xref: /openbmc/linux/arch/powerpc/kernel/irq.c (revision ae40e94f)
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_PPC_8xx
28  * to reduce code space and undefined function references.
29  */
30 
31 #undef DEBUG
32 
33 #include <linux/export.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/pci.h>
54 #include <linux/debugfs.h>
55 #include <linux/of.h>
56 #include <linux/of_irq.h>
57 
58 #include <linux/uaccess.h>
59 #include <asm/io.h>
60 #include <asm/pgtable.h>
61 #include <asm/irq.h>
62 #include <asm/cache.h>
63 #include <asm/prom.h>
64 #include <asm/ptrace.h>
65 #include <asm/machdep.h>
66 #include <asm/udbg.h>
67 #include <asm/smp.h>
68 #include <asm/livepatch.h>
69 #include <asm/asm-prototypes.h>
70 #include <asm/hw_irq.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 #include <asm/cpu_has_feature.h>
80 
81 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
82 EXPORT_PER_CPU_SYMBOL(irq_stat);
83 
84 int __irq_offset_value;
85 
86 #ifdef CONFIG_PPC32
87 EXPORT_SYMBOL(__irq_offset_value);
88 atomic_t ppc_n_lost_interrupts;
89 
90 #ifdef CONFIG_TAU_INT
91 extern int tau_initialized;
92 u32 tau_interrupts(unsigned long cpu);
93 #endif
94 #endif /* CONFIG_PPC32 */
95 
96 #ifdef CONFIG_PPC64
97 
98 int distribute_irqs = 1;
99 
100 static inline notrace unsigned long get_irq_happened(void)
101 {
102 	unsigned long happened;
103 
104 	__asm__ __volatile__("lbz %0,%1(13)"
105 	: "=r" (happened) : "i" (offsetof(struct paca_struct, irq_happened)));
106 
107 	return happened;
108 }
109 
110 static inline notrace int decrementer_check_overflow(void)
111 {
112  	u64 now = get_tb_or_rtc();
113 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
114 
115 	return now >= *next_tb;
116 }
117 
118 /* This is called whenever we are re-enabling interrupts
119  * and returns either 0 (nothing to do) or 500/900/280/a00/e80 if
120  * there's an EE, DEC or DBELL to generate.
121  *
122  * This is called in two contexts: From arch_local_irq_restore()
123  * before soft-enabling interrupts, and from the exception exit
124  * path when returning from an interrupt from a soft-disabled to
125  * a soft enabled context. In both case we have interrupts hard
126  * disabled.
127  *
128  * We take care of only clearing the bits we handled in the
129  * PACA irq_happened field since we can only re-emit one at a
130  * time and we don't want to "lose" one.
131  */
132 notrace unsigned int __check_irq_replay(void)
133 {
134 	/*
135 	 * We use local_paca rather than get_paca() to avoid all
136 	 * the debug_smp_processor_id() business in this low level
137 	 * function
138 	 */
139 	unsigned char happened = local_paca->irq_happened;
140 
141 	/*
142 	 * We are responding to the next interrupt, so interrupt-off
143 	 * latencies should be reset here.
144 	 */
145 	trace_hardirqs_on();
146 	trace_hardirqs_off();
147 
148 	/*
149 	 * We are always hard disabled here, but PACA_IRQ_HARD_DIS may
150 	 * not be set, which means interrupts have only just been hard
151 	 * disabled as part of the local_irq_restore or interrupt return
152 	 * code. In that case, skip the decrementr check becaus it's
153 	 * expensive to read the TB.
154 	 *
155 	 * HARD_DIS then gets cleared here, but it's reconciled later.
156 	 * Either local_irq_disable will replay the interrupt and that
157 	 * will reconcile state like other hard interrupts. Or interrupt
158 	 * retur will replay the interrupt and in that case it sets
159 	 * PACA_IRQ_HARD_DIS by hand (see comments in entry_64.S).
160 	 */
161 	if (happened & PACA_IRQ_HARD_DIS) {
162 		local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
163 
164 		/*
165 		 * We may have missed a decrementer interrupt if hard disabled.
166 		 * Check the decrementer register in case we had a rollover
167 		 * while hard disabled.
