rcu: Make rcu_read_lock_sched_held() take boot time into account
[linux-2.6.git] / kernel / irq / handle.c
1 /*
2  * linux/kernel/irq/handle.c
3  *
4  * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
5  * Copyright (C) 2005-2006, Thomas Gleixner, Russell King
6  *
7  * This file contains the core interrupt handling code.
8  *
9  * Detailed information is available in Documentation/DocBook/genericirq
10  *
11  */
12
13 #include <linux/irq.h>
14 #include <linux/sched.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/random.h>
18 #include <linux/interrupt.h>
19 #include <linux/kernel_stat.h>
20 #include <linux/rculist.h>
21 #include <linux/hash.h>
22 #include <linux/bootmem.h>
23 #include <trace/events/irq.h>
24
25 #include "internals.h"
26
27 /*
28  * lockdep: we want to handle all irq_desc locks as a single lock-class:
29  */
30 struct lock_class_key irq_desc_lock_class;
31
32 /**
33  * handle_bad_irq - handle spurious and unhandled irqs
34  * @irq:       the interrupt number
35  * @desc:      description of the interrupt
36  *
37  * Handles spurious and unhandled IRQ's. It also prints a debugmessage.
38  */
39 void handle_bad_irq(unsigned int irq, struct irq_desc *desc)
40 {
41         print_irq_desc(irq, desc);
42         kstat_incr_irqs_this_cpu(irq, desc);
43         ack_bad_irq(irq);
44 }
45
46 #if defined(CONFIG_SMP) && defined(CONFIG_GENERIC_HARDIRQS)
47 static void __init init_irq_default_affinity(void)
48 {
49         alloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
50         cpumask_setall(irq_default_affinity);
51 }
52 #else
53 static void __init init_irq_default_affinity(void)
54 {
55 }
56 #endif
57
58 /*
59  * Linux has a controller-independent interrupt architecture.
60  * Every controller has a 'controller-template', that is used
61  * by the main code to do the right thing. Each driver-visible
62  * interrupt source is transparently wired to the appropriate
63  * controller. Thus drivers need not be aware of the
64  * interrupt-controller.
65  *
66  * The code is designed to be easily extended with new/different
67  * interrupt controllers, without having to do assembly magic or
68  * having to touch the generic code.
69  *
70  * Controller mappings for all interrupt sources:
71  */
72 int nr_irqs = NR_IRQS;
73 EXPORT_SYMBOL_GPL(nr_irqs);
74
75 #ifdef CONFIG_SPARSE_IRQ
76
77 static struct irq_desc irq_desc_init = {
78         .irq        = -1,
79         .status     = IRQ_DISABLED,
80         .chip       = &no_irq_chip,
81         .handle_irq = handle_bad_irq,
82         .depth      = 1,
83         .lock       = __RAW_SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
84 };
85
86 void __ref init_kstat_irqs(struct irq_desc *desc, int node, int nr)
87 {
88         void *ptr;
89
90         if (slab_is_available())
91                 ptr = kzalloc_node(nr * sizeof(*desc->kstat_irqs),
92                                    GFP_ATOMIC, node);
93         else
94                 ptr = alloc_bootmem_node(NODE_DATA(node),
95                                 nr * sizeof(*desc->kstat_irqs));
96
97         /*
98          * don't overwite if can not get new one
99          * init_copy_kstat_irqs() could still use old one
100          */
101         if (ptr) {
102                 printk(KERN_DEBUG "  alloc kstat_irqs on node %d\n", node);
103                 desc->kstat_irqs = ptr;
104         }
105 }
106
107 static void init_one_irq_desc(int irq, struct irq_desc *desc, int node)
108 {
109         memcpy(desc, &irq_desc_init, sizeof(struct irq_desc));
110
111         raw_spin_lock_init(&desc->lock);
112         desc->irq = irq;
113 #ifdef CONFIG_SMP
114         desc->node = node;
115 #endif
116         lockdep_set_class(&desc->lock, &irq_desc_lock_class);
117         init_kstat_irqs(desc, node, nr_cpu_ids);
118         if (!desc->kstat_irqs) {
119                 printk(KERN_ERR "can not alloc kstat_irqs\n");
120                 BUG_ON(1);
121         }
122         if (!alloc_desc_masks(desc, node, false)) {
123                 printk(KERN_ERR "can not alloc irq_desc cpumasks\n");
124                 BUG_ON(1);
125         }
126         init_desc_masks(desc);
127         arch_init_chip_data(desc, node);
128 }
129
130 /*
131  * Protect the sparse_irqs:
132  */
133 DEFINE_RAW_SPINLOCK(sparse_irq_lock);
134
135 struct irq_desc **irq_desc_ptrs __read_mostly;
136
137 static struct irq_desc irq_desc_legacy[NR_IRQS_LEGACY] __cacheline_aligned_in_smp = {
138         [0 ... NR_IRQS_LEGACY-1] = {
139                 .irq        = -1,
140                 .status     = IRQ_DISABLED,
141                 .chip       = &no_irq_chip,
142                 .handle_irq = handle_bad_irq,
143                 .depth      = 1,
