1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2007, 2008 Rui Paulo <rpaulo@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 18 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 19 * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, 20 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 21 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 22 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 24 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 25 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 26 * POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 /* 30 * Device driver for Intel's On Die thermal sensor via MSR. 31 * First introduced in Intel's Core line of processors. 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include <sys/param.h> 38 #include <sys/bus.h> 39 #include <sys/conf.h> 40 #include <sys/kernel.h> 41 #include <sys/lock.h> 42 #include <sys/module.h> 43 #include <sys/mutex.h> 44 #include <sys/proc.h> /* for curthread */ 45 #include <sys/smp.h> 46 #include <sys/sysctl.h> 47 #include <sys/systm.h> 48 49 #include <machine/specialreg.h> 50 #include <machine/cpufunc.h> 51 #include <machine/cputypes.h> 52 #include <machine/md_var.h> 53 54 #define TZ_ZEROC 2731 55 56 #define THERM_STATUS_LOG 0x02 57 #define THERM_STATUS 0x01 58 #define THERM_STATUS_TEMP_SHIFT 16 59 #define THERM_STATUS_TEMP_MASK 0x7f 60 #define THERM_STATUS_RES_SHIFT 27 61 #define THERM_STATUS_RES_MASK 0x0f 62 #define THERM_STATUS_VALID_SHIFT 31 63 #define THERM_STATUS_VALID_MASK 0x01 64 65 struct coretemp_softc { 66 device_t sc_dev; 67 int sc_tjmax; 68 unsigned int sc_throttle_log; 69 }; 70 71 /* 72 * Device methods. 73 */ 74 static void coretemp_identify(driver_t *driver, device_t parent); 75 static int coretemp_probe(device_t dev); 76 static int coretemp_attach(device_t dev); 77 static int coretemp_detach(device_t dev); 78 79 static uint64_t coretemp_get_thermal_msr(int cpu); 80 static void coretemp_clear_thermal_msr(int cpu); 81 static int coretemp_get_val_sysctl(SYSCTL_HANDLER_ARGS); 82 static int coretemp_throttle_log_sysctl(SYSCTL_HANDLER_ARGS); 83 84 static device_method_t coretemp_methods[] = { 85 /* Device interface */ 86 DEVMETHOD(device_identify, coretemp_identify), 87 DEVMETHOD(device_probe, coretemp_probe), 88 DEVMETHOD(device_attach, coretemp_attach), 89 DEVMETHOD(device_detach, coretemp_detach), 90 91 DEVMETHOD_END 92 }; 93 94 static driver_t coretemp_driver = { 95 "coretemp", 96 coretemp_methods, 97 sizeof(struct coretemp_softc), 98 }; 99 100 enum therm_info { 101 CORETEMP_TEMP, 102 CORETEMP_DELTA, 103 CORETEMP_RESOLUTION, 104 CORETEMP_TJMAX, 105 }; 106 107 static devclass_t coretemp_devclass; 108 DRIVER_MODULE(coretemp, cpu, coretemp_driver, coretemp_devclass, NULL, 109 NULL); 110 111 static void 112 coretemp_identify(driver_t *driver, device_t parent) 113 { 114 device_t child; 115 u_int regs[4]; 116 117 /* Make sure we're not being doubly invoked. */ 118 if (device_find_child(parent, "coretemp", -1) != NULL) 119 return; 120 121 /* Check that CPUID 0x06 is supported and the vendor is Intel.*/ 122 if (cpu_high < 6 || cpu_vendor_id != CPU_VENDOR_INTEL) 123 return; 124 /* 125 * CPUID 0x06 returns 1 if the processor has on-die thermal 126 * sensors. EBX[0:3] contains the number of sensors. 