1 // SPDX-License-Identifier: GPL-2.0-only
3 * coretemp.c - Linux kernel module for hardware monitoring
5 * Copyright (C) 2007 Rudolf Marek <r.marek@assembler.cz>
7 * Inspired from many hwmon drivers
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/slab.h>
15 #include <linux/jiffies.h>
16 #include <linux/hwmon.h>
17 #include <linux/sysfs.h>
18 #include <linux/hwmon-sysfs.h>
19 #include <linux/err.h>
20 #include <linux/mutex.h>
21 #include <linux/list.h>
22 #include <linux/platform_device.h>
23 #include <linux/cpu.h>
24 #include <linux/smp.h>
25 #include <linux/moduleparam.h>
26 #include <linux/pci.h>
28 #include <asm/processor.h>
29 #include <asm/cpu_device_id.h>
31 #define DRVNAME "coretemp"
34 * force_tjmax only matters when TjMax can't be read from the CPU itself.
35 * When set, it replaces the driver's suboptimal heuristic.
37 static int force_tjmax;
38 module_param_named(tjmax, force_tjmax, int, 0444);
39 MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
41 #define PKG_SYSFS_ATTR_NO 1 /* Sysfs attribute for package temp */
42 #define BASE_SYSFS_ATTR_NO 2 /* Sysfs Base attr no for coretemp */
43 #define NUM_REAL_CORES 128 /* Number of Real cores per cpu */
44 #define CORETEMP_NAME_LENGTH 19 /* String Length of attrs */
45 #define MAX_CORE_ATTRS 4 /* Maximum no of basic attrs */
46 #define TOTAL_ATTRS (MAX_CORE_ATTRS + 1)
47 #define MAX_CORE_DATA (NUM_REAL_CORES + BASE_SYSFS_ATTR_NO)
49 #define TO_CORE_ID(cpu) (cpu_data(cpu).cpu_core_id)
50 #define TO_ATTR_NO(cpu) (TO_CORE_ID(cpu) + BASE_SYSFS_ATTR_NO)
53 #define for_each_sibling(i, cpu) \
54 for_each_cpu(i, topology_sibling_cpumask(cpu))
56 #define for_each_sibling(i, cpu) for (i = 0; false; )
60 * Per-Core Temperature Data
61 * @last_updated: The time when the current temperature value was updated
62 * earlier (in jiffies).
63 * @cpu_core_id: The CPU Core from which temperature values should be read
64 * This value is passed as "id" field to rdmsr/wrmsr functions.
65 * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
66 * from where the temperature values should be read.
67 * @attr_size: Total number of pre-core attrs displayed in the sysfs.
68 * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
69 * Otherwise, temp_data holds coretemp data.
70 * @valid: If this is 1, the current temperature is valid.
76 unsigned long last_updated;
83 struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
84 char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
85 struct attribute *attrs[TOTAL_ATTRS + 1];
86 struct attribute_group attr_group;
87 struct mutex update_lock;
90 /* Platform Data per Physical CPU */
91 struct platform_data {
92 struct device *hwmon_dev;
94 struct cpumask cpumask;
95 struct temp_data *core_data[MAX_CORE_DATA];
96 struct device_attribute name_attr;
99 /* Keep track of how many zone pointers we allocated in init() */
100 static int max_zones __read_mostly;
101 /* Array of zone pointers. Serialized by cpu hotplug lock */
102 static struct platform_device **zone_devices;
104 static ssize_t show_label(struct device *dev,
105 struct device_attribute *devattr, char *buf)
107 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
108 struct platform_data *pdata = dev_get_drvdata(dev);
109 struct temp_data *tdata = pdata->core_data[attr->index];
111 if (tdata->is_pkg_data)
112 return sprintf(buf, "Package id %u\n", pdata->pkg_id);
114 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
117 static ssize_t show_crit_alarm(struct device *dev,
118 struct device_attribute *devattr, char *buf)
121 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
122 struct platform_data *pdata = dev_get_drvdata(dev);
123 struct temp_data *tdata = pdata->core_data[attr->index];
125 mutex_lock(&tdata->update_lock);
126 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
127 mutex_unlock(&tdata->update_lock);
129 return sprintf(buf, "%d\n", (eax >> 5) & 1);
132 static ssize_t show_tjmax(struct device *dev,
133 struct device_attribute *devattr, char *buf)
135 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
136 struct platform_data *pdata = dev_get_drvdata(dev);
138 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
141 static ssize_t show_ttarget(struct device *dev,
142 struct device_attribute *devattr, char *buf)
144 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
145 struct platform_data *pdata = dev_get_drvdata(dev);
147 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
150 static ssize_t show_temp(struct device *dev,
151 struct device_attribute *devattr, char *buf)
154 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
155 struct platform_data *pdata = dev_get_drvdata(dev);
156 struct temp_data *tdata = pdata->core_data[attr->index];
158 mutex_lock(&tdata->update_lock);
160 /* Check whether the time interval has elapsed */
161 if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
162 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
164 * Ignore the valid bit. In all observed cases the register
165 * value is either low or zero if the valid bit is 0.
