1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * k10temp.c - AMD Family 10h/11h/12h/14h/15h/16h/17h 4 * processor hardware monitoring 5 * 6 * Copyright (c) 2009 Clemens Ladisch <clemens@ladisch.de> 7 * Copyright (c) 2020 Guenter Roeck <linux@roeck-us.net> 8 * 9 * Implementation notes: 10 * - CCD register address information as well as the calculation to 11 * convert raw register values is from https://github.com/ocerman/zenpower. 12 * The information is not confirmed from chip datasheets, but experiments 13 * suggest that it provides reasonable temperature values. 14 */ 15 16 #include <linux/bitops.h> 17 #include <linux/err.h> 18 #include <linux/hwmon.h> 19 #include <linux/init.h> 20 #include <linux/module.h> 21 #include <linux/pci.h> 22 #include <linux/pci_ids.h> 23 #include <asm/amd_nb.h> 24 #include <asm/processor.h> 25 26 MODULE_DESCRIPTION("AMD Family 10h+ CPU core temperature monitor"); 27 MODULE_AUTHOR("Clemens Ladisch <clemens@ladisch.de>"); 28 MODULE_LICENSE("GPL"); 29 30 static bool force; 31 module_param(force, bool, 0444); 32 MODULE_PARM_DESC(force, "force loading on processors with erratum 319"); 33 34 /* Provide lock for writing to NB_SMU_IND_ADDR */ 35 static DEFINE_MUTEX(nb_smu_ind_mutex); 36 37 #ifndef PCI_DEVICE_ID_AMD_15H_M70H_NB_F3 38 #define PCI_DEVICE_ID_AMD_15H_M70H_NB_F3 0x15b3 39 #endif 40 41 /* CPUID function 0x80000001, ebx */ 42 #define CPUID_PKGTYPE_MASK GENMASK(31, 28) 43 #define CPUID_PKGTYPE_F 0x00000000 44 #define CPUID_PKGTYPE_AM2R2_AM3 0x10000000 45 46 /* DRAM controller (PCI function 2) */ 47 #define REG_DCT0_CONFIG_HIGH 0x094 48 #define DDR3_MODE BIT(8) 49 50 /* miscellaneous (PCI function 3) */ 51 #define REG_HARDWARE_THERMAL_CONTROL 0x64 52 #define HTC_ENABLE BIT(0) 53 54 #define REG_REPORTED_TEMPERATURE 0xa4 55 56 #define REG_NORTHBRIDGE_CAPABILITIES 0xe8 57 #define NB_CAP_HTC BIT(10) 58 59 /* 60 * For F15h M60h and M70h, REG_HARDWARE_THERMAL_CONTROL 61 * and REG_REPORTED_TEMPERATURE have been moved to 62 * D0F0xBC_xD820_0C64 [Hardware Temperature Control] 63 * D0F0xBC_xD820_0CA4 [Reported Temperature Control] 64 */ 65 #define F15H_M60H_HARDWARE_TEMP_CTRL_OFFSET 0xd8200c64 66 #define F15H_M60H_REPORTED_TEMP_CTRL_OFFSET 0xd8200ca4 67 68 /* Common for Zen CPU families (Family 17h and 18h and 19h and 1Ah) */ 69 #define ZEN_REPORTED_TEMP_CTRL_BASE 0x00059800 70 71 #define ZEN_CCD_TEMP(offset, x) (ZEN_REPORTED_TEMP_CTRL_BASE + \ 72 (offset) + ((x) * 4)) 73 #define ZEN_CCD_TEMP_VALID BIT(11) 74 #define ZEN_CCD_TEMP_MASK GENMASK(10, 0) 75 76 #define ZEN_CUR_TEMP_SHIFT 21 77 #define ZEN_CUR_TEMP_RANGE_SEL_MASK BIT(19) 78 #define ZEN_CUR_TEMP_TJ_SEL_MASK GENMASK(17, 16) 79 80 /* 81 * AMD's Industrial processor 3255 supports temperature from -40 deg to 105 deg Celsius. 82 * Use the model name to identify 3255 CPUs and set a flag to display negative temperature. 83 * Do not round off to zero for negative Tctl or Tdie values if the flag is set 84 */ 85 #define AMD_I3255_STR "3255" 86 87 struct k10temp_data { 88 struct pci_dev *pdev; 89 void (*read_htcreg)(struct pci_dev *pdev, u32 *regval); 90 void (*read_tempreg)(struct pci_dev *pdev, u32 *regval); 91 int temp_offset; 92 u32 temp_adjust_mask; 93 u32 show_temp; 94 bool is_zen; 95 u32 ccd_offset; 96 bool disp_negative; 97 }; 98 99 #define TCTL_BIT 0 100 #define TDIE_BIT 1 101 #define TCCD_BIT(x) ((x) + 2) 102 103 #define HAVE_TEMP(d, channel) ((d)->show_temp & BIT(channel)) 104 #define HAVE_TDIE(d) HAVE_TEMP(d, TDIE_BIT) 105 106 struct tctl_offset { 107 u8 model; 108 char const *id; 109 int offset; 110 }; 111 112 static const struct tctl_offset tctl_offset_table[] = { 113 { 0x17, "AMD Ryzen 5 1600X", 20000 }, 114 { 0x17, "AMD Ryzen 7 1700X", 20000 }, 115 { 0x17, "AMD Ryzen 7 1800X", 20000 }, 116 { 0x17, "AMD Ryzen 7 2700X", 10000 }, 117 { 0x17, "AMD Ryzen Threadripper 19", 27000 }, /* 19{00,20,50}X */ 118 { 0x17, "AMD Ryzen Threadripper 29", 27000 }, /* 29{20,50,70,90}[W]X */ 119 }; 120 121 static void read_htcreg_pci(struct pci_dev *pdev, u32 *regval) 122 { 123 pci_read_config_dword(pdev, REG_HARDWARE_THERMAL_CONTROL, regval); 124 } 125 126 static void read_tempreg_pci(struct pci_dev *pdev, u32 *regval) 127 { 128 pci_read_config_dword(pdev, REG_REPORTED_TEMPERATURE, regval); 129 } 130 131 static void amd_nb_index_read(struct pci_dev *pdev, unsigned int devfn, 132 unsigned int base, int offset, u32 *val) 133 { 134 mutex_lock(&nb_smu_ind_mutex); 135 pci_bus_write_config_dword(pdev->bus, devfn, 136 base, offset); 137 pci_bus_read_config_dword(pdev->bus, devfn, 138 base + 4, val); 139 mutex_unlock(&nb_smu_ind_mutex); 140 } 141 142 static void read_htcreg_nb_f15(struct pci_dev *pdev, u32 *regval) 143 { 144 amd_nb_index_read(pdev, PCI_DEVFN(0, 0), 0xb8, 145 F15H_M60H_HARDWARE_TEMP_CTRL_OFFSET, regval); 146 } 147 148 static void read_tempreg_nb_f15(struct pci_dev *pdev, u32 *regval) 149 { 150 amd_nb_index_read(pdev, PCI_DEVFN(0, 0), 0xb8, 151 F15H_M60H_REPORTED_TEMP_CTRL_OFFSET, regval); 152 } 153 154 static void read_tempreg_nb_zen(struct pci_dev *pdev, u32 *regval) 155 { 156 if (amd_smn_read(amd_pci_dev_to_node_id(pdev), 157 ZEN_REPORTED_TEMP_CTRL_BASE, regval)) 158 *regval = 0; 159 } 160 161 static long get_raw_temp(struct k10temp_data *data) 162 { 163 u32 regval; 164 long temp; 165 166 data->read_tempreg(data->pdev, ®val); 167 temp = (regval >> ZEN_CUR_TEMP_SHIFT) * 125; 168 if ((regval & data->temp_adjust_mask) || 169 (regval & ZEN_CUR_TEMP_TJ_SEL_MASK) == ZEN_CUR_TEMP_TJ_SEL_MASK) 170 temp -= 49000; 171 return temp; 172 } 173 174 static const char *k10temp_temp_label[] = { 175 "Tctl", 176 "Tdie", 177 "Tccd1", 178 "Tccd2", 179 "Tccd3", 180 "Tccd4", 181 "Tccd5", 182 "Tccd6", 183 "Tccd7", 184 "Tccd8", 185 "Tccd9", 186 "Tccd10", 187 "Tccd11", 188 "Tccd12", 189 }; 190 191 static int k10temp_read_labels(struct device *dev, 192 enum hwmon_sensor_types type, 193 u32 attr, int channel, const char **str) 194 { 195 switch (type) { 196 case hwmon_temp: 197 *str = k10temp_temp_label[channel]; 198 break; 199 default: 200 return -EOPNOTSUPP; 201 } 202 return 0; 203 } 204 205 static int k10temp_read_temp(struct device *dev, u32 attr, int channel, 206 long *val) 207 { 208 struct k10temp_data *data = dev_get_drvdata(dev); 209 int ret = -EOPNOTSUPP; 210 u32 regval; 211 212 switch (attr) { 213 case hwmon_temp_input: 214 switch (channel) { 215 case 0: /* Tctl */ 216 *val = get_raw_temp(data); 217 if (*val < 0 && !data->disp_negative) 218 *val = 0; 219 break; 220 case 1: /* Tdie */ 221 *val = get_raw_temp(data) - data->temp_offset; 222 if (*val < 0 && !data->disp_negative) 223 *val = 0; 224 break; 225 case 2 ... 