xref: /openbmc/linux/arch/powerpc/kernel/irq.c (revision c1cf3d89)
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 	struct pt_regs regs;
184 
185 	/*
186 	 * Be careful here, calling these interrupt handlers can cause
187 	 * softirqs to be raised, which they may run when calling irq_exit,
188 	 * which will cause local_irq_enable() to be run, which can then
189 	 * recurse into this function. Don't keep any state across
190 	 * interrupt handler calls which may change underneath us.
191 	 *
192 	 * We use local_paca rather than get_paca() to avoid all the
193 	 * debug_smp_processor_id() business in this low level function.
194 	 */
195 
196 	ppc_save_regs(&regs);
197 	regs.softe = IRQS_ENABLED;
198 
199 again:
200 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
201 		WARN_ON_ONCE(mfmsr() & MSR_EE);
202 
203 	/*
204 	 * Force the delivery of pending soft-disabled interrupts on PS3.
205 	 * Any HV call will have this side effect.
206 	 */
207 	if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
208 		u64 tmp, tmp2;
209 		lv1_get_version_info(&tmp, &tmp2);
210 	}
211 
212 	/*
213 	 * Check if an hypervisor Maintenance interrupt happened.
214 	 * This is a higher priority interrupt than the others, so
215 	 * replay it first.
216 	 */
217 	if (IS_ENABLED(CONFIG_PPC_BOOK3S) && (local_paca->irq_happened & PACA_IRQ_HMI)) {
218 		local_paca->irq_happened &= ~PACA_IRQ_HMI;
219 		regs.trap = 0xe60;
220 		handle_hmi_exception(&regs);
221 		if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
222 			hard_irq_disable();
223 	}
224 
225 	if (local_paca->irq_happened & PACA_IRQ_DEC) {
226 		local_paca->irq_happened &= ~PACA_IRQ_DEC;
227 		regs.trap = 0x900;
228 		timer_interrupt(&regs);
229 		if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
230 			hard_irq_disable();
231 	}
232 
233 	if (local_paca->irq_happened & PACA_IRQ_EE) {
234 		local_paca->irq_happened &= ~PACA_IRQ_EE;
235 		regs.trap = 0x500;
236 		do_IRQ(&regs);
237 		if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
238 			hard_irq_disable();
239 	}
240 
241 	if (IS_ENABLED(CONFIG_PPC_DOORBELL) && (local_paca->irq_happened & PACA_IRQ_DBELL)) {
242 		local_paca->irq_happened &= ~PACA_IRQ_DBELL;
243 		if (IS_ENABLED(CONFIG_PPC_BOOK3E))
244 			regs.trap = 0x280;
245 		else
246 			regs.trap = 0xa00;
247 		doorbell_exception(&regs);
248 		if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
249 			hard_irq_disable();
250 	}
251 
252 	/* Book3E does not support soft-masking PMI interrupts */
253 	if (IS_ENABLED(CONFIG_PPC_BOOK3S) && (local_paca->irq_happened & PACA_IRQ_PMI)) {
254 		local_paca->irq_happened &= ~PACA_IRQ_PMI;
255 		regs.trap = 0xf00;
256 		performance_monitor_exception(&regs);
257 		if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
258 			hard_irq_disable();
259 	}
260 
261 	if (local_paca->irq_happened & ~PACA_IRQ_HARD_DIS) {
262 		/*
263 		 * We are responding to the next interrupt, so interrupt-off
264 		 * latencies should be reset here.
265 		 */
266 		trace_hardirqs_on();
267 		trace_hardirqs_off();
268 		goto again;
269 	}
270 }
271 
272 notrace void arch_local_irq_restore(unsigned long mask)
273 {
274 	unsigned char irq_happened;
275 
276 	/* Write the new soft-enabled value */
277 	irq_soft_mask_set(mask);
278 	if (mask)
279 		return;
280 
281 	/*
282 	 * From this point onward, we can take interrupts, preempt,
283 	 * etc... unless we got hard-disabled. We check if an event
284 	 * happened. If none happened, we know we can just return.
285 	 *
286 	 * We may have preempted before the check below, in which case
287 	 * we are checking the "new" CPU instead of the old one. This
288 	 * is only a problem if an event happened on the "old" CPU.
289 	 *
290 	 * External interrupt events will have caused interrupts to
291 	 * be hard-disabled, so there is no problem, we
292 	 * cannot have preempted.