144                 .lock       = __RAW_SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
145         }
146 };
147
148 static unsigned int *kstat_irqs_legacy;
149
150 int __init early_irq_init(void)
151 {
152         struct irq_desc *desc;
153         int legacy_count;
154         int node;
155         int i;
156
157         init_irq_default_affinity();
158
159          /* initialize nr_irqs based on nr_cpu_ids */
160         arch_probe_nr_irqs();
161         printk(KERN_INFO "NR_IRQS:%d nr_irqs:%d\n", NR_IRQS, nr_irqs);
162
163         desc = irq_desc_legacy;
164         legacy_count = ARRAY_SIZE(irq_desc_legacy);
165         node = first_online_node;
166
167         /* allocate irq_desc_ptrs array based on nr_irqs */
168         irq_desc_ptrs = kcalloc(nr_irqs, sizeof(void *), GFP_NOWAIT);
169
170         /* allocate based on nr_cpu_ids */
171         kstat_irqs_legacy = kzalloc_node(NR_IRQS_LEGACY * nr_cpu_ids *
172                                           sizeof(int), GFP_NOWAIT, node);
173
174         for (i = 0; i < legacy_count; i++) {
175                 desc[i].irq = i;
176 #ifdef CONFIG_SMP
177                 desc[i].node = node;
178 #endif
179                 desc[i].kstat_irqs = kstat_irqs_legacy + i * nr_cpu_ids;
180                 lockdep_set_class(&desc[i].lock, &irq_desc_lock_class);
181                 alloc_desc_masks(&desc[i], node, true);
182                 init_desc_masks(&desc[i]);
183                 irq_desc_ptrs[i] = desc + i;
184         }
185
186         for (i = legacy_count; i < nr_irqs; i++)
187                 irq_desc_ptrs[i] = NULL;
188
189         return arch_early_irq_init();
190 }
191
192 struct irq_desc *irq_to_desc(unsigned int irq)
193 {
194         if (irq_desc_ptrs && irq < nr_irqs)
195                 return irq_desc_ptrs[irq];
196
197         return NULL;
198 }
199
200 struct irq_desc * __ref irq_to_desc_alloc_node(unsigned int irq, int node)
201 {
202         struct irq_desc *desc;
203         unsigned long flags;
204
205         if (irq >= nr_irqs) {
206                 WARN(1, "irq (%d) >= nr_irqs (%d) in irq_to_desc_alloc\n",
207                         irq, nr_irqs);
208                 return NULL;
209         }
210
211         desc = irq_desc_ptrs[irq];
212         if (desc)
213                 return desc;
214
215         raw_spin_lock_irqsave(&sparse_irq_lock, flags);
216
217         /* We have to check it to avoid races with another CPU */
218         desc = irq_desc_ptrs[irq];
219         if (desc)
220                 goto out_unlock;
221
222         if (slab_is_available())
223                 desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
224         else
225                 desc = alloc_bootmem_node(NODE_DATA(node), sizeof(*desc));
226
227         printk(KERN_DEBUG "  alloc irq_desc for %d on node %d\n", irq, node);
228         if (!desc) {
229                 printk(KERN_ERR "can not alloc irq_desc\n");
230                 BUG_ON(1);
231         }
232         init_one_irq_desc(irq, desc, node);
233
234         irq_desc_ptrs[irq] = desc;
235
236 out_unlock:
237         raw_spin_unlock_irqrestore(&sparse_irq_lock, flags);
238
239         return desc;
240 }
241
242 #else /* !CONFIG_SPARSE_IRQ */
243
244 struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
245         [0 ... NR_IRQS-1] = {
246                 .status = IRQ_DISABLED,
247                 .chip = &no_irq_chip,
248                 .handle_irq = handle_bad_irq,
249                 .depth = 1,
250                 .lock = __RAW_SPIN_LOCK_UNLOCKED(irq_desc->lock),
251         }
252 };
253
254 static unsigned int kstat_irqs_all[NR_IRQS][NR_CPUS];
255 int __init early_irq_init(void)
256 {
257         struct irq_desc *desc;
258         int count;
259         int i;
260
261         init_irq_default_affinity();
262
263         printk(KERN_INFO "NR_IRQS:%d\n", NR_IRQS);
264
265         desc = irq_desc;
266         count = ARRAY_SIZE(irq_desc);
267
268         for (i = 0; i < count; i++) {
269                 desc[i].irq = i;
270                 alloc_desc_masks(&desc[i], 0, true);
271                 init_desc_masks(&desc[i]);
272                 desc[i].kstat_irqs = kstat_irqs_all[i];
273         }
274         return arch_early_irq_init();
275 }
276
277 struct irq_desc *irq_to_desc(unsigned int irq)
278 {
279         return (irq < NR_IRQS) ? irq_desc + irq : NULL;
280 }
281
282 struct irq_desc *irq_to_desc_alloc_node(unsigned int irq, int node)
283 {
284         return irq_to_desc(irq);
285 }
286 #endif /* !CONFIG_SPARSE_IRQ */
287
288 void clear_kstat_irqs(struct irq_desc *desc)
289 {
290         memset(desc->kstat_irqs, 0, nr_cpu_ids * sizeof(*(desc->kstat_irqs)));
291 }
292
293 /*
294  * What should we do if we get a hw irq event on an illegal vector?
