xref: /openbmc/linux/arch/powerpc/kernel/irq.c (revision 7d94ab16)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  Derived from arch/i386/kernel/irq.c
4  *    Copyright (C) 1992 Linus Torvalds
5  *  Adapted from arch/i386 by Gary Thomas
6  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
7  *  Updated and modified by Cort Dougan <cort@fsmlabs.com>
8  *    Copyright (C) 1996-2001 Cort Dougan
9  *  Adapted for Power Macintosh by Paul Mackerras
10  *    Copyright (C) 1996 Paul Mackerras (paulus@cs.anu.edu.au)
11  *
12  * This file contains the code used by various IRQ handling routines:
13  * asking for different IRQ's should be done through these routines
14  * instead of just grabbing them. Thus setups with different IRQ numbers
15  * shouldn't result in any weird surprises, and installing new handlers
16  * should be easier.
17  *
18  * The MPC8xx has an interrupt mask in the SIU.  If a bit is set, the
19  * interrupt is _enabled_.  As expected, IRQ0 is bit 0 in the 32-bit
20  * mask register (of which only 16 are defined), hence the weird shifting
21  * and complement of the cached_irq_mask.  I want to be able to stuff
22  * this right into the SIU SMASK register.
23  * Many of the prep/chrp functions are conditional compiled on CONFIG_PPC_8xx
24  * to reduce code space and undefined function references.
25  */
26 
27 #undef DEBUG
28 
29 #include <linux/export.h>
30 #include <linux/threads.h>
31 #include <linux/kernel_stat.h>
32 #include <linux/signal.h>
33 #include <linux/sched.h>
34 #include <linux/ptrace.h>
35 #include <linux/ioport.h>
36 #include <linux/interrupt.h>
37 #include <linux/timex.h>
38 #include <linux/init.h>
39 #include <linux/slab.h>
40 #include <linux/delay.h>
41 #include <linux/irq.h>
42 #include <linux/seq_file.h>
43 #include <linux/cpumask.h>
44 #include <linux/profile.h>
45 #include <linux/bitops.h>
46 #include <linux/list.h>
47 #include <linux/radix-tree.h>
48 #include <linux/mutex.h>
49 #include <linux/pci.h>
50 #include <linux/debugfs.h>
51 #include <linux/of.h>
52 #include <linux/of_irq.h>
53 #include <linux/vmalloc.h>
54 
55 #include <linux/uaccess.h>
56 #include <asm/io.h>
57 #include <asm/pgtable.h>
58 #include <asm/irq.h>
59 #include <asm/cache.h>
60 #include <asm/prom.h>
61 #include <asm/ptrace.h>
62 #include <asm/machdep.h>
63 #include <asm/udbg.h>
64 #include <asm/smp.h>
65 #include <asm/livepatch.h>
66 #include <asm/asm-prototypes.h>
67 #include <asm/hw_irq.h>
68 
69 #ifdef CONFIG_PPC64
70 #include <asm/paca.h>
71 #include <asm/firmware.h>
72 #include <asm/lv1call.h>
73 #endif
74 #define CREATE_TRACE_POINTS
75 #include <asm/trace.h>
76 #include <asm/cpu_has_feature.h>
77 
78 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
79 EXPORT_PER_CPU_SYMBOL(irq_stat);
80 
81 #ifdef CONFIG_PPC32
82 atomic_t ppc_n_lost_interrupts;
83 
84 #ifdef CONFIG_TAU_INT
85 extern int tau_initialized;
86 u32 tau_interrupts(unsigned long cpu);
87 #endif
88 #endif /* CONFIG_PPC32 */
89 
90 #ifdef CONFIG_PPC64
91 
92 int distribute_irqs = 1;
93 
94 static inline notrace unsigned long get_irq_happened(void)
95 {
96 	unsigned long happened;
97 
98 	__asm__ __volatile__("lbz %0,%1(13)"
99 	: "=r" (happened) : "i" (offsetof(struct paca_struct, irq_happened)));
100 
101 	return happened;
102 }
103 
104 static inline notrace int decrementer_check_overflow(void)
105 {
106  	u64 now = get_tb_or_rtc();
107 	u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
108 
109 	return now >= *next_tb;
110 }
111 
112 /* This is called whenever we are re-enabling interrupts
113  * and returns either 0 (nothing to do) or 500/900/280/a00/e80 if
114  * there's an EE, DEC or DBELL to generate.
