xref: /titanic_41/usr/src/uts/i86pc/os/mp_machdep.c (revision fd9cb95cbb2f626355a60efb9d02c5f0a33c10e6)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * 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 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #define	PSMI_1_5
30 #include <sys/smp_impldefs.h>
31 #include <sys/psm.h>
32 #include <sys/psm_modctl.h>
33 #include <sys/pit.h>
34 #include <sys/cmn_err.h>
35 #include <sys/strlog.h>
36 #include <sys/clock.h>
37 #include <sys/debug.h>
38 #include <sys/rtc.h>
39 #include <sys/x86_archext.h>
40 #include <sys/cpupart.h>
41 #include <sys/cpuvar.h>
42 #include <sys/chip.h>
43 #include <sys/disp.h>
44 #include <sys/cpu.h>
45 #include <sys/archsystm.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);
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 
90 /*
91  *	PSM functions initialization
92  */
93 void (*psm_shutdownf)(int, int)	= return_instr;
94 void (*psm_preshutdownf)(int, int) = return_instr;
95 void (*psm_notifyf)(int)	= return_instr;
96 void (*psm_set_idle_cpuf)(int)	= return_instr;
97 void (*psm_unset_idle_cpuf)(int) = return_instr;
98 void (*psminitf)()		= mach_init;
99 void (*picinitf)() 		= return_instr;
100 int (*clkinitf)(int, int *) 	= (int (*)(int, int *))return_instr;
101 void (*cpu_startf)() 		= return_instr;
102 int (*ap_mlsetup)() 		= (int (*)(void))return_instr;
103 void (*send_dirintf)() 		= return_instr;
104 void (*setspl)(int)		= return_instr;
105 int (*addspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
106 int (*delspl)(int, int, int, int) = (int (*)(int, int, int, int))return_instr;
107 void (*setsoftint)(int)		= (void (*)(int))return_instr;
108 int (*slvltovect)(int)		= (int (*)(int))return_instr;
109 int (*setlvl)(int, int *)	= (int (*)(int, int *))return_instr;
110 void (*setlvlx)(int, int)	= (void (*)(int, int))return_instr;
111 int (*psm_disable_intr)(int)	= mp_disable_intr;
112 void (*psm_enable_intr)(int)	= mp_enable_intr;
113 hrtime_t (*gethrtimef)(void)	= dummy_hrtime;
114 hrtime_t (*gethrtimeunscaledf)(void)	= dummy_hrtime;
115 void (*scalehrtimef)(hrtime_t *)	= dummy_scalehrtime;
116 int (*psm_translate_irq)(dev_info_t *, int) = mach_translate_irq;
117 void (*gethrestimef)(timestruc_t *) = pc_gethrestime;
118 int (*psm_todgetf)(todinfo_t *) = (int (*)(todinfo_t *))return_instr;
119 int (*psm_todsetf)(todinfo_t *) = (int (*)(todinfo_t *))return_instr;
120 void (*psm_notify_error)(int, char *) = (void (*)(int, char *))NULL;
121 int (*psm_get_clockirq)(int) = NULL;
122 int (*psm_get_ipivect)(int, int) = NULL;
123 
124 int (*psm_clkinit)(int) = NULL;
125 void (*psm_timer_reprogram)(hrtime_t) = NULL;
126 void (*psm_timer_enable)(void) = NULL;
127 void (*psm_timer_disable)(void) = NULL;
128 void (*psm_post_cyclic_setup)(void *arg) = NULL;
129 int (*psm_intr_ops)(dev_info_t *, ddi_intr_handle_impl_t *, psm_intr_op_t,
130     int *) = mach_intr_ops;
131 
132 void (*notify_error)(int, char *) = (void (*)(int, char *))return_instr;
133 void (*hrtime_tick)(void)	= return_instr;
134 
135 int tsc_gethrtime_enable = 1;
136 int tsc_gethrtime_initted = 0;
137 
138 /*
139  * Local Static Data
140  */
141 static struct psm_ops mach_ops;
142 static struct psm_ops *mach_set[4] = {&mach_ops, NULL, NULL, NULL};
143 static ushort_t mach_ver[4] = {0, 0, 0, 0};
144 
145 /*
146  * If non-zero, idle cpus will "halted" when there's
147  * no work to do.
148  */
149 int	halt_idle_cpus = 1;
150 
151 #if defined(__amd64)
152 /*
153  * If non-zero, will use cr8 for interrupt priority masking
154  * We declare this here since install_spl is called from here
155  * (where this is checked).
