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)
50 #define for_each_sibling(i, cpu) \
51 for_each_cpu(i, topology_sibling_cpumask(cpu))
53 #define for_each_sibling(i, cpu) for (i = 0; false; )
57 * Per-Core Temperature Data
58 * @last_updated: The time when the current temperature value was updated
59 * earlier (in jiffies).
60 * @cpu_core_id: The CPU Core from which temperature values should be read
61 * This value is passed as "id" field to rdmsr/wrmsr functions.
62 * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
63 * from where the temperature values should be read.
64 * @attr_size: Total number of pre-core attrs displayed in the sysfs.
65 * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
66 * Otherwise, temp_data holds coretemp data.
67 * @valid: If this is 1, the current temperature is valid.
73 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;
98 /* Keep track of how many zone pointers we allocated in init() */
99 static int max_zones __read_mostly;
100 /* Array of zone pointers. Serialized by cpu hotplug lock */
101 static struct platform_device **zone_devices;
103 static ssize_t show_label(struct device *dev,
104 struct device_attribute *devattr, char *buf)
106 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
107 struct platform_data *pdata = dev_get_drvdata(dev);
108 struct temp_data *tdata = pdata->core_data[attr->index];
110 if (tdata->is_pkg_data)
111 return sprintf(buf, "Package id %u\n", pdata->pkg_id);
113 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
116 static ssize_t show_crit_alarm(struct device *dev,
117 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 mutex_lock(&tdata->update_lock);
125 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
126 mutex_unlock(&tdata->update_lock);
128 return sprintf(buf, "%d\n", (eax >> 5) & 1);
131 static ssize_t show_tjmax(struct device *dev,
132 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);
137 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
140 static ssize_t show_ttarget(struct device *dev,
141 struct device_attribute *devattr, char *buf)
143 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
144 struct platform_data *pdata = dev_get_drvdata(dev);
146 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
149 static ssize_t show_temp(struct device *dev,
150 struct device_attribute *devattr, char *buf)
153 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
154 struct platform_data *pdata = dev_get_drvdata(dev);
155 struct temp_data *tdata = pdata->core_data[attr->index];
157 mutex_lock(&tdata->update_lock);
159 /* Check whether the time interval has elapsed */
160 if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
161 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
163 * Ignore the valid bit. In all observed cases the register
164 * value is either low or zero if the valid bit is 0.
165 * Return it instead of reporting an error which doesn't
166 * really help at all.
168 tdata->temp = tdata->tjmax - ((eax >> 16) & 0x7f) * 1000;
170 tdata->last_updated = jiffies;
173 mutex_unlock(&tdata->update_lock);
174 return sprintf(buf, "%d\n", tdata->temp);
182 static const struct tjmax_pci tjmax_pci_table[] = {
183 { 0x0708, 110000 }, /* CE41x0 (Sodaville ) */
184 { 0x0c72, 102000 }, /* Atom S1240 (Centerton) */
185 { 0x0c73, 95000 }, /* Atom S1220 (Centerton) */
186 { 0x0c75, 95000 }, /* Atom S1260 (Centerton) */
194 static const struct tjmax tjmax_table[] = {
195 { "CPU 230", 100000 }, /* Model 0x1c, stepping 2 */
196 { "CPU 330", 125000 }, /* Model 0x1c, stepping 2 */
207 static const struct tjmax_model tjmax_model_table[] = {
208 { 0x1c, 10, 100000 }, /* D4xx, K4xx, N4xx, D5xx, K5xx, N5xx */
209 { 0x1c, ANY, 90000 }, /* Z5xx, N2xx, possibly others
210 * Note: Also matches 230 and 330,
211 * which are covered by tjmax_table
213 { 0x26, ANY, 90000 }, /* Atom Tunnel Creek (Exx), Lincroft (Z6xx)
214 * Note: TjMax for E6xxT is 110C, but CPU type
215 * is undetectable by software
217 { 0x27, ANY, 90000 }, /* Atom Medfield (Z2460) */
218 { 0x35, ANY, 90000 }, /* Atom Clover Trail/Cloverview (Z27x0) */
219 { 0x36, ANY, 100000 }, /* Atom Cedar Trail/Cedarview (N2xxx, D2xxx)
220 * Also matches S12x0 (stepping 9), covered by
225 static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
227 /* The 100C is default for both mobile and non mobile CPUs */
230 int tjmax_ee = 85000;
235 u16 devfn = PCI_DEVFN(0, 0);
236 struct pci_dev *host_bridge = pci_get_domain_bus_and_slot(0, 0, devfn);
239 * Explicit tjmax table entries override heuristics.