13: /* Tccd{1-12} */ 226 ret = amd_smn_read(amd_pci_dev_to_node_id(data->pdev), 227 ZEN_CCD_TEMP(data->ccd_offset, channel - 2), 228 ®val); 229 230 if (ret) 231 return ret; 232 233 *val = (regval & ZEN_CCD_TEMP_MASK) * 125 - 49000; 234 break; 235 default: 236 return ret; 237 } 238 break; 239 case hwmon_temp_max: 240 *val = 70 * 1000; 241 break; 242 case hwmon_temp_crit: 243 data->read_htcreg(data->pdev, ®val); 244 *val = ((regval >> 16) & 0x7f) * 500 + 52000; 245 break; 246 case hwmon_temp_crit_hyst: 247 data->read_htcreg(data->pdev, ®val); 248 *val = (((regval >> 16) & 0x7f) 249 - ((regval >> 24) & 0xf)) * 500 + 52000; 250 break; 251 default: 252 return ret; 253 } 254 return 0; 255 } 256 257 static int k10temp_read(struct device *dev, enum hwmon_sensor_types type, 258 u32 attr, int channel, long *val) 259 { 260 switch (type) { 261 case hwmon_temp: 262 return k10temp_read_temp(dev, attr, channel, val); 263 default: 264 return -EOPNOTSUPP; 265 } 266 } 267 268 static umode_t k10temp_is_visible(const void *_data, 269 enum hwmon_sensor_types type, 270 u32 attr, int channel) 271 { 272 const struct k10temp_data *data = _data; 273 struct pci_dev *pdev = data->pdev; 274 u32 reg; 275 276 switch (type) { 277 case hwmon_temp: 278 switch (attr) { 279 case hwmon_temp_input: 280 if (!HAVE_TEMP(data, channel)) 281 return 0; 282 break; 283 case hwmon_temp_max: 284 if (channel || data->is_zen) 285 return 0; 286 break; 287 case hwmon_temp_crit: 288 case hwmon_temp_crit_hyst: 289 if (channel || !data->read_htcreg) 290 return 0; 291 292 pci_read_config_dword(pdev, 293 REG_NORTHBRIDGE_CAPABILITIES, 294 ®); 295 if (!(reg & NB_CAP_HTC)) 296 return 0; 297 298 data->read_htcreg(data->pdev, ®); 299 if (!(reg & HTC_ENABLE)) 300 return 0; 301 break; 302 case hwmon_temp_label: 303 /* Show temperature labels only on Zen CPUs */ 304 if (!data->is_zen || !HAVE_TEMP(data, channel)) 305 return 0; 306 break; 307 default: 308 return 0; 309 } 310 break; 311 default: 312 return 0; 313 } 314 return 0444; 315 } 316 317 static bool has_erratum_319(struct pci_dev *pdev) 318 { 319 u32 pkg_type, reg_dram_cfg; 320 321 if (boot_cpu_data.x86 != 0x10) 322 return false; 323 324 /* 325 * Erratum 319: The thermal sensor of Socket F/AM2+ processors 326 * may be unreliable. 327 */ 328 pkg_type = cpuid_ebx(0x80000001) & CPUID_PKGTYPE_MASK; 329 if (pkg_type == CPUID_PKGTYPE_F) 330 return true; 331 if (pkg_type != CPUID_PKGTYPE_AM2R2_AM3) 332 return false; 333 334 /* DDR3 memory implies socket AM3, which is good */ 335 pci_bus_read_config_dword(pdev->bus, 336 PCI_DEVFN(PCI_SLOT(pdev->devfn), 2), 337 REG_DCT0_CONFIG_HIGH, ®_dram_cfg); 338 if (reg_dram_cfg & DDR3_MODE) 339 return false; 340 341 /* 342 * Unfortunately it is possible to run a socket AM3 CPU with DDR2 343 * memory. We blacklist all the cores which do exist in socket AM2+ 344 * format. It still isn't perfect, as RB-C2 cores exist in both AM2+ 345 * and AM3 formats, but that's the best we can do. 346 */ 347 return boot_cpu_data.x86_model < 4 || 348 (boot_cpu_data.x86_model == 4 && boot_cpu_data.x86_stepping <= 2); 349 } 350 351 static const struct hwmon_channel_info * const k10temp_info[] = { 352 HWMON_CHANNEL_INFO(temp, 353 HWMON_T_INPUT | HWMON_T_MAX | 354 HWMON_T_CRIT | HWMON_T_CRIT_HYST | 355 HWMON_T_LABEL, 356 HWMON_T_INPUT | HWMON_T_LABEL, 357 HWMON_T_INPUT | HWMON_T_LABEL, 358 HWMON_T_INPUT | HWMON_T_LABEL, 359 HWMON_T_INPUT | HWMON_T_LABEL, 360 HWMON_T_INPUT | HWMON_T_LABEL, 361 HWMON_T_INPUT | HWMON_T_LABEL, 362 HWMON_T_INPUT | HWMON_T_LABEL, 363 HWMON_T_INPUT | HWMON_T_LABEL, 364 