xref: /linux/arch/powerpc/kernel/irq.c (revision d39d0ed196aa1685bb24771e92f78633c66ac9cb)
1 /*
2  *  Derived from arch/i386/kernel/irq.c
3  *    Copyright (C) 1992 Linus Torvalds
4  *  Adapted from arch/i386 by Gary Thomas
5  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
6  *  Updated and modified by Cort Dougan <cort@fsmlabs.com>
7  *    Copyright (C) 1996-2001 Cort Dougan
8  *  Adapted for Power Macintosh by Paul Mackerras
9  *    Copyright (C) 1996 Paul Mackerras (paulus@cs.anu.edu.au)
10  *
11  * This program is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU General Public License
13  * as published by the Free Software Foundation; either version
14  * 2 of the License, or (at your option) any later version.
15  *
16  * This file contains the code used by various IRQ handling routines:
17  * asking for different IRQ's should be done through these routines
18  * instead of just grabbing them. Thus setups with different IRQ numbers
19  * shouldn't result in any weird surprises, and installing new handlers
20  * should be easier.
21  *
22  * The MPC8xx has an interrupt mask in the SIU.  If a bit is set, the
23  * interrupt is _enabled_.  As expected, IRQ0 is bit 0 in the 32-bit
24  * mask register (of which only 16 are defined), hence the weird shifting
25  * and complement of the cached_irq_mask.  I want to be able to stuff
26  * this right into the SIU SMASK register.
27  * Many of the prep/chrp functions are conditional compiled on CONFIG_8xx
28  * to reduce code space and undefined function references.
29  */
30 
31 #undef DEBUG
32 
33 #include <linux/module.h>
34 #include <linux/threads.h>
35 #include <linux/kernel_stat.h>
36 #include <linux/signal.h>
37 #include <linux/sched.h>
38 #include <linux/ptrace.h>
39 #include <linux/ioport.h>
40 #include <linux/interrupt.h>
41 #include <linux/timex.h>
42 #include <linux/init.h>
43 #include <linux/slab.h>
44 #include <linux/delay.h>
45 #include <linux/irq.h>
46 #include <linux/seq_file.h>
47 #include <linux/cpumask.h>
48 #include <linux/profile.h>
49 #include <linux/bitops.h>
50 #include <linux/list.h>
51 #include <linux/radix-tree.h>
52 #include <linux/mutex.h>
53 #include <linux/bootmem.h>
54 #include <linux/pci.h>
55 #include <linux/debugfs.h>
56 #include <linux/of.h>
57 #include <linux/of_irq.h>
58 
59 #include <asm/uaccess.h>
60 #include <asm/system.h>
61 #include <asm/io.h>
62 #include <asm/pgtable.h>
63 #include <asm/irq.h>
64 #include <asm/cache.h>
65 #include <asm/prom.h>
66 #include <asm/ptrace.h>
67 #include <asm/machdep.h>
68 #include <asm/udbg.h>
69 #include <asm/dbell.h>
70 
71 #ifdef CONFIG_PPC64
72 #include <asm/paca.h>
73 #include <asm/firmware.h>
74 #include <asm/lv1call.h>
75 #endif
76 #define CREATE_TRACE_POINTS
77 #include <asm/trace.h>
78 
79 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
80 EXPORT_PER_CPU_SYMBOL(irq_stat);
81 
82 int __irq_offset_value;
83 
84 #ifdef CONFIG_PPC32
85 EXPORT_SYMBOL(__irq_offset_value);
86 atomic_t ppc_n_lost_interrupts;
87 
88 #ifdef CONFIG_TAU_INT
89 extern int tau_initialized;
90 extern int tau_interrupts(int);
91 #endif
92 #endif /* CONFIG_PPC32 */
93 
94 #ifdef CONFIG_PPC64
95 
96 #ifndef CONFIG_SPARSE_IRQ
97 EXPORT_SYMBOL(irq_desc);
98 #endif
99 
100 int distribute_irqs = 1;
101 
102 static inline notrace unsigned long get_hard_enabled(void)
103 {
104 	unsigned long enabled;
105 
106 	__asm__ __volatile__("lbz %0,%1(13)"
107 	: "=r" (enabled) : "i" (offsetof(struct paca_struct, hard_enabled)));
108 
109 	return enabled;
110 }
111 
112 static inline notrace void set_soft_enabled(unsigned long enable)
113 {
114 	__asm__ __volatile__("stb %0,%1(13)"
115 	: : "r" (enable), "i" (offsetof(struct paca_struct, soft_enabled)));
116 }
117 
118 notrace void raw_local_irq_restore(unsigned long en)
119 {
120 	/*
121 	 * get_paca()->soft_enabled = en;
122 	 * Is it ever valid to use local_irq_restore(0) when soft_enabled is 1?
123 	 * That was allowed before, and in such a case we do need to take care
124 	 * that gcc will set soft_enabled directly via r13, not choose to use
125 	 * an intermediate register, lest we're preempted to a different cpu.
126 	 */
127 	set_soft_enabled(en);
128 	if (!en)
129 		return;
130 
131 #ifdef CONFIG_PPC_STD_MMU_64
132 	if (firmware_has_feature(FW_FEATURE_ISERIES)) {
133 		/*
134 		 * Do we need to disable preemption here?  Not really: in the
135 		 * unlikely event that we're preempted to a different cpu in
136 		 * between getting r13, loading its lppaca_ptr, and loading
137 		 * its any_int, we might call iseries_handle_interrupts without
138 		 * an interrupt pending on the new cpu, but that's no disaster,
139 		 * is it?  And the business of preempting us off the old cpu
140 		 * would itself involve a local_irq_restore which handles the
141 		 * interrupt to that cpu.
