xref: /titanic_41/usr/src/uts/i86pc/os/mp_machdep.c (revision 6f45ec7b0b964c3be967c4880e8867ac1e7763a5)
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/archsystm.h>
44 #include <sys/machsystm.h>
45 #include <sys/param.h>
46 #include <sys/promif.h>
47 #include <sys/mach_intr.h>
48 
49 #define	OFFSETOF(s, m)		(size_t)(&(((s *)0)->m))
50 
51 /*
52  *	Local function prototypes
53  */
54 static int mp_disable_intr(processorid_t cpun);
55 static void mp_enable_intr(processorid_t cpun);
56 static void mach_init();
57 static void mach_picinit();
58 static uint64_t mach_calchz(uint32_t pit_counter, uint64_t *processor_clks);
59 static int machhztomhz(uint64_t cpu_freq_hz);
60 static uint64_t mach_getcpufreq(void);
61 static void mach_fixcpufreq(void);
62 static int mach_clkinit(int, int *);
63 static void mach_smpinit(void);
64 static void mach_set_softintr(int ipl, struct av_softinfo *);
65 static int mach_softlvl_to_vect(int ipl);
66 static void mach_get_platform(int owner);
67 static void mach_construct_info();
68 static int mach_translate_irq(dev_info_t *dip, int irqno);
69 static int mach_intr_ops(dev_info_t *, ddi_intr_handle_impl_t *,
70     psm_intr_op_t, int *);
71 static void mach_notify_error(int level, char *errmsg);
72 static hrtime_t dummy_hrtime(void);
73 static void dummy_scalehrtime(hrtime_t *);
74 static void cpu_idle(void);
75 static void cpu_wakeup(cpu_t *, int);
76 /*
77  *	External reference functions
78  */
79 extern void return_instr();
80 extern uint64_t freq_tsc(uint32_t *);
81 #if defined(__i386)
82 extern uint64_t freq_notsc(uint32_t *);
83 #endif
84 extern void pc_gethrestime(timestruc_t *);
85 extern int cpuid_get_coreid(cpu_t *);
86 extern int cpuid_get_chipid(cpu_t *);
87 
88 /*
89  *	PSM functions initialization
90  */
91 void (*psm_shutdownf)(int, int)	= (void (*)(int, int))return_instr;
92 void (*psm_preshutdownf)(int, int) = (void (*)(int, int))return_instr;
93 void (*psm_notifyf)(int)	= (void (*)(int))return_instr;
94 void (*psm_set_idle_cpuf)(int)	= (void (*)(int))return_instr;
95 void (*psm_unset_idle_cpuf)(int) = (void (*)(int))return_instr;
96 void (*psminitf)()		= mach_init;
97 void (*picinitf)() 		= return_instr;
98 int (*clkinitf)(int, int *) 	= (int (*)(int, int *))return_instr;
99 int (*ap_mlsetup)() 		= (int (*)(void))return_instr;
100 void (*send_dirintf)() 		= return_instr;
101 void (*setspl)(int)		= (void (*)(int))return_instr;
102 int (*addspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
103 int (*delspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
104 void (*setsoftint)(int, struct av_softinfo *)=
105 	(void (*)(int, struct av_softinfo *))return_instr;
106 int (*slvltovect)(int)		= (int (*)(int))return_instr;
107 int (*setlvl)(int, int *)	= (int (*)(int, int *))return_instr;
108 void (*setlvlx)(int, int)	= (void (*)(int, int))return_instr;
109 int (*psm_disable_intr)(int)	= mp_disable_intr;
110 void (*psm_enable_intr)(int)	= mp_enable_intr;
111 hrtime_t (*gethrtimef)(void)	= dummy_hrtime;
112 hrtime_t (*gethrtimeunscaledf)(void)	= dummy_hrtime;
113 void (*scalehrtimef)(hrtime_t *)	= dummy_scalehrtime;
114 int (*psm_translate_irq)(dev_info_t *, int) = mach_translate_irq;
115 void (*gethrestimef)(timestruc_t *) = pc_gethrestime;
116 void (*psm_notify_error)(int, char *) = (void (*)(int, char *))NULL;
117 int (*psm_get_clockirq)(int) = NULL;
118 int (*psm_get_ipivect)(int, int) = NULL;
119 
120 int (*psm_clkinit)(int) = NULL;
121 void (*psm_timer_reprogram)(hrtime_t) = NULL;
122 void (*psm_timer_enable)(void) = NULL;
123 void (*psm_timer_disable)(void) = NULL;
124 void (*psm_post_cyclic_setup)(void *arg) = NULL;
125 int (*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *, psm_intr_op_t,
126     int *) = mach_intr_ops;
127 
128 void (*notify_error)(int, char *) = (void (*)(int, char *))return_instr;
129 void (*hrtime_tick)(void)	= return_instr;
130 
131 int tsc_gethrtime_enable = 1;
132 int tsc_gethrtime_initted = 0;
133 
134 /*
135  * Local Static Data
136  */
137 static struct psm_ops mach_ops;
138 static struct psm_ops *mach_set[4] = {&mach_ops, NULL, NULL, NULL};
139 static ushort_t mach_ver[4] = {0, 0, 0, 0};
140 
141 /*
142  * If non-zero, idle cpus will become "halted" when there's
143  * no work to do.
