Linux Audio

Check our new training course

Loading...
v5.4
  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/mm.h>
 17#include <linux/notifier.h>
 18#include <linux/percpu.h>
 19#include <linux/cpu.h>
 20#include <linux/freezer.h>
 21#include <linux/kthread.h>
 22#include <linux/rcupdate.h>
 23#include <linux/ftrace.h>
 24#include <linux/smp.h>
 25#include <linux/smpboot.h>
 26#include <linux/tick.h>
 27#include <linux/irq.h>
 28
 29#define CREATE_TRACE_POINTS
 30#include <trace/events/irq.h>
 31
 
 32/*
 33   - No shared variables, all the data are CPU local.
 34   - If a softirq needs serialization, let it serialize itself
 35     by its own spinlocks.
 36   - Even if softirq is serialized, only local cpu is marked for
 37     execution. Hence, we get something sort of weak cpu binding.
 38     Though it is still not clear, will it result in better locality
 39     or will not.
 40
 41   Examples:
 42   - NET RX softirq. It is multithreaded and does not require
 43     any global serialization.
 44   - NET TX softirq. It kicks software netdevice queues, hence
 45     it is logically serialized per device, but this serialization
 46     is invisible to common code.
 47   - Tasklets: serialized wrt itself.
 48 */
 49
 50#ifndef __ARCH_IRQ_STAT
 51DEFINE_PER_CPU_ALIGNED(irq_cpustat_t, irq_stat);
 52EXPORT_PER_CPU_SYMBOL(irq_stat);
 53#endif
 54
 55static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp;
 56
 57DEFINE_PER_CPU(struct task_struct *, ksoftirqd);
 58
 59const char * const softirq_to_name[NR_SOFTIRQS] = {
 60	"HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "IRQ_POLL",
 61	"TASKLET", "SCHED", "HRTIMER", "RCU"
 62};
 63
 64/*
 65 * we cannot loop indefinitely here to avoid userspace starvation,
 66 * but we also don't want to introduce a worst case 1/HZ latency
 67 * to the pending events, so lets the scheduler to balance
 68 * the softirq load for us.
 69 */
 70static void wakeup_softirqd(void)
 71{
 72	/* Interrupts are disabled: no need to stop preemption */
 73	struct task_struct *tsk = __this_cpu_read(ksoftirqd);
 74
 75	if (tsk && tsk->state != TASK_RUNNING)
 76		wake_up_process(tsk);
 77}
 78
 79/*
 80 * If ksoftirqd is scheduled, we do not want to process pending softirqs
 81 * right now. Let ksoftirqd handle this at its own rate, to get fairness,
 82 * unless we're doing some of the synchronous softirqs.
 83 */
 84#define SOFTIRQ_NOW_MASK ((1 << HI_SOFTIRQ) | (1 << TASKLET_SOFTIRQ))
 85static bool ksoftirqd_running(unsigned long pending)
 86{
 87	struct task_struct *tsk = __this_cpu_read(ksoftirqd);
 88
 89	if (pending & SOFTIRQ_NOW_MASK)
 90		return false;
 91	return tsk && (tsk->state == TASK_RUNNING) &&
 92		!__kthread_should_park(tsk);
 93}
 94
 95/*
 96 * preempt_count and SOFTIRQ_OFFSET usage:
 97 * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving
 98 *   softirq processing.
 99 * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET)
100 *   on local_bh_disable or local_bh_enable.
101 * This lets us distinguish between whether we are currently processing
102 * softirq and whether we just have bh disabled.
103 */
104
105/*
106 * This one is for softirq.c-internal use,
107 * where hardirqs are disabled legitimately:
108 */
109#ifdef CONFIG_TRACE_IRQFLAGS
110void __local_bh_disable_ip(unsigned long ip, unsigned int cnt)
111{
112	unsigned long flags;
113
114	WARN_ON_ONCE(in_irq());
115
116	raw_local_irq_save(flags);
117	/*
118	 * The preempt tracer hooks into preempt_count_add and will break
119	 * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET
120	 * is set and before current->softirq_enabled is cleared.
121	 * We must manually increment preempt_count here and manually
122	 * call the trace_preempt_off later.
123	 */
124	__preempt_count_add(cnt);
125	/*
126	 * Were softirqs turned off above:
127	 */
128	if (softirq_count() == (cnt & SOFTIRQ_MASK))
129		trace_softirqs_off(ip);
130	raw_local_irq_restore(flags);
131
132	if (preempt_count() == cnt) {
133#ifdef CONFIG_DEBUG_PREEMPT
134		current->preempt_disable_ip = get_lock_parent_ip();
135#endif
136		trace_preempt_off(CALLER_ADDR0, get_lock_parent_ip());
137	}
 
 
138}
139EXPORT_SYMBOL(__local_bh_disable_ip);
140#endif /* CONFIG_TRACE_IRQFLAGS */
141
142static void __local_bh_enable(unsigned int cnt)
143{
144	lockdep_assert_irqs_disabled();
 
 
145
146	if (preempt_count() == cnt)
147		trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip());
148
149	if (softirq_count() == (cnt & SOFTIRQ_MASK))
150		trace_softirqs_on(_RET_IP_);
 
 
151
152	__preempt_count_sub(cnt);
 
 
153}
154
155/*
156 * Special-case - softirqs can safely be enabled by __do_softirq(),
 
157 * without processing still-pending softirqs:
158 */
159void _local_bh_enable(void)
160{
161	WARN_ON_ONCE(in_irq());
162	__local_bh_enable(SOFTIRQ_DISABLE_OFFSET);
163}
 
164EXPORT_SYMBOL(_local_bh_enable);
165
166void __local_bh_enable_ip(unsigned long ip, unsigned int cnt)
167{
168	WARN_ON_ONCE(in_irq());
169	lockdep_assert_irqs_enabled();
170#ifdef CONFIG_TRACE_IRQFLAGS
171	local_irq_disable();
172#endif
173	/*
174	 * Are softirqs going to be turned on now:
175	 */
176	if (softirq_count() == SOFTIRQ_DISABLE_OFFSET)
177		trace_softirqs_on(ip);
178	/*
179	 * Keep preemption disabled until we are done with
180	 * softirq processing:
181	 */
182	preempt_count_sub(cnt - 1);
183
184	if (unlikely(!in_interrupt() && local_softirq_pending())) {
185		/*
186		 * Run softirq if any pending. And do it in its own stack
187		 * as we may be calling this deep in a task call stack already.
188		 */
189		do_softirq();
190	}
191
192	preempt_count_dec();
193#ifdef CONFIG_TRACE_IRQFLAGS
194	local_irq_enable();
195#endif
196	preempt_check_resched();
197}
198EXPORT_SYMBOL(__local_bh_enable_ip);
 