142 		 *
143 		 * But use "local_paca->lppaca_ptr" instead of "get_lppaca()"
144 		 * to avoid any preemption checking added into get_paca().
145 		 */
146 		if (local_paca->lppaca_ptr->int_dword.any_int)
147 			iseries_handle_interrupts();
148 	}
149 #endif /* CONFIG_PPC_STD_MMU_64 */
150 
151 	/*
152 	 * if (get_paca()->hard_enabled) return;
153 	 * But again we need to take care that gcc gets hard_enabled directly
154 	 * via r13, not choose to use an intermediate register, lest we're
155 	 * preempted to a different cpu in between the two instructions.
156 	 */
157 	if (get_hard_enabled())
158 		return;
159 
160 #if defined(CONFIG_BOOKE) && defined(CONFIG_SMP)
161 	/* Check for pending doorbell interrupts and resend to ourself */
162 	doorbell_check_self();
163 #endif
164 
165 	/*
166 	 * Need to hard-enable interrupts here.  Since currently disabled,
167 	 * no need to take further asm precautions against preemption; but
168 	 * use local_paca instead of get_paca() to avoid preemption checking.
169 	 */
170 	local_paca->hard_enabled = en;
171 
172 #ifndef CONFIG_BOOKE
173 	/* On server, re-trigger the decrementer if it went negative since
174 	 * some processors only trigger on edge transitions of the sign bit.
175 	 *
176 	 * BookE has a level sensitive decrementer (latches in TSR) so we
177 	 * don't need that
178 	 */
179 	if ((int)mfspr(SPRN_DEC) < 0)
180 		mtspr(SPRN_DEC, 1);
181 #endif /* CONFIG_BOOKE */
182 
183 	/*
184 	 * Force the delivery of pending soft-disabled interrupts on PS3.
185 	 * Any HV call will have this side effect.
186 	 */
187 	if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
188 		u64 tmp;
189 		lv1_get_version_info(&tmp);
190 	}
191 
192 	__hard_irq_enable();
193 }
194 EXPORT_SYMBOL(raw_local_irq_restore);
195 #endif /* CONFIG_PPC64 */
196 
197 static int show_other_interrupts(struct seq_file *p, int prec)
198 {
199 	int j;
200 
201 #if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
202 	if (tau_initialized) {
203 		seq_printf(p, "%*s: ", prec, "TAU");
204 		for_each_online_cpu(j)
205 			seq_printf(p, "%10u ", tau_interrupts(j));
206 		seq_puts(p, "  PowerPC             Thermal Assist (cpu temp)\n");
207 	}
208 #endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
209 
210 	seq_printf(p, "%*s: ", prec, "LOC");
211 	for_each_online_cpu(j)
212 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs);
213         seq_printf(p, "  Local timer interrupts\n");
214 
215 	seq_printf(p, "%*s: ", prec, "SPU");
216 	for_each_online_cpu(j)
217 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
218 	seq_printf(p, "  Spurious interrupts\n");
219 
220 	seq_printf(p, "%*s: ", prec, "CNT");
221 	for_each_online_cpu(j)
222 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
223 	seq_printf(p, "  Performance monitoring interrupts\n");
224 
225 	seq_printf(p, "%*s: ", prec, "MCE");
226 	for_each_online_cpu(j)
227 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
228 	seq_printf(p, "  Machine check exceptions\n");
229 
230 	return 0;
231 }
232 
233 int show_interrupts(struct seq_file *p, void *v)
234 {
235 	unsigned long flags, any_count = 0;
236 	int i = *(loff_t *) v, j, prec;
237 	struct irqaction *action;
238 	struct irq_desc *desc;
239 
240 	if (i > nr_irqs)
241 		return 0;
242 
243 	for (prec = 3, j = 1000; prec < 10 && j <= nr_irqs; ++prec)
244 		j *= 10;
245 
246 	if (i == nr_irqs)
247 		return show_other_interrupts(p, prec);
248 
249 	/* print header */
250 	if (i == 0) {
251 		seq_printf(p, "%*s", prec + 8, "");
252 		for_each_online_cpu(j)
253 			seq_printf(p, "CPU%-8d", j);
254 		seq_putc(p, '\n');
255 	}
256 
257 	desc = irq_to_desc(i);
258 	if (!desc)
259 		return 0;
260 
261 	raw_spin_lock_irqsave(&desc->lock, flags);
262 	for_each_online_cpu(j)
263 		any_count |= kstat_irqs_cpu(i, j);
264 	action = desc->action;
265 	if (!action && !any_count)
266 		goto out;
267 
268 	seq_printf(p, "%*d: ", prec, i);
269 	for_each_online_cpu(j)
270 		seq_printf(p, "%10u ", kstat_irqs_cpu(i, j));
271 
272 	if (desc->chip)
273 		seq_printf(p, "  %-16s", desc->chip->name);
274 	else
275 		seq_printf(p, "  %-16s", "None");
