• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
get_irq_happened(void)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  */
__check_irq_replay(void)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  */
restore_interrupts(void)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 
replay_soft_interrupts(void)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 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_KUAP)
replay_soft_interrupts_irqrestore(void)273 static inline void replay_soft_interrupts_irqrestore(void)
274 {
275 	unsigned long kuap_state = get_kuap();
276 
277 	/*
278 	 * Check if anything calls local_irq_enable/restore() when KUAP is
279 	 * disabled (user access enabled). We handle that case here by saving
280 	 * and re-locking AMR but we shouldn't get here in the first place,
281 	 * hence the warning.
282 	 */
283 	kuap_check_amr();
284 
285 	if (kuap_state != AMR_KUAP_BLOCKED)
286 		set_kuap(AMR_KUAP_BLOCKED);
287 
288 	replay_soft_interrupts();
289 
290 	if (kuap_state != AMR_KUAP_BLOCKED)
291 		set_kuap(kuap_state);
292 }
293 #else
294 #define replay_soft_interrupts_irqrestore() replay_soft_interrupts()
295 #endif
296 
arch_local_irq_restore(unsigned long mask)297 notrace void arch_local_irq_restore(unsigned long mask)
298 {
299 	unsigned char irq_happened;
300 
301 	/* Write the new soft-enabled value */
302 	irq_soft_mask_set(mask);
303 	if (mask)
304 		return;
305 
306 	/*
307 	 * From this point onward, we can take interrupts, preempt,
308 	 * etc... unless we got hard-disabled. We check if an event
309 	 * happened. If none happened, we know we can just return.
310 	 *
311 	 * We may have preempted before the check below, in which case
312 	 * we are checking the "new" CPU instead of the old one. This
313 	 * is only a problem if an event happened on the "old" CPU.
314 	 *
315 	 * External interrupt events will have caused interrupts to
316 	 * be hard-disabled, so there is no problem, we
317 	 * cannot have preempted.
318 	 */
319 	irq_happened = get_irq_happened();
320 	if (!irq_happened) {
321 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
322 			WARN_ON_ONCE(!(mfmsr() & MSR_EE));
323 		return;
324 	}
325 
326 	/* We need to hard disable to replay. */
327 	if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
328 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
329 			WARN_ON_ONCE(!(mfmsr() & MSR_EE));
330 		__hard_irq_disable();
331 		local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
332 	} else {
333 		/*
334 		 * We should already be hard disabled here. We had bugs
335 		 * where that wasn't the case so let's dbl check it and
336 		 * warn if we are wrong. Only do that when IRQ tracing
337 		 * is enabled as mfmsr() can be costly.
338 		 */
339 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
340 			if (WARN_ON_ONCE(mfmsr() & MSR_EE))
341 				__hard_irq_disable();
342 		}
343 
344 		if (irq_happened == PACA_IRQ_HARD_DIS) {
345 			local_paca->irq_happened = 0;
346 			__hard_irq_enable();
347 			return;
348 		}
349 	}
350 
351 	/*
352 	 * Disable preempt here, so that the below preempt_enable will
353 	 * perform resched if required (a replayed interrupt may set
354 	 * need_resched).
355 	 */
356 	preempt_disable();
357 	irq_soft_mask_set(IRQS_ALL_DISABLED);
358 	trace_hardirqs_off();
359 
360 	replay_soft_interrupts_irqrestore();
361 	local_paca->irq_happened = 0;
362 
363 	trace_hardirqs_on();
364 	irq_soft_mask_set(IRQS_ENABLED);
365 	__hard_irq_enable();
366 	preempt_enable();
367 }
368 EXPORT_SYMBOL(arch_local_irq_restore);
369 
370 /*
371  * This is a helper to use when about to go into idle low-power
372  * when the latter has the side effect of re-enabling interrupts
373  * (such as calling H_CEDE under pHyp).
374  *
375  * You call this function with interrupts soft-disabled (this is
376  * already the case when ppc_md.power_save is called). The function
377  * will return whether to enter power save or just return.
