blob: dce38fac4f323f415cff156907b8ea44da1b9698 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/kernel/softirq.c
3 *
4 * Copyright (C) 1992 Linus Torvalds
5 *
Pavel Machekb10db7f2008-01-30 13:30:00 +01006 * Distribute under GPLv2.
7 *
8 * Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903)
David S. Miller54514a72008-09-23 22:15:57 -07009 *
10 * Remote softirq infrastructure is by Jens Axboe.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 */
12
Paul Gortmaker9984de12011-05-23 14:51:41 -040013#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#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>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070021#include <linux/freezer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/kthread.h>
23#include <linux/rcupdate.h>
Steven Rostedt7e49fcc2009-01-22 19:01:40 -050024#include <linux/ftrace.h>
Andrew Morton78eef012006-03-22 00:08:16 -080025#include <linux/smp.h>
Thomas Gleixner3e339b52012-07-16 10:42:37 +000026#include <linux/smpboot.h>
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -080027#include <linux/tick.h>
Heiko Carstensa0e39ed2009-04-29 13:51:39 +020028
29#define CREATE_TRACE_POINTS
Steven Rostedtad8d75f2009-04-14 19:39:12 -040030#include <trace/events/irq.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32#include <asm/irq.h>
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
Alexey Dobriyan978b0112008-09-06 20:04:36 +020056static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Venkatesh Pallipadi4dd53d82010-12-21 17:09:00 -080058DEFINE_PER_CPU(struct task_struct *, ksoftirqd);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Jason Baron5d592b42009-03-12 14:33:36 -040060char *softirq_to_name[NR_SOFTIRQS] = {
Li Zefan5dd4de52009-09-17 17:38:32 +080061 "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "BLOCK_IOPOLL",
Shaohua Li09223372011-06-14 13:26:25 +080062 "TASKLET", "SCHED", "HRTIMER", "RCU"
Jason Baron5d592b42009-03-12 14:33:36 -040063};
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/*
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 */
Thomas Gleixner676cb022009-07-20 23:33:49 +020071static void wakeup_softirqd(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072{
73 /* Interrupts are disabled: no need to stop preemption */
Christoph Lameter909ea962010-12-08 16:22:55 +010074 struct task_struct *tsk = __this_cpu_read(ksoftirqd);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76 if (tsk && tsk->state != TASK_RUNNING)
77 wake_up_process(tsk);
78}
79
80/*
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -070081 * preempt_count and SOFTIRQ_OFFSET usage:
82 * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving
83 * softirq processing.
84 * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET)
85 * on local_bh_disable or local_bh_enable.
86 * This lets us distinguish between whether we are currently processing
87 * softirq and whether we just have bh disabled.
88 */
89
90/*
Ingo Molnarde30a2b2006-07-03 00:24:42 -070091 * This one is for softirq.c-internal use,
92 * where hardirqs are disabled legitimately:
93 */
Tim Chen3c829c32006-07-30 03:04:02 -070094#ifdef CONFIG_TRACE_IRQFLAGS
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -070095static void __local_bh_disable(unsigned long ip, unsigned int cnt)
Ingo Molnarde30a2b2006-07-03 00:24:42 -070096{
97 unsigned long flags;
98
99 WARN_ON_ONCE(in_irq());
100
101 raw_local_irq_save(flags);
Steven Rostedt7e49fcc2009-01-22 19:01:40 -0500102 /*
103 * The preempt tracer hooks into add_preempt_count and will break
104 * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET
105 * is set and before current->softirq_enabled is cleared.
106 * We must manually increment preempt_count here and manually
107 * call the trace_preempt_off later.
