xref: /titanic_41/usr/src/uts/i86pc/os/mp_machdep.c (revision b885580b43755ee4ea1e280b85428893d2ba9291)
1 
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 /*
27  * Copyright (c) 2009, Intel Corporation.
28  * All rights reserved.
29  */
30 
31 #define	PSMI_1_6
32 #include <sys/smp_impldefs.h>
33 #include <sys/psm.h>
34 #include <sys/psm_modctl.h>
35 #include <sys/pit.h>
36 #include <sys/cmn_err.h>
37 #include <sys/strlog.h>
38 #include <sys/clock.h>
39 #include <sys/debug.h>
40 #include <sys/rtc.h>
41 #include <sys/x86_archext.h>
42 #include <sys/cpupart.h>
43 #include <sys/cpuvar.h>
44 #include <sys/cpu_event.h>
45 #include <sys/cmt.h>
46 #include <sys/cpu.h>
47 #include <sys/disp.h>
48 #include <sys/archsystm.h>
49 #include <sys/machsystm.h>
50 #include <sys/sysmacros.h>
51 #include <sys/memlist.h>
52 #include <sys/param.h>
53 #include <sys/promif.h>
54 #include <sys/cpu_pm.h>
55 #if defined(__xpv)
56 #include <sys/hypervisor.h>
57 #endif
58 #include <sys/mach_intr.h>
59 #include <vm/hat_i86.h>
60 #include <sys/kdi_machimpl.h>
61 #include <sys/sdt.h>
62 #include <sys/hpet.h>
63 #include <sys/sunddi.h>
64 #include <sys/sunndi.h>
65 #include <sys/cpc_pcbe.h>
66 
67 #define	OFFSETOF(s, m)		(size_t)(&(((s *)0)->m))
68 
69 /*
70  *	Local function prototypes
71  */
72 static int mp_disable_intr(processorid_t cpun);
73 static void mp_enable_intr(processorid_t cpun);
74 static void mach_init();
75 static void mach_picinit();
76 static int machhztomhz(uint64_t cpu_freq_hz);
77 static uint64_t mach_getcpufreq(void);
78 static void mach_fixcpufreq(void);
79 static int mach_clkinit(int, int *);
80 static void mach_smpinit(void);
81 static int mach_softlvl_to_vect(int ipl);
82 static void mach_get_platform(int owner);
83 static void mach_construct_info();
84 static int mach_translate_irq(dev_info_t *dip, int irqno);
85 static int mach_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
86     psm_intr_op_t, int *);
87 static void mach_notify_error(int level, char *errmsg);
88 static hrtime_t dummy_hrtime(void);
89 static void dummy_scalehrtime(hrtime_t *);
90 static uint64_t dummy_unscalehrtime(hrtime_t);
91 void cpu_idle(void);
92 static void cpu_wakeup(cpu_t *, int);
93 #ifndef __xpv
94 void cpu_idle_mwait(void);
95 static void cpu_wakeup_mwait(cpu_t *, int);
96 #endif
97 static int mach_cpu_create_devinfo(cpu_t *cp, dev_info_t **dipp);
98 
99 /*
100  *	External reference functions
101  */
102 extern void return_instr();
103 extern uint64_t freq_tsc(uint32_t *);
104 #if defined(__i386)
105 extern uint64_t freq_notsc(uint32_t *);
106 #endif
107 extern void pc_gethrestime(timestruc_t *);
108 extern int cpuid_get_coreid(cpu_t *);
109 extern int cpuid_get_chipid(cpu_t *);
110 
111 /*
112  *	PSM functions initialization
113  */
114 void (*psm_shutdownf)(int, int)	= (void (*)(int, int))return_instr;
115 void (*psm_preshutdownf)(int, int) = (void (*)(int, int))return_instr;
116 void (*psm_notifyf)(int)	= (void (*)(int))return_instr;
117 void (*psm_set_idle_cpuf)(int)	= (void (*)(int))return_instr;
118 void (*psm_unset_idle_cpuf)(int) = (void (*)(int))return_instr;
119 void (*psminitf)()		= mach_init;
120 void (*picinitf)() 		= return_instr;
121 int (*clkinitf)(int, int *) 	= (int (*)(int, int *))return_instr;
122 int (*ap_mlsetup)() 		= (int (*)(void))return_instr;
123 void (*send_dirintf)() 		= return_instr;
124 void (*setspl)(int)		= (void (*)(int))return_instr;
125 int (*addspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
126 int (*delspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
127 void (*kdisetsoftint)(int, struct av_softinfo *)=
128 	(void (*)(int, struct av_softinfo *))return_instr;
129 void (*setsoftint)(int, struct av_softinfo *)=
130 	(void (*)(int, struct av_softinfo *))return_instr;
131 int (*slvltovect)(int)		= (int (*)(int))return_instr;
132 int (*setlvl)(int, int *)	= (int (*)(int, int *))return_instr;
133 void (*setlvlx)(int, int)	= (void (*)(int, int))return_instr;
134 int (*psm_disable_intr)(int)	= mp_disable_intr;
135 void (*psm_enable_intr)(int)	= mp_enable_intr;
136 hrtime_t (*gethrtimef)(void)	= dummy_hrtime;
137 hrtime_t (*gethrtimeunscaledf)(void)	= dummy_hrtime;
138 void (*scalehrtimef)(hrtime_t *)	= dummy_scalehrtime;
139 uint64_t (*unscalehrtimef)(hrtime_t)	= dummy_unscalehrtime;
140 int (*psm_translate_irq)(dev_info_t *, int) = mach_translate_irq;
141 void (*gethrestimef)(timestruc_t *) = pc_gethrestime;
142 void (*psm_notify_error)(int, char *) = (void (*)(int, char *))NULL;
143 int (*psm_get_clockirq)(int) = NULL;
144 int (*psm_get_ipivect)(int, int) = NULL;
145 
146 int (*psm_clkinit)(int) = NULL;
147 void (*psm_timer_reprogram)(hrtime_t) = NULL;
148 void (*psm_timer_enable)(void) = NULL;
149 void (*psm_timer_disable)(void) = NULL;
150 void (*psm_post_cyclic_setup)(void *arg) = NULL;
151 int (*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *, psm_intr_op_t,
152     int *) = mach_intr_ops;
153 int (*psm_state)(psm_state_request_t *) = (int (*)(psm_state_request_t *))
154     return_instr;
155 
156 void (*notify_error)(int, char *) = (void (*)(int, char *))return_instr;
157 void (*hrtime_tick)(void)	= return_instr;
158 
159 int (*psm_cpu_create_devinfo)(cpu_t *, dev_info_t **) = mach_cpu_create_devinfo;
160 
161 /*
162  * True if the generic TSC code is our source of hrtime, rather than whatever
163  * the PSM can provide.
164  */
165 #ifdef __xpv
166 int tsc_gethrtime_enable = 0;
167 #else
168 int tsc_gethrtime_enable = 1;
169 #endif
170 int tsc_gethrtime_initted = 0;
171 
172 /*
173  * True if the hrtime implementation is "hires"; namely, better than microdata.
174  */
175 int gethrtime_hires = 0;
176 
177 /*
178  * Local Static Data
179  */
180 static struct psm_ops mach_ops;
181 static struct psm_ops *mach_set[4] = {&mach_ops, NULL, NULL, NULL};
182 static ushort_t mach_ver[4] = {0, 0, 0, 0};
183 
184 /*
185  * virtualization support for psm
186  */
187 void *psm_vt_ops = NULL;
188 /*
189  * If non-zero, idle cpus will become "halted" when there's
190  * no work to do.
191  */
192 int	idle_cpu_use_hlt = 1;
193 
194 #ifndef __xpv
195 /*
196  * If non-zero, idle cpus will use mwait if available to halt instead of hlt.
197  */
198 int	idle_cpu_prefer_mwait = 1;
199 /*
200  * Set to 0 to avoid MONITOR+CLFLUSH assertion.
201  */
202 int	idle_cpu_assert_cflush_monitor = 1;
203 
204 /*
205  * If non-zero, idle cpus will not use power saving Deep C-States idle loop.