276 	seq_printf(p, " %-8s", (desc->status & IRQ_LEVEL) ? "Level" : "Edge");
277 
278 	if (action) {
279 		seq_printf(p, "     %s", action->name);
280 		while ((action = action->next) != NULL)
281 			seq_printf(p, ", %s", action->name);
282 	}
283 
284 	seq_putc(p, '\n');
285 out:
286 	raw_spin_unlock_irqrestore(&desc->lock, flags);
287 	return 0;
288 }
289 
290 /*
291  * /proc/stat helpers
292  */
293 u64 arch_irq_stat_cpu(unsigned int cpu)
294 {
295 	u64 sum = per_cpu(irq_stat, cpu).timer_irqs;
296 
297 	sum += per_cpu(irq_stat, cpu).pmu_irqs;
298 	sum += per_cpu(irq_stat, cpu).mce_exceptions;
299 	sum += per_cpu(irq_stat, cpu).spurious_irqs;
300 
301 	return sum;
302 }
303 
304 #ifdef CONFIG_HOTPLUG_CPU
305 void fixup_irqs(const struct cpumask *map)
306 {
307 	struct irq_desc *desc;
308 	unsigned int irq;
309 	static int warned;
310 	cpumask_var_t mask;
311 
312 	alloc_cpumask_var(&mask, GFP_KERNEL);
313 
314 	for_each_irq(irq) {
315 		desc = irq_to_desc(irq);
316 		if (!desc)
317 			continue;
318 
319 		if (desc->status & IRQ_PER_CPU)
320 			continue;
321 
322 		cpumask_and(mask, desc->affinity, map);
323 		if (cpumask_any(mask) >= nr_cpu_ids) {
324 			printk("Breaking affinity for irq %i\n", irq);
325 			cpumask_copy(mask, map);
326 		}
327 		if (desc->chip->set_affinity)
328 			desc->chip->set_affinity(irq, mask);
329 		else if (desc->action && !(warned++))
330 			printk("Cannot set affinity for irq %i\n", irq);
331 	}
332 
333 	free_cpumask_var(mask);
334 
335 	local_irq_enable();
336 	mdelay(1);
337 	local_irq_disable();
338 }
339 #endif
340 
341 static inline void handle_one_irq(unsigned int irq)
342 {
343 	struct thread_info *curtp, *irqtp;
344 	unsigned long saved_sp_limit;
345 	struct irq_desc *desc;
346 
347 	/* Switch to the irq stack to handle this */
348 	curtp = current_thread_info();
349 	irqtp = hardirq_ctx[smp_processor_id()];
350 
351 	if (curtp == irqtp) {
352 		/* We're already on the irq stack, just handle it */
353 		generic_handle_irq(irq);
354 		return;
355 	}
356 
357 	desc = irq_to_desc(irq);
358 	saved_sp_limit = current->thread.ksp_limit;
359 
360 	irqtp->task = curtp->task;
361 	irqtp->flags = 0;
362 
363 	/* Copy the softirq bits in preempt_count so that the
364 	 * softirq checks work in the hardirq context. */
365 	irqtp->preempt_count = (irqtp->preempt_count & ~SOFTIRQ_MASK) |
366 			       (curtp->preempt_count & SOFTIRQ_MASK);
367 
368 	current->thread.ksp_limit = (unsigned long)irqtp +
369 		_ALIGN_UP(sizeof(struct thread_info), 16);
370 
371 	call_handle_irq(irq, desc, irqtp, desc->handle_irq);
372 	current->thread.ksp_limit = saved_sp_limit;
373 	irqtp->task = NULL;
374 
375 	/* Set any flag that may have been set on the
376 	 * alternate stack
377 	 */
378 	if (irqtp->flags)
379 		set_bits(irqtp->flags, &curtp->flags);
380 }
381 
382 static inline void check_stack_overflow(void)
383 {
384 #ifdef CONFIG_DEBUG_STACKOVERFLOW
385 	long sp;
386 
387 	sp = __get_SP() & (THREAD_SIZE-1);
388 
389 	/* check for stack overflow: is there less than 2KB free? */
390 	if (unlikely(sp < (sizeof(struct thread_info) + 2048))) {
391 		printk("do_IRQ: stack overflow: %ld\n",
392 			sp - sizeof(struct thread_info));
393 		dump_stack();
394 	}
395 #endif
396 }
397 
398 void do_IRQ(struct pt_regs *regs)
399 {
400 	struct pt_regs *old_regs = set_irq_regs(regs);
401 	unsigned int irq;
402 
403 	trace_irq_entry(regs);
404 
405 	irq_enter();
406 
407 	check_stack_overflow();
408 
409 	irq = ppc_md.get_irq();
410 
411 	if (irq != NO_IRQ && irq != NO_IRQ_IGNORE)
412 		handle_one_irq(irq);
413 	else if (irq != NO_IRQ_IGNORE)
414 		__get_cpu_var(irq_stat).spurious_irqs++;
415 
416 	irq_exit();
417 	set_irq_regs(old_regs);
418 
419 #ifdef CONFIG_PPC_ISERIES
420 	if (firmware_has_feature(FW_FEATURE_ISERIES) &&
421 			get_lppaca()->int_dword.fields.decr_int) {
422 		get_lppaca()->int_dword.fields.decr_int = 0;
423 		/* Signal a fake decrementer interrupt */
424 		timer_interrupt(regs);
425 	}
426 #endif
427 
428 	trace_irq_exit(regs);
429 }
430 
431 void __init init_IRQ(void)
432 {
433 	if (ppc_md.init_IRQ)
434 		ppc_md.init_IRQ();
435 
436 	exc_lvl_ctx_init();
