xref: /xnu-12377.61.12/osfmk/arm64/cpu.c (revision 4d495c6e23c53686cf65f45067f79024cf5dcee8)
1 /*
2  * Copyright (c) 2007-2021 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 /*
29  *	File:	arm64/cpu.c
30  *
31  *	cpu specific routines
32  */
33 
34 #include <pexpert/arm64/board_config.h>
35 #include <kern/kalloc.h>
36 #include <kern/machine.h>
37 #include <kern/cpu_number.h>
38 #include <kern/percpu.h>
39 #include <kern/thread.h>
40 #include <kern/timer_queue.h>
41 #include <arm/cpu_data.h>
42 #include <arm/cpuid.h>
43 #include <arm/caches_internal.h>
44 #include <arm/cpu_data_internal.h>
45 #include <arm/cpu_internal.h>
46 #include <arm/misc_protos.h>
47 #include <arm/machine_cpu.h>
48 #include <arm/rtclock.h>
49 #include <arm64/proc_reg.h>
50 #include <mach/processor_info.h>
51 #include <vm/pmap.h>
52 #include <vm/vm_kern_xnu.h>
53 #include <vm/vm_map.h>
54 #include <pexpert/arm/protos.h>
55 #include <pexpert/device_tree.h>
56 #include <sys/kdebug.h>
57 #include <arm/machine_routines.h>
58 
59 #include <machine/atomic.h>
60 
61 #include <san/kasan.h>
62 
63 #include <kern/kpc.h>
64 #if CONFIG_CPU_COUNTERS
65 #include <kern/monotonic.h>
66 #endif /* CONFIG_CPU_COUNTERS */
67 
68 #if KPERF
69 #include <kperf/kptimer.h>
70 #endif /* KPERF */
71 
72 #if HIBERNATION
73 #include <IOKit/IOPlatformExpert.h>
74 #include <IOKit/IOHibernatePrivate.h>
75 #endif /* HIBERNATION */
76 
77 
78 #include <libkern/section_keywords.h>
79 
80 extern boolean_t        idle_enable;
81 extern uint64_t         wake_abstime;
82 
83 #if WITH_CLASSIC_S2R
84 void sleep_token_buffer_init(void);
85 #endif
86 
87 #if !CONFIG_SPTM
88 extern uintptr_t resume_idle_cpu;
89 extern uintptr_t start_cpu;
90 vm_address_t   start_cpu_paddr;
91 #endif
92 
93 #if __ARM_KERNEL_PROTECT__
94 extern void exc_vectors_table;
95 #endif /* __ARM_KERNEL_PROTECT__ */
96 
97 #if APPLEVIRTUALPLATFORM
98 extern vm_offset_t reset_vector_vaddr;
99 #endif /* APPLEVIRTUALPLATFORM */
100 
101 #if APPLEVIRTUALPLATFORM
102 extern void __attribute__((noreturn)) arm64_prepare_for_sleep(boolean_t deep_sleep, unsigned int cpu, uint64_t entry_pa);
103 #else
104 extern void __attribute__((noreturn)) arm64_prepare_for_sleep(boolean_t deep_sleep);
105 #endif
106 extern void arm64_force_wfi_clock_gate(void);
107 #if defined(APPLETYPHOON)
108 // <rdar://problem/15827409>
109 extern void typhoon_prepare_for_wfi(void);
110 extern void typhoon_return_from_wfi(void);
111 #endif
112 
113 #if HAS_RETENTION_STATE
114 extern void arm64_retention_wfi(void);
115 #endif
116 
117 // wfi - wfi mode
118 //  0 : disabled
119 //  1 : normal
120 //  2 : overhead simulation (delay & flags)
121 TUNABLE(unsigned int, wfi, "wfi", 1);
122 #if DEVELOPMENT || DEBUG
123 
124 // wfi_flags
125 //  1 << 0 : flush L1s
126 //  1 << 1 : flush TLBs
127 static int wfi_flags = 0;
128 
129 // wfi_delay - delay ticks after wfi exit
130 static uint64_t wfi_delay = 0;
131 
132 #endif /* DEVELOPMENT || DEBUG */
133 
134 #define CPUPM_IDLE_WFE 0x5310300
135 #define CPUPM_IDLE_TIMER_WFE 0x5310304
136 
137 #define DEFAULT_EXPECTING_IPI_WFE_TIMEOUT_USEC (60ULL)
138 TUNABLE(uint32_t, expecting_ipi_wfe_timeout_usec,
139     "expecting_ipi_wfe_timeout_usec", DEFAULT_EXPECTING_IPI_WFE_TIMEOUT_USEC);
140 uint64_t expecting_ipi_wfe_timeout_mt = 0x0ULL; /* initialized to a non-zero value in sched_init */
141 
142 /* When recommended, issue WFE with [FI]IRQ unmasked in the idle
143  * loop. The default.
144  */
145 uint32_t idle_proximate_io_wfe_unmasked = 1;
146 #if DEVELOPMENT || DEBUG
147 uint32_t idle_proximate_timer_wfe = 1;
148 uint32_t idle_proximate_io_wfe_masked = 0;
149 #else
150 /* Issue WFE in lieu of WFI when awaiting a proximate timer. */
151 static uint32_t idle_proximate_timer_wfe = 1;
152 /* When recommended, issue WFE with [FI]IRQ masked in the idle loop.
153  * Non-default, retained for experimentation.
154  */
155 static uint32_t idle_proximate_io_wfe_masked = 0;
156 #endif
157 
158 #if __ARM_GLOBAL_SLEEP_BIT__
159 volatile boolean_t arm64_stall_sleep = TRUE;
160 #endif
161 
162 #if WITH_CLASSIC_S2R
163 /*
164  * These must be aligned to avoid issues with calling bcopy_phys on them before
165  * we are done with pmap initialization.
