xref: /openbmc/linux/arch/powerpc/kernel/irq.c (revision f5cc14e4)
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 #include <linux/pgtable.h>
55 
56 #include <linux/uaccess.h>
57 #include <asm/io.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 #include <asm/dbell.h>
74 #endif
75 #define CREATE_TRACE_POINTS
76 #include <asm/trace.h>
77 #include <asm/cpu_has_feature.h>
78 
79 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
80 EXPORT_PER_CPU_SYMBOL(irq_stat);
81 
82 #ifdef CONFIG_PPC32
83 atomic_t ppc_n_lost_interrupts;
84 
85 #ifdef CONFIG_TAU_INT
86 extern int tau_initialized;
87 u32 tau_interrupts(unsigned long cpu);
88 #endif
89 #endif /* CONFIG_PPC32 */
90 
91 #ifdef CONFIG_PPC64
92 
93 int distribute_irqs = 1;
94 
95 static inline notrace unsigned long get_irq_happened(void)
96 {
97 	unsigned long happened;
98 
99 	__asm__ __volatile__("lbz %0,%1(13)"
100 	: "=r" (happened) : "i" (offsetof(struct paca_struct, irq_happened)));
101 
102 	return happened;
103 }
104 
105 #ifdef CONFIG_PPC_BOOK3E
106 
107 /* This is called whenever we are re-enabling interrupts
108  * and returns either 0 (nothing to do) or 500/900/280 if
109  * there's an EE, DEC or DBELL to generate.
110  *
111  * This is called in two contexts: From arch_local_irq_restore()
112  * before soft-enabling interrupts, and from the exception exit
113  * path when returning from an interrupt from a soft-disabled to
114  * a soft enabled context. In both case we have interrupts hard
115  * disabled.
116  *
117  * We take care of only clearing the bits we handled in the
118  * PACA irq_happened field since we can only re-emit one at a
119  * time and we don't want to "lose" one.
120  */
121 notrace unsigned int __check_irq_replay(void)
122 {
123 	/*
124 	 * We use local_paca rather than get_paca() to avoid all
125 	 * the debug_smp_processor_id() business in this low level
126 	 * function
127 	 */
128 	unsigned char happened = local_paca->irq_happened;
129 
130 	/*
131 	 * We are responding to the next interrupt, so interrupt-off
132 	 * latencies should be reset here.
133 	 */
134 	trace_hardirqs_on();
135 	trace_hardirqs_off();
136 
137 	if (happened & PACA_IRQ_DEC) {
138 		local_paca->irq_happened &= ~PACA_IRQ_DEC;
139 		return 0x900;
140 	}
141 
142 	if (happened & PACA_IRQ_EE) {
143 		local_paca->irq_happened &= ~PACA_IRQ_EE;
144 		return 0x500;
145 	}
146 
147 	if (happened & PACA_IRQ_DBELL) {
148 		local_paca->irq_happened &= ~PACA_IRQ_DBELL;
149 		return 0x280;
150 	}
151 
152 	if (happened & PACA_IRQ_HARD_DIS)
153 		local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
154 
155 	/* There should be nothing left ! */
156 	BUG_ON(local_paca->irq_happened != 0);
157 
158 	return 0;
159 }
160 
161 /*
162  * This is specifically called by assembly code to re-enable interrupts
163  * if they are currently disabled. This is typically called before
164  * schedule() or do_signal() when returning to userspace. We do it
165  * in C to avoid the burden of dealing with lockdep etc...
166  *
167  * NOTE: This is called with interrupts hard disabled but not marked
168  * as such in paca->irq_happened, so we need to resync this.
169  */
170 void notrace restore_interrupts(void)
171 {
172 	if (irqs_disabled()) {
173 		local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
174 		local_irq_enable();
175 	} else
176 		__hard_irq_enable();
177 }
178 
179 #endif /* CONFIG_PPC_BOOK3E */
180 
181 void replay_soft_interrupts(void)
182 {
183 	/*
184 	 * We use local_paca rather than get_paca() to avoid all
185 	 * the debug_smp_processor_id() business in this low level
186 	 * function
187 	 */
188 	unsigned char happened = local_paca->irq_happened;
189 	struct pt_regs regs;
190 
191 	ppc_save_regs(&regs);
192 	regs.softe = IRQS_ENABLED;
193 
194 again:
195 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
196 		WARN_ON_ONCE(mfmsr() & MSR_EE);
197 
198 	/*
199 	 * Force the delivery of pending soft-disabled interrupts on PS3.