168 		 */
169 		if (!(happened & PACA_IRQ_DEC)) {
170 			if (decrementer_check_overflow()) {
171 				local_paca->irq_happened |= PACA_IRQ_DEC;
172 				happened |= PACA_IRQ_DEC;
173 			}
174 		}
175 	}
176 
177 	/*
178 	 * Force the delivery of pending soft-disabled interrupts on PS3.
179 	 * Any HV call will have this side effect.
180 	 */
181 	if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
182 		u64 tmp, tmp2;
183 		lv1_get_version_info(&tmp, &tmp2);
184 	}
185 
186 	/*
187 	 * Check if an hypervisor Maintenance interrupt happened.
188 	 * This is a higher priority interrupt than the others, so
189 	 * replay it first.
190 	 */
191 	if (happened & PACA_IRQ_HMI) {
192 		local_paca->irq_happened &= ~PACA_IRQ_HMI;
193 		return 0xe60;
194 	}
195 
196 	if (happened & PACA_IRQ_DEC) {
197 		local_paca->irq_happened &= ~PACA_IRQ_DEC;
198 		return 0x900;
199 	}
200 
201 	if (happened & PACA_IRQ_PMI) {
202 		local_paca->irq_happened &= ~PACA_IRQ_PMI;
203 		return 0xf00;
204 	}
205 
206 	if (happened & PACA_IRQ_EE) {
207 		local_paca->irq_happened &= ~PACA_IRQ_EE;
208 		return 0x500;
209 	}
210 
211 #ifdef CONFIG_PPC_BOOK3E
212 	/*
213 	 * Check if an EPR external interrupt happened this bit is typically
214 	 * set if we need to handle another "edge" interrupt from within the
215 	 * MPIC "EPR" handler.
216 	 */
217 	if (happened & PACA_IRQ_EE_EDGE) {
218 		local_paca->irq_happened &= ~PACA_IRQ_EE_EDGE;
219 		return 0x500;
220 	}
221 
222 	if (happened & PACA_IRQ_DBELL) {
223 		local_paca->irq_happened &= ~PACA_IRQ_DBELL;
224 		return 0x280;
225 	}
226 #else
227 	if (happened & PACA_IRQ_DBELL) {
228 		local_paca->irq_happened &= ~PACA_IRQ_DBELL;
229 		return 0xa00;
230 	}
231 #endif /* CONFIG_PPC_BOOK3E */
232 
233 	/* There should be nothing left ! */
234 	BUG_ON(local_paca->irq_happened != 0);
235 
236 	return 0;
237 }
238 
239 notrace void arch_local_irq_restore(unsigned long mask)
240 {
241 	unsigned char irq_happened;
242 	unsigned int replay;
243 
244 	/* Write the new soft-enabled value */
245 	irq_soft_mask_set(mask);
246 	if (mask)
247 		return;
248 
249 	/*
250 	 * From this point onward, we can take interrupts, preempt,
251 	 * etc... unless we got hard-disabled. We check if an event
252 	 * happened. If none happened, we know we can just return.
253 	 *
254 	 * We may have preempted before the check below, in which case
255 	 * we are checking the "new" CPU instead of the old one. This
256 	 * is only a problem if an event happened on the "old" CPU.
257 	 *
258 	 * External interrupt events will have caused interrupts to
259 	 * be hard-disabled, so there is no problem, we
260 	 * cannot have preempted.
261 	 */
262 	irq_happened = get_irq_happened();
263 	if (!irq_happened) {
264 		/*
265 		 * FIXME. Here we'd like to be able to do:
266 		 *
267 		 * #ifdef CONFIG_PPC_IRQ_SOFT_MASK_DEBUG
268 		 *   WARN_ON(!(mfmsr() & MSR_EE));
269 		 * #endif
270 		 *
271 		 * But currently it hits in a few paths, we should fix those and
272 		 * enable the warning.
273 		 */
274 		return;
275 	}
276 
277 	/*
278 	 * We need to hard disable to get a trusted value from
279 	 * __check_irq_replay(). We also need to soft-disable
280 	 * again to avoid warnings in there due to the use of
281 	 * per-cpu variables.
