xref: /xnu-8020.101.4/osfmk/arm64/cpu.c (revision e7776783b89a353188416a9a346c6cdb4928faad)
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.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 #if KPC
64 #include <kern/kpc.h>
65 #endif
66 
67 #if MONOTONIC
68 #include <kern/monotonic.h>
69 #endif /* MONOTONIC */
70 
71 #if KPERF
72 #include <kperf/kptimer.h>
73 #endif /* KPERF */
74 
75 #if HIBERNATION
76 #include <IOKit/IOPlatformExpert.h>
77 #include <IOKit/IOHibernatePrivate.h>
78 #endif /* HIBERNATION */
79 
80 
81 #include <libkern/section_keywords.h>
82 
83 extern boolean_t        idle_enable;
84 extern uint64_t         wake_abstime;
85 
86 #if WITH_CLASSIC_S2R
87 void sleep_token_buffer_init(void);
88 #endif
89 
90 
91 extern uintptr_t resume_idle_cpu;
92 extern uintptr_t start_cpu;
93 
94 #if __ARM_KERNEL_PROTECT__
95 extern void exc_vectors_table;
96 #endif /* __ARM_KERNEL_PROTECT__ */
97 
98 extern void __attribute__((noreturn)) arm64_prepare_for_sleep(boolean_t deep_sleep, unsigned int cpu, uint64_t entry_pa);
99 extern void arm64_force_wfi_clock_gate(void);
100 #if defined(APPLETYPHOON)
101 // <rdar://problem/15827409>
102 extern void typhoon_prepare_for_wfi(void);
103 extern void typhoon_return_from_wfi(void);
104 #endif
105 
106 #if HAS_RETENTION_STATE
107 extern void arm64_retention_wfi(void);
108 #endif
109 
110 vm_address_t   start_cpu_paddr;
111 
112 sysreg_restore_t sysreg_restore __attribute__((section("__DATA, __const"))) = {
113 	.tcr_el1 = TCR_EL1_BOOT,
114 };
115 
116 // wfi - wfi mode
117 //  0 : disabled
118 //  1 : normal
119 //  2 : overhead simulation (delay & flags)
120 TUNABLE(unsigned int, wfi, "wfi", 1);
121 #if DEVELOPMENT || DEBUG
122 
123 // wfi_flags
124 //  1 << 0 : flush L1s
125 //  1 << 1 : flush TLBs
126 static int wfi_flags = 0;
127 
128 // wfi_delay - delay ticks after wfi exit
129 static uint64_t wfi_delay = 0;
130 
131 #endif /* DEVELOPMENT || DEBUG */
132 
133 #define CPUPM_IDLE_WFE 0x5310300
134 #define CPUPM_IDLE_TIMER_WFE 0x5310304
135 
136 /* When recommended, issue WFE with [FI]IRQ unmasked in the idle
137  * loop. The default.
138  */
139 uint32_t idle_proximate_io_wfe_unmasked = 1;
140 #if DEVELOPMENT || DEBUG
141 uint32_t idle_proximate_timer_wfe = 1;
142 uint32_t idle_proximate_io_wfe_masked = 0;
143 #else
144 /* Issue WFE in lieu of WFI when awaiting a proximate timer. */
145 static uint32_t idle_proximate_timer_wfe = 1;
146 /* When recommended, issue WFE with [FI]IRQ masked in the idle loop.
147  * Non-default, retained for experimentation.
148  */
149 static uint32_t idle_proximate_io_wfe_masked = 0;
150 #endif
151 
152 #if __ARM_GLOBAL_SLEEP_BIT__
153 volatile boolean_t arm64_stall_sleep = TRUE;
154 #endif
155 
156 #if WITH_CLASSIC_S2R
157 /*
158  * These must be aligned to avoid issues with calling bcopy_phys on them before
159  * we are done with pmap initialization.
160  */
161 static const uint8_t __attribute__ ((aligned(8))) suspend_signature[] = {'X', 'S', 'O', 'M', 'P', 'S', 'U', 'S'};
162 static const uint8_t __attribute__ ((aligned(8))) running_signature[] = {'X', 'S', 'O', 'M', 'N', 'N', 'U', 'R'};
163 #endif
164 
165 #if WITH_CLASSIC_S2R
166 static vm_offset_t sleepTokenBuffer = (vm_offset_t)NULL;
167 #endif
168 static boolean_t coresight_debug_enabled = FALSE;
169 
170 #if defined(CONFIG_XNUPOST)
171 void arm64_ipi_test_callback(void *);
172 void arm64_immediate_ipi_test_callback(void *);
173 
174 void
arm64_ipi_test_callback(void * parm)175 arm64_ipi_test_callback(void *parm)
176 {
177 	volatile uint64_t *ipi_test_data = parm;
178 	cpu_data_t *cpu_data;
179 
180 	cpu_data = getCpuDatap();
181 
182 	*ipi_test_data = cpu_data->cpu_number;
183 }
184 
185 void
arm64_immediate_ipi_test_callback(void * parm)186 arm64_immediate_ipi_test_callback(void *parm)
187 {
188 	volatile uint64_t *ipi_test_data = parm;
189 	cpu_data_t *cpu_data;
190 
191 	cpu_data = getCpuDatap();
192 
193 	*ipi_test_data = cpu_data->cpu_number + MAX_CPUS;
194 }
195 
196 uint64_t arm64_ipi_test_data[MAX_CPUS * 2];
197 
198 MACHINE_TIMEOUT(arm64_ipi_test_timeout, "arm64-ipi-test", 100, MACHINE_TIMEOUT_UNIT_MSEC, NULL);
199 
200 void
arm64_ipi_test()201 arm64_ipi_test()
202 {
203 	volatile uint64_t *ipi_test_data, *immediate_ipi_test_data;
204 	uint64_t timeout_ms = os_atomic_load(&arm64_ipi_test_timeout, relaxed);
205 	uint64_t then, now, delta;
206 	int current_cpu_number = getCpuDatap()->cpu_number;
207 
208 	/*
209 	 * probably the only way to have this on most systems is with the
210 	 * cpus=1 boot-arg, but nonetheless, if we only have 1 CPU active,
211 	 * IPI is not available
212 	 */
213 	if (real_ncpus == 1) {
214 		return;
215 	}
216 
217 	const unsigned int max_cpu_id = ml_get_max_cpu_number();
218 	for (unsigned int i = 0; i <= max_cpu_id; ++i) {
