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/pci.h>
38 #include <linux/smp.h>
40 #include <asm/processor.h>
42 #define DRVNAME "coretemp"
44 #define BASE_SYSFS_ATTR_NO 2 /* Sysfs Base attr no for coretemp */
45 #define NUM_REAL_CORES 16 /* Number of Real cores per cpu */
46 #define CORETEMP_NAME_LENGTH 17 /* String Length of attrs */
47 #define MAX_ATTRS 5 /* Maximum no of per-core attrs */
48 #define MAX_CORE_DATA (NUM_REAL_CORES + BASE_SYSFS_ATTR_NO)
51 #define TO_PHYS_ID(cpu) cpu_data(cpu).phys_proc_id
52 #define TO_CORE_ID(cpu) cpu_data(cpu).cpu_core_id
53 #define TO_ATTR_NO(cpu) (TO_CORE_ID(cpu) + BASE_SYSFS_ATTR_NO)
54 #define for_each_sibling(i, cpu) for_each_cpu(i, cpu_sibling_mask(cpu))
56 #define TO_PHYS_ID(cpu) (cpu)
57 #define TO_CORE_ID(cpu) (cpu)
58 #define TO_ATTR_NO(cpu) (cpu)
59 #define for_each_sibling(i, cpu) for (i = 0; false; )
63 * Per-Core Temperature Data
64 * @last_updated: The time when the current temperature value was updated
65 * earlier (in jiffies).
66 * @cpu_core_id: The CPU Core from which temperature values should be read
67 * This value is passed as "id" field to rdmsr/wrmsr functions.
68 * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
69 * from where the temperature values should be read.
70 * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
71 * Otherwise, temp_data holds coretemp data.
72 * @valid: If this is 1, the current temperature is valid.
78 unsigned long last_updated;
84 struct sensor_device_attribute sd_attrs[MAX_ATTRS];
85 char attr_name[MAX_ATTRS][CORETEMP_NAME_LENGTH];
86 struct mutex update_lock;
89 /* Platform Data per Physical CPU */
90 struct platform_data {
91 struct device *hwmon_dev;
93 struct temp_data *core_data[MAX_CORE_DATA];
94 struct device_attribute name_attr;
98 struct list_head list;
99 struct platform_device *pdev;
105 static LIST_HEAD(pdev_list);
106 static DEFINE_MUTEX(pdev_list_mutex);
108 static ssize_t show_name(struct device *dev,
109 struct device_attribute *devattr, char *buf)
111 return sprintf(buf, "%s\n", DRVNAME);
114 static ssize_t show_label(struct device *dev,
115 struct device_attribute *devattr, char *buf)
117 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
118 struct platform_data *pdata = dev_get_drvdata(dev);
119 struct temp_data *tdata = pdata->core_data[attr->index];
121 if (tdata->is_pkg_data)
122 return sprintf(buf, "Physical id %u\n", pdata->phys_proc_id);
124 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
127 static ssize_t show_crit_alarm(struct device *dev,
128 struct device_attribute *devattr, char *buf)
131 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
132 struct platform_data *pdata = dev_get_drvdata(dev);
133 struct temp_data *tdata = pdata->core_data[attr->index];
135 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
137 return sprintf(buf, "%d\n", (eax >> 5) & 1);
140 static ssize_t show_tjmax(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]->tjmax);
149 static ssize_t show_ttarget(struct device *dev,
150 struct device_attribute *devattr, char *buf)
152 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
153 struct platform_data *pdata = dev_get_drvdata(dev);
155 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
158 static ssize_t show_temp(struct device *dev,
159 struct device_attribute *devattr, char *buf)
162 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
163 struct platform_data *pdata = dev_get_drvdata(dev);
164 struct temp_data *tdata = pdata->core_data[attr->index];
166 mutex_lock(&tdata->update_lock);
168 /* Check whether the time interval has elapsed */
169 if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
170 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
172 /* Check whether the data is valid */
173 if (eax & 0x80000000) {
174 tdata->temp = tdata->tjmax -
175 ((eax >> 16) & 0x7f) * 1000;
178 tdata->last_updated = jiffies;
181 mutex_unlock(&tdata->update_lock);
182 return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
185 static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
187 /* The 100C is default for both mobile and non mobile CPUs */
190 int tjmax_ee = 85000;
194 struct pci_dev *host_bridge;
196 /* Early chips have no MSR for TjMax */
198 if (c->x86_model == 0xf && c->x86_mask < 4)
203 if (c->x86_model == 0x1c) {