378  *
379  * In the former case, it will have notified lockdep of interrupts
380  * being re-enabled and generally sanitized the lazy irq state,
381  * and in the latter case it will leave with interrupts hard
382  * disabled and marked as such, so the local_irq_enable() call
383  * in arch_cpu_idle() will properly re-enable everything.
384  */
prep_irq_for_idle(void)385 bool prep_irq_for_idle(void)
386 {
387 	/*
388 	 * First we need to hard disable to ensure no interrupt
389 	 * occurs before we effectively enter the low power state
390 	 */
391 	__hard_irq_disable();
392 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
393 
394 	/*
395 	 * If anything happened while we were soft-disabled,
396 	 * we return now and do not enter the low power state.
397 	 */
398 	if (lazy_irq_pending())
399 		return false;
400 
401 	/* Tell lockdep we are about to re-enable */
402 	trace_hardirqs_on();
403 
404 	/*
405 	 * Mark interrupts as soft-enabled and clear the
406 	 * PACA_IRQ_HARD_DIS from the pending mask since we
407 	 * are about to hard enable as well as a side effect
408 	 * of entering the low power state.
409 	 */
410 	local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
411 	irq_soft_mask_set(IRQS_ENABLED);
412 
413 	/* Tell the caller to enter the low power state */
414 	return true;
415 }
416 
417 #ifdef CONFIG_PPC_BOOK3S
418 /*
419  * This is for idle sequences that return with IRQs off, but the
420  * idle state itself wakes on interrupt. Tell the irq tracer that
421  * IRQs are enabled for the duration of idle so it does not get long
422  * off times. Must be paired with fini_irq_for_idle_irqsoff.
423  */
prep_irq_for_idle_irqsoff(void)424 bool prep_irq_for_idle_irqsoff(void)
425 {
426 	WARN_ON(!irqs_disabled());
427 
428 	/*
429 	 * First we need to hard disable to ensure no interrupt
430 	 * occurs before we effectively enter the low power state
431 	 */
432 	__hard_irq_disable();
433 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
434 
435 	/*
436 	 * If anything happened while we were soft-disabled,
437 	 * we return now and do not enter the low power state.
438 	 */
439 	if (lazy_irq_pending())
440 		return false;
441 
442 	/* Tell lockdep we are about to re-enable */
443 	trace_hardirqs_on();
444 
445 	return true;
446 }
447 
448 /*
449  * Take the SRR1 wakeup reason, index into this table to find the
450  * appropriate irq_happened bit.
451  *
452  * Sytem reset exceptions taken in idle state also come through here,
453  * but they are NMI interrupts so do not need to wait for IRQs to be
454  * restored, and should be taken as early as practical. These are marked
455  * with 0xff in the table. The Power ISA specifies 0100b as the system
456  * reset interrupt reason.
457  */
458 #define IRQ_SYSTEM_RESET	0xff
459 
460 static const u8 srr1_to_lazyirq[0x10] = {
461 	0, 0, 0,
462 	PACA_IRQ_DBELL,
463 	IRQ_SYSTEM_RESET,
464 	PACA_IRQ_DBELL,
465 	PACA_IRQ_DEC,
466 	0,
467 	PACA_IRQ_EE,
468 	PACA_IRQ_EE,
469 	PACA_IRQ_HMI,
470 	0, 0, 0, 0, 0 };
471 
replay_system_reset(void)472 void replay_system_reset(void)
473 {
474 	struct pt_regs regs;
475 
476 	ppc_save_regs(&regs);
477 	regs.trap = 0x100;
478 	get_paca()->in_nmi = 1;
479 	system_reset_exception(&regs);
480 	get_paca()->in_nmi = 0;
481 }
482 EXPORT_SYMBOL_GPL(replay_system_reset);
483 
irq_set_pending_from_srr1(unsigned long srr1)484 void irq_set_pending_from_srr1(unsigned long srr1)
485 {
486 	unsigned int idx = (srr1 & SRR1_WAKEMASK_P8) >> 18;
487 	u8 reason = srr1_to_lazyirq[idx];
488 
489 	/*
490 	 * Take the system reset now, which is immediately after registers
491 	 * are restored from idle. It's an NMI, so interrupts need not be
492 	 * re-enabled before it is taken.
