xref: /titanic_41/usr/src/uts/i86pc/os/mp_machdep.c (revision bafec74292ca6805e5acb387856f4e60a5314b37)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /*
26  * Copyright (c) 2009, Intel Corporation.
27  * All rights reserved.
28  */
29 
30 #define	PSMI_1_6
31 #include <sys/smp_impldefs.h>
32 #include <sys/psm.h>
33 #include <sys/psm_modctl.h>
34 #include <sys/pit.h>
35 #include <sys/cmn_err.h>
36 #include <sys/strlog.h>
37 #include <sys/clock.h>
38 #include <sys/debug.h>
39 #include <sys/rtc.h>
40 #include <sys/x86_archext.h>
41 #include <sys/cpupart.h>
42 #include <sys/cpuvar.h>
43 #include <sys/cpu_event.h>
44 #include <sys/cmt.h>
45 #include <sys/cpu.h>
46 #include <sys/disp.h>
47 #include <sys/archsystm.h>
48 #include <sys/machsystm.h>
49 #include <sys/sysmacros.h>
50 #include <sys/memlist.h>
51 #include <sys/param.h>
52 #include <sys/promif.h>
53 #include <sys/cpu_pm.h>
54 #if defined(__xpv)
55 #include <sys/hypervisor.h>
56 #endif
57 #include <sys/mach_intr.h>
58 #include <vm/hat_i86.h>
59 #include <sys/kdi_machimpl.h>
60 #include <sys/sdt.h>
61 #include <sys/hpet.h>
62 #include <sys/sunddi.h>
63 #include <sys/sunndi.h>
64 
65 #define	OFFSETOF(s, m)		(size_t)(&(((s *)0)->m))
66 
67 /*
68  *	Local function prototypes
69  */
70 static int mp_disable_intr(processorid_t cpun);
71 static void mp_enable_intr(processorid_t cpun);
72 static void mach_init();
73 static void mach_picinit();
74 static int machhztomhz(uint64_t cpu_freq_hz);
75 static uint64_t mach_getcpufreq(void);
76 static void mach_fixcpufreq(void);
77 static int mach_clkinit(int, int *);
78 static void mach_smpinit(void);
79 static int mach_softlvl_to_vect(int ipl);
80 static void mach_get_platform(int owner);
81 static void mach_construct_info();
82 static int mach_translate_irq(dev_info_t *dip, int irqno);
83 static int mach_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
84     psm_intr_op_t, int *);
85 static void mach_notify_error(int level, char *errmsg);
86 static hrtime_t dummy_hrtime(void);
87 static void dummy_scalehrtime(hrtime_t *);
88 static uint64_t dummy_unscalehrtime(hrtime_t);
89 void cpu_idle(void);
90 static void cpu_wakeup(cpu_t *, int);
91 #ifndef __xpv
92 void cpu_idle_mwait(void);
93 static void cpu_wakeup_mwait(cpu_t *, int);
94 #endif
95 static int mach_cpu_create_devinfo(cpu_t *cp, dev_info_t **dipp);
96 
97 /*
98  *	External reference functions
99  */
100 extern void return_instr();
101 extern uint64_t freq_tsc(uint32_t *);
102 #if defined(__i386)
103 extern uint64_t freq_notsc(uint32_t *);
104 #endif
105 extern void pc_gethrestime(timestruc_t *);
106 extern int cpuid_get_coreid(cpu_t *);
107 extern int cpuid_get_chipid(cpu_t *);
108 
109 /*
110  *	PSM functions initialization
111  */
112 void (*psm_shutdownf)(int, int)	= (void (*)(int, int))return_instr;
113 void (*psm_preshutdownf)(int, int) = (void (*)(int, int))return_instr;
114 void (*psm_notifyf)(int)	= (void (*)(int))return_instr;
115 void (*psm_set_idle_cpuf)(int)	= (void (*)(int))return_instr;
116 void (*psm_unset_idle_cpuf)(int) = (void (*)(int))return_instr;
117 void (*psminitf)()		= mach_init;
118 void (*picinitf)() 		= return_instr;
119 int (*clkinitf)(int, int *) 	= (int (*)(int, int *))return_instr;
120 int (*ap_mlsetup)() 		= (int (*)(void))return_instr;
121 void (*send_dirintf)() 		= return_instr;
122 void (*setspl)(int)		= (void (*)(int))return_instr;
123 int (*addspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
124 int (*delspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
125 void (*kdisetsoftint)(int, struct av_softinfo *)=
126 	(void (*)(int, struct av_softinfo *))return_instr;
127 void (*setsoftint)(int, struct av_softinfo *)=
128 	(void (*)(int, struct av_softinfo *))return_instr;
129 int (*slvltovect)(int)		= (int (*)(int))return_instr;
130 int (*setlvl)(int, int *)	= (int (*)(int, int *))return_instr;
131 void (*setlvlx)(int, int)	= (void (*)(int, int))return_instr;
132 int (*psm_disable_intr)(int)	= mp_disable_intr;
133 void (*psm_enable_intr)(int)	= mp_enable_intr;
134 hrtime_t (*gethrtimef)(void)	= dummy_hrtime;
135 hrtime_t (*gethrtimeunscaledf)(void)	= dummy_hrtime;
136 void (*scalehrtimef)(hrtime_t *)	= dummy_scalehrtime;
137 uint64_t (*unscalehrtimef)(hrtime_t)	= dummy_unscalehrtime;
138 int (*psm_translate_irq)(dev_info_t *, int) = mach_translate_irq;
139 void (*gethrestimef)(timestruc_t *) = pc_gethrestime;
140 void (*psm_notify_error)(int, char *) = (void (*)(int, char *))NULL;
141 int (*psm_get_clockirq)(int) = NULL;
142 int (*psm_get_ipivect)(int, int) = NULL;
143 
144 int (*psm_clkinit)(int) = NULL;
145 void (*psm_timer_reprogram)(hrtime_t) = NULL;
146 void (*psm_timer_enable)(void) = NULL;
147 void (*psm_timer_disable)(void) = NULL;
148 void (*psm_post_cyclic_setup)(void *arg) = NULL;
149 int (*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *, psm_intr_op_t,
150     int *) = mach_intr_ops;
151 int (*psm_state)(psm_state_request_t *) = (int (*)(psm_state_request_t *))
152     return_instr;
153 
154 void (*notify_error)(int, char *) = (void (*)(int, char *))return_instr;
155 void (*hrtime_tick)(void)	= return_instr;
156 
157 int (*psm_cpu_create_devinfo)(cpu_t *, dev_info_t **) = mach_cpu_create_devinfo;
158 
159 /*
160  * True if the generic TSC code is our source of hrtime, rather than whatever
161  * the PSM can provide.
162  */
163 #ifdef __xpv
164 int tsc_gethrtime_enable = 0;
165 #else
166 int tsc_gethrtime_enable = 1;
167 #endif
168 int tsc_gethrtime_initted = 0;
169 
170 /*
171  * True if the hrtime implementation is "hires"; namely, better than microdata.
172  */
173 int gethrtime_hires = 0;
174 
175 /*
176  * Local Static Data
177  */
178 static struct psm_ops mach_ops;
179 static struct psm_ops *mach_set[4] = {&mach_ops, NULL, NULL, NULL};
180 static ushort_t mach_ver[4] = {0, 0, 0, 0};
181 
182 /*
183  * virtualization support for psm
184  */
185 void *psm_vt_ops = NULL;
186 /*
187  * If non-zero, idle cpus will become "halted" when there's
188  * no work to do.
189  */
190 int	idle_cpu_use_hlt = 1;
191 
192 #ifndef __xpv
193 /*
194  * If non-zero, idle cpus will use mwait if available to halt instead of hlt.
195  */
196 int	idle_cpu_prefer_mwait = 1;
197 /*
198  * Set to 0 to avoid MONITOR+CLFLUSH assertion.
