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