xref: /titanic_41/usr/src/uts/i86pc/os/mp_machdep.c (revision 0db3240d392634cfff2f95fb6da34b56b8dc574f)
1 
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 /*
27  * Copyright (c) 2009-2010, Intel Corporation.
28  * All rights reserved.
29  */
30 
31 #define	PSMI_1_7
32 #include <sys/smp_impldefs.h>
33 #include <sys/psm.h>
34 #include <sys/psm_modctl.h>
35 #include <sys/pit.h>
36 #include <sys/cmn_err.h>
37 #include <sys/strlog.h>
38 #include <sys/clock.h>
39 #include <sys/debug.h>
40 #include <sys/rtc.h>
41 #include <sys/x86_archext.h>
42 #include <sys/cpupart.h>
43 #include <sys/cpuvar.h>
44 #include <sys/cpu_event.h>
45 #include <sys/cmt.h>
46 #include <sys/cpu.h>
47 #include <sys/disp.h>
48 #include <sys/archsystm.h>
49 #include <sys/machsystm.h>
50 #include <sys/sysmacros.h>
51 #include <sys/memlist.h>
52 #include <sys/param.h>
53 #include <sys/promif.h>
54 #include <sys/cpu_pm.h>
55 #if defined(__xpv)
56 #include <sys/hypervisor.h>
57 #endif
58 #include <sys/mach_intr.h>
59 #include <vm/hat_i86.h>
60 #include <sys/kdi_machimpl.h>
61 #include <sys/sdt.h>
62 #include <sys/hpet.h>
63 #include <sys/sunddi.h>
64 #include <sys/sunndi.h>
65 #include <sys/cpc_pcbe.h>
66 
67 #define	OFFSETOF(s, m)		(size_t)(&(((s *)0)->m))
68 
69 /*
70  *	Local function prototypes
71  */
72 static int mp_disable_intr(processorid_t cpun);
73 static void mp_enable_intr(processorid_t cpun);
74 static void mach_init();
75 static void mach_picinit();
76 static int machhztomhz(uint64_t cpu_freq_hz);
77 static uint64_t mach_getcpufreq(void);
78 static void mach_fixcpufreq(void);
79 static int mach_clkinit(int, int *);
80 static void mach_smpinit(void);
81 static int mach_softlvl_to_vect(int ipl);
82 static void mach_get_platform(int owner);
83 static void mach_construct_info();
84 static int mach_translate_irq(dev_info_t *dip, int irqno);
85 static int mach_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
86     psm_intr_op_t, int *);
87 static void mach_notify_error(int level, char *errmsg);
88 static hrtime_t dummy_hrtime(void);
89 static void dummy_scalehrtime(hrtime_t *);
90 static uint64_t dummy_unscalehrtime(hrtime_t);
91 void cpu_idle(void);
92 static void cpu_wakeup(cpu_t *, int);
93 #ifndef __xpv
94 void cpu_idle_mwait(void);
95 static void cpu_wakeup_mwait(cpu_t *, int);
96 #endif
97 static int mach_cpu_create_devinfo(cpu_t *cp, dev_info_t **dipp);
98 
99 /*
100  *	External reference functions
101  */
102 extern void return_instr();
103 extern uint64_t freq_tsc(uint32_t *);
104 #if defined(__i386)
105 extern uint64_t freq_notsc(uint32_t *);
106 #endif
107 extern void pc_gethrestime(timestruc_t *);
108 extern int cpuid_get_coreid(cpu_t *);
109 extern int cpuid_get_chipid(cpu_t *);
110 
111 /*
112  *	PSM functions initialization
113  */
114 void (*psm_shutdownf)(int, int)	= (void (*)(int, int))return_instr;
115 void (*psm_preshutdownf)(int, int) = (void (*)(int, int))return_instr;
116 void (*psm_notifyf)(int)	= (void (*)(int))return_instr;
117 void (*psm_set_idle_cpuf)(int)	= (void (*)(int))return_instr;
118 void (*psm_unset_idle_cpuf)(int) = (void (*)(int))return_instr;
119 void (*psminitf)()		= mach_init;
120 void (*picinitf)() 		= return_instr;
121 int (*clkinitf)(int, int *) 	= (int (*)(int, int *))return_instr;
122 int (*ap_mlsetup)() 		= (int (*)(void))return_instr;
123 void (*send_dirintf)() 		= return_instr;
124 void (*setspl)(int)		= (void (*)(int))return_instr;
125 int (*addspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
126 int (*delspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
127 void (*kdisetsoftint)(int, struct av_softinfo *)=
128 	(void (*)(int, struct av_softinfo *))return_instr;
129 void (*setsoftint)(int, struct av_softinfo *)=
130 	(void (*)(int, struct av_softinfo *))return_instr;
131 int (*slvltovect)(int)		= (int (*)(int))return_instr;
132 int (*setlvl)(int, int *)	= (int (*)(int, int *))return_instr;
133 void (*setlvlx)(int, int)	= (void (*)(int, int))return_instr;
134 int (*psm_disable_intr)(int)	= mp_disable_intr;
135 void (*psm_enable_intr)(int)	= mp_enable_intr;
136 hrtime_t (*gethrtimef)(void)	= dummy_hrtime;
137 hrtime_t (*gethrtimeunscaledf)(void)	= dummy_hrtime;
138 void (*scalehrtimef)(hrtime_t *)	= dummy_scalehrtime;
139 uint64_t (*unscalehrtimef)(hrtime_t)	= dummy_unscalehrtime;
140 int (*psm_translate_irq)(dev_info_t *, int) = mach_translate_irq;
141 void (*gethrestimef)(timestruc_t *) = pc_gethrestime;
142 void (*psm_notify_error)(int, char *) = (void (*)(int, char *))NULL;
143 int (*psm_get_clockirq)(int) = NULL;
144 int (*psm_get_ipivect)(int, int) = NULL;
145 
146 int (*psm_clkinit)(int) = NULL;
147 void (*psm_timer_reprogram)(hrtime_t) = NULL;
148 void (*psm_timer_enable)(void) = NULL;
149 void (*psm_timer_disable)(void) = NULL;
150 void (*psm_post_cyclic_setup)(void *arg) = NULL;
151 int (*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *, psm_intr_op_t,
152     int *) = mach_intr_ops;
153 int (*psm_state)(psm_state_request_t *) = (int (*)(psm_state_request_t *))
154     return_instr;
155 
156 void (*notify_error)(int, char *) = (void (*)(int, char *))return_instr;
157 void (*hrtime_tick)(void)	= return_instr;
158 
159 int (*psm_cpu_create_devinfo)(cpu_t *, dev_info_t **) = mach_cpu_create_devinfo;
160 int (*psm_cpu_get_devinfo)(cpu_t *, dev_info_t **) = NULL;
161 
162 /*
163  * True if the generic TSC code is our source of hrtime, rather than whatever
164  * the PSM can provide.
165  */
166 #ifdef __xpv
167 int tsc_gethrtime_enable = 0;
168 #else
169 int tsc_gethrtime_enable = 1;
170 #endif
171 int tsc_gethrtime_initted = 0;
172 
173 /*
174  * True if the hrtime implementation is "hires"; namely, better than microdata.
175  */
176 int gethrtime_hires = 0;
177 
178 /*
179  * Local Static Data
180  */
181 static struct psm_ops mach_ops;
182 static struct psm_ops *mach_set[4] = {&mach_ops, NULL, NULL, NULL};
183 static ushort_t mach_ver[4] = {0, 0, 0, 0};
184 
185 /*
186  * virtualization support for psm
187  */
188 void *psm_vt_ops = NULL;
189 /*
190  * If non-zero, idle cpus will become "halted" when there's
191  * no work to do.
192  */
193 int	idle_cpu_use_hlt = 1;
194 
195 #ifndef __xpv
196 /*
197  * If non-zero, idle cpus will use mwait if available to halt instead of hlt.
198  */
199 int	idle_cpu_prefer_mwait = 1;
200 /*
201  * Set to 0 to avoid MONITOR+CLFLUSH assertion.
202  */
203 int	idle_cpu_assert_cflush_monitor = 1;
204 
205 /*
206  * If non-zero, idle cpus will not use power saving Deep C-States idle loop.
207  */
208 int	idle_cpu_no_deep_c = 0;
209 /*
210  * Non-power saving idle loop and wakeup pointers.
211  * Allows user to toggle Deep Idle power saving feature on/off.
212  */
213 void	(*non_deep_idle_cpu)() = cpu_idle;
214 void	(*non_deep_idle_disp_enq_thread)(cpu_t *, int);
215 
216 /*
217  * Object for the kernel to access the HPET.
