xref: /illumos-gate/usr/src/uts/i86pc/os/mp_machdep.c (revision 4fceebdf03eeac0d7c58a4f70cc19b00a8c40a73)
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/pghw.h>
42 #include <sys/disp.h>
43 #include <sys/cpu.h>
44 #include <sys/archsystm.h>
45 #include <sys/mach_intr.h>
46 
47 #define	OFFSETOF(s, m)		(size_t)(&(((s *)0)->m))
48 
49 /*
50  *	Local function prototypes
51  */
52 static int mp_disable_intr(processorid_t cpun);
53 static void mp_enable_intr(processorid_t cpun);
54 static void mach_init();
55 static void mach_picinit();
56 static uint64_t mach_calchz(uint32_t pit_counter, uint64_t *processor_clks);
57 static int machhztomhz(uint64_t cpu_freq_hz);
58 static uint64_t mach_getcpufreq(void);
59 static void mach_fixcpufreq(void);
60 static int mach_clkinit(int, int *);
61 static void mach_smpinit(void);
62 static void mach_set_softintr(int ipl, struct av_softinfo *);
63 static void mach_cpu_start(int cpun);
64 static int mach_softlvl_to_vect(int ipl);
65 static void mach_get_platform(int owner);
66 static void mach_construct_info();
67 static int mach_translate_irq(dev_info_t *dip, int irqno);
68 static int mach_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
69     psm_intr_op_t, int *);
70 static timestruc_t mach_tod_get(void);
71 static void mach_tod_set(timestruc_t ts);
72 static void mach_notify_error(int level, char *errmsg);
73 static hrtime_t dummy_hrtime(void);
74 static void dummy_scalehrtime(hrtime_t *);
75 static void cpu_halt(void);
76 static void cpu_wakeup(cpu_t *, int);
77 /*
78  *	External reference functions
79  */
80 extern void return_instr();
81 extern timestruc_t (*todgetf)(void);
82 extern void (*todsetf)(timestruc_t);
83 extern long gmt_lag;
84 extern uint64_t freq_tsc(uint32_t *);
85 #if defined(__i386)
86 extern uint64_t freq_notsc(uint32_t *);
87 #endif
88 extern void pc_gethrestime(timestruc_t *);
89 extern int cpuid_get_coreid(cpu_t *);
90 extern int cpuid_get_chipid(cpu_t *);
91 
92 /*
93  *	PSM functions initialization
94  */
95 void (*psm_shutdownf)(int, int)	= return_instr;
96 void (*psm_preshutdownf)(int, int) = return_instr;
97 void (*psm_notifyf)(int)	= return_instr;
98 void (*psm_set_idle_cpuf)(int)	= return_instr;
99 void (*psm_unset_idle_cpuf)(int) = return_instr;
100 void (*psminitf)()		= mach_init;
101 void (*picinitf)() 		= return_instr;
102 int (*clkinitf)(int, int *) 	= (int (*)(int, int *))return_instr;
103 void (*cpu_startf)() 		= return_instr;
104 int (*ap_mlsetup)() 		= (int (*)(void))return_instr;
105 void (*send_dirintf)() 		= return_instr;
106 void (*setspl)(int)		= return_instr;
107 int (*addspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
108 int (*delspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
109 void (*setsoftint)(int, struct av_softinfo *)=
110 	(void (*)(int, struct av_softinfo *))return_instr;
111 int (*slvltovect)(int)		= (int (*)(int))return_instr;
112 int (*setlvl)(int, int *)	= (int (*)(int, int *))return_instr;
113 void (*setlvlx)(int, int)	= (void (*)(int, int))return_instr;
114 int (*psm_disable_intr)(int)	= mp_disable_intr;
115 void (*psm_enable_intr)(int)	= mp_enable_intr;
116 hrtime_t (*gethrtimef)(void)	= dummy_hrtime;
117 hrtime_t (*gethrtimeunscaledf)(void)	= dummy_hrtime;
118 void (*scalehrtimef)(hrtime_t *)	= dummy_scalehrtime;
119 int (*psm_translate_irq)(dev_info_t *, int) = mach_translate_irq;
120 void (*gethrestimef)(timestruc_t *) = pc_gethrestime;
121 int (*psm_todgetf)(todinfo_t *) = (int (*)(todinfo_t *))return_instr;
122 int (*psm_todsetf)(todinfo_t *) = (int (*)(todinfo_t *))return_instr;
123 void (*psm_notify_error)(int, char *) = (void (*)(int, char *))NULL;
124 int (*psm_get_clockirq)(int) = NULL;
125 int (*psm_get_ipivect)(int, int) = NULL;
126 
127 int (*psm_clkinit)(int) = NULL;
128 void (*psm_timer_reprogram)(hrtime_t) = NULL;
129 void (*psm_timer_enable)(void) = NULL;
130 void (*psm_timer_disable)(void) = NULL;
131 void (*psm_post_cyclic_setup)(void *arg) = NULL;
132 int (*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *, psm_intr_op_t,
133     int *) = mach_intr_ops;
134 
135 void (*notify_error)(int, char *) = (void (*)(int, char *))return_instr;
136 void (*hrtime_tick)(void)	= return_instr;
137 
138 int tsc_gethrtime_enable = 1;
139 int tsc_gethrtime_initted = 0;
140 
141 /*
142  * Local Static Data
143  */
144 static struct psm_ops mach_ops;
145 static struct psm_ops *mach_set[4] = {&mach_ops, NULL, NULL, NULL};
146 static ushort_t mach_ver[4] = {0, 0, 0, 0};
147 
148 /*
149  * If non-zero, idle cpus will "halted" when there's
150  * no work to do.
