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
  43 #define DRVNAME "coretemp"
  44
  45 /*
  46  * force_tjmax only matters when TjMax can't be read from the CPU itself.
  47  * When set, it replaces the driver's suboptimal heuristic.
  48  */
  49 static int force_tjmax;
  50 module_param_named(tjmax, force_tjmax, int, 0444);
  51 MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
  52
  53 #define BASE_SYSFS_ATTR_NO      2       /* Sysfs Base attr no for coretemp */
  54 #define NUM_REAL_CORES          32      /* Number of Real cores per cpu */
  55 #define CORETEMP_NAME_LENGTH    17      /* String Length of attrs */
  56 #define MAX_CORE_ATTRS          4       /* Maximum no of basic attrs */
  57 #define TOTAL_ATTRS             (MAX_CORE_ATTRS + 1)
  58 #define MAX_CORE_DATA           (NUM_REAL_CORES + BASE_SYSFS_ATTR_NO)
  59
  60 #define TO_PHYS_ID(cpu)         cpu_data(cpu).phys_proc_id
  61 #define TO_CORE_ID(cpu)         cpu_data(cpu).cpu_core_id
  62 #define TO_ATTR_NO(cpu)         (TO_CORE_ID(cpu) + BASE_SYSFS_ATTR_NO)
  63
  64 #ifdef CONFIG_SMP
  65 #define for_each_sibling(i, cpu)        for_each_cpu(i, cpu_sibling_mask(cpu))
  66 #else
  67 #define for_each_sibling(i, cpu)        for (i = 0; false; )
  68 #endif
  69
  70 /*
  71  * Per-Core Temperature Data
  72  * @last_updated: The time when the current temperature value was updated
  73  *              earlier (in jiffies).
  74  * @cpu_core_id: The CPU Core from which temperature values should be read
  75  *              This value is passed as "id" field to rdmsr/wrmsr functions.
  76  * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
  77  *              from where the temperature values should be read.
  78  * @attr_size:  Total number of pre-core attrs displayed in the sysfs.
  79  * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
  80  *              Otherwise, temp_data holds coretemp data.
  81  * @valid: If this is 1, the current temperature is valid.
  82  */
  83 struct temp_data {
  84         int temp;
  85         int ttarget;
  86         int tjmax;
  87         unsigned long last_updated;
  88         unsigned int cpu;
  89         u32 cpu_core_id;
  90         u32 status_reg;
  91         int attr_size;
  92         bool is_pkg_data;
  93         bool valid;
  94         struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
  95         char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
  96         struct mutex update_lock;
  97 };
  98
  99 /* Platform Data per Physical CPU */
 100 struct platform_data {
 101         struct device *hwmon_dev;
 102         u16 phys_proc_id;
 103         struct temp_data *core_data[MAX_CORE_DATA];
 104         struct device_attribute name_attr;
 105 };
 106
 107 struct pdev_entry {
 108         struct list_head list;
 109         struct platform_device *pdev;
 110         u16 phys_proc_id;
 111 };
 112
 113 static LIST_HEAD(pdev_list);
 114 static DEFINE_MUTEX(pdev_list_mutex);
 115
 116 static ssize_t show_name(struct device *dev,
 117                         struct device_attribute *devattr, char *buf)
 118 {
 119         return sprintf(buf, "%s\n", DRVNAME);
 120 }
 121
 122 static ssize_t show_label(struct device *dev,
 123                                 struct device_attribute *devattr, char *buf)
 124 {
 125         struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
 126         struct platform_data *pdata = dev_get_drvdata(dev);
 127         struct temp_data *tdata = pdata->core_data[attr->index];
 128
 129         if (tdata->is_pkg_data)
 130                 return sprintf(buf, "Physical id %u\n", pdata->phys_proc_id);
 131
 132         return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
 133 }
 134
 135 static ssize_t show_crit_alarm(struct device *dev,
 136                                 struct device_attribute *devattr, char *buf)
 137 {
 138         u32 eax, edx;
 139         struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
 140         struct platform_data *pdata = dev_get_drvdata(dev);
 141         struct temp_data *tdata = pdata->core_data[attr->index];
 142
 143         rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
 144
 145         return sprintf(buf, "%d\n", (eax >> 5) & 1);
 146 }
 147
 148 static ssize_t show_tjmax(struct device *dev,
 149                         struct device_attribute *devattr, char *buf)
 150 {
 151         struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
 152         struct platform_data *pdata = dev_get_drvdata(dev);
 153
 154         return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
 155 }
 156
 157 static ssize_t show_ttarget(struct device *dev,