144  */
145 int	idle_cpu_use_hlt = 1;
146 
147 
148 /*ARGSUSED*/
149 int
150 pg_plat_hw_shared(cpu_t *cp, pghw_type_t hw)
151 {
152 	switch (hw) {
153 	case PGHW_IPIPE:
154 		if (x86_feature & (X86_HTT)) {
155 			/*
156 			 * Hyper-threading is SMT
157 			 */
158 			return (1);
159 		} else {
160 			return (0);
161 		}
162 	case PGHW_CHIP:
163 		if (x86_feature & (X86_CMP|X86_HTT))
164 			return (1);
165 		else
166 			return (0);
167 	default:
168 		return (0);
169 	}
170 }
171 
172 /*
173  * Compare two CPUs and see if they have a pghw_type_t sharing relationship
174  * If pghw_type_t is an unsupported hardware type, then return -1
175  */
176 int
177 pg_plat_cpus_share(cpu_t *cpu_a, cpu_t *cpu_b, pghw_type_t hw)
178 {
179 	id_t pgp_a, pgp_b;
180 
181 	pgp_a = pg_plat_hw_instance_id(cpu_a, hw);
182 	pgp_b = pg_plat_hw_instance_id(cpu_b, hw);
183 
184 	if (pgp_a == -1 || pgp_b == -1)
185 		return (-1);
186 
187 	return (pgp_a == pgp_b);
188 }
189 
190 /*
191  * Return a physical instance identifier for known hardware sharing
192  * relationships
193  */
194 id_t
195 pg_plat_hw_instance_id(cpu_t *cpu, pghw_type_t hw)
196 {
197 	switch (hw) {
198 	case PGHW_IPIPE:
199 		return (cpuid_get_coreid(cpu));
200 	case PGHW_CHIP:
201 		return (cpuid_get_chipid(cpu));
202 	default:
203 		return (-1);
204 	}
205 }
206 
207 int
208 pg_plat_hw_level(pghw_type_t hw)
209 {
210 	int i;
211 	static pghw_type_t hw_hier[] = {
212 		PGHW_IPIPE,
213 		PGHW_CHIP,
214 		PGHW_NUM_COMPONENTS
215 	};
216 
217 	for (i = 0; hw_hier[i] != PGHW_NUM_COMPONENTS; i++) {
218 		if (hw_hier[i] == hw)
219 			return (i);
220 	}
221 	return (-1);
222 }
223 
224 id_t
225 pg_plat_get_core_id(cpu_t *cpu)
226 {
227 	return ((id_t)cpuid_get_coreid(cpu));
228 }
229 
230 void
231 cmp_set_nosteal_interval(void)
232 {
233 	/* Set the nosteal interval (used by disp_getbest()) to 100us */
234 	nosteal_nsec = 100000UL;
235 }
236 
237 /*
238  * Routine to ensure initial callers to hrtime gets 0 as return
239  */
240 static hrtime_t
241 dummy_hrtime(void)
242 {
243 	return (0);
244 }
245 
246 /* ARGSUSED */
247 static void
248 dummy_scalehrtime(hrtime_t *ticks)
249 {}
250 
251 /*
252  * Idle the present CPU until awoken via an interrupt
253  */
254 static void
255 cpu_idle(void)
256 {
257 	cpu_t		*cpup = CPU;
258 	processorid_t	cpun = cpup->cpu_id;
259 	cpupart_t	*cp = cpup->cpu_part;
260 	int		hset_update = 1;
261 
262 	/*
263 	 * If this CPU is online, and there's multiple CPUs
264 	 * in the system, then we should notate our halting
265 	 * by adding ourselves to the partition's halted CPU
266 	 * bitmap. This allows other CPUs to find/awaken us when
267 	 * work becomes available.
268 	 */
269 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
270 		hset_update = 0;
271 
272 	/*
273 	 * Add ourselves to the partition's halted CPUs bitmask
274 	 * and set our HALTED flag, if necessary.
275 	 *
276 	 * When a thread becomes runnable, it is placed on the queue
277 	 * and then the halted cpuset is checked to determine who
278 	 * (if anyone) should be awoken. We therefore need to first
279 	 * add ourselves to the halted cpuset, and and then check if there
280 	 * is any work available.
281 	 *
282 	 * Note that memory barriers after updating the HALTED flag
283 	 * are not necessary since an atomic operation (updating the bitmap)
284 	 * immediately follows. On x86 the atomic operation acts as a
285 	 * memory barrier for the update of cpu_disp_flags.
286 	 */
287 	if (hset_update) {
288 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
289 		CPUSET_ATOMIC_ADD(cp->cp_mach->mc_haltset, cpun);
290 	}
291 
292 	/*
293 	 * Check to make sure there's really nothing to do.
294 	 * Work destined for this CPU may become available after
295 	 * this check. We'll be notified through the clearing of our
296 	 * bit in the halted CPU bitmask, and a poke.
