~hc/RK356X_SDK_RELEASE.git

..	..	@@ -20,11 +20,20 @@
20	20
21	21	static void posix_cpu_timer_rearm(struct k_itimer *timer);
22	22
	23	+void posix_cputimers_group_init(struct posix_cputimers *pct, u64 cpu_limit)
	24	+{
	25	+ posix_cputimers_init(pct);
	26	+ if (cpu_limit != RLIM_INFINITY) {
	27	+ pct->bases[CPUCLOCK_PROF].nextevt = cpu_limit * NSEC_PER_SEC;
	28	+ pct->timers_active = true;
	29	+ }
	30	+}
	31	+
23	32	/*
24	33	* Called after updating RLIMIT_CPU to run cpu timer and update
25		- * tsk->signal->cputime_expires expiration cache if necessary. Needs
26		- * siglock protection since other code may update expiration cache as
27		- * well.
	34	+ * tsk->signal->posix_cputimers.bases[clock].nextevt expiration cache if
	35	+ * necessary. Needs siglock protection since other code may update the
	36	+ * expiration cache as well.
28	37	*/
29	38	void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
30	39	{
..	..	@@ -35,46 +44,87 @@
35	44	spin_unlock_irq(&task->sighand->siglock);
36	45	}
37	46
38		-static int check_clock(const clockid_t which_clock)
	47	+/*
	48	+ * Functions for validating access to tasks.
	49	+ */
	50	+static struct pid *pid_for_clock(const clockid_t clock, bool gettime)
39	51	{
40		- int error = 0;
41		- struct task_struct *p;
42		- const pid_t pid = CPUCLOCK_PID(which_clock);
	52	+ const bool thread = !!CPUCLOCK_PERTHREAD(clock);
	53	+ const pid_t upid = CPUCLOCK_PID(clock);
	54	+ struct pid *pid;
43	55
44		- if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
45		- return -EINVAL;
	56	+ if (CPUCLOCK_WHICH(clock) >= CPUCLOCK_MAX)
	57	+ return NULL;
46	58
47		- if (pid == 0)
48		- return 0;
	59	+ /*
	60	+ * If the encoded PID is 0, then the timer is targeted at current
	61	+ * or the process to which current belongs.
	62	+ */
	63	+ if (upid == 0)
	64	+ return thread ? task_pid(current) : task_tgid(current);
	65	+
	66	+ pid = find_vpid(upid);
	67	+ if (!pid)
	68	+ return NULL;
	69	+
	70	+ if (thread) {
	71	+ struct task_struct *tsk = pid_task(pid, PIDTYPE_PID);
	72	+ return (tsk && same_thread_group(tsk, current)) ? pid : NULL;
	73	+ }
	74	+
	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);
	83	+
	84	+ /*
	85	+ * For processes require that pid identifies a process.
	86	+ */
	87	+ return pid_has_task(pid, PIDTYPE_TGID) ? pid : NULL;
	88	+}
	89	+
	90	+static inline int validate_clock_permissions(const clockid_t clock)
	91	+{
	92	+ int ret;
49	93
50	94	rcu_read_lock();
51		- p = find_task_by_vpid(pid);
52		- if (!p \|\| !(CPUCLOCK_PERTHREAD(which_clock) ?
53		- same_thread_group(p, current) : has_group_leader_pid(p))) {
54		- error = -EINVAL;
55		- }
	95	+ ret = pid_for_clock(clock, false) ? 0 : -EINVAL;
56	96	rcu_read_unlock();
57	97
58		- return error;
	98	+ return ret;
	99	+}
	100	+
	101	+static inline enum pid_type clock_pid_type(const clockid_t clock)
	102	+{
	103	+ return CPUCLOCK_PERTHREAD(clock) ? PIDTYPE_PID : PIDTYPE_TGID;
	104	+}
	105	+
	106	+static inline struct task_struct cpu_timer_task_rcu(struct k_itimer timer)
	107	+{
	108	+ return pid_task(timer->it.cpu.pid, clock_pid_type(timer->it_clock));
59	109	}
60	110
61	111	/*
62	112	* Update expiry time from increment, and increase overrun count,
63	113	* given the current clock sample.
64	114	*/
65		-static void bump_cpu_timer(struct k_itimer *timer, u64 now)
	115	+static u64 bump_cpu_timer(struct k_itimer *timer, u64 now)
66	116	{
	117	+ u64 delta, incr, expires = timer->it.cpu.node.expires;
67	118	int i;
68		- u64 delta, incr;
69	119
70		- if (timer->it.cpu.incr == 0)
71		- return;
	120	+ if (!timer->it_interval)
	121	+ return expires;
72	122
73		- if (now < timer->it.cpu.expires)
74		- return;
	123	+ if (now < expires)
	124	+ return expires;
75	125
76		- incr = timer->it.cpu.incr;
77		- delta = now + incr - timer->it.cpu.expires;
	126	+ incr = timer->it_interval;
	127	+ delta = now + incr - expires;
78	128
79	129	/* Don't use (incr2 < delta), incr2 might overflow. */
80	130	for (i = 0; incr < delta - incr; i++)
..	..	@@ -84,48 +134,26 @@
84	134	if (delta < incr)
85	135	continue;
86	136
87		- timer->it.cpu.expires += incr;
	137	+ timer->it.cpu.node.expires += incr;
88	138	timer->it_overrun += 1LL << i;
89	139	delta -= incr;
90	140	}
	141	+ return timer->it.cpu.node.expires;
91	142	}
92	143
93		-/**
94		- * task_cputime_zero - Check a task_cputime struct for all zero fields.
95		- *
96		- * @cputime: The struct to compare.
97		- *
98		- * Checks @cputime to see if all fields are zero. Returns true if all fields
99		- * are zero, false if any field is nonzero.
100		- */
101		-static inline int task_cputime_zero(const struct task_cputime *cputime)
	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)
102	146	{
103		- if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
104		- return 1;
105		- return 0;
106		-}
107		-
108		-static inline u64 prof_ticks(struct task_struct *p)
109		-{
110		- u64 utime, stime;
111		-
112		- task_cputime(p, &utime, &stime);
113		-
114		- return utime + stime;
115		-}
116		-static inline u64 virt_ticks(struct task_struct *p)
117		-{
118		- u64 utime, stime;
119		-
120		- task_cputime(p, &utime, &stime);
121		-
122		- return utime;
	147	+ return !(~pct->bases[CPUCLOCK_PROF].nextevt \|
	148	+ ~pct->bases[CPUCLOCK_VIRT].nextevt \|
	149	+ ~pct->bases[CPUCLOCK_SCHED].nextevt);
123	150	}
124	151
125	152	static int
126	153	posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
127	154	{
128		- int error = check_clock(which_clock);
	155	+ int error = validate_clock_permissions(which_clock);
	156	+
129	157	if (!error) {
130	158	tp->tv_sec = 0;
131	159	tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
..	..	@@ -142,40 +170,64 @@
142	170	}
143	171
144	172	static int
145		-posix_cpu_clock_set(const clockid_t which_clock, const struct timespec64 *tp)
	173	+posix_cpu_clock_set(const clockid_t clock, const struct timespec64 *tp)
146	174	{
	175	+ int error = validate_clock_permissions(clock);
	176	+
147	177	/*
148	178	* You can never reset a CPU clock, but we check for other errors
149	179	* in the call before failing with EPERM.
150	180	*/
151		- int error = check_clock(which_clock);
152		- if (error == 0) {
153		- error = -EPERM;
154		- }
155		- return error;
	181	+ return error ? : -EPERM;
156	182	}
157	183
158		-
159	184	/*
160		- * Sample a per-thread clock for the given task.
	185	+ * Sample a per-thread clock for the given task. clkid is validated.
