Home | History | Annotate | Line # | Download | only in acpi
acpi_cpu_md.c revision 1.38
      1 /* $NetBSD: acpi_cpu_md.c,v 1.38 2011/01/13 03:40:50 jruoho Exp $ */
      2 
      3 /*-
      4  * Copyright (c) 2010 Jukka Ruohonen <jruohonen (at) 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 1.38 2011/01/13 03:40:50 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 #include <machine/acpi_machdep.h>
     52 
     53 /*
     54  * AMD C1E.
     55  */
     56 #define MSR_CMPHALT		0xc0010055
     57 
     58 #define MSR_CMPHALT_SMI		__BIT(27)
     59 #define MSR_CMPHALT_C1E		__BIT(28)
     60 #define MSR_CMPHALT_BMSTS	__BIT(29)
     61 
     62 /*
     63  * AMD families 10h and 11h.
     64  */
     65 #define MSR_10H_LIMIT		0xc0010061
     66 #define MSR_10H_CONTROL		0xc0010062
     67 #define MSR_10H_STATUS		0xc0010063
     68 #define MSR_10H_CONFIG		0xc0010064
     69 
     70 /*
     71  * AMD family 0Fh.
     72  */
     73 #define MSR_0FH_CONTROL		0xc0010041
     74 #define MSR_0FH_STATUS		0xc0010042
     75 
     76 #define MSR_0FH_STATUS_CFID	__BITS( 0,  5)
     77 #define MSR_0FH_STATUS_CVID	__BITS(32, 36)
     78 #define MSR_0FH_STATUS_PENDING	__BITS(31, 31)
     79 
     80 #define MSR_0FH_CONTROL_FID	__BITS( 0,  5)
     81 #define MSR_0FH_CONTROL_VID	__BITS( 8, 12)
     82 #define MSR_0FH_CONTROL_CHG	__BITS(16, 16)
     83 #define MSR_0FH_CONTROL_CNT	__BITS(32, 51)
     84 
     85 #define ACPI_0FH_STATUS_FID	__BITS( 0,  5)
     86 #define ACPI_0FH_STATUS_VID	__BITS( 6, 10)
     87 
     88 #define ACPI_0FH_CONTROL_FID	__BITS( 0,  5)
     89 #define ACPI_0FH_CONTROL_VID	__BITS( 6, 10)
     90 #define ACPI_0FH_CONTROL_VST	__BITS(11, 17)
     91 #define ACPI_0FH_CONTROL_MVS	__BITS(18, 19)
     92 #define ACPI_0FH_CONTROL_PLL	__BITS(20, 26)
     93 #define ACPI_0FH_CONTROL_RVO	__BITS(28, 29)
     94 #define ACPI_0FH_CONTROL_IRT	__BITS(30, 31)
     95 
     96 #define FID_TO_VCO_FID(fidd)	(((fid) < 8) ? (8 + ((fid) << 1)) : (fid))
     97 
     98 static char	  native_idle_text[16];
     99 void		(*native_idle)(void) = NULL;
    100 
    101 static int	 acpicpu_md_quirks_piix4(struct pci_attach_args *);
    102 static void	 acpicpu_md_pstate_status(void *, void *);
    103 static int	 acpicpu_md_pstate_fidvid_get(struct acpicpu_softc *,
    104                                               uint32_t *);
    105 static int	 acpicpu_md_pstate_fidvid_set(struct acpicpu_pstate *);
    106 static int	 acpicpu_md_pstate_fidvid_read(uint32_t *, uint32_t *);
    107 static void	 acpicpu_md_pstate_fidvid_write(uint32_t, uint32_t,
    108 					        uint32_t, uint32_t);
    109 static void	 acpicpu_md_tstate_status(void *, void *);
    110 static int	 acpicpu_md_pstate_sysctl_init(void);
    111 static int	 acpicpu_md_pstate_sysctl_get(SYSCTLFN_PROTO);
    112 static int	 acpicpu_md_pstate_sysctl_set(SYSCTLFN_PROTO);
    113 static int	 acpicpu_md_pstate_sysctl_all(SYSCTLFN_PROTO);
    114 
    115 extern struct acpicpu_softc **acpicpu_sc;
    116 static bool acpicpu_pstate_status = false;
    117 static struct sysctllog *acpicpu_log = NULL;
    118 
    119 uint32_t
    120 acpicpu_md_cap(void)
    121 {
    122 	struct cpu_info *ci = curcpu();
    123 	uint32_t val = 0;
    124 
    125 	if (cpu_vendor != CPUVENDOR_IDT &&
    126 	    cpu_vendor != CPUVENDOR_INTEL)
    127 		return val;
    128 
    129 	/*
    130 	 * Basic SMP C-states (required for _CST).
    131 	 */
    132 	val |= ACPICPU_PDC_C_C1PT | ACPICPU_PDC_C_C2C3;
    133 
    134         /*
    135 	 * If MONITOR/MWAIT is available, announce
    136 	 * support for native instructions in all C-states.
