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 PKG_SYSFS_ATTR_NO 1 /* Sysfs attribute for package temp */
55 #define BASE_SYSFS_ATTR_NO 2 /* Sysfs Base attr no for coretemp */
56 #define NUM_REAL_CORES 128 /* Number of Real cores per cpu */
57 #define CORETEMP_NAME_LENGTH 19 /* String Length of attrs */
58 #define MAX_CORE_ATTRS 4 /* Maximum no of basic attrs */
59 #define TOTAL_ATTRS (MAX_CORE_ATTRS + 1)
60 #define MAX_CORE_DATA (NUM_REAL_CORES + BASE_SYSFS_ATTR_NO)
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 cpumask cpumask;
108 struct temp_data *core_data[MAX_CORE_DATA];
109 struct device_attribute name_attr;
112 /* Keep track of how many package pointers we allocated in init() */
113 static int max_packages __read_mostly;
114 /* Array of package pointers. Serialized by cpu hotplug lock */
115 static struct platform_device **pkg_devices;
117 static ssize_t show_label(struct device *dev,
118 struct device_attribute *devattr, char *buf)
120 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
121 struct platform_data *pdata = dev_get_drvdata(dev);
122 struct temp_data *tdata = pdata->core_data[attr->index];
124 if (tdata->is_pkg_data)
125 return sprintf(buf, "Package id %u\n", pdata->pkg_id);
127 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
130 static ssize_t show_crit_alarm(struct device *dev,
131 struct device_attribute *devattr, char *buf)
134 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
135 struct platform_data *pdata = dev_get_drvdata(dev);
136 struct temp_data *tdata = pdata->core_data[attr->index];
138 mutex_lock(&tdata->update_lock);
139 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
140 mutex_unlock(&tdata->update_lock);
142 return sprintf(buf, "%d\n", (eax >> 5) & 1);
145 static ssize_t show_tjmax(struct device *dev,
146 struct device_attribute *devattr, char *buf)
148 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
149 struct platform_data *pdata = dev_get_drvdata(dev);
151 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
154 static ssize_t show_ttarget(struct device *dev,
155 struct device_attribute *devattr, char *buf)
157 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
158 struct platform_data *pdata = dev_get_drvdata(dev);
160 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
163 static ssize_t show_temp(struct device *dev,
164 struct device_attribute *devattr, char *buf)
167 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
168 struct platform_data *pdata = dev_get_drvdata(dev);
169 struct temp_data *tdata = pdata->core_data[attr->index];
171 mutex_lock(&tdata->update_lock);
173 /* Check whether the time interval has elapsed */
174 if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
175 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
177 * Ignore the valid bit. In all observed cases the register
178 * value is either low or zero if the valid bit is 0.
179 * Return it instead of reporting an error which doesn't
180 * really help at all.
182 tdata->temp = tdata->tjmax - ((eax >> 16) & 0x7f) * 1000;
184 tdata->last_updated = jiffies;
187 mutex_unlock(&tdata->update_lock);
188 return sprintf(buf, "%d\n", tdata->temp);
196 static const struct tjmax_pci tjmax_pci_table[] = {
197 { 0x0708, 110000 }, /* CE41x0 (Sodaville ) */
198 { 0x0c72, 102000 }, /* Atom S1240 (Centerton) */
199 { 0x0c73, 95000 }, /* Atom S1220 (Centerton) */
200 { 0x0c75, 95000 }, /* Atom S1260 (Centerton) */
208 static const struct tjmax tjmax_table[] = {
209 { "CPU 230", 100000 }, /* Model 0x1c, stepping 2 */
210 { "CPU 330", 125000 }, /* Model 0x1c, stepping 2 */
221 static const struct tjmax_model tjmax_model_table[] = {
222 { 0x1c, 10, 100000 }, /* D4xx, K4xx, N4xx, D5xx, K5xx, N5xx */
223 { 0x1c, ANY, 90000 }, /* Z5xx, N2xx, possibly others
224 * Note: Also matches 230 and 330,
225 * which are covered by tjmax_table
227 { 0x26, ANY, 90000 }, /* Atom Tunnel Creek (Exx), Lincroft (Z6xx)
228 * Note: TjMax for E6xxT is 110C, but CPU type
229 * is undetectable by software
231 { 0x27, ANY, 90000 }, /* Atom Medfield (Z2460) */
232 { 0x35, ANY, 90000 }, /* Atom Clover Trail/Cloverview (Z27x0) */
233 { 0x36, ANY, 100000 }, /* Atom Cedar Trail/Cedarview (N2xxx, D2xxx)
234 * Also matches S12x0 (stepping 9), covered by
239 static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
241 /* The 100C is default for both mobile and non mobile CPUs */
244 int tjmax_ee = 85000;
249 struct pci_dev *host_bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));
252 * Explicit tjmax table entries override heuristics.
