2 * coretemp.c - Linux kernel module for hardware monitoring
4 * Copyright (C) 2007 Rudolf Marek <r.marek@assembler.cz>
6 * Inspired from many hwmon drivers
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.
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.
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
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
41 #include <asm/processor.h>
42 #include <asm/cpu_device_id.h>
44 #define DRVNAME "coretemp"
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.
50 static int force_tjmax;
51 module_param_named(tjmax, force_tjmax, int, 0444);
52 MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
54 #define BASE_SYSFS_ATTR_NO 2 /* Sysfs Base attr no for coretemp */
55 #define NUM_REAL_CORES 16 /* 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)
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)
66 #define for_each_sibling(i, cpu) for_each_cpu(i, cpu_sibling_mask(cpu))
68 #define for_each_sibling(i, cpu) for (i = 0; false; )
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.
88 unsigned long last_updated;
95 struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
96 char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
97 struct mutex update_lock;
100 /* Platform Data per Physical CPU */
101 struct platform_data {
102 struct device *hwmon_dev;
104 struct temp_data *core_data[MAX_CORE_DATA];
105 struct device_attribute name_attr;
109 struct list_head list;
110 struct platform_device *pdev;
114 static LIST_HEAD(pdev_list);
115 static DEFINE_MUTEX(pdev_list_mutex);
117 static ssize_t show_name(struct device *dev,
118 struct device_attribute *devattr, char *buf)
120 return sprintf(buf, "%s\n", DRVNAME);
123 static ssize_t show_label(struct device *dev,
124 struct device_attribute *devattr, char *buf)
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];
130 if (tdata->is_pkg_data)
131 return sprintf(buf, "Physical id %u\n", pdata->phys_proc_id);
133 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
136 static ssize_t show_crit_alarm(struct device *dev,
137 struct device_attribute *devattr, char *buf)
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];
144 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
146 return sprintf(buf, "%d\n", (eax >> 5) & 1);
149 static ssize_t show_tjmax(struct device *dev,
150 struct device_attribute *devattr, char *buf)
152 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
153 struct platform_data *pdata = dev_get_drvdata(dev);
155 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
158 static ssize_t show_ttarget(struct device *dev,
159 struct device_attribute *devattr, char *buf)
161 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
162 struct platform_data *pdata = dev_get_drvdata(dev);
164 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
167 static ssize_t show_temp(struct device *dev,
168 struct device_attribute *devattr, char *buf)
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];
175 mutex_lock(&tdata->update_lock);
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);
181 /* Check whether the data is valid */
182 if (eax & 0x80000000) {
183 tdata->temp = tdata->tjmax -
184 ((eax >> 16) & 0x7f) * 1000;
187 tdata->last_updated = jiffies;
190 mutex_unlock(&tdata->update_lock);
191 return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
194 static int __cpuinit adjust_tjmax(struct cpuinfo_x86 *c, u32 id,
197 /* The 100C is default for both mobile and non mobile CPUs */
200 int tjmax_ee = 85000;
204 struct pci_dev *host_bridge;
206 /* Early chips have no MSR for TjMax */
208 if (c->x86_model == 0xf && c->x86_mask < 4)
213 if (c->x86_model == 0x1c) {
216 host_bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));
218 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL
219 && (host_bridge->device == 0xa000 /* NM10 based nettop */
220 || host_bridge->device == 0xa010)) /* NM10 based netbook */
225 pci_dev_put(host_bridge);
228 if (c->x86_model > 0xe && usemsr_ee) {
232 * Now we can detect the mobile CPU using Intel provided table
233 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
234 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
236 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
239 "Unable to access MSR 0x17, assuming desktop"
242 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
244 * Trust bit 28 up to Penryn, I could not find any
245 * documentation on that; if you happen to know
246 * someone at Intel please ask
250 /* Platform ID bits 52:50 (EDX starts at bit 32) */
251 platform_id = (edx >> 18) & 0x7;
254 * Mobile Penryn CPU seems to be platform ID 7 or 5
257 if (c->x86_model == 0x17 &&
258 (platform_id == 5 || platform_id == 7)) {
260 * If MSR EE bit is set, set it to 90 degrees C,
261 * otherwise 105 degrees C
270 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
273 "Unable to access MSR 0xEE, for Tjmax, left"
275 } else if (eax & 0x40000000) {
278 } else if (tjmax == 100000) {
280 * If we don't use msr EE it means we are desktop CPU
281 * (with exeception of Atom)
283 dev_warn(dev, "Using relative temperature scale!\n");
289 static int __cpuinit get_tjmax(struct cpuinfo_x86 *c, u32 id,
297 * A new feature of current Intel(R) processors, the
298 * IA32_TEMPERATURE_TARGET contains the TjMax value
300 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
302 if (c->x86_model > 0xe && c->x86_model != 0x1c)
303 dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
305 val = (eax >> 16) & 0xff;
307 * If the TjMax is not plausible, an assumption
311 dev_dbg(dev, "TjMax is %d degrees C\n", val);
317 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
319 return force_tjmax * 1000;
323 * An assumption is made for early CPUs and unreadable MSR.