218  */
219 hpet_t hpet;
220 
221 #endif	/* ifndef __xpv */
222 
223 /*ARGSUSED*/
224 int
225 pg_plat_hw_shared(cpu_t *cp, pghw_type_t hw)
226 {
227 	switch (hw) {
228 	case PGHW_IPIPE:
229 		if (x86_feature & (X86_HTT)) {
230 			/*
231 			 * Hyper-threading is SMT
232 			 */
233 			return (1);
234 		} else {
235 			return (0);
236 		}
237 	case PGHW_PROCNODE:
238 		if (cpuid_get_procnodes_per_pkg(cp) > 1)
239 			return (1);
240 		else
241 			return (0);
242 	case PGHW_CHIP:
243 		if (x86_feature & (X86_CMP|X86_HTT))
244 			return (1);
245 		else
246 			return (0);
247 	case PGHW_CACHE:
248 		if (cpuid_get_ncpu_sharing_last_cache(cp) > 1)
249 			return (1);
250 		else
251 			return (0);
252 	case PGHW_POW_ACTIVE:
253 		if (cpupm_domain_id(cp, CPUPM_DTYPE_ACTIVE) != (id_t)-1)
254 			return (1);
255 		else
256 			return (0);
257 	case PGHW_POW_IDLE:
258 		if (cpupm_domain_id(cp, CPUPM_DTYPE_IDLE) != (id_t)-1)
259 			return (1);
260 		else
261 			return (0);
262 	default:
263 		return (0);
264 	}
265 }
266 
267 /*
268  * Compare two CPUs and see if they have a pghw_type_t sharing relationship
269  * If pghw_type_t is an unsupported hardware type, then return -1
270  */
271 int
272 pg_plat_cpus_share(cpu_t *cpu_a, cpu_t *cpu_b, pghw_type_t hw)
273 {
274 	id_t pgp_a, pgp_b;
275 
276 	pgp_a = pg_plat_hw_instance_id(cpu_a, hw);
277 	pgp_b = pg_plat_hw_instance_id(cpu_b, hw);
278 
279 	if (pgp_a == -1 || pgp_b == -1)
280 		return (-1);
281 
282 	return (pgp_a == pgp_b);
283 }
284 
285 /*
286  * Return a physical instance identifier for known hardware sharing
287  * relationships
288  */
289 id_t
290 pg_plat_hw_instance_id(cpu_t *cpu, pghw_type_t hw)
291 {
292 	switch (hw) {
293 	case PGHW_IPIPE:
294 		return (cpuid_get_coreid(cpu));
295 	case PGHW_CACHE:
296 		return (cpuid_get_last_lvl_cacheid(cpu));
297 	case PGHW_PROCNODE:
298 		return (cpuid_get_procnodeid(cpu));
299 	case PGHW_CHIP:
300 		return (cpuid_get_chipid(cpu));
301 	case PGHW_POW_ACTIVE:
302 		return (cpupm_domain_id(cpu, CPUPM_DTYPE_ACTIVE));
303 	case PGHW_POW_IDLE:
304 		return (cpupm_domain_id(cpu, CPUPM_DTYPE_IDLE));
305 	default:
306 		return (-1);
307 	}
308 }
309 
310 /*
311  * Express preference for optimizing for sharing relationship
312  * hw1 vs hw2
313  */
314 pghw_type_t
315 pg_plat_hw_rank(pghw_type_t hw1, pghw_type_t hw2)
316 {
317 	int i, rank1, rank2;
318 
319 	static pghw_type_t hw_hier[] = {
320 		PGHW_IPIPE,
321 		PGHW_CACHE,
322 		PGHW_PROCNODE,
323 		PGHW_CHIP,
324 		PGHW_POW_IDLE,
325 		PGHW_POW_ACTIVE,
326 		PGHW_NUM_COMPONENTS
327 	};
328 
329 	for (i = 0; hw_hier[i] != PGHW_NUM_COMPONENTS; i++) {
330 		if (hw_hier[i] == hw1)
331 			rank1 = i;
332 		if (hw_hier[i] == hw2)
333 			rank2 = i;
334 	}
335 
336 	if (rank1 > rank2)
337 		return (hw1);
338 	else
339 		return (hw2);
340 }
341 
342 /*
343  * Override the default CMT dispatcher policy for the specified
344  * hardware sharing relationship
345  */
346 pg_cmt_policy_t
347 pg_plat_cmt_policy(pghw_type_t hw)
348 {
349 	/*
350 	 * For shared caches, also load balance across them to
351 	 * maximize aggregate cache capacity
352 	 */
353 	switch (hw) {
354 	case PGHW_CACHE:
355 		return (CMT_BALANCE|CMT_AFFINITY);
356 	default:
357 		return (CMT_NO_POLICY);
358 	}
359 }
360 
361 id_t
362 pg_plat_get_core_id(cpu_t *cpu)
363 {
364 	return ((id_t)cpuid_get_coreid(cpu));
365 }
366 
367 void
368 cmp_set_nosteal_interval(void)
369 {
370 	/* Set the nosteal interval (used by disp_getbest()) to 100us */
371 	nosteal_nsec = 100000UL;
372 }
373 
374 /*
375  * Routine to ensure initial callers to hrtime gets 0 as return
376  */
377 static hrtime_t
378 dummy_hrtime(void)
379 {
380 	return (0);
381 }
382 
383 /* ARGSUSED */
384 static void
385 dummy_scalehrtime(hrtime_t *ticks)
386 {}
387 
388 static uint64_t
389 dummy_unscalehrtime(hrtime_t nsecs)
390 {
391 	return ((uint64_t)nsecs);
392 }
393 
394 /*
395  * Supports Deep C-State power saving idle loop.
396  */
397 void
398 cpu_idle_adaptive(void)
399 {
400 	(*CPU->cpu_m.mcpu_idle_cpu)();
401 }
402 
403 /*
404  * Function called by CPU idle notification framework to check whether CPU
405  * has been awakened. It will be called with interrupt disabled.
406  * If CPU has been awakened, call cpu_idle_exit() to notify CPU idle
407  * notification framework.
408  */
409 /*ARGSUSED*/
410 static void
411 cpu_idle_check_wakeup(void *arg)
412 {
413 	/*
414 	 * Toggle interrupt flag to detect pending interrupts.
415 	 * If interrupt happened, do_interrupt() will notify CPU idle
416 	 * notification framework so no need to call cpu_idle_exit() here.
417 	 */
418 	sti();
419 	SMT_PAUSE();
420 	cli();
421 }
422 
423 /*
424  * Idle the present CPU until wakened via an interrupt
425  */
426 void
427 cpu_idle(void)
428 {
429 	cpu_t		*cpup = CPU;
430 	processorid_t	cpu_sid = cpup->cpu_seqid;
431 	cpupart_t	*cp = cpup->cpu_part;
432 	int		hset_update = 1;
433 
434 	/*
435 	 * If this CPU is online, and there's multiple CPUs
436 	 * in the system, then we should notate our halting
437 	 * by adding ourselves to the partition's halted CPU
438 	 * bitmap. This allows other CPUs to find/awaken us when
439 	 * work becomes available.
440 	 */
441 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
442 		hset_update = 0;
443 
444 	/*
445 	 * Add ourselves to the partition's halted CPUs bitmap
446 	 * and set our HALTED flag, if necessary.
447 	 *
448 	 * When a thread becomes runnable, it is placed on the queue
449 	 * and then the halted CPU bitmap is checked to determine who
450 	 * (if anyone) should be awakened. We therefore need to first
451 	 * add ourselves to the bitmap, and and then check if there
452 	 * is any work available. The order is important to prevent a race
453 	 * that can lead to work languishing on a run queue somewhere while
454 	 * this CPU remains halted.
455 	 *
456 	 * Either the producing CPU will see we're halted and will awaken us,
457 	 * or this CPU will see the work available in disp_anywork().
458 	 *
459 	 * Note that memory barriers after updating the HALTED flag
460 	 * are not necessary since an atomic operation (updating the bitset)
461 	 * immediately follows. On x86 the atomic operation acts as a
462 	 * memory barrier for the update of cpu_disp_flags.
463 	 */
464 	if (hset_update) {
465 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
466 		bitset_atomic_add(&cp->cp_haltset, cpu_sid);
467 	}
468 
469 	/*
470 	 * Check to make sure there's really nothing to do.
