Loading...
1/*
2 * coretemp.c - Linux kernel module for hardware monitoring
3 *
4 * Copyright (C) 2007 Rudolf Marek <r.marek@assembler.cz>
5 *
6 * Inspired from many hwmon drivers
7 *
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.
11 *
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.
16 *
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
20 * 02110-1301 USA.
21 */
22
23#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24
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>
39#include <linux/moduleparam.h>
40#include <asm/msr.h>
41#include <asm/processor.h>
42
43#define DRVNAME "coretemp"
44
45/*
46 * force_tjmax only matters when TjMax can't be read from the CPU itself.
47 * When set, it replaces the driver's suboptimal heuristic.
48 */
49static int force_tjmax;
50module_param_named(tjmax, force_tjmax, int, 0444);
51MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
52
53#define BASE_SYSFS_ATTR_NO 2 /* Sysfs Base attr no for coretemp */
54#define NUM_REAL_CORES 16 /* Number of Real cores per cpu */
55#define CORETEMP_NAME_LENGTH 17 /* String Length of attrs */
56#define MAX_CORE_ATTRS 4 /* Maximum no of basic attrs */
57#define TOTAL_ATTRS (MAX_CORE_ATTRS + 1)
58#define MAX_CORE_DATA (NUM_REAL_CORES + BASE_SYSFS_ATTR_NO)
59
60#ifdef CONFIG_SMP
61#define TO_PHYS_ID(cpu) cpu_data(cpu).phys_proc_id
62#define TO_CORE_ID(cpu) cpu_data(cpu).cpu_core_id
63#define TO_ATTR_NO(cpu) (TO_CORE_ID(cpu) + BASE_SYSFS_ATTR_NO)
64#define for_each_sibling(i, cpu) for_each_cpu(i, cpu_sibling_mask(cpu))
65#else
66#define TO_PHYS_ID(cpu) (cpu)
67#define TO_CORE_ID(cpu) (cpu)
68#define TO_ATTR_NO(cpu) (cpu)
69#define for_each_sibling(i, cpu) for (i = 0; false; )
70#endif
71
72/*
73 * Per-Core Temperature Data
74 * @last_updated: The time when the current temperature value was updated
75 * earlier (in jiffies).
76 * @cpu_core_id: The CPU Core from which temperature values should be read
77 * This value is passed as "id" field to rdmsr/wrmsr functions.
78 * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
79 * from where the temperature values should be read.
80 * @attr_size: Total number of pre-core attrs displayed in the sysfs.
81 * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
82 * Otherwise, temp_data holds coretemp data.
83 * @valid: If this is 1, the current temperature is valid.
84 */
85struct temp_data {
86 int temp;
87 int ttarget;
88 int tjmax;
89 unsigned long last_updated;
90 unsigned int cpu;
91 u32 cpu_core_id;
92 u32 status_reg;
93 int attr_size;
94 bool is_pkg_data;
95 bool valid;
96 struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
97 char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
98 struct mutex update_lock;
99};
100
101/* Platform Data per Physical CPU */
102struct platform_data {
103 struct device *hwmon_dev;
104 u16 phys_proc_id;
105 struct temp_data *core_data[MAX_CORE_DATA];
106 struct device_attribute name_attr;
107};
108
109struct pdev_entry {
110 struct list_head list;
111 struct platform_device *pdev;
112 u16 phys_proc_id;
113};
114
115static LIST_HEAD(pdev_list);
116static DEFINE_MUTEX(pdev_list_mutex);
117
118static ssize_t show_name(struct device *dev,
119 struct device_attribute *devattr, char *buf)
120{
121 return sprintf(buf, "%s\n", DRVNAME);
122}
123
124static ssize_t show_label(struct device *dev,
125 struct device_attribute *devattr, char *buf)
126{
127 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
128 struct platform_data *pdata = dev_get_drvdata(dev);
129 struct temp_data *tdata = pdata->core_data[attr->index];
130
131 if (tdata->is_pkg_data)
132 return sprintf(buf, "Physical id %u\n", pdata->phys_proc_id);
133
134 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
135}
136
137static ssize_t show_crit_alarm(struct device *dev,
138 struct device_attribute *devattr, char *buf)
139{
140 u32 eax, edx;
141 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
142 struct platform_data *pdata = dev_get_drvdata(dev);
143 struct temp_data *tdata = pdata->core_data[attr->index];
144
145 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
146
147 return sprintf(buf, "%d\n", (eax >> 5) & 1);
148}
149
150static ssize_t show_tjmax(struct device *dev,
151 struct device_attribute *devattr, char *buf)
152{
153 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
154 struct platform_data *pdata = dev_get_drvdata(dev);
155
156 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
157}
158
159static ssize_t show_ttarget(struct device *dev,
160 struct device_attribute *devattr, char *buf)
161{
162 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
163 struct platform_data *pdata = dev_get_drvdata(dev);
164
165 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
166}
167
168static ssize_t show_temp(struct device *dev,
169 struct device_attribute *devattr, char *buf)
170{
171 u32 eax, edx;
172 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
173 struct platform_data *pdata = dev_get_drvdata(dev);
174 struct temp_data *tdata = pdata->core_data[attr->index];
175
176 mutex_lock(&tdata->update_lock);
177
178 /* Check whether the time interval has elapsed */
179 if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
180 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
181 tdata->valid = 0;
182 /* Check whether the data is valid */
183 if (eax & 0x80000000) {
184 tdata->temp = tdata->tjmax -
185 ((eax >> 16) & 0x7f) * 1000;
186 tdata->valid = 1;
187 }
188 tdata->last_updated = jiffies;
189 }
190
191 mutex_unlock(&tdata->update_lock);
192 return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
193}
194
195static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
196{
197 /* The 100C is default for both mobile and non mobile CPUs */
198
199 int tjmax = 100000;
200 int tjmax_ee = 85000;
201 int usemsr_ee = 1;
202 int err;
203 u32 eax, edx;
204 struct pci_dev *host_bridge;
205
206 /* Early chips have no MSR for TjMax */
207
208 if (c->x86_model == 0xf && c->x86_mask < 4)
209 usemsr_ee = 0;
210
211 /* Atom CPUs */
212
213 if (c->x86_model == 0x1c) {
214 usemsr_ee = 0;
215
216 host_bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));
217
218 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL
219 && (host_bridge->device == 0xa000 /* NM10 based nettop */
220 || host_bridge->device == 0xa010)) /* NM10 based netbook */
221 tjmax = 100000;
222 else
223 tjmax = 90000;
224
225 pci_dev_put(host_bridge);
226 }
227
228 if (c->x86_model > 0xe && usemsr_ee) {
229 u8 platform_id;
230
231 /*
232 * Now we can detect the mobile CPU using Intel provided table
