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)
63 #define for_each_sibling(i, cpu) \
64 for_each_cpu(i, topology_sibling_cpumask(cpu))
66 #define for_each_sibling(i, cpu) for (i = 0; false; )
70 * Per-Core Temperature Data
71 * @last_updated: The time when the current temperature value was updated
72 * earlier (in jiffies).
73 * @cpu_core_id: The CPU Core from which temperature values should be read
74 * This value is passed as "id" field to rdmsr/wrmsr functions.
75 * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
76 * from where the temperature values should be read.
77 * @attr_size: Total number of pre-core attrs displayed in the sysfs.
78 * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
79 * Otherwise, temp_data holds coretemp data.
80 * @valid: If this is 1, the current temperature is valid.
86 unsigned long last_updated;
93 struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
94 char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
95 struct attribute *attrs[TOTAL_ATTRS + 1];
96 struct attribute_group attr_group;
97 struct mutex update_lock;
100 /* Platform Data per Physical CPU */
101 struct platform_data {
102 struct device *hwmon_dev;
104 u16 cpu_map[NUM_REAL_CORES];
106 struct cpumask cpumask;
107 struct temp_data *core_data[MAX_CORE_DATA];
108 struct device_attribute name_attr;
111 /* Keep track of how many package pointers we allocated in init() */
112 static int max_packages __read_mostly;
113 /* Array of package pointers. Serialized by cpu hotplug lock */
114 static struct platform_device **pkg_devices;
116 static ssize_t show_label(struct device *dev,
117 struct device_attribute *devattr, char *buf)
119 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
120 struct platform_data *pdata = dev_get_drvdata(dev);
121 struct temp_data *tdata = pdata->core_data[attr->index];
123 if (tdata->is_pkg_data)
124 return sprintf(buf, "Package id %u\n", pdata->pkg_id);
126 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
129 static ssize_t show_crit_alarm(struct device *dev,
130 struct device_attribute *devattr, char *buf)
133 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
134 struct platform_data *pdata = dev_get_drvdata(dev);
135 struct temp_data *tdata = pdata->core_data[attr->index];
137 mutex_lock(&tdata->update_lock);
138 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
139 mutex_unlock(&tdata->update_lock);
141 return sprintf(buf, "%d\n", (eax >> 5) & 1);
144 static ssize_t show_tjmax(struct device *dev,
145 struct device_attribute *devattr, char *buf)
147 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
148 struct platform_data *pdata = dev_get_drvdata(dev);
150 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
153 static ssize_t show_ttarget(struct device *dev,
154 struct device_attribute *devattr, char *buf)
156 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
157 struct platform_data *pdata = dev_get_drvdata(dev);
159 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
162 static ssize_t show_temp(struct device *dev,
163 struct device_attribute *devattr, char *buf)
166 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
167 struct platform_data *pdata = dev_get_drvdata(dev);
168 struct temp_data *tdata = pdata->core_data[attr->index];
170 mutex_lock(&tdata->update_lock);
172 /* Check whether the time interval has elapsed */
173 if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
174 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
176 * Ignore the valid bit. In all observed cases the register
177 * value is either low or zero if the valid bit is 0.
178 * Return it instead of reporting an error which doesn't
179 * really help at all.
