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/smp.h>
38 #include <linux/moduleparam.h>
40 #include <asm/processor.h>
41 #include <asm/cpu_device_id.h>
43 #define DRVNAME "coretemp"
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.
49 static int force_tjmax;
50 module_param_named(tjmax, force_tjmax, int, 0444);
51 MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
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)
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)
65 #define for_each_sibling(i, cpu) for_each_cpu(i, cpu_sibling_mask(cpu))
67 #define for_each_sibling(i, cpu) for (i = 0; false; )
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.
87 unsigned long last_updated;
94 struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
95 char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
96 struct mutex update_lock;
99 /* Platform Data per Physical CPU */
100 struct platform_data {
101 struct device *hwmon_dev;
103 struct temp_data *core_data[MAX_CORE_DATA];
104 struct device_attribute name_attr;
108 struct list_head list;
109 struct platform_device *pdev;
113 static LIST_HEAD(pdev_list);
114 static DEFINE_MUTEX(pdev_list_mutex);
116 static ssize_t show_name(struct device *dev,
117 struct device_attribute *devattr, char *buf)
119 return sprintf(buf, "%s\n", DRVNAME);
122 static ssize_t show_label(struct device *dev,
123 struct device_attribute *devattr, char *buf)
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];
129 if (tdata->is_pkg_data)
130 return sprintf(buf, "Physical id %u\n", pdata->phys_proc_id);
132 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
135 static ssize_t show_crit_alarm(struct device *dev,
136 struct device_attribute *devattr, char *buf)
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];
143 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
145 return sprintf(buf, "%d\n", (eax >> 5) & 1);
148 static ssize_t show_tjmax(struct device *dev,
149 struct device_attribute *devattr, char *buf)
151 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
152 struct platform_data *pdata = dev_get_drvdata(dev);
154 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
157 static ssize_t show_ttarget(struct device *dev,
158 struct device_attribute *devattr, char *buf)
160 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
161 struct platform_data *pdata = dev_get_drvdata(dev);
163 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
166 static ssize_t show_temp(struct device *dev,
167 struct device_attribute *devattr, char *buf)
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];
174 mutex_lock(&tdata->update_lock);
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);
180 /* Check whether the data is valid */
181 if (eax & 0x80000000) {
182 tdata->temp = tdata->tjmax -
183 ((eax >> 16) & 0x7f) * 1000;
186 tdata->last_updated = jiffies;
189 mutex_unlock(&tdata->update_lock);
190 return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
198 static const struct tjmax __cpuinitconst tjmax_table[] = {
199 { "CPU 230", 100000 }, /* Model 0x1c, stepping 2 */
200 { "CPU 330", 125000 }, /* Model 0x1c, stepping 2 */
201 { "CPU CE4110", 110000 }, /* Model 0x1c, stepping 10 Sodaville */
202 { "CPU CE4150", 110000 }, /* Model 0x1c, stepping 10 */
203 { "CPU CE4170", 110000 }, /* Model 0x1c, stepping 10 */
214 static const struct tjmax_model __cpuinitconst tjmax_model_table[] = {
215 { 0x1c, 10, 100000 }, /* D4xx, K4xx, N4xx, D5xx, K5xx, N5xx */
216 { 0x1c, ANY, 90000 }, /* Z5xx, N2xx, possibly others
217 * Note: Also matches 230 and 330,
218 * which are covered by tjmax_table
220 { 0x26, ANY, 90000 }, /* Atom Tunnel Creek (Exx), Lincroft (Z6xx)
221 * Note: TjMax for E6xxT is 110C, but CPU type
222 * is undetectable by software
224 { 0x27, ANY, 90000 }, /* Atom Medfield (Z2460) */
225 { 0x35, ANY, 90000 }, /* Atom Clover Trail/Cloverview (Z2760) */
226 { 0x36, ANY, 100000 }, /* Atom Cedar Trail/Cedarview (N2xxx, D2xxx) */
229 static int __cpuinit adjust_tjmax(struct cpuinfo_x86 *c, u32 id,
232 /* The 100C is default for both mobile and non mobile CPUs */
235 int tjmax_ee = 85000;
241 /* explicit tjmax table entries override heuristics */
242 for (i = 0; i < ARRAY_SIZE(tjmax_table); i++) {
243 if (strstr(c->x86_model_id, tjmax_table[i].id))
244 return tjmax_table[i].tjmax;
247 for (i = 0; i < ARRAY_SIZE(tjmax_model_table); i++) {
248 const struct tjmax_model *tm = &tjmax_model_table[i];
249 if (c->x86_model == tm->model &&
250 (tm->mask == ANY || c->x86_mask == tm->mask))
254 /* Early chips have no MSR for TjMax */
256 if (c->x86_model == 0xf && c->x86_mask < 4)
259 if (c->x86_model > 0xe && usemsr_ee) {
263 * Now we can detect the mobile CPU using Intel provided table
264 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