297 	 */
298 	if (disp_anywork()) {
299 		if (hset_update) {
300 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
301 			CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
302 		}
303 		return;
304 	}
305 
306 	/*
307 	 * We're on our way to being halted.
308 	 *
309 	 * Disable interrupts now, so that we'll awaken immediately
310 	 * after halting if someone tries to poke us between now and
311 	 * the time we actually halt.
312 	 *
313 	 * We check for the presence of our bit after disabling interrupts.
314 	 * If it's cleared, we'll return. If the bit is cleared after
315 	 * we check then the poke will pop us out of the halted state.
316 	 *
317 	 * This means that the ordering of the poke and the clearing
318 	 * of the bit by cpu_wakeup is important.
319 	 * cpu_wakeup() must clear, then poke.
320 	 * cpu_idle() must disable interrupts, then check for the bit.
321 	 */
322 	cli();
323 
324 	if (hset_update && !CPU_IN_SET(cp->cp_mach->mc_haltset, cpun)) {
325 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
326 		sti();
327 		return;
328 	}
329 
330 	/*
331 	 * The check for anything locally runnable is here for performance
332 	 * and isn't needed for correctness. disp_nrunnable ought to be
333 	 * in our cache still, so it's inexpensive to check, and if there
334 	 * is anything runnable we won't have to wait for the poke.
335 	 */
336 	if (cpup->cpu_disp->disp_nrunnable != 0) {
337 		if (hset_update) {
338 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
339 			CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
340 		}
341 		sti();
342 		return;
343 	}
344 
345 	mach_cpu_idle();
346 
347 	/*
348 	 * We're no longer halted
349 	 */
350 	if (hset_update) {
351 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
352 		CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpun);
353 	}
354 }
355 
356 
357 /*
358  * If "cpu" is halted, then wake it up clearing its halted bit in advance.
359  * Otherwise, see if other CPUs in the cpu partition are halted and need to
360  * be woken up so that they can steal the thread we placed on this CPU.
361  * This function is only used on MP systems.
362  */
363 static void
364 cpu_wakeup(cpu_t *cpu, int bound)
365 {
366 	uint_t		cpu_found;
367 	int		result;
368 	cpupart_t	*cp;
369 
370 	cp = cpu->cpu_part;
371 	if (CPU_IN_SET(cp->cp_mach->mc_haltset, cpu->cpu_id)) {
372 		/*
373 		 * Clear the halted bit for that CPU since it will be
374 		 * poked in a moment.
375 		 */
376 		CPUSET_ATOMIC_DEL(cp->cp_mach->mc_haltset, cpu->cpu_id);
377 		/*
378 		 * We may find the current CPU present in the halted cpuset
379 		 * if we're in the context of an interrupt that occurred
380 		 * before we had a chance to clear our bit in cpu_idle().
381 		 * Poking ourself is obviously unnecessary, since if
382 		 * we're here, we're not halted.
383 		 */
384 		if (cpu != CPU)
385 			poke_cpu(cpu->cpu_id);
386 		return;
387 	} else {
388 		/*
389 		 * This cpu isn't halted, but it's idle or undergoing a
390 		 * context switch. No need to awaken anyone else.
391 		 */
392 		if (cpu->cpu_thread == cpu->cpu_idle_thread ||
393 		    cpu->cpu_disp_flags & CPU_DISP_DONTSTEAL)
394 			return;
395 	}
396 
397 	/*
398 	 * No need to wake up other CPUs if the thread we just enqueued
399 	 * is bound.
400 	 */
401 	if (bound)
402 		return;
403 
404 
405 	/*
406 	 * See if there's any other halted CPUs. If there are, then
407 	 * select one, and awaken it.
408 	 * It's possible that after we find a CPU, somebody else
409 	 * will awaken it before we get the chance.
410 	 * In that case, look again.
