Loading...
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/interrupt.h>
58#include <asm/io.h>
59#include <asm/irq.h>
60#include <asm/cache.h>
61#include <asm/prom.h>
62#include <asm/ptrace.h>
63#include <asm/machdep.h>
64#include <asm/udbg.h>
65#include <asm/smp.h>
66#include <asm/livepatch.h>
67#include <asm/asm-prototypes.h>
68#include <asm/hw_irq.h>
69#include <asm/softirq_stack.h>
70
71#ifdef CONFIG_PPC64
72#include <asm/paca.h>
73#include <asm/firmware.h>
74#include <asm/lv1call.h>
75#include <asm/dbell.h>
76#endif
77#define CREATE_TRACE_POINTS
78#include <asm/trace.h>
79#include <asm/cpu_has_feature.h>
80
81DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
82EXPORT_PER_CPU_SYMBOL(irq_stat);
83
84#ifdef CONFIG_PPC32
85atomic_t ppc_n_lost_interrupts;
86
87#ifdef CONFIG_TAU_INT
88extern int tau_initialized;
89u32 tau_interrupts(unsigned long cpu);
90#endif
91#endif /* CONFIG_PPC32 */
92
93#ifdef CONFIG_PPC64
94
95int distribute_irqs = 1;
96
97static inline notrace unsigned long get_irq_happened(void)
98{
99 unsigned long happened;
100
101 __asm__ __volatile__("lbz %0,%1(13)"
102 : "=r" (happened) : "i" (offsetof(struct paca_struct, irq_happened)));
103
104 return happened;
105}
106
107void replay_soft_interrupts(void)
108{
109 struct pt_regs regs;
110
111 /*
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.
117 *
118 * We use local_paca rather than get_paca() to avoid all the
119 * debug_smp_processor_id() business in this low level function.
120 */
121
122 ppc_save_regs(®s);
123 regs.softe = IRQS_ENABLED;
124 regs.msr |= MSR_EE;
125
126again:
127 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
128 WARN_ON_ONCE(mfmsr() & MSR_EE);
129
130 /*
131 * Force the delivery of pending soft-disabled interrupts on PS3.
132 * Any HV call will have this side effect.
133 */
134 if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
135 u64 tmp, tmp2;
136 lv1_get_version_info(&tmp, &tmp2);
137 }
138
139 /*
140 * Check if an hypervisor Maintenance interrupt happened.
141 * This is a higher priority interrupt than the others, so
142 * replay it first.
143 */
144 if (IS_ENABLED(CONFIG_PPC_BOOK3S) && (local_paca->irq_happened & PACA_IRQ_HMI)) {
145 local_paca->irq_happened &= ~PACA_IRQ_HMI;
146 regs.trap = INTERRUPT_HMI;
147 handle_hmi_exception(®s);
148 if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
149 hard_irq_disable();
150 }
151
152 if (local_paca->irq_happened & PACA_IRQ_DEC) {
153 local_paca->irq_happened &= ~PACA_IRQ_DEC;
154 regs.trap = INTERRUPT_DECREMENTER;
155 timer_interrupt(®s);
156 if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
157 hard_irq_disable();
158 }
159
160 if (local_paca->irq_happened & PACA_IRQ_EE) {
161 local_paca->irq_happened &= ~PACA_IRQ_EE;
162 regs.trap = INTERRUPT_EXTERNAL;
163 do_IRQ(®s);
164 if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
165 hard_irq_disable();
166 }
167
168 if (IS_ENABLED(CONFIG_PPC_DOORBELL) && (local_paca->irq_happened & PACA_IRQ_DBELL)) {
169 local_paca->irq_happened &= ~PACA_IRQ_DBELL;
170 regs.trap = INTERRUPT_DOORBELL;
171 doorbell_exception(®s);
172 if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
173 hard_irq_disable();
174 }
175
176 /* Book3E does not support soft-masking PMI interrupts */
177 if (IS_ENABLED(CONFIG_PPC_BOOK3S) && (local_paca->irq_happened & PACA_IRQ_PMI)) {
178 local_paca->irq_happened &= ~PACA_IRQ_PMI;
179 regs.trap = INTERRUPT_PERFMON;
180 performance_monitor_exception(®s);
181 if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
182 hard_irq_disable();
183 }
184
185 if (local_paca->irq_happened & ~PACA_IRQ_HARD_DIS) {
186 /*
187 * We are responding to the next interrupt, so interrupt-off
188 * latencies should be reset here.
189 */
190 trace_hardirqs_on();
191 trace_hardirqs_off();
192 goto again;
193 }
194}
195
196#if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_KUAP)
197static inline void replay_soft_interrupts_irqrestore(void)
198{
199 unsigned long kuap_state = get_kuap();
200
201 /*
202 * Check if anything calls local_irq_enable/restore() when KUAP is
203 * disabled (user access enabled). We handle that case here by saving
204 * and re-locking AMR but we shouldn't get here in the first place,
205 * hence the warning.
206 */
207 kuap_assert_locked();
208
209 if (kuap_state != AMR_KUAP_BLOCKED)
210 set_kuap(AMR_KUAP_BLOCKED);
211
212 replay_soft_interrupts();
213
214 if (kuap_state != AMR_KUAP_BLOCKED)
215 set_kuap(kuap_state);
216}
217#else
218#define replay_soft_interrupts_irqrestore() replay_soft_interrupts()
219#endif
220
221#ifdef CONFIG_CC_HAS_ASM_GOTO
222notrace void arch_local_irq_restore(unsigned long mask)
223{
224 unsigned char irq_happened;
225
226 /* Write the new soft-enabled value if it is a disable */
227 if (mask) {
228 irq_soft_mask_set(mask);
229 return;
230 }
231
232 /*
233 * After the stb, interrupts are unmasked and there are no interrupts
234 * pending replay. The restart sequence makes this atomic with
235 * respect to soft-masked interrupts. If this was just a simple code
236 * sequence, a soft-masked interrupt could become pending right after
237 * the comparison and before the stb.
238 *
239 * This allows interrupts to be unmasked without hard disabling, and
240 * also without new hard interrupts coming in ahead of pending ones.