161	186	*/
162		-static int cpu_clock_sample(const clockid_t which_clock,
163		- struct task_struct p, u64 sample)
	187	+static u64 cpu_clock_sample(const clockid_t clkid, struct task_struct *p)
164	188	{
165		- switch (CPUCLOCK_WHICH(which_clock)) {
166		- default:
167		- return -EINVAL;
	189	+ u64 utime, stime;
	190	+
	191	+ if (clkid == CPUCLOCK_SCHED)
	192	+ return task_sched_runtime(p);
	193	+
	194	+ task_cputime(p, &utime, &stime);
	195	+
	196	+ switch (clkid) {
168	197	case CPUCLOCK_PROF:
169		- *sample = prof_ticks(p);
170		- break;
	198	+ return utime + stime;
171	199	case CPUCLOCK_VIRT:
172		- *sample = virt_ticks(p);
173		- break;
174		- case CPUCLOCK_SCHED:
175		- *sample = task_sched_runtime(p);
176		- break;
	200	+ return utime;
	201	+ default:
	202	+ WARN_ON_ONCE(1);
177	203	}
178	204	return 0;
	205	+}
	206	+
	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);
179	231	}
180	232
181	233	/*
..	..	@@ -193,29 +245,56 @@
193	245	}
194	246	}
195	247
196		-static void update_gt_cputime(struct task_cputime_atomic cputime_atomic, struct task_cputime sum)
	248	+static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic,
	249	+ struct task_cputime *sum)
197	250	{
198	251	__update_gt_cputime(&cputime_atomic->utime, sum->utime);
199	252	__update_gt_cputime(&cputime_atomic->stime, sum->stime);
200	253	__update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
201	254	}
202	255
203		-/* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */
204		-static inline void sample_cputime_atomic(struct task_cputime *times,
205		- struct task_cputime_atomic *atomic_times)
206		-{
207		- times->utime = atomic64_read(&atomic_times->utime);
208		- times->stime = atomic64_read(&atomic_times->stime);
209		- times->sum_exec_runtime = atomic64_read(&atomic_times->sum_exec_runtime);
210		-}
211		-
212		-void thread_group_cputimer(struct task_struct tsk, struct task_cputime times)
	256	+/**
	257	+ * thread_group_sample_cputime - Sample cputime for a given task
	258	+ * @tsk: Task for which cputime needs to be started
	259	+ * @samples: Storage for time samples
	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	+ */
	267	+void thread_group_sample_cputime(struct task_struct tsk, u64 samples)
213	268	{
214	269	struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
215		- struct task_cputime sum;
	270	+ struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
	271	+
	272	+ WARN_ON_ONCE(!pct->timers_active);
	273	+
	274	+ proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
	275	+}
	276	+
	277	+/**
	278	+ * thread_group_start_cputime - Start cputime and return a sample
	279	+ * @tsk: Task for which cputime needs to be started
	280	+ * @samples: Storage for time samples
	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	+ */
	289	+static void thread_group_start_cputime(struct task_struct tsk, u64 samples)
	290	+{
	291	+ struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
	292	+ struct posix_cputimers *pct = &tsk->signal->posix_cputimers;
216	293
217	294	/* Check if cputimer isn't running. This is accessed without locking. */
218		- if (!READ_ONCE(cputimer->running)) {
	295	+ if (!READ_ONCE(pct->timers_active)) {
	296	+ struct task_cputime sum;
	297	+
219	298	/*
220	299	* The POSIX timer interface allows for absolute time expiry
221	300	* values through the TIMER_ABSTIME flag, therefore we have
..	..	@@ -225,94 +304,70 @@
225	304	update_gt_cputime(&cputimer->cputime_atomic, &sum);
226	305
227	306	/*
228		- * We're setting cputimer->running without a lock. Ensure
229		- * this only gets written to in one operation. We set
230		- * running after update_gt_cputime() as a small optimization,
231		- * but barriers are not required because update_gt_cputime()
	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()
232	311	* can handle concurrent updates.
233	312	*/
234		- WRITE_ONCE(cputimer->running, true);
	313	+ WRITE_ONCE(pct->timers_active, true);
235	314	}
236		- sample_cputime_atomic(times, &cputimer->cputime_atomic);
	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);
237	324	}
238	325
239	326	/*
240		- * Sample a process (thread group) clock for the given group_leader task.
241		- * Must be called with task sighand lock held for safe while_each_thread()
242		- * traversal.
	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
	329	+ * store. Otherwise a full update is required. clkid is already validated.
243	330	*/
244		-static int cpu_clock_sample_group(const clockid_t which_clock,
245		- struct task_struct *p,
246		- u64 *sample)
	331	+static u64 cpu_clock_sample_group(const clockid_t clkid, struct task_struct *p,
	332	+ bool start)
247	333	{
248		- struct task_cputime cputime;
	334	+ struct thread_group_cputimer *cputimer = &p->signal->cputimer;
	335	+ struct posix_cputimers *pct = &p->signal->posix_cputimers;
	336	+ u64 samples[CPUCLOCK_MAX];
249	337
250		- switch (CPUCLOCK_WHICH(which_clock)) {
251		- default:
252		- return -EINVAL;
253		- case CPUCLOCK_PROF:
254		- thread_group_cputime(p, &cputime);
255		- *sample = cputime.utime + cputime.stime;
256		- break;
257		- case CPUCLOCK_VIRT:
258		- thread_group_cputime(p, &cputime);
259		- *sample = cputime.utime;
260		- break;
261		- case CPUCLOCK_SCHED:
262		- thread_group_cputime(p, &cputime);
263		- *sample = cputime.sum_exec_runtime;
264		- break;
265		- }
266		- return 0;
267		-}
268		-
269		-static int posix_cpu_clock_get_task(struct task_struct *tsk,
270		- const clockid_t which_clock,
271		- struct timespec64 *tp)
272		-{
273		- int err = -EINVAL;
274		- u64 rtn;
275		-
276		- if (CPUCLOCK_PERTHREAD(which_clock)) {
277		- if (same_thread_group(tsk, current))
278		- err = cpu_clock_sample(which_clock, tsk, &rtn);
	338	+ if (!READ_ONCE(pct->timers_active)) {
	339	+ if (start)
	340	+ thread_group_start_cputime(p, samples);
	341	+ else
	342	+ __thread_group_cputime(p, samples);
279	343	} else {
280		- if (tsk == current \|\| thread_group_leader(tsk))
281		- err = cpu_clock_sample_group(which_clock, tsk, &rtn);
	344	+ proc_sample_cputime_atomic(&cputimer->cputime_atomic, samples);
282	345	}
283	346
284		- if (!err)
285		- *tp = ns_to_timespec64(rtn);
286		-
287		- return err;
	347	+ return samples[clkid];
288	348	}
289	349
290		-
291		-static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec64 *tp)
	350	+static int posix_cpu_clock_get(const clockid_t clock, struct timespec64 *tp)
292	351	{
293		- const pid_t pid = CPUCLOCK_PID(which_clock);
294		- int err = -EINVAL;
	352	+ const clockid_t clkid = CPUCLOCK_WHICH(clock);
	353	+ struct task_struct *tsk;
	354	+ u64 t;
295	355
296		- if (pid == 0) {
297		- /*
298		- * Special case constant value for our own clocks.
299		- * We don't have to do any lookup to find ourselves.
300		- */
301		- err = posix_cpu_clock_get_task(current, which_clock, tp);
302		- } else {
303		- /*
304		- * Find the given PID, and validate that the caller
305		- * should be able to see it.
306		- */
307		- struct task_struct *p;
308		- rcu_read_lock();
309		- p = find_task_by_vpid(pid);
310		- if (p)
311		- err = posix_cpu_clock_get_task(p, which_clock, tp);
	356	+ rcu_read_lock();
	357	+ tsk = pid_task(pid_for_clock(clock, true), clock_pid_type(clock));
	358	+ if (!tsk) {
312	359	rcu_read_unlock();
	360	+ return -EINVAL;
313	361	}
314	362
315		- return err;
	363	+ if (CPUCLOCK_PERTHREAD(clock))
	364	+ t = cpu_clock_sample(clkid, tsk);
	365	+ else
	366	+ t = cpu_clock_sample_group(clkid, tsk, false);
	367	+ rcu_read_unlock();
	368	+
	369	+ *tp = ns_to_timespec64(t);
	370	+ return 0;
316	371	}
317	372
318	373	/*
..	..	@@ -322,44 +377,32 @@
322	377	*/
323	378	static int posix_cpu_timer_create(struct k_itimer *new_timer)
324	379	{
325		- int ret = 0;
326		- const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
327		- struct task_struct *p;
328		-
329		- if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
330		- return -EINVAL;
331		-
332		- new_timer->kclock = &clock_posix_cpu;
333		-
334		- INIT_LIST_HEAD(&new_timer->it.cpu.entry);
	380	+ static struct lock_class_key posix_cpu_timers_key;
	381	+ struct pid *pid;
335	382
336	383	rcu_read_lock();
337		- if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
338		- if (pid == 0) {
339		- p = current;
340		- } else {
341		- p = find_task_by_vpid(pid);
342		- if (p && !same_thread_group(p, current))
343		- p = NULL;
344		- }
345		- } else {
346		- if (pid == 0) {
347		- p = current->group_leader;
348		- } else {
349		- p = find_task_by_vpid(pid);
350		- if (p && !has_group_leader_pid(p))
351		- p = NULL;
352		- }
	384	+ pid = pid_for_clock(new_timer->it_clock, false);
	385	+ if (!pid) {
	386	+ rcu_read_unlock();
	387	+ return -EINVAL;
353	388	}
354		- new_timer->it.cpu.task = p;
355		- if (p) {
356		- get_task_struct(p);
357		- } else {
358		- ret = -EINVAL;
359		- }
360		- rcu_read_unlock();
361	389
362		- return ret;
	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	+
	401	+ new_timer->kclock = &clock_posix_cpu;
	402	+ timerqueue_init(&new_timer->it.cpu.node);
	403	+ new_timer->it.cpu.pid = get_pid(pid);
	404	+ rcu_read_unlock();
	405	+ return 0;
363	406	}
364	407
365	408	/*
..	..	@@ -370,13 +413,16 @@
370	413	*/
371	414	static int posix_cpu_timer_del(struct k_itimer *timer)
372	415	{
373		- int ret = 0;
374		- unsigned long flags;
	416	+ struct cpu_timer *ctmr = &timer->it.cpu;
375	417	struct sighand_struct *sighand;
376		- struct task_struct *p = timer->it.cpu.task;
	418	+ struct task_struct *p;
	419	+ unsigned long flags;
	420	+ int ret = 0;
377	421
378		- if (WARN_ON_ONCE(!p))
379		- return -EINVAL;
	422	+ rcu_read_lock();
	423	+ p = cpu_timer_task_rcu(timer);
	424	+ if (!p)
	425	+ goto out;
380	426
381	427	/*
382	428	* Protect against sighand release/switch in exit/exec and process/
..	..	@@ -385,44 +431,51 @@
385	431	sighand = lock_task_sighand(p, &flags);
386	432	if (unlikely(sighand == NULL)) {
387	433	/*
388		- * We raced with the reaping of the task.
