xref: /netbsd-src/sys/arch/x86/acpi/acpi_cpu_md.c (revision b1c86f5f087524e68db12794ee9c3e3da1ab17a0)
1 /* $NetBSD: acpi_cpu_md.c,v 1.34 2010/08/25 05:07:43 jruoho Exp $ */
2 
3 /*-
4  * Copyright (c) 2010 Jukka Ruohonen <jruohonen@iki.fi>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 #include <sys/cdefs.h>
30 __KERNEL_RCSID(0, "$NetBSD: acpi_cpu_md.c,v 1.34 2010/08/25 05:07:43 jruoho Exp $");
31 
32 #include <sys/param.h>
33 #include <sys/bus.h>
34 #include <sys/kcore.h>
35 #include <sys/sysctl.h>
36 #include <sys/xcall.h>
37 
38 #include <x86/cpu.h>
39 #include <x86/cpufunc.h>
40 #include <x86/cputypes.h>
41 #include <x86/cpuvar.h>
42 #include <x86/cpu_msr.h>
43 #include <x86/machdep.h>
44 
45 #include <dev/acpi/acpica.h>
46 #include <dev/acpi/acpi_cpu.h>
47 
48 #include <dev/pci/pcivar.h>
49 #include <dev/pci/pcidevs.h>
50 
51 #define ACPICPU_P_STATE_STATUS	0
52 
53 /*
54  * AMD families 10h and 11h.
55  */
56 #define MSR_10H_LIMIT		0xc0010061
57 #define MSR_10H_CONTROL		0xc0010062
58 #define MSR_10H_STATUS		0xc0010063
59 #define MSR_10H_CONFIG		0xc0010064
60 
61 /*
62  * AMD family 0Fh.
63  */
64 #define MSR_0FH_CONTROL		0xc0010041
65 #define MSR_0FH_STATUS		0xc0010042
66 
67 #define MSR_0FH_STATUS_CFID	__BITS( 0,  5)
68 #define MSR_0FH_STATUS_CVID	__BITS(32, 36)
69 #define MSR_0FH_STATUS_PENDING	__BITS(31, 31)
70 
71 #define MSR_0FH_CONTROL_FID	__BITS( 0,  5)
72 #define MSR_0FH_CONTROL_VID	__BITS( 8, 12)
73 #define MSR_0FH_CONTROL_CHG	__BITS(16, 16)
74 #define MSR_0FH_CONTROL_CNT	__BITS(32, 51)
75 
76 #define ACPI_0FH_STATUS_FID	__BITS( 0,  5)
77 #define ACPI_0FH_STATUS_VID	__BITS( 6, 10)
78 
79 #define ACPI_0FH_CONTROL_FID	__BITS( 0,  5)
80 #define ACPI_0FH_CONTROL_VID	__BITS( 6, 10)
81 #define ACPI_0FH_CONTROL_VST	__BITS(11, 17)
82 #define ACPI_0FH_CONTROL_MVS	__BITS(18, 19)
83 #define ACPI_0FH_CONTROL_PLL	__BITS(20, 26)
84 #define ACPI_0FH_CONTROL_RVO	__BITS(28, 29)
85 #define ACPI_0FH_CONTROL_IRT	__BITS(30, 31)
86 
87 #define FID_TO_VCO_FID(fidd)	(((fid) < 8) ? (8 + ((fid) << 1)) : (fid))
88 
89 static char	  native_idle_text[16];
90 void		(*native_idle)(void) = NULL;
91 
92 static int	 acpicpu_md_quirks_piix4(struct pci_attach_args *);
93 static void	 acpicpu_md_pstate_status(void *, void *);
94 static int	 acpicpu_md_pstate_fidvid_get(struct acpicpu_softc *,
95                                               uint32_t *);
96 static int	 acpicpu_md_pstate_fidvid_set(struct acpicpu_pstate *);
97 static int	 acpicpu_md_pstate_fidvid_read(uint32_t *, uint32_t *);
98 static void	 acpicpu_md_pstate_fidvid_write(uint32_t, uint32_t,
99 					        uint32_t, uint32_t);
100 static void	 acpicpu_md_tstate_status(void *, void *);
101 static int	 acpicpu_md_pstate_sysctl_init(void);
102 static int	 acpicpu_md_pstate_sysctl_get(SYSCTLFN_PROTO);
103 static int	 acpicpu_md_pstate_sysctl_set(SYSCTLFN_PROTO);
104 static int	 acpicpu_md_pstate_sysctl_all(SYSCTLFN_PROTO);
105 
106 extern uint32_t cpus_running;
107 extern struct acpicpu_softc **acpicpu_sc;
108 static struct sysctllog *acpicpu_log = NULL;
109 
110 uint32_t
111 acpicpu_md_cap(void)
112 {
113 	struct cpu_info *ci = curcpu();
114 	uint32_t val = 0;
115 
116 	if (cpu_vendor != CPUVENDOR_IDT &&
117 	    cpu_vendor != CPUVENDOR_INTEL)
118 		return val;
119 
120 	/*
121 	 * Basic SMP C-states (required for _CST).
