb612013-06-28 00:06:42 +0000706 if (new_expires != 0 && !(val < new_expires)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 /*
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. Biederman55e8c8e2020-02-28 11:11:06 -0600717 rcu_read_unlock();
Frederic Weisbeckerebd7e7f2017-01-31 04:09:34 +0100718 if (old)
Deepa Dinamani5f252b32017-03-26 12:04:17 -0700719 old->it_interval = ns_to_timespec64(old_incr);
Frederic Weisbeckerb7878302015-07-17 22:25:49 +0200720
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 return ret;
722}
723
Deepa Dinamani5f252b32017-03-26 12:04:17 -0700724static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
Thomas Gleixner99093c52019-08-21 21:08:57 +0200726 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
Thomas Gleixner60bda032019-08-27 21:31:02 +0200727 struct cpu_timer *ctmr = &timer->it.cpu;
728 u64 now, expires = cpu_timer_getexpires(ctmr);
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600729 struct task_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600731 rcu_read_lock();
732 p = cpu_timer_task_rcu(timer);
733 if (!p)
734 goto out;
Frederic Weisbeckera3222f82013-10-11 00:37:39 +0200735
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 /*
737 * Easy part: convert the reload time.
738 */
Thomas Gleixner16118792019-01-11 14:33:17 +0100739 itp->it_interval = ktime_to_timespec64(timer->it_interval);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
Thomas Gleixner60bda032019-08-27 21:31:02 +0200741 if (!expires)
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600742 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 /*
745 * Sample the clock to take the difference with the expiry time.
746 */
Eric W. Biederman60f2cea2020-02-28 11:09:19 -0600747 if (CPUCLOCK_PERTHREAD(timer->it_clock))
Thomas Gleixner8c2d74f2019-08-21 21:09:01 +0200748 now = cpu_clock_sample(clkid, p);
Eric W. Biederman60f2cea2020-02-28 11:09:19 -0600749 else
750 now = cpu_clock_sample_group(clkid, p, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
Thomas Gleixner60bda032019-08-27 21:31:02 +0200752 if (now < expires) {
753 itp->it_value = ns_to_timespec64(expires - now);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 } 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. Biederman55e8c8e2020-02-28 11:11:06 -0600762out:
763 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764}
765
Thomas Gleixner60bda032019-08-27 21:31:02 +0200766#define MAX_COLLECTED 20
767
768static u64 collect_timerqueue(struct timerqueue_head *head,
769 struct list_head *firing, u64 now)
Frederic Weisbecker2473f3e2013-06-28 00:06:43 +0000770{
Thomas Gleixner60bda032019-08-27 21:31:02 +0200771 struct timerqueue_node *next;
772 int i = 0;
Frederic Weisbecker2473f3e2013-06-28 00:06:43 +0000773
Thomas Gleixner60bda032019-08-27 21:31:02 +0200774 while ((next = timerqueue_getnext(head))) {
775 struct cpu_timer *ctmr;
776 u64 expires;
Frederic Weisbecker2473f3e2013-06-28 00:06:43 +0000777
Thomas Gleixner60bda032019-08-27 21:31:02 +0200778 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 Weisbecker2473f3e2013-06-28 00:06:43 +0000783
Thomas Gleixner60bda032019-08-27 21:31:02 +0200784 ctmr->firing = 1;
785 cpu_timer_dequeue(ctmr);
786 list_add_tail(&ctmr->elist, firing);
Frederic Weisbecker2473f3e2013-06-28 00:06:43 +0000787 }
788
Thomas Gleixner2bbdbda2019-08-21 21:09:19 +0200789 return U64_MAX;
Frederic Weisbecker2473f3e2013-06-28 00:06:43 +0000790}
791
Thomas Gleixner60bda032019-08-27 21:31:02 +0200792static void collect_posix_cputimers(struct posix_cputimers *pct, u64 *samples,
793 struct list_head *firing)
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200794{
795 struct posix_cputimer_base *base = pct->bases;
796 int i;
797
798 for (i = 0; i < CPUCLOCK_MAX; i++, base++) {
Thomas Gleixner60bda032019-08-27 21:31:02 +0200799 base->nextevt = collect_timerqueue(&base->tqhead, firing,
800 samples[i]);
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200801 }
802}
803
Juri Lelli34be3932017-12-12 12:10:24 +0100804static 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 Gleixner8991afe2019-08-21 21:09:23 +0200812static 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 Torvalds1da177e2005-04-16 15:20:36 -0700826/*
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 */
831static void check_thread_timers(struct task_struct *tsk,
832 struct list_head *firing)
833{
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200834 struct posix_cputimers *pct = &tsk->posix_cputimers;
835 u64 samples[CPUCLOCK_MAX];
Jiri Slabyd4bb52742010-03-05 13:42:53 -0800836 unsigned long soft;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
Juri Lelli34be3932017-12-12 12:10:24 +0100838 if (dl_task(tsk))
839 check_dl_overrun(tsk);
840
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200841 if (expiry_cache_is_inactive(pct))
Jason Low934715a2015-10-14 12:07:54 -0700842 return;
843
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200844 task_sample_cputime(tsk, samples);
845 collect_posix_cputimers(pct, samples, firing);
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100846
847 /*
848 * Check for the special case thread timers.
