a81cf80554dbf4da46c53cada05fc747f7c8eeff
[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_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
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 node;
154         int i;
155
156         init_irq_default_affinity();
157
158          /* initialize nr_irqs based on nr_cpu_ids */
159         arch_probe_nr_irqs();
160         printk(KERN_INFO "NR_IRQS:%d nr_irqs:%d\n", NR_IRQS, nr_irqs);
161
162         desc = irq_desc_legacy;
163         legacy_count = ARRAY_SIZE(irq_desc_legacy);
164         node = first_online_node;
165
166         /* allocate irq_desc_ptrs array based on nr_irqs */
167         irq_desc_ptrs = kcalloc(nr_irqs, sizeof(void *), GFP_NOWAIT);
168
169         /* allocate based on nr_cpu_ids */
170         kstat_irqs_legacy = kzalloc_node(NR_IRQS_LEGACY * nr_cpu_ids *
171                                           sizeof(int), GFP_NOWAIT, node);
172
173         for (i = 0; i < legacy_count; i++) {
174                 desc[i].irq = i;
175 #ifdef CONFIG_SMP
176                 desc[i].node = node;
177 #endif
178                 desc[i].kstat_irqs = kstat_irqs_legacy + i * nr_cpu_ids;
179                 lockdep_set_class(&desc[i].lock, &irq_desc_lock_class);
180                 alloc_desc_masks(&desc[i], node, true);
181                 init_desc_masks(&desc[i]);
182                 irq_desc_ptrs[i] = desc + i;
183         }
184
185         for (i = legacy_count; i < nr_irqs; i++)
186                 irq_desc_ptrs[i] = NULL;
187
188         return arch_early_irq_init();
189 }
190
191 struct irq_desc *irq_to_desc(unsigned int irq)
192 {
193         if (irq_desc_ptrs && irq < nr_irqs)
194                 return irq_desc_ptrs[irq];
195
196         return NULL;
197 }
198
199 struct irq_desc * __ref irq_to_desc_alloc_node(unsigned int irq, int node)
200 {
201         struct irq_desc *desc;
202         unsigned long flags;
203
204         if (irq >= nr_irqs) {
205                 WARN(1, "irq (%d) >= nr_irqs (%d) in irq_to_desc_alloc\n",
206                         irq, nr_irqs);
207                 return NULL;
208         }
209
210         desc = irq_desc_ptrs[irq];
211         if (desc)
212                 return desc;
213
214         spin_lock_irqsave(&sparse_irq_lock, flags);
215
216         /* We have to check it to avoid races with another CPU */
217         desc = irq_desc_ptrs[irq];
218         if (desc)
219                 goto out_unlock;
220
221         if (slab_is_available())
222                 desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
223         else
224                 desc = alloc_bootmem_node(NODE_DATA(node), sizeof(*desc));
225
226         printk(KERN_DEBUG "  alloc irq_desc for %d on node %d\n", irq, node);
227         if (!desc) {
228                 printk(KERN_ERR "can not alloc irq_desc\n");
229                 BUG_ON(1);
230         }
231         init_one_irq_desc(irq, desc, node);
232
233         irq_desc_ptrs[irq] = desc;
234
235 out_unlock:
236         spin_unlock_irqrestore(&sparse_irq_lock, flags);
237
238         return desc;
239 }
240
241 #else /* !CONFIG_SPARSE_IRQ */
242
243 struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
244         [0 ... NR_IRQS-1] = {
245                 .status = IRQ_DISABLED,
246                 .chip = &no_irq_chip,
247                 .handle_irq = handle_bad_irq,
248                 .depth = 1,
249                 .lock = __SPIN_LOCK_UNLOCKED(irq_desc->lock),
250         }
251 };
252
253 static unsigned int kstat_irqs_all[NR_IRQS][NR_CPUS];
254 int __init early_irq_init(void)
255 {
256         struct irq_desc *desc;
257         int count;
258         int i;
259
260         init_irq_default_affinity();
261
262         printk(KERN_INFO "NR_IRQS:%d\n", NR_IRQS);
263
264         desc = irq_desc;
265         count = ARRAY_SIZE(irq_desc);
266
267         for (i = 0; i < count; i++) {
268                 desc[i].irq = i;
269                 alloc_desc_masks(&desc[i], 0, true);
270                 init_desc_masks(&desc[i]);
271                 desc[i].kstat_irqs = kstat_irqs_all[i];
272         }
273         return arch_early_irq_init();
274 }
275
276 struct irq_desc *irq_to_desc(unsigned int irq)
277 {
278         return (irq < NR_IRQS) ? irq_desc + irq : NULL;
279 }
280
281 struct irq_desc *irq_to_desc_alloc_node(unsigned int irq, int node)
282 {
283         return irq_to_desc(irq);
284 }
285 #endif /* !CONFIG_SPARSE_IRQ */
286
287 void clear_kstat_irqs(struct irq_desc *desc)
288 {
289         memset(desc->kstat_irqs, 0, nr_cpu_ids * sizeof(*(desc->kstat_irqs)));
290 }
291
292 /*
293  * What should we do if we get a hw irq event on an illegal vector?
