Linux Audio

Check our new training course

Loading...
v4.17
 
  1/*
  2 *	linux/kernel/softirq.c
  3 *
  4 *	Copyright (C) 1992 Linus Torvalds
  5 *
  6 *	Distribute under GPLv2.
  7 *
  8 *	Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903)
  9 */
 10
 11#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 12
 13#include <linux/export.h>
 14#include <linux/kernel_stat.h>
 15#include <linux/interrupt.h>
 16#include <linux/init.h>
 
 17#include <linux/mm.h>
 18#include <linux/notifier.h>
 19#include <linux/percpu.h>
 20#include <linux/cpu.h>
 21#include <linux/freezer.h>
 22#include <linux/kthread.h>
 23#include <linux/rcupdate.h>
 24#include <linux/ftrace.h>
 25#include <linux/smp.h>
 26#include <linux/smpboot.h>
 27#include <linux/tick.h>
 28#include <linux/irq.h>
 
 
 
 29
 30#define CREATE_TRACE_POINTS
 31#include <trace/events/irq.h>
 32
 33/*
 34   - No shared variables, all the data are CPU local.
 35   - If a softirq needs serialization, let it serialize itself
 36     by its own spinlocks.
 37   - Even if softirq is serialized, only local cpu is marked for
 38     execution. Hence, we get something sort of weak cpu binding.
 39     Though it is still not clear, will it result in better locality
 40     or will not.
 41
 42   Examples:
 43   - NET RX softirq. It is multithreaded and does not require
 44     any global serialization.
 45   - NET TX softirq. It kicks software netdevice queues, hence
 46     it is logically serialized per device, but this serialization
 47     is invisible to common code.
 48   - Tasklets: serialized wrt itself.
 49 */
 50
 51#ifndef __ARCH_IRQ_STAT
 52irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned;
 53EXPORT_SYMBOL(irq_stat);
 54#endif
 55
 56static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp;
 57
 58DEFINE_PER_CPU(struct task_struct *, ksoftirqd);
 59
 60const char * const softirq_to_name[NR_SOFTIRQS] = {
 61	"HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "IRQ_POLL",
 62	"TASKLET", "SCHED", "HRTIMER", "RCU"
 63};
 64
 65/*
 66 * we cannot loop indefinitely here to avoid userspace starvation,
 67 * but we also don't want to introduce a worst case 1/HZ latency
 68 * to the pending events, so lets the scheduler to balance
 69 * the softirq load for us.
 70 */
 71static void wakeup_softirqd(void)
 72{
 73	/* Interrupts are disabled: no need to stop preemption */
 74	struct task_struct *tsk = __this_cpu_read(ksoftirqd);
 75
 76	if (tsk && tsk->state != TASK_RUNNING)
 77		wake_up_process(tsk);
 78}
 79
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 80/*
 81 * If ksoftirqd is scheduled, we do not want to process pending softirqs
 82 * right now. Let ksoftirqd handle this at its own rate, to get fairness.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 83 */
 84static bool ksoftirqd_running(void)
 85{
 86	struct task_struct *tsk = __this_cpu_read(ksoftirqd);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 87
 88	return tsk && (tsk->state == TASK_RUNNING);
 
 
 
 
 
 
 89}
 90
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 91/*
 92 * preempt_count and SOFTIRQ_OFFSET usage:
 93 * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving
 94 *   softirq processing.
 95 * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET)
 96 *   on local_bh_disable or local_bh_enable.
 97 * This lets us distinguish between whether we are currently processing
 98 * softirq and whether we just have bh disabled.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 99 */
 
 
 
 
 
 
 
100
101/*
102 * This one is for softirq.c-internal use,
103 * where hardirqs are disabled legitimately:
104 */
105#ifdef CONFIG_TRACE_IRQFLAGS
106void __local_bh_disable_ip(unsigned long ip, unsigned int cnt)
107{
108	unsigned long flags;
109
110	WARN_ON_ONCE(in_irq());
111
112	raw_local_irq_save(flags);
113	/*
114	 * The preempt tracer hooks into preempt_count_add and will break
115	 * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET
116	 * is set and before current->softirq_enabled is cleared.
117	 * We must manually increment preempt_count here and manually
118	 * call the trace_preempt_off later.
119	 */
120	__preempt_count_add(cnt);
121	/*
122	 * Were softirqs turned off above:
123	 */
124	if (softirq_count() == (cnt & SOFTIRQ_MASK))
125		trace_softirqs_off(ip);
126	raw_local_irq_restore(flags);
127
128	if (preempt_count() == cnt) {
129#ifdef CONFIG_DEBUG_PREEMPT
130		current->preempt_disable_ip = get_lock_parent_ip();
131#endif
132		trace_preempt_off(CALLER_ADDR0, get_lock_parent_ip());
133	}
134}
135EXPORT_SYMBOL(__local_bh_disable_ip);
136#endif /* CONFIG_TRACE_IRQFLAGS */
137
138static void __local_bh_enable(unsigned int cnt)
139{
140	lockdep_assert_irqs_disabled();
141
 
 
 
142	if (softirq_count() == (cnt & SOFTIRQ_MASK))
143		trace_softirqs_on(_RET_IP_);
144	preempt_count_sub(cnt);
 