293 	 */
294 	irq_happened = get_irq_happened();
295 	if (!irq_happened) {
296 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
297 			WARN_ON_ONCE(!(mfmsr() & MSR_EE));
298 		return;
299 	}
300 
301 	/* We need to hard disable to replay. */
302 	if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
303 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
304 			WARN_ON_ONCE(!(mfmsr() & MSR_EE));
305 		__hard_irq_disable();
306 		local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
307 	} else {
308 		/*
309 		 * We should already be hard disabled here. We had bugs
310 		 * where that wasn't the case so let's dbl check it and
311 		 * warn if we are wrong. Only do that when IRQ tracing
312 		 * is enabled as mfmsr() can be costly.
313 		 */
314 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
315 			if (WARN_ON_ONCE(mfmsr() & MSR_EE))
316 				__hard_irq_disable();
317 		}
318 
319 		if (irq_happened == PACA_IRQ_HARD_DIS) {
320 			local_paca->irq_happened = 0;
321 			__hard_irq_enable();
322 			return;
323 		}
324 	}
325 
326 	/*
327 	 * Disable preempt here, so that the below preempt_enable will
328 	 * perform resched if required (a replayed interrupt may set
329 	 * need_resched).
330 	 */
331 	preempt_disable();
332 	irq_soft_mask_set(IRQS_ALL_DISABLED);
333 	trace_hardirqs_off();
334 
335 	replay_soft_interrupts();
336 	local_paca->irq_happened = 0;
337 
338 	trace_hardirqs_on();
339 	irq_soft_mask_set(IRQS_ENABLED);
340 	__hard_irq_enable();
341 	preempt_enable();
342 }
343 EXPORT_SYMBOL(arch_local_irq_restore);
344 
345 /*
346  * This is a helper to use when about to go into idle low-power
347  * when the latter has the side effect of re-enabling interrupts
348  * (such as calling H_CEDE under pHyp).
349  *
350  * You call this function with interrupts soft-disabled (this is
351  * already the case when ppc_md.power_save is called). The function
352  * will return whether to enter power save or just return.
353  *
354  * In the former case, it will have notified lockdep of interrupts
355  * being re-enabled and generally sanitized the lazy irq state,
356  * and in the latter case it will leave with interrupts hard
357  * disabled and marked as such, so the local_irq_enable() call
358  * in arch_cpu_idle() will properly re-enable everything.
359  */
360 bool prep_irq_for_idle(void)
361 {
362 	/*
363 	 * First we need to hard disable to ensure no interrupt
364 	 * occurs before we effectively enter the low power state
365 	 */
366 	__hard_irq_disable();
367 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
368 
369 	/*
370 	 * If anything happened while we were soft-disabled,
371 	 * we return now and do not enter the low power state.
372 	 */
373 	if (lazy_irq_pending())
374 		return false;
375 
376 	/* Tell lockdep we are about to re-enable */
377 	trace_hardirqs_on();
378 
379 	/*
380 	 * Mark interrupts as soft-enabled and clear the
381 	 * PACA_IRQ_HARD_DIS from the pending mask since we
382 	 * are about to hard enable as well as a side effect
383 	 * of entering the low power state.
384 	 */
385 	local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
386 	irq_soft_mask_set(IRQS_ENABLED);
387 
388 	/* Tell the caller to enter the low power state */
389 	return true;
390 }
391 
392 #ifdef CONFIG_PPC_BOOK3S
393 /*
394  * This is for idle sequences that return with IRQs off, but the
395  * idle state itself wakes on interrupt. Tell the irq tracer that
396  * IRQs are enabled for the duration of idle so it does not get long
397  * off times. Must be paired with fini_irq_for_idle_irqsoff.
398  */
399 bool prep_irq_for_idle_irqsoff(void)
400 {
401 	WARN_ON(!irqs_disabled());
402 
403 	/*
404 	 * First we need to hard disable to ensure no interrupt
405 	 * occurs before we effectively enter the low power state
406 	 */
407 	__hard_irq_disable();
408 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
409 
410 	/*
411 	 * If anything happened while we were soft-disabled,
412 	 * we return now and do not enter the low power state.
413 	 */
414 	if (lazy_irq_pending())
415 		return false;
416 
417 	/* Tell lockdep we are about to re-enable */
418 	trace_hardirqs_on();
419 
420 	return true;
421 }
422 
423 /*
424  * Take the SRR1 wakeup reason, index into this table to find the
425  * appropriate irq_happened bit.