411 	 */
412 	do {
413 		CPUSET_FIND(cp->cp_mach->mc_haltset, cpu_found);
414 		if (cpu_found == CPUSET_NOTINSET)
415 			return;
416 
417 		ASSERT(cpu_found >= 0 && cpu_found < NCPU);
418 		CPUSET_ATOMIC_XDEL(cp->cp_mach->mc_haltset, cpu_found, result);
419 	} while (result < 0);
420 
421 	if (cpu_found != CPU->cpu_id)
422 		poke_cpu(cpu_found);
423 }
424 
425 void (*cpu_pause_handler)(volatile char *) = NULL;
426 
427 static int
428 mp_disable_intr(int cpun)
429 {
430 	/*
431 	 * switch to the offline cpu
432 	 */
433 	affinity_set(cpun);
434 	/*
435 	 * raise ipl to just below cross call
436 	 */
437 	splx(XC_MED_PIL-1);
438 	/*
439 	 *	set base spl to prevent the next swtch to idle from
440 	 *	lowering back to ipl 0
441 	 */
442 	CPU->cpu_intr_actv |= (1 << (XC_MED_PIL-1));
443 	set_base_spl();
444 	affinity_clear();
445 	return (DDI_SUCCESS);
446 }
447 
448 static void
449 mp_enable_intr(int cpun)
450 {
451 	/*
452 	 * switch to the online cpu
453 	 */
454 	affinity_set(cpun);
455 	/*
456 	 * clear the interrupt active mask
457 	 */
458 	CPU->cpu_intr_actv &= ~(1 << (XC_MED_PIL-1));
459 	set_base_spl();
460 	(void) spl0();
461 	affinity_clear();
462 }
463 
464 static void
465 mach_get_platform(int owner)
466 {
467 	void		**srv_opsp;
468 	void		**clt_opsp;
469 	int		i;
470 	int		total_ops;
471 
472 	/* fix up psm ops */
473 	srv_opsp = (void **)mach_set[0];
474 	clt_opsp = (void **)mach_set[owner];
475 	if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01)
476 		total_ops = sizeof (struct psm_ops_ver01) /
477 				sizeof (void (*)(void));
478 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_1)
479 		/* no psm_notify_func */
480 		total_ops = OFFSETOF(struct psm_ops, psm_notify_func) /
481 		    sizeof (void (*)(void));
482 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_2)
483 		/* no psm_timer funcs */
484 		total_ops = OFFSETOF(struct psm_ops, psm_timer_reprogram) /
485 		    sizeof (void (*)(void));
486 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_3)
487 		/* no psm_preshutdown function */
488 		total_ops = OFFSETOF(struct psm_ops, psm_preshutdown) /
489 		    sizeof (void (*)(void));
490 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_4)
491 		/* no psm_preshutdown function */
492 		total_ops = OFFSETOF(struct psm_ops, psm_intr_ops) /
493 		    sizeof (void (*)(void));
494 	else
495 		total_ops = sizeof (struct psm_ops) / sizeof (void (*)(void));
496 
497 	/*
498 	 * Save the version of the PSM module, in case we need to
499 	 * bahave differently based on version.
500 	 */
501 	mach_ver[0] = mach_ver[owner];
502 
503 	for (i = 0; i < total_ops; i++)
504 		if (clt_opsp[i] != NULL)
505 			srv_opsp[i] = clt_opsp[i];
506 }
507 
508 static void
509 mach_construct_info()
510 {
511 	struct psm_sw *swp;
512 	int	mach_cnt[PSM_OWN_OVERRIDE+1] = {0};
513 	int	conflict_owner = 0;
514 
515 	if (psmsw->psw_forw == psmsw)
516 		panic("No valid PSM modules found");
517 	mutex_enter(&psmsw_lock);
518 	for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
519 		if (!(swp->psw_flag & PSM_MOD_IDENTIFY))
520 			continue;
521 		mach_set[swp->psw_infop->p_owner] = swp->psw_infop->p_ops;
522 		mach_ver[swp->psw_infop->p_owner] = swp->psw_infop->p_version;
523 		mach_cnt[swp->psw_infop->p_owner]++;
524 	}
525 	mutex_exit(&psmsw_lock);
526 
527 	mach_get_platform(PSM_OWN_SYS_DEFAULT);
528 
529 	/* check to see are there any conflicts */
530 	if (mach_cnt[PSM_OWN_EXCLUSIVE] > 1)
531 		conflict_owner = PSM_OWN_EXCLUSIVE;
532 	if (mach_cnt[PSM_OWN_OVERRIDE] > 1)
533 		conflict_owner = PSM_OWN_OVERRIDE;
534 	if (conflict_owner) {
535 		/* remove all psm modules except uppc */
536 		cmn_err(CE_WARN,
537 			"Conflicts detected on the following PSM modules:");
538 		mutex_enter(&psmsw_lock);
539 		for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
540 			if (swp->psw_infop->p_owner == conflict_owner)
541 				cmn_err(CE_WARN, "%s ",
542 					swp->psw_infop->p_mach_idstring);
543 		}
544 		mutex_exit(&psmsw_lock);
545 		cmn_err(CE_WARN,
546 			"Setting the system back to SINGLE processor mode!");
547 		cmn_err(CE_WARN,
548 		    "Please edit /etc/mach to remove the invalid PSM module.");
549 		return;
550 	}
551 
552 	if (mach_set[PSM_OWN_EXCLUSIVE])
553 		mach_get_platform(PSM_OWN_EXCLUSIVE);
554 
555 	if (mach_set[PSM_OWN_OVERRIDE])
556 		mach_get_platform(PSM_OWN_OVERRIDE);
557 }
558 
559 static void
560 mach_init()
561 {
562 	struct psm_ops  *pops;
563 
564 	mach_construct_info();
565 
566 	pops = mach_set[0];
567 
568 	/* register the interrupt and clock initialization rotuines */
569 	picinitf = mach_picinit;
570 	clkinitf = mach_clkinit;
571 	psm_get_clockirq = pops->psm_get_clockirq;
572 
573 	/* register the interrupt setup code */
574 	slvltovect = mach_softlvl_to_vect;
575 	addspl	= pops->psm_addspl;
576 	delspl	= pops->psm_delspl;
577 
578 	if (pops->psm_translate_irq)
579 		psm_translate_irq = pops->psm_translate_irq;
580 	if (pops->psm_intr_ops)
581 		psm_intr_ops = pops->psm_intr_ops;
582 
583 #if defined(PSMI_1_2) || defined(PSMI_1_3) || defined(PSMI_1_4)
584 	/*
585 	 * Time-of-day functionality now handled in TOD modules.