108 */
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700109 preempt_count() += cnt;
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700110 /*
111 * Were softirqs turned off above:
112 */
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700113 if (softirq_count() == cnt)
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700114 trace_softirqs_off(ip);
115 raw_local_irq_restore(flags);
Steven Rostedt7e49fcc2009-01-22 19:01:40 -0500116
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700117 if (preempt_count() == cnt)
Steven Rostedt7e49fcc2009-01-22 19:01:40 -0500118 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700119}
Tim Chen3c829c32006-07-30 03:04:02 -0700120#else /* !CONFIG_TRACE_IRQFLAGS */
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700121static inline void __local_bh_disable(unsigned long ip, unsigned int cnt)
Tim Chen3c829c32006-07-30 03:04:02 -0700122{
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700123 add_preempt_count(cnt);
Tim Chen3c829c32006-07-30 03:04:02 -0700124 barrier();
125}
126#endif /* CONFIG_TRACE_IRQFLAGS */
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700127
128void local_bh_disable(void)
129{
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700130 __local_bh_disable((unsigned long)__builtin_return_address(0),
131 SOFTIRQ_DISABLE_OFFSET);
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700132}
133
134EXPORT_SYMBOL(local_bh_disable);
135
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700136static void __local_bh_enable(unsigned int cnt)
137{
138 WARN_ON_ONCE(in_irq());
139 WARN_ON_ONCE(!irqs_disabled());
140
141 if (softirq_count() == cnt)
142 trace_softirqs_on((unsigned long)__builtin_return_address(0));
143 sub_preempt_count(cnt);
144}
145
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700146/*
147 * Special-case - softirqs can safely be enabled in
148 * cond_resched_softirq(), or by __do_softirq(),
149 * without processing still-pending softirqs:
150 */
151void _local_bh_enable(void)
152{
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700153 __local_bh_enable(SOFTIRQ_DISABLE_OFFSET);
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700154}
155
156EXPORT_SYMBOL(_local_bh_enable);
157
Johannes Berg0f476b6d2008-06-18 09:29:37 +0200158static inline void _local_bh_enable_ip(unsigned long ip)
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700159{
Johannes Berg0f476b6d2008-06-18 09:29:37 +0200160 WARN_ON_ONCE(in_irq() || irqs_disabled());
Tim Chen3c829c32006-07-30 03:04:02 -0700161#ifdef CONFIG_TRACE_IRQFLAGS
Johannes Berg0f476b6d2008-06-18 09:29:37 +0200162 local_irq_disable();
Tim Chen3c829c32006-07-30 03:04:02 -0700163#endif
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700164 /*
165 * Are softirqs going to be turned on now:
166 */
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700167 if (softirq_count() == SOFTIRQ_DISABLE_OFFSET)
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700168 trace_softirqs_on(ip);
169 /*
170 * Keep preemption disabled until we are done with
171 * softirq processing:
172 */
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700173 sub_preempt_count(SOFTIRQ_DISABLE_OFFSET - 1);
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700174
175 if (unlikely(!in_interrupt() && local_softirq_pending()))
176 do_softirq();
177
178 dec_preempt_count();
Tim Chen3c829c32006-07-30 03:04:02 -0700179#ifdef CONFIG_TRACE_IRQFLAGS
Johannes Berg0f476b6d2008-06-18 09:29:37 +0200180 local_irq_enable();
Tim Chen3c829c32006-07-30 03:04:02 -0700181#endif
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700182 preempt_check_resched();
183}
Johannes Berg0f476b6d2008-06-18 09:29:37 +0200184
185void local_bh_enable(void)
186{
187 _local_bh_enable_ip((unsigned long)__builtin_return_address(0));
188}
189EXPORT_SYMBOL(local_bh_enable);
190
191void local_bh_enable_ip(unsigned long ip)
192{
193 _local_bh_enable_ip(ip);
194}
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700195EXPORT_SYMBOL(local_bh_enable_ip);
196
197/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 * We restart softirq processing MAX_SOFTIRQ_RESTART times,
199 * and we fall back to softirqd after that.