426  *
427  * Sytem reset exceptions taken in idle state also come through here,
428  * but they are NMI interrupts so do not need to wait for IRQs to be
429  * restored, and should be taken as early as practical. These are marked
430  * with 0xff in the table. The Power ISA specifies 0100b as the system
431  * reset interrupt reason.
432  */
433 #define IRQ_SYSTEM_RESET	0xff
434 
435 static const u8 srr1_to_lazyirq[0x10] = {
436 	0, 0, 0,
437 	PACA_IRQ_DBELL,
438 	IRQ_SYSTEM_RESET,
439 	PACA_IRQ_DBELL,
440 	PACA_IRQ_DEC,
441 	0,
442 	PACA_IRQ_EE,
443 	PACA_IRQ_EE,
444 	PACA_IRQ_HMI,
445 	0, 0, 0, 0, 0 };
446 
447 void replay_system_reset(void)
448 {
449 	struct pt_regs regs;
450 
451 	ppc_save_regs(&regs);
452 	regs.trap = 0x100;
453 	get_paca()->in_nmi = 1;
454 	system_reset_exception(&regs);
455 	get_paca()->in_nmi = 0;
456 }
457 EXPORT_SYMBOL_GPL(replay_system_reset);
458 
459 void irq_set_pending_from_srr1(unsigned long srr1)
460 {
461 	unsigned int idx = (srr1 & SRR1_WAKEMASK_P8) >> 18;
462 	u8 reason = srr1_to_lazyirq[idx];
463 
464 	/*
465 	 * Take the system reset now, which is immediately after registers
466 	 * are restored from idle. It's an NMI, so interrupts need not be
467 	 * re-enabled before it is taken.
468 	 */
469 	if (unlikely(reason == IRQ_SYSTEM_RESET)) {
470 		replay_system_reset();
471 		return;
472 	}
473 
474 	if (reason == PACA_IRQ_DBELL) {
475 		/*
476 		 * When doorbell triggers a system reset wakeup, the message
477 		 * is not cleared, so if the doorbell interrupt is replayed
478 		 * and the IPI handled, the doorbell interrupt would still
479 		 * fire when EE is enabled.
480 		 *
481 		 * To avoid taking the superfluous doorbell interrupt,
482 		 * execute a msgclr here before the interrupt is replayed.
483 		 */
484 		ppc_msgclr(PPC_DBELL_MSGTYPE);
485 	}
486 
487 	/*
488 	 * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
489 	 * so this can be called unconditionally with the SRR1 wake
490 	 * reason as returned by the idle code, which uses 0 to mean no
491 	 * interrupt.
492 	 *
493 	 * If a future CPU was to designate this as an interrupt reason,
494 	 * then a new index for no interrupt must be assigned.
495 	 */
496 	local_paca->irq_happened |= reason;
497 }
498 #endif /* CONFIG_PPC_BOOK3S */
499 
500 /*
501  * Force a replay of the external interrupt handler on this CPU.
502  */
503 void force_external_irq_replay(void)
504 {
505 	/*
506 	 * This must only be called with interrupts soft-disabled,
507 	 * the replay will happen when re-enabling.
508 	 */
509 	WARN_ON(!arch_irqs_disabled());
510 
511 	/*
512 	 * Interrupts must always be hard disabled before irq_happened is
513 	 * modified (to prevent lost update in case of interrupt between
514 	 * load and store).