115  *
116  * This is called in two contexts: From arch_local_irq_restore()
117  * before soft-enabling interrupts, and from the exception exit
118  * path when returning from an interrupt from a soft-disabled to
119  * a soft enabled context. In both case we have interrupts hard
120  * disabled.
121  *
122  * We take care of only clearing the bits we handled in the
123  * PACA irq_happened field since we can only re-emit one at a
124  * time and we don't want to "lose" one.
125  */
126 notrace unsigned int __check_irq_replay(void)
127 {
128 	/*
129 	 * We use local_paca rather than get_paca() to avoid all
130 	 * the debug_smp_processor_id() business in this low level
131 	 * function
132 	 */
133 	unsigned char happened = local_paca->irq_happened;
134 
135 	/*
136 	 * We are responding to the next interrupt, so interrupt-off
137 	 * latencies should be reset here.
138 	 */
139 	trace_hardirqs_on();
140 	trace_hardirqs_off();
141 
142 	/*
143 	 * We are always hard disabled here, but PACA_IRQ_HARD_DIS may
144 	 * not be set, which means interrupts have only just been hard
145 	 * disabled as part of the local_irq_restore or interrupt return
146 	 * code. In that case, skip the decrementr check becaus it's
147 	 * expensive to read the TB.
148 	 *
149 	 * HARD_DIS then gets cleared here, but it's reconciled later.
150 	 * Either local_irq_disable will replay the interrupt and that
151 	 * will reconcile state like other hard interrupts. Or interrupt
152 	 * retur will replay the interrupt and in that case it sets
153 	 * PACA_IRQ_HARD_DIS by hand (see comments in entry_64.S).
154 	 */
155 	if (happened & PACA_IRQ_HARD_DIS) {
156 		local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
157 
158 		/*
159 		 * We may have missed a decrementer interrupt if hard disabled.
160 		 * Check the decrementer register in case we had a rollover
161 		 * while hard disabled.
162 		 */
163 		if (!(happened & PACA_IRQ_DEC)) {
164 			if (decrementer_check_overflow()) {
165 				local_paca->irq_happened |= PACA_IRQ_DEC;
166 				happened |= PACA_IRQ_DEC;
167 			}
168 		}
169 	}
170 
171 	/*
172 	 * Force the delivery of pending soft-disabled interrupts on PS3.
173 	 * Any HV call will have this side effect.
174 	 */
175 	if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
176 		u64 tmp, tmp2;
177 		lv1_get_version_info(&tmp, &tmp2);
178 	}
179 
180 	/*
181 	 * Check if an hypervisor Maintenance interrupt happened.
182 	 * This is a higher priority interrupt than the others, so
183 	 * replay it first.
184 	 */
185 	if (happened & PACA_IRQ_HMI) {
186 		local_paca->irq_happened &= ~PACA_IRQ_HMI;
187 		return 0xe60;
188 	}
189 
190 	if (happened & PACA_IRQ_DEC) {
191 		local_paca->irq_happened &= ~PACA_IRQ_DEC;
192 		return 0x900;
193 	}
194 
195 	if (happened & PACA_IRQ_PMI) {
196 		local_paca->irq_happened &= ~PACA_IRQ_PMI;
197 		return 0xf00;
198 	}
199 
200 	if (happened & PACA_IRQ_EE) {
201 		local_paca->irq_happened &= ~PACA_IRQ_EE;
202 		return 0x500;
203 	}
204 
205 #ifdef CONFIG_PPC_BOOK3E
206 	/*
207 	 * Check if an EPR external interrupt happened this bit is typically
208 	 * set if we need to handle another "edge" interrupt from within the
209 	 * MPIC "EPR" handler.