    137 	 */
    138         if ((ci->ci_feat_val[1] & CPUID2_MONITOR) != 0)
    139 		val |= ACPICPU_PDC_C_C1_FFH | ACPICPU_PDC_C_C2C3_FFH;
    140 
    141 	/*
    142 	 * Set native P- and T-states, if available.
    143 	 */
    144         if ((ci->ci_feat_val[1] & CPUID2_EST) != 0)
    145 		val |= ACPICPU_PDC_P_FFH;
    146 
    147 	if ((ci->ci_feat_val[0] & CPUID_ACPI) != 0)
    148 		val |= ACPICPU_PDC_T_FFH;
    149 
    150 	return val;
    151 }
    152 
    153 uint32_t
    154 acpicpu_md_quirks(void)
    155 {
    156 	struct cpu_info *ci = curcpu();
    157 	struct pci_attach_args pa;
    158 	uint32_t family, val = 0;
    159 	uint32_t regs[4];
    160 
    161 	if (acpi_md_ncpus() == 1)
    162 		val |= ACPICPU_FLAG_C_BM;
    163 
    164 	if ((ci->ci_feat_val[1] & CPUID2_MONITOR) != 0)
    165 		val |= ACPICPU_FLAG_C_FFH;
    166 
    167 	val |= ACPICPU_FLAG_C_APIC | ACPICPU_FLAG_C_TSC;
    168 
    169 	switch (cpu_vendor) {
    170 
    171 	case CPUVENDOR_IDT:
    172 
    173 		if ((ci->ci_feat_val[1] & CPUID2_EST) != 0)
    174 			val |= ACPICPU_FLAG_P_FFH;
    175 
    176 		if ((ci->ci_feat_val[0] & CPUID_ACPI) != 0)
    177 			val |= ACPICPU_FLAG_T_FFH;
    178 
    179 		break;
    180 
    181 	case CPUVENDOR_INTEL:
    182 
    183 		val |= ACPICPU_FLAG_C_BM | ACPICPU_FLAG_C_ARB;
    184 
    185 		if ((ci->ci_feat_val[1] & CPUID2_EST) != 0)
    186 			val |= ACPICPU_FLAG_P_FFH;
    187 
    188 		if ((ci->ci_feat_val[0] & CPUID_ACPI) != 0)
    189 			val |= ACPICPU_FLAG_T_FFH;
    190 
    191 		/*
    192 		 * Check whether MSR_APERF, MSR_MPERF, and Turbo
    193 		 * Boost are available. Also see if we might have
    194 		 * an invariant local APIC timer ("ARAT").
    195 		 */
    196 		if (cpuid_level >= 0x06) {
    197 
    198 			x86_cpuid(0x06, regs);
    199 
    200 			if ((regs[2] & CPUID_DSPM_HWF) != 0)
    201 				val |= ACPICPU_FLAG_P_HW;
    202 
    203 			if ((regs[0] & CPUID_DSPM_IDA) != 0)
    204 				val |= ACPICPU_FLAG_P_TURBO;
    205 
    206 			if ((regs[0] & CPUID_DSPM_ARAT) != 0)
    207 				val &= ~ACPICPU_FLAG_C_APIC;
    208 		}
    209 
    210 		/*
    211 		 * Detect whether TSC is invariant. If it is not,
    212 		 * we keep the flag to note that TSC will not run
    213 		 * at constant rate. Depending on the CPU, this may
    214 		 * affect P- and T-state changes, but especially
    215 		 * relevant are C-states; with variant TSC, states
    216 		 * larger than C1 may completely stop the counter.
    217 		 */
    218 		x86_cpuid(0x80000000, regs);
    219 
    220 		if (regs[0] >= 0x80000007) {
    221 
    222 			x86_cpuid(0x80000007, regs);
    223 
    224 			if ((regs[3] & __BIT(8)) != 0)
    225 				val &= ~ACPICPU_FLAG_C_TSC;
    226 		}
    227 
    228 		break;
    229 
    230 	case CPUVENDOR_AMD:
    231 
    232 		x86_cpuid(0x80000000, regs);
    233 
    234 		if (regs[0] < 0x80000007)
    235 			break;
    236 
    237 		x86_cpuid(0x80000007, regs);
    238 
    239 		family = CPUID2FAMILY(ci->ci_signature);
    240 
    241 		if (family == 0xf)
    242 			family += CPUID2EXTFAMILY(ci->ci_signature);
    243 
    244     		switch (family) {
    245 
    246 		case 0x0f:
    247 
    248 			if ((regs[3] & CPUID_APM_FID) == 0)
    249 				break;
    250 
    251 			if ((regs[3] & CPUID_APM_VID) == 0)
    252 				break;
    253 
    254 			val |= ACPICPU_FLAG_P_FFH | ACPICPU_FLAG_P_FIDVID;
    255 			break;
    256 
    257 		case 0x10:
    258 		case 0x11:
    259 
    260 			if ((regs[3] & CPUID_APM_TSC) != 0)
    261 				val &= ~ACPICPU_FLAG_C_TSC;
    262 
    263 			if ((regs[3] & CPUID_APM_HWP) != 0)
    264 				val |= ACPICPU_FLAG_P_FFH;
    265 
    266 			if ((regs[3] & CPUID_APM_CPB) != 0)
    267 				val |= ACPICPU_FLAG_P_TURBO;
    268 
    269 			val |= ACPICPU_FLAG_C_C1E;
    270 			break;
    271 		}
    272 
    273 		break;
    274 	}
    275 
    276 	/*
    277 	 * There are several erratums for PIIX4.