122 	 */
123 	val |= ACPICPU_PDC_C_C1PT | ACPICPU_PDC_C_C2C3;
124 
125         /*
126 	 * If MONITOR/MWAIT is available, announce
127 	 * support for native instructions in all C-states.
128 	 */
129         if ((ci->ci_feat_val[1] & CPUID2_MONITOR) != 0)
130 		val |= ACPICPU_PDC_C_C1_FFH | ACPICPU_PDC_C_C2C3_FFH;
131 
132 	/*
133 	 * Set native P- and T-states, if available.
134 	 */
135         if ((ci->ci_feat_val[1] & CPUID2_EST) != 0)
136 		val |= ACPICPU_PDC_P_FFH;
137 
138 	if ((ci->ci_feat_val[0] & CPUID_ACPI) != 0)
139 		val |= ACPICPU_PDC_T_FFH;
140 
141 	return val;
142 }
143 
144 uint32_t
145 acpicpu_md_quirks(void)
146 {
147 	struct cpu_info *ci = curcpu();
148 	struct pci_attach_args pa;
149 	uint32_t family, val = 0;
150 	uint32_t regs[4];
151 
152 	if (acpicpu_md_cpus_running() == 1)
153 		val |= ACPICPU_FLAG_C_BM;
154 
155 	if ((ci->ci_feat_val[1] & CPUID2_MONITOR) != 0)
156 		val |= ACPICPU_FLAG_C_FFH;
157 
158 	val |= ACPICPU_FLAG_C_APIC | ACPICPU_FLAG_C_TSC;
159 
160 	switch (cpu_vendor) {
161 
162 	case CPUVENDOR_IDT:
163 
164 		if ((ci->ci_feat_val[1] & CPUID2_EST) != 0)
165 			val |= ACPICPU_FLAG_P_FFH;
166 
167 		if ((ci->ci_feat_val[0] & CPUID_ACPI) != 0)
168 			val |= ACPICPU_FLAG_T_FFH;
169 
170 		break;
171 
172 	case CPUVENDOR_INTEL:
173 
174 		val |= ACPICPU_FLAG_C_BM | ACPICPU_FLAG_C_ARB;
175 
176 		if ((ci->ci_feat_val[1] & CPUID2_EST) != 0)
177 			val |= ACPICPU_FLAG_P_FFH;
178 
179 		if ((ci->ci_feat_val[0] & CPUID_ACPI) != 0)
180 			val |= ACPICPU_FLAG_T_FFH;
181 
182 		/*
183 		 * Check whether MSR_APERF, MSR_MPERF, and Turbo
184 		 * Boost are available. Also see if we might have
185 		 * an invariant local APIC timer ("ARAT").
186 		 */
187 		if (cpuid_level >= 0x06) {
188 
189 			x86_cpuid(0x06, regs);
190 
191 			if ((regs[2] & CPUID_DSPM_HWF) != 0)
192 				val |= ACPICPU_FLAG_P_HW;
193 
194 			if ((regs[0] & CPUID_DSPM_IDA) != 0)
195 				val |= ACPICPU_FLAG_P_TURBO;
196 
197 			if ((regs[0] & CPUID_DSPM_ARAT) != 0)
198 				val &= ~ACPICPU_FLAG_C_APIC;
199 		}
200 
201 		/*
202 		 * Detect whether TSC is invariant. If it is not,
203 		 * we keep the flag to note that TSC will not run
204 		 * at constant rate. Depending on the CPU, this may
205 		 * affect P- and T-state changes, but especially
206 		 * relevant are C-states; with variant TSC, states
207 		 * larger than C1 may completely stop the counter.
