xref: /titanic_51/usr/src/uts/i86pc/os/mp_machdep.c (revision c3ea284015697c0da5651ef400f3dc6bc430b290)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #define	PSMI_1_6
27 #include <sys/smp_impldefs.h>
28 #include <sys/psm.h>
29 #include <sys/psm_modctl.h>
30 #include <sys/pit.h>
31 #include <sys/cmn_err.h>
32 #include <sys/strlog.h>
33 #include <sys/clock.h>
34 #include <sys/debug.h>
35 #include <sys/rtc.h>
36 #include <sys/x86_archext.h>
37 #include <sys/cpupart.h>
38 #include <sys/cpuvar.h>
39 #include <sys/cmt.h>
40 #include <sys/cpu.h>
41 #include <sys/disp.h>
42 #include <sys/archsystm.h>
43 #include <sys/machsystm.h>
44 #include <sys/sysmacros.h>
45 #include <sys/memlist.h>
46 #include <sys/param.h>
47 #include <sys/promif.h>
48 #if defined(__xpv)
49 #include <sys/hypervisor.h>
50 #endif
51 #include <sys/mach_intr.h>
52 #include <vm/hat_i86.h>
53 #include <sys/kdi_machimpl.h>
54 #include <sys/sdt.h>
55 
56 #define	OFFSETOF(s, m)		(size_t)(&(((s *)0)->m))
57 
58 /*
59  *	Local function prototypes
60  */
61 static int mp_disable_intr(processorid_t cpun);
62 static void mp_enable_intr(processorid_t cpun);
63 static void mach_init();
64 static void mach_picinit();
65 static int machhztomhz(uint64_t cpu_freq_hz);
66 static uint64_t mach_getcpufreq(void);
67 static void mach_fixcpufreq(void);
68 static int mach_clkinit(int, int *);
69 static void mach_smpinit(void);
70 static int mach_softlvl_to_vect(int ipl);
71 static void mach_get_platform(int owner);
72 static void mach_construct_info();
73 static int mach_translate_irq(dev_info_t *dip, int irqno);
74 static int mach_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
75     psm_intr_op_t, int *);
76 static void mach_notify_error(int level, char *errmsg);
77 static hrtime_t dummy_hrtime(void);
78 static void dummy_scalehrtime(hrtime_t *);
79 static void cpu_idle(void);
80 static void cpu_wakeup(cpu_t *, int);
81 #ifndef __xpv
82 static void cpu_idle_mwait(void);
83 static void cpu_wakeup_mwait(cpu_t *, int);
84 #endif
85 /*
86  *	External reference functions
87  */
88 extern void return_instr();
89 extern uint64_t freq_tsc(uint32_t *);
90 #if defined(__i386)
91 extern uint64_t freq_notsc(uint32_t *);
92 #endif
93 extern void pc_gethrestime(timestruc_t *);
94 extern int cpuid_get_coreid(cpu_t *);
95 extern int cpuid_get_chipid(cpu_t *);
96 
97 /*
98  *	PSM functions initialization
99  */
100 void (*psm_shutdownf)(int, int)	= (void (*)(int, int))return_instr;
101 void (*psm_preshutdownf)(int, int) = (void (*)(int, int))return_instr;
102 void (*psm_notifyf)(int)	= (void (*)(int))return_instr;
103 void (*psm_set_idle_cpuf)(int)	= (void (*)(int))return_instr;
104 void (*psm_unset_idle_cpuf)(int) = (void (*)(int))return_instr;
105 void (*psminitf)()		= mach_init;
106 void (*picinitf)() 		= return_instr;
107 int (*clkinitf)(int, int *) 	= (int (*)(int, int *))return_instr;
108 int (*ap_mlsetup)() 		= (int (*)(void))return_instr;
109 void (*send_dirintf)() 		= return_instr;
110 void (*setspl)(int)		= (void (*)(int))return_instr;
111 int (*addspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
112 int (*delspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
113 void (*kdisetsoftint)(int, struct av_softinfo *)=
114 	(void (*)(int, struct av_softinfo *))return_instr;
115 void (*setsoftint)(int, struct av_softinfo *)=
116 	(void (*)(int, struct av_softinfo *))return_instr;
117 int (*slvltovect)(int)		= (int (*)(int))return_instr;
118 int (*setlvl)(int, int *)	= (int (*)(int, int *))return_instr;
119 void (*setlvlx)(int, int)	= (void (*)(int, int))return_instr;
120 int (*psm_disable_intr)(int)	= mp_disable_intr;
121 void (*psm_enable_intr)(int)	= mp_enable_intr;
122 hrtime_t (*gethrtimef)(void)	= dummy_hrtime;
123 hrtime_t (*gethrtimeunscaledf)(void)	= dummy_hrtime;
124 void (*scalehrtimef)(hrtime_t *)	= dummy_scalehrtime;
125 int (*psm_translate_irq)(dev_info_t *, int) = mach_translate_irq;
126 void (*gethrestimef)(timestruc_t *) = pc_gethrestime;
127 void (*psm_notify_error)(int, char *) = (void (*)(int, char *))NULL;
128 int (*psm_get_clockirq)(int) = NULL;
129 int (*psm_get_ipivect)(int, int) = NULL;
130 
131 int (*psm_clkinit)(int) = NULL;
132 void (*psm_timer_reprogram)(hrtime_t) = NULL;
133 void (*psm_timer_enable)(void) = NULL;
134 void (*psm_timer_disable)(void) = NULL;
135 void (*psm_post_cyclic_setup)(void *arg) = NULL;
136 int (*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *, psm_intr_op_t,
137     int *) = mach_intr_ops;
138 int (*psm_state)(psm_state_request_t *) = (int (*)(psm_state_request_t *))
139     return_instr;
140 
141 void (*notify_error)(int, char *) = (void (*)(int, char *))return_instr;
142 void (*hrtime_tick)(void)	= return_instr;
143 
144 /*
145  * True if the generic TSC code is our source of hrtime, rather than whatever
146  * the PSM can provide.
147  */
148 #ifdef __xpv
149 int tsc_gethrtime_enable = 0;
150 #else
151 int tsc_gethrtime_enable = 1;
152 #endif
153 int tsc_gethrtime_initted = 0;
154 
155 /*
156  * True if the hrtime implementation is "hires"; namely, better than microdata.
157  */
158 int gethrtime_hires = 0;
159 
160 /*
161  * Local Static Data
162  */
163 static struct psm_ops mach_ops;
164 static struct psm_ops *mach_set[4] = {&mach_ops, NULL, NULL, NULL};
165 static ushort_t mach_ver[4] = {0, 0, 0, 0};
166 
167 /*
168  * If non-zero, idle cpus will become "halted" when there's
169  * no work to do.