200 	 * Any HV call will have this side effect.
201 	 */
202 	if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
203 		u64 tmp, tmp2;
204 		lv1_get_version_info(&tmp, &tmp2);
205 	}
206 
207 	/*
208 	 * Check if an hypervisor Maintenance interrupt happened.
209 	 * This is a higher priority interrupt than the others, so
210 	 * replay it first.
211 	 */
212 	if (IS_ENABLED(CONFIG_PPC_BOOK3S) && (happened & PACA_IRQ_HMI)) {
213 		local_paca->irq_happened &= ~PACA_IRQ_HMI;
214 		regs.trap = 0xe60;
215 		handle_hmi_exception(&regs);
216 		if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
217 			hard_irq_disable();
218 	}
219 
220 	if (happened & PACA_IRQ_DEC) {
221 		local_paca->irq_happened &= ~PACA_IRQ_DEC;
222 		regs.trap = 0x900;
223 		timer_interrupt(&regs);
224 		if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
225 			hard_irq_disable();
226 	}
227 
228 	if (happened & PACA_IRQ_EE) {
229 		local_paca->irq_happened &= ~PACA_IRQ_EE;
230 		regs.trap = 0x500;
231 		do_IRQ(&regs);
232 		if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
233 			hard_irq_disable();
234 	}
235 
236 	if (IS_ENABLED(CONFIG_PPC_DOORBELL) && (happened & PACA_IRQ_DBELL)) {
237 		local_paca->irq_happened &= ~PACA_IRQ_DBELL;
238 		if (IS_ENABLED(CONFIG_PPC_BOOK3E))
239 			regs.trap = 0x280;
240 		else
241 			regs.trap = 0xa00;
242 		doorbell_exception(&regs);
243 		if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
244 			hard_irq_disable();
245 	}
246 
247 	/* Book3E does not support soft-masking PMI interrupts */
248 	if (IS_ENABLED(CONFIG_PPC_BOOK3S) && (happened & PACA_IRQ_PMI)) {
249 		local_paca->irq_happened &= ~PACA_IRQ_PMI;
250 		regs.trap = 0xf00;
251 		performance_monitor_exception(&regs);
252 		if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
253 			hard_irq_disable();
254 	}
255 
256 	happened = local_paca->irq_happened;
257 	if (happened & ~PACA_IRQ_HARD_DIS) {
258 		/*
259 		 * We are responding to the next interrupt, so interrupt-off
260 		 * latencies should be reset here.
261 		 */
262 		trace_hardirqs_on();
263 		trace_hardirqs_off();
264 		goto again;
265 	}
266 }
267 
268 notrace void arch_local_irq_restore(unsigned long mask)
269 {
270 	unsigned char irq_happened;
271 
272 	/* Write the new soft-enabled value */
273 	irq_soft_mask_set(mask);
274 	if (mask)
275 		return;
276 
277 	/*
278 	 * From this point onward, we can take interrupts, preempt,
279 	 * etc... unless we got hard-disabled. We check if an event
280 	 * happened. If none happened, we know we can just return.
281 	 *
282 	 * We may have preempted before the check below, in which case
283 	 * we are checking the "new" CPU instead of the old one. This
284 	 * is only a problem if an event happened on the "old" CPU.
285 	 *
286 	 * External interrupt events will have caused interrupts to
287 	 * be hard-disabled, so there is no problem, we
288 	 * cannot have preempted.
289 	 */
290 	irq_happened = get_irq_happened();
291 	if (!irq_happened) {
292 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
293 			WARN_ON_ONCE(!(mfmsr() & MSR_EE));
294 		return;
295 	}
296 
297 	/* We need to hard disable to replay. */
298 	if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
299 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
300 			WARN_ON_ONCE(!(mfmsr() & MSR_EE));
301 		__hard_irq_disable();
302 		local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
303 	} else {
304 		/*
305 		 * We should already be hard disabled here. We had bugs
306 		 * where that wasn't the case so let's dbl check it and
307 		 * warn if we are wrong. Only do that when IRQ tracing
308 		 * is enabled as mfmsr() can be costly.