241 */
242 asm_volatile_goto(
243"1: \n"
244" lbz 9,%0(13) \n"
245" cmpwi 9,0 \n"
246" bne %l[happened] \n"
247" stb 9,%1(13) \n"
248"2: \n"
249 RESTART_TABLE(1b, 2b, 1b)
250 : : "i" (offsetof(struct paca_struct, irq_happened)),
251 "i" (offsetof(struct paca_struct, irq_soft_mask))
252 : "cr0", "r9"
253 : happened);
254
255 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
256 WARN_ON_ONCE(!(mfmsr() & MSR_EE));
257
258 return;
259
260happened:
261 irq_happened = get_irq_happened();
262 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
263 WARN_ON_ONCE(!irq_happened);
264
265 if (irq_happened == PACA_IRQ_HARD_DIS) {
266 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
267 WARN_ON_ONCE(mfmsr() & MSR_EE);
268 irq_soft_mask_set(IRQS_ENABLED);
269 local_paca->irq_happened = 0;
270 __hard_irq_enable();
271 return;
272 }
273
274 /* Have interrupts to replay, need to hard disable first */
275 if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
276 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
277 if (!(mfmsr() & MSR_EE)) {
278 /*
279 * An interrupt could have come in and cleared
280 * MSR[EE] and set IRQ_HARD_DIS, so check
281 * IRQ_HARD_DIS again and warn if it is still
282 * clear.
283 */
284 irq_happened = get_irq_happened();
285 WARN_ON_ONCE(!(irq_happened & PACA_IRQ_HARD_DIS));
286 }
287 }
288 __hard_irq_disable();
289 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
290 } else {
291 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
292 if (WARN_ON_ONCE(mfmsr() & MSR_EE))
293 __hard_irq_disable();
294 }
295 }
296
297 /*
298 * Disable preempt here, so that the below preempt_enable will
299 * perform resched if required (a replayed interrupt may set
300 * need_resched).
301 */
302 preempt_disable();
303 irq_soft_mask_set(IRQS_ALL_DISABLED);
304 trace_hardirqs_off();
305
306 replay_soft_interrupts_irqrestore();
307 local_paca->irq_happened = 0;
308
309 trace_hardirqs_on();
310 irq_soft_mask_set(IRQS_ENABLED);
311 __hard_irq_enable();
312 preempt_enable();
313}
314#else
315notrace void arch_local_irq_restore(unsigned long mask)
316{
317 unsigned char irq_happened;
318
319 /* Write the new soft-enabled value */
320 irq_soft_mask_set(mask);
321 if (mask)
322 return;
323
324 /*
325 * From this point onward, we can take interrupts, preempt,
326 * etc... unless we got hard-disabled. We check if an event
327 * happened. If none happened, we know we can just return.
328 *
329 * We may have preempted before the check below, in which case
330 * we are checking the "new" CPU instead of the old one. This
331 * is only a problem if an event happened on the "old" CPU.
332 *
333 * External interrupt events will have caused interrupts to
334 * be hard-disabled, so there is no problem, we
335 * cannot have preempted.
336 */
337 irq_happened = get_irq_happened();
338 if (!irq_happened) {
339 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
340 WARN_ON_ONCE(!(mfmsr() & MSR_EE));
341 return;
342 }
343
344 /* We need to hard disable to replay. */
345 if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
346 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
347 WARN_ON_ONCE(!(mfmsr() & MSR_EE));
348 __hard_irq_disable();
349 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
350 } else {
351 /*
352 * We should already be hard disabled here. We had bugs
353 * where that wasn't the case so let's dbl check it and
354 * warn if we are wrong. Only do that when IRQ tracing
355 * is enabled as mfmsr() can be costly.
356 */
357 if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
358 if (WARN_ON_ONCE(mfmsr() & MSR_EE))
359 __hard_irq_disable();
360 }
361
362 if (irq_happened == PACA_IRQ_HARD_DIS) {
363 local_paca->irq_happened = 0;
364 __hard_irq_enable();
365 return;
366 }
367 }
368
369 /*
370 * Disable preempt here, so that the below preempt_enable will
371 * perform resched if required (a replayed interrupt may set
372 * need_resched).
373 */
374 preempt_disable();
375 irq_soft_mask_set(IRQS_ALL_DISABLED);
376 trace_hardirqs_off();
377
378 replay_soft_interrupts_irqrestore();
379 local_paca->irq_happened = 0;
380
381 trace_hardirqs_on();
382 irq_soft_mask_set(IRQS_ENABLED);
383 __hard_irq_enable();
384 preempt_enable();
385}
386#endif
387EXPORT_SYMBOL(arch_local_irq_restore);
388
389/*
390 * This is a helper to use when about to go into idle low-power
391 * when the latter has the side effect of re-enabling interrupts
392 * (such as calling H_CEDE under pHyp).
393 *
394 * You call this function with interrupts soft-disabled (this is
395 * already the case when ppc_md.power_save is called). The function
396 * will return whether to enter power save or just return.
397 *
398 * In the former case, it will have notified lockdep of interrupts
399 * being re-enabled and generally sanitized the lazy irq state,
400 * and in the latter case it will leave with interrupts hard
401 * disabled and marked as such, so the local_irq_enable() call
402 * in arch_cpu_idle() will properly re-enable everything.
403 */
404bool prep_irq_for_idle(void)
405{
406 /*
407 * First we need to hard disable to ensure no interrupt
408 * occurs before we effectively enter the low power state
409 */
410 __hard_irq_disable();
411 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
412
413 /*
414 * If anything happened while we were soft-disabled,
415 * we return now and do not enter the low power state.
416 */
417 if (lazy_irq_pending())
418 return false;
419
420 /* Tell lockdep we are about to re-enable */
421 trace_hardirqs_on();
422
423 /*
424 * Mark interrupts as soft-enabled and clear the
425 * PACA_IRQ_HARD_DIS from the pending mask since we
426 * are about to hard enable as well as a side effect
427 * of entering the low power state.
428 */
429 local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
430 irq_soft_mask_set(IRQS_ENABLED);
431
432 /* Tell the caller to enter the low power state */
433 return true;
434}
435
436#ifdef CONFIG_PPC_BOOK3S
437/*
438 * This is for idle sequences that return with IRQs off, but the
439 * idle state itself wakes on interrupt. Tell the irq tracer that
440 * IRQs are enabled for the duration of idle so it does not get long
441 * off times. Must be paired with fini_irq_for_idle_irqsoff.
