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