forked from ~ljy/RK356X_SDK_RELEASE

blame | history | patch | commit | commitdiff
hc
2023-12-09 958e46acc8e900e8569dd467c1af9b8d2d019394 
 kernel/kernel/sched/idle.c
....@@ -1,3 +1,4 @@
1
+// SPDX-License-Identifier: GPL-2.0-only
12 /*
23  * Generic entry points for the idle threads and
34  * implementation of the idle task scheduling class.
....@@ -9,6 +10,8 @@
910 
1011 #include <trace/events/power.h>
1112 
13
+#include <trace/hooks/sched.h>
14
+
1215 /* Linker adds these: start and end of __cpuidle functions */
1316 extern char __cpuidle_text_start[], __cpuidle_text_end[];
1417 
....@@ -16,10 +19,9 @@
1619  * sched_idle_set_state - Record idle state for the current CPU.
1720  * @idle_state: State to record.
1821  */
19
-void sched_idle_set_state(struct cpuidle_state *idle_state, int index)
22
+void sched_idle_set_state(struct cpuidle_state *idle_state)
2023 {
2124 	idle_set_state(this_rq(), idle_state);
22
-	idle_set_state_idx(this_rq(), index);
2325 }
2426 
2527 static int __read_mostly cpu_idle_force_poll;
....@@ -54,17 +56,18 @@
5456 
5557 static noinline int __cpuidle cpu_idle_poll(void)
5658 {
57
-	rcu_idle_enter();
58
-	trace_cpu_idle_rcuidle(0, smp_processor_id());
59
-	local_irq_enable();
59
+	trace_cpu_idle(0, smp_processor_id());
6060 	stop_critical_timings();
61
+	rcu_idle_enter();
62
+	local_irq_enable();
6163 
6264 	while (!tif_need_resched() &&
63
-		(cpu_idle_force_poll || tick_check_broadcast_expired()))
65
+	       (cpu_idle_force_poll || tick_check_broadcast_expired()))
6466 		cpu_relax();
65
-	start_critical_timings();
66
-	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
67
+
6768 	rcu_idle_exit();
69
+	start_critical_timings();
70
+	trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
6871 
6972 	return 1;
7073 }
....@@ -77,7 +80,7 @@
7780 void __weak arch_cpu_idle(void)
7881 {
7982 	cpu_idle_force_poll = 1;
80
-	local_irq_enable();
83
+	raw_local_irq_enable();
8184 }
8285 
8386 /**
....@@ -90,10 +93,50 @@
9093 	if (current_clr_polling_and_test()) {
9194 		local_irq_enable();
9295 	} else {
96
+
97
+		trace_cpu_idle(1, smp_processor_id());
9398 		stop_critical_timings();
99
+
100
+		/*
101
+		 * arch_cpu_idle() is supposed to enable IRQs, however
102
+		 * we can't do that because of RCU and tracing.
103
+		 *
104
+		 * Trace IRQs enable here, then switch off RCU, and have
105
+		 * arch_cpu_idle() use raw_local_irq_enable(). Note that
106
+		 * rcu_idle_enter() relies on lockdep IRQ state, so switch that
107
+		 * last -- this is very similar to the entry code.
108
+		 */
109
+		trace_hardirqs_on_prepare();
110
+		lockdep_hardirqs_on_prepare(_THIS_IP_);
111
+		rcu_idle_enter();
112
+		lockdep_hardirqs_on(_THIS_IP_);
113
+
94114 		arch_cpu_idle();
115
+
116
+		/*
117
+		 * OK, so IRQs are enabled here, but RCU needs them disabled to
118
+		 * turn itself back on.. funny thing is that disabling IRQs
119
+		 * will cause tracing, which needs RCU. Jump through hoops to
120
+		 * make it 'work'.
121
+		 */
122
+		raw_local_irq_disable();
123
+		lockdep_hardirqs_off(_THIS_IP_);
124
+		rcu_idle_exit();
125
+		lockdep_hardirqs_on(_THIS_IP_);
126
+		raw_local_irq_enable();
127
+
95128 		start_critical_timings();
129
+		trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
96130 	}
131
+}
132
+
133
+static int call_cpuidle_s2idle(struct cpuidle_driver *drv,
134
+			       struct cpuidle_device *dev)
135
+{
136
+	if (current_clr_polling_and_test())
137
+		return -EBUSY;
138
+
139
+	return cpuidle_enter_s2idle(drv, dev);
97140 }
98141 
99142 static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
....@@ -104,7 +147,7 @@
104147 	 * update no idle residency and return.
