xref: /titanic_44/usr/src/uts/i86pc/os/mp_machdep.c (revision 4ba70ed0e487727de98a6297bc6d0a827001a390)
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 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #define	PSMI_1_5
29 #include <sys/smp_impldefs.h>
30 #include <sys/psm.h>
31 #include <sys/psm_modctl.h>
32 #include <sys/pit.h>
33 #include <sys/cmn_err.h>
34 #include <sys/strlog.h>
35 #include <sys/clock.h>
36 #include <sys/debug.h>
37 #include <sys/rtc.h>
38 #include <sys/x86_archext.h>
39 #include <sys/cpupart.h>
40 #include <sys/cpuvar.h>
41 #include <sys/cmt.h>
42 #include <sys/cpu.h>
43 #include <sys/disp.h>
44 #include <sys/archsystm.h>
45 #include <sys/machsystm.h>
46 #include <sys/sysmacros.h>
47 #include <sys/memlist.h>
48 #include <sys/param.h>
49 #include <sys/promif.h>
50 #if defined(__xpv)
51 #include <sys/hypervisor.h>
52 #endif
53 #include <sys/mach_intr.h>
54 #include <vm/hat_i86.h>
55 #include <sys/kdi_machimpl.h>
56 
57 #define	OFFSETOF(s, m)		(size_t)(&(((s *)0)->m))
58 
59 /*
60  *	Local function prototypes
61  */
62 static int mp_disable_intr(processorid_t cpun);
63 static void mp_enable_intr(processorid_t cpun);
64 static void mach_init();
65 static void mach_picinit();
66 static int machhztomhz(uint64_t cpu_freq_hz);
67 static uint64_t mach_getcpufreq(void);
68 static void mach_fixcpufreq(void);
69 static int mach_clkinit(int, int *);
70 static void mach_smpinit(void);
71 static int mach_softlvl_to_vect(int ipl);
72 static void mach_get_platform(int owner);
73 static void mach_construct_info();
74 static int mach_translate_irq(dev_info_t *dip, int irqno);
75 static int mach_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
76     psm_intr_op_t, int *);
77 static void mach_notify_error(int level, char *errmsg);
78 static hrtime_t dummy_hrtime(void);
79 static void dummy_scalehrtime(hrtime_t *);
80 static void cpu_idle(void);
81 static void cpu_wakeup(cpu_t *, int);
82 #ifndef __xpv
83 static void cpu_idle_mwait(void);
84 static void cpu_wakeup_mwait(cpu_t *, int);
85 #endif
86 /*
87  *	External reference functions
88  */
89 extern void return_instr();
90 extern uint64_t freq_tsc(uint32_t *);
91 #if defined(__i386)
92 extern uint64_t freq_notsc(uint32_t *);
93 #endif
94 extern void pc_gethrestime(timestruc_t *);
95 extern int cpuid_get_coreid(cpu_t *);
96 extern int cpuid_get_chipid(cpu_t *);
97 
98 /*
99  *	PSM functions initialization
100  */
101 void (*psm_shutdownf)(int, int)	= (void (*)(int, int))return_instr;
102 void (*psm_preshutdownf)(int, int) = (void (*)(int, int))return_instr;
103 void (*psm_notifyf)(int)	= (void (*)(int))return_instr;
104 void (*psm_set_idle_cpuf)(int)	= (void (*)(int))return_instr;
105 void (*psm_unset_idle_cpuf)(int) = (void (*)(int))return_instr;
106 void (*psminitf)()		= mach_init;
107 void (*picinitf)() 		= return_instr;
108 int (*clkinitf)(int, int *) 	= (int (*)(int, int *))return_instr;
109 int (*ap_mlsetup)() 		= (int (*)(void))return_instr;
110 void (*send_dirintf)() 		= return_instr;
111 void (*setspl)(int)		= (void (*)(int))return_instr;
112 int (*addspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
113 int (*delspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
114 void (*kdisetsoftint)(int, struct av_softinfo *)=
115 	(void (*)(int, struct av_softinfo *))return_instr;
116 void (*setsoftint)(int, struct av_softinfo *)=
117 	(void (*)(int, struct av_softinfo *))return_instr;
118 int (*slvltovect)(int)		= (int (*)(int))return_instr;
119 int (*setlvl)(int, int *)	= (int (*)(int, int *))return_instr;
120 void (*setlvlx)(int, int)	= (void (*)(int, int))return_instr;
121 int (*psm_disable_intr)(int)	= mp_disable_intr;
122 void (*psm_enable_intr)(int)	= mp_enable_intr;
123 hrtime_t (*gethrtimef)(void)	= dummy_hrtime;
124 hrtime_t (*gethrtimeunscaledf)(void)	= dummy_hrtime;
125 void (*scalehrtimef)(hrtime_t *)	= dummy_scalehrtime;
126 int (*psm_translate_irq)(dev_info_t *, int) = mach_translate_irq;
127 void (*gethrestimef)(timestruc_t *) = pc_gethrestime;
128 void (*psm_notify_error)(int, char *) = (void (*)(int, char *))NULL;
129 int (*psm_get_clockirq)(int) = NULL;
130 int (*psm_get_ipivect)(int, int) = NULL;
131 
132 int (*psm_clkinit)(int) = NULL;
133 void (*psm_timer_reprogram)(hrtime_t) = NULL;
134 void (*psm_timer_enable)(void) = NULL;
135 void (*psm_timer_disable)(void) = NULL;
136 void (*psm_post_cyclic_setup)(void *arg) = NULL;
137 int (*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *, psm_intr_op_t,
138     int *) = mach_intr_ops;
139 
140 void (*notify_error)(int, char *) = (void (*)(int, char *))return_instr;
141 void (*hrtime_tick)(void)	= return_instr;
142 
143 /*
144  * True if the generic TSC code is our source of hrtime, rather than whatever
145  * the PSM can provide.
146  */
147 #ifdef __xpv
148 int tsc_gethrtime_enable = 0;
149 #else
150 int tsc_gethrtime_enable = 1;
151 #endif
152 int tsc_gethrtime_initted = 0;
153 
154 /*
155  * True if the hrtime implementation is "hires"; namely, better than microdata.
156  */
157 int gethrtime_hires = 0;
158 
159 /*
160  * Local Static Data
161  */
162 static struct psm_ops mach_ops;
163 static struct psm_ops *mach_set[4] = {&mach_ops, NULL, NULL, NULL};
164 static ushort_t mach_ver[4] = {0, 0, 0, 0};
165 
166 /*
167  * If non-zero, idle cpus will become "halted" when there's
168  * no work to do.
