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