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>
39 #include <linux/pci.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 128 /* Number of Real cores per cpu */
56 #define CORETEMP_NAME_LENGTH 19 /* 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) \
67 for_each_cpu(i, topology_sibling_cpumask(cpu))
69 #define for_each_sibling(i, cpu) for (i = 0; false; )
73 * Per-Core Temperature Data
74 * @last_updated: The time when the current temperature value was updated
75 * earlier (in jiffies).
76 * @cpu_core_id: The CPU Core from which temperature values should be read
77 * This value is passed as "id" field to rdmsr/wrmsr functions.
78 * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
79 * from where the temperature values should be read.
80 * @attr_size: Total number of pre-core attrs displayed in the sysfs.
81 * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
82 * Otherwise, temp_data holds coretemp data.
83 * @valid: If this is 1, the current temperature is valid.
89 unsigned long last_updated;
96 struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
97 char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
98 struct attribute *attrs[TOTAL_ATTRS + 1];
99 struct attribute_group attr_group;
100 struct mutex update_lock;
103 /* Platform Data per Physical CPU */
104 struct platform_data {
105 struct device *hwmon_dev;
107 struct temp_data *core_data[MAX_CORE_DATA];
108 struct device_attribute name_attr;
112 struct list_head list;
113 struct platform_device *pdev;
117 static LIST_HEAD(pdev_list);
118 static DEFINE_MUTEX(pdev_list_mutex);
120 static ssize_t show_label(struct device *dev,
121 struct device_attribute *devattr, char *buf)
123 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
124 struct platform_data *pdata = dev_get_drvdata(dev);
125 struct temp_data *tdata = pdata->core_data[attr->index];
127 if (tdata->is_pkg_data)
128 return sprintf(buf, "Physical id %u\n", pdata->phys_proc_id);
130 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
133 static ssize_t show_crit_alarm(struct device *dev,
134 struct device_attribute *devattr, char *buf)
137 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
138 struct platform_data *pdata = dev_get_drvdata(dev);
139 struct temp_data *tdata = pdata->core_data[attr->index];
141 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
143 return sprintf(buf, "%d\n", (eax >> 5) & 1);
146 static ssize_t show_tjmax(struct device *dev,
147 struct device_attribute *devattr, char *buf)
149 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
150 struct platform_data *pdata = dev_get_drvdata(dev);
152 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
155 static ssize_t show_ttarget(struct device *dev,
156 struct device_attribute *devattr, char *buf)
158 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
159 struct platform_data *pdata = dev_get_drvdata(dev);
161 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
164 static ssize_t show_temp(struct device *dev,
165 struct device_attribute *devattr, char *buf)
168 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
169 struct platform_data *pdata = dev_get_drvdata(dev);
170 struct temp_data *tdata = pdata->core_data[attr->index];
172 mutex_lock(&tdata->update_lock);
174 /* Check whether the time interval has elapsed */
175 if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
176 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
178 * Ignore the valid bit. In all observed cases the register
179 * value is either low or zero if the valid bit is 0.
180 * Return it instead of reporting an error which doesn't
181 * really help at all.
183 tdata->temp = tdata->tjmax - ((eax >> 16) & 0x7f) * 1000;
185 tdata->last_updated = jiffies;
188 mutex_unlock(&tdata->update_lock);
189 return sprintf(buf, "%d\n", tdata->temp);
197 static const struct tjmax_pci tjmax_pci_table[] = {
198 { 0x0708, 110000 }, /* CE41x0 (Sodaville ) */
199 { 0x0c72, 102000 }, /* Atom S1240 (Centerton) */
200 { 0x0c73, 95000 }, /* Atom S1220 (Centerton) */
201 { 0x0c75, 95000 }, /* Atom S1260 (Centerton) */
209 static const struct tjmax tjmax_table[] = {
210 { "CPU 230", 100000 }, /* Model 0x1c, stepping 2 */
211 { "CPU 330", 125000 }, /* Model 0x1c, stepping 2 */
222 static const struct tjmax_model tjmax_model_table[] = {
223 { 0x1c, 10, 100000 }, /* D4xx, K4xx, N4xx, D5xx, K5xx, N5xx */
224 { 0x1c, ANY, 90000 }, /* Z5xx, N2xx, possibly others
225 * Note: Also matches 230 and 330,
226 * which are covered by tjmax_table
228 { 0x26, ANY, 90000 }, /* Atom Tunnel Creek (Exx), Lincroft (Z6xx)
229 * Note: TjMax for E6xxT is 110C, but CPU type
230 * is undetectable by software
232 { 0x27, ANY, 90000 }, /* Atom Medfield (Z2460) */
233 { 0x35, ANY, 90000 }, /* Atom Clover Trail/Cloverview (Z27x0) */
234 { 0x36, ANY, 100000 }, /* Atom Cedar Trail/Cedarview (N2xxx, D2xxx)
235 * Also matches S12x0 (stepping 9), covered by
240 static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
242 /* The 100C is default for both mobile and non mobile CPUs */
245 int tjmax_ee = 85000;
250 struct pci_dev *host_bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));
253 * Explicit tjmax table entries override heuristics.
