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>
30 #include <linux/sched/isolation.h>
32 #define DRVNAME "coretemp"
35 * force_tjmax only matters when TjMax can't be read from the CPU itself.
36 * When set, it replaces the driver's suboptimal heuristic.
38 static int force_tjmax;
39 module_param_named(tjmax, force_tjmax, int, 0444);
40 MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
42 #define PKG_SYSFS_ATTR_NO 1 /* Sysfs attribute for package temp */
43 #define BASE_SYSFS_ATTR_NO 2 /* Sysfs Base attr no for coretemp */
44 #define NUM_REAL_CORES 128 /* Number of Real cores per cpu */
45 #define CORETEMP_NAME_LENGTH 19 /* String Length of attrs */
46 #define MAX_CORE_ATTRS 4 /* Maximum no of basic attrs */
47 #define TOTAL_ATTRS (MAX_CORE_ATTRS + 1)
48 #define MAX_CORE_DATA (NUM_REAL_CORES + BASE_SYSFS_ATTR_NO)
51 #define for_each_sibling(i, cpu) \
52 for_each_cpu(i, topology_sibling_cpumask(cpu))
54 #define for_each_sibling(i, cpu) for (i = 0; false; )
58 * Per-Core Temperature Data
59 * @tjmax: The static tjmax value when tjmax cannot be retrieved from
60 * IA32_TEMPERATURE_TARGET MSR.
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.
74 unsigned long last_updated;
80 struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
81 char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
82 struct attribute *attrs[TOTAL_ATTRS + 1];
83 struct attribute_group attr_group;
84 struct mutex update_lock;
87 /* Platform Data per Physical CPU */
88 struct platform_data {
89 struct device *hwmon_dev;
91 u16 cpu_map[NUM_REAL_CORES];
93 struct cpumask cpumask;
94 struct temp_data *core_data[MAX_CORE_DATA];
95 struct device_attribute name_attr;
103 static const struct tjmax_pci tjmax_pci_table[] = {
104 { 0x0708, 110000 }, /* CE41x0 (Sodaville ) */
105 { 0x0c72, 102000 }, /* Atom S1240 (Centerton) */
106 { 0x0c73, 95000 }, /* Atom S1220 (Centerton) */
107 { 0x0c75, 95000 }, /* Atom S1260 (Centerton) */
115 static const struct tjmax tjmax_table[] = {
116 { "CPU 230", 100000 }, /* Model 0x1c, stepping 2 */
117 { "CPU 330", 125000 }, /* Model 0x1c, stepping 2 */
128 static const struct tjmax_model tjmax_model_table[] = {
129 { 0x1c, 10, 100000 }, /* D4xx, K4xx, N4xx, D5xx, K5xx, N5xx */
130 { 0x1c, ANY, 90000 }, /* Z5xx, N2xx, possibly others
131 * Note: Also matches 230 and 330,
132 * which are covered by tjmax_table
134 { 0x26, ANY, 90000 }, /* Atom Tunnel Creek (Exx), Lincroft (Z6xx)
135 * Note: TjMax for E6xxT is 110C, but CPU type
136 * is undetectable by software
138 { 0x27, ANY, 90000 }, /* Atom Medfield (Z2460) */
139 { 0x35, ANY, 90000 }, /* Atom Clover Trail/Cloverview (Z27x0) */
140 { 0x36, ANY, 100000 }, /* Atom Cedar Trail/Cedarview (N2xxx, D2xxx)
141 * Also matches S12x0 (stepping 9), covered by
146 static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
148 /* The 100C is default for both mobile and non mobile CPUs */
151 int tjmax_ee = 85000;
156 u16 devfn = PCI_DEVFN(0, 0);
157 struct pci_dev *host_bridge = pci_get_domain_bus_and_slot(0, 0, devfn);
160 * Explicit tjmax table entries override heuristics.
161 * First try PCI host bridge IDs, followed by model ID strings
162 * and model/stepping information.
