xref: /openbmc/linux/arch/powerpc/kernel/irq.c (revision 476eb4912601a8c01e6702b9a029f476b4b131d2)
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 #include <asm/smp.h>
71 
72 #ifdef CONFIG_PPC64
73 #include <asm/paca.h>
74 #include <asm/firmware.h>
75 #include <asm/lv1call.h>
76 #endif
77 #define CREATE_TRACE_POINTS
78 #include <asm/trace.h>
79 
80 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
81 EXPORT_PER_CPU_SYMBOL(irq_stat);
82 
83 int __irq_offset_value;
84 
85 #ifdef CONFIG_PPC32
86 EXPORT_SYMBOL(__irq_offset_value);
87 atomic_t ppc_n_lost_interrupts;
88 
89 #ifdef CONFIG_TAU_INT
90 extern int tau_initialized;
91 extern int tau_interrupts(int);
92 #endif
93 #endif /* CONFIG_PPC32 */
94 
95 #ifdef CONFIG_PPC64
96 
97 #ifndef CONFIG_SPARSE_IRQ
98 EXPORT_SYMBOL(irq_desc);
99 #endif
100 
101 int distribute_irqs = 1;
102 
103 static inline notrace unsigned long get_hard_enabled(void)
104 {
105 	unsigned long enabled;
106 
107 	__asm__ __volatile__("lbz %0,%1(13)"
108 	: "=r" (enabled) : "i" (offsetof(struct paca_struct, hard_enabled)));
109 
110 	return enabled;
111 }
112 
113 static inline notrace void set_soft_enabled(unsigned long enable)
114 {
115 	__asm__ __volatile__("stb %0,%1(13)"
116 	: : "r" (enable), "i" (offsetof(struct paca_struct, soft_enabled)));
117 }
118 
119 notrace void arch_local_irq_restore(unsigned long en)
120 {
121 	/*
122 	 * get_paca()->soft_enabled = en;
123 	 * Is it ever valid to use local_irq_restore(0) when soft_enabled is 1?
124 	 * That was allowed before, and in such a case we do need to take care
125 	 * that gcc will set soft_enabled directly via r13, not choose to use
126 	 * an intermediate register, lest we're preempted to a different cpu.
127 	 */
128 	set_soft_enabled(en);
129 	if (!en)
130 		return;
131 
132 #ifdef CONFIG_PPC_STD_MMU_64
133 	if (firmware_has_feature(FW_FEATURE_ISERIES)) {
134 		/*
135 		 * Do we need to disable preemption here?  Not really: in the
136 		 * unlikely event that we're preempted to a different cpu in
137 		 * between getting r13, loading its lppaca_ptr, and loading
138 		 * its any_int, we might call iseries_handle_interrupts without
139 		 * an interrupt pending on the new cpu, but that's no disaster,
140 		 * is it?  And the business of preempting us off the old cpu
141 		 * would itself involve a local_irq_restore which handles the
142 		 * interrupt to that cpu.
143 		 *
144 		 * But use "local_paca->lppaca_ptr" instead of "get_lppaca()"
145 		 * to avoid any preemption checking added into get_paca().
146 		 */
147 		if (local_paca->lppaca_ptr->int_dword.any_int)
148 			iseries_handle_interrupts();
149 	}
150 #endif /* CONFIG_PPC_STD_MMU_64 */
151 
152 	/*
153 	 * if (get_paca()->hard_enabled) return;
154 	 * But again we need to take care that gcc gets hard_enabled directly
155 	 * via r13, not choose to use an intermediate register, lest we're
156 	 * preempted to a different cpu in between the two instructions.
157 	 */
158 	if (get_hard_enabled())
159 		return;
160 
161 #if defined(CONFIG_BOOKE) && defined(CONFIG_SMP)
162 	/* Check for pending doorbell interrupts and resend to ourself */
163 	doorbell_check_self();
164 #endif
165 
166 	/*
167 	 * Need to hard-enable interrupts here.  Since currently disabled,
168 	 * no need to take further asm precautions against preemption; but
169 	 * use local_paca instead of get_paca() to avoid preemption checking.
170 	 */
171 	local_paca->hard_enabled = en;
172 
173 #ifndef CONFIG_BOOKE
174 	/* On server, re-trigger the decrementer if it went negative since
175 	 * some processors only trigger on edge transitions of the sign bit.
176 	 *
177 	 * BookE has a level sensitive decrementer (latches in TSR) so we
178 	 * don't need that
179 	 */
180 	if ((int)mfspr(SPRN_DEC) < 0)
181 		mtspr(SPRN_DEC, 1);
182 #endif /* CONFIG_BOOKE */
183 
184 	/*
185 	 * Force the delivery of pending soft-disabled interrupts on PS3.
186 	 * Any HV call will have this side effect.