233 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
234 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
235 */
236 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
237 if (err) {
238 dev_warn(dev,
239 "Unable to access MSR 0x17, assuming desktop"
240 " CPU\n");
241 usemsr_ee = 0;
242 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
243 /*
244 * Trust bit 28 up to Penryn, I could not find any
245 * documentation on that; if you happen to know
246 * someone at Intel please ask
247 */
248 usemsr_ee = 0;
249 } else {
250 /* Platform ID bits 52:50 (EDX starts at bit 32) */
251 platform_id = (edx >> 18) & 0x7;
252
253 /*
254 * Mobile Penryn CPU seems to be platform ID 7 or 5
255 * (guesswork)
256 */
257 if (c->x86_model == 0x17 &&
258 (platform_id == 5 || platform_id == 7)) {
259 /*
260 * If MSR EE bit is set, set it to 90 degrees C,
261 * otherwise 105 degrees C
262 */
263 tjmax_ee = 90000;
264 tjmax = 105000;
265 }
266 }
267 }
268
269 if (usemsr_ee) {
270 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
271 if (err) {
272 dev_warn(dev,
273 "Unable to access MSR 0xEE, for Tjmax, left"
274 " at default\n");
275 } else if (eax & 0x40000000) {
276 tjmax = tjmax_ee;
277 }
278 } else if (tjmax == 100000) {
279 /*
280 * If we don't use msr EE it means we are desktop CPU
281 * (with exeception of Atom)
282 */
283 dev_warn(dev, "Using relative temperature scale!\n");
284 }
285
286 return tjmax;
287}
288
289static int get_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
290{
291 int err;
292 u32 eax, edx;
293 u32 val;
294
295 /*
296 * A new feature of current Intel(R) processors, the
297 * IA32_TEMPERATURE_TARGET contains the TjMax value
298 */
299 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
300 if (err) {
301 if (c->x86_model > 0xe && c->x86_model != 0x1c)
302 dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
303 } else {
304 val = (eax >> 16) & 0xff;
305 /*
306 * If the TjMax is not plausible, an assumption
307 * will be used
308 */
309 if (val) {
310 dev_dbg(dev, "TjMax is %d degrees C\n", val);
311 return val * 1000;
312 }
313 }
314
315 if (force_tjmax) {
316 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
317 force_tjmax);
318 return force_tjmax * 1000;
319 }
320
321 /*
322 * An assumption is made for early CPUs and unreadable MSR.
323 * NOTE: the calculated value may not be correct.
324 */
325 return adjust_tjmax(c, id, dev);
326}
327
328static void __devinit get_ucode_rev_on_cpu(void *edx)
329{
330 u32 eax;
331
332 wrmsr(MSR_IA32_UCODE_REV, 0, 0);
333 sync_core();
334 rdmsr(MSR_IA32_UCODE_REV, eax, *(u32 *)edx);
335}
336
337static int create_name_attr(struct platform_data *pdata, struct device *dev)
338{
339 sysfs_attr_init(&pdata->name_attr.attr);
340 pdata->name_attr.attr.name = "name";
341 pdata->name_attr.attr.mode = S_IRUGO;
342 pdata->name_attr.show = show_name;
343 return device_create_file(dev, &pdata->name_attr);
344}
345
346static int create_core_attrs(struct temp_data *tdata, struct device *dev,
347 int attr_no)
348{
349 int err, i;
350 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
351 struct device_attribute *devattr, char *buf) = {
352 show_label, show_crit_alarm, show_temp, show_tjmax,
353 show_ttarget };
354 static const char *const names[TOTAL_ATTRS] = {
355 "temp%d_label", "temp%d_crit_alarm",
356 "temp%d_input", "temp%d_crit",
357 "temp%d_max" };
358
359 for (i = 0; i < tdata->attr_size; i++) {
360 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH, names[i],
361 attr_no);
362 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
363 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
364 tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
365 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
366 tdata->sd_attrs[i].index = attr_no;
367 err = device_create_file(dev, &tdata->sd_attrs[i].dev_attr);
368 if (err)
369 goto exit_free;
370 }
371 return 0;
372
373exit_free:
374 while (--i >= 0)
375 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
376 return err;
377}
378
379
380static int __cpuinit chk_ucode_version(unsigned int cpu)
381{
382 struct cpuinfo_x86 *c = &cpu_data(cpu);
383 int err;
384 u32 edx;
385
386 /*
387 * Check if we have problem with errata AE18 of Core processors:
388 * Readings might stop update when processor visited too deep sleep,
389 * fixed for stepping D0 (6EC).
390 */
391 if (c->x86_model == 0xe && c->x86_mask < 0xc) {
392 /* check for microcode update */
393 err = smp_call_function_single(cpu, get_ucode_rev_on_cpu,
394 &edx, 1);
395 if (err) {
396 pr_err("Cannot determine microcode revision of "
397 "CPU#%u (%d)!\n", cpu, err);
398 return -ENODEV;
399 } else if (edx < 0x39) {
400 pr_err("Errata AE18 not fixed, update BIOS or "
401 "microcode of the CPU!\n");
402 return -ENODEV;
403 }
404 }
405 return 0;
406}
407
408static struct platform_device *coretemp_get_pdev(unsigned int cpu)
409{
410 u16 phys_proc_id = TO_PHYS_ID(cpu);
411 struct pdev_entry *p;
412
413 mutex_lock(&pdev_list_mutex);
414
415 list_for_each_entry(p, &pdev_list, list)
416 if (p->phys_proc_id == phys_proc_id) {
417 mutex_unlock(&pdev_list_mutex);
418 return p->pdev;
419 }
420
421 mutex_unlock(&pdev_list_mutex);
422 return NULL;
423}
424
425static struct temp_data *init_temp_data(unsigned int cpu, int pkg_flag)
426{
427 struct temp_data *tdata;
428
429 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
430 if (!tdata)
431 return NULL;
432
433 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
434 MSR_IA32_THERM_STATUS;
435 tdata->is_pkg_data = pkg_flag;
436 tdata->cpu = cpu;
437 tdata->cpu_core_id = TO_CORE_ID(cpu);
438 tdata->attr_size = MAX_CORE_ATTRS;
439 mutex_init(&tdata->update_lock);
440 return tdata;
441}
442
443static int create_core_data(struct platform_device *pdev,
444 unsigned int cpu, int pkg_flag)
445{
446 struct temp_data *tdata;
447 struct platform_data *pdata = platform_get_drvdata(pdev);
448 struct cpuinfo_x86 *c = &cpu_data(cpu);
449 u32 eax, edx;
450 int err, attr_no;
451
452 /*
453 * Find attr number for sysfs:
454 * We map the attr number to core id of the CPU
455 * The attr number is always core id + 2
456 * The Pkgtemp will always show up as temp1_*, if available
457 */
458 attr_no = pkg_flag ? 1 : TO_ATTR_NO(cpu);
459
460 if (attr_no > MAX_CORE_DATA - 1)
461 return -ERANGE;
462
463 /*
464 * Provide a single set of attributes for all HT siblings of a core
465 * to avoid duplicate sensors (the processor ID and core ID of all
466 * HT siblings of a core are the same).