 158                                 struct device_attribute *devattr, char *buf)
 159 {
 160         struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
 161         struct platform_data *pdata = dev_get_drvdata(dev);
 162
 163         return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
 164 }
 165
 166 static ssize_t show_temp(struct device *dev,
 167                         struct device_attribute *devattr, char *buf)
 168 {
 169         u32 eax, edx;
 170         struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
 171         struct platform_data *pdata = dev_get_drvdata(dev);
 172         struct temp_data *tdata = pdata->core_data[attr->index];
 173
 174         mutex_lock(&tdata->update_lock);
 175
 176         /* Check whether the time interval has elapsed */
 177         if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
 178                 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
 179                 tdata->valid = 0;
 180                 /* Check whether the data is valid */
 181                 if (eax & 0x80000000) {
 182                         tdata->temp = tdata->tjmax -
 183                                         ((eax >> 16) & 0x7f) * 1000;
 184                         tdata->valid = 1;
 185                 }
 186                 tdata->last_updated = jiffies;
 187         }
 188
 189         mutex_unlock(&tdata->update_lock);
 190         return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
 191 }
 192
 193 static int __cpuinit adjust_tjmax(struct cpuinfo_x86 *c, u32 id,
 194                                   struct device *dev)
 195 {
 196         /* The 100C is default for both mobile and non mobile CPUs */
 197
 198         int tjmax = 100000;
 199         int tjmax_ee = 85000;
 200         int usemsr_ee = 1;
 201         int err;
 202         u32 eax, edx;
 203         struct pci_dev *host_bridge;
 204
 205         /* Early chips have no MSR for TjMax */
 206
 207         if (c->x86_model == 0xf && c->x86_mask < 4)
 208                 usemsr_ee = 0;
 209
 210         /* Atom CPUs */
 211
 212         if (c->x86_model == 0x1c) {
 213                 usemsr_ee = 0;
 214
 215                 host_bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));
 216
 217                 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL
 218                     && (host_bridge->device == 0xa000   /* NM10 based nettop */
 219                     || host_bridge->device == 0xa010))  /* NM10 based netbook */
 220                         tjmax = 100000;
 221                 else
 222                         tjmax = 90000;
 223
 224                 pci_dev_put(host_bridge);
 225         }
 226
 227         if (c->x86_model > 0xe && usemsr_ee) {
 228                 u8 platform_id;
 229
 230                 /*
 231                  * Now we can detect the mobile CPU using Intel provided table
 232                  * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
 233                  * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
 234                  */
 235                 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
 236                 if (err) {
 237                         dev_warn(dev,
 238                                  "Unable to access MSR 0x17, assuming desktop"
 239                                  " CPU\n");
 240                         usemsr_ee = 0;
 241                 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
 242                         /*
 243                          * Trust bit 28 up to Penryn, I could not find any
 244                          * documentation on that; if you happen to know
 245                          * someone at Intel please ask
 246                          */
 247                         usemsr_ee = 0;
 248                 } else {
 249                         /* Platform ID bits 52:50 (EDX starts at bit 32) */
 250                         platform_id = (edx >> 18) & 0x7;
 251
 252                         /*
 253                          * Mobile Penryn CPU seems to be platform ID 7 or 5
 254                          * (guesswork)
 255                          */
 256                         if (c->x86_model == 0x17 &&
 257                             (platform_id == 5 || platform_id == 7)) {
 258                                 /*
 259                                  * If MSR EE bit is set, set it to 90 degrees C,
 260                                  * otherwise 105 degrees C
 261                                  */
 262                                 tjmax_ee = 90000;
 263                                 tjmax = 105000;
 264                         }
 265                 }
 266         }