145}
146
147/*
148 * Special-case - softirqs can safely be enabled in
149 * cond_resched_softirq(), or by __do_softirq(),
150 * without processing still-pending softirqs:
151 */
152void _local_bh_enable(void)
153{
154	WARN_ON_ONCE(in_irq());
155	__local_bh_enable(SOFTIRQ_DISABLE_OFFSET);
156}
157EXPORT_SYMBOL(_local_bh_enable);
158
159void __local_bh_enable_ip(unsigned long ip, unsigned int cnt)
160{
161	WARN_ON_ONCE(in_irq());
162	lockdep_assert_irqs_enabled();
163#ifdef CONFIG_TRACE_IRQFLAGS
164	local_irq_disable();
165#endif
166	/*
167	 * Are softirqs going to be turned on now:
168	 */
169	if (softirq_count() == SOFTIRQ_DISABLE_OFFSET)
170		trace_softirqs_on(ip);
171	/*
172	 * Keep preemption disabled until we are done with
173	 * softirq processing:
174	 */
175	preempt_count_sub(cnt - 1);
176
177	if (unlikely(!in_interrupt() && local_softirq_pending())) {
178		/*
179		 * Run softirq if any pending. And do it in its own stack
180		 * as we may be calling this deep in a task call stack already.
181		 */
182		do_softirq();
183	}
184
185	preempt_count_dec();
186#ifdef CONFIG_TRACE_IRQFLAGS
187	local_irq_enable();
188#endif
189	preempt_check_resched();
190}
191EXPORT_SYMBOL(__local_bh_enable_ip);
192
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
193/*
194 * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times,
195 * but break the loop if need_resched() is set or after 2 ms.
196 * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in
197 * certain cases, such as stop_machine(), jiffies may cease to
198 * increment and so we need the MAX_SOFTIRQ_RESTART limit as
199 * well to make sure we eventually return from this method.
200 *
201 * These limits have been established via experimentation.
202 * The two things to balance is latency against fairness -
203 * we want to handle softirqs as soon as possible, but they
204 * should not be able to lock up the box.
205 */
206#define MAX_SOFTIRQ_TIME  msecs_to_jiffies(2)
207#define MAX_SOFTIRQ_RESTART 10
208
209#ifdef CONFIG_TRACE_IRQFLAGS
210/*
211 * When we run softirqs from irq_exit() and thus on the hardirq stack we need
212 * to keep the lockdep irq context tracking as tight as possible in order to
213 * not miss-qualify lock contexts and miss possible deadlocks.
214 */
215
216static inline bool lockdep_softirq_start(void)
217{
218	bool in_hardirq = false;
219
220	if (trace_hardirq_context(current)) {
221		in_hardirq = true;
222		trace_hardirq_exit();
223	}
224
225	lockdep_softirq_enter();
226
227	return in_hardirq;
228}
229
230static inline void lockdep_softirq_end(bool in_hardirq)
231{
232	lockdep_softirq_exit();
233
234	if (in_hardirq)
235		trace_hardirq_enter();
236}
237#else
238static inline bool lockdep_softirq_start(void) { return false; }
239static inline void lockdep_softirq_end(bool in_hardirq) { }
240#endif
241
242asmlinkage __visible void __softirq_entry __do_softirq(void)
243{
244	unsigned long end = jiffies + MAX_SOFTIRQ_TIME;
245	unsigned long old_flags = current->flags;
246	int max_restart = MAX_SOFTIRQ_RESTART;
247	struct softirq_action *h;
248	bool in_hardirq;
249	__u32 pending;
250	int softirq_bit;
251
252	/*
253	 * Mask out PF_MEMALLOC s current task context is borrowed for the
254	 * softirq. A softirq handled such as network RX might set PF_MEMALLOC
255	 * again if the socket is related to swap
256	 */
257	current->flags &= ~PF_MEMALLOC;
258
259	pending = local_softirq_pending();
260	account_irq_enter_time(current);
261
262	__local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET);
263	in_hardirq = lockdep_softirq_start();
 
264
265restart:
266	/* Reset the pending bitmask before enabling irqs */
267	set_softirq_pending(0);
268
269	local_irq_enable();
270
271	h = softirq_vec;
272
273	while ((softirq_bit = ffs(pending))) {
274		unsigned int vec_nr;
275		int prev_count;
276
277		h += softirq_bit - 1;
278
279		vec_nr = h - softirq_vec;
280		prev_count = preempt_count();
281
282		kstat_incr_softirqs_this_cpu(vec_nr);
283
284		trace_softirq_entry(vec_nr);
285		h->action(h);
286		trace_softirq_exit(vec_nr);
287		if (unlikely(prev_count != preempt_count())) {
288			pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n",
289			       vec_nr, softirq_to_name[vec_nr], h->action,
290			       prev_count, preempt_count());
291			preempt_count_set(prev_count);
292		}
293		h++;
294		pending >>= softirq_bit;
295	}
296
297	rcu_bh_qs();
 
 
 
298	local_irq_disable();
299
300	pending = local_softirq_pending();
301	if (pending) {
302		if (time_before(jiffies, end) && !need_resched() &&
303		    --max_restart)
304			goto restart;
305
306		wakeup_softirqd();
307	}
308
 
309	lockdep_softirq_end(in_hardirq);
310	account_irq_exit_time(current);
311	__local_bh_enable(SOFTIRQ_OFFSET);
312	WARN_ON_ONCE(in_interrupt());
313	current_restore_flags(old_flags, PF_MEMALLOC);
314}
315
316asmlinkage __visible void do_softirq(void)
 
 
 
317{
318	__u32 pending;
319	unsigned long flags;
320
321	if (in_interrupt())
322		return;
323
324	local_irq_save(flags);
325
326	pending = local_softirq_pending();
327
328	if (pending && !ksoftirqd_running())
329		do_softirq_own_stack();
330
331	local_irq_restore(flags);
332}
333
334/*
335 * Enter an interrupt context.
336 */
337void irq_enter(void)
338{
339	rcu_irq_enter();
340	if (is_idle_task(current) && !in_interrupt()) {
341		/*
342		 * Prevent raise_softirq from needlessly waking up ksoftirqd
343		 * here, as softirq will be serviced on return from interrupt.
344		 */
345		local_bh_disable();
346		tick_irq_enter();
347		_local_bh_enable();
348	}
349
350	__irq_enter();
351}
352
353static inline void invoke_softirq(void)
354{
355	if (ksoftirqd_running())
356		return;
357
358	if (!force_irqthreads) {
359#ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK
360		/*
361		 * We can safely execute softirq on the current stack if
362		 * it is the irq stack, because it should be near empty
363		 * at this stage.
364		 */
365		__do_softirq();
366#else
367		/*
368		 * Otherwise, irq_exit() is called on the task stack that can
369		 * be potentially deep already. So call softirq in its own stack
370		 * to prevent from any overrun.
371		 */
372		do_softirq_own_stack();
373#endif
374	} else {
375		wakeup_softirqd();
376	}
377}
378
379static inline void tick_irq_exit(void)
380{
381#ifdef CONFIG_NO_HZ_COMMON
382	int cpu = smp_processor_id();
383
384	/* Make sure that timer wheel updates are propagated */
385	if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) {
386		if (!in_interrupt())
387			tick_nohz_irq_exit();
388	}
389#endif
390}
391
392/*
393 * Exit an interrupt context. Process softirqs if needed and possible:
394 */
395void irq_exit(void)
396{
397#ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED
398	local_irq_disable();
399#else
400	lockdep_assert_irqs_disabled();
401#endif
402	account_irq_exit_time(current);
403	preempt_count_sub(HARDIRQ_OFFSET);
404	if (!in_interrupt() && local_softirq_pending())
405		invoke_softirq();
406
407	tick_irq_exit();
408	rcu_irq_exit();
409	trace_hardirq_exit(); /* must be last! */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
410}
411
412/*
413 * This function must run with irqs disabled!
414 */
415inline void raise_softirq_irqoff(unsigned int nr)
416{
417	__raise_softirq_irqoff(nr);
418
419	/*
420	 * If we're in an interrupt or softirq, we're done
421	 * (this also catches softirq-disabled code). We will
422	 * actually run the softirq once we return from
423	 * the irq or softirq.
424	 *
425	 * Otherwise we wake up ksoftirqd to make sure we
426	 * schedule the softirq soon.
427	 */
428	if (!in_interrupt())
429		wakeup_softirqd();
430}
431
432void raise_softirq(unsigned int nr)
433{
434	unsigned long flags;
435
436	local_irq_save(flags);
437	raise_softirq_irqoff(nr);
438	local_irq_restore(flags);
439}
440
441void __raise_softirq_irqoff(unsigned int nr)
442{
 