265 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
267 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
270 "Unable to access MSR 0x17, assuming desktop"
273 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
275 * Trust bit 28 up to Penryn, I could not find any
276 * documentation on that; if you happen to know
277 * someone at Intel please ask
281 /* Platform ID bits 52:50 (EDX starts at bit 32) */
282 platform_id = (edx >> 18) & 0x7;
285 * Mobile Penryn CPU seems to be platform ID 7 or 5
288 if (c->x86_model == 0x17 &&
289 (platform_id == 5 || platform_id == 7)) {
291 * If MSR EE bit is set, set it to 90 degrees C,
292 * otherwise 105 degrees C
301 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
304 "Unable to access MSR 0xEE, for Tjmax, left"
306 } else if (eax & 0x40000000) {
309 } else if (tjmax == 100000) {
311 * If we don't use msr EE it means we are desktop CPU
312 * (with exeception of Atom)
314 dev_warn(dev, "Using relative temperature scale!\n");
320 static int __cpuinit get_tjmax(struct cpuinfo_x86 *c, u32 id,
328 * A new feature of current Intel(R) processors, the
329 * IA32_TEMPERATURE_TARGET contains the TjMax value
331 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
333 if (c->x86_model > 0xe && c->x86_model != 0x1c)
334 dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
336 val = (eax >> 16) & 0xff;
338 * If the TjMax is not plausible, an assumption
342 dev_dbg(dev, "TjMax is %d degrees C\n", val);
348 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
350 return force_tjmax * 1000;
354 * An assumption is made for early CPUs and unreadable MSR.
355 * NOTE: the calculated value may not be correct.
357 return adjust_tjmax(c, id, dev);
360 static int create_name_attr(struct platform_data *pdata,
363 sysfs_attr_init(&pdata->name_attr.attr);
364 pdata->name_attr.attr.name = "name";
365 pdata->name_attr.attr.mode = S_IRUGO;
366 pdata->name_attr.show = show_name;
367 return device_create_file(dev, &pdata->name_attr);
370 static int __cpuinit create_core_attrs(struct temp_data *tdata,
371 struct device *dev, int attr_no)
374 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
375 struct device_attribute *devattr, char *buf) = {
376 show_label, show_crit_alarm, show_temp, show_tjmax,
378 static const char *const names[TOTAL_ATTRS] = {
379 "temp%d_label", "temp%d_crit_alarm",
380 "temp%d_input", "temp%d_crit",
383 for (i = 0; i < tdata->attr_size; i++) {
384 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH, names[i],
386 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
387 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
388 tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
389 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
390 tdata->sd_attrs[i].index = attr_no;
391 err = device_create_file(dev, &tdata->sd_attrs[i].dev_attr);
399 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
404 static int __cpuinit chk_ucode_version(unsigned int cpu)
406 struct cpuinfo_x86 *c = &cpu_data(cpu);
409 * Check if we have problem with errata AE18 of Core processors:
410 * Readings might stop update when processor visited too deep sleep,
411 * fixed for stepping D0 (6EC).
413 if (c->x86_model == 0xe && c->x86_mask < 0xc && c->microcode < 0x39) {
414 pr_err("Errata AE18 not fixed, update BIOS or "
415 "microcode of the CPU!\n");
421 static struct platform_device __cpuinit *coretemp_get_pdev(unsigned int cpu)
423 u16 phys_proc_id = TO_PHYS_ID(cpu);
424 struct pdev_entry *p;
426 mutex_lock(&pdev_list_mutex);
428 list_for_each_entry(p, &pdev_list, list)
429 if (p->phys_proc_id == phys_proc_id) {
430 mutex_unlock(&pdev_list_mutex);
434 mutex_unlock(&pdev_list_mutex);
438 static struct temp_data __cpuinit *init_temp_data(unsigned int cpu,
441 struct temp_data *tdata;
443 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
447 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
448 MSR_IA32_THERM_STATUS;
449 tdata->is_pkg_data = pkg_flag;
451 tdata->cpu_core_id = TO_CORE_ID(cpu);
452 tdata->attr_size = MAX_CORE_ATTRS;
453 mutex_init(&tdata->update_lock);
457 static int __cpuinit create_core_data(struct platform_device *pdev,
458 unsigned int cpu, int pkg_flag)
460 struct temp_data *tdata;
461 struct platform_data *pdata = platform_get_drvdata(pdev);
462 struct cpuinfo_x86 *c = &cpu_data(cpu);
467 * Find attr number for sysfs:
468 * We map the attr number to core id of the CPU
469 * The attr number is always core id + 2
470 * The Pkgtemp will always show up as temp1_*, if available
472 attr_no = pkg_flag ? 1 : TO_ATTR_NO(cpu);
474 if (attr_no > MAX_CORE_DATA - 1)
478 * Provide a single set of attributes for all HT siblings of a core
479 * to avoid duplicate sensors (the processor ID and core ID of all
480 * HT siblings of a core are the same).