151  */
152 int	halt_idle_cpus = 1;
153 
154 #if defined(__amd64)
155 /*
156  * If non-zero, will use cr8 for interrupt priority masking
157  * We declare this here since install_spl is called from here
158  * (where this is checked).
159  */
160 int	intpri_use_cr8 = 0;
161 #endif	/* __amd64 */
162 
163 #ifdef	_SIMULATOR_SUPPORT
164 
165 int simulator_run = 0;	/* patch to non-zero if running under simics */
166 
167 #endif	/* _SIMULATOR_SUPPORT */
168 
169 
170 /*ARGSUSED*/
171 int
172 pg_plat_hw_shared(cpu_t *cp, pghw_type_t hw)
173 {
174 	switch (hw) {
175 	case PGHW_IPIPE:
176 		if (x86_feature & (X86_HTT)) {
177 			/*
178 			 * Hyper-threading is SMT
179 			 */
180 			return (1);
181 		} else {
182 			return (0);
183 		}
184 	case PGHW_CHIP:
185 		if (x86_feature & (X86_CMP|X86_HTT))
186 			return (1);
187 		else
188 			return (0);
189 	default:
190 		return (0);
191 	}
192 }
193 
194 /*
195  * Compare two CPUs and see if they have a pghw_type_t sharing relationship
196  * If pghw_type_t is an unsupported hardware type, then return -1
197  */
198 int
199 pg_plat_cpus_share(cpu_t *cpu_a, cpu_t *cpu_b, pghw_type_t hw)
200 {
201 	id_t pgp_a, pgp_b;
202 
203 	pgp_a = pg_plat_hw_instance_id(cpu_a, hw);
204 	pgp_b = pg_plat_hw_instance_id(cpu_b, hw);
205 
206 	if (pgp_a == -1 || pgp_b == -1)
207 		return (-1);
208 
209 	return (pgp_a == pgp_b);
210 }
211 
212 /*
213  * Return a physical instance identifier for known hardware sharing
214  * relationships
215  */
216 id_t
217 pg_plat_hw_instance_id(cpu_t *cpu, pghw_type_t hw)
218 {
219 	switch (hw) {
220 	case PGHW_IPIPE:
221 		return (cpuid_get_coreid(cpu));
222 	case PGHW_CHIP:
223 		return (cpuid_get_chipid(cpu));
224 	default:
225 		return (-1);
226 	}
227 }
228 
229 int
230 pg_plat_hw_level(pghw_type_t hw)
231 {
232 	int i;
233 	static pghw_type_t hw_hier[] = {
234 		PGHW_IPIPE,
235 		PGHW_CHIP,
236 		PGHW_NUM_COMPONENTS
237 	};
238 
239 	for (i = 0; hw_hier[i] != PGHW_NUM_COMPONENTS; i++) {
240 		if (hw_hier[i] == hw)
241 			return (i);
242 	}
243 	return (-1);
244 }
245 
246 id_t
247 pg_plat_get_core_id(cpu_t *cpu)
248 {
249 	return ((id_t)cpuid_get_coreid(cpu));
250 }
251 
252 void
253 cmp_set_nosteal_interval(void)
254 {
255 	/* Set the nosteal interval (used by disp_getbest()) to 100us */
256 	nosteal_nsec = 100000UL;
257 }
258 
259 /*
260  * Routine to ensure initial callers to hrtime gets 0 as return
261  */
262 static hrtime_t
263 dummy_hrtime(void)
264 {
265 	return (0);
266 }
267 
268 /* ARGSUSED */
269 static void
270 dummy_scalehrtime(hrtime_t *ticks)
271 {}
272 
273 /*
274  * Halt the present CPU until awoken via an interrupt
275  */
276 static void
277 cpu_halt(void)
278 {
279 	cpu_t		*cpup = CPU;
280 	processorid_t	cpun = cpup->cpu_id;
281 	cpupart_t	*cp = cpup->cpu_part;
282 	int		hset_update = 1;
283 
284 	/*
285 	 * If this CPU is online, and there's multiple CPUs
286 	 * in the system, then we should notate our halting
287 	 * by adding ourselves to the partition's halted CPU
288 	 * bitmap. This allows other CPUs to find/awaken us when
289 	 * work becomes available.
290 	 */
291 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
292 		hset_update = 0;
293 
294 	/*
295 	 * Add ourselves to the partition's halted CPUs bitmask
296 	 * and set our HALTED flag, if necessary.
297 	 *
298 	 * When a thread becomes runnable, it is placed on the queue
299 	 * and then the halted cpuset is checked to determine who
300 	 * (if anyone) should be awoken. We therefore need to first
301 	 * add ourselves to the halted cpuset, and and then check if there
302 	 * is any work available.
303 	 *
304 	 * Note that memory barriers after updating the HALTED flag
305 	 * are not necessary since an atomic operation (updating the bitmap)
306 	 * immediately follows. On x86 the atomic operation acts as a
307 	 * memory barrier for the update of cpu_disp_flags.
308 	 */
309 	if (hset_update) {
310 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
311 		CPUSET_ATOMIC_ADD(cp->cp_mach->mc_haltset, cpun);
312 	}
313 
314 	/*
315 	 * Check to make sure there's really nothing to do.
316 	 * Work destined for this CPU may become available after
317 	 * this check. We'll be notified through the clearing of our
318 	 * bit in the halted CPU bitmask, and a poke.
