1 /*- 2 * Copyright (c) 1998-2003 Poul-Henning Kamp 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include "opt_clock.h" 31 32 #include <sys/param.h> 33 #include <sys/bus.h> 34 #include <sys/cpu.h> 35 #include <sys/malloc.h> 36 #include <sys/systm.h> 37 #include <sys/sysctl.h> 38 #include <sys/time.h> 39 #include <sys/timetc.h> 40 #include <sys/kernel.h> 41 #include <sys/power.h> 42 #include <sys/smp.h> 43 #include <machine/clock.h> 44 #include <machine/cputypes.h> 45 #include <machine/md_var.h> 46 #include <machine/specialreg.h> 47 48 #include "cpufreq_if.h" 49 50 uint64_t tsc_freq; 51 int tsc_is_invariant; 52 static eventhandler_tag tsc_levels_tag, tsc_pre_tag, tsc_post_tag; 53 54 SYSCTL_INT(_kern_timecounter, OID_AUTO, invariant_tsc, CTLFLAG_RDTUN, 55 &tsc_is_invariant, 0, "Indicates whether the TSC is P-state invariant"); 56 TUNABLE_INT("kern.timecounter.invariant_tsc", &tsc_is_invariant); 57 58 #ifdef SMP 59 static int smp_tsc; 60 SYSCTL_INT(_kern_timecounter, OID_AUTO, smp_tsc, CTLFLAG_RDTUN, &smp_tsc, 0, 61 "Indicates whether the TSC is safe to use in SMP mode"); 62 TUNABLE_INT("kern.timecounter.smp_tsc", &smp_tsc); 63 #endif 64 65 static int tsc_disabled; 66 SYSCTL_INT(_machdep, OID_AUTO, disable_tsc, CTLFLAG_RDTUN, &tsc_disabled, 0, 67 "Disable x86 Time Stamp Counter"); 68 TUNABLE_INT("machdep.disable_tsc", &tsc_disabled); 69 70 static void tsc_freq_changed(void *arg, const struct cf_level *level, 71 int status); 72 static void tsc_freq_changing(void *arg, const struct cf_level *level, 73 int *status); 74 static unsigned tsc_get_timecount(struct timecounter *tc); 75 static void tsc_levels_changed(void *arg, int unit); 76 77 static struct timecounter tsc_timecounter = { 78 tsc_get_timecount, /* get_timecount */ 79 0, /* no poll_pps */ 80 ~0u, /* counter_mask */ 81 0, /* frequency */ 82 "TSC", /* name */ 83 800, /* quality (adjusted in code) */ 84 }; 85 86 void 87 init_TSC(void) 88 { 89 u_int64_t tscval[2]; 90 91 if ((cpu_feature & CPUID_TSC) == 0 || tsc_disabled) 92 return; 93 94 if (bootverbose) 95 printf("Calibrating TSC clock ... "); 96 97 tscval[0] = rdtsc(); 98 DELAY(1000000); 99 tscval[1] = rdtsc(); 100 101 tsc_freq = tscval[1] - tscval[0]; 102 if (bootverbose) 103 printf("TSC clock: %ju Hz\n", (intmax_t)tsc_freq); 104 105 switch (cpu_vendor_id) { 106 case CPU_VENDOR_AMD: 107 if ((amd_pminfo & AMDPM_TSC_INVARIANT) != 0 || 108 (vm_guest == VM_GUEST_NO && 109 CPUID_TO_FAMILY(cpu_id) >= 0x10)) 110 tsc_is_invariant = 1; 111 break; 112 case CPU_VENDOR_INTEL: 113 if ((amd_pminfo & AMDPM_TSC_INVARIANT) != 0 || 114 (vm_guest == VM_GUEST_NO && 115 ((CPUID_TO_FAMILY(cpu_id) == 0x6 && 116 CPUID_TO_MODEL(cpu_id) >= 0xe) || 117 (CPUID_TO_FAMILY(cpu_id) == 0xf && 118 CPUID_TO_MODEL(cpu_id) >= 0x3)))) 119 tsc_is_invariant = 1; 120 break; 121 case CPU_VENDOR_CENTAUR: 122 if (vm_guest == VM_GUEST_NO && 123 CPUID_TO_FAMILY(cpu_id) == 0x6 && 124 CPUID_TO_MODEL(cpu_id) >= 0xf && 125 (rdmsr(0x1203) & 0x100000000ULL) == 0) 126 tsc_is_invariant = 1; 127 break; 128 } 129 130 /* 131 * Inform CPU accounting about our boot-time clock rate. This will 132 * be updated if someone loads a cpufreq driver after boot that 133 * discovers a new max frequency. 134 */ 135 set_cputicker(rdtsc, tsc_freq, 1); 136 137 if (tsc_is_invariant) 138 return; 139 140 /* Register to find out about changes in CPU frequency. */ 141 tsc_pre_tag = EVENTHANDLER_REGISTER(cpufreq_pre_change, 142 tsc_freq_changing, NULL, EVENTHANDLER_PRI_FIRST); 143 tsc_post_tag = EVENTHANDLER_REGISTER(cpufreq_post_change, 144 tsc_freq_changed, NULL, EVENTHANDLER_PRI_FIRST); 145 tsc_levels_tag = EVENTHANDLER_REGISTER(cpufreq_levels_changed, 146 tsc_levels_changed, NULL, EVENTHANDLER_PRI_ANY); 147 } 148 149 void 150 init_TSC_tc(void) 151 { 152 153 if ((cpu_feature & CPUID_TSC) == 0 || tsc_disabled) 154 return; 155 156 /* 157 * We can not use the TSC if we support APM. Precise timekeeping 158 * on an APM'ed machine is at best a fools pursuit, since 159 * any and all of the time spent in various SMM code can't 160 * be reliably accounted for. Reading the RTC is your only 161 * source of reliable time info. The i8254 loses too, of course, 162 * but we need to have some kind of time... 163 * We don't know at this point whether APM is going to be used 164 * or not, nor when it might be activated. Play it safe. 165 */ 166 if (power_pm_get_type() == POWER_PM_TYPE_APM) { 167 tsc_timecounter.tc_quality = -1000; 168 if (bootverbose) 169 printf("TSC timecounter disabled: APM enabled.