206  */
207 int	idle_cpu_no_deep_c = 0;
208 /*
209  * Non-power saving idle loop and wakeup pointers.
210  * Allows user to toggle Deep Idle power saving feature on/off.
211  */
212 void	(*non_deep_idle_cpu)() = cpu_idle;
213 void	(*non_deep_idle_disp_enq_thread)(cpu_t *, int);
214 
215 /*
216  * Object for the kernel to access the HPET.
217  */
218 hpet_t hpet;
219 
220 #endif	/* ifndef __xpv */
221 
222 /*ARGSUSED*/
223 int
224 pg_plat_hw_shared(cpu_t *cp, pghw_type_t hw)
225 {
226 	switch (hw) {
227 	case PGHW_IPIPE:
228 		if (x86_feature & (X86_HTT)) {
229 			/*
230 			 * Hyper-threading is SMT
231 			 */
232 			return (1);
233 		} else {
234 			return (0);
235 		}
236 	case PGHW_PROCNODE:
237 		if (cpuid_get_procnodes_per_pkg(cp) > 1)
238 			return (1);
239 		else
240 			return (0);
241 	case PGHW_CHIP:
242 		if (x86_feature & (X86_CMP|X86_HTT))
243 			return (1);
244 		else
245 			return (0);
246 	case PGHW_CACHE:
247 		if (cpuid_get_ncpu_sharing_last_cache(cp) > 1)
248 			return (1);
249 		else
250 			return (0);
251 	case PGHW_POW_ACTIVE:
252 		if (cpupm_domain_id(cp, CPUPM_DTYPE_ACTIVE) != (id_t)-1)
253 			return (1);
254 		else
255 			return (0);
256 	case PGHW_POW_IDLE:
257 		if (cpupm_domain_id(cp, CPUPM_DTYPE_IDLE) != (id_t)-1)
258 			return (1);
259 		else
260 			return (0);
261 	default:
262 		return (0);
263 	}
264 }
265 
266 /*
267  * Compare two CPUs and see if they have a pghw_type_t sharing relationship
268  * If pghw_type_t is an unsupported hardware type, then return -1
269  */
270 int
271 pg_plat_cpus_share(cpu_t *cpu_a, cpu_t *cpu_b, pghw_type_t hw)
272 {
273 	id_t pgp_a, pgp_b;
274 
275 	pgp_a = pg_plat_hw_instance_id(cpu_a, hw);
276 	pgp_b = pg_plat_hw_instance_id(cpu_b, hw);
277 
278 	if (pgp_a == -1 || pgp_b == -1)
279 		return (-1);
280 
281 	return (pgp_a == pgp_b);
282 }
283 
284 /*
285  * Return a physical instance identifier for known hardware sharing
286  * relationships
287  */
288 id_t
289 pg_plat_hw_instance_id(cpu_t *cpu, pghw_type_t hw)
290 {
291 	switch (hw) {
292 	case PGHW_IPIPE:
293 		return (cpuid_get_coreid(cpu));
294 	case PGHW_CACHE:
295 		return (cpuid_get_last_lvl_cacheid(cpu));
296 	case PGHW_PROCNODE:
297 		return (cpuid_get_procnodeid(cpu));
298 	case PGHW_CHIP:
299 		return (cpuid_get_chipid(cpu));
300 	case PGHW_POW_ACTIVE:
301 		return (cpupm_domain_id(cpu, CPUPM_DTYPE_ACTIVE));
302 	case PGHW_POW_IDLE:
303 		return (cpupm_domain_id(cpu, CPUPM_DTYPE_IDLE));
304 	default:
305 		return (-1);
306 	}
307 }
308 
309 /*
310  * Express preference for optimizing for sharing relationship
311  * hw1 vs hw2
312  */
313 pghw_type_t
314 pg_plat_hw_rank(pghw_type_t hw1, pghw_type_t hw2)
315 {
316 	int i, rank1, rank2;
317 
318 	static pghw_type_t hw_hier[] = {
319 		PGHW_IPIPE,
320 		PGHW_CACHE,
321 		PGHW_PROCNODE,
322 		PGHW_CHIP,
323 		PGHW_POW_IDLE,
324 		PGHW_POW_ACTIVE,
325 		PGHW_NUM_COMPONENTS
326 	};
327 
328 	for (i = 0; hw_hier[i] != PGHW_NUM_COMPONENTS; i++) {
329 		if (hw_hier[i] == hw1)
330 			rank1 = i;
331 		if (hw_hier[i] == hw2)
332 			rank2 = i;
333 	}
334 
335 	if (rank1 > rank2)
336 		return (hw1);
337 	else
338 		return (hw2);
339 }
340 
341 /*
342  * Override the default CMT dispatcher policy for the specified
343  * hardware sharing relationship
344  */
345 pg_cmt_policy_t
346 pg_plat_cmt_policy(pghw_type_t hw)
347 {
348 	/*
349 	 * For shared caches, also load balance across them to
350 	 * maximize aggregate cache capacity
351 	 */
352 	switch (hw) {
353 	case PGHW_CACHE:
354 		return (CMT_BALANCE|CMT_AFFINITY);
355 	default:
356 		return (CMT_NO_POLICY);
357 	}
358 }
359 
360 id_t
361 pg_plat_get_core_id(cpu_t *cpu)
362 {
363 	return ((id_t)cpuid_get_coreid(cpu));
364 }
365 
366 void
367 cmp_set_nosteal_interval(void)
368 {
369 	/* Set the nosteal interval (used by disp_getbest()) to 100us */
370 	nosteal_nsec = 100000UL;
371 }
372 
373 /*
374  * Routine to ensure initial callers to hrtime gets 0 as return
375  */
376 static hrtime_t
377 dummy_hrtime(void)
378 {
379 	return (0);
380 }
381 
382 /* ARGSUSED */
383 static void
384 dummy_scalehrtime(hrtime_t *ticks)
385 {}
386 
387 static uint64_t
388 dummy_unscalehrtime(hrtime_t nsecs)
389 {
390 	return ((uint64_t)nsecs);
391 }
392 
393 /*
394  * Supports Deep C-State power saving idle loop.
395  */
396 void
397 cpu_idle_adaptive(void)
398 {
399 	(*CPU->cpu_m.mcpu_idle_cpu)();
400 }
401 
402 /*
403  * Function called by CPU idle notification framework to check whether CPU
404  * has been awakened. It will be called with interrupt disabled.
405  * If CPU has been awakened, call cpu_idle_exit() to notify CPU idle
406  * notification framework.
407  */
408 /*ARGSUSED*/
409 static void
410 cpu_idle_check_wakeup(void *arg)
411 {
412 	/*
413 	 * Toggle interrupt flag to detect pending interrupts.
414 	 * If interrupt happened, do_interrupt() will notify CPU idle
415 	 * notification framework so no need to call cpu_idle_exit() here.
416 	 */
417 	sti();
418 	SMT_PAUSE();
419 	cli();
420 }
421 
422 /*
423  * Idle the present CPU until wakened via an interrupt
424  */
425 void
426 cpu_idle(void)
427 {
428 	cpu_t		*cpup = CPU;
429 	processorid_t	cpu_sid = cpup->cpu_seqid;
430 	cpupart_t	*cp = cpup->cpu_part;
431 	int		hset_update = 1;
432 
433 	/*
434 	 * If this CPU is online, and there's multiple CPUs
435 	 * in the system, then we should notate our halting
436 	 * by adding ourselves to the partition's halted CPU
437 	 * bitmap. This allows other CPUs to find/awaken us when
438 	 * work becomes available.
439 	 */
440 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
441 		hset_update = 0;
442 
443 	/*
444 	 * Add ourselves to the partition's halted CPUs bitmap
445 	 * and set our HALTED flag, if necessary.
446 	 *
447 	 * When a thread becomes runnable, it is placed on the queue
448 	 * and then the halted CPU bitmap is checked to determine who
449 	 * (if anyone) should be awakened. We therefore need to first
450 	 * add ourselves to the bitmap, and and then check if there
451 	 * is any work available. The order is important to prevent a race
452 	 * that can lead to work languishing on a run queue somewhere while
453 	 * this CPU remains halted.