253 * First try PCI host bridge IDs, followed by model ID strings
254 * and model/stepping information.
256 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL) {
257 for (i = 0; i < ARRAY_SIZE(tjmax_pci_table); i++) {
258 if (host_bridge->device == tjmax_pci_table[i].device) {
259 pci_dev_put(host_bridge);
260 return tjmax_pci_table[i].tjmax;
264 pci_dev_put(host_bridge);
266 for (i = 0; i < ARRAY_SIZE(tjmax_table); i++) {
267 if (strstr(c->x86_model_id, tjmax_table[i].id))
268 return tjmax_table[i].tjmax;
271 for (i = 0; i < ARRAY_SIZE(tjmax_model_table); i++) {
272 const struct tjmax_model *tm = &tjmax_model_table[i];
273 if (c->x86_model == tm->model &&
274 (tm->mask == ANY || c->x86_stepping == tm->mask))
278 /* Early chips have no MSR for TjMax */
280 if (c->x86_model == 0xf && c->x86_stepping < 4)
283 if (c->x86_model > 0xe && usemsr_ee) {
287 * Now we can detect the mobile CPU using Intel provided table
288 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
289 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
291 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
294 "Unable to access MSR 0x17, assuming desktop"
297 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
299 * Trust bit 28 up to Penryn, I could not find any
300 * documentation on that; if you happen to know
301 * someone at Intel please ask
305 /* Platform ID bits 52:50 (EDX starts at bit 32) */
306 platform_id = (edx >> 18) & 0x7;
309 * Mobile Penryn CPU seems to be platform ID 7 or 5
312 if (c->x86_model == 0x17 &&
313 (platform_id == 5 || platform_id == 7)) {
315 * If MSR EE bit is set, set it to 90 degrees C,
316 * otherwise 105 degrees C
325 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
328 "Unable to access MSR 0xEE, for Tjmax, left"
330 } else if (eax & 0x40000000) {
333 } else if (tjmax == 100000) {
335 * If we don't use msr EE it means we are desktop CPU
336 * (with exeception of Atom)
338 dev_warn(dev, "Using relative temperature scale!\n");
344 static bool cpu_has_tjmax(struct cpuinfo_x86 *c)
346 u8 model = c->x86_model;
348 return model > 0xe &&
356 static int get_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
363 * A new feature of current Intel(R) processors, the
364 * IA32_TEMPERATURE_TARGET contains the TjMax value
366 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
368 if (cpu_has_tjmax(c))
369 dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
371 val = (eax >> 16) & 0xff;
373 * If the TjMax is not plausible, an assumption
377 dev_dbg(dev, "TjMax is %d degrees C\n", val);
383 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
385 return force_tjmax * 1000;
389 * An assumption is made for early CPUs and unreadable MSR.
390 * NOTE: the calculated value may not be correct.
392 return adjust_tjmax(c, id, dev);
395 static int create_core_attrs(struct temp_data *tdata, struct device *dev,
399 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
400 struct device_attribute *devattr, char *buf) = {
401 show_label, show_crit_alarm, show_temp, show_tjmax,
403 static const char *const suffixes[TOTAL_ATTRS] = {
404 "label", "crit_alarm", "input", "crit", "max"
407 for (i = 0; i < tdata->attr_size; i++) {
408 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH,
409 "temp%d_%s", attr_no, suffixes[i]);
410 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
411 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
412 tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
413 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
414 tdata->sd_attrs[i].index = attr_no;
415 tdata->attrs[i] = &tdata->sd_attrs[i].dev_attr.attr;
417 tdata->attr_group.attrs = tdata->attrs;
418 return sysfs_create_group(&dev->kobj, &tdata->attr_group);
422 static int chk_ucode_version(unsigned int cpu)
424 struct cpuinfo_x86 *c = &cpu_data(cpu);
427 * Check if we have problem with errata AE18 of Core processors:
428 * Readings might stop update when processor visited too deep sleep,
429 * fixed for stepping D0 (6EC).