208 		 */
209 		x86_cpuid(0x80000000, regs);
210 
211 		if (regs[0] >= 0x80000007) {
212 
213 			x86_cpuid(0x80000007, regs);
214 
215 			if ((regs[3] & __BIT(8)) != 0)
216 				val &= ~ACPICPU_FLAG_C_TSC;
217 		}
218 
219 		break;
220 
221 	case CPUVENDOR_AMD:
222 
223 		x86_cpuid(0x80000000, regs);
224 
225 		if (regs[0] < 0x80000007)
226 			break;
227 
228 		x86_cpuid(0x80000007, regs);
229 
230 		family = CPUID2FAMILY(ci->ci_signature);
231 
232 		if (family == 0xf)
233 			family += CPUID2EXTFAMILY(ci->ci_signature);
234 
235     		switch (family) {
236 
237 		case 0x0f:
238 
239 			if ((regs[3] & CPUID_APM_FID) == 0)
240 				break;
241 
242 			if ((regs[3] & CPUID_APM_VID) == 0)
243 				break;
244 
245 			val |= ACPICPU_FLAG_P_FFH | ACPICPU_FLAG_P_FIDVID;
246 			break;
247 
248 		case 0x10:
249 		case 0x11:
250 
251 			if ((regs[3] & CPUID_APM_TSC) != 0)
252 				val &= ~ACPICPU_FLAG_C_TSC;
253 
254 			if ((regs[3] & CPUID_APM_HWP) != 0)
255 				val |= ACPICPU_FLAG_P_FFH;
256 
257 			if ((regs[3] & CPUID_APM_CPB) != 0)
258 				val |= ACPICPU_FLAG_P_TURBO;
259 		}
260 
261 		break;
262 	}
263 
264 	/*
265 	 * There are several erratums for PIIX4.
266 	 */
267 	if (pci_find_device(&pa, acpicpu_md_quirks_piix4) != 0)
268 		val |= ACPICPU_FLAG_PIIX4;
269 
270 	return val;
271 }
272 
273 static int
274 acpicpu_md_quirks_piix4(struct pci_attach_args *pa)
275 {
276 
277 	/*
278 	 * XXX: The pci_find_device(9) function only
279 	 *	deals with attached devices. Change this
280 	 *	to use something like pci_device_foreach().
281 	 */
282 	if (PCI_VENDOR(pa->pa_id) != PCI_VENDOR_INTEL)
283 		return 0;
284 
285 	if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_INTEL_82371AB_ISA ||
286 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_INTEL_82440MX_PMC)
287 		return 1;
288 
289 	return 0;
290 }
291 
292 uint32_t
293 acpicpu_md_cpus_running(void)
294 {
295 
296 	return popcount32(cpus_running);
297 }
298 
299 int
300 acpicpu_md_idle_start(struct acpicpu_softc *sc)
301 {
302 	const size_t size = sizeof(native_idle_text);
303 	struct acpicpu_cstate *cs;
304 	bool ipi = false;
305 	int i;
306 
307 	x86_cpu_idle_get(&native_idle, native_idle_text, size);
308 
309 	for (i = 0; i < ACPI_C_STATE_COUNT; i++) {
310 
311 		cs = &sc->sc_cstate[i];
312 
313 		if (cs->cs_method == ACPICPU_C_STATE_HALT) {
314 			ipi = true;
315 			break;
316 		}
317 	}
318 
319 	x86_cpu_idle_set(acpicpu_cstate_idle, "acpi", ipi);
320 
321 	return 0;
322 }
323 
324 int
325 acpicpu_md_idle_stop(void)
326 {
327 	uint64_t xc;
328 	bool ipi;
329 
330 	ipi = (native_idle != x86_cpu_idle_halt) ? false : true;
331 	x86_cpu_idle_set(native_idle, native_idle_text, ipi);
332 
333 	/*
334 	 * Run a cross-call to ensure that all CPUs are
335 	 * out from the ACPI idle-loop before detachment.
336 	 */
337 	xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
338 	xc_wait(xc);
339 
340 	return 0;
341 }
342 
343 /*
344  * Called with interrupts disabled.
345  * Caller should enable interrupts after return.