    278 	 */
    279 	if (pci_find_device(&pa, acpicpu_md_quirks_piix4) != 0)
    280 		val |= ACPICPU_FLAG_PIIX4;
    281 
    282 	return val;
    283 }
    284 
    285 static int
    286 acpicpu_md_quirks_piix4(struct pci_attach_args *pa)
    287 {
    288 
    289 	/*
    290 	 * XXX: The pci_find_device(9) function only
    291 	 *	deals with attached devices. Change this
    292 	 *	to use something like pci_device_foreach().
    293 	 */
    294 	if (PCI_VENDOR(pa->pa_id) != PCI_VENDOR_INTEL)
    295 		return 0;
    296 
    297 	if (PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_INTEL_82371AB_ISA ||
    298 	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_INTEL_82440MX_PMC)
    299 		return 1;
    300 
    301 	return 0;
    302 }
    303 
    304 void
    305 acpicpu_md_quirks_c1e(void)
    306 {
    307 	const uint64_t c1e = MSR_CMPHALT_SMI | MSR_CMPHALT_C1E;
    308 	uint64_t val;
    309 
    310 	val = rdmsr(MSR_CMPHALT);
    311 
    312 	if ((val & c1e) != 0)
    313 		wrmsr(MSR_CMPHALT, val & ~c1e);
    314 }
    315 
    316 int
    317 acpicpu_md_idle_start(struct acpicpu_softc *sc)
    318 {
    319 	const size_t size = sizeof(native_idle_text);
    320 	struct acpicpu_cstate *cs;
    321 	bool ipi = false;
    322 	int i;
    323 
    324 	x86_cpu_idle_get(&native_idle, native_idle_text, size);
    325 
    326 	for (i = 0; i < ACPI_C_STATE_COUNT; i++) {
    327 
    328 		cs = &sc->sc_cstate[i];
    329 
    330 		if (cs->cs_method == ACPICPU_C_STATE_HALT) {
    331 			ipi = true;
    332 			break;
    333 		}
    334 	}
    335 
    336 	x86_cpu_idle_set(acpicpu_cstate_idle, "acpi", ipi);
    337 
    338 	return 0;
    339 }
    340 
    341 int
    342 acpicpu_md_idle_stop(void)
    343 {
    344 	uint64_t xc;
    345 	bool ipi;
    346 
    347 	ipi = (native_idle != x86_cpu_idle_halt) ? false : true;
    348 	x86_cpu_idle_set(native_idle, native_idle_text, ipi);
    349 
    350 	/*
    351 	 * Run a cross-call to ensure that all CPUs are
    352 	 * out from the ACPI idle-loop before detachment.
    353 	 */
    354 	xc = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
    355 	xc_wait(xc);
    356 
    357 	return 0;
    358 }
    359 
    360 /*
    361  * Called with interrupts disabled.
    362  * Caller should enable interrupts after return.
    363  */
    364 void
    365 acpicpu_md_idle_enter(int method, int state)
    366 {
    367 	struct cpu_info *ci = curcpu();
    368 
    369 	switch (method) {
    370 
    371 	case ACPICPU_C_STATE_FFH:
    372 
    373 		x86_enable_intr();
    374 		x86_monitor(&ci->ci_want_resched, 0, 0);
    375 
    376 		if (__predict_false(ci->ci_want_resched != 0))
    377 			return;
    378 
    379 		x86_mwait((state - 1) << 4, 0);
    380 		break;
    381 
    382 	case ACPICPU_C_STATE_HALT:
    383 
    384 		if (__predict_false(ci->ci_want_resched != 0))
    385 			return;
    386 
    387 		x86_stihlt();
    388 		break;
    389 	}
    390 }
    391 
    392 int
    393 acpicpu_md_pstate_start(void)
    394 {
    395 	const uint64_t est = __BIT(16);
    396 	uint64_t val;
    397 
    398 	switch (cpu_vendor) {
    399 
    400 	case CPUVENDOR_IDT:
    401 	case CPUVENDOR_INTEL:
    402 
    403 		val = rdmsr(MSR_MISC_ENABLE);
    404 
    405 		if ((val & est) == 0) {
    406 
    407 			val |= est;
    408 
    409 			wrmsr(MSR_MISC_ENABLE, val);
    410 			val = rdmsr(MSR_MISC_ENABLE);
    411 
    412 			if ((val & est) == 0)
    413 				return ENOTTY;
    414 		}
    415 	}
    416 
    417 	return acpicpu_md_pstate_sysctl_init();
    418 }
    419 
    420 int
    421 acpicpu_md_pstate_stop(void)
    422 {
    423 
    424 	if (acpicpu_log != NULL)
    425 		sysctl_teardown(&acpicpu_log);
    426 
    427 	return 0;
    428 }
    429 
    430 int
    431 acpicpu_md_pstate_pss(struct acpicpu_softc *sc)
    432 {
    433 	struct acpicpu_pstate *ps, msr;
    434 	struct cpu_info *ci = curcpu();
    435 	uint32_t family, i = 0;
    436 
    437 	(void)memset(&msr, 0, sizeof(struct acpicpu_pstate));
    438 
    439 	switch (cpu_vendor) {
    440 
    441 	case CPUVENDOR_IDT:
    442 	case CPUVENDOR_INTEL:
    443 
    444 		/*
    445 		 * If the so-called Turbo Boost is present,
    446 		 * the P0-state is always the "turbo state".
