tsc.c revision 216276
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/i386/i386/tsc.c 216276 2010-12-07 22:43:25Z 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; 51110370Sphkint tsc_is_broken; 52184102Sjkimint tsc_is_invariant; 53110370Sphku_int tsc_present; 54167905Snjlstatic eventhandler_tag tsc_levels_tag, tsc_pre_tag, tsc_post_tag; 551390Ssos 56184102SjkimSYSCTL_INT(_kern_timecounter, OID_AUTO, invariant_tsc, CTLFLAG_RDTUN, 57184108Sjkim &tsc_is_invariant, 0, "Indicates whether the TSC is P-state invariant"); 58184108SjkimTUNABLE_INT("kern.timecounter.invariant_tsc", &tsc_is_invariant); 59184102Sjkim 60113348Sdes#ifdef SMP 61113348Sdesstatic int smp_tsc; 62121307SsilbySYSCTL_INT(_kern_timecounter, OID_AUTO, smp_tsc, CTLFLAG_RDTUN, &smp_tsc, 0, 63113348Sdes "Indicates whether the TSC is safe to use in SMP mode"); 64113348SdesTUNABLE_INT("kern.timecounter.smp_tsc", &smp_tsc); 65113348Sdes#endif 66113348Sdes 67167905Snjlstatic void tsc_freq_changed(void *arg, const struct cf_level *level, 68167905Snjl int status); 69167905Snjlstatic void tsc_freq_changing(void *arg, const struct cf_level *level, 70167905Snjl int *status); 7192765Salfredstatic unsigned tsc_get_timecount(struct timecounter *tc); 72167905Snjlstatic void tsc_levels_changed(void *arg, int unit); 7317353Sbde 7440610Sphkstatic struct timecounter tsc_timecounter = { 7533690Sphk tsc_get_timecount, /* get_timecount */ 7636741Sphk 0, /* no poll_pps */ 77167905Snjl ~0u, /* counter_mask */ 7833690Sphk 0, /* frequency */ 79167905Snjl "TSC", /* name */ 80118987Sphk 800, /* quality (adjusted in code) */ 8133690Sphk}; 8233690Sphk 831390Ssosvoid 84110370Sphkinit_TSC(void) 851390Ssos{ 86110370Sphk u_int64_t tscval[2]; 871390Ssos 8832054Sphk if (cpu_feature & CPUID_TSC) 8932054Sphk tsc_present = 1; 9032054Sphk else 9132054Sphk tsc_present = 0; 9232054Sphk 93110370Sphk if (!tsc_present) 94110370Sphk return; 9515508Sbde 96110370Sphk if (bootverbose) 97110370Sphk printf("Calibrating TSC clock ... "); 9815508Sbde 99110370Sphk tscval[0] = rdtsc(); 100110370Sphk DELAY(1000000); 101110370Sphk tscval[1] = rdtsc(); 10215508Sbde 103110370Sphk tsc_freq = tscval[1] - tscval[0]; 104110370Sphk if (bootverbose) 105110370Sphk printf("TSC clock: %ju Hz\n", (intmax_t)tsc_freq); 106167905Snjl 107216272Sjkim switch (cpu_vendor_id) { 108216272Sjkim case CPU_VENDOR_AMD: 109216272Sjkim if ((amd_pminfo & AMDPM_TSC_INVARIANT) || 110216272Sjkim CPUID_TO_FAMILY(cpu_id) >= 0x10 || cpu_id == 0x60fb2) 111216272Sjkim tsc_is_invariant = 1; 112216272Sjkim break; 113216272Sjkim case CPU_VENDOR_INTEL: 114216272Sjkim if ((amd_pminfo & AMDPM_TSC_INVARIANT) || 115216272Sjkim (CPUID_TO_FAMILY(cpu_id) == 0x6 && 116216272Sjkim CPUID_TO_MODEL(cpu_id) >= 0xe) || 117216272Sjkim (CPUID_TO_FAMILY(cpu_id) == 0xf && 118216272Sjkim CPUID_TO_MODEL(cpu_id) >= 0x3)) 119216272Sjkim tsc_is_invariant = 1; 120216272Sjkim break; 121216272Sjkim case CPU_VENDOR_CENTAUR: 122216272Sjkim if (CPUID_TO_FAMILY(cpu_id) == 0x6 && 123216272Sjkim CPUID_TO_MODEL(cpu_id) >= 0xf && 124216272Sjkim (rdmsr(0x1203) & 0x100000000ULL) == 0) 125216272Sjkim tsc_is_invariant = 1; 126216272Sjkim break; 127216272Sjkim } 128216272Sjkim 129167905Snjl /* 130216274Sjkim * Inform CPU accounting about our boot-time clock rate. This will 131216274Sjkim * be updated if someone loads a cpufreq driver after boot that 132216274Sjkim * discovers a new max frequency. 