454 	 *
455 	 * Either the producing CPU will see we're halted and will awaken us,
456 	 * or this CPU will see the work available in disp_anywork().
457 	 *
458 	 * Note that memory barriers after updating the HALTED flag
459 	 * are not necessary since an atomic operation (updating the bitset)
460 	 * immediately follows. On x86 the atomic operation acts as a
461 	 * memory barrier for the update of cpu_disp_flags.
462 	 */
463 	if (hset_update) {
464 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
465 		bitset_atomic_add(&cp->cp_haltset, cpu_sid);
466 	}
467 
468 	/*
469 	 * Check to make sure there's really nothing to do.
470 	 * Work destined for this CPU may become available after
471 	 * this check. We'll be notified through the clearing of our
472 	 * bit in the halted CPU bitmap, and a poke.
473 	 */
474 	if (disp_anywork()) {
475 		if (hset_update) {
476 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
477 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
478 		}
479 		return;
480 	}
481 
482 	/*
483 	 * We're on our way to being halted.
484 	 *
485 	 * Disable interrupts now, so that we'll awaken immediately
486 	 * after halting if someone tries to poke us between now and
487 	 * the time we actually halt.
488 	 *
489 	 * We check for the presence of our bit after disabling interrupts.
490 	 * If it's cleared, we'll return. If the bit is cleared after
491 	 * we check then the poke will pop us out of the halted state.
492 	 *
493 	 * This means that the ordering of the poke and the clearing
494 	 * of the bit by cpu_wakeup is important.
495 	 * cpu_wakeup() must clear, then poke.
496 	 * cpu_idle() must disable interrupts, then check for the bit.
497 	 */
498 	cli();
499 
500 	if (hset_update && bitset_in_set(&cp->cp_haltset, cpu_sid) == 0) {
501 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
502 		sti();
503 		return;
504 	}
505 
506 	/*
507 	 * The check for anything locally runnable is here for performance
508 	 * and isn't needed for correctness. disp_nrunnable ought to be
509 	 * in our cache still, so it's inexpensive to check, and if there
510 	 * is anything runnable we won't have to wait for the poke.
511 	 */
512 	if (cpup->cpu_disp->disp_nrunnable != 0) {
513 		if (hset_update) {
514 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
515 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
516 		}
517 		sti();
518 		return;
519 	}
520 
521 	if (cpu_idle_enter(IDLE_STATE_C1, 0,
522 	    cpu_idle_check_wakeup, NULL) == 0) {
523 		mach_cpu_idle();
524 		cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
525 	}
526 
527 	/*
528 	 * We're no longer halted
529 	 */
530 	if (hset_update) {
531 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
532 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
533 	}
534 }
535 
536 
537 /*
538  * If "cpu" is halted, then wake it up clearing its halted bit in advance.
539  * Otherwise, see if other CPUs in the cpu partition are halted and need to
540  * be woken up so that they can steal the thread we placed on this CPU.
541  * This function is only used on MP systems.
542  */
543 static void
544 cpu_wakeup(cpu_t *cpu, int bound)
545 {
546 	uint_t		cpu_found;
547 	processorid_t	cpu_sid;
548 	cpupart_t	*cp;
549 
550 	cp = cpu->cpu_part;
551 	cpu_sid = cpu->cpu_seqid;
552 	if (bitset_in_set(&cp->cp_haltset, cpu_sid)) {
553 		/*
554 		 * Clear the halted bit for that CPU since it will be
555 		 * poked in a moment.
556 		 */
557 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
558 		/*
559 		 * We may find the current CPU present in the halted cpuset
560 		 * if we're in the context of an interrupt that occurred
561 		 * before we had a chance to clear our bit in cpu_idle().
562 		 * Poking ourself is obviously unnecessary, since if
563 		 * we're here, we're not halted.
564 		 */
565 		if (cpu != CPU)
566 			poke_cpu(cpu->cpu_id);
567 		return;
568 	} else {
569 		/*
570 		 * This cpu isn't halted, but it's idle or undergoing a
571 		 * context switch. No need to awaken anyone else.
572 		 */
573 		if (cpu->cpu_thread == cpu->cpu_idle_thread ||
574 		    cpu->cpu_disp_flags & CPU_DISP_DONTSTEAL)
575 			return;
576 	}
577 
578 	/*
579 	 * No need to wake up other CPUs if this is for a bound thread.
580 	 */
581 	if (bound)
582 		return;
583 
584 	/*
585 	 * The CPU specified for wakeup isn't currently halted, so check
586 	 * to see if there are any other halted CPUs in the partition,
587 	 * and if there are then awaken one.
588 	 */
589 	do {
590 		cpu_found = bitset_find(&cp->cp_haltset);
591 		if (cpu_found == (uint_t)-1)
592 			return;
593 	} while (bitset_atomic_test_and_del(&cp->cp_haltset, cpu_found) < 0);
594 
595 	if (cpu_found != CPU->cpu_seqid) {
596 		poke_cpu(cpu_seq[cpu_found]->cpu_id);
597 	}
598 }
599 
600 #ifndef __xpv
601 /*
602  * Function called by CPU idle notification framework to check whether CPU
603  * has been awakened. It will be called with interrupt disabled.
604  * If CPU has been awakened, call cpu_idle_exit() to notify CPU idle
605  * notification framework.
606  */
607 static void
608 cpu_idle_mwait_check_wakeup(void *arg)
609 {
610 	volatile uint32_t *mcpu_mwait = (volatile uint32_t *)arg;
611 
612 	ASSERT(arg != NULL);
613 	if (*mcpu_mwait != MWAIT_HALTED) {
614 		/*
615 		 * CPU has been awakened, notify CPU idle notification system.
616 		 */
617 		cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
618 	} else {
619 		/*
620 		 * Toggle interrupt flag to detect pending interrupts.
621 		 * If interrupt happened, do_interrupt() will notify CPU idle
622 		 * notification framework so no need to call cpu_idle_exit()
623 		 * here.
624 		 */
625 		sti();
626 		SMT_PAUSE();
627 		cli();
628 	}
629 }
630 
631 /*
632  * Idle the present CPU until awakened via touching its monitored line
633  */
634 void
635 cpu_idle_mwait(void)
636 {
637 	volatile uint32_t	*mcpu_mwait = CPU->cpu_m.mcpu_mwait;
638 	cpu_t			*cpup = CPU;
639 	processorid_t		cpu_sid = cpup->cpu_seqid;
640 	cpupart_t		*cp = cpup->cpu_part;
641 	int			hset_update = 1;
642 
643 	/*
644 	 * Set our mcpu_mwait here, so we can tell if anyone tries to
645 	 * wake us between now and when we call mwait.  No other cpu will
646 	 * attempt to set our mcpu_mwait until we add ourself to the halted
647 	 * CPU bitmap.
648 	 */
649 	*mcpu_mwait = MWAIT_HALTED;
650 
651 	/*
652 	 * If this CPU is online, and there's multiple CPUs
653 	 * in the system, then we should note our halting
654 	 * by adding ourselves to the partition's halted CPU
655 	 * bitmap. This allows other CPUs to find/awaken us when
656 	 * work becomes available.
657 	 */
658 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
659 		hset_update = 0;
660 
661 	/*
662 	 * Add ourselves to the partition's halted CPUs bitmap
663 	 * and set our HALTED flag, if necessary.
664 	 *
665 	 * When a thread becomes runnable, it is placed on the queue
666 	 * and then the halted CPU bitmap is checked to determine who
667 	 * (if anyone) should be awakened. We therefore need to first
668 	 * add ourselves to the bitmap, and and then check if there
669 	 * is any work available.
670 	 *
671 	 * Note that memory barriers after updating the HALTED flag
672 	 * are not necessary since an atomic operation (updating the bitmap)
673 	 * immediately follows. On x86 the atomic operation acts as a
674 	 * memory barrier for the update of cpu_disp_flags.
