2 * Derived from arch/i386/kernel/irq.c
3 * Copyright (C) 1992 Linus Torvalds
4 * Adapted from arch/i386 by Gary Thomas
5 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
6 * Updated and modified by Cort Dougan <cort@fsmlabs.com>
7 * Copyright (C) 1996-2001 Cort Dougan
8 * Adapted for Power Macintosh by Paul Mackerras
9 * Copyright (C) 1996 Paul Mackerras (paulus@cs.anu.edu.au)
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version
14 * 2 of the License, or (at your option) any later version.
16 * This file contains the code used by various IRQ handling routines:
17 * asking for different IRQ's should be done through these routines
18 * instead of just grabbing them. Thus setups with different IRQ numbers
19 * shouldn't result in any weird surprises, and installing new handlers
22 * The MPC8xx has an interrupt mask in the SIU. If a bit is set, the
23 * interrupt is _enabled_. As expected, IRQ0 is bit 0 in the 32-bit
24 * mask register (of which only 16 are defined), hence the weird shifting
25 * and complement of the cached_irq_mask. I want to be able to stuff
26 * this right into the SIU SMASK register.
27 * Many of the prep/chrp functions are conditional compiled on CONFIG_8xx
28 * to reduce code space and undefined function references.
33 #include <linux/export.h>
34 #include <linux/threads.h>
35 #include <linux/kernel_stat.h>
36 #include <linux/signal.h>
37 #include <linux/sched.h>
38 #include <linux/ptrace.h>
39 #include <linux/ioport.h>
40 #include <linux/interrupt.h>
41 #include <linux/timex.h>
42 #include <linux/init.h>
43 #include <linux/slab.h>
44 #include <linux/delay.h>
45 #include <linux/irq.h>
46 #include <linux/seq_file.h>
47 #include <linux/cpumask.h>
48 #include <linux/profile.h>
49 #include <linux/bitops.h>
50 #include <linux/list.h>
51 #include <linux/radix-tree.h>
52 #include <linux/mutex.h>
53 #include <linux/bootmem.h>
54 #include <linux/pci.h>
55 #include <linux/debugfs.h>
57 #include <linux/of_irq.h>
59 #include <asm/uaccess.h>
61 #include <asm/pgtable.h>
63 #include <asm/cache.h>
65 #include <asm/ptrace.h>
66 #include <asm/machdep.h>
69 #include <asm/debug.h>
73 #include <asm/firmware.h>
74 #include <asm/lv1call.h>
76 #define CREATE_TRACE_POINTS
77 #include <asm/trace.h>
79 DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
80 EXPORT_PER_CPU_SYMBOL(irq_stat);
82 int __irq_offset_value;
85 EXPORT_SYMBOL(__irq_offset_value);
86 atomic_t ppc_n_lost_interrupts;
89 extern int tau_initialized;
90 extern int tau_interrupts(int);
92 #endif /* CONFIG_PPC32 */
96 int distribute_irqs = 1;
98 static inline notrace unsigned long get_irq_happened(void)
100 unsigned long happened;
102 __asm__ __volatile__("lbz %0,%1(13)"
103 : "=r" (happened) : "i" (offsetof(struct paca_struct, irq_happened)));
108 static inline notrace void set_soft_enabled(unsigned long enable)
110 __asm__ __volatile__("stb %0,%1(13)"
111 : : "r" (enable), "i" (offsetof(struct paca_struct, soft_enabled)));
114 static inline notrace int decrementer_check_overflow(void)
116 u64 now = get_tb_or_rtc();
117 u64 *next_tb = &__get_cpu_var(decrementers_next_tb);
121 return now >= *next_tb;
124 /* This is called whenever we are re-enabling interrupts
125 * and returns either 0 (nothing to do) or 500/900/280/a00/e80 if
126 * there's an EE, DEC or DBELL to generate.
128 * This is called in two contexts: From arch_local_irq_restore()
129 * before soft-enabling interrupts, and from the exception exit
130 * path when returning from an interrupt from a soft-disabled to
131 * a soft enabled context. In both case we have interrupts hard
134 * We take care of only clearing the bits we handled in the
135 * PACA irq_happened field since we can only re-emit one at a
136 * time and we don't want to "lose" one.
138 notrace unsigned int __check_irq_replay(void)
141 * We use local_paca rather than get_paca() to avoid all
142 * the debug_smp_processor_id() business in this low level
145 unsigned char happened = local_paca->irq_happened;
147 /* Clear bit 0 which we wouldn't clear otherwise */
148 local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
151 * Force the delivery of pending soft-disabled interrupts on PS3.
152 * Any HV call will have this side effect.