164 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL) {
165 for (i = 0; i < ARRAY_SIZE(tjmax_pci_table); i++) {
166 if (host_bridge->device == tjmax_pci_table[i].device) {
167 pci_dev_put(host_bridge);
168 return tjmax_pci_table[i].tjmax;
172 pci_dev_put(host_bridge);
174 for (i = 0; i < ARRAY_SIZE(tjmax_table); i++) {
175 if (strstr(c->x86_model_id, tjmax_table[i].id))
176 return tjmax_table[i].tjmax;
179 for (i = 0; i < ARRAY_SIZE(tjmax_model_table); i++) {
180 const struct tjmax_model *tm = &tjmax_model_table[i];
181 if (c->x86_model == tm->model &&
182 (tm->mask == ANY || c->x86_stepping == tm->mask))
186 /* Early chips have no MSR for TjMax */
188 if (c->x86_model == 0xf && c->x86_stepping < 4)
191 if (c->x86_model > 0xe && usemsr_ee) {
195 * Now we can detect the mobile CPU using Intel provided table
196 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
197 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
199 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
202 "Unable to access MSR 0x17, assuming desktop"
205 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
207 * Trust bit 28 up to Penryn, I could not find any
208 * documentation on that; if you happen to know
209 * someone at Intel please ask
213 /* Platform ID bits 52:50 (EDX starts at bit 32) */
214 platform_id = (edx >> 18) & 0x7;
217 * Mobile Penryn CPU seems to be platform ID 7 or 5
220 if (c->x86_model == 0x17 &&
221 (platform_id == 5 || platform_id == 7)) {
223 * If MSR EE bit is set, set it to 90 degrees C,
224 * otherwise 105 degrees C
233 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
236 "Unable to access MSR 0xEE, for Tjmax, left"
238 } else if (eax & 0x40000000) {
241 } else if (tjmax == 100000) {
243 * If we don't use msr EE it means we are desktop CPU
244 * (with exeception of Atom)
246 dev_warn(dev, "Using relative temperature scale!\n");
252 static bool cpu_has_tjmax(struct cpuinfo_x86 *c)
254 u8 model = c->x86_model;
256 return model > 0xe &&
264 static int get_tjmax(struct temp_data *tdata, struct device *dev)
266 struct cpuinfo_x86 *c = &cpu_data(tdata->cpu);
271 /* use static tjmax once it is set */
276 * A new feature of current Intel(R) processors, the
277 * IA32_TEMPERATURE_TARGET contains the TjMax value
279 err = rdmsr_safe_on_cpu(tdata->cpu, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
281 if (cpu_has_tjmax(c))
282 dev_warn(dev, "Unable to read TjMax from CPU %u\n", tdata->cpu);
284 val = (eax >> 16) & 0xff;
290 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
292 tdata->tjmax = force_tjmax * 1000;
295 * An assumption is made for early CPUs and unreadable MSR.
296 * NOTE: the calculated value may not be correct.
298 tdata->tjmax = adjust_tjmax(c, tdata->cpu, dev);
303 static int get_ttarget(struct temp_data *tdata, struct device *dev)
306 int tjmax, ttarget_offset, ret;
309 * ttarget is valid only if tjmax can be retrieved from
310 * MSR_IA32_TEMPERATURE_TARGET
315 ret = rdmsr_safe_on_cpu(tdata->cpu, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
319 tjmax = (eax >> 16) & 0xff;
321 /* Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET. */
322 ttarget_offset = (eax >> 8) & 0xff;
324 return (tjmax - ttarget_offset) * 1000;
327 /* Keep track of how many zone pointers we allocated in init() */
328 static int max_zones __read_mostly;
329 /* Array of zone pointers. Serialized by cpu hotplug lock */
330 static struct platform_device **zone_devices;
332 static ssize_t show_label(struct device *dev,
333 struct device_attribute *devattr, char *buf)
335 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
336 struct platform_data *pdata = dev_get_drvdata(dev);
