xref: /linux/arch/powerpc/kernel/irq_64.c (revision be239684b18e1cdcafcf8c7face4a2f562c745ad)
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 
19 #undef DEBUG
20 
21 #include <linux/export.h>
22 #include <linux/threads.h>
23 #include <linux/kernel_stat.h>
24 #include <linux/signal.h>
25 #include <linux/sched.h>
26 #include <linux/ptrace.h>
27 #include <linux/ioport.h>
28 #include <linux/interrupt.h>
29 #include <linux/timex.h>
30 #include <linux/init.h>
31 #include <linux/slab.h>
32 #include <linux/delay.h>
33 #include <linux/irq.h>
34 #include <linux/seq_file.h>
35 #include <linux/cpumask.h>
36 #include <linux/profile.h>
37 #include <linux/bitops.h>
38 #include <linux/list.h>
39 #include <linux/radix-tree.h>
40 #include <linux/mutex.h>
41 #include <linux/pci.h>
42 #include <linux/debugfs.h>
43 #include <linux/of.h>
44 #include <linux/of_irq.h>
45 #include <linux/vmalloc.h>
46 #include <linux/pgtable.h>
47 #include <linux/static_call.h>
48 
49 #include <linux/uaccess.h>
50 #include <asm/interrupt.h>
51 #include <asm/io.h>
52 #include <asm/irq.h>
53 #include <asm/cache.h>
54 #include <asm/ptrace.h>
55 #include <asm/machdep.h>
56 #include <asm/udbg.h>
57 #include <asm/smp.h>
58 #include <asm/hw_irq.h>
59 #include <asm/softirq_stack.h>
60 #include <asm/ppc_asm.h>
61 
62 #include <asm/paca.h>
63 #include <asm/firmware.h>
64 #include <asm/lv1call.h>
65 #include <asm/dbell.h>
66 #include <asm/trace.h>
67 #include <asm/cpu_has_feature.h>
68 
69 int distribute_irqs = 1;
70 
71 static inline void next_interrupt(struct pt_regs *regs)
72 {
73 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
74 		WARN_ON(!(local_paca->irq_happened & PACA_IRQ_HARD_DIS));
75 		WARN_ON(irq_soft_mask_return() != IRQS_ALL_DISABLED);
76 	}
77 
78 	/*
79 	 * We are responding to the next interrupt, so interrupt-off
80 	 * latencies should be reset here.
81 	 */
82 	lockdep_hardirq_exit();
83 	trace_hardirqs_on();
84 	trace_hardirqs_off();
85 	lockdep_hardirq_enter();
86 }
87 
88 static inline bool irq_happened_test_and_clear(u8 irq)
89 {
90 	if (local_paca->irq_happened & irq) {
91 		local_paca->irq_happened &= ~irq;
92 		return true;
93 	}
94 	return false;
95 }
96 
97 static __no_kcsan void __replay_soft_interrupts(void)
98 {
99 	struct pt_regs regs;
100 
101 	/*
102 	 * We use local_paca rather than get_paca() to avoid all the
103 	 * debug_smp_processor_id() business in this low level function.
104 	 */
105 
106 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
107 		WARN_ON_ONCE(mfmsr() & MSR_EE);
108 		WARN_ON(!(local_paca->irq_happened & PACA_IRQ_HARD_DIS));
109 		WARN_ON(local_paca->irq_happened & PACA_IRQ_REPLAYING);
110 	}
111 
112 	/*
113 	 * PACA_IRQ_REPLAYING prevents interrupt handlers from enabling
114 	 * MSR[EE] to get PMIs, which can result in more IRQs becoming
115 	 * pending.
116 	 */
117 	local_paca->irq_happened |= PACA_IRQ_REPLAYING;
118 
119 	ppc_save_regs(&regs);
120 	regs.softe = IRQS_ENABLED;
121 	regs.msr |= MSR_EE;
122 
123 	/*
124 	 * Force the delivery of pending soft-disabled interrupts on PS3.
125 	 * Any HV call will have this side effect.
126 	 */
127 	if (firmware_has_feature(FW_FEATURE_PS3_LV1)) {
128 		u64 tmp, tmp2;
129 		lv1_get_version_info(&tmp, &tmp2);
130 	}
131 
132 	/*
133 	 * Check if an hypervisor Maintenance interrupt happened.
134 	 * This is a higher priority interrupt than the others, so
135 	 * replay it first.