187 	 */
188 	if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
189 		u64 tmp;
190 		lv1_get_version_info(&tmp);
191 	}
192 
193 	__hard_irq_enable();
194 }
195 EXPORT_SYMBOL(arch_local_irq_restore);
196 #endif /* CONFIG_PPC64 */
197 
198 int arch_show_interrupts(struct seq_file *p, int prec)
199 {
200 	int j;
201 
202 #if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
203 	if (tau_initialized) {
204 		seq_printf(p, "%*s: ", prec, "TAU");
205 		for_each_online_cpu(j)
206 			seq_printf(p, "%10u ", tau_interrupts(j));
207 		seq_puts(p, "  PowerPC             Thermal Assist (cpu temp)\n");
208 	}
209 #endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
210 
211 	seq_printf(p, "%*s: ", prec, "LOC");
212 	for_each_online_cpu(j)
213 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs);
214         seq_printf(p, "  Local timer interrupts\n");
215 
216 	seq_printf(p, "%*s: ", prec, "SPU");
217 	for_each_online_cpu(j)
218 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
219 	seq_printf(p, "  Spurious interrupts\n");
220 
221 	seq_printf(p, "%*s: ", prec, "CNT");
222 	for_each_online_cpu(j)
223 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
224 	seq_printf(p, "  Performance monitoring interrupts\n");
225 
226 	seq_printf(p, "%*s: ", prec, "MCE");
227 	for_each_online_cpu(j)
228 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
229 	seq_printf(p, "  Machine check exceptions\n");
230 
231 	return 0;
232 }
233 
234 /*
235  * /proc/stat helpers
236  */
237 u64 arch_irq_stat_cpu(unsigned int cpu)
238 {
239 	u64 sum = per_cpu(irq_stat, cpu).timer_irqs;
240 
241 	sum += per_cpu(irq_stat, cpu).pmu_irqs;
242 	sum += per_cpu(irq_stat, cpu).mce_exceptions;
243 	sum += per_cpu(irq_stat, cpu).spurious_irqs;
244 
245 	return sum;
246 }
247 
248 #ifdef CONFIG_HOTPLUG_CPU
249 void migrate_irqs(void)
250 {
251 	struct irq_desc *desc;
252 	unsigned int irq;
253 	static int warned;
254 	cpumask_var_t mask;
255 	const struct cpumask *map = cpu_online_mask;
256 
257 	alloc_cpumask_var(&mask, GFP_KERNEL);
258 
259 	for_each_irq(irq) {
260 		struct irq_data *data;
261 		struct irq_chip *chip;
262 
263 		desc = irq_to_desc(irq);
264 		if (!desc)
265 			continue;
266 
267 		data = irq_desc_get_irq_data(desc);
268 		if (irqd_is_per_cpu(data))
269 			continue;
270 
271 		chip = irq_data_get_irq_chip(data);
272 
273 		cpumask_and(mask, data->affinity, map);
274 		if (cpumask_any(mask) >= nr_cpu_ids) {
275 			printk("Breaking affinity for irq %i\n", irq);
276 			cpumask_copy(mask, map);
277 		}
278 		if (chip->irq_set_affinity)
279 			chip->irq_set_affinity(data, mask, true);
280 		else if (desc->action && !(warned++))
281 			printk("Cannot set affinity for irq %i\n", irq);
282 	}
283 
284 	free_cpumask_var(mask);
285 
286 	local_irq_enable();
287 	mdelay(1);
288 	local_irq_disable();
289 }
290 #endif
291 
292 static inline void handle_one_irq(unsigned int irq)
293 {
294 	struct thread_info *curtp, *irqtp;
295 	unsigned long saved_sp_limit;
296 	struct irq_desc *desc;
297 
298 	/* Switch to the irq stack to handle this */
299 	curtp = current_thread_info();
300 	irqtp = hardirq_ctx[smp_processor_id()];
301 
302 	if (curtp == irqtp) {
303 		/* We're already on the irq stack, just handle it */
304 		generic_handle_irq(irq);
305 		return;
306 	}
307 
308 	desc = irq_to_desc(irq);
309 	saved_sp_limit = current->thread.ksp_limit;
310 
311 	irqtp->task = curtp->task;
312 	irqtp->flags = 0;
313 
314 	/* Copy the softirq bits in preempt_count so that the
315 	 * softirq checks work in the hardirq context. */
316 	irqtp->preempt_count = (irqtp->preempt_count & ~SOFTIRQ_MASK) |
317 			       (curtp->preempt_count & SOFTIRQ_MASK);
318 
319 	current->thread.ksp_limit = (unsigned long)irqtp +
320 		_ALIGN_UP(sizeof(struct thread_info), 16);
321 
322 	call_handle_irq(irq, desc, irqtp, desc->handle_irq);
323 	current->thread.ksp_limit = saved_sp_limit;
324 	irqtp->task = NULL;
325 
326 	/* Set any flag that may have been set on the