156  */
157 int	intpri_use_cr8 = 0;
158 #endif	/* __amd64 */
159 
160 #ifdef	_SIMULATOR_SUPPORT
161 
162 int simulator_run = 0;	/* patch to non-zero if running under simics */
163 
164 #endif	/* _SIMULATOR_SUPPORT */
165 
166 /* ARGSUSED */
167 void
168 chip_plat_define_chip(cpu_t *cp, chip_def_t *cd)
169 {
170 	if (x86_feature & (X86_HTT|X86_CMP))
171 		/*
172 		 * Hyperthreading is SMT
173 		 */
174 		cd->chipd_type = CHIP_SMT;
175 	else
176 		cd->chipd_type = CHIP_DEFAULT;
177 
178 	cd->chipd_rechoose_adj = 0;
179 }
180 
181 /*
182  * Routine to ensure initial callers to hrtime gets 0 as return
183  */
184 static hrtime_t
185 dummy_hrtime(void)
186 {
187 	return (0);
188 }
189 
190 /* ARGSUSED */
191 static void
192 dummy_scalehrtime(hrtime_t *ticks)
193 {}
194 
195 /*
196  * Halt the present CPU until awoken via an interrupt
197  */
198 static void
199 cpu_halt(void)
200 {
201 	cpu_t		*cpup = CPU;
202 	processorid_t	cpun = cpup->cpu_id;
203 	cpupart_t	*cp = cpup->cpu_part;
204 	int		hset_update = 1;
205 
206 	/*
207 	 * If this CPU is online, and there's multiple CPUs
208 	 * in the system, then we should notate our halting
209 	 * by adding ourselves to the partition's halted CPU
210 	 * bitmap. This allows other CPUs to find/awaken us when
211 	 * work becomes available.
212 	 */
213 	if (cpup->cpu_flags & CPU_OFFLINE || ncpus == 1)
214 		hset_update = 0;
215 
216 	/*
217 	 * Add ourselves to the partition's halted CPUs bitmask
218 	 * and set our HALTED flag, if necessary.
219 	 *
220 	 * When a thread becomes runnable, it is placed on the queue
221 	 * and then the halted cpuset is checked to determine who
222 	 * (if anyone) should be awoken. We therefore need to first
223 	 * add ourselves to the halted cpuset, and and then check if there
224 	 * is any work available.
225 	 *
226 	 * Note that memory barriers after updating the HALTED flag
227 	 * are not necessary since an atomic operation (updating the bitmap)
228 	 * immediately follows. On x86 the atomic operation acts as a
229 	 * memory barrier for the update of cpu_disp_flags.
230 	 */
231 	if (hset_update) {
232 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
233 		CPUSET_ATOMIC_ADD(cp->cp_haltset, cpun);
234 	}
235 
236 	/*
237 	 * Check to make sure there's really nothing to do.
238 	 * Work destined for this CPU may become available after
239 	 * this check. We'll be notified through the clearing of our
240 	 * bit in the halted CPU bitmask, and a poke.
241 	 */
242 	if (disp_anywork()) {
243 		if (hset_update) {
244 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
245 			CPUSET_ATOMIC_DEL(cp->cp_haltset, cpun);
246 		}
247 		return;
248 	}
249 
250 	/*
251 	 * We're on our way to being halted.
252 	 *
253 	 * Disable interrupts now, so that we'll awaken immediately
254 	 * after halting if someone tries to poke us between now and
255 	 * the time we actually halt.
256 	 *
257 	 * We check for the presence of our bit after disabling interrupts.
258 	 * If it's cleared, we'll return. If the bit is cleared after
259 	 * we check then the poke will pop us out of the halted state.
260 	 *
261 	 * This means that the ordering of the poke and the clearing
262 	 * of the bit by cpu_wakeup is important.
263 	 * cpu_wakeup() must clear, then poke.
264 	 * cpu_halt() must disable interrupts, then check for the bit.
265 	 */
266 	cli();
267 
268 	if (hset_update && !CPU_IN_SET(cp->cp_haltset, cpun)) {
269 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
270 		sti();
271 		return;
272 	}
273 
274 	/*
275 	 * The check for anything locally runnable is here for performance
276 	 * and isn't needed for correctness. disp_nrunnable ought to be
277 	 * in our cache still, so it's inexpensive to check, and if there
278 	 * is anything runnable we won't have to wait for the poke.
279 	 */
280 	if (cpup->cpu_disp->disp_nrunnable != 0) {
281 		if (hset_update) {
282 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
283 			CPUSET_ATOMIC_DEL(cp->cp_haltset, cpun);
284 		}
285 		sti();
286 		return;
287 	}
288 
289 	/*
290 	 * Call the halt sequence:
291 	 * sti
292 	 * hlt
293 	 */
294 	i86_halt();
295 
296 	/*
297 	 * We're no longer halted
298 	 */
299 	if (hset_update) {
300 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
301 		CPUSET_ATOMIC_DEL(cp->cp_haltset, cpun);
302 	}
303 }
304 
305 
306 /*
307  * If "cpu" is halted, then wake it up clearing its halted bit in advance.
308  * Otherwise, see if other CPUs in the cpu partition are halted and need to
309  * be woken up so that they can steal the thread we placed on this CPU.
310  * This function is only used on MP systems.
311  */
312 static void
313 cpu_wakeup(cpu_t *cpu, int bound)
314 {
315 	uint_t		cpu_found;
316 	int		result;
317 	cpupart_t	*cp;
318 
319 	cp = cpu->cpu_part;
320 	if (CPU_IN_SET(cp->cp_haltset, cpu->cpu_id)) {
321 		/*
322 		 * Clear the halted bit for that CPU since it will be
323 		 * poked in a moment.