 
 
 
 
 
 
 
 
 
 
199
200/*
201 * We restart softirq processing for at most MAX_SOFTIRQ_RESTART times,
202 * but break the loop if need_resched() is set or after 2 ms.
203 * The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in
204 * certain cases, such as stop_machine(), jiffies may cease to
205 * increment and so we need the MAX_SOFTIRQ_RESTART limit as
206 * well to make sure we eventually return from this method.
207 *
208 * These limits have been established via experimentation.
209 * The two things to balance is latency against fairness -
210 * we want to handle softirqs as soon as possible, but they
211 * should not be able to lock up the box.
212 */
213#define MAX_SOFTIRQ_TIME  msecs_to_jiffies(2)
214#define MAX_SOFTIRQ_RESTART 10
215
216#ifdef CONFIG_TRACE_IRQFLAGS
217/*
218 * When we run softirqs from irq_exit() and thus on the hardirq stack we need
219 * to keep the lockdep irq context tracking as tight as possible in order to
220 * not miss-qualify lock contexts and miss possible deadlocks.
221 */
222
223static inline bool lockdep_softirq_start(void)
224{
225	bool in_hardirq = false;
226
227	if (trace_hardirq_context(current)) {
228		in_hardirq = true;
229		trace_hardirq_exit();
230	}
231
232	lockdep_softirq_enter();
233
234	return in_hardirq;
235}
236
237static inline void lockdep_softirq_end(bool in_hardirq)
238{
239	lockdep_softirq_exit();
240
241	if (in_hardirq)
242		trace_hardirq_enter();
243}
244#else
245static inline bool lockdep_softirq_start(void) { return false; }
246static inline void lockdep_softirq_end(bool in_hardirq) { }
247#endif
248
249asmlinkage __visible void __softirq_entry __do_softirq(void)
250{
251	unsigned long end = jiffies + MAX_SOFTIRQ_TIME;
252	unsigned long old_flags = current->flags;
253	int max_restart = MAX_SOFTIRQ_RESTART;
254	struct softirq_action *h;
255	bool in_hardirq;
256	__u32 pending;
257	int softirq_bit;
258
259	/*
260	 * Mask out PF_MEMALLOC as the current task context is borrowed for the
261	 * softirq. A softirq handled, such as network RX, might set PF_MEMALLOC
262	 * again if the socket is related to swapping.
263	 */
264	current->flags &= ~PF_MEMALLOC;
265
266	pending = local_softirq_pending();
267	account_irq_enter_time(current);
268
269	__local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET);
270	in_hardirq = lockdep_softirq_start();
 
271
 
272restart:
273	/* Reset the pending bitmask before enabling irqs */
274	set_softirq_pending(0);
275
276	local_irq_enable();
277
278	h = softirq_vec;
279
280	while ((softirq_bit = ffs(pending))) {
281		unsigned int vec_nr;
282		int prev_count;
283
284		h += softirq_bit - 1;
285
286		vec_nr = h - softirq_vec;
287		prev_count = preempt_count();
288
289		kstat_incr_softirqs_this_cpu(vec_nr);
290
291		trace_softirq_entry(vec_nr);
292		h->action(h);
293		trace_softirq_exit(vec_nr);
294		if (unlikely(prev_count != preempt_count())) {
295			pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n",
296			       vec_nr, softirq_to_name[vec_nr], h->action,
297			       prev_count, preempt_count());
298			preempt_count_set(prev_count);
 
299		}
300		h++;
301		pending >>= softirq_bit;
302	}
303
304	if (__this_cpu_read(ksoftirqd) == current)
305		rcu_softirq_qs();
306	local_irq_disable();
307
308	pending = local_softirq_pending();
309	if (pending) {
310		if (time_before(jiffies, end) && !need_resched() &&
311		    --max_restart)
312			goto restart;
313
 
314		wakeup_softirqd();
315	}
316
317	lockdep_softirq_end(in_hardirq);
318	account_irq_exit_time(current);
 
319	__local_bh_enable(SOFTIRQ_OFFSET);
320	WARN_ON_ONCE(in_interrupt());
321	current_restore_flags(old_flags, PF_MEMALLOC);
322}
323
324asmlinkage __visible void do_softirq(void)
 
 
325{
326	__u32 pending;
327	unsigned long flags;
328
329	if (in_interrupt())
330		return;
331
332	local_irq_save(flags);
333
334	pending = local_softirq_pending();
335
336	if (pending && !ksoftirqd_running(pending))
337		do_softirq_own_stack();
338
339	local_irq_restore(flags);
340}
341
 
 
342/*
343 * Enter an interrupt context.
344 */
345void irq_enter(void)
346{
 
 
347	rcu_irq_enter();
348	if (is_idle_task(current) && !in_interrupt()) {
349		/*
350		 * Prevent raise_softirq from needlessly waking up ksoftirqd
351		 * here, as softirq will be serviced on return from interrupt.
352		 */
353		local_bh_disable();
354		tick_irq_enter();
355		_local_bh_enable();
356	}
357
358	__irq_enter();
359}
360
361static inline void invoke_softirq(void)
362{
363	if (ksoftirqd_running(local_softirq_pending()))
364		return;
365
366	if (!force_irqthreads) {
367#ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK
368		/*
369		 * We can safely execute softirq on the current stack if
370		 * it is the irq stack, because it should be near empty
371		 * at this stage.
372		 */
373		__do_softirq();
374#else
375		/*
376		 * Otherwise, irq_exit() is called on the task stack that can
377		 * be potentially deep already. So call softirq in its own stack
378		 * to prevent from any overrun.
379		 */
380		do_softirq_own_stack();
381#endif
382	} else {
 
 
383		wakeup_softirqd();
 