 267
 268         if (usemsr_ee) {
 269                 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
 270                 if (err) {
 271                         dev_warn(dev,
 272                                  "Unable to access MSR 0xEE, for Tjmax, left"
 273                                  " at default\n");
 274                 } else if (eax & 0x40000000) {
 275                         tjmax = tjmax_ee;
 276                 }
 277         } else if (tjmax == 100000) {
 278                 /*
 279                  * If we don't use msr EE it means we are desktop CPU
 280                  * (with exeception of Atom)
 281                  */
 282                 dev_warn(dev, "Using relative temperature scale!\n");
 283         }
 284
 285         return tjmax;
 286 }
 287
 288 static int __cpuinit get_tjmax(struct cpuinfo_x86 *c, u32 id,
 289                                struct device *dev)
 290 {
 291         int err;
 292         u32 eax, edx;
 293         u32 val;
 294
 295         /*
 296          * A new feature of current Intel(R) processors, the
 297          * IA32_TEMPERATURE_TARGET contains the TjMax value
 298          */
 299         err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
 300         if (err) {
 301                 if (c->x86_model > 0xe && c->x86_model != 0x1c)
 302                         dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
 303         } else {
 304                 val = (eax >> 16) & 0xff;
 305                 /*
 306                  * If the TjMax is not plausible, an assumption
 307                  * will be used
 308                  */
 309                 if (val) {
 310                         dev_dbg(dev, "TjMax is %d degrees C\n", val);
 311                         return val * 1000;
 312                 }
 313         }
 314
 315         if (force_tjmax) {
 316                 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
 317                            force_tjmax);
 318                 return force_tjmax * 1000;
 319         }
 320
 321         /*
 322          * An assumption is made for early CPUs and unreadable MSR.
 323          * NOTE: the calculated value may not be correct.
 324          */
 325         return adjust_tjmax(c, id, dev);
 326 }
 327
 328 static int __devinit create_name_attr(struct platform_data *pdata,
 329                                       struct device *dev)
 330 {
 331         sysfs_attr_init(&pdata->name_attr.attr);
 332         pdata->name_attr.attr.name = "name";
 333         pdata->name_attr.attr.mode = S_IRUGO;
 334         pdata->name_attr.show = show_name;
 335         return device_create_file(dev, &pdata->name_attr);
 336 }
 337
 338 static int __cpuinit create_core_attrs(struct temp_data *tdata,
 339                                        struct device *dev, int attr_no)
 340 {
 341         int err, i;
 342         static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
 343                         struct device_attribute *devattr, char *buf) = {
 344                         show_label, show_crit_alarm, show_temp, show_tjmax,
 345                         show_ttarget };
 346         static const char *const names[TOTAL_ATTRS] = {
 347                                         "temp%d_label", "temp%d_crit_alarm",
 348                                         "temp%d_input", "temp%d_crit",
 349                                         "temp%d_max" };
 350
 351         for (i = 0; i < tdata->attr_size; i++) {
 352                 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH, names[i],
 353                         attr_no);
 354                 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
 355                 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
 356                 tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
 357                 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
 358                 tdata->sd_attrs[i].index = attr_no;
 359                 err = device_create_file(dev, &tdata->sd_attrs[i].dev_attr);
 360                 if (err)
 361                         goto exit_free;
 362         }
 363         return 0;
 364
 365 exit_free:
 366         while (--i >= 0)
 367                 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
 368         return err;
 369 }
 370
 371
 372 static int __cpuinit chk_ucode_version(unsigned int cpu)
 373 {
 374         struct cpuinfo_x86 *c = &cpu_data(cpu);
 375
 376         /*
 377          * Check if we have problem with errata AE18 of Core processors:
 378          * Readings might stop update when processor visited too deep sleep,
 379          * fixed for stepping D0 (6EC).