199  */
200 int	idle_cpu_assert_cflush_monitor = 1;
201 
202 /*
203  * If non-zero, idle cpus will not use power saving Deep C-States idle loop.
204  */
205 int	idle_cpu_no_deep_c = 0;
206 /*
207  * Non-power saving idle loop and wakeup pointers.
208  * Allows user to toggle Deep Idle power saving feature on/off.
209  */
210 void	(*non_deep_idle_cpu)() = cpu_idle;
211 void	(*non_deep_idle_disp_enq_thread)(cpu_t *, int);
212 
213 /*
214  * Object for the kernel to access the HPET.
215  */
216 hpet_t hpet;
217 
218 #endif	/* ifndef __xpv */
219 
220 /*ARGSUSED*/
221 int
222 pg_plat_hw_shared(cpu_t *cp, pghw_type_t hw)
223 {
224 	switch (hw) {
225 	case PGHW_IPIPE:
226 		if (x86_feature & (X86_HTT)) {
227 			/*
228 			 * Hyper-threading is SMT
229 			 */
230 			return (1);
231 		} else {
232 			return (0);
233 		}
234 	case PGHW_PROCNODE:
235 		if (cpuid_get_procnodes_per_pkg(cp) > 1)
236 			return (1);
237 		else
238 			return (0);
239 	case PGHW_CHIP:
240 		if (x86_feature & (X86_CMP|X86_HTT))
241 			return (1);
242 		else
243 			return (0);
244 	case PGHW_CACHE:
245 		if (cpuid_get_ncpu_sharing_last_cache(cp) > 1)
246 			return (1);
247 		else
248 			return (0);
249 	case PGHW_POW_ACTIVE:
250 		if (cpupm_domain_id(cp, CPUPM_DTYPE_ACTIVE) != (id_t)-1)
251 			return (1);
252 		else
253 			return (0);
254 	case PGHW_POW_IDLE:
255 		if (cpupm_domain_id(cp, CPUPM_DTYPE_IDLE) != (id_t)-1)
256 			return (1);
257 		else
258 			return (0);
259 	default:
260 		return (0);
261 	}
262 }
263 
264 /*
265  * Compare two CPUs and see if they have a pghw_type_t sharing relationship
266  * If pghw_type_t is an unsupported hardware type, then return -1
267  */
268 int
269 pg_plat_cpus_share(cpu_t *cpu_a, cpu_t *cpu_b, pghw_type_t hw)
270 {
271 	id_t pgp_a, pgp_b;
272 
273 	pgp_a = pg_plat_hw_instance_id(cpu_a, hw);
274 	pgp_b = pg_plat_hw_instance_id(cpu_b, hw);
275 
276 	if (pgp_a == -1 || pgp_b == -1)
277 		return (-1);
278 
279 	return (pgp_a == pgp_b);
280 }
281 
282 /*
283  * Return a physical instance identifier for known hardware sharing
284  * relationships
285  */
286 id_t
287 pg_plat_hw_instance_id(cpu_t *cpu, pghw_type_t hw)
288 {
289 	switch (hw) {
290 	case PGHW_IPIPE:
291 		return (cpuid_get_coreid(cpu));
292 	case PGHW_CACHE:
293 		return (cpuid_get_last_lvl_cacheid(cpu));
294 	case PGHW_PROCNODE:
295 		return (cpuid_get_procnodeid(cpu));
296 	case PGHW_CHIP:
297 		return (cpuid_get_chipid(cpu));
298 	case PGHW_POW_ACTIVE:
299 		return (cpupm_domain_id(cpu, CPUPM_DTYPE_ACTIVE));
300 	case PGHW_POW_IDLE:
301 		return (cpupm_domain_id(cpu, CPUPM_DTYPE_IDLE));
302 	default:
303 		return (-1);
304 	}
305 }
306 
307 /*
308  * Express preference for optimizing for sharing relationship
309  * hw1 vs hw2
310  */
311 pghw_type_t
312 pg_plat_hw_rank(pghw_type_t hw1, pghw_type_t hw2)
313 {
314 	int i, rank1, rank2;
315 
316 	static pghw_type_t hw_hier[] = {
317 		PGHW_IPIPE,
318 		PGHW_CACHE,
319 		PGHW_PROCNODE,
320 		PGHW_CHIP,
321 		PGHW_POW_IDLE,
322 		PGHW_POW_ACTIVE,
323 		PGHW_NUM_COMPONENTS
324 	};
325 
326 	for (i = 0; hw_hier[i] != PGHW_NUM_COMPONENTS; i++) {
327 		if (hw_hier[i] == hw1)
328 			rank1 = i;
329 		if (hw_hier[i] == hw2)
330 			rank2 = i;
331 	}
332 
333 	if (rank1 > rank2)
334 		return (hw1);
335 	else
336 		return (hw2);
337 }
338 
339 /*
340  * Override the default CMT dispatcher policy for the specified
341  * hardware sharing relationship
342  */
343 pg_cmt_policy_t
344 pg_plat_cmt_policy(pghw_type_t hw)
345 {
346 	/*
347 	 * For shared caches, also load balance across them to
348 	 * maximize aggregate cache capacity
349 	 */
350 	switch (hw) {
351 	case PGHW_CACHE:
352 		return (CMT_BALANCE|CMT_AFFINITY);
353 	default:
354 		return (CMT_NO_POLICY);
355 	}
356 }
357 
358 id_t
359 pg_plat_get_core_id(cpu_t *cpu)
360 {
361 	return ((id_t)cpuid_get_coreid(cpu));
362 }
363 
364 void
365 cmp_set_nosteal_interval(void)
366 {
367 	/* Set the nosteal interval (used by disp_getbest()) to 100us */
368 	nosteal_nsec = 100000UL;
369 }
370 
371 /*
372  * Routine to ensure initial callers to hrtime gets 0 as return
373  */
374 static hrtime_t
375 dummy_hrtime(void)
376 {
377 	return (0);
378 }
379 
380 /* ARGSUSED */
381 static void
382 dummy_scalehrtime(hrtime_t *ticks)
383 {}
384 
385 static uint64_t
386 dummy_unscalehrtime(hrtime_t nsecs)
387 {
388 	return ((uint64_t)nsecs);
389 }
390 
391 /*
392  * Supports Deep C-State power saving idle loop.
393  */
394 void
395 cpu_idle_adaptive(void)
396 {
397 	(*CPU->cpu_m.mcpu_idle_cpu)();
398 }
399 
400 /*
401  * Function called by CPU idle notification framework to check whether CPU
402  * has been awakened. It will be called with interrupt disabled.
403  * If CPU has been awakened, call cpu_idle_exit() to notify CPU idle
404  * notification framework.
405  */
406 /*ARGSUSED*/
407 static void
408 cpu_idle_check_wakeup(void *arg)
409 {
410 	/*
411 	 * Toggle interrupt flag to detect pending interrupts.
412 	 * If interrupt happened, do_interrupt() will notify CPU idle
413 	 * notification framework so no need to call cpu_idle_exit() here.
414 	 */
415 	sti();
416 	SMT_PAUSE();
417 	cli();
418 }
419 
420 /*
421  * Idle the present CPU until wakened via an interrupt
422  */
423 void
424 cpu_idle(void)
425 {
426 	cpu_t		*cpup = CPU;
427 	processorid_t	cpu_sid = cpup->cpu_seqid;
428 	cpupart_t	*cp = cpup->cpu_part;
429 	int		hset_update = 1;
430 
431 	/*
432 	 * If this CPU is online, and there's multiple CPUs
433 	 * in the system, then we should notate our halting
434 	 * by adding ourselves to the partition's halted CPU
435 	 * bitmap. This allows other CPUs to find/awaken us when
436 	 * work becomes available.
437 	 */
438 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
439 		hset_update = 0;
440 
441 	/*
442 	 * Add ourselves to the partition's halted CPUs bitmap
443 	 * and set our HALTED flag, if necessary.