327 	 * alternate stack
328 	 */
329 	if (irqtp->flags)
330 		set_bits(irqtp->flags, &curtp->flags);
331 }
332 
333 static inline void check_stack_overflow(void)
334 {
335 #ifdef CONFIG_DEBUG_STACKOVERFLOW
336 	long sp;
337 
338 	sp = __get_SP() & (THREAD_SIZE-1);
339 
340 	/* check for stack overflow: is there less than 2KB free? */
341 	if (unlikely(sp < (sizeof(struct thread_info) + 2048))) {
342 		printk("do_IRQ: stack overflow: %ld\n",
343 			sp - sizeof(struct thread_info));
344 		dump_stack();
345 	}
346 #endif
347 }
348 
349 void do_IRQ(struct pt_regs *regs)
350 {
351 	struct pt_regs *old_regs = set_irq_regs(regs);
352 	unsigned int irq;
353 
354 	trace_irq_entry(regs);
355 
356 	irq_enter();
357 
358 	check_stack_overflow();
359 
360 	irq = ppc_md.get_irq();
361 
362 	if (irq != NO_IRQ && irq != NO_IRQ_IGNORE)
363 		handle_one_irq(irq);
364 	else if (irq != NO_IRQ_IGNORE)
365 		__get_cpu_var(irq_stat).spurious_irqs++;
366 
367 	irq_exit();
368 	set_irq_regs(old_regs);
369 
370 #ifdef CONFIG_PPC_ISERIES
371 	if (firmware_has_feature(FW_FEATURE_ISERIES) &&
372 			get_lppaca()->int_dword.fields.decr_int) {
373 		get_lppaca()->int_dword.fields.decr_int = 0;
374 		/* Signal a fake decrementer interrupt */
375 		timer_interrupt(regs);
376 	}
377 #endif
378 
379 	trace_irq_exit(regs);
380 }
381 
382 void __init init_IRQ(void)
383 {
384 	if (ppc_md.init_IRQ)
385 		ppc_md.init_IRQ();
386 
387 	exc_lvl_ctx_init();
388 
389 	irq_ctx_init();
390 }
391 
392 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
393 struct thread_info   *critirq_ctx[NR_CPUS] __read_mostly;
394 struct thread_info    *dbgirq_ctx[NR_CPUS] __read_mostly;
395 struct thread_info *mcheckirq_ctx[NR_CPUS] __read_mostly;
396 
397 void exc_lvl_ctx_init(void)
398 {
399 	struct thread_info *tp;
400 	int i, cpu_nr;
401 
402 	for_each_possible_cpu(i) {
403 #ifdef CONFIG_PPC64
404 		cpu_nr = i;
405 #else
406 		cpu_nr = get_hard_smp_processor_id(i);
407 #endif
408 		memset((void *)critirq_ctx[cpu_nr], 0, THREAD_SIZE);
409 		tp = critirq_ctx[cpu_nr];
410 		tp->cpu = cpu_nr;
411 		tp->preempt_count = 0;
412 
413 #ifdef CONFIG_BOOKE
414 		memset((void *)dbgirq_ctx[cpu_nr], 0, THREAD_SIZE);
415 		tp = dbgirq_ctx[cpu_nr];
416 		tp->cpu = cpu_nr;
417 		tp->preempt_count = 0;
418 
419 		memset((void *)mcheckirq_ctx[cpu_nr], 0, THREAD_SIZE);
420 		tp = mcheckirq_ctx[cpu_nr];
421 		tp->cpu = cpu_nr;
422 		tp->preempt_count = HARDIRQ_OFFSET;
423 #endif
424 	}
425 }
426 #endif
427 
428 struct thread_info *softirq_ctx[NR_CPUS] __read_mostly;
429 struct thread_info *hardirq_ctx[NR_CPUS] __read_mostly;
430 
431 void irq_ctx_init(void)
432 {
433 	struct thread_info *tp;
434 	int i;
435 
436 	for_each_possible_cpu(i) {
437 		memset((void *)softirq_ctx[i], 0, THREAD_SIZE);
438 		tp = softirq_ctx[i];
439 		tp->cpu = i;
440 		tp->preempt_count = 0;
441 
442 		memset((void *)hardirq_ctx[i], 0, THREAD_SIZE);
443 		tp = hardirq_ctx[i];
444 		tp->cpu = i;
445 		tp->preempt_count = HARDIRQ_OFFSET;
446 	}
447 }
448 
449 static inline void do_softirq_onstack(void)
450 {
451 	struct thread_info *curtp, *irqtp;
452 	unsigned long saved_sp_limit = current->thread.ksp_limit;
453 
454 	curtp = current_thread_info();
455 	irqtp = softirq_ctx[smp_processor_id()];
456 	irqtp->task = curtp->task;
457 	current->thread.ksp_limit = (unsigned long)irqtp +
458 				    _ALIGN_UP(sizeof(struct thread_info), 16);
459 	call_do_softirq(irqtp);
460 	current->thread.ksp_limit = saved_sp_limit;
461 	irqtp->task = NULL;
462 }
463 
464 void do_softirq(void)
465 {
466 	unsigned long flags;
467 
468 	if (in_interrupt())
469 		return;
470 
471 	local_irq_save(flags);
472 
473 	if (local_softirq_pending())
474 		do_softirq_onstack();
475 
476 	local_irq_restore(flags);
477 }
478 
479 
480 /*
481  * IRQ controller and virtual interrupts
482  */
483 
484 /* The main irq map itself is an array of NR_IRQ entries containing the
485  * associate host and irq number. An entry with a host of NULL is free.