431 if (c->x86_model == 0xe && c->x86_stepping < 0xc && c->microcode < 0x39) {
432 pr_err("Errata AE18 not fixed/*(DEBLOBBED)*/\n");
438 static struct platform_device *coretemp_get_pdev(unsigned int cpu)
440 int pkgid = topology_logical_package_id(cpu);
442 if (pkgid >= 0 && pkgid < max_packages)
443 return pkg_devices[pkgid];
447 static struct temp_data *init_temp_data(unsigned int cpu, int pkg_flag)
449 struct temp_data *tdata;
451 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
455 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
456 MSR_IA32_THERM_STATUS;
457 tdata->is_pkg_data = pkg_flag;
459 tdata->cpu_core_id = TO_CORE_ID(cpu);
460 tdata->attr_size = MAX_CORE_ATTRS;
461 mutex_init(&tdata->update_lock);
465 static int create_core_data(struct platform_device *pdev, unsigned int cpu,
468 struct temp_data *tdata;
469 struct platform_data *pdata = platform_get_drvdata(pdev);
470 struct cpuinfo_x86 *c = &cpu_data(cpu);
475 * Find attr number for sysfs:
476 * We map the attr number to core id of the CPU
477 * The attr number is always core id + 2
478 * The Pkgtemp will always show up as temp1_*, if available
480 attr_no = pkg_flag ? PKG_SYSFS_ATTR_NO : TO_ATTR_NO(cpu);
482 if (attr_no > MAX_CORE_DATA - 1)
485 tdata = init_temp_data(cpu, pkg_flag);
489 /* Test if we can access the status register */
490 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
494 /* We can access status register. Get Critical Temperature */
495 tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
498 * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
499 * The target temperature is available on older CPUs but not in this
500 * register. Atoms don't have the register at all.
502 if (c->x86_model > 0xe && c->x86_model != 0x1c) {
503 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
507 = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
512 pdata->core_data[attr_no] = tdata;
514 /* Create sysfs interfaces */
515 err = create_core_attrs(tdata, pdata->hwmon_dev, attr_no);
521 pdata->core_data[attr_no] = NULL;
527 coretemp_add_core(struct platform_device *pdev, unsigned int cpu, int pkg_flag)
529 if (create_core_data(pdev, cpu, pkg_flag))
530 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
533 static void coretemp_remove_core(struct platform_data *pdata, int indx)
535 struct temp_data *tdata = pdata->core_data[indx];
537 /* if we errored on add then this is already gone */
541 /* Remove the sysfs attributes */
542 sysfs_remove_group(&pdata->hwmon_dev->kobj, &tdata->attr_group);
544 kfree(pdata->core_data[indx]);
545 pdata->core_data[indx] = NULL;
548 static int coretemp_probe(struct platform_device *pdev)
550 struct device *dev = &pdev->dev;
551 struct platform_data *pdata;
553 /* Initialize the per-package data structures */
554 pdata = devm_kzalloc(dev, sizeof(struct platform_data), GFP_KERNEL);
558 pdata->pkg_id = pdev->id;
559 platform_set_drvdata(pdev, pdata);
561 pdata->hwmon_dev = devm_hwmon_device_register_with_groups(dev, DRVNAME,
563 return PTR_ERR_OR_ZERO(pdata->hwmon_dev);
566 static int coretemp_remove(struct platform_device *pdev)
568 struct platform_data *pdata = platform_get_drvdata(pdev);
571 for (i = MAX_CORE_DATA - 1; i >= 0; --i)
572 if (pdata->core_data[i])
573 coretemp_remove_core(pdata, i);
578 static struct platform_driver coretemp_driver = {
582 .probe = coretemp_probe,
583 .remove = coretemp_remove,
586 static struct platform_device *coretemp_device_add(unsigned int cpu)
588 int err, pkgid = topology_logical_package_id(cpu);
589 struct platform_device *pdev;
592 return ERR_PTR(-ENOMEM);
594 pdev = platform_device_alloc(DRVNAME, pkgid);
596 return ERR_PTR(-ENOMEM);
598 err = platform_device_add(pdev);
600 platform_device_put(pdev);
604 pkg_devices[pkgid] = pdev;
608 static int coretemp_cpu_online(unsigned int cpu)
610 struct platform_device *pdev = coretemp_get_pdev(cpu);
611 struct cpuinfo_x86 *c = &cpu_data(cpu);
612 struct platform_data *pdata;
615 * Don't execute this on resume as the offline callback did
616 * not get executed on suspend.