309 		 */
310 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
311 			if (WARN_ON_ONCE(mfmsr() & MSR_EE))
312 				__hard_irq_disable();
313 		}
314 
315 		if (irq_happened == PACA_IRQ_HARD_DIS) {
316 			local_paca->irq_happened = 0;
317 			__hard_irq_enable();
318 			return;
319 		}
320 	}
321 
322 	/*
323 	 * Disable preempt here, so that the below preempt_enable will
324 	 * perform resched if required (a replayed interrupt may set
325 	 * need_resched).
326 	 */
327 	preempt_disable();
328 	irq_soft_mask_set(IRQS_ALL_DISABLED);
329 	trace_hardirqs_off();
330 
331 	replay_soft_interrupts();
332 	local_paca->irq_happened = 0;
333 
334 	trace_hardirqs_on();
335 	irq_soft_mask_set(IRQS_ENABLED);
336 	__hard_irq_enable();
337 	preempt_enable();
338 }
339 EXPORT_SYMBOL(arch_local_irq_restore);
340 
341 /*
342  * This is a helper to use when about to go into idle low-power
343  * when the latter has the side effect of re-enabling interrupts
344  * (such as calling H_CEDE under pHyp).
345  *
346  * You call this function with interrupts soft-disabled (this is
347  * already the case when ppc_md.power_save is called). The function
348  * will return whether to enter power save or just return.
349  *
350  * In the former case, it will have notified lockdep of interrupts
351  * being re-enabled and generally sanitized the lazy irq state,
352  * and in the latter case it will leave with interrupts hard
353  * disabled and marked as such, so the local_irq_enable() call
354  * in arch_cpu_idle() will properly re-enable everything.
355  */
356 bool prep_irq_for_idle(void)
357 {
358 	/*
359 	 * First we need to hard disable to ensure no interrupt
360 	 * occurs before we effectively enter the low power state
361 	 */
362 	__hard_irq_disable();
363 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
364 
365 	/*
366 	 * If anything happened while we were soft-disabled,
367 	 * we return now and do not enter the low power state.
368 	 */
369 	if (lazy_irq_pending())
370 		return false;
371 
372 	/* Tell lockdep we are about to re-enable */
373 	trace_hardirqs_on();
374 
375 	/*
376 	 * Mark interrupts as soft-enabled and clear the
377 	 * PACA_IRQ_HARD_DIS from the pending mask since we
378 	 * are about to hard enable as well as a side effect
379 	 * of entering the low power state.
380 	 */
381 	local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
382 	irq_soft_mask_set(IRQS_ENABLED);
383 
384 	/* Tell the caller to enter the low power state */
385 	return true;
386 }
387 
388 #ifdef CONFIG_PPC_BOOK3S
389 /*
390  * This is for idle sequences that return with IRQs off, but the
391  * idle state itself wakes on interrupt. Tell the irq tracer that
392  * IRQs are enabled for the duration of idle so it does not get long
393  * off times. Must be paired with fini_irq_for_idle_irqsoff.
394  */
395 bool prep_irq_for_idle_irqsoff(void)
396 {
397 	WARN_ON(!irqs_disabled());
398 
399 	/*
400 	 * First we need to hard disable to ensure no interrupt
401 	 * occurs before we effectively enter the low power state
402 	 */
403 	__hard_irq_disable();
404 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
405 
406 	/*
407 	 * If anything happened while we were soft-disabled,
408 	 * we return now and do not enter the low power state.
409 	 */
410 	if (lazy_irq_pending())
411 		return false;
412 
413 	/* Tell lockdep we are about to re-enable */
414 	trace_hardirqs_on();
415 
416 	return true;
417 }
418 
419 /*
420  * Take the SRR1 wakeup reason, index into this table to find the
421  * appropriate irq_happened bit.
422  *
423  * Sytem reset exceptions taken in idle state also come through here,
424  * but they are NMI interrupts so do not need to wait for IRQs to be
425  * restored, and should be taken as early as practical. These are marked
426  * with 0xff in the table. The Power ISA specifies 0100b as the system
427  * reset interrupt reason.