210 	 */
211 	if (happened & PACA_IRQ_EE_EDGE) {
212 		local_paca->irq_happened &= ~PACA_IRQ_EE_EDGE;
213 		return 0x500;
214 	}
215 
216 	if (happened & PACA_IRQ_DBELL) {
217 		local_paca->irq_happened &= ~PACA_IRQ_DBELL;
218 		return 0x280;
219 	}
220 #else
221 	if (happened & PACA_IRQ_DBELL) {
222 		local_paca->irq_happened &= ~PACA_IRQ_DBELL;
223 		return 0xa00;
224 	}
225 #endif /* CONFIG_PPC_BOOK3E */
226 
227 	/* There should be nothing left ! */
228 	BUG_ON(local_paca->irq_happened != 0);
229 
230 	return 0;
231 }
232 
233 notrace void arch_local_irq_restore(unsigned long mask)
234 {
235 	unsigned char irq_happened;
236 	unsigned int replay;
237 
238 	/* Write the new soft-enabled value */
239 	irq_soft_mask_set(mask);
240 	if (mask)
241 		return;
242 
243 	/*
244 	 * From this point onward, we can take interrupts, preempt,
245 	 * etc... unless we got hard-disabled. We check if an event
246 	 * happened. If none happened, we know we can just return.
247 	 *
248 	 * We may have preempted before the check below, in which case
249 	 * we are checking the "new" CPU instead of the old one. This
250 	 * is only a problem if an event happened on the "old" CPU.
251 	 *
252 	 * External interrupt events will have caused interrupts to
253 	 * be hard-disabled, so there is no problem, we
254 	 * cannot have preempted.
255 	 */
256 	irq_happened = get_irq_happened();
257 	if (!irq_happened) {
258 #ifdef CONFIG_PPC_IRQ_SOFT_MASK_DEBUG
259 		WARN_ON_ONCE(!(mfmsr() & MSR_EE));
260 #endif
261 		return;
262 	}
263 
264 	/*
265 	 * We need to hard disable to get a trusted value from
266 	 * __check_irq_replay(). We also need to soft-disable
267 	 * again to avoid warnings in there due to the use of
268 	 * per-cpu variables.
269 	 */
270 	if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
271 #ifdef CONFIG_PPC_IRQ_SOFT_MASK_DEBUG
272 		WARN_ON_ONCE(!(mfmsr() & MSR_EE));
273 #endif
274 		__hard_irq_disable();
275 #ifdef CONFIG_PPC_IRQ_SOFT_MASK_DEBUG
276 	} else {
277 		/*
278 		 * We should already be hard disabled here. We had bugs
279 		 * where that wasn't the case so let's dbl check it and
280 		 * warn if we are wrong. Only do that when IRQ tracing
281 		 * is enabled as mfmsr() can be costly.
282 		 */
283 		if (WARN_ON_ONCE(mfmsr() & MSR_EE))
284 			__hard_irq_disable();
285 #endif
286 	}
287 
288 	irq_soft_mask_set(IRQS_ALL_DISABLED);
289 	trace_hardirqs_off();
290 
291 	/*
292 	 * Check if anything needs to be re-emitted. We haven't
293 	 * soft-enabled yet to avoid warnings in decrementer_check_overflow
294 	 * accessing per-cpu variables
295 	 */
296 	replay = __check_irq_replay();
297 
298 	/* We can soft-enable now */
299 	trace_hardirqs_on();
300 	irq_soft_mask_set(IRQS_ENABLED);
301 
302 	/*
303 	 * And replay if we have to. This will return with interrupts
304 	 * hard-enabled.
305 	 */
306 	if (replay) {
307 		__replay_interrupt(replay);
308 		return;
309 	}
310 
311 	/* Finally, let's ensure we are hard enabled */
312 	__hard_irq_enable();
313 }
314 EXPORT_SYMBOL(arch_local_irq_restore);
315 
316 /*
317  * This is specifically called by assembly code to re-enable interrupts
318  * if they are currently disabled. This is typically called before
319  * schedule() or do_signal() when returning to userspace. We do it
320  * in C to avoid the burden of dealing with lockdep etc...