486  * An entry can be allocated if it's free, the allocator always then sets
487  * hwirq first to the host's invalid irq number and then fills ops.
488  */
489 struct irq_map_entry {
490 	irq_hw_number_t	hwirq;
491 	struct irq_host	*host;
492 };
493 
494 static LIST_HEAD(irq_hosts);
495 static DEFINE_RAW_SPINLOCK(irq_big_lock);
496 static unsigned int revmap_trees_allocated;
497 static DEFINE_MUTEX(revmap_trees_mutex);
498 static struct irq_map_entry irq_map[NR_IRQS];
499 static unsigned int irq_virq_count = NR_IRQS;
500 static struct irq_host *irq_default_host;
501 
502 irq_hw_number_t irqd_to_hwirq(struct irq_data *d)
503 {
504 	return irq_map[d->irq].hwirq;
505 }
506 EXPORT_SYMBOL_GPL(irqd_to_hwirq);
507 
508 irq_hw_number_t virq_to_hw(unsigned int virq)
509 {
510 	return irq_map[virq].hwirq;
511 }
512 EXPORT_SYMBOL_GPL(virq_to_hw);
513 
514 struct irq_host *virq_to_host(unsigned int virq)
515 {
516 	return irq_map[virq].host;
517 }
518 EXPORT_SYMBOL_GPL(virq_to_host);
519 
520 static int default_irq_host_match(struct irq_host *h, struct device_node *np)
521 {
522 	return h->of_node != NULL && h->of_node == np;
523 }
524 
525 struct irq_host *irq_alloc_host(struct device_node *of_node,
526 				unsigned int revmap_type,
527 				unsigned int revmap_arg,
528 				struct irq_host_ops *ops,
529 				irq_hw_number_t inval_irq)
530 {
531 	struct irq_host *host;
532 	unsigned int size = sizeof(struct irq_host);
533 	unsigned int i;
534 	unsigned int *rmap;
535 	unsigned long flags;
536 
537 	/* Allocate structure and revmap table if using linear mapping */
538 	if (revmap_type == IRQ_HOST_MAP_LINEAR)
539 		size += revmap_arg * sizeof(unsigned int);
540 	host = zalloc_maybe_bootmem(size, GFP_KERNEL);
541 	if (host == NULL)
542 		return NULL;
543 
544 	/* Fill structure */
545 	host->revmap_type = revmap_type;
546 	host->inval_irq = inval_irq;
547 	host->ops = ops;
548 	host->of_node = of_node_get(of_node);
549 
550 	if (host->ops->match == NULL)
551 		host->ops->match = default_irq_host_match;
552 
553 	raw_spin_lock_irqsave(&irq_big_lock, flags);
554 
555 	/* If it's a legacy controller, check for duplicates and
556 	 * mark it as allocated (we use irq 0 host pointer for that
557 	 */
558 	if (revmap_type == IRQ_HOST_MAP_LEGACY) {
559 		if (irq_map[0].host != NULL) {
560 			raw_spin_unlock_irqrestore(&irq_big_lock, flags);
561 			/* If we are early boot, we can't free the structure,
562 			 * too bad...
563 			 * this will be fixed once slab is made available early
564 			 * instead of the current cruft
565 			 */
566 			if (mem_init_done) {
567 				of_node_put(host->of_node);
568 				kfree(host);
569 			}
570 			return NULL;
571 		}
572 		irq_map[0].host = host;
573 	}
574 
575 	list_add(&host->link, &irq_hosts);
576 	raw_spin_unlock_irqrestore(&irq_big_lock, flags);
577 
578 	/* Additional setups per revmap type */
579 	switch(revmap_type) {
580 	case IRQ_HOST_MAP_LEGACY:
581 		/* 0 is always the invalid number for legacy */
582 		host->inval_irq = 0;
583 		/* setup us as the host for all legacy interrupts */
584 		for (i = 1; i < NUM_ISA_INTERRUPTS; i++) {
585 			irq_map[i].hwirq = i;
586 			smp_wmb();
587 			irq_map[i].host = host;
588 			smp_wmb();
589 
590 			/* Clear norequest flags */
591 			irq_clear_status_flags(i, IRQ_NOREQUEST);
592 
593 			/* Legacy flags are left to default at this point,
594 			 * one can then use irq_create_mapping() to
595 			 * explicitly change them
596 			 */
597 			ops->map(host, i, i);
598 		}
599 		break;
600 	case IRQ_HOST_MAP_LINEAR:
601 		rmap = (unsigned int *)(host + 1);
602 		for (i = 0; i < revmap_arg; i++)
603 			rmap[i] = NO_IRQ;
604 		host->revmap_data.linear.size = revmap_arg;
605 		smp_wmb();
606 		host->revmap_data.linear.revmap = rmap;
607 		break;
608 	default:
609 		break;
610 	}
611 
612 	pr_debug("irq: Allocated host of type %d @0x%p\n", revmap_type, host);
613 
614 	return host;
615 }
616 
617 struct irq_host *irq_find_host(struct device_node *node)
618 {
619 	struct irq_host *h, *found = NULL;
620 	unsigned long flags;
621 
622 	/* We might want to match the legacy controller last since
623 	 * it might potentially be set to match all interrupts in
624 	 * the absence of a device node. This isn't a problem so far
625 	 * yet though...