428  */
429 #define IRQ_SYSTEM_RESET	0xff
430 
431 static const u8 srr1_to_lazyirq[0x10] = {
432 	0, 0, 0,
433 	PACA_IRQ_DBELL,
434 	IRQ_SYSTEM_RESET,
435 	PACA_IRQ_DBELL,
436 	PACA_IRQ_DEC,
437 	0,
438 	PACA_IRQ_EE,
439 	PACA_IRQ_EE,
440 	PACA_IRQ_HMI,
441 	0, 0, 0, 0, 0 };
442 
443 void replay_system_reset(void)
444 {
445 	struct pt_regs regs;
446 
447 	ppc_save_regs(&regs);
448 	regs.trap = 0x100;
449 	get_paca()->in_nmi = 1;
450 	system_reset_exception(&regs);
451 	get_paca()->in_nmi = 0;
452 }
453 EXPORT_SYMBOL_GPL(replay_system_reset);
454 
455 void irq_set_pending_from_srr1(unsigned long srr1)
456 {
457 	unsigned int idx = (srr1 & SRR1_WAKEMASK_P8) >> 18;
458 	u8 reason = srr1_to_lazyirq[idx];
459 
460 	/*
461 	 * Take the system reset now, which is immediately after registers
462 	 * are restored from idle. It's an NMI, so interrupts need not be
463 	 * re-enabled before it is taken.
464 	 */
465 	if (unlikely(reason == IRQ_SYSTEM_RESET)) {
466 		replay_system_reset();
467 		return;
468 	}
469 
470 	if (reason == PACA_IRQ_DBELL) {
471 		/*
472 		 * When doorbell triggers a system reset wakeup, the message
473 		 * is not cleared, so if the doorbell interrupt is replayed
474 		 * and the IPI handled, the doorbell interrupt would still
475 		 * fire when EE is enabled.
476 		 *
477 		 * To avoid taking the superfluous doorbell interrupt,
478 		 * execute a msgclr here before the interrupt is replayed.
479 		 */
480 		ppc_msgclr(PPC_DBELL_MSGTYPE);
481 	}
482 
483 	/*
484 	 * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
485 	 * so this can be called unconditionally with the SRR1 wake
486 	 * reason as returned by the idle code, which uses 0 to mean no
487 	 * interrupt.
488 	 *
489 	 * If a future CPU was to designate this as an interrupt reason,
490 	 * then a new index for no interrupt must be assigned.
491 	 */
492 	local_paca->irq_happened |= reason;
493 }
494 #endif /* CONFIG_PPC_BOOK3S */
495 
496 /*
497  * Force a replay of the external interrupt handler on this CPU.
498  */
499 void force_external_irq_replay(void)
500 {
501 	/*
502 	 * This must only be called with interrupts soft-disabled,
503 	 * the replay will happen when re-enabling.
504 	 */
505 	WARN_ON(!arch_irqs_disabled());
506 
507 	/*
508 	 * Interrupts must always be hard disabled before irq_happened is
509 	 * modified (to prevent lost update in case of interrupt between
510 	 * load and store).
511 	 */
512 	__hard_irq_disable();
513 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
514 
515 	/* Indicate in the PACA that we have an interrupt to replay */
516 	local_paca->irq_happened |= PACA_IRQ_EE;
517 }
518 
519 #endif /* CONFIG_PPC64 */
520 
521 int arch_show_interrupts(struct seq_file *p, int prec)
522 {
523 	int j;
524 
525 #if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
526 	if (tau_initialized) {
527 		seq_printf(p, "%*s: ", prec, "TAU");
528 		for_each_online_cpu(j)
529 			seq_printf(p, "%10u ", tau_interrupts(j));
530 		seq_puts(p, "  PowerPC             Thermal Assist (cpu temp)\n");
531 	}
532 #endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
533 
534 	seq_printf(p, "%*s: ", prec, "LOC");
535 	for_each_online_cpu(j)
536 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_event);
537         seq_printf(p, "  Local timer interrupts for timer event device\n");
538 
539 	seq_printf(p, "%*s: ", prec, "BCT");