389		- * The deletion should have cleared us off the list.
	434	+ * This raced with the reaping of the task. The exit cleanup
	435	+ * should have removed this timer from the timer queue.
390	436	*/
391		- WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
	437	+ WARN_ON_ONCE(ctmr->head \|\| timerqueue_node_queued(&ctmr->node));
392	438	} else {
393	439	if (timer->it.cpu.firing)
394	440	ret = TIMER_RETRY;
395	441	else
396		- list_del(&timer->it.cpu.entry);
	442	+ cpu_timer_dequeue(ctmr);
397	443
398	444	unlock_task_sighand(p, &flags);
399	445	}
400	446
	447	+out:
	448	+ rcu_read_unlock();
401	449	if (!ret)
402		- put_task_struct(p);
	450	+ put_pid(ctmr->pid);
403	451
404	452	return ret;
405	453	}
406	454
407		-static void cleanup_timers_list(struct list_head *head)
	455	+static void cleanup_timerqueue(struct timerqueue_head *head)
408	456	{
409		- struct cpu_timer_list timer, next;
	457	+ struct timerqueue_node *node;
	458	+ struct cpu_timer *ctmr;
410	459
411		- list_for_each_entry_safe(timer, next, head, entry)
412		- list_del_init(&timer->entry);
	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	+ }
413	465	}
414	466
415	467	/*
416		- * Clean out CPU timers still ticking when a thread exited. The task
417		- * pointer is cleared, and the expiry time is replaced with the residual
418		- * time for later timer_gettime calls to return.
	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	+ *
419	472	* This must be called with the siglock held.
420	473	*/
421		-static void cleanup_timers(struct list_head *head)
	474	+static void cleanup_timers(struct posix_cputimers *pct)
422	475	{
423		- cleanup_timers_list(head);
424		- cleanup_timers_list(++head);
425		- cleanup_timers_list(++head);
	476	+ cleanup_timerqueue(&pct->bases[CPUCLOCK_PROF].tqhead);
	477	+ cleanup_timerqueue(&pct->bases[CPUCLOCK_VIRT].tqhead);
	478	+ cleanup_timerqueue(&pct->bases[CPUCLOCK_SCHED].tqhead);
426	479	}
427	480
428	481	/*
..	..	@@ -432,76 +485,45 @@
432	485	*/
433	486	void posix_cpu_timers_exit(struct task_struct *tsk)
434	487	{
435		- cleanup_timers(tsk->cpu_timers);
	488	+ cleanup_timers(&tsk->posix_cputimers);
436	489	}
437	490	void posix_cpu_timers_exit_group(struct task_struct *tsk)
438	491	{
439		- cleanup_timers(tsk->signal->cpu_timers);
440		-}
441		-
442		-static inline int expires_gt(u64 expires, u64 new_exp)
443		-{
444		- return expires == 0 \|\| expires > new_exp;
	492	+ cleanup_timers(&tsk->signal->posix_cputimers);
445	493	}
446	494
447	495	/*
448	496	* Insert the timer on the appropriate list before any timers that
449	497	* expire later. This must be called with the sighand lock held.
450	498	*/
451		-static void arm_timer(struct k_itimer *timer)
	499	+static void arm_timer(struct k_itimer timer, struct task_struct p)
452	500	{
453		- struct task_struct *p = timer->it.cpu.task;
454		- struct list_head head, listpos;
455		- struct task_cputime *cputime_expires;
456		- struct cpu_timer_list *const nt = &timer->it.cpu;
457		- struct cpu_timer_list *next;
	501	+ int clkidx = CPUCLOCK_WHICH(timer->it_clock);
	502	+ struct cpu_timer *ctmr = &timer->it.cpu;
	503	+ u64 newexp = cpu_timer_getexpires(ctmr);
	504	+ struct posix_cputimer_base *base;
458	505
459		- if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
460		- head = p->cpu_timers;
461		- cputime_expires = &p->cputime_expires;
462		- } else {
463		- head = p->signal->cpu_timers;
464		- cputime_expires = &p->signal->cputime_expires;
465		- }
466		- head += CPUCLOCK_WHICH(timer->it_clock);
	506	+ if (CPUCLOCK_PERTHREAD(timer->it_clock))
	507	+ base = p->posix_cputimers.bases + clkidx;
	508	+ else
	509	+ base = p->signal->posix_cputimers.bases + clkidx;
467	510
468		- listpos = head;
469		- list_for_each_entry(next, head, entry) {
470		- if (nt->expires < next->expires)
471		- break;
472		- listpos = &next->entry;
473		- }
474		- list_add(&nt->entry, listpos);
	511	+ if (!cpu_timer_enqueue(&base->tqhead, ctmr))
	512	+ return;
475	513
476		- if (listpos == head) {
477		- u64 exp = nt->expires;
	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	+ */
	520	+ if (newexp < base->nextevt)
	521	+ base->nextevt = newexp;
478	522
479		- /*
480		- * We are the new earliest-expiring POSIX 1.b timer, hence
481		- * need to update expiration cache. Take into account that
482		- * for process timers we share expiration cache with itimers
483		- * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
484		- */
485		-
486		- switch (CPUCLOCK_WHICH(timer->it_clock)) {
487		- case CPUCLOCK_PROF:
488		- if (expires_gt(cputime_expires->prof_exp, exp))
489		- cputime_expires->prof_exp = exp;
490		- break;
491		- case CPUCLOCK_VIRT:
492		- if (expires_gt(cputime_expires->virt_exp, exp))
493		- cputime_expires->virt_exp = exp;
494		- break;
495		- case CPUCLOCK_SCHED:
496		- if (expires_gt(cputime_expires->sched_exp, exp))
497		- cputime_expires->sched_exp = exp;
498		- break;
499		- }
500		- if (CPUCLOCK_PERTHREAD(timer->it_clock))
501		- tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
502		- else
503		- tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
504		- }
	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);
505	527	}
506	528
507	529	/*
..	..	@@ -509,24 +531,26 @@
509	531	*/
510	532	static void cpu_timer_fire(struct k_itimer *timer)
511	533	{
	534	+ struct cpu_timer *ctmr = &timer->it.cpu;
	535	+
512	536	if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
513	537	/*
514	538	* User don't want any signal.
515	539	*/
516		- timer->it.cpu.expires = 0;
	540	+ cpu_timer_setexpires(ctmr, 0);
517	541	} else if (unlikely(timer->sigq == NULL)) {
518	542	/*
519	543	* This a special case for clock_nanosleep,
520	544	* not a normal timer from sys_timer_create.
521	545	*/
522	546	wake_up_process(timer->it_process);
523		- timer->it.cpu.expires = 0;
524		- } else if (timer->it.cpu.incr == 0) {
	547	+ cpu_timer_setexpires(ctmr, 0);
	548	+ } else if (!timer->it_interval) {
525	549	/*
526	550	* One-shot timer. Clear it as soon as it's fired.
527	551	*/
528	552	posix_timer_event(timer, 0);
529		- timer->it.cpu.expires = 0;
	553	+ cpu_timer_setexpires(ctmr, 0);
530	554	} else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
531	555	/*
532	556	* The signal did not get queued because the signal
..	..	@@ -540,33 +564,6 @@
540	564	}
541	565
542	566	/*
543		- * Sample a process (thread group) timer for the given group_leader task.
544		- * Must be called with task sighand lock held for safe while_each_thread()
545		- * traversal.
546		- */
547		-static int cpu_timer_sample_group(const clockid_t which_clock,
548		- struct task_struct p, u64 sample)
549		-{
550		- struct task_cputime cputime;
551		-
552		- thread_group_cputimer(p, &cputime);
553		- switch (CPUCLOCK_WHICH(which_clock)) {
554		- default:
555		- return -EINVAL;
556		- case CPUCLOCK_PROF:
557		- *sample = cputime.utime + cputime.stime;
558		- break;
559		- case CPUCLOCK_VIRT:
560		- *sample = cputime.utime;
561		- break;
562		- case CPUCLOCK_SCHED:
563		- *sample = cputime.sum_exec_runtime;
564		- break;
565		- }
566		- return 0;
567		-}
568		-
569		-/*
570	567	* Guts of sys_timer_settime for CPU timers.
571	568	* This is called with the timer locked and interrupts disabled.
572	569	* If we return TIMER_RETRY, it's necessary to release the timer's lock
..	..	@@ -575,14 +572,24 @@
575	572	static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
576	573	struct itimerspec64 new, struct itimerspec64 old)
577	574	{
578		- unsigned long flags;
579		- struct sighand_struct *sighand;
580		- struct task_struct *p = timer->it.cpu.task;
	575	+ clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
581	576	u64 old_expires, new_expires, old_incr, val;
582		- int ret;
	577	+ struct cpu_timer *ctmr = &timer->it.cpu;
	578	+ struct sighand_struct *sighand;
	579	+ struct task_struct *p;
	580	+ unsigned long flags;
	581	+ int ret = 0;
583	582
584		- if (WARN_ON_ONCE(!p))
585		- return -EINVAL;
	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	+ }
586	593
587	594	/*
588	595	* Use the to_ktime conversion because that clamps the maximum
..	..	@@ -600,21 +607,22 @@
600	607	* longer get any information about it at all.