166  */
167 static const uint8_t __attribute__ ((aligned(8))) suspend_signature[] = {'X', 'S', 'O', 'M', 'P', 'S', 'U', 'S'};
168 static const uint8_t __attribute__ ((aligned(8))) running_signature[] = {'X', 'S', 'O', 'M', 'N', 'N', 'U', 'R'};
169 #endif
170 
171 #if WITH_CLASSIC_S2R
172 static vm_offset_t sleepTokenBuffer = (vm_offset_t)NULL;
173 #endif
174 static boolean_t coresight_debug_enabled = FALSE;
175 
176 #if defined(CONFIG_XNUPOST)
177 void arm64_ipi_test_callback(void *);
178 void arm64_immediate_ipi_test_callback(void *);
179 
180 void
arm64_ipi_test_callback(void * parm)181 arm64_ipi_test_callback(void *parm)
182 {
183 	volatile uint64_t *ipi_test_data = parm;
184 	cpu_data_t *cpu_data;
185 
186 	cpu_data = getCpuDatap();
187 
188 	*ipi_test_data = cpu_data->cpu_number;
189 }
190 
191 void
arm64_immediate_ipi_test_callback(void * parm)192 arm64_immediate_ipi_test_callback(void *parm)
193 {
194 	volatile uint64_t *ipi_test_data = parm;
195 	cpu_data_t *cpu_data;
196 
197 	cpu_data = getCpuDatap();
198 
199 	*ipi_test_data = cpu_data->cpu_number + MAX_CPUS;
200 }
201 
202 uint64_t arm64_ipi_test_data[MAX_CPUS * 2];
203 
204 MACHINE_TIMEOUT(arm64_ipi_test_timeout, "arm64-ipi-test", 100, MACHINE_TIMEOUT_UNIT_MSEC, NULL);
205 
206 void
arm64_ipi_test()207 arm64_ipi_test()
208 {
209 	volatile uint64_t *ipi_test_data, *immediate_ipi_test_data;
210 	uint64_t timeout_ms = os_atomic_load(&arm64_ipi_test_timeout, relaxed);
211 	uint64_t then, now, delta;
212 	int current_cpu_number = getCpuDatap()->cpu_number;
213 
214 	/*
215 	 * probably the only way to have this on most systems is with the
216 	 * cpus=1 boot-arg, but nonetheless, if we only have 1 CPU active,
217 	 * IPI is not available
218 	 */
219 	if (real_ncpus == 1) {
220 		return;
221 	}
222 
223 	const unsigned int max_cpu_id = ml_get_max_cpu_number();
224 	for (unsigned int i = 0; i <= max_cpu_id; ++i) {
225 		ipi_test_data = &arm64_ipi_test_data[i];
226 		immediate_ipi_test_data = &arm64_ipi_test_data[i + MAX_CPUS];
227 		*ipi_test_data = ~i;
228 		kern_return_t error = cpu_xcall((int)i, (void *)arm64_ipi_test_callback, (void *)(uintptr_t)ipi_test_data);
229 		if (error != KERN_SUCCESS) {
230 			panic("CPU %d was unable to IPI CPU %u: error %d", current_cpu_number, i, error);
231 		}
232 
233 		while ((error = cpu_immediate_xcall((int)i, (void *)arm64_immediate_ipi_test_callback,
234 		    (void *)(uintptr_t)immediate_ipi_test_data)) == KERN_ALREADY_WAITING) {
235 			now = mach_absolute_time();
236 			absolutetime_to_nanoseconds(now - then, &delta);
237 			if ((delta / NSEC_PER_MSEC) > timeout_ms) {
238 				panic("CPU %d was unable to immediate-IPI CPU %u within %lldms", current_cpu_number, i, timeout_ms);
239 			}
240 		}
241 
242 		if (error != KERN_SUCCESS) {
243 			panic("CPU %d was unable to immediate-IPI CPU %u: error %d", current_cpu_number, i, error);
244 		}
245 
246 		then = mach_absolute_time();
247 
248 		while ((*ipi_test_data != i) || (*immediate_ipi_test_data != (i + MAX_CPUS))) {
249 			now = mach_absolute_time();
250 			absolutetime_to_nanoseconds(now - then, &delta);
251 			if ((delta / NSEC_PER_MSEC) > timeout_ms) {
252 				panic("CPU %d tried to IPI CPU %d but didn't get correct responses within %lldms, responses: %llx, %llx",
253 				    current_cpu_number, i, timeout_ms, *ipi_test_data, *immediate_ipi_test_data);
254 			}
255 		}
256 	}
257 }
258 #endif /* defined(CONFIG_XNUPOST) */
259 
260 static void
configure_coresight_registers(cpu_data_t * cdp)261 configure_coresight_registers(cpu_data_t *cdp)
262 {
263 	int             i;
264 
265 	assert(cdp);
266 	vm_offset_t     coresight_regs = ml_get_topology_info()->cpus[cdp->cpu_number].coresight_regs;
267 
268 	/*
269 	 * ARMv8 coresight registers are optional. If the device tree did not
270 	 * provide either cpu_regmap_paddr (from the legacy "reg-private" EDT property)
271 	 * or coresight_regs (from the new "coresight-reg" property), assume that
272 	 * coresight registers are not supported.
273 	 */
274 	if (cdp->cpu_regmap_paddr || coresight_regs) {
275 		for (i = 0; i < CORESIGHT_REGIONS; ++i) {
276 			if (i == CORESIGHT_CTI || i == CORESIGHT_PMU) {
277 				continue;
278 			}
279 			/* Skip debug-only registers on production chips */
280 			if (((i == CORESIGHT_ED) || (i == CORESIGHT_UTT)) && !coresight_debug_enabled) {
281 				continue;
282 			}
283 
284 			if (!cdp->coresight_base[i]) {
285 				if (coresight_regs) {
286 					cdp->coresight_base[i] = coresight_regs + CORESIGHT_OFFSET(i);
287 				} else {
288 					uint64_t addr = cdp->cpu_regmap_paddr + CORESIGHT_OFFSET(i);
289 					cdp->coresight_base[i] = (vm_offset_t)ml_io_map(addr, CORESIGHT_SIZE);
290 				}
291 			}
292 			/* Unlock EDLAR (CTI and PMU are skipped above). */
293 			if (i != CORESIGHT_UTT) {
294 				*(volatile uint32_t *)(cdp->coresight_base[i] + ARM_DEBUG_OFFSET_DBGLAR) = ARM_DBG_LOCK_ACCESS_KEY;
295 			}
296 		}
297 	}
298 }
299 
300 
301 /*
302  *	Routine:	cpu_bootstrap
303  *	Function:
304  */
305 void
cpu_bootstrap(void)306 cpu_bootstrap(void)
307 {
308 }
309 
310 /*
311  *	Routine:	cpu_sleep
312  *	Function:
313  */
314 void
cpu_sleep(void)315 cpu_sleep(void)
316 {
317 	cpu_data_t     *cpu_data_ptr = getCpuDatap();
318 
319 	cpu_data_ptr->cpu_active_thread = current_thread();
320 #if CONFIG_SPTM
321 	cpu_data_ptr->cpu_reset_handler = (uintptr_t) VM_KERNEL_STRIP_PTR(arm_init_cpu);
322 #else
323 	cpu_data_ptr->cpu_reset_handler = (uintptr_t) start_cpu_paddr;
324 #endif
325 	os_atomic_or(&cpu_data_ptr->cpu_flags, SleepState, relaxed);
326 
327 	if (cpu_data_ptr->cpu_user_debug != NULL) {
328 		arm_debug_set(NULL);
329 	}
330 
331 #if CONFIG_CPU_COUNTERS
332 	kpc_idle();
333 	mt_cpu_down(cpu_data_ptr);
334 #endif /* CONFIG_CPU_COUNTERS */
335 #if KPERF
336 	kptimer_stop_curcpu();
337 #endif /* KPERF */
338 
339 	CleanPoC_Dcache();
340 
341 #if USE_APPLEARMSMP
342 	if (ml_is_quiescing()) {
343 		PE_cpu_machine_quiesce(cpu_data_ptr->cpu_id);
344 	} else {
345 		bool deep_sleep = PE_cpu_down(cpu_data_ptr->cpu_id);
346 		cpu_data_ptr->cpu_sleep_token = ARM_CPU_ON_SLEEP_PATH;
347 		// hang CPU on spurious wakeup
348 		cpu_data_ptr->cpu_reset_handler = (uintptr_t)0;
349 		__builtin_arm_dsb(DSB_ISH);
350 		CleanPoU_Dcache();
351 #if APPLEVIRTUALPLATFORM
352 		arm64_prepare_for_sleep(deep_sleep, cpu_data_ptr->cpu_number, ml_vtophys(reset_vector_vaddr));
353 #else /* APPLEVIRTUALPLATFORM */
354 		arm64_prepare_for_sleep(deep_sleep);
355 #endif /* APPLEVIRTUALPLATFORM */
356 	}
357 #else
358 	PE_cpu_machine_quiesce(cpu_data_ptr->cpu_id);
359 #endif
360 	/*NOTREACHED*/
361 }
362 
363 /*
364  *	Routine:	cpu_interrupt_is_pending
365  *	Function:	Returns a bool signifying a non-zero ISR_EL1,
366  *			indicating a pending IRQ, FIQ or external abort.