295  * Each architecture has to answer this themself.
296  */
297 static void ack_bad(unsigned int irq)
298 {
299         struct irq_desc *desc = irq_to_desc(irq);
300
301         print_irq_desc(irq, desc);
302         ack_bad_irq(irq);
303 }
304
305 /*
306  * NOP functions
307  */
308 static void noop(unsigned int irq)
309 {
310 }
311
312 static unsigned int noop_ret(unsigned int irq)
313 {
314         return 0;
315 }
316
317 /*
318  * Generic no controller implementation
319  */
320 struct irq_chip no_irq_chip = {
321         .name           = "none",
322         .startup        = noop_ret,
323         .shutdown       = noop,
324         .enable         = noop,
325         .disable        = noop,
326         .ack            = ack_bad,
327         .end            = noop,
328 };
329
330 /*
331  * Generic dummy implementation which can be used for
332  * real dumb interrupt sources
333  */
334 struct irq_chip dummy_irq_chip = {
335         .name           = "dummy",
336         .startup        = noop_ret,
337         .shutdown       = noop,
338         .enable         = noop,
339         .disable        = noop,
340         .ack            = noop,
341         .mask           = noop,
342         .unmask         = noop,
343         .end            = noop,
344 };
345
346 /*
347  * Special, empty irq handler:
348  */
349 irqreturn_t no_action(int cpl, void *dev_id)
350 {
351         return IRQ_NONE;
352 }
353
354 static void warn_no_thread(unsigned int irq, struct irqaction *action)
355 {
356         if (test_and_set_bit(IRQTF_WARNED, &action->thread_flags))
357                 return;
358
359         printk(KERN_WARNING "IRQ %d device %s returned IRQ_WAKE_THREAD "
360                "but no thread function available.", irq, action->name);
361 }
362
363 /**
364  * handle_IRQ_event - irq action chain handler
365  * @irq:        the interrupt number
366  * @action:     the interrupt action chain for this irq
367  *
368  * Handles the action chain of an irq event
369  */
370 irqreturn_t handle_IRQ_event(unsigned int irq, struct irqaction *action)
371 {
372         irqreturn_t ret, retval = IRQ_NONE;
373         unsigned int status = 0;
374
375         if (!(action->flags & IRQF_DISABLED))
376                 local_irq_enable_in_hardirq();
377
378         do {
379                 trace_irq_handler_entry(irq, action);
380                 ret = action->handler(irq, action->dev_id);
381                 trace_irq_handler_exit(irq, action, ret);
382
383                 switch (ret) {
384                 case IRQ_WAKE_THREAD:
385                         /*
386                          * Set result to handled so the spurious check
387                          * does not trigger.
388                          */
389                         ret = IRQ_HANDLED;
390
391                         /*
392                          * Catch drivers which return WAKE_THREAD but
393                          * did not set up a thread function
394                          */
395                         if (unlikely(!action->thread_fn)) {
396                                 warn_no_thread(irq, action);
397                                 break;
398                         }
399
400                         /*
401                          * Wake up the handler thread for this
402                          * action. In case the thread crashed and was
403                          * killed we just pretend that we handled the
404                          * interrupt. The hardirq handler above has
405                          * disabled the device interrupt, so no irq
406                          * storm is lurking.