586 	 * (Warn about PSM modules that think that we're going to use
587 	 * their ops vectors.)
588 	 */
589 	if (pops->psm_tod_get)
590 		cmn_err(CE_WARN, "obsolete psm_tod_get op %p",
591 		    (void *)pops->psm_tod_get);
592 
593 	if (pops->psm_tod_set)
594 		cmn_err(CE_WARN, "obsolete psm_tod_set op %p",
595 		    (void *)pops->psm_tod_set);
596 #endif
597 
598 	if (pops->psm_notify_error) {
599 		psm_notify_error = mach_notify_error;
600 		notify_error = pops->psm_notify_error;
601 	}
602 
603 	(*pops->psm_softinit)();
604 
605 	/*
606 	 * Initialize the dispatcher's function hooks
607 	 * to enable CPU halting when idle
608 	 */
609 	if (idle_cpu_use_hlt)
610 		idle_cpu = cpu_idle;
611 
612 	mach_smpinit();
613 }
614 
615 static void
616 mach_smpinit(void)
617 {
618 	struct psm_ops  *pops;
619 	processorid_t cpu_id;
620 	int cnt;
621 	cpuset_t cpumask;
622 
623 	pops = mach_set[0];
624 
625 	cpu_id = -1;
626 	cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
627 	for (cnt = 0, CPUSET_ZERO(cpumask); cpu_id != -1; cnt++) {
628 		CPUSET_ADD(cpumask, cpu_id);
629 		cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
630 	}
631 
632 	mp_cpus = cpumask;
633 
634 	/* MP related routines */
635 	ap_mlsetup = pops->psm_post_cpu_start;
636 	send_dirintf = pops->psm_send_ipi;
637 
638 	/* optional MP related routines */
639 	if (pops->psm_shutdown)
640 		psm_shutdownf = pops->psm_shutdown;
641 	if (pops->psm_preshutdown)
642 		psm_preshutdownf = pops->psm_preshutdown;
643 	if (pops->psm_notify_func)
644 		psm_notifyf = pops->psm_notify_func;
645 	if (pops->psm_set_idlecpu)
646 		psm_set_idle_cpuf = pops->psm_set_idlecpu;
647 	if (pops->psm_unset_idlecpu)
648 		psm_unset_idle_cpuf = pops->psm_unset_idlecpu;
649 
650 	psm_clkinit = pops->psm_clkinit;
651 
652 	if (pops->psm_timer_reprogram)
653 		psm_timer_reprogram = pops->psm_timer_reprogram;
654 
655 	if (pops->psm_timer_enable)
656 		psm_timer_enable = pops->psm_timer_enable;
657 
658 	if (pops->psm_timer_disable)
659 		psm_timer_disable = pops->psm_timer_disable;
660 
661 	if (pops->psm_post_cyclic_setup)
662 		psm_post_cyclic_setup = pops->psm_post_cyclic_setup;
663 
664 	/* check for multiple cpu's */
665 	if (cnt < 2)
666 		return;
667 
668 	/* check for MP platforms */
669 	if (pops->psm_cpu_start == NULL)
670 		return;
671 
672 	/*
673 	 * Set the dispatcher hook to enable cpu "wake up"
674 	 * when a thread becomes runnable.
675 	 */
676 	if (idle_cpu_use_hlt)
677 		disp_enq_thread = cpu_wakeup;
678 
679 	if (pops->psm_disable_intr)
680 		psm_disable_intr = pops->psm_disable_intr;
681 	if (pops->psm_enable_intr)
682 		psm_enable_intr  = pops->psm_enable_intr;
683 
684 	psm_get_ipivect = pops->psm_get_ipivect;
685 
686 	(void) add_avintr((void *)NULL, XC_HI_PIL, xc_serv, "xc_hi_intr",
687 		(*pops->psm_get_ipivect)(XC_HI_PIL, PSM_INTR_IPI_HI),
688 		(caddr_t)X_CALL_HIPRI, NULL, NULL, NULL);
689 	(void) add_avintr((void *)NULL, XC_MED_PIL, xc_serv, "xc_med_intr",
690 		(*pops->psm_get_ipivect)(XC_MED_PIL, PSM_INTR_IPI_LO),
691 		(caddr_t)X_CALL_MEDPRI, NULL, NULL, NULL);
692 
693 	(void) (*pops->psm_get_ipivect)(XC_CPUPOKE_PIL, PSM_INTR_POKE);
694 }
695 
696 static void
697 mach_picinit()
698 {
699 	struct psm_ops  *pops;
700 
701 	pops = mach_set[0];
702 
703 	/* register the interrupt handlers */
704 	setlvl = pops->psm_intr_enter;
705 	setlvlx = pops->psm_intr_exit;
706 
707 	/* initialize the interrupt hardware */
708 	(*pops->psm_picinit)();
709 
710 	/* set interrupt mask for current ipl */
711 	setspl = pops->psm_setspl;
712 	cli();
713 	setspl(CPU->cpu_pri);
714 }
715 
716 uint_t	cpu_freq;	/* MHz */
717 uint64_t cpu_freq_hz;	/* measured (in hertz) */
718 
719 #define	MEGA_HZ		1000000
720 
721 static uint64_t
722 mach_calchz(uint32_t pit_counter, uint64_t *processor_clks)
723 {
724 	uint64_t cpu_hz;
725 
726 	if ((pit_counter == 0) || (*processor_clks == 0) ||
727 	    (*processor_clks > (((uint64_t)-1) / PIT_HZ)))
728 		return (0);
729 
730 	cpu_hz = ((uint64_t)PIT_HZ * *processor_clks) / pit_counter;
731 
732 	return (cpu_hz);
733 }
734 
735 static uint64_t
736 mach_getcpufreq(void)
737 {
738 	uint32_t pit_counter;
739 	uint64_t processor_clks;
740 
741 	if (x86_feature & X86_TSC) {
742 		/*
743 		 * We have a TSC. freq_tsc() knows how to measure the number
744 		 * of clock cycles sampled against the PIT.