675 	 */
676 	if (hset_update) {
677 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
678 		bitset_atomic_add(&cp->cp_haltset, cpu_sid);
679 	}
680 
681 	/*
682 	 * Check to make sure there's really nothing to do.
683 	 * Work destined for this CPU may become available after
684 	 * this check. We'll be notified through the clearing of our
685 	 * bit in the halted CPU bitmap, and a write to our mcpu_mwait.
686 	 *
687 	 * disp_anywork() checks disp_nrunnable, so we do not have to later.
688 	 */
689 	if (disp_anywork()) {
690 		if (hset_update) {
691 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
692 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
693 		}
694 		return;
695 	}
696 
697 	/*
698 	 * We're on our way to being halted.
699 	 * To avoid a lost wakeup, arm the monitor before checking if another
700 	 * cpu wrote to mcpu_mwait to wake us up.
701 	 */
702 	i86_monitor(mcpu_mwait, 0, 0);
703 	if (*mcpu_mwait == MWAIT_HALTED) {
704 		if (cpu_idle_enter(IDLE_STATE_C1, 0,
705 		    cpu_idle_mwait_check_wakeup, (void *)mcpu_mwait) == 0) {
706 			if (*mcpu_mwait == MWAIT_HALTED) {
707 				i86_mwait(0, 0);
708 			}
709 			cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
710 		}
711 	}
712 
713 	/*
714 	 * We're no longer halted
715 	 */
716 	if (hset_update) {
717 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
718 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
719 	}
720 }
721 
722 /*
723  * If "cpu" is halted in mwait, then wake it up clearing its halted bit in
724  * advance.  Otherwise, see if other CPUs in the cpu partition are halted and
725  * need to be woken up so that they can steal the thread we placed on this CPU.
726  * This function is only used on MP systems.
727  */
728 static void
729 cpu_wakeup_mwait(cpu_t *cp, int bound)
730 {
731 	cpupart_t	*cpu_part;
732 	uint_t		cpu_found;
733 	processorid_t	cpu_sid;
734 
735 	cpu_part = cp->cpu_part;
736 	cpu_sid = cp->cpu_seqid;
737 
738 	/*
739 	 * Clear the halted bit for that CPU since it will be woken up
740 	 * in a moment.
741 	 */
742 	if (bitset_in_set(&cpu_part->cp_haltset, cpu_sid)) {
743 		/*
744 		 * Clear the halted bit for that CPU since it will be
745 		 * poked in a moment.
746 		 */
747 		bitset_atomic_del(&cpu_part->cp_haltset, cpu_sid);
748 		/*
749 		 * We may find the current CPU present in the halted cpuset
750 		 * if we're in the context of an interrupt that occurred
751 		 * before we had a chance to clear our bit in cpu_idle().
752 		 * Waking ourself is obviously unnecessary, since if
753 		 * we're here, we're not halted.
754 		 *
755 		 * monitor/mwait wakeup via writing to our cache line is
756 		 * harmless and less expensive than always checking if we
757 		 * are waking ourself which is an uncommon case.
758 		 */
759 		MWAIT_WAKEUP(cp);	/* write to monitored line */
760 		return;
761 	} else {
762 		/*
763 		 * This cpu isn't halted, but it's idle or undergoing a
764 		 * context switch. No need to awaken anyone else.
765 		 */
766 		if (cp->cpu_thread == cp->cpu_idle_thread ||
767 		    cp->cpu_disp_flags & CPU_DISP_DONTSTEAL)
768 			return;
769 	}
770 
771 	/*
772 	 * No need to wake up other CPUs if the thread we just enqueued
773 	 * is bound.
774 	 */
775 	if (bound || ncpus == 1)
776 		return;
777 
778 	/*
779 	 * See if there's any other halted CPUs. If there are, then
780 	 * select one, and awaken it.
781 	 * It's possible that after we find a CPU, somebody else
782 	 * will awaken it before we get the chance.
783 	 * In that case, look again.
784 	 */
785 	do {
786 		cpu_found = bitset_find(&cpu_part->cp_haltset);
787 		if (cpu_found == (uint_t)-1)
788 			return;
789 	} while (bitset_atomic_test_and_del(&cpu_part->cp_haltset,
790 	    cpu_found) < 0);
791 
792 	/*
793 	 * Do not check if cpu_found is ourself as monitor/mwait
794 	 * wakeup is cheap.
795 	 */
796 	MWAIT_WAKEUP(cpu_seq[cpu_found]); /* write to monitored line */
797 }
798 
799 #endif
800 
801 void (*cpu_pause_handler)(volatile char *) = NULL;
802 
803 static int
804 mp_disable_intr(int cpun)
805 {
806 	/*
807 	 * switch to the offline cpu
808 	 */
809 	affinity_set(cpun);
810 	/*
811 	 * raise ipl to just below cross call
812 	 */
813 	splx(XC_SYS_PIL - 1);
814 	/*
815 	 *	set base spl to prevent the next swtch to idle from
816 	 *	lowering back to ipl 0
817 	 */
818 	CPU->cpu_intr_actv |= (1 << (XC_SYS_PIL - 1));
819 	set_base_spl();
820 	affinity_clear();
821 	return (DDI_SUCCESS);
822 }
823 
824 static void
825 mp_enable_intr(int cpun)
826 {
827 	/*
828 	 * switch to the online cpu
829 	 */
830 	affinity_set(cpun);
831 	/*
832 	 * clear the interrupt active mask
833 	 */
834 	CPU->cpu_intr_actv &= ~(1 << (XC_SYS_PIL - 1));
835 	set_base_spl();
836 	(void) spl0();
837 	affinity_clear();
838 }
839 
840 static void
841 mach_get_platform(int owner)
842 {
843 	void		**srv_opsp;
844 	void		**clt_opsp;
845 	int		i;
846 	int		total_ops;
847 
848 	/* fix up psm ops */
849 	srv_opsp = (void **)mach_set[0];
850 	clt_opsp = (void **)mach_set[owner];
851 	if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01)
852 		total_ops = sizeof (struct psm_ops_ver01) /
853 		    sizeof (void (*)(void));
854 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_1)
855 		/* no psm_notify_func */
856 		total_ops = OFFSETOF(struct psm_ops, psm_notify_func) /
857 		    sizeof (void (*)(void));
858 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_2)
859 		/* no psm_timer funcs */
860 		total_ops = OFFSETOF(struct psm_ops, psm_timer_reprogram) /
861 		    sizeof (void (*)(void));
862 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_3)
863 		/* no psm_preshutdown function */
864 		total_ops = OFFSETOF(struct psm_ops, psm_preshutdown) /
865 		    sizeof (void (*)(void));
866 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_4)
867 		/* no psm_preshutdown function */
868 		total_ops = OFFSETOF(struct psm_ops, psm_intr_ops) /
869 		    sizeof (void (*)(void));
870 	else
871 		total_ops = sizeof (struct psm_ops) / sizeof (void (*)(void));
872 
873 	/*
874 	 * Save the version of the PSM module, in case we need to
875 	 * behave differently based on version.
