drivers/hwmon/coretemp.c

   1 /*
   2  * coretemp.c - Linux kernel module for hardware monitoring
   3  *
   4  * Copyright (C) 2007 Rudolf Marek <r.marek@assembler.cz>
   5  *
   6  * Inspired from many hwmon drivers
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; version 2 of the License.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  20  * 02110-1301 USA.
  21  */
  22
  23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  24
  25 #include <linux/module.h>
  26 #include <linux/init.h>
  27 #include <linux/slab.h>
  28 #include <linux/jiffies.h>
  29 #include <linux/hwmon.h>
  30 #include <linux/sysfs.h>
  31 #include <linux/hwmon-sysfs.h>
  32 #include <linux/err.h>
  33 #include <linux/mutex.h>
  34 #include <linux/list.h>
  35 #include <linux/platform_device.h>
  36 #include <linux/cpu.h>
  37 #include <linux/pci.h>
  38 #include <linux/smp.h>
  39 #include <linux/moduleparam.h>
  40 #include <asm/msr.h>
  41 #include <asm/processor.h>
  42 #include <asm/cpu_device_id.h>
  43
  44 #define DRVNAME "coretemp"
  45
  46 /*
  47  * force_tjmax only matters when TjMax can't be read from the CPU itself.
  48  * When set, it replaces the driver's suboptimal heuristic.
  49  */
  50 static int force_tjmax;
  51 module_param_named(tjmax, force_tjmax, int, 0444);
  52 MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
  53
  54 #define BASE_SYSFS_ATTR_NO      2       /* Sysfs Base attr no for coretemp */
  55 #define NUM_REAL_CORES          32      /* Number of Real cores per cpu */
  56 #define CORETEMP_NAME_LENGTH    17      /* String Length of attrs */
  57 #define MAX_CORE_ATTRS          4       /* Maximum no of basic attrs */
  58 #define TOTAL_ATTRS             (MAX_CORE_ATTRS + 1)
  59 #define MAX_CORE_DATA           (NUM_REAL_CORES + BASE_SYSFS_ATTR_NO)
  60
  61 #define TO_PHYS_ID(cpu)         (cpu_data(cpu).phys_proc_id)
  62 #define TO_CORE_ID(cpu)         (cpu_data(cpu).cpu_core_id)
  63 #define TO_ATTR_NO(cpu)         (TO_CORE_ID(cpu) + BASE_SYSFS_ATTR_NO)
  64
  65 #ifdef CONFIG_SMP
  66 #define for_each_sibling(i, cpu)        for_each_cpu(i, cpu_sibling_mask(cpu))
  67 #else
  68 #define for_each_sibling(i, cpu)        for (i = 0; false; )
  69 #endif
  70
  71 /*
  72  * Per-Core Temperature Data
  73  * @last_updated: The time when the current temperature value was updated
  74  *              earlier (in jiffies).
  75  * @cpu_core_id: The CPU Core from which temperature values should be read
  76  *              This value is passed as "id" field to rdmsr/wrmsr functions.
  77  * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
  78  *              from where the temperature values should be read.
  79  * @attr_size:  Total number of pre-core attrs displayed in the sysfs.
  80  * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
  81  *              Otherwise, temp_data holds coretemp data.
  82  * @valid: If this is 1, the current temperature is valid.