442 */
443bool prep_irq_for_idle_irqsoff(void)
444{
445 WARN_ON(!irqs_disabled());
446
447 /*
448 * First we need to hard disable to ensure no interrupt
449 * occurs before we effectively enter the low power state
450 */
451 __hard_irq_disable();
452 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
453
454 /*
455 * If anything happened while we were soft-disabled,
456 * we return now and do not enter the low power state.
457 */
458 if (lazy_irq_pending())
459 return false;
460
461 /* Tell lockdep we are about to re-enable */
462 trace_hardirqs_on();
463
464 return true;
465}
466
467/*
468 * Take the SRR1 wakeup reason, index into this table to find the
469 * appropriate irq_happened bit.
470 *
471 * Sytem reset exceptions taken in idle state also come through here,
472 * but they are NMI interrupts so do not need to wait for IRQs to be
473 * restored, and should be taken as early as practical. These are marked
474 * with 0xff in the table. The Power ISA specifies 0100b as the system
475 * reset interrupt reason.
476 */
477#define IRQ_SYSTEM_RESET 0xff
478
479static const u8 srr1_to_lazyirq[0x10] = {
480 0, 0, 0,
481 PACA_IRQ_DBELL,
482 IRQ_SYSTEM_RESET,
483 PACA_IRQ_DBELL,
484 PACA_IRQ_DEC,
485 0,
486 PACA_IRQ_EE,
487 PACA_IRQ_EE,
488 PACA_IRQ_HMI,
489 0, 0, 0, 0, 0 };
490
491void replay_system_reset(void)
492{
493 struct pt_regs regs;
494
495 ppc_save_regs(®s);
496 regs.trap = 0x100;
497 get_paca()->in_nmi = 1;
498 system_reset_exception(®s);
499 get_paca()->in_nmi = 0;
500}
501EXPORT_SYMBOL_GPL(replay_system_reset);
502
503void irq_set_pending_from_srr1(unsigned long srr1)
504{
505 unsigned int idx = (srr1 & SRR1_WAKEMASK_P8) >> 18;
506 u8 reason = srr1_to_lazyirq[idx];
507
508 /*
509 * Take the system reset now, which is immediately after registers
510 * are restored from idle. It's an NMI, so interrupts need not be
511 * re-enabled before it is taken.
512 */
513 if (unlikely(reason == IRQ_SYSTEM_RESET)) {
514 replay_system_reset();
515 return;
516 }
517
518 if (reason == PACA_IRQ_DBELL) {
519 /*
520 * When doorbell triggers a system reset wakeup, the message
521 * is not cleared, so if the doorbell interrupt is replayed
522 * and the IPI handled, the doorbell interrupt would still
523 * fire when EE is enabled.
524 *
525 * To avoid taking the superfluous doorbell interrupt,
526 * execute a msgclr here before the interrupt is replayed.
527 */
528 ppc_msgclr(PPC_DBELL_MSGTYPE);
529 }
530
531 /*
532 * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
533 * so this can be called unconditionally with the SRR1 wake
534 * reason as returned by the idle code, which uses 0 to mean no
535 * interrupt.
536 *
537 * If a future CPU was to designate this as an interrupt reason,
538 * then a new index for no interrupt must be assigned.
539 */
540 local_paca->irq_happened |= reason;
541}
542#endif /* CONFIG_PPC_BOOK3S */
543
544/*
545 * Force a replay of the external interrupt handler on this CPU.
546 */
547void force_external_irq_replay(void)
548{
549 /*
550 * This must only be called with interrupts soft-disabled,
551 * the replay will happen when re-enabling.
552 */
553 WARN_ON(!arch_irqs_disabled());
554
555 /*
556 * Interrupts must always be hard disabled before irq_happened is
557 * modified (to prevent lost update in case of interrupt between
558 * load and store).
559 */
560 __hard_irq_disable();
561 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
562
563 /* Indicate in the PACA that we have an interrupt to replay */
564 local_paca->irq_happened |= PACA_IRQ_EE;
565}
566
567#endif /* CONFIG_PPC64 */
568
569int arch_show_interrupts(struct seq_file *p, int prec)
570{
571 int j;
572
573#if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
574 if (tau_initialized) {
575 seq_printf(p, "%*s: ", prec, "TAU");
576 for_each_online_cpu(j)
577 seq_printf(p, "%10u ", tau_interrupts(j));
578 seq_puts(p, " PowerPC Thermal Assist (cpu temp)\n");
579 }
580#endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
581
582 seq_printf(p, "%*s: ", prec, "LOC");
583 for_each_online_cpu(j)
584 seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_event);
585 seq_printf(p, " Local timer interrupts for timer event device\n");
586
587 seq_printf(p, "%*s: ", prec, "BCT");
588 for_each_online_cpu(j)
589 seq_printf(p, "%10u ", per_cpu(irq_stat, j).broadcast_irqs_event);