876 	 */
877 	mach_ver[0] = mach_ver[owner];
878 
879 	for (i = 0; i < total_ops; i++)
880 		if (clt_opsp[i] != NULL)
881 			srv_opsp[i] = clt_opsp[i];
882 }
883 
884 static void
885 mach_construct_info()
886 {
887 	struct psm_sw *swp;
888 	int	mach_cnt[PSM_OWN_OVERRIDE+1] = {0};
889 	int	conflict_owner = 0;
890 
891 	if (psmsw->psw_forw == psmsw)
892 		panic("No valid PSM modules found");
893 	mutex_enter(&psmsw_lock);
894 	for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
895 		if (!(swp->psw_flag & PSM_MOD_IDENTIFY))
896 			continue;
897 		mach_set[swp->psw_infop->p_owner] = swp->psw_infop->p_ops;
898 		mach_ver[swp->psw_infop->p_owner] = swp->psw_infop->p_version;
899 		mach_cnt[swp->psw_infop->p_owner]++;
900 	}
901 	mutex_exit(&psmsw_lock);
902 
903 	mach_get_platform(PSM_OWN_SYS_DEFAULT);
904 
905 	/* check to see are there any conflicts */
906 	if (mach_cnt[PSM_OWN_EXCLUSIVE] > 1)
907 		conflict_owner = PSM_OWN_EXCLUSIVE;
908 	if (mach_cnt[PSM_OWN_OVERRIDE] > 1)
909 		conflict_owner = PSM_OWN_OVERRIDE;
910 	if (conflict_owner) {
911 		/* remove all psm modules except uppc */
912 		cmn_err(CE_WARN,
913 		    "Conflicts detected on the following PSM modules:");
914 		mutex_enter(&psmsw_lock);
915 		for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
916 			if (swp->psw_infop->p_owner == conflict_owner)
917 				cmn_err(CE_WARN, "%s ",
918 				    swp->psw_infop->p_mach_idstring);
919 		}
920 		mutex_exit(&psmsw_lock);
921 		cmn_err(CE_WARN,
922 		    "Setting the system back to SINGLE processor mode!");
923 		cmn_err(CE_WARN,
924 		    "Please edit /etc/mach to remove the invalid PSM module.");
925 		return;
926 	}
927 
928 	if (mach_set[PSM_OWN_EXCLUSIVE])
929 		mach_get_platform(PSM_OWN_EXCLUSIVE);
930 
931 	if (mach_set[PSM_OWN_OVERRIDE])
932 		mach_get_platform(PSM_OWN_OVERRIDE);
933 }
934 
935 static void
936 mach_init()
937 {
938 	struct psm_ops  *pops;
939 
940 	mach_construct_info();
941 
942 	pops = mach_set[0];
943 
944 	/* register the interrupt and clock initialization rotuines */
945 	picinitf = mach_picinit;
946 	clkinitf = mach_clkinit;
947 	psm_get_clockirq = pops->psm_get_clockirq;
948 
949 	/* register the interrupt setup code */
950 	slvltovect = mach_softlvl_to_vect;
951 	addspl	= pops->psm_addspl;
952 	delspl	= pops->psm_delspl;
953 
954 	if (pops->psm_translate_irq)
955 		psm_translate_irq = pops->psm_translate_irq;
956 	if (pops->psm_intr_ops)
957 		psm_intr_ops = pops->psm_intr_ops;
958 
959 #if defined(PSMI_1_2) || defined(PSMI_1_3) || defined(PSMI_1_4)
960 	/*
961 	 * Time-of-day functionality now handled in TOD modules.
962 	 * (Warn about PSM modules that think that we're going to use
963 	 * their ops vectors.)
964 	 */
965 	if (pops->psm_tod_get)
966 		cmn_err(CE_WARN, "obsolete psm_tod_get op %p",
967 		    (void *)pops->psm_tod_get);
968 
969 	if (pops->psm_tod_set)
970 		cmn_err(CE_WARN, "obsolete psm_tod_set op %p",
971 		    (void *)pops->psm_tod_set);
972 #endif
973 
974 	if (pops->psm_notify_error) {
975 		psm_notify_error = mach_notify_error;
976 		notify_error = pops->psm_notify_error;
977 	}
978 
979 	(*pops->psm_softinit)();
980 
981 	/*
982 	 * Initialize the dispatcher's function hooks to enable CPU halting
983 	 * when idle.  Set both the deep-idle and non-deep-idle hooks.
984 	 *
985 	 * Assume we can use power saving deep-idle loop cpu_idle_adaptive.
986 	 * Platform deep-idle driver will reset our idle loop to
987 	 * non_deep_idle_cpu if power saving deep-idle feature is not available.
988 	 *
989 	 * Do not use monitor/mwait if idle_cpu_use_hlt is not set(spin idle)
990 	 * or idle_cpu_prefer_mwait is not set.
991 	 * Allocate monitor/mwait buffer for cpu0.
992 	 */
993 #ifndef __xpv
994 	non_deep_idle_disp_enq_thread = disp_enq_thread;
995 #endif
996 	if (idle_cpu_use_hlt) {
997 		idle_cpu = cpu_idle_adaptive;
998 		CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
999 #ifndef __xpv
1000 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait) {
1001 			CPU->cpu_m.mcpu_mwait = cpuid_mwait_alloc(CPU);
1002 			/*
1003 			 * Protect ourself from insane mwait size.
1004 			 */
1005 			if (CPU->cpu_m.mcpu_mwait == NULL) {
1006 #ifdef DEBUG
1007 				cmn_err(CE_NOTE, "Using hlt idle.  Cannot "
1008 				    "handle cpu 0 mwait size.");
1009 #endif
1010 				idle_cpu_prefer_mwait = 0;
1011 				CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
1012 			} else {
1013 				CPU->cpu_m.mcpu_idle_cpu = cpu_idle_mwait;
1014 			}
1015 		} else {
1016 			CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
1017 		}
1018 		non_deep_idle_cpu = CPU->cpu_m.mcpu_idle_cpu;
1019 
1020 		/*
1021 		 * Disable power saving deep idle loop?
1022 		 */
1023 		if (idle_cpu_no_deep_c) {
1024 			idle_cpu = non_deep_idle_cpu;
1025 		}
1026 #endif
1027 	}
1028 
1029 	mach_smpinit();
1030 }
1031 
1032 static void
1033 mach_smpinit(void)
1034 {
1035 	struct psm_ops  *pops;
1036 	processorid_t cpu_id;
1037 	int cnt;
1038 	cpuset_t cpumask;
1039 
1040 	pops = mach_set[0];
1041 	CPUSET_ZERO(cpumask);
1042 
1043 	cpu_id = -1;
1044 	cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
1045 	for (cnt = 0; cpu_id != -1; cnt++) {
1046 		CPUSET_ADD(cpumask, cpu_id);
1047 		cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
1048 	}
1049 
1050 	mp_cpus = cpumask;
1051 
1052 	/* MP related routines */
1053 	ap_mlsetup = pops->psm_post_cpu_start;
1054 	send_dirintf = pops->psm_send_ipi;
1055 
1056 	/* optional MP related routines */
1057 	if (pops->psm_shutdown)
1058 		psm_shutdownf = pops->psm_shutdown;
1059 	if (pops->psm_preshutdown)
1060 		psm_preshutdownf = pops->psm_preshutdown;
1061 	if (pops->psm_notify_func)
1062 		psm_notifyf = pops->psm_notify_func;
1063 	if (pops->psm_set_idlecpu)
1064 		psm_set_idle_cpuf = pops->psm_set_idlecpu;
1065 	if (pops->psm_unset_idlecpu)
1066 		psm_unset_idle_cpuf = pops->psm_unset_idlecpu;
1067 
1068 	psm_clkinit = pops->psm_clkinit;
1069 
1070 	if (pops->psm_timer_reprogram)
1071 		psm_timer_reprogram = pops->psm_timer_reprogram;
1072 
1073 	if (pops->psm_timer_enable)
1074 		psm_timer_enable = pops->psm_timer_enable;
1075 
1076 	if (pops->psm_timer_disable)
1077 		psm_timer_disable = pops->psm_timer_disable;
1078 
1079 	if (pops->psm_post_cyclic_setup)
1080 		psm_post_cyclic_setup = pops->psm_post_cyclic_setup;
1081 
1082 	if (pops->psm_state)
1083 		psm_state = pops->psm_state;
1084 
1085 	/*
1086 	 * Set these vectors here so they can be used by Suspend/Resume
1087 	 * on UP machines.
1088 	 */
1089 	if (pops->psm_disable_intr)
1090 		psm_disable_intr = pops->psm_disable_intr;
1091 	if (pops->psm_enable_intr)
1092 		psm_enable_intr  = pops->psm_enable_intr;
1093 
1094 	/* check for multiple CPUs */
1095 	if (cnt < 2)
1096 		return;
1097 
1098 	/* check for MP platforms */
1099 	if (pops->psm_cpu_start == NULL)
1100 		return;
1101 
1102 	/*
1103 	 * Set the dispatcher hook to enable cpu "wake up"
1104 	 * when a thread becomes runnable.