    447 		 *
    448 		 * For discussion, see:
    449 		 *
    450 		 *	Intel Corporation: Intel Turbo Boost Technology
    451 		 *	in Intel Core(tm) Microarchitectures (Nehalem)
    452 		 *	Based Processors. White Paper, November 2008.
    453 		 */
    454 		if ((sc->sc_flags & ACPICPU_FLAG_P_TURBO) != 0)
    455 			sc->sc_pstate[0].ps_flags |= ACPICPU_FLAG_P_TURBO;
    456 
    457 		msr.ps_control_addr = MSR_PERF_CTL;
    458 		msr.ps_control_mask = __BITS(0, 15);
    459 
    460 		msr.ps_status_addr  = MSR_PERF_STATUS;
    461 		msr.ps_status_mask  = __BITS(0, 15);
    462 		break;
    463 
    464 	case CPUVENDOR_AMD:
    465 
    466 		if ((sc->sc_flags & ACPICPU_FLAG_P_FIDVID) != 0)
    467 			msr.ps_flags |= ACPICPU_FLAG_P_FIDVID;
    468 
    469 		family = CPUID2FAMILY(ci->ci_signature);
    470 
    471 		if (family == 0xf)
    472 			family += CPUID2EXTFAMILY(ci->ci_signature);
    473 
    474 		switch (family) {
    475 
    476 		case 0x0f:
    477 			msr.ps_control_addr = MSR_0FH_CONTROL;
    478 			msr.ps_status_addr  = MSR_0FH_STATUS;
    479 			break;
    480 
    481 		case 0x10:
    482 		case 0x11:
    483 			msr.ps_control_addr = MSR_10H_CONTROL;
    484 			msr.ps_control_mask = __BITS(0, 2);
    485 
    486 			msr.ps_status_addr  = MSR_10H_STATUS;
    487 			msr.ps_status_mask  = __BITS(0, 2);
    488 			break;
    489 
    490 		default:
    491 
    492 			if ((sc->sc_flags & ACPICPU_FLAG_P_XPSS) == 0)
    493 				return EOPNOTSUPP;
    494 		}
    495 
    496 		break;
    497 
    498 	default:
    499 		return ENODEV;
    500 	}
    501 
    502 	/*
    503 	 * Fill the P-state structures with MSR addresses that are
    504 	 * known to be correct. If we do not know the addresses,
    505 	 * leave the values intact. If a vendor uses XPSS, we do
    506 	 * not necessary need to do anything to support new CPUs.