493 	 */
494 	if (unlikely(reason == IRQ_SYSTEM_RESET)) {
495 		replay_system_reset();
496 		return;
497 	}
498 
499 	if (reason == PACA_IRQ_DBELL) {
500 		/*
501 		 * When doorbell triggers a system reset wakeup, the message
502 		 * is not cleared, so if the doorbell interrupt is replayed
503 		 * and the IPI handled, the doorbell interrupt would still
504 		 * fire when EE is enabled.
505 		 *
506 		 * To avoid taking the superfluous doorbell interrupt,
507 		 * execute a msgclr here before the interrupt is replayed.
508 		 */
509 		ppc_msgclr(PPC_DBELL_MSGTYPE);
510 	}
511 
512 	/*
513 	 * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
514 	 * so this can be called unconditionally with the SRR1 wake
515 	 * reason as returned by the idle code, which uses 0 to mean no
516 	 * interrupt.
517 	 *
518 	 * If a future CPU was to designate this as an interrupt reason,
519 	 * then a new index for no interrupt must be assigned.
520 	 */
521 	local_paca->irq_happened |= reason;
522 }
523 #endif /* CONFIG_PPC_BOOK3S */
524 
525 /*
526  * Force a replay of the external interrupt handler on this CPU.
527  */
force_external_irq_replay(void)528 void force_external_irq_replay(void)
529 {
530 	/*
531 	 * This must only be called with interrupts soft-disabled,
532 	 * the replay will happen when re-enabling.
533 	 */
534 	WARN_ON(!arch_irqs_disabled());
535 
536 	/*
537 	 * Interrupts must always be hard disabled before irq_happened is
538 	 * modified (to prevent lost update in case of interrupt between
539 	 * load and store).
540 	 */
541 	__hard_irq_disable();
542 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
543 
544 	/* Indicate in the PACA that we have an interrupt to replay */
545 	local_paca->irq_happened |= PACA_IRQ_EE;
546 }
547 
548 #endif /* CONFIG_PPC64 */
549 
arch_show_interrupts(struct seq_file * p,int prec)550 int arch_show_interrupts(struct seq_file *p, int prec)
551 {
552 	int j;
553 
554 #if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
555 	if (tau_initialized) {
556 		seq_printf(p, "%*s: ", prec, "TAU");
557 		for_each_online_cpu(j)
558 			seq_printf(p, "%10u ", tau_interrupts(j));
559 		seq_puts(p, "  PowerPC             Thermal Assist (cpu temp)\n");
560 	}
561 #endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
562 
563 	seq_printf(p, "%*s: ", prec, "LOC");
564 	for_each_online_cpu(j)
565 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_event);
566         seq_printf(p, "  Local timer interrupts for timer event device\n");
567 
568 	seq_printf(p, "%*s: ", prec, "BCT");
569 	for_each_online_cpu(j)
570 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).broadcast_irqs_event);
571 	seq_printf(p, "  Broadcast timer interrupts for timer event device\n");
572 
573 	seq_printf(p, "%*s: ", prec, "LOC");
574 	for_each_online_cpu(j)
575 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_others);
576         seq_printf(p, "  Local timer interrupts for others\n");
577 
578 	seq_printf(p, "%*s: ", prec, "SPU");