324 		 */
325 		CPUSET_ATOMIC_DEL(cp->cp_haltset, cpu->cpu_id);
326 		/*
327 		 * We may find the current CPU present in the halted cpuset
328 		 * if we're in the context of an interrupt that occurred
329 		 * before we had a chance to clear our bit in cpu_halt().
330 		 * Poking ourself is obviously unnecessary, since if
331 		 * we're here, we're not halted.
332 		 */
333 		if (cpu != CPU)
334 			poke_cpu(cpu->cpu_id);
335 		return;
336 	} else {
337 		/*
338 		 * This cpu isn't halted, but it's idle or undergoing a
339 		 * context switch. No need to awaken anyone else.
340 		 */
341 		if (cpu->cpu_thread == cpu->cpu_idle_thread ||
342 		    cpu->cpu_disp_flags & CPU_DISP_DONTSTEAL)
343 			return;
344 	}
345 
346 	/*
347 	 * No need to wake up other CPUs if the thread we just enqueued
348 	 * is bound.
349 	 */
350 	if (bound)
351 		return;
352 
353 
354 	/*
355 	 * See if there's any other halted CPUs. If there are, then
356 	 * select one, and awaken it.
357 	 * It's possible that after we find a CPU, somebody else
358 	 * will awaken it before we get the chance.
359 	 * In that case, look again.
360 	 */
361 	do {
362 		CPUSET_FIND(cp->cp_haltset, cpu_found);
363 		if (cpu_found == CPUSET_NOTINSET)
364 			return;
365 
366 		ASSERT(cpu_found >= 0 && cpu_found < NCPU);
367 		CPUSET_ATOMIC_XDEL(cp->cp_haltset, cpu_found, result);
368 	} while (result < 0);
369 
370 	if (cpu_found != CPU->cpu_id)
371 		poke_cpu(cpu_found);
372 }
373 
374 static int
375 mp_disable_intr(int cpun)
376 {
377 	/*
378 	 * switch to the offline cpu
379 	 */
380 	affinity_set(cpun);
381 	/*
382 	 * raise ipl to just below cross call
383 	 */
384 	splx(XC_MED_PIL-1);
385 	/*
386 	 *	set base spl to prevent the next swtch to idle from
387 	 *	lowering back to ipl 0
388 	 */
389 	CPU->cpu_intr_actv |= (1 << (XC_MED_PIL-1));
390 	set_base_spl();
391 	affinity_clear();
392 	return (DDI_SUCCESS);
393 }
394 
395 static void
396 mp_enable_intr(int cpun)
397 {
398 	/*
399 	 * switch to the online cpu
400 	 */
401 	affinity_set(cpun);
402 	/*
403 	 * clear the interrupt active mask
404 	 */
405 	CPU->cpu_intr_actv &= ~(1 << (XC_MED_PIL-1));
406 	set_base_spl();
407 	(void) spl0();
408 	affinity_clear();
409 }
410 
411 static void
412 mach_get_platform(int owner)
413 {
414 	void		**srv_opsp;
415 	void		**clt_opsp;
416 	int		i;
417 	int		total_ops;
418 
419 	/* fix up psm ops */
420 	srv_opsp = (void **)mach_set[0];
421 	clt_opsp = (void **)mach_set[owner];
422 	if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01)
423 		total_ops = sizeof (struct psm_ops_ver01) /
424 				sizeof (void (*)(void));
425 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_1)
426 		/* no psm_notify_func */
427 		total_ops = OFFSETOF(struct psm_ops, psm_notify_func) /
428 		    sizeof (void (*)(void));
429 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_2)
430 		/* no psm_timer funcs */
431 		total_ops = OFFSETOF(struct psm_ops, psm_timer_reprogram) /
432 		    sizeof (void (*)(void));
433 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_3)
434 		/* no psm_preshutdown function */
435 		total_ops = OFFSETOF(struct psm_ops, psm_preshutdown) /
436 		    sizeof (void (*)(void));
437 	else if (mach_ver[owner] == (ushort_t)PSM_INFO_VER01_4)
438 		/* no psm_preshutdown function */
439 		total_ops = OFFSETOF(struct psm_ops, psm_intr_ops) /
440 		    sizeof (void (*)(void));
441 	else
442 		total_ops = sizeof (struct psm_ops) / sizeof (void (*)(void));
443 
444 	/*
445 	 * Save the version of the PSM module, in case we need to
446 	 * bahave differently based on version.