170  */
171 int	idle_cpu_use_hlt = 1;
172 
173 #ifndef __xpv
174 /*
175  * If non-zero, idle cpus will use mwait if available to halt instead of hlt.
176  */
177 int	idle_cpu_prefer_mwait = 1;
178 /*
179  * Set to 0 to avoid MONITOR+CLFLUSH assertion.
180  */
181 int	idle_cpu_assert_cflush_monitor = 1;
182 
183 #endif
184 
185 /*ARGSUSED*/
186 int
187 pg_plat_hw_shared(cpu_t *cp, pghw_type_t hw)
188 {
189 	switch (hw) {
190 	case PGHW_IPIPE:
191 		if (x86_feature & (X86_HTT)) {
192 			/*
193 			 * Hyper-threading is SMT
194 			 */
195 			return (1);
196 		} else {
197 			return (0);
198 		}
199 	case PGHW_CHIP:
200 		if (x86_feature & (X86_CMP|X86_HTT))
201 			return (1);
202 		else
203 			return (0);
204 	case PGHW_CACHE:
205 		if (cpuid_get_ncpu_sharing_last_cache(cp) > 1)
206 			return (1);
207 		else
208 			return (0);
209 	default:
210 		return (0);
211 	}
212 }
213 
214 /*
215  * Compare two CPUs and see if they have a pghw_type_t sharing relationship
216  * If pghw_type_t is an unsupported hardware type, then return -1
217  */
218 int
219 pg_plat_cpus_share(cpu_t *cpu_a, cpu_t *cpu_b, pghw_type_t hw)
220 {
221 	id_t pgp_a, pgp_b;
222 
223 	pgp_a = pg_plat_hw_instance_id(cpu_a, hw);
224 	pgp_b = pg_plat_hw_instance_id(cpu_b, hw);
225 
226 	if (pgp_a == -1 || pgp_b == -1)
227 		return (-1);
228 
229 	return (pgp_a == pgp_b);
230 }
231 
232 /*
233  * Return a physical instance identifier for known hardware sharing
234  * relationships
235  */
236 id_t
237 pg_plat_hw_instance_id(cpu_t *cpu, pghw_type_t hw)
238 {
239 	switch (hw) {
240 	case PGHW_IPIPE:
241 		return (cpuid_get_coreid(cpu));
242 	case PGHW_CACHE:
243 		return (cpuid_get_last_lvl_cacheid(cpu));
244 	case PGHW_CHIP:
245 		return (cpuid_get_chipid(cpu));
246 	default:
247 		return (-1);
248 	}
249 }
250 
251 int
252 pg_plat_hw_level(pghw_type_t hw)
253 {
254 	int i;
255 	static pghw_type_t hw_hier[] = {
256 		PGHW_IPIPE,
257 		PGHW_CACHE,
258 		PGHW_CHIP,
259 		PGHW_NUM_COMPONENTS
260 	};
261 
262 	for (i = 0; hw_hier[i] != PGHW_NUM_COMPONENTS; i++) {
263 		if (hw_hier[i] == hw)
264 			return (i);
265 	}
266 	return (-1);
267 }
268 
269 /*
270  * Return 1 if CMT load balancing policies should be
271  * implemented across instances of the specified hardware
272  * sharing relationship.
273  */
274 int
275 pg_plat_cmt_load_bal_hw(pghw_type_t hw)
276 {
277 	if (hw == PGHW_IPIPE ||
278 	    hw == PGHW_FPU ||
279 	    hw == PGHW_CHIP ||
280 	    hw == PGHW_CACHE)
281 		return (1);
282 	else
283 		return (0);
284 }
285 
286 
287 /*
288  * Return 1 if thread affinity polices should be implemented
289  * for instances of the specifed hardware sharing relationship.
290  */
291 int
292 pg_plat_cmt_affinity_hw(pghw_type_t hw)
293 {
294 	if (hw == PGHW_CACHE)
295 		return (1);
296 	else
297 		return (0);
298 }
299 
300 id_t
301 pg_plat_get_core_id(cpu_t *cpu)
302 {
303 	return ((id_t)cpuid_get_coreid(cpu));
304 }
305 
306 void
307 cmp_set_nosteal_interval(void)
308 {
309 	/* Set the nosteal interval (used by disp_getbest()) to 100us */
310 	nosteal_nsec = 100000UL;
311 }
312 
313 /*
314  * Routine to ensure initial callers to hrtime gets 0 as return
315  */
316 static hrtime_t
317 dummy_hrtime(void)
318 {
319 	return (0);
320 }
321 
322 /* ARGSUSED */
323 static void
324 dummy_scalehrtime(hrtime_t *ticks)
325 {}
326 
327 /*
328  * Idle the present CPU until awoken via an interrupt
329  */
330 static void
331 cpu_idle(void)
332 {
333 	cpu_t		*cpup = CPU;
334 	processorid_t	cpu_sid = cpup->cpu_seqid;
335 	cpupart_t	*cp = cpup->cpu_part;
336 	int		hset_update = 1;
337 
338 	/*
339 	 * If this CPU is online, and there's multiple CPUs
340 	 * in the system, then we should notate our halting
341 	 * by adding ourselves to the partition's halted CPU
342 	 * bitmap. This allows other CPUs to find/awaken us when
343 	 * work becomes available.
344 	 */
345 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
346 		hset_update = 0;
347 
348 	/*
349 	 * Add ourselves to the partition's halted CPUs bitmap
350 	 * and set our HALTED flag, if necessary.
351 	 *
352 	 * When a thread becomes runnable, it is placed on the queue
353 	 * and then the halted CPU bitmap is checked to determine who
354 	 * (if anyone) should be awoken. We therefore need to first
355 	 * add ourselves to the bitmap, and and then check if there
356 	 * is any work available. The order is important to prevent a race
357 	 * that can lead to work languishing on a run queue somewhere while
358 	 * this CPU remains halted.
359 	 *
360 	 * Either the producing CPU will see we're halted and will awaken us,
361 	 * or this CPU will see the work available in disp_anywork().
362 	 *
363 	 * Note that memory barriers after updating the HALTED flag
364 	 * are not necessary since an atomic operation (updating the bitset)
365 	 * immediately follows. On x86 the atomic operation acts as a
366 	 * memory barrier for the update of cpu_disp_flags.
367 	 */
368 	if (hset_update) {
369 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
370 		bitset_atomic_add(&cp->cp_haltset, cpu_sid);
371 	}
372 
373 	/*
374 	 * Check to make sure there's really nothing to do.
375 	 * Work destined for this CPU may become available after
376 	 * this check. We'll be notified through the clearing of our
377 	 * bit in the halted CPU bitmap, and a poke.
378 	 */
379 	if (disp_anywork()) {
380 		if (hset_update) {
381 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
382 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
383 		}
384 		return;
385 	}
386 
387 	/*
388 	 * We're on our way to being halted.