471 	 * Work destined for this CPU may become available after
472 	 * this check. We'll be notified through the clearing of our
473 	 * bit in the halted CPU bitmap, and a poke.
474 	 */
475 	if (disp_anywork()) {
476 		if (hset_update) {
477 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
478 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
479 		}
480 		return;
481 	}
482 
483 	/*
484 	 * We're on our way to being halted.
485 	 *
486 	 * Disable interrupts now, so that we'll awaken immediately
487 	 * after halting if someone tries to poke us between now and
488 	 * the time we actually halt.
489 	 *
490 	 * We check for the presence of our bit after disabling interrupts.
491 	 * If it's cleared, we'll return. If the bit is cleared after
492 	 * we check then the poke will pop us out of the halted state.
493 	 *
494 	 * This means that the ordering of the poke and the clearing
495 	 * of the bit by cpu_wakeup is important.
496 	 * cpu_wakeup() must clear, then poke.
497 	 * cpu_idle() must disable interrupts, then check for the bit.
498 	 */
499 	cli();
500 
501 	if (hset_update && bitset_in_set(&cp->cp_haltset, cpu_sid) == 0) {
502 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
503 		sti();
504 		return;
505 	}
506 
507 	/*
508 	 * The check for anything locally runnable is here for performance
509 	 * and isn't needed for correctness. disp_nrunnable ought to be
510 	 * in our cache still, so it's inexpensive to check, and if there
511 	 * is anything runnable we won't have to wait for the poke.
512 	 */
513 	if (cpup->cpu_disp->disp_nrunnable != 0) {
514 		if (hset_update) {
515 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
516 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
517 		}
518 		sti();
519 		return;
520 	}
521 
522 	if (cpu_idle_enter(IDLE_STATE_C1, 0,
523 	    cpu_idle_check_wakeup, NULL) == 0) {
524 		mach_cpu_idle();
525 		cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
526 	}
527 
528 	/*
529 	 * We're no longer halted
530 	 */
531 	if (hset_update) {
532 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
533 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
534 	}
535 }
536 
537 
538 /*
539  * If "cpu" is halted, then wake it up clearing its halted bit in advance.
540  * Otherwise, see if other CPUs in the cpu partition are halted and need to
541  * be woken up so that they can steal the thread we placed on this CPU.
542  * This function is only used on MP systems.
543  */
544 static void
545 cpu_wakeup(cpu_t *cpu, int bound)
546 {
547 	uint_t		cpu_found;
548 	processorid_t	cpu_sid;
549 	cpupart_t	*cp;
550 
551 	cp = cpu->cpu_part;
552 	cpu_sid = cpu->cpu_seqid;
553 	if (bitset_in_set(&cp->cp_haltset, cpu_sid)) {
554 		/*
555 		 * Clear the halted bit for that CPU since it will be
556 		 * poked in a moment.
557 		 */
558 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
559 		/*
560 		 * We may find the current CPU present in the halted cpuset
561 		 * if we're in the context of an interrupt that occurred
562 		 * before we had a chance to clear our bit in cpu_idle().
563 		 * Poking ourself is obviously unnecessary, since if
564 		 * we're here, we're not halted.
565 		 */
566 		if (cpu != CPU)
567 			poke_cpu(cpu->cpu_id);
568 		return;
569 	} else {
570 		/*
571 		 * This cpu isn't halted, but it's idle or undergoing a
572 		 * context switch. No need to awaken anyone else.
573 		 */
574 		if (cpu->cpu_thread == cpu->cpu_idle_thread ||
575 		    cpu->cpu_disp_flags & CPU_DISP_DONTSTEAL)
576 			return;
577 	}
578 
579 	/*
580 	 * No need to wake up other CPUs if this is for a bound thread.
581 	 */
582 	if (bound)
583 		return;
584 
585 	/*
586 	 * The CPU specified for wakeup isn't currently halted, so check
587 	 * to see if there are any other halted CPUs in the partition,
588 	 * and if there are then awaken one.
589 	 */
590 	do {
591 		cpu_found = bitset_find(&cp->cp_haltset);
592 		if (cpu_found == (uint_t)-1)
593 			return;
594 	} while (bitset_atomic_test_and_del(&cp->cp_haltset, cpu_found) < 0);
595 
596 	if (cpu_found != CPU->cpu_seqid) {
597 		poke_cpu(cpu_seq[cpu_found]->cpu_id);
598 	}
599 }
600 
601 #ifndef __xpv
602 /*
603  * Function called by CPU idle notification framework to check whether CPU
604  * has been awakened. It will be called with interrupt disabled.
605  * If CPU has been awakened, call cpu_idle_exit() to notify CPU idle
606  * notification framework.
607  */
608 static void
609 cpu_idle_mwait_check_wakeup(void *arg)
610 {
611 	volatile uint32_t *mcpu_mwait = (volatile uint32_t *)arg;
612 
613 	ASSERT(arg != NULL);
614 	if (*mcpu_mwait != MWAIT_HALTED) {
615 		/*
616 		 * CPU has been awakened, notify CPU idle notification system.
617 		 */
618 		cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
619 	} else {
620 		/*
621 		 * Toggle interrupt flag to detect pending interrupts.
622 		 * If interrupt happened, do_interrupt() will notify CPU idle
623 		 * notification framework so no need to call cpu_idle_exit()
624 		 * here.
625 		 */
626 		sti();
627 		SMT_PAUSE();
628 		cli();
629 	}
630 }
631 
632 /*
633  * Idle the present CPU until awakened via touching its monitored line
634  */
635 void
636 cpu_idle_mwait(void)
637 {
638 	volatile uint32_t	*mcpu_mwait = CPU->cpu_m.mcpu_mwait;
639 	cpu_t			*cpup = CPU;
640 	processorid_t		cpu_sid = cpup->cpu_seqid;
641 	cpupart_t		*cp = cpup->cpu_part;
642 	int			hset_update = 1;
643 
644 	/*
645 	 * Set our mcpu_mwait here, so we can tell if anyone tries to
646 	 * wake us between now and when we call mwait.  No other cpu will
647 	 * attempt to set our mcpu_mwait until we add ourself to the halted
648 	 * CPU bitmap.
649 	 */
650 	*mcpu_mwait = MWAIT_HALTED;
651 
652 	/*
653 	 * If this CPU is online, and there's multiple CPUs
654 	 * in the system, then we should note our halting
655 	 * by adding ourselves to the partition's halted CPU
656 	 * bitmap. This allows other CPUs to find/awaken us when
657 	 * work becomes available.
658 	 */
659 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
660 		hset_update = 0;
661 
662 	/*
663 	 * Add ourselves to the partition's halted CPUs bitmap
664 	 * and set our HALTED flag, if necessary.
665 	 *
666 	 * When a thread becomes runnable, it is placed on the queue
667 	 * and then the halted CPU bitmap is checked to determine who
668 	 * (if anyone) should be awakened. We therefore need to first
669 	 * add ourselves to the bitmap, and and then check if there
670 	 * is any work available.
671 	 *
672 	 * Note that memory barriers after updating the HALTED flag
673 	 * are not necessary since an atomic operation (updating the bitmap)
674 	 * immediately follows. On x86 the atomic operation acts as a
675 	 * memory barrier for the update of cpu_disp_flags.
676 	 */
677 	if (hset_update) {
678 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
679 		bitset_atomic_add(&cp->cp_haltset, cpu_sid);
680 	}
681 
682 	/*
683 	 * Check to make sure there's really nothing to do.
684 	 * Work destined for this CPU may become available after
685 	 * this check. We'll be notified through the clearing of our
686 	 * bit in the halted CPU bitmap, and a write to our mcpu_mwait.
687 	 *
688 	 * disp_anywork() checks disp_nrunnable, so we do not have to later.
689 	 */
690 	if (disp_anywork()) {
691 		if (hset_update) {
692 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
693 			bitset_atomic_del(&cp->cp_haltset, cpu_sid);
694 		}
695 		return;
696 	}
697 
698 	/*
699 	 * We're on our way to being halted.
700 	 * To avoid a lost wakeup, arm the monitor before checking if another
701 	 * cpu wrote to mcpu_mwait to wake us up.