HWMON_T_INPUT | HWMON_T_LABEL, 365 HWMON_T_INPUT | HWMON_T_LABEL, 366 HWMON_T_INPUT | HWMON_T_LABEL, 367 HWMON_T_INPUT | HWMON_T_LABEL, 368 HWMON_T_INPUT | HWMON_T_LABEL), 369 NULL 370 }; 371 372 static const struct hwmon_ops k10temp_hwmon_ops = { 373 .is_visible = k10temp_is_visible, 374 .read = k10temp_read, 375 .read_string = k10temp_read_labels, 376 }; 377 378 static const struct hwmon_chip_info k10temp_chip_info = { 379 .ops = &k10temp_hwmon_ops, 380 .info = k10temp_info, 381 }; 382 383 static void k10temp_get_ccd_support(struct pci_dev *pdev, 384 struct k10temp_data *data, int limit) 385 { 386 u32 regval; 387 int i; 388 389 for (i = 0; i < limit; i++) { 390 /* 391 * Ignore inaccessible CCDs. 392 * 393 * Some systems will return a register value of 0, and the TEMP_VALID 394 * bit check below will naturally fail. 395 * 396 * Other systems will return a PCI_ERROR_RESPONSE (0xFFFFFFFF) for 397 * the register value. And this will incorrectly pass the TEMP_VALID 398 * bit check. 399 */ 400 if (amd_smn_read(amd_pci_dev_to_node_id(pdev), 401 ZEN_CCD_TEMP(data->ccd_offset, i), ®val)) 402 continue; 403 404 if (regval & ZEN_CCD_TEMP_VALID) 405 data->show_temp |= BIT(TCCD_BIT(i)); 406 } 407 } 408 409 static int k10temp_probe(struct pci_dev *pdev, const struct pci_device_id *id) 410 { 411 int unreliable = has_erratum_319(pdev); 412 struct device *dev = &pdev->dev; 413 struct k10temp_data *data; 414 struct device *hwmon_dev; 415 int i; 416 417 if (unreliable) { 418 if (!force) { 419 dev_err(dev, 420 "unreliable CPU thermal sensor; monitoring disabled\n"); 421 return -ENODEV; 422 } 423 dev_warn(dev, 424 "unreliable CPU thermal sensor; check erratum 319\n"); 425 } 426 427 data = devm_kzalloc(dev, sizeof(*data), GFP_KERNEL); 428 if (!data) 429 return -ENOMEM; 430 431 data->pdev = pdev; 432 data->show_temp |= BIT(TCTL_BIT); /* Always show Tctl */ 433 434 if (boot_cpu_data.x86 == 0x17 && 435 strstr(boot_cpu_data.x86_model_id, AMD_I3255_STR)) { 436 data->disp_negative = true; 437 } 438 439 if (boot_cpu_data.x86 == 0x15 && 440 ((boot_cpu_data.x86_model & 0xf0) == 0x60 || 441 (boot_cpu_data.x86_model & 0xf0) == 0x70)) { 442 data->read_htcreg = read_htcreg_nb_f15; 443 data->read_tempreg = read_tempreg_nb_f15; 444 } else if (boot_cpu_data.x86 == 0x17 || boot_cpu_data.x86 == 0x18) { 445 data->temp_adjust_mask = ZEN_CUR_TEMP_RANGE_SEL_MASK; 446 data->read_tempreg = read_tempreg_nb_zen; 447 data->is_zen = true; 448 449 switch (boot_cpu_data.x86_model) { 450 case 0x1: /* Zen */ 451 case 0x8: /* Zen+ */ 452 case 0x11: /* Zen APU */ 453 case 0x18: /* Zen+ APU */ 454 data->ccd_offset = 0x154; 455 k10temp_get_ccd_support(pdev, data, 4); 456 break; 457 case 0x31: /* Zen2 Threadripper */ 458 case 0x60: /* Renoir */ 459 case 0x68: /* Lucienne */ 460 case 0x71: /* Zen2 */ 461 data->ccd_offset = 0x154; 462 k10temp_get_ccd_support(pdev, data, 8); 463 break; 464 case 0xa0 ... 0xaf: 465 data->ccd_offset = 0x300; 466 k10temp_get_ccd_support(pdev, data, 8); 467 break; 468 } 469 } else if (boot_cpu_data.x86 == 0x19) { 470 data->temp_adjust_mask = ZEN_CUR_TEMP_RANGE_SEL_MASK; 471 data->read_tempreg = read_tempreg_nb_zen; 472 data->is_zen = true; 473 474 switch (boot_cpu_data.x86_model) { 475 case 0x0 ... 0x1: /* Zen3 SP3/TR */ 476 case 0x21: /* Zen3 Ryzen Desktop */ 477 case 0x50 ... 0x5f: /* Green Sardine */ 478 data->ccd_offset = 0x154; 479 k10temp_get_ccd_support(pdev, data, 8); 480 break; 481 case 0x40 ... 