346  */
347 void
348 acpicpu_md_idle_enter(int method, int state)
349 {
350 	struct cpu_info *ci = curcpu();
351 
352 	switch (method) {
353 
354 	case ACPICPU_C_STATE_FFH:
355 
356 		x86_enable_intr();
357 		x86_monitor(&ci->ci_want_resched, 0, 0);
358 
359 		if (__predict_false(ci->ci_want_resched != 0))
360 			return;
361 
362 		x86_mwait((state - 1) << 4, 0);
363 		break;
364 
365 	case ACPICPU_C_STATE_HALT:
366 
367 		if (__predict_false(ci->ci_want_resched != 0))
368 			return;
369 
370 		x86_stihlt();
371 		break;
372 	}
373 }
374 
375 int
376 acpicpu_md_pstate_start(void)
377 {
378 	const uint64_t est = __BIT(16);
379 	uint64_t val;
380 
381 	switch (cpu_vendor) {
382 
383 	case CPUVENDOR_IDT:
384 	case CPUVENDOR_INTEL:
385 
386 		val = rdmsr(MSR_MISC_ENABLE);
387 
388 		if ((val & est) == 0) {
389 
390 			val |= est;
391 
392 			wrmsr(MSR_MISC_ENABLE, val);
393 			val = rdmsr(MSR_MISC_ENABLE);
394 
395 			if ((val & est) == 0)
396 				return ENOTTY;
397 		}
398 	}
399 
400 	return acpicpu_md_pstate_sysctl_init();
401 }
402 
403 int
404 acpicpu_md_pstate_stop(void)
405 {
406 
407 	if (acpicpu_log != NULL)
408 		sysctl_teardown(&acpicpu_log);
409 
410 	return 0;
411 }
412 
413 int
414 acpicpu_md_pstate_pss(struct acpicpu_softc *sc)
415 {
416 	struct acpicpu_pstate *ps, msr;
417 	struct cpu_info *ci = curcpu();
418 	uint32_t family, i = 0;
419 
420 	(void)memset(&msr, 0, sizeof(struct acpicpu_pstate));
421 
422 	switch (cpu_vendor) {
423 
424 	case CPUVENDOR_IDT:
425 	case CPUVENDOR_INTEL:
426 
427 		/*
428 		 * If the so-called Turbo Boost is present,
429 		 * the P0-state is always the "turbo state".
430 		 *
431 		 * For discussion, see:
432 		 *
433 		 *	Intel Corporation: Intel Turbo Boost Technology
434 		 *	in Intel Core(tm) Microarchitectures (Nehalem)
435 		 *	Based Processors. White Paper, November 2008.
436 		 */
437 		if ((sc->sc_flags & ACPICPU_FLAG_P_TURBO) != 0)
438 			sc->sc_pstate[0].ps_flags |= ACPICPU_FLAG_P_TURBO;
439 
440 		msr.ps_control_addr = MSR_PERF_CTL;
441 		msr.ps_control_mask = __BITS(0, 15);
442 
443 		msr.ps_status_addr  = MSR_PERF_STATUS;
444 		msr.ps_status_mask  = __BITS(0, 15);
445 		break;
446 
447 	case CPUVENDOR_AMD:
448 
449 		if ((sc->sc_flags & ACPICPU_FLAG_P_FIDVID) != 0)
450 			msr.ps_flags |= ACPICPU_FLAG_P_FIDVID;
451 
452 		family = CPUID2FAMILY(ci->ci_signature);
453 
454 		if (family == 0xf)
455 			family += CPUID2EXTFAMILY(ci->ci_signature);
456 
457 		switch (family) {
458 
459 		case 0x0f:
460 			msr.ps_control_addr = MSR_0FH_CONTROL;
461 			msr.ps_status_addr  = MSR_0FH_STATUS;
462 			break;
463 
464 		case 0x10:
465 		case 0x11:
466 			msr.ps_control_addr = MSR_10H_CONTROL;
467 			msr.ps_control_mask = __BITS(0, 2);
468 
469 			msr.ps_status_addr  = MSR_10H_STATUS;
470 			msr.ps_status_mask  = __BITS(0, 2);
471 			break;
472 
473 		default:
474 
475 			if ((sc->sc_flags & ACPICPU_FLAG_P_XPSS) == 0)
476 				return EOPNOTSUPP;
477 		}
478 
479 		break;
480 
481 	default:
482 		return ENODEV;
483 	}
484 
485 	/*
486 	 * Fill the P-state structures with MSR addresses that are
487 	 * known to be correct. If we do not know the addresses,
488 	 * leave the values intact. If a vendor uses XPSS, we do
489 	 * not necessary need to do anything to support new CPUs.