626 	 */
627 	raw_spin_lock_irqsave(&irq_big_lock, flags);
628 	list_for_each_entry(h, &irq_hosts, link)
629 		if (h->ops->match(h, node)) {
630 			found = h;
631 			break;
632 		}
633 	raw_spin_unlock_irqrestore(&irq_big_lock, flags);
634 	return found;
635 }
636 EXPORT_SYMBOL_GPL(irq_find_host);
637 
638 void irq_set_default_host(struct irq_host *host)
639 {
640 	pr_debug("irq: Default host set to @0x%p\n", host);
641 
642 	irq_default_host = host;
643 }
644 
645 void irq_set_virq_count(unsigned int count)
646 {
647 	pr_debug("irq: Trying to set virq count to %d\n", count);
648 
649 	BUG_ON(count < NUM_ISA_INTERRUPTS);
650 	if (count < NR_IRQS)
651 		irq_virq_count = count;
652 }
653 
654 static int irq_setup_virq(struct irq_host *host, unsigned int virq,
655 			    irq_hw_number_t hwirq)
656 {
657 	int res;
658 
659 	res = irq_alloc_desc_at(virq, 0);
660 	if (res != virq) {
661 		pr_debug("irq: -> allocating desc failed\n");
662 		goto error;
663 	}
664 
665 	irq_clear_status_flags(virq, IRQ_NOREQUEST);
666 
667 	/* map it */
668 	smp_wmb();
669 	irq_map[virq].hwirq = hwirq;
670 	smp_mb();
671 
672 	if (host->ops->map(host, virq, hwirq)) {
673 		pr_debug("irq: -> mapping failed, freeing\n");
674 		goto errdesc;
675 	}
676 
677 	return 0;
678 
679 errdesc:
680 	irq_free_descs(virq, 1);
681 error:
682 	irq_free_virt(virq, 1);
683 	return -1;
684 }
685 
686 unsigned int irq_create_direct_mapping(struct irq_host *host)
687 {
688 	unsigned int virq;
689 
690 	if (host == NULL)
691 		host = irq_default_host;
692 
693 	BUG_ON(host == NULL);
694 	WARN_ON(host->revmap_type != IRQ_HOST_MAP_NOMAP);
695 
696 	virq = irq_alloc_virt(host, 1, 0);
697 	if (virq == NO_IRQ) {
698 		pr_debug("irq: create_direct virq allocation failed\n");
699 		return NO_IRQ;
700 	}
701 
702 	pr_debug("irq: create_direct obtained virq %d\n", virq);
703 
704 	if (irq_setup_virq(host, virq, virq))
705 		return NO_IRQ;
706 
707 	return virq;
708 }
709 
710 unsigned int irq_create_mapping(struct irq_host *host,
711 				irq_hw_number_t hwirq)
712 {
713 	unsigned int virq, hint;
714 
715 	pr_debug("irq: irq_create_mapping(0x%p, 0x%lx)\n", host, hwirq);
716 
717 	/* Look for default host if nececssary */
718 	if (host == NULL)
719 		host = irq_default_host;
720 	if (host == NULL) {
721 		printk(KERN_WARNING "irq_create_mapping called for"
722 		       " NULL host, hwirq=%lx\n", hwirq);
723 		WARN_ON(1);
724 		return NO_IRQ;
725 	}
726 	pr_debug("irq: -> using host @%p\n", host);
727 
728 	/* Check if mapping already exist, if it does, call
729 	 * host->ops->map() to update the flags
730 	 */
731 	virq = irq_find_mapping(host, hwirq);
732 	if (virq != NO_IRQ) {
733 		if (host->ops->remap)
734 			host->ops->remap(host, virq, hwirq);
735 		pr_debug("irq: -> existing mapping on virq %d\n", virq);
736 		return virq;
737 	}
738 
739 	/* Get a virtual interrupt number */
740 	if (host->revmap_type == IRQ_HOST_MAP_LEGACY) {
741 		/* Handle legacy */
742 		virq = (unsigned int)hwirq;
743 		if (virq == 0 || virq >= NUM_ISA_INTERRUPTS)
744 			return NO_IRQ;
745 		return virq;
746 	} else {
747 		/* Allocate a virtual interrupt number */
748 		hint = hwirq % irq_virq_count;
749 		virq = irq_alloc_virt(host, 1, hint);
750 		if (virq == NO_IRQ) {
751 			pr_debug("irq: -> virq allocation failed\n");
752 			return NO_IRQ;
753 		}
754 	}
755 
756 	if (irq_setup_virq(host, virq, hwirq))
757 		return NO_IRQ;
758 
759 	printk(KERN_DEBUG "irq: irq %lu on host %s mapped to virtual irq %u\n",
760 		hwirq, host->of_node ? host->of_node->full_name : "null", virq);
761 
762 	return virq;
763 }
764 EXPORT_SYMBOL_GPL(irq_create_mapping);
765 
766 unsigned int irq_create_of_mapping(struct device_node *controller,
767 				   const u32 *intspec, unsigned int intsize)