136 	 */
137 	if (IS_ENABLED(CONFIG_PPC_BOOK3S) &&
138 	    irq_happened_test_and_clear(PACA_IRQ_HMI)) {
139 		regs.trap = INTERRUPT_HMI;
140 		handle_hmi_exception(&regs);
141 		next_interrupt(&regs);
142 	}
143 
144 	if (irq_happened_test_and_clear(PACA_IRQ_DEC)) {
145 		regs.trap = INTERRUPT_DECREMENTER;
146 		timer_interrupt(&regs);
147 		next_interrupt(&regs);
148 	}
149 
150 	if (irq_happened_test_and_clear(PACA_IRQ_EE)) {
151 		regs.trap = INTERRUPT_EXTERNAL;
152 		do_IRQ(&regs);
153 		next_interrupt(&regs);
154 	}
155 
156 	if (IS_ENABLED(CONFIG_PPC_DOORBELL) &&
157 	    irq_happened_test_and_clear(PACA_IRQ_DBELL)) {
158 		regs.trap = INTERRUPT_DOORBELL;
159 		doorbell_exception(&regs);
160 		next_interrupt(&regs);
161 	}
162 
163 	/* Book3E does not support soft-masking PMI interrupts */
164 	if (IS_ENABLED(CONFIG_PPC_BOOK3S) &&
165 	    irq_happened_test_and_clear(PACA_IRQ_PMI)) {
166 		regs.trap = INTERRUPT_PERFMON;
167 		performance_monitor_exception(&regs);
168 		next_interrupt(&regs);
169 	}
170 
171 	local_paca->irq_happened &= ~PACA_IRQ_REPLAYING;
172 }
173 
174 __no_kcsan void replay_soft_interrupts(void)
175 {
176 	irq_enter(); /* See comment in arch_local_irq_restore */
177 	__replay_soft_interrupts();
178 	irq_exit();
179 }
180 
181 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_KUAP)
182 static inline __no_kcsan void replay_soft_interrupts_irqrestore(void)
183 {
184 	unsigned long kuap_state = get_kuap();
185 
186 	/*
187 	 * Check if anything calls local_irq_enable/restore() when KUAP is
188 	 * disabled (user access enabled). We handle that case here by saving
189 	 * and re-locking AMR but we shouldn't get here in the first place,
190 	 * hence the warning.
191 	 */
192 	kuap_assert_locked();
193 
194 	if (kuap_state != AMR_KUAP_BLOCKED)
195 		set_kuap(AMR_KUAP_BLOCKED);
196 
197 	__replay_soft_interrupts();
198 
199 	if (kuap_state != AMR_KUAP_BLOCKED)
200 		set_kuap(kuap_state);
201 }
202 #else
203 #define replay_soft_interrupts_irqrestore() __replay_soft_interrupts()
204 #endif
205 
206 notrace __no_kcsan void arch_local_irq_restore(unsigned long mask)
207 {
208 	unsigned char irq_happened;
209 
210 	/* Write the new soft-enabled value if it is a disable */
211 	if (mask) {
212 		irq_soft_mask_set(mask);
213 		return;
214 	}
215 
216 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
217 		WARN_ON_ONCE(in_nmi());
218 		WARN_ON_ONCE(in_hardirq());
219 		WARN_ON_ONCE(local_paca->irq_happened & PACA_IRQ_REPLAYING);
220 	}
221 
222 again:
223 	/*
224 	 * After the stb, interrupts are unmasked and there are no interrupts
225 	 * pending replay. The restart sequence makes this atomic with
226 	 * respect to soft-masked interrupts. If this was just a simple code
227 	 * sequence, a soft-masked interrupt could become pending right after
228 	 * the comparison and before the stb.
229 	 *
230 	 * This allows interrupts to be unmasked without hard disabling, and
231 	 * also without new hard interrupts coming in ahead of pending ones.
232 	 */
233 	asm goto(
234 "1:					\n"
235 "		lbz	9,%0(13)	\n"
236 "		cmpwi	9,0		\n"
237 "		bne	%l[happened]	\n"
238 "		stb	9,%1(13)	\n"
239 "2:					\n"
240 		RESTART_TABLE(1b, 2b, 1b)
241 	: : "i" (offsetof(struct paca_struct, irq_happened)),
242 	    "i" (offsetof(struct paca_struct, irq_soft_mask))
243 	: "cr0", "r9"
244 	: happened);
245 
246 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
247 		WARN_ON_ONCE(!(mfmsr() & MSR_EE));
248 
249 	/*
250 	 * If we came here from the replay below, we might have a preempt
251 	 * pending (due to preempt_enable_no_resched()). Have to check now.