133167905Snjl */ 134155534Sphk set_cputicker(rdtsc, tsc_freq, 1); 135167905Snjl 136216274Sjkim if (tsc_is_invariant) 137216274Sjkim return; 138216274Sjkim 139167905Snjl /* Register to find out about changes in CPU frequency. */ 140184108Sjkim tsc_pre_tag = EVENTHANDLER_REGISTER(cpufreq_pre_change, 141184108Sjkim tsc_freq_changing, NULL, EVENTHANDLER_PRI_FIRST); 142167905Snjl tsc_post_tag = EVENTHANDLER_REGISTER(cpufreq_post_change, 143167905Snjl tsc_freq_changed, NULL, EVENTHANDLER_PRI_FIRST); 144167905Snjl tsc_levels_tag = EVENTHANDLER_REGISTER(cpufreq_levels_changed, 145167905Snjl tsc_levels_changed, NULL, EVENTHANDLER_PRI_ANY); 146118550Sphk} 14734617Sphk 148118550Sphkvoid 149118550Sphkinit_TSC_tc(void) 150118550Sphk{ 151209103Smav 152209103Smav if (!tsc_present) 153209103Smav return; 154209103Smav 15534617Sphk /* 156160964Syar * We can not use the TSC if we support APM. Precise timekeeping 15749186Smsmith * on an APM'ed machine is at best a fools pursuit, since 15834617Sphk * any and all of the time spent in various SMM code can't 15934617Sphk * be reliably accounted for. Reading the RTC is your only 160160964Syar * source of reliable time info. The i8254 loses too, of course, 16134617Sphk * but we need to have some kind of time... 16249186Smsmith * We don't know at this point whether APM is going to be used 16349186Smsmith * or not, nor when it might be activated. Play it safe. 16434617Sphk */ 16585835Siwasaki if (power_pm_get_type() == POWER_PM_TYPE_APM) { 166118987Sphk tsc_timecounter.tc_quality = -1000; 16785835Siwasaki if (bootverbose) 168110370Sphk printf("TSC timecounter disabled: APM enabled.\n"); 16964031Sphk } 17034617Sphk 171118987Sphk#ifdef SMP 172118987Sphk /* 173118987Sphk * We can not use the TSC in SMP mode unless the TSCs on all CPUs 174118987Sphk * are somehow synchronized. Some hardware configurations do 175118987Sphk * this, but we have no way of determining whether this is the 176118987Sphk * case, so we do not use the TSC in multi-processor systems 177118987Sphk * unless the user indicated (by setting kern.timecounter.smp_tsc 178118987Sphk * to 1) that he believes that his TSCs are synchronized. 179118987Sphk */ 180118987Sphk if (mp_ncpus > 1 && !smp_tsc) 181118987Sphk tsc_timecounter.tc_quality = -100; 182118987Sphk#endif 183118987Sphk 184209103Smav if (tsc_freq != 0 && !tsc_is_broken) { 18540610Sphk tsc_timecounter.tc_frequency = tsc_freq; 18658377Sphk tc_init(&tsc_timecounter); 18733690Sphk } 1884Srgrimes} 1894Srgrimes 190167905Snjl/* 191167905Snjl * When cpufreq levels change, find out about the (new) max frequency. We 192167905Snjl * use this to update CPU accounting in case it got a lower estimate at boot. 193167905Snjl */ 194167905Snjlstatic void 195167905Snjltsc_levels_changed(void *arg, int unit) 196167905Snjl{ 197167905Snjl device_t cf_dev; 198167905Snjl struct cf_level *levels; 199167905Snjl int count, error; 200167905Snjl uint64_t max_freq; 201167905Snjl 202167905Snjl /* Only use values from the first CPU, assuming all are equal. */ 203167905Snjl if (unit != 0) 204167905Snjl return; 205167905Snjl 206167905Snjl /* Find the appropriate cpufreq device instance. */ 207167905Snjl cf_dev = devclass_get_device(devclass_find("cpufreq"), unit); 208167905Snjl if (cf_dev == NULL) { 209167905Snjl printf("tsc_levels_changed() called but no cpufreq device?