319 	 */
320 	if (disp_anywork()) {
321 		if (hset_update) {
322 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
323 			CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
324 		}
325 		return;
326 	}
327 
328 	/*
329 	 * We're on our way to being halted.
330 	 *
331 	 * Disable interrupts now, so that we'll awaken immediately
332 	 * after halting if someone tries to poke us between now and
333 	 * the time we actually halt.
334 	 *
335 	 * We check for the presence of our bit after disabling interrupts.
336 	 * If it's cleared, we'll return. If the bit is cleared after
337 	 * we check then the poke will pop us out of the halted state.
338 	 *
339 	 * This means that the ordering of the poke and the clearing
340 	 * of the bit by cpu_wakeup is important.
341 	 * cpu_wakeup() must clear, then poke.
342 	 * cpu_halt() must disable interrupts, then check for the bit.
343 	 */
344 	cli();
345 
346 	if (hset_update && !CPU_IN_SET(cp->cp_mach->mc_haltset, cpun)) {
347 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
348 		sti();
349 		return;
350 	}
351 
352 	/*
353 	 * The check for anything locally runnable is here for performance
354 	 * and isn't needed for correctness. disp_nrunnable ought to be
355 	 * in our cache still, so it's inexpensive to check, and if there
356 	 * is anything runnable we won't have to wait for the poke.
357 	 */
358 	if (cpup->cpu_disp->disp_nrunnable != 0) {
359 		if (hset_update) {
360 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
361 			CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
362 		}
363 		sti();
364 		return;
365 	}
366 
367 	/*
368 	 * Call the halt sequence:
369 	 * sti
370 	 * hlt
371 	 */
372 	i86_halt();
373 
374 	/*
375 	 * We're no longer halted
376 	 */
377 	if (hset_update) {
378 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
379 		CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
380 	}
381 }
382 
383 
384 /*
385  * If "cpu" is halted, then wake it up clearing its halted bit in advance.
386  * Otherwise, see if other CPUs in the cpu partition are halted and need to
387  * be woken up so that they can steal the thread we placed on this CPU.
388  * This function is only used on MP systems.
389  */
390 static void
391 cpu_wakeup(cpu_t *cpu, int bound)
392 {
393 	uint_t		cpu_found;
394 	int		result;
395 	cpupart_t	*cp;
396 
397 	cp = cpu->cpu_part;
398 	if (CPU_IN_SET(cp->cp_mach->mc_haltset, cpu->cpu_id)) {
399 		/*
400 		 * Clear the halted bit for that CPU since it will be
401 		 * poked in a moment.
402 		 */
403 		CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpu->cpu_id);
404 		/*
405 		 * We may find the current CPU present in the halted cpuset
406 		 * if we're in the context of an interrupt that occurred
407 		 * before we had a chance to clear our bit in cpu_halt().
408 		 * Poking ourself is obviously unnecessary, since if
409 		 * we're here, we're not halted.
410 		 */
411 		if (cpu != CPU)
412 			poke_cpu(cpu->cpu_id);
413 		return;
414 	} else {
415 		/*
416 		 * This cpu isn't halted, but it's idle or undergoing a
417 		 * context switch. No need to awaken anyone else.
418 		 */
419 		if (cpu->cpu_thread == cpu->cpu_idle_thread ||
420 		    cpu->cpu_disp_flags & CPU_DISP_DONTSTEAL)
421 			return;
422 	}
423 
424 	/*
425 	 * No need to wake up other CPUs if the thread we just enqueued
426 	 * is bound.
427 	 */
428 	if (bound)
429 		return;
430 
431 
432 	/*
433 	 * See if there's any other halted CPUs. If there are, then
434 	 * select one, and awaken it.
435 	 * It's possible that after we find a CPU, somebody else
436 	 * will awaken it before we get the chance.
437 	 * In that case, look again.
438 	 */
439 	do {
440 		CPUSET_FIND(cp->cp_mach->mc_haltset, cpu_found);
441 		if (cpu_found == CPUSET_NOTINSET)
442 			return;
443 
444 		ASSERT(cpu_found >= 0 && cpu_found < NCPU);
445 		CPUSET_ATOMIC_XDEL(cp->cp_mach->mc_haltset, cpu_found, result);
446 	} while (result < 0);
447 
448 	if (cpu_found != CPU->cpu_id)
449 		poke_cpu(cpu_found);
450 }
451 
452 static int
453 mp_disable_intr(int cpun)
454 {
455 	/*
456 	 * switch to the offline cpu
457 	 */
458 	affinity_set(cpun);
459 	/*
460 	 * raise ipl to just below cross call
461 	 */
462 	splx(XC_MED_PIL-1);
463 	/*
464 	 *	set base spl to prevent the next swtch to idle from
465 	 *	lowering back to ipl 0
466 	 */
467 	CPU->cpu_intr_actv |= (1 << (XC_MED_PIL-1));
468 	set_base_spl();
469 	affinity_clear();
470 	return (DDI_SUCCESS);
471 }
472 
473 static void
474 mp_enable_intr(int cpun)
475 {
476 	/*
477 	 * switch to the online cpu
478 	 */
479 	affinity_set(cpun);
480 	/*
481 	 * clear the interrupt active mask
482 	 */
483 	CPU->cpu_intr_actv &= ~(1 << (XC_MED_PIL-1));
484 	set_base_spl();
485 	(void) spl0();
486 	affinity_clear();
487 }
488 
489 static void
490 mach_get_platform(int owner)
491 {
492 	void		**srv_opsp;
493 	void		**clt_opsp;
494 	int		i;
495 	int		total_ops;
496 
497 	/* fix up psm ops */
498 	srv_opsp = (void **)mach_set[0];
499 	clt_opsp = (void **)mach_set[owner];
500 	if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01)
501 		total_ops = sizeof (struct psm_ops_ver01) /
502 				sizeof (void (*)(void));
503 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_1)
504 		/* no psm_notify_func */
505 		total_ops = OFFSETOF(struct psm_ops, psm_notify_func) /
506 		    sizeof (void (*)(void));
507 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_2)
508 		/* no psm_timer funcs */
509 		total_ops = OFFSETOF(struct psm_ops, psm_timer_reprogram) /
510 		    sizeof (void (*)(void));
511 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_3)
512 		/* no psm_preshutdown function */
513 		total_ops = OFFSETOF(struct psm_ops, psm_preshutdown) /
514 		    sizeof (void (*)(void));
515 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_4)
516 		/* no psm_preshutdown function */
517 		total_ops = OFFSETOF(struct psm_ops, psm_intr_ops) /
518 		    sizeof (void (*)(void));
519 	else
520 		total_ops = sizeof (struct psm_ops) / sizeof (void (*)(void));
521 
522 	/*
523 	 * Save the version of the PSM module, in case we need to
524 	 * bahave differently based on version.