200 *
201 * This number has 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_RESTART 10
207
208asmlinkage void __do_softirq(void)
209{
210 struct softirq_action *h;
211 __u32 pending;
212 int max_restart = MAX_SOFTIRQ_RESTART;
213 int cpu;
Mel Gorman907aed42012-07-31 16:44:07 -0700214 unsigned long old_flags = current->flags;
215
216 /*
217 * Mask out PF_MEMALLOC s current task context is borrowed for the
218 * softirq. A softirq handled such as network RX might set PF_MEMALLOC
219 * again if the socket is related to swap
220 */
221 current->flags &= ~PF_MEMALLOC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
223 pending = local_softirq_pending();
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100224 account_irq_enter_time(current);
Paul Mackerras829035fd2006-07-03 00:25:40 -0700225
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700226 __local_bh_disable((unsigned long)__builtin_return_address(0),
227 SOFTIRQ_OFFSET);
Ingo Molnard820ac42009-03-13 01:30:40 +0100228 lockdep_softirq_enter();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 cpu = smp_processor_id();
231restart:
232 /* Reset the pending bitmask before enabling irqs */
Andi Kleen3f744782005-09-12 18:49:24 +0200233 set_softirq_pending(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
Andrew Mortonc70f5d62005-07-30 10:22:49 -0700235 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
237 h = softirq_vec;
238
239 do {
240 if (pending & 1) {
Thomas Gleixnerf4bc6bb2010-10-19 15:00:13 +0200241 unsigned int vec_nr = h - softirq_vec;
Thomas Gleixner8e85b4b2008-10-02 10:50:53 +0200242 int prev_count = preempt_count();
243
Thomas Gleixnerf4bc6bb2010-10-19 15:00:13 +0200244 kstat_incr_softirqs_this_cpu(vec_nr);
245
246 trace_softirq_entry(vec_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 h->action(h);
Thomas Gleixnerf4bc6bb2010-10-19 15:00:13 +0200248 trace_softirq_exit(vec_nr);
Thomas Gleixner8e85b4b2008-10-02 10:50:53 +0200249 if (unlikely(prev_count != preempt_count())) {
Thomas Gleixnerf4bc6bb2010-10-19 15:00:13 +0200250 printk(KERN_ERR "huh, entered softirq %u %s %p"
Thomas Gleixner8e85b4b2008-10-02 10:50:53 +0200251 "with preempt_count %08x,"
Thomas Gleixnerf4bc6bb2010-10-19 15:00:13 +0200252 " exited with %08x?\n", vec_nr,
253 softirq_to_name[vec_nr], h->action,
254 prev_count, preempt_count());
Thomas Gleixner8e85b4b2008-10-02 10:50:53 +0200255 preempt_count() = prev_count;
256 }
257
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700258 rcu_bh_qs(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 }
260 h++;
261 pending >>= 1;
262 } while (pending);
263
Andrew Mortonc70f5d62005-07-30 10:22:49 -0700264 local_irq_disable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
266 pending = local_softirq_pending();
267 if (pending && --max_restart)
268 goto restart;
269
270 if (pending)
271 wakeup_softirqd();
272
Ingo Molnard820ac42009-03-13 01:30:40 +0100273 lockdep_softirq_exit();
Paul Mackerras829035fd2006-07-03 00:25:40 -0700274
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100275 account_irq_exit_time(current);
Venkatesh Pallipadi75e10562010-10-04 17:03:16 -0700276 __local_bh_enable(SOFTIRQ_OFFSET);
Mel Gorman907aed42012-07-31 16:44:07 -0700277 tsk_restore_flags(current, old_flags, PF_MEMALLOC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278}
279
280#ifndef __ARCH_HAS_DO_SOFTIRQ
281
282asmlinkage void do_softirq(void)
283{
284 __u32 pending;
285 unsigned long flags;
286
287 if (in_interrupt())
288 return;
289
290 local_irq_save(flags);
291
292 pending = local_softirq_pending();
293
294 if (pending)
295 __do_softirq();
296
297 local_irq_restore(flags);
298}
299
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300#endif
301
Ingo Molnardde4b2b2007-02-16 01:27:45 -0800302/*
303 * Enter an interrupt context.
304 */
305void irq_enter(void)
306{
Venki Pallipadi6378ddb2008-01-30 13:30:04 +0100307 int cpu = smp_processor_id();
Thomas Gleixner719254f2008-10-17 09:59:47 +0200308
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100309 rcu_irq_enter();
Frederic Weisbecker0a8a2e72012-01-24 18:59:44 +0100310 if (is_idle_task(current) && !in_interrupt()) {
Venkatesh Pallipadid267f872010-10-04 17:03:23 -0700311 /*
312 * Prevent raise_softirq from needlessly waking up ksoftirqd
313 * here, as softirq will be serviced on return from interrupt.