443	trace_softirq_raise(nr);
444	or_softirq_pending(1UL << nr);
445}
446
447void open_softirq(int nr, void (*action)(struct softirq_action *))
448{
449	softirq_vec[nr].action = action;
450}
451
452/*
453 * Tasklets
454 */
455struct tasklet_head {
456	struct tasklet_struct *head;
457	struct tasklet_struct **tail;
458};
459
460static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec);
461static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec);
462
463static void __tasklet_schedule_common(struct tasklet_struct *t,
464				      struct tasklet_head __percpu *headp,
465				      unsigned int softirq_nr)
466{
467	struct tasklet_head *head;
468	unsigned long flags;
469
470	local_irq_save(flags);
471	head = this_cpu_ptr(headp);
472	t->next = NULL;
473	*head->tail = t;
474	head->tail = &(t->next);
475	raise_softirq_irqoff(softirq_nr);
476	local_irq_restore(flags);
477}
478
479void __tasklet_schedule(struct tasklet_struct *t)
480{
481	__tasklet_schedule_common(t, &tasklet_vec,
482				  TASKLET_SOFTIRQ);
483}
484EXPORT_SYMBOL(__tasklet_schedule);
485
486void __tasklet_hi_schedule(struct tasklet_struct *t)
487{
488	__tasklet_schedule_common(t, &tasklet_hi_vec,
489				  HI_SOFTIRQ);
490}
491EXPORT_SYMBOL(__tasklet_hi_schedule);
492
 
 
 
 
 
 
 
 
 
 
 
 
 
 
493static void tasklet_action_common(struct softirq_action *a,
494				  struct tasklet_head *tl_head,
495				  unsigned int softirq_nr)
496{
497	struct tasklet_struct *list;
498
499	local_irq_disable();
500	list = tl_head->head;
501	tl_head->head = NULL;
502	tl_head->tail = &tl_head->head;
503	local_irq_enable();
504
505	while (list) {
506		struct tasklet_struct *t = list;
507
508		list = list->next;
509
510		if (tasklet_trylock(t)) {
511			if (!atomic_read(&t->count)) {
512				if (!test_and_clear_bit(TASKLET_STATE_SCHED,
513							&t->state))
514					BUG();
515				t->func(t->data);
 
 
 
 
 
 
 
516				tasklet_unlock(t);
517				continue;
518			}
519			tasklet_unlock(t);
520		}
521
522		local_irq_disable();
523		t->next = NULL;
524		*tl_head->tail = t;
525		tl_head->tail = &t->next;
526		__raise_softirq_irqoff(softirq_nr);
527		local_irq_enable();
528	}
529}
530
531static __latent_entropy void tasklet_action(struct softirq_action *a)
532{
533	tasklet_action_common(a, this_cpu_ptr(&tasklet_vec), TASKLET_SOFTIRQ);
534}
535
536static __latent_entropy void tasklet_hi_action(struct softirq_action *a)
537{
538	tasklet_action_common(a, this_cpu_ptr(&tasklet_hi_vec), HI_SOFTIRQ);
539}
540
 
 
 
 
 
 
 
 
 
 
 
 
541void tasklet_init(struct tasklet_struct *t,
542		  void (*func)(unsigned long), unsigned long data)
543{
544	t->next = NULL;
545	t->state = 0;
546	atomic_set(&t->count, 0);
547	t->func = func;
 
548	t->data = data;
549}
550EXPORT_SYMBOL(tasklet_init);
551
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
552void tasklet_kill(struct tasklet_struct *t)
553{
554	if (in_interrupt())
555		pr_notice("Attempt to kill tasklet from interrupt\n");
556
557	while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
558		do {
559			yield();
560		} while (test_bit(TASKLET_STATE_SCHED, &t->state));
561	}
562	tasklet_unlock_wait(t);
563	clear_bit(TASKLET_STATE_SCHED, &t->state);
564}
565EXPORT_SYMBOL(tasklet_kill);
566
567/*
568 * tasklet_hrtimer
569 */
570
571/*
572 * The trampoline is called when the hrtimer expires. It schedules a tasklet
573 * to run __tasklet_hrtimer_trampoline() which in turn will call the intended
574 * hrtimer callback, but from softirq context.
575 */
576static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer)
577{
578	struct tasklet_hrtimer *ttimer =
579		container_of(timer, struct tasklet_hrtimer, timer);
580
581	tasklet_hi_schedule(&ttimer->tasklet);
582	return HRTIMER_NORESTART;
583}
 