444 	 *
445 	 * When a thread becomes runnable, it is placed on the queue
446 	 * and then the halted CPU bitmap is checked to determine who
447 	 * (if anyone) should be awakened. We therefore need to first
448 	 * add ourselves to the bitmap, and and then check if there
449 	 * is any work available. The order is important to prevent a race
450 	 * that can lead to work languishing on a run queue somewhere while
451 	 * this CPU remains halted.
452 	 *
453 	 * Either the producing CPU will see we're halted and will awaken us,
454 	 * or this CPU will see the work available in disp_anywork().
455 	 *
456 	 * Note that memory barriers after updating the HALTED flag
457 	 * are not necessary since an atomic operation (updating the bitset)
458 	 * immediately follows. On x86 the atomic operation acts as a
459 	 * memory barrier for the update of cpu_disp_flags.
460 	 */
461 	if (hset_update) {
462 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
463 		bitset_atomic_add(&cp->cp_haltset, cpu_sid);
464 	}
465 
466 	/*
467 	 * Check to make sure there's really nothing to do.
468 	 * Work destined for this CPU may become available after
469 	 * this check. We'll be notified through the clearing of our
470 	 * bit in the halted CPU bitmap, and a poke.
471 	 */
472 	if (disp_anywork()) {
473 		if (hset_update) {
474 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
475 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
476 		}
477 		return;
478 	}
479 
480 	/*
481 	 * We're on our way to being halted.
482 	 *
483 	 * Disable interrupts now, so that we'll awaken immediately
484 	 * after halting if someone tries to poke us between now and
485 	 * the time we actually halt.
486 	 *
487 	 * We check for the presence of our bit after disabling interrupts.
488 	 * If it's cleared, we'll return. If the bit is cleared after
489 	 * we check then the poke will pop us out of the halted state.
490 	 *
491 	 * This means that the ordering of the poke and the clearing
492 	 * of the bit by cpu_wakeup is important.
493 	 * cpu_wakeup() must clear, then poke.
494 	 * cpu_idle() must disable interrupts, then check for the bit.
495 	 */
496 	cli();
497 
498 	if (hset_update && bitset_in_set(&cp->cp_haltset, cpu_sid) == 0) {
499 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
500 		sti();
501 		return;
502 	}
503 
504 	/*
505 	 * The check for anything locally runnable is here for performance
506 	 * and isn't needed for correctness. disp_nrunnable ought to be
507 	 * in our cache still, so it's inexpensive to check, and if there
508 	 * is anything runnable we won't have to wait for the poke.
509 	 */
510 	if (cpup->cpu_disp->disp_nrunnable != 0) {
511 		if (hset_update) {
512 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
513 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
514 		}
515 		sti();
516 		return;
517 	}
518 
519 	if (cpu_idle_enter(IDLE_STATE_C1, 0,
520 	    cpu_idle_check_wakeup, NULL) == 0) {
521 		mach_cpu_idle();
522 		cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
523 	}
524 
525 	/*
526 	 * We're no longer halted
527 	 */
528 	if (hset_update) {
529 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
530 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
531 	}
532 }
533 
534 
535 /*
536  * If "cpu" is halted, then wake it up clearing its halted bit in advance.
537  * Otherwise, see if other CPUs in the cpu partition are halted and need to
538  * be woken up so that they can steal the thread we placed on this CPU.
539  * This function is only used on MP systems.
540  */
541 static void
542 cpu_wakeup(cpu_t *cpu, int bound)
543 {
544 	uint_t		cpu_found;
545 	processorid_t	cpu_sid;
546 	cpupart_t	*cp;
547 
548 	cp = cpu->cpu_part;
549 	cpu_sid = cpu->cpu_seqid;
550 	if (bitset_in_set(&cp->cp_haltset, cpu_sid)) {
551 		/*
552 		 * Clear the halted bit for that CPU since it will be
553 		 * poked in a moment.
554 		 */
555 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
556 		/*
557 		 * We may find the current CPU present in the halted cpuset
558 		 * if we're in the context of an interrupt that occurred
559 		 * before we had a chance to clear our bit in cpu_idle().
560 		 * Poking ourself is obviously unnecessary, since if
561 		 * we're here, we're not halted.
562 		 */
563 		if (cpu != CPU)
564 			poke_cpu(cpu->cpu_id);
565 		return;
566 	} else {
567 		/*
568 		 * This cpu isn't halted, but it's idle or undergoing a
569 		 * context switch. No need to awaken anyone else.
570 		 */
571 		if (cpu->cpu_thread == cpu->cpu_idle_thread ||
572 		    cpu->cpu_disp_flags & CPU_DISP_DONTSTEAL)
573 			return;
574 	}
575 
576 	/*
577 	 * No need to wake up other CPUs if this is for a bound thread.
578 	 */
579 	if (bound)
580 		return;
581 
582 	/*
583 	 * The CPU specified for wakeup isn't currently halted, so check
584 	 * to see if there are any other halted CPUs in the partition,
585 	 * and if there are then awaken one.
586 	 */
587 	do {
588 		cpu_found = bitset_find(&cp->cp_haltset);
589 		if (cpu_found == (uint_t)-1)
590 			return;
591 	} while (bitset_atomic_test_and_del(&cp->cp_haltset, cpu_found) < 0);
592 
593 	if (cpu_found != CPU->cpu_seqid) {
594 		poke_cpu(cpu_seq[cpu_found]->cpu_id);
595 	}
596 }
597 
598 #ifndef __xpv
599 /*
600  * Function called by CPU idle notification framework to check whether CPU
601  * has been awakened. It will be called with interrupt disabled.
602  * If CPU has been awakened, call cpu_idle_exit() to notify CPU idle
603  * notification framework.
604  */
605 static void
606 cpu_idle_mwait_check_wakeup(void *arg)
607 {
608 	volatile uint32_t *mcpu_mwait = (volatile uint32_t *)arg;
609 
610 	ASSERT(arg != NULL);
611 	if (*mcpu_mwait != MWAIT_HALTED) {
612 		/*
613 		 * CPU has been awakened, notify CPU idle notification system.
614 		 */
615 		cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
616 	} else {
617 		/*
618 		 * Toggle interrupt flag to detect pending interrupts.
619 		 * If interrupt happened, do_interrupt() will notify CPU idle
620 		 * notification framework so no need to call cpu_idle_exit()
621 		 * here.
622 		 */
623 		sti();
624 		SMT_PAUSE();
625 		cli();
626 	}
627 }
628 
629 /*
630  * Idle the present CPU until awakened via touching its monitored line
631  */
632 void
633 cpu_idle_mwait(void)
634 {
635 	volatile uint32_t	*mcpu_mwait = CPU->cpu_m.mcpu_mwait;
636 	cpu_t			*cpup = CPU;
637 	processorid_t		cpu_sid = cpup->cpu_seqid;
638 	cpupart_t		*cp = cpup->cpu_part;
639 	int			hset_update = 1;
640 
641 	/*
642 	 * Set our mcpu_mwait here, so we can tell if anyone tries to
643 	 * wake us between now and when we call mwait.  No other cpu will
644 	 * attempt to set our mcpu_mwait until we add ourself to the halted
645 	 * CPU bitmap.
646 	 */
647 	*mcpu_mwait = MWAIT_HALTED;
648 
649 	/*
650 	 * If this CPU is online, and there's multiple CPUs
651 	 * in the system, then we should note our halting
652 	 * by adding ourselves to the partition's halted CPU
653 	 * bitmap. This allows other CPUs to find/awaken us when
654 	 * work becomes available.
655 	 */
656 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
657 		hset_update = 0;
658 
659 	/*
660 	 * Add ourselves to the partition's halted CPUs bitmap
661 	 * and set our HALTED flag, if necessary.