768 {
769 	struct irq_host *host;
770 	irq_hw_number_t hwirq;
771 	unsigned int type = IRQ_TYPE_NONE;
772 	unsigned int virq;
773 
774 	if (controller == NULL)
775 		host = irq_default_host;
776 	else
777 		host = irq_find_host(controller);
778 	if (host == NULL) {
779 		printk(KERN_WARNING "irq: no irq host found for %s !\n",
780 		       controller->full_name);
781 		return NO_IRQ;
782 	}
783 
784 	/* If host has no translation, then we assume interrupt line */
785 	if (host->ops->xlate == NULL)
786 		hwirq = intspec[0];
787 	else {
788 		if (host->ops->xlate(host, controller, intspec, intsize,
789 				     &hwirq, &type))
790 			return NO_IRQ;
791 	}
792 
793 	/* Create mapping */
794 	virq = irq_create_mapping(host, hwirq);
795 	if (virq == NO_IRQ)
796 		return virq;
797 
798 	/* Set type if specified and different than the current one */
799 	if (type != IRQ_TYPE_NONE &&
800 	    type != (irqd_get_trigger_type(irq_get_irq_data(virq))))
801 		irq_set_irq_type(virq, type);
802 	return virq;
803 }
804 EXPORT_SYMBOL_GPL(irq_create_of_mapping);
805 
806 void irq_dispose_mapping(unsigned int virq)
807 {
808 	struct irq_host *host;
809 	irq_hw_number_t hwirq;
810 
811 	if (virq == NO_IRQ)
812 		return;
813 
814 	host = irq_map[virq].host;
815 	WARN_ON (host == NULL);
816 	if (host == NULL)
817 		return;
818 
819 	/* Never unmap legacy interrupts */
820 	if (host->revmap_type == IRQ_HOST_MAP_LEGACY)
821 		return;
822 
823 	/* remove chip and handler */
824 	irq_set_chip_and_handler(virq, NULL, NULL);
825 
826 	/* Make sure it's completed */
827 	synchronize_irq(virq);
828 
829 	/* Tell the PIC about it */
830 	if (host->ops->unmap)
831 		host->ops->unmap(host, virq);
832 	smp_mb();
833 
834 	/* Clear reverse map */
835 	hwirq = irq_map[virq].hwirq;
836 	switch(host->revmap_type) {
837 	case IRQ_HOST_MAP_LINEAR:
838 		if (hwirq < host->revmap_data.linear.size)
839 			host->revmap_data.linear.revmap[hwirq] = NO_IRQ;
840 		break;
841 	case IRQ_HOST_MAP_TREE:
842 		/*
843 		 * Check if radix tree allocated yet, if not then nothing to
844 		 * remove.
845 		 */
846 		smp_rmb();
847 		if (revmap_trees_allocated < 1)
848 			break;
849 		mutex_lock(&revmap_trees_mutex);
850 		radix_tree_delete(&host->revmap_data.tree, hwirq);
851 		mutex_unlock(&revmap_trees_mutex);
852 		break;
853 	}
854 
855 	/* Destroy map */
856 	smp_mb();
857 	irq_map[virq].hwirq = host->inval_irq;
858 
859 	irq_set_status_flags(virq, IRQ_NOREQUEST);
860 
861 	irq_free_descs(virq, 1);
862 	/* Free it */
863 	irq_free_virt(virq, 1);
864 }
865 EXPORT_SYMBOL_GPL(irq_dispose_mapping);
866 
867 unsigned int irq_find_mapping(struct irq_host *host,
868 			      irq_hw_number_t hwirq)
869 {
870 	unsigned int i;
871 	unsigned int hint = hwirq % irq_virq_count;
872 
873 	/* Look for default host if nececssary */
874 	if (host == NULL)
875 		host = irq_default_host;
876 	if (host == NULL)
877 		return NO_IRQ;
878 
879 	/* legacy -> bail early */
880 	if (host->revmap_type == IRQ_HOST_MAP_LEGACY)
881 		return hwirq;
882 
883 	/* Slow path does a linear search of the map */
884 	if (hint < NUM_ISA_INTERRUPTS)
885 		hint = NUM_ISA_INTERRUPTS;
886 	i = hint;
887 	do  {
888 		if (irq_map[i].host == host &&
889 		    irq_map[i].hwirq == hwirq)
890 			return i;
891 		i++;
892 		if (i >= irq_virq_count)
893 			i = NUM_ISA_INTERRUPTS;
894 	} while(i != hint);
895 	return NO_IRQ;
896 }
897 EXPORT_SYMBOL_GPL(irq_find_mapping);
898 
899 
900 unsigned int irq_radix_revmap_lookup(struct irq_host *host,
901 				     irq_hw_number_t hwirq)
902 {
903 	struct irq_map_entry *ptr;
904 	unsigned int virq;
905 
906 	WARN_ON(host->revmap_type != IRQ_HOST_MAP_TREE);
907 
908 	/*
909 	 * Check if the radix tree exists and has bee initialized.
