1 // SPDX-License-Identifier: GPL-2.0-or-later
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)
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
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.
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>
52 #include <linux/of_irq.h>
54 #include <linux/uaccess.h>
56 #include <asm/pgtable.h>
58 #include <asm/cache.h>
60 #include <asm/ptrace.h>
61 #include <asm/machdep.h>
64 #include <asm/livepatch.h>
65 #include <asm/asm-prototypes.h>
66 #include <asm/hw_irq.h>
70 #include <asm/firmware.h>
71 #include <asm/lv1call.h>
73 #define CREATE_TRACE_POINTS
74 #include <asm/trace.h>
75 #include <asm/cpu_has_feature.h>
77 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
78 EXPORT_PER_CPU_SYMBOL(irq_stat);
81 atomic_t ppc_n_lost_interrupts;
84 extern int tau_initialized;
85 u32 tau_interrupts(unsigned long cpu);
87 #endif /* CONFIG_PPC32 */
91 int distribute_irqs = 1;
93 static inline notrace unsigned long get_irq_happened(void)
95 unsigned long happened;
97 __asm__ __volatile__("lbz %0,%1(13)"
98 : "=r" (happened) : "i" (offsetof(struct paca_struct, irq_happened)));
103 static inline notrace int decrementer_check_overflow(void)
105 u64 now = get_tb_or_rtc();
106 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
108 return now >= *next_tb;
111 /* This is called whenever we are re-enabling interrupts
112 * and returns either 0 (nothing to do) or 500/900/280/a00/e80 if
113 * there's an EE, DEC or DBELL to generate.
115 * This is called in two contexts: From arch_local_irq_restore()
116 * before soft-enabling interrupts, and from the exception exit
117 * path when returning from an interrupt from a soft-disabled to
118 * a soft enabled context. In both case we have interrupts hard
121 * We take care of only clearing the bits we handled in the
122 * PACA irq_happened field since we can only re-emit one at a
123 * time and we don't want to "lose" one.
125 notrace unsigned int __check_irq_replay(void)
128 * We use local_paca rather than get_paca() to avoid all
129 * the debug_smp_processor_id() business in this low level
132 unsigned char happened = local_paca->irq_happened;
135 * We are responding to the next interrupt, so interrupt-off
136 * latencies should be reset here.
139 trace_hardirqs_off();
142 * We are always hard disabled here, but PACA_IRQ_HARD_DIS may
143 * not be set, which means interrupts have only just been hard
144 * disabled as part of the local_irq_restore or interrupt return
145 * code. In that case, skip the decrementr check becaus it's
146 * expensive to read the TB.
148 * HARD_DIS then gets cleared here, but it's reconciled later.
149 * Either local_irq_disable will replay the interrupt and that
150 * will reconcile state like other hard interrupts. Or interrupt
151 * retur will replay the interrupt and in that case it sets
152 * PACA_IRQ_HARD_DIS by hand (see comments in entry_64.S).
154 if (happened & PACA_IRQ_HARD_DIS) {
155 local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
158 * We may have missed a decrementer interrupt if hard disabled.
159 * Check the decrementer register in case we had a rollover
160 * while hard disabled.
162 if (!(happened & PACA_IRQ_DEC)) {
163 if (decrementer_check_overflow()) {
164 local_paca->irq_happened |= PACA_IRQ_DEC;
165 happened |= PACA_IRQ_DEC;
171 * Force the delivery of pending soft-disabled interrupts on PS3.
172 * Any HV call will have this side effect.
174 if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
176 lv1_get_version_info(&tmp, &tmp2);
180 * Check if an hypervisor Maintenance interrupt happened.
