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
8 * Copyright (C) 1996-2001 Cort Dougan
9 * Adapted for Power Macintosh by Paul Mackerras
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>
53 #include <linux/vmalloc.h>
54 #include <linux/pgtable.h>
56 #include <linux/uaccess.h>
57 #include <asm/interrupt.h>
60 #include <asm/cache.h>
62 #include <asm/ptrace.h>
63 #include <asm/machdep.h>
66 #include <asm/livepatch.h>
67 #include <asm/asm-prototypes.h>
68 #include <asm/hw_irq.h>
69 #include <asm/softirq_stack.h>
73 #include <asm/firmware.h>
74 #include <asm/lv1call.h>
75 #include <asm/dbell.h>
77 #define CREATE_TRACE_POINTS
78 #include <asm/trace.h>
79 #include <asm/cpu_has_feature.h>
81 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
82 EXPORT_PER_CPU_SYMBOL(irq_stat);
85 atomic_t ppc_n_lost_interrupts;
88 extern int tau_initialized;
89 u32 tau_interrupts(unsigned long cpu);
91 #endif /* CONFIG_PPC32 */
95 int distribute_irqs = 1;
97 static inline notrace unsigned long get_irq_happened(void)
99 unsigned long happened;
101 __asm__ __volatile__("lbz %0,%1(13)"
102 : "=r" (happened) : "i" (offsetof(struct paca_struct, irq_happened)));
107 void replay_soft_interrupts(void)
112 * Be careful here, calling these interrupt handlers can cause
113 * softirqs to be raised, which they may run when calling irq_exit,
114 * which will cause local_irq_enable() to be run, which can then
115 * recurse into this function. Don't keep any state across
116 * interrupt handler calls which may change underneath us.
118 * We use local_paca rather than get_paca() to avoid all the
119 * debug_smp_processor_id() business in this low level function.
122 ppc_save_regs(®s);
123 regs.softe = IRQS_ENABLED;
126 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
127 WARN_ON_ONCE(mfmsr() & MSR_EE);
130 * Force the delivery of pending soft-disabled interrupts on PS3.
131 * Any HV call will have this side effect.
133 if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
135 lv1_get_version_info(&tmp, &tmp2);
139 * Check if an hypervisor Maintenance interrupt happened.
140 * This is a higher priority interrupt than the others, so
143 if (IS_ENABLED(CONFIG_PPC_BOOK3S) && (local_paca->irq_happened & PACA_IRQ_HMI)) {
144 local_paca->irq_happened &= ~PACA_IRQ_HMI;
145 regs.trap = INTERRUPT_HMI;
146 handle_hmi_exception(®s);
147 if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
151 if (local_paca->irq_happened & PACA_IRQ_DEC) {
152 local_paca->irq_happened &= ~PACA_IRQ_DEC;
153 regs.trap = INTERRUPT_DECREMENTER;
154 timer_interrupt(®s);
155 if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
159 if (local_paca->irq_happened & PACA_IRQ_EE) {
160 local_paca->irq_happened &= ~PACA_IRQ_EE;
161 regs.trap = INTERRUPT_EXTERNAL;
163 if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
167 if (IS_ENABLED(CONFIG_PPC_DOORBELL) && (local_paca->irq_happened & PACA_IRQ_DBELL)) {
168 local_paca->irq_happened &= ~PACA_IRQ_DBELL;
169 regs.trap = INTERRUPT_DOORBELL;
170 doorbell_exception(®s);
171 if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
175 /* Book3E does not support soft-masking PMI interrupts */
176 if (IS_ENABLED(CONFIG_PPC_BOOK3S) && (local_paca->irq_happened & PACA_IRQ_PMI)) {
177 local_paca->irq_happened &= ~PACA_IRQ_PMI;
178 regs.trap = INTERRUPT_PERFMON;
179 performance_monitor_exception(®s);
180 if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
184 if (local_paca->irq_happened & ~PACA_IRQ_HARD_DIS) {
186 * We are responding to the next interrupt, so interrupt-off
187 * latencies should be reset here.