490 	 */
491 	while (i < sc->sc_pstate_count) {
492 
493 		ps = &sc->sc_pstate[i];
494 
495 		if (msr.ps_flags != 0)
496 			ps->ps_flags |= msr.ps_flags;
497 
498 		if (msr.ps_status_addr != 0)
499 			ps->ps_status_addr = msr.ps_status_addr;
500 
501 		if (msr.ps_status_mask != 0)
502 			ps->ps_status_mask = msr.ps_status_mask;
503 
504 		if (msr.ps_control_addr != 0)
505 			ps->ps_control_addr = msr.ps_control_addr;
506 
507 		if (msr.ps_control_mask != 0)
508 			ps->ps_control_mask = msr.ps_control_mask;
509 
510 		i++;
511 	}
512 
513 	return 0;
514 }
515 
516 int
517 acpicpu_md_pstate_get(struct acpicpu_softc *sc, uint32_t *freq)
518 {
519 	struct acpicpu_pstate *ps = NULL;
520 	uint64_t val;
521 	uint32_t i;
522 
523 	if ((sc->sc_flags & ACPICPU_FLAG_P_FIDVID) != 0)
524 		return acpicpu_md_pstate_fidvid_get(sc, freq);
525 
526 	for (i = 0; i < sc->sc_pstate_count; i++) {
527 
528 		ps = &sc->sc_pstate[i];
529 
530 		if (__predict_true(ps->ps_freq != 0))
531 			break;
532 	}
533 
534 	if (__predict_false(ps == NULL))
535 		return ENODEV;
536 
537 	if (__predict_false(ps->ps_status_addr == 0))
538 		return EINVAL;
539 
540 	val = rdmsr(ps->ps_status_addr);
541 
542 	if (__predict_true(ps->ps_status_mask != 0))
543 		val = val & ps->ps_status_mask;
544 
545 	for (i = 0; i < sc->sc_pstate_count; i++) {
546 
547 		ps = &sc->sc_pstate[i];
548 
549 		if (__predict_false(ps->ps_freq == 0))
550 			continue;
551 
552 		if (val == ps->ps_status) {
553 			*freq = ps->ps_freq;
554 			return 0;
555 		}
556 	}
557 
558 	return EIO;
559 }
560 
561 int
562 acpicpu_md_pstate_set(struct acpicpu_pstate *ps)
563 {
564 	struct msr_rw_info msr;
565 	uint64_t xc;
566 	int rv = 0;
567 
568 	if ((ps->ps_flags & ACPICPU_FLAG_P_FIDVID) != 0)
569 		return acpicpu_md_pstate_fidvid_set(ps);
570 
571 	msr.msr_read  = false;
572 	msr.msr_type  = ps->ps_control_addr;
573 	msr.msr_value = ps->ps_control;
574 
575 	if (__predict_true(ps->ps_control_mask != 0)) {
576 		msr.msr_mask = ps->ps_control_mask;
577 		msr.msr_read = true;
578 	}
579 
580 	xc = xc_broadcast(0, (xcfunc_t)x86_msr_xcall, &msr, NULL);
581 	xc_wait(xc);
582 
583 	if (ACPICPU_P_STATE_STATUS == 0) {
584 		DELAY(ps->ps_latency);
585 		return 0;
586 	}
587 
588 	xc = xc_broadcast(0, (xcfunc_t)acpicpu_md_pstate_status, ps, &rv);
589 	xc_wait(xc);
590 
591 	return rv;
592 }
593 
594 static void
595 acpicpu_md_pstate_status(void *arg1, void *arg2)
596 {
597 	struct acpicpu_pstate *ps = arg1;
598 	uint64_t val;
599 	int i;
600 
601 	for (i = val = 0; i < ACPICPU_P_STATE_RETRY; i++) {
602 
603 		val = rdmsr(ps->ps_status_addr);
604 
605 		if (__predict_true(ps->ps_status_mask != 0))
606 			val = val & ps->ps_status_mask;
607 
608 		if (val == ps->ps_status)
609 			return;
610 
611 		DELAY(ps->ps_latency);
612 	}
613 
614 	*(uintptr_t *)arg2 = EAGAIN;
615 }
616 
617 static int
618 acpicpu_md_pstate_fidvid_get(struct acpicpu_softc *sc, uint32_t *freq)
619 {
620 	struct acpicpu_pstate *ps;
621 	uint32_t fid, i, vid;
622 	uint32_t cfid, cvid;
623 	int rv;
624 
625 	/*
626 	 * AMD family 0Fh needs special treatment.
627 	 * While it wants to use ACPI, it does not
628 	 * comply with the ACPI specifications.