910 	 * If not, we fallback to slow mode
911 	 */
912 	if (revmap_trees_allocated < 2)
913 		return irq_find_mapping(host, hwirq);
914 
915 	/* Now try to resolve */
916 	/*
917 	 * No rcu_read_lock(ing) needed, the ptr returned can't go under us
918 	 * as it's referencing an entry in the static irq_map table.
919 	 */
920 	ptr = radix_tree_lookup(&host->revmap_data.tree, hwirq);
921 
922 	/*
923 	 * If found in radix tree, then fine.
924 	 * Else fallback to linear lookup - this should not happen in practice
925 	 * as it means that we failed to insert the node in the radix tree.
926 	 */
927 	if (ptr)
928 		virq = ptr - irq_map;
929 	else
930 		virq = irq_find_mapping(host, hwirq);
931 
932 	return virq;
933 }
934 
935 void irq_radix_revmap_insert(struct irq_host *host, unsigned int virq,
936 			     irq_hw_number_t hwirq)
937 {
938 
939 	WARN_ON(host->revmap_type != IRQ_HOST_MAP_TREE);
940 
941 	/*
942 	 * Check if the radix tree exists yet.
943 	 * If not, then the irq will be inserted into the tree when it gets
944 	 * initialized.
945 	 */
946 	smp_rmb();
947 	if (revmap_trees_allocated < 1)
948 		return;
949 
950 	if (virq != NO_IRQ) {
951 		mutex_lock(&revmap_trees_mutex);
952 		radix_tree_insert(&host->revmap_data.tree, hwirq,
953 				  &irq_map[virq]);
954 		mutex_unlock(&revmap_trees_mutex);
955 	}
956 }
957 
958 unsigned int irq_linear_revmap(struct irq_host *host,
959 			       irq_hw_number_t hwirq)
960 {
961 	unsigned int *revmap;
962 
963 	WARN_ON(host->revmap_type != IRQ_HOST_MAP_LINEAR);
964 
965 	/* Check revmap bounds */
966 	if (unlikely(hwirq >= host->revmap_data.linear.size))
967 		return irq_find_mapping(host, hwirq);
968 
969 	/* Check if revmap was allocated */
970 	revmap = host->revmap_data.linear.revmap;
971 	if (unlikely(revmap == NULL))
972 		return irq_find_mapping(host, hwirq);
973 
974 	/* Fill up revmap with slow path if no mapping found */
975 	if (unlikely(revmap[hwirq] == NO_IRQ))
976 		revmap[hwirq] = irq_find_mapping(host, hwirq);
977 
978 	return revmap[hwirq];
979 }
980 
981 unsigned int irq_alloc_virt(struct irq_host *host,
982 			    unsigned int count,
983 			    unsigned int hint)
984 {
985 	unsigned long flags;
986 	unsigned int i, j, found = NO_IRQ;
987 
988 	if (count == 0 || count > (irq_virq_count - NUM_ISA_INTERRUPTS))
989 		return NO_IRQ;
990 
991 	raw_spin_lock_irqsave(&irq_big_lock, flags);
992 
993 	/* Use hint for 1 interrupt if any */
994 	if (count == 1 && hint >= NUM_ISA_INTERRUPTS &&
995 	    hint < irq_virq_count && irq_map[hint].host == NULL) {
996 		found = hint;
997 		goto hint_found;
998 	}
999 
1000 	/* Look for count consecutive numbers in the allocatable
1001 	 * (non-legacy) space
1002 	 */
1003 	for (i = NUM_ISA_INTERRUPTS, j = 0; i < irq_virq_count; i++) {
1004 		if (irq_map[i].host != NULL)
1005 			j = 0;
1006 		else
1007 			j++;
1008 
1009 		if (j == count) {
1010 			found = i - count + 1;
1011 			break;
1012 		}
1013 	}
1014 	if (found == NO_IRQ) {
1015 		raw_spin_unlock_irqrestore(&irq_big_lock, flags);
1016 		return NO_IRQ;
1017 	}
1018  hint_found:
1019 	for (i = found; i < (found + count); i++) {
1020 		irq_map[i].hwirq = host->inval_irq;
1021 		smp_wmb();
1022 		irq_map[i].host = host;
1023 	}
1024 	raw_spin_unlock_irqrestore(&irq_big_lock, flags);
1025 	return found;
1026 }
1027 
1028 void irq_free_virt(unsigned int virq, unsigned int count)
1029 {
1030 	unsigned long flags;
1031 	unsigned int i;
1032 
1033 	WARN_ON (virq < NUM_ISA_INTERRUPTS);
1034 	WARN_ON (count == 0 || (virq + count) > irq_virq_count);
1035 
1036 	raw_spin_lock_irqsave(&irq_big_lock, flags);
1037 	for (i = virq; i < (virq + count); i++) {
1038 		struct irq_host *host;
1039 
1040 		if (i < NUM_ISA_INTERRUPTS ||
1041 		    (virq + count) > irq_virq_count)
1042 			continue;
1043 
1044 		host = irq_map[i].host;
1045 		irq_map[i].hwirq = host->inval_irq;
1046 		smp_wmb();
1047 		irq_map[i].host = NULL;
1048 	}
1049 	raw_spin_unlock_irqrestore(&irq_big_lock, flags);
1050 }
1051 
1052 int arch_early_irq_init(void)
1053 {
1054 	return 0;
1055 }
1056 
1057 /* We need to create the radix trees late */
1058 static int irq_late_init(void)
1059 {
1060 	struct irq_host *h;
1061 	unsigned int i;
1062 
1063 	/*
1064 	 * No mutual exclusion with respect to accessors of the tree is needed
1065 	 * here as the synchronization is done via the state variable
1066 	 * revmap_trees_allocated.