282 	 */
283 	if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
284 #ifdef CONFIG_PPC_IRQ_SOFT_MASK_DEBUG
285 		WARN_ON(!(mfmsr() & MSR_EE));
286 #endif
287 		__hard_irq_disable();
288 #ifdef CONFIG_PPC_IRQ_SOFT_MASK_DEBUG
289 	} else {
290 		/*
291 		 * We should already be hard disabled here. We had bugs
292 		 * where that wasn't the case so let's dbl check it and
293 		 * warn if we are wrong. Only do that when IRQ tracing
294 		 * is enabled as mfmsr() can be costly.
295 		 */
296 		if (WARN_ON(mfmsr() & MSR_EE))
297 			__hard_irq_disable();
298 #endif
299 	}
300 
301 	irq_soft_mask_set(IRQS_ALL_DISABLED);
302 	trace_hardirqs_off();
303 
304 	/*
305 	 * Check if anything needs to be re-emitted. We haven't
306 	 * soft-enabled yet to avoid warnings in decrementer_check_overflow
307 	 * accessing per-cpu variables
308 	 */
309 	replay = __check_irq_replay();
310 
311 	/* We can soft-enable now */
312 	trace_hardirqs_on();
313 	irq_soft_mask_set(IRQS_ENABLED);
314 
315 	/*
316 	 * And replay if we have to. This will return with interrupts
317 	 * hard-enabled.
318 	 */
319 	if (replay) {
320 		__replay_interrupt(replay);
321 		return;
322 	}
323 
324 	/* Finally, let's ensure we are hard enabled */
325 	__hard_irq_enable();
326 }
327 EXPORT_SYMBOL(arch_local_irq_restore);
328 
329 /*
330  * This is specifically called by assembly code to re-enable interrupts
331  * if they are currently disabled. This is typically called before
332  * schedule() or do_signal() when returning to userspace. We do it
333  * in C to avoid the burden of dealing with lockdep etc...
334  *
335  * NOTE: This is called with interrupts hard disabled but not marked
336  * as such in paca->irq_happened, so we need to resync this.
337  */
338 void notrace restore_interrupts(void)
339 {
340 	if (irqs_disabled()) {
341 		local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
342 		local_irq_enable();
343 	} else
344 		__hard_irq_enable();
345 }
346 
347 /*
348  * This is a helper to use when about to go into idle low-power
349  * when the latter has the side effect of re-enabling interrupts
350  * (such as calling H_CEDE under pHyp).
351  *
352  * You call this function with interrupts soft-disabled (this is
353  * already the case when ppc_md.power_save is called). The function
354  * will return whether to enter power save or just return.
355  *
356  * In the former case, it will have notified lockdep of interrupts
357  * being re-enabled and generally sanitized the lazy irq state,
358  * and in the latter case it will leave with interrupts hard
359  * disabled and marked as such, so the local_irq_enable() call
360  * in arch_cpu_idle() will properly re-enable everything.
361  */
362 bool prep_irq_for_idle(void)
363 {
364 	/*
365 	 * First we need to hard disable to ensure no interrupt
366 	 * occurs before we effectively enter the low power state
367 	 */
368 	__hard_irq_disable();
369 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
370 
371 	/*
372 	 * If anything happened while we were soft-disabled,
373 	 * we return now and do not enter the low power state.
374 	 */
375 	if (lazy_irq_pending())
376 		return false;
377 
378 	/* Tell lockdep we are about to re-enable */
379 	trace_hardirqs_on();
380 
381 	/*
382 	 * Mark interrupts as soft-enabled and clear the
383 	 * PACA_IRQ_HARD_DIS from the pending mask since we
384 	 * are about to hard enable as well as a side effect
385 	 * of entering the low power state.
386 	 */
387 	local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
388 	irq_soft_mask_set(IRQS_ENABLED);
389 
390 	/* Tell the caller to enter the low power state */
391 	return true;
392 }
393 
394 #ifdef CONFIG_PPC_BOOK3S
395 /*
396  * This is for idle sequences that return with IRQs off, but the
397  * idle state itself wakes on interrupt. Tell the irq tracer that
398  * IRQs are enabled for the duration of idle so it does not get long
399  * off times. Must be paired with fini_irq_for_idle_irqsoff.