584
585/*
586 * Helper function which calls the hrtimer callback from
587 * tasklet/softirq context
588 */
589static void __tasklet_hrtimer_trampoline(unsigned long data)
590{
591	struct tasklet_hrtimer *ttimer = (void *)data;
592	enum hrtimer_restart restart;
593
594	restart = ttimer->function(&ttimer->timer);
595	if (restart != HRTIMER_NORESTART)
596		hrtimer_restart(&ttimer->timer);
597}
598
599/**
600 * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks
601 * @ttimer:	 tasklet_hrtimer which is initialized
602 * @function:	 hrtimer callback function which gets called from softirq context
603 * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME)
604 * @mode:	 hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL)
605 */
606void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer,
607			  enum hrtimer_restart (*function)(struct hrtimer *),
608			  clockid_t which_clock, enum hrtimer_mode mode)
609{
610	hrtimer_init(&ttimer->timer, which_clock, mode);
611	ttimer->timer.function = __hrtimer_tasklet_trampoline;
612	tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline,
613		     (unsigned long)ttimer);
614	ttimer->function = function;
615}
616EXPORT_SYMBOL_GPL(tasklet_hrtimer_init);
617
618void __init softirq_init(void)
619{
620	int cpu;
621
622	for_each_possible_cpu(cpu) {
623		per_cpu(tasklet_vec, cpu).tail =
624			&per_cpu(tasklet_vec, cpu).head;
625		per_cpu(tasklet_hi_vec, cpu).tail =
626			&per_cpu(tasklet_hi_vec, cpu).head;
627	}
628
629	open_softirq(TASKLET_SOFTIRQ, tasklet_action);
630	open_softirq(HI_SOFTIRQ, tasklet_hi_action);
631}
632
633static int ksoftirqd_should_run(unsigned int cpu)
634{
635	return local_softirq_pending();
636}
637
638static void run_ksoftirqd(unsigned int cpu)
639{
640	local_irq_disable();
641	if (local_softirq_pending()) {
642		/*
643		 * We can safely run softirq on inline stack, as we are not deep
644		 * in the task stack here.
645		 */
646		__do_softirq();
647		local_irq_enable();
648		cond_resched();
649		return;
650	}
651	local_irq_enable();
652}
653
654#ifdef CONFIG_HOTPLUG_CPU
655/*
656 * tasklet_kill_immediate is called to remove a tasklet which can already be
657 * scheduled for execution on @cpu.
658 *
659 * Unlike tasklet_kill, this function removes the tasklet
660 * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state.
661 *
662 * When this function is called, @cpu must be in the CPU_DEAD state.
663 */
664void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu)
665{
666	struct tasklet_struct **i;
667
668	BUG_ON(cpu_online(cpu));
669	BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state));
670
671	if (!test_bit(TASKLET_STATE_SCHED, &t->state))
672		return;
673
674	/* CPU is dead, so no lock needed. */
675	for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) {
676		if (*i == t) {
677			*i = t->next;
678			/* If this was the tail element, move the tail ptr */
679			if (*i == NULL)
680				per_cpu(tasklet_vec, cpu).tail = i;
681			return;
682		}
683	}
684	BUG();
685}
686
687static int takeover_tasklets(unsigned int cpu)
688{
689	/* CPU is dead, so no lock needed. */
690	local_irq_disable();
691
692	/* Find end, append list for that CPU. */
693	if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) {
694		*__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head;
695		this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail);
696		per_cpu(tasklet_vec, cpu).head = NULL;
697		per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head;
698	}
699	raise_softirq_irqoff(TASKLET_SOFTIRQ);
700
701	if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) {
702		*__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head;
703		__this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail);
704		per_cpu(tasklet_hi_vec, cpu).head = NULL;
705		per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head;
706	}
707	raise_softirq_irqoff(HI_SOFTIRQ);
708
709	local_irq_enable();
710	return 0;
711}
712#else
713#define takeover_tasklets	NULL
714#endif /* CONFIG_HOTPLUG_CPU */
715
716static struct smp_hotplug_thread softirq_threads = {
717	.store			= &ksoftirqd,
718	.thread_should_run	= ksoftirqd_should_run,
719	.thread_fn		= run_ksoftirqd,
720	.thread_comm		= "ksoftirqd/%u",
721};
722
723static __init int spawn_ksoftirqd(void)
724{
725	cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL,
726				  takeover_tasklets);
727	BUG_ON(smpboot_register_percpu_thread(&softirq_threads));
728
729	return 0;
730}
731early_initcall(spawn_ksoftirqd);
732
733/*
734 * [ These __weak aliases are kept in a separate compilation unit, so that
735 *   GCC does not inline them incorrectly. ]
736 */
737
738int __init __weak early_irq_init(void)
739{
740	return 0;
741}
742
743int __init __weak arch_probe_nr_irqs(void)
744{
745	return NR_IRQS_LEGACY;
746}
747
748int __init __weak arch_early_irq_init(void)
749{
750	return 0;
751}
752
753unsigned int __weak arch_dynirq_lower_bound(unsigned int from)
754{
755	return from;
756}
v6.8
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 *	linux/kernel/softirq.c
  4 *
  5 *	Copyright (C) 1992 Linus Torvalds
  6 *
 