590 seq_printf(p, " Broadcast timer interrupts for timer event device\n");
591
592 seq_printf(p, "%*s: ", prec, "LOC");
593 for_each_online_cpu(j)
594 seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_others);
595 seq_printf(p, " Local timer interrupts for others\n");
596
597 seq_printf(p, "%*s: ", prec, "SPU");
598 for_each_online_cpu(j)
599 seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
600 seq_printf(p, " Spurious interrupts\n");
601
602 seq_printf(p, "%*s: ", prec, "PMI");
603 for_each_online_cpu(j)
604 seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
605 seq_printf(p, " Performance monitoring interrupts\n");
606
607 seq_printf(p, "%*s: ", prec, "MCE");
608 for_each_online_cpu(j)
609 seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
610 seq_printf(p, " Machine check exceptions\n");
611
612#ifdef CONFIG_PPC_BOOK3S_64
613 if (cpu_has_feature(CPU_FTR_HVMODE)) {
614 seq_printf(p, "%*s: ", prec, "HMI");
615 for_each_online_cpu(j)
616 seq_printf(p, "%10u ", paca_ptrs[j]->hmi_irqs);
617 seq_printf(p, " Hypervisor Maintenance Interrupts\n");
618 }
619#endif
620
621 seq_printf(p, "%*s: ", prec, "NMI");
622 for_each_online_cpu(j)
623 seq_printf(p, "%10u ", per_cpu(irq_stat, j).sreset_irqs);
624 seq_printf(p, " System Reset interrupts\n");
625
626#ifdef CONFIG_PPC_WATCHDOG
627 seq_printf(p, "%*s: ", prec, "WDG");
628 for_each_online_cpu(j)
629 seq_printf(p, "%10u ", per_cpu(irq_stat, j).soft_nmi_irqs);
630 seq_printf(p, " Watchdog soft-NMI interrupts\n");
631#endif
632
633#ifdef CONFIG_PPC_DOORBELL
634 if (cpu_has_feature(CPU_FTR_DBELL)) {
635 seq_printf(p, "%*s: ", prec, "DBL");
636 for_each_online_cpu(j)
637 seq_printf(p, "%10u ", per_cpu(irq_stat, j).doorbell_irqs);
638 seq_printf(p, " Doorbell interrupts\n");
639 }
640#endif
641
642 return 0;
643}
644
645/*
646 * /proc/stat helpers
647 */
648u64 arch_irq_stat_cpu(unsigned int cpu)
649{
650 u64 sum = per_cpu(irq_stat, cpu).timer_irqs_event;
651
652 sum += per_cpu(irq_stat, cpu).broadcast_irqs_event;
653 sum += per_cpu(irq_stat, cpu).pmu_irqs;
654 sum += per_cpu(irq_stat, cpu).mce_exceptions;
655 sum += per_cpu(irq_stat, cpu).spurious_irqs;
656 sum += per_cpu(irq_stat, cpu).timer_irqs_others;
657#ifdef CONFIG_PPC_BOOK3S_64
658 sum += paca_ptrs[cpu]->hmi_irqs;
659#endif
660 sum += per_cpu(irq_stat, cpu).sreset_irqs;
661#ifdef CONFIG_PPC_WATCHDOG
662 sum += per_cpu(irq_stat, cpu).soft_nmi_irqs;
663#endif
664#ifdef CONFIG_PPC_DOORBELL
665 sum += per_cpu(irq_stat, cpu).doorbell_irqs;
666#endif
667
668 return sum;
669}
670
671static inline void check_stack_overflow(void)
672{
673 long sp;
674
675 if (!IS_ENABLED(CONFIG_DEBUG_STACKOVERFLOW))
676 return;
677
678 sp = current_stack_pointer & (THREAD_SIZE - 1);
679
680 /* check for stack overflow: is there less than 2KB free? */
681 if (unlikely(sp < 2048)) {
682 pr_err("do_IRQ: stack overflow: %ld\n", sp);
683 dump_stack();
684 }
685}
686
687static __always_inline void call_do_softirq(const void *sp)
688{
689 /* Temporarily switch r1 to sp, call __do_softirq() then restore r1. */
690 asm volatile (
691 PPC_STLU " %%r1, %[offset](%[sp]) ;"
692 "mr %%r1, %[sp] ;"
693 "bl %[callee] ;"
694 PPC_LL " %%r1, 0(%%r1) ;"
695 : // Outputs
696 : // Inputs
697 [sp] "b" (sp), [offset] "i" (THREAD_SIZE - STACK_FRAME_OVERHEAD),
698 [callee] "i" (__do_softirq)
699 : // Clobbers
700 "lr", "xer", "ctr", "memory", "cr0", "cr1", "cr5", "cr6",
701 "cr7", "r0", "r3", "r4", "r5", "r6", "r7", "r8", "r9", "r10",
702 "r11", "r12"
703 );
704}
705
706static __always_inline void call_do_irq(struct pt_regs *regs, void *sp)
707{
708 register unsigned long r3 asm("r3") = (unsigned long)regs;
709
710 /* Temporarily switch r1 to sp, call __do_irq() then restore r1. */
711 asm volatile (
712 PPC_STLU " %%r1, %[offset](%[sp]) ;"
713 "mr %%r1, %[sp] ;"
714 "bl %[callee] ;"
715 PPC_LL " %%r1, 0(%%r1) ;"
716 : // Outputs
717 "+r" (r3)
718 : // Inputs
719 [sp] "b" (sp), [offset] "i" (THREAD_SIZE - STACK_FRAME_OVERHEAD),
720 [callee] "i" (__do_irq)
721 : // Clobbers
722 "lr", "xer", "ctr", "memory", "cr0", "cr1", "cr5", "cr6",
723 "cr7", "r0", "r4", "r5", "r6", "r7", "r8", "r9", "r10",
724 "r11", "r12"
725 );
726}
727
728void __do_irq(struct pt_regs *regs)
729{
730 unsigned int irq;
731
732 trace_irq_entry(regs);
733
734 /*
735 * Query the platform PIC for the interrupt & ack it.