166 * Return it instead of reporting an error which doesn't
167 * really help at all.
169 tdata->temp = tdata->tjmax - ((eax >> 16) & 0x7f) * 1000;
171 tdata->last_updated = jiffies;
174 mutex_unlock(&tdata->update_lock);
175 return sprintf(buf, "%d\n", tdata->temp);
183 static const struct tjmax_pci tjmax_pci_table[] = {
184 { 0x0708, 110000 }, /* CE41x0 (Sodaville ) */
185 { 0x0c72, 102000 }, /* Atom S1240 (Centerton) */
186 { 0x0c73, 95000 }, /* Atom S1220 (Centerton) */
187 { 0x0c75, 95000 }, /* Atom S1260 (Centerton) */
195 static const struct tjmax tjmax_table[] = {
196 { "CPU 230", 100000 }, /* Model 0x1c, stepping 2 */
197 { "CPU 330", 125000 }, /* Model 0x1c, stepping 2 */
208 static const struct tjmax_model tjmax_model_table[] = {
209 { 0x1c, 10, 100000 }, /* D4xx, K4xx, N4xx, D5xx, K5xx, N5xx */
210 { 0x1c, ANY, 90000 }, /* Z5xx, N2xx, possibly others
211 * Note: Also matches 230 and 330,
212 * which are covered by tjmax_table
214 { 0x26, ANY, 90000 }, /* Atom Tunnel Creek (Exx), Lincroft (Z6xx)
215 * Note: TjMax for E6xxT is 110C, but CPU type
216 * is undetectable by software
218 { 0x27, ANY, 90000 }, /* Atom Medfield (Z2460) */
219 { 0x35, ANY, 90000 }, /* Atom Clover Trail/Cloverview (Z27x0) */
220 { 0x36, ANY, 100000 }, /* Atom Cedar Trail/Cedarview (N2xxx, D2xxx)
221 * Also matches S12x0 (stepping 9), covered by
226 static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
228 /* The 100C is default for both mobile and non mobile CPUs */
231 int tjmax_ee = 85000;
236 u16 devfn = PCI_DEVFN(0, 0);
237 struct pci_dev *host_bridge = pci_get_domain_bus_and_slot(0, 0, devfn);
240 * Explicit tjmax table entries override heuristics.
241 * First try PCI host bridge IDs, followed by model ID strings
242 * and model/stepping information.