467 * Skip if a HT sibling of this core is already registered.
468 * This is not an error.
469 */
470 if (pdata->core_data[attr_no] != NULL)
471 return 0;
472
473 tdata = init_temp_data(cpu, pkg_flag);
474 if (!tdata)
475 return -ENOMEM;
476
477 /* Test if we can access the status register */
478 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
479 if (err)
480 goto exit_free;
481
482 /* We can access status register. Get Critical Temperature */
483 tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
484
485 /*
486 * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
487 * The target temperature is available on older CPUs but not in this
488 * register. Atoms don't have the register at all.
489 */
490 if (c->x86_model > 0xe && c->x86_model != 0x1c) {
491 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
492 &eax, &edx);
493 if (!err) {
494 tdata->ttarget
495 = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
496 tdata->attr_size++;
497 }
498 }
499
500 pdata->core_data[attr_no] = tdata;
501
502 /* Create sysfs interfaces */
503 err = create_core_attrs(tdata, &pdev->dev, attr_no);
504 if (err)
505 goto exit_free;
506
507 return 0;
508exit_free:
509 pdata->core_data[attr_no] = NULL;
510 kfree(tdata);
511 return err;
512}
513
514static void coretemp_add_core(unsigned int cpu, int pkg_flag)
515{
516 struct platform_device *pdev = coretemp_get_pdev(cpu);
517 int err;
518
519 if (!pdev)
520 return;
521
522 err = create_core_data(pdev, cpu, pkg_flag);
523 if (err)
524 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
525}
526
527static void coretemp_remove_core(struct platform_data *pdata,
528 struct device *dev, int indx)
529{
530 int i;
531 struct temp_data *tdata = pdata->core_data[indx];
532
533 /* Remove the sysfs attributes */
534 for (i = 0; i < tdata->attr_size; i++)
535 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
536
537 kfree(pdata->core_data[indx]);
538 pdata->core_data[indx] = NULL;
539}
540
541static int __devinit coretemp_probe(struct platform_device *pdev)
542{
543 struct platform_data *pdata;
544 int err;
545
546 /* Initialize the per-package data structures */
547 pdata = kzalloc(sizeof(struct platform_data), GFP_KERNEL);
548 if (!pdata)
549 return -ENOMEM;
550
551 err = create_name_attr(pdata, &pdev->dev);
552 if (err)
553 goto exit_free;
554
555 pdata->phys_proc_id = pdev->id;
556 platform_set_drvdata(pdev, pdata);
557
558 pdata->hwmon_dev = hwmon_device_register(&pdev->dev);
559 if (IS_ERR(pdata->hwmon_dev)) {
560 err = PTR_ERR(pdata->hwmon_dev);
561 dev_err(&pdev->dev, "Class registration failed (%d)\n", err);
562 goto exit_name;
563 }
564 return 0;
565
566exit_name:
567 device_remove_file(&pdev->dev, &pdata->name_attr);
568 platform_set_drvdata(pdev, NULL);
569exit_free:
570 kfree(pdata);
571 return err;
572}
573
574static int __devexit coretemp_remove(struct platform_device *pdev)
575{
576 struct platform_data *pdata = platform_get_drvdata(pdev);
577 int i;
578
579 for (i = MAX_CORE_DATA - 1; i >= 0; --i)
580 if (pdata->core_data[i])
581 coretemp_remove_core(pdata, &pdev->dev, i);
582
583 device_remove_file(&pdev->dev, &pdata->name_attr);
584 hwmon_device_unregister(pdata->hwmon_dev);
585 platform_set_drvdata(pdev, NULL);
586 kfree(pdata);
587 return 0;
588}
589
590static struct platform_driver coretemp_driver = {
591 .driver = {
592 .owner = THIS_MODULE,
593 .name = DRVNAME,
594 },
595 .probe = coretemp_probe,
596 .remove = __devexit_p(coretemp_remove),
597};
598
599static int __cpuinit coretemp_device_add(unsigned int cpu)
600{
601 int err;
602 struct platform_device *pdev;
603 struct pdev_entry *pdev_entry;
604
605 mutex_lock(&pdev_list_mutex);
606
607 pdev = platform_device_alloc(DRVNAME, TO_PHYS_ID(cpu));
608 if (!pdev) {
609 err = -ENOMEM;
610 pr_err("Device allocation failed\n");
611 goto exit;
612 }
613
614 pdev_entry = kzalloc(sizeof(struct pdev_entry), GFP_KERNEL);
615 if (!pdev_entry) {
616 err = -ENOMEM;
617 goto exit_device_put;
618 }
619
620 err = platform_device_add(pdev);
621 if (err) {
622 pr_err("Device addition failed (%d)\n", err);
623 goto exit_device_free;
624 }
625
626 pdev_entry->pdev = pdev;
627 pdev_entry->phys_proc_id = pdev->id;
628
629 list_add_tail(&pdev_entry->list, &pdev_list);
630 mutex_unlock(&pdev_list_mutex);
631
632 return 0;
633
634exit_device_free:
635 kfree(pdev_entry);
636exit_device_put:
637 platform_device_put(pdev);
638exit:
639 mutex_unlock(&pdev_list_mutex);
640 return err;
641}
642
643static void coretemp_device_remove(unsigned int cpu)
644{
645 struct pdev_entry *p, *n;
646 u16 phys_proc_id = TO_PHYS_ID(cpu);
647
648 mutex_lock(&pdev_list_mutex);
649 list_for_each_entry_safe(p, n, &pdev_list, list) {
650 if (p->phys_proc_id != phys_proc_id)
651 continue;
652 platform_device_unregister(p->pdev);
653 list_del(&p->list);
654 kfree(p);
655 }
656 mutex_unlock(&pdev_list_mutex);
657}
658
659static bool is_any_core_online(struct platform_data *pdata)
660{
661 int i;
662
663 /* Find online cores, except pkgtemp data */
664 for (i = MAX_CORE_DATA - 1; i >= 0; --i) {
665 if (pdata->core_data[i] &&
666 !pdata->core_data[i]->is_pkg_data) {
667 return true;
668 }
669 }
670 return false;
671}
672
673static void __cpuinit get_core_online(unsigned int cpu)
674{
675 struct cpuinfo_x86 *c = &cpu_data(cpu);
676 struct platform_device *pdev = coretemp_get_pdev(cpu);
677 int err;
678
679 /*
680 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
681 * sensors. We check this bit only, all the early CPUs
682 * without thermal sensors will be filtered out.