702 	 */
703 	i86_monitor(mcpu_mwait, 0, 0);
704 	if (*mcpu_mwait == MWAIT_HALTED) {
705 		if (cpu_idle_enter(IDLE_STATE_C1, 0,
706 		    cpu_idle_mwait_check_wakeup, (void *)mcpu_mwait) == 0) {
707 			if (*mcpu_mwait == MWAIT_HALTED) {
708 				i86_mwait(0, 0);
709 			}
710 			cpu_idle_exit(CPU_IDLE_CB_FLAG_IDLE);
711 		}
712 	}
713 
714 	/*
715 	 * We're no longer halted
716 	 */
717 	if (hset_update) {
718 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
719 		bitset_atomic_del(&cp->cp_haltset, cpu_sid);
720 	}
721 }
722 
723 /*
724  * If "cpu" is halted in mwait, then wake it up clearing its halted bit in
725  * advance.  Otherwise, see if other CPUs in the cpu partition are halted and
726  * need to be woken up so that they can steal the thread we placed on this CPU.
727  * This function is only used on MP systems.
728  */
729 static void
730 cpu_wakeup_mwait(cpu_t *cp, int bound)
731 {
732 	cpupart_t	*cpu_part;
733 	uint_t		cpu_found;
734 	processorid_t	cpu_sid;
735 
736 	cpu_part = cp->cpu_part;
737 	cpu_sid = cp->cpu_seqid;
738 
739 	/*
740 	 * Clear the halted bit for that CPU since it will be woken up
741 	 * in a moment.
742 	 */
743 	if (bitset_in_set(&cpu_part->cp_haltset, cpu_sid)) {
744 		/*
745 		 * Clear the halted bit for that CPU since it will be
746 		 * poked in a moment.
747 		 */
748 		bitset_atomic_del(&cpu_part->cp_haltset, cpu_sid);
749 		/*
750 		 * We may find the current CPU present in the halted cpuset
751 		 * if we're in the context of an interrupt that occurred
752 		 * before we had a chance to clear our bit in cpu_idle().
753 		 * Waking ourself is obviously unnecessary, since if
754 		 * we're here, we're not halted.
755 		 *
756 		 * monitor/mwait wakeup via writing to our cache line is
757 		 * harmless and less expensive than always checking if we
758 		 * are waking ourself which is an uncommon case.
759 		 */
760 		MWAIT_WAKEUP(cp);	/* write to monitored line */
761 		return;
762 	} else {
763 		/*
764 		 * This cpu isn't halted, but it's idle or undergoing a
765 		 * context switch. No need to awaken anyone else.
766 		 */
767 		if (cp->cpu_thread == cp->cpu_idle_thread ||
768 		    cp->cpu_disp_flags & CPU_DISP_DONTSTEAL)
769 			return;
770 	}
771 
772 	/*
773 	 * No need to wake up other CPUs if the thread we just enqueued
774 	 * is bound.
775 	 */
776 	if (bound || ncpus == 1)
777 		return;
778 
779 	/*
780 	 * See if there's any other halted CPUs. If there are, then
781 	 * select one, and awaken it.
782 	 * It's possible that after we find a CPU, somebody else
783 	 * will awaken it before we get the chance.
784 	 * In that case, look again.
785 	 */
786 	do {
787 		cpu_found = bitset_find(&cpu_part->cp_haltset);
788 		if (cpu_found == (uint_t)-1)
789 			return;
790 	} while (bitset_atomic_test_and_del(&cpu_part->cp_haltset,
791 	    cpu_found) < 0);
792 
793 	/*
794 	 * Do not check if cpu_found is ourself as monitor/mwait
795 	 * wakeup is cheap.
796 	 */
797 	MWAIT_WAKEUP(cpu_seq[cpu_found]); /* write to monitored line */
798 }
799 
800 #endif
801 
802 void (*cpu_pause_handler)(volatile char *) = NULL;
803 
804 static int
805 mp_disable_intr(int cpun)
806 {
807 	/*
808 	 * switch to the offline cpu
809 	 */
810 	affinity_set(cpun);
811 	/*
812 	 * raise ipl to just below cross call
813 	 */
814 	splx(XC_SYS_PIL - 1);
815 	/*
816 	 *	set base spl to prevent the next swtch to idle from
817 	 *	lowering back to ipl 0
818 	 */
819 	CPU->cpu_intr_actv |= (1 << (XC_SYS_PIL - 1));
820 	set_base_spl();
821 	affinity_clear();
822 	return (DDI_SUCCESS);
823 }
824 
825 static void
826 mp_enable_intr(int cpun)
827 {
828 	/*
829 	 * switch to the online cpu
830 	 */
831 	affinity_set(cpun);
832 	/*
833 	 * clear the interrupt active mask
834 	 */
835 	CPU->cpu_intr_actv &= ~(1 << (XC_SYS_PIL - 1));
836 	set_base_spl();
837 	(void) spl0();
838 	affinity_clear();
839 }
840 
841 static void
842 mach_get_platform(int owner)
843 {
844 	void		**srv_opsp;
845 	void		**clt_opsp;
846 	int		i;
847 	int		total_ops;
848 
849 	/* fix up psm ops */
850 	srv_opsp = (void **)mach_set[0];
851 	clt_opsp = (void **)mach_set[owner];
852 	if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01)
853 		total_ops = sizeof (struct psm_ops_ver01) /
854 		    sizeof (void (*)(void));
855 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_1)
856 		/* no psm_notify_func */
857 		total_ops = OFFSETOF(struct psm_ops, psm_notify_func) /
858 		    sizeof (void (*)(void));
859 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_2)
860 		/* no psm_timer funcs */
861 		total_ops = OFFSETOF(struct psm_ops, psm_timer_reprogram) /
862 		    sizeof (void (*)(void));
863 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_3)
864 		/* no psm_preshutdown function */
865 		total_ops = OFFSETOF(struct psm_ops, psm_preshutdown) /
866 		    sizeof (void (*)(void));
867 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_4)
868 		/* no psm_intr_ops function */
869 		total_ops = OFFSETOF(struct psm_ops, psm_intr_ops) /
870 		    sizeof (void (*)(void));
871 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_5)
872 		/* no psm_state function */
873 		total_ops = OFFSETOF(struct psm_ops, psm_state) /
874 		    sizeof (void (*)(void));
875 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_6)
876 		/* no psm_cpu_ops function */
877 		total_ops = OFFSETOF(struct psm_ops, psm_cpu_ops) /
878 		    sizeof (void (*)(void));
879 	else
880 		total_ops = sizeof (struct psm_ops) / sizeof (void (*)(void));
881 
882 	/*
883 	 * Save the version of the PSM module, in case we need to
884 	 * behave differently based on version.
885 	 */
886 	mach_ver[0] = mach_ver[owner];
887 
888 	for (i = 0; i < total_ops; i++)
889 		if (clt_opsp[i] != NULL)
890 			srv_opsp[i] = clt_opsp[i];
891 }
892 
893 static void
894 mach_construct_info()
895 {
896 	struct psm_sw *swp;
897 	int	mach_cnt[PSM_OWN_OVERRIDE+1] = {0};
898 	int	conflict_owner = 0;
899 
900 	if (psmsw->psw_forw == psmsw)
901 		panic("No valid PSM modules found");
902 	mutex_enter(&psmsw_lock);
903 	for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
904 		if (!(swp->psw_flag & PSM_MOD_IDENTIFY))
905 			continue;
906 		mach_set[swp->psw_infop->p_owner] = swp->psw_infop->p_ops;
907 		mach_ver[swp->psw_infop->p_owner] = swp->psw_infop->p_version;
908 		mach_cnt[swp->psw_infop->p_owner]++;
909 	}
910 	mutex_exit(&psmsw_lock);
911 
912 	mach_get_platform(PSM_OWN_SYS_DEFAULT);
913 
914 	/* check to see are there any conflicts */
915 	if (mach_cnt[PSM_OWN_EXCLUSIVE] > 1)
916 		conflict_owner = PSM_OWN_EXCLUSIVE;
917 	if (mach_cnt[PSM_OWN_OVERRIDE] > 1)
918 		conflict_owner = PSM_OWN_OVERRIDE;
919 	if (conflict_owner) {
920 		/* remove all psm modules except uppc */
921 		cmn_err(CE_WARN,
922 		    "Conflicts detected on the following PSM modules:");
923 		mutex_enter(&psmsw_lock);
924 		for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
925 			if (swp->psw_infop->p_owner == conflict_owner)
926 				cmn_err(CE_WARN, "%s ",
927 				    swp->psw_infop->p_mach_idstring);
928 		}
929 		mutex_exit(&psmsw_lock);
930 		cmn_err(CE_WARN,
931 		    "Setting the system back to SINGLE processor mode!");
932 		cmn_err(CE_WARN,
933 		    "Please edit /etc/mach to remove the invalid PSM module.");
934 		return;
935 	}
936 
937 	if (mach_set[PSM_OWN_EXCLUSIVE])
938 		mach_get_platform(PSM_OWN_EXCLUSIVE);
939 
940 	if (mach_set[PSM_OWN_OVERRIDE])
941 		mach_get_platform(PSM_OWN_OVERRIDE);
942 }
943 
944 static void
945 mach_init()
946 {
947 	struct psm_ops  *pops;
948 
949 	mach_construct_info();
950 
951 	pops = mach_set[0];
952 
953 	/* register the interrupt and clock initialization rotuines */
954 	picinitf = mach_picinit;
955 	clkinitf = mach_clkinit;
956 	psm_get_clockirq = pops->psm_get_clockirq;
957 
958 	/* register the interrupt setup code */
959 	slvltovect = mach_softlvl_to_vect;
960 	addspl	= pops->psm_addspl;
961 	delspl	= pops->psm_delspl;
962 
963 	if (pops->psm_translate_irq)
964 		psm_translate_irq = pops->psm_translate_irq;
965 	if (pops->psm_intr_ops)
966 		psm_intr_ops = pops->psm_intr_ops;
967 
968 #if defined(PSMI_1_2) || defined(PSMI_1_3) || defined(PSMI_1_4)
969 	/*
970 	 * Time-of-day functionality now handled in TOD modules.
