tsc.c revision 219461
14Srgrimes/*- 2110379Sphk * Copyright (c) 1998-2003 Poul-Henning Kamp 34Srgrimes * All rights reserved. 44Srgrimes * 54Srgrimes * Redistribution and use in source and binary forms, with or without 64Srgrimes * modification, are permitted provided that the following conditions 74Srgrimes * are met: 84Srgrimes * 1. Redistributions of source code must retain the above copyright 94Srgrimes * notice, this list of conditions and the following disclaimer. 104Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 114Srgrimes * notice, this list of conditions and the following disclaimer in the 124Srgrimes * documentation and/or other materials provided with the distribution. 134Srgrimes * 14110379Sphk * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 154Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 164Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17110379Sphk * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 184Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 194Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 204Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 214Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 224Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 234Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 244Srgrimes * SUCH DAMAGE. 254Srgrimes */ 264Srgrimes 27115683Sobrien#include <sys/cdefs.h> 28115683Sobrien__FBSDID("$FreeBSD: head/sys/x86/x86/tsc.c 219461 2011-03-10 20:02:58Z jkim $"); 29115683Sobrien 3016299Spst#include "opt_clock.h" 3113228Swollman 322056Swollman#include <sys/param.h> 33167905Snjl#include <sys/bus.h> 34167905Snjl#include <sys/cpu.h> 35167905Snjl#include <sys/malloc.h> 362056Swollman#include <sys/systm.h> 37113348Sdes#include <sys/sysctl.h> 382056Swollman#include <sys/time.h> 3958377Sphk#include <sys/timetc.h> 402056Swollman#include <sys/kernel.h> 4185835Siwasaki#include <sys/power.h> 42113348Sdes#include <sys/smp.h> 434180Sbde#include <machine/clock.h> 44216272Sjkim#include <machine/cputypes.h> 4532054Sphk#include <machine/md_var.h> 4632054Sphk#include <machine/specialreg.h> 4715508Sbde 48167905Snjl#include "cpufreq_if.h" 49167905Snjl 50216163Sjkimuint64_t tsc_freq; 51184102Sjkimint tsc_is_invariant; 52216279Sjkimint tsc_present; 53167905Snjlstatic eventhandler_tag tsc_levels_tag, tsc_pre_tag, tsc_post_tag; 541390Ssos 55184102SjkimSYSCTL_INT(_kern_timecounter, OID_AUTO, invariant_tsc, CTLFLAG_RDTUN, 56184108Sjkim &tsc_is_invariant, 0, "Indicates whether the TSC is P-state invariant"); 57184108SjkimTUNABLE_INT("kern.timecounter.invariant_tsc", &tsc_is_invariant); 58184102Sjkim 59113348Sdes#ifdef SMP 60113348Sdesstatic int smp_tsc; 61121307SsilbySYSCTL_INT(_kern_timecounter, OID_AUTO, smp_tsc, CTLFLAG_RDTUN, &smp_tsc, 0, 62113348Sdes "Indicates whether the TSC is safe to use in SMP mode"); 63113348SdesTUNABLE_INT("kern.timecounter.smp_tsc", &smp_tsc); 64113348Sdes#endif 65113348Sdes 66167905Snjlstatic void tsc_freq_changed(void *arg, const struct cf_level *level, 67167905Snjl int status); 68167905Snjlstatic void tsc_freq_changing(void *arg, const struct cf_level *level, 69167905Snjl int *status); 7092765Salfredstatic unsigned tsc_get_timecount(struct timecounter *tc); 71167905Snjlstatic void tsc_levels_changed(void *arg, int unit); 7217353Sbde 7340610Sphkstatic struct timecounter tsc_timecounter = { 7433690Sphk tsc_get_timecount, /* get_timecount */ 7536741Sphk 0, /* no poll_pps */ 76167905Snjl ~0u, /* counter_mask */ 7733690Sphk 0, /* frequency */ 78167905Snjl "TSC", /* name */ 79118987Sphk 800, /* quality (adjusted in code) */ 8033690Sphk}; 8133690Sphk 821390Ssosvoid 83110370Sphkinit_TSC(void) 841390Ssos{ 85110370Sphk u_int64_t tscval[2]; 861390Ssos 8732054Sphk if (cpu_feature & CPUID_TSC) 8832054Sphk tsc_present = 1; 8932054Sphk else 9032054Sphk