337 struct temp_data *tdata = pdata->core_data[attr->index];
339 if (tdata->is_pkg_data)
340 return sprintf(buf, "Package id %u\n", pdata->pkg_id);
342 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
345 static ssize_t show_crit_alarm(struct device *dev,
346 struct device_attribute *devattr, char *buf)
349 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
350 struct platform_data *pdata = dev_get_drvdata(dev);
351 struct temp_data *tdata = pdata->core_data[attr->index];
353 mutex_lock(&tdata->update_lock);
354 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
355 mutex_unlock(&tdata->update_lock);
357 return sprintf(buf, "%d\n", (eax >> 5) & 1);
360 static ssize_t show_tjmax(struct device *dev,
361 struct device_attribute *devattr, char *buf)
363 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
364 struct platform_data *pdata = dev_get_drvdata(dev);
365 struct temp_data *tdata = pdata->core_data[attr->index];
368 mutex_lock(&tdata->update_lock);
369 tjmax = get_tjmax(tdata, dev);
370 mutex_unlock(&tdata->update_lock);
372 return sprintf(buf, "%d\n", tjmax);
375 static ssize_t show_ttarget(struct device *dev,
376 struct device_attribute *devattr, char *buf)
378 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
379 struct platform_data *pdata = dev_get_drvdata(dev);
380 struct temp_data *tdata = pdata->core_data[attr->index];
383 mutex_lock(&tdata->update_lock);
384 ttarget = get_ttarget(tdata, dev);
385 mutex_unlock(&tdata->update_lock);
389 return sprintf(buf, "%d\n", ttarget);
392 static ssize_t show_temp(struct device *dev,
393 struct device_attribute *devattr, char *buf)
396 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
397 struct platform_data *pdata = dev_get_drvdata(dev);
398 struct temp_data *tdata = pdata->core_data[attr->index];
401 mutex_lock(&tdata->update_lock);
403 tjmax = get_tjmax(tdata, dev);
404 /* Check whether the time interval has elapsed */
405 if (time_after(jiffies, tdata->last_updated + HZ)) {
406 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
408 * Ignore the valid bit. In all observed cases the register
409 * value is either low or zero if the valid bit is 0.
410 * Return it instead of reporting an error which doesn't
411 * really help at all.
413 tdata->temp = tjmax - ((eax >> 16) & 0x7f) * 1000;
414 tdata->last_updated = jiffies;
417 mutex_unlock(&tdata->update_lock);
418 return sprintf(buf, "%d\n", tdata->temp);
421 static int create_core_attrs(struct temp_data *tdata, struct device *dev,
425 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
426 struct device_attribute *devattr, char *buf) = {
427 show_label, show_crit_alarm, show_temp, show_tjmax,
429 static const char *const suffixes[TOTAL_ATTRS] = {
430 "label", "crit_alarm", "input", "crit", "max"
433 for (i = 0; i < tdata->attr_size; i++) {
434 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH,
435 "temp%d_%s", attr_no, suffixes[i]);
436 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
437 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
438 tdata->sd_attrs[i].dev_attr.attr.mode = 0444;
439 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
440 tdata->sd_attrs[i].index = attr_no;
441 tdata->attrs[i] = &tdata->sd_attrs[i].dev_attr.attr;
443 tdata->attr_group.attrs = tdata->attrs;
444 return sysfs_create_group(&dev->kobj, &tdata->attr_group);
448 static int chk_ucode_version(unsigned int cpu)
450 struct cpuinfo_x86 *c = &cpu_data(cpu);
453 * Check if we have problem with errata AE18 of Core processors:
454 * Readings might stop update when processor visited too deep sleep,
455 * fixed for stepping D0 (6EC).