0x4f: /* Yellow Carp */ 482 data->ccd_offset = 0x300; 483 k10temp_get_ccd_support(pdev, data, 8); 484 break; 485 case 0x60 ... 0x6f: 486 case 0x70 ... 0x7f: 487 data->ccd_offset = 0x308; 488 k10temp_get_ccd_support(pdev, data, 8); 489 break; 490 case 0x10 ... 0x1f: 491 case 0xa0 ... 0xaf: 492 data->ccd_offset = 0x300; 493 k10temp_get_ccd_support(pdev, data, 12); 494 break; 495 } 496 } else if (boot_cpu_data.x86 == 0x1a) { 497 data->temp_adjust_mask = ZEN_CUR_TEMP_RANGE_SEL_MASK; 498 data->read_tempreg = read_tempreg_nb_zen; 499 data->is_zen = true; 500 } else { 501 data->read_htcreg = read_htcreg_pci; 502 data->read_tempreg = read_tempreg_pci; 503 } 504 505 for (i = 0; i < ARRAY_SIZE(tctl_offset_table); i++) { 506 const struct tctl_offset *entry = &tctl_offset_table[i]; 507 508 if (boot_cpu_data.x86 == entry->model && 509 strstr(boot_cpu_data.x86_model_id, entry->id)) { 510 data->show_temp |= BIT(TDIE_BIT); /* show Tdie */ 511 data->temp_offset = entry->offset; 512 break; 513 } 514 } 515 516 hwmon_dev = devm_hwmon_device_register_with_info(dev, "k10temp", data, 517 &k10temp_chip_info, 518 NULL); 519 return PTR_ERR_OR_ZERO(hwmon_dev); 520 } 521 522 static const struct pci_device_id k10temp_id_table[] = { 523 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_10H_NB_MISC) }, 524 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_11H_NB_MISC) }, 525 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_CNB17H_F3) }, 526 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_NB_F3) }, 527 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M10H_F3) }, 528 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M30H_NB_F3) }, 529 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M60H_NB_F3) }, 530 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M70H_NB_F3) }, 531 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_16H_NB_F3) }, 532 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_16H_M30H_NB_F3) }, 533 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_DF_F3) }, 534 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M10H_DF_F3) }, 535 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M30H_DF_F3) }, 536 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M60H_DF_F3) }, 537 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M70H_DF_F3) }, 538 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_MA0H_DF_F3) }, 539 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_DF_F3) }, 540 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M10H_DF_F3) }, 541 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M40H_DF_F3) }, 542 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M50H_DF_F3) }, 543 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M60H_DF_F3) }, 544 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M70H_DF_F3) }, 545 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M78H_DF_F3) }, 546 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_1AH_M00H_DF_F3) }, 547 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_1AH_M20H_DF_F3) }, 548 { PCI_VDEVICE(HYGON, PCI_DEVICE_ID_AMD_17H_DF_F3) }, 549 {} 550 }; 551 MODULE_DEVICE_TABLE(pci, k10temp_id_table); 552 553 static struct pci_driver k10temp_driver = { 554 .name = "k10temp", 555 .id_table = k10temp_id_table, 556 .probe = k10temp_probe, 557 }; 558 559 module_pci_driver(k10temp_driver); 560