525 	 */
526 	mach_ver[0] = mach_ver[owner];
527 
528 	for (i = 0; i < total_ops; i++)
529 		if (clt_opsp[i] != NULL)
530 			srv_opsp[i] = clt_opsp[i];
531 }
532 
533 static void
534 mach_construct_info()
535 {
536 	register struct psm_sw *swp;
537 	int	mach_cnt[PSM_OWN_OVERRIDE+1] = {0};
538 	int	conflict_owner = 0;
539 
540 	if (psmsw->psw_forw == psmsw)
541 		panic("No valid PSM modules found");
542 	mutex_enter(&psmsw_lock);
543 	for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
544 		if (!(swp->psw_flag & PSM_MOD_IDENTIFY))
545 			continue;
546 		mach_set[swp->psw_infop->p_owner] = swp->psw_infop->p_ops;
547 		mach_ver[swp->psw_infop->p_owner] = swp->psw_infop->p_version;
548 		mach_cnt[swp->psw_infop->p_owner]++;
549 	}
550 	mutex_exit(&psmsw_lock);
551 
552 	mach_get_platform(PSM_OWN_SYS_DEFAULT);
553 
554 	/* check to see are there any conflicts */
555 	if (mach_cnt[PSM_OWN_EXCLUSIVE] > 1)
556 		conflict_owner = PSM_OWN_EXCLUSIVE;
557 	if (mach_cnt[PSM_OWN_OVERRIDE] > 1)
558 		conflict_owner = PSM_OWN_OVERRIDE;
559 	if (conflict_owner) {
560 		/* remove all psm modules except uppc */
561 		cmn_err(CE_WARN,
562 			"Conflicts detected on the following PSM modules:");
563 		mutex_enter(&psmsw_lock);
564 		for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
565 			if (swp->psw_infop->p_owner == conflict_owner)
566 				cmn_err(CE_WARN, "%s ",
567 					swp->psw_infop->p_mach_idstring);
568 		}
569 		mutex_exit(&psmsw_lock);
570 		cmn_err(CE_WARN,
571 			"Setting the system back to SINGLE processor mode!");
572 		cmn_err(CE_WARN,
573 		    "Please edit /etc/mach to remove the invalid PSM module.");
574 		return;
575 	}
576 
577 	if (mach_set[PSM_OWN_EXCLUSIVE])
578 		mach_get_platform(PSM_OWN_EXCLUSIVE);
579 
580 	if (mach_set[PSM_OWN_OVERRIDE])
581 		mach_get_platform(PSM_OWN_OVERRIDE);
582 }
583 
584 static void
585 mach_init()
586 {
587 	register struct psm_ops  *pops;
588 
589 	mach_construct_info();
590 
591 	pops = mach_set[0];
592 
593 	/* register the interrupt and clock initialization rotuines */
594 	picinitf = mach_picinit;
595 	clkinitf = mach_clkinit;
596 	psm_get_clockirq = pops->psm_get_clockirq;
597 
598 	/* register the interrupt setup code */
599 	slvltovect = mach_softlvl_to_vect;
600 	addspl	= pops->psm_addspl;
601 	delspl	= pops->psm_delspl;
602 
603 	if (pops->psm_translate_irq)
604 		psm_translate_irq = pops->psm_translate_irq;
605 	if (pops->psm_intr_ops)
606 		psm_intr_ops = pops->psm_intr_ops;
607 	if (pops->psm_tod_get) {
608 		todgetf = mach_tod_get;
609 		psm_todgetf = pops->psm_tod_get;
610 	}
611 	if (pops->psm_tod_set) {
612 		todsetf = mach_tod_set;
613 		psm_todsetf = pops->psm_tod_set;
614 	}
615 	if (pops->psm_notify_error) {
616 		psm_notify_error = mach_notify_error;
617 		notify_error = pops->psm_notify_error;
618 	}
619 
620 	(*pops->psm_softinit)();
621 
622 	/*
623 	 * Initialize the dispatcher's function hooks
624 	 * to enable CPU halting when idle
625 	 */
626 #if defined(_SIMULATOR_SUPPORT)
627 	if (halt_idle_cpus && !simulator_run)
628 		idle_cpu = cpu_halt;
629 #else
630 	if (halt_idle_cpus)
631 		idle_cpu = cpu_halt;
632 #endif	/* _SIMULATOR_SUPPORT */
633 
634 	mach_smpinit();
635 }
636 
637 static void
638 mach_smpinit(void)
639 {
640 	struct psm_ops  *pops;
641 	processorid_t cpu_id;
642 	int cnt;
643 	cpuset_t cpumask;
644 
645 	pops = mach_set[0];
646 
647 	cpu_id = -1;
648 	cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
649 	for (cnt = 0, CPUSET_ZERO(cpumask); cpu_id != -1; cnt++) {
650 		CPUSET_ADD(cpumask, cpu_id);
651 		cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
652 	}
653 
654 	mp_cpus = cpumask;
655 
656 	/* MP related routines */
657 	cpu_startf = mach_cpu_start;
658 	ap_mlsetup = pops->psm_post_cpu_start;
659 	send_dirintf = pops->psm_send_ipi;