447 	 */
448 	mach_ver[0] = mach_ver[owner];
449 
450 	for (i = 0; i < total_ops; i++)
451 		if (clt_opsp[i] != NULL)
452 			srv_opsp[i] = clt_opsp[i];
453 }
454 
455 static void
456 mach_construct_info()
457 {
458 	register struct psm_sw *swp;
459 	int	mach_cnt[PSM_OWN_OVERRIDE+1] = {0};
460 	int	conflict_owner = 0;
461 
462 	if (psmsw->psw_forw == psmsw)
463 		panic("No valid PSM modules found");
464 	mutex_enter(&psmsw_lock);
465 	for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
466 		if (!(swp->psw_flag & PSM_MOD_IDENTIFY))
467 			continue;
468 		mach_set[swp->psw_infop->p_owner] = swp->psw_infop->p_ops;
469 		mach_ver[swp->psw_infop->p_owner] = swp->psw_infop->p_version;
470 		mach_cnt[swp->psw_infop->p_owner]++;
471 	}
472 	mutex_exit(&psmsw_lock);
473 
474 	mach_get_platform(PSM_OWN_SYS_DEFAULT);
475 
476 	/* check to see are there any conflicts */
477 	if (mach_cnt[PSM_OWN_EXCLUSIVE] > 1)
478 		conflict_owner = PSM_OWN_EXCLUSIVE;
479 	if (mach_cnt[PSM_OWN_OVERRIDE] > 1)
480 		conflict_owner = PSM_OWN_OVERRIDE;
481 	if (conflict_owner) {
482 		/* remove all psm modules except uppc */
483 		cmn_err(CE_WARN,
484 			"Conflicts detected on the following PSM modules:");
485 		mutex_enter(&psmsw_lock);
486 		for (swp = psmsw->psw_forw; swp != psmsw; swp = swp->psw_forw) {
487 			if (swp->psw_infop->p_owner == conflict_owner)
488 				cmn_err(CE_WARN, "%s ",
489 					swp->psw_infop->p_mach_idstring);
490 		}
491 		mutex_exit(&psmsw_lock);
492 		cmn_err(CE_WARN,
493 			"Setting the system back to SINGLE processor mode!");
494 		cmn_err(CE_WARN,
495 		    "Please edit /etc/mach to remove the invalid PSM module.");
496 		return;
497 	}
498 
499 	if (mach_set[PSM_OWN_EXCLUSIVE])
500 		mach_get_platform(PSM_OWN_EXCLUSIVE);
501 
502 	if (mach_set[PSM_OWN_OVERRIDE])
503 		mach_get_platform(PSM_OWN_OVERRIDE);
504 }
505 
506 static void
507 mach_init()
508 {
509 	register struct psm_ops  *pops;
510 
511 	mach_construct_info();
512 
513 	pops = mach_set[0];
514 
515 	/* register the interrupt and clock initialization rotuines */
516 	picinitf = mach_picinit;
517 	clkinitf = mach_clkinit;
518 	psm_get_clockirq = pops->psm_get_clockirq;
519 
520 	/* register the interrupt setup code */
521 	slvltovect = mach_softlvl_to_vect;
522 	addspl	= pops->psm_addspl;
523 	delspl	= pops->psm_delspl;
524 
525 	if (pops->psm_translate_irq)
526 		psm_translate_irq = pops->psm_translate_irq;
527 	if (pops->psm_intr_ops)
528 		psm_intr_ops = pops->psm_intr_ops;
529 	if (pops->psm_tod_get) {
530 		todgetf = mach_tod_get;
531 		psm_todgetf = pops->psm_tod_get;
532 	}
533 	if (pops->psm_tod_set) {
534 		todsetf = mach_tod_set;
535 		psm_todsetf = pops->psm_tod_set;
536 	}
537 	if (pops->psm_notify_error) {
538 		psm_notify_error = mach_notify_error;
539 		notify_error = pops->psm_notify_error;
540 	}
541 
542 	(*pops->psm_softinit)();
543 
544 	/*
545 	 * Initialize the dispatcher's function hooks
546 	 * to enable CPU halting when idle
547 	 */
548 #if defined(_SIMULATOR_SUPPORT)
549 	if (halt_idle_cpus && !simulator_run)
550 		idle_cpu = cpu_halt;
551 #else
552 	if (halt_idle_cpus)
553 		idle_cpu = cpu_halt;
554 #endif	/* _SIMULATOR_SUPPORT */
555 
556 	mach_smpinit();
557 }
558 
559 static void
560 mach_smpinit(void)
561 {
562 	register struct psm_ops  *pops;
563 	register processorid_t cpu_id;
564 	int	 cnt;
565 	int	 cpumask;
566 
567 	pops = mach_set[0];
568 
569 	cpu_id = -1;
570 	cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
571 	for (cnt = 0, cpumask = 0; cpu_id != -1; cnt++) {
572 		cpumask |= 1 << cpu_id;
573 		cpu_id = (*pops->psm_get_next_processorid)(cpu_id);
574 	}
575 
576 	mp_cpus = cpumask;
577 
578 	/* MP related routines */
579 	cpu_startf = mach_cpu_start;
580 	ap_mlsetup = pops->psm_post_cpu_start;
581 	send_dirintf = pops->psm_send_ipi;
582 
583 	/* optional MP related routines */
584 	if (pops->psm_shutdown)
585 		psm_shutdownf = pops->psm_shutdown;
586 	if (pops->psm_preshutdown)
587 		psm_preshutdownf = pops->psm_preshutdown;
588 	if (pops->psm_notify_func)
589 		psm_notifyf = pops->psm_notify_func;
590 	if (pops->psm_set_idlecpu)
591 		psm_set_idle_cpuf = pops->psm_set_idlecpu;
592 	if (pops->psm_unset_idlecpu)
593 		psm_unset_idle_cpuf = pops->psm_unset_idlecpu;
594 
595 	psm_clkinit = pops->psm_clkinit;
596 
597 	if (pops->psm_timer_reprogram)
598 		psm_timer_reprogram = pops->psm_timer_reprogram;
599 
600 	if (pops->psm_timer_enable)
601 		psm_timer_enable = pops->psm_timer_enable;
602 
603 	if (pops->psm_timer_disable)
604 		psm_timer_disable = pops->psm_timer_disable;
605 
606 	if (pops->psm_post_cyclic_setup)
607 		psm_post_cyclic_setup = pops->psm_post_cyclic_setup;
608 
609 	/* check for multiple cpu's */
610 	if (cnt < 2)
611 		return;
612 
613 	/* check for MP platforms */
614 	if (pops->psm_cpu_start == NULL)
615 		return;
616 
617 	/*
618 	 * Set the dispatcher hook to enable cpu "wake up"
619 	 * when a thread becomes runnable.