190 trace_hardirqs_off();
195 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_KUAP)
196 static inline void replay_soft_interrupts_irqrestore(void)
198 unsigned long kuap_state = get_kuap();
201 * Check if anything calls local_irq_enable/restore() when KUAP is
202 * disabled (user access enabled). We handle that case here by saving
203 * and re-locking AMR but we shouldn't get here in the first place,
206 kuap_assert_locked();
208 if (kuap_state != AMR_KUAP_BLOCKED)
209 set_kuap(AMR_KUAP_BLOCKED);
211 replay_soft_interrupts();
213 if (kuap_state != AMR_KUAP_BLOCKED)
214 set_kuap(kuap_state);
217 #define replay_soft_interrupts_irqrestore() replay_soft_interrupts()
220 notrace void arch_local_irq_restore(unsigned long mask)
222 unsigned char irq_happened;
224 /* Write the new soft-enabled value */
225 irq_soft_mask_set(mask);
230 * From this point onward, we can take interrupts, preempt,
231 * etc... unless we got hard-disabled. We check if an event
232 * happened. If none happened, we know we can just return.
234 * We may have preempted before the check below, in which case
235 * we are checking the "new" CPU instead of the old one. This
236 * is only a problem if an event happened on the "old" CPU.
238 * External interrupt events will have caused interrupts to
239 * be hard-disabled, so there is no problem, we
240 * cannot have preempted.
242 irq_happened = get_irq_happened();
244 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
245 WARN_ON_ONCE(!(mfmsr() & MSR_EE));
249 /* We need to hard disable to replay. */
250 if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
251 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
252 WARN_ON_ONCE(!(mfmsr() & MSR_EE));
253 __hard_irq_disable();
254 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
257 * We should already be hard disabled here. We had bugs
258 * where that wasn't the case so let's dbl check it and
259 * warn if we are wrong. Only do that when IRQ tracing
260 * is enabled as mfmsr() can be costly.
262 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
263 if (WARN_ON_ONCE(mfmsr() & MSR_EE))
264 __hard_irq_disable();
267 if (irq_happened == PACA_IRQ_HARD_DIS) {
268 local_paca->irq_happened = 0;
275 * Disable preempt here, so that the below preempt_enable will
276 * perform resched if required (a replayed interrupt may set
280 irq_soft_mask_set(IRQS_ALL_DISABLED);
281 trace_hardirqs_off();
283 replay_soft_interrupts_irqrestore();
284 local_paca->irq_happened = 0;
287 irq_soft_mask_set(IRQS_ENABLED);
291 EXPORT_SYMBOL(arch_local_irq_restore);
294 * This is a helper to use when about to go into idle low-power
295 * when the latter has the side effect of re-enabling interrupts
296 * (such as calling H_CEDE under pHyp).
298 * You call this function with interrupts soft-disabled (this is
299 * already the case when ppc_md.power_save is called). The function
300 * will return whether to enter power save or just return.
302 * In the former case, it will have notified lockdep of interrupts
303 * being re-enabled and generally sanitized the lazy irq state,
304 * and in the latter case it will leave with interrupts hard
305 * disabled and marked as such, so the local_irq_enable() call
306 * in arch_cpu_idle() will properly re-enable everything.
308 bool prep_irq_for_idle(void)
311 * First we need to hard disable to ensure no interrupt
312 * occurs before we effectively enter the low power state
314 __hard_irq_disable();
315 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
318 * If anything happened while we were soft-disabled,
319 * we return now and do not enter the low power state.
321 if (lazy_irq_pending())
324 /* Tell lockdep we are about to re-enable */
328 * Mark interrupts as soft-enabled and clear the
329 * PACA_IRQ_HARD_DIS from the pending mask since we
330 * are about to hard enable as well as a side effect
331 * of entering the low power state.
333 local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
334 irq_soft_mask_set(IRQS_ENABLED);
336 /* Tell the caller to enter the low power state */
340 #ifdef CONFIG_PPC_BOOK3S
342 * This is for idle sequences that return with IRQs off, but the
343 * idle state itself wakes on interrupt. Tell the irq tracer that
344 * IRQs are enabled for the duration of idle so it does not get long
345 * off times. Must be paired with fini_irq_for_idle_irqsoff.