745 		 */
746 		ulong_t flags = clear_int_flag();
747 		processor_clks = freq_tsc(&pit_counter);
748 		restore_int_flag(flags);
749 		return (mach_calchz(pit_counter, &processor_clks));
750 	} else if (x86_vendor == X86_VENDOR_Cyrix || x86_type == X86_TYPE_P5) {
751 #if defined(__amd64)
752 		panic("mach_getcpufreq: no TSC!");
753 #elif defined(__i386)
754 		/*
755 		 * We are a Cyrix based on a 6x86 core or an Intel Pentium
756 		 * for which freq_notsc() knows how to measure the number of
757 		 * elapsed clock cycles sampled against the PIT
758 		 */
759 		ulong_t flags = clear_int_flag();
760 		processor_clks = freq_notsc(&pit_counter);
761 		restore_int_flag(flags);
762 		return (mach_calchz(pit_counter, &processor_clks));
763 #endif	/* __i386 */
764 	}
765 
766 	/* We do not know how to calculate cpu frequency for this cpu. */
767 	return (0);
768 }
769 
770 /*
771  * If the clock speed of a cpu is found to be reported incorrectly, do not add
772  * to this array, instead improve the accuracy of the algorithm that determines
773  * the clock speed of the processor or extend the implementation to support the
774  * vendor as appropriate. This is here only to support adjusting the speed on
775  * older slower processors that mach_fixcpufreq() would not be able to account
776  * for otherwise.
777  */
778 static int x86_cpu_freq[] = { 60, 75, 80, 90, 120, 160, 166, 175, 180, 233 };
779 
780 /*
781  * On fast processors the clock frequency that is measured may be off by
782  * a few MHz from the value printed on the part. This is a combination of
783  * the factors that for such fast parts being off by this much is within
784  * the tolerances for manufacture and because of the difficulties in the
785  * measurement that can lead to small error. This function uses some
786  * heuristics in order to tweak the value that was measured to match what
787  * is most likely printed on the part.
788  *
789  * Some examples:
790  * 	AMD Athlon 1000 mhz measured as 998 mhz
791  * 	Intel Pentium III Xeon 733 mhz measured as 731 mhz
792  * 	Intel Pentium IV 1500 mhz measured as 1495mhz
793  *
794  * If in the future this function is no longer sufficient to correct
795  * for the error in the measurement, then the algorithm used to perform
796  * the measurement will have to be improved in order to increase accuracy
797  * rather than adding horrible and questionable kludges here.
798  *
799  * This is called after the cyclics subsystem because of the potential
800  * that the heuristics within may give a worse estimate of the clock
801  * frequency than the value that was measured.
802  */
803 static void
804 mach_fixcpufreq(void)
805 {
806 	uint32_t freq, mul, near66, delta66, near50, delta50, fixed, delta, i;
807 
808 	freq = (uint32_t)cpu_freq;
809 
810 	/*
811 	 * Find the nearest integer multiple of 200/3 (about 66) MHz to the
812 	 * measured speed taking into account that the 667 MHz parts were
813 	 * the first to round-up.
814 	 */
815 	mul = (uint32_t)((3 * (uint64_t)freq + 100) / 200);
816 	near66 = (uint32_t)((200 * (uint64_t)mul + ((mul >= 10) ? 1 : 0)) / 3);
817 	delta66 = (near66 > freq) ? (near66 - freq) : (freq - near66);
818 
819 	/* Find the nearest integer multiple of 50 MHz to the measured speed */
820 	mul = (freq + 25) / 50;
821 	near50 = mul * 50;
822 	delta50 = (near50 > freq) ? (near50 - freq) : (freq - near50);
823 
824 	/* Find the closer of the two */
825 	if (delta66 < delta50) {
826 		fixed = near66;
827 		delta = delta66;
828 	} else {
829 		fixed = near50;
830 		delta = delta50;
831 	}
832 
833 	if (fixed > INT_MAX)
834 		return;
835 
836 	/*
837 	 * Some older parts have a core clock frequency that is not an
838 	 * integral multiple of 50 or 66 MHz. Check if one of the old
839 	 * clock frequencies is closer to the measured value than any
840 	 * of the integral multiples of 50 an 66, and if so set fixed
841 	 * and delta appropriately to represent the closest value.