254 * First try PCI host bridge IDs, followed by model ID strings
255 * and model/stepping information.
257 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL) {
258 for (i = 0; i < ARRAY_SIZE(tjmax_pci_table); i++) {
259 if (host_bridge->device == tjmax_pci_table[i].device)
260 return tjmax_pci_table[i].tjmax;
264 for (i = 0; i < ARRAY_SIZE(tjmax_table); i++) {
265 if (strstr(c->x86_model_id, tjmax_table[i].id))
266 return tjmax_table[i].tjmax;
269 for (i = 0; i < ARRAY_SIZE(tjmax_model_table); i++) {
270 const struct tjmax_model *tm = &tjmax_model_table[i];
271 if (c->x86_model == tm->model &&
272 (tm->mask == ANY || c->x86_stepping == tm->mask))
276 /* Early chips have no MSR for TjMax */
278 if (c->x86_model == 0xf && c->x86_stepping < 4)
281 if (c->x86_model > 0xe && usemsr_ee) {
285 * Now we can detect the mobile CPU using Intel provided table
286 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
287 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
289 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
292 "Unable to access MSR 0x17, assuming desktop"
295 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
297 * Trust bit 28 up to Penryn, I could not find any
298 * documentation on that; if you happen to know
299 * someone at Intel please ask
303 /* Platform ID bits 52:50 (EDX starts at bit 32) */
304 platform_id = (edx >> 18) & 0x7;
307 * Mobile Penryn CPU seems to be platform ID 7 or 5
310 if (c->x86_model == 0x17 &&
311 (platform_id == 5 || platform_id == 7)) {
313 * If MSR EE bit is set, set it to 90 degrees C,
314 * otherwise 105 degrees C
323 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
326 "Unable to access MSR 0xEE, for Tjmax, left"
328 } else if (eax & 0x40000000) {
331 } else if (tjmax == 100000) {
333 * If we don't use msr EE it means we are desktop CPU
334 * (with exeception of Atom)
336 dev_warn(dev, "Using relative temperature scale!\n");
342 static bool cpu_has_tjmax(struct cpuinfo_x86 *c)
344 u8 model = c->x86_model;
346 return model > 0xe &&
354 static int get_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
361 * A new feature of current Intel(R) processors, the
362 * IA32_TEMPERATURE_TARGET contains the TjMax value
364 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
366 if (cpu_has_tjmax(c))
367 dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
369 val = (eax >> 16) & 0xff;
371 * If the TjMax is not plausible, an assumption
375 dev_dbg(dev, "TjMax is %d degrees C\n", val);
381 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
383 return force_tjmax * 1000;
387 * An assumption is made for early CPUs and unreadable MSR.
388 * NOTE: the calculated value may not be correct.
390 return adjust_tjmax(c, id, dev);
393 static int create_core_attrs(struct temp_data *tdata, struct device *dev,
397 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
398 struct device_attribute *devattr, char *buf) = {
399 show_label, show_crit_alarm, show_temp, show_tjmax,
401 static const char *const suffixes[TOTAL_ATTRS] = {
402 "label", "crit_alarm", "input", "crit", "max"
405 for (i = 0; i < tdata->attr_size; i++) {
406 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH,
407 "temp%d_%s", attr_no, suffixes[i]);
408 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
409 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
410 tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
411 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
412 tdata->sd_attrs[i].index = attr_no;
413 tdata->attrs[i] = &tdata->sd_attrs[i].dev_attr.attr;
415 tdata->attr_group.attrs = tdata->attrs;
416 return sysfs_create_group(&dev->kobj, &tdata->attr_group);
420 static int chk_ucode_version(unsigned int cpu)
422 struct cpuinfo_x86 *c = &cpu_data(cpu);
425 * Check if we have problem with errata AE18 of Core processors:
426 * Readings might stop update when processor visited too deep sleep,
427 * fixed for stepping D0 (6EC).