736 *
737 * This will typically lower the interrupt line to the CPU
738 */
739 irq = ppc_md.get_irq();
740
741 /* We can hard enable interrupts now to allow perf interrupts */
742 may_hard_irq_enable();
743
744 /* And finally process it */
745 if (unlikely(!irq))
746 __this_cpu_inc(irq_stat.spurious_irqs);
747 else
748 generic_handle_irq(irq);
749
750 trace_irq_exit(regs);
751}
752
753void __do_IRQ(struct pt_regs *regs)
754{
755 struct pt_regs *old_regs = set_irq_regs(regs);
756 void *cursp, *irqsp, *sirqsp;
757
758 /* Switch to the irq stack to handle this */
759 cursp = (void *)(current_stack_pointer & ~(THREAD_SIZE - 1));
760 irqsp = hardirq_ctx[raw_smp_processor_id()];
761 sirqsp = softirq_ctx[raw_smp_processor_id()];
762
763 check_stack_overflow();
764
765 /* Already there ? */
766 if (unlikely(cursp == irqsp || cursp == sirqsp)) {
767 __do_irq(regs);
768 set_irq_regs(old_regs);
769 return;
770 }
771 /* Switch stack and call */
772 call_do_irq(regs, irqsp);
773
774 set_irq_regs(old_regs);
775}
776
777DEFINE_INTERRUPT_HANDLER_ASYNC(do_IRQ)
778{
779 __do_IRQ(regs);
780}
781
782static void *__init alloc_vm_stack(void)
783{
784 return __vmalloc_node(THREAD_SIZE, THREAD_ALIGN, THREADINFO_GFP,
785 NUMA_NO_NODE, (void *)_RET_IP_);
786}
787
788static void __init vmap_irqstack_init(void)
789{
790 int i;
791
792 for_each_possible_cpu(i) {
793 softirq_ctx[i] = alloc_vm_stack();
794 hardirq_ctx[i] = alloc_vm_stack();
795 }
796}
797
798
799void __init init_IRQ(void)
800{
801 if (IS_ENABLED(CONFIG_VMAP_STACK))
802 vmap_irqstack_init();
803
804 if (ppc_md.init_IRQ)
805 ppc_md.init_IRQ();
806}
807
808#if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
809void *critirq_ctx[NR_CPUS] __read_mostly;
810void *dbgirq_ctx[NR_CPUS] __read_mostly;
811void *mcheckirq_ctx[NR_CPUS] __read_mostly;
812#endif
813
814void *softirq_ctx[NR_CPUS] __read_mostly;
815void *hardirq_ctx[NR_CPUS] __read_mostly;
816
817void do_softirq_own_stack(void)
818{
819 call_do_softirq(softirq_ctx[smp_processor_id()]);
820}
821
822irq_hw_number_t virq_to_hw(unsigned int virq)
823{
824 struct irq_data *irq_data = irq_get_irq_data(virq);
825 return WARN_ON(!irq_data) ? 0 : irq_data->hwirq;
826}
827EXPORT_SYMBOL_GPL(virq_to_hw);
828
829#ifdef CONFIG_SMP
830int irq_choose_cpu(const struct cpumask *mask)
831{
832 int cpuid;
833
834 if (cpumask_equal(mask, cpu_online_mask)) {
835 static int irq_rover;
836 static DEFINE_RAW_SPINLOCK(irq_rover_lock);
837 unsigned long flags;
838
839 /* Round-robin distribution... */
840do_round_robin:
841 raw_spin_lock_irqsave(&irq_rover_lock, flags);
842
843 irq_rover = cpumask_next(irq_rover, cpu_online_mask);
844 if (irq_rover >= nr_cpu_ids)
845 irq_rover = cpumask_first(cpu_online_mask);
846
847 cpuid = irq_rover;
848
849 raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
850 } else {
851 cpuid = cpumask_first_and(mask, cpu_online_mask);
852 if (cpuid >= nr_cpu_ids)
853 goto do_round_robin;
854 }
855
856 return get_hard_smp_processor_id(cpuid);
857}
858#else
859int irq_choose_cpu(const struct cpumask *mask)
860{
861 return hard_smp_processor_id();
862}
863#endif
864
865#ifdef CONFIG_PPC64
866static int __init setup_noirqdistrib(char *str)
867{
868 distribute_irqs = 0;
869 return 1;
870}
871
872__setup("noirqdistrib", setup_noirqdistrib);
873#endif /* CONFIG_PPC64 */
1/*
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)
10 *
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.
15 *
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
20 * should be easier.
21 *
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.
29 */
30
31#undef DEBUG
32
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/pci.h>
54#include <linux/debugfs.h>
55#include <linux/of.h>
56#include <linux/of_irq.h>
57
58#include <linux/uaccess.h>
59#include <asm/io.h>
60#include <asm/pgtable.h>
61#include <asm/irq.h>
62#include <asm/cache.h>
63#include <asm/prom.h>
64#include <asm/ptrace.h>
65#include <asm/machdep.h>
66#include <asm/udbg.h>
67#include <asm/smp.h>
68#include <asm/debug.h>
69#include <asm/livepatch.h>
70#include <asm/asm-prototypes.h>
71
72#ifdef CONFIG_PPC64
73#include <asm/paca.h>
74#include <asm/firmware.h>
75#include <asm/lv1call.h>
76#endif
77#define CREATE_TRACE_POINTS
78#include <asm/trace.h>
79#include <asm/cpu_has_feature.h>
80
81DEFINE_PER_CPU_SHARED_ALIGNED(irq_cpustat_t, irq_stat);
82EXPORT_PER_CPU_SYMBOL(irq_stat);
83
84int __irq_offset_value;
85
86#ifdef CONFIG_PPC32
87EXPORT_SYMBOL(__irq_offset_value);
88atomic_t ppc_n_lost_interrupts;
89
90#ifdef CONFIG_TAU_INT
91extern int tau_initialized;
92extern int tau_interrupts(int);
93#endif
94#endif /* CONFIG_PPC32 */
95
96#ifdef CONFIG_PPC64
97
98int distribute_irqs = 1;
99
100static inline notrace unsigned long get_irq_happened(void)
101{
102 unsigned long happened;
103
104 __asm__ __volatile__("lbz %0,%1(13)"
105 : "=r" (happened) : "i" (offsetof(struct paca_struct, irq_happened)));
106
107 return happened;
108}
109
110static inline notrace void set_soft_enabled(unsigned long enable)
111{
112 __asm__ __volatile__("stb %0,%1(13)"
113 : : "r" (enable), "i" (offsetof(struct paca_struct, soft_enabled)));
114}
115
116static inline notrace int decrementer_check_overflow(void)
117{
118 u64 now = get_tb_or_rtc();
119 u64 *next_tb = this_cpu_ptr(&decrementers_next_tb);
120
121 return now >= *next_tb;
122}
123
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.
127 *
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
132 * disabled.
133 *
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.
137 */
138notrace unsigned int __check_irq_replay(void)
139{
140 /*
141 * We use local_paca rather than get_paca() to avoid all
142 * the debug_smp_processor_id() business in this low level
143 * function
144 */
145 unsigned char happened = local_paca->irq_happened;
146
147 /* Clear bit 0 which we wouldn't clear otherwise */
148 local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
149
150 /*
151 * Force the delivery of pending soft-disabled interrupts on PS3.
152 * Any HV call will have this side effect.
153 */
154 if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
155 u64 tmp, tmp2;
156 lv1_get_version_info(&tmp, &tmp2);
157 }
158
159 /*
160 * Check if an hypervisor Maintenance interrupt happened.
161 * This is a higher priority interrupt than the others, so
162 * replay it first.