1067 	 */
1068 	list_for_each_entry(h, &irq_hosts, link) {
1069 		if (h->revmap_type == IRQ_HOST_MAP_TREE)
1070 			INIT_RADIX_TREE(&h->revmap_data.tree, GFP_KERNEL);
1071 	}
1072 
1073 	/*
1074 	 * Make sure the radix trees inits are visible before setting
1075 	 * the flag
1076 	 */
1077 	smp_wmb();
1078 	revmap_trees_allocated = 1;
1079 
1080 	/*
1081 	 * Insert the reverse mapping for those interrupts already present
1082 	 * in irq_map[].
1083 	 */
1084 	mutex_lock(&revmap_trees_mutex);
1085 	for (i = 0; i < irq_virq_count; i++) {
1086 		if (irq_map[i].host &&
1087 		    (irq_map[i].host->revmap_type == IRQ_HOST_MAP_TREE))
1088 			radix_tree_insert(&irq_map[i].host->revmap_data.tree,
1089 					  irq_map[i].hwirq, &irq_map[i]);
1090 	}
1091 	mutex_unlock(&revmap_trees_mutex);
1092 
1093 	/*
1094 	 * Make sure the radix trees insertions are visible before setting
1095 	 * the flag
1096 	 */
1097 	smp_wmb();
1098 	revmap_trees_allocated = 2;
1099 
1100 	return 0;
1101 }
1102 arch_initcall(irq_late_init);
1103 
1104 #ifdef CONFIG_VIRQ_DEBUG
1105 static int virq_debug_show(struct seq_file *m, void *private)
1106 {
1107 	unsigned long flags;
1108 	struct irq_desc *desc;
1109 	const char *p;
1110 	static const char none[] = "none";
1111 	void *data;
1112 	int i;
1113 
1114 	seq_printf(m, "%-5s  %-7s  %-15s  %-18s  %s\n", "virq", "hwirq",
1115 		      "chip name", "chip data", "host name");
1116 
1117 	for (i = 1; i < nr_irqs; i++) {
1118 		desc = irq_to_desc(i);
1119 		if (!desc)
1120 			continue;
1121 
1122 		raw_spin_lock_irqsave(&desc->lock, flags);
1123 
1124 		if (desc->action && desc->action->handler) {
1125 			struct irq_chip *chip;
1126 
1127 			seq_printf(m, "%5d  ", i);
1128 			seq_printf(m, "0x%05lx  ", irq_map[i].hwirq);
1129 
1130 			chip = irq_desc_get_chip(desc);
1131 			if (chip && chip->name)
1132 				p = chip->name;
1133 			else
1134 				p = none;
1135 			seq_printf(m, "%-15s  ", p);
1136 
1137 			data = irq_desc_get_chip_data(desc);
1138 			seq_printf(m, "0x%16p  ", data);
1139 
1140 			if (irq_map[i].host && irq_map[i].host->of_node)
1141 				p = irq_map[i].host->of_node->full_name;
1142 			else
1143 				p = none;
1144 			seq_printf(m, "%s\n", p);
1145 		}
1146 
1147 		raw_spin_unlock_irqrestore(&desc->lock, flags);
1148 	}
1149 
1150 	return 0;
1151 }
1152 
1153 static int virq_debug_open(struct inode *inode, struct file *file)
1154 {
1155 	return single_open(file, virq_debug_show, inode->i_private);
1156 }
1157 
1158 static const struct file_operations virq_debug_fops = {
1159 	.open = virq_debug_open,
1160 	.read = seq_read,
1161 	.llseek = seq_lseek,
1162 	.release = single_release,
1163 };
1164 
1165 static int __init irq_debugfs_init(void)
1166 {
1167 	if (debugfs_create_file("virq_mapping", S_IRUGO, powerpc_debugfs_root,
1168 				 NULL, &virq_debug_fops) == NULL)
1169 		return -ENOMEM;
1170 
1171 	return 0;
1172 }
1173 __initcall(irq_debugfs_init);
1174 #endif /* CONFIG_VIRQ_DEBUG */
1175 
1176 #ifdef CONFIG_PPC64
1177 static int __init setup_noirqdistrib(char *str)
1178 {
1179 	distribute_irqs = 0;
1180 	return 1;
1181 }
1182 
1183 __setup("noirqdistrib", setup_noirqdistrib);
1184 #endif /* CONFIG_PPC64 */
1185