240 * First try PCI host bridge IDs, followed by model ID strings
241 * and model/stepping information.
243 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL) {
244 for (i = 0; i < ARRAY_SIZE(tjmax_pci_table); i++) {
245 if (host_bridge->device == tjmax_pci_table[i].device) {
246 pci_dev_put(host_bridge);
247 return tjmax_pci_table[i].tjmax;
251 pci_dev_put(host_bridge);
253 for (i = 0; i < ARRAY_SIZE(tjmax_table); i++) {
254 if (strstr(c->x86_model_id, tjmax_table[i].id))
255 return tjmax_table[i].tjmax;
258 for (i = 0; i < ARRAY_SIZE(tjmax_model_table); i++) {
259 const struct tjmax_model *tm = &tjmax_model_table[i];
260 if (c->x86_model == tm->model &&
261 (tm->mask == ANY || c->x86_stepping == tm->mask))
265 /* Early chips have no MSR for TjMax */
267 if (c->x86_model == 0xf && c->x86_stepping < 4)
270 if (c->x86_model > 0xe && usemsr_ee) {
274 * Now we can detect the mobile CPU using Intel provided table
275 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
276 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
278 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
281 "Unable to access MSR 0x17, assuming desktop"
284 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
286 * Trust bit 28 up to Penryn, I could not find any
287 * documentation on that; if you happen to know
288 * someone at Intel please ask
292 /* Platform ID bits 52:50 (EDX starts at bit 32) */
293 platform_id = (edx >> 18) & 0x7;
296 * Mobile Penryn CPU seems to be platform ID 7 or 5
299 if (c->x86_model == 0x17 &&
300 (platform_id == 5 || platform_id == 7)) {
302 * If MSR EE bit is set, set it to 90 degrees C,
303 * otherwise 105 degrees C
312 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
315 "Unable to access MSR 0xEE, for Tjmax, left"
317 } else if (eax & 0x40000000) {
320 } else if (tjmax == 100000) {
322 * If we don't use msr EE it means we are desktop CPU
323 * (with exeception of Atom)
325 dev_warn(dev, "Using relative temperature scale!\n");
331 static bool cpu_has_tjmax(struct cpuinfo_x86 *c)
333 u8 model = c->x86_model;
335 return model > 0xe &&
343 static int get_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
350 * A new feature of current Intel(R) processors, the
351 * IA32_TEMPERATURE_TARGET contains the TjMax value
353 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
355 if (cpu_has_tjmax(c))
356 dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
358 val = (eax >> 16) & 0xff;
360 * If the TjMax is not plausible, an assumption
364 dev_dbg(dev, "TjMax is %d degrees C\n", val);
370 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
372 return force_tjmax * 1000;
376 * An assumption is made for early CPUs and unreadable MSR.
377 * NOTE: the calculated value may not be correct.
379 return adjust_tjmax(c, id, dev);
382 static int create_core_attrs(struct temp_data *tdata, struct device *dev,
386 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
387 struct device_attribute *devattr, char *buf) = {
388 show_label, show_crit_alarm, show_temp, show_tjmax,
390 static const char *const suffixes[TOTAL_ATTRS] = {
391 "label", "crit_alarm", "input", "crit", "max"
394 for (i = 0; i < tdata->attr_size; i++) {
395 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH,
396 "temp%d_%s", attr_no, suffixes[i]);
397 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
398 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
399 tdata->sd_attrs[i].dev_attr.attr.mode = 0444;
400 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
401 tdata->sd_attrs[i].index = attr_no;
402 tdata->attrs[i] = &tdata->sd_attrs[i].dev_attr.attr;
404 tdata->attr_group.attrs = tdata->attrs;
405 return sysfs_create_group(&dev->kobj, &tdata->attr_group);
409 static int chk_ucode_version(unsigned int cpu)
411 struct cpuinfo_x86 *c = &cpu_data(cpu);
414 * Check if we have problem with errata AE18 of Core processors:
415 * Readings might stop update when processor visited too deep sleep,
416 * fixed for stepping D0 (6EC).