389 	 *
390 	 * Disable interrupts now, so that we'll awaken immediately
391 	 * after halting if someone tries to poke us between now and
392 	 * the time we actually halt.
393 	 *
394 	 * We check for the presence of our bit after disabling interrupts.
395 	 * If it's cleared, we'll return. If the bit is cleared after
396 	 * we check then the poke will pop us out of the halted state.
397 	 *
398 	 * This means that the ordering of the poke and the clearing
399 	 * of the bit by cpu_wakeup is important.
400 	 * cpu_wakeup() must clear, then poke.
401 	 * cpu_idle() must disable interrupts, then check for the bit.
402 	 */
403 	cli();
404 
405 	if (hset_update && bitset_in_set(&cp->cp_haltset, cpu_sid) == 0) {
406 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
407 		sti();
408 		return;
409 	}
410 
411 	/*
412 	 * The check for anything locally runnable is here for performance
413 	 * and isn't needed for correctness. disp_nrunnable ought to be
414 	 * in our cache still, so it's inexpensive to check, and if there
415 	 * is anything runnable we won't have to wait for the poke.
416 	 */
417 	if (cpup->cpu_disp->disp_nrunnable != 0) {
418 		if (hset_update) {
419 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
420 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
421 		}
422 		sti();
423 		return;
424 	}
425 
426 	DTRACE_PROBE1(idle__state__transition, uint_t, IDLE_STATE_C1);
427 
428 	mach_cpu_idle();
429 
430 	DTRACE_PROBE1(idle__state__transition, uint_t, IDLE_STATE_C0);
431 
432 	/*
433 	 * We're no longer halted
434 	 */
435 	if (hset_update) {
436 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
437 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
438 	}
439 }
440 
441 
442 /*
443  * If "cpu" is halted, then wake it up clearing its halted bit in advance.
444  * Otherwise, see if other CPUs in the cpu partition are halted and need to
445  * be woken up so that they can steal the thread we placed on this CPU.
446  * This function is only used on MP systems.
447  */
448 static void
449 cpu_wakeup(cpu_t *cpu, int bound)
450 {
451 	uint_t		cpu_found;
452 	processorid_t	cpu_sid;
453 	cpupart_t	*cp;
454 
455 	cp = cpu->cpu_part;
456 	cpu_sid = cpu->cpu_seqid;
457 	if (bitset_in_set(&cp->cp_haltset, cpu_sid)) {
458 		/*
459 		 * Clear the halted bit for that CPU since it will be
460 		 * poked in a moment.
461 		 */
462 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
463 		/*
464 		 * We may find the current CPU present in the halted cpuset
465 		 * if we're in the context of an interrupt that occurred
466 		 * before we had a chance to clear our bit in cpu_idle().
467 		 * Poking ourself is obviously unnecessary, since if
468 		 * we're here, we're not halted.
469 		 */
470 		if (cpu != CPU)
471 			poke_cpu(cpu->cpu_id);
472 		return;
473 	} else {
474 		/*
475 		 * This cpu isn't halted, but it's idle or undergoing a
476 		 * context switch. No need to awaken anyone else.
477 		 */
478 		if (cpu->cpu_thread == cpu->cpu_idle_thread ||
479 		    cpu->cpu_disp_flags & CPU_DISP_DONTSTEAL)
480 			return;
481 	}
482 
483 	/*
484 	 * No need to wake up other CPUs if this is for a bound thread.
485 	 */
486 	if (bound)
487 		return;
488 
489 	/*
490 	 * The CPU specified for wakeup isn't currently halted, so check
491 	 * to see if there are any other halted CPUs in the partition,
492 	 * and if there are then awaken one.
493 	 */
494 	do {
495 		cpu_found = bitset_find(&cp->cp_haltset);
496 		if (cpu_found == (uint_t)-1)
497 			return;
498 	} while (bitset_atomic_test_and_del(&cp->cp_haltset, cpu_found) < 0);
499 
500 	if (cpu_found != CPU->cpu_seqid) {
501 		poke_cpu(cpu_seq[cpu_found]->cpu_id);
502 	}
503 }
504 
505 #ifndef __xpv
506 /*
507  * Idle the present CPU until awoken via touching its monitored line
508  */
509 static void
510 cpu_idle_mwait(void)
511 {
512 	volatile uint32_t	*mcpu_mwait = CPU->cpu_m.mcpu_mwait;
513 	cpu_t			*cpup = CPU;
514 	processorid_t		cpu_sid = cpup->cpu_seqid;
515 	cpupart_t		*cp = cpup->cpu_part;
516 	int			hset_update = 1;
517 
518 	/*
519 	 * Set our mcpu_mwait here, so we can tell if anyone trys to
520 	 * wake us between now and when we call mwait.  No other cpu will
521 	 * attempt to set our mcpu_mwait until we add ourself to the halted
522 	 * CPU bitmap.
523 	 */
524 	*mcpu_mwait = MWAIT_HALTED;
525 
526 	/*
527 	 * If this CPU is online, and there's multiple CPUs
528 	 * in the system, then we should notate our halting
529 	 * by adding ourselves to the partition's halted CPU
530 	 * bitmap. This allows other CPUs to find/awaken us when
531 	 * work becomes available.
532 	 */
533 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
534 		hset_update = 0;
535 
536 	/*
537 	 * Add ourselves to the partition's halted CPUs bitmap
538 	 * and set our HALTED flag, if necessary.
539 	 *
540 	 * When a thread becomes runnable, it is placed on the queue
541 	 * and then the halted CPU bitmap is checked to determine who
542 	 * (if anyone) should be awoken. We therefore need to first
543 	 * add ourselves to the bitmap, and and then check if there
544 	 * is any work available.
545 	 *
546 	 * Note that memory barriers after updating the HALTED flag
547 	 * are not necessary since an atomic operation (updating the bitmap)
548 	 * immediately follows. On x86 the atomic operation acts as a
549 	 * memory barrier for the update of cpu_disp_flags.
550 	 */
551 	if (hset_update) {
552 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
553 		bitset_atomic_add(&cp->cp_haltset, cpu_sid);
554 	}
555 
556 	/*
557 	 * Check to make sure there's really nothing to do.
558 	 * Work destined for this CPU may become available after
559 	 * this check. We'll be notified through the clearing of our
560 	 * bit in the halted CPU bitmap, and a write to our mcpu_mwait.
561 	 *
562 	 * disp_anywork() checks disp_nrunnable, so we do not have to later.
563 	 */
564 	if (disp_anywork()) {
565 		if (hset_update) {
566 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
567 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
568 		}
569 		return;
570 	}
571 
572 	/*
573 	 * We're on our way to being halted.
574 	 * To avoid a lost wakeup, arm the monitor before checking if another
575 	 * cpu wrote to mcpu_mwait to wake us up.