    507 	 */
    508 	while (i < sc->sc_pstate_count) {
    509 
    510 		ps = &sc->sc_pstate[i];
    511 
    512 		if (msr.ps_flags != 0)
    513 			ps->ps_flags |= msr.ps_flags;
    514 
    515 		if (msr.ps_status_addr != 0)
    516 			ps->ps_status_addr = msr.ps_status_addr;
    517 
    518 		if (msr.ps_status_mask != 0)
    519 			ps->ps_status_mask = msr.ps_status_mask;
    520 
    521 		if (msr.ps_control_addr != 0)
    522 			ps->ps_control_addr = msr.ps_control_addr;
    523 
    524 		if (msr.ps_control_mask != 0)
    525 			ps->ps_control_mask = msr.ps_control_mask;
    526 
    527 		i++;
    528 	}
    529 
    530 	return 0;
    531 }
    532 
    533 int
    534 acpicpu_md_pstate_get(struct acpicpu_softc *sc, uint32_t *freq)
    535 {
    536 	struct acpicpu_pstate *ps = NULL;
    537 	uint64_t val;
    538 	uint32_t i;
    539 
    540 	if ((sc->sc_flags & ACPICPU_FLAG_P_FIDVID) != 0)
    541 		return acpicpu_md_pstate_fidvid_get(sc, freq);
    542 
    543 	for (i = 0; i < sc->sc_pstate_count; i++) {
    544 
    545 		ps = &sc->sc_pstate[i];
    546 
    547 		if (__predict_true(ps->ps_freq != 0))
    548 			break;
    549 	}
    550 
    551 	if (__predict_false(ps == NULL))
    552 		return ENODEV;
    553 
    554 	if (__predict_false(ps->ps_status_addr == 0))
    555 		return EINVAL;
    556 
    557 	val = rdmsr(ps->ps_status_addr);
    558 
    559 	if (__predict_true(ps->ps_status_mask != 0))
    560 		val = val & ps->ps_status_mask;
    561 
    562 	for (i = 0; i < sc->sc_pstate_count; i++) {
    563 
    564 		ps = &sc->sc_pstate[i];
    565 
    566 		if (__predict_false(ps->ps_freq == 0))
    567 			continue;
    568 
    569 		if (val == ps->ps_status) {
    570 			*freq = ps->ps_freq;
    571 			return 0;
    572 		}
    573 	}
    574 
    575 	return EIO;
    576 }
    577 
    578 int
    579 acpicpu_md_pstate_set(struct acpicpu_pstate *ps)
    580 {
    581 	struct msr_rw_info msr;
    582 	uint64_t xc;
    583 	int rv = 0;
    584 
    585 	if (__predict_false(ps->ps_control_addr == 0))
    586 		return EINVAL;
    587 
    588 	if ((ps->ps_flags & ACPICPU_FLAG_P_FIDVID) != 0)
    589 		return acpicpu_md_pstate_fidvid_set(ps);
    590 
    591 	msr.msr_read  = false;
    592 	msr.msr_type  = ps->ps_control_addr;
    593 	msr.msr_value = ps->ps_control;
    594 
    595 	if (__predict_true(ps->ps_control_mask != 0)) {
    596 		msr.msr_mask = ps->ps_control_mask;
    597 		msr.msr_read = true;
    598 	}
    599 
    600 	xc = xc_broadcast(0, (xcfunc_t)x86_msr_xcall, &msr, NULL);
    601 	xc_wait(xc);
    602 
    603 	/*
    604 	 * Due several problems, we bypass the
    605 	 * relatively expensive status check.
    606 	 */
    607 	if (acpicpu_pstate_status != true) {
    608 		DELAY(ps->ps_latency);
    609 		return 0;
    610 	}
    611 
    612 	xc = xc_broadcast(0, (xcfunc_t)acpicpu_md_pstate_status, ps, &rv);
    613 	xc_wait(xc);
    614 
    615 	return rv;
    616 }
    617 
    618 static void
    619 acpicpu_md_pstate_status(void *arg1, void *arg2)
    620 {
    621 	struct acpicpu_pstate *ps = arg1;
    622 	uint64_t val;
    623 	int i;
    624 
    625 	for (i = val = 0; i < ACPICPU_P_STATE_RETRY; i++) {
    626 
    627 		val = rdmsr(ps->ps_status_addr);
    628 
    629 		if (__predict_true(ps->ps_status_mask != 0))
    630 			val = val & ps->ps_status_mask;
    631 
    632 		if (val == ps->ps_status)
    633 			return;
    634 
    635 		DELAY(ps->ps_latency);
    636 	}
    637 
    638 	*(uintptr_t *)arg2 = EAGAIN;
    639 }
    640 
    641 static int
    642 acpicpu_md_pstate_fidvid_get(struct acpicpu_softc *sc, uint32_t *freq)
    643 {
    644 	struct acpicpu_pstate *ps;
    645 	uint32_t fid, i, vid;
    646 	uint32_t cfid, cvid;
    647 	int rv;
    648 
    649 	/*
    650 	 * AMD family 0Fh needs special treatment.
    651 	 * While it wants to use ACPI, it does not
    652 	 * comply with the ACPI specifications.
    653 	 */
    654 	rv = acpicpu_md_pstate_fidvid_read(&cfid, &cvid);
    655 
    656 	if (rv != 0)
    657 		return rv;
    658 
    659 	for (i = 0; i < sc->sc_pstate_count; i++) {
    660 
    661 		ps = &sc->sc_pstate[i];
    662 
    663 		if (__predict_false(ps->ps_freq == 0))
    664 			continue;
    665 
    666 		fid = __SHIFTOUT(ps->ps_status, ACPI_0FH_STATUS_FID);
    667 		vid = __SHIFTOUT(ps->ps_status, ACPI_0FH_STATUS_VID);
    668 
    669 		if (cfid == fid && cvid == vid) {
    670 			*freq = ps->ps_freq;
    671 			return 0;
    672 		}
    673 	}
    674 
    675 	return EIO;
    676 }
    677 
    678 static int
    679 acpicpu_md_pstate_fidvid_set(struct acpicpu_pstate *ps)
    680 {
    681 	const uint64_t ctrl = ps->ps_control;
    682 	uint32_t cfid, cvid, fid, i, irt;
    683 	uint32_t pll, vco_cfid, vco_fid;
    684 	uint32_t val, vid, vst;
    685 	int rv;
    686 
    687 	rv = acpicpu_md_pstate_fidvid_read(&cfid, &cvid);
    688 
    689 	if (rv != 0)
    690 		return rv;
    691 
    692 	fid = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_FID);
    693 	vid = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_VID);
    694 	irt = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_IRT);
    695 	vst = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_VST);
    696 	pll = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_PLL);
    697 
    698 	vst = vst * 20;
    699 	pll = pll * 1000 / 5;
    700 	irt = 10 * __BIT(irt);
    701 
    702 	/*
    703 	 * Phase 1.