154 if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
156 lv1_get_version_info(&tmp, &tmp2);
160 * We may have missed a decrementer interrupt. We check the
161 * decrementer itself rather than the paca irq_happened field
162 * in case we also had a rollover while hard disabled
164 local_paca->irq_happened &= ~PACA_IRQ_DEC;
165 if ((happened & PACA_IRQ_DEC) || decrementer_check_overflow())
168 /* Finally check if an external interrupt happened */
169 local_paca->irq_happened &= ~PACA_IRQ_EE;
170 if (happened & PACA_IRQ_EE)
173 #ifdef CONFIG_PPC_BOOK3E
174 /* Finally check if an EPR external interrupt happened
175 * this bit is typically set if we need to handle another
176 * "edge" interrupt from within the MPIC "EPR" handler
178 local_paca->irq_happened &= ~PACA_IRQ_EE_EDGE;
179 if (happened & PACA_IRQ_EE_EDGE)
182 local_paca->irq_happened &= ~PACA_IRQ_DBELL;
183 if (happened & PACA_IRQ_DBELL)
186 local_paca->irq_happened &= ~PACA_IRQ_DBELL;
187 if (happened & PACA_IRQ_DBELL) {
188 if (cpu_has_feature(CPU_FTR_HVMODE))
192 #endif /* CONFIG_PPC_BOOK3E */
194 /* There should be nothing left ! */
195 BUG_ON(local_paca->irq_happened != 0);
200 notrace void arch_local_irq_restore(unsigned long en)
202 unsigned char irq_happened;
205 /* Write the new soft-enabled value */
206 set_soft_enabled(en);
210 * From this point onward, we can take interrupts, preempt,
211 * etc... unless we got hard-disabled. We check if an event
212 * happened. If none happened, we know we can just return.
214 * We may have preempted before the check below, in which case
215 * we are checking the "new" CPU instead of the old one. This
216 * is only a problem if an event happened on the "old" CPU.
218 * External interrupt events will have caused interrupts to
219 * be hard-disabled, so there is no problem, we
220 * cannot have preempted.
222 irq_happened = get_irq_happened();
227 * We need to hard disable to get a trusted value from
228 * __check_irq_replay(). We also need to soft-disable
229 * again to avoid warnings in there due to the use of
232 * We know that if the value in irq_happened is exactly 0x01
233 * then we are already hard disabled (there are other less
234 * common cases that we'll ignore for now), so we skip the
235 * (expensive) mtmsrd.
237 if (unlikely(irq_happened != PACA_IRQ_HARD_DIS))
238 __hard_irq_disable();
239 #ifdef CONFIG_TRACE_IRQFLAGS
242 * We should already be hard disabled here. We had bugs
243 * where that wasn't the case so let's dbl check it and
244 * warn if we are wrong. Only do that when IRQ tracing
245 * is enabled as mfmsr() can be costly.
247 if (WARN_ON(mfmsr() & MSR_EE))
248 __hard_irq_disable();
250 #endif /* CONFIG_TRACE_IRQFLAG */
255 * Check if anything needs to be re-emitted. We haven't
256 * soft-enabled yet to avoid warnings in decrementer_check_overflow
257 * accessing per-cpu variables
259 replay = __check_irq_replay();
261 /* We can soft-enable now */
265 * And replay if we have to. This will return with interrupts
269 __replay_interrupt(replay);
273 /* Finally, let's ensure we are hard enabled */
276 EXPORT_SYMBOL(arch_local_irq_restore);
279 * This is specifically called by assembly code to re-enable interrupts
280 * if they are currently disabled. This is typically called before
281 * schedule() or do_signal() when returning to userspace. We do it
282 * in C to avoid the burden of dealing with lockdep etc...
284 * NOTE: This is called with interrupts hard disabled but not marked
285 * as such in paca->irq_happened, so we need to resync this.
287 void notrace restore_interrupts(void)
289 if (irqs_disabled()) {
290 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
297 * This is a helper to use when about to go into idle low-power
298 * when the latter has the side effect of re-enabling interrupts
299 * (such as calling H_CEDE under pHyp).
301 * You call this function with interrupts soft-disabled (this is
302 * already the case when ppc_md.power_save is called). The function
303 * will return whether to enter power save or just return.
305 * In the former case, it will have notified lockdep of interrupts
306 * being re-enabled and generally sanitized the lazy irq state,
307 * and in the latter case it will leave with interrupts hard
308 * disabled and marked as such, so the local_irq_enable() call
309 * in cpu_idle() will properly re-enable everything.
311 bool prep_irq_for_idle(void)
314 * First we need to hard disable to ensure no interrupt
315 * occurs before we effectively enter the low power state
320 * If anything happened while we were soft-disabled,
321 * we return now and do not enter the low power state.