457 if (c->x86_model == 0xe && c->x86_stepping < 0xc && c->microcode < 0x39) {
458 pr_err("Errata AE18 not fixed, update BIOS or microcode of the CPU!\n");
464 static struct platform_device *coretemp_get_pdev(unsigned int cpu)
466 int id = topology_logical_die_id(cpu);
468 if (id >= 0 && id < max_zones)
469 return zone_devices[id];
473 static struct temp_data *init_temp_data(unsigned int cpu, int pkg_flag)
475 struct temp_data *tdata;
477 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
481 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
482 MSR_IA32_THERM_STATUS;
483 tdata->is_pkg_data = pkg_flag;
485 tdata->cpu_core_id = topology_core_id(cpu);
486 tdata->attr_size = MAX_CORE_ATTRS;
487 mutex_init(&tdata->update_lock);
491 static int create_core_data(struct platform_device *pdev, unsigned int cpu,
494 struct temp_data *tdata;
495 struct platform_data *pdata = platform_get_drvdata(pdev);
496 struct cpuinfo_x86 *c = &cpu_data(cpu);
498 int err, index, attr_no;
500 if (!housekeeping_cpu(cpu, HK_TYPE_MISC))
504 * Find attr number for sysfs:
505 * We map the attr number to core id of the CPU
506 * The attr number is always core id + 2
507 * The Pkgtemp will always show up as temp1_*, if available
510 attr_no = PKG_SYSFS_ATTR_NO;
512 index = ida_alloc(&pdata->ida, GFP_KERNEL);
515 pdata->cpu_map[index] = topology_core_id(cpu);
516 attr_no = index + BASE_SYSFS_ATTR_NO;
519 if (attr_no > MAX_CORE_DATA - 1) {
524 tdata = init_temp_data(cpu, pkg_flag);
530 /* Test if we can access the status register */
531 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
535 /* Make sure tdata->tjmax is a valid indicator for dynamic/static tjmax */
536 get_tjmax(tdata, &pdev->dev);
539 * The target temperature is available on older CPUs but not in the
540 * MSR_IA32_TEMPERATURE_TARGET register. Atoms don't have the register
543 if (c->x86_model > 0xe && c->x86_model != 0x1c)
544 if (get_ttarget(tdata, &pdev->dev) >= 0)
547 pdata->core_data[attr_no] = tdata;
549 /* Create sysfs interfaces */
550 err = create_core_attrs(tdata, pdata->hwmon_dev, attr_no);
556 pdata->core_data[attr_no] = NULL;
560 ida_free(&pdata->ida, index);
565 coretemp_add_core(struct platform_device *pdev, unsigned int cpu, int pkg_flag)
567 if (create_core_data(pdev, cpu, pkg_flag))
568 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
571 static void coretemp_remove_core(struct platform_data *pdata, int indx)
573 struct temp_data *tdata = pdata->core_data[indx];
575 /* if we errored on add then this is already gone */
579 /* Remove the sysfs attributes */
580 sysfs_remove_group(&pdata->hwmon_dev->kobj, &tdata->attr_group);
582 kfree(pdata->core_data[indx]);
583 pdata->core_data[indx] = NULL;
585 if (indx >= BASE_SYSFS_ATTR_NO)
586 ida_free(&pdata->ida, indx - BASE_SYSFS_ATTR_NO);
589 static int coretemp_device_add(int zoneid)
591 struct platform_device *pdev;
592 struct platform_data *pdata;
595 /* Initialize the per-zone data structures */
596 pdata = kzalloc(sizeof(*pdata), GFP_KERNEL);
600 pdata->pkg_id = zoneid;
601 ida_init(&pdata->ida);
603 pdev = platform_device_alloc(DRVNAME, zoneid);
609 err = platform_device_add(pdev);
613 platform_set_drvdata(pdev, pdata);
614 zone_devices[zoneid] = pdev;
618 platform_device_put(pdev);
624 static void coretemp_device_remove(int zoneid)
626 struct platform_device *pdev = zone_devices[zoneid];
627 struct platform_data *pdata = platform_get_drvdata(pdev);
629 ida_destroy(&pdata->ida);
631 platform_device_unregister(pdev);
634 static int coretemp_cpu_online(unsigned int cpu)
636 struct platform_device *pdev = coretemp_get_pdev(cpu);
637 struct cpuinfo_x86 *c = &cpu_data(cpu);
638 struct platform_data *pdata;
641 * Don't execute this on resume as the offline callback did
642 * not get executed on suspend.
644 if (cpuhp_tasks_frozen)
648 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
649 * sensors. We check this bit only, all the early CPUs
650 * without thermal sensors will be filtered out.
652 if (!cpu_has(c, X86_FEATURE_DTHERM))
655 pdata = platform_get_drvdata(pdev);
656 if (!pdata->hwmon_dev) {
657 struct device *hwmon;
659 /* Check the microcode version of the CPU */
660 if (chk_ucode_version(cpu))
664 * Alright, we have DTS support.