849 */
Krzysztof Opasiak3cf29492017-07-05 19:25:48 +0200850 soft = task_rlimit(tsk, RLIMIT_RTTIME);
Jiri Slabyd4bb52742010-03-05 13:42:53 -0800851 if (soft != RLIM_INFINITY) {
Thomas Gleixner8ea1de92019-08-21 21:09:21 +0200852 /* Task RT timeout is accounted in jiffies. RTTIME is usec */
Thomas Gleixner8991afe2019-08-21 21:09:23 +0200853 unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ);
Krzysztof Opasiak3cf29492017-07-05 19:25:48 +0200854 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100855
Thomas Gleixner8991afe2019-08-21 21:09:23 +0200856 /* At the hard limit, send SIGKILL. No further action. */
857 if (hard != RLIM_INFINITY &&
858 check_rlimit(rttime, hard, SIGKILL, true, true))
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100859 return;
Thomas Gleixnerdd670222019-08-21 21:09:22 +0200860
Thomas Gleixner8991afe2019-08-21 21:09:23 +0200861 /* At the soft limit, send a SIGXCPU every second */
862 if (check_rlimit(rttime, soft, SIGXCPU, true, false)) {
Thomas Gleixnerdd670222019-08-21 21:09:22 +0200863 soft += USEC_PER_SEC;
864 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft;
Peter Zijlstra78f2c7d2008-01-25 21:08:27 +0100865 }
866 }
Thomas Gleixnerc02b0782019-08-21 21:09:10 +0200867
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200868 if (expiry_cache_is_inactive(pct))
Frederic Weisbeckerb7878302015-07-17 22:25:49 +0200869 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870}
871
Jason Low10180162015-04-28 13:00:22 -0700872static inline void stop_process_timers(struct signal_struct *sig)
Peter Zijlstra3fccfd62009-02-10 16:37:31 +0100873{
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200874 struct posix_cputimers *pct = &sig->posix_cputimers;
Peter Zijlstra3fccfd62009-02-10 16:37:31 +0100875
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200876 /* Turn off the active flag. This is done without locking. */
877 WRITE_ONCE(pct->timers_active, false);
Frederic Weisbeckerb7878302015-07-17 22:25:49 +0200878 tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
Peter Zijlstra3fccfd62009-02-10 16:37:31 +0100879}
880
Stanislaw Gruszka42c4ab42009-07-29 12:15:26 +0200881static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
Frederic Weisbeckerebd7e7f2017-01-31 04:09:34 +0100882 u64 *expires, u64 cur_time, int signo)
Stanislaw Gruszka42c4ab42009-07-29 12:15:26 +0200883{
Martin Schwidefsky64861632011-12-15 14:56:09 +0100884 if (!it->expires)
Stanislaw Gruszka42c4ab42009-07-29 12:15:26 +0200885 return;
886
Frederic Weisbecker858cf3a2017-01-31 04:09:35 +0100887 if (cur_time >= it->expires) {
888 if (it->incr)
Martin Schwidefsky64861632011-12-15 14:56:09 +0100889 it->expires += it->incr;
Frederic Weisbecker858cf3a2017-01-31 04:09:35 +0100890 else
Martin Schwidefsky64861632011-12-15 14:56:09 +0100891 it->expires = 0;
Stanislaw Gruszka42c4ab42009-07-29 12:15:26 +0200892
Xiao Guangrong3f0a5252009-08-10 10:52:30 +0800893 trace_itimer_expire(signo == SIGPROF ?