321  *
322  * NOTE: This is called with interrupts hard disabled but not marked
323  * as such in paca->irq_happened, so we need to resync this.
324  */
325 void notrace restore_interrupts(void)
326 {
327 	if (irqs_disabled()) {
328 		local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
329 		local_irq_enable();
330 	} else
331 		__hard_irq_enable();
332 }
333 
334 /*
335  * This is a helper to use when about to go into idle low-power
336  * when the latter has the side effect of re-enabling interrupts
337  * (such as calling H_CEDE under pHyp).
338  *
339  * You call this function with interrupts soft-disabled (this is
340  * already the case when ppc_md.power_save is called). The function
341  * will return whether to enter power save or just return.
342  *
343  * In the former case, it will have notified lockdep of interrupts
344  * being re-enabled and generally sanitized the lazy irq state,
345  * and in the latter case it will leave with interrupts hard
346  * disabled and marked as such, so the local_irq_enable() call
347  * in arch_cpu_idle() will properly re-enable everything.
348  */
349 bool prep_irq_for_idle(void)
350 {
351 	/*
352 	 * First we need to hard disable to ensure no interrupt
353 	 * occurs before we effectively enter the low power state
354 	 */
355 	__hard_irq_disable();
356 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
357 
358 	/*
359 	 * If anything happened while we were soft-disabled,
360 	 * we return now and do not enter the low power state.
361 	 */
362 	if (lazy_irq_pending())
363 		return false;
364 
365 	/* Tell lockdep we are about to re-enable */
366 	trace_hardirqs_on();
367 
368 	/*
369 	 * Mark interrupts as soft-enabled and clear the
370 	 * PACA_IRQ_HARD_DIS from the pending mask since we
371 	 * are about to hard enable as well as a side effect
372 	 * of entering the low power state.
373 	 */
374 	local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
375 	irq_soft_mask_set(IRQS_ENABLED);
376 
377 	/* Tell the caller to enter the low power state */
378 	return true;
379 }
380 
381 #ifdef CONFIG_PPC_BOOK3S
382 /*
383  * This is for idle sequences that return with IRQs off, but the
384  * idle state itself wakes on interrupt. Tell the irq tracer that
385  * IRQs are enabled for the duration of idle so it does not get long
386  * off times. Must be paired with fini_irq_for_idle_irqsoff.
387  */
388 bool prep_irq_for_idle_irqsoff(void)
389 {
390 	WARN_ON(!irqs_disabled());
391 
392 	/*
393 	 * First we need to hard disable to ensure no interrupt
394 	 * occurs before we effectively enter the low power state
395 	 */
396 	__hard_irq_disable();
397 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
398 
399 	/*
400 	 * If anything happened while we were soft-disabled,
401 	 * we return now and do not enter the low power state.
402 	 */
403 	if (lazy_irq_pending())
404 		return false;
405 
406 	/* Tell lockdep we are about to re-enable */
407 	trace_hardirqs_on();
408 
409 	return true;
410 }
411 
412 /*
413  * Take the SRR1 wakeup reason, index into this table to find the
414  * appropriate irq_happened bit.
415  *
416  * Sytem reset exceptions taken in idle state also come through here,
417  * but they are NMI interrupts so do not need to wait for IRQs to be
418  * restored, and should be taken as early as practical. These are marked
419  * with 0xff in the table. The Power ISA specifies 0100b as the system
420  * reset interrupt reason.