127 */ 128 do_cpuid(0x06, regs); 129 if ((regs[0] & 0x1) != 1) 130 return; 131 132 /* 133 * We add a child for each CPU since settings must be performed 134 * on each CPU in the SMP case. 135 */ 136 child = device_add_child(parent, "coretemp", -1); 137 if (child == NULL) 138 device_printf(parent, "add coretemp child failed\n"); 139 } 140 141 static int 142 coretemp_probe(device_t dev) 143 { 144 if (resource_disabled("coretemp", 0)) 145 return (ENXIO); 146 147 device_set_desc(dev, "CPU On-Die Thermal Sensors"); 148 149 if (!bootverbose && device_get_unit(dev) != 0) 150 device_quiet(dev); 151 152 return (BUS_PROBE_GENERIC); 153 } 154 155 static int 156 coretemp_attach(device_t dev) 157 { 158 struct coretemp_softc *sc = device_get_softc(dev); 159 device_t pdev; 160 uint64_t msr; 161 int cpu_model, cpu_stepping; 162 int ret, tjtarget; 163 struct sysctl_oid *oid; 164 struct sysctl_ctx_list *ctx; 165 166 sc->sc_dev = dev; 167 pdev = device_get_parent(dev); 168 cpu_model = CPUID_TO_MODEL(cpu_id); 169 cpu_stepping = CPUID_TO_STEPPING(cpu_id); 170 171 /* 172 * Some CPUs, namely the PIII, don't have thermal sensors, but 173 * report them when the CPUID check is performed in 174 * coretemp_identify(). This leads to a later GPF when the sensor 175 * is queried via a MSR, so we stop here. 176 */ 177 if (cpu_model < 0xe) 178 return (ENXIO); 179 180 #if 0 /* 181 * XXXrpaulo: I have this CPU model and when it returns from C3 182 * coretemp continues to function properly. 183 */ 184 185 /* 186 * Check for errata AE18. 187 * "Processor Digital Thermal Sensor (DTS) Readout stops 188 * updating upon returning from C3/C4 state." 189 * 190 * Adapted from the Linux coretemp driver. 191 */ 192 if (cpu_model == 0xe && cpu_stepping < 0xc) { 193 msr = rdmsr(MSR_BIOS_SIGN); 194 msr = msr >> 32; 195 if (msr < 0x39) { 196 device_printf(dev, "not supported (Intel errata " 197 "AE18), try updating your BIOS\n"); 198 return (ENXIO); 199 } 200 } 201 #endif 202 203 /* 204 * Use 100C as the initial value. 205 */ 206 sc->sc_tjmax = 100; 207 208 if ((cpu_model == 0xf && cpu_stepping >= 2) || cpu_model == 0xe) { 209 /* 210 * On some Core 2 CPUs, there's an undocumented MSR that 211 * can tell us if Tj(max) is 100 or 85. 212 * 213 * The if-clause for CPUs having the MSR_IA32_EXT_CONFIG was adapted 214 * from the Linux coretemp driver. 215 */ 216 msr = rdmsr(MSR_IA32_EXT_CONFIG); 217 if (msr & (1 << 30)) 218 sc->sc_tjmax = 85; 219 } else if (cpu_model == 0x17) { 220 switch (cpu_stepping) { 221 case 0x6: /* Mobile Core 2 Duo */ 222 sc->sc_tjmax = 105; 223 break; 224 default: /* Unknown stepping */ 225 break; 226 } 227 } else if (cpu_model == 0x1c) { 228 switch (cpu_stepping) { 229 case 0xa: /* 45nm Atom D400, N400 and D500 series */ 230 sc->sc_tjmax = 100; 231 break; 232 default: 233 sc->sc_tjmax = 90; 234 break; 235 } 236 } else { 237 /* 238 * Attempt to get Tj(max) from MSR IA32_TEMPERATURE_TARGET. 