105148 	 */
106149 	if (current_clr_polling_and_test()) {
107
-		dev->last_residency = 0;
150
+		dev->last_residency_ns = 0;
108151 		local_irq_enable();
109152 		return -EBUSY;
110153 	}
....@@ -149,7 +192,6 @@
149192 
150193 	if (cpuidle_not_available(drv, dev)) {
151194 		tick_nohz_idle_stop_tick();
152
-		rcu_idle_enter();
153195 
154196 		default_idle_call();
155197 		goto exit_idle;
....@@ -158,30 +200,30 @@
158200 	/*
159201 	 * Suspend-to-idle ("s2idle") is a system state in which all user space
160202 	 * has been frozen, all I/O devices have been suspended and the only
161
-	 * activity happens here and in iterrupts (if any).  In that case bypass
203
+	 * activity happens here and in interrupts (if any). In that case bypass
162204 	 * the cpuidle governor and go stratight for the deepest idle state
163205 	 * available.  Possibly also suspend the local tick and the entire
164206 	 * timekeeping to prevent timer interrupts from kicking us out of idle
165207 	 * until a proper wakeup interrupt happens.
166208 	 */
167209 
168
-	if (idle_should_enter_s2idle() || dev->use_deepest_state) {
210
+	if (idle_should_enter_s2idle() || dev->forced_idle_latency_limit_ns) {
211
+		u64 max_latency_ns;
212
+
169213 		if (idle_should_enter_s2idle()) {
170
-			rcu_idle_enter();
171214 
172
-			entered_state = cpuidle_enter_s2idle(drv, dev);
173
-			if (entered_state > 0) {
174
-				local_irq_enable();
215
+			entered_state = call_cpuidle_s2idle(drv, dev);
216
+			if (entered_state > 0)
175217 				goto exit_idle;
176
-			}
177218 
178
-			rcu_idle_exit();
219
+			max_latency_ns = U64_MAX;
220
+		} else {
221
+			max_latency_ns = dev->forced_idle_latency_limit_ns;
179222 		}
180223 
181224 		tick_nohz_idle_stop_tick();
182
-		rcu_idle_enter();
183225 
184
-		next_state = cpuidle_find_deepest_state(drv, dev);
226
+		next_state = cpuidle_find_deepest_state(drv, dev, max_latency_ns);
185227 		call_cpuidle(drv, dev, next_state);
186228 	} else {
187229 		bool stop_tick = true;
....@@ -195,8 +237,6 @@
195237 			tick_nohz_idle_stop_tick();
196238 		else
197239 			tick_nohz_idle_retain_tick();
198
-
199
-		rcu_idle_enter();
200240 
201241 		entered_state = call_cpuidle(drv, dev, next_state);
202242 		/*
....@@ -213,8 +253,6 @@
213253 	 */
214254 	if (WARN_ON_ONCE(irqs_disabled()))
215255 		local_irq_enable();
216
-
217
-	rcu_idle_exit();
218256 }
219257 
220258 /*
....@@ -238,7 +276,6 @@
238276 	tick_nohz_idle_enter();
239277 
240278 	while (!need_resched()) {
241
-		check_pgt_cache();
242279 		rmb();
243280 
244281 		local_irq_disable();
....@@ -250,6 +287,7 @@
250287 		}
251288 
252289 		arch_cpu_idle_enter();
290
+		rcu_nocb_flush_deferred_wakeup();
253291 
254292 		/*
255293 		 * In poll mode we reenable interrupts and spin. Also if we
....@@ -284,7 +322,11 @@
284322 	 */
285323 	smp_mb__after_atomic();
286324 
287
-	sched_ttwu_pending();
325
+	/*
326
+	 * RCU relies on this call to be done outside of an RCU read-side
327
+	 * critical section.
328
+	 */
329
+	flush_smp_call_function_from_idle();
288330 	schedule_idle();
289331 
290332 	if (unlikely(klp_patch_pending(current)))
....@@ -312,7 +354,7 @@
312354 	return HRTIMER_NORESTART;
313355 }
314356 
315
-void play_idle(unsigned long duration_ms)
357
+void play_idle_precise(u64 duration_ns, u64 latency_ns)
316358 {
317359 	struct idle_timer it;
318360 
....@@ -324,46 +366,32 @@
324366 	WARN_ON_ONCE(current->nr_cpus_allowed != 1);
325367 	WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
326368 	WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
327
-	WARN_ON_ONCE(!duration_ms);
369
+	WARN_ON_ONCE(!duration_ns);
328370 
329371 	rcu_sleep_check();
330372 	preempt_disable();
331373 	current->flags |= PF_IDLE;
332
-	cpuidle_use_deepest_state(true);
374
+	cpuidle_use_deepest_state(latency_ns);
333375 
334376 	it.done = 0;
335
-	hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
377
+	hrtimer_init_on_stack(&it.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
336378 	it.timer.function = idle_inject_timer_fn;
337
-	hrtimer_start(&it.timer, ms_to_ktime(duration_ms), HRTIMER_MODE_REL_PINNED);
379
+	hrtimer_start(&it.timer, ns_to_ktime(duration_ns),
380
+		      HRTIMER_MODE_REL_PINNED_HARD);
338381 
339382 	while (!READ_ONCE(it.done))
340383 		do_idle();
341384 
342
-	cpuidle_use_deepest_state(false);
385
+	cpuidle_use_deepest_state(0);
343386 	current->flags &= ~PF_IDLE;
344387 
345388 	preempt_fold_need_resched();
346389 	preempt_enable();
347390 }
348
-EXPORT_SYMBOL_GPL(play_idle);
391
+EXPORT_SYMBOL_GPL(play_idle_precise);
349392 
350393 void cpu_startup_entry(enum cpuhp_state state)
351394 {
352
-	/*
353
-	 * This #ifdef needs to die, but it's too late in the cycle to
354
-	 * make this generic (ARM and SH have never invoked the canary
355
-	 * init for the non boot CPUs!). Will be fixed in 3.11
356
-	 */
357
-#ifdef CONFIG_X86
358
-	/*
359
-	 * If we're the non-boot CPU, nothing set the stack canary up
360
-	 * for us. The boot CPU already has it initialized but no harm
361
-	 * in doing it again. This is a good place for updating it, as
362
-	 * we wont ever return from this function (so the invalid
363
-	 * canaries already on the stack wont ever trigger).