421  */
422 #define IRQ_SYSTEM_RESET	0xff
423 
424 static const u8 srr1_to_lazyirq[0x10] = {
425 	0, 0, 0,
426 	PACA_IRQ_DBELL,
427 	IRQ_SYSTEM_RESET,
428 	PACA_IRQ_DBELL,
429 	PACA_IRQ_DEC,
430 	0,
431 	PACA_IRQ_EE,
432 	PACA_IRQ_EE,
433 	PACA_IRQ_HMI,
434 	0, 0, 0, 0, 0 };
435 
436 void replay_system_reset(void)
437 {
438 	struct pt_regs regs;
439 
440 	ppc_save_regs(&regs);
441 	regs.trap = 0x100;
442 	get_paca()->in_nmi = 1;
443 	system_reset_exception(&regs);
444 	get_paca()->in_nmi = 0;
445 }
446 EXPORT_SYMBOL_GPL(replay_system_reset);
447 
448 void irq_set_pending_from_srr1(unsigned long srr1)
449 {
450 	unsigned int idx = (srr1 & SRR1_WAKEMASK_P8) >> 18;
451 	u8 reason = srr1_to_lazyirq[idx];
452 
453 	/*
454 	 * Take the system reset now, which is immediately after registers
455 	 * are restored from idle. It's an NMI, so interrupts need not be
456 	 * re-enabled before it is taken.
457 	 */
458 	if (unlikely(reason == IRQ_SYSTEM_RESET)) {
459 		replay_system_reset();
460 		return;
461 	}
462 
463 	/*
464 	 * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
465 	 * so this can be called unconditionally with the SRR1 wake
466 	 * reason as returned by the idle code, which uses 0 to mean no
467 	 * interrupt.
468 	 *
469 	 * If a future CPU was to designate this as an interrupt reason,
470 	 * then a new index for no interrupt must be assigned.
471 	 */
472 	local_paca->irq_happened |= reason;
473 }
474 #endif /* CONFIG_PPC_BOOK3S */
475 
476 /*
477  * Force a replay of the external interrupt handler on this CPU.
478  */
479 void force_external_irq_replay(void)
480 {
481 	/*
482 	 * This must only be called with interrupts soft-disabled,
483 	 * the replay will happen when re-enabling.
484 	 */
485 	WARN_ON(!arch_irqs_disabled());
486 
487 	/*
488 	 * Interrupts must always be hard disabled before irq_happened is
489 	 * modified (to prevent lost update in case of interrupt between
490 	 * load and store).
491 	 */
492 	__hard_irq_disable();
493 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
494 
495 	/* Indicate in the PACA that we have an interrupt to replay */
496 	local_paca->irq_happened |= PACA_IRQ_EE;
497 }
498 
499 #endif /* CONFIG_PPC64 */
500 
501 int arch_show_interrupts(struct seq_file *p, int prec)
502 {
503 	int j;
504 
505 #if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
506 	if (tau_initialized) {
507 		seq_printf(p, "%*s: ", prec, "TAU");
508 		for_each_online_cpu(j)
509 			seq_printf(p, "%10u ", tau_interrupts(j));
510 		seq_puts(p, "  PowerPC             Thermal Assist (cpu temp)\n");
511 	}
512 #endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
513 
514 	seq_printf(p, "%*s: ", prec, "LOC");
515 	for_each_online_cpu(j)
516 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_event);
517         seq_printf(p, "  Local timer interrupts for timer event device\n");
518 
519 	seq_printf(p, "%*s: ", prec, "BCT");
520 	for_each_online_cpu(j)
521 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).broadcast_irqs_event);
522 	seq_printf(p, "  Broadcast timer interrupts for timer event device\n");