 380          */
 381         if (c->x86_model == 0xe && c->x86_mask < 0xc && c->microcode < 0x39) {
 382                 pr_err("Errata AE18 not fixed, update BIOS or "
 383                        "microcode of the CPU!\n");
 384                 return -ENODEV;
 385         }
 386         return 0;
 387 }
 388
 389 static struct platform_device __cpuinit *coretemp_get_pdev(unsigned int cpu)
 390 {
 391         u16 phys_proc_id = TO_PHYS_ID(cpu);
 392         struct pdev_entry *p;
 393
 394         mutex_lock(&pdev_list_mutex);
 395
 396         list_for_each_entry(p, &pdev_list, list)
 397                 if (p->phys_proc_id == phys_proc_id) {
 398                         mutex_unlock(&pdev_list_mutex);
 399                         return p->pdev;
 400                 }
 401
 402         mutex_unlock(&pdev_list_mutex);
 403         return NULL;
 404 }
 405
 406 static struct temp_data __cpuinit *init_temp_data(unsigned int cpu,
 407                                                   int pkg_flag)
 408 {
 409         struct temp_data *tdata;
 410
 411         tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
 412         if (!tdata)
 413                 return NULL;
 414
 415         tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
 416                                                         MSR_IA32_THERM_STATUS;
 417         tdata->is_pkg_data = pkg_flag;
 418         tdata->cpu = cpu;
 419         tdata->cpu_core_id = TO_CORE_ID(cpu);
 420         tdata->attr_size = MAX_CORE_ATTRS;
 421         mutex_init(&tdata->update_lock);
 422         return tdata;
 423 }
 424
 425 static int __cpuinit create_core_data(struct platform_device *pdev,
 426                                 unsigned int cpu, int pkg_flag)
 427 {
 428         struct temp_data *tdata;
 429         struct platform_data *pdata = platform_get_drvdata(pdev);
 430         struct cpuinfo_x86 *c = &cpu_data(cpu);
 431         u32 eax, edx;
 432         int err, attr_no;
 433
 434         /*
 435          * Find attr number for sysfs:
 436          * We map the attr number to core id of the CPU
 437          * The attr number is always core id + 2
 438          * The Pkgtemp will always show up as temp1_*, if available
 439          */
 440         attr_no = pkg_flag ? 1 : TO_ATTR_NO(cpu);
 441
 442         if (attr_no > MAX_CORE_DATA - 1)
 443                 return -ERANGE;
 444
 445         /*
 446          * Provide a single set of attributes for all HT siblings of a core
 447          * to avoid duplicate sensors (the processor ID and core ID of all
 448          * HT siblings of a core are the same).
 449          * Skip if a HT sibling of this core is already registered.
 450          * This is not an error.
 451          */
 452         if (pdata->core_data[attr_no] != NULL)
 453                 return 0;
 454
 455         tdata = init_temp_data(cpu, pkg_flag);
 456         if (!tdata)
 457                 return -ENOMEM;
 458
 459         /* Test if we can access the status register */
 460         err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
 461         if (err)
 462                 goto exit_free;
 463
 464         /* We can access status register. Get Critical Temperature */
 465         tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
 466
 467         /*
 468          * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
 469          * The target temperature is available on older CPUs but not in this
 470          * register. Atoms don't have the register at all.