219 		ipi_test_data = &arm64_ipi_test_data[i];
220 		immediate_ipi_test_data = &arm64_ipi_test_data[i + MAX_CPUS];
221 		*ipi_test_data = ~i;
222 		kern_return_t error = cpu_xcall((int)i, (void *)arm64_ipi_test_callback, (void *)(uintptr_t)ipi_test_data);
223 		if (error != KERN_SUCCESS) {
224 			panic("CPU %d was unable to IPI CPU %u: error %d", current_cpu_number, i, error);
225 		}
226 
227 		while ((error = cpu_immediate_xcall((int)i, (void *)arm64_immediate_ipi_test_callback,
228 		    (void *)(uintptr_t)immediate_ipi_test_data)) == KERN_ALREADY_WAITING) {
229 			now = mach_absolute_time();
230 			absolutetime_to_nanoseconds(now - then, &delta);
231 			if ((delta / NSEC_PER_MSEC) > timeout_ms) {
232 				panic("CPU %d was unable to immediate-IPI CPU %u within %lldms", current_cpu_number, i, timeout_ms);
233 			}
234 		}
235 
236 		if (error != KERN_SUCCESS) {
237 			panic("CPU %d was unable to immediate-IPI CPU %u: error %d", current_cpu_number, i, error);
238 		}
239 
240 		then = mach_absolute_time();
241 
242 		while ((*ipi_test_data != i) || (*immediate_ipi_test_data != (i + MAX_CPUS))) {
243 			now = mach_absolute_time();
244 			absolutetime_to_nanoseconds(now - then, &delta);
245 			if ((delta / NSEC_PER_MSEC) > timeout_ms) {
246 				panic("CPU %d tried to IPI CPU %d but didn't get correct responses within %lldms, responses: %llx, %llx",
247 				    current_cpu_number, i, timeout_ms, *ipi_test_data, *immediate_ipi_test_data);
248 			}
249 		}
250 	}
251 }
252 #endif /* defined(CONFIG_XNUPOST) */
253 
254 static void
configure_coresight_registers(cpu_data_t * cdp)255 configure_coresight_registers(cpu_data_t *cdp)
256 {
257 	int             i;
258 
259 	assert(cdp);
260 	vm_offset_t     coresight_regs = ml_get_topology_info()->cpus[cdp->cpu_number].coresight_regs;
261 
262 	/*
263 	 * ARMv8 coresight registers are optional. If the device tree did not
264 	 * provide either cpu_regmap_paddr (from the legacy "reg-private" EDT property)
265 	 * or coresight_regs (from the new "coresight-reg" property), assume that
266 	 * coresight registers are not supported.
267 	 */
268 	if (cdp->cpu_regmap_paddr || coresight_regs) {
269 		for (i = 0; i < CORESIGHT_REGIONS; ++i) {
270 			if (i == CORESIGHT_CTI) {
271 				continue;
272 			}
273 			/* Skip debug-only registers on production chips */
274 			if (((i == CORESIGHT_ED) || (i == CORESIGHT_UTT)) && !coresight_debug_enabled) {
275 				continue;
276 			}
277 
278 			if (!cdp->coresight_base[i]) {
279 				if (coresight_regs) {
280 					cdp->coresight_base[i] = coresight_regs + CORESIGHT_OFFSET(i);
281 				} else {
282 					uint64_t addr = cdp->cpu_regmap_paddr + CORESIGHT_OFFSET(i);
283 					cdp->coresight_base[i] = (vm_offset_t)ml_io_map(addr, CORESIGHT_SIZE);
284 				}
285 
286 				/*
287 				 * At this point, failing to io map the
288 				 * registers is considered as an error.
289 				 */
290 				if (!cdp->coresight_base[i]) {
291 					panic("unable to ml_io_map coresight regions");
292 				}
293 			}
294 			/* Unlock EDLAR, CTILAR, PMLAR */
295 			if (i != CORESIGHT_UTT) {
296 				*(volatile uint32_t *)(cdp->coresight_base[i] + ARM_DEBUG_OFFSET_DBGLAR) = ARM_DBG_LOCK_ACCESS_KEY;
297 			}
298 		}
299 	}
300 }
301 
302 
303 /*
304  *	Routine:	cpu_bootstrap
305  *	Function:
306  */
307 void
cpu_bootstrap(void)308 cpu_bootstrap(void)
309 {
310 }
311 
312 /*
313  *	Routine:	cpu_sleep
314  *	Function:
315  */
316 void
cpu_sleep(void)317 cpu_sleep(void)
318 {
319 	cpu_data_t     *cpu_data_ptr = getCpuDatap();
320 
321 	cpu_data_ptr->cpu_active_thread = current_thread();
322 	cpu_data_ptr->cpu_reset_handler = (uintptr_t) start_cpu_paddr;
323 	cpu_data_ptr->cpu_flags |= SleepState;
324 	cpu_data_ptr->cpu_user_debug = NULL;
325 #if KPC
326 	kpc_idle();
327 #endif /* KPC */
328 #if MONOTONIC
329 	mt_cpu_down(cpu_data_ptr);
330 #endif /* MONOTONIC */
331 #if KPERF
332 	kptimer_stop_curcpu();
333 #endif /* KPERF */
334 
335 	CleanPoC_Dcache();
336 
337 #if USE_APPLEARMSMP
338 	if (ml_is_quiescing()) {
339 		PE_cpu_machine_quiesce(cpu_data_ptr->cpu_id);
340 	} else {
341 		bool deep_sleep = PE_cpu_down(cpu_data_ptr->cpu_id);
342 		cpu_data_ptr->cpu_sleep_token = ARM_CPU_ON_SLEEP_PATH;
343 		// hang CPU on spurious wakeup
344 		cpu_data_ptr->cpu_reset_handler = (uintptr_t)0;
345 		__builtin_arm_dsb(DSB_ISH);
346 		CleanPoU_Dcache();
347 		arm64_prepare_for_sleep(deep_sleep, cpu_data_ptr->cpu_number, ml_vtophys((vm_offset_t)&LowResetVectorBase));
348 	}
349 #else
350 	PE_cpu_machine_quiesce(cpu_data_ptr->cpu_id);
351 #endif
352 	/*NOTREACHED*/
353 }
354 
355 /*
356  *	Routine:	cpu_interrupt_is_pending
357  *	Function:	Returns a bool signifying a non-zero ISR_EL1,
358  *			indicating a pending IRQ, FIQ or external abort.