206 host_bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));
208 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL
209 && (host_bridge->device == 0xa000 /* NM10 based nettop */
210 || host_bridge->device == 0xa010)) /* NM10 based netbook */
215 pci_dev_put(host_bridge);
218 if (c->x86_model > 0xe && usemsr_ee) {
222 * Now we can detect the mobile CPU using Intel provided table
223 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
224 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
226 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
229 "Unable to access MSR 0x17, assuming desktop"
232 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
234 * Trust bit 28 up to Penryn, I could not find any
235 * documentation on that; if you happen to know
236 * someone at Intel please ask
240 /* Platform ID bits 52:50 (EDX starts at bit 32) */
241 platform_id = (edx >> 18) & 0x7;
244 * Mobile Penryn CPU seems to be platform ID 7 or 5
247 if (c->x86_model == 0x17 &&
248 (platform_id == 5 || platform_id == 7)) {
250 * If MSR EE bit is set, set it to 90 degrees C,
251 * otherwise 105 degrees C
260 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
263 "Unable to access MSR 0xEE, for Tjmax, left"
265 } else if (eax & 0x40000000) {
268 } else if (tjmax == 100000) {
270 * If we don't use msr EE it means we are desktop CPU
271 * (with exeception of Atom)
273 dev_warn(dev, "Using relative temperature scale!\n");
279 static int get_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
281 /* The 100C is default for both mobile and non mobile CPUs */
287 * A new feature of current Intel(R) processors, the
288 * IA32_TEMPERATURE_TARGET contains the TjMax value
290 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
292 dev_warn(dev, "Unable to read TjMax from CPU.\n");
294 val = (eax >> 16) & 0xff;
296 * If the TjMax is not plausible, an assumption
299 if (val > 80 && val < 120) {
300 dev_info(dev, "TjMax is %d C.\n", val);
306 * An assumption is made for early CPUs and unreadable MSR.
307 * NOTE: the given value may not be correct.
310 switch (c->x86_model) {
315 dev_warn(dev, "TjMax is assumed as 100 C!\n");
318 case 0x1c: /* Atom CPUs */
319 return adjust_tjmax(c, id, dev);
321 dev_warn(dev, "CPU (model=0x%x) is not supported yet,"
322 " using default TjMax of 100C.\n", c->x86_model);
327 static void __devinit get_ucode_rev_on_cpu(void *edx)
331 wrmsr(MSR_IA32_UCODE_REV, 0, 0);
333 rdmsr(MSR_IA32_UCODE_REV, eax, *(u32 *)edx);
336 static int get_pkg_tjmax(unsigned int cpu, struct device *dev)
341 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
343 val = (eax >> 16) & 0xff;
344 if (val > 80 && val < 120)
347 dev_warn(dev, "Unable to read Pkg-TjMax from CPU:%u\n", cpu);
348 return 100000; /* Default TjMax: 100 degree celsius */
351 static int create_name_attr(struct platform_data *pdata, struct device *dev)
353 sysfs_attr_init(&pdata->name_attr.attr);
354 pdata->name_attr.attr.name = "name";
355 pdata->name_attr.attr.mode = S_IRUGO;
356 pdata->name_attr.show = show_name;
357 return device_create_file(dev, &pdata->name_attr);
360 static int create_core_attrs(struct temp_data *tdata, struct device *dev,
364 static ssize_t (*rd_ptr[MAX_ATTRS]) (struct device *dev,
365 struct device_attribute *devattr, char *buf) = {
366 show_label, show_crit_alarm, show_ttarget,
367 show_temp, show_tjmax };
368 static const char *names[MAX_ATTRS] = {
369 "temp%d_label", "temp%d_crit_alarm",
370 "temp%d_max", "temp%d_input",
373 for (i = 0; i < MAX_ATTRS; i++) {
374 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH, names[i],
376 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
377 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
378 tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
379 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
380 tdata->sd_attrs[i].dev_attr.store = NULL;
381 tdata->sd_attrs[i].index = attr_no;
382 err = device_create_file(dev, &tdata->sd_attrs[i].dev_attr);
390 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
394 static void update_ttarget(__u8 cpu_model, struct temp_data *tdata,
401 * Initialize ttarget value. Eventually this will be
402 * initialized with the value from MSR_IA32_THERM_INTERRUPT
403 * register. If IA32_TEMPERATURE_TARGET is supported, this
404 * value will be over written below.