323 if (lazy_irq_pending())
326 /* Tell lockdep we are about to re-enable */
330 * Mark interrupts as soft-enabled and clear the
331 * PACA_IRQ_HARD_DIS from the pending mask since we
332 * are about to hard enable as well as a side effect
333 * of entering the low power state.
335 local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
336 local_paca->soft_enabled = 1;
338 /* Tell the caller to enter the low power state */
342 #endif /* CONFIG_PPC64 */
344 int arch_show_interrupts(struct seq_file *p, int prec)
348 #if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
349 if (tau_initialized) {
350 seq_printf(p, "%*s: ", prec, "TAU");
351 for_each_online_cpu(j)
352 seq_printf(p, "%10u ", tau_interrupts(j));
353 seq_puts(p, " PowerPC Thermal Assist (cpu temp)\n");
355 #endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
357 seq_printf(p, "%*s: ", prec, "LOC");
358 for_each_online_cpu(j)
359 seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs);
360 seq_printf(p, " Local timer interrupts\n");
362 seq_printf(p, "%*s: ", prec, "SPU");
363 for_each_online_cpu(j)
364 seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
365 seq_printf(p, " Spurious interrupts\n");
367 seq_printf(p, "%*s: ", prec, "CNT");
368 for_each_online_cpu(j)
369 seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
370 seq_printf(p, " Performance monitoring interrupts\n");
372 seq_printf(p, "%*s: ", prec, "MCE");
373 for_each_online_cpu(j)
374 seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
375 seq_printf(p, " Machine check exceptions\n");
377 #ifdef CONFIG_PPC_DOORBELL
378 if (cpu_has_feature(CPU_FTR_DBELL)) {
379 seq_printf(p, "%*s: ", prec, "DBL");
380 for_each_online_cpu(j)
381 seq_printf(p, "%10u ", per_cpu(irq_stat, j).doorbell_irqs);
382 seq_printf(p, " Doorbell interrupts\n");
392 u64 arch_irq_stat_cpu(unsigned int cpu)
394 u64 sum = per_cpu(irq_stat, cpu).timer_irqs;
396 sum += per_cpu(irq_stat, cpu).pmu_irqs;
397 sum += per_cpu(irq_stat, cpu).mce_exceptions;
398 sum += per_cpu(irq_stat, cpu).spurious_irqs;
399 #ifdef CONFIG_PPC_DOORBELL
400 sum += per_cpu(irq_stat, cpu).doorbell_irqs;
406 #ifdef CONFIG_HOTPLUG_CPU
407 void migrate_irqs(void)
409 struct irq_desc *desc;
413 const struct cpumask *map = cpu_online_mask;
415 alloc_cpumask_var(&mask, GFP_KERNEL);
417 for_each_irq_desc(irq, desc) {
418 struct irq_data *data;
419 struct irq_chip *chip;
421 data = irq_desc_get_irq_data(desc);
422 if (irqd_is_per_cpu(data))
425 chip = irq_data_get_irq_chip(data);
427 cpumask_and(mask, data->affinity, map);
428 if (cpumask_any(mask) >= nr_cpu_ids) {
429 printk("Breaking affinity for irq %i\n", irq);
430 cpumask_copy(mask, map);
432 if (chip->irq_set_affinity)
433 chip->irq_set_affinity(data, mask, true);
434 else if (desc->action && !(warned++))
435 printk("Cannot set affinity for irq %i\n", irq);
438 free_cpumask_var(mask);
446 static inline void handle_one_irq(unsigned int irq)
448 struct thread_info *curtp, *irqtp;
449 unsigned long saved_sp_limit;
450 struct irq_desc *desc;
452 desc = irq_to_desc(irq);
456 /* Switch to the irq stack to handle this */
457 curtp = current_thread_info();
458 irqtp = hardirq_ctx[smp_processor_id()];
460 if (curtp == irqtp) {
461 /* We're already on the irq stack, just handle it */
462 desc->handle_irq(irq, desc);
466 saved_sp_limit = current->thread.ksp_limit;
468 irqtp->task = curtp->task;
471 /* Copy the softirq bits in preempt_count so that the
472 * softirq checks work in the hardirq context. */
473 irqtp->preempt_count = (irqtp->preempt_count & ~SOFTIRQ_MASK) |
474 (curtp->preempt_count & SOFTIRQ_MASK);
476 current->thread.ksp_limit = (unsigned long)irqtp +
477 _ALIGN_UP(sizeof(struct thread_info), 16);
479 call_handle_irq(irq, desc, irqtp, desc->handle_irq);
480 current->thread.ksp_limit = saved_sp_limit;
483 /* Set any flag that may have been set on the
487 set_bits(irqtp->flags, &curtp->flags);
490 static inline void check_stack_overflow(void)
492 #ifdef CONFIG_DEBUG_STACKOVERFLOW
495 sp = __get_SP() & (THREAD_SIZE-1);
497 /* check for stack overflow: is there less than 2KB free? */
498 if (unlikely(sp < (sizeof(struct thread_info) + 2048))) {
499 printk("do_IRQ: stack overflow: %ld\n",
500 sp - sizeof(struct thread_info));
506 void do_IRQ(struct pt_regs *regs)
508 struct pt_regs *old_regs = set_irq_regs(regs);
513 trace_irq_entry(regs);
515 check_stack_overflow();
518 * Query the platform PIC for the interrupt & ack it.