579 	for_each_online_cpu(j)
580 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
581 	seq_printf(p, "  Spurious interrupts\n");
582 
583 	seq_printf(p, "%*s: ", prec, "PMI");
584 	for_each_online_cpu(j)
585 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
586 	seq_printf(p, "  Performance monitoring interrupts\n");
587 
588 	seq_printf(p, "%*s: ", prec, "MCE");
589 	for_each_online_cpu(j)
590 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
591 	seq_printf(p, "  Machine check exceptions\n");
592 
593 #ifdef CONFIG_PPC_BOOK3S_64
594 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
595 		seq_printf(p, "%*s: ", prec, "HMI");
596 		for_each_online_cpu(j)
597 			seq_printf(p, "%10u ", paca_ptrs[j]->hmi_irqs);
598 		seq_printf(p, "  Hypervisor Maintenance Interrupts\n");
599 	}
600 #endif
601 
602 	seq_printf(p, "%*s: ", prec, "NMI");
603 	for_each_online_cpu(j)
604 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).sreset_irqs);
605 	seq_printf(p, "  System Reset interrupts\n");
606 
607 #ifdef CONFIG_PPC_WATCHDOG
608 	seq_printf(p, "%*s: ", prec, "WDG");
609 	for_each_online_cpu(j)
610 		seq_printf(p, "%10u ", per_cpu(irq_stat, j).soft_nmi_irqs);
611 	seq_printf(p, "  Watchdog soft-NMI interrupts\n");
612 #endif
613 
614 #ifdef CONFIG_PPC_DOORBELL
615 	if (cpu_has_feature(CPU_FTR_DBELL)) {
616 		seq_printf(p, "%*s: ", prec, "DBL");
617 		for_each_online_cpu(j)
618 			seq_printf(p, "%10u ", per_cpu(irq_stat, j).doorbell_irqs);
619 		seq_printf(p, "  Doorbell interrupts\n");
620 	}
621 #endif
622 
623 	return 0;
624 }
625 
626 /*
627  * /proc/stat helpers
628  */
arch_irq_stat_cpu(unsigned int cpu)629 u64 arch_irq_stat_cpu(unsigned int cpu)
630 {
631 	u64 sum = per_cpu(irq_stat, cpu).timer_irqs_event;
632 
633 	sum += per_cpu(irq_stat, cpu).broadcast_irqs_event;
634 	sum += per_cpu(irq_stat, cpu).pmu_irqs;
635 	sum += per_cpu(irq_stat, cpu).mce_exceptions;
636 	sum += per_cpu(irq_stat, cpu).spurious_irqs;
637 	sum += per_cpu(irq_stat, cpu).timer_irqs_others;
638 #ifdef CONFIG_PPC_BOOK3S_64
639 	sum += paca_ptrs[cpu]->hmi_irqs;
640 #endif
641 	sum += per_cpu(irq_stat, cpu).sreset_irqs;
642 #ifdef CONFIG_PPC_WATCHDOG
643 	sum += per_cpu(irq_stat, cpu).soft_nmi_irqs;
644 #endif
645 #ifdef CONFIG_PPC_DOORBELL
646 	sum += per_cpu(irq_stat, cpu).doorbell_irqs;
647 #endif
648 
649 	return sum;
650 }
651 
check_stack_overflow(void)652 static inline void check_stack_overflow(void)
653 {
654 	long sp;
655 
656 	if (!IS_ENABLED(CONFIG_DEBUG_STACKOVERFLOW))
657 		return;
658 
659 	sp = current_stack_pointer & (THREAD_SIZE - 1);
660 
661 	/* check for stack overflow: is there less than 2KB free? */
662 	if (unlikely(sp < 2048)) {
663 		pr_err("do_IRQ: stack overflow: %ld\n", sp);
664 		dump_stack();
665 	}
666 }
667 
__do_irq(struct pt_regs * regs)668 void __do_irq(struct pt_regs *regs)
669 {
670 	unsigned int irq;
671 
672 	irq_enter();
673 
674 	trace_irq_entry(regs);
675 
676 	/*
677 	 * Query the platform PIC for the interrupt & ack it.