405 * To Do: Patch to initialize ttarget from MSR_IA32_THERM_INTERRUPT
407 tdata->ttarget = tdata->tjmax - 20000;
410 * Read the still undocumented IA32_TEMPERATURE_TARGET. It exists
411 * on older CPUs but not in this register,
412 * Atoms don't have it either.
414 if (cpu_model > 0xe && cpu_model != 0x1c) {
415 err = rdmsr_safe_on_cpu(tdata->cpu,
416 MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
419 "Unable to read IA32_TEMPERATURE_TARGET MSR\n");
421 tdata->ttarget = tdata->tjmax -
422 ((eax >> 8) & 0xff) * 1000;
427 static int __devinit chk_ucode_version(struct platform_device *pdev)
429 struct cpuinfo_x86 *c = &cpu_data(pdev->id);
434 * Check if we have problem with errata AE18 of Core processors:
435 * Readings might stop update when processor visited too deep sleep,
436 * fixed for stepping D0 (6EC).
438 if (c->x86_model == 0xe && c->x86_mask < 0xc) {
439 /* check for microcode update */
440 err = smp_call_function_single(pdev->id, get_ucode_rev_on_cpu,
444 "Cannot determine microcode revision of "
445 "CPU#%u (%d)!\n", pdev->id, err);
447 } else if (edx < 0x39) {
449 "Errata AE18 not fixed, update BIOS or "
450 "microcode of the CPU!\n");
457 static struct platform_device *coretemp_get_pdev(unsigned int cpu)
459 u16 phys_proc_id = TO_PHYS_ID(cpu);
460 struct pdev_entry *p;
462 mutex_lock(&pdev_list_mutex);
464 list_for_each_entry(p, &pdev_list, list)
465 if (p->phys_proc_id == phys_proc_id) {
466 mutex_unlock(&pdev_list_mutex);
470 mutex_unlock(&pdev_list_mutex);
474 static struct temp_data *init_temp_data(unsigned int cpu, int pkg_flag)
476 struct temp_data *tdata;
478 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
482 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
483 MSR_IA32_THERM_STATUS;
484 tdata->is_pkg_data = pkg_flag;
486 tdata->cpu_core_id = TO_CORE_ID(cpu);
487 mutex_init(&tdata->update_lock);
491 static int create_core_data(struct platform_data *pdata,
492 struct platform_device *pdev,
493 unsigned int cpu, int pkg_flag)
495 struct temp_data *tdata;
496 struct cpuinfo_x86 *c = &cpu_data(cpu);
501 * Find attr number for sysfs:
502 * We map the attr number to core id of the CPU
503 * The attr number is always core id + 2
504 * The Pkgtemp will always show up as temp1_*, if available
506 attr_no = pkg_flag ? 1 : TO_ATTR_NO(cpu);
508 if (attr_no > MAX_CORE_DATA - 1)
512 * Provide a single set of attributes for all HT siblings of a core
513 * to avoid duplicate sensors (the processor ID and core ID of all
514 * HT siblings of a core are the same).
515 * Skip if a HT sibling of this core is already registered.
516 * This is not an error.