 
  7 *	Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903)
  8 */
  9
 10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 11
 12#include <linux/export.h>
 13#include <linux/kernel_stat.h>
 14#include <linux/interrupt.h>
 15#include <linux/init.h>
 16#include <linux/local_lock.h>
 17#include <linux/mm.h>
 18#include <linux/notifier.h>
 19#include <linux/percpu.h>
 20#include <linux/cpu.h>
 21#include <linux/freezer.h>
 22#include <linux/kthread.h>
 23#include <linux/rcupdate.h>
 24#include <linux/ftrace.h>
 25#include <linux/smp.h>
 26#include <linux/smpboot.h>
 27#include <linux/tick.h>
 28#include <linux/irq.h>
 29#include <linux/wait_bit.h>
 30
 31#include <asm/softirq_stack.h>
 32
 33#define CREATE_TRACE_POINTS
 34#include <trace/events/irq.h>
 35
 36/*
 37   - No shared variables, all the data are CPU local.
 38   - If a softirq needs serialization, let it serialize itself
 39     by its own spinlocks.
 40   - Even if softirq is serialized, only local cpu is marked for
 41     execution. Hence, we get something sort of weak cpu binding.
 42     Though it is still not clear, will it result in better locality
 43     or will not.
 44
 45   Examples:
 46   - NET RX softirq. It is multithreaded and does not require
 47     any global serialization.
 48   - NET TX softirq. It kicks software netdevice queues, hence
 49     it is logically serialized per device, but this serialization
 50     is invisible to common code.
 51   - Tasklets: serialized wrt itself.
 52 */
 53
 54#ifndef __ARCH_IRQ_STAT
 55DEFINE_PER_CPU_ALIGNED(irq_cpustat_t, irq_stat);
 56EXPORT_PER_CPU_SYMBOL(irq_stat);
 57#endif
 58
 59static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp;
 60
 61DEFINE_PER_CPU(struct task_struct *, ksoftirqd);
 62
 63const char * const softirq_to_name[NR_SOFTIRQS] = {
 64	"HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "IRQ_POLL",
 65	"TASKLET", "SCHED", "HRTIMER", "RCU"
 66};
 67
 68/*
 69 * we cannot loop indefinitely here to avoid userspace starvation,
 70 * but we also don't want to introduce a worst case 1/HZ latency
 71 * to the pending events, so lets the scheduler to balance
 72 * the softirq load for us.
 73 */
 74static void wakeup_softirqd(void)
 75{
 76	/* Interrupts are disabled: no need to stop preemption */
 77	struct task_struct *tsk = __this_cpu_read(ksoftirqd);
 78
 79	if (tsk)
 80		wake_up_process(tsk);
 81}
 82
 83#ifdef CONFIG_TRACE_IRQFLAGS
 84DEFINE_PER_CPU(int, hardirqs_enabled);
 85DEFINE_PER_CPU(int, hardirq_context);
 86EXPORT_PER_CPU_SYMBOL_GPL(hardirqs_enabled);
 87EXPORT_PER_CPU_SYMBOL_GPL(hardirq_context);
 88#endif
 89
 90/*
 91 * SOFTIRQ_OFFSET usage:
 92 *
 93 * On !RT kernels 'count' is the preempt counter, on RT kernels this applies
 94 * to a per CPU counter and to task::softirqs_disabled_cnt.
 95 *
 96 * - count is changed by SOFTIRQ_OFFSET on entering or leaving softirq
 97 *   processing.
 98 *
 99 * - count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET)
100 *   on local_bh_disable or local_bh_enable.
101 *
102 * This lets us distinguish between whether we are currently processing
103 * softirq and whether we just have bh disabled.
104 */
105#ifdef CONFIG_PREEMPT_RT
106
107/*
108 * RT accounts for BH disabled sections in task::softirqs_disabled_cnt and
109 * also in per CPU softirq_ctrl::cnt. This is necessary to allow tasks in a
110 * softirq disabled section to be preempted.
111 *
112 * The per task counter is used for softirq_count(), in_softirq() and
113 * in_serving_softirqs() because these counts are only valid when the task
114 * holding softirq_ctrl::lock is running.
115 *
116 * The per CPU counter prevents pointless wakeups of ksoftirqd in case that
117 * the task which is in a softirq disabled section is preempted or blocks.
118 */
119struct softirq_ctrl {
120	local_lock_t	lock;
121	int		cnt;
122};
123
124static DEFINE_PER_CPU(struct softirq_ctrl, softirq_ctrl) = {
125	.lock	= INIT_LOCAL_LOCK(softirq_ctrl.lock),
126};
127
128/**
129 * local_bh_blocked() - Check for idle whether BH processing is blocked
130 *
131 * Returns false if the per CPU softirq::cnt is 0 otherwise true.
132 *
133 * This is invoked from the idle task to guard against false positive
134 * softirq pending warnings, which would happen when the task which holds
135 * softirq_ctrl::lock was the only running task on the CPU and blocks on
136 * some other lock.
137 */
138bool local_bh_blocked(void)
139{
140	return __this_cpu_read(softirq_ctrl.cnt) != 0;
141}
142
143void __local_bh_disable_ip(unsigned long ip, unsigned int cnt)
144{
145	unsigned long flags;
146	int newcnt;
147
148	WARN_ON_ONCE(in_hardirq());
149
150	/* First entry of a task into a BH disabled section? */
151	if (!current->softirq_disable_cnt) {
152		if (preemptible()) {
153			local_lock(&softirq_ctrl.lock);
154			/* Required to meet the RCU bottomhalf requirements. */
155			rcu_read_lock();
156		} else {
157			DEBUG_LOCKS_WARN_ON(this_cpu_read(softirq_ctrl.cnt));
158		}
159	}
160
161	/*
162	 * Track the per CPU softirq disabled state. On RT this is per CPU
163	 * state to allow preemption of bottom half disabled sections.
164	 */
165	newcnt = __this_cpu_add_return(softirq_ctrl.cnt, cnt);
166	/*
167	 * Reflect the result in the task state to prevent recursion on the
168	 * local lock and to make softirq_count() & al work.
169	 */
170	current->softirq_disable_cnt = newcnt;
171
172	if (IS_ENABLED(CONFIG_TRACE_IRQFLAGS) && newcnt == cnt) {
173		raw_local_irq_save(flags);
174		lockdep_softirqs_off(ip);
175		raw_local_irq_restore(flags);
176	}
177}
178EXPORT_SYMBOL(__local_bh_disable_ip);
179
180static void __local_bh_enable(unsigned int cnt, bool unlock)
181{
182	unsigned long flags;
183	int newcnt;
184
185	DEBUG_LOCKS_WARN_ON(current->softirq_disable_cnt !=
186			    this_cpu_read(softirq_ctrl.cnt));
187
188	if (IS_ENABLED(CONFIG_TRACE_IRQFLAGS) && softirq_count() == cnt) {
189		raw_local_irq_save(flags);
190		lockdep_softirqs_on(_RET_IP_);
191		raw_local_irq_restore(flags);
192	}
193
194	newcnt = __this_cpu_sub_return(softirq_ctrl.cnt, cnt);
195	current->softirq_disable_cnt = newcnt;
196
197	if (!newcnt && unlock) {
198		rcu_read_unlock();
199		local_unlock(&softirq_ctrl.lock);
200	}
201}
202
203void __local_bh_enable_ip(unsigned long ip, unsigned int cnt)
204{
205	bool preempt_on = preemptible();
206	unsigned long flags;
207	u32 pending;
208	int curcnt;
209
210	WARN_ON_ONCE(in_hardirq());
211	lockdep_assert_irqs_enabled();
212
213	local_irq_save(flags);
214	curcnt = __this_cpu_read(softirq_ctrl.cnt);
215
216	/*
217	 * If this is not reenabling soft interrupts, no point in trying to
218	 * run pending ones.
219	 */
220	if (curcnt != cnt)
221		goto out;
222
223	pending = local_softirq_pending();
224	if (!pending)
225		goto out;
226
227	/*
228	 * If this was called from non preemptible context, wake up the
229	 * softirq daemon.
230	 */
231	if (!preempt_on) {
232		wakeup_softirqd();
233		goto out;
234	}
235
236	/*
237	 * Adjust softirq count to SOFTIRQ_OFFSET which makes
238	 * in_serving_softirq() become true.
239	 */
240	cnt = SOFTIRQ_OFFSET;
241	__local_bh_enable(cnt, false);
242	__do_softirq();
243
244out:
245	__local_bh_enable(cnt, preempt_on);
246	local_irq_restore(flags);
247}
248EXPORT_SYMBOL(__local_bh_enable_ip);
249
250/*
251 * Invoked from ksoftirqd_run() outside of the interrupt disabled section
252 * to acquire the per CPU local lock for reentrancy protection.
253 */
254static inline void ksoftirqd_run_begin(void)
255{
256	__local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET);
257	local_irq_disable();
258}
259
260/* Counterpart to ksoftirqd_run_begin() */
261static inline void ksoftirqd_run_end(void)
262{
263	__local_bh_enable(SOFTIRQ_OFFSET, true);
264	WARN_ON_ONCE(in_interrupt());
265	local_irq_enable();
266}
267
268static inline void softirq_handle_begin(void) { }
269static inline void softirq_handle_end(void) { }
270
271static inline bool should_wake_ksoftirqd(void)
272{
273	return !this_cpu_read(softirq_ctrl.cnt);
274}
275
276static inline void invoke_softirq(void)
277{
278	if (should_wake_ksoftirqd())
279		wakeup_softirqd();
280}
281
282/*
283 * flush_smp_call_function_queue() can raise a soft interrupt in a function
284 * call. On RT kernels this is undesired and the only known functionality
285 * in the block layer which does this is disabled on RT. If soft interrupts
286 * get raised which haven't been raised before the flush, warn so it can be
287 * investigated.
288 */
289void do_softirq_post_smp_call_flush(unsigned int was_pending)
290{
291	if (WARN_ON_ONCE(was_pending != local_softirq_pending()))
292		invoke_softirq();
293}
294
295#else /* CONFIG_PREEMPT_RT */
296
297/*
298 * This one is for softirq.c-internal use, where hardirqs are disabled
299 * legitimately:
300 */
301#ifdef CONFIG_TRACE_IRQFLAGS
302void __local_bh_disable_ip(unsigned long ip, unsigned int cnt)
303{
304	unsigned long flags;
305
306	WARN_ON_ONCE(in_hardirq());
307
308	raw_local_irq_save(flags);
309	/*
310	 * The preempt tracer hooks into preempt_count_add and will break
311	 * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET
312	 * is set and before current->softirq_enabled is cleared.
313	 * We must manually increment preempt_count here and manually
314	 * call the trace_preempt_off later.
315	 */
316	__preempt_count_add(cnt);
317	/*
318	 * Were softirqs turned off above:
319	 */
320	if (softirq_count() == (cnt & SOFTIRQ_MASK))
321		lockdep_softirqs_off(ip);
322	raw_local_irq_restore(flags);
323
324	if (preempt_count() == cnt) {
325#ifdef CONFIG_DEBUG_PREEMPT
326		current->preempt_disable_ip = get_lock_parent_ip();
327#endif
328		trace_preempt_off(CALLER_ADDR0, get_lock_parent_ip());
329	}
330}
331EXPORT_SYMBOL(__local_bh_disable_ip);
332#endif /* CONFIG_TRACE_IRQFLAGS */
333
334static void __local_bh_enable(unsigned int cnt)
335{
336	lockdep_assert_irqs_disabled();
337
338	if (preempt_count() == cnt)
339		trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip());
340
341	if (softirq_count() == (cnt & SOFTIRQ_MASK))
342		lockdep_softirqs_on(_RET_IP_);
343
344	__preempt_count_sub(cnt);
345}
346
347/*
348 * Special-case - softirqs can safely be enabled by __do_softirq(),
 