601	608	*/
602	609	if (unlikely(sighand == NULL)) {
	610	+ rcu_read_unlock();
603	611	return -ESRCH;
604	612	}
605	613
606	614	/*
607	615	* Disarm any old timer after extracting its expiry time.
608	616	*/
	617	+ old_incr = timer->it_interval;
	618	+ old_expires = cpu_timer_getexpires(ctmr);
609	619
610		- ret = 0;
611		- old_incr = timer->it.cpu.incr;
612		- old_expires = timer->it.cpu.expires;
613	620	if (unlikely(timer->it.cpu.firing)) {
614	621	timer->it.cpu.firing = -1;
615	622	ret = TIMER_RETRY;
616		- } else
617		- list_del_init(&timer->it.cpu.entry);
	623	+ } else {
	624	+ cpu_timer_dequeue(ctmr);
	625	+ }
618	626
619	627	/*
620	628	* We need to sample the current value to convert the new
..	..	@@ -624,11 +632,10 @@
624	632	* times (in arm_timer). With an absolute time, we must
625	633	* check if it's already passed. In short, we need a sample.
626	634	*/
627		- if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
628		- cpu_clock_sample(timer->it_clock, p, &val);
629		- } else {
630		- cpu_timer_sample_group(timer->it_clock, p, &val);
631		- }
	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);
632	639
633	640	if (old) {
634	641	if (old_expires == 0) {
..	..	@@ -636,18 +643,16 @@
636	643	old->it_value.tv_nsec = 0;
637	644	} else {
638	645	/*
639		- * Update the timer in case it has
640		- * overrun already. If it has,
641		- * we'll report it as having overrun
642		- * and with the next reloaded timer
643		- * already ticking, though we are
644		- * swallowing that pending
645		- * notification here to install the
646		- * new setting.
	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.
647	651	*/
648		- bump_cpu_timer(timer, val);
649		- if (val < timer->it.cpu.expires) {
650		- old_expires = timer->it.cpu.expires - val;
	652	+ u64 exp = bump_cpu_timer(timer, val);
	653	+
	654	+ if (val < exp) {
	655	+ old_expires = exp - val;
651	656	old->it_value = ns_to_timespec64(old_expires);
652	657	} else {
653	658	old->it_value.tv_nsec = 1;
..	..	@@ -676,9 +681,9 @@
676	681	* For a timer with no notification action, we don't actually
677	682	* arm the timer (we'll just fake it for timer_gettime).
678	683	*/
679		- timer->it.cpu.expires = new_expires;
	684	+ cpu_timer_setexpires(ctmr, new_expires);
680	685	if (new_expires != 0 && val < new_expires) {
681		- arm_timer(timer);
	686	+ arm_timer(timer, p);
682	687	}
683	688
684	689	unlock_task_sighand(p, &flags);
..	..	@@ -686,8 +691,7 @@
686	691	* Install the new reload setting, and
687	692	* set up the signal and overrun bookkeeping.
688	693	*/
689		- timer->it.cpu.incr = timespec64_to_ns(&new->it_interval);
690		- timer->it_interval = ns_to_ktime(timer->it.cpu.incr);
	694	+ timer->it_interval = timespec64_to_ktime(new->it_interval);
691	695
692	696	/*
693	697	* This acts as a modification timestamp for the timer,
..	..	@@ -710,6 +714,7 @@
710	714
711	715	ret = 0;
712	716	out:
	717	+ rcu_read_unlock();
713	718	if (old)
714	719	old->it_interval = ns_to_timespec64(old_incr);
715	720
..	..	@@ -718,51 +723,34 @@
718	723
719	724	static void posix_cpu_timer_get(struct k_itimer timer, struct itimerspec64 itp)
720	725	{
721		- struct task_struct *p = timer->it.cpu.task;
722		- u64 now;
	726	+ clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
	727	+ struct cpu_timer *ctmr = &timer->it.cpu;
	728	+ u64 now, expires = cpu_timer_getexpires(ctmr);
	729	+ struct task_struct *p;
723	730
724		- if (WARN_ON_ONCE(!p))
725		- return;
	731	+ rcu_read_lock();
	732	+ p = cpu_timer_task_rcu(timer);
	733	+ if (!p)
	734	+ goto out;
726	735
727	736	/*
728	737	* Easy part: convert the reload time.
729	738	*/
730		- itp->it_interval = ns_to_timespec64(timer->it.cpu.incr);
	739	+ itp->it_interval = ktime_to_timespec64(timer->it_interval);
731	740
732		- if (!timer->it.cpu.expires)
733		- return;
	741	+ if (!expires)
	742	+ goto out;
734	743
735	744	/*
736	745	* Sample the clock to take the difference with the expiry time.
737	746	*/
738		- if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
739		- cpu_clock_sample(timer->it_clock, p, &now);
740		- } else {
741		- struct sighand_struct *sighand;
742		- unsigned long flags;
	747	+ if (CPUCLOCK_PERTHREAD(timer->it_clock))
	748	+ now = cpu_clock_sample(clkid, p);
	749	+ else
	750	+ now = cpu_clock_sample_group(clkid, p, false);
743	751
744		- /*
745		- * Protect against sighand release/switch in exit/exec and
746		- * also make timer sampling safe if it ends up calling
747		- * thread_group_cputime().
748		- */
749		- sighand = lock_task_sighand(p, &flags);
750		- if (unlikely(sighand == NULL)) {
751		- /*
752		- * The process has been reaped.
753		- * We can't even collect a sample any more.
754		- * Call the timer disarmed, nothing else to do.
755		- */
756		- timer->it.cpu.expires = 0;
757		- return;
758		- } else {
759		- cpu_timer_sample_group(timer->it_clock, p, &now);
760		- unlock_task_sighand(p, &flags);
761		- }
762		- }
763		-
764		- if (now < timer->it.cpu.expires) {
765		- itp->it_value = ns_to_timespec64(timer->it.cpu.expires - now);
	752	+ if (now < expires) {
	753	+ itp->it_value = ns_to_timespec64(expires - now);
766	754	} else {
767	755	/*
768	756	* The timer should have expired already, but the firing
..	..	@@ -771,28 +759,48 @@
771	759	itp->it_value.tv_nsec = 1;
772	760	itp->it_value.tv_sec = 0;
773	761	}
	762	+out:
	763	+ rcu_read_unlock();
774	764	}
775	765
776		-static unsigned long long
777		-check_timers_list(struct list_head *timers,
778		- struct list_head *firing,
779		- unsigned long long curr)
	766	+#define MAX_COLLECTED 20
	767	+
	768	+static u64 collect_timerqueue(struct timerqueue_head *head,
	769	+ struct list_head *firing, u64 now)
780	770	{
781		- int maxfire = 20;
	771	+ struct timerqueue_node *next;
	772	+ int i = 0;
782	773
783		- while (!list_empty(timers)) {
784		- struct cpu_timer_list *t;
	774	+ while ((next = timerqueue_getnext(head))) {
	775	+ struct cpu_timer *ctmr;
	776	+ u64 expires;
785	777
786		- t = list_first_entry(timers, struct cpu_timer_list, entry);
	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;
787	783
788		- if (!--maxfire \|\| curr < t->expires)
789		- return t->expires;
790		-
791		- t->firing = 1;
792		- list_move_tail(&t->entry, firing);
	784	+ ctmr->firing = 1;
	785	+ /* See posix_cpu_timer_wait_running() */
	786	+ rcu_assign_pointer(ctmr->handling, current);
	787	+ cpu_timer_dequeue(ctmr);
	788	+ list_add_tail(&ctmr->elist, firing);
793	789	}
794	790
795		- return 0;
	791	+ return U64_MAX;
	792	+}
	793	+
	794	+static void collect_posix_cputimers(struct posix_cputimers pct, u64 samples,
	795	+ struct list_head *firing)
	796	+{
	797	+ struct posix_cputimer_base *base = pct->bases;
	798	+ int i;
	799	+
	800	+ for (i = 0; i < CPUCLOCK_MAX; i++, base++) {
	801	+ base->nextevt = collect_timerqueue(&base->tqhead, firing,
	802	+ samples[i]);
	803	+ }
796	804	}
797	805
798	806	static inline void check_dl_overrun(struct task_struct *tsk)
..	..	@@ -803,6 +811,20 @@
803	811	}
804	812	}
805	813
	814	+static bool check_rlimit(u64 time, u64 limit, int signo, bool rt, bool hard)
	815	+{
	816	+ if (time < limit)
	817	+ return false;
	818	+
	819	+ if (print_fatal_signals) {
	820	+ pr_info("%s Watchdog Timeout (%s): %s[%d]\n",
	821	+ rt ? "RT" : "CPU", hard ? "hard" : "soft",
	822	+ current->comm, task_pid_nr(current));
	823	+ }
	824	+ __group_send_sig_info(signo, SEND_SIG_PRIV, current);
	825	+ return true;
	826	+}
	827	+
806	828	/*
807	829	* Check for any per-thread CPU timers that have fired and move them off
808	830	* the tsk->cpu_timers[N] list onto the firing list. Here we update the
..	..	@@ -811,76 +833,50 @@
811	833	static void check_thread_timers(struct task_struct *tsk,
812	834	struct list_head *firing)
813	835	{
814		- struct list_head *timers = tsk->cpu_timers;
815		- struct task_cputime *tsk_expires = &tsk->cputime_expires;
816		- u64 expires;
	836	+ struct posix_cputimers *pct = &tsk->posix_cputimers;
	837	+ u64 samples[CPUCLOCK_MAX];
817	838	unsigned long soft;
818	839
819	840	if (dl_task(tsk))
820	841	check_dl_overrun(tsk);
821	842
822		- /*
823		- * If cputime_expires is zero, then there are no active
824		- * per thread CPU timers.