518 if (pdata->core_data[attr_no] != NULL)
521 tdata = init_temp_data(cpu, pkg_flag);
525 /* Test if we can access the status register */
526 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
530 /* We can access status register. Get Critical Temperature */
532 tdata->tjmax = get_pkg_tjmax(pdev->id, &pdev->dev);
534 tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
536 update_ttarget(c->x86_model, tdata, &pdev->dev);
537 pdata->core_data[attr_no] = tdata;
539 /* Create sysfs interfaces */
540 err = create_core_attrs(tdata, &pdev->dev, attr_no);
550 static void coretemp_add_core(unsigned int cpu, int pkg_flag)
552 struct platform_data *pdata;
553 struct platform_device *pdev = coretemp_get_pdev(cpu);
559 pdata = platform_get_drvdata(pdev);
561 err = create_core_data(pdata, pdev, cpu, pkg_flag);
563 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
566 static void coretemp_remove_core(struct platform_data *pdata,
567 struct device *dev, int indx)
570 struct temp_data *tdata = pdata->core_data[indx];
572 /* Remove the sysfs attributes */
573 for (i = 0; i < MAX_ATTRS; i++)
574 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
576 kfree(pdata->core_data[indx]);
577 pdata->core_data[indx] = NULL;
580 static int __devinit coretemp_probe(struct platform_device *pdev)
582 struct platform_data *pdata;
585 /* Check the microcode version of the CPU */
586 err = chk_ucode_version(pdev);
590 /* Initialize the per-package data structures */
591 pdata = kzalloc(sizeof(struct platform_data), GFP_KERNEL);
595 err = create_name_attr(pdata, &pdev->dev);
599 pdata->phys_proc_id = TO_PHYS_ID(pdev->id);
600 platform_set_drvdata(pdev, pdata);
602 pdata->hwmon_dev = hwmon_device_register(&pdev->dev);
603 if (IS_ERR(pdata->hwmon_dev)) {
604 err = PTR_ERR(pdata->hwmon_dev);
605 dev_err(&pdev->dev, "Class registration failed (%d)\n", err);
611 device_remove_file(&pdev->dev, &pdata->name_attr);
612 platform_set_drvdata(pdev, NULL);
618 static int __devexit coretemp_remove(struct platform_device *pdev)
620 struct platform_data *pdata = platform_get_drvdata(pdev);
623 for (i = MAX_CORE_DATA - 1; i >= 0; --i)
624 if (pdata->core_data[i])
625 coretemp_remove_core(pdata, &pdev->dev, i);
627 device_remove_file(&pdev->dev, &pdata->name_attr);
628 hwmon_device_unregister(pdata->hwmon_dev);
629 platform_set_drvdata(pdev, NULL);
634 static struct platform_driver coretemp_driver = {
636 .owner = THIS_MODULE,
639 .probe = coretemp_probe,
640 .remove = __devexit_p(coretemp_remove),
643 static int __cpuinit coretemp_device_add(unsigned int cpu)
646 struct platform_device *pdev;
647 struct pdev_entry *pdev_entry;
649 mutex_lock(&pdev_list_mutex);
651 pdev = platform_device_alloc(DRVNAME, cpu);
654 pr_err("Device allocation failed\n");
658 pdev_entry = kzalloc(sizeof(struct pdev_entry), GFP_KERNEL);
661 goto exit_device_put;
664 err = platform_device_add(pdev);
666 pr_err("Device addition failed (%d)\n", err);
667 goto exit_device_free;
670 pdev_entry->pdev = pdev;
671 pdev_entry->cpu = cpu;
672 pdev_entry->phys_proc_id = TO_PHYS_ID(cpu);
673 pdev_entry->cpu_core_id = TO_CORE_ID(cpu);
675 list_add_tail(&pdev_entry->list, &pdev_list);
676 mutex_unlock(&pdev_list_mutex);
683 platform_device_put(pdev);
685 mutex_unlock(&pdev_list_mutex);
689 static void coretemp_device_remove(unsigned int cpu)
691 struct pdev_entry *p, *n;
692 u16 phys_proc_id = TO_PHYS_ID(cpu);
694 mutex_lock(&pdev_list_mutex);
695 list_for_each_entry_safe(p, n, &pdev_list, list) {
696 if (p->phys_proc_id != phys_proc_id)
698 platform_device_unregister(p->pdev);
702 mutex_unlock(&pdev_list_mutex);
705 static bool is_any_core_online(struct platform_data *pdata)
709 /* Find online cores, except pkgtemp data */
710 for (i = MAX_CORE_DATA - 1; i >= 0; --i) {
711 if (pdata->core_data[i] &&
712 !pdata->core_data[i]->is_pkg_data) {
719 static void __cpuinit get_core_online(unsigned int cpu)
721 struct cpuinfo_x86 *c = &cpu_data(cpu);
722 struct platform_device *pdev = coretemp_get_pdev(cpu);
726 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
727 * sensors. We check this bit only, all the early CPUs
728 * without thermal sensors will be filtered out.