894 ITIMER_PROF : ITIMER_VIRTUAL,
Eric W. Biederman6883f812017-06-04 04:32:13 -0500895 task_tgid(tsk), cur_time);
Stanislaw Gruszka42c4ab42009-07-29 12:15:26 +0200896 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
897 }
898
Thomas Gleixner2bbdbda2019-08-21 21:09:19 +0200899 if (it->expires && it->expires < *expires)
Frederic Weisbecker858cf3a2017-01-31 04:09:35 +0100900 *expires = it->expires;
Stanislaw Gruszka42c4ab42009-07-29 12:15:26 +0200901}
902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903/*
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 */
908static void check_process_timers(struct task_struct *tsk,
909 struct list_head *firing)
910{
911 struct signal_struct *const sig = tsk->signal;
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200912 struct posix_cputimers *pct = &sig->posix_cputimers;
913 u64 samples[CPUCLOCK_MAX];
Jiri Slabyd4bb52742010-03-05 13:42:53 -0800914 unsigned long soft;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 /*
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200917 * If there are no active process wide timers (POSIX 1.b, itimers,
Thomas Gleixnera2ed4fd2019-08-29 12:52:28 +0200918 * RLIMIT_CPU) nothing to check. Also skip the process wide timer
919 * processing when there is already another task handling them.
Jason Low934715a2015-10-14 12:07:54 -0700920 */
Thomas Gleixnera2ed4fd2019-08-29 12:52:28 +0200921 if (!READ_ONCE(pct->timers_active) || pct->expiry_active)
Jason Low934715a2015-10-14 12:07:54 -0700922 return;
923
Thomas Gleixnera2ed4fd2019-08-29 12:52:28 +0200924 /*
Jason Lowc8d75aa2015-10-14 12:07:56 -0700925 * Signify that a thread is checking for process timers.
926 * Write access to this field is protected by the sighand lock.
927 */
Thomas Gleixnera2ed4fd2019-08-29 12:52:28 +0200928 pct->expiry_active = true;
Jason Lowc8d75aa2015-10-14 12:07:56 -0700929
Jason Low934715a2015-10-14 12:07:54 -0700930 /*
Thomas Gleixnera3249562019-08-21 21:08:53 +0200931 * Collect the current process totals. Group accounting is active
932 * so the sample can be taken directly.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 */
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200934 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples);
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200935 collect_posix_cputimers(pct, samples, firing);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937 /*
938 * Check for the special case process timers.
939 */
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200940 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF],
941 &pct->bases[CPUCLOCK_PROF].nextevt,
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200942 samples[CPUCLOCK_PROF], SIGPROF);
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200943 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT],
944 &pct->bases[CPUCLOCK_VIRT].nextevt,
945 samples[CPUCLOCK_VIRT], SIGVTALRM);
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200946
Krzysztof Opasiak3cf29492017-07-05 19:25:48 +0200947 soft = task_rlimit(tsk, RLIMIT_CPU);
Jiri Slabyd4bb52742010-03-05 13:42:53 -0800948 if (soft != RLIM_INFINITY) {
Thomas Gleixner8ea1de92019-08-21 21:09:21 +0200949 /* RLIMIT_CPU is in seconds. Samples are nanoseconds */
Krzysztof Opasiak3cf29492017-07-05 19:25:48 +0200950 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
Thomas Gleixner8ea1de92019-08-21 21:09:21 +0200951 u64 ptime = samples[CPUCLOCK_PROF];
952 u64 softns = (u64)soft * NSEC_PER_SEC;
953 u64 hardns = (u64)hard * NSEC_PER_SEC;
Thomas Gleixnerb7be4ef2019-08-21 21:09:16 +0200954
Thomas Gleixner8991afe2019-08-21 21:09:23 +0200955 /* At the hard limit, send SIGKILL. No further action. */
956 if (hard != RLIM_INFINITY &&
957 check_rlimit(ptime, hardns, SIGKILL, false, true))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 return;
Thomas Gleixnerdd670222019-08-21 21:09:22 +0200959
Thomas Gleixner8991afe2019-08-21 21:09:23 +0200960 /* At the soft limit, send a SIGXCPU every second */
961 if (check_rlimit(ptime, softns, SIGXCPU, false, false)) {
Thomas Gleixnerdd670222019-08-21 21:09:22 +0200962 sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1;
963 softns += NSEC_PER_SEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 }
Thomas Gleixner8ea1de92019-08-21 21:09:21 +0200965
966 /* Update the expiry cache */
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200967 if (softns < pct->bases[CPUCLOCK_PROF].nextevt)
968 pct->bases[CPUCLOCK_PROF].nextevt = softns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 }
970
Thomas Gleixner1cd07c0b2019-08-21 21:09:20 +0200971 if (expiry_cache_is_inactive(pct))
Stanislaw Gruszka29f87b72010-04-27 14:12:15 -0700972 stop_process_timers(sig);
Jason Lowc8d75aa2015-10-14 12:07:56 -0700973
Thomas Gleixner244d49e2019-08-21 21:09:24 +0200974 pct->expiry_active = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975}
976
977/*
Thomas Gleixner96fe3b02017-05-30 23:15:46 +0200978 * This is called from the signal code (via posixtimer_rearm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 * when the last timer signal was delivered and we have to reload the timer.