662 	 *
663 	 * When a thread becomes runnable, it is placed on the queue
664 	 * and then the halted CPU bitmap is checked to determine who
665 	 * (if anyone) should be awakened. We therefore need to first
666 	 * add ourselves to the bitmap, and and then check if there
667 	 * is any work available.
668 	 *
669 	 * Note that memory barriers after updating the HALTED flag
670 	 * are not necessary since an atomic operation (updating the bitmap)
671 	 * immediately follows. On x86 the atomic operation acts as a
672 	 * memory barrier for the update of cpu_disp_flags.
673 	 */
674 	if (hset_update) {
675 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
676 		bitset_atomic_add(&cp->cp_haltset, cpu_sid);
677 	}
678 
679 	/*
680 	 * Check to make sure there's really nothing to do.
681 	 * Work destined for this CPU may become available after
682 	 * this check. We'll be notified through the clearing of our
683 	 * bit in the halted CPU bitmap, and a write to our mcpu_mwait.
684 	 *
685 	 * disp_anywork() checks disp_nrunnable, so we do not have to later.
686 	 */
687 	if (disp_anywork()) {
688 		if (hset_update) {
689 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
690 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
691 		}
692 		return;
693 	}
694 
695 	/*
696 	 * We're on our way to being halted.
697 	 * To avoid a lost wakeup, arm the monitor before checking if another
698 	 * cpu wrote to mcpu_mwait to wake us up.
699 	 */
700 	i86_monitor(mcpu_mwait, 0, 0);
701 	if (*mcpu_mwait == MWAIT_HALTED) {
702 		if (cpu_idle_enter(IDLE_STATE_C1, 0,
703 		    cpu_idle_mwait_check_wakeup, (void *)mcpu_mwait) == 0) {
704 			if (*mcpu_mwait == MWAIT_HALTED) {
705 				i86_mwait(0, 0);
706 			}
707 			cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
708 		}
709 	}
710 
711 	/*
712 	 * We're no longer halted
713 	 */
714 	if (hset_update) {
715 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
716 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
717 	}
718 }
719 
720 /*
721  * If "cpu" is halted in mwait, then wake it up clearing its halted bit in
722  * advance.  Otherwise, see if other CPUs in the cpu partition are halted and
723  * need to be woken up so that they can steal the thread we placed on this CPU.
724  * This function is only used on MP systems.
725  */
726 static void
727 cpu_wakeup_mwait(cpu_t *cp, int bound)
728 {
729 	cpupart_t	*cpu_part;
730 	uint_t		cpu_found;
731 	processorid_t	cpu_sid;
732 
733 	cpu_part = cp->cpu_part;
734 	cpu_sid = cp->cpu_seqid;
735 
736 	/*
737 	 * Clear the halted bit for that CPU since it will be woken up
738 	 * in a moment.
739 	 */
740 	if (bitset_in_set(&cpu_part->cp_haltset, cpu_sid)) {
741 		/*
742 		 * Clear the halted bit for that CPU since it will be
743 		 * poked in a moment.
744 		 */
745 		bitset_atomic_del(&cpu_part->cp_haltset, cpu_sid);
746 		/*
747 		 * We may find the current CPU present in the halted cpuset
748 		 * if we're in the context of an interrupt that occurred
749 		 * before we had a chance to clear our bit in cpu_idle().
750 		 * Waking ourself is obviously unnecessary, since if
751 		 * we're here, we're not halted.
752 		 *
753 		 * monitor/mwait wakeup via writing to our cache line is
754 		 * harmless and less expensive than always checking if we
755 		 * are waking ourself which is an uncommon case.
756 		 */
757 		MWAIT_WAKEUP(cp);	/* write to monitored line */
758 		return;
759 	} else {
760 		/*
761 		 * This cpu isn't halted, but it's idle or undergoing a
762 		 * context switch. No need to awaken anyone else.
763 		 */
764 		if (cp->cpu_thread == cp->cpu_idle_thread ||
765 		    cp->cpu_disp_flags & CPU_DISP_DONTSTEAL)
766 			return;
767 	}
768 
769 	/*
770 	 * No need to wake up other CPUs if the thread we just enqueued
771 	 * is bound.
772 	 */
773 	if (bound || ncpus == 1)
774 		return;
775 
776 	/*
777 	 * See if there's any other halted CPUs. If there are, then
778 	 * select one, and awaken it.
779 	 * It's possible that after we find a CPU, somebody else
780 	 * will awaken it before we get the chance.
781 	 * In that case, look again.
782 	 */
783 	do {
784 		cpu_found = bitset_find(&cpu_part->cp_haltset);
785 		if (cpu_found == (uint_t)-1)
786 			return;
787 	} while (bitset_atomic_test_and_del(&cpu_part->cp_haltset,
788 	    cpu_found) < 0);
789 
790 	/*
791 	 * Do not check if cpu_found is ourself as monitor/mwait
792 	 * wakeup is cheap.
793 	 */
794 	MWAIT_WAKEUP(cpu_seq[cpu_found]); /* write to monitored line */
795 }
796 
797 #endif
798 
799 void (*cpu_pause_handler)(volatile char *) = NULL;
800 
801 static int
802 mp_disable_intr(int cpun)
803 {
804 	/*
805 	 * switch to the offline cpu
806 	 */
807 	affinity_set(cpun);
808 	/*
809 	 * raise ipl to just below cross call
810 	 */
811 	splx(XC_SYS_PIL - 1);
812 	/*
813 	 *	set base spl to prevent the next swtch to idle from
814 	 *	lowering back to ipl 0
815 	 */
816 	CPU->cpu_intr_actv |= (1 << (XC_SYS_PIL - 1));
817 	set_base_spl();
818 	affinity_clear();
819 	return (DDI_SUCCESS);
820 }
821 
822 static void
823 mp_enable_intr(int cpun)
824 {
825 	/*
826 	 * switch to the online cpu
827 	 */
828 	affinity_set(cpun);
829 	/*
830 	 * clear the interrupt active mask
831 	 */
832 	CPU->cpu_intr_actv &= ~(1 << (XC_SYS_PIL - 1));
833 	set_base_spl();
834 	(void) spl0();
835 	affinity_clear();
836 }
837 
838 static void
839 mach_get_platform(int owner)
840 {
841 	void		**srv_opsp;
842 	void		**clt_opsp;
843 	int		i;
844 	int		total_ops;
845 
846 	/* fix up psm ops */
847 	srv_opsp = (void **)mach_set[0];
848 	clt_opsp = (void **)mach_set[owner];
849 	if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01)
850 		total_ops = sizeof (struct psm_ops_ver01) /
851 		    sizeof (void (*)(void));
852 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_1)
853 		/* no psm_notify_func */
854 		total_ops = OFFSETOF(struct psm_ops, psm_notify_func) /
855 		    sizeof (void (*)(void));
856 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_2)
857 		/* no psm_timer funcs */
858 		total_ops = OFFSETOF(struct psm_ops, psm_timer_reprogram) /
859 		    sizeof (void (*)(void));
860 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_3)
861 		/* no psm_preshutdown function */
862 		total_ops = OFFSETOF(struct psm_ops, psm_preshutdown) /
863 		    sizeof (void (*)(void));
864 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_4)
865 		/* no psm_preshutdown function */
866 		total_ops = OFFSETOF(struct psm_ops, psm_intr_ops) /
867 		    sizeof (void (*)(void));
868 	else
869 		total_ops = sizeof (struct psm_ops) / sizeof (void (*)(void));
870 
871 	/*
872 	 * Save the version of the PSM module, in case we need to
873 	 * behave differently based on version.