 471          */
 472         if (c->x86_model > 0xe && c->x86_model != 0x1c) {
 473                 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
 474                                         &eax, &edx);
 475                 if (!err) {
 476                         tdata->ttarget
 477                           = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
 478                         tdata->attr_size++;
 479                 }
 480         }
 481
 482         pdata->core_data[attr_no] = tdata;
 483
 484         /* Create sysfs interfaces */
 485         err = create_core_attrs(tdata, &pdev->dev, attr_no);
 486         if (err)
 487                 goto exit_free;
 488
 489         return 0;
 490 exit_free:
 491         pdata->core_data[attr_no] = NULL;
 492         kfree(tdata);
 493         return err;
 494 }
 495
 496 static void __cpuinit coretemp_add_core(unsigned int cpu, int pkg_flag)
 497 {
 498         struct platform_device *pdev = coretemp_get_pdev(cpu);
 499         int err;
 500
 501         if (!pdev)
 502                 return;
 503
 504         err = create_core_data(pdev, cpu, pkg_flag);
 505         if (err)
 506                 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
 507 }
 508
 509 static void coretemp_remove_core(struct platform_data *pdata,
 510                                 struct device *dev, int indx)
 511 {
 512         int i;
 513         struct temp_data *tdata = pdata->core_data[indx];
 514
 515         /* Remove the sysfs attributes */
 516         for (i = 0; i < tdata->attr_size; i++)
 517                 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
 518
 519         kfree(pdata->core_data[indx]);
 520         pdata->core_data[indx] = NULL;
 521 }
 522
 523 static int __devinit coretemp_probe(struct platform_device *pdev)
 524 {
 525         struct platform_data *pdata;
 526         int err;
 527
 528         /* Initialize the per-package data structures */
 529         pdata = kzalloc(sizeof(struct platform_data), GFP_KERNEL);
 530         if (!pdata)
 531                 return -ENOMEM;
 532
 533         err = create_name_attr(pdata, &pdev->dev);
 534         if (err)
 535                 goto exit_free;
 536
 537         pdata->phys_proc_id = pdev->id;
 538         platform_set_drvdata(pdev, pdata);
 539
 540         pdata->hwmon_dev = hwmon_device_register(&pdev->dev);
 541         if (IS_ERR(pdata->hwmon_dev)) {
 542                 err = PTR_ERR(pdata->hwmon_dev);
 543                 dev_err(&pdev->dev, "Class registration failed (%d)\n", err);
 544                 goto exit_name;
 545         }
 546         return 0;
 547
 548 exit_name:
 549         device_remove_file(&pdev->dev, &pdata->name_attr);
 550         platform_set_drvdata(pdev, NULL);
 551 exit_free:
 552         kfree(pdata);
 553         return err;
 554 }
 555
 556 static int __devexit coretemp_remove(struct platform_device *pdev)
 557 {
 558         struct platform_data *pdata = platform_get_drvdata(pdev);
 559         int i;
 560
 561         for (i = MAX_CORE_DATA - 1; i >= 0; --i)
 562                 if (pdata->core_data[i])
 563                         coretemp_remove_core(pdata, &pdev->dev, i);
 564
 565         device_remove_file(&pdev->dev, &pdata->name_attr);
 566         hwmon_device_unregister(pdata->hwmon_dev);
 567         platform_set_drvdata(pdev, NULL);
 568         kfree(pdata);
 569         return 0;
 570 }
 571
 572 static struct platform_driver coretemp_driver = {
 573         .driver = {
 574                 .owner = THIS_MODULE,
 575                 .name = DRVNAME,
 576         },
 577         .probe = coretemp_probe,
 578         .remove = __devexit_p(coretemp_remove),
 579 };
 580
 581 static int __cpuinit coretemp_device_add(unsigned int cpu)
 582 {
 583         int err;
