xref: /openbmc/linux/arch/powerpc/kernel/irq_64.c (revision 83b975b5)
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 	/*
74 	 * Softirq processing can enable/disable irqs, which will leave
75 	 * MSR[EE] enabled and the soft mask set to IRQS_DISABLED. Fix
76 	 * this up.
77 	 */
78 	if (!(local_paca->irq_happened & PACA_IRQ_HARD_DIS))
79 		hard_irq_disable();
80 	else
81 		irq_soft_mask_set(IRQS_ALL_DISABLED);
82 
83 	/*
84 	 * We are responding to the next interrupt, so interrupt-off
85 	 * latencies should be reset here.
86 	 */
87 	trace_hardirqs_on();
88 	trace_hardirqs_off();
89 }
90 
91 static inline bool irq_happened_test_and_clear(u8 irq)
92 {
93 	if (local_paca->irq_happened & irq) {
94 		local_paca->irq_happened &= ~irq;
95 		return true;
96 	}
97 	return false;
98 }
99 
100 void replay_soft_interrupts(void)
101 {
102 	struct pt_regs regs;
103 
104 	/*
105 	 * Be careful here, calling these interrupt handlers can cause
106 	 * softirqs to be raised, which they may run when calling irq_exit,
107 	 * which will cause local_irq_enable() to be run, which can then
108 	 * recurse into this function. Don't keep any state across
109 	 * interrupt handler calls which may change underneath us.
110 	 *
111 	 * Softirqs can not be disabled over replay to stop this recursion
112 	 * because interrupts taken in idle code may require RCU softirq
113 	 * to run in the irq RCU tracking context. This is a hard problem
114 	 * to fix without changes to the softirq or idle layer.
115 	 *
116 	 * We use local_paca rather than get_paca() to avoid all the
117 	 * debug_smp_processor_id() business in this low level function.
118 	 */
119 
120 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
121 		WARN_ON_ONCE(mfmsr() & MSR_EE);
122 		WARN_ON(!(local_paca->irq_happened & PACA_IRQ_HARD_DIS));
123 	}
124 
125 	ppc_save_regs(&regs);
126 	regs.softe = IRQS_ENABLED;
127 	regs.msr |= MSR_EE;
128 
129 again:
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) &&
145 	    irq_happened_test_and_clear(PACA_IRQ_HMI)) {
146 		regs.trap = INTERRUPT_HMI;
147 		handle_hmi_exception(&regs);
148 		next_interrupt(&regs);
149 	}
150 
151 	if (irq_happened_test_and_clear(PACA_IRQ_DEC)) {
152 		regs.trap = INTERRUPT_DECREMENTER;
153 		timer_interrupt(&regs);
154 		next_interrupt(&regs);
155 	}
156 
157 	if (irq_happened_test_and_clear(PACA_IRQ_EE)) {
158 		regs.trap = INTERRUPT_EXTERNAL;
159 		do_IRQ(&regs);
160 		next_interrupt(&regs);
161 	}
162 
163 	if (IS_ENABLED(CONFIG_PPC_DOORBELL) &&
164 	    irq_happened_test_and_clear(PACA_IRQ_DBELL)) {
165 		regs.trap = INTERRUPT_DOORBELL;
166 		doorbell_exception(&regs);
167 		next_interrupt(&regs);
168 	}
169 
170 	/* Book3E does not support soft-masking PMI interrupts */
171 	if (IS_ENABLED(CONFIG_PPC_BOOK3S) &&
172 	    irq_happened_test_and_clear(PACA_IRQ_PMI)) {
173 		regs.trap = INTERRUPT_PERFMON;
174 		performance_monitor_exception(&regs);
175 		next_interrupt(&regs);
176 	}
177 
178 	/*
179 	 * Softirq processing can enable and disable interrupts, which can
180 	 * result in new irqs becoming pending. Must keep looping until we
181 	 * have cleared out all pending interrupts.
182 	 */
183 	if (local_paca->irq_happened & ~PACA_IRQ_HARD_DIS)
184 		goto again;
185 }
186 
187 #if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_KUAP)
188 static inline void replay_soft_interrupts_irqrestore(void)
189 {
190 	unsigned long kuap_state = get_kuap();
191 
192 	/*
193 	 * Check if anything calls local_irq_enable/restore() when KUAP is
194 	 * disabled (user access enabled). We handle that case here by saving
195 	 * and re-locking AMR but we shouldn't get here in the first place,
196 	 * hence the warning.
197 	 */
198 	kuap_assert_locked();
199 
200 	if (kuap_state != AMR_KUAP_BLOCKED)
201 		set_kuap(AMR_KUAP_BLOCKED);
202 
203 	replay_soft_interrupts();
204 
205 	if (kuap_state != AMR_KUAP_BLOCKED)
206 		set_kuap(kuap_state);
207 }
208 #else
209 #define replay_soft_interrupts_irqrestore() replay_soft_interrupts()
210 #endif
211 
212 notrace void arch_local_irq_restore(unsigned long mask)
213 {
214 	unsigned char irq_happened;
215 
216 	/* Write the new soft-enabled value if it is a disable */
217 	if (mask) {
218 		irq_soft_mask_set(mask);
219 		return;
220 	}
221 
222 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
223 		WARN_ON_ONCE(in_nmi() || in_hardirq());
224 
225 	/*
226 	 * After the stb, interrupts are unmasked and there are no interrupts
227 	 * pending replay. The restart sequence makes this atomic with
228 	 * respect to soft-masked interrupts. If this was just a simple code
229 	 * sequence, a soft-masked interrupt could become pending right after
230 	 * the comparison and before the stb.