1105 	 */
1106 	if (idle_cpu_use_hlt) {
1107 		disp_enq_thread = cpu_wakeup;
1108 #ifndef __xpv
1109 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait)
1110 			disp_enq_thread = cpu_wakeup_mwait;
1111 		non_deep_idle_disp_enq_thread = disp_enq_thread;
1112 #endif
1113 	}
1114 
1115 	psm_get_ipivect = pops->psm_get_ipivect;
1116 
1117 	(void) add_avintr((void *)NULL, XC_HI_PIL, xc_serv, "xc_intr",
1118 	    (*pops->psm_get_ipivect)(XC_HI_PIL, PSM_INTR_IPI_HI),
1119 	    NULL, NULL, NULL, NULL);
1120 
1121 	(void) (*pops->psm_get_ipivect)(XC_CPUPOKE_PIL, PSM_INTR_POKE);
1122 }
1123 
1124 static void
1125 mach_picinit()
1126 {
1127 	struct psm_ops  *pops;
1128 
1129 	pops = mach_set[0];
1130 
1131 	/* register the interrupt handlers */
1132 	setlvl = pops->psm_intr_enter;
1133 	setlvlx = pops->psm_intr_exit;
1134 
1135 	/* initialize the interrupt hardware */
1136 	(*pops->psm_picinit)();
1137 
1138 	/* set interrupt mask for current ipl */
1139 	setspl = pops->psm_setspl;
1140 	cli();
1141 	setspl(CPU->cpu_pri);
1142 }
1143 
1144 uint_t	cpu_freq;	/* MHz */
1145 uint64_t cpu_freq_hz;	/* measured (in hertz) */
1146 
1147 #define	MEGA_HZ		1000000
1148 
1149 #ifdef __xpv
1150 
1151 int xpv_cpufreq_workaround = 1;
1152 int xpv_cpufreq_verbose = 0;
1153 
1154 #else	/* __xpv */
1155 
1156 static uint64_t
1157 mach_calchz(uint32_t pit_counter, uint64_t *processor_clks)
1158 {
1159 	uint64_t cpu_hz;
1160 
1161 	if ((pit_counter == 0) || (*processor_clks == 0) ||
1162 	    (*processor_clks > (((uint64_t)-1) / PIT_HZ)))
1163 		return (0);
1164 
1165 	cpu_hz = ((uint64_t)PIT_HZ * *processor_clks) / pit_counter;
1166 
1167 	return (cpu_hz);
1168 }
1169 
1170 #endif	/* __xpv */
1171 
1172 static uint64_t
1173 mach_getcpufreq(void)
1174 {
1175 #if defined(__xpv)
1176 	vcpu_time_info_t *vti = &CPU->cpu_m.mcpu_vcpu_info->time;
1177 	uint64_t cpu_hz;
1178 
1179 	/*
1180 	 * During dom0 bringup, it was noted that on at least one older
1181 	 * Intel HT machine, the hypervisor initially gives a tsc_to_system_mul
1182 	 * value that is quite wrong (the 3.06GHz clock was reported
1183 	 * as 4.77GHz)
1184 	 *
1185 	 * The curious thing is, that if you stop the kernel at entry,
1186 	 * breakpoint here and inspect the value with kmdb, the value
1187 	 * is correct - but if you don't stop and simply enable the
1188 	 * printf statement (below), you can see the bad value printed
1189 	 * here.  Almost as if something kmdb did caused the hypervisor to
1190 	 * figure it out correctly.  And, note that the hypervisor
1191 	 * eventually -does- figure it out correctly ... if you look at
1192 	 * the field later in the life of dom0, it is correct.
1193 	 *
1194 	 * For now, on dom0, we employ a slightly cheesy workaround of
1195 	 * using the DOM0_PHYSINFO hypercall.
1196 	 */
1197 	if (DOMAIN_IS_INITDOMAIN(xen_info) && xpv_cpufreq_workaround) {
1198 		cpu_hz = 1000 * xpv_cpu_khz();
1199 	} else {
1200 		cpu_hz = (UINT64_C(1000000000) << 32) / vti->tsc_to_system_mul;
1201 
1202 		if (vti->tsc_shift < 0)
1203 			cpu_hz <<= -vti->tsc_shift;
1204 		else
1205 			cpu_hz >>= vti->tsc_shift;
1206 	}
1207 
1208 	if (xpv_cpufreq_verbose)
1209 		printf("mach_getcpufreq: system_mul 0x%x, shift %d, "
1210 		    "cpu_hz %" PRId64 "Hz\n",
1211 		    vti->tsc_to_system_mul, vti->tsc_shift, cpu_hz);
1212 
1213 	return (cpu_hz);
1214 #else	/* __xpv */
1215 	uint32_t pit_counter;
1216 	uint64_t processor_clks;
1217 
1218 	if (x86_feature & X86_TSC) {
1219 		/*
1220 		 * We have a TSC. freq_tsc() knows how to measure the number
1221 		 * of clock cycles sampled against the PIT.
1222 		 */
1223 		ulong_t flags = clear_int_flag();
1224 		processor_clks = freq_tsc(&pit_counter);
1225 		restore_int_flag(flags);
1226 		return (mach_calchz(pit_counter, &processor_clks));
1227 	} else if (x86_vendor == X86_VENDOR_Cyrix || x86_type == X86_TYPE_P5) {
1228 #if defined(__amd64)
1229 		panic("mach_getcpufreq: no TSC!");
1230 #elif defined(__i386)
1231 		/*
1232 		 * We are a Cyrix based on a 6x86 core or an Intel Pentium
1233 		 * for which freq_notsc() knows how to measure the number of
1234 		 * elapsed clock cycles sampled against the PIT
1235 		 */
1236 		ulong_t flags = clear_int_flag();
1237 		processor_clks = freq_notsc(&pit_counter);
1238 		restore_int_flag(flags);
1239 		return (mach_calchz(pit_counter, &processor_clks));
1240 #endif	/* __i386 */
1241 	}
1242 
1243 	/* We do not know how to calculate cpu frequency for this cpu. */
1244 	return (0);
1245 #endif	/* __xpv */
1246 }
1247 
1248 /*
1249  * If the clock speed of a cpu is found to be reported incorrectly, do not add
1250  * to this array, instead improve the accuracy of the algorithm that determines
1251  * the clock speed of the processor or extend the implementation to support the
1252  * vendor as appropriate. This is here only to support adjusting the speed on
1253  * older slower processors that mach_fixcpufreq() would not be able to account
1254  * for otherwise.
1255  */
1256 static int x86_cpu_freq[] = { 60, 75, 80, 90, 120, 160, 166, 175, 180, 233 };
1257 
1258 /*
1259  * On fast processors the clock frequency that is measured may be off by
1260  * a few MHz from the value printed on the part. This is a combination of
1261  * the factors that for such fast parts being off by this much is within
1262  * the tolerances for manufacture and because of the difficulties in the
1263  * measurement that can lead to small error. This function uses some
1264  * heuristics in order to tweak the value that was measured to match what
1265  * is most likely printed on the part.
1266  *
1267  * Some examples:
1268  * 	AMD Athlon 1000 mhz measured as 998 mhz
1269  * 	Intel Pentium III Xeon 733 mhz measured as 731 mhz
1270  * 	Intel Pentium IV 1500 mhz measured as 1495mhz
1271  *
1272  * If in the future this function is no longer sufficient to correct
1273  * for the error in the measurement, then the algorithm used to perform
1274  * the measurement will have to be improved in order to increase accuracy
1275  * rather than adding horrible and questionable kludges here.
1276  *
1277  * This is called after the cyclics subsystem because of the potential
1278  * that the heuristics within may give a worse estimate of the clock
1279  * frequency than the value that was measured.
1280  */
1281 static void
1282 mach_fixcpufreq(void)
1283 {
1284 	uint32_t freq, mul, near66, delta66, near50, delta50, fixed, delta, i;
1285 
1286 	freq = (uint32_t)cpu_freq;
1287 
1288 	/*
1289 	 * Find the nearest integer multiple of 200/3 (about 66) MHz to the
1290 	 * measured speed taking into account that the 667 MHz parts were
1291 	 * the first to round-up.