515 	 */
516 	__hard_irq_disable();
517 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
518 
519 	/* Indicate in the PACA that we have an interrupt to replay */
520 	local_paca->irq_happened |= PACA_IRQ_EE;
521 }
522 
523 #endif /* CONFIG_PPC64 */
524 
525 int arch_show_interrupts(struct seq_file *p, int prec)
526 {
527 	int j;
528 
529 #if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
530 	if (tau_initialized) {
531 		seq_printf(p, "%*s: ", prec, "TAU");
532 		for_each_online_cpu(j)
533 			seq_printf(p, "%10u ", tau_interrupts(j));
534 		seq_puts(p, "  PowerPC             Thermal Assist (cpu temp)\n");
535 	}
536 #endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
537 
538 	seq_printf(p, "%*s: ", prec, "LOC");
539 	for_each_online_cpu(j)
540 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_event);
541         seq_printf(p, "  Local timer interrupts for timer event device\n");
542 
543 	seq_printf(p, "%*s: ", prec, "BCT");
544 	for_each_online_cpu(j)
545 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).broadcast_irqs_event);
546 	seq_printf(p, "  Broadcast timer interrupts for timer event device\n");
547 
548 	seq_printf(p, "%*s: ", prec, "LOC");
549 	for_each_online_cpu(j)
550 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_others);
551         seq_printf(p, "  Local timer interrupts for others\n");
552 
553 	seq_printf(p, "%*s: ", prec, "SPU");
554 	for_each_online_cpu(j)
555 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
556 	seq_printf(p, "  Spurious interrupts\n");
557 
558 	seq_printf(p, "%*s: ", prec, "PMI");
559 	for_each_online_cpu(j)
560 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
561 	seq_printf(p, "  Performance monitoring interrupts\n");
562 
563 	seq_printf(p, "%*s: ", prec, "MCE");
564 	for_each_online_cpu(j)
565 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
566 	seq_printf(p, "  Machine check exceptions\n");
567 
568 #ifdef CONFIG_PPC_BOOK3S_64
569 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
570 		seq_printf(p, "%*s: ", prec, "HMI");
571 		for_each_online_cpu(j)
572 			seq_printf(p, "%10u ", paca_ptrs[j]->hmi_irqs);
573 		seq_printf(p, "  Hypervisor Maintenance Interrupts\n");
574 	}
575 #endif
576 
577 	seq_printf(p, "%*s: ", prec, "NMI");
578 	for_each_online_cpu(j)
579 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).sreset_irqs);
580 	seq_printf(p, "  System Reset interrupts\n");
581 
582 #ifdef CONFIG_PPC_WATCHDOG
583 	seq_printf(p, "%*s: ", prec, "WDG");
584 	for_each_online_cpu(j)
585 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).soft_nmi_irqs);
586 	seq_printf(p, "  Watchdog soft-NMI interrupts\n");
587 #endif
588 
589 #ifdef CONFIG_PPC_DOORBELL
590 	if (cpu_has_feature(CPU_FTR_DBELL)) {
591 		seq_printf(p, "%*s: ", prec, "DBL");
592 		for_each_online_cpu(j)
593 			seq_printf(p, "%10u ", per_cpu(irq_stat, j).doorbell_irqs);
594 		seq_printf(p, "  Doorbell interrupts\n");
595 	}
596 #endif
597 
598 	return 0;
599 }
600 
601 /*
602  * /proc/stat helpers
603  */
604 u64 arch_irq_stat_cpu(unsigned int cpu)
605 {
606 	u64 sum = per_cpu(irq_stat, cpu).timer_irqs_event;
607 
608 	sum += per_cpu(irq_stat, cpu).broadcast_irqs_event;
609 	sum += per_cpu(irq_stat, cpu).pmu_irqs;
610 	sum += per_cpu(irq_stat, cpu).mce_exceptions;
611 	sum += per_cpu(irq_stat, cpu).spurious_irqs;
612 	sum += per_cpu(irq_stat, cpu).timer_irqs_others;
613 #ifdef CONFIG_PPC_BOOK3S_64
614 	sum += paca_ptrs[cpu]->hmi_irqs;
615 #endif
616 	sum += per_cpu(irq_stat, cpu).sreset_irqs;
617 #ifdef CONFIG_PPC_WATCHDOG
618 	sum += per_cpu(irq_stat, cpu).soft_nmi_irqs;
619 #endif
620 #ifdef CONFIG_PPC_DOORBELL
621 	sum += per_cpu(irq_stat, cpu).doorbell_irqs;
622 #endif
623 
624 	return sum;
625 }
626 
627 static inline void check_stack_overflow(void)
628 {
629 	long sp;
630 
631 	if (!IS_ENABLED(CONFIG_DEBUG_STACKOVERFLOW))
632 		return;
633 
634 	sp = current_stack_pointer & (THREAD_SIZE - 1);
635 
636 	/* check for stack overflow: is there less than 2KB free? */
637 	if (unlikely(sp < 2048)) {
638 		pr_err("do_IRQ: stack overflow: %ld\n", sp);
639 		dump_stack();
640 	}
641 }
642 
643 void __do_irq(struct pt_regs *regs)
644 {
645 	unsigned int irq;
646 
647 	irq_enter();
648 
649 	trace_irq_entry(regs);
650 
651 	/*
652 	 * Query the platform PIC for the interrupt & ack it.