349 * without processing still-pending softirqs:
350 */
351void _local_bh_enable(void)
352{
353	WARN_ON_ONCE(in_hardirq());
354	__local_bh_enable(SOFTIRQ_DISABLE_OFFSET);
355}
356EXPORT_SYMBOL(_local_bh_enable);
357
358void __local_bh_enable_ip(unsigned long ip, unsigned int cnt)
359{
360	WARN_ON_ONCE(in_hardirq());
361	lockdep_assert_irqs_enabled();
362#ifdef CONFIG_TRACE_IRQFLAGS
363	local_irq_disable();
364#endif
365	/*
366	 * Are softirqs going to be turned on now:
367	 */
368	if (softirq_count() == SOFTIRQ_DISABLE_OFFSET)
369		lockdep_softirqs_on(ip);
370	/*
371	 * Keep preemption disabled until we are done with
372	 * softirq processing:
373	 */
374	__preempt_count_sub(cnt - 1);
375
376	if (unlikely(!in_interrupt() && local_softirq_pending())) {
377		/*
378		 * Run softirq if any pending. And do it in its own stack
379		 * as we may be calling this deep in a task call stack already.
380		 */
381		do_softirq();
382	}
383
384	preempt_count_dec();
385#ifdef CONFIG_TRACE_IRQFLAGS
386	local_irq_enable();
387#endif
388	preempt_check_resched();
389}
390EXPORT_SYMBOL(__local_bh_enable_ip);
391
392static inline void softirq_handle_begin(void)
393{
394	__local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET);
395}
396
397static inline void softirq_handle_end(void)
398{
399	__local_bh_enable(SOFTIRQ_OFFSET);
400	WARN_ON_ONCE(in_interrupt());
401}
402
403static inline void ksoftirqd_run_begin(void)
404{
405	local_irq_disable();
406}
407
408static inline void ksoftirqd_run_end(void)
409{
410	local_irq_enable();
411}
412
413static inline bool should_wake_ksoftirqd(void)
414{
415	return true;
416}
417
418static inline void invoke_softirq(void)
419{
420	if (!force_irqthreads() || !__this_cpu_read(ksoftirqd)) {
421#ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK
422		/*
423		 * We can safely execute softirq on the current stack if
424		 * it is the irq stack, because it should be near empty
425		 * at this stage.
426		 */
427		__do_softirq();
428#else
429		/*
430		 * Otherwise, irq_exit() is called on the task stack that can
431		 * be potentially deep already. So call softirq in its own stack
432		 * to prevent from any overrun.
433		 */
434		do_softirq_own_stack();
435#endif
436	} else {
437		wakeup_softirqd();
438	}
439}
440
441asmlinkage __visible void do_softirq(void)
442{
443	__u32 pending;
444	unsigned long flags;
445
446	if (in_interrupt())
447		return;
448
449	local_irq_save(flags);
450
451	pending = local_softirq_pending();
452
453	if (pending)
454		do_softirq_own_stack();
455
456	local_irq_restore(flags);
457}
458
459#endif /* !CONFIG_PREEMPT_RT */
460
461/*
462 * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times,
463 * but break the loop if need_resched() is set or after 2 ms.
464 * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in
465 * certain cases, such as stop_machine(), jiffies may cease to
466 * increment and so we need the MAX_SOFTIRQ_RESTART limit as
467 * well to make sure we eventually return from this method.
468 *
469 * These limits have been established via experimentation.
470 * The two things to balance is latency against fairness -
471 * we want to handle softirqs as soon as possible, but they
472 * should not be able to lock up the box.
473 */
474#define MAX_SOFTIRQ_TIME  msecs_to_jiffies(2)
475#define MAX_SOFTIRQ_RESTART 10
476
477#ifdef CONFIG_TRACE_IRQFLAGS
478/*
479 * When we run softirqs from irq_exit() and thus on the hardirq stack we need
480 * to keep the lockdep irq context tracking as tight as possible in order to
481 * not miss-qualify lock contexts and miss possible deadlocks.
482 */
483
484static inline bool lockdep_softirq_start(void)
485{
486	bool in_hardirq = false;
487
488	if (lockdep_hardirq_context()) {
489		in_hardirq = true;
490		lockdep_hardirq_exit();
491	}
492
493	lockdep_softirq_enter();
494
495	return in_hardirq;
496}
497
498static inline void lockdep_softirq_end(bool in_hardirq)
499{
500	lockdep_softirq_exit();
501
502	if (in_hardirq)
503		lockdep_hardirq_enter();
504}
505#else
506static inline bool lockdep_softirq_start(void) { return false; }
507static inline void lockdep_softirq_end(bool in_hardirq) { }
508#endif
509
510asmlinkage __visible void __softirq_entry __do_softirq(void)
511{
512	unsigned long end = jiffies + MAX_SOFTIRQ_TIME;
513	unsigned long old_flags = current->flags;
514	int max_restart = MAX_SOFTIRQ_RESTART;
515	struct softirq_action *h;
516	bool in_hardirq;
517	__u32 pending;
518	int softirq_bit;
519
520	/*
521	 * Mask out PF_MEMALLOC as the current task context is borrowed for the
522	 * softirq. A softirq handled, such as network RX, might set PF_MEMALLOC
523	 * again if the socket is related to swapping.
524	 */
525	current->flags &= ~PF_MEMALLOC;
526
527	pending = local_softirq_pending();
 