576 	 */
577 	i86_monitor(mcpu_mwait, 0, 0);
578 	if (*mcpu_mwait == MWAIT_HALTED) {
579 		DTRACE_PROBE1(idle__state__transition, uint_t, IDLE_STATE_C1);
580 
581 		tlb_going_idle();
582 		i86_mwait(0, 0);
583 		tlb_service();
584 
585 		DTRACE_PROBE1(idle__state__transition, uint_t, IDLE_STATE_C0);
586 	}
587 
588 	/*
589 	 * We're no longer halted
590 	 */
591 	if (hset_update) {
592 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
593 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
594 	}
595 }
596 
597 /*
598  * If "cpu" is halted in mwait, then wake it up clearing its halted bit in
599  * advance.  Otherwise, see if other CPUs in the cpu partition are halted and
600  * need to be woken up so that they can steal the thread we placed on this CPU.
601  * This function is only used on MP systems.
602  */
603 static void
604 cpu_wakeup_mwait(cpu_t *cp, int bound)
605 {
606 	cpupart_t	*cpu_part;
607 	uint_t		cpu_found;
608 	processorid_t	cpu_sid;
609 
610 	cpu_part = cp->cpu_part;
611 	cpu_sid = cp->cpu_seqid;
612 
613 	/*
614 	 * Clear the halted bit for that CPU since it will be woken up
615 	 * in a moment.
616 	 */
617 	if (bitset_in_set(&cpu_part->cp_haltset, cpu_sid)) {
618 		/*
619 		 * Clear the halted bit for that CPU since it will be
620 		 * poked in a moment.
621 		 */
622 		bitset_atomic_del(&cpu_part->cp_haltset, cpu_sid);
623 		/*
624 		 * We may find the current CPU present in the halted cpuset
625 		 * if we're in the context of an interrupt that occurred
626 		 * before we had a chance to clear our bit in cpu_idle().
627 		 * Waking ourself is obviously unnecessary, since if
628 		 * we're here, we're not halted.
629 		 *
630 		 * monitor/mwait wakeup via writing to our cache line is
631 		 * harmless and less expensive than always checking if we
632 		 * are waking ourself which is an uncommon case.
633 		 */
634 		MWAIT_WAKEUP(cp);	/* write to monitored line */
635 		return;
636 	} else {
637 		/*
638 		 * This cpu isn't halted, but it's idle or undergoing a
639 		 * context switch. No need to awaken anyone else.
640 		 */
641 		if (cp->cpu_thread == cp->cpu_idle_thread ||
642 		    cp->cpu_disp_flags & CPU_DISP_DONTSTEAL)
643 			return;
644 	}
645 
646 	/*
647 	 * No need to wake up other CPUs if the thread we just enqueued
648 	 * is bound.
649 	 */
650 	if (bound || ncpus == 1)
651 		return;
652 
653 	/*
654 	 * See if there's any other halted CPUs. If there are, then
655 	 * select one, and awaken it.
656 	 * It's possible that after we find a CPU, somebody else
657 	 * will awaken it before we get the chance.
658 	 * In that case, look again.
659 	 */
660 	do {
661 		cpu_found = bitset_find(&cpu_part->cp_haltset);
662 		if (cpu_found == (uint_t)-1)
663 			return;
664 	} while (bitset_atomic_test_and_del(&cpu_part->cp_haltset,
665 	    cpu_found) < 0);
666 
667 	/*
668 	 * Do not check if cpu_found is ourself as monitor/mwait
669 	 * wakeup is cheap.
670 	 */
671 	MWAIT_WAKEUP(cpu_seq[cpu_found]); /* write to monitored line */
672 }
673 
674 #endif
675 
676 void (*cpu_pause_handler)(volatile char *) = NULL;
677 
678 static int
679 mp_disable_intr(int cpun)
680 {
681 	/*
682 	 * switch to the offline cpu
683 	 */
684 	affinity_set(cpun);
685 	/*
686 	 * raise ipl to just below cross call
687 	 */
688 	splx(XC_MED_PIL-1);
689 	/*
690 	 *	set base spl to prevent the next swtch to idle from
691 	 *	lowering back to ipl 0
692 	 */
693 	CPU->cpu_intr_actv |= (1 << (XC_MED_PIL-1));
694 	set_base_spl();
695 	affinity_clear();
696 	return (DDI_SUCCESS);
697 }
698 
699 static void
700 mp_enable_intr(int cpun)
701 {
702 	/*
703 	 * switch to the online cpu
704 	 */
705 	affinity_set(cpun);
706 	/*
707 	 * clear the interrupt active mask
708 	 */
709 	CPU->cpu_intr_actv &= ~(1 << (XC_MED_PIL-1));
710 	set_base_spl();
711 	(void) spl0();
712 	affinity_clear();
713 }
714 
715 static void
716 mach_get_platform(int owner)
717 {
718 	void		**srv_opsp;
719 	void		**clt_opsp;
720 	int		i;
721 	int		total_ops;
722 
723 	/* fix up psm ops */
724 	srv_opsp = (void **)mach_set[0];
725 	clt_opsp = (void **)mach_set[owner];
726 	if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01)
727 		total_ops = sizeof (struct psm_ops_ver01) /
728 		    sizeof (void (*)(void));
729 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_1)
730 		/* no psm_notify_func */
731 		total_ops = OFFSETOF(struct psm_ops, psm_notify_func) /
732 		    sizeof (void (*)(void));
733 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_2)
734 		/* no psm_timer funcs */
735 		total_ops = OFFSETOF(struct psm_ops, psm_timer_reprogram) /
736 		    sizeof (void (*)(void));
737 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_3)
738 		/* no psm_preshutdown function */
739 		total_ops = OFFSETOF(struct psm_ops, psm_preshutdown) /
740 		    sizeof (void (*)(void));
741 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_4)
742 		/* no psm_preshutdown function */
743 		total_ops = OFFSETOF(struct psm_ops, psm_intr_ops) /
744 		    sizeof (void (*)(void));
745 	else
746 		total_ops = sizeof (struct psm_ops) / sizeof (void (*)(void));
747 
748 	/*
749 	 * Save the version of the PSM module, in case we need to
750 	 * bahave differently based on version.