252 	 */
253 	preempt_check_resched();
254 
255 	return;
256 
257 happened:
258 	irq_happened = READ_ONCE(local_paca->irq_happened);
259 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
260 		WARN_ON_ONCE(!irq_happened);
261 
262 	if (irq_happened == PACA_IRQ_HARD_DIS) {
263 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
264 			WARN_ON_ONCE(mfmsr() & MSR_EE);
265 		irq_soft_mask_set(IRQS_ENABLED);
266 		local_paca->irq_happened = 0;
267 		__hard_irq_enable();
268 		preempt_check_resched();
269 		return;
270 	}
271 
272 	/* Have interrupts to replay, need to hard disable first */
273 	if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
274 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
275 			if (!(mfmsr() & MSR_EE)) {
276 				/*
277 				 * An interrupt could have come in and cleared
278 				 * MSR[EE] and set IRQ_HARD_DIS, so check
279 				 * IRQ_HARD_DIS again and warn if it is still
280 				 * clear.
281 				 */
282 				irq_happened = READ_ONCE(local_paca->irq_happened);
283 				WARN_ON_ONCE(!(irq_happened & PACA_IRQ_HARD_DIS));
284 			}
285 		}
286 		__hard_irq_disable();
287 		local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
288 	} else {
289 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
290 			if (WARN_ON_ONCE(mfmsr() & MSR_EE))
291 				__hard_irq_disable();
292 		}
293 	}
294 
295 	/*
296 	 * Disable preempt here, so that the below preempt_enable will
297 	 * perform resched if required (a replayed interrupt may set
298 	 * need_resched).
299 	 */
300 	preempt_disable();
301 	irq_soft_mask_set(IRQS_ALL_DISABLED);
302 	trace_hardirqs_off();
303 
304 	/*
305 	 * Now enter interrupt context. The interrupt handlers themselves
306 	 * also call irq_enter/exit (which is okay, they can nest). But call
307 	 * it here now to hold off softirqs until the below irq_exit(). If
308 	 * we allowed replayed handlers to run softirqs, that enables irqs,
309 	 * which must replay interrupts, which recurses in here and makes
310 	 * things more complicated. The recursion is limited to 2, and it can
311 	 * be made to work, but it's complicated.
312 	 *
313 	 * local_bh_disable can not be used here because interrupts taken in
314 	 * idle are not in the right context (RCU, tick, etc) to run softirqs
315 	 * so irq_enter must be called.
316 	 */
317 	irq_enter();
318 
319 	replay_soft_interrupts_irqrestore();
320 
321 	irq_exit();
322 
323 	if (unlikely(local_paca->irq_happened != PACA_IRQ_HARD_DIS)) {
324 		/*
325 		 * The softirq processing in irq_exit() may enable interrupts
326 		 * temporarily, which can result in MSR[EE] being enabled and
327 		 * more irqs becoming pending. Go around again if that happens.
328 		 */
329 		trace_hardirqs_on();
330 		preempt_enable_no_resched();
331 		goto again;
332 	}
333 
334 	trace_hardirqs_on();
335 	irq_soft_mask_set(IRQS_ENABLED);
336 	local_paca->irq_happened = 0;
337 	__hard_irq_enable();
338 	preempt_enable();
339 }
340 EXPORT_SYMBOL(arch_local_irq_restore);
341 
342 /*
343  * This is a helper to use when about to go into idle low-power
344  * when the latter has the side effect of re-enabling interrupts
345  * (such as calling H_CEDE under pHyp).
346  *
347  * You call this function with interrupts soft-disabled (this is
348  * already the case when ppc_md.power_save is called). The function
349  * will return whether to enter power save or just return.
350  *
351  * In the former case, it will have generally sanitized the lazy irq
352  * state, and in the latter case it will leave with interrupts hard
353  * disabled and marked as such, so the local_irq_enable() call
354  * in arch_cpu_idle() will properly re-enable everything.
355  */
356 __cpuidle bool prep_irq_for_idle(void)
357 {
358 	/*
359 	 * First we need to hard disable to ensure no interrupt
360 	 * occurs before we effectively enter the low power state
361 	 */
362 	__hard_irq_disable();
363 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
364 
365 	/*
366 	 * If anything happened while we were soft-disabled,
367 	 * we return now and do not enter the low power state.
368 	 */
369 	if (lazy_irq_pending())
370 		return false;
371 
372 	/*
373 	 * Mark interrupts as soft-enabled and clear the
374 	 * PACA_IRQ_HARD_DIS from the pending mask since we
375 	 * are about to hard enable as well as a side effect
376 	 * of entering the low power state.
377 	 */
378 	local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
379 	irq_soft_mask_set(IRQS_ENABLED);
380 
381 	/* Tell the caller to enter the low power state */
382 	return true;
383 }
384 
385 #ifdef CONFIG_PPC_BOOK3S
386 /*
387  * This is for idle sequences that return with IRQs off, but the
388  * idle state itself wakes on interrupt. Tell the irq tracer that
389  * IRQs are enabled for the duration of idle so it does not get long
390  * off times. Must be paired with fini_irq_for_idle_irqsoff.