  83  */
  84 struct temp_data {
  85         int temp;
  86         int ttarget;
  87         int tjmax;
  88         unsigned long last_updated;
  89         unsigned int cpu;
  90         u32 cpu_core_id;
  91         u32 status_reg;
  92         int attr_size;
  93         bool is_pkg_data;
  94         bool valid;
  95         struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
  96         char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
  97         struct mutex update_lock;
  98 };
  99
 100 /* Platform Data per Physical CPU */
 101 struct platform_data {
 102         struct device *hwmon_dev;
 103         u16 phys_proc_id;
 104         struct temp_data *core_data[MAX_CORE_DATA];
 105         struct device_attribute name_attr;
 106 };
 107
 108 struct pdev_entry {
 109         struct list_head list;
 110         struct platform_device *pdev;
 111         u16 phys_proc_id;
 112 };
 113
 114 static LIST_HEAD(pdev_list);
 115 static DEFINE_MUTEX(pdev_list_mutex);
 116
 117 static ssize_t show_name(struct device *dev,
 118                         struct device_attribute *devattr, char *buf)
 119 {
 120         return sprintf(buf, "%s\n", DRVNAME);
 121 }
 122
 123 static ssize_t show_label(struct device *dev,
 124                                 struct device_attribute *devattr, char *buf)
 125 {
 126         struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
 127         struct platform_data *pdata = dev_get_drvdata(dev);
 128         struct temp_data *tdata = pdata->core_data[attr->index];
 129
 130         if (tdata->is_pkg_data)
 131                 return sprintf(buf, "Physical id %u\n", pdata->phys_proc_id);
 132
 133         return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
 134 }
 135
 136 static ssize_t show_crit_alarm(struct device *dev,
 137                                 struct device_attribute *devattr, char *buf)
 138 {
 139         u32 eax, edx;
 140         struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
 141         struct platform_data *pdata = dev_get_drvdata(dev);
 142         struct temp_data *tdata = pdata->core_data[attr->index];
 143
 144         rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
 145
 146         return sprintf(buf, "%d\n", (eax >> 5) & 1);
 147 }
 148
 149 static ssize_t show_tjmax(struct device *dev,
 150                         struct device_attribute *devattr, char *buf)
 151 {
 152         struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
 153         struct platform_data *pdata = dev_get_drvdata(dev);
 154
 155         return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
 156 }
 157
 158 static ssize_t show_ttarget(struct device *dev,
 159                                 struct device_attribute *devattr, char *buf)
 160 {
 161         struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
 162         struct platform_data *pdata = dev_get_drvdata(dev);
 163
 164         return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
 165 }
 166
 167 static ssize_t show_temp(struct device *dev,
 168                         struct device_attribute *devattr, char *buf)
 169 {
 170         u32 eax, edx;
 171         struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
 172         struct platform_data *pdata = dev_get_drvdata(dev);
 173         struct temp_data *tdata = pdata->core_data[attr->index];
 174
 175         mutex_lock(&tdata->update_lock);
 176
 177         /* Check whether the time interval has elapsed */
 178         if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
 179                 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
 180                 tdata->valid = 0;
 181                 /* Check whether the data is valid */
 182                 if (eax & 0x80000000) {
 183                         tdata->temp = tdata->tjmax -
 184                                         ((eax >> 16) & 0x7f) * 1000;
 185                         tdata->valid = 1;
 186                 }
 187                 tdata->last_updated = jiffies;
 188         }
 189
 190         mutex_unlock(&tdata->update_lock);
 191         return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
 192 }
 193
 194 static int __cpuinit adjust_tjmax(struct cpuinfo_x86 *c, u32 id,
 195                                   struct device *dev)
 196 {
 197         /* The 100C is default for both mobile and non mobile CPUs */
 198
 199         int tjmax = 100000;
 200         int tjmax_ee = 85000;
 201         int usemsr_ee = 1;
 202         int err;
 203         u32 eax, edx;
 204         struct pci_dev *host_bridge;
 205
 206         /* Early chips have no MSR for TjMax */
 207
 208         if (c->x86_model == 0xf && c->x86_mask < 4)
 209                 usemsr_ee = 0;
 210
 211         /* Atom CPUs */
 212
 213         if (c->x86_model == 0x1c || c->x86_model == 0x26
 214             || c->x86_model == 0x27) {
 215                 usemsr_ee = 0;
 216
 217                 host_bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));
 218
 219                 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL
 220                     && (host_bridge->device == 0xa000   /* NM10 based nettop */
 221                     || host_bridge->device == 0xa010))  /* NM10 based netbook */
 222                         tjmax = 100000;
 223                 else
 224                         tjmax = 90000;
 225
 226                 pci_dev_put(host_bridge);
 227         }
 228
 229         if (c->x86_model > 0xe && usemsr_ee) {
 230                 u8 platform_id;
 231
 232                 /*
 233                  * Now we can detect the mobile CPU using Intel provided table
 234                  * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
 235                  * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
 236                  */
 237                 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
 238                 if (err) {
 239                         dev_warn(dev,
 240                                  "Unable to access MSR 0x17, assuming desktop"