400  */
401 bool prep_irq_for_idle_irqsoff(void)
402 {
403 	WARN_ON(!irqs_disabled());
404 
405 	/*
406 	 * First we need to hard disable to ensure no interrupt
407 	 * occurs before we effectively enter the low power state
408 	 */
409 	__hard_irq_disable();
410 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
411 
412 	/*
413 	 * If anything happened while we were soft-disabled,
414 	 * we return now and do not enter the low power state.
415 	 */
416 	if (lazy_irq_pending())
417 		return false;
418 
419 	/* Tell lockdep we are about to re-enable */
420 	trace_hardirqs_on();
421 
422 	return true;
423 }
424 
425 /*
426  * Take the SRR1 wakeup reason, index into this table to find the
427  * appropriate irq_happened bit.
428  *
429  * Sytem reset exceptions taken in idle state also come through here,
430  * but they are NMI interrupts so do not need to wait for IRQs to be
431  * restored, and should be taken as early as practical. These are marked
432  * with 0xff in the table. The Power ISA specifies 0100b as the system
433  * reset interrupt reason.
434  */
435 #define IRQ_SYSTEM_RESET	0xff
436 
437 static const u8 srr1_to_lazyirq[0x10] = {
438 	0, 0, 0,
439 	PACA_IRQ_DBELL,
440 	IRQ_SYSTEM_RESET,
441 	PACA_IRQ_DBELL,
442 	PACA_IRQ_DEC,
443 	0,
444 	PACA_IRQ_EE,
445 	PACA_IRQ_EE,
446 	PACA_IRQ_HMI,
447 	0, 0, 0, 0, 0 };
448 
449 void replay_system_reset(void)
450 {
451 	struct pt_regs regs;
452 
453 	ppc_save_regs(&regs);
454 	regs.trap = 0x100;
455 	get_paca()->in_nmi = 1;
456 	system_reset_exception(&regs);
457 	get_paca()->in_nmi = 0;
458 }
459 EXPORT_SYMBOL_GPL(replay_system_reset);
460 
461 void irq_set_pending_from_srr1(unsigned long srr1)
462 {
463 	unsigned int idx = (srr1 & SRR1_WAKEMASK_P8) >> 18;
464 	u8 reason = srr1_to_lazyirq[idx];
465 
466 	/*
467 	 * Take the system reset now, which is immediately after registers
468 	 * are restored from idle. It's an NMI, so interrupts need not be
469 	 * re-enabled before it is taken.
470 	 */
471 	if (unlikely(reason == IRQ_SYSTEM_RESET)) {
472 		replay_system_reset();
473 		return;
474 	}
475 
476 	/*
477 	 * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
478 	 * so this can be called unconditionally with the SRR1 wake
479 	 * reason as returned by the idle code, which uses 0 to mean no
480 	 * interrupt.
481 	 *
482 	 * If a future CPU was to designate this as an interrupt reason,
483 	 * then a new index for no interrupt must be assigned.
484 	 */
485 	local_paca->irq_happened |= reason;
486 }
487 #endif /* CONFIG_PPC_BOOK3S */
488 
489 /*
490  * Force a replay of the external interrupt handler on this CPU.
491  */
492 void force_external_irq_replay(void)
493 {
494 	/*
495 	 * This must only be called with interrupts soft-disabled,
496 	 * the replay will happen when re-enabling.
497 	 */
498 	WARN_ON(!arch_irqs_disabled());
499 
500 	/*
501 	 * Interrupts must always be hard disabled before irq_happened is
502 	 * modified (to prevent lost update in case of interrupt between
503 	 * load and store).