tsc_present = 0; 9132054Sphk 92110370Sphk if (!tsc_present) 93110370Sphk return; 9415508Sbde 95110370Sphk if (bootverbose) 96110370Sphk printf("Calibrating TSC clock ... "); 9715508Sbde 98110370Sphk tscval[0] = rdtsc(); 99110370Sphk DELAY(1000000); 100110370Sphk tscval[1] = rdtsc(); 10115508Sbde 102110370Sphk tsc_freq = tscval[1] - tscval[0]; 103110370Sphk if (bootverbose) 104110370Sphk printf("TSC clock: %ju Hz\n", (intmax_t)tsc_freq); 105167905Snjl 106216272Sjkim switch (cpu_vendor_id) { 107216272Sjkim case CPU_VENDOR_AMD: 108216272Sjkim if ((amd_pminfo & AMDPM_TSC_INVARIANT) || 109216337Sjkim CPUID_TO_FAMILY(cpu_id) >= 0x10) 110216272Sjkim tsc_is_invariant = 1; 111216272Sjkim break; 112216272Sjkim case CPU_VENDOR_INTEL: 113216272Sjkim if ((amd_pminfo & AMDPM_TSC_INVARIANT) || 114216272Sjkim (CPUID_TO_FAMILY(cpu_id) == 0x6 && 115216272Sjkim CPUID_TO_MODEL(cpu_id) >= 0xe) || 116216272Sjkim (CPUID_TO_FAMILY(cpu_id) == 0xf && 117216272Sjkim CPUID_TO_MODEL(cpu_id) >= 0x3)) 118216272Sjkim tsc_is_invariant = 1; 119216272Sjkim break; 120216272Sjkim case CPU_VENDOR_CENTAUR: 121216272Sjkim if (CPUID_TO_FAMILY(cpu_id) == 0x6 && 122216272Sjkim CPUID_TO_MODEL(cpu_id) >= 0xf && 123216272Sjkim (rdmsr(0x1203) & 0x100000000ULL) == 0) 124216272Sjkim tsc_is_invariant = 1; 125216272Sjkim break; 126216272Sjkim } 127216272Sjkim 128167905Snjl /* 129216274Sjkim * Inform CPU accounting about our boot-time clock rate. This will 130216274Sjkim * be updated if someone loads a cpufreq driver after boot that 131216274Sjkim * discovers a new max frequency. 132167905Snjl */ 133155534Sphk set_cputicker(rdtsc, tsc_freq, 1); 134167905Snjl 135216274Sjkim if (tsc_is_invariant) 136216274Sjkim return; 137216274Sjkim 138167905Snjl /* Register to find out about changes in CPU frequency. */ 139184108Sjkim tsc_pre_tag = EVENTHANDLER_REGISTER(cpufreq_pre_change, 140184108Sjkim tsc_freq_changing, NULL, EVENTHANDLER_PRI_FIRST); 141167905Snjl tsc_post_tag = EVENTHANDLER_REGISTER(cpufreq_post_change, 142167905Snjl tsc_freq_changed, NULL, EVENTHANDLER_PRI_FIRST); 143167905Snjl tsc_levels_tag = EVENTHANDLER_REGISTER(cpufreq_levels_changed, 144167905Snjl tsc_levels_changed, NULL, EVENTHANDLER_PRI_ANY); 145118550Sphk} 14634617Sphk 147118550Sphkvoid 148118550Sphkinit_TSC_tc(void) 149118550Sphk{ 150209103Smav 151209103Smav if (!tsc_present) 152209103Smav return; 153209103Smav 15434617Sphk /* 155160964Syar * We can not use the TSC if we support APM. Precise timekeeping 15649186Smsmith * on an APM'ed machine is at best a fools pursuit, since 15734617Sphk * any and all of the time spent in various SMM code can't 15834617Sphk * be reliably accounted for. Reading the RTC is your only 159160964Syar * source of reliable time info. The i8254 loses too, of course, 16034617Sphk * but we need to have some kind of time... 16149186Smsmith * We don't know at this point whether APM is going to be used 16249186Smsmith * or not, nor when it might be activated. Play it safe. 16334617Sphk */ 16485835Siwasaki if (power_pm_get_type() == POWER_PM_TYPE_APM) { 165118987Sphk tsc_timecounter.tc_quality = -1000; 16685835Siwasaki if (bootverbose) 167110370Sphk printf("TSC timecounter disabled: APM enabled.\n"); 16864031Sphk } 16934617Sphk 170118987Sphk#ifdef SMP 171118987Sphk /* 172118987Sphk * We can not use the TSC in SMP mode unless the TSCs on all CPUs 173118987Sphk * are somehow synchronized. Some hardware configurations do 174118987Sphk * this, but we have no way of determining whether this is the 175118987Sphk * case, so we do not use the TSC in multi-processor systems 176118987Sphk * unless the user indicated (by setting kern.timecounter.smp_tsc 177118987Sphk * to 1) that he believes that his TSCs are synchronized. 