842 	 */
843 	i = sizeof (x86_cpu_freq) / sizeof (int);
844 	while (i > 0) {
845 		i--;
846 
847 		if (x86_cpu_freq[i] <= freq) {
848 			mul = freq - x86_cpu_freq[i];
849 
850 			if (mul < delta) {
851 				fixed = x86_cpu_freq[i];
852 				delta = mul;
853 			}
854 
855 			break;
856 		}
857 
858 		mul = x86_cpu_freq[i] - freq;
859 
860 		if (mul < delta) {
861 			fixed = x86_cpu_freq[i];
862 			delta = mul;
863 		}
864 	}
865 
866 	/*
867 	 * Set a reasonable maximum for how much to correct the measured
868 	 * result by. This check is here to prevent the adjustment made
869 	 * by this function from being more harm than good. It is entirely
870 	 * possible that in the future parts will be made that are not
871 	 * integral multiples of 66 or 50 in clock frequency or that
872 	 * someone may overclock a part to some odd frequency. If the
873 	 * measured value is farther from the corrected value than
874 	 * allowed, then assume the corrected value is in error and use
875 	 * the measured value.
876 	 */
877 	if (6 < delta)
878 		return;
879 
880 	cpu_freq = (int)fixed;
881 }
882 
883 
884 static int
885 machhztomhz(uint64_t cpu_freq_hz)
886 {
887 	uint64_t cpu_mhz;
888 
889 	/* Round to nearest MHZ */
890 	cpu_mhz = (cpu_freq_hz + (MEGA_HZ / 2)) / MEGA_HZ;
891 
892 	if (cpu_mhz > INT_MAX)
893 		return (0);
894 
895 	return ((int)cpu_mhz);
896 
897 }
898 
899 
900 static int
901 mach_clkinit(int preferred_mode, int *set_mode)
902 {
903 	struct psm_ops  *pops;
904 	int resolution;
905 
906 	pops = mach_set[0];
907 
908 	cpu_freq_hz = mach_getcpufreq();
909 
910 	cpu_freq = machhztomhz(cpu_freq_hz);
911 
912 	if (!(x86_feature & X86_TSC) || (cpu_freq == 0))
913 		tsc_gethrtime_enable = 0;
914 
915 	if (tsc_gethrtime_enable) {
916 		tsc_hrtimeinit(cpu_freq_hz);
917 		gethrtimef = tsc_gethrtime;
918 		gethrtimeunscaledf = tsc_gethrtimeunscaled;
919 		scalehrtimef = tsc_scalehrtime;
920 		hrtime_tick = tsc_tick;
921 		tsc_gethrtime_initted = 1;
922 	} else {
923 		if (pops->psm_hrtimeinit)
924 			(*pops->psm_hrtimeinit)();
925 		gethrtimef = pops->psm_gethrtime;
926 		gethrtimeunscaledf = gethrtimef;
927 		/* scalehrtimef will remain dummy */
928 	}
929 
930 	mach_fixcpufreq();
931 
932 	if (mach_ver[0] >= PSM_INFO_VER01_3) {
933 		if ((preferred_mode == TIMER_ONESHOT) &&
934 		    (tsc_gethrtime_enable)) {
935 
936 			resolution = (*pops->psm_clkinit)(0);
937 			if (resolution != 0)  {
938 				*set_mode = TIMER_ONESHOT;
939 				return (resolution);
940 			}
941 
942 		}
943 
944 		/*
945 		 * either periodic mode was requested or could not set to
946 		 * one-shot mode
947 		 */
948 		resolution = (*pops->psm_clkinit)(hz);
949 		/*
950 		 * psm should be able to do periodic, so we do not check
951 		 * for return value of psm_clkinit here.
952 		 */
953 		*set_mode = TIMER_PERIODIC;
954 		return (resolution);
955 	} else {
956 		/*
957 		 * PSMI interface prior to PSMI_3 does not define a return
958 		 * value for psm_clkinit, so the return value is ignored.