825		- */
826		- if (task_cputime_zero(&tsk->cputime_expires))
	843	+ if (expiry_cache_is_inactive(pct))
827	844	return;
828	845
829		- expires = check_timers_list(timers, firing, prof_ticks(tsk));
830		- tsk_expires->prof_exp = expires;
831		-
832		- expires = check_timers_list(++timers, firing, virt_ticks(tsk));
833		- tsk_expires->virt_exp = expires;
834		-
835		- tsk_expires->sched_exp = check_timers_list(++timers, firing,
836		- tsk->se.sum_exec_runtime);
	846	+ task_sample_cputime(tsk, samples);
	847	+ collect_posix_cputimers(pct, samples, firing);
837	848
838	849	/*
839	850	* Check for the special case thread timers.
840	851	*/
841	852	soft = task_rlimit(tsk, RLIMIT_RTTIME);
842	853	if (soft != RLIM_INFINITY) {
	854	+ /* Task RT timeout is accounted in jiffies. RTTIME is usec */
	855	+ unsigned long rttime = tsk->rt.timeout * (USEC_PER_SEC / HZ);
843	856	unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
844	857
	858	+ /* At the hard limit, send SIGKILL. No further action. */
845	859	if (hard != RLIM_INFINITY &&
846		- tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
847		- /*
848		- * At the hard limit, we just die.
849		- * No need to calculate anything else now.
850		- */
851		- if (print_fatal_signals) {
852		- pr_info("CPU Watchdog Timeout (hard): %s[%d]\n",
853		- tsk->comm, task_pid_nr(tsk));
854		- }
855		- __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
	860	+ check_rlimit(rttime, hard, SIGKILL, true, true))
856	861	return;
857		- }
858		- if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
859		- /*
860		- * At the soft limit, send a SIGXCPU every second.
861		- */
862		- if (soft < hard) {
863		- soft += USEC_PER_SEC;
864		- tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur =
865		- soft;
866		- }
867		- if (print_fatal_signals) {
868		- pr_info("RT Watchdog Timeout (soft): %s[%d]\n",
869		- tsk->comm, task_pid_nr(tsk));
870		- }
871		- __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
	862	+
	863	+ /* At the soft limit, send a SIGXCPU every second */
	864	+ if (check_rlimit(rttime, soft, SIGXCPU, true, false)) {
	865	+ soft += USEC_PER_SEC;
	866	+ tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur = soft;
872	867	}
873	868	}
874		- if (task_cputime_zero(tsk_expires))
	869	+
	870	+ if (expiry_cache_is_inactive(pct))
875	871	tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
876	872	}
877	873
878	874	static inline void stop_process_timers(struct signal_struct *sig)
879	875	{
880		- struct thread_group_cputimer *cputimer = &sig->cputimer;
	876	+ struct posix_cputimers *pct = &sig->posix_cputimers;
881	877
882		- /* Turn off cputimer->running. This is done without locking. */
883		- WRITE_ONCE(cputimer->running, false);
	878	+ /* Turn off the active flag. This is done without locking. */
	879	+ WRITE_ONCE(pct->timers_active, false);
884	880	tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
885	881	}
886	882
..	..	@@ -902,7 +898,7 @@
902	898	__group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
903	899	}
904	900
905		- if (it->expires && (!expires \|\| it->expires < expires))
	901	+ if (it->expires && it->expires < *expires)
906	902	*expires = it->expires;
907	903	}
908	904
..	..	@@ -915,90 +911,69 @@
915	911	struct list_head *firing)
916	912	{
917	913	struct signal_struct *const sig = tsk->signal;
918		- u64 utime, ptime, virt_expires, prof_expires;
919		- u64 sum_sched_runtime, sched_expires;
920		- struct list_head *timers = sig->cpu_timers;
921		- struct task_cputime cputime;
	914	+ struct posix_cputimers *pct = &sig->posix_cputimers;
	915	+ u64 samples[CPUCLOCK_MAX];
922	916	unsigned long soft;
923	917
924		- if (dl_task(tsk))
925		- check_dl_overrun(tsk);
926		-
927	918	/*
928		- * If cputimer is not running, then there are no active
929		- * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU).
	919	+ * If there are no active process wide timers (POSIX 1.b, itimers,
	920	+ * RLIMIT_CPU) nothing to check. Also skip the process wide timer
	921	+ * processing when there is already another task handling them.
930	922	*/
931		- if (!READ_ONCE(tsk->signal->cputimer.running))
	923	+ if (!READ_ONCE(pct->timers_active) \|\| pct->expiry_active)
932	924	return;
933	925
934		- /*
	926	+ /*
935	927	* Signify that a thread is checking for process timers.
936	928	* Write access to this field is protected by the sighand lock.
937	929	*/
938		- sig->cputimer.checking_timer = true;
	930	+ pct->expiry_active = true;
939	931
940	932	/*
941		- * Collect the current process totals.
	933	+ * Collect the current process totals. Group accounting is active
	934	+ * so the sample can be taken directly.
942	935	*/
943		- thread_group_cputimer(tsk, &cputime);
944		- utime = cputime.utime;
945		- ptime = utime + cputime.stime;
946		- sum_sched_runtime = cputime.sum_exec_runtime;
947		-
948		- prof_expires = check_timers_list(timers, firing, ptime);
949		- virt_expires = check_timers_list(++timers, firing, utime);
950		- sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
	936	+ proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic, samples);
	937	+ collect_posix_cputimers(pct, samples, firing);
951	938
952	939	/*
953	940	* Check for the special case process timers.
954	941	*/
955		- check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
956		- SIGPROF);
957		- check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
958		- SIGVTALRM);
	942	+ check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF],
	943	+ &pct->bases[CPUCLOCK_PROF].nextevt,
	944	+ samples[CPUCLOCK_PROF], SIGPROF);
	945	+ check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT],
	946	+ &pct->bases[CPUCLOCK_VIRT].nextevt,
	947	+ samples[CPUCLOCK_VIRT], SIGVTALRM);
	948	+
959	949	soft = task_rlimit(tsk, RLIMIT_CPU);
960	950	if (soft != RLIM_INFINITY) {
961		- unsigned long psecs = div_u64(ptime, NSEC_PER_SEC);
	951	+ /* RLIMIT_CPU is in seconds. Samples are nanoseconds */
962	952	unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
963		- u64 x;
964		- if (psecs >= hard) {
965		- /*
966		- * At the hard limit, we just die.
967		- * No need to calculate anything else now.
968		- */
969		- if (print_fatal_signals) {
970		- pr_info("RT Watchdog Timeout (hard): %s[%d]\n",
971		- tsk->comm, task_pid_nr(tsk));
972		- }
973		- __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
	953	+ u64 ptime = samples[CPUCLOCK_PROF];
	954	+ u64 softns = (u64)soft * NSEC_PER_SEC;
	955	+ u64 hardns = (u64)hard * NSEC_PER_SEC;
	956	+
	957	+ /* At the hard limit, send SIGKILL. No further action. */
	958	+ if (hard != RLIM_INFINITY &&
	959	+ check_rlimit(ptime, hardns, SIGKILL, false, true))
974	960	return;
	961	+
	962	+ /* At the soft limit, send a SIGXCPU every second */
	963	+ if (check_rlimit(ptime, softns, SIGXCPU, false, false)) {
	964	+ sig->rlim[RLIMIT_CPU].rlim_cur = soft + 1;
	965	+ softns += NSEC_PER_SEC;
975	966	}
976		- if (psecs >= soft) {
977		- /*
978		- * At the soft limit, send a SIGXCPU every second.
979		- */
980		- if (print_fatal_signals) {
981		- pr_info("CPU Watchdog Timeout (soft): %s[%d]\n",
982		- tsk->comm, task_pid_nr(tsk));
983		- }
984		- __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
985		- if (soft < hard) {
986		- soft++;
987		- sig->rlim[RLIMIT_CPU].rlim_cur = soft;
988		- }
989		- }
990		- x = soft * NSEC_PER_SEC;
991		- if (!prof_expires \|\| x < prof_expires)
992		- prof_expires = x;
	967	+
	968	+ /* Update the expiry cache */
	969	+ if (softns < pct->bases[CPUCLOCK_PROF].nextevt)
	970	+ pct->bases[CPUCLOCK_PROF].nextevt = softns;
993	971	}
994	972
995		- sig->cputime_expires.prof_exp = prof_expires;
996		- sig->cputime_expires.virt_exp = virt_expires;
997		- sig->cputime_expires.sched_exp = sched_expires;
998		- if (task_cputime_zero(&sig->cputime_expires))
	973	+ if (expiry_cache_is_inactive(pct))
999	974	stop_process_timers(sig);
1000	975
1001		- sig->cputimer.checking_timer = false;
	976	+ pct->expiry_active = false;
1002	977	}
1003	978
1004	979	/*
..	..	@@ -1007,78 +982,60 @@
1007	982	*/
1008	983	static void posix_cpu_timer_rearm(struct k_itimer *timer)
1009	984	{
1010		- struct task_struct *p = timer->it.cpu.task;
	985	+ clockid_t clkid = CPUCLOCK_WHICH(timer->it_clock);
	986	+ struct task_struct *p;
1011	987	struct sighand_struct *sighand;
1012	988	unsigned long flags;
1013	989	u64 now;
1014	990
1015		- if (WARN_ON_ONCE(!p))
1016		- return;
	991	+ rcu_read_lock();
	992	+ p = cpu_timer_task_rcu(timer);
	993	+ if (!p)
	994	+ goto out;
	995	+
	996	+ /* Protect timer list r/w in arm_timer() */
	997	+ sighand = lock_task_sighand(p, &flags);
	998	+ if (unlikely(sighand == NULL))
	999	+ goto out;
1017	1000
1018	1001	/*
1019	1002	* Fetch the current sample and update the timer's expiry time.