523 
524 	seq_printf(p, "%*s: ", prec, "LOC");
525 	for_each_online_cpu(j)
526 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_others);
527         seq_printf(p, "  Local timer interrupts for others\n");
528 
529 	seq_printf(p, "%*s: ", prec, "SPU");
530 	for_each_online_cpu(j)
531 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
532 	seq_printf(p, "  Spurious interrupts\n");
533 
534 	seq_printf(p, "%*s: ", prec, "PMI");
535 	for_each_online_cpu(j)
536 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
537 	seq_printf(p, "  Performance monitoring interrupts\n");
538 
539 	seq_printf(p, "%*s: ", prec, "MCE");
540 	for_each_online_cpu(j)
541 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
542 	seq_printf(p, "  Machine check exceptions\n");
543 
544 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
545 		seq_printf(p, "%*s: ", prec, "HMI");
546 		for_each_online_cpu(j)
547 			seq_printf(p, "%10u ",
548 					per_cpu(irq_stat, j).hmi_exceptions);
549 		seq_printf(p, "  Hypervisor Maintenance Interrupts\n");
550 	}
551 
552 	seq_printf(p, "%*s: ", prec, "NMI");
553 	for_each_online_cpu(j)
554 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).sreset_irqs);
555 	seq_printf(p, "  System Reset interrupts\n");
556 
557 #ifdef CONFIG_PPC_WATCHDOG
558 	seq_printf(p, "%*s: ", prec, "WDG");
559 	for_each_online_cpu(j)
560 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).soft_nmi_irqs);
561 	seq_printf(p, "  Watchdog soft-NMI interrupts\n");
562 #endif
563 
564 #ifdef CONFIG_PPC_DOORBELL
565 	if (cpu_has_feature(CPU_FTR_DBELL)) {
566 		seq_printf(p, "%*s: ", prec, "DBL");
567 		for_each_online_cpu(j)
568 			seq_printf(p, "%10u ", per_cpu(irq_stat, j).doorbell_irqs);
569 		seq_printf(p, "  Doorbell interrupts\n");
570 	}
571 #endif
572 
573 	return 0;
574 }
575 
576 /*
577  * /proc/stat helpers
578  */
579 u64 arch_irq_stat_cpu(unsigned int cpu)
580 {
581 	u64 sum = per_cpu(irq_stat, cpu).timer_irqs_event;
582 
583 	sum += per_cpu(irq_stat, cpu).broadcast_irqs_event;
584 	sum += per_cpu(irq_stat, cpu).pmu_irqs;
585 	sum += per_cpu(irq_stat, cpu).mce_exceptions;
586 	sum += per_cpu(irq_stat, cpu).spurious_irqs;
587 	sum += per_cpu(irq_stat, cpu).timer_irqs_others;
588 	sum += per_cpu(irq_stat, cpu).hmi_exceptions;
589 	sum += per_cpu(irq_stat, cpu).sreset_irqs;
590 #ifdef CONFIG_PPC_WATCHDOG
591 	sum += per_cpu(irq_stat, cpu).soft_nmi_irqs;
592 #endif
593 #ifdef CONFIG_PPC_DOORBELL
594 	sum += per_cpu(irq_stat, cpu).doorbell_irqs;
595 #endif
596 
597 	return sum;
598 }
599 
600 static inline void check_stack_overflow(void)
601 {
602 #ifdef CONFIG_DEBUG_STACKOVERFLOW
603 	long sp;
604 
605 	sp = current_stack_pointer() & (THREAD_SIZE-1);
606 
607 	/* check for stack overflow: is there less than 2KB free? */
608 	if (unlikely(sp < 2048)) {
609 		pr_err("do_IRQ: stack overflow: %ld\n", sp);
610 		dump_stack();
611 	}
612 #endif
613 }
614 
615 void __do_irq(struct pt_regs *regs)
616 {
617 	unsigned int irq;
618 
619 	irq_enter();
620 
621 	trace_irq_entry(regs);
622 
623 	/*
624 	 * Query the platform PIC for the interrupt & ack it.