359  */
360 
361 bool
cpu_interrupt_is_pending(void)362 cpu_interrupt_is_pending(void)
363 {
364 	uint64_t isr_value;
365 	isr_value = __builtin_arm_rsr64("ISR_EL1");
366 	return isr_value != 0;
367 }
368 
369 static bool
cpu_proximate_timer(void)370 cpu_proximate_timer(void)
371 {
372 	return !SetIdlePop();
373 }
374 
375 #ifdef ARM64_BOARD_CONFIG_T6000
376 int wfe_allowed = 0;
377 #else
378 int wfe_allowed = 1;
379 #endif /* ARM64_BOARD_CONFIG_T6000 */
380 
381 #if DEVELOPMENT || DEBUG
382 #define WFE_STAT(x)     \
383 	do {            \
384 	        (x);    \
385 	} while(0)
386 #else
387 #define WFE_STAT(x)     do {} while(0)
388 #endif /* DEVELOPMENT || DEBUG */
389 
390 bool
wfe_to_deadline_or_interrupt(uint32_t cid,uint64_t wfe_deadline,cpu_data_t * cdp,bool unmask)391 wfe_to_deadline_or_interrupt(uint32_t cid, uint64_t wfe_deadline, cpu_data_t *cdp, bool unmask)
392 {
393 	bool ipending = false;
394 	uint64_t irqc = 0, nirqc = 0;
395 
396 	/* The ARMv8 architecture permits a processor dwelling in WFE
397 	 * with F/IRQ masked to ignore a pending interrupt, i.e.
398 	 * not classify it as an 'event'. This is potentially
399 	 * problematic with AICv2's IRQ distribution model, as
400 	 * a transient interrupt masked interval can cause an SIQ
401 	 * query rejection, possibly routing the interrupt to
402 	 * another core/cluster in a powergated state.
403 	 * Hence, optionally unmask IRQs+FIQs across WFE.
404 	 */
405 	if (unmask) {
406 		/* Latch SW IRQ+FIQ counter prior to unmasking
407 		 * interrupts.
408 		 */
409 		irqc = nirqc = os_atomic_load(&cdp->cpu_stat.irq_ex_cnt_wake, relaxed);
410 		/* Unmask IRQ+FIQ. Mirrors mask used by machine_idle()
411 		 * with ASYNCF omission. Consider that this could
412 		 * delay recognition of an async abort, including
413 		 * those triggered by ISRs
414 		 */
415 		__builtin_arm_wsr("DAIFClr", (DAIFSC_IRQF | DAIFSC_FIQF));
416 	}
417 
418 	while ((ipending = (cpu_interrupt_is_pending())) == false) {
419 		if (unmask) {
420 			/* If WFE was issued with IRQs unmasked, an
421 			 * interrupt may have been processed.
422 			 * Consult the SW IRQ counter to determine
423 			 * whether the 'idle loop' must be
424 			 * re-evaluated.
425 			 */
426 			nirqc = os_atomic_load(&cdp->cpu_stat.irq_ex_cnt_wake, relaxed);
427 			if (nirqc != irqc) {
428 				break;
429 			}
430 		}
431 
432 		if (__probable(wfe_allowed)) {
433 			/*
434 			 * If IRQs are unmasked, there's a small window
435 			 * where an 'extra' WFE may be issued after
436 			 * the consultation of the SW interrupt counter
437 			 * and new interrupt arrival. Hence this WFE
438 			 * relies on the [FI]RQ interrupt handler
439 			 * epilogue issuing a 'SEVL', to post an
440 			 * event which causes the next WFE on the same
441 			 * PE to retire immediately.
442 			 */
443 
444 			__builtin_arm_wfe();
445 		}
446 
447 		WFE_STAT(cdp->wfe_count++);
448 		if (wfe_deadline != ~0ULL) {
449 			WFE_STAT(cdp->wfe_deadline_checks++);
450 			/* Check if the WFE recommendation has expired.
451 			 * We do not recompute the deadline here.
452 			 */
453 			if ((ml_cluster_wfe_timeout(cid) == 0) ||
454 			    mach_absolute_time() >= wfe_deadline) {
455 				WFE_STAT(cdp->wfe_terminations++);
456 				break;
457 			}
458 		}
459 	}
460 
461 	if (unmask) {
462 		/* Re-mask IRQ+FIQ
463 		 * Mirrors mask used by machine_idle(), with ASYNCF
464 		 * omission
465 		 */
466 		__builtin_arm_wsr64("DAIFSet", (DAIFSC_IRQF | DAIFSC_FIQF));
467 		/* Refetch SW interrupt counter with IRQs masked
468 		 * It is important that this routine accurately flags
469 		 * any observed interrupts via its return value,
470 		 * inaccuracy may lead to an erroneous WFI fallback.
471 		 */
472 		nirqc = os_atomic_load(&cdp->cpu_stat.irq_ex_cnt_wake, relaxed);
473 	}
474 
475 	return ipending || (nirqc != irqc);
476 }
477 
478 /*
479  *	Routine:	cpu_idle
480  *	Function:
481  */
482 void __attribute__((noreturn))
cpu_idle(void)483 cpu_idle(void)
484 {
485 	cpu_data_t     *cpu_data_ptr = getCpuDatap();
486 	uint64_t        new_idle_timeout_ticks = 0x0ULL, lastPop;
487 	bool idle_disallowed = false;
488 
489 	if (__improbable((!idle_enable))) {
490 		idle_disallowed = true;
491 	} else if (__improbable(cpu_data_ptr->cpu_signal & SIGPdisabled)) {
492 		idle_disallowed = true;
493 	}
494 
495 	if (__improbable(idle_disallowed)) {
496 		Idle_load_context();
497 	}
498 
499 	bool ipending = false;
500 	uint32_t cid = cpu_data_ptr->cpu_cluster_id;
501 
502 	if (idle_proximate_io_wfe_masked == 1) {
503 		uint64_t wfe_deadline = 0;
504 		/* Check for an active perf. controller generated
505 		 * WFE recommendation for this cluster.
506 		 */
507 		uint64_t wfe_ttd = 0;
508 		if ((wfe_ttd = ml_cluster_wfe_timeout(cid)) != 0) {
509 			wfe_deadline = mach_absolute_time() + wfe_ttd;
510 		}
511 
512 		if (wfe_deadline != 0) {
513 			/* Poll issuing event-bounded WFEs until an interrupt
514 			 * arrives or the WFE recommendation expires
515 			 */
516 #if DEVELOPMENT || DEBUG
517 			KDBG(CPUPM_IDLE_WFE | DBG_FUNC_START, ipending, cpu_data_ptr->wfe_count, wfe_ttd, cid);
518 #endif
519 			ipending = wfe_to_deadline_or_interrupt(cid, wfe_deadline, cpu_data_ptr, false);
520 #if DEVELOPMENT || DEBUG
521 			KDBG(CPUPM_IDLE_WFE | DBG_FUNC_END, ipending, cpu_data_ptr->wfe_count, wfe_deadline, 0);
522 #endif
523 			if (ipending == true) {
524 				/* Back to machine_idle() */
525 				Idle_load_context();
526 			}
527 		}
528 	}
529 
530 	if (__improbable(cpu_proximate_timer())) {
531 		if (idle_proximate_timer_wfe == 1) {
532 			/* Poll issuing WFEs until the expected
533 			 * timer FIQ arrives.