504 	 */
505 	__hard_irq_disable();
506 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
507 
508 	/* Indicate in the PACA that we have an interrupt to replay */
509 	local_paca->irq_happened |= PACA_IRQ_EE;
510 }
511 
512 #endif /* CONFIG_PPC64 */
513 
514 int arch_show_interrupts(struct seq_file *p, int prec)
515 {
516 	int j;
517 
518 #if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
519 	if (tau_initialized) {
520 		seq_printf(p, "%*s: ", prec, "TAU");
521 		for_each_online_cpu(j)
522 			seq_printf(p, "%10u ", tau_interrupts(j));
523 		seq_puts(p, "  PowerPC             Thermal Assist (cpu temp)\n");
524 	}
525 #endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
526 
527 	seq_printf(p, "%*s: ", prec, "LOC");
528 	for_each_online_cpu(j)
529 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_event);
530         seq_printf(p, "  Local timer interrupts for timer event device\n");
531 
532 	seq_printf(p, "%*s: ", prec, "BCT");
533 	for_each_online_cpu(j)
534 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).broadcast_irqs_event);
535 	seq_printf(p, "  Broadcast timer interrupts for timer event device\n");
536 
537 	seq_printf(p, "%*s: ", prec, "LOC");
538 	for_each_online_cpu(j)
539 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_others);
540         seq_printf(p, "  Local timer interrupts for others\n");
541 
542 	seq_printf(p, "%*s: ", prec, "SPU");
543 	for_each_online_cpu(j)
544 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
545 	seq_printf(p, "  Spurious interrupts\n");
546 
547 	seq_printf(p, "%*s: ", prec, "PMI");
548 	for_each_online_cpu(j)
549 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
550 	seq_printf(p, "  Performance monitoring interrupts\n");
551 
552 	seq_printf(p, "%*s: ", prec, "MCE");
553 	for_each_online_cpu(j)
554 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
555 	seq_printf(p, "  Machine check exceptions\n");
556 
557 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
558 		seq_printf(p, "%*s: ", prec, "HMI");
559 		for_each_online_cpu(j)
560 			seq_printf(p, "%10u ",
561 					per_cpu(irq_stat, j).hmi_exceptions);
562 		seq_printf(p, "  Hypervisor Maintenance Interrupts\n");
563 	}
564 
565 	seq_printf(p, "%*s: ", prec, "NMI");
566 	for_each_online_cpu(j)
567 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).sreset_irqs);
568 	seq_printf(p, "  System Reset interrupts\n");
569 
570 #ifdef CONFIG_PPC_WATCHDOG
571 	seq_printf(p, "%*s: ", prec, "WDG");
572 	for_each_online_cpu(j)
573 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).soft_nmi_irqs);
574 	seq_printf(p, "  Watchdog soft-NMI interrupts\n");
575 #endif
576 
577 #ifdef CONFIG_PPC_DOORBELL
578 	if (cpu_has_feature(CPU_FTR_DBELL)) {
579 		seq_printf(p, "%*s: ", prec, "DBL");
580 		for_each_online_cpu(j)
581 			seq_printf(p, "%10u ", per_cpu(irq_stat, j).doorbell_irqs);
582 		seq_printf(p, "  Doorbell interrupts\n");
583 	}
584 #endif
585 
586 	return 0;
587 }
588 
589 /*
590  * /proc/stat helpers
591  */
592 u64 arch_irq_stat_cpu(unsigned int cpu)
593 {
594 	u64 sum = per_cpu(irq_stat, cpu).timer_irqs_event;
595 
596 	sum += per_cpu(irq_stat, cpu).broadcast_irqs_event;
597 	sum += per_cpu(irq_stat, cpu).pmu_irqs;
598 	sum += per_cpu(irq_stat, cpu).mce_exceptions;
599 	sum += per_cpu(irq_stat, cpu).spurious_irqs;
600 	sum += per_cpu(irq_stat, cpu).timer_irqs_others;
601 	sum += per_cpu(irq_stat, cpu).hmi_exceptions;
602 	sum += per_cpu(irq_stat, cpu).sreset_irqs;
603 #ifdef CONFIG_PPC_WATCHDOG
604 	sum += per_cpu(irq_stat, cpu).soft_nmi_irqs;
605 #endif
606 #ifdef CONFIG_PPC_DOORBELL
607 	sum += per_cpu(irq_stat, cpu).doorbell_irqs;
608 #endif
609 
610 	return sum;
611 }
612 
613 static inline void check_stack_overflow(void)
614 {
615 #ifdef CONFIG_DEBUG_STACKOVERFLOW
616 	long sp;
617 
618 	sp = current_stack_pointer() & (THREAD_SIZE-1);
619 
620 	/* check for stack overflow: is there less than 2KB free? */
621 	if (unlikely(sp < 2048)) {
622 		pr_err("do_IRQ: stack overflow: %ld\n", sp);
623 		dump_stack();
624 	}
625 #endif
626 }
627 
628 void __do_irq(struct pt_regs *regs)
629 {
630 	unsigned int irq;
631 
632 	irq_enter();
633 
634 	trace_irq_entry(regs);
635 
636 	check_stack_overflow();
637 
638 	/*
639 	 * Query the platform PIC for the interrupt & ack it.