324 * NOTE: the calculated value may not be correct.
326 return adjust_tjmax(c, id, dev);
329 static int __devinit create_name_attr(struct platform_data *pdata,
332 sysfs_attr_init(&pdata->name_attr.attr);
333 pdata->name_attr.attr.name = "name";
334 pdata->name_attr.attr.mode = S_IRUGO;
335 pdata->name_attr.show = show_name;
336 return device_create_file(dev, &pdata->name_attr);
339 static int __cpuinit create_core_attrs(struct temp_data *tdata,
340 struct device *dev, int attr_no)
343 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
344 struct device_attribute *devattr, char *buf) = {
345 show_label, show_crit_alarm, show_temp, show_tjmax,
347 static const char *const names[TOTAL_ATTRS] = {
348 "temp%d_label", "temp%d_crit_alarm",
349 "temp%d_input", "temp%d_crit",
352 for (i = 0; i < tdata->attr_size; i++) {
353 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH, names[i],
355 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
356 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
357 tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
358 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
359 tdata->sd_attrs[i].index = attr_no;
360 err = device_create_file(dev, &tdata->sd_attrs[i].dev_attr);
368 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
373 static int __cpuinit chk_ucode_version(unsigned int cpu)
375 struct cpuinfo_x86 *c = &cpu_data(cpu);
378 * Check if we have problem with errata AE18 of Core processors:
379 * Readings might stop update when processor visited too deep sleep,
380 * fixed for stepping D0 (6EC).
382 if (c->x86_model == 0xe && c->x86_mask < 0xc && c->microcode < 0x39) {
383 pr_err("Errata AE18 not fixed, update BIOS or "
384 "microcode of the CPU!\n");
390 static struct platform_device __cpuinit *coretemp_get_pdev(unsigned int cpu)
392 u16 phys_proc_id = TO_PHYS_ID(cpu);
393 struct pdev_entry *p;
395 mutex_lock(&pdev_list_mutex);
397 list_for_each_entry(p, &pdev_list, list)
398 if (p->phys_proc_id == phys_proc_id) {
399 mutex_unlock(&pdev_list_mutex);
403 mutex_unlock(&pdev_list_mutex);
407 static struct temp_data __cpuinit *init_temp_data(unsigned int cpu,
410 struct temp_data *tdata;
412 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
416 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
417 MSR_IA32_THERM_STATUS;
418 tdata->is_pkg_data = pkg_flag;
420 tdata->cpu_core_id = TO_CORE_ID(cpu);
421 tdata->attr_size = MAX_CORE_ATTRS;
422 mutex_init(&tdata->update_lock);
426 static int __cpuinit create_core_data(struct platform_device *pdev,
427 unsigned int cpu, int pkg_flag)
429 struct temp_data *tdata;
430 struct platform_data *pdata = platform_get_drvdata(pdev);
431 struct cpuinfo_x86 *c = &cpu_data(cpu);
436 * Find attr number for sysfs:
437 * We map the attr number to core id of the CPU
438 * The attr number is always core id + 2
439 * The Pkgtemp will always show up as temp1_*, if available
441 attr_no = pkg_flag ? 1 : TO_ATTR_NO(cpu);
443 if (attr_no > MAX_CORE_DATA - 1)
447 * Provide a single set of attributes for all HT siblings of a core
448 * to avoid duplicate sensors (the processor ID and core ID of all
449 * HT siblings of a core are the same).
450 * Skip if a HT sibling of this core is already registered.
451 * This is not an error.