429 if (c->x86_model == 0xe && c->x86_stepping < 0xc && c->microcode < 0x39) {
430 pr_err("Errata AE18 not fixed/*(DEBLOBBED)*/\n");
436 static struct platform_device *coretemp_get_pdev(unsigned int cpu)
438 u16 phys_proc_id = TO_PHYS_ID(cpu);
439 struct pdev_entry *p;
441 mutex_lock(&pdev_list_mutex);
443 list_for_each_entry(p, &pdev_list, list)
444 if (p->phys_proc_id == phys_proc_id) {
445 mutex_unlock(&pdev_list_mutex);
449 mutex_unlock(&pdev_list_mutex);
453 static struct temp_data *init_temp_data(unsigned int cpu, int pkg_flag)
455 struct temp_data *tdata;
457 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
461 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
462 MSR_IA32_THERM_STATUS;
463 tdata->is_pkg_data = pkg_flag;
465 tdata->cpu_core_id = TO_CORE_ID(cpu);
466 tdata->attr_size = MAX_CORE_ATTRS;
467 mutex_init(&tdata->update_lock);
471 static int create_core_data(struct platform_device *pdev, unsigned int cpu,
474 struct temp_data *tdata;
475 struct platform_data *pdata = platform_get_drvdata(pdev);
476 struct cpuinfo_x86 *c = &cpu_data(cpu);
481 * Find attr number for sysfs:
482 * We map the attr number to core id of the CPU
483 * The attr number is always core id + 2
484 * The Pkgtemp will always show up as temp1_*, if available
486 attr_no = pkg_flag ? 1 : TO_ATTR_NO(cpu);
488 if (attr_no > MAX_CORE_DATA - 1)
492 * Provide a single set of attributes for all HT siblings of a core
493 * to avoid duplicate sensors (the processor ID and core ID of all
494 * HT siblings of a core are the same).
495 * Skip if a HT sibling of this core is already registered.
496 * This is not an error.
498 if (pdata->core_data[attr_no] != NULL)
501 tdata = init_temp_data(cpu, pkg_flag);
505 /* Test if we can access the status register */
506 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
510 /* We can access status register. Get Critical Temperature */
511 tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
514 * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
515 * The target temperature is available on older CPUs but not in this
516 * register. Atoms don't have the register at all.
518 if (c->x86_model > 0xe && c->x86_model != 0x1c) {
519 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
523 = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
528 pdata->core_data[attr_no] = tdata;
530 /* Create sysfs interfaces */
531 err = create_core_attrs(tdata, pdata->hwmon_dev, attr_no);
537 pdata->core_data[attr_no] = NULL;
542 static void coretemp_add_core(unsigned int cpu, int pkg_flag)
544 struct platform_device *pdev = coretemp_get_pdev(cpu);
550 err = create_core_data(pdev, cpu, pkg_flag);
552 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
555 static void coretemp_remove_core(struct platform_data *pdata,
558 struct temp_data *tdata = pdata->core_data[indx];
560 /* Remove the sysfs attributes */
561 sysfs_remove_group(&pdata->hwmon_dev->kobj, &tdata->attr_group);
563 kfree(pdata->core_data[indx]);
564 pdata->core_data[indx] = NULL;
567 static int coretemp_probe(struct platform_device *pdev)
569 struct device *dev = &pdev->dev;
570 struct platform_data *pdata;
572 /* Initialize the per-package data structures */
573 pdata = devm_kzalloc(dev, sizeof(struct platform_data), GFP_KERNEL);
577 pdata->phys_proc_id = pdev->id;
578 platform_set_drvdata(pdev, pdata);
580 pdata->hwmon_dev = devm_hwmon_device_register_with_groups(dev, DRVNAME,
582 return PTR_ERR_OR_ZERO(pdata->hwmon_dev);
585 static int coretemp_remove(struct platform_device *pdev)
587 struct platform_data *pdata = platform_get_drvdata(pdev);
590 for (i = MAX_CORE_DATA - 1; i >= 0; --i)
591 if (pdata->core_data[i])
592 coretemp_remove_core(pdata, i);
597 static struct platform_driver coretemp_driver = {
601 .probe = coretemp_probe,
602 .remove = coretemp_remove,
605 static int coretemp_device_add(unsigned int cpu)
608 struct platform_device *pdev;
609 struct pdev_entry *pdev_entry;
611 mutex_lock(&pdev_list_mutex);
613 pdev = platform_device_alloc(DRVNAME, TO_PHYS_ID(cpu));
616 pr_err("Device allocation failed\n");
620 pdev_entry = kzalloc(sizeof(struct pdev_entry), GFP_KERNEL);
623 goto exit_device_put;
626 err = platform_device_add(pdev);
628 pr_err("Device addition failed (%d)\n", err);
629 goto exit_device_free;
632 pdev_entry->pdev = pdev;
633 pdev_entry->phys_proc_id = pdev->id;
635 list_add_tail(&pdev_entry->list, &pdev_list);
636 mutex_unlock(&pdev_list_mutex);
643 platform_device_put(pdev);
645 mutex_unlock(&pdev_list_mutex);
649 static void coretemp_device_remove(unsigned int cpu)
651 struct pdev_entry *p, *n;
652 u16 phys_proc_id = TO_PHYS_ID(cpu);
654 mutex_lock(&pdev_list_mutex);
655 list_for_each_entry_safe(p, n, &pdev_list, list) {
656 if (p->phys_proc_id != phys_proc_id)
658 platform_device_unregister(p->pdev);
662 mutex_unlock(&pdev_list_mutex);
665 static bool is_any_core_online(struct platform_data *pdata)
669 /* Find online cores, except pkgtemp data */
670 for (i = MAX_CORE_DATA - 1; i >= 0; --i) {
671 if (pdata->core_data[i] &&
672 !pdata->core_data[i]->is_pkg_data) {
679 static void get_core_online(unsigned int cpu)
681 struct cpuinfo_x86 *c = &cpu_data(cpu);
682 struct platform_device *pdev = coretemp_get_pdev(cpu);
686 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
687 * sensors. We check this bit only, all the early CPUs
688 * without thermal sensors will be filtered out.