481 * Skip if a HT sibling of this core is already registered.
482 * This is not an error.
484 if (pdata->core_data[attr_no] != NULL)
487 tdata = init_temp_data(cpu, pkg_flag);
491 /* Test if we can access the status register */
492 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
496 /* We can access status register. Get Critical Temperature */
497 tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
500 * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
501 * The target temperature is available on older CPUs but not in this
502 * register. Atoms don't have the register at all.
504 if (c->x86_model > 0xe && c->x86_model != 0x1c) {
505 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
509 = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
514 pdata->core_data[attr_no] = tdata;
516 /* Create sysfs interfaces */
517 err = create_core_attrs(tdata, &pdev->dev, attr_no);
523 pdata->core_data[attr_no] = NULL;
528 static void __cpuinit coretemp_add_core(unsigned int cpu, int pkg_flag)
530 struct platform_device *pdev = coretemp_get_pdev(cpu);
536 err = create_core_data(pdev, cpu, pkg_flag);
538 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
541 static void coretemp_remove_core(struct platform_data *pdata,
542 struct device *dev, int indx)
545 struct temp_data *tdata = pdata->core_data[indx];
547 /* Remove the sysfs attributes */
548 for (i = 0; i < tdata->attr_size; i++)
549 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
551 kfree(pdata->core_data[indx]);
552 pdata->core_data[indx] = NULL;
555 static int coretemp_probe(struct platform_device *pdev)
557 struct platform_data *pdata;
560 /* Initialize the per-package data structures */
561 pdata = kzalloc(sizeof(struct platform_data), GFP_KERNEL);
565 err = create_name_attr(pdata, &pdev->dev);
569 pdata->phys_proc_id = pdev->id;
570 platform_set_drvdata(pdev, pdata);
572 pdata->hwmon_dev = hwmon_device_register(&pdev->dev);
573 if (IS_ERR(pdata->hwmon_dev)) {
574 err = PTR_ERR(pdata->hwmon_dev);
575 dev_err(&pdev->dev, "Class registration failed (%d)\n", err);
581 device_remove_file(&pdev->dev, &pdata->name_attr);
582 platform_set_drvdata(pdev, NULL);
588 static int coretemp_remove(struct platform_device *pdev)
590 struct platform_data *pdata = platform_get_drvdata(pdev);
593 for (i = MAX_CORE_DATA - 1; i >= 0; --i)
594 if (pdata->core_data[i])
595 coretemp_remove_core(pdata, &pdev->dev, i);
597 device_remove_file(&pdev->dev, &pdata->name_attr);
598 hwmon_device_unregister(pdata->hwmon_dev);
599 platform_set_drvdata(pdev, NULL);
604 static struct platform_driver coretemp_driver = {
606 .owner = THIS_MODULE,
609 .probe = coretemp_probe,
610 .remove = coretemp_remove,
613 static int __cpuinit coretemp_device_add(unsigned int cpu)
616 struct platform_device *pdev;
617 struct pdev_entry *pdev_entry;
619 mutex_lock(&pdev_list_mutex);
621 pdev = platform_device_alloc(DRVNAME, TO_PHYS_ID(cpu));
624 pr_err("Device allocation failed\n");
628 pdev_entry = kzalloc(sizeof(struct pdev_entry), GFP_KERNEL);
631 goto exit_device_put;
634 err = platform_device_add(pdev);
636 pr_err("Device addition failed (%d)\n", err);
637 goto exit_device_free;
640 pdev_entry->pdev = pdev;
641 pdev_entry->phys_proc_id = pdev->id;
643 list_add_tail(&pdev_entry->list, &pdev_list);
644 mutex_unlock(&pdev_list_mutex);
651 platform_device_put(pdev);
653 mutex_unlock(&pdev_list_mutex);
657 static void __cpuinit coretemp_device_remove(unsigned int cpu)
659 struct pdev_entry *p, *n;
660 u16 phys_proc_id = TO_PHYS_ID(cpu);
662 mutex_lock(&pdev_list_mutex);
663 list_for_each_entry_safe(p, n, &pdev_list, list) {
664 if (p->phys_proc_id != phys_proc_id)
666 platform_device_unregister(p->pdev);
670 mutex_unlock(&pdev_list_mutex);
673 static bool __cpuinit is_any_core_online(struct platform_data *pdata)
677 /* Find online cores, except pkgtemp data */
678 for (i = MAX_CORE_DATA - 1; i >= 0; --i) {
679 if (pdata->core_data[i] &&
680 !pdata->core_data[i]->is_pkg_data) {
687 static void __cpuinit get_core_online(unsigned int cpu)
689 struct cpuinfo_x86 *c = &cpu_data(cpu);
690 struct platform_device *pdev = coretemp_get_pdev(cpu);
694 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
695 * sensors. We check this bit only, all the early CPUs
696 * without thermal sensors will be filtered out.