314 */
315 local_bh_disable();
Thomas Gleixner719254f2008-10-17 09:59:47 +0200316 tick_check_idle(cpu);
Venkatesh Pallipadid267f872010-10-04 17:03:23 -0700317 _local_bh_enable();
318 }
319
320 __irq_enter();
Ingo Molnardde4b2b2007-02-16 01:27:45 -0800321}
322
Heiko Carstensb2a00172012-03-05 15:07:25 -0800323static inline void invoke_softirq(void)
324{
Thomas Gleixnerfacd8b82013-02-21 18:17:42 +0100325 if (!force_irqthreads)
Thomas Gleixner8d32a302011-02-23 23:52:23 +0000326 __do_softirq();
Thomas Gleixnerfacd8b82013-02-21 18:17:42 +0100327 else
Thomas Gleixner8d32a302011-02-23 23:52:23 +0000328 wakeup_softirqd();
329}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
331/*
332 * Exit an interrupt context. Process softirqs if needed and possible:
333 */
334void irq_exit(void)
335{
Thomas Gleixner74eed012013-02-20 22:00:48 +0100336#ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED
Frederic Weisbecker4cd5d112013-02-28 20:00:43 +0100337 local_irq_disable();
Thomas Gleixner74eed012013-02-20 22:00:48 +0100338#else
339 WARN_ON_ONCE(!irqs_disabled());
340#endif
341
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100342 account_irq_exit_time(current);
Ingo Molnarde30a2b2006-07-03 00:24:42 -0700343 trace_hardirq_exit();
Frederic Weisbecker4d4c4e22013-02-22 00:05:07 +0100344 sub_preempt_count(HARDIRQ_OFFSET);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 if (!in_interrupt() && local_softirq_pending())
346 invoke_softirq();
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -0800347
348#ifdef CONFIG_NO_HZ
349 /* Make sure that timer wheel updates are propagated */
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100350 if (idle_cpu(smp_processor_id()) && !in_interrupt() && !need_resched())
Frederic Weisbecker280f0672011-10-07 18:22:06 +0200351 tick_nohz_irq_exit();
Thomas Gleixner79bf2bb2007-02-16 01:28:03 -0800352#endif
Frederic Weisbecker416eb332011-10-07 16:31:02 -0700353 rcu_irq_exit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354}
355
356/*
357 * This function must run with irqs disabled!
358 */
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800359inline void raise_softirq_irqoff(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360{
361 __raise_softirq_irqoff(nr);
362
363 /*
364 * If we're in an interrupt or softirq, we're done
365 * (this also catches softirq-disabled code). We will
366 * actually run the softirq once we return from
367 * the irq or softirq.
368 *
369 * Otherwise we wake up ksoftirqd to make sure we
370 * schedule the softirq soon.
371 */
372 if (!in_interrupt())
373 wakeup_softirqd();
374}
375
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800376void raise_softirq(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377{
378 unsigned long flags;
379
380 local_irq_save(flags);
381 raise_softirq_irqoff(nr);
382 local_irq_restore(flags);
383}
384
Steven Rostedtf0696862012-01-25 20:18:55 -0500385void __raise_softirq_irqoff(unsigned int nr)
386{
387 trace_softirq_raise(nr);
388 or_softirq_pending(1UL << nr);
389}
390
Carlos R. Mafra962cf362008-05-15 11:15:37 -0300391void open_softirq(int nr, void (*action)(struct softirq_action *))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 softirq_vec[nr].action = action;
394}
395
Peter Zijlstra9ba5f002009-07-22 14:18:35 +0200396/*
397 * Tasklets
398 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399struct tasklet_head
400{
Olof Johansson48f20a92008-03-04 15:23:25 -0800401 struct tasklet_struct *head;
402 struct tasklet_struct **tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403};
404
Vegard Nossum4620b492008-06-12 23:21:53 +0200405static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec);
406static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800408void __tasklet_schedule(struct tasklet_struct *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409{
410 unsigned long flags;
411
412 local_irq_save(flags);
Olof Johansson48f20a92008-03-04 15:23:25 -0800413 t->next = NULL;
Christoph Lameter909ea962010-12-08 16:22:55 +0100414 *__this_cpu_read(tasklet_vec.tail) = t;
415 __this_cpu_write(tasklet_vec.tail, &(t->next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 raise_softirq_irqoff(TASKLET_SOFTIRQ);
417 local_irq_restore(flags);
418}
419
420EXPORT_SYMBOL(__tasklet_schedule);
421
Harvey Harrison7ad5b3a2008-02-08 04:19:53 -0800422void __tasklet_hi_schedule(struct tasklet_struct *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423{