534 			 */
535 #if DEVELOPMENT || DEBUG
536 			KDBG(CPUPM_IDLE_TIMER_WFE | DBG_FUNC_START, ipending, cpu_data_ptr->wfe_count, ~0ULL, 0);
537 #endif
538 			ipending = wfe_to_deadline_or_interrupt(cid, ~0ULL, cpu_data_ptr, false);
539 #if DEVELOPMENT || DEBUG
540 			KDBG(CPUPM_IDLE_TIMER_WFE | DBG_FUNC_END, ipending, cpu_data_ptr->wfe_count, ~0ULL, 0);
541 #endif
542 			assert(ipending == true);
543 		}
544 		Idle_load_context();
545 	}
546 
547 	lastPop = cpu_data_ptr->rtcPop;
548 
549 	cpu_data_ptr->cpu_active_thread = current_thread();
550 
551 	if (wfi && (cpu_data_ptr->cpu_idle_notify != NULL)) {
552 		cpu_data_ptr->cpu_idle_notify(cpu_data_ptr->cpu_id, TRUE, &new_idle_timeout_ticks);
553 	}
554 
555 	if (cpu_data_ptr->idle_timer_notify != NULL) {
556 		if (new_idle_timeout_ticks == 0x0ULL) {
557 			/* turn off the idle timer */
558 			cpu_data_ptr->idle_timer_deadline = 0x0ULL;
559 		} else {
560 			/* set the new idle timeout */
561 			clock_absolutetime_interval_to_deadline(new_idle_timeout_ticks, &cpu_data_ptr->idle_timer_deadline);
562 		}
563 		timer_resync_deadlines();
564 		if (cpu_data_ptr->rtcPop != lastPop) {
565 			SetIdlePop();
566 		}
567 	}
568 
569 #if KPC
570 	kpc_idle();
571 #endif
572 #if MONOTONIC
573 	mt_cpu_idle(cpu_data_ptr);
574 #endif /* MONOTONIC */
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 #if !defined(APPLE_ARM64_ARCH_FAMILY)
629 		platform_cache_idle_exit();
630 #endif
631 	}
632 
633 	ClearIdlePop(TRUE);
634 
635 	cpu_idle_exit(FALSE);
636 }
637 
638 /*
639  *	Routine:	cpu_idle_exit
640  *	Function:
641  */
642 void
cpu_idle_exit(boolean_t from_reset)643 cpu_idle_exit(boolean_t from_reset)
644 {
645 	uint64_t        new_idle_timeout_ticks = 0x0ULL;
646 	cpu_data_t     *cpu_data_ptr = getCpuDatap();
647 
648 	assert(exception_stack_pointer() != 0);
649 
650 	/* Back from WFI, unlock OSLAR and EDLAR. */
651 	if (from_reset) {
652 		configure_coresight_registers(cpu_data_ptr);
653 	}
654 
655 #if KPC
656 	kpc_idle_exit();
657 #endif
658 
659 #if MONOTONIC
660 	mt_cpu_run(cpu_data_ptr);
661 #endif /* MONOTONIC */
662 
663 	if (wfi && (cpu_data_ptr->cpu_idle_notify != NULL)) {
664 		cpu_data_ptr->cpu_idle_notify(cpu_data_ptr->cpu_id, FALSE, &new_idle_timeout_ticks);
665 	}
666 
667 	if (cpu_data_ptr->idle_timer_notify != NULL) {
668 		if (new_idle_timeout_ticks == 0x0ULL) {
669 			/* turn off the idle timer */
670 			cpu_data_ptr->idle_timer_deadline = 0x0ULL;
671 		} else {
672 			/* set the new idle timeout */
673 			clock_absolutetime_interval_to_deadline(new_idle_timeout_ticks, &cpu_data_ptr->idle_timer_deadline);
674 		}
675 		timer_resync_deadlines();
676 	}
677 
678 	Idle_load_context();
679 }
680 
681 void
cpu_init(void)682 cpu_init(void)
683 {
684 	cpu_data_t     *cdp = getCpuDatap();
685 	arm_cpu_info_t *cpu_info_p;
686 
687 	assert(exception_stack_pointer() != 0);
688 
689 	if (cdp->cpu_type != CPU_TYPE_ARM64) {
690 		cdp->cpu_type = CPU_TYPE_ARM64;
691 
692 		timer_call_queue_init(&cdp->rtclock_timer.queue);
693 		cdp->rtclock_timer.deadline = EndOfAllTime;
694 
695 		if (cdp == &BootCpuData) {
696 			do_cpuid();
697 			do_mvfpid();
698 		} else {
699 			/*
700 			 * We initialize non-boot CPUs here; the boot CPU is
701 			 * dealt with as part of pmap_bootstrap.