690 if (!cpu_has(c, X86_FEATURE_DTHERM))
694 /* Check the microcode version of the CPU */
695 if (chk_ucode_version(cpu))
699 * Alright, we have DTS support.
700 * We are bringing the _first_ core in this pkg
701 * online. So, initialize per-pkg data structures and
702 * then bring this core online.
704 err = coretemp_device_add(cpu);
708 * Check whether pkgtemp support is available.
709 * If so, add interfaces for pkgtemp.
711 if (cpu_has(c, X86_FEATURE_PTS))
712 coretemp_add_core(cpu, 1);
715 * Physical CPU device already exists.
716 * So, just add interfaces for this core.
718 coretemp_add_core(cpu, 0);
721 static void put_core_offline(unsigned int cpu)
724 struct platform_data *pdata;
725 struct platform_device *pdev = coretemp_get_pdev(cpu);
727 /* If the physical CPU device does not exist, just return */
731 pdata = platform_get_drvdata(pdev);
733 indx = TO_ATTR_NO(cpu);
735 /* The core id is too big, just return */
736 if (indx > MAX_CORE_DATA - 1)
739 if (pdata->core_data[indx] && pdata->core_data[indx]->cpu == cpu)
740 coretemp_remove_core(pdata, indx);
743 * If a HT sibling of a core is taken offline, but another HT sibling
744 * of the same core is still online, register the alternate sibling.
745 * This ensures that exactly one set of attributes is provided as long
746 * as at least one HT sibling of a core is online.
748 for_each_sibling(i, cpu) {
752 * Display temperature sensor data for one HT sibling
753 * per core only, so abort the loop after one such
754 * sibling has been found.
760 * If all cores in this pkg are offline, remove the device.
761 * coretemp_device_remove calls unregister_platform_device,
762 * which in turn calls coretemp_remove. This removes the
763 * pkgtemp entry and does other clean ups.
765 if (!is_any_core_online(pdata))
766 coretemp_device_remove(cpu);
769 static int coretemp_cpu_callback(struct notifier_block *nfb,
770 unsigned long action, void *hcpu)
772 unsigned int cpu = (unsigned long) hcpu;
776 case CPU_DOWN_FAILED:
777 get_core_online(cpu);
779 case CPU_DOWN_PREPARE:
780 put_core_offline(cpu);
786 static struct notifier_block coretemp_cpu_notifier __refdata = {
787 .notifier_call = coretemp_cpu_callback,
790 static const struct x86_cpu_id __initconst coretemp_ids[] = {
791 { X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_DTHERM },
794 MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
796 static int __init coretemp_init(void)
801 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
802 * sensors. We check this bit only, all the early CPUs
803 * without thermal sensors will be filtered out.
805 if (!x86_match_cpu(coretemp_ids))
808 err = platform_driver_register(&coretemp_driver);
812 cpu_notifier_register_begin();
813 for_each_online_cpu(i)
816 #ifndef CONFIG_HOTPLUG_CPU
817 if (list_empty(&pdev_list)) {
818 cpu_notifier_register_done();
820 goto exit_driver_unreg;
824 __register_hotcpu_notifier(&coretemp_cpu_notifier);
825 cpu_notifier_register_done();
828 #ifndef CONFIG_HOTPLUG_CPU
830 platform_driver_unregister(&coretemp_driver);
836 static void __exit coretemp_exit(void)
838 struct pdev_entry *p, *n;
840 cpu_notifier_register_begin();
841 __unregister_hotcpu_notifier(&coretemp_cpu_notifier);
842 mutex_lock(&pdev_list_mutex);
843 list_for_each_entry_safe(p, n, &pdev_list, list) {
844 platform_device_unregister(p->pdev);
848 mutex_unlock(&pdev_list_mutex);
849 cpu_notifier_register_done();
850 platform_driver_unregister(&coretemp_driver);
853 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
854 MODULE_DESCRIPTION("Intel Core temperature monitor");
855 MODULE_LICENSE("GPL");
857 module_init(coretemp_init)
858 module_exit(coretemp_exit)