629 	 */
630 	rv = acpicpu_md_pstate_fidvid_read(&cfid, &cvid);
631 
632 	if (rv != 0)
633 		return rv;
634 
635 	for (i = 0; i < sc->sc_pstate_count; i++) {
636 
637 		ps = &sc->sc_pstate[i];
638 
639 		if (__predict_false(ps->ps_freq == 0))
640 			continue;
641 
642 		fid = __SHIFTOUT(ps->ps_status, ACPI_0FH_STATUS_FID);
643 		vid = __SHIFTOUT(ps->ps_status, ACPI_0FH_STATUS_VID);
644 
645 		if (cfid == fid && cvid == vid) {
646 			*freq = ps->ps_freq;
647 			return 0;
648 		}
649 	}
650 
651 	return EIO;
652 }
653 
654 static int
655 acpicpu_md_pstate_fidvid_set(struct acpicpu_pstate *ps)
656 {
657 	const uint64_t ctrl = ps->ps_control;
658 	uint32_t cfid, cvid, fid, i, irt;
659 	uint32_t pll, vco_cfid, vco_fid;
660 	uint32_t val, vid, vst;
661 	int rv;
662 
663 	rv = acpicpu_md_pstate_fidvid_read(&cfid, &cvid);
664 
665 	if (rv != 0)
666 		return rv;
667 
668 	fid = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_FID);
669 	vid = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_VID);
670 	irt = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_IRT);
671 	vst = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_VST);
672 	pll = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_PLL);
673 
674 	vst = vst * 20;
675 	pll = pll * 1000 / 5;
676 	irt = 10 * __BIT(irt);
677 
678 	/*
679 	 * Phase 1.
680 	 */
681 	while (cvid > vid) {
682 
683 		val = 1 << __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_MVS);
684 		val = (val > cvid) ? 0 : cvid - val;
685 
686 		acpicpu_md_pstate_fidvid_write(cfid, val, 1, vst);
687 		rv = acpicpu_md_pstate_fidvid_read(NULL, &cvid);
688 
689 		if (rv != 0)
690 			return rv;
691 	}
692 
693 	i = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_RVO);
694 
695 	for (; i > 0 && cvid > 0; --i) {
696 
697 		acpicpu_md_pstate_fidvid_write(cfid, cvid - 1, 1, vst);
698 		rv = acpicpu_md_pstate_fidvid_read(NULL, &cvid);
699 
700 		if (rv != 0)
701 			return rv;
702 	}
703 
704 	/*
705 	 * Phase 2.
706 	 */
707 	if (cfid != fid) {
708 
709 		vco_fid  = FID_TO_VCO_FID(fid);
710 		vco_cfid = FID_TO_VCO_FID(cfid);
711 
712 		while (abs(vco_fid - vco_cfid) > 2) {
713 
714 			if (fid <= cfid)
715 				val = cfid - 2;
716 			else {
717 				val = (cfid > 6) ? cfid + 2 :
718 				    FID_TO_VCO_FID(cfid) + 2;
719 			}
720 
721 			acpicpu_md_pstate_fidvid_write(val, cvid, pll, irt);
722 			rv = acpicpu_md_pstate_fidvid_read(&cfid, NULL);
723 
724 			if (rv != 0)
725 				return rv;
726 
727 			vco_cfid = FID_TO_VCO_FID(cfid);
728 		}
729 
730 		acpicpu_md_pstate_fidvid_write(fid, cvid, pll, irt);
731 		rv = acpicpu_md_pstate_fidvid_read(&cfid, NULL);
732 
733 		if (rv != 0)
734 			return rv;
735 	}
736 
737 	/*
738 	 * Phase 3.
739 	 */
740 	if (cvid != vid) {
741 
742 		acpicpu_md_pstate_fidvid_write(cfid, vid, 1, vst);
743 		rv = acpicpu_md_pstate_fidvid_read(NULL, &cvid);
744 
745 		if (rv != 0)
746 			return rv;
747 	}
748 
749 	if (cfid != fid || cvid != vid)
750 		return EIO;
751 
752 	return 0;
753 }
754 
755 static int
756 acpicpu_md_pstate_fidvid_read(uint32_t *cfid, uint32_t *cvid)
757 {
758 	int i = ACPICPU_P_STATE_RETRY * 100;
759 	uint64_t val;
760 
761 	do {
762 		val = rdmsr(MSR_0FH_STATUS);
763 
764 	} while (__SHIFTOUT(val, MSR_0FH_STATUS_PENDING) != 0 && --i >= 0);