1020	1003	*/
1021		- if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1022		- cpu_clock_sample(timer->it_clock, p, &now);
1023		- bump_cpu_timer(timer, now);
1024		- if (unlikely(p->exit_state))
1025		- return;
	1004	+ if (CPUCLOCK_PERTHREAD(timer->it_clock))
	1005	+ now = cpu_clock_sample(clkid, p);
	1006	+ else
	1007	+ now = cpu_clock_sample_group(clkid, p, true);
1026	1008
1027		- /* Protect timer list r/w in arm_timer() */
1028		- sighand = lock_task_sighand(p, &flags);
1029		- if (!sighand)
1030		- return;
1031		- } else {
1032		- /*
1033		- * Protect arm_timer() and timer sampling in case of call to
1034		- * thread_group_cputime().
1035		- */
1036		- sighand = lock_task_sighand(p, &flags);
1037		- if (unlikely(sighand == NULL)) {
1038		- /*
1039		- * The process has been reaped.
1040		- * We can't even collect a sample any more.
1041		- */
1042		- timer->it.cpu.expires = 0;
1043		- return;
1044		- } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1045		- /* If the process is dying, no need to rearm */
1046		- goto unlock;
1047		- }
1048		- cpu_timer_sample_group(timer->it_clock, p, &now);
1049		- bump_cpu_timer(timer, now);
1050		- /* Leave the sighand locked for the call below. */
1051		- }
	1009	+ bump_cpu_timer(timer, now);
1052	1010
1053	1011	/*
1054	1012	* Now re-arm for the new expiry time.
1055	1013	*/
1056		- arm_timer(timer);
1057		-unlock:
	1014	+ arm_timer(timer, p);
1058	1015	unlock_task_sighand(p, &flags);
	1016	+out:
	1017	+ rcu_read_unlock();
1059	1018	}
1060	1019
1061	1020	/**
1062		- * task_cputime_expired - Compare two task_cputime entities.
	1021	+ * task_cputimers_expired - Check whether posix CPU timers are expired
1063	1022	*
1064		- * @sample: The task_cputime structure to be checked for expiration.
1065		- * @expires: Expiration times, against which @sample will be checked.
	1023	+ * @samples: Array of current samples for the CPUCLOCK clocks
	1024	+ * @pct: Pointer to a posix_cputimers container
1066	1025	*
1067		- * Checks @sample against @expires to see if any field of @sample has expired.
1068		- * Returns true if any field of the former is greater than the corresponding
1069		- * field of the latter if the latter field is set. Otherwise returns false.
	1026	+ * Returns true if any member of @samples is greater than the corresponding
	1027	+ * member of @pct->bases[CLK].nextevt. False otherwise
1070	1028	*/
1071		-static inline int task_cputime_expired(const struct task_cputime *sample,
1072		- const struct task_cputime *expires)
	1029	+static inline bool
	1030	+task_cputimers_expired(const u64 samples, struct posix_cputimers pct)
1073	1031	{
1074		- if (expires->utime && sample->utime >= expires->utime)
1075		- return 1;
1076		- if (expires->stime && sample->utime + sample->stime >= expires->stime)
1077		- return 1;
1078		- if (expires->sum_exec_runtime != 0 &&
1079		- sample->sum_exec_runtime >= expires->sum_exec_runtime)
1080		- return 1;
1081		- return 0;
	1032	+ int i;
	1033	+
	1034	+ for (i = 0; i < CPUCLOCK_MAX; i++) {
	1035	+ if (samples[i] >= pct->bases[i].nextevt)
	1036	+ return true;
	1037	+ }
	1038	+ return false;
1082	1039	}
1083	1040
1084	1041	/**
..	..	@@ -1091,83 +1048,279 @@
1091	1048	* timers and compare them with the corresponding expiration times. Return
1092	1049	* true if a timer has expired, else return false.
1093	1050	*/
1094		-static inline int fastpath_timer_check(struct task_struct *tsk)
	1051	+static inline bool fastpath_timer_check(struct task_struct *tsk)
1095	1052	{
	1053	+ struct posix_cputimers *pct = &tsk->posix_cputimers;
1096	1054	struct signal_struct *sig;
1097	1055
1098		- if (!task_cputime_zero(&tsk->cputime_expires)) {
1099		- struct task_cputime task_sample;
	1056	+ if (!expiry_cache_is_inactive(pct)) {
	1057	+ u64 samples[CPUCLOCK_MAX];
1100	1058
1101		- task_cputime(tsk, &task_sample.utime, &task_sample.stime);
1102		- task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
1103		- if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1104		- return 1;
	1059	+ task_sample_cputime(tsk, samples);
	1060	+ if (task_cputimers_expired(samples, pct))
	1061	+ return true;
1105	1062	}
1106	1063
1107	1064	sig = tsk->signal;
	1065	+ pct = &sig->posix_cputimers;
1108	1066	/*
1109		- * Check if thread group timers expired when the cputimer is
1110		- * running and no other thread in the group is already checking
1111		- * for thread group cputimers. These fields are read without the
1112		- * sighand lock. However, this is fine because this is meant to
1113		- * be a fastpath heuristic to determine whether we should try to
1114		- * acquire the sighand lock to check/handle timers.
	1067	+ * Check if thread group timers expired when timers are active and
	1068	+ * no other thread in the group is already handling expiry for
	1069	+ * thread group cputimers. These fields are read without the
	1070	+ * sighand lock. However, this is fine because this is meant to be
	1071	+ * a fastpath heuristic to determine whether we should try to
	1072	+ * acquire the sighand lock to handle timer expiry.
1115	1073	*
1116		- * In the worst case scenario, if 'running' or 'checking_timer' gets
1117		- * set but the current thread doesn't see the change yet, we'll wait
1118		- * until the next thread in the group gets a scheduler interrupt to
1119		- * handle the timer. This isn't an issue in practice because these
1120		- * types of delays with signals actually getting sent are expected.
	1074	+ * In the worst case scenario, if concurrently timers_active is set
	1075	+ * or expiry_active is cleared, but the current thread doesn't see
	1076	+ * the change yet, the timer checks are delayed until the next
	1077	+ * thread in the group gets a scheduler interrupt to handle the
	1078	+ * timer. This isn't an issue in practice because these types of
	1079	+ * delays with signals actually getting sent are expected.
1121	1080	*/
1122		- if (READ_ONCE(sig->cputimer.running) &&
1123		- !READ_ONCE(sig->cputimer.checking_timer)) {
1124		- struct task_cputime group_sample;
	1081	+ if (READ_ONCE(pct->timers_active) && !READ_ONCE(pct->expiry_active)) {
	1082	+ u64 samples[CPUCLOCK_MAX];
1125	1083
1126		- sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
	1084	+ proc_sample_cputime_atomic(&sig->cputimer.cputime_atomic,
	1085	+ samples);
1127	1086
1128		- if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1129		- return 1;
	1087	+ if (task_cputimers_expired(samples, pct))
	1088	+ return true;
1130	1089	}
1131	1090
1132	1091	if (dl_task(tsk) && tsk->dl.dl_overrun)
1133		- return 1;
	1092	+ return true;
1134	1093
1135		- return 0;
	1094	+ return false;
	1095	+}
	1096	+
	1097	+static void handle_posix_cpu_timers(struct task_struct *tsk);
	1098	+
	1099	+#ifdef CONFIG_POSIX_CPU_TIMERS_TASK_WORK
	1100	+static void posix_cpu_timers_work(struct callback_head *work)
	1101	+{
	1102	+ struct posix_cputimers_work cw = container_of(work, typeof(cw), work);
	1103	+
	1104	+ mutex_lock(&cw->mutex);
	1105	+ handle_posix_cpu_timers(current);
	1106	+ mutex_unlock(&cw->mutex);
1136	1107	}
1137	1108
1138	1109	/*
1139		- * This is called from the timer interrupt handler. The irq handler has
1140		- * already updated our counts. We need to check if any timers fire now.
1141		- * Interrupts are disabled.
	1110	+ * Invoked from the posix-timer core when a cancel operation failed because
	1111	+ * the timer is marked firing. The caller holds rcu_read_lock(), which
	1112	+ * protects the timer and the task which is expiring it from being freed.