 241                                  " CPU\n");
 242                         usemsr_ee = 0;
 243                 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
 244                         /*
 245                          * Trust bit 28 up to Penryn, I could not find any
 246                          * documentation on that; if you happen to know
 247                          * someone at Intel please ask
 248                          */
 249                         usemsr_ee = 0;
 250                 } else {
 251                         /* Platform ID bits 52:50 (EDX starts at bit 32) */
 252                         platform_id = (edx >> 18) & 0x7;
 253
 254                         /*
 255                          * Mobile Penryn CPU seems to be platform ID 7 or 5
 256                          * (guesswork)
 257                          */
 258                         if (c->x86_model == 0x17 &&
 259                             (platform_id == 5 || platform_id == 7)) {
 260                                 /*
 261                                  * If MSR EE bit is set, set it to 90 degrees C,
 262                                  * otherwise 105 degrees C
 263                                  */
 264                                 tjmax_ee = 90000;
 265                                 tjmax = 105000;
 266                         }
 267                 }
 268         }
 269
 270         if (usemsr_ee) {
 271                 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
 272                 if (err) {
 273                         dev_warn(dev,
 274                                  "Unable to access MSR 0xEE, for Tjmax, left"
 275                                  " at default\n");
 276                 } else if (eax & 0x40000000) {
 277                         tjmax = tjmax_ee;
 278                 }
 279         } else if (tjmax == 100000) {
 280                 /*
 281                  * If we don't use msr EE it means we are desktop CPU
 282                  * (with exeception of Atom)
 283                  */
 284                 dev_warn(dev, "Using relative temperature scale!\n");
 285         }
 286
 287         return tjmax;
 288 }
 289
 290 static int __cpuinit get_tjmax(struct cpuinfo_x86 *c, u32 id,
 291                                struct device *dev)
 292 {
 293         int err;
 294         u32 eax, edx;
 295         u32 val;
 296
 297         /*
 298          * A new feature of current Intel(R) processors, the
 299          * IA32_TEMPERATURE_TARGET contains the TjMax value
 300          */
 301         err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
 302         if (err) {
 303                 if (c->x86_model > 0xe && c->x86_model != 0x1c)
 304                         dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
 305         } else {
 306                 val = (eax >> 16) & 0xff;
 307                 /*
 308                  * If the TjMax is not plausible, an assumption
 309                  * will be used
 310                  */
 311                 if (val) {
 312                         dev_dbg(dev, "TjMax is %d degrees C\n", val);
 313                         return val * 1000;
 314                 }
 315         }
 316
 317         if (force_tjmax) {
 318                 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
 319                            force_tjmax);
 320                 return force_tjmax * 1000;
 321         }
 322
 323         /*
 324          * An assumption is made for early CPUs and unreadable MSR.
 325          * NOTE: the calculated value may not be correct.
 326          */
 327         return adjust_tjmax(c, id, dev);
 328 }
 329
 330 static int __devinit create_name_attr(struct platform_data *pdata,
 331                                       struct device *dev)
 332 {
 333         sysfs_attr_init(&pdata->name_attr.attr);
 334         pdata->name_attr.attr.name = "name";
 335         pdata->name_attr.attr.mode = S_IRUGO;
 336         pdata->name_attr.show = show_name;
 337         return device_create_file(dev, &pdata->name_attr);
 338 }
 339
 340 static int __cpuinit create_core_attrs(struct temp_data *tdata,
 341                                        struct device *dev, int attr_no)
 342 {
 343         int err, i;
 344         static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
 345                         struct device_attribute *devattr, char *buf) = {
 346                         show_label, show_crit_alarm, show_temp, show_tjmax,
 347                         show_ttarget };
 348         static const char *const names[TOTAL_ATTRS] = {
 349                                         "temp%d_label", "temp%d_crit_alarm",
 350                                         "temp%d_input", "temp%d_crit",
 351                                         "temp%d_max" };
 352
 353         for (i = 0; i < tdata->attr_size; i++) {
 354                 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH, names[i],
 355                         attr_no);
 356                 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
 357                 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
 358                 tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
 359                 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
 360                 tdata->sd_attrs[i].index = attr_no;
 361                 err = device_create_file(dev, &tdata->sd_attrs[i].dev_attr);
 362                 if (err)
 363                         goto exit_free;
 364         }
 365         return 0;
 366
 367 exit_free:
 368         while (--i >= 0)
 369                 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
 370         return err;
 371 }
 372
 373
 374 static int __cpuinit chk_ucode_version(unsigned int cpu)
 375 {
 376         struct cpuinfo_x86 *c = &cpu_data(cpu);
 377
 378         /*
 379          * Check if we have problem with errata AE18 of Core processors:
 380          * Readings might stop update when processor visited too deep sleep,
 381          * fixed for stepping D0 (6EC).