751 	 */
752 	mach_ver[0] = mach_ver[owner];
753 
754 	for (i = 0; i < total_ops; i++)
755 		if (clt_opsp[i] != NULL)
756 			srv_opsp[i] = clt_opsp[i];
757 }
758 
759 static void
760 mach_construct_info()
761 {
762 	struct psm_sw *swp;
763 	int	mach_cnt[PSM_OWN_OVERRIDE+1] = {0};
764 	int	conflict_owner = 0;
765 
766 	if (psmsw->psw_forw == psmsw)
767 		panic("No valid PSM modules found");
768 	mutex_enter(&psmsw_lock);
769 	for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
770 		if (!(swp->psw_flag & PSM_MOD_IDENTIFY))
771 			continue;
772 		mach_set[swp->psw_infop->p_owner] = swp->psw_infop->p_ops;
773 		mach_ver[swp->psw_infop->p_owner] = swp->psw_infop->p_version;
774 		mach_cnt[swp->psw_infop->p_owner]++;
775 	}
776 	mutex_exit(&psmsw_lock);
777 
778 	mach_get_platform(PSM_OWN_SYS_DEFAULT);
779 
780 	/* check to see are there any conflicts */
781 	if (mach_cnt[PSM_OWN_EXCLUSIVE] > 1)
782 		conflict_owner = PSM_OWN_EXCLUSIVE;
783 	if (mach_cnt[PSM_OWN_OVERRIDE] > 1)
784 		conflict_owner = PSM_OWN_OVERRIDE;
785 	if (conflict_owner) {
786 		/* remove all psm modules except uppc */
787 		cmn_err(CE_WARN,
788 		    "Conflicts detected on the following PSM modules:");
789 		mutex_enter(&psmsw_lock);
790 		for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
791 			if (swp->psw_infop->p_owner == conflict_owner)
792 				cmn_err(CE_WARN, "%s ",
793 				    swp->psw_infop->p_mach_idstring);
794 		}
795 		mutex_exit(&psmsw_lock);
796 		cmn_err(CE_WARN,
797 		    "Setting the system back to SINGLE processor mode!");
798 		cmn_err(CE_WARN,
799 		    "Please edit /etc/mach to remove the invalid PSM module.");
800 		return;
801 	}
802 
803 	if (mach_set[PSM_OWN_EXCLUSIVE])
804 		mach_get_platform(PSM_OWN_EXCLUSIVE);
805 
806 	if (mach_set[PSM_OWN_OVERRIDE])
807 		mach_get_platform(PSM_OWN_OVERRIDE);
808 }
809 
810 static void
811 mach_init()
812 {
813 	struct psm_ops  *pops;
814 
815 	mach_construct_info();
816 
817 	pops = mach_set[0];
818 
819 	/* register the interrupt and clock initialization rotuines */
820 	picinitf = mach_picinit;
821 	clkinitf = mach_clkinit;
822 	psm_get_clockirq = pops->psm_get_clockirq;
823 
824 	/* register the interrupt setup code */
825 	slvltovect = mach_softlvl_to_vect;
826 	addspl	= pops->psm_addspl;
827 	delspl	= pops->psm_delspl;
828 
829 	if (pops->psm_translate_irq)
830 		psm_translate_irq = pops->psm_translate_irq;
831 	if (pops->psm_intr_ops)
832 		psm_intr_ops = pops->psm_intr_ops;
833 
834 #if defined(PSMI_1_2) || defined(PSMI_1_3) || defined(PSMI_1_4)
835 	/*
836 	 * Time-of-day functionality now handled in TOD modules.
837 	 * (Warn about PSM modules that think that we're going to use
838 	 * their ops vectors.)
839 	 */
840 	if (pops->psm_tod_get)
841 		cmn_err(CE_WARN, "obsolete psm_tod_get op %p",
842 		    (void *)pops->psm_tod_get);
843 
844 	if (pops->psm_tod_set)
845 		cmn_err(CE_WARN, "obsolete psm_tod_set op %p",
846 		    (void *)pops->psm_tod_set);
847 #endif
848 
849 	if (pops->psm_notify_error) {
850 		psm_notify_error = mach_notify_error;
851 		notify_error = pops->psm_notify_error;
852 	}
853 
854 	(*pops->psm_softinit)();
855 
856 	/*
857 	 * Initialize the dispatcher's function hooks
858 	 * to enable CPU halting when idle.
859 	 * Do not use monitor/mwait if idle_cpu_use_hlt is not set(spin idle)
860 	 * or idle_cpu_prefer_mwait is not set.
861 	 * Allocate monitor/mwait buffer for cpu0.
862 	 */
863 	if (idle_cpu_use_hlt) {
864 		idle_cpu = cpu_idle;
865 #ifndef __xpv
866 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait) {
867 			CPU->cpu_m.mcpu_mwait = cpuid_mwait_alloc(CPU);
868 			/*
869 			 * Protect ourself from insane mwait size.
870 			 */
871 			if (CPU->cpu_m.mcpu_mwait == NULL) {
872 #ifdef DEBUG
873 				cmn_err(CE_NOTE, "Using hlt idle.  Cannot "
874 				    "handle cpu 0 mwait size.");
875 #endif
876 				idle_cpu_prefer_mwait = 0;
877 				idle_cpu = cpu_idle;
878 			} else {
879 				idle_cpu = cpu_idle_mwait;
880 			}
881 		} else {
882 			idle_cpu = cpu_idle;
883 		}
884 #endif
885 	}
886 
887 	mach_smpinit();
888 }
889 
890 static void
891 mach_smpinit(void)
892 {
893 	struct psm_ops  *pops;
894 	processorid_t cpu_id;
895 	int cnt;
896 	cpuset_t cpumask;
897 
898 	pops = mach_set[0];
899 	CPUSET_ZERO(cpumask);
900 
901 	cpu_id = -1;
902 	cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
903 	for (cnt = 0; cpu_id != -1; cnt++) {
904 		CPUSET_ADD(cpumask, cpu_id);
905 		cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
906 	}
907 
908 	mp_cpus = cpumask;
909 
910 	/* MP related routines */
911 	ap_mlsetup = pops->psm_post_cpu_start;
912 	send_dirintf = pops->psm_send_ipi;
913 
914 	/* optional MP related routines */
915 	if (pops->psm_shutdown)
916 		psm_shutdownf = pops->psm_shutdown;
917 	if (pops->psm_preshutdown)
918 		psm_preshutdownf = pops->psm_preshutdown;
919 	if (pops->psm_notify_func)
920 		psm_notifyf = pops->psm_notify_func;
921 	if (pops->psm_set_idlecpu)
922 		psm_set_idle_cpuf = pops->psm_set_idlecpu;
923 	if (pops->psm_unset_idlecpu)
924 		psm_unset_idle_cpuf = pops->psm_unset_idlecpu;
925 
926 	psm_clkinit = pops->psm_clkinit;
927 
928 	if (pops->psm_timer_reprogram)
929 		psm_timer_reprogram = pops->psm_timer_reprogram;
930 
931 	if (pops->psm_timer_enable)
932 		psm_timer_enable = pops->psm_timer_enable;
933 
934 	if (pops->psm_timer_disable)
935 		psm_timer_disable = pops->psm_timer_disable;
936 
937 	if (pops->psm_post_cyclic_setup)
938 		psm_post_cyclic_setup = pops->psm_post_cyclic_setup;
939 
940 	if (pops->psm_state)
941 		psm_state = pops->psm_state;
942 
943 	/*
944 	 * Set these vectors here so they can be used by Suspend/Resume
945 	 * on UP machines.