683 */
684 if (!cpu_has(c, X86_FEATURE_DTS))
685 return;
686
687 if (!pdev) {
688 /* Check the microcode version of the CPU */
689 if (chk_ucode_version(cpu))
690 return;
691
692 /*
693 * Alright, we have DTS support.
694 * We are bringing the _first_ core in this pkg
695 * online. So, initialize per-pkg data structures and
696 * then bring this core online.
697 */
698 err = coretemp_device_add(cpu);
699 if (err)
700 return;
701 /*
702 * Check whether pkgtemp support is available.
703 * If so, add interfaces for pkgtemp.
704 */
705 if (cpu_has(c, X86_FEATURE_PTS))
706 coretemp_add_core(cpu, 1);
707 }
708 /*
709 * Physical CPU device already exists.
710 * So, just add interfaces for this core.
711 */
712 coretemp_add_core(cpu, 0);
713}
714
715static void __cpuinit put_core_offline(unsigned int cpu)
716{
717 int i, indx;
718 struct platform_data *pdata;
719 struct platform_device *pdev = coretemp_get_pdev(cpu);
720
721 /* If the physical CPU device does not exist, just return */
722 if (!pdev)
723 return;
724
725 pdata = platform_get_drvdata(pdev);
726
727 indx = TO_ATTR_NO(cpu);
728
729 if (pdata->core_data[indx] && pdata->core_data[indx]->cpu == cpu)
730 coretemp_remove_core(pdata, &pdev->dev, indx);
731
732 /*
733 * If a HT sibling of a core is taken offline, but another HT sibling
734 * of the same core is still online, register the alternate sibling.
735 * This ensures that exactly one set of attributes is provided as long
736 * as at least one HT sibling of a core is online.
737 */
738 for_each_sibling(i, cpu) {
739 if (i != cpu) {
740 get_core_online(i);
741 /*
742 * Display temperature sensor data for one HT sibling
743 * per core only, so abort the loop after one such
744 * sibling has been found.
745 */
746 break;
747 }
748 }
749 /*
750 * If all cores in this pkg are offline, remove the device.
751 * coretemp_device_remove calls unregister_platform_device,
752 * which in turn calls coretemp_remove. This removes the
753 * pkgtemp entry and does other clean ups.
754 */
755 if (!is_any_core_online(pdata))
756 coretemp_device_remove(cpu);
757}
758
759static int __cpuinit coretemp_cpu_callback(struct notifier_block *nfb,
760 unsigned long action, void *hcpu)
761{
762 unsigned int cpu = (unsigned long) hcpu;
763
764 switch (action) {
765 case CPU_ONLINE:
766 case CPU_DOWN_FAILED:
767 get_core_online(cpu);
768 break;
769 case CPU_DOWN_PREPARE:
770 put_core_offline(cpu);
771 break;
772 }
773 return NOTIFY_OK;
774}
775
776static struct notifier_block coretemp_cpu_notifier __refdata = {
777 .notifier_call = coretemp_cpu_callback,
778};
779
780static int __init coretemp_init(void)
781{
782 int i, err = -ENODEV;
783
784 /* quick check if we run Intel */
785 if (cpu_data(0).x86_vendor != X86_VENDOR_INTEL)
786 goto exit;
787
788 err = platform_driver_register(&coretemp_driver);
789 if (err)
790 goto exit;
791
792 for_each_online_cpu(i)
793 get_core_online(i);
794
795#ifndef CONFIG_HOTPLUG_CPU
796 if (list_empty(&pdev_list)) {
797 err = -ENODEV;
798 goto exit_driver_unreg;
799 }
800#endif
801
802 register_hotcpu_notifier(&coretemp_cpu_notifier);
803 return 0;
804
805#ifndef CONFIG_HOTPLUG_CPU
806exit_driver_unreg:
807 platform_driver_unregister(&coretemp_driver);
808#endif
809exit:
810 return err;
811}
812
813static void __exit coretemp_exit(void)
814{
815 struct pdev_entry *p, *n;
816
817 unregister_hotcpu_notifier(&coretemp_cpu_notifier);
818 mutex_lock(&pdev_list_mutex);
819 list_for_each_entry_safe(p, n, &pdev_list, list) {
820 platform_device_unregister(p->pdev);
821 list_del(&p->list);
822 kfree(p);
823 }
824 mutex_unlock(&pdev_list_mutex);
825 platform_driver_unregister(&coretemp_driver);
826}
827
828MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
829MODULE_DESCRIPTION("Intel Core temperature monitor");
830MODULE_LICENSE("GPL");
831
832module_init(coretemp_init)
833module_exit(coretemp_exit)
1/*
2 * coretemp.c - Linux kernel module for hardware monitoring
3 *
4 * Copyright (C) 2007 Rudolf Marek <r.marek@assembler.cz>
5 *
6 * Inspired from many hwmon drivers
7 *
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.
11 *
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.
16 *
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
20 * 02110-1301 USA.