 382          */
 383         if (c->x86_model == 0xe && c->x86_mask < 0xc && c->microcode < 0x39) {
 384                 pr_err("Errata AE18 not fixed, update BIOS or "
 385                        "microcode of the CPU!\n");
 386                 return -ENODEV;
 387         }
 388         return 0;
 389 }
 390
 391 static struct platform_device __cpuinit *coretemp_get_pdev(unsigned int cpu)
 392 {
 393         u16 phys_proc_id = TO_PHYS_ID(cpu);
 394         struct pdev_entry *p;
 395
 396         mutex_lock(&pdev_list_mutex);
 397
 398         list_for_each_entry(p, &pdev_list, list)
 399                 if (p->phys_proc_id == phys_proc_id) {
 400                         mutex_unlock(&pdev_list_mutex);
 401                         return p->pdev;
 402                 }
 403
 404         mutex_unlock(&pdev_list_mutex);
 405         return NULL;
 406 }
 407
 408 static struct temp_data __cpuinit *init_temp_data(unsigned int cpu,
 409                                                   int pkg_flag)
 410 {
 411         struct temp_data *tdata;
 412
 413         tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
 414         if (!tdata)
 415                 return NULL;
 416
 417         tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
 418                                                         MSR_IA32_THERM_STATUS;
 419         tdata->is_pkg_data = pkg_flag;
 420         tdata->cpu = cpu;
 421         tdata->cpu_core_id = TO_CORE_ID(cpu);
 422         tdata->attr_size = MAX_CORE_ATTRS;
 423         mutex_init(&tdata->update_lock);
 424         return tdata;
 425 }
 426
 427 static int __cpuinit create_core_data(struct platform_device *pdev,
 428                                 unsigned int cpu, int pkg_flag)
 429 {
 430         struct temp_data *tdata;
 431         struct platform_data *pdata = platform_get_drvdata(pdev);
 432         struct cpuinfo_x86 *c = &cpu_data(cpu);
 433         u32 eax, edx;
 434         int err, attr_no;
 435
 436         /*
 437          * Find attr number for sysfs:
 438          * We map the attr number to core id of the CPU
 439          * The attr number is always core id + 2
 440          * The Pkgtemp will always show up as temp1_*, if available
 441          */
 442         attr_no = pkg_flag ? 1 : TO_ATTR_NO(cpu);
 443
 444         if (attr_no > MAX_CORE_DATA - 1)
 445                 return -ERANGE;
 446
 447         /*
 448          * Provide a single set of attributes for all HT siblings of a core
 449          * to avoid duplicate sensors (the processor ID and core ID of all
 450          * HT siblings of a core are the same).
 451          * Skip if a HT sibling of this core is already registered.
 452          * This is not an error.
 453          */
 454         if (pdata->core_data[attr_no] != NULL)
 455                 return 0;
 456
 457         tdata = init_temp_data(cpu, pkg_flag);
 458         if (!tdata)
 459                 return -ENOMEM;
 460
 461         /* Test if we can access the status register */
 462         err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
 463         if (err)
 464                 goto exit_free;
 465
 466         /* We can access status register. Get Critical Temperature */
 467         tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
 468
 469         /*
 470          * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
 471          * The target temperature is available on older CPUs but not in this
 472          * register. Atoms don't have the register at all.