    704 	 */
    705 	while (cvid > vid) {
    706 
    707 		val = 1 << __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_MVS);
    708 		val = (val > cvid) ? 0 : cvid - val;
    709 
    710 		acpicpu_md_pstate_fidvid_write(cfid, val, 1, vst);
    711 		rv = acpicpu_md_pstate_fidvid_read(NULL, &cvid);
    712 
    713 		if (rv != 0)
    714 			return rv;
    715 	}
    716 
    717 	i = __SHIFTOUT(ctrl, ACPI_0FH_CONTROL_RVO);
    718 
    719 	for (; i > 0 && cvid > 0; --i) {
    720 
    721 		acpicpu_md_pstate_fidvid_write(cfid, cvid - 1, 1, vst);
    722 		rv = acpicpu_md_pstate_fidvid_read(NULL, &cvid);
    723 
    724 		if (rv != 0)
    725 			return rv;
    726 	}
    727 
    728 	/*
    729 	 * Phase 2.
    730 	 */
    731 	if (cfid != fid) {
    732 
    733 		vco_fid  = FID_TO_VCO_FID(fid);
    734 		vco_cfid = FID_TO_VCO_FID(cfid);
    735 
    736 		while (abs(vco_fid - vco_cfid) > 2) {
    737 
    738 			if (fid <= cfid)
    739 				val = cfid - 2;
    740 			else {
    741 				val = (cfid > 6) ? cfid + 2 :
    742 				    FID_TO_VCO_FID(cfid) + 2;
    743 			}
    744 
    745 			acpicpu_md_pstate_fidvid_write(val, cvid, pll, irt);
    746 			rv = acpicpu_md_pstate_fidvid_read(&cfid, NULL);
    747 
    748 			if (rv != 0)
    749 				return rv;
    750 
    751 			vco_cfid = FID_TO_VCO_FID(cfid);
    752 		}
    753 
    754 		acpicpu_md_pstate_fidvid_write(fid, cvid, pll, irt);
    755 		rv = acpicpu_md_pstate_fidvid_read(&cfid, NULL);
    756 
    757 		if (rv != 0)
    758 			return rv;
    759 	}
    760 
    761 	/*
    762 	 * Phase 3.
    763 	 */
    764 	if (cvid != vid) {
    765 
    766 		acpicpu_md_pstate_fidvid_write(cfid, vid, 1, vst);
    767 		rv = acpicpu_md_pstate_fidvid_read(NULL, &cvid);
    768 
    769 		if (rv != 0)
    770 			return rv;
    771 	}
    772 
    773 	if (cfid != fid || cvid != vid)
    774 		return EIO;
    775 
    776 	return 0;
    777 }
    778 
    779 static int
    780 acpicpu_md_pstate_fidvid_read(uint32_t *cfid, uint32_t *cvid)
    781 {
    782 	int i = ACPICPU_P_STATE_RETRY * 100;
    783 	uint64_t val;
    784 
    785 	do {
    786 		val = rdmsr(MSR_0FH_STATUS);
    787 
    788 	} while (__SHIFTOUT(val, MSR_0FH_STATUS_PENDING) != 0 && --i >= 0);
    789 
    790 	if (i == 0)
    791 		return EAGAIN;
    792 
    793 	if (cfid != NULL)
    794 		*cfid = __SHIFTOUT(val, MSR_0FH_STATUS_CFID);
    795 
    796 	if (cvid != NULL)
    797 		*cvid = __SHIFTOUT(val, MSR_0FH_STATUS_CVID);
    798 
    799 	return 0;
    800 }
    801 
    802 static void
    803 acpicpu_md_pstate_fidvid_write(uint32_t fid,
    804     uint32_t vid, uint32_t cnt, uint32_t tmo)
    805 {
    806 	struct msr_rw_info msr;
    807 	uint64_t xc;
    808 
    809 	msr.msr_read  = false;
    810 	msr.msr_type  = MSR_0FH_CONTROL;
    811 	msr.msr_value = 0;
    812 
    813 	msr.msr_value |= __SHIFTIN(fid, MSR_0FH_CONTROL_FID);
    814 	msr.msr_value |= __SHIFTIN(vid, MSR_0FH_CONTROL_VID);
    815 	msr.msr_value |= __SHIFTIN(cnt, MSR_0FH_CONTROL_CNT);
    816 	msr.msr_value |= __SHIFTIN(0x1, MSR_0FH_CONTROL_CHG);
    817 
    818 	xc = xc_broadcast(0, (xcfunc_t)x86_msr_xcall, &msr, NULL);
    819 	xc_wait(xc);
    820 
    821 	DELAY(tmo);
    822 }
    823 
    824 int