239 * 240 * This method is described in Intel white paper "CPU 241 * Monitoring With DTS/PECI". (#322683) 242 */ 243 ret = rdmsr_safe(MSR_IA32_TEMPERATURE_TARGET, &msr); 244 if (ret == 0) { 245 tjtarget = (msr >> 16) & 0xff; 246 247 /* 248 * On earlier generation of processors, the value 249 * obtained from IA32_TEMPERATURE_TARGET register is 250 * an offset that needs to be summed with a model 251 * specific base. It is however not clear what 252 * these numbers are, with the publicly available 253 * documents from Intel. 254 * 255 * For now, we consider [70, 110]C range, as 256 * described in #322683, as "reasonable" and accept 257 * these values whenever the MSR is available for 258 * read, regardless the CPU model. 259 */ 260 if (tjtarget >= 70 && tjtarget <= 110) 261 sc->sc_tjmax = tjtarget; 262 else 263 device_printf(dev, "Tj(target) value %d " 264 "does not seem right.\n", tjtarget); 265 } else 266 device_printf(dev, "Can not get Tj(target) " 267 "from your CPU, using 100C.\n"); 268 } 269 270 if (bootverbose) 271 device_printf(dev, "Setting TjMax=%d\n", sc->sc_tjmax); 272 273 ctx = device_get_sysctl_ctx(dev); 274 275 oid = SYSCTL_ADD_NODE(ctx, 276 SYSCTL_CHILDREN(device_get_sysctl_tree(pdev)), OID_AUTO, 277 "coretemp", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 278 "Per-CPU thermal information"); 279 280 /* 281 * Add the MIBs to dev.cpu.N and dev.cpu.N.coretemp. 282 */ 283 SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(pdev)), 284 OID_AUTO, "temperature", CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, 285 dev, CORETEMP_TEMP, coretemp_get_val_sysctl, "IK", 286 "Current temperature"); 287 SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO, "delta", 288 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, dev, CORETEMP_DELTA, 289 coretemp_get_val_sysctl, "I", 290 "Delta between TCC activation and current temperature"); 291 SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO, "resolution", 292 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, dev, CORETEMP_RESOLUTION, 293 coretemp_get_val_sysctl, "I", 294 "Resolution of CPU thermal sensor"); 295 SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO, "tjmax", 296 CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, dev, CORETEMP_TJMAX, 297 coretemp_get_val_sysctl, "IK", 298 "TCC activation temperature"); 299 SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO, 300 "throttle_log", CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, dev, 0, 301 coretemp_throttle_log_sysctl, "I", 302 "Set to 1 if the thermal sensor has tripped"); 303 304 return (0); 305 } 306 307 static int 308 coretemp_detach(device_t dev) 309 { 310 return (0); 311 } 312 313 struct coretemp_args { 314 u_int msr; 315 uint64_t val; 316 }; 317 318 static void 319 coretemp_rdmsr(void *arg) 320 { 321 struct coretemp_args *args = arg; 322 323 args->val = rdmsr(args->msr); 324 } 325 326 static void 327 coretemp_wrmsr(void *arg) 328 { 329 struct coretemp_args *args = arg; 330 331 wrmsr(args->msr, args->val); 332 } 333 334 static uint64_t 335 coretemp_get_thermal_msr(int cpu) 336 { 337 struct coretemp_args args; 338 cpuset_t cpus; 339 340 /* 341 * The digital temperature reading is located at bit 16 342 * of MSR_THERM_STATUS. 343 * 344 * There is a bit on that MSR that indicates whether the 345 * temperature is valid or not. 346 * 347 * The temperature is computed by subtracting the temperature 348 * reading by Tj(max). 