21 */
22
23#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24
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>
39#include <linux/moduleparam.h>
40#include <asm/msr.h>
41#include <asm/processor.h>
42#include <asm/cpu_device_id.h>
43
44#define DRVNAME "coretemp"
45
46/*
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.
49 */
50static int force_tjmax;
51module_param_named(tjmax, force_tjmax, int, 0444);
52MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
53
54#define BASE_SYSFS_ATTR_NO 2 /* Sysfs Base attr no for coretemp */
55#define NUM_REAL_CORES 32 /* Number of Real cores per cpu */
56#define CORETEMP_NAME_LENGTH 17 /* String Length of attrs */
57#define MAX_CORE_ATTRS 4 /* Maximum no of basic attrs */
58#define TOTAL_ATTRS (MAX_CORE_ATTRS + 1)
59#define MAX_CORE_DATA (NUM_REAL_CORES + BASE_SYSFS_ATTR_NO)
60
61#define TO_PHYS_ID(cpu) (cpu_data(cpu).phys_proc_id)
62#define TO_CORE_ID(cpu) (cpu_data(cpu).cpu_core_id)
63#define TO_ATTR_NO(cpu) (TO_CORE_ID(cpu) + BASE_SYSFS_ATTR_NO)
64
65#ifdef CONFIG_SMP
66#define for_each_sibling(i, cpu) for_each_cpu(i, cpu_sibling_mask(cpu))
67#else
68#define for_each_sibling(i, cpu) for (i = 0; false; )
69#endif
70
71/*
72 * Per-Core Temperature Data
73 * @last_updated: The time when the current temperature value was updated
74 * earlier (in jiffies).
75 * @cpu_core_id: The CPU Core from which temperature values should be read
76 * This value is passed as "id" field to rdmsr/wrmsr functions.
77 * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
78 * from where the temperature values should be read.
79 * @attr_size: Total number of pre-core attrs displayed in the sysfs.
80 * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
81 * Otherwise, temp_data holds coretemp data.
82 * @valid: If this is 1, the current temperature is valid.
83 */
84struct temp_data {
85 int temp;
86 int ttarget;
87 int tjmax;
88 unsigned long last_updated;
89 unsigned int cpu;
90 u32 cpu_core_id;
91 u32 status_reg;
92 int attr_size;
93 bool is_pkg_data;
94 bool valid;
95 struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
96 char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
97 struct mutex update_lock;
98};
99
100/* Platform Data per Physical CPU */
101struct platform_data {
102 struct device *hwmon_dev;
103 u16 phys_proc_id;
104 struct temp_data *core_data[MAX_CORE_DATA];
105 struct device_attribute name_attr;
106};
107
108struct pdev_entry {
109 struct list_head list;
110 struct platform_device *pdev;
111 u16 phys_proc_id;
112};
113
114static LIST_HEAD(pdev_list);
115static DEFINE_MUTEX(pdev_list_mutex);
116
117static ssize_t show_name(struct device *dev,
118 struct device_attribute *devattr, char *buf)
119{
120 return sprintf(buf, "%s\n", DRVNAME);
121}
122
123static ssize_t show_label(struct device *dev,
124 struct device_attribute *devattr, char *buf)
125{
126 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
127 struct platform_data *pdata = dev_get_drvdata(dev);
128 struct temp_data *tdata = pdata->core_data[attr->index];
129
130 if (tdata->is_pkg_data)
131 return sprintf(buf, "Physical id %u\n", pdata->phys_proc_id);
132
133 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
134}
135
136static ssize_t show_crit_alarm(struct device *dev,
137 struct device_attribute *devattr, char *buf)
138{
139 u32 eax, edx;
140 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
141 struct platform_data *pdata = dev_get_drvdata(dev);
142 struct temp_data *tdata = pdata->core_data[attr->index];
143
144 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
145
146 return sprintf(buf, "%d\n", (eax >> 5) & 1);
147}
148
149static ssize_t show_tjmax(struct device *dev,
150 struct device_attribute *devattr, char *buf)
151{
152 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
153 struct platform_data *pdata = dev_get_drvdata(dev);
154
155 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
156}
157
158static ssize_t show_ttarget(struct device *dev,
159 struct device_attribute *devattr, char *buf)
160{
161 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
162 struct platform_data *pdata = dev_get_drvdata(dev);
163
164 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
165}
166
167static ssize_t show_temp(struct device *dev,
168 struct device_attribute *devattr, char *buf)
169{
170 u32 eax, edx;
171 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
172 struct platform_data *pdata = dev_get_drvdata(dev);
173 struct temp_data *tdata = pdata->core_data[attr->index];
174
175 mutex_lock(&tdata->update_lock);
176
177 /* Check whether the time interval has elapsed */
178 if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
179 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
180 tdata->valid = 0;
181 /* Check whether the data is valid */
182 if (eax & 0x80000000) {
183 tdata->temp = tdata->tjmax -
184 ((eax >> 16) & 0x7f) * 1000;
185 tdata->valid = 1;
186 }
187 tdata->last_updated = jiffies;
188 }
189
190 mutex_unlock(&tdata->update_lock);
191 return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
192}
193
194struct tjmax {
195 char const *id;
196 int tjmax;
197};
198
199static struct tjmax __cpuinitconst tjmax_table[] = {
200 { "CPU D410", 100000 },
201 { "CPU D425", 100000 },
202 { "CPU D510", 100000 },
203 { "CPU D525", 100000 },
204 { "CPU N450", 100000 },
205 { "CPU N455", 100000 },
206 { "CPU N470", 100000 },
207 { "CPU N475", 100000 },
208 { "CPU 230", 100000 },
209 { "CPU 330", 125000 },
210};
211