528
529	softirq_handle_begin();
530	in_hardirq = lockdep_softirq_start();
531	account_softirq_enter(current);
532
533restart:
534	/* Reset the pending bitmask before enabling irqs */
535	set_softirq_pending(0);
536
537	local_irq_enable();
538
539	h = softirq_vec;
540
541	while ((softirq_bit = ffs(pending))) {
542		unsigned int vec_nr;
543		int prev_count;
544
545		h += softirq_bit - 1;
546
547		vec_nr = h - softirq_vec;
548		prev_count = preempt_count();
549
550		kstat_incr_softirqs_this_cpu(vec_nr);
551
552		trace_softirq_entry(vec_nr);
553		h->action(h);
554		trace_softirq_exit(vec_nr);
555		if (unlikely(prev_count != preempt_count())) {
556			pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n",
557			       vec_nr, softirq_to_name[vec_nr], h->action,
558			       prev_count, preempt_count());
559			preempt_count_set(prev_count);
560		}
561		h++;
562		pending >>= softirq_bit;
563	}
564
565	if (!IS_ENABLED(CONFIG_PREEMPT_RT) &&
566	    __this_cpu_read(ksoftirqd) == current)
567		rcu_softirq_qs();
568
569	local_irq_disable();
570
571	pending = local_softirq_pending();
572	if (pending) {
573		if (time_before(jiffies, end) && !need_resched() &&
574		    --max_restart)
575			goto restart;
576
577		wakeup_softirqd();
578	}
579
580	account_softirq_exit(current);
581	lockdep_softirq_end(in_hardirq);
582	softirq_handle_end();
 
 
583	current_restore_flags(old_flags, PF_MEMALLOC);
584}
585
586/**
587 * irq_enter_rcu - Enter an interrupt context with RCU watching
588 */
589void irq_enter_rcu(void)
590{
591	__irq_enter_raw();
 
 
 
 
592
593	if (tick_nohz_full_cpu(smp_processor_id()) ||
594	    (is_idle_task(current) && (irq_count() == HARDIRQ_OFFSET)))
595		tick_irq_enter();
 
 
 
596
597	account_hardirq_enter(current);
598}
599
600/**
601 * irq_enter - Enter an interrupt context including RCU update
602 */
603void irq_enter(void)
604{
605	ct_irq_enter();
606	irq_enter_rcu();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
607}
608
609static inline void tick_irq_exit(void)
610{
611#ifdef CONFIG_NO_HZ_COMMON
612	int cpu = smp_processor_id();
613
614	/* Make sure that timer wheel updates are propagated */
615	if ((sched_core_idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) {
616		if (!in_hardirq())
617			tick_nohz_irq_exit();
618	}
619#endif
620}
621
622static inline void __irq_exit_rcu(void)
 
 
 
623{
624#ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED
625	local_irq_disable();
626#else
627	lockdep_assert_irqs_disabled();
628#endif
629	account_hardirq_exit(current);
630	preempt_count_sub(HARDIRQ_OFFSET);
631	if (!in_interrupt() && local_softirq_pending())
632		invoke_softirq();
633
634	tick_irq_exit();
635}
636
637/**
638 * irq_exit_rcu() - Exit an interrupt context without updating RCU
639 *
640 * Also processes softirqs if needed and possible.
641 */
642void irq_exit_rcu(void)
643{
644	__irq_exit_rcu();
645	 /* must be last! */
646	lockdep_hardirq_exit();
647}
648
649/**
650 * irq_exit - Exit an interrupt context, update RCU and lockdep
651 *
652 * Also processes softirqs if needed and possible.
653 */
654void irq_exit(void)
655{
656	__irq_exit_rcu();
657	ct_irq_exit();
658	 /* must be last! */
659	lockdep_hardirq_exit();
660}
661
662/*
663 * This function must run with irqs disabled!
664 */
665inline void raise_softirq_irqoff(unsigned int nr)
666{
667	__raise_softirq_irqoff(nr);
668
669	/*
670	 * If we're in an interrupt or softirq, we're done
671	 * (this also catches softirq-disabled code). We will
672	 * actually run the softirq once we return from
673	 * the irq or softirq.
674	 *
675	 * Otherwise we wake up ksoftirqd to make sure we
676	 * schedule the softirq soon.
677	 */
678	if (!in_interrupt() && should_wake_ksoftirqd())
679		wakeup_softirqd();
680}
681
682void raise_softirq(unsigned int nr)
683{
684	unsigned long flags;
685
686	local_irq_save(flags);
687	raise_softirq_irqoff(nr);
688	local_irq_restore(flags);
689}
690
691void __raise_softirq_irqoff(unsigned int nr)
692{
693	lockdep_assert_irqs_disabled();
694	trace_softirq_raise(nr);
695	or_softirq_pending(1UL << nr);
696}
697
698void open_softirq(int nr, void (*action)(struct softirq_action *))
699{
700	softirq_vec[nr].action = action;
701}
702
703/*
704 * Tasklets
705 */
706struct tasklet_head {
707	struct tasklet_struct *head;
708	struct tasklet_struct **tail;
709};
710
711static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec);
712static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec);
713
714static void __tasklet_schedule_common(struct tasklet_struct *t,
715				      struct tasklet_head __percpu *headp,
716				      unsigned int softirq_nr)
717{
718	struct tasklet_head *head;
719	unsigned long flags;
720
721	local_irq_save(flags);
722	head = this_cpu_ptr(headp);
723	t->next = NULL;
724	*head->tail = t;
725	head->tail = &(t->next);
726	raise_softirq_irqoff(softirq_nr);
727	local_irq_restore(flags);
728}
729
730void __tasklet_schedule(struct tasklet_struct *t)
731{
732	__tasklet_schedule_common(t, &tasklet_vec,
733				  TASKLET_SOFTIRQ);
734}
735EXPORT_SYMBOL(__tasklet_schedule);
736
737void __tasklet_hi_schedule(struct tasklet_struct *t)
738{
739	__tasklet_schedule_common(t, &tasklet_hi_vec,
740				  HI_SOFTIRQ);
741}
742EXPORT_SYMBOL(__tasklet_hi_schedule);
743
744static bool tasklet_clear_sched(struct tasklet_struct *t)
745{
746	if (test_and_clear_bit(TASKLET_STATE_SCHED, &t->state)) {
747		wake_up_var(&t->state);
748		return true;
749	}
750
751	WARN_ONCE(1, "tasklet SCHED state not set: %s %pS\n",
752		  t->use_callback ? "callback" : "func",
753		  t->use_callback ? (void *)t->callback : (void *)t->func);
754
755	return false;
756}
757
758static void tasklet_action_common(struct softirq_action *a,
759				  struct tasklet_head *tl_head,
760				  unsigned int softirq_nr)
761{
762	struct tasklet_struct *list;
763
764	local_irq_disable();
765	list = tl_head->head;
766	tl_head->head = NULL;
767	tl_head->tail = &tl_head->head;
768	local_irq_enable();
769
770	while (list) {
771		struct tasklet_struct *t = list;
772
773		list = list->next;
774
775		if (tasklet_trylock(t)) {
776			if (!atomic_read(&t->count)) {
777				if (tasklet_clear_sched(t)) {
778					if (t->use_callback) {
779						trace_tasklet_entry(t, t->callback);
780						t->callback(t);
781						trace_tasklet_exit(t, t->callback);
782					} else {
783						trace_tasklet_entry(t, t->func);
784						t->func(t->data);
785						trace_tasklet_exit(t, t->func);
786					}
787				}
788				tasklet_unlock(t);
789				continue;
790			}
791			tasklet_unlock(t);
792		}
793
794		local_irq_disable();
795		t->next = NULL;
796		*tl_head->tail = t;
797		tl_head->tail = &t->next;
798		__raise_softirq_irqoff(softirq_nr);
799		local_irq_enable();
800	}
801}
802
803static __latent_entropy void tasklet_action(struct softirq_action *a)
804{
805	tasklet_action_common(a, this_cpu_ptr(&tasklet_vec), TASKLET_SOFTIRQ);
806}
807
808static __latent_entropy void tasklet_hi_action(struct softirq_action *a)
809{
810	tasklet_action_common(a, this_cpu_ptr(&tasklet_hi_vec), HI_SOFTIRQ);
811}
812
813void tasklet_setup(struct tasklet_struct *t,
814		   void (*callback)(struct tasklet_struct *))
815{
816	t->next = NULL;
817	t->state = 0;
818	atomic_set(&t->count, 0);
819	t->callback = callback;
820	t->use_callback = true;
821	t->data = 0;
822}
823EXPORT_SYMBOL(tasklet_setup);
824
825void tasklet_init(struct tasklet_struct *t,
826		  void (*func)(unsigned long), unsigned long data)
827{
828	t->next = NULL;
829	t->state = 0;
830	atomic_set(&t->count, 0);
831	t->func = func;
832	t->use_callback = false;
833	t->data = data;
834}
835EXPORT_SYMBOL(tasklet_init);
836
837#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
838/*
839 * Do not use in new code. Waiting for tasklets from atomic contexts is
840 * error prone and should be avoided.
841 */
842void tasklet_unlock_spin_wait(struct tasklet_struct *t)
843{
844	while (test_bit(TASKLET_STATE_RUN, &(t)->state)) {
845		if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
846			/*
847			 * Prevent a live lock when current preempted soft
848			 * interrupt processing or prevents ksoftirqd from
849			 * running. If the tasklet runs on a different CPU
850			 * then this has no effect other than doing the BH
851			 * disable/enable dance for nothing.
852			 */
853			local_bh_disable();
854			local_bh_enable();
855		} else {
856			cpu_relax();
857		}
858	}
859}
860EXPORT_SYMBOL(tasklet_unlock_spin_wait);
861#endif
862
863void tasklet_kill(struct tasklet_struct *t)
864{
865	if (in_interrupt())
866		pr_notice("Attempt to kill tasklet from interrupt\n");
867
868	while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state))
869		wait_var_event(&t->state, !test_bit(TASKLET_STATE_SCHED, &t->state));
870
 