\n"); 170 } 171 172 #ifdef SMP 173 /* 174 * We can not use the TSC in SMP mode unless the TSCs on all CPUs 175 * are somehow synchronized. Some hardware configurations do 176 * this, but we have no way of determining whether this is the 177 * case, so we do not use the TSC in multi-processor systems 178 * unless the user indicated (by setting kern.timecounter.smp_tsc 179 * to 1) that he believes that his TSCs are synchronized. 180 */ 181 if (mp_ncpus > 1 && !smp_tsc) 182 tsc_timecounter.tc_quality = -100; 183 #endif 184 185 if (tsc_freq != 0) { 186 tsc_timecounter.tc_frequency = tsc_freq; 187 tc_init(&tsc_timecounter); 188 } 189 } 190 191 /* 192 * When cpufreq levels change, find out about the (new) max frequency. We 193 * use this to update CPU accounting in case it got a lower estimate at boot. 194 */ 195 static void 196 tsc_levels_changed(void *arg, int unit) 197 { 198 device_t cf_dev; 199 struct cf_level *levels; 200 int count, error; 201 uint64_t max_freq; 202 203 /* Only use values from the first CPU, assuming all are equal. */ 204 if (unit != 0) 205 return; 206 207 /* Find the appropriate cpufreq device instance. */ 208 cf_dev = devclass_get_device(devclass_find("cpufreq"), unit); 209 if (cf_dev == NULL) { 210 printf("tsc_levels_changed() called but no cpufreq device?\n"); 211 return; 212 } 213 214 /* Get settings from the device and find the max frequency. */ 215 count = 64; 216 levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 217 if (levels == NULL) 218 return; 219 error = CPUFREQ_LEVELS(cf_dev, levels, &count); 220 if (error == 0 && count != 0) { 221 max_freq = (uint64_t)levels[0].total_set.freq * 1000000; 222 set_cputicker(rdtsc, max_freq, 1); 223 } else 224 printf("tsc_levels_changed: no max freq found\n"); 225 free(levels, M_TEMP); 226 } 227 228 /* 229 * If the TSC timecounter is in use, veto the pending change. It may be 230 * possible in the future to handle a dynamically-changing timecounter rate. 231 */ 232 static void 233 tsc_freq_changing(void *arg, const struct cf_level *level, int *status) 234 { 235 236 if (*status != 0 || timecounter != &tsc_timecounter) 237 return; 238 239 printf("timecounter TSC must not be in use when " 240 "changing frequencies; change denied\n"); 241 *status = EBUSY; 242 } 243 244 /* Update TSC freq with the value indicated by the caller. */ 245 static void 246 tsc_freq_changed(void *arg, const struct cf_level *level, int status) 247 { 248 249 /* If there was an error during the transition, don't do anything. */ 250 if (tsc_disabled || status != 0) 251 return; 252 253 /* Total setting for this level gives the new frequency in MHz. */ 254 tsc_freq = (uint64_t)level->total_set.freq * 1000000; 255 tsc_timecounter.tc_frequency = tsc_freq; 256 } 257 258 static int 259 sysctl_machdep_tsc_freq(SYSCTL_HANDLER_ARGS) 260 { 261 int error; 262 uint64_t freq; 263 264 if (tsc_timecounter.tc_frequency == 0) 265 return (EOPNOTSUPP); 266 freq = tsc_freq; 267 error = sysctl_handle_64(oidp, &freq, 0, req); 268 if (error == 0 && req->newptr != NULL) { 269 tsc_freq = freq; 270 tsc_timecounter.tc_frequency = tsc_freq; 271 } 272 return (error); 273 } 274 275 SYSCTL_PROC(_machdep, OID_AUTO, tsc_freq, CTLTYPE_U64 | CTLFLAG_RW, 276 0, 0, sysctl_machdep_tsc_freq, "QU", ""); 277 278 static unsigned 279 tsc_get_timecount(struct timecounter *tc) 280 { 281 return (rdtsc()); 282 } 283