730 if (!cpu_has(c, X86_FEATURE_DTS))
735 * Alright, we have DTS support.
736 * We are bringing the _first_ core in this pkg
737 * online. So, initialize per-pkg data structures and
738 * then bring this core online.
740 err = coretemp_device_add(cpu);
744 * Check whether pkgtemp support is available.
745 * If so, add interfaces for pkgtemp.
747 if (cpu_has(c, X86_FEATURE_PTS))
748 coretemp_add_core(cpu, 1);
751 * Physical CPU device already exists.
752 * So, just add interfaces for this core.
754 coretemp_add_core(cpu, 0);
757 static void __cpuinit put_core_offline(unsigned int cpu)
760 struct platform_data *pdata;
761 struct platform_device *pdev = coretemp_get_pdev(cpu);
763 /* If the physical CPU device does not exist, just return */
767 pdata = platform_get_drvdata(pdev);
769 indx = TO_ATTR_NO(cpu);
771 if (pdata->core_data[indx] && pdata->core_data[indx]->cpu == cpu)
772 coretemp_remove_core(pdata, &pdev->dev, indx);
775 * If a HT sibling of a core is taken offline, but another HT sibling
776 * of the same core is still online, register the alternate sibling.
777 * This ensures that exactly one set of attributes is provided as long
778 * as at least one HT sibling of a core is online.
780 for_each_sibling(i, cpu) {
784 * Display temperature sensor data for one HT sibling
785 * per core only, so abort the loop after one such
786 * sibling has been found.
792 * If all cores in this pkg are offline, remove the device.
793 * coretemp_device_remove calls unregister_platform_device,
794 * which in turn calls coretemp_remove. This removes the
795 * pkgtemp entry and does other clean ups.
797 if (!is_any_core_online(pdata))
798 coretemp_device_remove(cpu);
801 static int __cpuinit coretemp_cpu_callback(struct notifier_block *nfb,
802 unsigned long action, void *hcpu)
804 unsigned int cpu = (unsigned long) hcpu;
808 case CPU_DOWN_FAILED:
809 get_core_online(cpu);
811 case CPU_DOWN_PREPARE:
812 put_core_offline(cpu);
818 static struct notifier_block coretemp_cpu_notifier __refdata = {
819 .notifier_call = coretemp_cpu_callback,
822 static int __init coretemp_init(void)
824 int i, err = -ENODEV;
826 /* quick check if we run Intel */
827 if (cpu_data(0).x86_vendor != X86_VENDOR_INTEL)
830 err = platform_driver_register(&coretemp_driver);
834 for_each_online_cpu(i)
837 #ifndef CONFIG_HOTPLUG_CPU
838 if (list_empty(&pdev_list)) {
840 goto exit_driver_unreg;
844 register_hotcpu_notifier(&coretemp_cpu_notifier);
847 #ifndef CONFIG_HOTPLUG_CPU
849 platform_driver_unregister(&coretemp_driver);
855 static void __exit coretemp_exit(void)
857 struct pdev_entry *p, *n;
859 unregister_hotcpu_notifier(&coretemp_cpu_notifier);
860 mutex_lock(&pdev_list_mutex);
861 list_for_each_entry_safe(p, n, &pdev_list, list) {
862 platform_device_unregister(p->pdev);
866 mutex_unlock(&pdev_list_mutex);
867 platform_driver_unregister(&coretemp_driver);
870 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
871 MODULE_DESCRIPTION("Intel Core temperature monitor");
872 MODULE_LICENSE("GPL");
874 module_init(coretemp_init)
875 module_exit(coretemp_exit)