384	}
385}
386
387static inline void tick_irq_exit(void)
388{
389#ifdef CONFIG_NO_HZ_COMMON
390	int cpu = smp_processor_id();
391
392	/* Make sure that timer wheel updates are propagated */
393	if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) {
394		if (!in_irq())
395			tick_nohz_irq_exit();
396	}
397#endif
398}
399
400/*
401 * Exit an interrupt context. Process softirqs if needed and possible:
402 */
403void irq_exit(void)
404{
405#ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED
406	local_irq_disable();
407#else
408	lockdep_assert_irqs_disabled();
409#endif
410	account_irq_exit_time(current);
411	preempt_count_sub(HARDIRQ_OFFSET);
412	if (!in_interrupt() && local_softirq_pending())
413		invoke_softirq();
414
415	tick_irq_exit();
 
 
 
 
416	rcu_irq_exit();
417	trace_hardirq_exit(); /* must be last! */
418}
419
420/*
421 * This function must run with irqs disabled!
422 */
423inline void raise_softirq_irqoff(unsigned int nr)
424{
425	__raise_softirq_irqoff(nr);
426
427	/*
428	 * If we're in an interrupt or softirq, we're done
429	 * (this also catches softirq-disabled code). We will
430	 * actually run the softirq once we return from
431	 * the irq or softirq.
432	 *
433	 * Otherwise we wake up ksoftirqd to make sure we
434	 * schedule the softirq soon.
435	 */
436	if (!in_interrupt())
437		wakeup_softirqd();
438}
439
440void raise_softirq(unsigned int nr)
441{
442	unsigned long flags;
443
444	local_irq_save(flags);
445	raise_softirq_irqoff(nr);
446	local_irq_restore(flags);
447}
448
449void __raise_softirq_irqoff(unsigned int nr)
450{
451	trace_softirq_raise(nr);
452	or_softirq_pending(1UL << nr);
453}
454
455void open_softirq(int nr, void (*action)(struct softirq_action *))
456{
457	softirq_vec[nr].action = action;
458}
459
460/*
461 * Tasklets
462 */
463struct tasklet_head {
 
464	struct tasklet_struct *head;
465	struct tasklet_struct **tail;
466};
467
468static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec);
469static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec);
470
471static void __tasklet_schedule_common(struct tasklet_struct *t,
472				      struct tasklet_head __percpu *headp,
473				      unsigned int softirq_nr)
474{
475	struct tasklet_head *head;
476	unsigned long flags;
477
478	local_irq_save(flags);
479	head = this_cpu_ptr(headp);
480	t->next = NULL;
481	*head->tail = t;
482	head->tail = &(t->next);
483	raise_softirq_irqoff(softirq_nr);
484	local_irq_restore(flags);
485}
486
487void __tasklet_schedule(struct tasklet_struct *t)
488{
489	__tasklet_schedule_common(t, &tasklet_vec,
490				  TASKLET_SOFTIRQ);
491}
492EXPORT_SYMBOL(__tasklet_schedule);
493
494void __tasklet_hi_schedule(struct tasklet_struct *t)
495{
496	__tasklet_schedule_common(t, &tasklet_hi_vec,
497				  HI_SOFTIRQ);
 
 
 
 
 
 
498}
 
499EXPORT_SYMBOL(__tasklet_hi_schedule);
500
501static void tasklet_action_common(struct softirq_action *a,
502				  struct tasklet_head *tl_head,
503				  unsigned int softirq_nr)
 
 
 
 
 
 
 
 
 
504{
505	struct tasklet_struct *list;
506
507	local_irq_disable();
508	list = tl_head->head;
509	tl_head->head = NULL;
510	tl_head->tail = &tl_head->head;
511	local_irq_enable();
512
513	while (list) {
514		struct tasklet_struct *t = list;
515
516		list = list->next;
517
518		if (tasklet_trylock(t)) {
519			if (!atomic_read(&t->count)) {
520				if (!test_and_clear_bit(TASKLET_STATE_SCHED,
521							&t->state))
522					BUG();
523				t->func(t->data);
524				tasklet_unlock(t);
525				continue;
526			}
527			tasklet_unlock(t);
528		}
529
530		local_irq_disable();
531		t->next = NULL;
532		*tl_head->tail = t;
533		tl_head->tail = &t->next;
534		__raise_softirq_irqoff(softirq_nr);
535		local_irq_enable();
536	}
537}
538
539static __latent_entropy void tasklet_action(struct softirq_action *a)
540{
541	tasklet_action_common(a, this_cpu_ptr(&tasklet_vec), TASKLET_SOFTIRQ);
542}
543
544static __latent_entropy void tasklet_hi_action(struct softirq_action *a)
545{
546	tasklet_action_common(a, this_cpu_ptr(&tasklet_hi_vec), HI_SOFTIRQ);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
547}
548
 
549void tasklet_init(struct tasklet_struct *t,
550		  void (*func)(unsigned long), unsigned long data)
551{
552	t->next = NULL;
553	t->state = 0;
554	atomic_set(&t->count, 0);
555	t->func = func;
556	t->data = data;
557}
 
558EXPORT_SYMBOL(tasklet_init);
559
560void tasklet_kill(struct tasklet_struct *t)
561{
562	if (in_interrupt())
563		pr_notice("Attempt to kill tasklet from interrupt\n");
564
565	while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
566		do {
567			yield();
568		} while (test_bit(TASKLET_STATE_SCHED, &t->state));
569	}
570	tasklet_unlock_wait(t);
571	clear_bit(TASKLET_STATE_SCHED, &t->state);
572}
 
573EXPORT_SYMBOL(tasklet_kill);
574
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
575void __init softirq_init(void)
576{
577	int cpu;
578
579	for_each_possible_cpu(cpu) {
 
 
580		per_cpu(tasklet_vec, cpu).tail =
581			&per_cpu(tasklet_vec, cpu).head;
582		per_cpu(tasklet_hi_vec, cpu).tail =
583			&per_cpu(tasklet_hi_vec, cpu).head;
 
 
584	}
585
 
 
586	open_softirq(TASKLET_SOFTIRQ, tasklet_action);
587	open_softirq(HI_SOFTIRQ, tasklet_hi_action);
588}
589
590static int ksoftirqd_should_run(unsigned int cpu)
591{
592	return local_softirq_pending();
593}
594
595static void run_ksoftirqd(unsigned int cpu)
596{
597	local_irq_disable();
598	if (local_softirq_pending()) {
599		/*
600		 * We can safely run softirq on inline stack, as we are not deep
601		 * in the task stack here.
602		 */
603		__do_softirq();
604		local_irq_enable();
605		cond_resched();
606		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
607	}
608	local_irq_enable();
 