980 */
Thomas Gleixnerf37fb0a2017-05-30 23:15:47 +0200981static void posix_cpu_timer_rearm(struct k_itimer *timer)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982{
Thomas Gleixnerda020ce2019-08-21 21:08:58 +0200983 clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600984 struct task_struct *p;
Frederic Weisbeckere73d84e2013-10-11 18:56:49 +0200985 struct sighand_struct *sighand;
986 unsigned long flags;
Frederic Weisbeckerebd7e7f2017-01-31 04:09:34 +0100987 u64 now;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -0600989 rcu_read_lock();
990 p = cpu_timer_task_rcu(timer);
991 if (!p)
992 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
994 /*
995 * Fetch the current sample and update the timer's expiry time.
996 */
Eric W. Biederman60f2cea2020-02-28 11:09:19 -0600997 if (CPUCLOCK_PERTHREAD(timer->it_clock))
Thomas Gleixner8c2d74f2019-08-21 21:09:01 +0200998 now = cpu_clock_sample(clkid, p);
Eric W. Biederman60f2cea2020-02-28 11:09:19 -0600999 else
Thomas Gleixner8c2d74f2019-08-21 21:09:01 +02001000 now = cpu_clock_sample_group(clkid, p, true);
Eric W. Biederman60f2cea2020-02-28 11:09:19 -06001001
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. Biederman55e8c8e2020-02-28 11:11:06 -06001007 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009 /*
1010 * Now re-arm for the new expiry time.
1011 */
Eric W. Biedermanbeb41d92020-02-28 11:09:46 -06001012 arm_timer(timer, p);
Frederic Weisbeckere73d84e2013-10-11 18:56:49 +02001013 unlock_task_sighand(p, &flags);
Eric W. Biederman55e8c8e2020-02-28 11:11:06 -06001014out:
1015 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016}
1017
Frank Mayharf06febc2008-09-12 09:54:39 -07001018/**
Thomas Gleixner87dc6442019-08-26 20:22:24 +02001019 * task_cputimers_expired - Check whether posix CPU timers are expired
Frank Mayharf06febc2008-09-12 09:54:39 -07001020 *
Thomas Gleixner001f7972019-08-21 21:09:13 +02001021 * @samples: Array of current samples for the CPUCLOCK clocks
Thomas Gleixner87dc6442019-08-26 20:22:24 +02001022 * @pct: Pointer to a posix_cputimers container
Frank Mayharf06febc2008-09-12 09:54:39 -07001023 *
Thomas Gleixner87dc6442019-08-26 20:22:24 +02001024 * Returns true if any member of @samples is greater than the corresponding
1025 * member of @pct->bases[CLK].nextevt. False otherwise
Frank Mayharf06febc2008-09-12 09:54:39 -07001026 */
Thomas Gleixner87dc6442019-08-26 20:22:24 +02001027static inline bool
Yi Wang7f2cbcb2019-10-21 15:44:12 +08001028task_cputimers_expired(const u64 *samples, struct posix_cputimers *pct)
Frank Mayharf06febc2008-09-12 09:54:39 -07001029{
Thomas Gleixner001f7972019-08-21 21:09:13 +02001030 int i;
1031
1032 for (i = 0; i < CPUCLOCK_MAX; i++) {
Yi Wang7f2cbcb2019-10-21 15:44:12 +08001033 if (samples[i] >= pct->bases[i].nextevt)
Thomas Gleixner001f7972019-08-21 21:09:13 +02001034 return true;
1035 }
1036 return false;
Frank Mayharf06febc2008-09-12 09:54:39 -07001037}
1038
1039/**
1040 * fastpath_timer_check - POSIX CPU timers fast path.