212static int __cpuinit adjust_tjmax(struct cpuinfo_x86 *c, u32 id,
213 struct device *dev)
214{
215 /* The 100C is default for both mobile and non mobile CPUs */
216
217 int tjmax = 100000;
218 int tjmax_ee = 85000;
219 int usemsr_ee = 1;
220 int err;
221 u32 eax, edx;
222 struct pci_dev *host_bridge;
223 int i;
224
225 /* explicit tjmax table entries override heuristics */
226 for (i = 0; i < ARRAY_SIZE(tjmax_table); i++) {
227 if (strstr(c->x86_model_id, tjmax_table[i].id))
228 return tjmax_table[i].tjmax;
229 }
230
231 /* Early chips have no MSR for TjMax */
232
233 if (c->x86_model == 0xf && c->x86_mask < 4)
234 usemsr_ee = 0;
235
236 /* Atom CPUs */
237
238 if (c->x86_model == 0x1c || c->x86_model == 0x26
239 || c->x86_model == 0x27) {
240 usemsr_ee = 0;
241
242 host_bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));
243
244 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL
245 && (host_bridge->device == 0xa000 /* NM10 based nettop */
246 || host_bridge->device == 0xa010)) /* NM10 based netbook */
247 tjmax = 100000;
248 else
249 tjmax = 90000;
250
251 pci_dev_put(host_bridge);
252 } else if (c->x86_model == 0x36) {
253 usemsr_ee = 0;
254 tjmax = 100000;
255 }
256
257 if (c->x86_model > 0xe && usemsr_ee) {
258 u8 platform_id;
259
260 /*
261 * Now we can detect the mobile CPU using Intel provided table
262 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
263 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
264 */
265 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
266 if (err) {
267 dev_warn(dev,
268 "Unable to access MSR 0x17, assuming desktop"
269 " CPU\n");
270 usemsr_ee = 0;
271 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
272 /*
273 * Trust bit 28 up to Penryn, I could not find any
274 * documentation on that; if you happen to know
275 * someone at Intel please ask
276 */
277 usemsr_ee = 0;
278 } else {
279 /* Platform ID bits 52:50 (EDX starts at bit 32) */
280 platform_id = (edx >> 18) & 0x7;
281
282 /*
283 * Mobile Penryn CPU seems to be platform ID 7 or 5
284 * (guesswork)
285 */
286 if (c->x86_model == 0x17 &&
287 (platform_id == 5 || platform_id == 7)) {
288 /*
289 * If MSR EE bit is set, set it to 90 degrees C,
290 * otherwise 105 degrees C
291 */
292 tjmax_ee = 90000;
293 tjmax = 105000;
294 }
295 }
296 }
297
298 if (usemsr_ee) {
299 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
300 if (err) {
301 dev_warn(dev,
302 "Unable to access MSR 0xEE, for Tjmax, left"
303 " at default\n");
304 } else if (eax & 0x40000000) {
305 tjmax = tjmax_ee;
306 }
307 } else if (tjmax == 100000) {
308 /*
309 * If we don't use msr EE it means we are desktop CPU
310 * (with exeception of Atom)
311 */
312 dev_warn(dev, "Using relative temperature scale!\n");
313 }
314
315 return tjmax;
316}
317
318static int __cpuinit get_tjmax(struct cpuinfo_x86 *c, u32 id,
319 struct device *dev)
320{
321 int err;
322 u32 eax, edx;
323 u32 val;
324
325 /*
326 * A new feature of current Intel(R) processors, the
327 * IA32_TEMPERATURE_TARGET contains the TjMax value
328 */
329 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
330 if (err) {
331 if (c->x86_model > 0xe && c->x86_model != 0x1c)
332 dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
333 } else {
334 val = (eax >> 16) & 0xff;
335 /*
336 * If the TjMax is not plausible, an assumption
337 * will be used
338 */
339 if (val) {
340 dev_dbg(dev, "TjMax is %d degrees C\n", val);
341 return val * 1000;
342 }
343 }
344
345 if (force_tjmax) {
346 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
347 force_tjmax);
348 return force_tjmax * 1000;
349 }
350
351 /*
352 * An assumption is made for early CPUs and unreadable MSR.
353 * NOTE: the calculated value may not be correct.
354 */
355 return adjust_tjmax(c, id, dev);
356}
357
358static int __devinit create_name_attr(struct platform_data *pdata,
359 struct device *dev)
360{
361 sysfs_attr_init(&pdata->name_attr.attr);
362 pdata->name_attr.attr.name = "name";
363 pdata->name_attr.attr.mode = S_IRUGO;
364 pdata->name_attr.show = show_name;
365 return device_create_file(dev, &pdata->name_attr);
366}
367
368static int __cpuinit create_core_attrs(struct temp_data *tdata,
369 struct device *dev, int attr_no)
370{
371 int err, i;
372 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
373 struct device_attribute *devattr, char *buf) = {
374 show_label, show_crit_alarm, show_temp, show_tjmax,
375 show_ttarget };
376 static const char *const names[TOTAL_ATTRS] = {
377 "temp%d_label", "temp%d_crit_alarm",
378 "temp%d_input", "temp%d_crit",
379 "temp%d_max" };
380
381 for (i = 0; i < tdata->attr_size; i++) {
382 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH, names[i],
383 attr_no);
384 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
385 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
386 tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
387 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
388 tdata->sd_attrs[i].index = attr_no;
389 err = device_create_file(dev, &tdata->sd_attrs[i].dev_attr);
390 if (err)
391 goto exit_free;
392 }
393 return 0;
394
395exit_free:
396 while (--i >= 0)
397 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
398 return err;
399}
400
401
402static int __cpuinit chk_ucode_version(unsigned int cpu)
403{
404 struct cpuinfo_x86 *c = &cpu_data(cpu);
405
406 /*
407 * Check if we have problem with errata AE18 of Core processors:
408 * Readings might stop update when processor visited too deep sleep,
409 * fixed for stepping D0 (6EC).
