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 int tsc_present; 53 static eventhandler_tag tsc_levels_tag, tsc_pre_tag, tsc_post_tag; 54 55 SYSCTL_INT(_kern_timecounter, OID_AUTO, invariant_tsc, CTLFLAG_RDTUN, 56 &tsc_is_invariant, 0, "Indicates whether the TSC is P-state invariant"); 57 TUNABLE_INT("kern.timecounter.invariant_tsc", &tsc_is_invariant); 58 59 #ifdef SMP 60 static int smp_tsc; 61 SYSCTL_INT(_kern_timecounter, OID_AUTO, smp_tsc, CTLFLAG_RDTUN, &smp_tsc, 0, 62 "Indicates whether the TSC is safe to use in SMP mode"); 63 TUNABLE_INT("kern.timecounter.smp_tsc", &smp_tsc); 64 #endif 65 66 static int tsc_disabled; 67 SYSCTL_INT(_machdep, OID_AUTO, disable_tsc, CTLFLAG_RDTUN, &tsc_disabled, 0, 68 "Disable x86 Time Stamp Counter"); 69 TUNABLE_INT("machdep.disable_tsc", &tsc_disabled); 70 71 static void tsc_freq_changed(void *arg, const struct cf_level *level, 72 int status); 73 static void tsc_freq_changing(void *arg, const struct cf_level *level, 74 int *status); 75 static unsigned tsc_get_timecount(struct timecounter *tc); 76 static void tsc_levels_changed(void *arg, int unit); 77 78 static struct timecounter tsc_timecounter = { 79 tsc_get_timecount, /* get_timecount */ 80 0, /* no poll_pps */ 81 ~0u, /* counter_mask */ 82 0, /* frequency */ 83 "TSC", /* name */ 84 800, /* quality (adjusted in code) */ 85 }; 86 87 void 88 init_TSC(void) 89 { 90 u_int64_t tscval[2]; 91 92 if ((cpu_feature & CPUID_TSC) == 0) 93 return; 94 tsc_present = 1; 95 96 if (tsc_disabled) 97 return; 98 99 if (bootverbose) 100 printf("Calibrating TSC clock ... "); 101 102 tscval[0] = rdtsc(); 103 DELAY(1000000); 104 tscval[1] = rdtsc(); 105 106 tsc_freq = tscval[1] - tscval[0]; 107 if (bootverbose) 108 printf("TSC clock: %ju Hz\n", (intmax_t)tsc_freq); 109 110 switch (cpu_vendor_id) { 111 case CPU_VENDOR_AMD: 112 if ((amd_pminfo & AMDPM_TSC_INVARIANT) != 0 || 113 (vm_guest == VM_GUEST_NO && 114 CPUID_TO_FAMILY(cpu_id) >= 0x10)) 115 tsc_is_invariant = 1; 116 break; 117 case CPU_VENDOR_INTEL: 118 if ((amd_pminfo & AMDPM_TSC_INVARIANT) != 0 || 119 (vm_guest == VM_GUEST_NO && 120 ((CPUID_TO_FAMILY(cpu_id) == 0x6 && 121 CPUID_TO_MODEL(cpu_id) >= 0xe) || 122 (CPUID_TO_FAMILY(cpu_id) == 0xf && 123 CPUID_TO_MODEL(cpu_id) >= 0x3)))) 124 tsc_is_invariant = 1; 125 break; 126 case CPU_VENDOR_CENTAUR: 127 if (vm_guest == VM_GUEST_NO && 128 CPUID_TO_FAMILY(cpu_id) == 0x6 && 129 CPUID_TO_MODEL(cpu_id) >= 0xf && 130 (rdmsr(0x1203) & 0x100000000ULL) == 0) 131 tsc_is_invariant = 1; 132 break; 133 } 134 135 /* 136 * Inform CPU accounting about our boot-time clock rate. This will 137 * be updated if someone loads a cpufreq driver after boot that 138 * discovers a new max frequency. 139 */ 140 set_cputicker(rdtsc, tsc_freq, 1); 141 142 if (tsc_is_invariant) 143 return; 144 145 /* Register to find out about changes in CPU frequency. */ 146 tsc_pre_tag = EVENTHANDLER_REGISTER(cpufreq_pre_change, 147 tsc_freq_changing, NULL, EVENTHANDLER_PRI_FIRST); 148 tsc_post_tag = EVENTHANDLER_REGISTER(cpufreq_post_change, 149 tsc_freq_changed, NULL, EVENTHANDLER_PRI_FIRST); 150 tsc_levels_tag = EVENTHANDLER_REGISTER(cpufreq_levels_changed, 151 tsc_levels_changed, NULL, EVENTHANDLER_PRI_ANY); 152 } 153 154 void 155 init_TSC_tc(void) 156 { 157 158 if (!tsc_present || tsc_disabled) 159 return; 160 161 /* 162 * We can not use the TSC if we support APM. Precise timekeeping 163 * on an APM'ed machine is at best a fools pursuit, since 164 * any and all of the time spent in various SMM code can't 165 * be reliably accounted for. Reading the RTC is your only 166 * source of reliable time info. The i8254 loses too, of course, 167 * but we need to have some kind of time... 168 * We don't know at this point whether APM is going to be used 169 * or not, nor when it might be activated. Play it safe. 170 */ 171 if (power_pm_get_type() == POWER_PM_TYPE_APM) { 172 tsc_timecounter.tc_quality = -1000; 173 if (bootverbose) 174 printf("TSC timecounter disabled: APM enabled.