163 */
164 local_paca->irq_happened &= ~PACA_IRQ_HMI;
165 if (happened & PACA_IRQ_HMI)
166 return 0xe60;
167
168 /*
169 * We may have missed a decrementer interrupt. We check the
170 * decrementer itself rather than the paca irq_happened field
171 * in case we also had a rollover while hard disabled
172 */
173 local_paca->irq_happened &= ~PACA_IRQ_DEC;
174 if ((happened & PACA_IRQ_DEC) || decrementer_check_overflow())
175 return 0x900;
176
177 /* Finally check if an external interrupt happened */
178 local_paca->irq_happened &= ~PACA_IRQ_EE;
179 if (happened & PACA_IRQ_EE)
180 return 0x500;
181
182#ifdef CONFIG_PPC_BOOK3E
183 /* Finally check if an EPR external interrupt happened
184 * this bit is typically set if we need to handle another
185 * "edge" interrupt from within the MPIC "EPR" handler
186 */
187 local_paca->irq_happened &= ~PACA_IRQ_EE_EDGE;
188 if (happened & PACA_IRQ_EE_EDGE)
189 return 0x500;
190
191 local_paca->irq_happened &= ~PACA_IRQ_DBELL;
192 if (happened & PACA_IRQ_DBELL)
193 return 0x280;
194#else
195 local_paca->irq_happened &= ~PACA_IRQ_DBELL;
196 if (happened & PACA_IRQ_DBELL) {
197 if (cpu_has_feature(CPU_FTR_HVMODE))
198 return 0xe80;
199 return 0xa00;
200 }
201#endif /* CONFIG_PPC_BOOK3E */
202
203 /* There should be nothing left ! */
204 BUG_ON(local_paca->irq_happened != 0);
205
206 return 0;
207}
208
209notrace void arch_local_irq_restore(unsigned long en)
210{
211 unsigned char irq_happened;
212 unsigned int replay;
213
214 /* Write the new soft-enabled value */
215 set_soft_enabled(en);
216 if (!en)
217 return;
218 /*
219 * From this point onward, we can take interrupts, preempt,
220 * etc... unless we got hard-disabled. We check if an event
221 * happened. If none happened, we know we can just return.
222 *
223 * We may have preempted before the check below, in which case
224 * we are checking the "new" CPU instead of the old one. This
225 * is only a problem if an event happened on the "old" CPU.
226 *
227 * External interrupt events will have caused interrupts to
228 * be hard-disabled, so there is no problem, we
229 * cannot have preempted.
230 */
231 irq_happened = get_irq_happened();
232 if (!irq_happened)
233 return;
234
235 /*
236 * We need to hard disable to get a trusted value from
237 * __check_irq_replay(). We also need to soft-disable
238 * again to avoid warnings in there due to the use of
239 * per-cpu variables.
240 *
241 * We know that if the value in irq_happened is exactly 0x01
242 * then we are already hard disabled (there are other less
243 * common cases that we'll ignore for now), so we skip the
244 * (expensive) mtmsrd.
245 */
246 if (unlikely(irq_happened != PACA_IRQ_HARD_DIS))
247 __hard_irq_disable();
248#ifdef CONFIG_TRACE_IRQFLAGS
249 else {
250 /*
251 * We should already be hard disabled here. We had bugs
252 * where that wasn't the case so let's dbl check it and
253 * warn if we are wrong. Only do that when IRQ tracing
254 * is enabled as mfmsr() can be costly.
255 */
256 if (WARN_ON(mfmsr() & MSR_EE))
257 __hard_irq_disable();
258 }
259#endif /* CONFIG_TRACE_IRQFLAGS */
260
261 set_soft_enabled(0);
262
263 /*
264 * Check if anything needs to be re-emitted. We haven't
265 * soft-enabled yet to avoid warnings in decrementer_check_overflow
266 * accessing per-cpu variables
267 */
268 replay = __check_irq_replay();
269
270 /* We can soft-enable now */
271 set_soft_enabled(1);
272
273 /*
274 * And replay if we have to. This will return with interrupts
275 * hard-enabled.
276 */
277 if (replay) {
278 __replay_interrupt(replay);
279 return;
280 }
281
282 /* Finally, let's ensure we are hard enabled */
283 __hard_irq_enable();
284}
285EXPORT_SYMBOL(arch_local_irq_restore);
286
287/*
288 * This is specifically called by assembly code to re-enable interrupts
289 * if they are currently disabled. This is typically called before
290 * schedule() or do_signal() when returning to userspace. We do it
291 * in C to avoid the burden of dealing with lockdep etc...
292 *
293 * NOTE: This is called with interrupts hard disabled but not marked
294 * as such in paca->irq_happened, so we need to resync this.
295 */
296void notrace restore_interrupts(void)
297{
298 if (irqs_disabled()) {
299 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
300 local_irq_enable();
301 } else
302 __hard_irq_enable();
303}
304
305/*
306 * This is a helper to use when about to go into idle low-power
307 * when the latter has the side effect of re-enabling interrupts
308 * (such as calling H_CEDE under pHyp).
309 *
310 * You call this function with interrupts soft-disabled (this is
311 * already the case when ppc_md.power_save is called). The function
312 * will return whether to enter power save or just return.
313 *
314 * In the former case, it will have notified lockdep of interrupts
315 * being re-enabled and generally sanitized the lazy irq state,
316 * and in the latter case it will leave with interrupts hard
317 * disabled and marked as such, so the local_irq_enable() call
318 * in arch_cpu_idle() will properly re-enable everything.
319 */
320bool prep_irq_for_idle(void)
321{
322 /*
323 * First we need to hard disable to ensure no interrupt
324 * occurs before we effectively enter the low power state
325 */
326 hard_irq_disable();
327
328 /*
329 * If anything happened while we were soft-disabled,
330 * we return now and do not enter the low power state.
331 */
332 if (lazy_irq_pending())
333 return false;
334
335 /* Tell lockdep we are about to re-enable */
336 trace_hardirqs_on();
337
338 /*
339 * Mark interrupts as soft-enabled and clear the
340 * PACA_IRQ_HARD_DIS from the pending mask since we
341 * are about to hard enable as well as a side effect
342 * of entering the low power state.
343 */
344 local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
345 local_paca->soft_enabled = 1;
346
347 /* Tell the caller to enter the low power state */
348 return true;
349}
350
351/*
352 * Force a replay of the external interrupt handler on this CPU.