620 	 */
621 #if defined(_SIMULATOR_SUPPORT)
622 	if (halt_idle_cpus && !simulator_run) {
623 		disp_enq_thread = cpu_wakeup;
624 	}
625 #else
626 	if (halt_idle_cpus) {
627 		disp_enq_thread = cpu_wakeup;
628 	}
629 #endif	/* _SIMULATOR_SUPPORT */
630 
631 	if (pops->psm_disable_intr)
632 		psm_disable_intr = pops->psm_disable_intr;
633 	if (pops->psm_enable_intr)
634 		psm_enable_intr  = pops->psm_enable_intr;
635 
636 	psm_get_ipivect = pops->psm_get_ipivect;
637 
638 	(void) add_avintr((void *)NULL, XC_HI_PIL, xc_serv, "xc_hi_intr",
639 		(*pops->psm_get_ipivect)(XC_HI_PIL, PSM_INTR_IPI_HI),
640 		(caddr_t)X_CALL_HIPRI, NULL, NULL);
641 	(void) add_avintr((void *)NULL, XC_MED_PIL, xc_serv, "xc_med_intr",
642 		(*pops->psm_get_ipivect)(XC_MED_PIL, PSM_INTR_IPI_LO),
643 		(caddr_t)X_CALL_MEDPRI, NULL, NULL);
644 
645 	(void) (*pops->psm_get_ipivect)(XC_CPUPOKE_PIL, PSM_INTR_POKE);
646 }
647 
648 static void
649 mach_picinit()
650 {
651 	register struct psm_ops  *pops;
652 	extern void install_spl(void);	/* XXX: belongs in a header file */
653 #if defined(__amd64) && defined(DEBUG)
654 	extern void *spl_patch, *slow_spl, *setsplhi_patch, *slow_setsplhi;
655 #endif
656 
657 	pops = mach_set[0];
658 
659 	/* register the interrupt handlers */
660 	setlvl = pops->psm_intr_enter;
661 	setlvlx = pops->psm_intr_exit;
662 
663 	/* initialize the interrupt hardware */
664 	(*pops->psm_picinit)();
665 
666 	/* set interrupt mask for current ipl */
667 	setspl = pops->psm_setspl;
668 	setspl(CPU->cpu_pri);
669 
670 	/* Install proper spl routine now that we can Program the PIC   */
671 #if defined(__amd64)
672 	/*
673 	 * It would be better if we could check this at compile time
674 	 */
675 	ASSERT(((uintptr_t)&slow_setsplhi - (uintptr_t)&setsplhi_patch < 128) &&
676 		((uintptr_t)&slow_spl - (uintptr_t)&spl_patch < 128));
677 #endif
678 	install_spl();
679 }
680 
681 uint_t	cpu_freq;	/* MHz */
682 uint64_t cpu_freq_hz;	/* measured (in hertz) */
683 
684 #define	MEGA_HZ		1000000
685 
686 static uint64_t
687 mach_calchz(uint32_t pit_counter, uint64_t *processor_clks)
688 {
689 	uint64_t cpu_hz;
690 
691 	if ((pit_counter == 0) || (*processor_clks == 0) ||
692 	    (*processor_clks > (((uint64_t)-1) / PIT_HZ)))
693 		return (0);
694 
695 	cpu_hz = ((uint64_t)PIT_HZ * *processor_clks) / pit_counter;
696 
697 	return (cpu_hz);
698 }
699 
700 static uint64_t
701 mach_getcpufreq(void)
702 {
703 	uint32_t pit_counter;
704 	uint64_t processor_clks;
705 
706 	if (x86_feature & X86_TSC) {
707 		/*
708 		 * We have a TSC. freq_tsc() knows how to measure the number
709 		 * of clock cycles sampled against the PIT.