665 * We are bringing the _first_ core in this pkg
666 * online. So, initialize per-pkg data structures and
667 * then bring this core online.
669 hwmon = hwmon_device_register_with_groups(&pdev->dev, DRVNAME,
672 return PTR_ERR(hwmon);
673 pdata->hwmon_dev = hwmon;
676 * Check whether pkgtemp support is available.
677 * If so, add interfaces for pkgtemp.
679 if (cpu_has(c, X86_FEATURE_PTS))
680 coretemp_add_core(pdev, cpu, 1);
684 * Check whether a thread sibling is already online. If not add the
685 * interface for this CPU core.
687 if (!cpumask_intersects(&pdata->cpumask, topology_sibling_cpumask(cpu)))
688 coretemp_add_core(pdev, cpu, 0);
690 cpumask_set_cpu(cpu, &pdata->cpumask);
694 static int coretemp_cpu_offline(unsigned int cpu)
696 struct platform_device *pdev = coretemp_get_pdev(cpu);
697 struct platform_data *pd;
698 struct temp_data *tdata;
699 int i, indx = -1, target;
701 /* No need to tear down any interfaces for suspend */
702 if (cpuhp_tasks_frozen)
705 /* If the physical CPU device does not exist, just return */
706 pd = platform_get_drvdata(pdev);
710 for (i = 0; i < NUM_REAL_CORES; i++) {
711 if (pd->cpu_map[i] == topology_core_id(cpu)) {
712 indx = i + BASE_SYSFS_ATTR_NO;
717 /* Too many cores and this core is not populated, just return */
721 tdata = pd->core_data[indx];
723 cpumask_clear_cpu(cpu, &pd->cpumask);
726 * If this is the last thread sibling, remove the CPU core
727 * interface, If there is still a sibling online, transfer the
728 * target cpu of that core interface to it.
730 target = cpumask_any_and(&pd->cpumask, topology_sibling_cpumask(cpu));
731 if (target >= nr_cpu_ids) {
732 coretemp_remove_core(pd, indx);
733 } else if (tdata && tdata->cpu == cpu) {
734 mutex_lock(&tdata->update_lock);
736 mutex_unlock(&tdata->update_lock);
740 * If all cores in this pkg are offline, remove the interface.
742 tdata = pd->core_data[PKG_SYSFS_ATTR_NO];
743 if (cpumask_empty(&pd->cpumask)) {
745 coretemp_remove_core(pd, PKG_SYSFS_ATTR_NO);
746 hwmon_device_unregister(pd->hwmon_dev);
747 pd->hwmon_dev = NULL;
752 * Check whether this core is the target for the package
753 * interface. We need to assign it to some other cpu.
755 if (tdata && tdata->cpu == cpu) {
756 target = cpumask_first(&pd->cpumask);
757 mutex_lock(&tdata->update_lock);
759 mutex_unlock(&tdata->update_lock);
763 static const struct x86_cpu_id __initconst coretemp_ids[] = {
764 X86_MATCH_VENDOR_FEATURE(INTEL, X86_FEATURE_DTHERM, NULL),
767 MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
769 static enum cpuhp_state coretemp_hp_online;
771 static int __init coretemp_init(void)
776 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
777 * sensors. We check this bit only, all the early CPUs
778 * without thermal sensors will be filtered out.
780 if (!x86_match_cpu(coretemp_ids))
783 max_zones = topology_max_packages() * topology_max_die_per_package();
784 zone_devices = kcalloc(max_zones, sizeof(struct platform_device *),
789 for (i = 0; i < max_zones; i++) {
790 err = coretemp_device_add(i);
795 err = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hwmon/coretemp:online",
796 coretemp_cpu_online, coretemp_cpu_offline);
799 coretemp_hp_online = err;
804 coretemp_device_remove(i);
808 module_init(coretemp_init)
810 static void __exit coretemp_exit(void)
814 cpuhp_remove_state(coretemp_hp_online);
815 for (i = 0; i < max_zones; i++)
816 coretemp_device_remove(i);
819 module_exit(coretemp_exit)
821 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
822 MODULE_DESCRIPTION("Intel Core temperature monitor");
823 MODULE_LICENSE("GPL");