169  */
170 int	idle_cpu_use_hlt = 1;
171 
172 #ifndef __xpv
173 /*
174  * If non-zero, idle cpus will use mwait if available to halt instead of hlt.
175  */
176 int	idle_cpu_prefer_mwait = 1;
177 #endif
178 
179 /*ARGSUSED*/
180 int
181 pg_plat_hw_shared(cpu_t *cp, pghw_type_t hw)
182 {
183 	switch (hw) {
184 	case PGHW_IPIPE:
185 		if (x86_feature & (X86_HTT)) {
186 			/*
187 			 * Hyper-threading is SMT
188 			 */
189 			return (1);
190 		} else {
191 			return (0);
192 		}
193 	case PGHW_CHIP:
194 		if (x86_feature & (X86_CMP|X86_HTT))
195 			return (1);
196 		else
197 			return (0);
198 	case PGHW_CACHE:
199 		if (cpuid_get_ncpu_sharing_last_cache(cp) > 1)
200 			return (1);
201 		else
202 			return (0);
203 	default:
204 		return (0);
205 	}
206 }
207 
208 /*
209  * Compare two CPUs and see if they have a pghw_type_t sharing relationship
210  * If pghw_type_t is an unsupported hardware type, then return -1
211  */
212 int
213 pg_plat_cpus_share(cpu_t *cpu_a, cpu_t *cpu_b, pghw_type_t hw)
214 {
215 	id_t pgp_a, pgp_b;
216 
217 	pgp_a = pg_plat_hw_instance_id(cpu_a, hw);
218 	pgp_b = pg_plat_hw_instance_id(cpu_b, hw);
219 
220 	if (pgp_a == -1 || pgp_b == -1)
221 		return (-1);
222 
223 	return (pgp_a == pgp_b);
224 }
225 
226 /*
227  * Return a physical instance identifier for known hardware sharing
228  * relationships
229  */
230 id_t
231 pg_plat_hw_instance_id(cpu_t *cpu, pghw_type_t hw)
232 {
233 	switch (hw) {
234 	case PGHW_IPIPE:
235 		return (cpuid_get_coreid(cpu));
236 	case PGHW_CACHE:
237 		return (cpuid_get_last_lvl_cacheid(cpu));
238 	case PGHW_CHIP:
239 		return (cpuid_get_chipid(cpu));
240 	default:
241 		return (-1);
242 	}
243 }
244 
245 int
246 pg_plat_hw_level(pghw_type_t hw)
247 {
248 	int i;
249 	static pghw_type_t hw_hier[] = {
250 		PGHW_IPIPE,
251 		PGHW_CACHE,
252 		PGHW_CHIP,
253 		PGHW_NUM_COMPONENTS
254 	};
255 
256 	for (i = 0; hw_hier[i] != PGHW_NUM_COMPONENTS; i++) {
257 		if (hw_hier[i] == hw)
258 			return (i);
259 	}
260 	return (-1);
261 }
262 
263 /*
264  * Return 1 if CMT load balancing policies should be
265  * implemented across instances of the specified hardware
266  * sharing relationship.
267  */
268 int
269 pg_plat_cmt_load_bal_hw(pghw_type_t hw)
270 {
271 	if (hw == PGHW_IPIPE ||
272 	    hw == PGHW_FPU ||
273 	    hw == PGHW_CHIP ||
274 	    hw == PGHW_CACHE)
275 		return (1);
276 	else
277 		return (0);
278 }
279 
280 
281 /*
282  * Return 1 if thread affinity polices should be implemented
283  * for instances of the specifed hardware sharing relationship.
284  */
285 int
286 pg_plat_cmt_affinity_hw(pghw_type_t hw)
287 {
288 	if (hw == PGHW_CACHE)
289 		return (1);
290 	else
291 		return (0);
292 }
293 
294 id_t
295 pg_plat_get_core_id(cpu_t *cpu)
296 {
297 	return ((id_t)cpuid_get_coreid(cpu));
298 }
299 
300 void
301 cmp_set_nosteal_interval(void)
302 {
303 	/* Set the nosteal interval (used by disp_getbest()) to 100us */
304 	nosteal_nsec = 100000UL;
305 }
306 
307 /*
308  * Routine to ensure initial callers to hrtime gets 0 as return
309  */
310 static hrtime_t
311 dummy_hrtime(void)
312 {
313 	return (0);
314 }
315 
316 /* ARGSUSED */
317 static void
318 dummy_scalehrtime(hrtime_t *ticks)
319 {}
320 
321 /*
322  * Idle the present CPU until awoken via an interrupt
323  */
324 static void
325 cpu_idle(void)
326 {
327 	cpu_t		*cpup = CPU;
328 	processorid_t	cpun = cpup->cpu_id;
329 	cpupart_t	*cp = cpup->cpu_part;
330 	int		hset_update = 1;
331 
332 	/*
333 	 * If this CPU is online, and there's multiple CPUs
334 	 * in the system, then we should notate our halting
335 	 * by adding ourselves to the partition's halted CPU
336 	 * bitmap. This allows other CPUs to find/awaken us when
337 	 * work becomes available.
338 	 */
339 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
340 		hset_update = 0;
341 
342 	/*
343 	 * Add ourselves to the partition's halted CPUs bitmask
344 	 * and set our HALTED flag, if necessary.
345 	 *
346 	 * When a thread becomes runnable, it is placed on the queue
347 	 * and then the halted cpuset is checked to determine who
348 	 * (if anyone) should be awoken. We therefore need to first
349 	 * add ourselves to the halted cpuset, and and then check if there
350 	 * is any work available.
351 	 *
352 	 * Note that memory barriers after updating the HALTED flag
353 	 * are not necessary since an atomic operation (updating the bitmap)
354 	 * immediately follows. On x86 the atomic operation acts as a
355 	 * memory barrier for the update of cpu_disp_flags.
356 	 */
357 	if (hset_update) {
358 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
359 		CPUSET_ATOMIC_ADD(cp->cp_mach->mc_haltset, cpun);
360 	}
361 
362 	/*
363 	 * Check to make sure there's really nothing to do.
364 	 * Work destined for this CPU may become available after
365 	 * this check. We'll be notified through the clearing of our
366 	 * bit in the halted CPU bitmask, and a poke.
367 	 */
368 	if (disp_anywork()) {
369 		if (hset_update) {
370 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
371 			CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
372 		}
373 		return;
374 	}
375 
376 	/*
377 	 * We're on our way to being halted.
378 	 *
379 	 * Disable interrupts now, so that we'll awaken immediately
380 	 * after halting if someone tries to poke us between now and
381 	 * the time we actually halt.
382 	 *
383 	 * We check for the presence of our bit after disabling interrupts.