702 			 */
703 			pmap_cpu_data_init();
704 		}
705 
706 		do_cacheid();
707 
708 		/* ARM_SMP: Assuming identical cpu */
709 		do_debugid();
710 
711 		cpu_info_p = cpuid_info();
712 
713 		/* switch based on CPU's reported architecture */
714 		switch (cpu_info_p->arm_info.arm_arch) {
715 		case CPU_ARCH_ARMv8:
716 			cdp->cpu_subtype = CPU_SUBTYPE_ARM64_V8;
717 			break;
718 		case CPU_ARCH_ARMv8E:
719 			cdp->cpu_subtype = CPU_SUBTYPE_ARM64E;
720 			break;
721 		default:
722 			//cdp->cpu_subtype = CPU_SUBTYPE_ARM64_ALL;
723 			/* this panic doesn't work this early in startup */
724 			panic("Unknown CPU subtype...");
725 			break;
726 		}
727 
728 		cdp->cpu_threadtype = CPU_THREADTYPE_NONE;
729 	}
730 	cdp->cpu_stat.irq_ex_cnt_wake = 0;
731 	cdp->cpu_stat.ipi_cnt_wake = 0;
732 #if MONOTONIC
733 	cdp->cpu_stat.pmi_cnt_wake = 0;
734 #endif /* MONOTONIC */
735 	cdp->cpu_running = TRUE;
736 	cdp->cpu_sleep_token_last = cdp->cpu_sleep_token;
737 	cdp->cpu_sleep_token = 0x0UL;
738 #if KPC
739 	kpc_idle_exit();
740 #endif /* KPC */
741 #if MONOTONIC
742 	mt_cpu_up(cdp);
743 #endif /* MONOTONIC */
744 }
745 
746 void
cpu_stack_alloc(cpu_data_t * cpu_data_ptr)747 cpu_stack_alloc(cpu_data_t *cpu_data_ptr)
748 {
749 	vm_offset_t             irq_stack = 0;
750 	vm_offset_t             exc_stack = 0;
751 
752 	kern_return_t kr = kernel_memory_allocate(kernel_map, &irq_stack,
753 	    INTSTACK_SIZE + (2 * PAGE_SIZE),
754 	    PAGE_MASK,
755 	    KMA_GUARD_FIRST | KMA_GUARD_LAST | KMA_KSTACK | KMA_KOBJECT,
756 	    VM_KERN_MEMORY_STACK);
757 	if (kr != KERN_SUCCESS) {
758 		panic("Unable to allocate cpu interrupt stack");
759 	}
760 
761 	cpu_data_ptr->intstack_top = irq_stack + PAGE_SIZE + INTSTACK_SIZE;
762 	cpu_data_ptr->istackptr = cpu_data_ptr->intstack_top;
763 
764 	kr = kernel_memory_allocate(kernel_map, &exc_stack,
765 	    EXCEPSTACK_SIZE + (2 * PAGE_SIZE),
766 	    PAGE_MASK,
767 	    KMA_GUARD_FIRST | KMA_GUARD_LAST | KMA_KSTACK | KMA_KOBJECT,
768 	    VM_KERN_MEMORY_STACK);
769 	if (kr != KERN_SUCCESS) {
770 		panic("Unable to allocate cpu exception stack");
771 	}
772 
773 	cpu_data_ptr->excepstack_top = exc_stack + PAGE_SIZE + EXCEPSTACK_SIZE;
774 	cpu_data_ptr->excepstackptr = cpu_data_ptr->excepstack_top;
775 }
776 
777 void
cpu_data_free(cpu_data_t * cpu_data_ptr)778 cpu_data_free(cpu_data_t *cpu_data_ptr)
779 {
780 	if ((cpu_data_ptr == NULL) || (cpu_data_ptr == &BootCpuData)) {
781 		return;
782 	}
783 
784 	int cpu_number = cpu_data_ptr->cpu_number;
785 
786 	if (CpuDataEntries[cpu_number].cpu_data_vaddr == cpu_data_ptr) {
787 		CpuDataEntries[cpu_number].cpu_data_vaddr = NULL;
788 		CpuDataEntries[cpu_number].cpu_data_paddr = 0;
789 		__builtin_arm_dmb(DMB_ISH); // Ensure prior stores to cpu array are visible
790 	}
791 	kmem_free(kernel_map,
792 	    cpu_data_ptr->intstack_top - INTSTACK_SIZE - PAGE_SIZE,
793 	    INTSTACK_SIZE + 2 * PAGE_SIZE);
794 	kmem_free(kernel_map,
795 	    cpu_data_ptr->excepstack_top - EXCEPSTACK_SIZE - PAGE_SIZE,
796 	    EXCEPSTACK_SIZE + 2 * PAGE_SIZE);
797 }
798 
799 void
cpu_data_init(cpu_data_t * cpu_data_ptr)800 cpu_data_init(cpu_data_t *cpu_data_ptr)
801 {
802 	uint32_t i;
803 
804 	cpu_data_ptr->cpu_flags = 0;
805 	cpu_data_ptr->cpu_int_state = 0;
806 	cpu_data_ptr->cpu_pending_ast = AST_NONE;
807 	cpu_data_ptr->cpu_cache_dispatch = NULL;
808 	cpu_data_ptr->rtcPop = EndOfAllTime;
809 	cpu_data_ptr->rtclock_datap = &RTClockData;
810 	cpu_data_ptr->cpu_user_debug = NULL;
811 
812 
813 	cpu_data_ptr->cpu_base_timebase = 0;
814 	cpu_data_ptr->cpu_idle_notify = NULL;
815 	cpu_data_ptr->cpu_idle_latency = 0x0ULL;
816 	cpu_data_ptr->cpu_idle_pop = 0x0ULL;
817 	cpu_data_ptr->cpu_reset_type = 0x0UL;
818 	cpu_data_ptr->cpu_reset_handler = 0x0UL;
819 	cpu_data_ptr->cpu_reset_assist = 0x0UL;
820 	cpu_data_ptr->cpu_regmap_paddr = 0x0ULL;
821 	cpu_data_ptr->cpu_phys_id = 0x0UL;
822 	cpu_data_ptr->cpu_l2_access_penalty = 0;
823 	cpu_data_ptr->cpu_cluster_type = CLUSTER_TYPE_SMP;
824 	cpu_data_ptr->cpu_cluster_id = 0;
825 	cpu_data_ptr->cpu_l2_id = 0;
826 	cpu_data_ptr->cpu_l2_size = 0;
827 	cpu_data_ptr->cpu_l3_id = 0;
828 	cpu_data_ptr->cpu_l3_size = 0;
829 
830 	cpu_data_ptr->cpu_signal = SIGPdisabled;
831 
832 	cpu_data_ptr->cpu_get_fiq_handler = NULL;
833 	cpu_data_ptr->cpu_tbd_hardware_addr = NULL;
834 	cpu_data_ptr->cpu_tbd_hardware_val = NULL;
835 	cpu_data_ptr->cpu_get_decrementer_func = NULL;
836 	cpu_data_ptr->cpu_set_decrementer_func = NULL;
837 	cpu_data_ptr->cpu_sleep_token = ARM_CPU_ON_SLEEP_PATH;
838 	cpu_data_ptr->cpu_sleep_token_last = 0x00000000UL;
839 	cpu_data_ptr->cpu_xcall_p0 = NULL;
840 	cpu_data_ptr->cpu_xcall_p1 = NULL;
841 	cpu_data_ptr->cpu_imm_xcall_p0 = NULL;
842 	cpu_data_ptr->cpu_imm_xcall_p1 = NULL;
843 
844 	for (i = 0; i < CORESIGHT_REGIONS; ++i) {
845 		cpu_data_ptr->coresight_base[i] = 0;
846 	}
847 
848 #if !XNU_MONITOR
849 	pmap_cpu_data_t * pmap_cpu_data_ptr = &cpu_data_ptr->cpu_pmap_cpu_data;
850 