971 	 * (Warn about PSM modules that think that we're going to use
972 	 * their ops vectors.)
973 	 */
974 	if (pops->psm_tod_get)
975 		cmn_err(CE_WARN, "obsolete psm_tod_get op %p",
976 		    (void *)pops->psm_tod_get);
977 
978 	if (pops->psm_tod_set)
979 		cmn_err(CE_WARN, "obsolete psm_tod_set op %p",
980 		    (void *)pops->psm_tod_set);
981 #endif
982 
983 	if (pops->psm_notify_error) {
984 		psm_notify_error = mach_notify_error;
985 		notify_error = pops->psm_notify_error;
986 	}
987 
988 	(*pops->psm_softinit)();
989 
990 	/*
991 	 * Initialize the dispatcher's function hooks to enable CPU halting
992 	 * when idle.  Set both the deep-idle and non-deep-idle hooks.
993 	 *
994 	 * Assume we can use power saving deep-idle loop cpu_idle_adaptive.
995 	 * Platform deep-idle driver will reset our idle loop to
996 	 * non_deep_idle_cpu if power saving deep-idle feature is not available.
997 	 *
998 	 * Do not use monitor/mwait if idle_cpu_use_hlt is not set(spin idle)
999 	 * or idle_cpu_prefer_mwait is not set.
1000 	 * Allocate monitor/mwait buffer for cpu0.
1001 	 */
1002 #ifndef __xpv
1003 	non_deep_idle_disp_enq_thread = disp_enq_thread;
1004 #endif
1005 	if (idle_cpu_use_hlt) {
1006 		idle_cpu = cpu_idle_adaptive;
1007 		CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
1008 #ifndef __xpv
1009 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait) {
1010 			CPU->cpu_m.mcpu_mwait = cpuid_mwait_alloc(CPU);
1011 			/*
1012 			 * Protect ourself from insane mwait size.
1013 			 */
1014 			if (CPU->cpu_m.mcpu_mwait == NULL) {
1015 #ifdef DEBUG
1016 				cmn_err(CE_NOTE, "Using hlt idle.  Cannot "
1017 				    "handle cpu 0 mwait size.");
1018 #endif
1019 				idle_cpu_prefer_mwait = 0;
1020 				CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
1021 			} else {
1022 				CPU->cpu_m.mcpu_idle_cpu = cpu_idle_mwait;
1023 			}
1024 		} else {
1025 			CPU->cpu_m.mcpu_idle_cpu = cpu_idle;
1026 		}
1027 		non_deep_idle_cpu = CPU->cpu_m.mcpu_idle_cpu;
1028 
1029 		/*
1030 		 * Disable power saving deep idle loop?
1031 		 */
1032 		if (idle_cpu_no_deep_c) {
1033 			idle_cpu = non_deep_idle_cpu;
1034 		}
1035 #endif
1036 	}
1037 
1038 	mach_smpinit();
1039 }
1040 
1041 static void
1042 mach_smpinit(void)
1043 {
1044 	struct psm_ops  *pops;
1045 	processorid_t cpu_id;
1046 	int cnt;
1047 	cpuset_t cpumask;
1048 
1049 	pops = mach_set[0];
1050 	CPUSET_ZERO(cpumask);
1051 
1052 	cpu_id = -1;
1053 	cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
1054 	/*
1055 	 * Only add boot_ncpus CPUs to mp_cpus. Other CPUs will be handled
1056 	 * by CPU DR driver at runtime.
1057 	 */
1058 	for (cnt = 0; cpu_id != -1 && cnt < boot_ncpus; cnt++) {
1059 		CPUSET_ADD(cpumask, cpu_id);
1060 		cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
1061 	}
1062 
1063 	mp_cpus = cpumask;
1064 
1065 	/* MP related routines */
1066 	ap_mlsetup = pops->psm_post_cpu_start;
1067 	send_dirintf = pops->psm_send_ipi;
1068 
1069 	/* optional MP related routines */
1070 	if (pops->psm_shutdown)
1071 		psm_shutdownf = pops->psm_shutdown;
1072 	if (pops->psm_preshutdown)
1073 		psm_preshutdownf = pops->psm_preshutdown;
1074 	if (pops->psm_notify_func)
1075 		psm_notifyf = pops->psm_notify_func;
1076 	if (pops->psm_set_idlecpu)
1077 		psm_set_idle_cpuf = pops->psm_set_idlecpu;
1078 	if (pops->psm_unset_idlecpu)
1079 		psm_unset_idle_cpuf = pops->psm_unset_idlecpu;
1080 
1081 	psm_clkinit = pops->psm_clkinit;
1082 
1083 	if (pops->psm_timer_reprogram)
1084 		psm_timer_reprogram = pops->psm_timer_reprogram;
1085 
1086 	if (pops->psm_timer_enable)
1087 		psm_timer_enable = pops->psm_timer_enable;
1088 
1089 	if (pops->psm_timer_disable)
1090 		psm_timer_disable = pops->psm_timer_disable;
1091 
1092 	if (pops->psm_post_cyclic_setup)
1093 		psm_post_cyclic_setup = pops->psm_post_cyclic_setup;
1094 
1095 	if (pops->psm_state)
1096 		psm_state = pops->psm_state;
1097 
1098 	/*
1099 	 * Set these vectors here so they can be used by Suspend/Resume
1100 	 * on UP machines.
1101 	 */
1102 	if (pops->psm_disable_intr)
1103 		psm_disable_intr = pops->psm_disable_intr;
1104 	if (pops->psm_enable_intr)
1105 		psm_enable_intr  = pops->psm_enable_intr;
1106 
1107 	/* check for multiple CPUs */
1108 	if (cnt < 2 && plat_dr_support_cpu() == B_FALSE)
1109 		return;
1110 
1111 	/* check for MP platforms */
1112 	if (pops->psm_cpu_start == NULL)
1113 		return;
1114 
1115 	/*
1116 	 * Set the dispatcher hook to enable cpu "wake up"
1117 	 * when a thread becomes runnable.