540 	for_each_online_cpu(j)
541 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).broadcast_irqs_event);
542 	seq_printf(p, "  Broadcast timer interrupts for timer event device\n");
543 
544 	seq_printf(p, "%*s: ", prec, "LOC");
545 	for_each_online_cpu(j)
546 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_others);
547         seq_printf(p, "  Local timer interrupts for others\n");
548 
549 	seq_printf(p, "%*s: ", prec, "SPU");
550 	for_each_online_cpu(j)
551 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
552 	seq_printf(p, "  Spurious interrupts\n");
553 
554 	seq_printf(p, "%*s: ", prec, "PMI");
555 	for_each_online_cpu(j)
556 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
557 	seq_printf(p, "  Performance monitoring interrupts\n");
558 
559 	seq_printf(p, "%*s: ", prec, "MCE");
560 	for_each_online_cpu(j)
561 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
562 	seq_printf(p, "  Machine check exceptions\n");
563 
564 #ifdef CONFIG_PPC_BOOK3S_64
565 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
566 		seq_printf(p, "%*s: ", prec, "HMI");
567 		for_each_online_cpu(j)
568 			seq_printf(p, "%10u ", paca_ptrs[j]->hmi_irqs);
569 		seq_printf(p, "  Hypervisor Maintenance Interrupts\n");
570 	}
571 #endif
572 
573 	seq_printf(p, "%*s: ", prec, "NMI");
574 	for_each_online_cpu(j)
575 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).sreset_irqs);
576 	seq_printf(p, "  System Reset interrupts\n");
577 
578 #ifdef CONFIG_PPC_WATCHDOG
579 	seq_printf(p, "%*s: ", prec, "WDG");
580 	for_each_online_cpu(j)
581 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).soft_nmi_irqs);
582 	seq_printf(p, "  Watchdog soft-NMI interrupts\n");
583 #endif
584 
585 #ifdef CONFIG_PPC_DOORBELL
586 	if (cpu_has_feature(CPU_FTR_DBELL)) {
587 		seq_printf(p, "%*s: ", prec, "DBL");
588 		for_each_online_cpu(j)
589 			seq_printf(p, "%10u ", per_cpu(irq_stat, j).doorbell_irqs);
590 		seq_printf(p, "  Doorbell interrupts\n");
591 	}
592 #endif
593 
594 	return 0;
595 }
596 
597 /*
598  * /proc/stat helpers
599  */
600 u64 arch_irq_stat_cpu(unsigned int cpu)
601 {
602 	u64 sum = per_cpu(irq_stat, cpu).timer_irqs_event;
603 
604 	sum += per_cpu(irq_stat, cpu).broadcast_irqs_event;
605 	sum += per_cpu(irq_stat, cpu).pmu_irqs;
606 	sum += per_cpu(irq_stat, cpu).mce_exceptions;
607 	sum += per_cpu(irq_stat, cpu).spurious_irqs;
608 	sum += per_cpu(irq_stat, cpu).timer_irqs_others;
609 #ifdef CONFIG_PPC_BOOK3S_64
610 	sum += paca_ptrs[cpu]->hmi_irqs;
611 #endif
612 	sum += per_cpu(irq_stat, cpu).sreset_irqs;
613 #ifdef CONFIG_PPC_WATCHDOG
614 	sum += per_cpu(irq_stat, cpu).soft_nmi_irqs;
615 #endif
616 #ifdef CONFIG_PPC_DOORBELL
617 	sum += per_cpu(irq_stat, cpu).doorbell_irqs;
618 #endif
619 
620 	return sum;
621 }
622 
623 static inline void check_stack_overflow(void)
624 {
625 	long sp;
626 
627 	if (!IS_ENABLED(CONFIG_DEBUG_STACKOVERFLOW))
628 		return;
629 
630 	sp = current_stack_pointer & (THREAD_SIZE - 1);
631 
632 	/* check for stack overflow: is there less than 2KB free? */
633 	if (unlikely(sp < 2048)) {
634 		pr_err("do_IRQ: stack overflow: %ld\n", sp);
635 		dump_stack();
636 	}
637 }
638 
639 void __do_irq(struct pt_regs *regs)
640 {
641 	unsigned int irq;
642 
643 	irq_enter();
644 
645 	trace_irq_entry(regs);
646 
647 	/*
648 	 * Query the platform PIC for the interrupt & ack it.