698 if (!cpu_has(c, X86_FEATURE_DTHERM))
702 /* Check the microcode version of the CPU */
703 if (chk_ucode_version(cpu))
707 * Alright, we have DTS support.
708 * We are bringing the _first_ core in this pkg
709 * online. So, initialize per-pkg data structures and
710 * then bring this core online.
712 err = coretemp_device_add(cpu);
716 * Check whether pkgtemp support is available.
717 * If so, add interfaces for pkgtemp.
719 if (cpu_has(c, X86_FEATURE_PTS))
720 coretemp_add_core(cpu, 1);
723 * Physical CPU device already exists.
724 * So, just add interfaces for this core.
726 coretemp_add_core(cpu, 0);
729 static void __cpuinit put_core_offline(unsigned int cpu)
732 struct platform_data *pdata;
733 struct platform_device *pdev = coretemp_get_pdev(cpu);
735 /* If the physical CPU device does not exist, just return */
739 pdata = platform_get_drvdata(pdev);
741 indx = TO_ATTR_NO(cpu);
743 /* The core id is too big, just return */
744 if (indx > MAX_CORE_DATA - 1)
747 if (pdata->core_data[indx] && pdata->core_data[indx]->cpu == cpu)
748 coretemp_remove_core(pdata, &pdev->dev, indx);
751 * If a HT sibling of a core is taken offline, but another HT sibling
752 * of the same core is still online, register the alternate sibling.
753 * This ensures that exactly one set of attributes is provided as long
754 * as at least one HT sibling of a core is online.
756 for_each_sibling(i, cpu) {
760 * Display temperature sensor data for one HT sibling
761 * per core only, so abort the loop after one such
762 * sibling has been found.
768 * If all cores in this pkg are offline, remove the device.
769 * coretemp_device_remove calls unregister_platform_device,
770 * which in turn calls coretemp_remove. This removes the
771 * pkgtemp entry and does other clean ups.
773 if (!is_any_core_online(pdata))
774 coretemp_device_remove(cpu);
777 static int __cpuinit coretemp_cpu_callback(struct notifier_block *nfb,
778 unsigned long action, void *hcpu)
780 unsigned int cpu = (unsigned long) hcpu;
784 case CPU_DOWN_FAILED:
785 get_core_online(cpu);
787 case CPU_DOWN_PREPARE:
788 put_core_offline(cpu);
794 static struct notifier_block coretemp_cpu_notifier __refdata = {
795 .notifier_call = coretemp_cpu_callback,
798 static const struct x86_cpu_id __initconst coretemp_ids[] = {
799 { X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_DTHERM },
802 MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
804 static int __init coretemp_init(void)
809 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
810 * sensors. We check this bit only, all the early CPUs
811 * without thermal sensors will be filtered out.
813 if (!x86_match_cpu(coretemp_ids))
816 err = platform_driver_register(&coretemp_driver);
821 for_each_online_cpu(i)
824 #ifndef CONFIG_HOTPLUG_CPU
825 if (list_empty(&pdev_list)) {
828 goto exit_driver_unreg;
832 register_hotcpu_notifier(&coretemp_cpu_notifier);
836 #ifndef CONFIG_HOTPLUG_CPU
838 platform_driver_unregister(&coretemp_driver);
844 static void __exit coretemp_exit(void)
846 struct pdev_entry *p, *n;
849 unregister_hotcpu_notifier(&coretemp_cpu_notifier);
850 mutex_lock(&pdev_list_mutex);
851 list_for_each_entry_safe(p, n, &pdev_list, list) {
852 platform_device_unregister(p->pdev);
856 mutex_unlock(&pdev_list_mutex);
858 platform_driver_unregister(&coretemp_driver);
861 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
862 MODULE_DESCRIPTION("Intel Core temperature monitor");
863 MODULE_LICENSE("GPL");
865 module_init(coretemp_init)
866 module_exit(coretemp_exit)