660 
661 	/* optional MP related routines */
662 	if (pops->psm_shutdown)
663 		psm_shutdownf = pops->psm_shutdown;
664 	if (pops->psm_preshutdown)
665 		psm_preshutdownf = pops->psm_preshutdown;
666 	if (pops->psm_notify_func)
667 		psm_notifyf = pops->psm_notify_func;
668 	if (pops->psm_set_idlecpu)
669 		psm_set_idle_cpuf = pops->psm_set_idlecpu;
670 	if (pops->psm_unset_idlecpu)
671 		psm_unset_idle_cpuf = pops->psm_unset_idlecpu;
672 
673 	psm_clkinit = pops->psm_clkinit;
674 
675 	if (pops->psm_timer_reprogram)
676 		psm_timer_reprogram = pops->psm_timer_reprogram;
677 
678 	if (pops->psm_timer_enable)
679 		psm_timer_enable = pops->psm_timer_enable;
680 
681 	if (pops->psm_timer_disable)
682 		psm_timer_disable = pops->psm_timer_disable;
683 
684 	if (pops->psm_post_cyclic_setup)
685 		psm_post_cyclic_setup = pops->psm_post_cyclic_setup;
686 
687 	/* check for multiple cpu's */
688 	if (cnt < 2)
689 		return;
690 
691 	/* check for MP platforms */
692 	if (pops->psm_cpu_start == NULL)
693 		return;
694 
695 	/*
696 	 * Set the dispatcher hook to enable cpu "wake up"
697 	 * when a thread becomes runnable.
698 	 */
699 #if defined(_SIMULATOR_SUPPORT)
700 	if (halt_idle_cpus && !simulator_run) {
701 		disp_enq_thread = cpu_wakeup;
702 	}
703 #else
704 	if (halt_idle_cpus) {
705 		disp_enq_thread = cpu_wakeup;
706 	}
707 #endif	/* _SIMULATOR_SUPPORT */
708 
709 	if (pops->psm_disable_intr)
710 		psm_disable_intr = pops->psm_disable_intr;
711 	if (pops->psm_enable_intr)
712 		psm_enable_intr  = pops->psm_enable_intr;
713 
714 	psm_get_ipivect = pops->psm_get_ipivect;
715 
716 	(void) add_avintr((void *)NULL, XC_HI_PIL, xc_serv, "xc_hi_intr",
717 		(*pops->psm_get_ipivect)(XC_HI_PIL, PSM_INTR_IPI_HI),
718 		(caddr_t)X_CALL_HIPRI, NULL, NULL, NULL);
719 	(void) add_avintr((void *)NULL, XC_MED_PIL, xc_serv, "xc_med_intr",
720 		(*pops->psm_get_ipivect)(XC_MED_PIL, PSM_INTR_IPI_LO),
721 		(caddr_t)X_CALL_MEDPRI, NULL, NULL, NULL);
722 
723 	(void) (*pops->psm_get_ipivect)(XC_CPUPOKE_PIL, PSM_INTR_POKE);
724 }
725 
726 static void
727 mach_picinit()
728 {
729 	struct psm_ops  *pops;
730 	extern void install_spl(void);	/* XXX: belongs in a header file */
731 #if defined(__amd64) && defined(DEBUG)
732 	extern void *spl_patch, *slow_spl, *setsplhi_patch, *slow_setsplhi;
733 #endif
734 
735 	pops = mach_set[0];
736 
737 	/* register the interrupt handlers */
738 	setlvl = pops->psm_intr_enter;
739 	setlvlx = pops->psm_intr_exit;
740 
741 	/* initialize the interrupt hardware */
742 	(*pops->psm_picinit)();
743 
744 	/* set interrupt mask for current ipl */
745 	setspl = pops->psm_setspl;
746 	setspl(CPU->cpu_pri);
747 
748 	/* Install proper spl routine now that we can Program the PIC   */
749 #if defined(__amd64)
750 	/*
751 	 * It would be better if we could check this at compile time
752 	 */
753 	ASSERT(((uintptr_t)&slow_setsplhi - (uintptr_t)&setsplhi_patch < 128) &&
754 		((uintptr_t)&slow_spl - (uintptr_t)&spl_patch < 128));
755 #endif
756 	install_spl();
757 }
758 
759 uint_t	cpu_freq;	/* MHz */
760 uint64_t cpu_freq_hz;	/* measured (in hertz) */
761 
762 #define	MEGA_HZ		1000000
763 
764 static uint64_t
765 mach_calchz(uint32_t pit_counter, uint64_t *processor_clks)
766 {
767 	uint64_t cpu_hz;
768 
769 	if ((pit_counter == 0) || (*processor_clks == 0) ||
770 	    (*processor_clks > (((uint64_t)-1) / PIT_HZ)))
771 		return (0);
772 
773 	cpu_hz = ((uint64_t)PIT_HZ * *processor_clks) / pit_counter;
774 
775 	return (cpu_hz);
776 }
777 
778 static uint64_t
779 mach_getcpufreq(void)
780 {
781 	uint32_t pit_counter;
782 	uint64_t processor_clks;
783 
784 	if (x86_feature & X86_TSC) {
785 		/*
786 		 * We have a TSC. freq_tsc() knows how to measure the number
787 		 * of clock cycles sampled against the PIT.