    825 acpicpu_md_tstate_get(struct acpicpu_softc *sc, uint32_t *percent)
    826 {
    827 	struct acpicpu_tstate *ts;
    828 	uint64_t val;
    829 	uint32_t i;
    830 
    831 	val = rdmsr(MSR_THERM_CONTROL);
    832 
    833 	for (i = 0; i < sc->sc_tstate_count; i++) {
    834 
    835 		ts = &sc->sc_tstate[i];
    836 
    837 		if (ts->ts_percent == 0)
    838 			continue;
    839 
    840 		if (val == ts->ts_status) {
    841 			*percent = ts->ts_percent;
    842 			return 0;
    843 		}
    844 	}
    845 
    846 	return EIO;
    847 }
    848 
    849 int
    850 acpicpu_md_tstate_set(struct acpicpu_tstate *ts)
    851 {
    852 	struct msr_rw_info msr;
    853 	uint64_t xc;
    854 	int rv = 0;
    855 
    856 	msr.msr_read  = true;
    857 	msr.msr_type  = MSR_THERM_CONTROL;
    858 	msr.msr_value = ts->ts_control;
    859 	msr.msr_mask = __BITS(1, 4);
    860 
    861 	xc = xc_broadcast(0, (xcfunc_t)x86_msr_xcall, &msr, NULL);
    862 	xc_wait(xc);
    863 
    864 	if (ts->ts_status == 0) {
    865 		DELAY(ts->ts_latency);
    866 		return 0;
    867 	}
    868 
    869 	xc = xc_broadcast(0, (xcfunc_t)acpicpu_md_tstate_status, ts, &rv);
    870 	xc_wait(xc);
    871 
    872 	return rv;
    873 }
    874 
    875 static void
    876 acpicpu_md_tstate_status(void *arg1, void *arg2)
    877 {
    878 	struct acpicpu_tstate *ts = arg1;
    879 	uint64_t val;
    880 	int i;
    881 
    882 	for (i = val = 0; i < ACPICPU_T_STATE_RETRY; i++) {
    883 
    884 		val = rdmsr(MSR_THERM_CONTROL);
    885 
    886 		if (val == ts->ts_status)
    887 			return;
    888 
    889 		DELAY(ts->ts_latency);
    890 	}
    891 
    892 	*(uintptr_t *)arg2 = EAGAIN;
    893 }
    894 
    895 /*
    896  * A kludge for backwards compatibility.
    897  */
    898 static int
    899 acpicpu_md_pstate_sysctl_init(void)
    900 {
    901 	const struct sysctlnode	*fnode, *mnode, *rnode;
    902 	const char *str;
    903 	int rv;
    904 
    905 	switch (cpu_vendor) {
    906 
    907 	case CPUVENDOR_IDT:
    908 	case CPUVENDOR_INTEL:
    909 		str = "est";
    910 		break;
    911 
    912 	case CPUVENDOR_AMD:
    913 		str = "powernow";
    914 		break;
    915 
    916 	default:
    917 		return ENODEV;
    918 	}
    919 
    920 
    921 	rv = sysctl_createv(&acpicpu_log, 0, NULL, &rnode,
    922 	    CTLFLAG_PERMANENT, CTLTYPE_NODE, "machdep", NULL,
    923 	    NULL, 0, NULL, 0, CTL_MACHDEP, CTL_EOL);
    924 
    925 	if (rv != 0)
    926 		goto fail;
    927 
    928 	rv = sysctl_createv(&acpicpu_log, 0, &rnode, &mnode,
    929 	    0, CTLTYPE_NODE, str, NULL,
    930 	    NULL, 0, NULL, 0, CTL_CREATE, CTL_EOL);
    931 
    932 	if (rv != 0)
    933 		goto fail;
    934 
    935 	rv = sysctl_createv(&acpicpu_log, 0, &mnode, &fnode,
    936 	    0, CTLTYPE_NODE, "frequency", NULL,
    937 	    NULL, 0, NULL, 0, CTL_CREATE, CTL_EOL);
    938 
    939 	if (rv != 0)
    940 		goto fail;
    941 
    942 	rv = sysctl_createv(&acpicpu_log, 0, &fnode, &rnode,
    943 	    CTLFLAG_READWRITE, CTLTYPE_INT, "target", NULL,
    944 	    acpicpu_md_pstate_sysctl_set, 0, NULL, 0, CTL_CREATE, CTL_EOL);
    945 
    946 	if (rv != 0)
    947 		goto fail;
    948 
    949 	rv = sysctl_createv(&acpicpu_log, 0, &fnode, &rnode,