1142	1113	*/
1143		-void run_posix_cpu_timers(struct task_struct *tsk)
	1114	+static void posix_cpu_timer_wait_running(struct k_itimer *timr)
1144	1115	{
1145		- LIST_HEAD(firing);
1146		- struct k_itimer timer, next;
1147		- unsigned long flags;
	1116	+ struct task_struct *tsk = rcu_dereference(timr->it.cpu.handling);
1148	1117
1149		- lockdep_assert_irqs_disabled();
	1118	+ /* Has the handling task completed expiry already? */
	1119	+ if (!tsk)
	1120	+ return;
	1121	+
	1122	+ /* Ensure that the task cannot go away */
	1123	+ get_task_struct(tsk);
	1124	+ /* Now drop the RCU protection so the mutex can be locked */
	1125	+ rcu_read_unlock();
	1126	+ /* Wait on the expiry mutex */
	1127	+ mutex_lock(&tsk->posix_cputimers_work.mutex);
	1128	+ /* Release it immediately again. */
	1129	+ mutex_unlock(&tsk->posix_cputimers_work.mutex);
	1130	+ /* Drop the task reference. */
	1131	+ put_task_struct(tsk);
	1132	+ /* Relock RCU so the callsite is balanced */
	1133	+ rcu_read_lock();
	1134	+}
	1135	+
	1136	+static void posix_cpu_timer_wait_running_nsleep(struct k_itimer *timr)
	1137	+{
	1138	+ /* Ensure that timr->it.cpu.handling task cannot go away */
	1139	+ rcu_read_lock();
	1140	+ spin_unlock_irq(&timr->it_lock);
	1141	+ posix_cpu_timer_wait_running(timr);
	1142	+ rcu_read_unlock();
	1143	+ /* @timr is on stack and is valid */
	1144	+ spin_lock_irq(&timr->it_lock);
	1145	+}
	1146	+
	1147	+/*
	1148	+ * Clear existing posix CPU timers task work.
	1149	+ */
	1150	+void clear_posix_cputimers_work(struct task_struct *p)
	1151	+{
	1152	+ /*
	1153	+ * A copied work entry from the old task is not meaningful, clear it.
	1154	+ * N.B. init_task_work will not do this.
	1155	+ */
	1156	+ memset(&p->posix_cputimers_work.work, 0,
	1157	+ sizeof(p->posix_cputimers_work.work));
	1158	+ init_task_work(&p->posix_cputimers_work.work,
	1159	+ posix_cpu_timers_work);
	1160	+ mutex_init(&p->posix_cputimers_work.mutex);
	1161	+ p->posix_cputimers_work.scheduled = false;
	1162	+}
	1163	+
	1164	+/*
	1165	+ * Initialize posix CPU timers task work in init task. Out of line to
	1166	+ * keep the callback static and to avoid header recursion hell.
	1167	+ */
	1168	+void __init posix_cputimers_init_work(void)
	1169	+{
	1170	+ clear_posix_cputimers_work(current);
	1171	+}
	1172	+
	1173	+/*
	1174	+ * Note: All operations on tsk->posix_cputimer_work.scheduled happen either
	1175	+ * in hard interrupt context or in task context with interrupts
	1176	+ * disabled. Aside of that the writer/reader interaction is always in the
	1177	+ * context of the current task, which means they are strict per CPU.
	1178	+ */
	1179	+static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
	1180	+{
	1181	+ return tsk->posix_cputimers_work.scheduled;
	1182	+}
	1183	+
	1184	+static inline void __run_posix_cpu_timers(struct task_struct *tsk)
	1185	+{
	1186	+ if (WARN_ON_ONCE(tsk->posix_cputimers_work.scheduled))
	1187	+ return;
	1188	+
	1189	+ /* Schedule task work to actually expire the timers */
	1190	+ tsk->posix_cputimers_work.scheduled = true;
	1191	+ task_work_add(tsk, &tsk->posix_cputimers_work.work, TWA_RESUME);
	1192	+}
	1193	+
	1194	+static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
	1195	+ unsigned long start)
	1196	+{
	1197	+ bool ret = true;
1150	1198
1151	1199	/*
1152		- * The fast path checks that there are no expired thread or thread
1153		- * group timers. If that's so, just return.
	1200	+ * On !RT kernels interrupts are disabled while collecting expired
	1201	+ * timers, so no tick can happen and the fast path check can be
	1202	+ * reenabled without further checks.
1154	1203	*/
1155		- if (!fastpath_timer_check(tsk))
1156		- return;
	1204	+ if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
	1205	+ tsk->posix_cputimers_work.scheduled = false;
	1206	+ return true;
	1207	+ }
	1208	+
	1209	+ /*
	1210	+ * On RT enabled kernels ticks can happen while the expired timers
	1211	+ * are collected under sighand lock. But any tick which observes
	1212	+ * the CPUTIMERS_WORK_SCHEDULED bit set, does not run the fastpath
	1213	+ * checks. So reenabling the tick work has do be done carefully:
	1214	+ *
	1215	+ * Disable interrupts and run the fast path check if jiffies have
	1216	+ * advanced since the collecting of expired timers started. If
	1217	+ * jiffies have not advanced or the fast path check did not find
	1218	+ * newly expired timers, reenable the fast path check in the timer
	1219	+ * interrupt. If there are newly expired timers, return false and
	1220	+ * let the collection loop repeat.
	1221	+ */
	1222	+ local_irq_disable();
	1223	+ if (start != jiffies && fastpath_timer_check(tsk))
	1224	+ ret = false;
	1225	+ else
	1226	+ tsk->posix_cputimers_work.scheduled = false;
	1227	+ local_irq_enable();
	1228	+
	1229	+ return ret;
	1230	+}
	1231	+#else /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
	1232	+static inline void __run_posix_cpu_timers(struct task_struct *tsk)
	1233	+{
	1234	+ lockdep_posixtimer_enter();
	1235	+ handle_posix_cpu_timers(tsk);
	1236	+ lockdep_posixtimer_exit();
	1237	+}
	1238	+
	1239	+static void posix_cpu_timer_wait_running(struct k_itimer *timr)
	1240	+{
	1241	+ cpu_relax();
	1242	+}
	1243	+
	1244	+static void posix_cpu_timer_wait_running_nsleep(struct k_itimer *timr)
	1245	+{
	1246	+ spin_unlock_irq(&timr->it_lock);
	1247	+ cpu_relax();
	1248	+ spin_lock_irq(&timr->it_lock);
	1249	+}
	1250	+
	1251	+static inline bool posix_cpu_timers_work_scheduled(struct task_struct *tsk)
	1252	+{
	1253	+ return false;
	1254	+}
	1255	+
	1256	+static inline bool posix_cpu_timers_enable_work(struct task_struct *tsk,
	1257	+ unsigned long start)
	1258	+{
	1259	+ return true;
	1260	+}
	1261	+#endif /* CONFIG_POSIX_CPU_TIMERS_TASK_WORK */
	1262	+
	1263	+static void handle_posix_cpu_timers(struct task_struct *tsk)
	1264	+{
	1265	+ struct k_itimer timer, next;
	1266	+ unsigned long flags, start;
	1267	+ LIST_HEAD(firing);
1157	1268
1158	1269	if (!lock_task_sighand(tsk, &flags))
1159	1270	return;
1160		- /*
1161		- * Here we take off tsk->signal->cpu_timers[N] and
1162		- * tsk->cpu_timers[N] all the timers that are firing, and
1163		- * put them on the firing list.
1164		- */
1165		- check_thread_timers(tsk, &firing);
1166	1271
1167		- check_process_timers(tsk, &firing);
	1272	+ do {
	1273	+ /*
	1274	+ * On RT locking sighand lock does not disable interrupts,
	1275	+ * so this needs to be careful vs. ticks. Store the current
	1276	+ * jiffies value.
	1277	+ */
	1278	+ start = READ_ONCE(jiffies);
	1279	+ barrier();
	1280	+
	1281	+ /*
	1282	+ * Here we take off tsk->signal->cpu_timers[N] and
	1283	+ * tsk->cpu_timers[N] all the timers that are firing, and
	1284	+ * put them on the firing list.
	1285	+ */
	1286	+ check_thread_timers(tsk, &firing);
	1287	+
	1288	+ check_process_timers(tsk, &firing);
	1289	+
	1290	+ /*
	1291	+ * The above timer checks have updated the exipry cache and
	1292	+ * because nothing can have queued or modified timers after
	1293	+ * sighand lock was taken above it is guaranteed to be
	1294	+ * consistent. So the next timer interrupt fastpath check
	1295	+ * will find valid data.
	1296	+ *
	1297	+ * If timer expiry runs in the timer interrupt context then
	1298	+ * the loop is not relevant as timers will be directly
	1299	+ * expired in interrupt context. The stub function below
	1300	+ * returns always true which allows the compiler to
	1301	+ * optimize the loop out.
	1302	+ *
	1303	+ * If timer expiry is deferred to task work context then
	1304	+ * the following rules apply:
	1305	+ *
	1306	+ * - On !RT kernels no tick can have happened on this CPU
	1307	+ * after sighand lock was acquired because interrupts are
	1308	+ * disabled. So reenabling task work before dropping
	1309	+ * sighand lock and reenabling interrupts is race free.
	1310	+ *
	1311	+ * - On RT kernels ticks might have happened but the tick
	1312	+ * work ignored posix CPU timer handling because the
	1313	+ * CPUTIMERS_WORK_SCHEDULED bit is set. Reenabling work
	1314	+ * must be done very carefully including a check whether
	1315	+ * ticks have happened since the start of the timer
	1316	+ * expiry checks. posix_cpu_timers_enable_work() takes
	1317	+ * care of that and eventually lets the expiry checks
	1318	+ * run again.