410 */
411 if (c->x86_model == 0xe && c->x86_mask < 0xc && c->microcode < 0x39) {
412 pr_err("Errata AE18 not fixed, update BIOS or "
413 "microcode of the CPU!\n");
414 return -ENODEV;
415 }
416 return 0;
417}
418
419static struct platform_device __cpuinit *coretemp_get_pdev(unsigned int cpu)
420{
421 u16 phys_proc_id = TO_PHYS_ID(cpu);
422 struct pdev_entry *p;
423
424 mutex_lock(&pdev_list_mutex);
425
426 list_for_each_entry(p, &pdev_list, list)
427 if (p->phys_proc_id == phys_proc_id) {
428 mutex_unlock(&pdev_list_mutex);
429 return p->pdev;
430 }
431
432 mutex_unlock(&pdev_list_mutex);
433 return NULL;
434}
435
436static struct temp_data __cpuinit *init_temp_data(unsigned int cpu,
437 int pkg_flag)
438{
439 struct temp_data *tdata;
440
441 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
442 if (!tdata)
443 return NULL;
444
445 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
446 MSR_IA32_THERM_STATUS;
447 tdata->is_pkg_data = pkg_flag;
448 tdata->cpu = cpu;
449 tdata->cpu_core_id = TO_CORE_ID(cpu);
450 tdata->attr_size = MAX_CORE_ATTRS;
451 mutex_init(&tdata->update_lock);
452 return tdata;
453}
454
455static int __cpuinit create_core_data(struct platform_device *pdev,
456 unsigned int cpu, int pkg_flag)
457{
458 struct temp_data *tdata;
459 struct platform_data *pdata = platform_get_drvdata(pdev);
460 struct cpuinfo_x86 *c = &cpu_data(cpu);
461 u32 eax, edx;
462 int err, attr_no;
463
464 /*
465 * Find attr number for sysfs:
466 * We map the attr number to core id of the CPU
467 * The attr number is always core id + 2
468 * The Pkgtemp will always show up as temp1_*, if available
469 */
470 attr_no = pkg_flag ? 1 : TO_ATTR_NO(cpu);
471
472 if (attr_no > MAX_CORE_DATA - 1)
473 return -ERANGE;
474
475 /*
476 * Provide a single set of attributes for all HT siblings of a core
477 * to avoid duplicate sensors (the processor ID and core ID of all
478 * HT siblings of a core are the same).
479 * Skip if a HT sibling of this core is already registered.
480 * This is not an error.
481 */
482 if (pdata->core_data[attr_no] != NULL)
483 return 0;
484
485 tdata = init_temp_data(cpu, pkg_flag);
486 if (!tdata)
487 return -ENOMEM;
488
489 /* Test if we can access the status register */
490 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
491 if (err)
492 goto exit_free;
493
494 /* We can access status register. Get Critical Temperature */
495 tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
496
497 /*
498 * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
499 * The target temperature is available on older CPUs but not in this
500 * register. Atoms don't have the register at all.
501 */
502 if (c->x86_model > 0xe && c->x86_model != 0x1c) {
503 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
504 &eax, &edx);
505 if (!err) {
506 tdata->ttarget
507 = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
508 tdata->attr_size++;
509 }
510 }
511
512 pdata->core_data[attr_no] = tdata;
513
514 /* Create sysfs interfaces */
515 err = create_core_attrs(tdata, &pdev->dev, attr_no);
516 if (err)
517 goto exit_free;
518
519 return 0;
520exit_free:
521 pdata->core_data[attr_no] = NULL;
522 kfree(tdata);
523 return err;
524}
525
526static void __cpuinit coretemp_add_core(unsigned int cpu, int pkg_flag)
527{
528 struct platform_device *pdev = coretemp_get_pdev(cpu);
529 int err;
530
531 if (!pdev)
532 return;
533
534 err = create_core_data(pdev, cpu, pkg_flag);
535 if (err)
536 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
537}
538
539static void coretemp_remove_core(struct platform_data *pdata,
540 struct device *dev, int indx)
541{
542 int i;
543 struct temp_data *tdata = pdata->core_data[indx];
544
545 /* Remove the sysfs attributes */
546 for (i = 0; i < tdata->attr_size; i++)
547 device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
548
549 kfree(pdata->core_data[indx]);
550 pdata->core_data[indx] = NULL;
551}
552
553static int __devinit coretemp_probe(struct platform_device *pdev)
554{
555 struct platform_data *pdata;
556 int err;
557
558 /* Initialize the per-package data structures */
559 pdata = kzalloc(sizeof(struct platform_data), GFP_KERNEL);
560 if (!pdata)
561 return -ENOMEM;
562
563 err = create_name_attr(pdata, &pdev->dev);
564 if (err)
565 goto exit_free;
566
567 pdata->phys_proc_id = pdev->id;
568 platform_set_drvdata(pdev, pdata);
569
570 pdata->hwmon_dev = hwmon_device_register(&pdev->dev);
571 if (IS_ERR(pdata->hwmon_dev)) {
572 err = PTR_ERR(pdata->hwmon_dev);
573 dev_err(&pdev->dev, "Class registration failed (%d)\n", err);
574 goto exit_name;
575 }
576 return 0;
577
578exit_name:
579 device_remove_file(&pdev->dev, &pdata->name_attr);
580 platform_set_drvdata(pdev, NULL);
581exit_free:
582 kfree(pdata);
583 return err;
584}
585
586static int __devexit coretemp_remove(struct platform_device *pdev)
587{
588 struct platform_data *pdata = platform_get_drvdata(pdev);
589 int i;
590
591 for (i = MAX_CORE_DATA - 1; i >= 0; --i)
592 if (pdata->core_data[i])
593 coretemp_remove_core(pdata, &pdev->dev, i);
594
595 device_remove_file(&pdev->dev, &pdata->name_attr);
596 hwmon_device_unregister(pdata->hwmon_dev);
597 platform_set_drvdata(pdev, NULL);
598 kfree(pdata);
599 return 0;
600}
601
602static struct platform_driver coretemp_driver = {
603 .driver = {
604 .owner = THIS_MODULE,
605 .name = DRVNAME,
606 },
607 .probe = coretemp_probe,
608 .remove = __devexit_p(coretemp_remove),
609};
610
611static int __cpuinit coretemp_device_add(unsigned int cpu)
612{
613 int err;
614 struct platform_device *pdev;
615 struct pdev_entry *pdev_entry;
616
617 mutex_lock(&pdev_list_mutex);
618
619 pdev = platform_device_alloc(DRVNAME, TO_PHYS_ID(cpu));
620 if (!pdev) {
621 err = -ENOMEM;
622 pr_err("Device allocation failed\n");
623 goto exit;
624 }
625
626 pdev_entry = kzalloc(sizeof(struct pdev_entry), GFP_KERNEL);
627 if (!pdev_entry) {
628 err = -ENOMEM;
629 goto exit_device_put;
630 }
631
632 err = platform_device_add(pdev);
633 if (err) {
634 pr_err("Device addition failed (%d)\n", err);
635 goto exit_device_free;
636 }
637
638 pdev_entry->pdev = pdev;
639 pdev_entry->phys_proc_id = pdev->id;
640
641 list_add_tail(&pdev_entry->list, &pdev_list);
642 mutex_unlock(&pdev_list_mutex);
643
644 return 0;
645
646exit_device_free:
647 kfree(pdev_entry);
648exit_device_put:
649 platform_device_put(pdev);
650exit:
651 mutex_unlock(&pdev_list_mutex);
652 return err;
653}
654
655static void __cpuinit coretemp_device_remove(unsigned int cpu)
656{
657 struct pdev_entry *p, *n;
658 u16 phys_proc_id = TO_PHYS_ID(cpu);
659
660 mutex_lock(&pdev_list_mutex);
661 list_for_each_entry_safe(p, n, &pdev_list, list) {
662 if (p->phys_proc_id != phys_proc_id)
663 continue;
664 platform_device_unregister(p->pdev);
665 list_del(&p->list);
666 kfree(p);
667 }
668 mutex_unlock(&pdev_list_mutex);
669}
670
671static bool __cpuinit is_any_core_online(struct platform_data *pdata)
672{
673 int i;
674
675 /* Find online cores, except pkgtemp data */
676 for (i = MAX_CORE_DATA - 1; i >= 0; --i) {
677 if (pdata->core_data[i] &&
678 !pdata->core_data[i]->is_pkg_data) {
679 return true;
680 }
681 }
682 return false;
683}
684
685static void __cpuinit get_core_online(unsigned int cpu)
686{
687 struct cpuinfo_x86 *c = &cpu_data(cpu);
688 struct platform_device *pdev = coretemp_get_pdev(cpu);
689 int err;
690
691 /*
692 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
693 * sensors. We check this bit only, all the early CPUs
694 * without thermal sensors will be filtered out.