181 tdata->temp = tdata->tjmax - ((eax >> 16) & 0x7f) * 1000;
183 tdata->last_updated = jiffies;
186 mutex_unlock(&tdata->update_lock);
187 return sprintf(buf, "%d\n", tdata->temp);
195 static const struct tjmax_pci tjmax_pci_table[] = {
196 { 0x0708, 110000 }, /* CE41x0 (Sodaville ) */
197 { 0x0c72, 102000 }, /* Atom S1240 (Centerton) */
198 { 0x0c73, 95000 }, /* Atom S1220 (Centerton) */
199 { 0x0c75, 95000 }, /* Atom S1260 (Centerton) */
207 static const struct tjmax tjmax_table[] = {
208 { "CPU 230", 100000 }, /* Model 0x1c, stepping 2 */
209 { "CPU 330", 125000 }, /* Model 0x1c, stepping 2 */
220 static const struct tjmax_model tjmax_model_table[] = {
221 { 0x1c, 10, 100000 }, /* D4xx, K4xx, N4xx, D5xx, K5xx, N5xx */
222 { 0x1c, ANY, 90000 }, /* Z5xx, N2xx, possibly others
223 * Note: Also matches 230 and 330,
224 * which are covered by tjmax_table
226 { 0x26, ANY, 90000 }, /* Atom Tunnel Creek (Exx), Lincroft (Z6xx)
227 * Note: TjMax for E6xxT is 110C, but CPU type
228 * is undetectable by software
230 { 0x27, ANY, 90000 }, /* Atom Medfield (Z2460) */
231 { 0x35, ANY, 90000 }, /* Atom Clover Trail/Cloverview (Z27x0) */
232 { 0x36, ANY, 100000 }, /* Atom Cedar Trail/Cedarview (N2xxx, D2xxx)
233 * Also matches S12x0 (stepping 9), covered by
238 static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
240 /* The 100C is default for both mobile and non mobile CPUs */
243 int tjmax_ee = 85000;
248 u16 devfn = PCI_DEVFN(0, 0);
249 struct pci_dev *host_bridge = pci_get_domain_bus_and_slot(0, 0, devfn);
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 return tjmax_pci_table[i].tjmax;
263 for (i = 0; i < ARRAY_SIZE(tjmax_table); i++) {
264 if (strstr(c->x86_model_id, tjmax_table[i].id))
265 return tjmax_table[i].tjmax;
268 for (i = 0; i < ARRAY_SIZE(tjmax_model_table); i++) {
269 const struct tjmax_model *tm = &tjmax_model_table[i];
270 if (c->x86_model == tm->model &&
271 (tm->mask == ANY || c->x86_stepping == tm->mask))
275 /* Early chips have no MSR for TjMax */
277 if (c->x86_model == 0xf && c->x86_stepping < 4)
280 if (c->x86_model > 0xe && usemsr_ee) {
284 * Now we can detect the mobile CPU using Intel provided table
285 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
286 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
288 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
291 "Unable to access MSR 0x17, assuming desktop"
294 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
296 * Trust bit 28 up to Penryn, I could not find any
297 * documentation on that; if you happen to know
298 * someone at Intel please ask
302 /* Platform ID bits 52:50 (EDX starts at bit 32) */
303 platform_id = (edx >> 18) & 0x7;
306 * Mobile Penryn CPU seems to be platform ID 7 or 5
309 if (c->x86_model == 0x17 &&
310 (platform_id == 5 || platform_id == 7)) {
312 * If MSR EE bit is set, set it to 90 degrees C,
313 * otherwise 105 degrees C
322 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
325 "Unable to access MSR 0xEE, for Tjmax, left"
327 } else if (eax & 0x40000000) {
330 } else if (tjmax == 100000) {
332 * If we don't use msr EE it means we are desktop CPU
333 * (with exeception of Atom)
335 dev_warn(dev, "Using relative temperature scale!\n");
341 static bool cpu_has_tjmax(struct cpuinfo_x86 *c)
343 u8 model = c->x86_model;
345 return model > 0xe &&
353 static int get_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
360 * A new feature of current Intel(R) processors, the
361 * IA32_TEMPERATURE_TARGET contains the TjMax value
363 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
365 if (cpu_has_tjmax(c))
366 dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
368 val = (eax >> 16) & 0xff;
370 * If the TjMax is not plausible, an assumption
374 dev_dbg(dev, "TjMax is %d degrees C\n", val);
380 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
382 return force_tjmax * 1000;
386 * An assumption is made for early CPUs and unreadable MSR.
387 * NOTE: the calculated value may not be correct.
389 return adjust_tjmax(c, id, dev);
392 static int create_core_attrs(struct temp_data *tdata, struct device *dev,
396 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
397 struct device_attribute *devattr, char *buf) = {
398 show_label, show_crit_alarm, show_temp, show_tjmax,
400 static const char *const suffixes[TOTAL_ATTRS] = {
401 "label", "crit_alarm", "input", "crit", "max"
404 for (i = 0; i < tdata->attr_size; i++) {
405 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH,
406 "temp%d_%s", attr_no, suffixes[i]);
407 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
408 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
409 tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
410 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
411 tdata->sd_attrs[i].index = attr_no;
412 tdata->attrs[i] = &tdata->sd_attrs[i].dev_attr.attr;
414 tdata->attr_group.attrs = tdata->attrs;
415 return sysfs_create_group(&dev->kobj, &tdata->attr_group);
419 static int chk_ucode_version(unsigned int cpu)
421 struct cpuinfo_x86 *c = &cpu_data(cpu);
424 * Check if we have problem with errata AE18 of Core processors:
425 * Readings might stop update when processor visited too deep sleep,
426 * fixed for stepping D0 (6EC).