367  */
368 
369 bool
cpu_interrupt_is_pending(void)370 cpu_interrupt_is_pending(void)
371 {
372 	uint64_t isr_value;
373 	isr_value = __builtin_arm_rsr64("ISR_EL1");
374 	return isr_value != 0;
375 }
376 
377 static bool
cpu_proximate_timer(void)378 cpu_proximate_timer(void)
379 {
380 	return !SetIdlePop();
381 }
382 
383 #ifdef ARM64_BOARD_CONFIG_T6000
384 int wfe_allowed = 0;
385 #else
386 int wfe_allowed = 1;
387 #endif /* ARM64_BOARD_CONFIG_T6000 */
388 
389 #if DEVELOPMENT || DEBUG
390 #define WFE_STAT(x)     \
391 	do {            \
392 	        (x);    \
393 	} while(0)
394 #else
395 #define WFE_STAT(x)     do {} while(0)
396 #endif /* DEVELOPMENT || DEBUG */
397 
398 bool
wfe_to_deadline_or_interrupt(uint32_t cid,uint64_t wfe_deadline,cpu_data_t * cdp,bool unmask,bool check_cluster_recommendation)399 wfe_to_deadline_or_interrupt(uint32_t cid, uint64_t wfe_deadline, cpu_data_t *cdp, bool unmask, bool check_cluster_recommendation)
400 {
401 	bool ipending = false;
402 	uint64_t irqc = 0, nirqc = 0;
403 
404 	/* The ARMv8 architecture permits a processor dwelling in WFE
405 	 * with F/IRQ masked to ignore a pending interrupt, i.e.
406 	 * not classify it as an 'event'. This is potentially
407 	 * problematic with AICv2's IRQ distribution model, as
408 	 * a transient interrupt masked interval can cause an SIQ
409 	 * query rejection, possibly routing the interrupt to
410 	 * another core/cluster in a powergated state.
411 	 * Hence, optionally unmask IRQs+FIQs across WFE.
412 	 */
413 	if (unmask) {
414 		/* Latch SW IRQ+FIQ counter prior to unmasking
415 		 * interrupts.
416 		 */
417 		irqc = nirqc = os_atomic_load(&cdp->cpu_stat.irq_ex_cnt_wake, relaxed);
418 		__builtin_arm_wsr("DAIFClr", DAIFSC_STANDARD_DISABLE);
419 	}
420 
421 	while ((ipending = (cpu_interrupt_is_pending())) == false) {
422 		if (unmask) {
423 			/* If WFE was issued with IRQs unmasked, an
424 			 * interrupt may have been processed.
425 			 * Consult the SW IRQ counter to determine
426 			 * whether the 'idle loop' must be
427 			 * re-evaluated.
428 			 */
429 			nirqc = os_atomic_load(&cdp->cpu_stat.irq_ex_cnt_wake, relaxed);
430 			if (nirqc != irqc) {
431 				break;
432 			}
433 		}
434 
435 		if (__probable(wfe_allowed)) {
436 			/*
437 			 * If IRQs are unmasked, there's a small window
438 			 * where an 'extra' WFE may be issued after
439 			 * the consultation of the SW interrupt counter
440 			 * and new interrupt arrival. Hence this WFE
441 			 * relies on the [FI]RQ interrupt handler
442 			 * epilogue issuing a 'SEVL', to post an
443 			 * event which causes the next WFE on the same
444 			 * PE to retire immediately.
445 			 */
446 
447 			__builtin_arm_wfe();
448 		}
449 
450 		WFE_STAT(cdp->wfe_count++);
451 		if (wfe_deadline != ~0ULL) {
452 			WFE_STAT(cdp->wfe_deadline_checks++);
453 			/* Check if the WFE recommendation has expired.
454 			 * We do not recompute the deadline here.
455 			 */
456 			if ((check_cluster_recommendation && ml_cluster_wfe_timeout(cid) == 0) ||
457 			    mach_absolute_time() >= wfe_deadline) {
458 				WFE_STAT(cdp->wfe_terminations++);
459 				break;
460 			}
461 		}
462 	}
463 
464 	if (unmask) {
465 		__builtin_arm_wsr64("DAIFSet", DAIFSC_STANDARD_DISABLE);
466 		/* Refetch SW interrupt counter with IRQs masked
467 		 * It is important that this routine accurately flags
468 		 * any observed interrupts via its return value,
469 		 * inaccuracy may lead to an erroneous WFI fallback.
470 		 */
471 		nirqc = os_atomic_load(&cdp->cpu_stat.irq_ex_cnt_wake, relaxed);
472 	}
473 
474 	return ipending || (nirqc != irqc);
475 }
476 
477 /*
478  *	Routine:	cpu_idle
479  *	Function:
480  */
481 void __attribute__((noreturn))
cpu_idle(void)482 cpu_idle(void)
483 {
484 	cpu_data_t     *cpu_data_ptr = getCpuDatap();
485 	processor_t     processor = current_processor();
486 	uint64_t        new_idle_timeout_ticks = 0x0ULL, lastPop;
487 	bool idle_disallowed = false;
488 	/* Read and reset the next_idle_short flag */
489 	bool next_idle_short = processor->next_idle_short;
490 	processor->next_idle_short = false;
491 
492 	if (__improbable((!idle_enable))) {
493 		idle_disallowed = true;
494 	} else if (__improbable(cpu_data_ptr->cpu_signal & SIGPdisabled)) {
495 		idle_disallowed = true;
496 	}
497 
498 	if (__improbable(idle_disallowed)) {
499 		Idle_load_context();
500 	}
501 
502 	bool ipending = false;
503 	uint32_t cid = cpu_data_ptr->cpu_cluster_id;
504 	bool check_cluster_recommendation = true;
505 	uint64_t wfe_timeout = 0;
506 
507 	if (idle_proximate_io_wfe_masked == 1) {
508 		/* Check for an active perf. controller generated
509 		 * WFE recommendation for this cluster.
510 		 */
511 		wfe_timeout = ml_cluster_wfe_timeout(cid);
512 	}
513 
514 	if (next_idle_short && expecting_ipi_wfe_timeout_mt > wfe_timeout) {
515 		/* In this case we should WFE because a response IPI
516 		 * is expected soon.
517 		 */
518 		wfe_timeout = expecting_ipi_wfe_timeout_mt;
519 		check_cluster_recommendation = false;
520 	}
521 
522 	if (wfe_timeout != 0) {
523 		uint64_t wfe_deadline = mach_absolute_time() + wfe_timeout;
524 		/* Poll issuing event-bounded WFEs until an interrupt
525 		 * arrives or the WFE recommendation expires
526 		 */
527 		KDBG(CPUPM_IDLE_WFE | DBG_FUNC_START, ipending, cpu_data_ptr->wfe_count, wfe_timeout, !check_cluster_recommendation);
528 		ipending = wfe_to_deadline_or_interrupt(cid, wfe_deadline, cpu_data_ptr, false, check_cluster_recommendation);
529 		KDBG(CPUPM_IDLE_WFE | DBG_FUNC_END, ipending, cpu_data_ptr->wfe_count, wfe_deadline);
530 		if (ipending == true) {
531 			/* Back to machine_idle() */
532 			Idle_load_context();
533 		}
534 	}
535 
536 	if (__improbable(cpu_proximate_timer())) {
537 		if (idle_proximate_timer_wfe == 1) {
538 			/* Poll issuing WFEs until the expected
539 			 * timer FIQ arrives.