946 	 */
947 	if (pops->psm_disable_intr)
948 		psm_disable_intr = pops->psm_disable_intr;
949 	if (pops->psm_enable_intr)
950 		psm_enable_intr  = pops->psm_enable_intr;
951 
952 	/* check for multiple CPUs */
953 	if (cnt < 2)
954 		return;
955 
956 	/* check for MP platforms */
957 	if (pops->psm_cpu_start == NULL)
958 		return;
959 
960 	/*
961 	 * Set the dispatcher hook to enable cpu "wake up"
962 	 * when a thread becomes runnable.
963 	 */
964 	if (idle_cpu_use_hlt) {
965 		disp_enq_thread = cpu_wakeup;
966 #ifndef __xpv
967 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait)
968 			disp_enq_thread = cpu_wakeup_mwait;
969 #endif
970 	}
971 
972 	psm_get_ipivect = pops->psm_get_ipivect;
973 
974 	(void) add_avintr((void *)NULL, XC_HI_PIL, xc_serv, "xc_hi_intr",
975 	    (*pops->psm_get_ipivect)(XC_HI_PIL, PSM_INTR_IPI_HI),
976 	    (caddr_t)X_CALL_HIPRI, NULL, NULL, NULL);
977 	(void) add_avintr((void *)NULL, XC_MED_PIL, xc_serv, "xc_med_intr",
978 	    (*pops->psm_get_ipivect)(XC_MED_PIL, PSM_INTR_IPI_LO),
979 	    (caddr_t)X_CALL_MEDPRI, NULL, NULL, NULL);
980 
981 	(void) (*pops->psm_get_ipivect)(XC_CPUPOKE_PIL, PSM_INTR_POKE);
982 }
983 
984 static void
985 mach_picinit()
986 {
987 	struct psm_ops  *pops;
988 
989 	pops = mach_set[0];
990 
991 	/* register the interrupt handlers */
992 	setlvl = pops->psm_intr_enter;
993 	setlvlx = pops->psm_intr_exit;
994 
995 	/* initialize the interrupt hardware */
996 	(*pops->psm_picinit)();
997 
998 	/* set interrupt mask for current ipl */
999 	setspl = pops->psm_setspl;
1000 	cli();
1001 	setspl(CPU->cpu_pri);
1002 }
1003 
1004 uint_t	cpu_freq;	/* MHz */
1005 uint64_t cpu_freq_hz;	/* measured (in hertz) */
1006 
1007 #define	MEGA_HZ		1000000
1008 
1009 #ifdef __xpv
1010 
1011 int xpv_cpufreq_workaround = 1;
1012 int xpv_cpufreq_verbose = 0;
1013 
1014 #else	/* __xpv */
1015 
1016 static uint64_t
1017 mach_calchz(uint32_t pit_counter, uint64_t *processor_clks)
1018 {
1019 	uint64_t cpu_hz;
1020 
1021 	if ((pit_counter == 0) || (*processor_clks == 0) ||
1022 	    (*processor_clks > (((uint64_t)-1) / PIT_HZ)))
1023 		return (0);
1024 
1025 	cpu_hz = ((uint64_t)PIT_HZ * *processor_clks) / pit_counter;
1026 
1027 	return (cpu_hz);
1028 }
1029 
1030 #endif	/* __xpv */
1031 
1032 static uint64_t
1033 mach_getcpufreq(void)
1034 {
1035 #if defined(__xpv)
1036 	vcpu_time_info_t *vti = &CPU->cpu_m.mcpu_vcpu_info->time;
1037 	uint64_t cpu_hz;
1038 
1039 	/*
1040 	 * During dom0 bringup, it was noted that on at least one older
1041 	 * Intel HT machine, the hypervisor initially gives a tsc_to_system_mul
1042 	 * value that is quite wrong (the 3.06GHz clock was reported
1043 	 * as 4.77GHz)
1044 	 *
1045 	 * The curious thing is, that if you stop the kernel at entry,
1046 	 * breakpoint here and inspect the value with kmdb, the value
1047 	 * is correct - but if you don't stop and simply enable the
1048 	 * printf statement (below), you can see the bad value printed
1049 	 * here.  Almost as if something kmdb did caused the hypervisor to
1050 	 * figure it out correctly.  And, note that the hypervisor
1051 	 * eventually -does- figure it out correctly ... if you look at
1052 	 * the field later in the life of dom0, it is correct.
1053 	 *
1054 	 * For now, on dom0, we employ a slightly cheesy workaround of
1055 	 * using the DOM0_PHYSINFO hypercall.
1056 	 */
1057 	if (DOMAIN_IS_INITDOMAIN(xen_info) && xpv_cpufreq_workaround) {
1058 		xen_sysctl_t op0, *op = &op0;
1059 
1060 		op->cmd = XEN_SYSCTL_physinfo;
1061 		op->interface_version = XEN_SYSCTL_INTERFACE_VERSION;
1062 		if (HYPERVISOR_sysctl(op) != 0)
1063 			panic("physinfo op refused");
1064 
1065 		cpu_hz = 1000 * (uint64_t)op->u.physinfo.cpu_khz;
1066 	} else {
1067 		cpu_hz = (UINT64_C(1000000000) << 32) / vti->tsc_to_system_mul;
1068 
1069 		if (vti->tsc_shift < 0)
1070 			cpu_hz <<= -vti->tsc_shift;
1071 		else
1072 			cpu_hz >>= vti->tsc_shift;
1073 	}
1074 
1075 	if (xpv_cpufreq_verbose)
1076 		printf("mach_getcpufreq: system_mul 0x%x, shift %d, "
1077 		    "cpu_hz %" PRId64 "Hz\n",
1078 		    vti->tsc_to_system_mul, vti->tsc_shift, cpu_hz);
1079 
1080 	return (cpu_hz);
1081 #else	/* __xpv */
1082 	uint32_t pit_counter;
1083 	uint64_t processor_clks;
1084 
1085 	if (x86_feature & X86_TSC) {
1086 		/*
1087 		 * We have a TSC. freq_tsc() knows how to measure the number
1088 		 * of clock cycles sampled against the PIT.