437 
438 	irq_ctx_init();
439 }
440 
441 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
442 struct thread_info   *critirq_ctx[NR_CPUS] __read_mostly;
443 struct thread_info    *dbgirq_ctx[NR_CPUS] __read_mostly;
444 struct thread_info *mcheckirq_ctx[NR_CPUS] __read_mostly;
445 
446 void exc_lvl_ctx_init(void)
447 {
448 	struct thread_info *tp;
449 	int i;
450 
451 	for_each_possible_cpu(i) {
452 		memset((void *)critirq_ctx[i], 0, THREAD_SIZE);
453 		tp = critirq_ctx[i];
454 		tp->cpu = i;
455 		tp->preempt_count = 0;
456 
457 #ifdef CONFIG_BOOKE
458 		memset((void *)dbgirq_ctx[i], 0, THREAD_SIZE);
459 		tp = dbgirq_ctx[i];
460 		tp->cpu = i;
461 		tp->preempt_count = 0;
462 
463 		memset((void *)mcheckirq_ctx[i], 0, THREAD_SIZE);
464 		tp = mcheckirq_ctx[i];
465 		tp->cpu = i;
466 		tp->preempt_count = HARDIRQ_OFFSET;
467 #endif
468 	}
469 }
470 #endif
471 
472 struct thread_info *softirq_ctx[NR_CPUS] __read_mostly;
473 struct thread_info *hardirq_ctx[NR_CPUS] __read_mostly;
474 
475 void irq_ctx_init(void)
476 {
477 	struct thread_info *tp;
478 	int i;
479 
480 	for_each_possible_cpu(i) {
481 		memset((void *)softirq_ctx[i], 0, THREAD_SIZE);
482 		tp = softirq_ctx[i];
483 		tp->cpu = i;
484 		tp->preempt_count = 0;
485 
486 		memset((void *)hardirq_ctx[i], 0, THREAD_SIZE);
487 		tp = hardirq_ctx[i];
488 		tp->cpu = i;
489 		tp->preempt_count = HARDIRQ_OFFSET;
490 	}
491 }
492 
493 static inline void do_softirq_onstack(void)
494 {
495 	struct thread_info *curtp, *irqtp;
496 	unsigned long saved_sp_limit = current->thread.ksp_limit;
497 
498 	curtp = current_thread_info();
499 	irqtp = softirq_ctx[smp_processor_id()];
500 	irqtp->task = curtp->task;
501 	current->thread.ksp_limit = (unsigned long)irqtp +
502 				    _ALIGN_UP(sizeof(struct thread_info), 16);
503 	call_do_softirq(irqtp);
504 	current->thread.ksp_limit = saved_sp_limit;
505 	irqtp->task = NULL;
506 }
507 
508 void do_softirq(void)
509 {
510 	unsigned long flags;
511 
512 	if (in_interrupt())
513 		return;
514 
515 	local_irq_save(flags);
516 
517 	if (local_softirq_pending())
518 		do_softirq_onstack();
519 
520 	local_irq_restore(flags);
521 }
522 
523 
524 /*
525  * IRQ controller and virtual interrupts
526  */
527 
528 static LIST_HEAD(irq_hosts);
529 static DEFINE_RAW_SPINLOCK(irq_big_lock);
530 static unsigned int revmap_trees_allocated;
531 static DEFINE_MUTEX(revmap_trees_mutex);
532 struct irq_map_entry irq_map[NR_IRQS];
533 static unsigned int irq_virq_count = NR_IRQS;
534 static struct irq_host *irq_default_host;
535 
536 irq_hw_number_t virq_to_hw(unsigned int virq)
537 {
538 	return irq_map[virq].hwirq;
539 }
540 EXPORT_SYMBOL_GPL(virq_to_hw);
541 
542 static int default_irq_host_match(struct irq_host *h, struct device_node *np)
543 {
544 	return h->of_node != NULL && h->of_node == np;
545 }
546 
547 struct irq_host *irq_alloc_host(struct device_node *of_node,
548 				unsigned int revmap_type,
549 				unsigned int revmap_arg,
550 				struct irq_host_ops *ops,
551 				irq_hw_number_t inval_irq)
552 {
553 	struct irq_host *host;
554 	unsigned int size = sizeof(struct irq_host);
555 	unsigned int i;
556 	unsigned int *rmap;
557 	unsigned long flags;
558 
559 	/* Allocate structure and revmap table if using linear mapping */
560 	if (revmap_type == IRQ_HOST_MAP_LINEAR)
561 		size += revmap_arg * sizeof(unsigned int);
562 	host = zalloc_maybe_bootmem(size, GFP_KERNEL);
563 	if (host == NULL)
564 		return NULL;
565 
566 	/* Fill structure */
567 	host->revmap_type = revmap_type;
568 	host->inval_irq = inval_irq;
569 	host->ops = ops;
570 	host->of_node = of_node_get(of_node);
571 
572 	if (host->ops->match == NULL)
573 		host->ops->match = default_irq_host_match;
574 
575 	raw_spin_lock_irqsave(&irq_big_lock, flags);
576 
577 	/* If it's a legacy controller, check for duplicates and
578 	 * mark it as allocated (we use irq 0 host pointer for that
579 	 */
580 	if (revmap_type == IRQ_HOST_MAP_LEGACY) {
581 		if (irq_map[0].host != NULL) {
582 			raw_spin_unlock_irqrestore(&irq_big_lock, flags);
583 			/* If we are early boot, we can't free the structure,
584 			 * too bad...