851 	pmap_cpu_data_ptr->cpu_nested_pmap = (struct pmap *) NULL;
852 	pmap_cpu_data_ptr->cpu_number = PMAP_INVALID_CPU_NUM;
853 	pmap_cpu_data_ptr->pv_free.list = NULL;
854 	pmap_cpu_data_ptr->pv_free.count = 0;
855 	pmap_cpu_data_ptr->pv_free_spill_marker = NULL;
856 
857 	bzero(&(pmap_cpu_data_ptr->cpu_sw_asids[0]), sizeof(pmap_cpu_data_ptr->cpu_sw_asids));
858 #endif
859 	cpu_data_ptr->halt_status = CPU_NOT_HALTED;
860 #if __ARM_KERNEL_PROTECT__
861 	cpu_data_ptr->cpu_exc_vectors = (vm_offset_t)&exc_vectors_table;
862 #endif /* __ARM_KERNEL_PROTECT__ */
863 
864 #if defined(HAS_APPLE_PAC)
865 	cpu_data_ptr->rop_key = 0;
866 	cpu_data_ptr->jop_key = ml_default_jop_pid();
867 #endif
868 }
869 
870 kern_return_t
cpu_data_register(cpu_data_t * cpu_data_ptr)871 cpu_data_register(cpu_data_t *cpu_data_ptr)
872 {
873 	int     cpu = cpu_data_ptr->cpu_number;
874 
875 #if KASAN
876 	for (int i = 0; i < CPUWINDOWS_MAX; i++) {
877 		kasan_notify_address_nopoison(pmap_cpu_windows_copy_addr(cpu, i), PAGE_SIZE);
878 	}
879 #endif
880 
881 	__builtin_arm_dmb(DMB_ISH); // Ensure prior stores to cpu data are visible
882 	CpuDataEntries[cpu].cpu_data_vaddr = cpu_data_ptr;
883 	CpuDataEntries[cpu].cpu_data_paddr = (void *)ml_vtophys((vm_offset_t)cpu_data_ptr);
884 	return KERN_SUCCESS;
885 }
886 
887 #if defined(KERNEL_INTEGRITY_CTRR)
888 /* Hibernation needs to reset this state, so data and text are in the hib segment;
889  * this allows them be accessed and executed early.
890  */
891 LCK_GRP_DECLARE(ctrr_cpu_start_lock_grp, "ctrr_cpu_start_lock");
892 LCK_SPIN_DECLARE(ctrr_cpu_start_lck, &ctrr_cpu_start_lock_grp);
893 enum ctrr_cluster_states ctrr_cluster_locked[MAX_CPU_CLUSTERS] MARK_AS_HIBERNATE_DATA;
894 
895 MARK_AS_HIBERNATE_TEXT
896 void
init_ctrr_cluster_states(void)897 init_ctrr_cluster_states(void)
898 {
899 	for (int i = 0; i < MAX_CPU_CLUSTERS; i++) {
900 		ctrr_cluster_locked[i] = CTRR_UNLOCKED;
901 	}
902 }
903 #endif
904 
905 kern_return_t
cpu_start(int cpu)906 cpu_start(int cpu)
907 {
908 	cpu_data_t *cpu_data_ptr = CpuDataEntries[cpu].cpu_data_vaddr;
909 
910 	kprintf("cpu_start() cpu: %d\n", cpu);
911 
912 	if (cpu == cpu_number()) {
913 		cpu_machine_init();
914 		configure_coresight_registers(cpu_data_ptr);
915 	} else {
916 		thread_t first_thread;
917 		processor_t processor;
918 
919 		cpu_data_ptr->cpu_reset_handler = (vm_offset_t) start_cpu_paddr;
920 
921 #if !XNU_MONITOR
922 		cpu_data_ptr->cpu_pmap_cpu_data.cpu_nested_pmap = NULL;
923 #endif
924 
925 		processor = PERCPU_GET_RELATIVE(processor, cpu_data, cpu_data_ptr);
926 		if (processor->startup_thread != THREAD_NULL) {
927 			first_thread = processor->startup_thread;
928 		} else {
929 			first_thread = processor->idle_thread;
930 		}
931 		cpu_data_ptr->cpu_active_thread = first_thread;
932 		first_thread->machine.CpuDatap = cpu_data_ptr;
933 		first_thread->machine.pcpu_data_base =
934 		    (vm_address_t)cpu_data_ptr - __PERCPU_ADDR(cpu_data);
935 
936 		configure_coresight_registers(cpu_data_ptr);
937 
938 		flush_dcache((vm_offset_t)&CpuDataEntries[cpu], sizeof(cpu_data_entry_t), FALSE);
939 		flush_dcache((vm_offset_t)cpu_data_ptr, sizeof(cpu_data_t), FALSE);
940 #if defined(KERNEL_INTEGRITY_CTRR)
941 
942 		/* First CPU being started within a cluster goes ahead to lock CTRR for cluster;
943 		 * other CPUs block until cluster is locked. */
944 		lck_spin_lock(&ctrr_cpu_start_lck);
945 		switch (ctrr_cluster_locked[cpu_data_ptr->cpu_cluster_id]) {
946 		case CTRR_UNLOCKED:
947 			ctrr_cluster_locked[cpu_data_ptr->cpu_cluster_id] = CTRR_LOCKING;
948 			lck_spin_unlock(&ctrr_cpu_start_lck);
949 			break;
950 		case CTRR_LOCKING:
951 			assert_wait(&ctrr_cluster_locked[cpu_data_ptr->cpu_cluster_id], THREAD_UNINT);
952 			lck_spin_unlock(&ctrr_cpu_start_lck);
953 			thread_block(THREAD_CONTINUE_NULL);
954 			assert(ctrr_cluster_locked[cpu_data_ptr->cpu_cluster_id] != CTRR_LOCKING);
955 			break;
956 		default:         // CTRR_LOCKED
957 			lck_spin_unlock(&ctrr_cpu_start_lck);
958 			break;
959 		}
960 #endif
961 		(void) PE_cpu_start(cpu_data_ptr->cpu_id, (vm_offset_t)NULL, (vm_offset_t)NULL);
962 	}
963 
964 	return KERN_SUCCESS;
965 }
966 
967 
968 void
cpu_timebase_init(boolean_t from_boot)969 cpu_timebase_init(boolean_t from_boot)
970 {
971 	cpu_data_t *cdp = getCpuDatap();
972 
973 	if (cdp->cpu_get_fiq_handler == NULL) {
974 		cdp->cpu_get_fiq_handler = rtclock_timebase_func.tbd_fiq_handler;
975 		cdp->cpu_get_decrementer_func = rtclock_timebase_func.tbd_get_decrementer;
976 		cdp->cpu_set_decrementer_func = rtclock_timebase_func.tbd_set_decrementer;
977 		cdp->cpu_tbd_hardware_addr = (void *)rtclock_timebase_addr;
978 		cdp->cpu_tbd_hardware_val = (void *)rtclock_timebase_val;
979 	}
980 
981 	if (!from_boot && (cdp == &BootCpuData)) {
982 		/*
983 		 * When we wake from sleep, we have no guarantee about the state
984 		 * of the hardware timebase.  It may have kept ticking across sleep, or
985 		 * it may have reset.