347 bool prep_irq_for_idle_irqsoff(void)
349 WARN_ON(!irqs_disabled());
352 * First we need to hard disable to ensure no interrupt
353 * occurs before we effectively enter the low power state
355 __hard_irq_disable();
356 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
359 * If anything happened while we were soft-disabled,
360 * we return now and do not enter the low power state.
362 if (lazy_irq_pending())
365 /* Tell lockdep we are about to re-enable */
372 * Take the SRR1 wakeup reason, index into this table to find the
373 * appropriate irq_happened bit.
375 * Sytem reset exceptions taken in idle state also come through here,
376 * but they are NMI interrupts so do not need to wait for IRQs to be
377 * restored, and should be taken as early as practical. These are marked
378 * with 0xff in the table. The Power ISA specifies 0100b as the system
379 * reset interrupt reason.
381 #define IRQ_SYSTEM_RESET 0xff
383 static const u8 srr1_to_lazyirq[0x10] = {
395 void replay_system_reset(void)
399 ppc_save_regs(®s);
401 get_paca()->in_nmi = 1;
402 system_reset_exception(®s);
403 get_paca()->in_nmi = 0;
405 EXPORT_SYMBOL_GPL(replay_system_reset);
407 void irq_set_pending_from_srr1(unsigned long srr1)
409 unsigned int idx = (srr1 & SRR1_WAKEMASK_P8) >> 18;
410 u8 reason = srr1_to_lazyirq[idx];
413 * Take the system reset now, which is immediately after registers
414 * are restored from idle. It's an NMI, so interrupts need not be
415 * re-enabled before it is taken.
417 if (unlikely(reason == IRQ_SYSTEM_RESET)) {
418 replay_system_reset();
422 if (reason == PACA_IRQ_DBELL) {
424 * When doorbell triggers a system reset wakeup, the message
425 * is not cleared, so if the doorbell interrupt is replayed
426 * and the IPI handled, the doorbell interrupt would still
427 * fire when EE is enabled.
429 * To avoid taking the superfluous doorbell interrupt,
430 * execute a msgclr here before the interrupt is replayed.
432 ppc_msgclr(PPC_DBELL_MSGTYPE);
436 * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
437 * so this can be called unconditionally with the SRR1 wake
438 * reason as returned by the idle code, which uses 0 to mean no
441 * If a future CPU was to designate this as an interrupt reason,
442 * then a new index for no interrupt must be assigned.
444 local_paca->irq_happened |= reason;
446 #endif /* CONFIG_PPC_BOOK3S */
449 * Force a replay of the external interrupt handler on this CPU.
451 void force_external_irq_replay(void)
454 * This must only be called with interrupts soft-disabled,
455 * the replay will happen when re-enabling.
457 WARN_ON(!arch_irqs_disabled());
460 * Interrupts must always be hard disabled before irq_happened is
461 * modified (to prevent lost update in case of interrupt between
464 __hard_irq_disable();
465 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
467 /* Indicate in the PACA that we have an interrupt to replay */
468 local_paca->irq_happened |= PACA_IRQ_EE;
471 #endif /* CONFIG_PPC64 */
473 int arch_show_interrupts(struct seq_file *p, int prec)
477 #if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
478 if (tau_initialized) {
479 seq_printf(p, "%*s: ", prec, "TAU");
480 for_each_online_cpu(j)
481 seq_printf(p, "%10u ", tau_interrupts(j));
482 seq_puts(p, " PowerPC Thermal Assist (cpu temp)\n");
484 #endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
486 seq_printf(p, "%*s: ", prec, "LOC");
487 for_each_online_cpu(j)
488 seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_event);
489 seq_printf(p, " Local timer interrupts for timer event device\n");
491 seq_printf(p, "%*s: ", prec, "BCT");
492 for_each_online_cpu(j)
493 seq_printf(p, "%10u ", per_cpu(irq_stat, j).broadcast_irqs_event);