585 			 * this will be fixed once slab is made available early
586 			 * instead of the current cruft
587 			 */
588 			if (mem_init_done)
589 				kfree(host);
590 			return NULL;
591 		}
592 		irq_map[0].host = host;
593 	}
594 
595 	list_add(&host->link, &irq_hosts);
596 	raw_spin_unlock_irqrestore(&irq_big_lock, flags);
597 
598 	/* Additional setups per revmap type */
599 	switch(revmap_type) {
600 	case IRQ_HOST_MAP_LEGACY:
601 		/* 0 is always the invalid number for legacy */
602 		host->inval_irq = 0;
603 		/* setup us as the host for all legacy interrupts */
604 		for (i = 1; i < NUM_ISA_INTERRUPTS; i++) {
605 			irq_map[i].hwirq = i;
606 			smp_wmb();
607 			irq_map[i].host = host;
608 			smp_wmb();
609 
610 			/* Clear norequest flags */
611 			irq_to_desc(i)->status &= ~IRQ_NOREQUEST;
612 
613 			/* Legacy flags are left to default at this point,
614 			 * one can then use irq_create_mapping() to
615 			 * explicitly change them
616 			 */
617 			ops->map(host, i, i);
618 		}
619 		break;
620 	case IRQ_HOST_MAP_LINEAR:
621 		rmap = (unsigned int *)(host + 1);
622 		for (i = 0; i < revmap_arg; i++)
623 			rmap[i] = NO_IRQ;
624 		host->revmap_data.linear.size = revmap_arg;
625 		smp_wmb();
626 		host->revmap_data.linear.revmap = rmap;
627 		break;
628 	default:
629 		break;
630 	}
631 
632 	pr_debug("irq: Allocated host of type %d @0x%p\n", revmap_type, host);
633 
634 	return host;
635 }
636 
637 struct irq_host *irq_find_host(struct device_node *node)
638 {
639 	struct irq_host *h, *found = NULL;
640 	unsigned long flags;
641 
642 	/* We might want to match the legacy controller last since
643 	 * it might potentially be set to match all interrupts in
644 	 * the absence of a device node. This isn't a problem so far
645 	 * yet though...
646 	 */
647 	raw_spin_lock_irqsave(&irq_big_lock, flags);
648 	list_for_each_entry(h, &irq_hosts, link)
649 		if (h->ops->match(h, node)) {
650 			found = h;
651 			break;
652 		}
653 	raw_spin_unlock_irqrestore(&irq_big_lock, flags);
654 	return found;
655 }
656 EXPORT_SYMBOL_GPL(irq_find_host);
657 
658 void irq_set_default_host(struct irq_host *host)
659 {
660 	pr_debug("irq: Default host set to @0x%p\n", host);
661 
662 	irq_default_host = host;
663 }
664 
665 void irq_set_virq_count(unsigned int count)
666 {
667 	pr_debug("irq: Trying to set virq count to %d\n", count);
668 
669 	BUG_ON(count < NUM_ISA_INTERRUPTS);
670 	if (count < NR_IRQS)
671 		irq_virq_count = count;
672 }
673 
674 static int irq_setup_virq(struct irq_host *host, unsigned int virq,
675 			    irq_hw_number_t hwirq)
676 {
677 	struct irq_desc *desc;
678 
679 	desc = irq_to_desc_alloc_node(virq, 0);
680 	if (!desc) {
681 		pr_debug("irq: -> allocating desc failed\n");
682 		goto error;
683 	}
684 
685 	/* Clear IRQ_NOREQUEST flag */
686 	desc->status &= ~IRQ_NOREQUEST;
687 
688 	/* map it */
689 	smp_wmb();
690 	irq_map[virq].hwirq = hwirq;
691 	smp_mb();
692 
693 	if (host->ops->map(host, virq, hwirq)) {
694 		pr_debug("irq: -> mapping failed, freeing\n");
695 		goto error;
696 	}
697 
698 	return 0;
699 
700 error:
701 	irq_free_virt(virq, 1);
702 	return -1;
703 }
704 
705 unsigned int irq_create_direct_mapping(struct irq_host *host)
706 {
707 	unsigned int virq;
708 
709 	if (host == NULL)
710 		host = irq_default_host;
711 
712 	BUG_ON(host == NULL);
713 	WARN_ON(host->revmap_type != IRQ_HOST_MAP_NOMAP);
714 
715 	virq = irq_alloc_virt(host, 1, 0);
716 	if (virq == NO_IRQ) {
717 		pr_debug("irq: create_direct virq allocation failed\n");
718 		return NO_IRQ;
719 	}
720 
721 	pr_debug("irq: create_direct obtained virq %d\n", virq);
722 
723 	if (irq_setup_virq(host, virq, virq))
724 		return NO_IRQ;
725 
726 	return virq;
727 }
728 
729 unsigned int irq_create_mapping(struct irq_host *host,
730 				irq_hw_number_t hwirq)
731 {
732 	unsigned int virq, hint;
733 
734 	pr_debug("irq: irq_create_mapping(0x%p, 0x%lx)\n", host, hwirq);
735 
736 	/* Look for default host if nececssary */
737 	if (host == NULL)
738 		host = irq_default_host;
739 	if (host == NULL) {
740 		printk(KERN_WARNING "irq_create_mapping called for"
741 		       " NULL host, hwirq=%lx\n", hwirq);
742 		WARN_ON(1);
743 		return NO_IRQ;
744 	}
745 	pr_debug("irq: -> using host @%p\n", host);
746 
747 	/* Check if mapping already exist, if it does, call
748 	 * host->ops->map() to update the flags
749 	 */
750 	virq = irq_find_mapping(host, hwirq);
751 	if (virq != NO_IRQ) {
752 		if (host->ops->remap)
753 			host->ops->remap(host, virq, hwirq);
754 		pr_debug("irq: -> existing mapping on virq %d\n", virq);
755 		return virq;
756 	}
757 
758 	/* Get a virtual interrupt number */