788 		 */
789 		processor_clks = freq_tsc(&pit_counter);
790 		return (mach_calchz(pit_counter, &processor_clks));
791 	} else if (x86_vendor == X86_VENDOR_Cyrix || x86_type == X86_TYPE_P5) {
792 #if defined(__amd64)
793 		panic("mach_getcpufreq: no TSC!");
794 #elif defined(__i386)
795 		/*
796 		 * We are a Cyrix based on a 6x86 core or an Intel Pentium
797 		 * for which freq_notsc() knows how to measure the number of
798 		 * elapsed clock cycles sampled against the PIT
799 		 */
800 		processor_clks = freq_notsc(&pit_counter);
801 		return (mach_calchz(pit_counter, &processor_clks));
802 #endif	/* __i386 */
803 	}
804 
805 	/* We do not know how to calculate cpu frequency for this cpu. */
806 	return (0);
807 }
808 
809 /*
810  * If the clock speed of a cpu is found to be reported incorrectly, do not add
811  * to this array, instead improve the accuracy of the algorithm that determines
812  * the clock speed of the processor or extend the implementation to support the
813  * vendor as appropriate. This is here only to support adjusting the speed on
814  * older slower processors that mach_fixcpufreq() would not be able to account
815  * for otherwise.
816  */
817 static int x86_cpu_freq[] = { 60, 75, 80, 90, 120, 160, 166, 175, 180, 233 };
818 
819 /*
820  * On fast processors the clock frequency that is measured may be off by
821  * a few MHz from the value printed on the part. This is a combination of
822  * the factors that for such fast parts being off by this much is within
823  * the tolerances for manufacture and because of the difficulties in the
824  * measurement that can lead to small error. This function uses some
825  * heuristics in order to tweak the value that was measured to match what
826  * is most likely printed on the part.
827  *
828  * Some examples:
829  * 	AMD Athlon 1000 mhz measured as 998 mhz
830  * 	Intel Pentium III Xeon 733 mhz measured as 731 mhz
831  * 	Intel Pentium IV 1500 mhz measured as 1495mhz
832  *
833  * If in the future this function is no longer sufficient to correct
834  * for the error in the measurement, then the algorithm used to perform
835  * the measurement will have to be improved in order to increase accuracy
836  * rather than adding horrible and questionable kludges here.
837  *
838  * This is called after the cyclics subsystem because of the potential
839  * that the heuristics within may give a worse estimate of the clock
840  * frequency than the value that was measured.
841  */
842 static void
843 mach_fixcpufreq(void)
844 {
845 	uint32_t freq, mul, near66, delta66, near50, delta50, fixed, delta, i;
846 
847 	freq = (uint32_t)cpu_freq;
848 
849 	/*
850 	 * Find the nearest integer multiple of 200/3 (about 66) MHz to the
851 	 * measured speed taking into account that the 667 MHz parts were
852 	 * the first to round-up.
853 	 */
854 	mul = (uint32_t)((3 * (uint64_t)freq + 100) / 200);
855 	near66 = (uint32_t)((200 * (uint64_t)mul + ((mul >= 10) ? 1 : 0)) / 3);
856 	delta66 = (near66 > freq) ? (near66 - freq) : (freq - near66);
857 
858 	/* Find the nearest integer multiple of 50 MHz to the measured speed */
859 	mul = (freq + 25) / 50;
860 	near50 = mul * 50;
861 	delta50 = (near50 > freq) ? (near50 - freq) : (freq - near50);
862 
863 	/* Find the closer of the two */
864 	if (delta66 < delta50) {
865 		fixed = near66;
866 		delta = delta66;
867 	} else {
868 		fixed = near50;
869 		delta = delta50;
870 	}
871 
872 	if (fixed > INT_MAX)
873 		return;
874 
875 	/*
876 	 * Some older parts have a core clock frequency that is not an
877 	 * integral multiple of 50 or 66 MHz. Check if one of the old
878 	 * clock frequencies is closer to the measured value than any
879 	 * of the integral multiples of 50 an 66, and if so set fixed
880 	 * and delta appropriately to represent the closest value.
881 	 */
882 	i = sizeof (x86_cpu_freq) / sizeof (int);
883 	while (i > 0) {
884 		i--;
885 
886 		if (x86_cpu_freq[i] <= freq) {
887 			mul = freq - x86_cpu_freq[i];
888 
889 			if (mul < delta) {
890 				fixed = x86_cpu_freq[i];
891 				delta = mul;
892 			}
893 
894 			break;
895 		}
896 
897 		mul = x86_cpu_freq[i] - freq;
898 
899 		if (mul < delta) {
900 			fixed = x86_cpu_freq[i];
901 			delta = mul;
902 		}
903 	}
904 
905 	/*
906 	 * Set a reasonable maximum for how much to correct the measured
907 	 * result by. This check is here to prevent the adjustment made
908 	 * by this function from being more harm than good. It is entirely
909 	 * possible that in the future parts will be made that are not
910 	 * integral multiples of 66 or 50 in clock frequency or that
911 	 * someone may overclock a part to some odd frequency. If the
912 	 * measured value is farther from the corrected value than
913 	 * allowed, then assume the corrected value is in error and use
914 	 * the measured value.