418 if (c->x86_model == 0xe && c->x86_stepping < 0xc && c->microcode < 0x39) {
419 pr_err("Errata AE18 not fixed, update BIOS or microcode of the CPU!\n");
425 static struct platform_device *coretemp_get_pdev(unsigned int cpu)
427 int id = topology_logical_die_id(cpu);
429 if (id >= 0 && id < max_zones)
430 return zone_devices[id];
434 static struct temp_data *init_temp_data(unsigned int cpu, int pkg_flag)
436 struct temp_data *tdata;
438 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
442 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
443 MSR_IA32_THERM_STATUS;
444 tdata->is_pkg_data = pkg_flag;
446 tdata->cpu_core_id = topology_core_id(cpu);
447 tdata->attr_size = MAX_CORE_ATTRS;
448 mutex_init(&tdata->update_lock);
452 static int create_core_data(struct platform_device *pdev, unsigned int cpu,
455 struct temp_data *tdata;
456 struct platform_data *pdata = platform_get_drvdata(pdev);
457 struct cpuinfo_x86 *c = &cpu_data(cpu);
459 int err, index, attr_no;
462 * Find attr number for sysfs:
463 * We map the attr number to core id of the CPU
464 * The attr number is always core id + 2
465 * The Pkgtemp will always show up as temp1_*, if available
468 attr_no = PKG_SYSFS_ATTR_NO;
470 index = ida_alloc(&pdata->ida, GFP_KERNEL);
473 pdata->cpu_map[index] = topology_core_id(cpu);
474 attr_no = index + BASE_SYSFS_ATTR_NO;
477 if (attr_no > MAX_CORE_DATA - 1) {
482 tdata = init_temp_data(cpu, pkg_flag);
488 /* Test if we can access the status register */
489 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
493 /* We can access status register. Get Critical Temperature */
494 tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
497 * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
498 * The target temperature is available on older CPUs but not in this
499 * register. Atoms don't have the register at all.
501 if (c->x86_model > 0xe && c->x86_model != 0x1c) {
502 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
506 = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
511 pdata->core_data[attr_no] = tdata;
513 /* Create sysfs interfaces */
514 err = create_core_attrs(tdata, pdata->hwmon_dev, attr_no);
520 pdata->core_data[attr_no] = NULL;
524 ida_free(&pdata->ida, index);
529 coretemp_add_core(struct platform_device *pdev, unsigned int cpu, int pkg_flag)
531 if (create_core_data(pdev, cpu, pkg_flag))
532 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
535 static void coretemp_remove_core(struct platform_data *pdata, int indx)
537 struct temp_data *tdata = pdata->core_data[indx];
539 /* if we errored on add then this is already gone */
543 /* Remove the sysfs attributes */
544 sysfs_remove_group(&pdata->hwmon_dev->kobj, &tdata->attr_group);
546 kfree(pdata->core_data[indx]);
547 pdata->core_data[indx] = NULL;
549 if (indx >= BASE_SYSFS_ATTR_NO)
550 ida_free(&pdata->ida, indx - BASE_SYSFS_ATTR_NO);
553 static int coretemp_device_add(int zoneid)
555 struct platform_device *pdev;
556 struct platform_data *pdata;
559 /* Initialize the per-zone data structures */
560 pdata = kzalloc(sizeof(*pdata), GFP_KERNEL);
564 pdata->pkg_id = zoneid;
565 ida_init(&pdata->ida);
567 pdev = platform_device_alloc(DRVNAME, zoneid);
573 err = platform_device_add(pdev);
577 platform_set_drvdata(pdev, pdata);
578 zone_devices[zoneid] = pdev;
582 platform_device_put(pdev);
588 static void coretemp_device_remove(int zoneid)
590 struct platform_device *pdev = zone_devices[zoneid];
591 struct platform_data *pdata = platform_get_drvdata(pdev);
593 ida_destroy(&pdata->ida);
595 platform_device_unregister(pdev);
598 static int coretemp_cpu_online(unsigned int cpu)
600 struct platform_device *pdev = coretemp_get_pdev(cpu);
601 struct cpuinfo_x86 *c = &cpu_data(cpu);
602 struct platform_data *pdata;
605 * Don't execute this on resume as the offline callback did
606 * not get executed on suspend.
608 if (cpuhp_tasks_frozen)
612 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
613 * sensors. We check this bit only, all the early CPUs
614 * without thermal sensors will be filtered out.