364
-	 */
365
-	boot_init_stack_canary();
366
-#endif
367395 	arch_cpu_idle_prepare();
368396 	cpuhp_online_idle(state);
369397 	while (1)
....@@ -376,10 +404,15 @@
376404 
377405 #ifdef CONFIG_SMP
378406 static int
379
-select_task_rq_idle(struct task_struct *p, int cpu, int sd_flag, int flags,
380
-		    int sibling_count_hint)
407
+select_task_rq_idle(struct task_struct *p, int cpu, int sd_flag, int flags)
381408 {
382409 	return task_cpu(p); /* IDLE tasks as never migrated */
410
+}
411
+
412
+static int
413
+balance_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
414
+{
415
+	return WARN_ON_ONCE(1);
383416 }
384417 #endif
385418 
....@@ -391,14 +424,23 @@
391424 	resched_curr(rq);
392425 }
393426 
394
-static struct task_struct *
395
-pick_next_task_idle(struct rq *rq, struct task_struct *prev, struct rq_flags *rf)
427
+static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
396428 {
397
-	put_prev_task(rq, prev);
429
+}
430
+
431
+static void set_next_task_idle(struct rq *rq, struct task_struct *next, bool first)
432
+{
398433 	update_idle_core(rq);
399434 	schedstat_inc(rq->sched_goidle);
435
+}
400436 
401
-	return rq->idle;
437
+struct task_struct *pick_next_task_idle(struct rq *rq)
438
+{
439
+	struct task_struct *next = rq->idle;
440
+
441
+	set_next_task_idle(rq, next, true);
442
+
443
+	return next;
402444 }
403445 
404446 /*
....@@ -410,12 +452,10 @@
410452 {
411453 	raw_spin_unlock_irq(&rq->lock);
412454 	printk(KERN_ERR "bad: scheduling from the idle thread!\n");
455
+
456
+	trace_android_rvh_dequeue_task_idle(p);
413457 	dump_stack();
414458 	raw_spin_lock_irq(&rq->lock);
415
-}
416
-
417
-static void put_prev_task_idle(struct rq *rq, struct task_struct *prev)
418
-{
419459 }
420460 
421461 /*
....@@ -430,10 +470,6 @@
430470 {
431471 }
432472 
433
-static void set_curr_task_idle(struct rq *rq)
434
-{
435
-}
436
-
437473 static void switched_to_idle(struct rq *rq, struct task_struct *p)
438474 {
439475 	BUG();
....@@ -445,11 +481,6 @@
445481 	BUG();
446482 }
447483 
448
-static unsigned int get_rr_interval_idle(struct rq *rq, struct task_struct *task)
449
-{
450
-	return 0;
451
-}
452
-
453484 static void update_curr_idle(struct rq *rq)
454485 {
455486 }
....@@ -457,8 +488,8 @@
457488 /*
458489  * Simple, special scheduling class for the per-CPU idle tasks:
459490  */
460
-const struct sched_class idle_sched_class = {
461
-	/* .next is NULL */
491
+const struct sched_class idle_sched_class
492
+	__section("__idle_sched_class") = {
462493 	/* no enqueue/yield_task for idle tasks */
463494 
464495 	/* dequeue is not valid, we print a debug message there: */
....@@ -468,16 +499,15 @@
468499 
469500 	.pick_next_task		= pick_next_task_idle,
470501 	.put_prev_task		= put_prev_task_idle,
502
+	.set_next_task          = set_next_task_idle,
471503 
472504 #ifdef CONFIG_SMP
505
+	.balance		= balance_idle,
473506 	.select_task_rq		= select_task_rq_idle,
474507 	.set_cpus_allowed	= set_cpus_allowed_common,
475508 #endif
476509 
477
-	.set_curr_task          = set_curr_task_idle,
478510 	.task_tick		= task_tick_idle,
479
-
480
-	.get_rr_interval	= get_rr_interval_idle,
481511 
482512 	.prio_changed		= prio_changed_idle,
483513 	.switched_to		= switched_to_idle,