520 * This will typically lower the interrupt line to the CPU
522 irq = ppc_md.get_irq();
524 /* We can hard enable interrupts now */
525 may_hard_irq_enable();
527 /* And finally process it */
531 __get_cpu_var(irq_stat).spurious_irqs++;
533 trace_irq_exit(regs);
536 set_irq_regs(old_regs);
539 void __init init_IRQ(void)
549 #if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
550 struct thread_info *critirq_ctx[NR_CPUS] __read_mostly;
551 struct thread_info *dbgirq_ctx[NR_CPUS] __read_mostly;
552 struct thread_info *mcheckirq_ctx[NR_CPUS] __read_mostly;
554 void exc_lvl_ctx_init(void)
556 struct thread_info *tp;
559 for_each_possible_cpu(i) {
563 cpu_nr = get_hard_smp_processor_id(i);
565 memset((void *)critirq_ctx[cpu_nr], 0, THREAD_SIZE);
566 tp = critirq_ctx[cpu_nr];
568 tp->preempt_count = 0;
571 memset((void *)dbgirq_ctx[cpu_nr], 0, THREAD_SIZE);
572 tp = dbgirq_ctx[cpu_nr];
574 tp->preempt_count = 0;
576 memset((void *)mcheckirq_ctx[cpu_nr], 0, THREAD_SIZE);
577 tp = mcheckirq_ctx[cpu_nr];
579 tp->preempt_count = HARDIRQ_OFFSET;
585 struct thread_info *softirq_ctx[NR_CPUS] __read_mostly;
586 struct thread_info *hardirq_ctx[NR_CPUS] __read_mostly;
588 void irq_ctx_init(void)
590 struct thread_info *tp;
593 for_each_possible_cpu(i) {
594 memset((void *)softirq_ctx[i], 0, THREAD_SIZE);
597 tp->preempt_count = 0;
599 memset((void *)hardirq_ctx[i], 0, THREAD_SIZE);
602 tp->preempt_count = HARDIRQ_OFFSET;
606 static inline void do_softirq_onstack(void)
608 struct thread_info *curtp, *irqtp;
609 unsigned long saved_sp_limit = current->thread.ksp_limit;
611 curtp = current_thread_info();
612 irqtp = softirq_ctx[smp_processor_id()];
613 irqtp->task = curtp->task;
615 current->thread.ksp_limit = (unsigned long)irqtp +
616 _ALIGN_UP(sizeof(struct thread_info), 16);
617 call_do_softirq(irqtp);
618 current->thread.ksp_limit = saved_sp_limit;
621 /* Set any flag that may have been set on the
625 set_bits(irqtp->flags, &curtp->flags);
628 void do_softirq(void)
635 local_irq_save(flags);
637 if (local_softirq_pending())
638 do_softirq_onstack();
640 local_irq_restore(flags);
643 irq_hw_number_t virq_to_hw(unsigned int virq)
645 struct irq_data *irq_data = irq_get_irq_data(virq);
646 return WARN_ON(!irq_data) ? 0 : irq_data->hwirq;
648 EXPORT_SYMBOL_GPL(virq_to_hw);
651 int irq_choose_cpu(const struct cpumask *mask)
655 if (cpumask_equal(mask, cpu_online_mask)) {
656 static int irq_rover;
657 static DEFINE_RAW_SPINLOCK(irq_rover_lock);
660 /* Round-robin distribution... */
662 raw_spin_lock_irqsave(&irq_rover_lock, flags);
664 irq_rover = cpumask_next(irq_rover, cpu_online_mask);
665 if (irq_rover >= nr_cpu_ids)
666 irq_rover = cpumask_first(cpu_online_mask);
670 raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
672 cpuid = cpumask_first_and(mask, cpu_online_mask);
673 if (cpuid >= nr_cpu_ids)
677 return get_hard_smp_processor_id(cpuid);
680 int irq_choose_cpu(const struct cpumask *mask)
682 return hard_smp_processor_id();
686 int arch_early_irq_init(void)
692 static int __init setup_noirqdistrib(char *str)
698 __setup("noirqdistrib", setup_noirqdistrib);
699 #endif /* CONFIG_PPC64 */