986 		 *
987 		 * To deal with this, we calculate an offset to the clock that will
988 		 * produce a timebase value wake_abstime at the point the boot
989 		 * CPU calls cpu_timebase_init on wake.
990 		 *
991 		 * This ensures that mach_absolute_time() stops ticking across sleep.
992 		 */
993 		rtclock_base_abstime = wake_abstime - ml_get_hwclock();
994 	} else if (from_boot) {
995 		/* On initial boot, initialize time_since_reset to CNTPCT_EL0. */
996 		ml_set_reset_time(ml_get_hwclock());
997 	}
998 
999 	cdp->cpu_decrementer = 0x7FFFFFFFUL;
1000 	cdp->cpu_timebase = 0x0UL;
1001 	cdp->cpu_base_timebase = rtclock_base_abstime;
1002 }
1003 
1004 int
cpu_cluster_id(void)1005 cpu_cluster_id(void)
1006 {
1007 	return getCpuDatap()->cpu_cluster_id;
1008 }
1009 
1010 __attribute__((noreturn))
1011 void
ml_arm_sleep(void)1012 ml_arm_sleep(void)
1013 {
1014 	cpu_data_t              *cpu_data_ptr = getCpuDatap();
1015 
1016 	if (cpu_data_ptr == &BootCpuData) {
1017 		cpu_data_t      *target_cdp;
1018 		int             cpu;
1019 		int             max_cpu;
1020 
1021 		max_cpu = ml_get_max_cpu_number();
1022 		for (cpu = 0; cpu <= max_cpu; cpu++) {
1023 			target_cdp = (cpu_data_t *)CpuDataEntries[cpu].cpu_data_vaddr;
1024 
1025 			if ((target_cdp == NULL) || (target_cdp == cpu_data_ptr)) {
1026 				continue;
1027 			}
1028 
1029 			while (target_cdp->cpu_sleep_token != ARM_CPU_ON_SLEEP_PATH) {
1030 				;
1031 			}
1032 		}
1033 
1034 		/*
1035 		 * Now that the other cores have entered the sleep path, set
1036 		 * the abstime value we'll use when we resume.
1037 		 */
1038 		wake_abstime = ml_get_timebase();
1039 		ml_set_reset_time(UINT64_MAX);
1040 	} else {
1041 		CleanPoU_Dcache();
1042 	}
1043 
1044 	cpu_data_ptr->cpu_sleep_token = ARM_CPU_ON_SLEEP_PATH;
1045 
1046 	if (cpu_data_ptr == &BootCpuData) {
1047 #if WITH_CLASSIC_S2R
1048 		// Classic suspend to RAM writes the suspend signature into the
1049 		// sleep token buffer so that iBoot knows that it's on the warm
1050 		// boot (wake) path (as opposed to the cold boot path). Newer SoC
1051 		// do not go through SecureROM/iBoot on the warm boot path. The
1052 		// reconfig engine script brings the CPU out of reset at the kernel's
1053 		// reset vector which points to the warm boot initialization code.
1054 		if (sleepTokenBuffer != (vm_offset_t) NULL) {
1055 			platform_cache_shutdown();
1056 			bcopy((const void *)suspend_signature, (void *)sleepTokenBuffer, sizeof(SleepToken));
1057 		} else {
1058 			panic("No sleep token buffer");
1059 		}
1060 #endif
1061 
1062 #if __ARM_GLOBAL_SLEEP_BIT__
1063 		/* Allow other CPUs to go to sleep. */
1064 		arm64_stall_sleep = FALSE;
1065 		__builtin_arm_dmb(DMB_ISH);
1066 #endif
1067 
1068 		/* Architectural debug state: <rdar://problem/12390433>:
1069 		 *      Grab debug lock EDLAR and clear bit 0 in EDPRCR,
1070 		 *      tell debugger to not prevent power gating .
1071 		 */
1072 		if (cpu_data_ptr->coresight_base[CORESIGHT_ED]) {
1073 			*(volatile uint32_t *)(cpu_data_ptr->coresight_base[CORESIGHT_ED] + ARM_DEBUG_OFFSET_DBGLAR) = ARM_DBG_LOCK_ACCESS_KEY;
1074 			*(volatile uint32_t *)(cpu_data_ptr->coresight_base[CORESIGHT_ED] + ARM_DEBUG_OFFSET_DBGPRCR) = 0;
1075 		}
1076 
1077 #if HIBERNATION
1078 		uint32_t mode = hibernate_write_image();
1079 		if (mode == kIOHibernatePostWriteHalt) {
1080 			HIBLOG("powering off after writing hibernation image\n");
1081 			int halt_result = -1;
1082 			if (PE_halt_restart) {
1083 				/**
1084 				 * Drain serial FIFOs now as the normal call further down won't
1085 				 * be hit when the CPU halts here for hibernation. Here, it'll
1086 				 * make sure the preceding HIBLOG is flushed as well.