384 	 * If it's cleared, we'll return. If the bit is cleared after
385 	 * we check then the poke will pop us out of the halted state.
386 	 *
387 	 * This means that the ordering of the poke and the clearing
388 	 * of the bit by cpu_wakeup is important.
389 	 * cpu_wakeup() must clear, then poke.
390 	 * cpu_idle() must disable interrupts, then check for the bit.
391 	 */
392 	cli();
393 
394 	if (hset_update && !CPU_IN_SET(cp->cp_mach->mc_haltset, cpun)) {
395 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
396 		sti();
397 		return;
398 	}
399 
400 	/*
401 	 * The check for anything locally runnable is here for performance
402 	 * and isn't needed for correctness. disp_nrunnable ought to be
403 	 * in our cache still, so it's inexpensive to check, and if there
404 	 * is anything runnable we won't have to wait for the poke.
405 	 */
406 	if (cpup->cpu_disp->disp_nrunnable != 0) {
407 		if (hset_update) {
408 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
409 			CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
410 		}
411 		sti();
412 		return;
413 	}
414 
415 	mach_cpu_idle();
416 
417 	/*
418 	 * We're no longer halted
419 	 */
420 	if (hset_update) {
421 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
422 		CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
423 	}
424 }
425 
426 
427 /*
428  * If "cpu" is halted, then wake it up clearing its halted bit in advance.
429  * Otherwise, see if other CPUs in the cpu partition are halted and need to
430  * be woken up so that they can steal the thread we placed on this CPU.
431  * This function is only used on MP systems.
432  */
433 static void
434 cpu_wakeup(cpu_t *cpu, int bound)
435 {
436 	uint_t		cpu_found;
437 	int		result;
438 	cpupart_t	*cp;
439 
440 	cp = cpu->cpu_part;
441 	if (CPU_IN_SET(cp->cp_mach->mc_haltset, cpu->cpu_id)) {
442 		/*
443 		 * Clear the halted bit for that CPU since it will be
444 		 * poked in a moment.
445 		 */
446 		CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpu->cpu_id);
447 		/*
448 		 * We may find the current CPU present in the halted cpuset
449 		 * if we're in the context of an interrupt that occurred
450 		 * before we had a chance to clear our bit in cpu_idle().
451 		 * Poking ourself is obviously unnecessary, since if
452 		 * we're here, we're not halted.
453 		 */
454 		if (cpu != CPU)
455 			poke_cpu(cpu->cpu_id);
456 		return;
457 	} else {
458 		/*
459 		 * This cpu isn't halted, but it's idle or undergoing a
460 		 * context switch. No need to awaken anyone else.
461 		 */
462 		if (cpu->cpu_thread == cpu->cpu_idle_thread ||
463 		    cpu->cpu_disp_flags & CPU_DISP_DONTSTEAL)
464 			return;
465 	}
466 
467 	/*
468 	 * No need to wake up other CPUs if the thread we just enqueued
469 	 * is bound.
470 	 */
471 	if (bound)
472 		return;
473 
474 
475 	/*
476 	 * See if there's any other halted CPUs. If there are, then
477 	 * select one, and awaken it.
478 	 * It's possible that after we find a CPU, somebody else
479 	 * will awaken it before we get the chance.
480 	 * In that case, look again.
481 	 */
482 	do {
483 		CPUSET_FIND(cp->cp_mach->mc_haltset, cpu_found);
484 		if (cpu_found == CPUSET_NOTINSET)
485 			return;
486 
487 		ASSERT(cpu_found >= 0 && cpu_found < NCPU);
488 		CPUSET_ATOMIC_XDEL(cp->cp_mach->mc_haltset, cpu_found, result);
489 	} while (result < 0);
490 
491 	if (cpu_found != CPU->cpu_id)
492 		poke_cpu(cpu_found);
493 }
494 
495 #ifndef __xpv
496 /*
497  * Idle the present CPU until awoken via touching its monitored line
498  */
499 static void
500 cpu_idle_mwait(void)
501 {
502 	volatile uint32_t	*mcpu_mwait = CPU->cpu_m.mcpu_mwait;
503 	cpu_t			*cpup = CPU;
504 	processorid_t		cpun = cpup->cpu_id;
505 	cpupart_t		*cp = cpup->cpu_part;
506 	int			hset_update = 1;
507 
508 	/*
509 	 * Set our mcpu_mwait here, so we can tell if anyone trys to
510 	 * wake us between now and when we call mwait.  No other cpu will
511 	 * attempt to set our mcpu_mwait until we add ourself to the haltset.
512 	 */
513 	*mcpu_mwait = MWAIT_HALTED;
514 
515 	/*
516 	 * If this CPU is online, and there's multiple CPUs
517 	 * in the system, then we should notate our halting
518 	 * by adding ourselves to the partition's halted CPU
519 	 * bitmap. This allows other CPUs to find/awaken us when
520 	 * work becomes available.
521 	 */
522 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
523 		hset_update = 0;
524 
525 	/*
526 	 * Add ourselves to the partition's halted CPUs bitmask
527 	 * and set our HALTED flag, if necessary.
528 	 *
529 	 * When a thread becomes runnable, it is placed on the queue
530 	 * and then the halted cpuset is checked to determine who
531 	 * (if anyone) should be awoken. We therefore need to first
532 	 * add ourselves to the halted cpuset, and and then check if there
533 	 * is any work available.
534 	 *
535 	 * Note that memory barriers after updating the HALTED flag
536 	 * are not necessary since an atomic operation (updating the bitmap)
537 	 * immediately follows. On x86 the atomic operation acts as a
538 	 * memory barrier for the update of cpu_disp_flags.
539 	 */
540 	if (hset_update) {
541 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
542 		CPUSET_ATOMIC_ADD(cp->cp_mach->mc_haltset, cpun);
543 	}
544 
545 	/*
546 	 * Check to make sure there's really nothing to do.
547 	 * Work destined for this CPU may become available after
548 	 * this check. We'll be notified through the clearing of our
549 	 * bit in the halted CPU bitmask, and a write to our mcpu_mwait.
550 	 *
551 	 * disp_anywork() checks disp_nrunnable, so we do not have to later.
552 	 */
553 	if (disp_anywork()) {
554 		if (hset_update) {
555 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
556 			CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
557 		}
558 		return;
559 	}
560 
561 	/*
562 	 * We're on our way to being halted.
563 	 * To avoid a lost wakeup, arm the monitor before checking if another
564 	 * cpu wrote to mcpu_mwait to wake us up.