618 if (cpuhp_tasks_frozen)
622 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
623 * sensors. We check this bit only, all the early CPUs
624 * without thermal sensors will be filtered out.
626 if (!cpu_has(c, X86_FEATURE_DTHERM))
630 /* Check the microcode version of the CPU */
631 if (chk_ucode_version(cpu))
635 * Alright, we have DTS support.
636 * We are bringing the _first_ core in this pkg
637 * online. So, initialize per-pkg data structures and
638 * then bring this core online.
640 pdev = coretemp_device_add(cpu);
642 return PTR_ERR(pdev);
645 * Check whether pkgtemp support is available.
646 * If so, add interfaces for pkgtemp.
648 if (cpu_has(c, X86_FEATURE_PTS))
649 coretemp_add_core(pdev, cpu, 1);
652 pdata = platform_get_drvdata(pdev);
654 * Check whether a thread sibling is already online. If not add the
655 * interface for this CPU core.
657 if (!cpumask_intersects(&pdata->cpumask, topology_sibling_cpumask(cpu)))
658 coretemp_add_core(pdev, cpu, 0);
660 cpumask_set_cpu(cpu, &pdata->cpumask);
664 static int coretemp_cpu_offline(unsigned int cpu)
666 struct platform_device *pdev = coretemp_get_pdev(cpu);
667 struct platform_data *pd;
668 struct temp_data *tdata;
672 * Don't execute this on suspend as the device remove locks
675 if (cpuhp_tasks_frozen)
678 /* If the physical CPU device does not exist, just return */
682 /* The core id is too big, just return */
683 indx = TO_ATTR_NO(cpu);
684 if (indx > MAX_CORE_DATA - 1)
687 pd = platform_get_drvdata(pdev);
688 tdata = pd->core_data[indx];
690 cpumask_clear_cpu(cpu, &pd->cpumask);
693 * If this is the last thread sibling, remove the CPU core
694 * interface, If there is still a sibling online, transfer the
695 * target cpu of that core interface to it.
697 target = cpumask_any_and(&pd->cpumask, topology_sibling_cpumask(cpu));
698 if (target >= nr_cpu_ids) {
699 coretemp_remove_core(pd, indx);
700 } else if (tdata && tdata->cpu == cpu) {
701 mutex_lock(&tdata->update_lock);
703 mutex_unlock(&tdata->update_lock);
707 * If all cores in this pkg are offline, remove the device. This
708 * will invoke the platform driver remove function, which cleans up
711 if (cpumask_empty(&pd->cpumask)) {
712 pkg_devices[topology_logical_package_id(cpu)] = NULL;
713 platform_device_unregister(pdev);
718 * Check whether this core is the target for the package
719 * interface. We need to assign it to some other cpu.
721 tdata = pd->core_data[PKG_SYSFS_ATTR_NO];
722 if (tdata && tdata->cpu == cpu) {
723 target = cpumask_first(&pd->cpumask);
724 mutex_lock(&tdata->update_lock);
726 mutex_unlock(&tdata->update_lock);
730 static const struct x86_cpu_id __initconst coretemp_ids[] = {
731 { X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_DTHERM },
734 MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
736 static enum cpuhp_state coretemp_hp_online;
738 static int __init coretemp_init(void)
743 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
744 * sensors. We check this bit only, all the early CPUs
745 * without thermal sensors will be filtered out.
747 if (!x86_match_cpu(coretemp_ids))
750 max_packages = topology_max_packages();
751 pkg_devices = kzalloc(max_packages * sizeof(struct platform_device *),
756 err = platform_driver_register(&coretemp_driver);
760 err = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hwmon/coretemp:online",
761 coretemp_cpu_online, coretemp_cpu_offline);
764 coretemp_hp_online = err;
768 platform_driver_unregister(&coretemp_driver);
772 module_init(coretemp_init)
774 static void __exit coretemp_exit(void)
776 cpuhp_remove_state(coretemp_hp_online);
777 platform_driver_unregister(&coretemp_driver);
780 module_exit(coretemp_exit)
782 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
783 MODULE_DESCRIPTION("Intel Core temperature monitor");
784 MODULE_LICENSE("GPL");