540 			 */
541 			KDBG(CPUPM_IDLE_TIMER_WFE | DBG_FUNC_START, ipending, cpu_data_ptr->wfe_count, ~0ULL);
542 			ipending = wfe_to_deadline_or_interrupt(cid, ~0ULL, cpu_data_ptr, false, false);
543 			KDBG(CPUPM_IDLE_TIMER_WFE | DBG_FUNC_END, ipending, cpu_data_ptr->wfe_count, ~0ULL);
544 			assert(ipending == true);
545 		}
546 		Idle_load_context();
547 	}
548 
549 	lastPop = cpu_data_ptr->rtcPop;
550 
551 	cpu_data_ptr->cpu_active_thread = current_thread();
552 
553 	if (wfi && (cpu_data_ptr->cpu_idle_notify != NULL)) {
554 		cpu_data_ptr->cpu_idle_notify(cpu_data_ptr->cpu_id, TRUE, &new_idle_timeout_ticks);
555 	}
556 
557 	if (cpu_data_ptr->idle_timer_notify != NULL) {
558 		if (new_idle_timeout_ticks == 0x0ULL) {
559 			/* turn off the idle timer */
560 			cpu_data_ptr->idle_timer_deadline = 0x0ULL;
561 		} else {
562 			/* set the new idle timeout */
563 			clock_absolutetime_interval_to_deadline(new_idle_timeout_ticks, &cpu_data_ptr->idle_timer_deadline);
564 		}
565 		timer_resync_deadlines();
566 		if (cpu_data_ptr->rtcPop != lastPop) {
567 			SetIdlePop();
568 		}
569 	}
570 
571 #if CONFIG_CPU_COUNTERS
572 	kpc_idle();
573 	mt_cpu_idle(cpu_data_ptr);
574 #endif /* CONFIG_CPU_COUNTERS */
575 
576 	if (wfi) {
577 #if !defined(APPLE_ARM64_ARCH_FAMILY)
578 		platform_cache_idle_enter();
579 #endif
580 
581 #if DEVELOPMENT || DEBUG
582 		// When simulating wfi overhead,
583 		// force wfi to clock gating only
584 		if (wfi == 2) {
585 			arm64_force_wfi_clock_gate();
586 		}
587 #endif /* DEVELOPMENT || DEBUG */
588 
589 #if defined(APPLETYPHOON)
590 		// <rdar://problem/15827409> CPU1 Stuck in WFIWT Because of MMU Prefetch
591 		typhoon_prepare_for_wfi();
592 #endif
593 		__builtin_arm_dsb(DSB_SY);
594 #if HAS_RETENTION_STATE
595 		arm64_retention_wfi();
596 #else
597 		__builtin_arm_wfi();
598 #endif
599 
600 #if defined(APPLETYPHOON)
601 		// <rdar://problem/15827409> CPU1 Stuck in WFIWT Because of MMU Prefetch
602 		typhoon_return_from_wfi();
603 #endif
604 
605 #if DEVELOPMENT || DEBUG
606 		// Handle wfi overhead simulation
607 		if (wfi == 2) {
608 			uint64_t deadline;
609 
610 			// Calculate wfi delay deadline
611 			clock_absolutetime_interval_to_deadline(wfi_delay, &deadline);
612 
613 			// Flush L1 caches
614 			if ((wfi_flags & 1) != 0) {
615 				InvalidatePoU_Icache();
616 				FlushPoC_Dcache();
617 			}
618 
619 			// Flush TLBs
620 			if ((wfi_flags & 2) != 0) {
621 				flush_core_tlb();
622 			}
623 
624 			// Wait for the ballance of the wfi delay
625 			clock_delay_until(deadline);
626 		}
627 #endif /* DEVELOPMENT || DEBUG */
628 	}
629 
630 	ClearIdlePop(TRUE);
631 
632 	cpu_idle_exit(FALSE);
633 }
634 
635 /*
636  *	Routine:	cpu_idle_exit
637  *	Function:
638  */
639 void
cpu_idle_exit(boolean_t from_reset)640 cpu_idle_exit(boolean_t from_reset)
641 {
642 	uint64_t        new_idle_timeout_ticks = 0x0ULL;
643 	cpu_data_t     *cpu_data_ptr = getCpuDatap();
644 
645 	assert(exception_stack_pointer() != 0);
646 
647 	/* Back from WFI, unlock OSLAR and EDLAR. */
648 	if (from_reset) {
649 		configure_coresight_registers(cpu_data_ptr);
650 	}
651 
652 #if CONFIG_CPU_COUNTERS
653 	kpc_idle_exit();
654 	mt_cpu_run(cpu_data_ptr);
655 #endif /* CONFIG_CPU_COUNTERS */
656 
657 	if (wfi && (cpu_data_ptr->cpu_idle_notify != NULL)) {
658 		cpu_data_ptr->cpu_idle_notify(cpu_data_ptr->cpu_id, FALSE, &new_idle_timeout_ticks);
659 	}
660 
661 	if (cpu_data_ptr->idle_timer_notify != NULL) {
662 		if (new_idle_timeout_ticks == 0x0ULL) {
663 			/* turn off the idle timer */
664 			cpu_data_ptr->idle_timer_deadline = 0x0ULL;
665 		} else {
666 			/* set the new idle timeout */
667 			clock_absolutetime_interval_to_deadline(new_idle_timeout_ticks, &cpu_data_ptr->idle_timer_deadline);
668 		}
669 		timer_resync_deadlines();
670 	}
671 
672 #if KASAN_TBI
673 	kasan_unpoison_curstack(false);
674 #endif /* KASAN_TBI */
675 
676 	Idle_load_context();
677 }
678 
679 void
cpu_init(void)680 cpu_init(void)
681 {
682 	cpu_data_t     *cdp = getCpuDatap();
683 	arm_cpu_info_t *cpu_info_p;
684 
685 	assert(exception_stack_pointer() != 0);
686 
687 	if (cdp->cpu_type != CPU_TYPE_ARM64) {
688 		cdp->cpu_type = CPU_TYPE_ARM64;
689 
690 		timer_call_queue_init(&cdp->rtclock_timer.queue);
691 		cdp->rtclock_timer.deadline = EndOfAllTime;
692 
693 		if (cdp == &BootCpuData) {
694 			do_cpuid();
695 			do_mvfpid();
696 		} else {
697 			/*
698 			 * We initialize non-boot CPUs here; the boot CPU is
699 			 * dealt with as part of pmap_bootstrap.