 584         struct platform_device *pdev;
 585         struct pdev_entry *pdev_entry;
 586
 587         mutex_lock(&pdev_list_mutex);
 588
 589         pdev = platform_device_alloc(DRVNAME, TO_PHYS_ID(cpu));
 590         if (!pdev) {
 591                 err = -ENOMEM;
 592                 pr_err("Device allocation failed\n");
 593                 goto exit;
 594         }
 595
 596         pdev_entry = kzalloc(sizeof(struct pdev_entry), GFP_KERNEL);
 597         if (!pdev_entry) {
 598                 err = -ENOMEM;
 599                 goto exit_device_put;
 600         }
 601
 602         err = platform_device_add(pdev);
 603         if (err) {
 604                 pr_err("Device addition failed (%d)\n", err);
 605                 goto exit_device_free;
 606         }
 607
 608         pdev_entry->pdev = pdev;
 609         pdev_entry->phys_proc_id = pdev->id;
 610
 611         list_add_tail(&pdev_entry->list, &pdev_list);
 612         mutex_unlock(&pdev_list_mutex);
 613
 614         return 0;
 615
 616 exit_device_free:
 617         kfree(pdev_entry);
 618 exit_device_put:
 619         platform_device_put(pdev);
 620 exit:
 621         mutex_unlock(&pdev_list_mutex);
 622         return err;
 623 }
 624
 625 static void __cpuinit coretemp_device_remove(unsigned int cpu)
 626 {
 627         struct pdev_entry *p, *n;
 628         u16 phys_proc_id = TO_PHYS_ID(cpu);
 629
 630         mutex_lock(&pdev_list_mutex);
 631         list_for_each_entry_safe(p, n, &pdev_list, list) {
 632                 if (p->phys_proc_id != phys_proc_id)
 633                         continue;
 634                 platform_device_unregister(p->pdev);
 635                 list_del(&p->list);
 636                 kfree(p);
 637         }
 638         mutex_unlock(&pdev_list_mutex);
 639 }
 640
 641 static bool __cpuinit is_any_core_online(struct platform_data *pdata)
 642 {
 643         int i;
 644
 645         /* Find online cores, except pkgtemp data */
 646         for (i = MAX_CORE_DATA - 1; i >= 0; --i) {
 647                 if (pdata->core_data[i] &&
 648                         !pdata->core_data[i]->is_pkg_data) {
 649                         return true;
 650                 }
 651         }
 652         return false;
 653 }
 654
 655 static void __cpuinit get_core_online(unsigned int cpu)
 656 {
 657         struct cpuinfo_x86 *c = &cpu_data(cpu);
 658         struct platform_device *pdev = coretemp_get_pdev(cpu);
 659         int err;
 660
 661         /*
 662          * CPUID.06H.EAX[0] indicates whether the CPU has thermal
 663          * sensors. We check this bit only, all the early CPUs
 664          * without thermal sensors will be filtered out.
 665          */
 666         if (!cpu_has(c, X86_FEATURE_DTS))
 667                 return;
 668
 669         if (!pdev) {
 670                 /* Check the microcode version of the CPU */
 671                 if (chk_ucode_version(cpu))
 672                         return;
 673
 674                 /*
 675                  * Alright, we have DTS support.
 676                  * We are bringing the _first_ core in this pkg
 677                  * online. So, initialize per-pkg data structures and
 678                  * then bring this core online.
 679                  */
 680                 err = coretemp_device_add(cpu);
 681                 if (err)
 682                         return;
 683                 /*
 684                  * Check whether pkgtemp support is available.
 685                  * If so, add interfaces for pkgtemp.
 686                  */
 687                 if (cpu_has(c, X86_FEATURE_PTS))
 688                         coretemp_add_core(cpu, 1);
 689         }
 690         /*
 691          * Physical CPU device already exists.
 692          * So, just add interfaces for this core.