710 		 */
711 		processor_clks = freq_tsc(&pit_counter);
712 		return (mach_calchz(pit_counter, &processor_clks));
713 	} else if (x86_vendor == X86_VENDOR_Cyrix || x86_type == X86_TYPE_P5) {
714 #if defined(__amd64)
715 		panic("mach_getcpufreq: no TSC!");
716 #elif defined(__i386)
717 		/*
718 		 * We are a Cyrix based on a 6x86 core or an Intel Pentium
719 		 * for which freq_notsc() knows how to measure the number of
720 		 * elapsed clock cycles sampled against the PIT
721 		 */
722 		processor_clks = freq_notsc(&pit_counter);
723 		return (mach_calchz(pit_counter, &processor_clks));
724 #endif	/* __i386 */
725 	}
726 
727 	/* We do not know how to calculate cpu frequency for this cpu. */
728 	return (0);
729 }
730 
731 /*
732  * If the clock speed of a cpu is found to be reported incorrectly, do not add
733  * to this array, instead improve the accuracy of the algorithm that determines
734  * the clock speed of the processor or extend the implementation to support the
735  * vendor as appropriate. This is here only to support adjusting the speed on
736  * older slower processors that mach_fixcpufreq() would not be able to account
737  * for otherwise.
738  */
739 static int x86_cpu_freq[] = { 60, 75, 80, 90, 120, 160, 166, 175, 180, 233 };
740 
741 /*
742  * On fast processors the clock frequency that is measured may be off by
743  * a few MHz from the value printed on the part. This is a combination of
744  * the factors that for such fast parts being off by this much is within
745  * the tolerances for manufacture and because of the difficulties in the
746  * measurement that can lead to small error. This function uses some
747  * heuristics in order to tweak the value that was measured to match what
748  * is most likely printed on the part.
749  *
750  * Some examples:
751  * 	AMD Athlon 1000 mhz measured as 998 mhz
752  * 	Intel Pentium III Xeon 733 mhz measured as 731 mhz
753  * 	Intel Pentium IV 1500 mhz measured as 1495mhz
754  *
755  * If in the future this function is no longer sufficient to correct
756  * for the error in the measurement, then the algorithm used to perform
757  * the measurement will have to be improved in order to increase accuracy
758  * rather than adding horrible and questionable kludges here.
759  *
760  * This is called after the cyclics subsystem because of the potential
761  * that the heuristics within may give a worse estimate of the clock
762  * frequency than the value that was measured.
763  */
764 static void
765 mach_fixcpufreq(void)
766 {
767 	uint32_t freq, mul, near66, delta66, near50, delta50, fixed, delta, i;
768 
769 	freq = (uint32_t)cpu_freq;
770 
771 	/*
772 	 * Find the nearest integer multiple of 200/3 (about 66) MHz to the
773 	 * measured speed taking into account that the 667 MHz parts were
774 	 * the first to round-up.
775 	 */
776 	mul = (uint32_t)((3 * (uint64_t)freq + 100) / 200);
777 	near66 = (uint32_t)((200 * (uint64_t)mul + ((mul >= 10) ? 1 : 0)) / 3);
778 	delta66 = (near66 > freq) ? (near66 - freq) : (freq - near66);
779 
780 	/* Find the nearest integer multiple of 50 MHz to the measured speed */
781 	mul = (freq + 25) / 50;
782 	near50 = mul * 50;
783 	delta50 = (near50 > freq) ? (near50 - freq) : (freq - near50);
784 
785 	/* Find the closer of the two */
786 	if (delta66 < delta50) {
787 		fixed = near66;
788 		delta = delta66;
789 	} else {
790 		fixed = near50;
791 		delta = delta50;
792 	}
793 
794 	if (fixed > INT_MAX)
795 		return;
796 
797 	/*
798 	 * Some older parts have a core clock frequency that is not an
799 	 * integral multiple of 50 or 66 MHz. Check if one of the old
800 	 * clock frequencies is closer to the measured value than any
801 	 * of the integral multiples of 50 an 66, and if so set fixed
802 	 * and delta appropriately to represent the closest value.
803 	 */
804 	i = sizeof (x86_cpu_freq) / sizeof (int);
805 	while (i > 0) {
806 		i--;
807 
808 		if (x86_cpu_freq[i] <= freq) {
809 			mul = freq - x86_cpu_freq[i];
810 
811 			if (mul < delta) {
812 				fixed = x86_cpu_freq[i];
813 				delta = mul;
814 			}
815 
816 			break;
817 		}
818 
819 		mul = x86_cpu_freq[i] - freq;
820 
821 		if (mul < delta) {
822 			fixed = x86_cpu_freq[i];
823 			delta = mul;
824 		}
825 	}
826 
827 	/*
828 	 * Set a reasonable maximum for how much to correct the measured
829 	 * result by. This check is here to prevent the adjustment made
830 	 * by this function from being more harm than good. It is entirely
831 	 * possible that in the future parts will be made that are not
832 	 * integral multiples of 66 or 50 in clock frequency or that
833 	 * someone may overclock a part to some odd frequency. If the
834 	 * measured value is farther from the corrected value than
835 	 * allowed, then assume the corrected value is in error and use
836 	 * the measured value.