 
 
 
 
 
 
 
 
 
 
 
609}
610
611#ifdef CONFIG_HOTPLUG_CPU
612/*
613 * tasklet_kill_immediate is called to remove a tasklet which can already be
614 * scheduled for execution on @cpu.
615 *
616 * Unlike tasklet_kill, this function removes the tasklet
617 * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state.
618 *
619 * When this function is called, @cpu must be in the CPU_DEAD state.
620 */
621void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu)
622{
623	struct tasklet_struct **i;
624
625	BUG_ON(cpu_online(cpu));
626	BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state));
627
628	if (!test_bit(TASKLET_STATE_SCHED, &t->state))
629		return;
630
631	/* CPU is dead, so no lock needed. */
632	for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) {
633		if (*i == t) {
634			*i = t->next;
635			/* If this was the tail element, move the tail ptr */
636			if (*i == NULL)
637				per_cpu(tasklet_vec, cpu).tail = i;
638			return;
639		}
640	}
641	BUG();
642}
643
644static int takeover_tasklets(unsigned int cpu)
645{
646	/* CPU is dead, so no lock needed. */
647	local_irq_disable();
648
649	/* Find end, append list for that CPU. */
650	if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) {
651		*__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head;
652		__this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail);
653		per_cpu(tasklet_vec, cpu).head = NULL;
654		per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head;
655	}
656	raise_softirq_irqoff(TASKLET_SOFTIRQ);
657
658	if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) {
659		*__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head;
660		__this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail);
661		per_cpu(tasklet_hi_vec, cpu).head = NULL;
662		per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head;
663	}
664	raise_softirq_irqoff(HI_SOFTIRQ);
665
666	local_irq_enable();
667	return 0;
668}
669#else
670#define takeover_tasklets	NULL
671#endif /* CONFIG_HOTPLUG_CPU */
672
673static struct smp_hotplug_thread softirq_threads = {
674	.store			= &ksoftirqd,
675	.thread_should_run	= ksoftirqd_should_run,
676	.thread_fn		= run_ksoftirqd,
677	.thread_comm		= "ksoftirqd/%u",
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
678};
679
680static __init int spawn_ksoftirqd(void)
681{
682	cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL,
683				  takeover_tasklets);
684	BUG_ON(smpboot_register_percpu_thread(&softirq_threads));
685
 
 
 
686	return 0;
687}
688early_initcall(spawn_ksoftirqd);
689
690/*
691 * [ These __weak aliases are kept in a separate compilation unit, so that
692 *   GCC does not inline them incorrectly. ]
693 */
694
695int __init __weak early_irq_init(void)
696{
697	return 0;
698}
699
 
700int __init __weak arch_probe_nr_irqs(void)
701{
702	return NR_IRQS_LEGACY;
703}
704
705int __init __weak arch_early_irq_init(void)
706{
707	return 0;
708}
709
710unsigned int __weak arch_dynirq_lower_bound(unsigned int from)
711{
712	return from;
713}
v3.5.6
 
  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 *	Remote softirq infrastructure is by Jens Axboe.
 11 */
 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/tick.h>
 
 27
 28#define CREATE_TRACE_POINTS
 29#include <trace/events/irq.h>
 30
 31#include <asm/irq.h>
 32/*
 33   - No shared variables, all the data are CPU local.
 34   - If a softirq needs serialization, let it serialize itself
 35     by its own spinlocks.
 36   - Even if softirq is serialized, only local cpu is marked for
 37     execution. Hence, we get something sort of weak cpu binding.
 38     Though it is still not clear, will it result in better locality
 39     or will not.
 40
 41   Examples:
 42   - NET RX softirq. It is multithreaded and does not require
 43     any global serialization.
 44   - NET TX softirq. It kicks software netdevice queues, hence
 45     it is logically serialized per device, but this serialization
 46     is invisible to common code.
 47   - Tasklets: serialized wrt itself.
 48 */
 49
 50#ifndef __ARCH_IRQ_STAT
 51irq_cpustat_t irq_stat[NR_CPUS] ____cacheline_aligned;
 52EXPORT_SYMBOL(irq_stat);
 53#endif
 54
 55static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp;
 56
 57DEFINE_PER_CPU(struct task_struct *, ksoftirqd);
 58
 59char *softirq_to_name[NR_SOFTIRQS] = {
 60	"HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "BLOCK_IOPOLL",
 61	"TASKLET", "SCHED", "HRTIMER", "RCU"
 62};
 63
 64/*
 65 * we cannot loop indefinitely here to avoid userspace starvation,
 66 * but we also don't want to introduce a worst case 1/HZ latency
 67 * to the pending events, so lets the scheduler to balance
 68 * the softirq load for us.
 69 */
 70static void wakeup_softirqd(void)
 71{
 72	/* Interrupts are disabled: no need to stop preemption */
 73	struct task_struct *tsk = __this_cpu_read(ksoftirqd);
 74
 75	if (tsk && tsk->state != TASK_RUNNING)
 76		wake_up_process(tsk);
 77}
 78
 79/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 80 * preempt_count and SOFTIRQ_OFFSET usage:
 81 * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving
 82 *   softirq processing.
 83 * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET)
 84 *   on local_bh_disable or local_bh_enable.
 85 * This lets us distinguish between whether we are currently processing
 86 * softirq and whether we just have bh disabled.
 87 */
 88
 89/*
 90 * This one is for softirq.c-internal use,
 91 * where hardirqs are disabled legitimately:
 92 */
 93#ifdef CONFIG_TRACE_IRQFLAGS
 94static void __local_bh_disable(unsigned long ip, unsigned int cnt)
 95{
 96	unsigned long flags;
 97
 98	WARN_ON_ONCE(in_irq());
 99
100	raw_local_irq_save(flags);
101	/*
102	 * The preempt tracer hooks into add_preempt_count and will break
103	 * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET
104	 * is set and before current->softirq_enabled is cleared.
105	 * We must manually increment preempt_count here and manually
106	 * call the trace_preempt_off later.
107	 */
108	preempt_count() += cnt;
109	/*
110	 * Were softirqs turned off above:
111	 */
112	if (softirq_count() == cnt)
113		trace_softirqs_off(ip);
114	raw_local_irq_restore(flags);
115
116	if (preempt_count() == cnt)
117		trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
118}
119#else /* !CONFIG_TRACE_IRQFLAGS */
120static inline void __local_bh_disable(unsigned long ip, unsigned int cnt)
121{
122	add_preempt_count(cnt);
123	barrier();
124}
 