759 	if (host->revmap_type == IRQ_HOST_MAP_LEGACY) {
760 		/* Handle legacy */
761 		virq = (unsigned int)hwirq;
762 		if (virq == 0 || virq >= NUM_ISA_INTERRUPTS)
763 			return NO_IRQ;
764 		return virq;
765 	} else {
766 		/* Allocate a virtual interrupt number */
767 		hint = hwirq % irq_virq_count;
768 		virq = irq_alloc_virt(host, 1, hint);
769 		if (virq == NO_IRQ) {
770 			pr_debug("irq: -> virq allocation failed\n");
771 			return NO_IRQ;
772 		}
773 	}
774 
775 	if (irq_setup_virq(host, virq, hwirq))
776 		return NO_IRQ;
777 
778 	printk(KERN_DEBUG "irq: irq %lu on host %s mapped to virtual irq %u\n",
779 		hwirq, host->of_node ? host->of_node->full_name : "null", virq);
780 
781 	return virq;
782 }
783 EXPORT_SYMBOL_GPL(irq_create_mapping);
784 
785 unsigned int irq_create_of_mapping(struct device_node *controller,
786 				   const u32 *intspec, unsigned int intsize)
787 {
788 	struct irq_host *host;
789 	irq_hw_number_t hwirq;
790 	unsigned int type = IRQ_TYPE_NONE;
791 	unsigned int virq;
792 
793 	if (controller == NULL)
794 		host = irq_default_host;
795 	else
796 		host = irq_find_host(controller);
797 	if (host == NULL) {
798 		printk(KERN_WARNING "irq: no irq host found for %s !\n",
799 		       controller->full_name);
800 		return NO_IRQ;
801 	}
802 
803 	/* If host has no translation, then we assume interrupt line */
804 	if (host->ops->xlate == NULL)
805 		hwirq = intspec[0];
806 	else {
807 		if (host->ops->xlate(host, controller, intspec, intsize,
808 				     &hwirq, &type))
809 			return NO_IRQ;
810 	}
811 
812 	/* Create mapping */
813 	virq = irq_create_mapping(host, hwirq);
814 	if (virq == NO_IRQ)
815 		return virq;
816 
817 	/* Set type if specified and different than the current one */
818 	if (type != IRQ_TYPE_NONE &&
819 	    type != (irq_to_desc(virq)->status & IRQF_TRIGGER_MASK))
820 		set_irq_type(virq, type);
821 	return virq;
822 }
823 EXPORT_SYMBOL_GPL(irq_create_of_mapping);
824 
825 void irq_dispose_mapping(unsigned int virq)
826 {
827 	struct irq_host *host;
828 	irq_hw_number_t hwirq;
829 
830 	if (virq == NO_IRQ)
831 		return;
832 
833 	host = irq_map[virq].host;
834 	WARN_ON (host == NULL);
835 	if (host == NULL)
836 		return;
837 
838 	/* Never unmap legacy interrupts */
839 	if (host->revmap_type == IRQ_HOST_MAP_LEGACY)
840 		return;
841 
842 	/* remove chip and handler */
843 	set_irq_chip_and_handler(virq, NULL, NULL);
844 
845 	/* Make sure it's completed */
846 	synchronize_irq(virq);
847 
848 	/* Tell the PIC about it */
849 	if (host->ops->unmap)
850 		host->ops->unmap(host, virq);
851 	smp_mb();
852 
853 	/* Clear reverse map */
854 	hwirq = irq_map[virq].hwirq;
855 	switch(host->revmap_type) {
856 	case IRQ_HOST_MAP_LINEAR:
857 		if (hwirq < host->revmap_data.linear.size)
858 			host->revmap_data.linear.revmap[hwirq] = NO_IRQ;
859 		break;
860 	case IRQ_HOST_MAP_TREE:
861 		/*
862 		 * Check if radix tree allocated yet, if not then nothing to
863 		 * remove.
864 		 */
865 		smp_rmb();
866 		if (revmap_trees_allocated < 1)
867 			break;
868 		mutex_lock(&revmap_trees_mutex);
869 		radix_tree_delete(&host->revmap_data.tree, hwirq);
870 		mutex_unlock(&revmap_trees_mutex);
871 		break;
872 	}
873 
874 	/* Destroy map */
875 	smp_mb();
876 	irq_map[virq].hwirq = host->inval_irq;
877 
878 	/* Set some flags */
879 	irq_to_desc(virq)->status |= IRQ_NOREQUEST;
880 
881 	/* Free it */
882 	irq_free_virt(virq, 1);
883 }
884 EXPORT_SYMBOL_GPL(irq_dispose_mapping);
885 
886 unsigned int irq_find_mapping(struct irq_host *host,
887 			      irq_hw_number_t hwirq)
888 {
889 	unsigned int i;
890 	unsigned int hint = hwirq % irq_virq_count;
891 
892 	/* Look for default host if nececssary */
893 	if (host == NULL)
894 		host = irq_default_host;
895 	if (host == NULL)
896 		return NO_IRQ;
897 
898 	/* legacy -> bail early */
899 	if (host->revmap_type == IRQ_HOST_MAP_LEGACY)
900 		return hwirq;
901 
902 	/* Slow path does a linear search of the map */
903 	if (hint < NUM_ISA_INTERRUPTS)
904 		hint = NUM_ISA_INTERRUPTS;
905 	i = hint;
906 	do  {
907 		if (irq_map[i].host == host &&
908 		    irq_map[i].hwirq == hwirq)
909 			return i;
910 		i++;
911 		if (i >= irq_virq_count)
912 			i = NUM_ISA_INTERRUPTS;
913 	} while(i != hint);
914 	return NO_IRQ;
915 }
916 EXPORT_SYMBOL_GPL(irq_find_mapping);
917 
918 
919 unsigned int irq_radix_revmap_lookup(struct irq_host *host,
920 				     irq_hw_number_t hwirq)
921 {
922 	struct irq_map_entry *ptr;
923 	unsigned int virq;
924 
925 	WARN_ON(host->revmap_type != IRQ_HOST_MAP_TREE);
926 
927 	/*
928 	 * Check if the radix tree exists and has bee initialized.