353 */
354void force_external_irq_replay(void)
355{
356 /*
357 * This must only be called with interrupts soft-disabled,
358 * the replay will happen when re-enabling.
359 */
360 WARN_ON(!arch_irqs_disabled());
361
362 /* Indicate in the PACA that we have an interrupt to replay */
363 local_paca->irq_happened |= PACA_IRQ_EE;
364}
365
366#endif /* CONFIG_PPC64 */
367
368int arch_show_interrupts(struct seq_file *p, int prec)
369{
370 int j;
371
372#if defined(CONFIG_PPC32) && defined(CONFIG_TAU_INT)
373 if (tau_initialized) {
374 seq_printf(p, "%*s: ", prec, "TAU");
375 for_each_online_cpu(j)
376 seq_printf(p, "%10u ", tau_interrupts(j));
377 seq_puts(p, " PowerPC Thermal Assist (cpu temp)\n");
378 }
379#endif /* CONFIG_PPC32 && CONFIG_TAU_INT */
380
381 seq_printf(p, "%*s: ", prec, "LOC");
382 for_each_online_cpu(j)
383 seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_event);
384 seq_printf(p, " Local timer interrupts for timer event device\n");
385
386 seq_printf(p, "%*s: ", prec, "LOC");
387 for_each_online_cpu(j)
388 seq_printf(p, "%10u ", per_cpu(irq_stat, j).timer_irqs_others);
389 seq_printf(p, " Local timer interrupts for others\n");
390
391 seq_printf(p, "%*s: ", prec, "SPU");
392 for_each_online_cpu(j)
393 seq_printf(p, "%10u ", per_cpu(irq_stat, j).spurious_irqs);
394 seq_printf(p, " Spurious interrupts\n");
395
396 seq_printf(p, "%*s: ", prec, "PMI");
397 for_each_online_cpu(j)
398 seq_printf(p, "%10u ", per_cpu(irq_stat, j).pmu_irqs);
399 seq_printf(p, " Performance monitoring interrupts\n");
400
401 seq_printf(p, "%*s: ", prec, "MCE");
402 for_each_online_cpu(j)
403 seq_printf(p, "%10u ", per_cpu(irq_stat, j).mce_exceptions);
404 seq_printf(p, " Machine check exceptions\n");
405
406 if (cpu_has_feature(CPU_FTR_HVMODE)) {
407 seq_printf(p, "%*s: ", prec, "HMI");
408 for_each_online_cpu(j)
409 seq_printf(p, "%10u ",
410 per_cpu(irq_stat, j).hmi_exceptions);
411 seq_printf(p, " Hypervisor Maintenance Interrupts\n");
412 }
413
414#ifdef CONFIG_PPC_DOORBELL
415 if (cpu_has_feature(CPU_FTR_DBELL)) {
416 seq_printf(p, "%*s: ", prec, "DBL");
417 for_each_online_cpu(j)
418 seq_printf(p, "%10u ", per_cpu(irq_stat, j).doorbell_irqs);
419 seq_printf(p, " Doorbell interrupts\n");
420 }
421#endif
422
423 return 0;
424}
425
426/*
427 * /proc/stat helpers
428 */
429u64 arch_irq_stat_cpu(unsigned int cpu)
430{
431 u64 sum = per_cpu(irq_stat, cpu).timer_irqs_event;
432
433 sum += per_cpu(irq_stat, cpu).pmu_irqs;
434 sum += per_cpu(irq_stat, cpu).mce_exceptions;
435 sum += per_cpu(irq_stat, cpu).spurious_irqs;
436 sum += per_cpu(irq_stat, cpu).timer_irqs_others;
437 sum += per_cpu(irq_stat, cpu).hmi_exceptions;
438#ifdef CONFIG_PPC_DOORBELL
439 sum += per_cpu(irq_stat, cpu).doorbell_irqs;
440#endif
441
442 return sum;
443}
444
445#ifdef CONFIG_HOTPLUG_CPU
446void migrate_irqs(void)
447{
448 struct irq_desc *desc;
449 unsigned int irq;
450 static int warned;
451 cpumask_var_t mask;
452 const struct cpumask *map = cpu_online_mask;
453
454 alloc_cpumask_var(&mask, GFP_KERNEL);
455
456 for_each_irq_desc(irq, desc) {
457 struct irq_data *data;
458 struct irq_chip *chip;
459
460 data = irq_desc_get_irq_data(desc);
461 if (irqd_is_per_cpu(data))
462 continue;
463
464 chip = irq_data_get_irq_chip(data);
465
466 cpumask_and(mask, irq_data_get_affinity_mask(data), map);
467 if (cpumask_any(mask) >= nr_cpu_ids) {
468 pr_warn("Breaking affinity for irq %i\n", irq);
469 cpumask_copy(mask, map);
470 }
471 if (chip->irq_set_affinity)
472 chip->irq_set_affinity(data, mask, true);
473 else if (desc->action && !(warned++))
474 pr_err("Cannot set affinity for irq %i\n", irq);
475 }
476
477 free_cpumask_var(mask);
478
479 local_irq_enable();
480 mdelay(1);
481 local_irq_disable();
482}
483#endif
484
485static inline void check_stack_overflow(void)
486{
487#ifdef CONFIG_DEBUG_STACKOVERFLOW
488 long sp;
489
490 sp = current_stack_pointer() & (THREAD_SIZE-1);
491
492 /* check for stack overflow: is there less than 2KB free? */
493 if (unlikely(sp < (sizeof(struct thread_info) + 2048))) {
494 pr_err("do_IRQ: stack overflow: %ld\n",
495 sp - sizeof(struct thread_info));
496 dump_stack();
497 }
498#endif
499}
500
501void __do_irq(struct pt_regs *regs)
502{
503 unsigned int irq;
504
505 irq_enter();
506
507 trace_irq_entry(regs);
508
509 check_stack_overflow();
510
511 /*
512 * Query the platform PIC for the interrupt & ack it.