125#endif /* CONFIG_TRACE_IRQFLAGS */
126
127void local_bh_disable(void)
128{
129	__local_bh_disable((unsigned long)__builtin_return_address(0),
130				SOFTIRQ_DISABLE_OFFSET);
131}
132
133EXPORT_SYMBOL(local_bh_disable);
 
134
135static void __local_bh_enable(unsigned int cnt)
136{
137	WARN_ON_ONCE(in_irq());
138	WARN_ON_ONCE(!irqs_disabled());
139
140	if (softirq_count() == cnt)
141		trace_softirqs_on((unsigned long)__builtin_return_address(0));
142	sub_preempt_count(cnt);
143}
144
145/*
146 * Special-case - softirqs can safely be enabled in
147 * cond_resched_softirq(), or by __do_softirq(),
148 * without processing still-pending softirqs:
149 */
150void _local_bh_enable(void)
151{
 
152	__local_bh_enable(SOFTIRQ_DISABLE_OFFSET);
153}
154
155EXPORT_SYMBOL(_local_bh_enable);
156
157static inline void _local_bh_enable_ip(unsigned long ip)
158{
159	WARN_ON_ONCE(in_irq() || irqs_disabled());
 
160#ifdef CONFIG_TRACE_IRQFLAGS
161	local_irq_disable();
162#endif
163	/*
164	 * Are softirqs going to be turned on now:
165	 */
166	if (softirq_count() == SOFTIRQ_DISABLE_OFFSET)
167		trace_softirqs_on(ip);
168	/*
169	 * Keep preemption disabled until we are done with
170	 * softirq processing:
171 	 */
172	sub_preempt_count(SOFTIRQ_DISABLE_OFFSET - 1);
173
174	if (unlikely(!in_interrupt() && local_softirq_pending()))
 
 
 
 
175		do_softirq();
 
176
177	dec_preempt_count();
178#ifdef CONFIG_TRACE_IRQFLAGS
179	local_irq_enable();
180#endif
181	preempt_check_resched();
182}
183
184void local_bh_enable(void)
185{
186	_local_bh_enable_ip((unsigned long)__builtin_return_address(0));
187}
188EXPORT_SYMBOL(local_bh_enable);
189
190void local_bh_enable_ip(unsigned long ip)
191{
192	_local_bh_enable_ip(ip);
193}
194EXPORT_SYMBOL(local_bh_enable_ip);
195
196/*
197 * We restart softirq processing MAX_SOFTIRQ_RESTART times,
198 * and we fall back to softirqd after that.
 
 
 
 
199 *
200 * This number has been established via experimentation.
201 * The two things to balance is latency against fairness -
202 * we want to handle softirqs as soon as possible, but they
203 * should not be able to lock up the box.
204 */
 
205#define MAX_SOFTIRQ_RESTART 10
206
207asmlinkage void __do_softirq(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
208{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
209	struct softirq_action *h;
 
210	__u32 pending;
211	int max_restart = MAX_SOFTIRQ_RESTART;
212	int cpu;
 
 
 
 
 
 
213
214	pending = local_softirq_pending();
215	account_system_vtime(current);
216
217	__local_bh_disable((unsigned long)__builtin_return_address(0),
218				SOFTIRQ_OFFSET);
219	lockdep_softirq_enter();
220
221	cpu = smp_processor_id();
222restart:
223	/* Reset the pending bitmask before enabling irqs */
224	set_softirq_pending(0);
225
226	local_irq_enable();
227
228	h = softirq_vec;
229
230	do {
231		if (pending & 1) {
232			unsigned int vec_nr = h - softirq_vec;
233			int prev_count = preempt_count();
234
235			kstat_incr_softirqs_this_cpu(vec_nr);
236
237			trace_softirq_entry(vec_nr);
238			h->action(h);
239			trace_softirq_exit(vec_nr);
240			if (unlikely(prev_count != preempt_count())) {
241				printk(KERN_ERR "huh, entered softirq %u %s %p"
242				       "with preempt_count %08x,"
243				       " exited with %08x?\n", vec_nr,
244				       softirq_to_name[vec_nr], h->action,
245				       prev_count, preempt_count());
246				preempt_count() = prev_count;
247			}
248
249			rcu_bh_qs(cpu);
250		}
251		h++;
252		pending >>= 1;
253	} while (pending);
254
 
 
255	local_irq_disable();
256
257	pending = local_softirq_pending();
258	if (pending && --max_restart)
259		goto restart;
 
 
260
261	if (pending)
262		wakeup_softirqd();
 
263
264	lockdep_softirq_exit();
265
266	account_system_vtime(current);
267	__local_bh_enable(SOFTIRQ_OFFSET);
 
 
268}
269
270#ifndef __ARCH_HAS_DO_SOFTIRQ
271
272asmlinkage void do_softirq(void)
273{
274	__u32 pending;
275	unsigned long flags;
276
277	if (in_interrupt())
278		return;
279
280	local_irq_save(flags);
281
282	pending = local_softirq_pending();
283
284	if (pending)
285		__do_softirq();
286
287	local_irq_restore(flags);
288}
289
290#endif
291
292/*
293 * Enter an interrupt context.
294 */
295void irq_enter(void)
296{
297	int cpu = smp_processor_id();
298
299	rcu_irq_enter();
300	if (is_idle_task(current) && !in_interrupt()) {
301		/*
302		 * Prevent raise_softirq from needlessly waking up ksoftirqd
303		 * here, as softirq will be serviced on return from interrupt.
304		 */
305		local_bh_disable();
306		tick_check_idle(cpu);
307		_local_bh_enable();
308	}
309
310	__irq_enter();
311}
312
313static inline void invoke_softirq(void)
314{
 
 
 
315	if (!force_irqthreads) {
316#ifdef __ARCH_IRQ_EXIT_IRQS_DISABLED
 
 
 
 
 
317		__do_softirq();
318#else
319		do_softirq();
 
 
 
 
 
320#endif
321	} else {
322		__local_bh_disable((unsigned long)__builtin_return_address(0),
323				SOFTIRQ_OFFSET);
324		wakeup_softirqd();
325		__local_bh_enable(SOFTIRQ_OFFSET);
326	}
327}
328
 
 
 
 
 
 
 
 
 
 
 