\n"); 210167905Snjl return; 211167905Snjl } 212167905Snjl 213167905Snjl /* Get settings from the device and find the max frequency. */ 214167905Snjl count = 64; 215167905Snjl levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 216167905Snjl if (levels == NULL) 217167905Snjl return; 218167905Snjl error = CPUFREQ_LEVELS(cf_dev, levels, &count); 219167905Snjl if (error == 0 && count != 0) { 220167905Snjl max_freq = (uint64_t)levels[0].total_set.freq * 1000000; 221167905Snjl set_cputicker(rdtsc, max_freq, 1); 222167905Snjl } else 223167905Snjl printf("tsc_levels_changed: no max freq found\n"); 224167905Snjl free(levels, M_TEMP); 225167905Snjl} 226167905Snjl 227167905Snjl/* 228167905Snjl * If the TSC timecounter is in use, veto the pending change. It may be 229167905Snjl * possible in the future to handle a dynamically-changing timecounter rate. 230167905Snjl */ 231167905Snjlstatic void 232167905Snjltsc_freq_changing(void *arg, const struct cf_level *level, int *status) 233167905Snjl{ 234167905Snjl 235216274Sjkim if (*status != 0 || timecounter != &tsc_timecounter) 236167905Snjl return; 237167905Snjl 238167905Snjl printf("timecounter TSC must not be in use when " 239184102Sjkim "changing frequencies; change denied\n"); 240167905Snjl *status = EBUSY; 241167905Snjl} 242167905Snjl 243167905Snjl/* Update TSC freq with the value indicated by the caller. */ 244167905Snjlstatic void 245167905Snjltsc_freq_changed(void *arg, const struct cf_level *level, int status) 246167905Snjl{ 247216276Sjkim 248216276Sjkim /* If there was an error during the transition, don't do anything. */ 249216274Sjkim if (status != 0) 250167905Snjl return; 251167905Snjl 252167905Snjl /* Total setting for this level gives the new frequency in MHz. */ 253167905Snjl tsc_freq = (uint64_t)level->total_set.freq * 1000000; 254167905Snjl tsc_timecounter.tc_frequency = tsc_freq; 255167905Snjl} 256167905Snjl 25715508Sbdestatic int 25862573Sphksysctl_machdep_tsc_freq(SYSCTL_HANDLER_ARGS) 25915508Sbde{ 26015508Sbde int error; 261110039Sphk uint64_t freq; 26215508Sbde 26348888Sbde if (tsc_timecounter.tc_frequency == 0) 26415508Sbde return (EOPNOTSUPP); 26532005Sphk freq = tsc_freq; 266170289Sdwmalone error = sysctl_handle_quad(oidp, &freq, 0, req); 26733690Sphk if (error == 0 && req->newptr != NULL) { 26833690Sphk tsc_freq = freq; 26940610Sphk tsc_timecounter.tc_frequency = tsc_freq; 27033690Sphk } 27115508Sbde return (error); 27215508Sbde} 27315508Sbde 274110039SphkSYSCTL_PROC(_machdep, OID_AUTO, tsc_freq, CTLTYPE_QUAD | CTLFLAG_RW, 275211082Sdwmalone 0, 0, sysctl_machdep_tsc_freq, "QU", ""); 27633690Sphk 27736441Sphkstatic unsigned 27836719Sphktsc_get_timecount(struct timecounter *tc) 27933690Sphk{ 28036198Sphk return (rdtsc()); 28133690Sphk} 282