494 seq_printf(p, " Broadcast timer interrupts for timer event device\n");
496 seq_printf(p, "%*s: ", prec, "LOC");
497 for_each_online_cpu(j)
498 seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_others);
499 seq_printf(p, " Local timer interrupts for others\n");
501 seq_printf(p, "%*s: ", prec, "SPU");
502 for_each_online_cpu(j)
503 seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
504 seq_printf(p, " Spurious interrupts\n");
506 seq_printf(p, "%*s: ", prec, "PMI");
507 for_each_online_cpu(j)
508 seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
509 seq_printf(p, " Performance monitoring interrupts\n");
511 seq_printf(p, "%*s: ", prec, "MCE");
512 for_each_online_cpu(j)
513 seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
514 seq_printf(p, " Machine check exceptions\n");
516 #ifdef CONFIG_PPC_BOOK3S_64
517 if (cpu_has_feature(CPU_FTR_HVMODE)) {
518 seq_printf(p, "%*s: ", prec, "HMI");
519 for_each_online_cpu(j)
520 seq_printf(p, "%10u ", paca_ptrs[j]->hmi_irqs);
521 seq_printf(p, " Hypervisor Maintenance Interrupts\n");
525 seq_printf(p, "%*s: ", prec, "NMI");
526 for_each_online_cpu(j)
527 seq_printf(p, "%10u ", per_cpu(irq_stat, j).sreset_irqs);
528 seq_printf(p, " System Reset interrupts\n");
530 #ifdef CONFIG_PPC_WATCHDOG
531 seq_printf(p, "%*s: ", prec, "WDG");
532 for_each_online_cpu(j)
533 seq_printf(p, "%10u ", per_cpu(irq_stat, j).soft_nmi_irqs);
534 seq_printf(p, " Watchdog soft-NMI interrupts\n");
537 #ifdef CONFIG_PPC_DOORBELL
538 if (cpu_has_feature(CPU_FTR_DBELL)) {
539 seq_printf(p, "%*s: ", prec, "DBL");
540 for_each_online_cpu(j)
541 seq_printf(p, "%10u ", per_cpu(irq_stat, j).doorbell_irqs);
542 seq_printf(p, " Doorbell interrupts\n");
552 u64 arch_irq_stat_cpu(unsigned int cpu)
554 u64 sum = per_cpu(irq_stat, cpu).timer_irqs_event;
556 sum += per_cpu(irq_stat, cpu).broadcast_irqs_event;
557 sum += per_cpu(irq_stat, cpu).pmu_irqs;
558 sum += per_cpu(irq_stat, cpu).mce_exceptions;
559 sum += per_cpu(irq_stat, cpu).spurious_irqs;
560 sum += per_cpu(irq_stat, cpu).timer_irqs_others;
561 #ifdef CONFIG_PPC_BOOK3S_64
562 sum += paca_ptrs[cpu]->hmi_irqs;
564 sum += per_cpu(irq_stat, cpu).sreset_irqs;
565 #ifdef CONFIG_PPC_WATCHDOG
566 sum += per_cpu(irq_stat, cpu).soft_nmi_irqs;
568 #ifdef CONFIG_PPC_DOORBELL
569 sum += per_cpu(irq_stat, cpu).doorbell_irqs;
575 static inline void check_stack_overflow(void)
579 if (!IS_ENABLED(CONFIG_DEBUG_STACKOVERFLOW))
582 sp = current_stack_pointer & (THREAD_SIZE - 1);
584 /* check for stack overflow: is there less than 2KB free? */
585 if (unlikely(sp < 2048)) {
586 pr_err("do_IRQ: stack overflow: %ld\n", sp);
591 static __always_inline void call_do_softirq(const void *sp)
593 /* Temporarily switch r1 to sp, call __do_softirq() then restore r1. */
595 PPC_STLU " %%r1, %[offset](%[sp]) ;"
598 PPC_LL " %%r1, 0(%%r1) ;"
601 [sp] "b" (sp), [offset] "i" (THREAD_SIZE - STACK_FRAME_OVERHEAD),
602 [callee] "i" (__do_softirq)
604 "lr", "xer", "ctr", "memory", "cr0", "cr1", "cr5", "cr6",
605 "cr7", "r0", "r3", "r4", "r5", "r6", "r7", "r8", "r9", "r10",
610 static __always_inline void call_do_irq(struct pt_regs *regs, void *sp)
612 register unsigned long r3 asm("r3") = (unsigned long)regs;
614 /* Temporarily switch r1 to sp, call __do_irq() then restore r1. */
616 PPC_STLU " %%r1, %[offset](%[sp]) ;"
619 PPC_LL " %%r1, 0(%%r1) ;"
623 [sp] "b" (sp), [offset] "i" (THREAD_SIZE - STACK_FRAME_OVERHEAD),
624 [callee] "i" (__do_irq)
626 "lr", "xer", "ctr", "memory", "cr0", "cr1", "cr5", "cr6",
627 "cr7", "r0", "r4", "r5", "r6", "r7", "r8", "r9", "r10",
632 void __do_irq(struct pt_regs *regs)
636 trace_irq_entry(regs);
639 * Query the platform PIC for the interrupt & ack it.