565 	 */
566 	i86_monitor(mcpu_mwait, 0, 0);
567 	if (*mcpu_mwait == MWAIT_HALTED) {
568 		tlb_going_idle();
569 		i86_mwait(0, 0);
570 		tlb_service();
571 	}
572 
573 	/*
574 	 * We're no longer halted
575 	 */
576 	if (hset_update) {
577 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
578 		CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
579 	}
580 }
581 
582 /*
583  * If "cpu" is halted in mwait, then wake it up clearing its halted bit in
584  * advance.  Otherwise, see if other CPUs in the cpu partition are halted and
585  * need to be woken up so that they can steal the thread we placed on this CPU.
586  * This function is only used on MP systems.
587  */
588 static void
589 cpu_wakeup_mwait(cpu_t *cp, int bound)
590 {
591 	cpupart_t	*cpu_part;
592 	uint_t		cpu_found;
593 	int		result;
594 
595 	cpu_part = cp->cpu_part;
596 
597 	/*
598 	 * Clear the halted bit for that CPU since it will be woken up
599 	 * in a moment.
600 	 */
601 	if (CPU_IN_SET(cpu_part->cp_mach->mc_haltset, cp->cpu_id)) {
602 		/*
603 		 * Clear the halted bit for that CPU since it will be
604 		 * poked in a moment.
605 		 */
606 		CPUSET_ATOMIC_DEL(cpu_part->cp_mach->mc_haltset, cp->cpu_id);
607 		/*
608 		 * We may find the current CPU present in the halted cpuset
609 		 * if we're in the context of an interrupt that occurred
610 		 * before we had a chance to clear our bit in cpu_idle().
611 		 * Waking ourself is obviously unnecessary, since if
612 		 * we're here, we're not halted.
613 		 *
614 		 * monitor/mwait wakeup via writing to our cache line is
615 		 * harmless and less expensive than always checking if we
616 		 * are waking ourself which is an uncommon case.
617 		 */
618 		MWAIT_WAKEUP(cp);	/* write to monitored line */
619 		return;
620 	} else {
621 		/*
622 		 * This cpu isn't halted, but it's idle or undergoing a
623 		 * context switch. No need to awaken anyone else.
624 		 */
625 		if (cp->cpu_thread == cp->cpu_idle_thread ||
626 		    cp->cpu_disp_flags & CPU_DISP_DONTSTEAL)
627 			return;
628 	}
629 
630 	/*
631 	 * No need to wake up other CPUs if the thread we just enqueued
632 	 * is bound.
633 	 */
634 	if (bound)
635 		return;
636 
637 
638 	/*
639 	 * See if there's any other halted CPUs. If there are, then
640 	 * select one, and awaken it.
641 	 * It's possible that after we find a CPU, somebody else
642 	 * will awaken it before we get the chance.
643 	 * In that case, look again.
644 	 */
645 	do {
646 		CPUSET_FIND(cpu_part->cp_mach->mc_haltset, cpu_found);
647 		if (cpu_found == CPUSET_NOTINSET)
648 			return;
649 
650 		ASSERT(cpu_found >= 0 && cpu_found < NCPU);
651 		CPUSET_ATOMIC_XDEL(cpu_part->cp_mach->mc_haltset, cpu_found,
652 		    result);
653 	} while (result < 0);
654 
655 	/*
656 	 * Do not check if cpu_found is ourself as monitor/mwait wakeup is
657 	 * cheap.
658 	 */
659 	MWAIT_WAKEUP(cpu[cpu_found]);	/* write to monitored line */
660 }
661 #endif
662 
663 void (*cpu_pause_handler)(volatile char *) = NULL;
664 
665 static int
666 mp_disable_intr(int cpun)
667 {
668 	/*
669 	 * switch to the offline cpu
670 	 */
671 	affinity_set(cpun);
672 	/*
673 	 * raise ipl to just below cross call
674 	 */
675 	splx(XC_MED_PIL-1);
676 	/*
677 	 *	set base spl to prevent the next swtch to idle from
678 	 *	lowering back to ipl 0
679 	 */
680 	CPU->cpu_intr_actv |= (1 << (XC_MED_PIL-1));
681 	set_base_spl();
682 	affinity_clear();
683 	return (DDI_SUCCESS);
684 }
685 
686 static void
687 mp_enable_intr(int cpun)
688 {
689 	/*
690 	 * switch to the online cpu
691 	 */
692 	affinity_set(cpun);
693 	/*
694 	 * clear the interrupt active mask
695 	 */
696 	CPU->cpu_intr_actv &= ~(1 << (XC_MED_PIL-1));
697 	set_base_spl();
698 	(void) spl0();
699 	affinity_clear();
700 }
701 
702 static void
703 mach_get_platform(int owner)
704 {
705 	void		**srv_opsp;
706 	void		**clt_opsp;
707 	int		i;
708 	int		total_ops;
709 
710 	/* fix up psm ops */
711 	srv_opsp = (void **)mach_set[0];
712 	clt_opsp = (void **)mach_set[owner];
713 	if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01)
714 		total_ops = sizeof (struct psm_ops_ver01) /
715 		    sizeof (void (*)(void));
716 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_1)
717 		/* no psm_notify_func */
718 		total_ops = OFFSETOF(struct psm_ops, psm_notify_func) /
719 		    sizeof (void (*)(void));
720 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_2)
721 		/* no psm_timer funcs */
722 		total_ops = OFFSETOF(struct psm_ops, psm_timer_reprogram) /
723 		    sizeof (void (*)(void));
724 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_3)
725 		/* no psm_preshutdown function */
726 		total_ops = OFFSETOF(struct psm_ops, psm_preshutdown) /
727 		    sizeof (void (*)(void));
728 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_4)
729 		/* no psm_preshutdown function */
730 		total_ops = OFFSETOF(struct psm_ops, psm_intr_ops) /
731 		    sizeof (void (*)(void));
732 	else
733 		total_ops = sizeof (struct psm_ops) / sizeof (void (*)(void));
734 
735 	/*
736 	 * Save the version of the PSM module, in case we need to
737 	 * bahave differently based on version.