874 	 */
875 	mach_ver[0] = mach_ver[owner];
876 
877 	for (i = 0; i < total_ops; i++)
878 		if (clt_opsp[i] != NULL)
879 			srv_opsp[i] = clt_opsp[i];
880 }
881 
882 static void
883 mach_construct_info()
884 {
885 	struct psm_sw *swp;
886 	int	mach_cnt[PSM_OWN_OVERRIDE+1] = {0};
887 	int	conflict_owner = 0;
888 
889 	if (psmsw->psw_forw == psmsw)
890 		panic("No valid PSM modules found");
891 	mutex_enter(&psmsw_lock);
892 	for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
893 		if (!(swp->psw_flag & PSM_MOD_IDENTIFY))
894 			continue;
895 		mach_set[swp->psw_infop->p_owner] = swp->psw_infop->p_ops;
896 		mach_ver[swp->psw_infop->p_owner] = swp->psw_infop->p_version;
897 		mach_cnt[swp->psw_infop->p_owner]++;
898 	}
899 	mutex_exit(&psmsw_lock);
900 
901 	mach_get_platform(PSM_OWN_SYS_DEFAULT);
902 
903 	/* check to see are there any conflicts */
904 	if (mach_cnt[PSM_OWN_EXCLUSIVE] > 1)
905 		conflict_owner = PSM_OWN_EXCLUSIVE;
906 	if (mach_cnt[PSM_OWN_OVERRIDE] > 1)
907 		conflict_owner = PSM_OWN_OVERRIDE;
908 	if (conflict_owner) {
909 		/* remove all psm modules except uppc */
910 		cmn_err(CE_WARN,
911 		    "Conflicts detected on the following PSM modules:");
912 		mutex_enter(&psmsw_lock);
913 		for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
914 			if (swp->psw_infop->p_owner == conflict_owner)
915 				cmn_err(CE_WARN, "%s ",
916 				    swp->psw_infop->p_mach_idstring);
917 		}
918 		mutex_exit(&psmsw_lock);
919 		cmn_err(CE_WARN,
920 		    "Setting the system back to SINGLE processor mode!");
921 		cmn_err(CE_WARN,
922 		    "Please edit /etc/mach to remove the invalid PSM module.");
923 		return;
924 	}
925 
926 	if (mach_set[PSM_OWN_EXCLUSIVE])
927 		mach_get_platform(PSM_OWN_EXCLUSIVE);
928 
929 	if (mach_set[PSM_OWN_OVERRIDE])
930 		mach_get_platform(PSM_OWN_OVERRIDE);
931 }
932 
933 static void
934 mach_init()
935 {
936 	struct psm_ops  *pops;
937 
938 	mach_construct_info();
939 
940 	pops = mach_set[0];
941 
942 	/* register the interrupt and clock initialization rotuines */
943 	picinitf = mach_picinit;
944 	clkinitf = mach_clkinit;
945 	psm_get_clockirq = pops->psm_get_clockirq;
946 
947 	/* register the interrupt setup code */
948 	slvltovect = mach_softlvl_to_vect;
949 	addspl	= pops->psm_addspl;
950 	delspl	= pops->psm_delspl;
951 
952 	if (pops->psm_translate_irq)
953 		psm_translate_irq = pops->psm_translate_irq;
954 	if (pops->psm_intr_ops)
955 		psm_intr_ops = pops->psm_intr_ops;
956 
957 #if defined(PSMI_1_2) || defined(PSMI_1_3) || defined(PSMI_1_4)
958 	/*
959 	 * Time-of-day functionality now handled in TOD modules.
960 	 * (Warn about PSM modules that think that we're going to use
961 	 * their ops vectors.)
962 	 */
963 	if (pops->psm_tod_get)
964 		cmn_err(CE_WARN, "obsolete psm_tod_get op %p",
965 		    (void *)pops->psm_tod_get);
966 
967 	if (pops->psm_tod_set)
968 		cmn_err(CE_WARN, "obsolete psm_tod_set op %p",
969 		    (void *)pops->psm_tod_set);
970 #endif
971 
972 	if (pops->psm_notify_error) {
973 		psm_notify_error = mach_notify_error;
974 		notify_error = pops->psm_notify_error;
975 	}
976 
977 	(*pops->psm_softinit)();
978 
979 	/*
980 	 * Initialize the dispatcher's function hooks to enable CPU halting
981 	 * when idle.  Set both the deep-idle and non-deep-idle hooks.
982 	 *
983 	 * Assume we can use power saving deep-idle loop cpu_idle_adaptive.
984 	 * Platform deep-idle driver will reset our idle loop to
985 	 * non_deep_idle_cpu if power saving deep-idle feature is not available.
986 	 *
987 	 * Do not use monitor/mwait if idle_cpu_use_hlt is not set(spin idle)
988 	 * or idle_cpu_prefer_mwait is not set.
989 	 * Allocate monitor/mwait buffer for cpu0.
990 	 */
991 #ifndef __xpv
992 	non_deep_idle_disp_enq_thread = disp_enq_thread;
993 #endif
994 	if (idle_cpu_use_hlt) {
995 		idle_cpu = cpu_idle_adaptive;
996 		CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
997 #ifndef __xpv
998 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait) {
999 			CPU->cpu_m.mcpu_mwait = cpuid_mwait_alloc(CPU);
1000 			/*
1001 			 * Protect ourself from insane mwait size.
1002 			 */
1003 			if (CPU->cpu_m.mcpu_mwait == NULL) {
1004 #ifdef DEBUG
1005 				cmn_err(CE_NOTE, "Using hlt idle.  Cannot "
1006 				    "handle cpu 0 mwait size.");
1007 #endif
1008 				idle_cpu_prefer_mwait = 0;
1009 				CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
1010 			} else {
1011 				CPU->cpu_m.mcpu_idle_cpu = cpu_idle_mwait;
1012 			}
1013 		} else {
1014 			CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
1015 		}
1016 		non_deep_idle_cpu = CPU->cpu_m.mcpu_idle_cpu;
1017 
1018 		/*
1019 		 * Disable power saving deep idle loop?
1020 		 */
1021 		if (idle_cpu_no_deep_c) {
1022 			idle_cpu = non_deep_idle_cpu;
1023 		}
1024 #endif
1025 	}
1026 
1027 	mach_smpinit();
1028 }
1029 
1030 static void
1031 mach_smpinit(void)
1032 {
1033 	struct psm_ops  *pops;
1034 	processorid_t cpu_id;
1035 	int cnt;
1036 	cpuset_t cpumask;
1037 
1038 	pops = mach_set[0];
1039 	CPUSET_ZERO(cpumask);
1040 
1041 	cpu_id = -1;
1042 	cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
1043 	for (cnt = 0; cpu_id != -1; cnt++) {
1044 		CPUSET_ADD(cpumask, cpu_id);
1045 		cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
1046 	}
1047 
1048 	mp_cpus = cpumask;
1049 
1050 	/* MP related routines */
1051 	ap_mlsetup = pops->psm_post_cpu_start;
1052 	send_dirintf = pops->psm_send_ipi;
1053 
1054 	/* optional MP related routines */
1055 	if (pops->psm_shutdown)
1056 		psm_shutdownf = pops->psm_shutdown;
1057 	if (pops->psm_preshutdown)
1058 		psm_preshutdownf = pops->psm_preshutdown;
1059 	if (pops->psm_notify_func)
1060 		psm_notifyf = pops->psm_notify_func;
1061 	if (pops->psm_set_idlecpu)
1062 		psm_set_idle_cpuf = pops->psm_set_idlecpu;
1063 	if (pops->psm_unset_idlecpu)
1064 		psm_unset_idle_cpuf = pops->psm_unset_idlecpu;
1065 
1066 	psm_clkinit = pops->psm_clkinit;
1067 
1068 	if (pops->psm_timer_reprogram)
1069 		psm_timer_reprogram = pops->psm_timer_reprogram;
1070 
1071 	if (pops->psm_timer_enable)
1072 		psm_timer_enable = pops->psm_timer_enable;
1073 
1074 	if (pops->psm_timer_disable)
1075 		psm_timer_disable = pops->psm_timer_disable;
1076 
1077 	if (pops->psm_post_cyclic_setup)
1078 		psm_post_cyclic_setup = pops->psm_post_cyclic_setup;
1079 
1080 	if (pops->psm_state)
1081 		psm_state = pops->psm_state;
1082 
1083 	/*
1084 	 * Set these vectors here so they can be used by Suspend/Resume
1085 	 * on UP machines.