1292 	 */
1293 	mul = (uint32_t)((3 * (uint64_t)freq + 100) / 200);
1294 	near66 = (uint32_t)((200 * (uint64_t)mul + ((mul >= 10) ? 1 : 0)) / 3);
1295 	delta66 = (near66 > freq) ? (near66 - freq) : (freq - near66);
1296 
1297 	/* Find the nearest integer multiple of 50 MHz to the measured speed */
1298 	mul = (freq + 25) / 50;
1299 	near50 = mul * 50;
1300 	delta50 = (near50 > freq) ? (near50 - freq) : (freq - near50);
1301 
1302 	/* Find the closer of the two */
1303 	if (delta66 < delta50) {
1304 		fixed = near66;
1305 		delta = delta66;
1306 	} else {
1307 		fixed = near50;
1308 		delta = delta50;
1309 	}
1310 
1311 	if (fixed > INT_MAX)
1312 		return;
1313 
1314 	/*
1315 	 * Some older parts have a core clock frequency that is not an
1316 	 * integral multiple of 50 or 66 MHz. Check if one of the old
1317 	 * clock frequencies is closer to the measured value than any
1318 	 * of the integral multiples of 50 an 66, and if so set fixed
1319 	 * and delta appropriately to represent the closest value.
1320 	 */
1321 	i = sizeof (x86_cpu_freq) / sizeof (int);
1322 	while (i > 0) {
1323 		i--;
1324 
1325 		if (x86_cpu_freq[i] <= freq) {
1326 			mul = freq - x86_cpu_freq[i];
1327 
1328 			if (mul < delta) {
1329 				fixed = x86_cpu_freq[i];
1330 				delta = mul;
1331 			}
1332 
1333 			break;
1334 		}
1335 
1336 		mul = x86_cpu_freq[i] - freq;
1337 
1338 		if (mul < delta) {
1339 			fixed = x86_cpu_freq[i];
1340 			delta = mul;
1341 		}
1342 	}
1343 
1344 	/*
1345 	 * Set a reasonable maximum for how much to correct the measured
1346 	 * result by. This check is here to prevent the adjustment made
1347 	 * by this function from being more harm than good. It is entirely
1348 	 * possible that in the future parts will be made that are not
1349 	 * integral multiples of 66 or 50 in clock frequency or that
1350 	 * someone may overclock a part to some odd frequency. If the
1351 	 * measured value is farther from the corrected value than
1352 	 * allowed, then assume the corrected value is in error and use
1353 	 * the measured value.
1354 	 */
1355 	if (6 < delta)
1356 		return;
1357 
1358 	cpu_freq = (int)fixed;
1359 }
1360 
1361 
1362 static int
1363 machhztomhz(uint64_t cpu_freq_hz)
1364 {
1365 	uint64_t cpu_mhz;
1366 
1367 	/* Round to nearest MHZ */
1368 	cpu_mhz = (cpu_freq_hz + (MEGA_HZ / 2)) / MEGA_HZ;
1369 
1370 	if (cpu_mhz > INT_MAX)
1371 		return (0);
1372 
1373 	return ((int)cpu_mhz);
1374 
1375 }
1376 
1377 
1378 static int
1379 mach_clkinit(int preferred_mode, int *set_mode)
1380 {
1381 	struct psm_ops  *pops;
1382 	int resolution;
1383 
1384 	pops = mach_set[0];
1385 
1386 	cpu_freq_hz = mach_getcpufreq();
1387 
1388 	cpu_freq = machhztomhz(cpu_freq_hz);
1389 
1390 	if (!(x86_feature & X86_TSC) || (cpu_freq == 0))
1391 		tsc_gethrtime_enable = 0;
1392 
1393 #ifndef __xpv
1394 	if (tsc_gethrtime_enable) {
1395 		tsc_hrtimeinit(cpu_freq_hz);
1396 	} else
1397 #endif
1398 	{
1399 		if (pops->psm_hrtimeinit)
1400 			(*pops->psm_hrtimeinit)();
1401 		gethrtimef = pops->psm_gethrtime;
1402 		gethrtimeunscaledf = gethrtimef;
1403 		/* scalehrtimef will remain dummy */
1404 	}
1405 
1406 	mach_fixcpufreq();
1407 
1408 	if (mach_ver[0] >= PSM_INFO_VER01_3) {
1409 		if (preferred_mode == TIMER_ONESHOT) {
1410 
1411 			resolution = (*pops->psm_clkinit)(0);
1412 			if (resolution != 0)  {
1413 				*set_mode = TIMER_ONESHOT;
1414 				return (resolution);
1415 			}
1416 		}
1417 
1418 		/*
1419 		 * either periodic mode was requested or could not set to
1420 		 * one-shot mode
1421 		 */
1422 		resolution = (*pops->psm_clkinit)(hz);
1423 		/*
1424 		 * psm should be able to do periodic, so we do not check
1425 		 * for return value of psm_clkinit here.
1426 		 */
1427 		*set_mode = TIMER_PERIODIC;
1428 		return (resolution);
1429 	} else {
1430 		/*
1431 		 * PSMI interface prior to PSMI_3 does not define a return
1432 		 * value for psm_clkinit, so the return value is ignored.
1433 		 */
1434 		(void) (*pops->psm_clkinit)(hz);
1435 		*set_mode = TIMER_PERIODIC;
1436 		return (nsec_per_tick);
1437 	}
1438 }
1439 
1440 
1441 /*ARGSUSED*/
1442 static int
1443 mach_softlvl_to_vect(int ipl)
1444 {
1445 	setsoftint = av_set_softint_pending;
1446 	kdisetsoftint = kdi_av_set_softint_pending;
1447 
1448 	return (PSM_SV_SOFTWARE);
1449 }
1450 
1451 #ifdef DEBUG
1452 /*
1453  * This is here to allow us to simulate cpus that refuse to start.
1454  */
1455 cpuset_t cpufailset;
1456 #endif
1457 
1458 int
1459 mach_cpu_start(struct cpu *cp, void *ctx)
1460 {
1461 	struct psm_ops *pops = mach_set[0];
1462 	processorid_t id = cp->cpu_id;
1463 
1464 #ifdef DEBUG
1465 	if (CPU_IN_SET(cpufailset, id))
1466 		return (0);
1467 #endif
1468 	return ((*pops->psm_cpu_start)(id, ctx));
1469 }
1470 
1471 int
1472 mach_cpuid_start(processorid_t id, void *ctx)
1473 {
1474 	struct psm_ops *pops = mach_set[0];
1475 
1476 #ifdef DEBUG
1477 	if (CPU_IN_SET(cpufailset, id))
1478 		return (0);
1479 #endif
1480 	return ((*pops->psm_cpu_start)(id, ctx));
1481 }
1482 
1483 /*
1484  * Default handler to create device node for CPU.
1485  * One reference count will be held on created device node.