700 			 */
701 			pmap_cpu_data_init();
702 		}
703 
704 		do_cacheid();
705 
706 		/* ARM_SMP: Assuming identical cpu */
707 		do_debugid();
708 
709 		cpu_info_p = cpuid_info();
710 
711 		/* switch based on CPU's reported architecture */
712 		switch (cpu_info_p->arm_info.arm_arch) {
713 		case CPU_ARCH_ARMv8:
714 			cdp->cpu_subtype = CPU_SUBTYPE_ARM64_V8;
715 			break;
716 		case CPU_ARCH_ARMv8E:
717 			cdp->cpu_subtype = CPU_SUBTYPE_ARM64E;
718 			break;
719 		default:
720 			//cdp->cpu_subtype = CPU_SUBTYPE_ARM64_ALL;
721 			/* this panic doesn't work this early in startup */
722 			panic("Unknown CPU subtype...");
723 			break;
724 		}
725 
726 		cdp->cpu_threadtype = CPU_THREADTYPE_NONE;
727 	}
728 	cdp->cpu_stat.irq_ex_cnt_wake = 0;
729 	cdp->cpu_stat.ipi_cnt_wake = 0;
730 #if CONFIG_CPU_COUNTERS
731 	cdp->cpu_stat.pmi_cnt_wake = 0;
732 #endif /* CONFIG_CPU_COUNTERS */
733 	cdp->cpu_running = TRUE;
734 	cdp->cpu_sleep_token_last = cdp->cpu_sleep_token;
735 	cdp->cpu_sleep_token = 0x0UL;
736 #if CONFIG_CPU_COUNTERS
737 	kpc_idle_exit();
738 	mt_cpu_up(cdp);
739 #endif /* CONFIG_CPU_COUNTERS */
740 }
741 
742 void
cpu_stack_alloc(cpu_data_t * cpu_data_ptr)743 cpu_stack_alloc(cpu_data_t *cpu_data_ptr)
744 {
745 	vm_offset_t irq_stack = 0;
746 	vm_offset_t exc_stack = 0;
747 
748 	kmem_alloc(kernel_map, &irq_stack,
749 	    INTSTACK_SIZE + ptoa(2), KMA_NOFAIL | KMA_PERMANENT | KMA_ZERO |
750 	    KMA_GUARD_FIRST | KMA_GUARD_LAST | KMA_KSTACK | KMA_KOBJECT,
751 	    VM_KERN_MEMORY_STACK);
752 
753 	cpu_data_ptr->intstack_top = irq_stack + PAGE_SIZE + INTSTACK_SIZE;
754 	cpu_data_ptr->istackptr = (void *)cpu_data_ptr->intstack_top;
755 
756 	kmem_alloc(kernel_map, &exc_stack,
757 	    EXCEPSTACK_SIZE + ptoa(2), KMA_NOFAIL | KMA_PERMANENT | KMA_ZERO |
758 	    KMA_GUARD_FIRST | KMA_GUARD_LAST | KMA_KSTACK | KMA_KOBJECT,
759 	    VM_KERN_MEMORY_STACK);
760 
761 	cpu_data_ptr->excepstack_top = exc_stack + PAGE_SIZE + EXCEPSTACK_SIZE;
762 	cpu_data_ptr->excepstackptr = (void *)cpu_data_ptr->excepstack_top;
763 }
764 
765 void
cpu_data_init(cpu_data_t * cpu_data_ptr)766 cpu_data_init(cpu_data_t *cpu_data_ptr)
767 {
768 	uint32_t i;
769 
770 	os_atomic_store(&cpu_data_ptr->cpu_flags, 0, relaxed);
771 	cpu_data_ptr->cpu_int_state = 0;
772 	cpu_data_ptr->cpu_pending_ast = AST_NONE;
773 	cpu_data_ptr->cpu_cache_dispatch = NULL;
774 	cpu_data_ptr->rtcPop = EndOfAllTime;
775 	cpu_data_ptr->rtclock_datap = &RTClockData;
776 	cpu_data_ptr->cpu_user_debug = NULL;
777 
778 
779 	cpu_data_ptr->cpu_base_timebase = 0;
780 	cpu_data_ptr->cpu_idle_notify = NULL;
781 	cpu_data_ptr->cpu_idle_latency = 0x0ULL;
782 	cpu_data_ptr->cpu_idle_pop = 0x0ULL;
783 	cpu_data_ptr->cpu_reset_type = 0x0UL;
784 	cpu_data_ptr->cpu_reset_handler = 0x0UL;
785 	cpu_data_ptr->cpu_reset_assist = 0x0UL;
786 	cpu_data_ptr->cpu_regmap_paddr = 0x0ULL;
787 	cpu_data_ptr->cpu_phys_id = 0x0UL;
788 	cpu_data_ptr->cpu_cluster_type = CLUSTER_TYPE_SMP;
789 	cpu_data_ptr->cpu_cluster_id = 0;
790 	cpu_data_ptr->cpu_l2_id = 0;
791 	cpu_data_ptr->cpu_l2_size = 0;
792 	cpu_data_ptr->cpu_l3_id = 0;
793 	cpu_data_ptr->cpu_l3_size = 0;
794 
795 	cpu_data_ptr->cpu_signal = SIGPdisabled;
796 
797 	cpu_data_ptr->cpu_get_fiq_handler = NULL;
798 	cpu_data_ptr->cpu_tbd_hardware_addr = NULL;
799 	cpu_data_ptr->cpu_tbd_hardware_val = NULL;
800 	cpu_data_ptr->cpu_get_decrementer_func = NULL;
801 	cpu_data_ptr->cpu_set_decrementer_func = NULL;
802 	cpu_data_ptr->cpu_sleep_token = ARM_CPU_ON_SLEEP_PATH;
803 	cpu_data_ptr->cpu_sleep_token_last = 0x00000000UL;
804 	cpu_data_ptr->cpu_xcall_p0 = NULL;
805 	cpu_data_ptr->cpu_xcall_p1 = NULL;
806 	cpu_data_ptr->cpu_imm_xcall_p0 = NULL;
807 	cpu_data_ptr->cpu_imm_xcall_p1 = NULL;
808 
809 	for (i = 0; i < CORESIGHT_REGIONS; ++i) {
810 		cpu_data_ptr->coresight_base[i] = 0;
811 	}
812 
813 #if !XNU_MONITOR
814 	pmap_cpu_data_t * pmap_cpu_data_ptr = &cpu_data_ptr->cpu_pmap_cpu_data;
815 
816 	pmap_cpu_data_ptr->cpu_number = PMAP_INVALID_CPU_NUM;
817 	pmap_cpu_data_ptr->pv_free.list = NULL;
818 	pmap_cpu_data_ptr->pv_free.count = 0;
819 	pmap_cpu_data_ptr->pv_free_spill_marker = NULL;
820 #if !CONFIG_SPTM
821 	pmap_cpu_data_ptr->cpu_nested_pmap = (struct pmap *) NULL;
822 	bzero(&(pmap_cpu_data_ptr->cpu_sw_asids[0]), sizeof(pmap_cpu_data_ptr->cpu_sw_asids));
823 #endif
824 #endif /* !XNU_MONITOR */
825 	cpu_data_ptr->halt_status = CPU_NOT_HALTED;
826 #if __ARM_KERNEL_PROTECT__
827 	cpu_data_ptr->cpu_exc_vectors = (vm_offset_t)&exc_vectors_table;
828 #endif /* __ARM_KERNEL_PROTECT__ */
829 
830 #if defined(HAS_APPLE_PAC)
831 	cpu_data_ptr->rop_key = 0;
832 	cpu_data_ptr->jop_key = ml_default_jop_pid();
833 #endif
834 #if NEEDS_MTE_IRG_RESEED
835 	cpu_data_ptr->cpu_irg_reseed_counter = 0;
836 #endif
837 }
838 
839 void
cpu_data_register(cpu_data_t * cpu_data_ptr)840 cpu_data_register(cpu_data_t *cpu_data_ptr)
841 {
842 	int     cpu = cpu_data_ptr->cpu_number;
843 
844 #if KASAN
845 	for (int i = 0; i < CPUWINDOWS_MAX; i++) {
846 		kasan_notify_address_nopoison(pmap_cpu_windows_copy_addr(cpu, i), PAGE_SIZE);
847 	}
848 #endif
849 
850 	__builtin_arm_dmb(DMB_ISH); // Ensure prior stores to cpu data are visible
851 	CpuDataEntries[cpu].cpu_data_vaddr = cpu_data_ptr;
852 	CpuDataEntries[cpu].cpu_data_paddr = (void *)ml_vtophys((vm_offset_t)cpu_data_ptr);
853 }
854 
855 #if defined(KERNEL_INTEGRITY_CTRR)
856 /* Hibernation needs to reset this state, so data and text are in the hib segment;
857  * this allows them be accessed and executed early.