 473          */
 474         if (c->x86_model > 0xe && c->x86_model != 0x1c) {
 475                 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
 476                                         &eax, &edx);
 477                 if (!err) {
 478                         tdata->ttarget
 479                           = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
 480                         tdata->attr_size++;
 481                 }
 482         }
 483
 484         pdata->core_data[attr_no] = tdata;
 485
 486         /* Create sysfs interfaces */
 487         err = create_core_attrs(tdata, &pdev->dev, attr_no);
 488         if (err)
 489                 goto exit_free;
 490
 491         return 0;
 492 exit_free:
 493         pdata->core_data[attr_no] = NULL;
 494         kfree(tdata);
 495         return err;
 496 }
 497
 498 static void __cpuinit coretemp_add_core(unsigned int cpu, int pkg_flag)
 499 {
 500         struct platform_device *pdev = coretemp_get_pdev(cpu);
 501         int err;
 502
 503         if (!pdev)
 504                 return;
 505
 506         err = create_core_data(pdev, cpu, pkg_flag);
 507         if (err)
 508                 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
 509 }
 510
 511 static void coretemp_remove_core(struct platform_data *pdata,
 512                                 struct device *dev, int indx)
 513 {
 514         int i;
 515         struct temp_data *tdata = pdata->core_data[indx];
 516
 517         /* Remove the sysfs attributes */
 518         for (i = 0; i < tdata->attr_size; i++)
 519                 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
 520
 521         kfree(pdata->core_data[indx]);
 522         pdata->core_data[indx] = NULL;
 523 }
 524
 525 static int __devinit coretemp_probe(struct platform_device *pdev)
 526 {
 527         struct platform_data *pdata;
 528         int err;
 529
 530         /* Initialize the per-package data structures */
 531         pdata = kzalloc(sizeof(struct platform_data), GFP_KERNEL);
 532         if (!pdata)
 533                 return -ENOMEM;
 534
 535         err = create_name_attr(pdata, &pdev->dev);
 536         if (err)
 537                 goto exit_free;
 538
 539         pdata->phys_proc_id = pdev->id;
 540         platform_set_drvdata(pdev, pdata);
 541
 542         pdata->hwmon_dev = hwmon_device_register(&pdev->dev);
 543         if (IS_ERR(pdata->hwmon_dev)) {
 544                 err = PTR_ERR(pdata->hwmon_dev);
 545                 dev_err(&pdev->dev, "Class registration failed (%d)\n", err);
 546                 goto exit_name;
 547         }
 548         return 0;
 549
 550 exit_name:
 551         device_remove_file(&pdev->dev, &pdata->name_attr);
 552         platform_set_drvdata(pdev, NULL);
 553 exit_free:
 554         kfree(pdata);
 555         return err;
 556 }
 557
 558 static int __devexit coretemp_remove(struct platform_device *pdev)
 559 {
 560         struct platform_data *pdata = platform_get_drvdata(pdev);
 561         int i;
 562
 563         for (i = MAX_CORE_DATA - 1; i >= 0; --i)
 564                 if (pdata->core_data[i])
 565                         coretemp_remove_core(pdata, &pdev->dev, i);
 566
 567         device_remove_file(&pdev->dev, &pdata->name_attr);
 568         hwmon_device_unregister(pdata->hwmon_dev);
 569         platform_set_drvdata(pdev, NULL);
 570         kfree(pdata);
 571         return 0;
 572 }
 573
 574 static struct platform_driver coretemp_driver = {
 575         .driver = {
 576                 .owner = THIS_MODULE,
 577                 .name = DRVNAME,
 578         },
 579         .probe = coretemp_probe,
 580         .remove = __devexit_p(coretemp_remove),
 581 };
 582
 583 static int __cpuinit coretemp_device_add(unsigned int cpu)
 584 {
 585         int err;
 586         struct platform_device *pdev;
 587         struct pdev_entry *pdev_entry;
 588
 589         mutex_lock(&pdev_list_mutex);
 590
 591         pdev = platform_device_alloc(DRVNAME, TO_PHYS_ID(cpu));
 592         if (!pdev) {
 593                 err = -ENOMEM;
 594                 pr_err("Device allocation failed\n");
 595                 goto exit;
 596         }
 597
 598         pdev_entry = kzalloc(sizeof(struct pdev_entry), GFP_KERNEL);
 599         if (!pdev_entry) {