837 	 */
838 	if (6 < delta)
839 		return;
840 
841 	cpu_freq = (int)fixed;
842 }
843 
844 
845 static int
846 machhztomhz(uint64_t cpu_freq_hz)
847 {
848 	uint64_t cpu_mhz;
849 
850 	/* Round to nearest MHZ */
851 	cpu_mhz = (cpu_freq_hz + (MEGA_HZ / 2)) / MEGA_HZ;
852 
853 	if (cpu_mhz > INT_MAX)
854 		return (0);
855 
856 	return ((int)cpu_mhz);
857 
858 }
859 
860 
861 static int
862 mach_clkinit(int preferred_mode, int *set_mode)
863 {
864 	register struct psm_ops  *pops;
865 	int resolution;
866 
867 	pops = mach_set[0];
868 
869 #ifdef	_SIMULATOR_SUPPORT
870 	if (!simulator_run)
871 		cpu_freq_hz = mach_getcpufreq();
872 	else
873 		cpu_freq_hz = 40000000; /* use 40 Mhz (hack for simulator) */
874 #else
875 	cpu_freq_hz = mach_getcpufreq();
876 #endif	/* _SIMULATOR_SUPPORT */
877 
878 	cpu_freq = machhztomhz(cpu_freq_hz);
879 
880 	if (!(x86_feature & X86_TSC) || (cpu_freq == 0))
881 		tsc_gethrtime_enable = 0;
882 
883 	if (tsc_gethrtime_enable) {
884 		tsc_hrtimeinit(cpu_freq_hz);
885 		gethrtimef = tsc_gethrtime;
886 		gethrtimeunscaledf = tsc_gethrtimeunscaled;
887 		scalehrtimef = tsc_scalehrtime;
888 		hrtime_tick = tsc_tick;
889 		tsc_gethrtime_initted = 1;
890 	} else {
891 		if (pops->psm_hrtimeinit)
892 			(*pops->psm_hrtimeinit)();
893 		gethrtimef = pops->psm_gethrtime;
894 		gethrtimeunscaledf = gethrtimef;
895 		/* scalehrtimef will remain dummy */
896 	}
897 
898 	mach_fixcpufreq();
899 
900 	if (mach_ver[0] >= PSM_INFO_VER01_3) {
901 		if ((preferred_mode == TIMER_ONESHOT) &&
902 		    (tsc_gethrtime_enable)) {
903 
904 			resolution = (*pops->psm_clkinit)(0);
905 			if (resolution != 0)  {
906 				*set_mode = TIMER_ONESHOT;
907 				return (resolution);
908 			}
909 
910 		}
911 
912 		/*
913 		 * either periodic mode was requested or could not set to
914 		 * one-shot mode
915 		 */
916 		resolution = (*pops->psm_clkinit)(hz);
917 		/*
918 		 * psm should be able to do periodic, so we do not check
919 		 * for return value of psm_clkinit here.
920 		 */
921 		*set_mode = TIMER_PERIODIC;
922 		return (resolution);
923 	} else {
924 		/*
925 		 * PSMI interface prior to PSMI_3 does not define a return
926 		 * value for psm_clkinit, so the return value is ignored.
927 		 */
928 		(void) (*pops->psm_clkinit)(hz);
929 		*set_mode = TIMER_PERIODIC;
930 		return (nsec_per_tick);
931 	}
932 }
933 
934 static int
935 mach_softlvl_to_vect(register int ipl)
936 {
937 	register int softvect;
938 	register struct psm_ops  *pops;
939 
940 	pops = mach_set[0];
941 
942 	/* check for null handler for set soft interrupt call		*/
943 	if (pops->psm_set_softintr == NULL) {
944 		setsoftint = set_pending;
945 		return (PSM_SV_SOFTWARE);
946 	}
947 
948 	softvect = (*pops->psm_softlvl_to_irq)(ipl);
949 	/* check for hardware scheme					*/
950 	if (softvect > PSM_SV_SOFTWARE) {
951 		setsoftint = pops->psm_set_softintr;
952 		return (softvect);
953 	}
954 
955 	if (softvect == PSM_SV_SOFTWARE)
956 		setsoftint = set_pending;
957 	else	/* hardware and software mixed scheme			*/
958 		setsoftint = mach_set_softintr;
959 
960 	return (PSM_SV_SOFTWARE);
961 }
962 
963 static void
964 mach_set_softintr(register int ipl)
965 {
966 	register struct psm_ops  *pops;
967 
968 	/* set software pending bits					*/
969 	set_pending(ipl);
970 
971 	/*	check if dosoftint will be called at the end of intr	*/
972 	if (CPU_ON_INTR(CPU) || (curthread->t_intr))
973 		return;
974 
975 	/* invoke hardware interrupt					*/
976 	pops = mach_set[0];
977 	(*pops->psm_set_softintr)(ipl);
978 }
979 
980 static void
981 mach_cpu_start(register int cpun)
982 {
983 	register struct psm_ops  *pops;
984 	int	i;
985 
986 	pops = mach_set[0];
987 
988 	(*pops->psm_cpu_start)(cpun, rm_platter_va);
989 
990 	/* wait for the auxillary cpu to be ready			*/
991 	for (i = 20000; i; i--) {