858  */
859 LCK_GRP_DECLARE(ctrr_cpu_start_lock_grp, "ctrr_cpu_start_lock");
860 LCK_SPIN_DECLARE(ctrr_cpu_start_lck, &ctrr_cpu_start_lock_grp);
861 enum ctrr_cluster_states ctrr_cluster_locked[MAX_CPU_CLUSTERS] MARK_AS_HIBERNATE_DATA;
862 
863 MARK_AS_HIBERNATE_TEXT
864 void
init_ctrr_cluster_states(void)865 init_ctrr_cluster_states(void)
866 {
867 	for (int i = 0; i < MAX_CPU_CLUSTERS; i++) {
868 		ctrr_cluster_locked[i] = CTRR_UNLOCKED;
869 	}
870 }
871 #endif
872 
873 void
cpu_start(int cpu)874 cpu_start(int cpu)
875 {
876 	cpu_data_t *cpu_data_ptr = CpuDataEntries[cpu].cpu_data_vaddr;
877 	processor_t processor = PERCPU_GET_RELATIVE(processor, cpu_data, cpu_data_ptr);
878 
879 	if (processor_should_kprintf(processor, true)) {
880 		kprintf("cpu_start() cpu: %d\n", cpu);
881 	}
882 
883 	if (cpu == cpu_number()) {
884 		/* Current CPU is already running, just needs initialization */
885 		cpu_machine_init();
886 		configure_coresight_registers(cpu_data_ptr);
887 	} else {
888 		thread_t first_thread;
889 #if CONFIG_SPTM
890 		cpu_data_ptr->cpu_reset_handler = (vm_offset_t) VM_KERNEL_STRIP_PTR(arm_init_cpu);
891 #else
892 		cpu_data_ptr->cpu_reset_handler = (vm_offset_t) start_cpu_paddr;
893 #if !XNU_MONITOR
894 		cpu_data_ptr->cpu_pmap_cpu_data.cpu_nested_pmap = NULL;
895 #endif
896 #endif /* !CONFIG_SPTM */
897 
898 		if (processor->startup_thread != THREAD_NULL) {
899 			first_thread = processor->startup_thread;
900 		} else {
901 			first_thread = processor->idle_thread;
902 		}
903 		cpu_data_ptr->cpu_active_thread = first_thread;
904 		first_thread->machine.CpuDatap = cpu_data_ptr;
905 		first_thread->machine.pcpu_data_base_and_cpu_number =
906 		    ml_make_pcpu_base_and_cpu_number((vm_address_t)cpu_data_ptr - __PERCPU_ADDR(cpu_data),
907 		    cpu_data_ptr->cpu_number);
908 
909 		configure_coresight_registers(cpu_data_ptr);
910 
911 		flush_dcache((vm_offset_t)&CpuDataEntries[cpu], sizeof(cpu_data_entry_t), FALSE);
912 		flush_dcache((vm_offset_t)cpu_data_ptr, sizeof(cpu_data_t), FALSE);
913 #if CONFIG_SPTM
914 		/**
915 		 * On SPTM devices, CTRR is configured entirely by the SPTM. Due to this, this logic
916 		 * is no longer required in XNU.
917 		 */
918 #else
919 #if defined(KERNEL_INTEGRITY_CTRR)
920 
921 		/* First CPU being started within a cluster goes ahead to lock CTRR for cluster;
922 		 * other CPUs block until cluster is locked. */
923 		lck_spin_lock(&ctrr_cpu_start_lck);
924 		switch (ctrr_cluster_locked[cpu_data_ptr->cpu_cluster_id]) {
925 		case CTRR_UNLOCKED:
926 			ctrr_cluster_locked[cpu_data_ptr->cpu_cluster_id] = CTRR_LOCKING;
927 			lck_spin_unlock(&ctrr_cpu_start_lck);
928 			break;
929 		case CTRR_LOCKING:
930 
931 			lck_spin_sleep(&ctrr_cpu_start_lck, LCK_SLEEP_UNLOCK,
932 			    &ctrr_cluster_locked[cpu_data_ptr->cpu_cluster_id],
933 			    THREAD_UNINT | THREAD_WAIT_NOREPORT);
934 
935 			assert(ctrr_cluster_locked[cpu_data_ptr->cpu_cluster_id] != CTRR_LOCKING);
936 			break;
937 		default:         // CTRR_LOCKED
938 			lck_spin_unlock(&ctrr_cpu_start_lck);
939 			break;
940 		}
941 #endif
942 #endif /* CONFIG_SPTM */
943 
944 		PE_cpu_start_internal(cpu_data_ptr->cpu_id, (vm_offset_t)NULL, (vm_offset_t)NULL);
945 
946 	}
947 }
948 
949 
950 void
cpu_timebase_init(boolean_t from_boot)951 cpu_timebase_init(boolean_t from_boot)
952 {
953 	cpu_data_t *cdp = getCpuDatap();
954 	uint64_t timebase_offset = 0;
955 
956 	if (cdp->cpu_get_fiq_handler == NULL) {
957 		cdp->cpu_get_fiq_handler = rtclock_timebase_func.tbd_fiq_handler;
958 		cdp->cpu_get_decrementer_func = rtclock_timebase_func.tbd_get_decrementer;
959 		cdp->cpu_set_decrementer_func = rtclock_timebase_func.tbd_set_decrementer;
960 		cdp->cpu_tbd_hardware_addr = (void *)rtclock_timebase_addr;
961 		cdp->cpu_tbd_hardware_val = (void *)rtclock_timebase_val;
962 	}
963 
964 	if (!from_boot && (cdp == &BootCpuData) && ml_is_quiescing()) {
965 		/*
966 		 * When we wake from sleep, we have no guarantee about the state
967 		 * of the hardware timebase.  It may have kept ticking across sleep, or
968 		 * it may have reset.
969 		 *
970 		 * To deal with this, we calculate an offset to the clock that will
971 		 * produce a timebase value wake_abstime at the point the boot
972 		 * CPU calls cpu_timebase_init on wake.
973 		 *
974 		 * This ensures that mach_absolute_time() stops ticking across sleep.