294  * Each architecture has to answer this themself.
295  */
296 static void ack_bad(unsigned int irq)
297 {
298         struct irq_desc *desc = irq_to_desc(irq);
299
300         print_irq_desc(irq, desc);
301         ack_bad_irq(irq);
302 }
303
304 /*
305  * NOP functions
306  */
307 static void noop(unsigned int irq)
308 {
309 }
310
311 static unsigned int noop_ret(unsigned int irq)
312 {
313         return 0;
314 }
315
316 /*
317  * Generic no controller implementation
318  */
319 struct irq_chip no_irq_chip = {
320         .name           = "none",
321         .startup        = noop_ret,
322         .shutdown       = noop,
323         .enable         = noop,
324         .disable        = noop,
325         .ack            = ack_bad,
326         .end            = noop,
327 };
328
329 /*
330  * Generic dummy implementation which can be used for
331  * real dumb interrupt sources
332  */
333 struct irq_chip dummy_irq_chip = {
334         .name           = "dummy",
335         .startup        = noop_ret,
336         .shutdown       = noop,
337         .enable         = noop,
338         .disable        = noop,
339         .ack            = noop,
340         .mask           = noop,
341         .unmask         = noop,
342         .end            = noop,
343 };
344
345 /*
346  * Special, empty irq handler:
347  */
348 irqreturn_t no_action(int cpl, void *dev_id)
349 {
350         return IRQ_NONE;
351 }
352
353 static void warn_no_thread(unsigned int irq, struct irqaction *action)
354 {
355         if (test_and_set_bit(IRQTF_WARNED, &action->thread_flags))
356                 return;
357
358         printk(KERN_WARNING "IRQ %d device %s returned IRQ_WAKE_THREAD "
359                "but no thread function available.", irq, action->name);
360 }
361
362 /**
363  * handle_IRQ_event - irq action chain handler
364  * @irq:        the interrupt number
365  * @action:     the interrupt action chain for this irq
366  *
367  * Handles the action chain of an irq event
368  */
369 irqreturn_t handle_IRQ_event(unsigned int irq, struct irqaction *action)
370 {
371         irqreturn_t ret, retval = IRQ_NONE;
372         unsigned int status = 0;
373
374         if (!(action->flags & IRQF_DISABLED))
375                 local_irq_enable_in_hardirq();
376
377         do {
378                 trace_irq_handler_entry(irq, action);
379                 ret = action->handler(irq, action->dev_id);
380                 trace_irq_handler_exit(irq, action, ret);
381
382                 switch (ret) {
383                 case IRQ_WAKE_THREAD:
384                         /*
385                          * Set result to handled so the spurious check
386                          * does not trigger.
387                          */
388                         ret = IRQ_HANDLED;
389
390                         /*
391                          * Catch drivers which return WAKE_THREAD but
392                          * did not set up a thread function
393                          */
394                         if (unlikely(!action->thread_fn)) {
395                                 warn_no_thread(irq, action);
396                                 break;
397                         }
398
399                         /*
400                          * Wake up the handler thread for this
401                          * action. In case the thread crashed and was
402                          * killed we just pretend that we handled the
403                          * interrupt. The hardirq handler above has
404                          * disabled the device interrupt, so no irq
405                          * storm is lurking.