1089 		 */
1090 		ulong_t flags = clear_int_flag();
1091 		processor_clks = freq_tsc(&pit_counter);
1092 		restore_int_flag(flags);
1093 		return (mach_calchz(pit_counter, &processor_clks));
1094 	} else if (x86_vendor == X86_VENDOR_Cyrix || x86_type == X86_TYPE_P5) {
1095 #if defined(__amd64)
1096 		panic("mach_getcpufreq: no TSC!");
1097 #elif defined(__i386)
1098 		/*
1099 		 * We are a Cyrix based on a 6x86 core or an Intel Pentium
1100 		 * for which freq_notsc() knows how to measure the number of
1101 		 * elapsed clock cycles sampled against the PIT
1102 		 */
1103 		ulong_t flags = clear_int_flag();
1104 		processor_clks = freq_notsc(&pit_counter);
1105 		restore_int_flag(flags);
1106 		return (mach_calchz(pit_counter, &processor_clks));
1107 #endif	/* __i386 */
1108 	}
1109 
1110 	/* We do not know how to calculate cpu frequency for this cpu. */
1111 	return (0);
1112 #endif	/* __xpv */
1113 }
1114 
1115 /*
1116  * If the clock speed of a cpu is found to be reported incorrectly, do not add
1117  * to this array, instead improve the accuracy of the algorithm that determines
1118  * the clock speed of the processor or extend the implementation to support the
1119  * vendor as appropriate. This is here only to support adjusting the speed on
1120  * older slower processors that mach_fixcpufreq() would not be able to account
1121  * for otherwise.
1122  */
1123 static int x86_cpu_freq[] = { 60, 75, 80, 90, 120, 160, 166, 175, 180, 233 };
1124 
1125 /*
1126  * On fast processors the clock frequency that is measured may be off by
1127  * a few MHz from the value printed on the part. This is a combination of
1128  * the factors that for such fast parts being off by this much is within
1129  * the tolerances for manufacture and because of the difficulties in the
1130  * measurement that can lead to small error. This function uses some
1131  * heuristics in order to tweak the value that was measured to match what
1132  * is most likely printed on the part.
1133  *
1134  * Some examples:
1135  * 	AMD Athlon 1000 mhz measured as 998 mhz
1136  * 	Intel Pentium III Xeon 733 mhz measured as 731 mhz
1137  * 	Intel Pentium IV 1500 mhz measured as 1495mhz
1138  *
1139  * If in the future this function is no longer sufficient to correct
1140  * for the error in the measurement, then the algorithm used to perform
1141  * the measurement will have to be improved in order to increase accuracy
1142  * rather than adding horrible and questionable kludges here.
1143  *
1144  * This is called after the cyclics subsystem because of the potential
1145  * that the heuristics within may give a worse estimate of the clock
1146  * frequency than the value that was measured.
1147  */
1148 static void
1149 mach_fixcpufreq(void)
1150 {
1151 	uint32_t freq, mul, near66, delta66, near50, delta50, fixed, delta, i;
1152 
1153 	freq = (uint32_t)cpu_freq;
1154 
1155 	/*
1156 	 * Find the nearest integer multiple of 200/3 (about 66) MHz to the
1157 	 * measured speed taking into account that the 667 MHz parts were
1158 	 * the first to round-up.
1159 	 */
1160 	mul = (uint32_t)((3 * (uint64_t)freq + 100) / 200);
1161 	near66 = (uint32_t)((200 * (uint64_t)mul + ((mul >= 10) ? 1 : 0)) / 3);
1162 	delta66 = (near66 > freq) ? (near66 - freq) : (freq - near66);
1163 
1164 	/* Find the nearest integer multiple of 50 MHz to the measured speed */
1165 	mul = (freq + 25) / 50;
1166 	near50 = mul * 50;
1167 	delta50 = (near50 > freq) ? (near50 - freq) : (freq - near50);
1168 
1169 	/* Find the closer of the two */
1170 	if (delta66 < delta50) {
1171 		fixed = near66;
1172 		delta = delta66;
1173 	} else {
1174 		fixed = near50;
1175 		delta = delta50;
1176 	}
1177 
1178 	if (fixed > INT_MAX)
1179 		return;
1180 
1181 	/*
1182 	 * Some older parts have a core clock frequency that is not an
1183 	 * integral multiple of 50 or 66 MHz. Check if one of the old
1184 	 * clock frequencies is closer to the measured value than any
1185 	 * of the integral multiples of 50 an 66, and if so set fixed
1186 	 * and delta appropriately to represent the closest value.
1187 	 */
1188 	i = sizeof (x86_cpu_freq) / sizeof (int);
1189 	while (i > 0) {
1190 		i--;
1191 
1192 		if (x86_cpu_freq[i] <= freq) {
1193 			mul = freq - x86_cpu_freq[i];
1194 
1195 			if (mul < delta) {
1196 				fixed = x86_cpu_freq[i];
1197 				delta = mul;
1198 			}
1199 
1200 			break;
1201 		}
1202 
1203 		mul = x86_cpu_freq[i] - freq;
1204 
1205 		if (mul < delta) {
1206 			fixed = x86_cpu_freq[i];
1207 			delta = mul;
1208 		}
1209 	}
1210 
1211 	/*
1212 	 * Set a reasonable maximum for how much to correct the measured
1213 	 * result by. This check is here to prevent the adjustment made
1214 	 * by this function from being more harm than good. It is entirely
1215 	 * possible that in the future parts will be made that are not
1216 	 * integral multiples of 66 or 50 in clock frequency or that
1217 	 * someone may overclock a part to some odd frequency. If the
1218 	 * measured value is farther from the corrected value than
1219 	 * allowed, then assume the corrected value is in error and use
1220 	 * the measured value.
1221 	 */
1222 	if (6 < delta)
1223 		return;
1224 
1225 	cpu_freq = (int)fixed;
1226 }
1227 
1228 
1229 static int
1230 machhztomhz(uint64_t cpu_freq_hz)
1231 {
1232 	uint64_t cpu_mhz;
1233 
1234 	/* Round to nearest MHZ */
1235 	cpu_mhz = (cpu_freq_hz + (MEGA_HZ / 2)) / MEGA_HZ;
1236 
1237 	if (cpu_mhz > INT_MAX)
1238 		return (0);
1239 
1240 	return ((int)cpu_mhz);
1241 
1242 }
1243 
1244 
1245 static int
1246 mach_clkinit(int preferred_mode, int *set_mode)
1247 {
1248 	struct psm_ops  *pops;
1249 	int resolution;
1250 
1251 	pops = mach_set[0];
1252 
1253 	cpu_freq_hz = mach_getcpufreq();
1254 
1255 	cpu_freq = machhztomhz(cpu_freq_hz);
1256 
1257 	if (!(x86_feature & X86_TSC) || (cpu_freq == 0))
1258 		tsc_gethrtime_enable = 0;
1259 
1260 #ifndef __xpv
1261 	if (tsc_gethrtime_enable) {
1262 		tsc_hrtimeinit(cpu_freq_hz);
1263 	} else
1264 #endif
1265 	{
1266 		if (pops->psm_hrtimeinit)
1267 			(*pops->psm_hrtimeinit)();
1268 		gethrtimef = pops->psm_gethrtime;
1269 		gethrtimeunscaledf = gethrtimef;
1270 		/* scalehrtimef will remain dummy */
1271 	}
1272 
1273 	mach_fixcpufreq();
1274 
1275 	if (mach_ver[0] >= PSM_INFO_VER01_3) {
1276 		if (preferred_mode == TIMER_ONESHOT) {
1277 
1278 			resolution = (*pops->psm_clkinit)(0);
1279 			if (resolution != 0)  {
1280 				*set_mode = TIMER_ONESHOT;
1281 				return (resolution);
1282 			}
1283 		}
1284 
1285 		/*
1286 		 * either periodic mode was requested or could not set to
1287 		 * one-shot mode
1288 		 */
1289 		resolution = (*pops->psm_clkinit)(hz);
1290 		/*
1291 		 * psm should be able to do periodic, so we do not check
1292 		 * for return value of psm_clkinit here.