 
 
329/*
330 * Exit an interrupt context. Process softirqs if needed and possible:
331 */
332void irq_exit(void)
333{
334	account_system_vtime(current);
335	trace_hardirq_exit();
336	sub_preempt_count(IRQ_EXIT_OFFSET);
 
 
 
 
337	if (!in_interrupt() && local_softirq_pending())
338		invoke_softirq();
339
340#ifdef CONFIG_NO_HZ
341	/* Make sure that timer wheel updates are propagated */
342	if (idle_cpu(smp_processor_id()) && !in_interrupt() && !need_resched())
343		tick_nohz_irq_exit();
344#endif
345	rcu_irq_exit();
346	sched_preempt_enable_no_resched();
347}
348
349/*
350 * This function must run with irqs disabled!
351 */
352inline void raise_softirq_irqoff(unsigned int nr)
353{
354	__raise_softirq_irqoff(nr);
355
356	/*
357	 * If we're in an interrupt or softirq, we're done
358	 * (this also catches softirq-disabled code). We will
359	 * actually run the softirq once we return from
360	 * the irq or softirq.
361	 *
362	 * Otherwise we wake up ksoftirqd to make sure we
363	 * schedule the softirq soon.
364	 */
365	if (!in_interrupt())
366		wakeup_softirqd();
367}
368
369void raise_softirq(unsigned int nr)
370{
371	unsigned long flags;
372
373	local_irq_save(flags);
374	raise_softirq_irqoff(nr);
375	local_irq_restore(flags);
376}
377
378void __raise_softirq_irqoff(unsigned int nr)
379{
380	trace_softirq_raise(nr);
381	or_softirq_pending(1UL << nr);
382}
383
384void open_softirq(int nr, void (*action)(struct softirq_action *))
385{
386	softirq_vec[nr].action = action;
387}
388
389/*
390 * Tasklets
391 */
392struct tasklet_head
393{
394	struct tasklet_struct *head;
395	struct tasklet_struct **tail;
396};
397
398static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec);
399static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec);
400
401void __tasklet_schedule(struct tasklet_struct *t)
 
 
402{
 
403	unsigned long flags;
404
405	local_irq_save(flags);
 
406	t->next = NULL;
407	*__this_cpu_read(tasklet_vec.tail) = t;
408	__this_cpu_write(tasklet_vec.tail, &(t->next));
409	raise_softirq_irqoff(TASKLET_SOFTIRQ);
410	local_irq_restore(flags);
411}
412
 
 
 
 
 
413EXPORT_SYMBOL(__tasklet_schedule);
414
415void __tasklet_hi_schedule(struct tasklet_struct *t)
416{
417	unsigned long flags;
418
419	local_irq_save(flags);
420	t->next = NULL;
421	*__this_cpu_read(tasklet_hi_vec.tail) = t;
422	__this_cpu_write(tasklet_hi_vec.tail,  &(t->next));
423	raise_softirq_irqoff(HI_SOFTIRQ);
424	local_irq_restore(flags);
425}
426
427EXPORT_SYMBOL(__tasklet_hi_schedule);
428
429void __tasklet_hi_schedule_first(struct tasklet_struct *t)
430{
431	BUG_ON(!irqs_disabled());
432
433	t->next = __this_cpu_read(tasklet_hi_vec.head);
434	__this_cpu_write(tasklet_hi_vec.head, t);
435	__raise_softirq_irqoff(HI_SOFTIRQ);
436}
437
438EXPORT_SYMBOL(__tasklet_hi_schedule_first);
439
440static void tasklet_action(struct softirq_action *a)
441{
442	struct tasklet_struct *list;
443
444	local_irq_disable();
445	list = __this_cpu_read(tasklet_vec.head);
446	__this_cpu_write(tasklet_vec.head, NULL);
447	__this_cpu_write(tasklet_vec.tail, &__get_cpu_var(tasklet_vec).head);
448	local_irq_enable();
449
450	while (list) {
451		struct tasklet_struct *t = list;
452
453		list = list->next;
454
455		if (tasklet_trylock(t)) {
456			if (!atomic_read(&t->count)) {
457				if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state))
 
458					BUG();
459				t->func(t->data);
460				tasklet_unlock(t);
461				continue;
462			}
463			tasklet_unlock(t);
464		}
465
466		local_irq_disable();
467		t->next = NULL;
468		*__this_cpu_read(tasklet_vec.tail) = t;
469		__this_cpu_write(tasklet_vec.tail, &(t->next));
470		__raise_softirq_irqoff(TASKLET_SOFTIRQ);
471		local_irq_enable();
472	}
473}
474
475static void tasklet_hi_action(struct softirq_action *a)
476{
477	struct tasklet_struct *list;
 