765 
766 	if (i == 0)
767 		return EAGAIN;
768 
769 	if (cfid != NULL)
770 		*cfid = __SHIFTOUT(val, MSR_0FH_STATUS_CFID);
771 
772 	if (cvid != NULL)
773 		*cvid = __SHIFTOUT(val, MSR_0FH_STATUS_CVID);
774 
775 	return 0;
776 }
777 
778 static void
779 acpicpu_md_pstate_fidvid_write(uint32_t fid,
780     uint32_t vid, uint32_t cnt, uint32_t tmo)
781 {
782 	struct msr_rw_info msr;
783 	uint64_t xc;
784 
785 	msr.msr_read  = false;
786 	msr.msr_type  = MSR_0FH_CONTROL;
787 	msr.msr_value = 0;
788 
789 	msr.msr_value |= __SHIFTIN(fid, MSR_0FH_CONTROL_FID);
790 	msr.msr_value |= __SHIFTIN(vid, MSR_0FH_CONTROL_VID);
791 	msr.msr_value |= __SHIFTIN(cnt, MSR_0FH_CONTROL_CNT);
792 	msr.msr_value |= __SHIFTIN(0x1, MSR_0FH_CONTROL_CHG);
793 
794 	xc = xc_broadcast(0, (xcfunc_t)x86_msr_xcall, &msr, NULL);
795 	xc_wait(xc);
796 
797 	DELAY(tmo);
798 }
799 
800 int
801 acpicpu_md_tstate_get(struct acpicpu_softc *sc, uint32_t *percent)
802 {
803 	struct acpicpu_tstate *ts;
804 	uint64_t val;
805 	uint32_t i;
806 
807 	val = rdmsr(MSR_THERM_CONTROL);
808 
809 	for (i = 0; i < sc->sc_tstate_count; i++) {
810 
811 		ts = &sc->sc_tstate[i];
812 
813 		if (ts->ts_percent == 0)
814 			continue;
815 
816 		if (val == ts->ts_status) {
817 			*percent = ts->ts_percent;
818 			return 0;
819 		}
820 	}
821 
822 	return EIO;
823 }
824 
825 int
826 acpicpu_md_tstate_set(struct acpicpu_tstate *ts)
827 {
828 	struct msr_rw_info msr;
829 	uint64_t xc;
830 	int rv = 0;
831 
832 	msr.msr_read  = true;
833 	msr.msr_type  = MSR_THERM_CONTROL;
834 	msr.msr_value = ts->ts_control;
835 	msr.msr_mask = __BITS(1, 4);
836 
837 	xc = xc_broadcast(0, (xcfunc_t)x86_msr_xcall, &msr, NULL);
838 	xc_wait(xc);
839 
840 	if (ts->ts_status == 0) {
841 		DELAY(ts->ts_latency);
842 		return 0;
843 	}
844 
845 	xc = xc_broadcast(0, (xcfunc_t)acpicpu_md_tstate_status, ts, &rv);
846 	xc_wait(xc);
847 
848 	return rv;
849 }
850 
851 static void
852 acpicpu_md_tstate_status(void *arg1, void *arg2)
853 {
854 	struct acpicpu_tstate *ts = arg1;
855 	uint64_t val;
856 	int i;
857 
858 	for (i = val = 0; i < ACPICPU_T_STATE_RETRY; i++) {
859 
860 		val = rdmsr(MSR_THERM_CONTROL);
861 
862 		if (val == ts->ts_status)
863 			return;
864 
865 		DELAY(ts->ts_latency);
866 	}
867 
868 	*(uintptr_t *)arg2 = EAGAIN;
869 }
870 
871 /*
872  * A kludge for backwards compatibility.
873  */
874 static int
875 acpicpu_md_pstate_sysctl_init(void)
876 {
877 	const struct sysctlnode	*fnode, *mnode, *rnode;
878 	const char *str;
879 	int rv;
880 
881 	switch (cpu_vendor) {
882 
883 	case CPUVENDOR_IDT:
884 	case CPUVENDOR_INTEL:
885 		str = "est";
886 		break;
887 
888 	case CPUVENDOR_AMD:
889 		str = "powernow";
890 		break;
891 
892 	default:
893 		return ENODEV;
894 	}
895 
896 
897 	rv = sysctl_createv(&acpicpu_log, 0, NULL, &rnode,
898 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "machdep", NULL,
899 	    NULL, 0, NULL, 0, CTL_MACHDEP, CTL_EOL);
900 
901 	if (rv != 0)
902 		goto fail;
903 
904 	rv = sysctl_createv(&acpicpu_log, 0, &rnode, &mnode,
905 	    0, CTLTYPE_NODE, str, NULL,
906 	    NULL, 0, NULL, 0, CTL_CREATE, CTL_EOL);
907 
908 	if (rv != 0)
909 		goto fail;
910 
911 	rv = sysctl_createv(&acpicpu_log, 0, &mnode, &fnode,