929 	 * If not, we fallback to slow mode
930 	 */
931 	if (revmap_trees_allocated < 2)
932 		return irq_find_mapping(host, hwirq);
933 
934 	/* Now try to resolve */
935 	/*
936 	 * No rcu_read_lock(ing) needed, the ptr returned can't go under us
937 	 * as it's referencing an entry in the static irq_map table.
938 	 */
939 	ptr = radix_tree_lookup(&host->revmap_data.tree, hwirq);
940 
941 	/*
942 	 * If found in radix tree, then fine.
943 	 * Else fallback to linear lookup - this should not happen in practice
944 	 * as it means that we failed to insert the node in the radix tree.
945 	 */
946 	if (ptr)
947 		virq = ptr - irq_map;
948 	else
949 		virq = irq_find_mapping(host, hwirq);
950 
951 	return virq;
952 }
953 
954 void irq_radix_revmap_insert(struct irq_host *host, unsigned int virq,
955 			     irq_hw_number_t hwirq)
956 {
957 
958 	WARN_ON(host->revmap_type != IRQ_HOST_MAP_TREE);
959 
960 	/*
961 	 * Check if the radix tree exists yet.
962 	 * If not, then the irq will be inserted into the tree when it gets
963 	 * initialized.
964 	 */
965 	smp_rmb();
966 	if (revmap_trees_allocated < 1)
967 		return;
968 
969 	if (virq != NO_IRQ) {
970 		mutex_lock(&revmap_trees_mutex);
971 		radix_tree_insert(&host->revmap_data.tree, hwirq,
972 				  &irq_map[virq]);
973 		mutex_unlock(&revmap_trees_mutex);
974 	}
975 }
976 
977 unsigned int irq_linear_revmap(struct irq_host *host,
978 			       irq_hw_number_t hwirq)
979 {
980 	unsigned int *revmap;
981 
982 	WARN_ON(host->revmap_type != IRQ_HOST_MAP_LINEAR);
983 
984 	/* Check revmap bounds */
985 	if (unlikely(hwirq >= host->revmap_data.linear.size))
986 		return irq_find_mapping(host, hwirq);
987 
988 	/* Check if revmap was allocated */
989 	revmap = host->revmap_data.linear.revmap;
990 	if (unlikely(revmap == NULL))
991 		return irq_find_mapping(host, hwirq);
992 
993 	/* Fill up revmap with slow path if no mapping found */
994 	if (unlikely(revmap[hwirq] == NO_IRQ))
995 		revmap[hwirq] = irq_find_mapping(host, hwirq);
996 
997 	return revmap[hwirq];
998 }
999 
1000 unsigned int irq_alloc_virt(struct irq_host *host,
1001 			    unsigned int count,
1002 			    unsigned int hint)
1003 {
1004 	unsigned long flags;
1005 	unsigned int i, j, found = NO_IRQ;
1006 
1007 	if (count == 0 || count > (irq_virq_count - NUM_ISA_INTERRUPTS))
1008 		return NO_IRQ;
1009 
1010 	raw_spin_lock_irqsave(&irq_big_lock, flags);
1011 
1012 	/* Use hint for 1 interrupt if any */
1013 	if (count == 1 && hint >= NUM_ISA_INTERRUPTS &&
1014 	    hint < irq_virq_count && irq_map[hint].host == NULL) {
1015 		found = hint;
1016 		goto hint_found;
1017 	}
1018 
1019 	/* Look for count consecutive numbers in the allocatable
1020 	 * (non-legacy) space
1021 	 */
1022 	for (i = NUM_ISA_INTERRUPTS, j = 0; i < irq_virq_count; i++) {
1023 		if (irq_map[i].host != NULL)
1024 			j = 0;
1025 		else
1026 			j++;
1027 
1028 		if (j == count) {
1029 			found = i - count + 1;
1030 			break;
1031 		}
1032 	}
1033 	if (found == NO_IRQ) {
1034 		raw_spin_unlock_irqrestore(&irq_big_lock, flags);
1035 		return NO_IRQ;
1036 	}
1037  hint_found:
1038 	for (i = found; i < (found + count); i++) {
1039 		irq_map[i].hwirq = host->inval_irq;
1040 		smp_wmb();
1041 		irq_map[i].host = host;
1042 	}
1043 	raw_spin_unlock_irqrestore(&irq_big_lock, flags);
1044 	return found;
1045 }
1046 
1047 void irq_free_virt(unsigned int virq, unsigned int count)
1048 {
1049 	unsigned long flags;
1050 	unsigned int i;
1051 
1052 	WARN_ON (virq < NUM_ISA_INTERRUPTS);
1053 	WARN_ON (count == 0 || (virq + count) > irq_virq_count);
1054 
1055 	raw_spin_lock_irqsave(&irq_big_lock, flags);
1056 	for (i = virq; i < (virq + count); i++) {
1057 		struct irq_host *host;
1058 
1059 		if (i < NUM_ISA_INTERRUPTS ||
1060 		    (virq + count) > irq_virq_count)
1061 			continue;
1062 
1063 		host = irq_map[i].host;
1064 		irq_map[i].hwirq = host->inval_irq;
1065 		smp_wmb();
1066 		irq_map[i].host = NULL;
1067 	}
1068 	raw_spin_unlock_irqrestore(&irq_big_lock, flags);
1069 }
1070 
1071 int arch_early_irq_init(void)
1072 {
1073 	struct irq_desc *desc;
1074 	int i;
1075 
1076 	for (i = 0; i < NR_IRQS; i++) {
1077 		desc = irq_to_desc(i);
1078 		if (desc)
1079 			desc->status |= IRQ_NOREQUEST;
1080 	}
1081 
1082 	return 0;
1083 }
1084 
1085 int arch_init_chip_data(struct irq_desc *desc, int node)
1086 {
1087 	desc->status |= IRQ_NOREQUEST;
1088 	return 0;
1089 }
1090 
1091 /* We need to create the radix trees late */
1092 static int irq_late_init(void)
1093 {
1094 	struct irq_host *h;
1095 	unsigned int i;
1096 
1097 	/*
1098 	 * No mutual exclusion with respect to accessors of the tree is needed
1099 	 * here as the synchronization is done via the state variable
1100 	 * revmap_trees_allocated.