424 unsigned long flags;
425
426 local_irq_save(flags);
Olof Johansson48f20a92008-03-04 15:23:25 -0800427 t->next = NULL;
Christoph Lameter909ea962010-12-08 16:22:55 +0100428 *__this_cpu_read(tasklet_hi_vec.tail) = t;
429 __this_cpu_write(tasklet_hi_vec.tail, &(t->next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 raise_softirq_irqoff(HI_SOFTIRQ);
431 local_irq_restore(flags);
432}
433
434EXPORT_SYMBOL(__tasklet_hi_schedule);
435
Vegard Nossum7c692cb2008-05-21 22:53:13 +0200436void __tasklet_hi_schedule_first(struct tasklet_struct *t)
437{
438 BUG_ON(!irqs_disabled());
439
Christoph Lameter909ea962010-12-08 16:22:55 +0100440 t->next = __this_cpu_read(tasklet_hi_vec.head);
441 __this_cpu_write(tasklet_hi_vec.head, t);
Vegard Nossum7c692cb2008-05-21 22:53:13 +0200442 __raise_softirq_irqoff(HI_SOFTIRQ);
443}
444
445EXPORT_SYMBOL(__tasklet_hi_schedule_first);
446
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447static void tasklet_action(struct softirq_action *a)
448{
449 struct tasklet_struct *list;
450
451 local_irq_disable();
Christoph Lameter909ea962010-12-08 16:22:55 +0100452 list = __this_cpu_read(tasklet_vec.head);
453 __this_cpu_write(tasklet_vec.head, NULL);
454 __this_cpu_write(tasklet_vec.tail, &__get_cpu_var(tasklet_vec).head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 local_irq_enable();
456
457 while (list) {
458 struct tasklet_struct *t = list;
459
460 list = list->next;
461
462 if (tasklet_trylock(t)) {
463 if (!atomic_read(&t->count)) {
464 if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state))
465 BUG();
466 t->func(t->data);
467 tasklet_unlock(t);
468 continue;
469 }
470 tasklet_unlock(t);
471 }
472
473 local_irq_disable();
Olof Johansson48f20a92008-03-04 15:23:25 -0800474 t->next = NULL;
Christoph Lameter909ea962010-12-08 16:22:55 +0100475 *__this_cpu_read(tasklet_vec.tail) = t;
476 __this_cpu_write(tasklet_vec.tail, &(t->next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 __raise_softirq_irqoff(TASKLET_SOFTIRQ);
478 local_irq_enable();
479 }
480}
481
482static void tasklet_hi_action(struct softirq_action *a)
483{
484 struct tasklet_struct *list;
485
486 local_irq_disable();
Christoph Lameter909ea962010-12-08 16:22:55 +0100487 list = __this_cpu_read(tasklet_hi_vec.head);
488 __this_cpu_write(tasklet_hi_vec.head, NULL);
489 __this_cpu_write(tasklet_hi_vec.tail, &__get_cpu_var(tasklet_hi_vec).head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 local_irq_enable();
491
492 while (list) {
493 struct tasklet_struct *t = list;
494
495 list = list->next;
496
497 if (tasklet_trylock(t)) {
498 if (!atomic_read(&t->count)) {
499 if (!test_and_clear_bit(TASKLET_STATE_SCHED, &t->state))
500 BUG();
501 t->func(t->data);
502 tasklet_unlock(t);
503 continue;
504 }
505 tasklet_unlock(t);
506 }
507
508 local_irq_disable();
Olof Johansson48f20a92008-03-04 15:23:25 -0800509 t->next = NULL;
Christoph Lameter909ea962010-12-08 16:22:55 +0100510 *__this_cpu_read(tasklet_hi_vec.tail) = t;
511 __this_cpu_write(tasklet_hi_vec.tail, &(t->next));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 __raise_softirq_irqoff(HI_SOFTIRQ);
513 local_irq_enable();
514 }
515}
516
517
518void tasklet_init(struct tasklet_struct *t,
519 void (*func)(unsigned long), unsigned long data)
520{
521 t->next = NULL;
522 t->state = 0;
523 atomic_set(&t->count, 0);
524 t->func = func;
525 t->data = data;
526}
527
528EXPORT_SYMBOL(tasklet_init);
529
530void tasklet_kill(struct tasklet_struct *t)
531{
532 if (in_interrupt())
533 printk("Attempt to kill tasklet from interrupt\n");
534
535 while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
H Hartley Sweeten79d381c2009-04-16 19:30:18 -0400536 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 yield();
H Hartley Sweeten79d381c2009-04-16 19:30:18 -0400538 } while (test_bit(TASKLET_STATE_SCHED, &t->state));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 }
540 tasklet_unlock_wait(t);
541 clear_bit(TASKLET_STATE_SCHED, &t->state);
542}
543
544EXPORT_SYMBOL(tasklet_kill);
545
Peter Zijlstra9ba5f002009-07-22 14:18:35 +0200546/*
547 * tasklet_hrtimer
548 */
549
550/*
Peter Zijlstrab9c30322010-02-03 18:08:52 +0100551 * The trampoline is called when the hrtimer expires. It schedules a tasklet
552 * to run __tasklet_hrtimer_trampoline() which in turn will call the intended
553 * hrtimer callback, but from softirq context.