1041 *
1042 * @tsk: The task (thread) being checked.
Frank Mayharf06febc2008-09-12 09:54:39 -07001043 *
Frank Mayharbb34d922008-09-12 09:54:39 -07001044 * 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 Mayharf06febc2008-09-12 09:54:39 -07001048 */
Thomas Gleixner001f7972019-08-21 21:09:13 +02001049static inline bool fastpath_timer_check(struct task_struct *tsk)
Frank Mayharf06febc2008-09-12 09:54:39 -07001050{
Thomas Gleixner244d49e2019-08-21 21:09:24 +02001051 struct posix_cputimers *pct = &tsk->posix_cputimers;
Oleg Nesterovad133ba2008-11-17 15:39:47 +01001052 struct signal_struct *sig;
Frank Mayharf06febc2008-09-12 09:54:39 -07001053
Thomas Gleixner244d49e2019-08-21 21:09:24 +02001054 if (!expiry_cache_is_inactive(pct)) {
Thomas Gleixner001f7972019-08-21 21:09:13 +02001055 u64 samples[CPUCLOCK_MAX];
Frank Mayharbb34d922008-09-12 09:54:39 -07001056
Thomas Gleixner001f7972019-08-21 21:09:13 +02001057 task_sample_cputime(tsk, samples);
Thomas Gleixner244d49e2019-08-21 21:09:24 +02001058 if (task_cputimers_expired(samples, pct))
Thomas Gleixner001f7972019-08-21 21:09:13 +02001059 return true;
Frank Mayharbb34d922008-09-12 09:54:39 -07001060 }
Oleg Nesterovad133ba2008-11-17 15:39:47 +01001061
1062 sig = tsk->signal;
Thomas Gleixner244d49e2019-08-21 21:09:24 +02001063 pct = &sig->posix_cputimers;
Jason Lowc8d75aa2015-10-14 12:07:56 -07001064 /*
Thomas Gleixner244d49e2019-08-21 21:09:24 +02001065 * 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 Lowc8d75aa2015-10-14 12:07:56 -07001071 *
Thomas Gleixner244d49e2019-08-21 21:09:24 +02001072 * 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 Lowc8d75aa2015-10-14 12:07:56 -07001078 */
Thomas Gleixner244d49e2019-08-21 21:09:24 +02001079 if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) {
Thomas Gleixner001f7972019-08-21 21:09:13 +02001080 u64 samples[CPUCLOCK_MAX];
Frank Mayharbb34d922008-09-12 09:54:39 -07001081
Thomas Gleixner001f7972019-08-21 21:09:13 +02001082 proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic,
1083 samples);
Oleg Nesterov8d1f4312010-06-11 20:04:46 +02001084
Thomas Gleixner244d49e2019-08-21 21:09:24 +02001085 if (task_cputimers_expired(samples, pct))
Thomas Gleixner001f7972019-08-21 21:09:13 +02001086 return true;
Frank Mayharbb34d922008-09-12 09:54:39 -07001087 }
Oleg Nesterov37bebc72009-03-23 20:34:11 +01001088
Juri Lelli34be3932017-12-12 12:10:24 +01001089 if (dl_task(tsk) && tsk->dl.dl_overrun)
Thomas Gleixner001f7972019-08-21 21:09:13 +02001090 return true;
Juri Lelli34be3932017-12-12 12:10:24 +01001091
Thomas Gleixner001f7972019-08-21 21:09:13 +02001092 return false;
Frank Mayharf06febc2008-09-12 09:54:39 -07001093}
1094
Thomas Gleixner1fb497d2020-07-30 12:14:06 +02001095static void handle_posix_cpu_timers(struct task_struct *tsk);
1096
1097#ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
1098static 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 */
1107void __init posix_cputimers_init_work(void)
1108{
1109 init_task_work(&current->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 */
1119static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1120{
1121 return tsk->posix_cputimers_work.scheduled;
1122}
1123
1124static 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
1134static 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 */
1172static 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
1179static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
1180{
1181 return false;
1182}
1183
1184static 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
1191static void handle_posix_cpu_timers(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 struct k_itimer *timer, *next;
Thomas Gleixner1fb497d2020-07-30 12:14:06 +02001194 unsigned long flags, start;
Thomas Gleixnerdce3e8f2019-08-19 16:31:47 +02001195 LIST_HEAD(firing);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Thomas Gleixner820903c72020-07-30 12:14:05 +02001197 if (!lock_task_sighand(tsk, &flags))
Frank Mayharf06febc2008-09-12 09:54:39 -07001198 return;
Ingo Molnar5ce73a42008-09-14 17:11:46 +02001199
Thomas Gleixner1fb497d2020-07-30 12:14:06 +02001200 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 Low934715a2015-10-14 12:07:54 -07001208
Thomas Gleixner1fb497d2020-07-30 12:14:06 +02001209 /*
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 Torvalds1da177e2005-04-16 15:20:36 -07001249
Frank Mayharbb34d922008-09-12 09:54:39 -07001250 /*
Thomas Gleixner1fb497d2020-07-30 12:14:06 +02001251 * We must release sighand lock before taking any timer's lock.