 600                 err = -ENOMEM;
 601                 goto exit_device_put;
 602         }
 603
 604         err = platform_device_add(pdev);
 605         if (err) {
 606                 pr_err("Device addition failed (%d)\n", err);
 607                 goto exit_device_free;
 608         }
 609
 610         pdev_entry->pdev = pdev;
 611         pdev_entry->phys_proc_id = pdev->id;
 612
 613         list_add_tail(&pdev_entry->list, &pdev_list);
 614         mutex_unlock(&pdev_list_mutex);
 615
 616         return 0;
 617
 618 exit_device_free:
 619         kfree(pdev_entry);
 620 exit_device_put:
 621         platform_device_put(pdev);
 622 exit:
 623         mutex_unlock(&pdev_list_mutex);
 624         return err;
 625 }
 626
 627 static void __cpuinit coretemp_device_remove(unsigned int cpu)
 628 {
 629         struct pdev_entry *p, *n;
 630         u16 phys_proc_id = TO_PHYS_ID(cpu);
 631
 632         mutex_lock(&pdev_list_mutex);
 633         list_for_each_entry_safe(p, n, &pdev_list, list) {
 634                 if (p->phys_proc_id != phys_proc_id)
 635                         continue;
 636                 platform_device_unregister(p->pdev);
 637                 list_del(&p->list);
 638                 kfree(p);
 639         }
 640         mutex_unlock(&pdev_list_mutex);
 641 }
 642
 643 static bool __cpuinit is_any_core_online(struct platform_data *pdata)
 644 {
 645         int i;
 646
 647         /* Find online cores, except pkgtemp data */
 648         for (i = MAX_CORE_DATA - 1; i >= 0; --i) {
 649                 if (pdata->core_data[i] &&
 650                         !pdata->core_data[i]->is_pkg_data) {
 651                         return true;
 652                 }
 653         }
 654         return false;
 655 }
 656
 657 static void __cpuinit get_core_online(unsigned int cpu)
 658 {
 659         struct cpuinfo_x86 *c = &cpu_data(cpu);
 660         struct platform_device *pdev = coretemp_get_pdev(cpu);
 661         int err;
 662
 663         /*
 664          * CPUID.06H.EAX[0] indicates whether the CPU has thermal
 665          * sensors. We check this bit only, all the early CPUs
 666          * without thermal sensors will be filtered out.
 667          */
 668         if (!cpu_has(c, X86_FEATURE_DTS))
 669                 return;
 670
 671         if (!pdev) {
 672                 /* Check the microcode version of the CPU */
 673                 if (chk_ucode_version(cpu))
 674                         return;
 675
 676                 /*
 677                  * Alright, we have DTS support.
 678                  * We are bringing the _first_ core in this pkg
 679                  * online. So, initialize per-pkg data structures and
 680                  * then bring this core online.
 681                  */
 682                 err = coretemp_device_add(cpu);
 683                 if (err)
 684                         return;
 685                 /*
 686                  * Check whether pkgtemp support is available.
 687                  * If so, add interfaces for pkgtemp.
 688                  */
 689                 if (cpu_has(c, X86_FEATURE_PTS))
 690                         coretemp_add_core(cpu, 1);
 691         }
 692         /*
 693          * Physical CPU device already exists.
 694          * So, just add interfaces for this core.
 695          */
 696         coretemp_add_core(cpu, 0);
 697 }
 698
 699 static void __cpuinit put_core_offline(unsigned int cpu)
 700 {
 701         int i, indx;
 702         struct platform_data *pdata;
 703         struct platform_device *pdev = coretemp_get_pdev(cpu);
 704
 705         /* If the physical CPU device does not exist, just return */
 706         if (!pdev)
 707                 return;
 708
 709         pdata = platform_get_drvdata(pdev);
 710
 711         indx = TO_ATTR_NO(cpu);
 712
 713         /* The core id is too big, just return */
 714         if (indx > MAX_CORE_DATA - 1)
 715                 return;
 716
 717         if (pdata->core_data[indx] && pdata->core_data[indx]->cpu == cpu)
 718                 coretemp_remove_core(pdata, &pdev->dev, indx);
 719
 720         /*
 721          * If a HT sibling of a core is taken offline, but another HT sibling
 722          * of the same core is still online, register the alternate sibling.