915 	 */
916 	if (6 < delta)
917 		return;
918 
919 	cpu_freq = (int)fixed;
920 }
921 
922 
923 static int
924 machhztomhz(uint64_t cpu_freq_hz)
925 {
926 	uint64_t cpu_mhz;
927 
928 	/* Round to nearest MHZ */
929 	cpu_mhz = (cpu_freq_hz + (MEGA_HZ / 2)) / MEGA_HZ;
930 
931 	if (cpu_mhz > INT_MAX)
932 		return (0);
933 
934 	return ((int)cpu_mhz);
935 
936 }
937 
938 
939 static int
940 mach_clkinit(int preferred_mode, int *set_mode)
941 {
942 	register struct psm_ops  *pops;
943 	int resolution;
944 
945 	pops = mach_set[0];
946 
947 #ifdef	_SIMULATOR_SUPPORT
948 	if (!simulator_run)
949 		cpu_freq_hz = mach_getcpufreq();
950 	else
951 		cpu_freq_hz = 40000000; /* use 40 Mhz (hack for simulator) */
952 #else
953 	cpu_freq_hz = mach_getcpufreq();
954 #endif	/* _SIMULATOR_SUPPORT */
955 
956 	cpu_freq = machhztomhz(cpu_freq_hz);
957 
958 	if (!(x86_feature & X86_TSC) || (cpu_freq == 0))
959 		tsc_gethrtime_enable = 0;
960 
961 	if (tsc_gethrtime_enable) {
962 		tsc_hrtimeinit(cpu_freq_hz);
963 		gethrtimef = tsc_gethrtime;
964 		gethrtimeunscaledf = tsc_gethrtimeunscaled;
965 		scalehrtimef = tsc_scalehrtime;
966 		hrtime_tick = tsc_tick;
967 		tsc_gethrtime_initted = 1;
968 	} else {
969 		if (pops->psm_hrtimeinit)
970 			(*pops->psm_hrtimeinit)();
971 		gethrtimef = pops->psm_gethrtime;
972 		gethrtimeunscaledf = gethrtimef;
973 		/* scalehrtimef will remain dummy */
974 	}
975 
976 	mach_fixcpufreq();
977 
978 	if (mach_ver[0] >= PSM_INFO_VER01_3) {
979 		if ((preferred_mode == TIMER_ONESHOT) &&
980 		    (tsc_gethrtime_enable)) {
981 
982 			resolution = (*pops->psm_clkinit)(0);
983 			if (resolution != 0)  {
984 				*set_mode = TIMER_ONESHOT;
985 				return (resolution);
986 			}
987 
988 		}
989 
990 		/*
991 		 * either periodic mode was requested or could not set to
992 		 * one-shot mode
993 		 */
994 		resolution = (*pops->psm_clkinit)(hz);
995 		/*
996 		 * psm should be able to do periodic, so we do not check
997 		 * for return value of psm_clkinit here.
998 		 */
999 		*set_mode = TIMER_PERIODIC;
1000 		return (resolution);
1001 	} else {
1002 		/*
1003 		 * PSMI interface prior to PSMI_3 does not define a return
1004 		 * value for psm_clkinit, so the return value is ignored.
1005 		 */
1006 		(void) (*pops->psm_clkinit)(hz);
1007 		*set_mode = TIMER_PERIODIC;
1008 		return (nsec_per_tick);
1009 	}
1010 }
1011 
1012 /*ARGSUSED*/
1013 static void
1014 mach_psm_set_softintr(int ipl, struct av_softinfo *pending)
1015 {
1016 	register struct psm_ops  *pops;
1017 
1018 	/* invoke hardware interrupt					*/
1019 	pops = mach_set[0];
1020 	(*pops->psm_set_softintr)(ipl);
1021 }
1022 
1023 static int
1024 mach_softlvl_to_vect(register int ipl)
1025 {
1026 	register int softvect;
1027 	register struct psm_ops  *pops;
1028 
1029 	pops = mach_set[0];
1030 
1031 	/* check for null handler for set soft interrupt call		*/
1032 	if (pops->psm_set_softintr == NULL) {
1033 		setsoftint = av_set_softint_pending;
1034 		return (PSM_SV_SOFTWARE);
1035 	}
1036 
1037 	softvect = (*pops->psm_softlvl_to_irq)(ipl);
1038 	/* check for hardware scheme					*/
1039 	if (softvect > PSM_SV_SOFTWARE) {
1040 		setsoftint = mach_psm_set_softintr;
1041 		return (softvect);
1042 	}
1043 
1044 	if (softvect == PSM_SV_SOFTWARE)
1045 		setsoftint = av_set_softint_pending;
1046 	else	/* hardware and software mixed scheme			*/
1047 		setsoftint = mach_set_softintr;
1048 
1049 	return (PSM_SV_SOFTWARE);
1050 }
1051 
1052 static void
1053 mach_set_softintr(register int ipl, struct av_softinfo *pending)
1054 {
1055 	register struct psm_ops  *pops;
1056 
1057 	/* set software pending bits					*/
1058 	av_set_softint_pending(ipl, pending);
1059 
1060 	/*	check if dosoftint will be called at the end of intr	*/
1061 	if (CPU_ON_INTR(CPU) || (curthread->t_intr))
1062 		return;
1063 
1064 	/* invoke hardware interrupt					*/
1065 	pops = mach_set[0];
1066 	(*pops->psm_set_softintr)(ipl);
1067 }
1068 
1069 static void