738 	 */
739 	mach_ver[0] = mach_ver[owner];
740 
741 	for (i = 0; i < total_ops; i++)
742 		if (clt_opsp[i] != NULL)
743 			srv_opsp[i] = clt_opsp[i];
744 }
745 
746 static void
747 mach_construct_info()
748 {
749 	struct psm_sw *swp;
750 	int	mach_cnt[PSM_OWN_OVERRIDE+1] = {0};
751 	int	conflict_owner = 0;
752 
753 	if (psmsw->psw_forw == psmsw)
754 		panic("No valid PSM modules found");
755 	mutex_enter(&psmsw_lock);
756 	for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
757 		if (!(swp->psw_flag & PSM_MOD_IDENTIFY))
758 			continue;
759 		mach_set[swp->psw_infop->p_owner] = swp->psw_infop->p_ops;
760 		mach_ver[swp->psw_infop->p_owner] = swp->psw_infop->p_version;
761 		mach_cnt[swp->psw_infop->p_owner]++;
762 	}
763 	mutex_exit(&psmsw_lock);
764 
765 	mach_get_platform(PSM_OWN_SYS_DEFAULT);
766 
767 	/* check to see are there any conflicts */
768 	if (mach_cnt[PSM_OWN_EXCLUSIVE] > 1)
769 		conflict_owner = PSM_OWN_EXCLUSIVE;
770 	if (mach_cnt[PSM_OWN_OVERRIDE] > 1)
771 		conflict_owner = PSM_OWN_OVERRIDE;
772 	if (conflict_owner) {
773 		/* remove all psm modules except uppc */
774 		cmn_err(CE_WARN,
775 		    "Conflicts detected on the following PSM modules:");
776 		mutex_enter(&psmsw_lock);
777 		for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
778 			if (swp->psw_infop->p_owner == conflict_owner)
779 				cmn_err(CE_WARN, "%s ",
780 				    swp->psw_infop->p_mach_idstring);
781 		}
782 		mutex_exit(&psmsw_lock);
783 		cmn_err(CE_WARN,
784 		    "Setting the system back to SINGLE processor mode!");
785 		cmn_err(CE_WARN,
786 		    "Please edit /etc/mach to remove the invalid PSM module.");
787 		return;
788 	}
789 
790 	if (mach_set[PSM_OWN_EXCLUSIVE])
791 		mach_get_platform(PSM_OWN_EXCLUSIVE);
792 
793 	if (mach_set[PSM_OWN_OVERRIDE])
794 		mach_get_platform(PSM_OWN_OVERRIDE);
795 }
796 
797 static void
798 mach_init()
799 {
800 	struct psm_ops  *pops;
801 
802 	mach_construct_info();
803 
804 	pops = mach_set[0];
805 
806 	/* register the interrupt and clock initialization rotuines */
807 	picinitf = mach_picinit;
808 	clkinitf = mach_clkinit;
809 	psm_get_clockirq = pops->psm_get_clockirq;
810 
811 	/* register the interrupt setup code */
812 	slvltovect = mach_softlvl_to_vect;
813 	addspl	= pops->psm_addspl;
814 	delspl	= pops->psm_delspl;
815 
816 	if (pops->psm_translate_irq)
817 		psm_translate_irq = pops->psm_translate_irq;
818 	if (pops->psm_intr_ops)
819 		psm_intr_ops = pops->psm_intr_ops;
820 
821 #if defined(PSMI_1_2) || defined(PSMI_1_3) || defined(PSMI_1_4)
822 	/*
823 	 * Time-of-day functionality now handled in TOD modules.
824 	 * (Warn about PSM modules that think that we're going to use
825 	 * their ops vectors.)
826 	 */
827 	if (pops->psm_tod_get)
828 		cmn_err(CE_WARN, "obsolete psm_tod_get op %p",
829 		    (void *)pops->psm_tod_get);
830 
831 	if (pops->psm_tod_set)
832 		cmn_err(CE_WARN, "obsolete psm_tod_set op %p",
833 		    (void *)pops->psm_tod_set);
834 #endif
835 
836 	if (pops->psm_notify_error) {
837 		psm_notify_error = mach_notify_error;
838 		notify_error = pops->psm_notify_error;
839 	}
840 
841 	(*pops->psm_softinit)();
842 
843 	/*
844 	 * Initialize the dispatcher's function hooks
845 	 * to enable CPU halting when idle.
846 	 * Do not use monitor/mwait if idle_cpu_use_hlt is not set(spin idle)
847 	 * or idle_cpu_prefer_mwait is not set.
848 	 * Allocate monitor/mwait buffer for cpu0.
849 	 */
850 	if (idle_cpu_use_hlt) {
851 		idle_cpu = cpu_idle;
852 #ifndef __xpv
853 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait) {
854 			CPU->cpu_m.mcpu_mwait = cpuid_mwait_alloc(CPU);
855 			/*
856 			 * Protect ourself from insane mwait size.
857 			 */
858 			if (CPU->cpu_m.mcpu_mwait == NULL) {
859 #ifdef DEBUG
860 				cmn_err(CE_NOTE, "Using hlt idle.  Cannot "
861 				    "handle cpu 0 mwait size.");
862 #endif
863 				idle_cpu_prefer_mwait = 0;
864 				idle_cpu = cpu_idle;
865 			} else {
866 				idle_cpu = cpu_idle_mwait;
867 			}
868 		} else {
869 			idle_cpu = cpu_idle;
870 		}
871 #endif
872 	}
873 
874 	mach_smpinit();
875 }
876 
877 static void
878 mach_smpinit(void)
879 {
880 	struct psm_ops  *pops;
881 	processorid_t cpu_id;
882 	int cnt;
883 	cpuset_t cpumask;
884 
885 	pops = mach_set[0];
886 
887 	cpu_id = -1;
888 	cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
889 	for (cnt = 0, CPUSET_ZERO(cpumask); cpu_id != -1; cnt++) {
890 		CPUSET_ADD(cpumask, cpu_id);
891 		cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
892 	}
893 
894 	mp_cpus = cpumask;
895 
896 	/* MP related routines */
897 	ap_mlsetup = pops->psm_post_cpu_start;
898 	send_dirintf = pops->psm_send_ipi;
899 
900 	/* optional MP related routines */
901 	if (pops->psm_shutdown)
902 		psm_shutdownf = pops->psm_shutdown;
903 	if (pops->psm_preshutdown)
904 		psm_preshutdownf = pops->psm_preshutdown;
905 	if (pops->psm_notify_func)
906 		psm_notifyf = pops->psm_notify_func;
907 	if (pops->psm_set_idlecpu)
908 		psm_set_idle_cpuf = pops->psm_set_idlecpu;
909 	if (pops->psm_unset_idlecpu)
910 		psm_unset_idle_cpuf = pops->psm_unset_idlecpu;
911 
912 	psm_clkinit = pops->psm_clkinit;
913 
914 	if (pops->psm_timer_reprogram)
915 		psm_timer_reprogram = pops->psm_timer_reprogram;
916 
917 	if (pops->psm_timer_enable)
918 		psm_timer_enable = pops->psm_timer_enable;
919 
920 	if (pops->psm_timer_disable)
921 		psm_timer_disable = pops->psm_timer_disable;
922 
923 	if (pops->psm_post_cyclic_setup)
924 		psm_post_cyclic_setup = pops->psm_post_cyclic_setup;
925 
926 	/* check for multiple cpu's */
927 	if (cnt < 2)
928 		return;
929 
930 	/* check for MP platforms */
931 	if (pops->psm_cpu_start == NULL)
932 		return;
933 
934 	/*
935 	 * Set the dispatcher hook to enable cpu "wake up"
936 	 * when a thread becomes runnable.