1118 	 */
1119 	if (idle_cpu_use_hlt) {
1120 		disp_enq_thread = cpu_wakeup;
1121 #ifndef __xpv
1122 		if ((x86_feature & X86_MWAIT) && idle_cpu_prefer_mwait)
1123 			disp_enq_thread = cpu_wakeup_mwait;
1124 		non_deep_idle_disp_enq_thread = disp_enq_thread;
1125 #endif
1126 	}
1127 
1128 	psm_get_ipivect = pops->psm_get_ipivect;
1129 
1130 	(void) add_avintr((void *)NULL, XC_HI_PIL, xc_serv, "xc_intr",
1131 	    (*pops->psm_get_ipivect)(XC_HI_PIL, PSM_INTR_IPI_HI),
1132 	    NULL, NULL, NULL, NULL);
1133 
1134 	(void) (*pops->psm_get_ipivect)(XC_CPUPOKE_PIL, PSM_INTR_POKE);
1135 }
1136 
1137 static void
1138 mach_picinit()
1139 {
1140 	struct psm_ops  *pops;
1141 
1142 	pops = mach_set[0];
1143 
1144 	/* register the interrupt handlers */
1145 	setlvl = pops->psm_intr_enter;
1146 	setlvlx = pops->psm_intr_exit;
1147 
1148 	/* initialize the interrupt hardware */
1149 	(*pops->psm_picinit)();
1150 
1151 	/* set interrupt mask for current ipl */
1152 	setspl = pops->psm_setspl;
1153 	cli();
1154 	setspl(CPU->cpu_pri);
1155 }
1156 
1157 uint_t	cpu_freq;	/* MHz */
1158 uint64_t cpu_freq_hz;	/* measured (in hertz) */
1159 
1160 #define	MEGA_HZ		1000000
1161 
1162 #ifdef __xpv
1163 
1164 int xpv_cpufreq_workaround = 1;
1165 int xpv_cpufreq_verbose = 0;
1166 
1167 #else	/* __xpv */
1168 
1169 static uint64_t
1170 mach_calchz(uint32_t pit_counter, uint64_t *processor_clks)
1171 {
1172 	uint64_t cpu_hz;
1173 
1174 	if ((pit_counter == 0) || (*processor_clks == 0) ||
1175 	    (*processor_clks > (((uint64_t)-1) / PIT_HZ)))
1176 		return (0);
1177 
1178 	cpu_hz = ((uint64_t)PIT_HZ * *processor_clks) / pit_counter;
1179 
1180 	return (cpu_hz);
1181 }
1182 
1183 #endif	/* __xpv */
1184 
1185 static uint64_t
1186 mach_getcpufreq(void)
1187 {
1188 #if defined(__xpv)
1189 	vcpu_time_info_t *vti = &CPU->cpu_m.mcpu_vcpu_info->time;
1190 	uint64_t cpu_hz;
1191 
1192 	/*
1193 	 * During dom0 bringup, it was noted that on at least one older
1194 	 * Intel HT machine, the hypervisor initially gives a tsc_to_system_mul
1195 	 * value that is quite wrong (the 3.06GHz clock was reported
1196 	 * as 4.77GHz)
1197 	 *
1198 	 * The curious thing is, that if you stop the kernel at entry,
1199 	 * breakpoint here and inspect the value with kmdb, the value
1200 	 * is correct - but if you don't stop and simply enable the
1201 	 * printf statement (below), you can see the bad value printed
1202 	 * here.  Almost as if something kmdb did caused the hypervisor to
1203 	 * figure it out correctly.  And, note that the hypervisor
1204 	 * eventually -does- figure it out correctly ... if you look at
1205 	 * the field later in the life of dom0, it is correct.
1206 	 *
1207 	 * For now, on dom0, we employ a slightly cheesy workaround of
1208 	 * using the DOM0_PHYSINFO hypercall.
1209 	 */
1210 	if (DOMAIN_IS_INITDOMAIN(xen_info) && xpv_cpufreq_workaround) {
1211 		cpu_hz = 1000 * xpv_cpu_khz();
1212 	} else {
1213 		cpu_hz = (UINT64_C(1000000000) << 32) / vti->tsc_to_system_mul;
1214 
1215 		if (vti->tsc_shift < 0)
1216 			cpu_hz <<= -vti->tsc_shift;
1217 		else
1218 			cpu_hz >>= vti->tsc_shift;
1219 	}
1220 
1221 	if (xpv_cpufreq_verbose)
1222 		printf("mach_getcpufreq: system_mul 0x%x, shift %d, "
1223 		    "cpu_hz %" PRId64 "Hz\n",
1224 		    vti->tsc_to_system_mul, vti->tsc_shift, cpu_hz);
1225 
1226 	return (cpu_hz);
1227 #else	/* __xpv */
1228 	uint32_t pit_counter;
1229 	uint64_t processor_clks;
1230 
1231 	if (x86_feature & X86_TSC) {
1232 		/*
1233 		 * We have a TSC. freq_tsc() knows how to measure the number
1234 		 * of clock cycles sampled against the PIT.
1235 		 */
1236 		ulong_t flags = clear_int_flag();
1237 		processor_clks = freq_tsc(&pit_counter);
1238 		restore_int_flag(flags);
1239 		return (mach_calchz(pit_counter, &processor_clks));
1240 	} else if (x86_vendor == X86_VENDOR_Cyrix || x86_type == X86_TYPE_P5) {
1241 #if defined(__amd64)
1242 		panic("mach_getcpufreq: no TSC!");
1243 #elif defined(__i386)
1244 		/*
1245 		 * We are a Cyrix based on a 6x86 core or an Intel Pentium
1246 		 * for which freq_notsc() knows how to measure the number of
1247 		 * elapsed clock cycles sampled against the PIT
1248 		 */
1249 		ulong_t flags = clear_int_flag();
1250 		processor_clks = freq_notsc(&pit_counter);
1251 		restore_int_flag(flags);
1252 		return (mach_calchz(pit_counter, &processor_clks));
1253 #endif	/* __i386 */
1254 	}
1255 
1256 	/* We do not know how to calculate cpu frequency for this cpu. */
1257 	return (0);
1258 #endif	/* __xpv */
1259 }
1260 
1261 /*
1262  * If the clock speed of a cpu is found to be reported incorrectly, do not add
1263  * to this array, instead improve the accuracy of the algorithm that determines
1264  * the clock speed of the processor or extend the implementation to support the
1265  * vendor as appropriate. This is here only to support adjusting the speed on
1266  * older slower processors that mach_fixcpufreq() would not be able to account
1267  * for otherwise.
1268  */
1269 static int x86_cpu_freq[] = { 60, 75, 80, 90, 120, 160, 166, 175, 180, 233 };
1270 
1271 /*
1272  * On fast processors the clock frequency that is measured may be off by
1273  * a few MHz from the value printed on the part. This is a combination of
1274  * the factors that for such fast parts being off by this much is within
1275  * the tolerances for manufacture and because of the difficulties in the
1276  * measurement that can lead to small error. This function uses some
1277  * heuristics in order to tweak the value that was measured to match what
1278  * is most likely printed on the part.
1279  *
1280  * Some examples:
1281  * 	AMD Athlon 1000 mhz measured as 998 mhz
1282  * 	Intel Pentium III Xeon 733 mhz measured as 731 mhz
1283  * 	Intel Pentium IV 1500 mhz measured as 1495mhz
1284  *
1285  * If in the future this function is no longer sufficient to correct
1286  * for the error in the measurement, then the algorithm used to perform
1287  * the measurement will have to be improved in order to increase accuracy
1288  * rather than adding horrible and questionable kludges here.
1289  *
1290  * This is called after the cyclics subsystem because of the potential
1291  * that the heuristics within may give a worse estimate of the clock
1292  * frequency than the value that was measured.
1293  */
1294 static void
1295 mach_fixcpufreq(void)
1296 {
1297 	uint32_t freq, mul, near66, delta66, near50, delta50, fixed, delta, i;
1298 
1299 	freq = (uint32_t)cpu_freq;
1300 
1301 	/*
1302 	 * Find the nearest integer multiple of 200/3 (about 66) MHz to the
1303 	 * measured speed taking into account that the 667 MHz parts were
1304 	 * the first to round-up.
1305 	 */
1306 	mul = (uint32_t)((3 * (uint64_t)freq + 100) / 200);
1307 	near66 = (uint32_t)((200 * (uint64_t)mul + ((mul >= 10) ? 1 : 0)) / 3);
1308 	delta66 = (near66 > freq) ? (near66 - freq) : (freq - near66);
1309 
1310 	/* Find the nearest integer multiple of 50 MHz to the measured speed */
1311 	mul = (freq + 25) / 50;
1312 	near50 = mul * 50;
1313 	delta50 = (near50 > freq) ? (near50 - freq) : (freq - near50);
1314 
1315 	/* Find the closer of the two */
1316 	if (delta66 < delta50) {
1317 		fixed = near66;
1318 		delta = delta66;
1319 	} else {
1320 		fixed = near50;
1321 		delta = delta50;
1322 	}
1323 
1324 	if (fixed > INT_MAX)
1325 		return;
1326 
1327 	/*
1328 	 * Some older parts have a core clock frequency that is not an
1329 	 * integral multiple of 50 or 66 MHz. Check if one of the old
1330 	 * clock frequencies is closer to the measured value than any
1331 	 * of the integral multiples of 50 an 66, and if so set fixed
1332 	 * and delta appropriately to represent the closest value.