 693          */
 694         coretemp_add_core(cpu, 0);
 695 }
 696
 697 static void __cpuinit put_core_offline(unsigned int cpu)
 698 {
 699         int i, indx;
 700         struct platform_data *pdata;
 701         struct platform_device *pdev = coretemp_get_pdev(cpu);
 702
 703         /* If the physical CPU device does not exist, just return */
 704         if (!pdev)
 705                 return;
 706
 707         pdata = platform_get_drvdata(pdev);
 708
 709         indx = TO_ATTR_NO(cpu);
 710
 711         /* The core id is too big, just return */
 712         if (indx > MAX_CORE_DATA - 1)
 713                 return;
 714
 715         if (pdata->core_data[indx] && pdata->core_data[indx]->cpu == cpu)
 716                 coretemp_remove_core(pdata, &pdev->dev, indx);
 717
 718         /*
 719          * If a HT sibling of a core is taken offline, but another HT sibling
 720          * of the same core is still online, register the alternate sibling.
 721          * This ensures that exactly one set of attributes is provided as long
 722          * as at least one HT sibling of a core is online.
 723          */
 724         for_each_sibling(i, cpu) {
 725                 if (i != cpu) {
 726                         get_core_online(i);
 727                         /*
 728                          * Display temperature sensor data for one HT sibling
 729                          * per core only, so abort the loop after one such
 730                          * sibling has been found.
 731                          */
 732                         break;
 733                 }
 734         }
 735         /*
 736          * If all cores in this pkg are offline, remove the device.
 737          * coretemp_device_remove calls unregister_platform_device,
 738          * which in turn calls coretemp_remove. This removes the
 739          * pkgtemp entry and does other clean ups.
 740          */
 741         if (!is_any_core_online(pdata))
 742                 coretemp_device_remove(cpu);
 743 }
 744
 745 static int __cpuinit coretemp_cpu_callback(struct notifier_block *nfb,
 746                                  unsigned long action, void *hcpu)
 747 {
 748         unsigned int cpu = (unsigned long) hcpu;
 749
 750         switch (action) {
 751         case CPU_ONLINE:
 752         case CPU_DOWN_FAILED:
 753                 get_core_online(cpu);
 754                 break;
 755         case CPU_DOWN_PREPARE:
 756                 put_core_offline(cpu);
 757                 break;
 758         }
 759         return NOTIFY_OK;
 760 }
 761
 762 static struct notifier_block coretemp_cpu_notifier __refdata = {
 763         .notifier_call = coretemp_cpu_callback,
 764 };
 765
 766 static int __init coretemp_init(void)
 767 {
 768         int i, err = -ENODEV;
 769
 770         /* quick check if we run Intel */
 771         if (cpu_data(0).x86_vendor != X86_VENDOR_INTEL)
 772                 goto exit;
 773
 774         err = platform_driver_register(&coretemp_driver);
 775         if (err)
 776                 goto exit;
 777
 778         for_each_online_cpu(i)
 779                 get_core_online(i);
 780
 781 #ifndef CONFIG_HOTPLUG_CPU
 782         if (list_empty(&pdev_list)) {
 783                 err = -ENODEV;
 784                 goto exit_driver_unreg;
 785         }
 786 #endif
 787
 788         register_hotcpu_notifier(&coretemp_cpu_notifier);
 789         return 0;
 790
 791 #ifndef CONFIG_HOTPLUG_CPU
 792 exit_driver_unreg:
 793         platform_driver_unregister(&coretemp_driver);
 794 #endif
 795 exit:
 796         return err;
 797 }
 798
 799 static void __exit coretemp_exit(void)
 800 {
 801         struct pdev_entry *p, *n;
 802
 803         unregister_hotcpu_notifier(&coretemp_cpu_notifier);
 804         mutex_lock(&pdev_list_mutex);
 805         list_for_each_entry_safe(p, n, &pdev_list, list) {
 806                 platform_device_unregister(p->pdev);
 807                 list_del(&p->list);
 808                 kfree(p);
 809         }
 810         mutex_unlock(&pdev_list_mutex);
 811         platform_driver_unregister(&coretemp_driver);
 812 }
 813
 814 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
 815 MODULE_DESCRIPTION("Intel Core temperature monitor");
 816 MODULE_LICENSE("GPL");
 817
 818 module_init(coretemp_init)
 819 module_exit(coretemp_exit)