625 	 *
626 	 * This will typically lower the interrupt line to the CPU
627 	 */
628 	irq = ppc_md.get_irq();
629 
630 	/* We can hard enable interrupts now to allow perf interrupts */
631 	may_hard_irq_enable();
632 
633 	/* And finally process it */
634 	if (unlikely(!irq))
635 		__this_cpu_inc(irq_stat.spurious_irqs);
636 	else
637 		generic_handle_irq(irq);
638 
639 	trace_irq_exit(regs);
640 
641 	irq_exit();
642 }
643 
644 void do_IRQ(struct pt_regs *regs)
645 {
646 	struct pt_regs *old_regs = set_irq_regs(regs);
647 	void *cursp, *irqsp, *sirqsp;
648 
649 	/* Switch to the irq stack to handle this */
650 	cursp = (void *)(current_stack_pointer() & ~(THREAD_SIZE - 1));
651 	irqsp = hardirq_ctx[raw_smp_processor_id()];
652 	sirqsp = softirq_ctx[raw_smp_processor_id()];
653 
654 	check_stack_overflow();
655 
656 	/* Already there ? */
657 	if (unlikely(cursp == irqsp || cursp == sirqsp)) {
658 		__do_irq(regs);
659 		set_irq_regs(old_regs);
660 		return;
661 	}
662 	/* Switch stack and call */
663 	call_do_irq(regs, irqsp);
664 
665 	set_irq_regs(old_regs);
666 }
667 
668 static void *__init alloc_vm_stack(void)
669 {
670 	return __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN, VMALLOC_START,
671 				    VMALLOC_END, THREADINFO_GFP, PAGE_KERNEL,
672 				     0, NUMA_NO_NODE, (void*)_RET_IP_);
673 }
674 
675 static void __init vmap_irqstack_init(void)
676 {
677 	int i;
678 
679 	for_each_possible_cpu(i) {
680 		softirq_ctx[i] = alloc_vm_stack();
681 		hardirq_ctx[i] = alloc_vm_stack();
682 	}
683 }
684 
685 
686 void __init init_IRQ(void)
687 {
688 	if (IS_ENABLED(CONFIG_VMAP_STACK))
689 		vmap_irqstack_init();
690 
691 	if (ppc_md.init_IRQ)
692 		ppc_md.init_IRQ();
693 }
694 
695 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
696 void   *critirq_ctx[NR_CPUS] __read_mostly;
697 void    *dbgirq_ctx[NR_CPUS] __read_mostly;
698 void *mcheckirq_ctx[NR_CPUS] __read_mostly;
699 #endif
700 
701 void *softirq_ctx[NR_CPUS] __read_mostly;
702 void *hardirq_ctx[NR_CPUS] __read_mostly;
703 
704 void do_softirq_own_stack(void)
705 {
706 	call_do_softirq(softirq_ctx[smp_processor_id()]);
707 }
708 
709 irq_hw_number_t virq_to_hw(unsigned int virq)
710 {
711 	struct irq_data *irq_data = irq_get_irq_data(virq);
712 	return WARN_ON(!irq_data) ? 0 : irq_data->hwirq;
713 }
714 EXPORT_SYMBOL_GPL(virq_to_hw);
715 
716 #ifdef CONFIG_SMP
717 int irq_choose_cpu(const struct cpumask *mask)
718 {
719 	int cpuid;
720 
721 	if (cpumask_equal(mask, cpu_online_mask)) {
722 		static int irq_rover;
723 		static DEFINE_RAW_SPINLOCK(irq_rover_lock);
724 		unsigned long flags;
725 
726 		/* Round-robin distribution... */
727 do_round_robin:
728 		raw_spin_lock_irqsave(&irq_rover_lock, flags);
729 
730 		irq_rover = cpumask_next(irq_rover, cpu_online_mask);
731 		if (irq_rover >= nr_cpu_ids)
732 			irq_rover = cpumask_first(cpu_online_mask);
733 
734 		cpuid = irq_rover;
735 
736 		raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
737 	} else {
738 		cpuid = cpumask_first_and(mask, cpu_online_mask);
739 		if (cpuid >= nr_cpu_ids)
740 			goto do_round_robin;
741 	}
742 
743 	return get_hard_smp_processor_id(cpuid);
744 }
745 #else
746 int irq_choose_cpu(const struct cpumask *mask)
747 {
748 	return hard_smp_processor_id();
749 }
750 #endif
751 
752 #ifdef CONFIG_PPC64
753 static int __init setup_noirqdistrib(char *str)
754 {
755 	distribute_irqs = 0;
756 	return 1;
757 }
758 
759 __setup("noirqdistrib", setup_noirqdistrib);
760 #endif /* CONFIG_PPC64 */
761