1070 mach_cpu_start(register int cpun)
1071 {
1072 	struct psm_ops  *pops;
1073 
1074 	pops = mach_set[0];
1075 
1076 	(*pops->psm_cpu_start)(cpun, rm_platter_va);
1077 }
1078 
1079 /*ARGSUSED*/
1080 static int
1081 mach_translate_irq(dev_info_t *dip, int irqno)
1082 {
1083 	return (irqno);	/* default to NO translation */
1084 }
1085 
1086 static timestruc_t
1087 mach_tod_get(void)
1088 {
1089 	timestruc_t ts;
1090 	todinfo_t tod;
1091 	static int mach_range_warn = 1;	/* warn only once */
1092 
1093 	ASSERT(MUTEX_HELD(&tod_lock));
1094 
1095 	/* The year returned from is the last 2 digit only */
1096 	if ((*psm_todgetf)(&tod)) {
1097 		ts.tv_sec = 0;
1098 		ts.tv_nsec = 0;
1099 		tod_fault_reset();
1100 		return (ts);
1101 	}
1102 
1103 	/* assume that we wrap the rtc year back to zero at 2000 */
1104 	if (tod.tod_year < 69) {
1105 		if (mach_range_warn && tod.tod_year > 38) {
1106 			cmn_err(CE_WARN, "hardware real-time clock is out "
1107 				"of range -- time needs to be reset");
1108 			mach_range_warn = 0;
1109 		}
1110 		tod.tod_year += 100;
1111 	}
1112 
1113 	/* tod_to_utc uses 1900 as base for the year */
1114 	ts.tv_sec = tod_to_utc(tod) + gmt_lag;
1115 	ts.tv_nsec = 0;
1116 
1117 	return (ts);
1118 }
1119 
1120 static void
1121 mach_tod_set(timestruc_t ts)
1122 {
1123 	todinfo_t tod = utc_to_tod(ts.tv_sec - gmt_lag);
1124 
1125 	ASSERT(MUTEX_HELD(&tod_lock));
1126 
1127 	if (tod.tod_year >= 100)
1128 		tod.tod_year -= 100;
1129 
1130 	(*psm_todsetf)(&tod);
1131 }
1132 
1133 static void
1134 mach_notify_error(int level, char *errmsg)
1135 {
1136 	/*
1137 	 * SL_FATAL is pass in once panicstr is set, deliver it
1138 	 * as CE_PANIC.  Also, translate SL_ codes back to CE_
1139 	 * codes for the psmi handler
1140 	 */
1141 	if (level & SL_FATAL)
1142 		(*notify_error)(CE_PANIC, errmsg);
1143 	else if (level & SL_WARN)
1144 		(*notify_error)(CE_WARN, errmsg);
1145 	else if (level & SL_NOTE)
1146 		(*notify_error)(CE_NOTE, errmsg);
1147 	else if (level & SL_CONSOLE)
1148 		(*notify_error)(CE_CONT, errmsg);
1149 }
1150 
1151 /*
1152  * It provides the default basic intr_ops interface for the new DDI
1153  * interrupt framework if the PSM doesn't have one.
1154  *
1155  * Input:
1156  * dip     - pointer to the dev_info structure of the requested device
1157  * hdlp    - pointer to the internal interrupt handle structure for the
1158  *	     requested interrupt
1159  * intr_op - opcode for this call
1160  * result  - pointer to the integer that will hold the result to be
1161  *	     passed back if return value is PSM_SUCCESS
1162  *
1163  * Output:
1164  * return value is either PSM_SUCCESS or PSM_FAILURE
1165  */
1166 static int
1167 mach_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1168     psm_intr_op_t intr_op, int *result)
1169 {
1170 	struct intrspec *ispec;
1171 
1172 	switch (intr_op) {
1173 	case PSM_INTR_OP_CHECK_MSI:
1174 		*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1175 			    DDI_INTR_TYPE_MSIX);
1176 		break;
1177 	case PSM_INTR_OP_ALLOC_VECTORS:
1178 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1179 			*result = 1;
1180 		else
1181 			*result = 0;
1182 		break;
1183 	case PSM_INTR_OP_FREE_VECTORS:
1184 		break;
1185 	case PSM_INTR_OP_NAVAIL_VECTORS:
1186 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1187 			*result = 1;
1188 		else
1189 			*result = 0;
1190 		break;
1191 	case PSM_INTR_OP_XLATE_VECTOR:
1192 		ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1193 		*result = psm_translate_irq(dip, ispec->intrspec_vec);
1194 		break;
1195 	case PSM_INTR_OP_GET_CAP:
1196 		*result = 0;
1197 		break;
1198 	case PSM_INTR_OP_GET_PENDING:
1199 	case PSM_INTR_OP_CLEAR_MASK:
1200 	case PSM_INTR_OP_SET_MASK:
1201 	case PSM_INTR_OP_GET_SHARED:
1202 	case PSM_INTR_OP_SET_PRI:
1203 	case PSM_INTR_OP_SET_CAP:
1204 	case PSM_INTR_OP_SET_CPU:
1205 	case PSM_INTR_OP_GET_INTR:
1206 	default:
1207 		return (PSM_FAILURE);
1208 	}
1209 	return (PSM_SUCCESS);
1210 }
1211