178118987Sphk */ 179118987Sphk if (mp_ncpus > 1 && !smp_tsc) 180118987Sphk tsc_timecounter.tc_quality = -100; 181118987Sphk#endif 182118987Sphk 183219461Sjkim if (tsc_freq != 0) { 18440610Sphk tsc_timecounter.tc_frequency = tsc_freq; 18558377Sphk tc_init(&tsc_timecounter); 18633690Sphk } 1874Srgrimes} 1884Srgrimes 189167905Snjl/* 190167905Snjl * When cpufreq levels change, find out about the (new) max frequency. We 191167905Snjl * use this to update CPU accounting in case it got a lower estimate at boot. 192167905Snjl */ 193167905Snjlstatic void 194167905Snjltsc_levels_changed(void *arg, int unit) 195167905Snjl{ 196167905Snjl device_t cf_dev; 197167905Snjl struct cf_level *levels; 198167905Snjl int count, error; 199167905Snjl uint64_t max_freq; 200167905Snjl 201167905Snjl /* Only use values from the first CPU, assuming all are equal. */ 202167905Snjl if (unit != 0) 203167905Snjl return; 204167905Snjl 205167905Snjl /* Find the appropriate cpufreq device instance. */ 206167905Snjl cf_dev = devclass_get_device(devclass_find("cpufreq"), unit); 207167905Snjl if (cf_dev == NULL) { 208167905Snjl printf("tsc_levels_changed() called but no cpufreq device?\n"); 209167905Snjl return; 210167905Snjl } 211167905Snjl 212167905Snjl /* Get settings from the device and find the max frequency. */ 213167905Snjl count = 64; 214167905Snjl levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 215167905Snjl if (levels == NULL) 216167905Snjl return; 217167905Snjl error = CPUFREQ_LEVELS(cf_dev, levels, &count); 218167905Snjl if (error == 0 && count != 0) { 219167905Snjl max_freq = (uint64_t)levels[0].total_set.freq * 1000000; 220167905Snjl set_cputicker(rdtsc, max_freq, 1); 221167905Snjl } else 222167905Snjl printf("tsc_levels_changed: no max freq found\n"); 223167905Snjl free(levels, M_TEMP); 224167905Snjl} 225167905Snjl 226167905Snjl/* 227167905Snjl * If the TSC timecounter is in use, veto the pending change. It may be 228167905Snjl * possible in the future to handle a dynamically-changing timecounter rate. 229167905Snjl */ 230167905Snjlstatic void 231167905Snjltsc_freq_changing(void *arg, const struct cf_level *level, int *status) 232167905Snjl{ 233167905Snjl 234216274Sjkim if (*status != 0 || timecounter != &tsc_timecounter) 235167905Snjl return; 236167905Snjl 237167905Snjl printf("timecounter TSC must not be in use when " 238184102Sjkim "changing frequencies; change denied\n"); 239167905Snjl *status = EBUSY; 240167905Snjl} 241167905Snjl 242167905Snjl/* Update TSC freq with the value indicated by the caller. */ 243167905Snjlstatic void 244167905Snjltsc_freq_changed(void *arg, const struct cf_level *level, int status) 245167905Snjl{ 246216276Sjkim 247216276Sjkim /* If there was an error during the transition, don't do anything. */ 248216274Sjkim if (status != 0) 249167905Snjl return; 250167905Snjl 251167905Snjl /* Total setting for this level gives the new frequency in MHz. */ 252167905Snjl tsc_freq = (uint64_t)level->total_set.freq * 1000000; 253167905Snjl tsc_timecounter.tc_frequency = tsc_freq; 254167905Snjl} 255167905Snjl 25615508Sbdestatic int 25762573Sphksysctl_machdep_tsc_freq(SYSCTL_HANDLER_ARGS) 25815508Sbde{ 25915508Sbde int error; 260110039Sphk uint64_t freq; 26115508Sbde 26248888Sbde if (tsc_timecounter.tc_frequency == 0) 26315508Sbde return (EOPNOTSUPP); 26432005Sphk freq = tsc_freq; 265217616Smdf error = sysctl_handle_64(oidp, &freq, 0, req); 26633690Sphk if (error == 0 && req->newptr != NULL) { 26733690Sphk tsc_freq = freq; 26840610Sphk tsc_timecounter.tc_frequency = tsc_freq; 26933690Sphk } 27015508Sbde return (error); 27115508Sbde} 27215508Sbde 273217616SmdfSYSCTL_PROC(_machdep, OID_AUTO, tsc_freq, CTLTYPE_U64 | CTLFLAG_RW, 274211082Sdwmalone 0, 0, sysctl_machdep_tsc_freq, "QU", ""); 27533690Sphk 27636441Sphkstatic unsigned 27736719Sphktsc_get_timecount(struct timecounter *tc) 27833690Sphk{ 27936198Sphk return (rdtsc()); 28033690Sphk} 281