231 	 *
232 	 * This allows interrupts to be unmasked without hard disabling, and
233 	 * also without new hard interrupts coming in ahead of pending ones.
234 	 */
235 	asm_volatile_goto(
236 "1:					\n"
237 "		lbz	9,%0(13)	\n"
238 "		cmpwi	9,0		\n"
239 "		bne	%l[happened]	\n"
240 "		stb	9,%1(13)	\n"
241 "2:					\n"
242 		RESTART_TABLE(1b, 2b, 1b)
243 	: : "i" (offsetof(struct paca_struct, irq_happened)),
244 	    "i" (offsetof(struct paca_struct, irq_soft_mask))
245 	: "cr0", "r9"
246 	: happened);
247 
248 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
249 		WARN_ON_ONCE(!(mfmsr() & MSR_EE));
250 
251 	return;
252 
253 happened:
254 	irq_happened = READ_ONCE(local_paca->irq_happened);
255 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
256 		WARN_ON_ONCE(!irq_happened);
257 
258 	if (irq_happened == PACA_IRQ_HARD_DIS) {
259 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
260 			WARN_ON_ONCE(mfmsr() & MSR_EE);
261 		irq_soft_mask_set(IRQS_ENABLED);
262 		local_paca->irq_happened = 0;
263 		__hard_irq_enable();
264 		return;
265 	}
266 
267 	/* Have interrupts to replay, need to hard disable first */
268 	if (!(irq_happened & PACA_IRQ_HARD_DIS)) {
269 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
270 			if (!(mfmsr() & MSR_EE)) {
271 				/*
272 				 * An interrupt could have come in and cleared
273 				 * MSR[EE] and set IRQ_HARD_DIS, so check
274 				 * IRQ_HARD_DIS again and warn if it is still
275 				 * clear.
276 				 */
277 				irq_happened = READ_ONCE(local_paca->irq_happened);
278 				WARN_ON_ONCE(!(irq_happened & PACA_IRQ_HARD_DIS));
279 			}
280 		}
281 		__hard_irq_disable();
282 		local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
283 	} else {
284 		if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG)) {
285 			if (WARN_ON_ONCE(mfmsr() & MSR_EE))
286 				__hard_irq_disable();
287 		}
288 	}
289 
290 	/*
291 	 * Disable preempt here, so that the below preempt_enable will
292 	 * perform resched if required (a replayed interrupt may set
293 	 * need_resched).
294 	 */
295 	preempt_disable();
296 	irq_soft_mask_set(IRQS_ALL_DISABLED);
297 	trace_hardirqs_off();
298 
299 	replay_soft_interrupts_irqrestore();
300 
301 	trace_hardirqs_on();
302 	irq_soft_mask_set(IRQS_ENABLED);
303 	if (IS_ENABLED(CONFIG_PPC_IRQ_SOFT_MASK_DEBUG))
304 		WARN_ON(local_paca->irq_happened != PACA_IRQ_HARD_DIS);
305 	local_paca->irq_happened = 0;
306 	__hard_irq_enable();
307 	preempt_enable();
308 }
309 EXPORT_SYMBOL(arch_local_irq_restore);
310 
311 /*
312  * This is a helper to use when about to go into idle low-power
313  * when the latter has the side effect of re-enabling interrupts
314  * (such as calling H_CEDE under pHyp).
315  *
316  * You call this function with interrupts soft-disabled (this is
317  * already the case when ppc_md.power_save is called). The function
318  * will return whether to enter power save or just return.
319  *
320  * In the former case, it will have notified lockdep of interrupts
321  * being re-enabled and generally sanitized the lazy irq state,
322  * and in the latter case it will leave with interrupts hard
323  * disabled and marked as such, so the local_irq_enable() call
324  * in arch_cpu_idle() will properly re-enable everything.
325  */
326 bool prep_irq_for_idle(void)
327 {
328 	/*
329 	 * First we need to hard disable to ensure no interrupt
330 	 * occurs before we effectively enter the low power state
331 	 */
332 	__hard_irq_disable();
333 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
334 
335 	/*
336 	 * If anything happened while we were soft-disabled,
337 	 * we return now and do not enter the low power state.
338 	 */
339 	if (lazy_irq_pending())
340 		return false;
341 
342 	/* Tell lockdep we are about to re-enable */
343 	trace_hardirqs_on();
344 
345 	/*
346 	 * Mark interrupts as soft-enabled and clear the
347 	 * PACA_IRQ_HARD_DIS from the pending mask since we
348 	 * are about to hard enable as well as a side effect
349 	 * of entering the low power state.