Frank Mayharbb34d922008-09-12 09:54:39 -07001252 * 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 Nesterov0bdd2ed2010-06-11 01:10:18 +02001258 unlock_task_sighand(tsk, &flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
1260 /*
1261 * Now that all the timers on our list have the firing flag,
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001262 * no one will touch their list entries but us. We'll take
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 * each timer's lock before clearing its firing flag, so no
1264 * timer call will interfere.
1265 */
Thomas Gleixner60bda032019-08-27 21:31:02 +02001266 list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) {
H Hartley Sweeten6e85c5b2009-04-29 19:14:32 -04001267 int cpu_firing;
1268
Thomas Gleixner1fb497d2020-07-30 12:14:06 +02001269 /*
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 Torvalds1da177e2005-04-16 15:20:36 -07001276 spin_lock(&timer->it_lock);
Thomas Gleixner60bda032019-08-27 21:31:02 +02001277 list_del_init(&timer->it.cpu.elist);
H Hartley Sweeten6e85c5b2009-04-29 19:14:32 -04001278 cpu_firing = timer->it.cpu.firing;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 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 Sweeten6e85c5b2009-04-29 19:14:32 -04001285 if (likely(cpu_firing >= 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 cpu_timer_fire(timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 spin_unlock(&timer->it_lock);
1288 }
Thomas Gleixner820903c72020-07-30 12:14:05 +02001289}
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 */
1296void run_posix_cpu_timers(void)
1297{
1298 struct task_struct *tsk = current;
1299
1300 lockdep_assert_irqs_disabled();
1301
1302 /*
Thomas Gleixner1fb497d2020-07-30 12:14:06 +02001303 * 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 Gleixner820903c72020-07-30 12:14:05 +02001310 * 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 Gleixner820903c72020-07-30 12:14:05 +02001316 __run_posix_cpu_timers(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317}
1318
1319/*
Stanislaw Gruszkaf55db602010-03-11 14:04:37 -08001320 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
Frank Mayharf06febc2008-09-12 09:54:39 -07001321 * The tsk->sighand->siglock must be held by the caller.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 */
Thomas Gleixner1b0dd962019-08-21 21:09:09 +02001323void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid,
Frederic Weisbecker858cf3a2017-01-31 04:09:35 +01001324 u64 *newval, u64 *oldval)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
Thomas Gleixner87dc6442019-08-26 20:22:24 +02001326 u64 now, *nextevt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
Thomas Gleixner1b0dd962019-08-21 21:09:09 +02001328 if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED))
Thomas Gleixner692117c2019-08-19 16:31:46 +02001329 return;
1330
Thomas Gleixner87dc6442019-08-26 20:22:24 +02001331 nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt;
Thomas Gleixner1b0dd962019-08-21 21:09:09 +02001332 now = cpu_clock_sample_group(clkid, tsk, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Thomas Gleixner5405d002019-08-21 21:08:59 +02001334 if (oldval) {
Stanislaw Gruszkaf55db602010-03-11 14:04:37 -08001335 /*
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 Schwidefsky64861632011-12-15 14:56:09 +01001340 if (*oldval) {
Frederic Weisbecker858cf3a2017-01-31 04:09:35 +01001341 if (*oldval <= now) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 /* Just about to fire. */
Frederic Weisbecker858cf3a2017-01-31 04:09:35 +01001343 *oldval = TICK_NSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 } else {
Frederic Weisbecker858cf3a2017-01-31 04:09:35 +01001345 *oldval -= now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 }
1347 }
1348
Martin Schwidefsky64861632011-12-15 14:56:09 +01001349 if (!*newval)
Frederic Weisbeckerb7878302015-07-17 22:25:49 +02001350 return;
Frederic Weisbecker858cf3a2017-01-31 04:09:35 +01001351 *newval += now;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 }
1353
1354 /*
Thomas Gleixner1b0dd962019-08-21 21:09:09 +02001355 * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF
1356 * expiry cache is also used by RLIMIT_CPU!.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 */
Thomas Gleixner2bbdbda2019-08-21 21:09:19 +02001358 if (*newval < *nextevt)
Thomas Gleixner87dc6442019-08-26 20:22:24 +02001359 *nextevt = *newval;
Frederic Weisbeckerb7878302015-07-17 22:25:49 +02001360
1361 tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362}
1363
Toyo Abee4b76552006-09-29 02:00:29 -07001364static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
Thomas Gleixner343d8fc2017-06-13 23:29:14 +02001365 const struct timespec64 *rqtp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
Al Viro86a9c442017-06-07 09:42:26 +01001367 struct itimerspec64 it;
Thomas Gleixner343d8fc2017-06-13 23:29:14 +02001368 struct k_itimer timer;
1369 u64 expires;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 int error;
1371
1372 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 * 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 Gleixner60bda032019-08-27 21:31:02 +02001381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 if (!error) {
Deepa Dinamani5f252b32017-03-26 12:04:17 -07001383 static struct itimerspec64 zero_it;
Al Viroedbeda42017-06-07 09:42:31 +01001384 struct restart_block *restart;
Toyo Abee4b76552006-09-29 02:00:29 -07001385
Al Viroedbeda42017-06-07 09:42:31 +01001386 memset(&it, 0, sizeof(it));
Al Viro86a9c442017-06-07 09:42:26 +01001387 it.it_value = *rqtp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
1389 spin_lock_irq(&timer.it_lock);
Al Viro86a9c442017-06-07 09:42:26 +01001390 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 if (error) {
1392 spin_unlock_irq(&timer.it_lock);
1393 return error;
1394 }
1395
1396 while (!signal_pending(current)) {
Thomas Gleixner60bda032019-08-27 21:31:02 +02001397 if (!cpu_timer_getexpires(&timer.it.cpu)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 /*
Stanislaw Gruszkae6c42c22013-02-15 11:08:11 +01001399 * Our timer fired and was reset, below
1400 * deletion can not fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 */
Stanislaw Gruszkae6c42c22013-02-15 11:08:11 +01001402 posix_cpu_timer_del(&timer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 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 Gleixner60bda032019-08-27 21:31:02 +02001419 expires = cpu_timer_getexpires(&timer.it.cpu);
Al Viro86a9c442017-06-07 09:42:26 +01001420 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
Stanislaw Gruszkae6c42c22013-02-15 11:08:11 +01001421 if (!error) {
1422 /*
1423 * Timer is now unarmed, deletion can not fail.
1424 */
1425 posix_cpu_timer_del(&timer);
1426 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 spin_unlock_irq(&timer.it_lock);
1428
Stanislaw Gruszkae6c42c22013-02-15 11:08:11 +01001429 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 Viro86a9c442017-06-07 09:42:26 +01001440 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 /*
1442 * It actually did fire already.
1443 */
1444 return 0;
1445 }
1446
Toyo Abee4b76552006-09-29 02:00:29 -07001447 error = -ERESTART_RESTARTBLOCK;
Al Viro86a9c442017-06-07 09:42:26 +01001448 /*
1449 * Report back to the user the time still remaining.
1450 */
Al Viroedbeda42017-06-07 09:42:31 +01001451 restart = &current->restart_block;
Thomas Gleixner343d8fc2017-06-13 23:29:14 +02001452 restart->nanosleep.expires = expires;
Deepa Dinamanic0edd7c2017-06-24 11:45:06 -07001453 if (restart->nanosleep.type != TT_NONE)
1454 error = nanosleep_copyout(restart, &it.it_value);
Toyo Abee4b76552006-09-29 02:00:29 -07001455 }
1456
1457 return error;
1458}
1459
Thomas Gleixnerbc2c8ea2011-02-01 13:52:12 +00001460static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1461
1462static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
Thomas Gleixner938e7cf2017-06-13 23:34:33 +02001463 const struct timespec64 *rqtp)
Toyo Abee4b76552006-09-29 02:00:29 -07001464{
Andy Lutomirskif56141e2015-02-12 15:01:14 -08001465 struct restart_block *restart_block = &current->restart_block;
Toyo Abee4b76552006-09-29 02:00:29 -07001466 int error;
1467
1468 /*
1469 * Diagnose required errors first.