453 if (pdata->core_data[attr_no] != NULL)
456 tdata = init_temp_data(cpu, pkg_flag);
460 /* Test if we can access the status register */
461 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
465 /* We can access status register. Get Critical Temperature */
466 tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
469 * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
470 * The target temperature is available on older CPUs but not in this
471 * register. Atoms don't have the register at all.
473 if (c->x86_model > 0xe && c->x86_model != 0x1c) {
474 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
478 = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
483 pdata->core_data[attr_no] = tdata;
485 /* Create sysfs interfaces */
486 err = create_core_attrs(tdata, &pdev->dev, attr_no);
492 pdata->core_data[attr_no] = NULL;
497 static void __cpuinit coretemp_add_core(unsigned int cpu, int pkg_flag)
499 struct platform_device *pdev = coretemp_get_pdev(cpu);
505 err = create_core_data(pdev, cpu, pkg_flag);
507 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
510 static void coretemp_remove_core(struct platform_data *pdata,
511 struct device *dev, int indx)
514 struct temp_data *tdata = pdata->core_data[indx];
516 /* Remove the sysfs attributes */
517 for (i = 0; i < tdata->attr_size; i++)
518 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
520 kfree(pdata->core_data[indx]);
521 pdata->core_data[indx] = NULL;
524 static int __devinit coretemp_probe(struct platform_device *pdev)
526 struct platform_data *pdata;
529 /* Initialize the per-package data structures */
530 pdata = kzalloc(sizeof(struct platform_data), GFP_KERNEL);
534 err = create_name_attr(pdata, &pdev->dev);
538 pdata->phys_proc_id = pdev->id;
539 platform_set_drvdata(pdev, pdata);
541 pdata->hwmon_dev = hwmon_device_register(&pdev->dev);
542 if (IS_ERR(pdata->hwmon_dev)) {
543 err = PTR_ERR(pdata->hwmon_dev);
544 dev_err(&pdev->dev, "Class registration failed (%d)\n", err);
550 device_remove_file(&pdev->dev, &pdata->name_attr);
551 platform_set_drvdata(pdev, NULL);
557 static int __devexit coretemp_remove(struct platform_device *pdev)
559 struct platform_data *pdata = platform_get_drvdata(pdev);
562 for (i = MAX_CORE_DATA - 1; i >= 0; --i)
563 if (pdata->core_data[i])
564 coretemp_remove_core(pdata, &pdev->dev, i);
566 device_remove_file(&pdev->dev, &pdata->name_attr);
567 hwmon_device_unregister(pdata->hwmon_dev);
568 platform_set_drvdata(pdev, NULL);
573 static struct platform_driver coretemp_driver = {
575 .owner = THIS_MODULE,
578 .probe = coretemp_probe,
579 .remove = __devexit_p(coretemp_remove),
582 static int __cpuinit coretemp_device_add(unsigned int cpu)
585 struct platform_device *pdev;
586 struct pdev_entry *pdev_entry;
588 mutex_lock(&pdev_list_mutex);
590 pdev = platform_device_alloc(DRVNAME, TO_PHYS_ID(cpu));
593 pr_err("Device allocation failed\n");
597 pdev_entry = kzalloc(sizeof(struct pdev_entry), GFP_KERNEL);
600 goto exit_device_put;
603 err = platform_device_add(pdev);
605 pr_err("Device addition failed (%d)\n", err);
606 goto exit_device_free;
609 pdev_entry->pdev = pdev;
610 pdev_entry->phys_proc_id = pdev->id;
612 list_add_tail(&pdev_entry->list, &pdev_list);
613 mutex_unlock(&pdev_list_mutex);
620 platform_device_put(pdev);
622 mutex_unlock(&pdev_list_mutex);
626 static void __cpuinit coretemp_device_remove(unsigned int cpu)
628 struct pdev_entry *p, *n;
629 u16 phys_proc_id = TO_PHYS_ID(cpu);
631 mutex_lock(&pdev_list_mutex);
632 list_for_each_entry_safe(p, n, &pdev_list, list) {
633 if (p->phys_proc_id != phys_proc_id)
635 platform_device_unregister(p->pdev);
639 mutex_unlock(&pdev_list_mutex);
642 static bool __cpuinit is_any_core_online(struct platform_data *pdata)
646 /* Find online cores, except pkgtemp data */
647 for (i = MAX_CORE_DATA - 1; i >= 0; --i) {
648 if (pdata->core_data[i] &&
649 !pdata->core_data[i]->is_pkg_data) {
656 static void __cpuinit get_core_online(unsigned int cpu)
658 struct cpuinfo_x86 *c = &cpu_data(cpu);
659 struct platform_device *pdev = coretemp_get_pdev(cpu);
663 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
664 * sensors. We check this bit only, all the early CPUs
665 * without thermal sensors will be filtered out.