1333 	 */
1334 	i = sizeof (x86_cpu_freq) / sizeof (int);
1335 	while (i > 0) {
1336 		i--;
1337 
1338 		if (x86_cpu_freq[i] <= freq) {
1339 			mul = freq - x86_cpu_freq[i];
1340 
1341 			if (mul < delta) {
1342 				fixed = x86_cpu_freq[i];
1343 				delta = mul;
1344 			}
1345 
1346 			break;
1347 		}
1348 
1349 		mul = x86_cpu_freq[i] - freq;
1350 
1351 		if (mul < delta) {
1352 			fixed = x86_cpu_freq[i];
1353 			delta = mul;
1354 		}
1355 	}
1356 
1357 	/*
1358 	 * Set a reasonable maximum for how much to correct the measured
1359 	 * result by. This check is here to prevent the adjustment made
1360 	 * by this function from being more harm than good. It is entirely
1361 	 * possible that in the future parts will be made that are not
1362 	 * integral multiples of 66 or 50 in clock frequency or that
1363 	 * someone may overclock a part to some odd frequency. If the
1364 	 * measured value is farther from the corrected value than
1365 	 * allowed, then assume the corrected value is in error and use
1366 	 * the measured value.
1367 	 */
1368 	if (6 < delta)
1369 		return;
1370 
1371 	cpu_freq = (int)fixed;
1372 }
1373 
1374 
1375 static int
1376 machhztomhz(uint64_t cpu_freq_hz)
1377 {
1378 	uint64_t cpu_mhz;
1379 
1380 	/* Round to nearest MHZ */
1381 	cpu_mhz = (cpu_freq_hz + (MEGA_HZ / 2)) / MEGA_HZ;
1382 
1383 	if (cpu_mhz > INT_MAX)
1384 		return (0);
1385 
1386 	return ((int)cpu_mhz);
1387 
1388 }
1389 
1390 
1391 static int
1392 mach_clkinit(int preferred_mode, int *set_mode)
1393 {
1394 	struct psm_ops  *pops;
1395 	int resolution;
1396 
1397 	pops = mach_set[0];
1398 
1399 	cpu_freq_hz = mach_getcpufreq();
1400 
1401 	cpu_freq = machhztomhz(cpu_freq_hz);
1402 
1403 	if (!(x86_feature & X86_TSC) || (cpu_freq == 0))
1404 		tsc_gethrtime_enable = 0;
1405 
1406 #ifndef __xpv
1407 	if (tsc_gethrtime_enable) {
1408 		tsc_hrtimeinit(cpu_freq_hz);
1409 	} else
1410 #endif
1411 	{
1412 		if (pops->psm_hrtimeinit)
1413 			(*pops->psm_hrtimeinit)();
1414 		gethrtimef = pops->psm_gethrtime;
1415 		gethrtimeunscaledf = gethrtimef;
1416 		/* scalehrtimef will remain dummy */
1417 	}
1418 
1419 	mach_fixcpufreq();
1420 
1421 	if (mach_ver[0] >= PSM_INFO_VER01_3) {
1422 		if (preferred_mode == TIMER_ONESHOT) {
1423 
1424 			resolution = (*pops->psm_clkinit)(0);
1425 			if (resolution != 0)  {
1426 				*set_mode = TIMER_ONESHOT;
1427 				return (resolution);
1428 			}
1429 		}
1430 
1431 		/*
1432 		 * either periodic mode was requested or could not set to
1433 		 * one-shot mode
1434 		 */
1435 		resolution = (*pops->psm_clkinit)(hz);
1436 		/*
1437 		 * psm should be able to do periodic, so we do not check
1438 		 * for return value of psm_clkinit here.
1439 		 */
1440 		*set_mode = TIMER_PERIODIC;
1441 		return (resolution);
1442 	} else {
1443 		/*
1444 		 * PSMI interface prior to PSMI_3 does not define a return
1445 		 * value for psm_clkinit, so the return value is ignored.
1446 		 */
1447 		(void) (*pops->psm_clkinit)(hz);
1448 		*set_mode = TIMER_PERIODIC;
1449 		return (nsec_per_tick);
1450 	}
1451 }
1452 
1453 
1454 /*ARGSUSED*/
1455 static int
1456 mach_softlvl_to_vect(int ipl)
1457 {
1458 	setsoftint = av_set_softint_pending;
1459 	kdisetsoftint = kdi_av_set_softint_pending;
1460 
1461 	return (PSM_SV_SOFTWARE);
1462 }
1463 
1464 #ifdef DEBUG
1465 /*
1466  * This is here to allow us to simulate cpus that refuse to start.
1467  */
1468 cpuset_t cpufailset;
1469 #endif
1470 
1471 int
1472 mach_cpu_start(struct cpu *cp, void *ctx)
1473 {
1474 	struct psm_ops *pops = mach_set[0];
1475 	processorid_t id = cp->cpu_id;
1476 
1477 #ifdef DEBUG
1478 	if (CPU_IN_SET(cpufailset, id))
1479 		return (0);
1480 #endif
1481 	return ((*pops->psm_cpu_start)(id, ctx));
1482 }
1483 
1484 int
1485 mach_cpuid_start(processorid_t id, void *ctx)
1486 {
1487 	struct psm_ops *pops = mach_set[0];
1488 
1489 #ifdef DEBUG
1490 	if (CPU_IN_SET(cpufailset, id))
1491 		return (0);
1492 #endif
1493 	return ((*pops->psm_cpu_start)(id, ctx));
1494 }
1495 
1496 int
1497 mach_cpu_stop(cpu_t *cp, void *ctx)
1498 {
1499 	struct psm_ops *pops = mach_set[0];
1500 	psm_cpu_request_t request;
1501 
1502 	if (pops->psm_cpu_ops == NULL) {
1503 		return (ENOTSUP);
1504 	}
1505 
1506 	ASSERT(cp->cpu_id != -1);
1507 	request.pcr_cmd = PSM_CPU_STOP;
1508 	request.req.cpu_stop.cpuid = cp->cpu_id;
1509 	request.req.cpu_stop.ctx = ctx;
1510 
1511 	return ((*pops->psm_cpu_ops)(&request));
1512 }
1513 
1514 int
1515 mach_cpu_add(mach_cpu_add_arg_t *argp, processorid_t *cpuidp)
1516 {
1517 	int rc;
1518 	struct psm_ops *pops = mach_set[0];
1519 	psm_cpu_request_t request;
1520 
1521 	if (pops->psm_cpu_ops == NULL) {
1522 		return (ENOTSUP);
1523 	}
1524 
1525 	request.pcr_cmd = PSM_CPU_ADD;
1526 	request.req.cpu_add.argp = argp;
1527 	request.req.cpu_add.cpuid = -1;
1528 	rc = (*pops->psm_cpu_ops)(&request);
1529 	if (rc == 0) {
1530 		ASSERT(request.req.cpu_add.cpuid != -1);
1531 		*cpuidp = request.req.cpu_add.cpuid;
1532 	}
1533 
1534 	return (rc);
1535 }
1536 
1537 int
1538 mach_cpu_remove(processorid_t cpuid)
1539 {
1540 	struct psm_ops *pops = mach_set[0];
1541 	psm_cpu_request_t request;
1542 
1543 	if (pops->psm_cpu_ops == NULL) {
1544 		return (ENOTSUP);
1545 	}
1546 
1547 	request.pcr_cmd = PSM_CPU_REMOVE;
1548 	request.req.cpu_remove.cpuid = cpuid;
1549 
1550 	return ((*pops->psm_cpu_ops)(&request));
1551 }
1552 
1553 /*
1554  * Default handler to create device node for CPU.
1555  * One reference count will be held on created device node.
1556  */
1557 static int
1558 mach_cpu_create_devinfo(cpu_t *cp, dev_info_t **dipp)
1559 {
1560 	int rv, circ;
1561 	dev_info_t *dip;
1562 	static kmutex_t cpu_node_lock;
1563 	static dev_info_t *cpu_nex_devi = NULL;
1564 
1565 	ASSERT(cp != NULL);
1566 	ASSERT(dipp != NULL);
1567 	*dipp = NULL;
1568 
1569 	if (cpu_nex_devi == NULL) {
1570 		mutex_enter(&cpu_node_lock);
1571 		/* First check whether cpus exists. */
1572 		cpu_nex_devi = ddi_find_devinfo("cpus", -1, 0);
1573 		/* Create cpus if it doesn't exist. */
1574 		if (cpu_nex_devi == NULL) {
1575 			ndi_devi_enter(ddi_root_node(), &circ);
1576 			rv = ndi_devi_alloc(ddi_root_node(), "cpus",
1577 			    (pnode_t)DEVI_SID_NODEID, &dip);
1578 			if (rv != NDI_SUCCESS) {
1579 				mutex_exit(&cpu_node_lock);
1580 				cmn_err(CE_CONT,
1581 				    "?failed to create cpu nexus device.\n");
1582 				return (PSM_FAILURE);
1583 			}
1584 			ASSERT(dip != NULL);
1585 			(void) ndi_devi_online(dip, 0);
1586 			ndi_devi_exit(ddi_root_node(), circ);
1587 			cpu_nex_devi = dip;
1588 		}
1589 		mutex_exit(&cpu_node_lock);
1590 	}
1591 
1592 	/*
1593 	 * create a child node for cpu identified as 'cpu_id'
1594 	 */
1595 	ndi_devi_enter(cpu_nex_devi, &circ);
1596 	dip = ddi_add_child(cpu_nex_devi, "cpu", DEVI_SID_NODEID, -1);
1597 	if (dip == NULL) {
1598 		cmn_err(CE_CONT,
1599 		    "?failed to create device node for cpu%d.\n", cp->cpu_id);
1600 		rv = PSM_FAILURE;
1601 	} else {
1602 		*dipp = dip;
1603 		(void) ndi_hold_devi(dip);
1604 		rv = PSM_SUCCESS;
1605 	}
1606 	ndi_devi_exit(cpu_nex_devi, circ);
1607 
1608 	return (rv);
1609 }
1610 
1611 /*
1612  * Create cpu device node in device tree and online it.