406                          */
407                         if (likely(!test_bit(IRQTF_DIED,
408                                              &action->thread_flags))) {
409                                 set_bit(IRQTF_RUNTHREAD, &action->thread_flags);
410                                 wake_up_process(action->thread);
411                         }
412
413                         /* Fall through to add to randomness */
414                 case IRQ_HANDLED:
415                         status |= action->flags;
416                         break;
417
418                 default:
419                         break;
420                 }
421
422                 retval |= ret;
423                 action = action->next;
424         } while (action);
425
426         if (status & IRQF_SAMPLE_RANDOM)
427                 add_interrupt_randomness(irq);
428         local_irq_disable();
429
430         return retval;
431 }
432
433 #ifndef CONFIG_GENERIC_HARDIRQS_NO__DO_IRQ
434
435 #ifdef CONFIG_ENABLE_WARN_DEPRECATED
436 # warning __do_IRQ is deprecated. Please convert to proper flow handlers
437 #endif
438
439 /**
440  * __do_IRQ - original all in one highlevel IRQ handler
441  * @irq:        the interrupt number
442  *
443  * __do_IRQ handles all normal device IRQ's (the special
444  * SMP cross-CPU interrupts have their own specific
445  * handlers).
446  *
447  * This is the original x86 implementation which is used for every
448  * interrupt type.
449  */
450 unsigned int __do_IRQ(unsigned int irq)
451 {
452         struct irq_desc *desc = irq_to_desc(irq);
453         struct irqaction *action;
454         unsigned int status;
455
456         kstat_incr_irqs_this_cpu(irq, desc);
457
458         if (CHECK_IRQ_PER_CPU(desc->status)) {
459                 irqreturn_t action_ret;
460
461                 /*
462                  * No locking required for CPU-local interrupts:
463                  */
464                 if (desc->chip->ack)
465                         desc->chip->ack(irq);
466                 if (likely(!(desc->status & IRQ_DISABLED))) {
467                         action_ret = handle_IRQ_event(irq, desc->action);
468                         if (!noirqdebug)
469                                 note_interrupt(irq, desc, action_ret);
470                 }
471                 desc->chip->end(irq);
472                 return 1;
473         }
474
475         spin_lock(&desc->lock);
476         if (desc->chip->ack)
477                 desc->chip->ack(irq);
478         /*
479          * REPLAY is when Linux resends an IRQ that was dropped earlier
480          * WAITING is used by probe to mark irqs that are being tested
481          */
482         status = desc->status & ~(IRQ_REPLAY | IRQ_WAITING);
483         status |= IRQ_PENDING; /* we _want_ to handle it */
484
485         /*
486          * If the IRQ is disabled for whatever reason, we cannot
487          * use the action we have.
488          */
489         action = NULL;
490         if (likely(!(status & (IRQ_DISABLED | IRQ_INPROGRESS)))) {
491                 action = desc->action;
492                 status &= ~IRQ_PENDING; /* we commit to handling */
493                 status |= IRQ_INPROGRESS; /* we are handling it */
494         }
495         desc->status = status;
496
497         /*
498          * If there is no IRQ handler or it was disabled, exit early.
499          * Since we set PENDING, if another processor is handling
500          * a different instance of this same irq, the other processor
501          * will take care of it.
502          */
503         if (unlikely(!action))
504                 goto out;
505
506         /*
507          * Edge triggered interrupts need to remember
508          * pending events.
509          * This applies to any hw interrupts that allow a second
510          * instance of the same irq to arrive while we are in do_IRQ
511          * or in the handler. But the code here only handles the _second_
512          * instance of the irq, not the third or fourth. So it is mostly
513          * useful for irq hardware that does not mask cleanly in an
514          * SMP environment.
515          */
516         for (;;) {
517                 irqreturn_t action_ret;
518
519                 spin_unlock(&desc->lock);
520
521                 action_ret = handle_IRQ_event(irq, action);
522                 if (!noirqdebug)
523                         note_interrupt(irq, desc, action_ret);
524
525                 spin_lock(&desc->lock);
526                 if (likely(!(desc->status & IRQ_PENDING)))
527                         break;
528                 desc->status &= ~IRQ_PENDING;
529         }
530         desc->status &= ~IRQ_INPROGRESS;
531
532 out:
533         /*
534          * The ->end() handler has to deal with interrupts which got
535          * disabled while the handler was running.
536          */
537         desc->chip->end(irq);
538         spin_unlock(&desc->lock);
539
540         return 1;
541 }
542 #endif
543
544 void early_init_irq_lock_class(void)
545 {
546         struct irq_desc *desc;
547         int i;
548
549         for_each_irq_desc(i, desc) {
550                 lockdep_set_class(&desc->lock, &irq_desc_lock_class);
551         }
552 }
553
554 unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
555 {
556         struct irq_desc *desc = irq_to_desc(irq);
557         return desc ? desc->kstat_irqs[cpu] : 0;
558 }
559 EXPORT_SYMBOL(kstat_irqs_cpu);
560