181 * This is a higher priority interrupt than the others, so
184 if (happened & PACA_IRQ_HMI) {
185 local_paca->irq_happened &= ~PACA_IRQ_HMI;
189 if (happened & PACA_IRQ_DEC) {
190 local_paca->irq_happened &= ~PACA_IRQ_DEC;
194 if (happened & PACA_IRQ_PMI) {
195 local_paca->irq_happened &= ~PACA_IRQ_PMI;
199 if (happened & PACA_IRQ_EE) {
200 local_paca->irq_happened &= ~PACA_IRQ_EE;
204 #ifdef CONFIG_PPC_BOOK3E
206 * Check if an EPR external interrupt happened this bit is typically
207 * set if we need to handle another "edge" interrupt from within the
208 * MPIC "EPR" handler.
210 if (happened & PACA_IRQ_EE_EDGE) {
211 local_paca->irq_happened &= ~PACA_IRQ_EE_EDGE;
215 if (happened & PACA_IRQ_DBELL) {
216 local_paca->irq_happened &= ~PACA_IRQ_DBELL;
220 if (happened & PACA_IRQ_DBELL) {
221 local_paca->irq_happened &= ~PACA_IRQ_DBELL;
224 #endif /* CONFIG_PPC_BOOK3E */
226 /* There should be nothing left ! */
227 BUG_ON(local_paca->irq_happened != 0);
232 notrace void arch_local_irq_restore(unsigned long mask)
234 unsigned char irq_happened;
237 /* Write the new soft-enabled value */
238 irq_soft_mask_set(mask);
243 * From this point onward, we can take interrupts, preempt,
244 * etc... unless we got hard-disabled. We check if an event
245 * happened. If none happened, we know we can just return.
247 * We may have preempted before the check below, in which case
248 * we are checking the "new" CPU instead of the old one. This
249 * is only a problem if an event happened on the "old" CPU.
251 * External interrupt events will have caused interrupts to
252 * be hard-disabled, so there is no problem, we
253 * cannot have preempted.
255 irq_happened = get_irq_happened();
257 #ifdef CONFIG_PPC_IRQ_SOFT_MASK_DEBUG
258 WARN_ON_ONCE(!(mfmsr() & MSR_EE));
264 * We need to hard disable to get a trusted value from
265 * __check_irq_replay(). We also need to soft-disable
266 * again to avoid warnings in there due to the use of
269 if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
270 #ifdef CONFIG_PPC_IRQ_SOFT_MASK_DEBUG
271 WARN_ON_ONCE(!(mfmsr() & MSR_EE));
273 __hard_irq_disable();
274 #ifdef CONFIG_PPC_IRQ_SOFT_MASK_DEBUG
277 * We should already be hard disabled here. We had bugs
278 * where that wasn't the case so let's dbl check it and
279 * warn if we are wrong. Only do that when IRQ tracing
280 * is enabled as mfmsr() can be costly.
282 if (WARN_ON_ONCE(mfmsr() & MSR_EE))
283 __hard_irq_disable();
287 irq_soft_mask_set(IRQS_ALL_DISABLED);
288 trace_hardirqs_off();
291 * Check if anything needs to be re-emitted. We haven't
292 * soft-enabled yet to avoid warnings in decrementer_check_overflow
293 * accessing per-cpu variables
295 replay = __check_irq_replay();
297 /* We can soft-enable now */
299 irq_soft_mask_set(IRQS_ENABLED);
302 * And replay if we have to. This will return with interrupts
306 __replay_interrupt(replay);
310 /* Finally, let's ensure we are hard enabled */
313 EXPORT_SYMBOL(arch_local_irq_restore);
316 * This is specifically called by assembly code to re-enable interrupts
317 * if they are currently disabled. This is typically called before
318 * schedule() or do_signal() when returning to userspace. We do it
319 * in C to avoid the burden of dealing with lockdep etc...
321 * NOTE: This is called with interrupts hard disabled but not marked
322 * as such in paca->irq_happened, so we need to resync this.
324 void notrace restore_interrupts(void)
326 if (irqs_disabled()) {
327 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
334 * This is a helper to use when about to go into idle low-power
335 * when the latter has the side effect of re-enabling interrupts
336 * (such as calling H_CEDE under pHyp).