937 	 */
938 	if (idle_cpu_use_hlt) {
939 		disp_enq_thread = cpu_wakeup;
940 #ifndef __xpv
941 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait)
942 			disp_enq_thread = cpu_wakeup_mwait;
943 #endif
944 	}
945 
946 	if (pops->psm_disable_intr)
947 		psm_disable_intr = pops->psm_disable_intr;
948 	if (pops->psm_enable_intr)
949 		psm_enable_intr  = pops->psm_enable_intr;
950 
951 	psm_get_ipivect = pops->psm_get_ipivect;
952 
953 	(void) add_avintr((void *)NULL, XC_HI_PIL, xc_serv, "xc_hi_intr",
954 	    (*pops->psm_get_ipivect)(XC_HI_PIL, PSM_INTR_IPI_HI),
955 	    (caddr_t)X_CALL_HIPRI, NULL, NULL, NULL);
956 	(void) add_avintr((void *)NULL, XC_MED_PIL, xc_serv, "xc_med_intr",
957 	    (*pops->psm_get_ipivect)(XC_MED_PIL, PSM_INTR_IPI_LO),
958 	    (caddr_t)X_CALL_MEDPRI, NULL, NULL, NULL);
959 
960 	(void) (*pops->psm_get_ipivect)(XC_CPUPOKE_PIL, PSM_INTR_POKE);
961 }
962 
963 static void
964 mach_picinit()
965 {
966 	struct psm_ops  *pops;
967 
968 	pops = mach_set[0];
969 
970 	/* register the interrupt handlers */
971 	setlvl = pops->psm_intr_enter;
972 	setlvlx = pops->psm_intr_exit;
973 
974 	/* initialize the interrupt hardware */
975 	(*pops->psm_picinit)();
976 
977 	/* set interrupt mask for current ipl */
978 	setspl = pops->psm_setspl;
979 	cli();
980 	setspl(CPU->cpu_pri);
981 }
982 
983 uint_t	cpu_freq;	/* MHz */
984 uint64_t cpu_freq_hz;	/* measured (in hertz) */
985 
986 #define	MEGA_HZ		1000000
987 
988 #ifdef __xpv
989 
990 int xpv_cpufreq_workaround = 1;
991 int xpv_cpufreq_verbose = 0;
992 
993 #else	/* __xpv */
994 
995 static uint64_t
996 mach_calchz(uint32_t pit_counter, uint64_t *processor_clks)
997 {
998 	uint64_t cpu_hz;
999 
1000 	if ((pit_counter == 0) || (*processor_clks == 0) ||
1001 	    (*processor_clks > (((uint64_t)-1) / PIT_HZ)))
1002 		return (0);
1003 
1004 	cpu_hz = ((uint64_t)PIT_HZ * *processor_clks) / pit_counter;
1005 
1006 	return (cpu_hz);
1007 }
1008 
1009 #endif	/* __xpv */
1010 
1011 static uint64_t
1012 mach_getcpufreq(void)
1013 {
1014 #if defined(__xpv)
1015 	vcpu_time_info_t *vti = &CPU->cpu_m.mcpu_vcpu_info->time;
1016 	uint64_t cpu_hz;
1017 
1018 	/*
1019 	 * During dom0 bringup, it was noted that on at least one older
1020 	 * Intel HT machine, the hypervisor initially gives a tsc_to_system_mul
1021 	 * value that is quite wrong (the 3.06GHz clock was reported
1022 	 * as 4.77GHz)
1023 	 *
1024 	 * The curious thing is, that if you stop the kernel at entry,
1025 	 * breakpoint here and inspect the value with kmdb, the value
1026 	 * is correct - but if you don't stop and simply enable the
1027 	 * printf statement (below), you can see the bad value printed
1028 	 * here.  Almost as if something kmdb did caused the hypervisor to
1029 	 * figure it out correctly.  And, note that the hypervisor
1030 	 * eventually -does- figure it out correctly ... if you look at
1031 	 * the field later in the life of dom0, it is correct.
1032 	 *
1033 	 * For now, on dom0, we employ a slightly cheesy workaround of
1034 	 * using the DOM0_PHYSINFO hypercall.
1035 	 */
1036 	if (DOMAIN_IS_INITDOMAIN(xen_info) && xpv_cpufreq_workaround) {
1037 		xen_sysctl_t op0, *op = &op0;
1038 
1039 		op->cmd = XEN_SYSCTL_physinfo;
1040 		op->interface_version = XEN_SYSCTL_INTERFACE_VERSION;
1041 		if (HYPERVISOR_sysctl(op) != 0)
1042 			panic("physinfo op refused");
1043 
1044 		cpu_hz = 1000 * (uint64_t)op->u.physinfo.cpu_khz;
1045 	} else {
1046 		cpu_hz = (UINT64_C(1000000000) << 32) / vti->tsc_to_system_mul;
1047 
1048 		if (vti->tsc_shift < 0)
1049 			cpu_hz <<= -vti->tsc_shift;
1050 		else
1051 			cpu_hz >>= vti->tsc_shift;
1052 	}
1053 
1054 	if (xpv_cpufreq_verbose)
1055 		printf("mach_getcpufreq: system_mul 0x%x, shift %d, "
1056 		    "cpu_hz %" PRId64 "Hz\n",
1057 		    vti->tsc_to_system_mul, vti->tsc_shift, cpu_hz);
1058 
1059 	return (cpu_hz);
1060 #else	/* __xpv */
1061 	uint32_t pit_counter;
1062 	uint64_t processor_clks;
1063 
1064 	if (x86_feature & X86_TSC) {
1065 		/*
1066 		 * We have a TSC. freq_tsc() knows how to measure the number
1067 		 * of clock cycles sampled against the PIT.