641 * This will typically lower the interrupt line to the CPU
643 irq = ppc_md.get_irq();
645 /* We can hard enable interrupts now to allow perf interrupts */
646 may_hard_irq_enable();
648 /* And finally process it */
650 __this_cpu_inc(irq_stat.spurious_irqs);
652 generic_handle_irq(irq);
654 trace_irq_exit(regs);
657 DEFINE_INTERRUPT_HANDLER_ASYNC(do_IRQ)
659 struct pt_regs *old_regs = set_irq_regs(regs);
660 void *cursp, *irqsp, *sirqsp;
662 /* Switch to the irq stack to handle this */
663 cursp = (void *)(current_stack_pointer & ~(THREAD_SIZE - 1));
664 irqsp = hardirq_ctx[raw_smp_processor_id()];
665 sirqsp = softirq_ctx[raw_smp_processor_id()];
667 check_stack_overflow();
669 /* Already there ? */
670 if (unlikely(cursp == irqsp || cursp == sirqsp)) {
672 set_irq_regs(old_regs);
675 /* Switch stack and call */
676 call_do_irq(regs, irqsp);
678 set_irq_regs(old_regs);
681 static void *__init alloc_vm_stack(void)
683 return __vmalloc_node(THREAD_SIZE, THREAD_ALIGN, THREADINFO_GFP,
684 NUMA_NO_NODE, (void *)_RET_IP_);
687 static void __init vmap_irqstack_init(void)
691 for_each_possible_cpu(i) {
692 softirq_ctx[i] = alloc_vm_stack();
693 hardirq_ctx[i] = alloc_vm_stack();
698 void __init init_IRQ(void)
700 if (IS_ENABLED(CONFIG_VMAP_STACK))
701 vmap_irqstack_init();
707 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
708 void *critirq_ctx[NR_CPUS] __read_mostly;
709 void *dbgirq_ctx[NR_CPUS] __read_mostly;
710 void *mcheckirq_ctx[NR_CPUS] __read_mostly;
713 void *softirq_ctx[NR_CPUS] __read_mostly;
714 void *hardirq_ctx[NR_CPUS] __read_mostly;
716 void do_softirq_own_stack(void)
718 call_do_softirq(softirq_ctx[smp_processor_id()]);
721 irq_hw_number_t virq_to_hw(unsigned int virq)
723 struct irq_data *irq_data = irq_get_irq_data(virq);
724 return WARN_ON(!irq_data) ? 0 : irq_data->hwirq;
726 EXPORT_SYMBOL_GPL(virq_to_hw);
729 int irq_choose_cpu(const struct cpumask *mask)
733 if (cpumask_equal(mask, cpu_online_mask)) {
734 static int irq_rover;
735 static DEFINE_RAW_SPINLOCK(irq_rover_lock);
738 /* Round-robin distribution... */
740 raw_spin_lock_irqsave(&irq_rover_lock, flags);
742 irq_rover = cpumask_next(irq_rover, cpu_online_mask);
743 if (irq_rover >= nr_cpu_ids)
744 irq_rover = cpumask_first(cpu_online_mask);
748 raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
750 cpuid = cpumask_first_and(mask, cpu_online_mask);
751 if (cpuid >= nr_cpu_ids)
755 return get_hard_smp_processor_id(cpuid);
758 int irq_choose_cpu(const struct cpumask *mask)
760 return hard_smp_processor_id();
765 static int __init setup_noirqdistrib(char *str)
771 __setup("noirqdistrib", setup_noirqdistrib);
772 #endif /* CONFIG_PPC64 */