667 if (!cpu_has(c, X86_FEATURE_DTS))
671 /* Check the microcode version of the CPU */
672 if (chk_ucode_version(cpu))
676 * Alright, we have DTS support.
677 * We are bringing the _first_ core in this pkg
678 * online. So, initialize per-pkg data structures and
679 * then bring this core online.
681 err = coretemp_device_add(cpu);
685 * Check whether pkgtemp support is available.
686 * If so, add interfaces for pkgtemp.
688 if (cpu_has(c, X86_FEATURE_PTS))
689 coretemp_add_core(cpu, 1);
692 * Physical CPU device already exists.
693 * So, just add interfaces for this core.
695 coretemp_add_core(cpu, 0);
698 static void __cpuinit put_core_offline(unsigned int cpu)
701 struct platform_data *pdata;
702 struct platform_device *pdev = coretemp_get_pdev(cpu);
704 /* If the physical CPU device does not exist, just return */
708 pdata = platform_get_drvdata(pdev);
710 indx = TO_ATTR_NO(cpu);
712 if (pdata->core_data[indx] && pdata->core_data[indx]->cpu == cpu)
713 coretemp_remove_core(pdata, &pdev->dev, indx);
716 * If a HT sibling of a core is taken offline, but another HT sibling
717 * of the same core is still online, register the alternate sibling.
718 * This ensures that exactly one set of attributes is provided as long
719 * as at least one HT sibling of a core is online.
721 for_each_sibling(i, cpu) {
725 * Display temperature sensor data for one HT sibling
726 * per core only, so abort the loop after one such
727 * sibling has been found.
733 * If all cores in this pkg are offline, remove the device.
734 * coretemp_device_remove calls unregister_platform_device,
735 * which in turn calls coretemp_remove. This removes the
736 * pkgtemp entry and does other clean ups.
738 if (!is_any_core_online(pdata))
739 coretemp_device_remove(cpu);
742 static int __cpuinit coretemp_cpu_callback(struct notifier_block *nfb,
743 unsigned long action, void *hcpu)
745 unsigned int cpu = (unsigned long) hcpu;
749 case CPU_DOWN_FAILED:
750 get_core_online(cpu);
752 case CPU_DOWN_PREPARE:
753 put_core_offline(cpu);
759 static struct notifier_block coretemp_cpu_notifier __refdata = {
760 .notifier_call = coretemp_cpu_callback,
763 static const struct x86_cpu_id coretemp_ids[] = {
764 { X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_DTS },
767 MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
769 static int __init coretemp_init(void)
771 int i, err = -ENODEV;
774 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
775 * sensors. We check this bit only, all the early CPUs
776 * without thermal sensors will be filtered out.
778 if (!x86_match_cpu(coretemp_ids))
781 err = platform_driver_register(&coretemp_driver);
785 for_each_online_cpu(i)
788 #ifndef CONFIG_HOTPLUG_CPU
789 if (list_empty(&pdev_list)) {
791 goto exit_driver_unreg;
795 register_hotcpu_notifier(&coretemp_cpu_notifier);
798 #ifndef CONFIG_HOTPLUG_CPU
800 platform_driver_unregister(&coretemp_driver);
806 static void __exit coretemp_exit(void)
808 struct pdev_entry *p, *n;
810 unregister_hotcpu_notifier(&coretemp_cpu_notifier);
811 mutex_lock(&pdev_list_mutex);
812 list_for_each_entry_safe(p, n, &pdev_list, list) {
813 platform_device_unregister(p->pdev);
817 mutex_unlock(&pdev_list_mutex);
818 platform_driver_unregister(&coretemp_driver);
821 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
822 MODULE_DESCRIPTION("Intel Core temperature monitor");
823 MODULE_LICENSE("GPL");
825 module_init(coretemp_init)
826 module_exit(coretemp_exit)