 
871	tasklet_unlock_wait(t);
872	tasklet_clear_sched(t);
873}
874EXPORT_SYMBOL(tasklet_kill);
875
876#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT_RT)
877void tasklet_unlock(struct tasklet_struct *t)
 
 
 
 
 
 
 
 
878{
879	smp_mb__before_atomic();
880	clear_bit(TASKLET_STATE_RUN, &t->state);
881	smp_mb__after_atomic();
882	wake_up_var(&t->state);
 
883}
884EXPORT_SYMBOL_GPL(tasklet_unlock);
885
886void tasklet_unlock_wait(struct tasklet_struct *t)
 
 
 
 
887{
888	wait_var_event(&t->state, !test_bit(TASKLET_STATE_RUN, &t->state));
 
 
 
 
 
889}
890EXPORT_SYMBOL_GPL(tasklet_unlock_wait);
891#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
892
893void __init softirq_init(void)
894{
895	int cpu;
896
897	for_each_possible_cpu(cpu) {
898		per_cpu(tasklet_vec, cpu).tail =
899			&per_cpu(tasklet_vec, cpu).head;
900		per_cpu(tasklet_hi_vec, cpu).tail =
901			&per_cpu(tasklet_hi_vec, cpu).head;
902	}
903
904	open_softirq(TASKLET_SOFTIRQ, tasklet_action);
905	open_softirq(HI_SOFTIRQ, tasklet_hi_action);
906}
907
908static int ksoftirqd_should_run(unsigned int cpu)
909{
910	return local_softirq_pending();
911}
912
913static void run_ksoftirqd(unsigned int cpu)
914{
915	ksoftirqd_run_begin();
916	if (local_softirq_pending()) {
917		/*
918		 * We can safely run softirq on inline stack, as we are not deep
919		 * in the task stack here.
920		 */
921		__do_softirq();
922		ksoftirqd_run_end();
923		cond_resched();
924		return;
925	}
926	ksoftirqd_run_end();
927}
928
929#ifdef CONFIG_HOTPLUG_CPU
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
930static int takeover_tasklets(unsigned int cpu)
931{
932	/* CPU is dead, so no lock needed. */
933	local_irq_disable();
934
935	/* Find end, append list for that CPU. */
936	if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) {
937		*__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head;
938		__this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail);
939		per_cpu(tasklet_vec, cpu).head = NULL;
940		per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head;
941	}
942	raise_softirq_irqoff(TASKLET_SOFTIRQ);
943
944	if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) {
945		*__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head;
946		__this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail);
947		per_cpu(tasklet_hi_vec, cpu).head = NULL;
948		per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head;
949	}
950	raise_softirq_irqoff(HI_SOFTIRQ);
951
952	local_irq_enable();
953	return 0;
954}
955#else
956#define takeover_tasklets	NULL
957#endif /* CONFIG_HOTPLUG_CPU */
958
959static struct smp_hotplug_thread softirq_threads = {
960	.store			= &ksoftirqd,
961	.thread_should_run	= ksoftirqd_should_run,
962	.thread_fn		= run_ksoftirqd,
963	.thread_comm		= "ksoftirqd/%u",
964};
965
966static __init int spawn_ksoftirqd(void)
967{
968	cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL,
969				  takeover_tasklets);
970	BUG_ON(smpboot_register_percpu_thread(&softirq_threads));
971
972	return 0;
973}
974early_initcall(spawn_ksoftirqd);
975
976/*
977 * [ These __weak aliases are kept in a separate compilation unit, so that
978 *   GCC does not inline them incorrectly. ]
979 */
980
981int __init __weak early_irq_init(void)
982{
983	return 0;
984}
985
986int __init __weak arch_probe_nr_irqs(void)
987{
988	return NR_IRQS_LEGACY;
989}
990
991int __init __weak arch_early_irq_init(void)
992{
993	return 0;
994}
995
996unsigned int __weak arch_dynirq_lower_bound(unsigned int from)
997{
998	return from;
999}