1087 				 */
1088 				serial_go_to_sleep();
1089 				halt_result = (*PE_halt_restart)(kPEHaltCPU);
1090 			}
1091 			panic("can't shutdown: PE_halt_restart returned %d", halt_result);
1092 		}
1093 #endif /* HIBERNATION */
1094 
1095 		serial_go_to_sleep();
1096 
1097 #if MONOTONIC
1098 		mt_sleep();
1099 #endif /* MONOTONIC */
1100 		/* ARM64-specific preparation */
1101 		arm64_prepare_for_sleep(true, cpu_data_ptr->cpu_number, ml_vtophys((vm_offset_t)&LowResetVectorBase));
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 		arm64_prepare_for_sleep(true, cpu_data_ptr->cpu_number, ml_vtophys((vm_offset_t)&LowResetVectorBase));
1132 	}
1133 }
1134 
1135 void
cpu_machine_idle_init(boolean_t from_boot)1136 cpu_machine_idle_init(boolean_t from_boot)
1137 {
1138 	static vm_address_t     resume_idle_cpu_paddr = (vm_address_t)NULL;
1139 	cpu_data_t              *cpu_data_ptr   = getCpuDatap();
1140 
1141 	if (from_boot) {
1142 		uint32_t        production = 1;
1143 		DTEntry         entry;
1144 
1145 		unsigned long   jtag = 0;
1146 
1147 		if (PE_parse_boot_argn("jtag", &jtag, sizeof(jtag))) {
1148 			if (jtag != 0) {
1149 				idle_enable = FALSE;
1150 			} else {
1151 				idle_enable = TRUE;
1152 			}
1153 		} else {
1154 			idle_enable = TRUE;
1155 		}
1156 
1157 #if DEVELOPMENT || DEBUG
1158 		uint32_t wfe_mode = 0;
1159 		if (PE_parse_boot_argn("wfe_mode", &wfe_mode, sizeof(wfe_mode))) {
1160 			idle_proximate_timer_wfe = ((wfe_mode & 1) == 1);
1161 			idle_proximate_io_wfe_masked = ((wfe_mode & 2) == 2);
1162 			extern uint32_t idle_proximate_io_wfe_unmasked;
1163 			idle_proximate_io_wfe_unmasked = ((wfe_mode & 4) == 4);
1164 		}
1165 #endif
1166 
1167 		// bits 7..0 give the wfi type
1168 		switch (wfi & 0xff) {
1169 		case 0:
1170 			// disable wfi
1171 			wfi = 0;
1172 			break;
1173 
1174 #if DEVELOPMENT || DEBUG
1175 		case 2:
1176 			// wfi overhead simulation
1177 			// 31..16 - wfi delay is us
1178 			// 15..8  - flags
1179 			// 7..0   - 2
1180 			wfi = 2;
1181 			wfi_flags = (wfi >> 8) & 0xFF;
1182 			nanoseconds_to_absolutetime(((wfi >> 16) & 0xFFFF) * NSEC_PER_MSEC, &wfi_delay);
1183 			break;
1184 #endif /* DEVELOPMENT || DEBUG */
1185 
1186 		case 1:
1187 		default:
1188 			// do nothing
1189 			break;
1190 		}
1191 
1192 		ResetHandlerData.assist_reset_handler = 0;
1193 		ResetHandlerData.cpu_data_entries = ml_static_vtop((vm_offset_t)CpuDataEntries);
1194 
1195 #ifdef MONITOR
1196 		monitor_call(MONITOR_SET_ENTRY, (uintptr_t)ml_static_vtop((vm_offset_t)&LowResetVectorBase), 0, 0);
1197 #elif !defined(NO_MONITOR)
1198 #error MONITOR undefined, WFI power gating may not operate correctly
1199 #endif /* MONITOR */
1200 
1201 		// Determine if we are on production or debug chip
1202 		if (kSuccess == SecureDTLookupEntry(NULL, "/chosen", &entry)) {
1203 			unsigned int    size;
1204 			void const      *prop;
1205 
1206 			if (kSuccess == SecureDTGetProperty(entry, "effective-production-status-ap", &prop, &size)) {
1207 				if (size == 4) {
1208 					bcopy(prop, &production, size);
1209 				}
1210 			}
1211 		}
1212 		if (!production) {
1213 #if defined(APPLE_ARM64_ARCH_FAMILY)
1214 			// Enable coresight debug registers on debug-fused chips
1215 			coresight_debug_enabled = TRUE;
1216 #endif
1217 		}
1218 
1219 		start_cpu_paddr = ml_static_vtop((vm_offset_t)&start_cpu);
1220 		resume_idle_cpu_paddr = ml_static_vtop((vm_offset_t)&resume_idle_cpu);
1221 	}
1222 
1223 #if WITH_CLASSIC_S2R
1224 	if (cpu_data_ptr == &BootCpuData) {
1225 		static addr64_t SleepToken_low_paddr = (addr64_t)NULL;
1226 		if (sleepTokenBuffer != (vm_offset_t) NULL) {
1227 			SleepToken_low_paddr = ml_vtophys(sleepTokenBuffer);
1228 		} else {
1229 			panic("No sleep token buffer");
1230 		}
1231 
1232 		bcopy_phys((addr64_t)ml_static_vtop((vm_offset_t)running_signature),
1233 		    SleepToken_low_paddr, sizeof(SleepToken));
1234 		flush_dcache((vm_offset_t)SleepToken, sizeof(SleepToken), TRUE);
1235 	}
1236 	;
1237 #endif
1238 
1239 	cpu_data_ptr->cpu_reset_handler = resume_idle_cpu_paddr;
1240 	clean_dcache((vm_offset_t)cpu_data_ptr, sizeof(cpu_data_t), FALSE);
1241 }
1242 
1243 _Atomic uint32_t cpu_idle_count = 0;
1244 
1245 void
machine_track_platform_idle(boolean_t entry)1246 machine_track_platform_idle(boolean_t entry)
1247 {
1248 	if (entry) {
1249 		os_atomic_inc(&cpu_idle_count, relaxed);
1250 	} else {
1251 		os_atomic_dec(&cpu_idle_count, relaxed);
1252 	}
1253 }
1254 
1255 #if WITH_CLASSIC_S2R
1256 void
sleep_token_buffer_init(void)1257 sleep_token_buffer_init(void)
1258 {
1259 	cpu_data_t      *cpu_data_ptr = getCpuDatap();
1260 	DTEntry         entry;
1261 	size_t          size;
1262 	void const * const *prop;
1263 
1264 	if ((cpu_data_ptr == &BootCpuData) && (sleepTokenBuffer == (vm_offset_t) NULL)) {
1265 		/* Find the stpage node in the device tree */
1266 		if (kSuccess != SecureDTLookupEntry(0, "stram", &entry)) {
1267 			return;
1268 		}
1269 
1270 		if (kSuccess != SecureDTGetProperty(entry, "reg", (const void **)&prop, (unsigned int *)&size)) {
1271 			return;
1272 		}
1273 
1274 		/* Map the page into the kernel space */
1275 		sleepTokenBuffer = ml_io_map(((vm_offset_t const *)prop)[0], ((vm_size_t const *)prop)[1]);
1276 	}
1277 }
1278 #endif
1279