Peter Zijlstra9ba5f002009-07-22 14:18:35 +0200554 */
555static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer)
556{
557 struct tasklet_hrtimer *ttimer =
558 container_of(timer, struct tasklet_hrtimer, timer);
559
Peter Zijlstrab9c30322010-02-03 18:08:52 +0100560 tasklet_hi_schedule(&ttimer->tasklet);
561 return HRTIMER_NORESTART;
Peter Zijlstra9ba5f002009-07-22 14:18:35 +0200562}
563
564/*
565 * Helper function which calls the hrtimer callback from
566 * tasklet/softirq context
567 */
568static void __tasklet_hrtimer_trampoline(unsigned long data)
569{
570 struct tasklet_hrtimer *ttimer = (void *)data;
571 enum hrtimer_restart restart;
572
573 restart = ttimer->function(&ttimer->timer);
574 if (restart != HRTIMER_NORESTART)
575 hrtimer_restart(&ttimer->timer);
576}
577
578/**
579 * tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks
580 * @ttimer: tasklet_hrtimer which is initialized
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300581 * @function: hrtimer callback function which gets called from softirq context
Peter Zijlstra9ba5f002009-07-22 14:18:35 +0200582 * @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME)
583 * @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL)
584 */
585void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer,
586 enum hrtimer_restart (*function)(struct hrtimer *),
587 clockid_t which_clock, enum hrtimer_mode mode)
588{
589 hrtimer_init(&ttimer->timer, which_clock, mode);
590 ttimer->timer.function = __hrtimer_tasklet_trampoline;
591 tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline,
592 (unsigned long)ttimer);
593 ttimer->function = function;
594}
595EXPORT_SYMBOL_GPL(tasklet_hrtimer_init);
596
597/*
598 * Remote softirq bits
599 */
600
David S. Miller54514a72008-09-23 22:15:57 -0700601DEFINE_PER_CPU(struct list_head [NR_SOFTIRQS], softirq_work_list);
602EXPORT_PER_CPU_SYMBOL(softirq_work_list);
603
604static void __local_trigger(struct call_single_data *cp, int softirq)
605{
606 struct list_head *head = &__get_cpu_var(softirq_work_list[softirq]);
607
608 list_add_tail(&cp->list, head);
609
610 /* Trigger the softirq only if the list was previously empty. */
611 if (head->next == &cp->list)
612 raise_softirq_irqoff(softirq);
613}
614
615#ifdef CONFIG_USE_GENERIC_SMP_HELPERS
616static void remote_softirq_receive(void *data)
617{
618 struct call_single_data *cp = data;
619 unsigned long flags;
620 int softirq;
621
622 softirq = cp->priv;
623
624 local_irq_save(flags);
625 __local_trigger(cp, softirq);
626 local_irq_restore(flags);
627}
628
629static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq)
630{
631 if (cpu_online(cpu)) {
632 cp->func = remote_softirq_receive;
633 cp->info = cp;
634 cp->flags = 0;
635 cp->priv = softirq;
636
Peter Zijlstra6e275632009-02-25 13:59:48 +0100637 __smp_call_function_single(cpu, cp, 0);
David S. Miller54514a72008-09-23 22:15:57 -0700638 return 0;
639 }
640 return 1;
641}
642#else /* CONFIG_USE_GENERIC_SMP_HELPERS */
643static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softirq)
644{
645 return 1;
646}
647#endif
648
649/**
650 * __send_remote_softirq - try to schedule softirq work on a remote cpu
651 * @cp: private SMP call function data area
652 * @cpu: the remote cpu
653 * @this_cpu: the currently executing cpu
654 * @softirq: the softirq for the work
655 *
656 * Attempt to schedule softirq work on a remote cpu. If this cannot be
657 * done, the work is instead queued up on the local cpu.