616 if (!cpu_has(c, X86_FEATURE_DTHERM))
619 pdata = platform_get_drvdata(pdev);
620 if (!pdata->hwmon_dev) {
621 struct device *hwmon;
623 /* Check the microcode version of the CPU */
624 if (chk_ucode_version(cpu))
628 * Alright, we have DTS support.
629 * We are bringing the _first_ core in this pkg
630 * online. So, initialize per-pkg data structures and
631 * then bring this core online.
633 hwmon = hwmon_device_register_with_groups(&pdev->dev, DRVNAME,
636 return PTR_ERR(hwmon);
637 pdata->hwmon_dev = hwmon;
640 * Check whether pkgtemp support is available.
641 * If so, add interfaces for pkgtemp.
643 if (cpu_has(c, X86_FEATURE_PTS))
644 coretemp_add_core(pdev, cpu, 1);
648 * Check whether a thread sibling is already online. If not add the
649 * interface for this CPU core.
651 if (!cpumask_intersects(&pdata->cpumask, topology_sibling_cpumask(cpu)))
652 coretemp_add_core(pdev, cpu, 0);
654 cpumask_set_cpu(cpu, &pdata->cpumask);
658 static int coretemp_cpu_offline(unsigned int cpu)
660 struct platform_device *pdev = coretemp_get_pdev(cpu);
661 struct platform_data *pd;
662 struct temp_data *tdata;
663 int i, indx = -1, target;
665 /* No need to tear down any interfaces for suspend */
666 if (cpuhp_tasks_frozen)
669 /* If the physical CPU device does not exist, just return */
670 pd = platform_get_drvdata(pdev);
674 for (i = 0; i < NUM_REAL_CORES; i++) {
675 if (pd->cpu_map[i] == topology_core_id(cpu)) {
676 indx = i + BASE_SYSFS_ATTR_NO;
681 /* Too many cores and this core is not populated, just return */
685 tdata = pd->core_data[indx];
687 cpumask_clear_cpu(cpu, &pd->cpumask);
690 * If this is the last thread sibling, remove the CPU core
691 * interface, If there is still a sibling online, transfer the
692 * target cpu of that core interface to it.
694 target = cpumask_any_and(&pd->cpumask, topology_sibling_cpumask(cpu));
695 if (target >= nr_cpu_ids) {
696 coretemp_remove_core(pd, indx);
697 } else if (tdata && tdata->cpu == cpu) {
698 mutex_lock(&tdata->update_lock);
700 mutex_unlock(&tdata->update_lock);
704 * If all cores in this pkg are offline, remove the interface.
706 tdata = pd->core_data[PKG_SYSFS_ATTR_NO];
707 if (cpumask_empty(&pd->cpumask)) {
709 coretemp_remove_core(pd, PKG_SYSFS_ATTR_NO);
710 hwmon_device_unregister(pd->hwmon_dev);
711 pd->hwmon_dev = NULL;
716 * Check whether this core is the target for the package
717 * interface. We need to assign it to some other cpu.
719 if (tdata && tdata->cpu == cpu) {
720 target = cpumask_first(&pd->cpumask);
721 mutex_lock(&tdata->update_lock);
723 mutex_unlock(&tdata->update_lock);
727 static const struct x86_cpu_id __initconst coretemp_ids[] = {
728 X86_MATCH_VENDOR_FEATURE(INTEL, X86_FEATURE_DTHERM, NULL),
731 MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
733 static enum cpuhp_state coretemp_hp_online;
735 static int __init coretemp_init(void)
740 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
741 * sensors. We check this bit only, all the early CPUs
742 * without thermal sensors will be filtered out.
744 if (!x86_match_cpu(coretemp_ids))
747 max_zones = topology_max_packages() * topology_max_die_per_package();
748 zone_devices = kcalloc(max_zones, sizeof(struct platform_device *),
753 for (i = 0; i < max_zones; i++) {
754 err = coretemp_device_add(i);
759 err = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hwmon/coretemp:online",
760 coretemp_cpu_online, coretemp_cpu_offline);
763 coretemp_hp_online = err;
768 coretemp_device_remove(i);
772 module_init(coretemp_init)
774 static void __exit coretemp_exit(void)
778 cpuhp_remove_state(coretemp_hp_online);
779 for (i = 0; i < max_zones; i++)
780 coretemp_device_remove(i);
783 module_exit(coretemp_exit)
785 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
786 MODULE_DESCRIPTION("Intel Core temperature monitor");
787 MODULE_LICENSE("GPL");