350 	 */
351 	local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
352 	irq_soft_mask_set(IRQS_ENABLED);
353 
354 	/* Tell the caller to enter the low power state */
355 	return true;
356 }
357 
358 #ifdef CONFIG_PPC_BOOK3S
359 /*
360  * This is for idle sequences that return with IRQs off, but the
361  * idle state itself wakes on interrupt. Tell the irq tracer that
362  * IRQs are enabled for the duration of idle so it does not get long
363  * off times. Must be paired with fini_irq_for_idle_irqsoff.
364  */
365 bool prep_irq_for_idle_irqsoff(void)
366 {
367 	WARN_ON(!irqs_disabled());
368 
369 	/*
370 	 * First we need to hard disable to ensure no interrupt
371 	 * occurs before we effectively enter the low power state
372 	 */
373 	__hard_irq_disable();
374 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
375 
376 	/*
377 	 * If anything happened while we were soft-disabled,
378 	 * we return now and do not enter the low power state.
379 	 */
380 	if (lazy_irq_pending())
381 		return false;
382 
383 	/* Tell lockdep we are about to re-enable */
384 	trace_hardirqs_on();
385 
386 	return true;
387 }
388 
389 /*
390  * Take the SRR1 wakeup reason, index into this table to find the
391  * appropriate irq_happened bit.
392  *
393  * Sytem reset exceptions taken in idle state also come through here,
394  * but they are NMI interrupts so do not need to wait for IRQs to be
395  * restored, and should be taken as early as practical. These are marked
396  * with 0xff in the table. The Power ISA specifies 0100b as the system
397  * reset interrupt reason.
398  */
399 #define IRQ_SYSTEM_RESET	0xff
400 
401 static const u8 srr1_to_lazyirq[0x10] = {
402 	0, 0, 0,
403 	PACA_IRQ_DBELL,
404 	IRQ_SYSTEM_RESET,
405 	PACA_IRQ_DBELL,
406 	PACA_IRQ_DEC,
407 	0,
408 	PACA_IRQ_EE,
409 	PACA_IRQ_EE,
410 	PACA_IRQ_HMI,
411 	0, 0, 0, 0, 0 };
412 
413 void replay_system_reset(void)
414 {
415 	struct pt_regs regs;
416 
417 	ppc_save_regs(&regs);
418 	regs.trap = 0x100;
419 	get_paca()->in_nmi = 1;
420 	system_reset_exception(&regs);
421 	get_paca()->in_nmi = 0;
422 }
423 EXPORT_SYMBOL_GPL(replay_system_reset);
424 
425 void irq_set_pending_from_srr1(unsigned long srr1)
426 {
427 	unsigned int idx = (srr1 & SRR1_WAKEMASK_P8) >> 18;
428 	u8 reason = srr1_to_lazyirq[idx];
429 
430 	/*
431 	 * Take the system reset now, which is immediately after registers
432 	 * are restored from idle. It's an NMI, so interrupts need not be
433 	 * re-enabled before it is taken.
434 	 */
435 	if (unlikely(reason == IRQ_SYSTEM_RESET)) {
436 		replay_system_reset();
437 		return;
438 	}
439 
440 	if (reason == PACA_IRQ_DBELL) {
441 		/*
442 		 * When doorbell triggers a system reset wakeup, the message
443 		 * is not cleared, so if the doorbell interrupt is replayed
444 		 * and the IPI handled, the doorbell interrupt would still
445 		 * fire when EE is enabled.
446 		 *
447 		 * To avoid taking the superfluous doorbell interrupt,
448 		 * execute a msgclr here before the interrupt is replayed.
449 		 */
450 		ppc_msgclr(PPC_DBELL_MSGTYPE);
451 	}
452 
453 	/*
454 	 * The 0 index (SRR1[42:45]=b0000) must always evaluate to 0,
455 	 * so this can be called unconditionally with the SRR1 wake
456 	 * reason as returned by the idle code, which uses 0 to mean no
457 	 * interrupt.
458 	 *
459 	 * If a future CPU was to designate this as an interrupt reason,
460 	 * then a new index for no interrupt must be assigned.
461 	 */
462 	local_paca->irq_happened |= reason;
463 }
464 #endif /* CONFIG_PPC_BOOK3S */
465 
466 /*
467  * Force a replay of the external interrupt handler on this CPU.
468  */
469 void force_external_irq_replay(void)
470 {
471 	/*
472 	 * This must only be called with interrupts soft-disabled,
473 	 * the replay will happen when re-enabling.
474 	 */
475 	WARN_ON(!arch_irqs_disabled());
476 
477 	/*
478 	 * Interrupts must always be hard disabled before irq_happened is
479 	 * modified (to prevent lost update in case of interrupt between
480 	 * load and store).
481 	 */
482 	__hard_irq_disable();
483 	local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
484 
485 	/* Indicate in the PACA that we have an interrupt to replay */
486 	local_paca->irq_happened |= PACA_IRQ_EE;
487 }
488 
489 static int __init setup_noirqdistrib(char *str)
490 {
491 	distribute_irqs = 0;
492 	return 1;
493 }
494 
495 __setup("noirqdistrib", setup_noirqdistrib);
496