658 *
659 * Interrupts must be disabled.
660 */
661void __send_remote_softirq(struct call_single_data *cp, int cpu, int this_cpu, int softirq)
662{
663 if (cpu == this_cpu || __try_remote_softirq(cp, cpu, softirq))
664 __local_trigger(cp, softirq);
665}
666EXPORT_SYMBOL(__send_remote_softirq);
667
668/**
669 * send_remote_softirq - try to schedule softirq work on a remote cpu
670 * @cp: private SMP call function data area
671 * @cpu: the remote cpu
672 * @softirq: the softirq for the work
673 *
674 * Like __send_remote_softirq except that disabling interrupts and
675 * computing the current cpu is done for the caller.
676 */
677void send_remote_softirq(struct call_single_data *cp, int cpu, int softirq)
678{
679 unsigned long flags;
680 int this_cpu;
681
682 local_irq_save(flags);
683 this_cpu = smp_processor_id();
684 __send_remote_softirq(cp, cpu, this_cpu, softirq);
685 local_irq_restore(flags);
686}
687EXPORT_SYMBOL(send_remote_softirq);
688
689static int __cpuinit remote_softirq_cpu_notify(struct notifier_block *self,
690 unsigned long action, void *hcpu)
691{
692 /*
693 * If a CPU goes away, splice its entries to the current CPU
694 * and trigger a run of the softirq
695 */
696 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
697 int cpu = (unsigned long) hcpu;
698 int i;
699
700 local_irq_disable();
701 for (i = 0; i < NR_SOFTIRQS; i++) {
702 struct list_head *head = &per_cpu(softirq_work_list[i], cpu);
703 struct list_head *local_head;
704
705 if (list_empty(head))
706 continue;
707
708 local_head = &__get_cpu_var(softirq_work_list[i]);
709 list_splice_init(head, local_head);
710 raise_softirq_irqoff(i);
711 }
712 local_irq_enable();
713 }
714
715 return NOTIFY_OK;
716}
717
718static struct notifier_block __cpuinitdata remote_softirq_cpu_notifier = {
719 .notifier_call = remote_softirq_cpu_notify,
720};
721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722void __init softirq_init(void)
723{
Olof Johansson48f20a92008-03-04 15:23:25 -0800724 int cpu;
725
726 for_each_possible_cpu(cpu) {
David S. Miller54514a72008-09-23 22:15:57 -0700727 int i;
728
Olof Johansson48f20a92008-03-04 15:23:25 -0800729 per_cpu(tasklet_vec, cpu).tail =
730 &per_cpu(tasklet_vec, cpu).head;
731 per_cpu(tasklet_hi_vec, cpu).tail =
732 &per_cpu(tasklet_hi_vec, cpu).head;
David S. Miller54514a72008-09-23 22:15:57 -0700733 for (i = 0; i < NR_SOFTIRQS; i++)
734 INIT_LIST_HEAD(&per_cpu(softirq_work_list[i], cpu));
Olof Johansson48f20a92008-03-04 15:23:25 -0800735 }
736
David S. Miller54514a72008-09-23 22:15:57 -0700737 register_hotcpu_notifier(&remote_softirq_cpu_notifier);
738
Carlos R. Mafra962cf362008-05-15 11:15:37 -0300739 open_softirq(TASKLET_SOFTIRQ, tasklet_action);
740 open_softirq(HI_SOFTIRQ, tasklet_hi_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741}
742
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000743static int ksoftirqd_should_run(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000745 return local_softirq_pending();
746}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000748static void run_ksoftirqd(unsigned int cpu)
749{
750 local_irq_disable();
751 if (local_softirq_pending()) {
752 __do_softirq();
753 rcu_note_context_switch(cpu);
754 local_irq_enable();
755 cond_resched();
756 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 }
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000758 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759}
760
761#ifdef CONFIG_HOTPLUG_CPU
762/*
763 * tasklet_kill_immediate is called to remove a tasklet which can already be
764 * scheduled for execution on @cpu.
765 *
766 * Unlike tasklet_kill, this function removes the tasklet
767 * _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state.
768 *
769 * When this function is called, @cpu must be in the CPU_DEAD state.