    950 	    CTLFLAG_READONLY, CTLTYPE_INT, "current", NULL,
    951 	    acpicpu_md_pstate_sysctl_get, 0, NULL, 0, CTL_CREATE, CTL_EOL);
    952 
    953 	if (rv != 0)
    954 		goto fail;
    955 
    956 	rv = sysctl_createv(&acpicpu_log, 0, &fnode, &rnode,
    957 	    CTLFLAG_READONLY, CTLTYPE_STRING, "available", NULL,
    958 	    acpicpu_md_pstate_sysctl_all, 0, NULL, 0, CTL_CREATE, CTL_EOL);
    959 
    960 	if (rv != 0)
    961 		goto fail;
    962 
    963 	return 0;
    964 
    965 fail:
    966 	if (acpicpu_log != NULL) {
    967 		sysctl_teardown(&acpicpu_log);
    968 		acpicpu_log = NULL;
    969 	}
    970 
    971 	return rv;
    972 }
    973 
    974 static int
    975 acpicpu_md_pstate_sysctl_get(SYSCTLFN_ARGS)
    976 {
    977 	struct cpu_info *ci = curcpu();
    978 	struct acpicpu_softc *sc;
    979 	struct sysctlnode node;
    980 	uint32_t freq;
    981 	int err;
    982 
    983 	sc = acpicpu_sc[ci->ci_acpiid];
    984 
    985 	if (sc == NULL)
    986 		return ENXIO;
    987 
    988 	err = acpicpu_pstate_get(sc, &freq);
    989 
    990 	if (err != 0)
    991 		return err;
    992 
    993 	node = *rnode;
    994 	node.sysctl_data = &freq;
    995 
    996 	err = sysctl_lookup(SYSCTLFN_CALL(&node));
    997 
    998 	if (err != 0 || newp == NULL)
    999 		return err;
   1000 
   1001 	return 0;
   1002 }
   1003 
   1004 static int
   1005 acpicpu_md_pstate_sysctl_set(SYSCTLFN_ARGS)
   1006 {
   1007 	struct cpu_info *ci = curcpu();
   1008 	struct acpicpu_softc *sc;
   1009 	struct sysctlnode node;
   1010 	uint32_t freq;
   1011 	int err;
   1012 
   1013 	sc = acpicpu_sc[ci->ci_acpiid];
   1014 
   1015 	if (sc == NULL)
   1016 		return ENXIO;
   1017 
   1018 	err = acpicpu_pstate_get(sc, &freq);
   1019 
   1020 	if (err != 0)
   1021 		return err;
   1022 
   1023 	node = *rnode;
   1024 	node.sysctl_data = &freq;
   1025 
   1026 	err = sysctl_lookup(SYSCTLFN_CALL(&node));
   1027 
   1028 	if (err != 0 || newp == NULL)
   1029 		return err;
   1030 
   1031 	err = acpicpu_pstate_set(sc, freq);
   1032 
   1033 	if (err != 0)
   1034 		return err;
   1035 
   1036 	return 0;
   1037 }
   1038 
   1039 static int
   1040 acpicpu_md_pstate_sysctl_all(SYSCTLFN_ARGS)
   1041 {
   1042 	struct cpu_info *ci = curcpu();
   1043 	struct acpicpu_softc *sc;
   1044 	struct sysctlnode node;
   1045 	char buf[1024];
   1046 	size_t len;
   1047 	uint32_t i;
   1048 	int err;
   1049 
   1050 	sc = acpicpu_sc[ci->ci_acpiid];
   1051 
   1052 	if (sc == NULL)
   1053 		return ENXIO;
   1054 
   1055 	(void)memset(&buf, 0, sizeof(buf));
   1056 
   1057 	mutex_enter(&sc->sc_mtx);
   1058 
   1059 	for (len = 0, i = sc->sc_pstate_max; i < sc->sc_pstate_count; i++) {
   1060 
   1061 		if (sc->sc_pstate[i].ps_freq == 0)
   1062 			continue;
   1063 
   1064 		len += snprintf(buf + len, sizeof(buf) - len, "%u%s",
   1065 		    sc->sc_pstate[i].ps_freq,
   1066 		    i < (sc->sc_pstate_count - 1) ? " " : "");
   1067 	}
   1068 
   1069 	mutex_exit(&sc->sc_mtx);
   1070 
   1071 	node = *rnode;
   1072 	node.sysctl_data = buf;
   1073 
   1074 	err = sysctl_lookup(SYSCTLFN_CALL(&node));
   1075 
   1076 	if (err != 0 || newp == NULL)
   1077 		return err;
   1078 
   1079 	return 0;
   1080 }
   1081 
   1082