513 *
514 * This will typically lower the interrupt line to the CPU
515 */
516 irq = ppc_md.get_irq();
517
518 /* We can hard enable interrupts now to allow perf interrupts */
519 may_hard_irq_enable();
520
521 /* And finally process it */
522 if (unlikely(!irq))
523 __this_cpu_inc(irq_stat.spurious_irqs);
524 else
525 generic_handle_irq(irq);
526
527 trace_irq_exit(regs);
528
529 irq_exit();
530}
531
532void do_IRQ(struct pt_regs *regs)
533{
534 struct pt_regs *old_regs = set_irq_regs(regs);
535 struct thread_info *curtp, *irqtp, *sirqtp;
536
537 /* Switch to the irq stack to handle this */
538 curtp = current_thread_info();
539 irqtp = hardirq_ctx[raw_smp_processor_id()];
540 sirqtp = softirq_ctx[raw_smp_processor_id()];
541
542 /* Already there ? */
543 if (unlikely(curtp == irqtp || curtp == sirqtp)) {
544 __do_irq(regs);
545 set_irq_regs(old_regs);
546 return;
547 }
548
549 /* Prepare the thread_info in the irq stack */
550 irqtp->task = curtp->task;
551 irqtp->flags = 0;
552
553 /* Copy the preempt_count so that the [soft]irq checks work. */
554 irqtp->preempt_count = curtp->preempt_count;
555
556 /* Switch stack and call */
557 call_do_irq(regs, irqtp);
558
559 /* Restore stack limit */
560 irqtp->task = NULL;
561
562 /* Copy back updates to the thread_info */
563 if (irqtp->flags)
564 set_bits(irqtp->flags, &curtp->flags);
565
566 set_irq_regs(old_regs);
567}
568
569void __init init_IRQ(void)
570{
571 if (ppc_md.init_IRQ)
572 ppc_md.init_IRQ();
573
574 exc_lvl_ctx_init();
575
576 irq_ctx_init();
577}
578
579#if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
580struct thread_info *critirq_ctx[NR_CPUS] __read_mostly;
581struct thread_info *dbgirq_ctx[NR_CPUS] __read_mostly;
582struct thread_info *mcheckirq_ctx[NR_CPUS] __read_mostly;
583
584void exc_lvl_ctx_init(void)
585{
586 struct thread_info *tp;
587 int i, cpu_nr;
588
589 for_each_possible_cpu(i) {
590#ifdef CONFIG_PPC64
591 cpu_nr = i;
592#else
593#ifdef CONFIG_SMP
594 cpu_nr = get_hard_smp_processor_id(i);
595#else
596 cpu_nr = 0;
597#endif
598#endif
599
600 memset((void *)critirq_ctx[cpu_nr], 0, THREAD_SIZE);
601 tp = critirq_ctx[cpu_nr];
602 tp->cpu = cpu_nr;
603 tp->preempt_count = 0;
604
605#ifdef CONFIG_BOOKE
606 memset((void *)dbgirq_ctx[cpu_nr], 0, THREAD_SIZE);
607 tp = dbgirq_ctx[cpu_nr];
608 tp->cpu = cpu_nr;
609 tp->preempt_count = 0;
610
611 memset((void *)mcheckirq_ctx[cpu_nr], 0, THREAD_SIZE);
612 tp = mcheckirq_ctx[cpu_nr];
613 tp->cpu = cpu_nr;
614 tp->preempt_count = HARDIRQ_OFFSET;
615#endif
616 }
617}
618#endif
619
620struct thread_info *softirq_ctx[NR_CPUS] __read_mostly;
621struct thread_info *hardirq_ctx[NR_CPUS] __read_mostly;
622
623void irq_ctx_init(void)
624{
625 struct thread_info *tp;
626 int i;
627
628 for_each_possible_cpu(i) {
629 memset((void *)softirq_ctx[i], 0, THREAD_SIZE);
630 tp = softirq_ctx[i];
631 tp->cpu = i;
632 klp_init_thread_info(tp);
633
634 memset((void *)hardirq_ctx[i], 0, THREAD_SIZE);
635 tp = hardirq_ctx[i];
636 tp->cpu = i;
637 klp_init_thread_info(tp);
638 }
639}
640
641void do_softirq_own_stack(void)
642{
643 struct thread_info *curtp, *irqtp;
644
645 curtp = current_thread_info();
646 irqtp = softirq_ctx[smp_processor_id()];
647 irqtp->task = curtp->task;
648 irqtp->flags = 0;
649 call_do_softirq(irqtp);
650 irqtp->task = NULL;
651
652 /* Set any flag that may have been set on the
653 * alternate stack
654 */
655 if (irqtp->flags)
656 set_bits(irqtp->flags, &curtp->flags);
657}
658
659irq_hw_number_t virq_to_hw(unsigned int virq)
660{
661 struct irq_data *irq_data = irq_get_irq_data(virq);
662 return WARN_ON(!irq_data) ? 0 : irq_data->hwirq;
663}
664EXPORT_SYMBOL_GPL(virq_to_hw);
665
666#ifdef CONFIG_SMP
667int irq_choose_cpu(const struct cpumask *mask)
668{
669 int cpuid;
670
671 if (cpumask_equal(mask, cpu_online_mask)) {
672 static int irq_rover;
673 static DEFINE_RAW_SPINLOCK(irq_rover_lock);
674 unsigned long flags;
675
676 /* Round-robin distribution... */
677do_round_robin:
678 raw_spin_lock_irqsave(&irq_rover_lock, flags);
679
680 irq_rover = cpumask_next(irq_rover, cpu_online_mask);
681 if (irq_rover >= nr_cpu_ids)
682 irq_rover = cpumask_first(cpu_online_mask);
683
684 cpuid = irq_rover;
685
686 raw_spin_unlock_irqrestore(&irq_rover_lock, flags);
687 } else {
688 cpuid = cpumask_first_and(mask, cpu_online_mask);
689 if (cpuid >= nr_cpu_ids)
690 goto do_round_robin;
691 }
692
693 return get_hard_smp_processor_id(cpuid);
694}
695#else
696int irq_choose_cpu(const struct cpumask *mask)
697{
698 return hard_smp_processor_id();
699}
700#endif
701
702int arch_early_irq_init(void)
703{
704 return 0;
705}
706
707#ifdef CONFIG_PPC64
708static int __init setup_noirqdistrib(char *str)
709{
710 distribute_irqs = 0;
711 return 1;
712}
713
714__setup("noirqdistrib", setup_noirqdistrib);
715#endif /* CONFIG_PPC64 */