478
479	local_irq_disable();
480	list = __this_cpu_read(tasklet_hi_vec.head);
481	__this_cpu_write(tasklet_hi_vec.head, NULL);
482	__this_cpu_write(tasklet_hi_vec.tail, &__get_cpu_var(tasklet_hi_vec).head);
483	local_irq_enable();
484
485	while (list) {
486		struct tasklet_struct *t = list;
487
488		list = list->next;
489
490		if (tasklet_trylock(t)) {
491			if (!atomic_read(&t->count)) {
492				if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state))
493					BUG();
494				t->func(t->data);
495				tasklet_unlock(t);
496				continue;
497			}
498			tasklet_unlock(t);
499		}
500
501		local_irq_disable();
502		t->next = NULL;
503		*__this_cpu_read(tasklet_hi_vec.tail) = t;
504		__this_cpu_write(tasklet_hi_vec.tail, &(t->next));
505		__raise_softirq_irqoff(HI_SOFTIRQ);
506		local_irq_enable();
507	}
508}
509
510
511void tasklet_init(struct tasklet_struct *t,
512		  void (*func)(unsigned long), unsigned long data)
513{
514	t->next = NULL;
515	t->state = 0;
516	atomic_set(&t->count, 0);
517	t->func = func;
518	t->data = data;
519}
520
521EXPORT_SYMBOL(tasklet_init);
522
523void tasklet_kill(struct tasklet_struct *t)
524{
525	if (in_interrupt())
526		printk("Attempt to kill tasklet from interrupt\n");
527
528	while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
529		do {
530			yield();
531		} while (test_bit(TASKLET_STATE_SCHED, &t->state));
532	}
533	tasklet_unlock_wait(t);
534	clear_bit(TASKLET_STATE_SCHED, &t->state);
535}
536
537EXPORT_SYMBOL(tasklet_kill);
538
539/*
540 * tasklet_hrtimer
541 */
542
543/*
544 * The trampoline is called when the hrtimer expires. It schedules a tasklet
545 * to run __tasklet_hrtimer_trampoline() which in turn will call the intended
546 * hrtimer callback, but from softirq context.
547 */
548static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer)
549{
550	struct tasklet_hrtimer *ttimer =
551		container_of(timer, struct tasklet_hrtimer, timer);
552
553	tasklet_hi_schedule(&ttimer->tasklet);
554	return HRTIMER_NORESTART;
555}
556
557/*
558 * Helper function which calls the hrtimer callback from
559 * tasklet/softirq context
560 */
561static void __tasklet_hrtimer_trampoline(unsigned long data)
562{
563	struct tasklet_hrtimer *ttimer = (void *)data;
564	enum hrtimer_restart restart;
565
566	restart = ttimer->function(&ttimer->timer);
567	if (restart != HRTIMER_NORESTART)
568		hrtimer_restart(&ttimer->timer);
569}
570
571/**
572 * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks
573 * @ttimer:	 tasklet_hrtimer which is initialized
574 * @function:	 hrtimer callback function which gets called from softirq context
575 * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME)
576 * @mode:	 hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL)
577 */
578void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer,
579			  enum hrtimer_restart (*function)(struct hrtimer *),
580			  clockid_t which_clock, enum hrtimer_mode mode)
581{
582	hrtimer_init(&ttimer->timer, which_clock, mode);
583	ttimer->timer.function = __hrtimer_tasklet_trampoline;
584	tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline,
585		     (unsigned long)ttimer);
586	ttimer->function = function;
587}
588EXPORT_SYMBOL_GPL(tasklet_hrtimer_init);
589
590/*
591 * Remote softirq bits
592 */
593
594DEFINE_PER_CPU(struct list_head [NR_SOFTIRQS], softirq_work_list);
595EXPORT_PER_CPU_SYMBOL(softirq_work_list);
596
597static void __local_trigger(struct call_single_data *cp, int softirq)
598{
599	struct list_head *head = &__get_cpu_var(softirq_work_list[softirq]);
600
601	list_add_tail(&cp->list, head);
602
603	/* Trigger the softirq only if the list was previously empty.  */
604	if (head->next == &cp->list)
605		raise_softirq_irqoff(softirq);
606}
607
608#ifdef CONFIG_USE_GENERIC_SMP_HELPERS
609static void remote_softirq_receive(void *data)
610{
611	struct call_single_data *cp = data;
612	unsigned long flags;
613	int softirq;
614
615	softirq = cp->priv;
616
617	local_irq_save(flags);
618	__local_trigger(cp, softirq);
619	local_irq_restore(flags);
620}
621
622static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq)
623{
624	if (cpu_online(cpu)) {
625		cp->func = remote_softirq_receive;
626		cp->info = cp;
627		cp->flags = 0;
628		cp->priv = softirq;
629
630		__smp_call_function_single(cpu, cp, 0);
631		return 0;
632	}
633	return 1;
634}
635#else /* CONFIG_USE_GENERIC_SMP_HELPERS */
636static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq)
637{
638	return 1;
639}
640#endif
641
642/**
643 * __send_remote_softirq - try to schedule softirq work on a remote cpu
644 * @cp: private SMP call function data area
645 * @cpu: the remote cpu
646 * @this_cpu: the currently executing cpu
647 * @softirq: the softirq for the work
648 *
649 * Attempt to schedule softirq work on a remote cpu.  If this cannot be
650 * done, the work is instead queued up on the local cpu.
651 *
652 * Interrupts must be disabled.
653 */
654void __send_remote_softirq(struct call_single_data *cp, int cpu, int this_cpu, int softirq)
655{
656	if (cpu == this_cpu || __try_remote_softirq(cp, cpu, softirq))
657		__local_trigger(cp, softirq);
658}
659EXPORT_SYMBOL(__send_remote_softirq);
660
661/**
662 * send_remote_softirq - try to schedule softirq work on a remote cpu
663 * @cp: private SMP call function data area
664 * @cpu: the remote cpu
665 * @softirq: the softirq for the work
666 *
667 * Like __send_remote_softirq except that disabling interrupts and
668 * computing the current cpu is done for the caller.
669 */
670void send_remote_softirq(struct call_single_data *cp, int cpu, int softirq)
671{
672	unsigned long flags;
673	int this_cpu;
674
675	local_irq_save(flags);
676	this_cpu = smp_processor_id();
677	__send_remote_softirq(cp, cpu, this_cpu, softirq);
678	local_irq_restore(flags);
679}
680EXPORT_SYMBOL(send_remote_softirq);
681
682static int __cpuinit remote_softirq_cpu_notify(struct notifier_block *self,
683					       unsigned long action, void *hcpu)
684{
685	/*
686	 * If a CPU goes away, splice its entries to the current CPU
687	 * and trigger a run of the softirq
688	 */
689	if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
690		int cpu = (unsigned long) hcpu;
691		int i;
692
693		local_irq_disable();
694		for (i = 0; i < NR_SOFTIRQS; i++) {
695			struct list_head *head = &per_cpu(softirq_work_list[i], cpu);
696			struct list_head *local_head;
697
698			if (list_empty(head))
699				continue;
700
701			local_head = &__get_cpu_var(softirq_work_list[i]);
702			list_splice_init(head, local_head);
703			raise_softirq_irqoff(i);
704		}
705		local_irq_enable();
706	}
707
708	return NOTIFY_OK;
709}
710
711static struct notifier_block __cpuinitdata remote_softirq_cpu_notifier = {
712	.notifier_call	= remote_softirq_cpu_notify,
713};
714
715void __init softirq_init(void)
716{
717	int cpu;
718
719	for_each_possible_cpu(cpu) {
720		int i;
721
722		per_cpu(tasklet_vec, cpu).tail =
723			&per_cpu(tasklet_vec, cpu).head;
724		per_cpu(tasklet_hi_vec, cpu).tail =
725			&per_cpu(tasklet_hi_vec, cpu).head;
726		for (i = 0; i < NR_SOFTIRQS; i++)
727			INIT_LIST_HEAD(&per_cpu(softirq_work_list[i], cpu));
728	}
729
730	register_hotcpu_notifier(&remote_softirq_cpu_notifier);
731
732	open_softirq(TASKLET_SOFTIRQ, tasklet_action);
733	open_softirq(HI_SOFTIRQ, tasklet_hi_action);
734}
735
736static int run_ksoftirqd(void * __bind_cpu)
737{
738	set_current_state(TASK_INTERRUPTIBLE);
 