\n"); 175 } 176 177 #ifdef SMP 178 /* 179 * We can not use the TSC in SMP mode unless the TSCs on all CPUs 180 * are somehow synchronized. Some hardware configurations do 181 * this, but we have no way of determining whether this is the 182 * case, so we do not use the TSC in multi-processor systems 183 * unless the user indicated (by setting kern.timecounter.smp_tsc 184 * to 1) that he believes that his TSCs are synchronized. 185 */ 186 if (mp_ncpus > 1 && !smp_tsc) 187 tsc_timecounter.tc_quality = -100; 188 #endif 189 190 if (tsc_freq != 0) { 191 tsc_timecounter.tc_frequency = tsc_freq; 192 tc_init(&tsc_timecounter); 193 } 194 } 195 196 /* 197 * When cpufreq levels change, find out about the (new) max frequency. We 198 * use this to update CPU accounting in case it got a lower estimate at boot. 199 */ 200 static void 201 tsc_levels_changed(void *arg, int unit) 202 { 203 device_t cf_dev; 204 struct cf_level *levels; 205 int count, error; 206 uint64_t max_freq; 207 208 /* Only use values from the first CPU, assuming all are equal. */ 209 if (unit != 0) 210 return; 211 212 /* Find the appropriate cpufreq device instance. */ 213 cf_dev = devclass_get_device(devclass_find("cpufreq"), unit); 214 if (cf_dev == NULL) { 215 printf("tsc_levels_changed() called but no cpufreq device?\n"); 216 return; 217 } 218 219 /* Get settings from the device and find the max frequency. */ 220 count = 64; 221 levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 222 if (levels == NULL) 223 return; 224 error = CPUFREQ_LEVELS(cf_dev, levels, &count); 225 if (error == 0 && count != 0) { 226 max_freq = (uint64_t)levels[0].total_set.freq * 1000000; 227 set_cputicker(rdtsc, max_freq, 1); 228 } else 229 printf("tsc_levels_changed: no max freq found\n"); 230 free(levels, M_TEMP); 231 } 232 233 /* 234 * If the TSC timecounter is in use, veto the pending change. It may be 235 * possible in the future to handle a dynamically-changing timecounter rate. 236 */ 237 static void 238 tsc_freq_changing(void *arg, const struct cf_level *level, int *status) 239 { 240 241 if (*status != 0 || timecounter != &tsc_timecounter) 242 return; 243 244 printf("timecounter TSC must not be in use when " 245 "changing frequencies; change denied\n"); 246 *status = EBUSY; 247 } 248 249 /* Update TSC freq with the value indicated by the caller. */ 250 static void 251 tsc_freq_changed(void *arg, const struct cf_level *level, int status) 252 { 253 254 /* If there was an error during the transition, don't do anything. */ 255 if (tsc_disabled || status != 0) 256 return; 257 258 /* Total setting for this level gives the new frequency in MHz. */ 259 tsc_freq = (uint64_t)level->total_set.freq * 1000000; 260 tsc_timecounter.tc_frequency = tsc_freq; 261 } 262 263 static int 264 sysctl_machdep_tsc_freq(SYSCTL_HANDLER_ARGS) 265 { 266 int error; 267 uint64_t freq; 268 269 if (tsc_timecounter.tc_frequency == 0) 270 return (EOPNOTSUPP); 271 freq = tsc_freq; 272 error = sysctl_handle_64(oidp, &freq, 0, req); 273 if (error == 0 && req->newptr != NULL) { 274 tsc_freq = freq; 275 tsc_timecounter.tc_frequency = tsc_freq; 276 } 277 return (error); 278 } 279 280 SYSCTL_PROC(_machdep, OID_AUTO, tsc_freq, CTLTYPE_U64 | CTLFLAG_RW, 281 0, 0, sysctl_machdep_tsc_freq, "QU", ""); 282 283 static unsigned 284 tsc_get_timecount(struct timecounter *tc) 285 { 286 return (rdtsc()); 287 } 288