407                          */
408                         if (likely(!test_bit(IRQTF_DIED,
409                                              &action->thread_flags))) {
410                                 set_bit(IRQTF_RUNTHREAD, &action->thread_flags);
411                                 wake_up_process(action->thread);
412                         }
413
414                         /* Fall through to add to randomness */
415                 case IRQ_HANDLED:
416                         status |= action->flags;
417                         break;
418
419                 default:
420                         break;
421                 }
422
423                 retval |= ret;
424                 action = action->next;
425         } while (action);
426
427         if (status & IRQF_SAMPLE_RANDOM)
428                 add_interrupt_randomness(irq);
429         local_irq_disable();
430
431         return retval;
432 }
433
434 #ifndef CONFIG_GENERIC_HARDIRQS_NO__DO_IRQ
435
436 #ifdef CONFIG_ENABLE_WARN_DEPRECATED
437 # warning __do_IRQ is deprecated. Please convert to proper flow handlers
438 #endif
439
440 /**
441  * __do_IRQ - original all in one highlevel IRQ handler
442  * @irq:        the interrupt number
443  *
444  * __do_IRQ handles all normal device IRQ's (the special
445  * SMP cross-CPU interrupts have their own specific
446  * handlers).
447  *
448  * This is the original x86 implementation which is used for every
449  * interrupt type.
450  */
451 unsigned int __do_IRQ(unsigned int irq)
452 {
453         struct irq_desc *desc = irq_to_desc(irq);
454         struct irqaction *action;
455         unsigned int status;
456
457         kstat_incr_irqs_this_cpu(irq, desc);
458
459         if (CHECK_IRQ_PER_CPU(desc->status)) {
460                 irqreturn_t action_ret;
461
462                 /*
463                  * No locking required for CPU-local interrupts:
464                  */
465                 if (desc->chip->ack)
466                         desc->chip->ack(irq);
467                 if (likely(!(desc->status & IRQ_DISABLED))) {
468                         action_ret = handle_IRQ_event(irq, desc->action);
469                         if (!noirqdebug)
470                                 note_interrupt(irq, desc, action_ret);
471                 }
472                 desc->chip->end(irq);
473                 return 1;
474         }
475
476         raw_spin_lock(&desc->lock);
477         if (desc->chip->ack)
478                 desc->chip->ack(irq);
479         /*
480          * REPLAY is when Linux resends an IRQ that was dropped earlier
481          * WAITING is used by probe to mark irqs that are being tested
482          */
483         status = desc->status & ~(IRQ_REPLAY | IRQ_WAITING);
484         status |= IRQ_PENDING; /* we _want_ to handle it */
485
486         /*
487          * If the IRQ is disabled for whatever reason, we cannot
488          * use the action we have.
489          */
490         action = NULL;
491         if (likely(!(status & (IRQ_DISABLED | IRQ_INPROGRESS)))) {
492                 action = desc->action;
493                 status &= ~IRQ_PENDING; /* we commit to handling */
494                 status |= IRQ_INPROGRESS; /* we are handling it */
495         }
496         desc->status = status;
497
498         /*
499          * If there is no IRQ handler or it was disabled, exit early.
500          * Since we set PENDING, if another processor is handling
501          * a different instance of this same irq, the other processor
502          * will take care of it.
503          */
504         if (unlikely(!action))
505                 goto out;
506
507         /*
508          * Edge triggered interrupts need to remember
509          * pending events.
510          * This applies to any hw interrupts that allow a second
511          * instance of the same irq to arrive while we are in do_IRQ
512          * or in the handler. But the code here only handles the _second_
513          * instance of the irq, not the third or fourth. So it is mostly
514          * useful for irq hardware that does not mask cleanly in an
515          * SMP environment.
516          */
517         for (;;) {
518                 irqreturn_t action_ret;
519
520                 raw_spin_unlock(&desc->lock);
521
522                 action_ret = handle_IRQ_event(irq, action);
523                 if (!noirqdebug)
524                         note_interrupt(irq, desc, action_ret);
525
526                 raw_spin_lock(&desc->lock);
527                 if (likely(!(desc->status & IRQ_PENDING)))
528                         break;
529                 desc->status &= ~IRQ_PENDING;
530         }
531         desc->status &= ~IRQ_INPROGRESS;
532
533 out:
534         /*
535          * The ->end() handler has to deal with interrupts which got
536          * disabled while the handler was running.
537          */
538         desc->chip->end(irq);
539         raw_spin_unlock(&desc->lock);
540
541         return 1;
542 }
543 #endif
544
545 void early_init_irq_lock_class(void)
546 {
547         struct irq_desc *desc;
548         int i;
549
550         for_each_irq_desc(i, desc) {
551                 lockdep_set_class(&desc->lock, &irq_desc_lock_class);
552         }
553 }
554
555 unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
556 {
557         struct irq_desc *desc = irq_to_desc(irq);
558         return desc ? desc->kstat_irqs[cpu] : 0;
559 }
560 EXPORT_SYMBOL(kstat_irqs_cpu);
561