975 		 */
976 		rtclock_base_abstime = wake_abstime - ml_get_hwclock();
977 	} else if (from_boot) {
978 #if DEBUG || DEVELOPMENT
979 		if (PE_parse_boot_argn("timebase_offset", &timebase_offset, sizeof(timebase_offset))) {
980 			rtclock_base_abstime += timebase_offset;
981 		}
982 #endif
983 		/* On initial boot, initialize time_since_reset to CNTPCT_EL0. */
984 		ml_set_reset_time(ml_get_hwclock());
985 	}
986 
987 	cdp->cpu_decrementer = 0x7FFFFFFFUL;
988 	cdp->cpu_timebase = timebase_offset;
989 	cdp->cpu_base_timebase = rtclock_base_abstime;
990 }
991 
992 int
cpu_cluster_id(void)993 cpu_cluster_id(void)
994 {
995 	return getCpuDatap()->cpu_cluster_id;
996 }
997 
998 __attribute__((noreturn))
999 void
ml_arm_sleep(void)1000 ml_arm_sleep(void)
1001 {
1002 	cpu_data_t              *cpu_data_ptr = getCpuDatap();
1003 
1004 	if (cpu_data_ptr == &BootCpuData) {
1005 		cpu_data_t      *target_cdp;
1006 		int             cpu;
1007 		int             max_cpu;
1008 
1009 		max_cpu = ml_get_max_cpu_number();
1010 		for (cpu = 0; cpu <= max_cpu; cpu++) {
1011 			target_cdp = (cpu_data_t *)CpuDataEntries[cpu].cpu_data_vaddr;
1012 
1013 			if ((target_cdp == NULL) || (target_cdp == cpu_data_ptr)) {
1014 				continue;
1015 			}
1016 
1017 			while (target_cdp->cpu_sleep_token != ARM_CPU_ON_SLEEP_PATH) {
1018 				;
1019 			}
1020 		}
1021 
1022 		/*
1023 		 * Now that the other cores have entered the sleep path, set
1024 		 * the abstime value we'll use when we resume.
1025 		 */
1026 		wake_abstime = ml_get_timebase();
1027 		ml_set_reset_time(UINT64_MAX);
1028 	} else {
1029 		CleanPoU_Dcache();
1030 	}
1031 
1032 	cpu_data_ptr->cpu_sleep_token = ARM_CPU_ON_SLEEP_PATH;
1033 
1034 	if (cpu_data_ptr == &BootCpuData) {
1035 #if WITH_CLASSIC_S2R
1036 		// Classic suspend to RAM writes the suspend signature into the
1037 		// sleep token buffer so that iBoot knows that it's on the warm
1038 		// boot (wake) path (as opposed to the cold boot path). Newer SoC
1039 		// do not go through SecureROM/iBoot on the warm boot path. The
1040 		// reconfig engine script brings the CPU out of reset at the kernel's
1041 		// reset vector which points to the warm boot initialization code.
1042 		if (sleepTokenBuffer != (vm_offset_t) NULL) {
1043 			platform_cache_shutdown();
1044 			bcopy((const void *)suspend_signature, (void *)sleepTokenBuffer, sizeof(SleepToken));
1045 		} else {
1046 			panic("No sleep token buffer");
1047 		}
1048 #endif
1049 
1050 #if __ARM_GLOBAL_SLEEP_BIT__
1051 		/* Allow other CPUs to go to sleep. */
1052 		arm64_stall_sleep = FALSE;
1053 		__builtin_arm_dmb(DMB_ISH);
1054 #endif
1055 
1056 		/* Architectural debug state: <rdar://problem/12390433>:
1057 		 *      Grab debug lock EDLAR and clear bit 0 in EDPRCR,
1058 		 *      tell debugger to not prevent power gating .
1059 		 */
1060 		if (cpu_data_ptr->coresight_base[CORESIGHT_ED]) {
1061 			*(volatile uint32_t *)(cpu_data_ptr->coresight_base[CORESIGHT_ED] + ARM_DEBUG_OFFSET_DBGLAR) = ARM_DBG_LOCK_ACCESS_KEY;
1062 			*(volatile uint32_t *)(cpu_data_ptr->coresight_base[CORESIGHT_ED] + ARM_DEBUG_OFFSET_DBGPRCR) = 0;
1063 		}
1064 
1065 #if HIBERNATION
1066 		uint32_t mode = hibernate_write_image();
1067 		if (mode == kIOHibernatePostWriteHalt) {
1068 			HIBLOG("powering off after writing hibernation image\n");
1069 			int halt_result = -1;
1070 			if (PE_halt_restart) {
1071 				/**
1072 				 * Drain serial FIFOs now as the normal call further down won't
1073 				 * be hit when the CPU halts here for hibernation. Here, it'll
1074 				 * make sure the preceding HIBLOG is flushed as well.
1075 				 */
1076 				serial_go_to_sleep();
1077 				halt_result = (*PE_halt_restart)(kPEHaltCPU);
1078 			}
1079 			panic("can't shutdown: PE_halt_restart returned %d", halt_result);
1080 		}
1081 #endif /* HIBERNATION */
1082 
1083 		serial_go_to_sleep();
1084 
1085 #if CONFIG_CPU_COUNTERS
1086 		mt_sleep();
1087 #endif /* CONFIG_CPU_COUNTERS */
1088 		/* ARM64-specific preparation */
1089 #if APPLEVIRTUALPLATFORM
1090 		extern bool test_sleep_in_vm;
1091 		if (test_sleep_in_vm) {
1092 			/*
1093 			 * Until sleep is supported on APPLEVIRTUALPLATFORM, use this
1094 			 * trick for testing sleep - just jump straight to the CPU resume point.
1095 			 */
1096 			arm_init_cpu(cpu_data_ptr, 0);
1097 		}
1098 		arm64_prepare_for_sleep(true, cpu_data_ptr->cpu_number, ml_vtophys(reset_vector_vaddr));
1099 #else /* APPLEVIRTUALPLATFORM */
1100 		arm64_prepare_for_sleep(true);
1101 #endif /* APPLEVIRTUALPLATFORM */
1102 	} else {
1103 #if __ARM_GLOBAL_SLEEP_BIT__
1104 		/*
1105 		 * With the exception of the CPU revisions listed above, our ARM64 CPUs have a
1106 		 * global register to manage entering deep sleep, as opposed to a per-CPU
1107 		 * register.  We cannot update this register until all CPUs are ready to enter
1108 		 * deep sleep, because if a CPU executes WFI outside of the deep sleep context
1109 		 * (by idling), it will hang (due to the side effects of enabling deep sleep),
1110 		 * which can hang the sleep process or cause memory corruption on wake.
1111 		 *
1112 		 * To avoid these issues, we'll stall on this global value, which CPU0 will
1113 		 * manage.
1114 		 */
1115 		while (arm64_stall_sleep) {
1116 			__builtin_arm_wfe();
1117 		}
1118 #endif
1119 		CleanPoU_DcacheRegion((vm_offset_t) cpu_data_ptr, sizeof(cpu_data_t));
1120 
1121 		/* Architectural debug state: <rdar://problem/12390433>:
1122 		 *      Grab debug lock EDLAR and clear bit 0 in EDPRCR,
1123 		 *      tell debugger to not prevent power gating .