912 	    0, CTLTYPE_NODE, "frequency", NULL,
913 	    NULL, 0, NULL, 0, CTL_CREATE, CTL_EOL);
914 
915 	if (rv != 0)
916 		goto fail;
917 
918 	rv = sysctl_createv(&acpicpu_log, 0, &fnode, &rnode,
919 	    CTLFLAG_READWRITE, CTLTYPE_INT, "target", NULL,
920 	    acpicpu_md_pstate_sysctl_set, 0, NULL, 0, CTL_CREATE, CTL_EOL);
921 
922 	if (rv != 0)
923 		goto fail;
924 
925 	rv = sysctl_createv(&acpicpu_log, 0, &fnode, &rnode,
926 	    CTLFLAG_READONLY, CTLTYPE_INT, "current", NULL,
927 	    acpicpu_md_pstate_sysctl_get, 0, NULL, 0, CTL_CREATE, CTL_EOL);
928 
929 	if (rv != 0)
930 		goto fail;
931 
932 	rv = sysctl_createv(&acpicpu_log, 0, &fnode, &rnode,
933 	    CTLFLAG_READONLY, CTLTYPE_STRING, "available", NULL,
934 	    acpicpu_md_pstate_sysctl_all, 0, NULL, 0, CTL_CREATE, CTL_EOL);
935 
936 	if (rv != 0)
937 		goto fail;
938 
939 	return 0;
940 
941 fail:
942 	if (acpicpu_log != NULL) {
943 		sysctl_teardown(&acpicpu_log);
944 		acpicpu_log = NULL;
945 	}
946 
947 	return rv;
948 }
949 
950 static int
951 acpicpu_md_pstate_sysctl_get(SYSCTLFN_ARGS)
952 {
953 	struct cpu_info *ci = curcpu();
954 	struct acpicpu_softc *sc;
955 	struct sysctlnode node;
956 	uint32_t freq;
957 	int err;
958 
959 	sc = acpicpu_sc[ci->ci_acpiid];
960 
961 	if (sc == NULL)
962 		return ENXIO;
963 
964 	err = acpicpu_pstate_get(sc, &freq);
965 
966 	if (err != 0)
967 		return err;
968 
969 	node = *rnode;
970 	node.sysctl_data = &freq;
971 
972 	err = sysctl_lookup(SYSCTLFN_CALL(&node));
973 
974 	if (err != 0 || newp == NULL)
975 		return err;
976 
977 	return 0;
978 }
979 
980 static int
981 acpicpu_md_pstate_sysctl_set(SYSCTLFN_ARGS)
982 {
983 	struct cpu_info *ci = curcpu();
984 	struct acpicpu_softc *sc;
985 	struct sysctlnode node;
986 	uint32_t freq;
987 	int err;
988 
989 	sc = acpicpu_sc[ci->ci_acpiid];
990 
991 	if (sc == NULL)
992 		return ENXIO;
993 
994 	err = acpicpu_pstate_get(sc, &freq);
995 
996 	if (err != 0)
997 		return err;
998 
999 	node = *rnode;
1000 	node.sysctl_data = &freq;
1001 
1002 	err = sysctl_lookup(SYSCTLFN_CALL(&node));
1003 
1004 	if (err != 0 || newp == NULL)
1005 		return err;
1006 
1007 	err = acpicpu_pstate_set(sc, freq);
1008 
1009 	if (err != 0)
1010 		return err;
1011 
1012 	return 0;
1013 }
1014 
1015 static int
1016 acpicpu_md_pstate_sysctl_all(SYSCTLFN_ARGS)
1017 {
1018 	struct cpu_info *ci = curcpu();
1019 	struct acpicpu_softc *sc;
1020 	struct sysctlnode node;
1021 	char buf[1024];
1022 	size_t len;
1023 	uint32_t i;
1024 	int err;
1025 
1026 	sc = acpicpu_sc[ci->ci_acpiid];
1027 
1028 	if (sc == NULL)
1029 		return ENXIO;
1030 
1031 	(void)memset(&buf, 0, sizeof(buf));
1032 
1033 	mutex_enter(&sc->sc_mtx);
1034 
1035 	for (len = 0, i = sc->sc_pstate_max; i < sc->sc_pstate_count; i++) {
1036 
1037 		if (sc->sc_pstate[i].ps_freq == 0)
1038 			continue;
1039 
1040 		len += snprintf(buf + len, sizeof(buf) - len, "%u%s",
1041 		    sc->sc_pstate[i].ps_freq,
1042 		    i < (sc->sc_pstate_count - 1) ? " " : "");
1043 	}
1044 
1045 	mutex_exit(&sc->sc_mtx);
1046 
1047 	node = *rnode;
1048 	node.sysctl_data = buf;
1049 
1050 	err = sysctl_lookup(SYSCTLFN_CALL(&node));
1051 
1052 	if (err != 0 || newp == NULL)
1053 		return err;
1054 
1055 	return 0;
1056 }
1057 
1058