1086 	 */
1087 	if (pops->psm_disable_intr)
1088 		psm_disable_intr = pops->psm_disable_intr;
1089 	if (pops->psm_enable_intr)
1090 		psm_enable_intr  = pops->psm_enable_intr;
1091 
1092 	/* check for multiple CPUs */
1093 	if (cnt < 2)
1094 		return;
1095 
1096 	/* check for MP platforms */
1097 	if (pops->psm_cpu_start == NULL)
1098 		return;
1099 
1100 	/*
1101 	 * Set the dispatcher hook to enable cpu "wake up"
1102 	 * when a thread becomes runnable.
1103 	 */
1104 	if (idle_cpu_use_hlt) {
1105 		disp_enq_thread = cpu_wakeup;
1106 #ifndef __xpv
1107 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait)
1108 			disp_enq_thread = cpu_wakeup_mwait;
1109 		non_deep_idle_disp_enq_thread = disp_enq_thread;
1110 #endif
1111 	}
1112 
1113 	psm_get_ipivect = pops->psm_get_ipivect;
1114 
1115 	(void) add_avintr((void *)NULL, XC_HI_PIL, xc_serv, "xc_intr",
1116 	    (*pops->psm_get_ipivect)(XC_HI_PIL, PSM_INTR_IPI_HI),
1117 	    NULL, NULL, NULL, NULL);
1118 
1119 	(void) (*pops->psm_get_ipivect)(XC_CPUPOKE_PIL, PSM_INTR_POKE);
1120 }
1121 
1122 static void
1123 mach_picinit()
1124 {
1125 	struct psm_ops  *pops;
1126 
1127 	pops = mach_set[0];
1128 
1129 	/* register the interrupt handlers */
1130 	setlvl = pops->psm_intr_enter;
1131 	setlvlx = pops->psm_intr_exit;
1132 
1133 	/* initialize the interrupt hardware */
1134 	(*pops->psm_picinit)();
1135 
1136 	/* set interrupt mask for current ipl */
1137 	setspl = pops->psm_setspl;
1138 	cli();
1139 	setspl(CPU->cpu_pri);
1140 }
1141 
1142 uint_t	cpu_freq;	/* MHz */
1143 uint64_t cpu_freq_hz;	/* measured (in hertz) */
1144 
1145 #define	MEGA_HZ		1000000
1146 
1147 #ifdef __xpv
1148 
1149 int xpv_cpufreq_workaround = 1;
1150 int xpv_cpufreq_verbose = 0;
1151 
1152 #else	/* __xpv */
1153 
1154 static uint64_t
1155 mach_calchz(uint32_t pit_counter, uint64_t *processor_clks)
1156 {
1157 	uint64_t cpu_hz;
1158 
1159 	if ((pit_counter == 0) || (*processor_clks == 0) ||
1160 	    (*processor_clks > (((uint64_t)-1) / PIT_HZ)))
1161 		return (0);
1162 
1163 	cpu_hz = ((uint64_t)PIT_HZ * *processor_clks) / pit_counter;
1164 
1165 	return (cpu_hz);
1166 }
1167 
1168 #endif	/* __xpv */
1169 
1170 static uint64_t
1171 mach_getcpufreq(void)
1172 {
1173 #if defined(__xpv)
1174 	vcpu_time_info_t *vti = &CPU->cpu_m.mcpu_vcpu_info->time;
1175 	uint64_t cpu_hz;
1176 
1177 	/*
1178 	 * During dom0 bringup, it was noted that on at least one older
1179 	 * Intel HT machine, the hypervisor initially gives a tsc_to_system_mul
1180 	 * value that is quite wrong (the 3.06GHz clock was reported
1181 	 * as 4.77GHz)
1182 	 *
1183 	 * The curious thing is, that if you stop the kernel at entry,
1184 	 * breakpoint here and inspect the value with kmdb, the value
1185 	 * is correct - but if you don't stop and simply enable the
1186 	 * printf statement (below), you can see the bad value printed
1187 	 * here.  Almost as if something kmdb did caused the hypervisor to
1188 	 * figure it out correctly.  And, note that the hypervisor
1189 	 * eventually -does- figure it out correctly ... if you look at
1190 	 * the field later in the life of dom0, it is correct.
1191 	 *
1192 	 * For now, on dom0, we employ a slightly cheesy workaround of
1193 	 * using the DOM0_PHYSINFO hypercall.
1194 	 */
1195 	if (DOMAIN_IS_INITDOMAIN(xen_info) && xpv_cpufreq_workaround) {
1196 		cpu_hz = 1000 * xpv_cpu_khz();
1197 	} else {
1198 		cpu_hz = (UINT64_C(1000000000) << 32) / vti->tsc_to_system_mul;
1199 
1200 		if (vti->tsc_shift < 0)
1201 			cpu_hz <<= -vti->tsc_shift;
1202 		else
1203 			cpu_hz >>= vti->tsc_shift;
1204 	}
1205 
1206 	if (xpv_cpufreq_verbose)
1207 		printf("mach_getcpufreq: system_mul 0x%x, shift %d, "
1208 		    "cpu_hz %" PRId64 "Hz\n",
1209 		    vti->tsc_to_system_mul, vti->tsc_shift, cpu_hz);
1210 
1211 	return (cpu_hz);
1212 #else	/* __xpv */
1213 	uint32_t pit_counter;
1214 	uint64_t processor_clks;
1215 
1216 	if (x86_feature & X86_TSC) {
1217 		/*
1218 		 * We have a TSC. freq_tsc() knows how to measure the number
1219 		 * of clock cycles sampled against the PIT.
1220 		 */
1221 		ulong_t flags = clear_int_flag();
1222 		processor_clks = freq_tsc(&pit_counter);
1223 		restore_int_flag(flags);
1224 		return (mach_calchz(pit_counter, &processor_clks));
1225 	} else if (x86_vendor == X86_VENDOR_Cyrix || x86_type == X86_TYPE_P5) {
1226 #if defined(__amd64)
1227 		panic("mach_getcpufreq: no TSC!");
1228 #elif defined(__i386)
1229 		/*
1230 		 * We are a Cyrix based on a 6x86 core or an Intel Pentium
1231 		 * for which freq_notsc() knows how to measure the number of
1232 		 * elapsed clock cycles sampled against the PIT
1233 		 */
1234 		ulong_t flags = clear_int_flag();
1235 		processor_clks = freq_notsc(&pit_counter);
1236 		restore_int_flag(flags);
1237 		return (mach_calchz(pit_counter, &processor_clks));
1238 #endif	/* __i386 */
1239 	}
1240 
1241 	/* We do not know how to calculate cpu frequency for this cpu. */
1242 	return (0);
1243 #endif	/* __xpv */
1244 }
1245 
1246 /*
1247  * If the clock speed of a cpu is found to be reported incorrectly, do not add
1248  * to this array, instead improve the accuracy of the algorithm that determines
1249  * the clock speed of the processor or extend the implementation to support the
1250  * vendor as appropriate. This is here only to support adjusting the speed on
1251  * older slower processors that mach_fixcpufreq() would not be able to account
1252  * for otherwise.
1253  */
1254 static int x86_cpu_freq[] = { 60, 75, 80, 90, 120, 160, 166, 175, 180, 233 };
1255 
1256 /*
1257  * On fast processors the clock frequency that is measured may be off by
1258  * a few MHz from the value printed on the part. This is a combination of
1259  * the factors that for such fast parts being off by this much is within
1260  * the tolerances for manufacture and because of the difficulties in the
1261  * measurement that can lead to small error. This function uses some
1262  * heuristics in order to tweak the value that was measured to match what
1263  * is most likely printed on the part.