640 	 *
641 	 * This will typically lower the interrupt line to the CPU
642 	 */
643 	irq = ppc_md.get_irq();
644 
645 	/* We can hard enable interrupts now to allow perf interrupts */
646 	may_hard_irq_enable();
647 
648 	/* And finally process it */
649 	if (unlikely(!irq))
650 		__this_cpu_inc(irq_stat.spurious_irqs);
651 	else
652 		generic_handle_irq(irq);
653 
654 	trace_irq_exit(regs);
655 
656 	irq_exit();
657 }
658 
659 void do_IRQ(struct pt_regs *regs)
660 {
661 	struct pt_regs *old_regs = set_irq_regs(regs);
662 	void *cursp, *irqsp, *sirqsp;
663 
664 	/* Switch to the irq stack to handle this */
665 	cursp = (void *)(current_stack_pointer() & ~(THREAD_SIZE - 1));
666 	irqsp = hardirq_ctx[raw_smp_processor_id()];
667 	sirqsp = softirq_ctx[raw_smp_processor_id()];
668 
669 	/* Already there ? */
670 	if (unlikely(cursp == irqsp || cursp == sirqsp)) {
671 		__do_irq(regs);
672 		set_irq_regs(old_regs);
673 		return;
674 	}
675 	/* Switch stack and call */
676 	call_do_irq(regs, irqsp);
677 
678 	set_irq_regs(old_regs);
679 }
680 
681 void __init init_IRQ(void)
682 {
683 	if (ppc_md.init_IRQ)
684 		ppc_md.init_IRQ();
685 }
686 
687 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
688 void   *critirq_ctx[NR_CPUS] __read_mostly;
689 void    *dbgirq_ctx[NR_CPUS] __read_mostly;
690 void *mcheckirq_ctx[NR_CPUS] __read_mostly;
691 #endif
692 
693 void *softirq_ctx[NR_CPUS] __read_mostly;
694 void *hardirq_ctx[NR_CPUS] __read_mostly;
695 
696 void do_softirq_own_stack(void)
697 {
698 	call_do_softirq(softirq_ctx[smp_processor_id()]);
699 }
700 
701 irq_hw_number_t virq_to_hw(unsigned int virq)
702 {
703 	struct irq_data *irq_data = irq_get_irq_data(virq);
704 	return WARN_ON(!irq_data) ? 0 : irq_data->hwirq;
705 }
706 EXPORT_SYMBOL_GPL(virq_to_hw);
707 
708 #ifdef CONFIG_SMP
709 int irq_choose_cpu(const struct cpumask *mask)
710 {
711 	int cpuid;
712 
713 	if (cpumask_equal(mask, cpu_online_mask)) {
714 		static int irq_rover;
715 		static DEFINE_RAW_SPINLOCK(irq_rover_lock);
716 		unsigned long flags;
717 
718 		/* Round-robin distribution... */
719 do_round_robin:
720 		raw_spin_lock_irqsave(&irq_rover_lock, flags);
721 
722 		irq_rover = cpumask_next(irq_rover, cpu_online_mask);
723 		if (irq_rover >= nr_cpu_ids)
724 			irq_rover = cpumask_first(cpu_online_mask);
725 
726 		cpuid = irq_rover;
727 
728 		raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
729 	} else {
730 		cpuid = cpumask_first_and(mask, cpu_online_mask);
731 		if (cpuid >= nr_cpu_ids)
732 			goto do_round_robin;
733 	}
734 
735 	return get_hard_smp_processor_id(cpuid);
736 }
737 #else
738 int irq_choose_cpu(const struct cpumask *mask)
739 {
740 	return hard_smp_processor_id();
741 }
742 #endif
743 
744 #ifdef CONFIG_PPC64
745 static int __init setup_noirqdistrib(char *str)
746 {
747 	distribute_irqs = 0;
748 	return 1;
749 }
750 
751 __setup("noirqdistrib", setup_noirqdistrib);
752 #endif /* CONFIG_PPC64 */
753