391  */
392 bool prep_irq_for_idle_irqsoff(void)
393 {
394 	WARN_ON(!irqs_disabled());
395 
396 	/*
397 	 * First we need to hard disable to ensure no interrupt
398 	 * occurs before we effectively enter the low power state
399 	 */
400 	__hard_irq_disable();
401 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
402 
403 	/*
404 	 * If anything happened while we were soft-disabled,
405 	 * we return now and do not enter the low power state.
406 	 */
407 	if (lazy_irq_pending())
408 		return false;
409 
410 	/* Tell lockdep we are about to re-enable */
411 	trace_hardirqs_on();
412 
413 	return true;
414 }
415 
416 /*
417  * Take the SRR1 wakeup reason, index into this table to find the
418  * appropriate irq_happened bit.
419  *
420  * Sytem reset exceptions taken in idle state also come through here,
421  * but they are NMI interrupts so do not need to wait for IRQs to be
422  * restored, and should be taken as early as practical. These are marked
423  * with 0xff in the table. The Power ISA specifies 0100b as the system
424  * reset interrupt reason.
425  */
426 #define IRQ_SYSTEM_RESET	0xff
427 
428 static const u8 srr1_to_lazyirq[0x10] = {
429 	0, 0, 0,
430 	PACA_IRQ_DBELL,
431 	IRQ_SYSTEM_RESET,
432 	PACA_IRQ_DBELL,
433 	PACA_IRQ_DEC,
434 	0,
435 	PACA_IRQ_EE,
436 	PACA_IRQ_EE,
437 	PACA_IRQ_HMI,
438 	0, 0, 0, 0, 0 };
439 
440 void replay_system_reset(void)
441 {
442 	struct pt_regs regs;
443 
444 	ppc_save_regs(&regs);
445 	regs.trap = 0x100;
446 	get_paca()->in_nmi = 1;
447 	system_reset_exception(&regs);
448 	get_paca()->in_nmi = 0;
449 }
450 EXPORT_SYMBOL_GPL(replay_system_reset);
451 
452 void irq_set_pending_from_srr1(unsigned long srr1)
453 {
454 	unsigned int idx = (srr1 & SRR1_WAKEMASK_P8) >> 18;
455 	u8 reason = srr1_to_lazyirq[idx];
456 
457 	/*
458 	 * Take the system reset now, which is immediately after registers
459 	 * are restored from idle. It's an NMI, so interrupts need not be
460 	 * re-enabled before it is taken.
461 	 */
462 	if (unlikely(reason == IRQ_SYSTEM_RESET)) {
463 		replay_system_reset();
464 		return;
465 	}
466 
467 	if (reason == PACA_IRQ_DBELL) {
468 		/*
469 		 * When doorbell triggers a system reset wakeup, the message
470 		 * is not cleared, so if the doorbell interrupt is replayed
471 		 * and the IPI handled, the doorbell interrupt would still
472 		 * fire when EE is enabled.
473 		 *
474 		 * To avoid taking the superfluous doorbell interrupt,
475 		 * execute a msgclr here before the interrupt is replayed.
476 		 */
477 		ppc_msgclr(PPC_DBELL_MSGTYPE);
478 	}
479 
480 	/*
481 	 * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
482 	 * so this can be called unconditionally with the SRR1 wake
483 	 * reason as returned by the idle code, which uses 0 to mean no
484 	 * interrupt.
485 	 *
486 	 * If a future CPU was to designate this as an interrupt reason,
487 	 * then a new index for no interrupt must be assigned.
488 	 */
489 	local_paca->irq_happened |= reason;
490 }
491 #endif /* CONFIG_PPC_BOOK3S */
492 
493 /*
494  * Force a replay of the external interrupt handler on this CPU.
495  */
496 void force_external_irq_replay(void)
497 {
498 	/*
499 	 * This must only be called with interrupts soft-disabled,
500 	 * the replay will happen when re-enabling.
501 	 */
502 	WARN_ON(!arch_irqs_disabled());
503 
504 	/*
505 	 * Interrupts must always be hard disabled before irq_happened is
506 	 * modified (to prevent lost update in case of interrupt between
507 	 * load and store).
508 	 */
509 	__hard_irq_disable();
510 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
511 
512 	/* Indicate in the PACA that we have an interrupt to replay */
513 	local_paca->irq_happened |= PACA_IRQ_EE;
514 }
515 
516 static int __init setup_noirqdistrib(char *str)
517 {
518 	distribute_irqs = 0;
519 	return 1;
520 }
521 
522 __setup("noirqdistrib", setup_noirqdistrib);
523