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