1613  * Return created dip with reference count held if requested.
1614  */
1615 int
1616 mach_cpu_create_device_node(struct cpu *cp, dev_info_t **dipp)
1617 {
1618 	int rv;
1619 	dev_info_t *dip = NULL;
1620 
1621 	ASSERT(psm_cpu_create_devinfo != NULL);
1622 	rv = psm_cpu_create_devinfo(cp, &dip);
1623 	if (rv == PSM_SUCCESS) {
1624 		cpuid_set_cpu_properties(dip, cp->cpu_id, cp->cpu_m.mcpu_cpi);
1625 		/* Recursively attach driver for parent nexus device. */
1626 		if (i_ddi_attach_node_hierarchy(ddi_get_parent(dip)) ==
1627 		    DDI_SUCCESS) {
1628 			/* Configure cpu itself and descendants. */
1629 			(void) ndi_devi_online(dip,
1630 			    NDI_ONLINE_ATTACH | NDI_CONFIG);
1631 		}
1632 		if (dipp != NULL) {
1633 			*dipp = dip;
1634 		} else {
1635 			(void) ndi_rele_devi(dip);
1636 		}
1637 	}
1638 
1639 	return (rv);
1640 }
1641 
1642 /*
1643  * The dipp contains one of following values on return:
1644  * - NULL if no device node found
1645  * - pointer to device node if found
1646  */
1647 int
1648 mach_cpu_get_device_node(struct cpu *cp, dev_info_t **dipp)
1649 {
1650 	*dipp = NULL;
1651 	if (psm_cpu_get_devinfo != NULL) {
1652 		if (psm_cpu_get_devinfo(cp, dipp) == PSM_SUCCESS) {
1653 			return (PSM_SUCCESS);
1654 		}
1655 	}
1656 
1657 	return (PSM_FAILURE);
1658 }
1659 
1660 /*ARGSUSED*/
1661 static int
1662 mach_translate_irq(dev_info_t *dip, int irqno)
1663 {
1664 	return (irqno);	/* default to NO translation */
1665 }
1666 
1667 static void
1668 mach_notify_error(int level, char *errmsg)
1669 {
1670 	/*
1671 	 * SL_FATAL is pass in once panicstr is set, deliver it
1672 	 * as CE_PANIC.  Also, translate SL_ codes back to CE_
1673 	 * codes for the psmi handler
1674 	 */
1675 	if (level & SL_FATAL)
1676 		(*notify_error)(CE_PANIC, errmsg);
1677 	else if (level & SL_WARN)
1678 		(*notify_error)(CE_WARN, errmsg);
1679 	else if (level & SL_NOTE)
1680 		(*notify_error)(CE_NOTE, errmsg);
1681 	else if (level & SL_CONSOLE)
1682 		(*notify_error)(CE_CONT, errmsg);
1683 }
1684 
1685 /*
1686  * It provides the default basic intr_ops interface for the new DDI
1687  * interrupt framework if the PSM doesn't have one.
1688  *
1689  * Input:
1690  * dip     - pointer to the dev_info structure of the requested device
1691  * hdlp    - pointer to the internal interrupt handle structure for the
1692  *	     requested interrupt
1693  * intr_op - opcode for this call
1694  * result  - pointer to the integer that will hold the result to be
1695  *	     passed back if return value is PSM_SUCCESS
1696  *
1697  * Output:
1698  * return value is either PSM_SUCCESS or PSM_FAILURE
1699  */
1700 static int
1701 mach_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1702     psm_intr_op_t intr_op, int *result)
1703 {
1704 	struct intrspec *ispec;
1705 
1706 	switch (intr_op) {
1707 	case PSM_INTR_OP_CHECK_MSI:
1708 		*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1709 		    DDI_INTR_TYPE_MSIX);
1710 		break;
1711 	case PSM_INTR_OP_ALLOC_VECTORS:
1712 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1713 			*result = 1;
1714 		else
1715 			*result = 0;
1716 		break;
1717 	case PSM_INTR_OP_FREE_VECTORS:
1718 		break;
1719 	case PSM_INTR_OP_NAVAIL_VECTORS:
1720 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1721 			*result = 1;
1722 		else
1723 			*result = 0;
1724 		break;
1725 	case PSM_INTR_OP_XLATE_VECTOR:
1726 		ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1727 		*result = psm_translate_irq(dip, ispec->intrspec_vec);
1728 		break;
1729 	case PSM_INTR_OP_GET_CAP:
1730 		*result = 0;
1731 		break;
1732 	case PSM_INTR_OP_GET_PENDING:
1733 	case PSM_INTR_OP_CLEAR_MASK:
1734 	case PSM_INTR_OP_SET_MASK:
1735 	case PSM_INTR_OP_GET_SHARED:
1736 	case PSM_INTR_OP_SET_PRI:
1737 	case PSM_INTR_OP_SET_CAP:
1738 	case PSM_INTR_OP_SET_CPU:
1739 	case PSM_INTR_OP_GET_INTR:
1740 	default:
1741 		return (PSM_FAILURE);
1742 	}
1743 	return (PSM_SUCCESS);
1744 }
1745 /*
1746  * Return 1 if CMT load balancing policies should be
1747  * implemented across instances of the specified hardware
1748  * sharing relationship.
1749  */
1750 int
1751 pg_cmt_load_bal_hw(pghw_type_t hw)
1752 {
1753 	if (hw == PGHW_IPIPE ||
1754 	    hw == PGHW_FPU ||
1755 	    hw == PGHW_PROCNODE ||
1756 	    hw == PGHW_CHIP)
1757 		return (1);
1758 	else
1759 		return (0);
1760 }
1761 /*
1762  * Return 1 if thread affinity polices should be implemented
1763  * for instances of the specifed hardware sharing relationship.
1764  */
1765 int
1766 pg_cmt_affinity_hw(pghw_type_t hw)
1767 {
1768 	if (hw == PGHW_CACHE)
1769 		return (1);
1770 	else
1771 		return (0);
1772 }
1773 
1774 /*
1775  * Return number of counter events requested to measure hardware capacity and
1776  * utilization and setup CPC requests for specified CPU as needed
1777  *
1778  * May return 0 when platform or processor specific code knows that no CPC
1779  * events should be programmed on this CPU or -1 when platform or processor
1780  * specific code doesn't know which counter events are best to use and common
1781  * code should decide for itself
1782  */
1783 int
1784 /* LINTED E_FUNC_ARG_UNUSED */
1785 cu_plat_cpc_init(cpu_t *cp, kcpc_request_list_t *reqs, int nreqs)
1786 {
1787 	const char	*impl_name;
1788 
1789 	/*
1790 	 * Return error if pcbe_ops not set
1791 	 */
1792 	if (pcbe_ops == NULL)
1793 		return (-1);
1794 
1795 	/*
1796 	 * Return that no CPC events should be programmed on hyperthreaded
1797 	 * Pentium 4 and return error for all other x86 processors to tell
1798 	 * common code to decide what counter events to program on those CPUs
1799 	 * for measuring hardware capacity and utilization
1800 	 */
1801 	impl_name = pcbe_ops->pcbe_impl_name();
1802 	if (impl_name != NULL && strcmp(impl_name, PCBE_IMPL_NAME_P4HT) == 0)
1803 		return (0);
1804 	else
1805 		return (-1);
1806 }
1807