349 */ 350 args.msr = MSR_THERM_STATUS; 351 CPU_SETOF(cpu, &cpus); 352 smp_rendezvous_cpus(cpus, smp_no_rendezvous_barrier, coretemp_rdmsr, 353 smp_no_rendezvous_barrier, &args); 354 return (args.val); 355 } 356 357 static void 358 coretemp_clear_thermal_msr(int cpu) 359 { 360 struct coretemp_args args; 361 cpuset_t cpus; 362 363 args.msr = MSR_THERM_STATUS; 364 args.val = 0; 365 CPU_SETOF(cpu, &cpus); 366 smp_rendezvous_cpus(cpus, smp_no_rendezvous_barrier, coretemp_wrmsr, 367 smp_no_rendezvous_barrier, &args); 368 } 369 370 static int 371 coretemp_get_val_sysctl(SYSCTL_HANDLER_ARGS) 372 { 373 device_t dev; 374 uint64_t msr; 375 int val, tmp; 376 struct coretemp_softc *sc; 377 enum therm_info type; 378 char stemp[16]; 379 380 dev = (device_t) arg1; 381 msr = coretemp_get_thermal_msr(device_get_unit(dev)); 382 sc = device_get_softc(dev); 383 type = arg2; 384 385 if (((msr >> THERM_STATUS_VALID_SHIFT) & THERM_STATUS_VALID_MASK) != 1) { 386 val = -1; 387 } else { 388 switch (type) { 389 case CORETEMP_TEMP: 390 tmp = (msr >> THERM_STATUS_TEMP_SHIFT) & 391 THERM_STATUS_TEMP_MASK; 392 val = (sc->sc_tjmax - tmp) * 10 + TZ_ZEROC; 393 break; 394 case CORETEMP_DELTA: 395 val = (msr >> THERM_STATUS_TEMP_SHIFT) & 396 THERM_STATUS_TEMP_MASK; 397 break; 398 case CORETEMP_RESOLUTION: 399 val = (msr >> THERM_STATUS_RES_SHIFT) & 400 THERM_STATUS_RES_MASK; 401 break; 402 case CORETEMP_TJMAX: 403 val = sc->sc_tjmax * 10 + TZ_ZEROC; 404 break; 405 } 406 } 407 408 if (msr & THERM_STATUS_LOG) { 409 coretemp_clear_thermal_msr(device_get_unit(dev)); 410 sc->sc_throttle_log = 1; 411 412 /* 413 * Check for Critical Temperature Status and Critical 414 * Temperature Log. It doesn't really matter if the 415 * current temperature is invalid because the "Critical 416 * Temperature Log" bit will tell us if the Critical 417 * Temperature has * been reached in past. It's not 418 * directly related to the current temperature. 419 * 420 * If we reach a critical level, allow devctl(4) 421 * to catch this and shutdown the system. 422 */ 423 if (msr & THERM_STATUS) { 424 tmp = (msr >> THERM_STATUS_TEMP_SHIFT) & 425 THERM_STATUS_TEMP_MASK; 426 tmp = (sc->sc_tjmax - tmp) * 10 + TZ_ZEROC; 427 device_printf(dev, "critical temperature detected, " 428 "suggest system shutdown\n"); 429 snprintf(stemp, sizeof(stemp), "%d", tmp); 430 devctl_notify("coretemp", "Thermal", stemp, 431 "notify=0xcc"); 432 } 433 } 434 435 return (sysctl_handle_int(oidp, &val, 0, req)); 436 } 437 438 static int 439 coretemp_throttle_log_sysctl(SYSCTL_HANDLER_ARGS) 440 { 441 device_t dev; 442 uint64_t msr; 443 int error, val; 444 struct coretemp_softc *sc; 445 446 dev = (device_t) arg1; 447 msr = coretemp_get_thermal_msr(device_get_unit(dev)); 448 sc = device_get_softc(dev); 449 450 if (msr & THERM_STATUS_LOG) { 451 coretemp_clear_thermal_msr(device_get_unit(dev)); 452 sc->sc_throttle_log = 1; 453 } 454 455 val = sc->sc_throttle_log; 456 457 error = sysctl_handle_int(oidp, &val, 0, req); 458 459 if (error || !req->newptr) 460 return (error); 461 else if (val != 0) 462 return (EINVAL); 463 464 coretemp_clear_thermal_msr(device_get_unit(dev)); 465 sc->sc_throttle_log = 0; 466 467 return (0); 468 } 469