653 	 *
654 	 * This will typically lower the interrupt line to the CPU
655 	 */
656 	irq = ppc_md.get_irq();
657 
658 	/* We can hard enable interrupts now to allow perf interrupts */
659 	may_hard_irq_enable();
660 
661 	/* And finally process it */
662 	if (unlikely(!irq))
663 		__this_cpu_inc(irq_stat.spurious_irqs);
664 	else
665 		generic_handle_irq(irq);
666 
667 	trace_irq_exit(regs);
668 
669 	irq_exit();
670 }
671 
672 void do_IRQ(struct pt_regs *regs)
673 {
674 	struct pt_regs *old_regs = set_irq_regs(regs);
675 	void *cursp, *irqsp, *sirqsp;
676 
677 	/* Switch to the irq stack to handle this */
678 	cursp = (void *)(current_stack_pointer & ~(THREAD_SIZE - 1));
679 	irqsp = hardirq_ctx[raw_smp_processor_id()];
680 	sirqsp = softirq_ctx[raw_smp_processor_id()];
681 
682 	check_stack_overflow();
683 
684 	/* Already there ? */
685 	if (unlikely(cursp == irqsp || cursp == sirqsp)) {
686 		__do_irq(regs);
687 		set_irq_regs(old_regs);
688 		return;
689 	}
690 	/* Switch stack and call */
691 	call_do_irq(regs, irqsp);
692 
693 	set_irq_regs(old_regs);
694 }
695 
696 static void *__init alloc_vm_stack(void)
697 {
698 	return __vmalloc_node(THREAD_SIZE, THREAD_ALIGN, THREADINFO_GFP,
699 			      NUMA_NO_NODE, (void *)_RET_IP_);
700 }
701 
702 static void __init vmap_irqstack_init(void)
703 {
704 	int i;
705 
706 	for_each_possible_cpu(i) {
707 		softirq_ctx[i] = alloc_vm_stack();
708 		hardirq_ctx[i] = alloc_vm_stack();
709 	}
710 }
711 
712 
713 void __init init_IRQ(void)
714 {
715 	if (IS_ENABLED(CONFIG_VMAP_STACK))
716 		vmap_irqstack_init();
717 
718 	if (ppc_md.init_IRQ)
719 		ppc_md.init_IRQ();
720 }
721 
722 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
723 void   *critirq_ctx[NR_CPUS] __read_mostly;
724 void    *dbgirq_ctx[NR_CPUS] __read_mostly;
725 void *mcheckirq_ctx[NR_CPUS] __read_mostly;
726 #endif
727 
728 void *softirq_ctx[NR_CPUS] __read_mostly;
729 void *hardirq_ctx[NR_CPUS] __read_mostly;
730 
731 void do_softirq_own_stack(void)
732 {
733 	call_do_softirq(softirq_ctx[smp_processor_id()]);
734 }
735 
736 irq_hw_number_t virq_to_hw(unsigned int virq)
737 {
738 	struct irq_data *irq_data = irq_get_irq_data(virq);
739 	return WARN_ON(!irq_data) ? 0 : irq_data->hwirq;
740 }
741 EXPORT_SYMBOL_GPL(virq_to_hw);
742 
743 #ifdef CONFIG_SMP
744 int irq_choose_cpu(const struct cpumask *mask)
745 {
746 	int cpuid;
747 
748 	if (cpumask_equal(mask, cpu_online_mask)) {
749 		static int irq_rover;
750 		static DEFINE_RAW_SPINLOCK(irq_rover_lock);
751 		unsigned long flags;
752 
753 		/* Round-robin distribution... */
754 do_round_robin:
755 		raw_spin_lock_irqsave(&irq_rover_lock, flags);
756 
757 		irq_rover = cpumask_next(irq_rover, cpu_online_mask);
758 		if (irq_rover >= nr_cpu_ids)
759 			irq_rover = cpumask_first(cpu_online_mask);
760 
761 		cpuid = irq_rover;
762 
763 		raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
764 	} else {
765 		cpuid = cpumask_first_and(mask, cpu_online_mask);
766 		if (cpuid >= nr_cpu_ids)
767 			goto do_round_robin;
768 	}
769 
770 	return get_hard_smp_processor_id(cpuid);
771 }
772 #else
773 int irq_choose_cpu(const struct cpumask *mask)
774 {
775 	return hard_smp_processor_id();
776 }
777 #endif
778 
779 #ifdef CONFIG_PPC64
780 static int __init setup_noirqdistrib(char *str)
781 {
782 	distribute_irqs = 0;
783 	return 1;
784 }
785 
786 __setup("noirqdistrib", setup_noirqdistrib);
787 #endif /* CONFIG_PPC64 */
788