1486  */
1487 static int
1488 mach_cpu_create_devinfo(cpu_t *cp, dev_info_t **dipp)
1489 {
1490 	int rv, circ;
1491 	dev_info_t *dip;
1492 	static kmutex_t cpu_node_lock;
1493 	static dev_info_t *cpu_nex_devi = NULL;
1494 
1495 	ASSERT(cp != NULL);
1496 	ASSERT(dipp != NULL);
1497 	*dipp = NULL;
1498 
1499 	if (cpu_nex_devi == NULL) {
1500 		mutex_enter(&cpu_node_lock);
1501 		/* First check whether cpus exists. */
1502 		cpu_nex_devi = ddi_find_devinfo("cpus", -1, 0);
1503 		/* Create cpus if it doesn't exist. */
1504 		if (cpu_nex_devi == NULL) {
1505 			ndi_devi_enter(ddi_root_node(), &circ);
1506 			rv = ndi_devi_alloc(ddi_root_node(), "cpus",
1507 			    (pnode_t)DEVI_SID_NODEID, &dip);
1508 			if (rv != NDI_SUCCESS) {
1509 				mutex_exit(&cpu_node_lock);
1510 				cmn_err(CE_CONT,
1511 				    "?failed to create cpu nexus device.\n");
1512 				return (PSM_FAILURE);
1513 			}
1514 			ASSERT(dip != NULL);
1515 			(void) ndi_devi_online(dip, 0);
1516 			ndi_devi_exit(ddi_root_node(), circ);
1517 			cpu_nex_devi = dip;
1518 		}
1519 		mutex_exit(&cpu_node_lock);
1520 	}
1521 
1522 	/*
1523 	 * create a child node for cpu identified as 'cpu_id'
1524 	 */
1525 	ndi_devi_enter(cpu_nex_devi, &circ);
1526 	dip = ddi_add_child(cpu_nex_devi, "cpu", DEVI_SID_NODEID, -1);
1527 	if (dip == NULL) {
1528 		cmn_err(CE_CONT,
1529 		    "?failed to create device node for cpu%d.\n", cp->cpu_id);
1530 		rv = PSM_FAILURE;
1531 	} else {
1532 		*dipp = dip;
1533 		(void) ndi_hold_devi(dip);
1534 		rv = PSM_SUCCESS;
1535 	}
1536 	ndi_devi_exit(cpu_nex_devi, circ);
1537 
1538 	return (rv);
1539 }
1540 
1541 /*
1542  * Create cpu device node in device tree and online it.
1543  * Return created dip with reference count held if requested.
1544  */
1545 int
1546 mach_cpu_create_device_node(struct cpu *cp, dev_info_t **dipp)
1547 {
1548 	int rv;
1549 	dev_info_t *dip = NULL;
1550 
1551 	ASSERT(psm_cpu_create_devinfo != NULL);
1552 	rv = psm_cpu_create_devinfo(cp, &dip);
1553 	if (rv == PSM_SUCCESS) {
1554 		cpuid_set_cpu_properties(dip, cp->cpu_id, cp->cpu_m.mcpu_cpi);
1555 		/* Recursively attach driver for parent nexus device. */
1556 		if (i_ddi_attach_node_hierarchy(ddi_get_parent(dip)) ==
1557 		    DDI_SUCCESS) {
1558 			/* Configure cpu itself and descendants. */
1559 			(void) ndi_devi_online(dip,
1560 			    NDI_ONLINE_ATTACH | NDI_CONFIG);
1561 		}
1562 		if (dipp != NULL) {
1563 			*dipp = dip;
1564 		} else {
1565 			(void) ndi_rele_devi(dip);
1566 		}
1567 	}
1568 
1569 	return (rv);
1570 }
1571 
1572 /*ARGSUSED*/
1573 static int
1574 mach_translate_irq(dev_info_t *dip, int irqno)
1575 {
1576 	return (irqno);	/* default to NO translation */
1577 }
1578 
1579 static void
1580 mach_notify_error(int level, char *errmsg)
1581 {
1582 	/*
1583 	 * SL_FATAL is pass in once panicstr is set, deliver it
1584 	 * as CE_PANIC.  Also, translate SL_ codes back to CE_
1585 	 * codes for the psmi handler
1586 	 */
1587 	if (level & SL_FATAL)
1588 		(*notify_error)(CE_PANIC, errmsg);
1589 	else if (level & SL_WARN)
1590 		(*notify_error)(CE_WARN, errmsg);
1591 	else if (level & SL_NOTE)
1592 		(*notify_error)(CE_NOTE, errmsg);
1593 	else if (level & SL_CONSOLE)
1594 		(*notify_error)(CE_CONT, errmsg);
1595 }
1596 
1597 /*
1598  * It provides the default basic intr_ops interface for the new DDI
1599  * interrupt framework if the PSM doesn't have one.
1600  *
1601  * Input:
1602  * dip     - pointer to the dev_info structure of the requested device
1603  * hdlp    - pointer to the internal interrupt handle structure for the
1604  *	     requested interrupt
1605  * intr_op - opcode for this call
1606  * result  - pointer to the integer that will hold the result to be
1607  *	     passed back if return value is PSM_SUCCESS
1608  *
1609  * Output:
1610  * return value is either PSM_SUCCESS or PSM_FAILURE
1611  */
1612 static int
1613 mach_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1614     psm_intr_op_t intr_op, int *result)
1615 {
1616 	struct intrspec *ispec;
1617 
1618 	switch (intr_op) {
1619 	case PSM_INTR_OP_CHECK_MSI:
1620 		*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1621 		    DDI_INTR_TYPE_MSIX);
1622 		break;
1623 	case PSM_INTR_OP_ALLOC_VECTORS:
1624 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1625 			*result = 1;
1626 		else
1627 			*result = 0;
1628 		break;
1629 	case PSM_INTR_OP_FREE_VECTORS:
1630 		break;
1631 	case PSM_INTR_OP_NAVAIL_VECTORS:
1632 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1633 			*result = 1;
1634 		else
1635 			*result = 0;
1636 		break;
1637 	case PSM_INTR_OP_XLATE_VECTOR:
1638 		ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1639 		*result = psm_translate_irq(dip, ispec->intrspec_vec);
1640 		break;
1641 	case PSM_INTR_OP_GET_CAP:
1642 		*result = 0;
1643 		break;
1644 	case PSM_INTR_OP_GET_PENDING:
1645 	case PSM_INTR_OP_CLEAR_MASK:
1646 	case PSM_INTR_OP_SET_MASK:
1647 	case PSM_INTR_OP_GET_SHARED:
1648 	case PSM_INTR_OP_SET_PRI:
1649 	case PSM_INTR_OP_SET_CAP:
1650 	case PSM_INTR_OP_SET_CPU:
1651 	case PSM_INTR_OP_GET_INTR:
1652 	default:
1653 		return (PSM_FAILURE);
1654 	}
1655 	return (PSM_SUCCESS);
1656 }
1657 /*
1658  * Return 1 if CMT load balancing policies should be
1659  * implemented across instances of the specified hardware
1660  * sharing relationship.
1661  */
1662 int
1663 pg_cmt_load_bal_hw(pghw_type_t hw)
1664 {
1665 	if (hw == PGHW_IPIPE ||
1666 	    hw == PGHW_FPU ||
1667 	    hw == PGHW_PROCNODE ||
1668 	    hw == PGHW_CHIP)
1669 		return (1);
1670 	else
1671 		return (0);
1672 }
1673 /*
1674  * Return 1 if thread affinity polices should be implemented
1675  * for instances of the specifed hardware sharing relationship.
1676  */
1677 int
1678 pg_cmt_affinity_hw(pghw_type_t hw)
1679 {
1680 	if (hw == PGHW_CACHE)
1681 		return (1);
1682 	else
1683 		return (0);
1684 }
1685 
1686 /*
1687  * Return number of counter events requested to measure hardware capacity and
1688  * utilization and setup CPC requests for specified CPU as needed
1689  *
1690  * May return 0 when platform or processor specific code knows that no CPC
1691  * events should be programmed on this CPU or -1 when platform or processor
1692  * specific code doesn't know which counter events are best to use and common
1693  * code should decide for itself
1694  */
1695 int
1696 /* LINTED E_FUNC_ARG_UNUSED */
1697 cu_plat_cpc_init(cpu_t *cp, kcpc_request_list_t *reqs, int nreqs)
1698 {
1699 	const char	*impl_name;
1700 
1701 	/*
1702 	 * Return error if pcbe_ops not set
1703 	 */
1704 	if (pcbe_ops == NULL)
1705 		return (-1);
1706 
1707 	/*
1708 	 * Return that no CPC events should be programmed on hyperthreaded
1709 	 * Pentium 4 and return error for all other x86 processors to tell
1710 	 * common code to decide what counter events to program on those CPUs
1711 	 * for measuring hardware capacity and utilization
1712 	 */
1713 	impl_name = pcbe_ops->pcbe_impl_name();
1714 	if (impl_name != NULL && strcmp(impl_name, PCBE_IMPL_NAME_P4HT) == 0)
1715 		return (0);
1716 	else
1717 		return (-1);
1718 }
1719