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_broken; 52 int tsc_is_invariant; 53 int tsc_present; 54 static eventhandler_tag tsc_levels_tag, tsc_pre_tag, tsc_post_tag; 55 56 SYSCTL_INT(_kern_timecounter, OID_AUTO, invariant_tsc, CTLFLAG_RDTUN, 57 &tsc_is_invariant, 0, "Indicates whether the TSC is P-state invariant"); 58 TUNABLE_INT("kern.timecounter.invariant_tsc", &tsc_is_invariant); 59 60 #ifdef SMP 61 static int smp_tsc; 62 SYSCTL_INT(_kern_timecounter, OID_AUTO, smp_tsc, CTLFLAG_RDTUN, &smp_tsc, 0, 63 "Indicates whether the TSC is safe to use in SMP mode"); 64 TUNABLE_INT("kern.timecounter.smp_tsc", &smp_tsc); 65 #endif 66 67 static void tsc_freq_changed(void *arg, const struct cf_level *level, 68 int status); 69 static void tsc_freq_changing(void *arg, const struct cf_level *level, 70 int *status); 71 static unsigned tsc_get_timecount(struct timecounter *tc); 72 static void tsc_levels_changed(void *arg, int unit); 73 74 static struct timecounter tsc_timecounter = { 75 tsc_get_timecount, /* get_timecount */ 76 0, /* no poll_pps */ 77 ~0u, /* counter_mask */ 78 0, /* frequency */ 79 "TSC", /* name */ 80 800, /* quality (adjusted in code) */ 81 }; 82 83 void 84 init_TSC(void) 85 { 86 u_int64_t tscval[2]; 87 88 if (cpu_feature & CPUID_TSC) 89 tsc_present = 1; 90 else 91 tsc_present = 0; 92 93 if (!tsc_present) 94 return; 95 96 if (bootverbose) 97 printf("Calibrating TSC clock ... "); 98 99 tscval[0] = rdtsc(); 100 DELAY(1000000); 101 tscval[1] = rdtsc(); 102 103 tsc_freq = tscval[1] - tscval[0]; 104 if (bootverbose) 105 printf("TSC clock: %ju Hz\n", (intmax_t)tsc_freq); 106 107 switch (cpu_vendor_id) { 108 case CPU_VENDOR_AMD: 109 if ((amd_pminfo & AMDPM_TSC_INVARIANT) || 110 CPUID_TO_FAMILY(cpu_id) >= 0x10) 111 tsc_is_invariant = 1; 112 break; 113 case CPU_VENDOR_INTEL: 114 if ((amd_pminfo & AMDPM_TSC_INVARIANT) || 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 (CPUID_TO_FAMILY(cpu_id) == 0x6 && 123 CPUID_TO_MODEL(cpu_id) >= 0xf && 124 (rdmsr(0x1203) & 0x100000000ULL) == 0) 125 tsc_is_invariant = 1; 126 break; 127 } 128 129 /* 130 * Inform CPU accounting about our boot-time clock rate. This will 131 * be updated if someone loads a cpufreq driver after boot that 132 * discovers a new max frequency. 133 */ 134 set_cputicker(rdtsc, tsc_freq, 1); 135 136 if (tsc_is_invariant) 137 return; 138 139 /* Register to find out about changes in CPU frequency. */ 140 tsc_pre_tag = EVENTHANDLER_REGISTER(cpufreq_pre_change, 141 tsc_freq_changing, NULL, EVENTHANDLER_PRI_FIRST); 142 tsc_post_tag = EVENTHANDLER_REGISTER(cpufreq_post_change, 143 tsc_freq_changed, NULL, EVENTHANDLER_PRI_FIRST); 144 tsc_levels_tag = EVENTHANDLER_REGISTER(cpufreq_levels_changed, 145 tsc_levels_changed, NULL, EVENTHANDLER_PRI_ANY); 146 } 147 148 void 149 init_TSC_tc(void) 150 { 151 152 if (!tsc_present) 153 return; 154 155 /* 156 * We can not use the TSC if we support APM. Precise timekeeping 157 * on an APM'ed machine is at best a fools pursuit, since 158 * any and all of the time spent in various SMM code can't 159 * be reliably accounted for. Reading the RTC is your only 160 * source of reliable time info. The i8254 loses too, of course, 161 * but we need to have some kind of time... 162 * We don't know at this point whether APM is going to be used 163 * or not, nor when it might be activated. Play it safe. 164 */ 165 if (power_pm_get_type() == POWER_PM_TYPE_APM) { 166 tsc_timecounter.tc_quality = -1000; 167 if (bootverbose) 168 printf("TSC timecounter disabled: APM enabled.