1068 		 */
1069 		ulong_t flags = clear_int_flag();
1070 		processor_clks = freq_tsc(&pit_counter);
1071 		restore_int_flag(flags);
1072 		return (mach_calchz(pit_counter, &processor_clks));
1073 	} else if (x86_vendor == X86_VENDOR_Cyrix || x86_type == X86_TYPE_P5) {
1074 #if defined(__amd64)
1075 		panic("mach_getcpufreq: no TSC!");
1076 #elif defined(__i386)
1077 		/*
1078 		 * We are a Cyrix based on a 6x86 core or an Intel Pentium
1079 		 * for which freq_notsc() knows how to measure the number of
1080 		 * elapsed clock cycles sampled against the PIT
1081 		 */
1082 		ulong_t flags = clear_int_flag();
1083 		processor_clks = freq_notsc(&pit_counter);
1084 		restore_int_flag(flags);
1085 		return (mach_calchz(pit_counter, &processor_clks));
1086 #endif	/* __i386 */
1087 	}
1088 
1089 	/* We do not know how to calculate cpu frequency for this cpu. */
1090 	return (0);
1091 #endif	/* __xpv */
1092 }
1093 
1094 /*
1095  * If the clock speed of a cpu is found to be reported incorrectly, do not add
1096  * to this array, instead improve the accuracy of the algorithm that determines
1097  * the clock speed of the processor or extend the implementation to support the
1098  * vendor as appropriate. This is here only to support adjusting the speed on
1099  * older slower processors that mach_fixcpufreq() would not be able to account
1100  * for otherwise.
1101  */
1102 static int x86_cpu_freq[] = { 60, 75, 80, 90, 120, 160, 166, 175, 180, 233 };
1103 
1104 /*
1105  * On fast processors the clock frequency that is measured may be off by
1106  * a few MHz from the value printed on the part. This is a combination of
1107  * the factors that for such fast parts being off by this much is within
1108  * the tolerances for manufacture and because of the difficulties in the
1109  * measurement that can lead to small error. This function uses some
1110  * heuristics in order to tweak the value that was measured to match what
1111  * is most likely printed on the part.
1112  *
1113  * Some examples:
1114  * 	AMD Athlon 1000 mhz measured as 998 mhz
1115  * 	Intel Pentium III Xeon 733 mhz measured as 731 mhz
1116  * 	Intel Pentium IV 1500 mhz measured as 1495mhz
1117  *
1118  * If in the future this function is no longer sufficient to correct
1119  * for the error in the measurement, then the algorithm used to perform
1120  * the measurement will have to be improved in order to increase accuracy
1121  * rather than adding horrible and questionable kludges here.
1122  *
1123  * This is called after the cyclics subsystem because of the potential
1124  * that the heuristics within may give a worse estimate of the clock
1125  * frequency than the value that was measured.
1126  */
1127 static void
1128 mach_fixcpufreq(void)
1129 {
1130 	uint32_t freq, mul, near66, delta66, near50, delta50, fixed, delta, i;
1131 
1132 	freq = (uint32_t)cpu_freq;
1133 
1134 	/*
1135 	 * Find the nearest integer multiple of 200/3 (about 66) MHz to the
1136 	 * measured speed taking into account that the 667 MHz parts were
1137 	 * the first to round-up.
1138 	 */
1139 	mul = (uint32_t)((3 * (uint64_t)freq + 100) / 200);
1140 	near66 = (uint32_t)((200 * (uint64_t)mul + ((mul >= 10) ? 1 : 0)) / 3);
1141 	delta66 = (near66 > freq) ? (near66 - freq) : (freq - near66);
1142 
1143 	/* Find the nearest integer multiple of 50 MHz to the measured speed */
1144 	mul = (freq + 25) / 50;
1145 	near50 = mul * 50;
1146 	delta50 = (near50 > freq) ? (near50 - freq) : (freq - near50);
1147 
1148 	/* Find the closer of the two */
1149 	if (delta66 < delta50) {
1150 		fixed = near66;
1151 		delta = delta66;
1152 	} else {
1153 		fixed = near50;
1154 		delta = delta50;
1155 	}
1156 
1157 	if (fixed > INT_MAX)
1158 		return;
1159 
1160 	/*
1161 	 * Some older parts have a core clock frequency that is not an
1162 	 * integral multiple of 50 or 66 MHz. Check if one of the old
1163 	 * clock frequencies is closer to the measured value than any
1164 	 * of the integral multiples of 50 an 66, and if so set fixed
1165 	 * and delta appropriately to represent the closest value.
1166 	 */
1167 	i = sizeof (x86_cpu_freq) / sizeof (int);
1168 	while (i > 0) {
1169 		i--;
1170 
1171 		if (x86_cpu_freq[i] <= freq) {
1172 			mul = freq - x86_cpu_freq[i];
1173 
1174 			if (mul < delta) {
1175 				fixed = x86_cpu_freq[i];
1176 				delta = mul;
1177 			}
1178 
1179 			break;
1180 		}
1181 
1182 		mul = x86_cpu_freq[i] - freq;
1183 
1184 		if (mul < delta) {
1185 			fixed = x86_cpu_freq[i];
1186 			delta = mul;
1187 		}
1188 	}
1189 
1190 	/*
1191 	 * Set a reasonable maximum for how much to correct the measured
1192 	 * result by. This check is here to prevent the adjustment made
1193 	 * by this function from being more harm than good. It is entirely
1194 	 * possible that in the future parts will be made that are not
1195 	 * integral multiples of 66 or 50 in clock frequency or that
1196 	 * someone may overclock a part to some odd frequency. If the
1197 	 * measured value is farther from the corrected value than
1198 	 * allowed, then assume the corrected value is in error and use
1199 	 * the measured value.
1200 	 */
1201 	if (6 < delta)
1202 		return;
1203 
1204 	cpu_freq = (int)fixed;
1205 }
1206 
1207 
1208 static int
1209 machhztomhz(uint64_t cpu_freq_hz)
1210 {
1211 	uint64_t cpu_mhz;
1212 
1213 	/* Round to nearest MHZ */
1214 	cpu_mhz = (cpu_freq_hz + (MEGA_HZ / 2)) / MEGA_HZ;
1215 
1216 	if (cpu_mhz > INT_MAX)
1217 		return (0);
1218 
1219 	return ((int)cpu_mhz);
1220 
1221 }
1222 
1223 
1224 static int
1225 mach_clkinit(int preferred_mode, int *set_mode)
1226 {
1227 	struct psm_ops  *pops;
1228 	int resolution;
1229 
1230 	pops = mach_set[0];
1231 
1232 	cpu_freq_hz = mach_getcpufreq();
1233 
1234 	cpu_freq = machhztomhz(cpu_freq_hz);
1235 
1236 	if (!(x86_feature & X86_TSC) || (cpu_freq == 0))
1237 		tsc_gethrtime_enable = 0;
1238 
1239 #ifndef __xpv
1240 	if (tsc_gethrtime_enable) {
1241 		tsc_hrtimeinit(cpu_freq_hz);
1242 	} else
1243 #endif
1244 	{
1245 		if (pops->psm_hrtimeinit)
1246 			(*pops->psm_hrtimeinit)();
1247 		gethrtimef = pops->psm_gethrtime;
1248 		gethrtimeunscaledf = gethrtimef;
1249 		/* scalehrtimef will remain dummy */
1250 	}
1251 
1252 	mach_fixcpufreq();
1253 
1254 	if (mach_ver[0] >= PSM_INFO_VER01_3) {
1255 		if (preferred_mode == TIMER_ONESHOT) {
1256 
1257 			resolution = (*pops->psm_clkinit)(0);
1258 			if (resolution != 0)  {
1259 				*set_mode = TIMER_ONESHOT;
1260 				return (resolution);
1261 			}
1262 		}
1263 
1264 		/*
1265 		 * either periodic mode was requested or could not set to
1266 		 * one-shot mode
1267 		 */
1268 		resolution = (*pops->psm_clkinit)(hz);
1269 		/*
1270 		 * psm should be able to do periodic, so we do not check
1271 		 * for return value of psm_clkinit here.