1470 */
1471 if (CPUCLOCK_PERTHREAD(which_clock) &&
1472 (CPUCLOCK_PID(which_clock) == 0 ||
Eric W. Biederman01a21972017-04-13 10:32:16 -05001473 CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
Toyo Abee4b76552006-09-29 02:00:29 -07001474 return -EINVAL;
1475
Al Viro86a9c442017-06-07 09:42:26 +01001476 error = do_cpu_nanosleep(which_clock, flags, rqtp);
Toyo Abee4b76552006-09-29 02:00:29 -07001477
1478 if (error == -ERESTART_RESTARTBLOCK) {
1479
Thomas Gleixner3751f9f2011-02-01 13:51:20 +00001480 if (flags & TIMER_ABSTIME)
Toyo Abee4b76552006-09-29 02:00:29 -07001481 return -ERESTARTNOHAND;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482
Toyo Abe1711ef32006-09-29 02:00:28 -07001483 restart_block->fn = posix_cpu_nsleep_restart;
Thomas Gleixnerab8177b2011-05-20 13:05:15 +02001484 restart_block->nanosleep.clockid = which_clock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 return error;
1487}
1488
Thomas Gleixnerbc2c8ea2011-02-01 13:52:12 +00001489static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490{
Thomas Gleixnerab8177b2011-05-20 13:05:15 +02001491 clockid_t which_clock = restart_block->nanosleep.clockid;
Deepa Dinamaniad196382017-03-26 12:04:18 -07001492 struct timespec64 t;
Thomas Gleixner97735f22006-01-09 20:52:37 -08001493
Deepa Dinamaniad196382017-03-26 12:04:18 -07001494 t = ns_to_timespec64(restart_block->nanosleep.expires);
Thomas Gleixner97735f22006-01-09 20:52:37 -08001495
Al Viro86a9c442017-06-07 09:42:26 +01001496 return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497}
1498
Nick Desaulniers29f1b2b2017-12-28 22:11:36 -05001499#define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED)
1500#define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
Thomas Gleixnera924b042006-01-09 20:52:27 -08001502static int process_cpu_clock_getres(const clockid_t which_clock,
Deepa Dinamanid2e3e0c2017-03-26 12:04:15 -07001503 struct timespec64 *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504{
1505 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1506}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001507static int process_cpu_clock_get(const clockid_t which_clock,
Deepa Dinamani3c9c12f2017-03-26 12:04:14 -07001508 struct timespec64 *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509{
1510 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1511}
1512static 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 Gleixnera924b042006-01-09 20:52:27 -08001517static int process_cpu_nsleep(const clockid_t which_clock, int flags,
Thomas Gleixner938e7cf2017-06-13 23:34:33 +02001518 const struct timespec64 *rqtp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519{
Al Viro99e6c0e2017-06-07 09:42:30 +01001520 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001522static int thread_cpu_clock_getres(const clockid_t which_clock,
Deepa Dinamanid2e3e0c2017-03-26 12:04:15 -07001523 struct timespec64 *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
1525 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1526}
Thomas Gleixnera924b042006-01-09 20:52:27 -08001527static int thread_cpu_clock_get(const clockid_t which_clock,
Deepa Dinamani3c9c12f2017-03-26 12:04:14 -07001528 struct timespec64 *tp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529{
1530 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1531}
1532static 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 Torvalds1da177e2005-04-16 15:20:36 -07001537
Christoph Hellwigd3ba5a92017-05-26 12:03:11 +03001538const struct k_clock clock_posix_cpu = {
Andrei Vagin819a95f2019-11-12 01:26:54 +00001539 .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 Gleixner19769452011-02-01 13:51:06 +00001548};
1549
Christoph Hellwigd3ba5a92017-05-26 12:03:11 +03001550const struct k_clock clock_process = {
Andrei Vagin819a95f2019-11-12 01:26:54 +00001551 .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 Hellwigd3ba5a92017-05-26 12:03:11 +03001555};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
Christoph Hellwigd3ba5a92017-05-26 12:03:11 +03001557const struct k_clock clock_thread = {
Andrei Vagin819a95f2019-11-12 01:26:54 +00001558 .clock_getres = thread_cpu_clock_getres,
1559 .clock_get_timespec = thread_cpu_clock_get,
1560 .timer_create = thread_cpu_timer_create,
Christoph Hellwigd3ba5a92017-05-26 12:03:11 +03001561};