1124 		 */
1125 		if (cpu_data_ptr->coresight_base[CORESIGHT_ED]) {
1126 			*(volatile uint32_t *)(cpu_data_ptr->coresight_base[CORESIGHT_ED] + ARM_DEBUG_OFFSET_DBGLAR) = ARM_DBG_LOCK_ACCESS_KEY;
1127 			*(volatile uint32_t *)(cpu_data_ptr->coresight_base[CORESIGHT_ED] + ARM_DEBUG_OFFSET_DBGPRCR) = 0;
1128 		}
1129 
1130 		/* ARM64-specific preparation */
1131 #if APPLEVIRTUALPLATFORM
1132 		arm64_prepare_for_sleep(true, cpu_data_ptr->cpu_number, ml_vtophys(reset_vector_vaddr));
1133 #else /* APPLEVIRTUALPLATFORM */
1134 		arm64_prepare_for_sleep(true);
1135 #endif /* APPLEVIRTUALPLATFORM */
1136 	}
1137 }
1138 
1139 void
cpu_machine_idle_init(boolean_t from_boot)1140 cpu_machine_idle_init(boolean_t from_boot)
1141 {
1142 #if !CONFIG_SPTM
1143 	static vm_address_t     resume_idle_cpu_paddr = (vm_address_t)NULL;
1144 #endif
1145 	cpu_data_t              *cpu_data_ptr   = getCpuDatap();
1146 
1147 	if (from_boot) {
1148 		uint32_t        production = 1;
1149 		DTEntry         entry;
1150 
1151 		unsigned long   jtag = 0;
1152 
1153 		if (PE_parse_boot_argn("jtag", &jtag, sizeof(jtag))) {
1154 			if (jtag != 0) {
1155 				idle_enable = FALSE;
1156 			} else {
1157 				idle_enable = TRUE;
1158 			}
1159 		} else {
1160 			idle_enable = TRUE;
1161 		}
1162 
1163 #if DEVELOPMENT || DEBUG
1164 		uint32_t wfe_mode = 0;
1165 		if (PE_parse_boot_argn("wfe_mode", &wfe_mode, sizeof(wfe_mode))) {
1166 			idle_proximate_timer_wfe = ((wfe_mode & 1) == 1);
1167 			idle_proximate_io_wfe_masked = ((wfe_mode & 2) == 2);
1168 			extern uint32_t idle_proximate_io_wfe_unmasked;
1169 			idle_proximate_io_wfe_unmasked = ((wfe_mode & 4) == 4);
1170 		}
1171 #endif
1172 
1173 		// bits 7..0 give the wfi type
1174 		switch (wfi & 0xff) {
1175 		case 0:
1176 			// disable wfi
1177 			wfi = 0;
1178 			break;
1179 
1180 #if DEVELOPMENT || DEBUG
1181 		case 2:
1182 			// wfi overhead simulation
1183 			// 31..16 - wfi delay is us
1184 			// 15..8  - flags
1185 			// 7..0   - 2
1186 			wfi = 2;
1187 			wfi_flags = (wfi >> 8) & 0xFF;
1188 			nanoseconds_to_absolutetime(((wfi >> 16) & 0xFFFF) * NSEC_PER_MSEC, &wfi_delay);
1189 			break;
1190 #endif /* DEVELOPMENT || DEBUG */
1191 
1192 		case 1:
1193 		default:
1194 			// do nothing
1195 			break;
1196 		}
1197 #if !CONFIG_SPTM
1198 		ResetHandlerData.assist_reset_handler = 0;
1199 		ResetHandlerData.cpu_data_entries = ml_static_vtop((vm_offset_t)CpuDataEntries);
1200 #endif
1201 
1202 #ifdef MONITOR
1203 		monitor_call(MONITOR_SET_ENTRY, (uintptr_t)ml_static_vtop((vm_offset_t)&LowResetVectorBase), 0, 0);
1204 #elif !defined(NO_MONITOR)
1205 #error MONITOR undefined, WFI power gating may not operate correctly
1206 #endif /* MONITOR */
1207 
1208 		// Determine if we are on production or debug chip
1209 		if (kSuccess == SecureDTLookupEntry(NULL, "/chosen", &entry)) {
1210 			unsigned int    size;
1211 			void const      *prop;
1212 
1213 			if (kSuccess == SecureDTGetProperty(entry, "effective-production-status-ap", &prop, &size)) {
1214 				if (size == 4) {
1215 					bcopy(prop, &production, size);
1216 				}
1217 			}
1218 		}
1219 		if (!production) {
1220 #if defined(APPLE_ARM64_ARCH_FAMILY)
1221 			// Enable coresight debug registers on debug-fused chips
1222 			coresight_debug_enabled = TRUE;
1223 #endif
1224 		}
1225 #if !CONFIG_SPTM
1226 		start_cpu_paddr = ml_static_vtop((vm_offset_t)&start_cpu);
1227 		resume_idle_cpu_paddr = ml_static_vtop((vm_offset_t)&resume_idle_cpu);
1228 #endif
1229 	}
1230 
1231 #if WITH_CLASSIC_S2R
1232 	if (cpu_data_ptr == &BootCpuData) {
1233 		static addr64_t SleepToken_low_paddr = (addr64_t)NULL;
1234 		if (sleepTokenBuffer != (vm_offset_t) NULL) {
1235 			SleepToken_low_paddr = ml_vtophys(sleepTokenBuffer);
1236 		} else {
1237 			panic("No sleep token buffer");
1238 		}
1239 
1240 		bcopy_phys((addr64_t)ml_static_vtop((vm_offset_t)running_signature),
1241 		    SleepToken_low_paddr, sizeof(SleepToken));
1242 		flush_dcache((vm_offset_t)SleepToken, sizeof(SleepToken), TRUE);
1243 	}
1244 	;
1245 #endif
1246 #if CONFIG_SPTM
1247 	cpu_data_ptr->cpu_reset_handler = (uintptr_t) VM_KERNEL_STRIP_PTR(arm_init_idle_cpu);
1248 #else
1249 	cpu_data_ptr->cpu_reset_handler = resume_idle_cpu_paddr;
1250 #endif
1251 	clean_dcache((vm_offset_t)cpu_data_ptr, sizeof(cpu_data_t), FALSE);
1252 }
1253 
1254 _Atomic uint32_t cpu_idle_count = 0;
1255 
1256 void
machine_track_platform_idle(boolean_t entry)1257 machine_track_platform_idle(boolean_t entry)
1258 {
1259 	if (entry) {
1260 		os_atomic_inc(&cpu_idle_count, relaxed);
1261 	} else {
1262 		os_atomic_dec(&cpu_idle_count, relaxed);
1263 	}
1264 }
1265 
1266 #if WITH_CLASSIC_S2R
1267 void
sleep_token_buffer_init(void)1268 sleep_token_buffer_init(void)
1269 {
1270 	cpu_data_t      *cpu_data_ptr = getCpuDatap();
1271 	DTEntry         entry;
1272 	size_t          size;
1273 	void const * const *prop;
1274 
1275 	if ((cpu_data_ptr == &BootCpuData) && (sleepTokenBuffer == (vm_offset_t) NULL)) {
1276 		/* Find the stpage node in the device tree */
1277 		if (kSuccess != SecureDTLookupEntry(0, "stram", &entry)) {
1278 			return;
1279 		}
1280 
1281 		if (kSuccess != SecureDTGetProperty(entry, "reg", (const void **)&prop, (unsigned int *)&size)) {
1282 			return;
1283 		}
1284 
1285 		/* Map the page into the kernel space */
1286 		sleepTokenBuffer = ml_io_map(((vm_offset_t const *)prop)[0], ((vm_size_t const *)prop)[1]);
1287 	}
1288 }
1289 #endif
1290