1101 	 */
1102 	list_for_each_entry(h, &irq_hosts, link) {
1103 		if (h->revmap_type == IRQ_HOST_MAP_TREE)
1104 			INIT_RADIX_TREE(&h->revmap_data.tree, GFP_KERNEL);
1105 	}
1106 
1107 	/*
1108 	 * Make sure the radix trees inits are visible before setting
1109 	 * the flag
1110 	 */
1111 	smp_wmb();
1112 	revmap_trees_allocated = 1;
1113 
1114 	/*
1115 	 * Insert the reverse mapping for those interrupts already present
1116 	 * in irq_map[].
1117 	 */
1118 	mutex_lock(&revmap_trees_mutex);
1119 	for (i = 0; i < irq_virq_count; i++) {
1120 		if (irq_map[i].host &&
1121 		    (irq_map[i].host->revmap_type == IRQ_HOST_MAP_TREE))
1122 			radix_tree_insert(&irq_map[i].host->revmap_data.tree,
1123 					  irq_map[i].hwirq, &irq_map[i]);
1124 	}
1125 	mutex_unlock(&revmap_trees_mutex);
1126 
1127 	/*
1128 	 * Make sure the radix trees insertions are visible before setting
1129 	 * the flag
1130 	 */
1131 	smp_wmb();
1132 	revmap_trees_allocated = 2;
1133 
1134 	return 0;
1135 }
1136 arch_initcall(irq_late_init);
1137 
1138 #ifdef CONFIG_VIRQ_DEBUG
1139 static int virq_debug_show(struct seq_file *m, void *private)
1140 {
1141 	unsigned long flags;
1142 	struct irq_desc *desc;
1143 	const char *p;
1144 	char none[] = "none";
1145 	int i;
1146 
1147 	seq_printf(m, "%-5s  %-7s  %-15s  %s\n", "virq", "hwirq",
1148 		      "chip name", "host name");
1149 
1150 	for (i = 1; i < nr_irqs; i++) {
1151 		desc = irq_to_desc(i);
1152 		if (!desc)
1153 			continue;
1154 
1155 		raw_spin_lock_irqsave(&desc->lock, flags);
1156 
1157 		if (desc->action && desc->action->handler) {
1158 			seq_printf(m, "%5d  ", i);
1159 			seq_printf(m, "0x%05lx  ", virq_to_hw(i));
1160 
1161 			if (desc->chip && desc->chip->name)
1162 				p = desc->chip->name;
1163 			else
1164 				p = none;
1165 			seq_printf(m, "%-15s  ", p);
1166 
1167 			if (irq_map[i].host && irq_map[i].host->of_node)
1168 				p = irq_map[i].host->of_node->full_name;
1169 			else
1170 				p = none;
1171 			seq_printf(m, "%s\n", p);
1172 		}
1173 
1174 		raw_spin_unlock_irqrestore(&desc->lock, flags);
1175 	}
1176 
1177 	return 0;
1178 }
1179 
1180 static int virq_debug_open(struct inode *inode, struct file *file)
1181 {
1182 	return single_open(file, virq_debug_show, inode->i_private);
1183 }
1184 
1185 static const struct file_operations virq_debug_fops = {
1186 	.open = virq_debug_open,
1187 	.read = seq_read,
1188 	.llseek = seq_lseek,
1189 	.release = single_release,
1190 };
1191 
1192 static int __init irq_debugfs_init(void)
1193 {
1194 	if (debugfs_create_file("virq_mapping", S_IRUGO, powerpc_debugfs_root,
1195 				 NULL, &virq_debug_fops) == NULL)
1196 		return -ENOMEM;
1197 
1198 	return 0;
1199 }
1200 __initcall(irq_debugfs_init);
1201 #endif /* CONFIG_VIRQ_DEBUG */
1202 
1203 #ifdef CONFIG_PPC64
1204 static int __init setup_noirqdistrib(char *str)
1205 {
1206 	distribute_irqs = 0;
1207 	return 1;
1208 }
1209 
1210 __setup("noirqdistrib", setup_noirqdistrib);
1211 #endif /* CONFIG_PPC64 */
1212