1264  *
1265  * Some examples:
1266  * 	AMD Athlon 1000 mhz measured as 998 mhz
1267  * 	Intel Pentium III Xeon 733 mhz measured as 731 mhz
1268  * 	Intel Pentium IV 1500 mhz measured as 1495mhz
1269  *
1270  * If in the future this function is no longer sufficient to correct
1271  * for the error in the measurement, then the algorithm used to perform
1272  * the measurement will have to be improved in order to increase accuracy
1273  * rather than adding horrible and questionable kludges here.
1274  *
1275  * This is called after the cyclics subsystem because of the potential
1276  * that the heuristics within may give a worse estimate of the clock
1277  * frequency than the value that was measured.
1278  */
1279 static void
1280 mach_fixcpufreq(void)
1281 {
1282 	uint32_t freq, mul, near66, delta66, near50, delta50, fixed, delta, i;
1283 
1284 	freq = (uint32_t)cpu_freq;
1285 
1286 	/*
1287 	 * Find the nearest integer multiple of 200/3 (about 66) MHz to the
1288 	 * measured speed taking into account that the 667 MHz parts were
1289 	 * the first to round-up.
1290 	 */
1291 	mul = (uint32_t)((3 * (uint64_t)freq + 100) / 200);
1292 	near66 = (uint32_t)((200 * (uint64_t)mul + ((mul >= 10) ? 1 : 0)) / 3);
1293 	delta66 = (near66 > freq) ? (near66 - freq) : (freq - near66);
1294 
1295 	/* Find the nearest integer multiple of 50 MHz to the measured speed */
1296 	mul = (freq + 25) / 50;
1297 	near50 = mul * 50;
1298 	delta50 = (near50 > freq) ? (near50 - freq) : (freq - near50);
1299 
1300 	/* Find the closer of the two */
1301 	if (delta66 < delta50) {
1302 		fixed = near66;
1303 		delta = delta66;
1304 	} else {
1305 		fixed = near50;
1306 		delta = delta50;
1307 	}
1308 
1309 	if (fixed > INT_MAX)
1310 		return;
1311 
1312 	/*
1313 	 * Some older parts have a core clock frequency that is not an
1314 	 * integral multiple of 50 or 66 MHz. Check if one of the old
1315 	 * clock frequencies is closer to the measured value than any
1316 	 * of the integral multiples of 50 an 66, and if so set fixed
1317 	 * and delta appropriately to represent the closest value.
1318 	 */
1319 	i = sizeof (x86_cpu_freq) / sizeof (int);
1320 	while (i > 0) {
1321 		i--;
1322 
1323 		if (x86_cpu_freq[i] <= freq) {
1324 			mul = freq - x86_cpu_freq[i];
1325 
1326 			if (mul < delta) {
1327 				fixed = x86_cpu_freq[i];
1328 				delta = mul;
1329 			}
1330 
1331 			break;
1332 		}
1333 
1334 		mul = x86_cpu_freq[i] - freq;
1335 
1336 		if (mul < delta) {
1337 			fixed = x86_cpu_freq[i];
1338 			delta = mul;
1339 		}
1340 	}
1341 
1342 	/*
1343 	 * Set a reasonable maximum for how much to correct the measured
1344 	 * result by. This check is here to prevent the adjustment made
1345 	 * by this function from being more harm than good. It is entirely
1346 	 * possible that in the future parts will be made that are not
1347 	 * integral multiples of 66 or 50 in clock frequency or that
1348 	 * someone may overclock a part to some odd frequency. If the
1349 	 * measured value is farther from the corrected value than
1350 	 * allowed, then assume the corrected value is in error and use
1351 	 * the measured value.
1352 	 */
1353 	if (6 < delta)
1354 		return;
1355 
1356 	cpu_freq = (int)fixed;
1357 }
1358 
1359 
1360 static int
1361 machhztomhz(uint64_t cpu_freq_hz)
1362 {
1363 	uint64_t cpu_mhz;
1364 
1365 	/* Round to nearest MHZ */
1366 	cpu_mhz = (cpu_freq_hz + (MEGA_HZ / 2)) / MEGA_HZ;
1367 
1368 	if (cpu_mhz > INT_MAX)
1369 		return (0);
1370 
1371 	return ((int)cpu_mhz);
1372 
1373 }
1374 
1375 
1376 static int
1377 mach_clkinit(int preferred_mode, int *set_mode)
1378 {
1379 	struct psm_ops  *pops;
1380 	int resolution;
1381 
1382 	pops = mach_set[0];
1383 
1384 	cpu_freq_hz = mach_getcpufreq();
1385 
1386 	cpu_freq = machhztomhz(cpu_freq_hz);
1387 
1388 	if (!(x86_feature & X86_TSC) || (cpu_freq == 0))
1389 		tsc_gethrtime_enable = 0;
1390 
1391 #ifndef __xpv
1392 	if (tsc_gethrtime_enable) {
1393 		tsc_hrtimeinit(cpu_freq_hz);
1394 	} else
1395 #endif
1396 	{
1397 		if (pops->psm_hrtimeinit)
1398 			(*pops->psm_hrtimeinit)();
1399 		gethrtimef = pops->psm_gethrtime;
1400 		gethrtimeunscaledf = gethrtimef;
1401 		/* scalehrtimef will remain dummy */
1402 	}
1403 
1404 	mach_fixcpufreq();
1405 
1406 	if (mach_ver[0] >= PSM_INFO_VER01_3) {
1407 		if (preferred_mode == TIMER_ONESHOT) {
1408 
1409 			resolution = (*pops->psm_clkinit)(0);
1410 			if (resolution != 0)  {
1411 				*set_mode = TIMER_ONESHOT;
1412 				return (resolution);
1413 			}
1414 		}
1415 
1416 		/*
1417 		 * either periodic mode was requested or could not set to
1418 		 * one-shot mode
1419 		 */
1420 		resolution = (*pops->psm_clkinit)(hz);
1421 		/*
1422 		 * psm should be able to do periodic, so we do not check
1423 		 * for return value of psm_clkinit here.
1424 		 */
1425 		*set_mode = TIMER_PERIODIC;
1426 		return (resolution);
1427 	} else {
1428 		/*
1429 		 * PSMI interface prior to PSMI_3 does not define a return
1430 		 * value for psm_clkinit, so the return value is ignored.
1431 		 */
1432 		(void) (*pops->psm_clkinit)(hz);
1433 		*set_mode = TIMER_PERIODIC;
1434 		return (nsec_per_tick);
1435 	}
1436 }
1437 
1438 
1439 /*ARGSUSED*/
1440 static int
1441 mach_softlvl_to_vect(int ipl)
1442 {
1443 	setsoftint = av_set_softint_pending;
1444 	kdisetsoftint = kdi_av_set_softint_pending;
1445 
1446 	return (PSM_SV_SOFTWARE);
1447 }
1448 
1449 #ifdef DEBUG
1450 /*
1451  * This is here to allow us to simulate cpus that refuse to start.
1452  */
1453 cpuset_t cpufailset;
1454 #endif
1455 
1456 int
1457 mach_cpu_start(struct cpu *cp, void *ctx)
1458 {
1459 	struct psm_ops *pops = mach_set[0];
1460 	processorid_t id = cp->cpu_id;
1461 
1462 #ifdef DEBUG
1463 	if (CPU_IN_SET(cpufailset, id))
1464 		return (0);
1465 #endif
1466 	return ((*pops->psm_cpu_start)(id, ctx));
1467 }
1468 
1469 int
1470 mach_cpuid_start(processorid_t id, void *ctx)
1471 {
1472 	struct psm_ops *pops = mach_set[0];
1473 
1474 #ifdef DEBUG
1475 	if (CPU_IN_SET(cpufailset, id))
1476 		return (0);
1477 #endif
1478 	return ((*pops->psm_cpu_start)(id, ctx));
1479 }
1480 
1481 /*
1482  * Default handler to create device node for CPU.