739
740	while (!kthread_should_stop()) {
741		preempt_disable();
742		if (!local_softirq_pending()) {
743			schedule_preempt_disabled();
744		}
745
746		__set_current_state(TASK_RUNNING);
747
748		while (local_softirq_pending()) {
749			/* Preempt disable stops cpu going offline.
750			   If already offline, we'll be on wrong CPU:
751			   don't process */
752			if (cpu_is_offline((long)__bind_cpu))
753				goto wait_to_die;
754			local_irq_disable();
755			if (local_softirq_pending())
756				__do_softirq();
757			local_irq_enable();
758			sched_preempt_enable_no_resched();
759			cond_resched();
760			preempt_disable();
761			rcu_note_context_switch((long)__bind_cpu);
762		}
763		preempt_enable();
764		set_current_state(TASK_INTERRUPTIBLE);
765	}
766	__set_current_state(TASK_RUNNING);
767	return 0;
768
769wait_to_die:
770	preempt_enable();
771	/* Wait for kthread_stop */
772	set_current_state(TASK_INTERRUPTIBLE);
773	while (!kthread_should_stop()) {
774		schedule();
775		set_current_state(TASK_INTERRUPTIBLE);
776	}
777	__set_current_state(TASK_RUNNING);
778	return 0;
779}
780
781#ifdef CONFIG_HOTPLUG_CPU
782/*
783 * tasklet_kill_immediate is called to remove a tasklet which can already be
784 * scheduled for execution on @cpu.
785 *
786 * Unlike tasklet_kill, this function removes the tasklet
787 * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state.
788 *
789 * When this function is called, @cpu must be in the CPU_DEAD state.
790 */
791void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu)
792{
793	struct tasklet_struct **i;
794
795	BUG_ON(cpu_online(cpu));
796	BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state));
797
798	if (!test_bit(TASKLET_STATE_SCHED, &t->state))
799		return;
800
801	/* CPU is dead, so no lock needed. */
802	for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) {
803		if (*i == t) {
804			*i = t->next;
805			/* If this was the tail element, move the tail ptr */
806			if (*i == NULL)
807				per_cpu(tasklet_vec, cpu).tail = i;
808			return;
809		}
810	}
811	BUG();
812}
813
814static void takeover_tasklets(unsigned int cpu)
815{
816	/* CPU is dead, so no lock needed. */
817	local_irq_disable();
818
819	/* Find end, append list for that CPU. */
820	if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) {
821		*__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head;
822		this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail);
823		per_cpu(tasklet_vec, cpu).head = NULL;
824		per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head;
825	}
826	raise_softirq_irqoff(TASKLET_SOFTIRQ);
827
828	if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) {
829		*__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head;
830		__this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail);
831		per_cpu(tasklet_hi_vec, cpu).head = NULL;
832		per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head;
833	}
834	raise_softirq_irqoff(HI_SOFTIRQ);
835
836	local_irq_enable();
 
837}
 
 
838#endif /* CONFIG_HOTPLUG_CPU */
839
840static int __cpuinit cpu_callback(struct notifier_block *nfb,
841				  unsigned long action,
842				  void *hcpu)
843{
844	int hotcpu = (unsigned long)hcpu;
845	struct task_struct *p;
846
847	switch (action) {
848	case CPU_UP_PREPARE:
849	case CPU_UP_PREPARE_FROZEN:
850		p = kthread_create_on_node(run_ksoftirqd,
851					   hcpu,
852					   cpu_to_node(hotcpu),
853					   "ksoftirqd/%d", hotcpu);
854		if (IS_ERR(p)) {
855			printk("ksoftirqd for %i failed\n", hotcpu);
856			return notifier_from_errno(PTR_ERR(p));
857		}
858		kthread_bind(p, hotcpu);
859  		per_cpu(ksoftirqd, hotcpu) = p;
860 		break;
861	case CPU_ONLINE:
862	case CPU_ONLINE_FROZEN:
863		wake_up_process(per_cpu(ksoftirqd, hotcpu));
864		break;
865#ifdef CONFIG_HOTPLUG_CPU
866	case CPU_UP_CANCELED:
867	case CPU_UP_CANCELED_FROZEN:
868		if (!per_cpu(ksoftirqd, hotcpu))
869			break;
870		/* Unbind so it can run.  Fall thru. */
871		kthread_bind(per_cpu(ksoftirqd, hotcpu),
872			     cpumask_any(cpu_online_mask));
873	case CPU_DEAD:
874	case CPU_DEAD_FROZEN: {
875		static const struct sched_param param = {
876			.sched_priority = MAX_RT_PRIO-1
877		};
878
879		p = per_cpu(ksoftirqd, hotcpu);
880		per_cpu(ksoftirqd, hotcpu) = NULL;
881		sched_setscheduler_nocheck(p, SCHED_FIFO, &param);
882		kthread_stop(p);
883		takeover_tasklets(hotcpu);
884		break;
885	}
886#endif /* CONFIG_HOTPLUG_CPU */
887 	}
888	return NOTIFY_OK;
889}
890
891static struct notifier_block __cpuinitdata cpu_nfb = {
892	.notifier_call = cpu_callback
893};
894
895static __init int spawn_ksoftirqd(void)
896{
897	void *cpu = (void *)(long)smp_processor_id();
898	int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
 
899
900	BUG_ON(err != NOTIFY_OK);
901	cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
902	register_cpu_notifier(&cpu_nfb);
903	return 0;
904}
905early_initcall(spawn_ksoftirqd);
906
907/*
908 * [ These __weak aliases are kept in a separate compilation unit, so that
909 *   GCC does not inline them incorrectly. ]
910 */
911
912int __init __weak early_irq_init(void)
913{
914	return 0;
915}
916
917#ifdef CONFIG_GENERIC_HARDIRQS
918int __init __weak arch_probe_nr_irqs(void)
919{
920	return NR_IRQS_LEGACY;
921}
922
923int __init __weak arch_early_irq_init(void)
924{
925	return 0;
926}
927#endif