959 		 */
960 		(void) (*pops->psm_clkinit)(hz);
961 		*set_mode = TIMER_PERIODIC;
962 		return (nsec_per_tick);
963 	}
964 }
965 
966 /*ARGSUSED*/
967 static void
968 mach_psm_set_softintr(int ipl, struct av_softinfo *pending)
969 {
970 	struct psm_ops  *pops;
971 
972 	/* invoke hardware interrupt					*/
973 	pops = mach_set[0];
974 	(*pops->psm_set_softintr)(ipl);
975 }
976 
977 static int
978 mach_softlvl_to_vect(int ipl)
979 {
980 	int softvect;
981 	struct psm_ops  *pops;
982 
983 	pops = mach_set[0];
984 
985 	/* check for null handler for set soft interrupt call		*/
986 	if (pops->psm_set_softintr == NULL) {
987 		setsoftint = av_set_softint_pending;
988 		return (PSM_SV_SOFTWARE);
989 	}
990 
991 	softvect = (*pops->psm_softlvl_to_irq)(ipl);
992 	/* check for hardware scheme					*/
993 	if (softvect > PSM_SV_SOFTWARE) {
994 		setsoftint = mach_psm_set_softintr;
995 		return (softvect);
996 	}
997 
998 	if (softvect == PSM_SV_SOFTWARE)
999 		setsoftint = av_set_softint_pending;
1000 	else	/* hardware and software mixed scheme			*/
1001 		setsoftint = mach_set_softintr;
1002 
1003 	return (PSM_SV_SOFTWARE);
1004 }
1005 
1006 static void
1007 mach_set_softintr(int ipl, struct av_softinfo *pending)
1008 {
1009 	struct psm_ops  *pops;
1010 
1011 	/* set software pending bits					*/
1012 	av_set_softint_pending(ipl, pending);
1013 
1014 	/*	check if dosoftint will be called at the end of intr	*/
1015 	if (CPU_ON_INTR(CPU) || (curthread->t_intr))
1016 		return;
1017 
1018 	/* invoke hardware interrupt					*/
1019 	pops = mach_set[0];
1020 	(*pops->psm_set_softintr)(ipl);
1021 }
1022 
1023 #ifdef DEBUG
1024 /*
1025  * This is here to allow us to simulate cpus that refuse to start.
1026  */
1027 cpuset_t cpufailset;
1028 #endif
1029 
1030 int
1031 mach_cpu_start(struct cpu *cp, void *ctx)
1032 {
1033 	struct psm_ops *pops = mach_set[0];
1034 	processorid_t id = cp->cpu_id;
1035 
1036 #ifdef DEBUG
1037 	if (CPU_IN_SET(cpufailset, id))
1038 		return (0);
1039 #endif
1040 	return ((*pops->psm_cpu_start)(id, ctx));
1041 }
1042 
1043 /*ARGSUSED*/
1044 static int
1045 mach_translate_irq(dev_info_t *dip, int irqno)
1046 {
1047 	return (irqno);	/* default to NO translation */
1048 }
1049 
1050 static void
1051 mach_notify_error(int level, char *errmsg)
1052 {
1053 	/*
1054 	 * SL_FATAL is pass in once panicstr is set, deliver it
1055 	 * as CE_PANIC.  Also, translate SL_ codes back to CE_
1056 	 * codes for the psmi handler
1057 	 */
1058 	if (level & SL_FATAL)
1059 		(*notify_error)(CE_PANIC, errmsg);
1060 	else if (level & SL_WARN)
1061 		(*notify_error)(CE_WARN, errmsg);
1062 	else if (level & SL_NOTE)
1063 		(*notify_error)(CE_NOTE, errmsg);
1064 	else if (level & SL_CONSOLE)
1065 		(*notify_error)(CE_CONT, errmsg);
1066 }
1067 
1068 /*
1069  * It provides the default basic intr_ops interface for the new DDI
1070  * interrupt framework if the PSM doesn't have one.
1071  *
1072  * Input:
1073  * dip     - pointer to the dev_info structure of the requested device
1074  * hdlp    - pointer to the internal interrupt handle structure for the
1075  *	     requested interrupt
1076  * intr_op - opcode for this call
1077  * result  - pointer to the integer that will hold the result to be
1078  *	     passed back if return value is PSM_SUCCESS
1079  *
1080  * Output:
1081  * return value is either PSM_SUCCESS or PSM_FAILURE
1082  */
1083 static int
1084 mach_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1085     psm_intr_op_t intr_op, int *result)
1086 {
1087 	struct intrspec *ispec;
1088 
1089 	switch (intr_op) {
1090 	case PSM_INTR_OP_CHECK_MSI:
1091 		*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1092 			    DDI_INTR_TYPE_MSIX);
1093 		break;
1094 	case PSM_INTR_OP_ALLOC_VECTORS:
1095 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1096 			*result = 1;
1097 		else
1098 			*result = 0;
1099 		break;
1100 	case PSM_INTR_OP_FREE_VECTORS:
1101 		break;
1102 	case PSM_INTR_OP_NAVAIL_VECTORS:
1103 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1104 			*result = 1;
1105 		else
1106 			*result = 0;
1107 		break;
1108 	case PSM_INTR_OP_XLATE_VECTOR:
1109 		ispec = ((ihdl_plat_t *)hdlp->ih_private)->ip_ispecp;
1110 		*result = psm_translate_irq(dip, ispec->intrspec_vec);
1111 		break;
1112 	case PSM_INTR_OP_GET_CAP:
1113 		*result = 0;
1114 		break;
1115 	case PSM_INTR_OP_GET_PENDING:
1116 	case PSM_INTR_OP_CLEAR_MASK:
1117 	case PSM_INTR_OP_SET_MASK:
1118 	case PSM_INTR_OP_GET_SHARED:
1119 	case PSM_INTR_OP_SET_PRI:
1120 	case PSM_INTR_OP_SET_CAP:
1121 	case PSM_INTR_OP_SET_CPU:
1122 	case PSM_INTR_OP_GET_INTR:
1123 	default:
1124 		return (PSM_FAILURE);
1125 	}
1126 	return (PSM_SUCCESS);
1127 }
1128