244 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL) {
245 for (i = 0; i < ARRAY_SIZE(tjmax_pci_table); i++) {
246 if (host_bridge->device == tjmax_pci_table[i].device)
247 return tjmax_pci_table[i].tjmax;
251 for (i = 0; i < ARRAY_SIZE(tjmax_table); i++) {
252 if (strstr(c->x86_model_id, tjmax_table[i].id))
253 return tjmax_table[i].tjmax;
256 for (i = 0; i < ARRAY_SIZE(tjmax_model_table); i++) {
257 const struct tjmax_model *tm = &tjmax_model_table[i];
258 if (c->x86_model == tm->model &&
259 (tm->mask == ANY || c->x86_stepping == tm->mask))
263 /* Early chips have no MSR for TjMax */
265 if (c->x86_model == 0xf && c->x86_stepping < 4)
268 if (c->x86_model > 0xe && usemsr_ee) {
272 * Now we can detect the mobile CPU using Intel provided table
273 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
274 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
276 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
279 "Unable to access MSR 0x17, assuming desktop"
282 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
284 * Trust bit 28 up to Penryn, I could not find any
285 * documentation on that; if you happen to know
286 * someone at Intel please ask
290 /* Platform ID bits 52:50 (EDX starts at bit 32) */
291 platform_id = (edx >> 18) & 0x7;
294 * Mobile Penryn CPU seems to be platform ID 7 or 5
297 if (c->x86_model == 0x17 &&
298 (platform_id == 5 || platform_id == 7)) {
300 * If MSR EE bit is set, set it to 90 degrees C,
301 * otherwise 105 degrees C
310 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
313 "Unable to access MSR 0xEE, for Tjmax, left"
315 } else if (eax & 0x40000000) {
318 } else if (tjmax == 100000) {
320 * If we don't use msr EE it means we are desktop CPU
321 * (with exeception of Atom)
323 dev_warn(dev, "Using relative temperature scale!\n");
329 static bool cpu_has_tjmax(struct cpuinfo_x86 *c)
331 u8 model = c->x86_model;
333 return model > 0xe &&
341 static int get_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
348 * A new feature of current Intel(R) processors, the
349 * IA32_TEMPERATURE_TARGET contains the TjMax value
351 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
353 if (cpu_has_tjmax(c))
354 dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
356 val = (eax >> 16) & 0xff;
358 * If the TjMax is not plausible, an assumption
362 dev_dbg(dev, "TjMax is %d degrees C\n", val);
368 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
370 return force_tjmax * 1000;
374 * An assumption is made for early CPUs and unreadable MSR.
375 * NOTE: the calculated value may not be correct.
377 return adjust_tjmax(c, id, dev);
380 static int create_core_attrs(struct temp_data *tdata, struct device *dev,
384 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
385 struct device_attribute *devattr, char *buf) = {
386 show_label, show_crit_alarm, show_temp, show_tjmax,
388 static const char *const suffixes[TOTAL_ATTRS] = {
389 "label", "crit_alarm", "input", "crit", "max"
392 for (i = 0; i < tdata->attr_size; i++) {
393 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH,
394 "temp%d_%s", attr_no, suffixes[i]);
395 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
396 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
397 tdata->sd_attrs[i].dev_attr.attr.mode = 0444;
398 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
399 tdata->sd_attrs[i].index = attr_no;
400 tdata->attrs[i] = &tdata->sd_attrs[i].dev_attr.attr;
402 tdata->attr_group.attrs = tdata->attrs;
403 return sysfs_create_group(&dev->kobj, &tdata->attr_group);
407 static int chk_ucode_version(unsigned int cpu)
409 struct cpuinfo_x86 *c = &cpu_data(cpu);
412 * Check if we have problem with errata AE18 of Core processors:
413 * Readings might stop update when processor visited too deep sleep,
414 * fixed for stepping D0 (6EC).