649 	 *
650 	 * This will typically lower the interrupt line to the CPU
651 	 */
652 	irq = ppc_md.get_irq();
653 
654 	/* We can hard enable interrupts now to allow perf interrupts */
655 	may_hard_irq_enable();
656 
657 	/* And finally process it */
658 	if (unlikely(!irq))
659 		__this_cpu_inc(irq_stat.spurious_irqs);
660 	else
661 		generic_handle_irq(irq);
662 
663 	trace_irq_exit(regs);
664 
665 	irq_exit();
666 }
667 
668 void do_IRQ(struct pt_regs *regs)
669 {
670 	struct pt_regs *old_regs = set_irq_regs(regs);
671 	void *cursp, *irqsp, *sirqsp;
672 
673 	/* Switch to the irq stack to handle this */
674 	cursp = (void *)(current_stack_pointer & ~(THREAD_SIZE - 1));
675 	irqsp = hardirq_ctx[raw_smp_processor_id()];
676 	sirqsp = softirq_ctx[raw_smp_processor_id()];
677 
678 	check_stack_overflow();
679 
680 	/* Already there ? */
681 	if (unlikely(cursp == irqsp || cursp == sirqsp)) {
682 		__do_irq(regs);
683 		set_irq_regs(old_regs);
684 		return;
685 	}
686 	/* Switch stack and call */
687 	call_do_irq(regs, irqsp);
688 
689 	set_irq_regs(old_regs);
690 }
691 
692 static void *__init alloc_vm_stack(void)
693 {
694 	return __vmalloc_node(THREAD_SIZE, THREAD_ALIGN, THREADINFO_GFP,
695 			      NUMA_NO_NODE, (void *)_RET_IP_);
696 }
697 
698 static void __init vmap_irqstack_init(void)
699 {
700 	int i;
701 
702 	for_each_possible_cpu(i) {
703 		softirq_ctx[i] = alloc_vm_stack();
704 		hardirq_ctx[i] = alloc_vm_stack();
705 	}
706 }
707 
708 
709 void __init init_IRQ(void)
710 {
711 	if (IS_ENABLED(CONFIG_VMAP_STACK))
712 		vmap_irqstack_init();
713 
714 	if (ppc_md.init_IRQ)
715 		ppc_md.init_IRQ();
716 }
717 
718 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
719 void   *critirq_ctx[NR_CPUS] __read_mostly;
720 void    *dbgirq_ctx[NR_CPUS] __read_mostly;
721 void *mcheckirq_ctx[NR_CPUS] __read_mostly;
722 #endif
723 
724 void *softirq_ctx[NR_CPUS] __read_mostly;
725 void *hardirq_ctx[NR_CPUS] __read_mostly;
726 
727 void do_softirq_own_stack(void)
728 {
729 	call_do_softirq(softirq_ctx[smp_processor_id()]);
730 }
731 
732 irq_hw_number_t virq_to_hw(unsigned int virq)
733 {
734 	struct irq_data *irq_data = irq_get_irq_data(virq);
735 	return WARN_ON(!irq_data) ? 0 : irq_data->hwirq;
736 }
737 EXPORT_SYMBOL_GPL(virq_to_hw);
738 
739 #ifdef CONFIG_SMP
740 int irq_choose_cpu(const struct cpumask *mask)
741 {
742 	int cpuid;
743 
744 	if (cpumask_equal(mask, cpu_online_mask)) {
745 		static int irq_rover;
746 		static DEFINE_RAW_SPINLOCK(irq_rover_lock);
747 		unsigned long flags;
748 
749 		/* Round-robin distribution... */
750 do_round_robin:
751 		raw_spin_lock_irqsave(&irq_rover_lock, flags);
752 
753 		irq_rover = cpumask_next(irq_rover, cpu_online_mask);
754 		if (irq_rover >= nr_cpu_ids)
755 			irq_rover = cpumask_first(cpu_online_mask);
756 
757 		cpuid = irq_rover;
758 
759 		raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
760 	} else {
761 		cpuid = cpumask_first_and(mask, cpu_online_mask);
762 		if (cpuid >= nr_cpu_ids)
763 			goto do_round_robin;
764 	}
765 
766 	return get_hard_smp_processor_id(cpuid);
767 }
768 #else
769 int irq_choose_cpu(const struct cpumask *mask)
770 {
771 	return hard_smp_processor_id();
772 }
773 #endif
774 
775 #ifdef CONFIG_PPC64
776 static int __init setup_noirqdistrib(char *str)
777 {
778 	distribute_irqs = 0;
779 	return 1;
780 }
781 
782 __setup("noirqdistrib", setup_noirqdistrib);
783 #endif /* CONFIG_PPC64 */
784