1293 		 */
1294 		*set_mode = TIMER_PERIODIC;
1295 		return (resolution);
1296 	} else {
1297 		/*
1298 		 * PSMI interface prior to PSMI_3 does not define a return
1299 		 * value for psm_clkinit, so the return value is ignored.
1300 		 */
1301 		(void) (*pops->psm_clkinit)(hz);
1302 		*set_mode = TIMER_PERIODIC;
1303 		return (nsec_per_tick);
1304 	}
1305 }
1306 
1307 
1308 /*ARGSUSED*/
1309 static int
1310 mach_softlvl_to_vect(int ipl)
1311 {
1312 	setsoftint = av_set_softint_pending;
1313 	kdisetsoftint = kdi_av_set_softint_pending;
1314 
1315 	return (PSM_SV_SOFTWARE);
1316 }
1317 
1318 #ifdef DEBUG
1319 /*
1320  * This is here to allow us to simulate cpus that refuse to start.
1321  */
1322 cpuset_t cpufailset;
1323 #endif
1324 
1325 int
1326 mach_cpu_start(struct cpu *cp, void *ctx)
1327 {
1328 	struct psm_ops *pops = mach_set[0];
1329 	processorid_t id = cp->cpu_id;
1330 
1331 #ifdef DEBUG
1332 	if (CPU_IN_SET(cpufailset, id))
1333 		return (0);
1334 #endif
1335 	return ((*pops->psm_cpu_start)(id, ctx));
1336 }
1337 
1338 int
1339 mach_cpuid_start(processorid_t id, void *ctx)
1340 {
1341 	struct psm_ops *pops = mach_set[0];
1342 
1343 #ifdef DEBUG
1344 	if (CPU_IN_SET(cpufailset, id))
1345 		return (0);
1346 #endif
1347 	return ((*pops->psm_cpu_start)(id, ctx));
1348 }
1349 
1350 /*ARGSUSED*/
1351 static int
1352 mach_translate_irq(dev_info_t *dip, int irqno)
1353 {
1354 	return (irqno);	/* default to NO translation */
1355 }
1356 
1357 static void
1358 mach_notify_error(int level, char *errmsg)
1359 {
1360 	/*
1361 	 * SL_FATAL is pass in once panicstr is set, deliver it
1362 	 * as CE_PANIC.  Also, translate SL_ codes back to CE_
1363 	 * codes for the psmi handler
1364 	 */
1365 	if (level & SL_FATAL)
1366 		(*notify_error)(CE_PANIC, errmsg);
1367 	else if (level & SL_WARN)
1368 		(*notify_error)(CE_WARN, errmsg);
1369 	else if (level & SL_NOTE)
1370 		(*notify_error)(CE_NOTE, errmsg);
1371 	else if (level & SL_CONSOLE)
1372 		(*notify_error)(CE_CONT, errmsg);
1373 }
1374 
1375 /*
1376  * It provides the default basic intr_ops interface for the new DDI
1377  * interrupt framework if the PSM doesn't have one.
1378  *
1379  * Input:
1380  * dip     - pointer to the dev_info structure of the requested device
1381  * hdlp    - pointer to the internal interrupt handle structure for the
1382  *	     requested interrupt
1383  * intr_op - opcode for this call
1384  * result  - pointer to the integer that will hold the result to be
1385  *	     passed back if return value is PSM_SUCCESS
1386  *
1387  * Output:
1388  * return value is either PSM_SUCCESS or PSM_FAILURE
1389  */
1390 static int
1391 mach_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1392     psm_intr_op_t intr_op, int *result)
1393 {
1394 	struct intrspec *ispec;
1395 
1396 	switch (intr_op) {
1397 	case PSM_INTR_OP_CHECK_MSI:
1398 		*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1399 		    DDI_INTR_TYPE_MSIX);
1400 		break;
1401 	case PSM_INTR_OP_ALLOC_VECTORS:
1402 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1403 			*result = 1;
1404 		else
1405 			*result = 0;
1406 		break;
1407 	case PSM_INTR_OP_FREE_VECTORS:
1408 		break;
1409 	case PSM_INTR_OP_NAVAIL_VECTORS:
1410 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1411 			*result = 1;
1412 		else
1413 			*result = 0;
1414 		break;
1415 	case PSM_INTR_OP_XLATE_VECTOR:
1416 		ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1417 		*result = psm_translate_irq(dip, ispec->intrspec_vec);
1418 		break;
1419 	case PSM_INTR_OP_GET_CAP:
1420 		*result = 0;
1421 		break;
1422 	case PSM_INTR_OP_GET_PENDING:
1423 	case PSM_INTR_OP_CLEAR_MASK:
1424 	case PSM_INTR_OP_SET_MASK:
1425 	case PSM_INTR_OP_GET_SHARED:
1426 	case PSM_INTR_OP_SET_PRI:
1427 	case PSM_INTR_OP_SET_CAP:
1428 	case PSM_INTR_OP_SET_CPU:
1429 	case PSM_INTR_OP_GET_INTR:
1430 	default:
1431 		return (PSM_FAILURE);
1432 	}
1433 	return (PSM_SUCCESS);
1434 }
1435 /*
1436  * Return 1 if CMT load balancing policies should be
1437  * implemented across instances of the specified hardware
1438  * sharing relationship.
1439  */
1440 int
1441 pg_cmt_load_bal_hw(pghw_type_t hw)
1442 {
1443 	if (hw == PGHW_IPIPE ||
1444 	    hw == PGHW_FPU ||
1445 	    hw == PGHW_CHIP)
1446 		return (1);
1447 	else
1448 		return (0);
1449 }
1450 /*
1451  * Return 1 if thread affinity polices should be implemented
1452  * for instances of the specifed hardware sharing relationship.
1453  */
1454 int
1455 pg_cmt_affinity_hw(pghw_type_t hw)
1456 {
1457 	if (hw == PGHW_CACHE)
1458 		return (1);
1459 	else
1460 		return (0);
1461 }
1462