416 if (c->x86_model == 0xe && c->x86_stepping < 0xc && c->microcode < 0x39) {
417 pr_err("Errata AE18 not fixed, update BIOS or microcode of the CPU!\n");
423 static struct platform_device *coretemp_get_pdev(unsigned int cpu)
425 int id = topology_logical_die_id(cpu);
427 if (id >= 0 && id < max_zones)
428 return zone_devices[id];
432 static struct temp_data *init_temp_data(unsigned int cpu, int pkg_flag)
434 struct temp_data *tdata;
436 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
440 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
441 MSR_IA32_THERM_STATUS;
442 tdata->is_pkg_data = pkg_flag;
444 tdata->cpu_core_id = TO_CORE_ID(cpu);
445 tdata->attr_size = MAX_CORE_ATTRS;
446 mutex_init(&tdata->update_lock);
450 static int create_core_data(struct platform_device *pdev, unsigned int cpu,
453 struct temp_data *tdata;
454 struct platform_data *pdata = platform_get_drvdata(pdev);
455 struct cpuinfo_x86 *c = &cpu_data(cpu);
460 * Find attr number for sysfs:
461 * We map the attr number to core id of the CPU
462 * The attr number is always core id + 2
463 * The Pkgtemp will always show up as temp1_*, if available
465 attr_no = pkg_flag ? PKG_SYSFS_ATTR_NO : TO_ATTR_NO(cpu);
467 if (attr_no > MAX_CORE_DATA - 1)
470 tdata = init_temp_data(cpu, pkg_flag);
474 /* Test if we can access the status register */
475 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
479 /* We can access status register. Get Critical Temperature */
480 tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
483 * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
484 * The target temperature is available on older CPUs but not in this
485 * register. Atoms don't have the register at all.
487 if (c->x86_model > 0xe && c->x86_model != 0x1c) {
488 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
492 = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
497 pdata->core_data[attr_no] = tdata;
499 /* Create sysfs interfaces */
500 err = create_core_attrs(tdata, pdata->hwmon_dev, attr_no);
506 pdata->core_data[attr_no] = NULL;
512 coretemp_add_core(struct platform_device *pdev, unsigned int cpu, int pkg_flag)
514 if (create_core_data(pdev, cpu, pkg_flag))
515 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
518 static void coretemp_remove_core(struct platform_data *pdata, int indx)
520 struct temp_data *tdata = pdata->core_data[indx];
522 /* Remove the sysfs attributes */
523 sysfs_remove_group(&pdata->hwmon_dev->kobj, &tdata->attr_group);
525 kfree(pdata->core_data[indx]);
526 pdata->core_data[indx] = NULL;
529 static int coretemp_probe(struct platform_device *pdev)
531 struct device *dev = &pdev->dev;
532 struct platform_data *pdata;
534 /* Initialize the per-zone data structures */
535 pdata = devm_kzalloc(dev, sizeof(struct platform_data), GFP_KERNEL);
539 pdata->pkg_id = pdev->id;
540 platform_set_drvdata(pdev, pdata);
542 pdata->hwmon_dev = devm_hwmon_device_register_with_groups(dev, DRVNAME,
544 return PTR_ERR_OR_ZERO(pdata->hwmon_dev);
547 static int coretemp_remove(struct platform_device *pdev)
549 struct platform_data *pdata = platform_get_drvdata(pdev);
552 for (i = MAX_CORE_DATA - 1; i >= 0; --i)
553 if (pdata->core_data[i])
554 coretemp_remove_core(pdata, i);
559 static struct platform_driver coretemp_driver = {
563 .probe = coretemp_probe,
564 .remove = coretemp_remove,
567 static struct platform_device *coretemp_device_add(unsigned int cpu)
569 int err, zoneid = topology_logical_die_id(cpu);
570 struct platform_device *pdev;
573 return ERR_PTR(-ENOMEM);
575 pdev = platform_device_alloc(DRVNAME, zoneid);
577 return ERR_PTR(-ENOMEM);
579 err = platform_device_add(pdev);
581 platform_device_put(pdev);
585 zone_devices[zoneid] = pdev;
589 static int coretemp_cpu_online(unsigned int cpu)
591 struct platform_device *pdev = coretemp_get_pdev(cpu);
592 struct cpuinfo_x86 *c = &cpu_data(cpu);
593 struct platform_data *pdata;
596 * Don't execute this on resume as the offline callback did
597 * not get executed on suspend.
599 if (cpuhp_tasks_frozen)
603 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
604 * sensors. We check this bit only, all the early CPUs
605 * without thermal sensors will be filtered out.
607 if (!cpu_has(c, X86_FEATURE_DTHERM))
611 /* Check the microcode version of the CPU */
612 if (chk_ucode_version(cpu))
616 * Alright, we have DTS support.