338 * You call this function with interrupts soft-disabled (this is
339 * already the case when ppc_md.power_save is called). The function
340 * will return whether to enter power save or just return.
342 * In the former case, it will have notified lockdep of interrupts
343 * being re-enabled and generally sanitized the lazy irq state,
344 * and in the latter case it will leave with interrupts hard
345 * disabled and marked as such, so the local_irq_enable() call
346 * in arch_cpu_idle() will properly re-enable everything.
348 bool prep_irq_for_idle(void)
351 * First we need to hard disable to ensure no interrupt
352 * occurs before we effectively enter the low power state
354 __hard_irq_disable();
355 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
358 * If anything happened while we were soft-disabled,
359 * we return now and do not enter the low power state.
361 if (lazy_irq_pending())
364 /* Tell lockdep we are about to re-enable */
368 * Mark interrupts as soft-enabled and clear the
369 * PACA_IRQ_HARD_DIS from the pending mask since we
370 * are about to hard enable as well as a side effect
371 * of entering the low power state.
373 local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
374 irq_soft_mask_set(IRQS_ENABLED);
376 /* Tell the caller to enter the low power state */
380 #ifdef CONFIG_PPC_BOOK3S
382 * This is for idle sequences that return with IRQs off, but the
383 * idle state itself wakes on interrupt. Tell the irq tracer that
384 * IRQs are enabled for the duration of idle so it does not get long
385 * off times. Must be paired with fini_irq_for_idle_irqsoff.
387 bool prep_irq_for_idle_irqsoff(void)
389 WARN_ON(!irqs_disabled());
392 * First we need to hard disable to ensure no interrupt
393 * occurs before we effectively enter the low power state
395 __hard_irq_disable();
396 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
399 * If anything happened while we were soft-disabled,
400 * we return now and do not enter the low power state.
402 if (lazy_irq_pending())
405 /* Tell lockdep we are about to re-enable */
412 * Take the SRR1 wakeup reason, index into this table to find the
413 * appropriate irq_happened bit.
415 * Sytem reset exceptions taken in idle state also come through here,
416 * but they are NMI interrupts so do not need to wait for IRQs to be
417 * restored, and should be taken as early as practical. These are marked
418 * with 0xff in the table. The Power ISA specifies 0100b as the system
419 * reset interrupt reason.
421 #define IRQ_SYSTEM_RESET 0xff
423 static const u8 srr1_to_lazyirq[0x10] = {
435 void replay_system_reset(void)
439 ppc_save_regs(®s);
441 get_paca()->in_nmi = 1;
442 system_reset_exception(®s);
443 get_paca()->in_nmi = 0;
445 EXPORT_SYMBOL_GPL(replay_system_reset);
447 void irq_set_pending_from_srr1(unsigned long srr1)
449 unsigned int idx = (srr1 & SRR1_WAKEMASK_P8) >> 18;
450 u8 reason = srr1_to_lazyirq[idx];
453 * Take the system reset now, which is immediately after registers
454 * are restored from idle. It's an NMI, so interrupts need not be
455 * re-enabled before it is taken.
457 if (unlikely(reason == IRQ_SYSTEM_RESET)) {
458 replay_system_reset();
463 * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
464 * so this can be called unconditionally with the SRR1 wake
465 * reason as returned by the idle code, which uses 0 to mean no
468 * If a future CPU was to designate this as an interrupt reason,
469 * then a new index for no interrupt must be assigned.
471 local_paca->irq_happened |= reason;
473 #endif /* CONFIG_PPC_BOOK3S */
476 * Force a replay of the external interrupt handler on this CPU.
478 void force_external_irq_replay(void)
481 * This must only be called with interrupts soft-disabled,
482 * the replay will happen when re-enabling.