	1319	+ */
	1320	+ } while (!posix_cpu_timers_enable_work(tsk, start));
1168	1321
1169	1322	/*
1170		- * We must release these locks before taking any timer's lock.
	1323	+ * We must release sighand lock before taking any timer's lock.
1171	1324	* There is a potential race with timer deletion here, as the
1172	1325	* siglock now protects our private firing list. We have set
1173	1326	* the firing flag in each timer, so that a deletion attempt
..	..	@@ -1182,11 +1335,18 @@
1182	1335	* each timer's lock before clearing its firing flag, so no
1183	1336	* timer call will interfere.
1184	1337	*/
1185		- list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
	1338	+ list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) {
1186	1339	int cpu_firing;
1187	1340
	1341	+ /*
	1342	+ * spin_lock() is sufficient here even independent of the
	1343	+ * expiry context. If expiry happens in hard interrupt
	1344	+ * context it's obvious. For task work context it's safe
	1345	+ * because all other operations on timer::it_lock happen in
	1346	+ * task context (syscall or exit).
	1347	+ */
1188	1348	spin_lock(&timer->it_lock);
1189		- list_del_init(&timer->it.cpu.entry);
	1349	+ list_del_init(&timer->it.cpu.elist);
1190	1350	cpu_firing = timer->it.cpu.firing;
1191	1351	timer->it.cpu.firing = 0;
1192	1352	/*
..	..	@@ -1196,26 +1356,56 @@
1196	1356	*/
1197	1357	if (likely(cpu_firing >= 0))
1198	1358	cpu_timer_fire(timer);
	1359	+ /* See posix_cpu_timer_wait_running() */
	1360	+ rcu_assign_pointer(timer->it.cpu.handling, NULL);
1199	1361	spin_unlock(&timer->it_lock);
1200	1362	}
	1363	+}
	1364	+
	1365	+/*
	1366	+ * This is called from the timer interrupt handler. The irq handler has
	1367	+ * already updated our counts. We need to check if any timers fire now.
	1368	+ * Interrupts are disabled.
	1369	+ */
	1370	+void run_posix_cpu_timers(void)
	1371	+{
	1372	+ struct task_struct *tsk = current;
	1373	+
	1374	+ lockdep_assert_irqs_disabled();
	1375	+
	1376	+ /*
	1377	+ * If the actual expiry is deferred to task work context and the
	1378	+ * work is already scheduled there is no point to do anything here.
	1379	+ */
	1380	+ if (posix_cpu_timers_work_scheduled(tsk))
	1381	+ return;
	1382	+
	1383	+ /*
	1384	+ * The fast path checks that there are no expired thread or thread
	1385	+ * group timers. If that's so, just return.
	1386	+ */
	1387	+ if (!fastpath_timer_check(tsk))
	1388	+ return;
	1389	+
	1390	+ __run_posix_cpu_timers(tsk);
1201	1391	}
1202	1392
1203	1393	/*
1204	1394	* Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1205	1395	* The tsk->sighand->siglock must be held by the caller.
1206	1396	*/
1207		-void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
	1397	+void set_process_cpu_timer(struct task_struct *tsk, unsigned int clkid,
1208	1398	u64 newval, u64 oldval)
1209	1399	{
1210		- u64 now;
1211		- int ret;
	1400	+ u64 now, *nextevt;
1212	1401
1213		- if (WARN_ON_ONCE(clock_idx >= CPUCLOCK_SCHED))
	1402	+ if (WARN_ON_ONCE(clkid >= CPUCLOCK_SCHED))
1214	1403	return;
1215	1404
1216		- ret = cpu_timer_sample_group(clock_idx, tsk, &now);
	1405	+ nextevt = &tsk->signal->posix_cputimers.bases[clkid].nextevt;
	1406	+ now = cpu_clock_sample_group(clkid, tsk, true);
1217	1407
1218		- if (oldval && ret != -EINVAL) {
	1408	+ if (oldval) {
1219	1409	/*
1220	1410	* We are setting itimer. The *oldval is absolute and we update
1221	1411	* it to be relative, *newval argument is relative and we update
..	..	@@ -1236,19 +1426,11 @@
1236	1426	}
1237	1427
1238	1428	/*
1239		- * Update expiration cache if we are the earliest timer, or eventually
1240		- * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
	1429	+ * Update expiration cache if this is the earliest timer. CPUCLOCK_PROF
	1430	+ * expiry cache is also used by RLIMIT_CPU!.
1241	1431	*/
1242		- switch (clock_idx) {
1243		- case CPUCLOCK_PROF:
1244		- if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1245		- tsk->signal->cputime_expires.prof_exp = *newval;
1246		- break;
1247		- case CPUCLOCK_VIRT:
1248		- if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1249		- tsk->signal->cputime_expires.virt_exp = *newval;
1250		- break;
1251		- }
	1432	+ if (newval < nextevt)
	1433	+ nextevt = newval;
1252	1434
1253	1435	tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1254	1436	}
..	..	@@ -1270,6 +1452,7 @@
1270	1452	timer.it_overrun = -1;
1271	1453	error = posix_cpu_timer_create(&timer);
1272	1454	timer.it_process = current;
	1455	+
1273	1456	if (!error) {
1274	1457	static struct itimerspec64 zero_it;
1275	1458	struct restart_block *restart;
..	..	@@ -1285,7 +1468,7 @@
1285	1468	}
1286	1469
1287	1470	while (!signal_pending(current)) {
1288		- if (timer.it.cpu.expires == 0) {
	1471	+ if (!cpu_timer_getexpires(&timer.it.cpu)) {
1289	1472	/*
1290	1473	* Our timer fired and was reset, below
1291	1474	* deletion can not fail.
..	..	@@ -1307,26 +1490,19 @@
1307	1490	/*
1308	1491	* We were interrupted by a signal.
1309	1492	*/
1310		- expires = timer.it.cpu.expires;
	1493	+ expires = cpu_timer_getexpires(&timer.it.cpu);
1311	1494	error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1312	1495	if (!error) {
1313		- /*
1314		- * Timer is now unarmed, deletion can not fail.
1315		- */
	1496	+ /* Timer is now unarmed, deletion can not fail. */
1316	1497	posix_cpu_timer_del(&timer);
	1498	+ } else {
	1499	+ while (error == TIMER_RETRY) {
	1500	+ posix_cpu_timer_wait_running_nsleep(&timer);
	1501	+ error = posix_cpu_timer_del(&timer);
	1502	+ }
1317	1503	}
1318		- spin_unlock_irq(&timer.it_lock);
1319	1504
1320		- while (error == TIMER_RETRY) {
1321		- /*
1322		- * We need to handle case when timer was or is in the
1323		- * middle of firing. In other cases we already freed
1324		- * resources.
1325		- */
1326		- spin_lock_irq(&timer.it_lock);
1327		- error = posix_cpu_timer_del(&timer);
1328		- spin_unlock_irq(&timer.it_lock);
1329		- }
	1505	+ spin_unlock_irq(&timer.it_lock);
1330	1506
1331	1507	if ((it.it_value.tv_sec \| it.it_value.tv_nsec) == 0) {
1332	1508	/*
..	..	@@ -1427,26 +1603,27 @@
1427	1603	}
1428	1604
1429	1605	const struct k_clock clock_posix_cpu = {
1430		- .clock_getres = posix_cpu_clock_getres,
1431		- .clock_set = posix_cpu_clock_set,
1432		- .clock_get = posix_cpu_clock_get,
1433		- .timer_create = posix_cpu_timer_create,
1434		- .nsleep = posix_cpu_nsleep,
1435		- .timer_set = posix_cpu_timer_set,
1436		- .timer_del = posix_cpu_timer_del,
1437		- .timer_get = posix_cpu_timer_get,
1438		- .timer_rearm = posix_cpu_timer_rearm,
	1606	+ .clock_getres = posix_cpu_clock_getres,
	1607	+ .clock_set = posix_cpu_clock_set,
	1608	+ .clock_get_timespec = posix_cpu_clock_get,
	1609	+ .timer_create = posix_cpu_timer_create,
	1610	+ .nsleep = posix_cpu_nsleep,
	1611	+ .timer_set = posix_cpu_timer_set,
	1612	+ .timer_del = posix_cpu_timer_del,
	1613	+ .timer_get = posix_cpu_timer_get,
	1614	+ .timer_rearm = posix_cpu_timer_rearm,
	1615	+ .timer_wait_running = posix_cpu_timer_wait_running,
1439	1616	};
1440	1617
1441	1618	const struct k_clock clock_process = {
1442		- .clock_getres = process_cpu_clock_getres,
1443		- .clock_get = process_cpu_clock_get,
1444		- .timer_create = process_cpu_timer_create,
1445		- .nsleep = process_cpu_nsleep,
	1619	+ .clock_getres = process_cpu_clock_getres,
	1620	+ .clock_get_timespec = process_cpu_clock_get,
	1621	+ .timer_create = process_cpu_timer_create,
	1622	+ .nsleep = process_cpu_nsleep,
1446	1623	};
1447	1624
1448	1625	const struct k_clock clock_thread = {
1449		- .clock_getres = thread_cpu_clock_getres,
1450		- .clock_get = thread_cpu_clock_get,
1451		- .timer_create = thread_cpu_timer_create,
	1626	+ .clock_getres = thread_cpu_clock_getres,
	1627	+ .clock_get_timespec = thread_cpu_clock_get,
	1628	+ .timer_create = thread_cpu_timer_create,
1452	1629	};