617 * We are bringing the _first_ core in this pkg
618 * online. So, initialize per-pkg data structures and
619 * then bring this core online.
621 pdev = coretemp_device_add(cpu);
623 return PTR_ERR(pdev);
626 * Check whether pkgtemp support is available.
627 * If so, add interfaces for pkgtemp.
629 if (cpu_has(c, X86_FEATURE_PTS))
630 coretemp_add_core(pdev, cpu, 1);
633 pdata = platform_get_drvdata(pdev);
635 * Check whether a thread sibling is already online. If not add the
636 * interface for this CPU core.
638 if (!cpumask_intersects(&pdata->cpumask, topology_sibling_cpumask(cpu)))
639 coretemp_add_core(pdev, cpu, 0);
641 cpumask_set_cpu(cpu, &pdata->cpumask);
645 static int coretemp_cpu_offline(unsigned int cpu)
647 struct platform_device *pdev = coretemp_get_pdev(cpu);
648 struct platform_data *pd;
649 struct temp_data *tdata;
653 * Don't execute this on suspend as the device remove locks
656 if (cpuhp_tasks_frozen)
659 /* If the physical CPU device does not exist, just return */
663 /* The core id is too big, just return */
664 indx = TO_ATTR_NO(cpu);
665 if (indx > MAX_CORE_DATA - 1)
668 pd = platform_get_drvdata(pdev);
669 tdata = pd->core_data[indx];
671 cpumask_clear_cpu(cpu, &pd->cpumask);
674 * If this is the last thread sibling, remove the CPU core
675 * interface, If there is still a sibling online, transfer the
676 * target cpu of that core interface to it.
678 target = cpumask_any_and(&pd->cpumask, topology_sibling_cpumask(cpu));
679 if (target >= nr_cpu_ids) {
680 coretemp_remove_core(pd, indx);
681 } else if (tdata && tdata->cpu == cpu) {
682 mutex_lock(&tdata->update_lock);
684 mutex_unlock(&tdata->update_lock);
688 * If all cores in this pkg are offline, remove the device. This
689 * will invoke the platform driver remove function, which cleans up
692 if (cpumask_empty(&pd->cpumask)) {
693 zone_devices[topology_logical_die_id(cpu)] = NULL;
694 platform_device_unregister(pdev);
699 * Check whether this core is the target for the package
700 * interface. We need to assign it to some other cpu.
702 tdata = pd->core_data[PKG_SYSFS_ATTR_NO];
703 if (tdata && tdata->cpu == cpu) {
704 target = cpumask_first(&pd->cpumask);
705 mutex_lock(&tdata->update_lock);
707 mutex_unlock(&tdata->update_lock);
711 static const struct x86_cpu_id __initconst coretemp_ids[] = {
712 X86_MATCH_VENDOR_FEATURE(INTEL, X86_FEATURE_DTHERM, NULL),
715 MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
717 static enum cpuhp_state coretemp_hp_online;
719 static int __init coretemp_init(void)
724 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
725 * sensors. We check this bit only, all the early CPUs
726 * without thermal sensors will be filtered out.
728 if (!x86_match_cpu(coretemp_ids))
731 max_zones = topology_max_packages() * topology_max_die_per_package();
732 zone_devices = kcalloc(max_zones, sizeof(struct platform_device *),
737 err = platform_driver_register(&coretemp_driver);
741 err = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hwmon/coretemp:online",
742 coretemp_cpu_online, coretemp_cpu_offline);
745 coretemp_hp_online = err;
749 platform_driver_unregister(&coretemp_driver);
754 module_init(coretemp_init)
756 static void __exit coretemp_exit(void)
758 cpuhp_remove_state(coretemp_hp_online);
759 platform_driver_unregister(&coretemp_driver);
762 module_exit(coretemp_exit)
764 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
765 MODULE_DESCRIPTION("Intel Core temperature monitor");
766 MODULE_LICENSE("GPL");