\n"); 169 } 170 171 #ifdef SMP 172 /* 173 * We can not use the TSC in SMP mode unless the TSCs on all CPUs 174 * are somehow synchronized. Some hardware configurations do 175 * this, but we have no way of determining whether this is the 176 * case, so we do not use the TSC in multi-processor systems 177 * unless the user indicated (by setting kern.timecounter.smp_tsc 178 * to 1) that he believes that his TSCs are synchronized. 179 */ 180 if (mp_ncpus > 1 && !smp_tsc) 181 tsc_timecounter.tc_quality = -100; 182 #endif 183 184 if (tsc_freq != 0 && !tsc_is_broken) { 185 tsc_timecounter.tc_frequency = tsc_freq; 186 tc_init(&tsc_timecounter); 187 } 188 } 189 190 /* 191 * When cpufreq levels change, find out about the (new) max frequency. We 192 * use this to update CPU accounting in case it got a lower estimate at boot. 193 */ 194 static void 195 tsc_levels_changed(void *arg, int unit) 196 { 197 device_t cf_dev; 198 struct cf_level *levels; 199 int count, error; 200 uint64_t max_freq; 201 202 /* Only use values from the first CPU, assuming all are equal. */ 203 if (unit != 0) 204 return; 205 206 /* Find the appropriate cpufreq device instance. */ 207 cf_dev = devclass_get_device(devclass_find("cpufreq"), unit); 208 if (cf_dev == NULL) { 209 printf("tsc_levels_changed() called but no cpufreq device?\n"); 210 return; 211 } 212 213 /* Get settings from the device and find the max frequency. */ 214 count = 64; 215 levels = malloc(count * sizeof(*levels), M_TEMP, M_NOWAIT); 216 if (levels == NULL) 217 return; 218 error = CPUFREQ_LEVELS(cf_dev, levels, &count); 219 if (error == 0 && count != 0) { 220 max_freq = (uint64_t)levels[0].total_set.freq * 1000000; 221 set_cputicker(rdtsc, max_freq, 1); 222 } else 223 printf("tsc_levels_changed: no max freq found\n"); 224 free(levels, M_TEMP); 225 } 226 227 /* 228 * If the TSC timecounter is in use, veto the pending change. It may be 229 * possible in the future to handle a dynamically-changing timecounter rate. 230 */ 231 static void 232 tsc_freq_changing(void *arg, const struct cf_level *level, int *status) 233 { 234 235 if (*status != 0 || timecounter != &tsc_timecounter) 236 return; 237 238 printf("timecounter TSC must not be in use when " 239 "changing frequencies; change denied\n"); 240 *status = EBUSY; 241 } 242 243 /* Update TSC freq with the value indicated by the caller. */ 244 static void 245 tsc_freq_changed(void *arg, const struct cf_level *level, int status) 246 { 247 248 /* If there was an error during the transition, don't do anything. */ 249 if (status != 0) 250 return; 251 252 /* Total setting for this level gives the new frequency in MHz. */ 253 tsc_freq = (uint64_t)level->total_set.freq * 1000000; 254 tsc_timecounter.tc_frequency = tsc_freq; 255 } 256 257 static int 258 sysctl_machdep_tsc_freq(SYSCTL_HANDLER_ARGS) 259 { 260 int error; 261 uint64_t freq; 262 263 if (tsc_timecounter.tc_frequency == 0) 264 return (EOPNOTSUPP); 265 freq = tsc_freq; 266 error = sysctl_handle_64(oidp, &freq, 0, req); 267 if (error == 0 && req->newptr != NULL) { 268 tsc_freq = freq; 269 tsc_timecounter.tc_frequency = tsc_freq; 270 } 271 return (error); 272 } 273 274 SYSCTL_PROC(_machdep, OID_AUTO, tsc_freq, CTLTYPE_U64 | CTLFLAG_RW, 275 0, 0, sysctl_machdep_tsc_freq, "QU", ""); 276 277 static unsigned 278 tsc_get_timecount(struct timecounter *tc) 279 { 280 return (rdtsc()); 281 } 282