428 if (c->x86_model == 0xe && c->x86_stepping < 0xc && c->microcode < 0x39) {
429 pr_err("Errata AE18 not fixed/*(DEBLOBBED)*/\n");
435 static struct platform_device *coretemp_get_pdev(unsigned int cpu)
437 int pkgid = topology_logical_package_id(cpu);
439 if (pkgid >= 0 && pkgid < max_packages)
440 return pkg_devices[pkgid];
444 static struct temp_data *init_temp_data(unsigned int cpu, int pkg_flag)
446 struct temp_data *tdata;
448 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
452 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
453 MSR_IA32_THERM_STATUS;
454 tdata->is_pkg_data = pkg_flag;
456 tdata->cpu_core_id = topology_core_id(cpu);
457 tdata->attr_size = MAX_CORE_ATTRS;
458 mutex_init(&tdata->update_lock);
462 static int create_core_data(struct platform_device *pdev, unsigned int cpu,
465 struct temp_data *tdata;
466 struct platform_data *pdata = platform_get_drvdata(pdev);
467 struct cpuinfo_x86 *c = &cpu_data(cpu);
469 int err, index, attr_no;
472 * Find attr number for sysfs:
473 * We map the attr number to core id of the CPU
474 * The attr number is always core id + 2
475 * The Pkgtemp will always show up as temp1_*, if available
478 attr_no = PKG_SYSFS_ATTR_NO;
480 index = ida_alloc(&pdata->ida, GFP_KERNEL);
483 pdata->cpu_map[index] = topology_core_id(cpu);
484 attr_no = index + BASE_SYSFS_ATTR_NO;
487 if (attr_no > MAX_CORE_DATA - 1) {
492 tdata = init_temp_data(cpu, pkg_flag);
498 /* Test if we can access the status register */
499 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
503 /* We can access status register. Get Critical Temperature */
504 tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
507 * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
508 * The target temperature is available on older CPUs but not in this
509 * register. Atoms don't have the register at all.
511 if (c->x86_model > 0xe && c->x86_model != 0x1c) {
512 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
516 = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
521 pdata->core_data[attr_no] = tdata;
523 /* Create sysfs interfaces */
524 err = create_core_attrs(tdata, pdata->hwmon_dev, attr_no);
530 pdata->core_data[attr_no] = NULL;
534 ida_free(&pdata->ida, index);
539 coretemp_add_core(struct platform_device *pdev, unsigned int cpu, int pkg_flag)
541 if (create_core_data(pdev, cpu, pkg_flag))
542 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
545 static void coretemp_remove_core(struct platform_data *pdata, int indx)
547 struct temp_data *tdata = pdata->core_data[indx];
549 /* Remove the sysfs attributes */
550 sysfs_remove_group(&pdata->hwmon_dev->kobj, &tdata->attr_group);
552 kfree(pdata->core_data[indx]);
553 pdata->core_data[indx] = NULL;
555 if (indx >= BASE_SYSFS_ATTR_NO)
556 ida_free(&pdata->ida, indx - BASE_SYSFS_ATTR_NO);
559 static int coretemp_probe(struct platform_device *pdev)
561 struct device *dev = &pdev->dev;
562 struct platform_data *pdata;
564 /* Initialize the per-package data structures */
565 pdata = devm_kzalloc(dev, sizeof(struct platform_data), GFP_KERNEL);
569 pdata->pkg_id = pdev->id;
570 ida_init(&pdata->ida);
571 platform_set_drvdata(pdev, pdata);
573 pdata->hwmon_dev = devm_hwmon_device_register_with_groups(dev, DRVNAME,
575 return PTR_ERR_OR_ZERO(pdata->hwmon_dev);
578 static int coretemp_remove(struct platform_device *pdev)
580 struct platform_data *pdata = platform_get_drvdata(pdev);
583 for (i = MAX_CORE_DATA - 1; i >= 0; --i)
584 if (pdata->core_data[i])
585 coretemp_remove_core(pdata, i);
587 ida_destroy(&pdata->ida);
591 static struct platform_driver coretemp_driver = {
595 .probe = coretemp_probe,
596 .remove = coretemp_remove,
599 static struct platform_device *coretemp_device_add(unsigned int cpu)
601 int err, pkgid = topology_logical_package_id(cpu);
602 struct platform_device *pdev;
605 return ERR_PTR(-ENOMEM);
607 pdev = platform_device_alloc(DRVNAME, pkgid);
609 return ERR_PTR(-ENOMEM);
611 err = platform_device_add(pdev);
613 platform_device_put(pdev);
617 pkg_devices[pkgid] = pdev;
621 static int coretemp_cpu_online(unsigned int cpu)
623 struct platform_device *pdev = coretemp_get_pdev(cpu);
624 struct cpuinfo_x86 *c = &cpu_data(cpu);
625 struct platform_data *pdata;
628 * Don't execute this on resume as the offline callback did
629 * not get executed on suspend.