992 		if (cpu[cpun]->cpu_flags & CPU_READY)
993 			return;
994 		drv_usecwait(100);
995 	}
996 }
997 
998 /*ARGSUSED*/
999 static int
1000 mach_translate_irq(dev_info_t *dip, int irqno)
1001 {
1002 	return (irqno);	/* default to NO translation */
1003 }
1004 
1005 static timestruc_t
1006 mach_tod_get(void)
1007 {
1008 	timestruc_t ts;
1009 	todinfo_t tod;
1010 	static int mach_range_warn = 1;	/* warn only once */
1011 
1012 	ASSERT(MUTEX_HELD(&tod_lock));
1013 
1014 	/* The year returned from is the last 2 digit only */
1015 	if ((*psm_todgetf)(&tod)) {
1016 		ts.tv_sec = 0;
1017 		ts.tv_nsec = 0;
1018 		tod_fault_reset();
1019 		return (ts);
1020 	}
1021 
1022 	/* assume that we wrap the rtc year back to zero at 2000 */
1023 	if (tod.tod_year < 69) {
1024 		if (mach_range_warn && tod.tod_year > 38) {
1025 			cmn_err(CE_WARN, "hardware real-time clock is out "
1026 				"of range -- time needs to be reset");
1027 			mach_range_warn = 0;
1028 		}
1029 		tod.tod_year += 100;
1030 	}
1031 
1032 	/* tod_to_utc uses 1900 as base for the year */
1033 	ts.tv_sec = tod_to_utc(tod) + gmt_lag;
1034 	ts.tv_nsec = 0;
1035 
1036 	return (ts);
1037 }
1038 
1039 static void
1040 mach_tod_set(timestruc_t ts)
1041 {
1042 	todinfo_t tod = utc_to_tod(ts.tv_sec - gmt_lag);
1043 
1044 	ASSERT(MUTEX_HELD(&tod_lock));
1045 
1046 	if (tod.tod_year >= 100)
1047 		tod.tod_year -= 100;
1048 
1049 	(*psm_todsetf)(&tod);
1050 }
1051 
1052 static void
1053 mach_notify_error(int level, char *errmsg)
1054 {
1055 	/*
1056 	 * SL_FATAL is pass in once panicstr is set, deliver it
1057 	 * as CE_PANIC.  Also, translate SL_ codes back to CE_
1058 	 * codes for the psmi handler
1059 	 */
1060 	if (level & SL_FATAL)
1061 		(*notify_error)(CE_PANIC, errmsg);
1062 	else if (level & SL_WARN)
1063 		(*notify_error)(CE_WARN, errmsg);
1064 	else if (level & SL_NOTE)
1065 		(*notify_error)(CE_NOTE, errmsg);
1066 	else if (level & SL_CONSOLE)
1067 		(*notify_error)(CE_CONT, errmsg);
1068 }
1069 
1070 /*
1071  * It provides the default basic intr_ops interface for the new DDI
1072  * interrupt framework if the PSM doesn't have one.
1073  *
1074  * Input:
1075  * dip     - pointer to the dev_info structure of the requested device
1076  * hdlp    - pointer to the internal interrupt handle structure for the
1077  *	     requested interrupt
1078  * intr_op - opcode for this call
1079  * result  - pointer to the integer that will hold the result to be
1080  *	     passed back if return value is PSM_SUCCESS
1081  *
1082  * Output:
1083  * return value is either PSM_SUCCESS or PSM_FAILURE
1084  */
1085 static int
1086 mach_intr_ops(dev_info_t *dip, ddi_intr_handle_impl_t *hdlp,
1087     psm_intr_op_t intr_op, int *result)
1088 {
1089 	struct intrspec *ispec;
1090 
1091 	switch (intr_op) {
1092 	case PSM_INTR_OP_CHECK_MSI:
1093 		*result = hdlp->ih_type & ~(DDI_INTR_TYPE_MSI |
1094 			    DDI_INTR_TYPE_MSIX);
1095 		break;
1096 	case PSM_INTR_OP_ALLOC_VECTORS:
1097 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1098 			*result = 1;
1099 		else
1100 			*result = 0;
1101 		break;
1102 	case PSM_INTR_OP_FREE_VECTORS:
1103 		break;
1104 	case PSM_INTR_OP_NAVAIL_VECTORS:
1105 		if (hdlp->ih_type == DDI_INTR_TYPE_FIXED)
1106 			*result = 1;
1107 		else
1108 			*result = 0;
1109 		break;
1110 	case PSM_INTR_OP_XLATE_VECTOR:
1111 		ispec = (struct intrspec *)hdlp->ih_private;
1112 		*result = psm_translate_irq(dip, ispec->intrspec_vec);
1113 		break;
1114 	case PSM_INTR_OP_GET_CAP:
1115 		*result = 0;
1116 		break;
1117 	case PSM_INTR_OP_GET_PENDING:
1118 	case PSM_INTR_OP_CLEAR_MASK:
1119 	case PSM_INTR_OP_SET_MASK:
1120 	case PSM_INTR_OP_GET_SHARED:
1121 	case PSM_INTR_OP_SET_PRI:
1122 	case PSM_INTR_OP_SET_CAP:
1123 	default:
1124 		return (PSM_FAILURE);
1125 	}
1126 	return (PSM_SUCCESS);
1127 }
1128