770 */
771void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu)
772{
773 struct tasklet_struct **i;
774
775 BUG_ON(cpu_online(cpu));
776 BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state));
777
778 if (!test_bit(TASKLET_STATE_SCHED, &t->state))
779 return;
780
781 /* CPU is dead, so no lock needed. */
Olof Johansson48f20a92008-03-04 15:23:25 -0800782 for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 if (*i == t) {
784 *i = t->next;
Olof Johansson48f20a92008-03-04 15:23:25 -0800785 /* If this was the tail element, move the tail ptr */
786 if (*i == NULL)
787 per_cpu(tasklet_vec, cpu).tail = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 return;
789 }
790 }
791 BUG();
792}
793
794static void takeover_tasklets(unsigned int cpu)
795{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 /* CPU is dead, so no lock needed. */
797 local_irq_disable();
798
799 /* Find end, append list for that CPU. */
Christian Borntraegere5e41722008-05-01 04:34:23 -0700800 if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) {
Christoph Lameter909ea962010-12-08 16:22:55 +0100801 *__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head;
802 this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail);
Christian Borntraegere5e41722008-05-01 04:34:23 -0700803 per_cpu(tasklet_vec, cpu).head = NULL;
804 per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head;
805 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 raise_softirq_irqoff(TASKLET_SOFTIRQ);
807
Christian Borntraegere5e41722008-05-01 04:34:23 -0700808 if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) {
Christoph Lameter909ea962010-12-08 16:22:55 +0100809 *__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head;
810 __this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail);
Christian Borntraegere5e41722008-05-01 04:34:23 -0700811 per_cpu(tasklet_hi_vec, cpu).head = NULL;
812 per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head;
813 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 raise_softirq_irqoff(HI_SOFTIRQ);
815
816 local_irq_enable();
817}
818#endif /* CONFIG_HOTPLUG_CPU */
819
Chandra Seetharaman8c78f302006-07-30 03:03:35 -0700820static int __cpuinit cpu_callback(struct notifier_block *nfb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 unsigned long action,
822 void *hcpu)
823{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 switch (action) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825#ifdef CONFIG_HOTPLUG_CPU
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 case CPU_DEAD:
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000827 case CPU_DEAD_FROZEN:
828 takeover_tasklets((unsigned long)hcpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 break;
830#endif /* CONFIG_HOTPLUG_CPU */
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000831 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 return NOTIFY_OK;
833}
834
Chandra Seetharaman8c78f302006-07-30 03:03:35 -0700835static struct notifier_block __cpuinitdata cpu_nfb = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 .notifier_call = cpu_callback
837};
838
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000839static struct smp_hotplug_thread softirq_threads = {
840 .store = &ksoftirqd,
841 .thread_should_run = ksoftirqd_should_run,
842 .thread_fn = run_ksoftirqd,
843 .thread_comm = "ksoftirqd/%u",
844};
845
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700846static __init int spawn_ksoftirqd(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 register_cpu_notifier(&cpu_nfb);
Thomas Gleixner3e339b52012-07-16 10:42:37 +0000849
850 BUG_ON(smpboot_register_percpu_thread(&softirq_threads));
851
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 return 0;
853}
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700854early_initcall(spawn_ksoftirqd);
Andrew Morton78eef012006-03-22 00:08:16 -0800855
Yinghai Lu43a25632008-12-28 16:01:13 -0800856/*
857 * [ These __weak aliases are kept in a separate compilation unit, so that
858 * GCC does not inline them incorrectly. ]
859 */
860
861int __init __weak early_irq_init(void)
862{
863 return 0;
864}
865
Thomas Gleixnerb683de22010-09-27 20:55:03 +0200866#ifdef CONFIG_GENERIC_HARDIRQS
Yinghai Lu4a046d12009-01-12 17:39:24 -0800867int __init __weak arch_probe_nr_irqs(void)
868{
Thomas Gleixnerb683de22010-09-27 20:55:03 +0200869 return NR_IRQS_LEGACY;
Yinghai Lu4a046d12009-01-12 17:39:24 -0800870}
871
Yinghai Lu43a25632008-12-28 16:01:13 -0800872int __init __weak arch_early_irq_init(void)
873{
874 return 0;
875}
Thomas Gleixnerb683de22010-09-27 20:55:03 +0200876#endif