484 WARN_ON(!arch_irqs_disabled());
487 * Interrupts must always be hard disabled before irq_happened is
488 * modified (to prevent lost update in case of interrupt between
491 __hard_irq_disable();
492 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
494 /* Indicate in the PACA that we have an interrupt to replay */
495 local_paca->irq_happened |= PACA_IRQ_EE;
498 #endif /* CONFIG_PPC64 */
500 int arch_show_interrupts(struct seq_file *p, int prec)
504 #if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
505 if (tau_initialized) {
506 seq_printf(p, "%*s: ", prec, "TAU");
507 for_each_online_cpu(j)
508 seq_printf(p, "%10u ", tau_interrupts(j));
509 seq_puts(p, " PowerPC Thermal Assist (cpu temp)\n");
511 #endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
513 seq_printf(p, "%*s: ", prec, "LOC");
514 for_each_online_cpu(j)
515 seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_event);
516 seq_printf(p, " Local timer interrupts for timer event device\n");
518 seq_printf(p, "%*s: ", prec, "BCT");
519 for_each_online_cpu(j)
520 seq_printf(p, "%10u ", per_cpu(irq_stat, j).broadcast_irqs_event);
521 seq_printf(p, " Broadcast timer interrupts for timer event device\n");
523 seq_printf(p, "%*s: ", prec, "LOC");
524 for_each_online_cpu(j)
525 seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_others);
526 seq_printf(p, " Local timer interrupts for others\n");
528 seq_printf(p, "%*s: ", prec, "SPU");
529 for_each_online_cpu(j)
530 seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
531 seq_printf(p, " Spurious interrupts\n");
533 seq_printf(p, "%*s: ", prec, "PMI");
534 for_each_online_cpu(j)
535 seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
536 seq_printf(p, " Performance monitoring interrupts\n");
538 seq_printf(p, "%*s: ", prec, "MCE");
539 for_each_online_cpu(j)
540 seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
541 seq_printf(p, " Machine check exceptions\n");
543 if (cpu_has_feature(CPU_FTR_HVMODE)) {
544 seq_printf(p, "%*s: ", prec, "HMI");
545 for_each_online_cpu(j)
546 seq_printf(p, "%10u ",
547 per_cpu(irq_stat, j).hmi_exceptions);
548 seq_printf(p, " Hypervisor Maintenance Interrupts\n");
551 seq_printf(p, "%*s: ", prec, "NMI");
552 for_each_online_cpu(j)
553 seq_printf(p, "%10u ", per_cpu(irq_stat, j).sreset_irqs);
554 seq_printf(p, " System Reset interrupts\n");
556 #ifdef CONFIG_PPC_WATCHDOG
557 seq_printf(p, "%*s: ", prec, "WDG");
558 for_each_online_cpu(j)
559 seq_printf(p, "%10u ", per_cpu(irq_stat, j).soft_nmi_irqs);
560 seq_printf(p, " Watchdog soft-NMI interrupts\n");
563 #ifdef CONFIG_PPC_DOORBELL
564 if (cpu_has_feature(CPU_FTR_DBELL)) {
565 seq_printf(p, "%*s: ", prec, "DBL");
566 for_each_online_cpu(j)
567 seq_printf(p, "%10u ", per_cpu(irq_stat, j).doorbell_irqs);
568 seq_printf(p, " Doorbell interrupts\n");
578 u64 arch_irq_stat_cpu(unsigned int cpu)
580 u64 sum = per_cpu(irq_stat, cpu).timer_irqs_event;
582 sum += per_cpu(irq_stat, cpu).broadcast_irqs_event;
583 sum += per_cpu(irq_stat, cpu).pmu_irqs;
584 sum += per_cpu(irq_stat, cpu).mce_exceptions;
585 sum += per_cpu(irq_stat, cpu).spurious_irqs;
586 sum += per_cpu(irq_stat, cpu).timer_irqs_others;
587 sum += per_cpu(irq_stat, cpu).hmi_exceptions;
588 sum += per_cpu(irq_stat, cpu).sreset_irqs;
589 #ifdef CONFIG_PPC_WATCHDOG
590 sum += per_cpu(irq_stat, cpu).soft_nmi_irqs;
592 #ifdef CONFIG_PPC_DOORBELL
593 sum += per_cpu(irq_stat, cpu).doorbell_irqs;
599 static inline void check_stack_overflow(void)
601 #ifdef CONFIG_DEBUG_STACKOVERFLOW
604 sp = current_stack_pointer() & (THREAD_SIZE-1);
606 /* check for stack overflow: is there less than 2KB free? */
607 if (unlikely(sp < 2048)) {
608 pr_err("do_IRQ: stack overflow: %ld\n", sp);
614 void __do_irq(struct pt_regs *regs)
620 trace_irq_entry(regs);
623 * Query the platform PIC for the interrupt & ack it.