631 if (cpuhp_tasks_frozen)
635 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
636 * sensors. We check this bit only, all the early CPUs
637 * without thermal sensors will be filtered out.
639 if (!cpu_has(c, X86_FEATURE_DTHERM))
643 /* Check the microcode version of the CPU */
644 if (chk_ucode_version(cpu))
648 * Alright, we have DTS support.
649 * We are bringing the _first_ core in this pkg
650 * online. So, initialize per-pkg data structures and
651 * then bring this core online.
653 pdev = coretemp_device_add(cpu);
655 return PTR_ERR(pdev);
658 * Check whether pkgtemp support is available.
659 * If so, add interfaces for pkgtemp.
661 if (cpu_has(c, X86_FEATURE_PTS))
662 coretemp_add_core(pdev, cpu, 1);
665 pdata = platform_get_drvdata(pdev);
667 * Check whether a thread sibling is already online. If not add the
668 * interface for this CPU core.
670 if (!cpumask_intersects(&pdata->cpumask, topology_sibling_cpumask(cpu)))
671 coretemp_add_core(pdev, cpu, 0);
673 cpumask_set_cpu(cpu, &pdata->cpumask);
677 static int coretemp_cpu_offline(unsigned int cpu)
679 struct platform_device *pdev = coretemp_get_pdev(cpu);
680 struct platform_data *pd;
681 struct temp_data *tdata;
682 int i, indx = -1, target;
685 * Don't execute this on suspend as the device remove locks
688 if (cpuhp_tasks_frozen)
691 /* If the physical CPU device does not exist, just return */
695 pd = platform_get_drvdata(pdev);
697 for (i = 0; i < NUM_REAL_CORES; i++) {
698 if (pd->cpu_map[i] == topology_core_id(cpu)) {
699 indx = i + BASE_SYSFS_ATTR_NO;
704 /* Too many cores and this core is not populated, just return */
708 tdata = pd->core_data[indx];
710 cpumask_clear_cpu(cpu, &pd->cpumask);
713 * If this is the last thread sibling, remove the CPU core
714 * interface, If there is still a sibling online, transfer the
715 * target cpu of that core interface to it.
717 target = cpumask_any_and(&pd->cpumask, topology_sibling_cpumask(cpu));
718 if (target >= nr_cpu_ids) {
719 coretemp_remove_core(pd, indx);
720 } else if (tdata && tdata->cpu == cpu) {
721 mutex_lock(&tdata->update_lock);
723 mutex_unlock(&tdata->update_lock);
727 * If all cores in this pkg are offline, remove the device. This
728 * will invoke the platform driver remove function, which cleans up
731 if (cpumask_empty(&pd->cpumask)) {
732 pkg_devices[topology_logical_package_id(cpu)] = NULL;
733 platform_device_unregister(pdev);
738 * Check whether this core is the target for the package
739 * interface. We need to assign it to some other cpu.
741 tdata = pd->core_data[PKG_SYSFS_ATTR_NO];
742 if (tdata && tdata->cpu == cpu) {
743 target = cpumask_first(&pd->cpumask);
744 mutex_lock(&tdata->update_lock);
746 mutex_unlock(&tdata->update_lock);
750 static const struct x86_cpu_id __initconst coretemp_ids[] = {
751 { X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_DTHERM },
754 MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
756 static enum cpuhp_state coretemp_hp_online;
758 static int __init coretemp_init(void)
763 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
764 * sensors. We check this bit only, all the early CPUs
765 * without thermal sensors will be filtered out.
767 if (!x86_match_cpu(coretemp_ids))
770 max_packages = topology_max_packages();
771 pkg_devices = kcalloc(max_packages, sizeof(struct platform_device *),
776 err = platform_driver_register(&coretemp_driver);
780 err = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hwmon/coretemp:online",
781 coretemp_cpu_online, coretemp_cpu_offline);
784 coretemp_hp_online = err;
788 platform_driver_unregister(&coretemp_driver);
792 module_init(coretemp_init)
794 static void __exit coretemp_exit(void)
796 cpuhp_remove_state(coretemp_hp_online);
797 platform_driver_unregister(&coretemp_driver);
800 module_exit(coretemp_exit)
802 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
803 MODULE_DESCRIPTION("Intel Core temperature monitor");
804 MODULE_LICENSE("GPL");