1483  * One reference count will be held on created device node.
1484  */
1485 static int
1486 mach_cpu_create_devinfo(cpu_t *cp, dev_info_t **dipp)
1487 {
1488 	int rv, circ;
1489 	dev_info_t *dip;
1490 	static kmutex_t cpu_node_lock;
1491 	static dev_info_t *cpu_nex_devi = NULL;
1492 
1493 	ASSERT(cp != NULL);
1494 	ASSERT(dipp != NULL);
1495 	*dipp = NULL;
1496 
1497 	if (cpu_nex_devi == NULL) {
1498 		mutex_enter(&cpu_node_lock);
1499 		/* First check whether cpus exists. */
1500 		cpu_nex_devi = ddi_find_devinfo("cpus", -1, 0);
1501 		/* Create cpus if it doesn't exist. */
1502 		if (cpu_nex_devi == NULL) {
1503 			ndi_devi_enter(ddi_root_node(), &circ);
1504 			rv = ndi_devi_alloc(ddi_root_node(), "cpus",
1505 			    (pnode_t)DEVI_SID_NODEID, &dip);
1506 			if (rv != NDI_SUCCESS) {
1507 				mutex_exit(&cpu_node_lock);
1508 				cmn_err(CE_CONT,
1509 				    "?failed to create cpu nexus device.\n");
1510 				return (PSM_FAILURE);
1511 			}
1512 			ASSERT(dip != NULL);
1513 			(void) ndi_devi_online(dip, 0);
1514 			ndi_devi_exit(ddi_root_node(), circ);
1515 			cpu_nex_devi = dip;
1516 		}
1517 		mutex_exit(&cpu_node_lock);
1518 	}
1519 
1520 	/*
1521 	 * create a child node for cpu identified as 'cpu_id'
1522 	 */
1523 	ndi_devi_enter(cpu_nex_devi, &circ);
1524 	dip = ddi_add_child(cpu_nex_devi, "cpu", DEVI_SID_NODEID, -1);
1525 	if (dip == NULL) {
1526 		cmn_err(CE_CONT,
1527 		    "?failed to create device node for cpu%d.\n", cp->cpu_id);
1528 		rv = PSM_FAILURE;
1529 	} else {
1530 		*dipp = dip;
1531 		(void) ndi_hold_devi(dip);
1532 		rv = PSM_SUCCESS;
1533 	}
1534 	ndi_devi_exit(cpu_nex_devi, circ);
1535 
1536 	return (rv);
1537 }
1538 
1539 /*
1540  * Create cpu device node in device tree and online it.
1541  * Return created dip with reference count held if requested.
1542  */
1543 int
1544 mach_cpu_create_device_node(struct cpu *cp, dev_info_t **dipp)
1545 {
1546 	int rv;
1547 	dev_info_t *dip = NULL;
1548 
1549 	ASSERT(psm_cpu_create_devinfo != NULL);
1550 	rv = psm_cpu_create_devinfo(cp, &dip);
1551 	if (rv == PSM_SUCCESS) {
1552 		cpuid_set_cpu_properties(dip, cp->cpu_id, cp->cpu_m.mcpu_cpi);
1553 		/* Recursively attach driver for parent nexus device. */
1554 		if (i_ddi_attach_node_hierarchy(ddi_get_parent(dip)) ==
1555 		    DDI_SUCCESS) {
1556 			/* Configure cpu itself and descendants. */
1557 			(void) ndi_devi_online(dip,
1558 			    NDI_ONLINE_ATTACH | NDI_CONFIG);
1559 		}
1560 		if (dipp != NULL) {
1561 			*dipp = dip;
1562 		} else {
1563 			(void) ndi_rele_devi(dip);
1564 		}
1565 	}
1566 
1567 	return (rv);
1568 }
1569 
1570 /*ARGSUSED*/
1571 static int
1572 mach_translate_irq(dev_info_t *dip, int irqno)
1573 {
1574 	return (irqno);	/* default to NO translation */
1575 }
1576 
1577 static void
1578 mach_notify_error(int level, char *errmsg)
1579 {
1580 	/*
1581 	 * SL_FATAL is pass in once panicstr is set, deliver it
1582 	 * as CE_PANIC.  Also, translate SL_ codes back to CE_
1583 	 * codes for the psmi handler
1584 	 */
1585 	if (level & SL_FATAL)
1586 		(*notify_error)(CE_PANIC, errmsg);
1587 	else if (level & SL_WARN)
1588 		(*notify_error)(CE_WARN, errmsg);
1589 	else if (level & SL_NOTE)
1590 		(*notify_error)(CE_NOTE, errmsg);
1591 	else if (level & SL_CONSOLE)
1592 		(*notify_error)(CE_CONT, errmsg);
1593 }
1594 
1595 /*
1596  * It provides the default basic intr_ops interface for the new DDI
1597  * interrupt framework if the PSM doesn't have one.
1598  *
1599  * Input:
1600  * dip     - pointer to the dev_info structure of the requested device
1601  * hdlp    - pointer to the internal interrupt handle structure for the
1602  *	     requested interrupt
1603  * intr_op - opcode for this call
1604  * result  - pointer to the integer that will hold the result to be
1605  *	     passed back if return value is PSM_SUCCESS
1606  *
1607  * Output:
1608  * return value is either PSM_SUCCESS or PSM_FAILURE
1609  */
1610 static int
1611 mach_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1612     psm_intr_op_t intr_op, int *result)
1613 {
1614 	struct intrspec *ispec;
1615 
1616 	switch (intr_op) {
1617 	case PSM_INTR_OP_CHECK_MSI:
1618 		*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1619 		    DDI_INTR_TYPE_MSIX);
1620 		break;
1621 	case PSM_INTR_OP_ALLOC_VECTORS:
1622 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1623 			*result = 1;
1624 		else
1625 			*result = 0;
1626 		break;
1627 	case PSM_INTR_OP_FREE_VECTORS:
1628 		break;
1629 	case PSM_INTR_OP_NAVAIL_VECTORS:
1630 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1631 			*result = 1;
1632 		else
1633 			*result = 0;
1634 		break;
1635 	case PSM_INTR_OP_XLATE_VECTOR:
1636 		ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1637 		*result = psm_translate_irq(dip, ispec->intrspec_vec);
1638 		break;
1639 	case PSM_INTR_OP_GET_CAP:
1640 		*result = 0;
1641 		break;
1642 	case PSM_INTR_OP_GET_PENDING:
1643 	case PSM_INTR_OP_CLEAR_MASK:
1644 	case PSM_INTR_OP_SET_MASK:
1645 	case PSM_INTR_OP_GET_SHARED:
1646 	case PSM_INTR_OP_SET_PRI:
1647 	case PSM_INTR_OP_SET_CAP:
1648 	case PSM_INTR_OP_SET_CPU:
1649 	case PSM_INTR_OP_GET_INTR:
1650 	default:
1651 		return (PSM_FAILURE);
1652 	}
1653 	return (PSM_SUCCESS);
1654 }
1655 /*
1656  * Return 1 if CMT load balancing policies should be
1657  * implemented across instances of the specified hardware
1658  * sharing relationship.
1659  */
1660 int
1661 pg_cmt_load_bal_hw(pghw_type_t hw)
1662 {
1663 	if (hw == PGHW_IPIPE ||
1664 	    hw == PGHW_FPU ||
1665 	    hw == PGHW_PROCNODE ||
1666 	    hw == PGHW_CHIP)
1667 		return (1);
1668 	else
1669 		return (0);
1670 }
1671 /*
1672  * Return 1 if thread affinity polices should be implemented
1673  * for instances of the specifed hardware sharing relationship.
1674  */
1675 int
1676 pg_cmt_affinity_hw(pghw_type_t hw)
1677 {
1678 	if (hw == PGHW_CACHE)
1679 		return (1);
1680 	else
1681 		return (0);
1682 }
1683