 723          * This ensures that exactly one set of attributes is provided as long
 724          * as at least one HT sibling of a core is online.
 725          */
 726         for_each_sibling(i, cpu) {
 727                 if (i != cpu) {
 728                         get_core_online(i);
 729                         /*
 730                          * Display temperature sensor data for one HT sibling
 731                          * per core only, so abort the loop after one such
 732                          * sibling has been found.
 733                          */
 734                         break;
 735                 }
 736         }
 737         /*
 738          * If all cores in this pkg are offline, remove the device.
 739          * coretemp_device_remove calls unregister_platform_device,
 740          * which in turn calls coretemp_remove. This removes the
 741          * pkgtemp entry and does other clean ups.
 742          */
 743         if (!is_any_core_online(pdata))
 744                 coretemp_device_remove(cpu);
 745 }
 746
 747 static int __cpuinit coretemp_cpu_callback(struct notifier_block *nfb,
 748                                  unsigned long action, void *hcpu)
 749 {
 750         unsigned int cpu = (unsigned long) hcpu;
 751
 752         switch (action) {
 753         case CPU_ONLINE:
 754         case CPU_DOWN_FAILED:
 755                 get_core_online(cpu);
 756                 break;
 757         case CPU_DOWN_PREPARE:
 758                 put_core_offline(cpu);
 759                 break;
 760         }
 761         return NOTIFY_OK;
 762 }
 763
 764 static struct notifier_block coretemp_cpu_notifier __refdata = {
 765         .notifier_call = coretemp_cpu_callback,
 766 };
 767
 768 static const struct x86_cpu_id coretemp_ids[] = {
 769         { X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_DTS },
 770         {}
 771 };
 772 MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
 773
 774 static int __init coretemp_init(void)
 775 {
 776         int i, err = -ENODEV;
 777
 778         /*
 779          * CPUID.06H.EAX[0] indicates whether the CPU has thermal
 780          * sensors. We check this bit only, all the early CPUs
 781          * without thermal sensors will be filtered out.
 782          */
 783         if (!x86_match_cpu(coretemp_ids))
 784                 return -ENODEV;
 785
 786         err = platform_driver_register(&coretemp_driver);
 787         if (err)
 788                 goto exit;
 789
 790         for_each_online_cpu(i)
 791                 get_core_online(i);
 792
 793 #ifndef CONFIG_HOTPLUG_CPU
 794         if (list_empty(&pdev_list)) {
 795                 err = -ENODEV;
 796                 goto exit_driver_unreg;
 797         }
 798 #endif
 799
 800         register_hotcpu_notifier(&coretemp_cpu_notifier);
 801         return 0;
 802
 803 #ifndef CONFIG_HOTPLUG_CPU
 804 exit_driver_unreg:
 805         platform_driver_unregister(&coretemp_driver);
 806 #endif
 807 exit:
 808         return err;
 809 }
 810
 811 static void __exit coretemp_exit(void)
 812 {
 813         struct pdev_entry *p, *n;
 814
 815         unregister_hotcpu_notifier(&coretemp_cpu_notifier);
 816         mutex_lock(&pdev_list_mutex);
 817         list_for_each_entry_safe(p, n, &pdev_list, list) {
 818                 platform_device_unregister(p->pdev);
 819                 list_del(&p->list);
 820                 kfree(p);
 821         }
 822         mutex_unlock(&pdev_list_mutex);
 823         platform_driver_unregister(&coretemp_driver);
 824 }
 825
 826 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
 827 MODULE_DESCRIPTION("Intel Core temperature monitor");
 828 MODULE_LICENSE("GPL");
 829
 830 module_init(coretemp_init)
 831 module_exit(coretemp_exit)