1272 		 */
1273 		*set_mode = TIMER_PERIODIC;
1274 		return (resolution);
1275 	} else {
1276 		/*
1277 		 * PSMI interface prior to PSMI_3 does not define a return
1278 		 * value for psm_clkinit, so the return value is ignored.
1279 		 */
1280 		(void) (*pops->psm_clkinit)(hz);
1281 		*set_mode = TIMER_PERIODIC;
1282 		return (nsec_per_tick);
1283 	}
1284 }
1285 
1286 
1287 /*ARGSUSED*/
1288 static int
1289 mach_softlvl_to_vect(int ipl)
1290 {
1291 	setsoftint = av_set_softint_pending;
1292 	kdisetsoftint = kdi_av_set_softint_pending;
1293 
1294 	return (PSM_SV_SOFTWARE);
1295 }
1296 
1297 #ifdef DEBUG
1298 /*
1299  * This is here to allow us to simulate cpus that refuse to start.
1300  */
1301 cpuset_t cpufailset;
1302 #endif
1303 
1304 int
1305 mach_cpu_start(struct cpu *cp, void *ctx)
1306 {
1307 	struct psm_ops *pops = mach_set[0];
1308 	processorid_t id = cp->cpu_id;
1309 
1310 #ifdef DEBUG
1311 	if (CPU_IN_SET(cpufailset, id))
1312 		return (0);
1313 #endif
1314 	return ((*pops->psm_cpu_start)(id, ctx));
1315 }
1316 
1317 /*ARGSUSED*/
1318 static int
1319 mach_translate_irq(dev_info_t *dip, int irqno)
1320 {
1321 	return (irqno);	/* default to NO translation */
1322 }
1323 
1324 static void
1325 mach_notify_error(int level, char *errmsg)
1326 {
1327 	/*
1328 	 * SL_FATAL is pass in once panicstr is set, deliver it
1329 	 * as CE_PANIC.  Also, translate SL_ codes back to CE_
1330 	 * codes for the psmi handler
1331 	 */
1332 	if (level & SL_FATAL)
1333 		(*notify_error)(CE_PANIC, errmsg);
1334 	else if (level & SL_WARN)
1335 		(*notify_error)(CE_WARN, errmsg);
1336 	else if (level & SL_NOTE)
1337 		(*notify_error)(CE_NOTE, errmsg);
1338 	else if (level & SL_CONSOLE)
1339 		(*notify_error)(CE_CONT, errmsg);
1340 }
1341 
1342 /*
1343  * It provides the default basic intr_ops interface for the new DDI
1344  * interrupt framework if the PSM doesn't have one.
1345  *
1346  * Input:
1347  * dip     - pointer to the dev_info structure of the requested device
1348  * hdlp    - pointer to the internal interrupt handle structure for the
1349  *	     requested interrupt
1350  * intr_op - opcode for this call
1351  * result  - pointer to the integer that will hold the result to be
1352  *	     passed back if return value is PSM_SUCCESS
1353  *
1354  * Output:
1355  * return value is either PSM_SUCCESS or PSM_FAILURE
1356  */
1357 static int
1358 mach_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1359     psm_intr_op_t intr_op, int *result)
1360 {
1361 	struct intrspec *ispec;
1362 
1363 	switch (intr_op) {
1364 	case PSM_INTR_OP_CHECK_MSI:
1365 		*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1366 		    DDI_INTR_TYPE_MSIX);
1367 		break;
1368 	case PSM_INTR_OP_ALLOC_VECTORS:
1369 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1370 			*result = 1;
1371 		else
1372 			*result = 0;
1373 		break;
1374 	case PSM_INTR_OP_FREE_VECTORS:
1375 		break;
1376 	case PSM_INTR_OP_NAVAIL_VECTORS:
1377 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1378 			*result = 1;
1379 		else
1380 			*result = 0;
1381 		break;
1382 	case PSM_INTR_OP_XLATE_VECTOR:
1383 		ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1384 		*result = psm_translate_irq(dip, ispec->intrspec_vec);
1385 		break;
1386 	case PSM_INTR_OP_GET_CAP:
1387 		*result = 0;
1388 		break;
1389 	case PSM_INTR_OP_GET_PENDING:
1390 	case PSM_INTR_OP_CLEAR_MASK:
1391 	case PSM_INTR_OP_SET_MASK:
1392 	case PSM_INTR_OP_GET_SHARED:
1393 	case PSM_INTR_OP_SET_PRI:
1394 	case PSM_INTR_OP_SET_CAP:
1395 	case PSM_INTR_OP_SET_CPU:
1396 	case PSM_INTR_OP_GET_INTR:
1397 	default:
1398 		return (PSM_FAILURE);
1399 	}
1400 	return (PSM_SUCCESS);
1401 }
1402 /*
1403  * Return 1 if CMT load balancing policies should be
1404  * implemented across instances of the specified hardware
1405  * sharing relationship.
1406  */
1407 int
1408 pg_cmt_load_bal_hw(pghw_type_t hw)
1409 {
1410 	if (hw == PGHW_IPIPE ||
1411 	    hw == PGHW_FPU ||
1412 	    hw == PGHW_CHIP)
1413 		return (1);
1414 	else
1415 		return (0);
1416 }
1417 /*
1418  * Return 1 if thread affinity polices should be implemented
1419  * for instances of the specifed hardware sharing relationship.
1420  */
1421 int
1422 pg_cmt_affinity_hw(pghw_type_t hw)
1423 {
1424 	if (hw == PGHW_CACHE)
1425 		return (1);
1426 	else
1427 		return (0);
1428 }
1429