695 */
696 if (!cpu_has(c, X86_FEATURE_DTHERM))
697 return;
698
699 if (!pdev) {
700 /* Check the microcode version of the CPU */
701 if (chk_ucode_version(cpu))
702 return;
703
704 /*
705 * Alright, we have DTS support.
706 * We are bringing the _first_ core in this pkg
707 * online. So, initialize per-pkg data structures and
708 * then bring this core online.
709 */
710 err = coretemp_device_add(cpu);
711 if (err)
712 return;
713 /*
714 * Check whether pkgtemp support is available.
715 * If so, add interfaces for pkgtemp.
716 */
717 if (cpu_has(c, X86_FEATURE_PTS))
718 coretemp_add_core(cpu, 1);
719 }
720 /*
721 * Physical CPU device already exists.
722 * So, just add interfaces for this core.
723 */
724 coretemp_add_core(cpu, 0);
725}
726
727static void __cpuinit put_core_offline(unsigned int cpu)
728{
729 int i, indx;
730 struct platform_data *pdata;
731 struct platform_device *pdev = coretemp_get_pdev(cpu);
732
733 /* If the physical CPU device does not exist, just return */
734 if (!pdev)
735 return;
736
737 pdata = platform_get_drvdata(pdev);
738
739 indx = TO_ATTR_NO(cpu);
740
741 /* The core id is too big, just return */
742 if (indx > MAX_CORE_DATA - 1)
743 return;
744
745 if (pdata->core_data[indx] && pdata->core_data[indx]->cpu == cpu)
746 coretemp_remove_core(pdata, &pdev->dev, indx);
747
748 /*
749 * If a HT sibling of a core is taken offline, but another HT sibling
750 * of the same core is still online, register the alternate sibling.
751 * This ensures that exactly one set of attributes is provided as long
752 * as at least one HT sibling of a core is online.
753 */
754 for_each_sibling(i, cpu) {
755 if (i != cpu) {
756 get_core_online(i);
757 /*
758 * Display temperature sensor data for one HT sibling
759 * per core only, so abort the loop after one such
760 * sibling has been found.
761 */
762 break;
763 }
764 }
765 /*
766 * If all cores in this pkg are offline, remove the device.
767 * coretemp_device_remove calls unregister_platform_device,
768 * which in turn calls coretemp_remove. This removes the
769 * pkgtemp entry and does other clean ups.
770 */
771 if (!is_any_core_online(pdata))
772 coretemp_device_remove(cpu);
773}
774
775static int __cpuinit coretemp_cpu_callback(struct notifier_block *nfb,
776 unsigned long action, void *hcpu)
777{
778 unsigned int cpu = (unsigned long) hcpu;
779
780 switch (action) {
781 case CPU_ONLINE:
782 case CPU_DOWN_FAILED:
783 get_core_online(cpu);
784 break;
785 case CPU_DOWN_PREPARE:
786 put_core_offline(cpu);
787 break;
788 }
789 return NOTIFY_OK;
790}
791
792static struct notifier_block coretemp_cpu_notifier __refdata = {
793 .notifier_call = coretemp_cpu_callback,
794};
795
796static const struct x86_cpu_id coretemp_ids[] = {
797 { X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_DTHERM },
798 {}
799};
800MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
801
802static int __init coretemp_init(void)
803{
804 int i, err;
805
806 /*
807 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
808 * sensors. We check this bit only, all the early CPUs
809 * without thermal sensors will be filtered out.
810 */
811 if (!x86_match_cpu(coretemp_ids))
812 return -ENODEV;
813
814 err = platform_driver_register(&coretemp_driver);
815 if (err)
816 goto exit;
817
818 for_each_online_cpu(i)
819 get_core_online(i);
820
821#ifndef CONFIG_HOTPLUG_CPU
822 if (list_empty(&pdev_list)) {
823 err = -ENODEV;
824 goto exit_driver_unreg;
825 }
826#endif
827
828 register_hotcpu_notifier(&coretemp_cpu_notifier);
829 return 0;
830
831#ifndef CONFIG_HOTPLUG_CPU
832exit_driver_unreg:
833 platform_driver_unregister(&coretemp_driver);
834#endif
835exit:
836 return err;
837}
838
839static void __exit coretemp_exit(void)
840{
841 struct pdev_entry *p, *n;
842
843 unregister_hotcpu_notifier(&coretemp_cpu_notifier);
844 mutex_lock(&pdev_list_mutex);
845 list_for_each_entry_safe(p, n, &pdev_list, list) {
846 platform_device_unregister(p->pdev);
847 list_del(&p->list);
848 kfree(p);
849 }
850 mutex_unlock(&pdev_list_mutex);
851 platform_driver_unregister(&coretemp_driver);
852}
853
854MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
855MODULE_DESCRIPTION("Intel Core temperature monitor");
856MODULE_LICENSE("GPL");
857
858module_init(coretemp_init)
859module_exit(coretemp_exit)