625 * This will typically lower the interrupt line to the CPU
627 irq = ppc_md.get_irq();
629 /* We can hard enable interrupts now to allow perf interrupts */
630 may_hard_irq_enable();
632 /* And finally process it */
634 __this_cpu_inc(irq_stat.spurious_irqs);
636 generic_handle_irq(irq);
638 trace_irq_exit(regs);
643 void do_IRQ(struct pt_regs *regs)
645 struct pt_regs *old_regs = set_irq_regs(regs);
646 void *cursp, *irqsp, *sirqsp;
648 /* Switch to the irq stack to handle this */
649 cursp = (void *)(current_stack_pointer() & ~(THREAD_SIZE - 1));
650 irqsp = hardirq_ctx[raw_smp_processor_id()];
651 sirqsp = softirq_ctx[raw_smp_processor_id()];
653 check_stack_overflow();
655 /* Already there ? */
656 if (unlikely(cursp == irqsp || cursp == sirqsp)) {
658 set_irq_regs(old_regs);
661 /* Switch stack and call */
662 call_do_irq(regs, irqsp);
664 set_irq_regs(old_regs);
667 void __init init_IRQ(void)
673 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
674 void *critirq_ctx[NR_CPUS] __read_mostly;
675 void *dbgirq_ctx[NR_CPUS] __read_mostly;
676 void *mcheckirq_ctx[NR_CPUS] __read_mostly;
679 void *softirq_ctx[NR_CPUS] __read_mostly;
680 void *hardirq_ctx[NR_CPUS] __read_mostly;
682 void do_softirq_own_stack(void)
684 call_do_softirq(softirq_ctx[smp_processor_id()]);
687 irq_hw_number_t virq_to_hw(unsigned int virq)
689 struct irq_data *irq_data = irq_get_irq_data(virq);
690 return WARN_ON(!irq_data) ? 0 : irq_data->hwirq;
692 EXPORT_SYMBOL_GPL(virq_to_hw);
695 int irq_choose_cpu(const struct cpumask *mask)
699 if (cpumask_equal(mask, cpu_online_mask)) {
700 static int irq_rover;
701 static DEFINE_RAW_SPINLOCK(irq_rover_lock);
704 /* Round-robin distribution... */
706 raw_spin_lock_irqsave(&irq_rover_lock, flags);
708 irq_rover = cpumask_next(irq_rover, cpu_online_mask);
709 if (irq_rover >= nr_cpu_ids)
710 irq_rover = cpumask_first(cpu_online_mask);
714 raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
716 cpuid = cpumask_first_and(mask, cpu_online_mask);
717 if (cpuid >= nr_cpu_ids)
721 return get_hard_smp_processor_id(cpuid);
724 int irq_choose_cpu(const struct cpumask *mask)
726 return hard_smp_processor_id();
731 static int __init setup_noirqdistrib(char *str)
737 __setup("noirqdistrib", setup_noirqdistrib);
738 #endif /* CONFIG_PPC64 */