678 	 *
679 	 * This will typically lower the interrupt line to the CPU
680 	 */
681 	irq = ppc_md.get_irq();
682 
683 	/* We can hard enable interrupts now to allow perf interrupts */
684 	may_hard_irq_enable();
685 
686 	/* And finally process it */
687 	if (unlikely(!irq))
688 		__this_cpu_inc(irq_stat.spurious_irqs);
689 	else
690 		generic_handle_irq(irq);
691 
692 	trace_irq_exit(regs);
693 
694 	irq_exit();
695 }
696 
do_IRQ(struct pt_regs * regs)697 void do_IRQ(struct pt_regs *regs)
698 {
699 	struct pt_regs *old_regs = set_irq_regs(regs);
700 	void *cursp, *irqsp, *sirqsp;
701 
702 	/* Switch to the irq stack to handle this */
703 	cursp = (void *)(current_stack_pointer & ~(THREAD_SIZE - 1));
704 	irqsp = hardirq_ctx[raw_smp_processor_id()];
705 	sirqsp = softirq_ctx[raw_smp_processor_id()];
706 
707 	check_stack_overflow();
708 
709 	/* Already there ? */
710 	if (unlikely(cursp == irqsp || cursp == sirqsp)) {
711 		__do_irq(regs);
712 		set_irq_regs(old_regs);
713 		return;
714 	}
715 	/* Switch stack and call */
716 	call_do_irq(regs, irqsp);
717 
718 	set_irq_regs(old_regs);
719 }
720 
alloc_vm_stack(void)721 static void *__init alloc_vm_stack(void)
722 {
723 	return __vmalloc_node(THREAD_SIZE, THREAD_ALIGN, THREADINFO_GFP,
724 			      NUMA_NO_NODE, (void *)_RET_IP_);
725 }
726 
vmap_irqstack_init(void)727 static void __init vmap_irqstack_init(void)
728 {
729 	int i;
730 
731 	for_each_possible_cpu(i) {
732 		softirq_ctx[i] = alloc_vm_stack();
733 		hardirq_ctx[i] = alloc_vm_stack();
734 	}
735 }
736 
737 
init_IRQ(void)738 void __init init_IRQ(void)
739 {
740 	if (IS_ENABLED(CONFIG_VMAP_STACK))
741 		vmap_irqstack_init();
742 
743 	if (ppc_md.init_IRQ)
744 		ppc_md.init_IRQ();
745 }
746 
747 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
748 void   *critirq_ctx[NR_CPUS] __read_mostly;
749 void    *dbgirq_ctx[NR_CPUS] __read_mostly;
750 void *mcheckirq_ctx[NR_CPUS] __read_mostly;
751 #endif
752 
753 void *softirq_ctx[NR_CPUS] __read_mostly;
754 void *hardirq_ctx[NR_CPUS] __read_mostly;
755 
do_softirq_own_stack(void)756 void do_softirq_own_stack(void)
757 {
758 	call_do_softirq(softirq_ctx[smp_processor_id()]);
759 }
760 
virq_to_hw(unsigned int virq)761 irq_hw_number_t virq_to_hw(unsigned int virq)
762 {
763 	struct irq_data *irq_data = irq_get_irq_data(virq);
764 	return WARN_ON(!irq_data) ? 0 : irq_data->hwirq;
765 }
766 EXPORT_SYMBOL_GPL(virq_to_hw);
767 
768 #ifdef CONFIG_SMP
irq_choose_cpu(const struct cpumask * mask)769 int irq_choose_cpu(const struct cpumask *mask)
770 {
771 	int cpuid;
772 
773 	if (cpumask_equal(mask, cpu_online_mask)) {
774 		static int irq_rover;
775 		static DEFINE_RAW_SPINLOCK(irq_rover_lock);
776 		unsigned long flags;
777 
778 		/* Round-robin distribution... */
779 do_round_robin:
780 		raw_spin_lock_irqsave(&irq_rover_lock, flags);
781 
782 		irq_rover = cpumask_next(irq_rover, cpu_online_mask);
783 		if (irq_rover >= nr_cpu_ids)
784 			irq_rover = cpumask_first(cpu_online_mask);
785 
786 		cpuid = irq_rover;
787 
788 		raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
789 	} else {
790 		cpuid = cpumask_first_and(mask, cpu_online_mask);
791 		if (cpuid >= nr_cpu_ids)
792 			goto do_round_robin;
793 	}
794 
795 	return get_hard_smp_processor_id(cpuid);
796 }
797 #else
irq_choose_cpu(const struct cpumask * mask)798 int irq_choose_cpu(const struct cpumask *mask)
799 {
800 	return hard_smp_processor_id();
801 }
802 #endif
803 
804 #ifdef CONFIG_PPC64
setup_noirqdistrib(char * str)805 static int __init setup_noirqdistrib(char *str)
806 {
807 	distribute_irqs = 0;
808 	return 1;
809 }
810 
811 __setup("noirqdistrib", setup_noirqdistrib);
812 #endif /* CONFIG_PPC64 */
813