Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * coretemp.c - Linux kernel module for hardware monitoring
4 *
5 * Copyright (C) 2007 Rudolf Marek <r.marek@assembler.cz>
6 *
7 * Inspired from many hwmon drivers
8 */
9
10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
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>
27#include <asm/msr.h>
28#include <asm/processor.h>
29#include <asm/cpu_device_id.h>
30#include <linux/sched/isolation.h>
31
32#define DRVNAME "coretemp"
33
34/*
35 * force_tjmax only matters when TjMax can't be read from the CPU itself.
36 * When set, it replaces the driver's suboptimal heuristic.
37 */
38static int force_tjmax;
39module_param_named(tjmax, force_tjmax, int, 0444);
40MODULE_PARM_DESC(tjmax, "TjMax value in degrees Celsius");
41
42#define PKG_SYSFS_ATTR_NO 1 /* Sysfs attribute for package temp */
43#define BASE_SYSFS_ATTR_NO 2 /* Sysfs Base attr no for coretemp */
44#define NUM_REAL_CORES 512 /* Number of Real cores per cpu */
45#define CORETEMP_NAME_LENGTH 28 /* String Length of attrs */
46#define MAX_CORE_ATTRS 4 /* Maximum no of basic attrs */
47#define TOTAL_ATTRS (MAX_CORE_ATTRS + 1)
48#define MAX_CORE_DATA (NUM_REAL_CORES + BASE_SYSFS_ATTR_NO)
49
50#ifdef CONFIG_SMP
51#define for_each_sibling(i, cpu) \
52 for_each_cpu(i, topology_sibling_cpumask(cpu))
53#else
54#define for_each_sibling(i, cpu) for (i = 0; false; )
55#endif
56
57/*
58 * Per-Core Temperature Data
59 * @tjmax: The static tjmax value when tjmax cannot be retrieved from
60 * IA32_TEMPERATURE_TARGET MSR.
61 * @last_updated: The time when the current temperature value was updated
62 * earlier (in jiffies).
63 * @cpu_core_id: The CPU Core from which temperature values should be read
64 * This value is passed as "id" field to rdmsr/wrmsr functions.
65 * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
66 * from where the temperature values should be read.
67 * @attr_size: Total number of pre-core attrs displayed in the sysfs.
68 * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
69 * Otherwise, temp_data holds coretemp data.
70 */
71struct temp_data {
72 int temp;
73 int tjmax;
74 unsigned long last_updated;
75 unsigned int cpu;
76 u32 cpu_core_id;
77 u32 status_reg;
78 int attr_size;
79 bool is_pkg_data;
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;
85};
86
87/* Platform Data per Physical CPU */
88struct platform_data {
89 struct device *hwmon_dev;
90 u16 pkg_id;
91 u16 cpu_map[NUM_REAL_CORES];
92 struct ida ida;
93 struct cpumask cpumask;
94 struct temp_data *core_data[MAX_CORE_DATA];
95 struct device_attribute name_attr;
96};
97
98struct tjmax_pci {
99 unsigned int device;
100 int tjmax;
101};
102
103static const struct tjmax_pci tjmax_pci_table[] = {
104 { 0x0708, 110000 }, /* CE41x0 (Sodaville ) */
105 { 0x0c72, 102000 }, /* Atom S1240 (Centerton) */
106 { 0x0c73, 95000 }, /* Atom S1220 (Centerton) */
107 { 0x0c75, 95000 }, /* Atom S1260 (Centerton) */
108};
109
110struct tjmax {
111 char const *id;
112 int tjmax;
113};
114
115static const struct tjmax tjmax_table[] = {
116 { "CPU 230", 100000 }, /* Model 0x1c, stepping 2 */
117 { "CPU 330", 125000 }, /* Model 0x1c, stepping 2 */
118};
119
120struct tjmax_model {
121 u8 model;
122 u8 mask;
123 int tjmax;
124};
125
126#define ANY 0xff
127
128static const struct tjmax_model tjmax_model_table[] = {
129 { 0x1c, 10, 100000 }, /* D4xx, K4xx, N4xx, D5xx, K5xx, N5xx */
130 { 0x1c, ANY, 90000 }, /* Z5xx, N2xx, possibly others
131 * Note: Also matches 230 and 330,
132 * which are covered by tjmax_table
133 */
134 { 0x26, ANY, 90000 }, /* Atom Tunnel Creek (Exx), Lincroft (Z6xx)
135 * Note: TjMax for E6xxT is 110C, but CPU type
136 * is undetectable by software
137 */
138 { 0x27, ANY, 90000 }, /* Atom Medfield (Z2460) */
139 { 0x35, ANY, 90000 }, /* Atom Clover Trail/Cloverview (Z27x0) */
140 { 0x36, ANY, 100000 }, /* Atom Cedar Trail/Cedarview (N2xxx, D2xxx)
141 * Also matches S12x0 (stepping 9), covered by
142 * PCI table
143 */
144};
145
146static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
147{
148 /* The 100C is default for both mobile and non mobile CPUs */
149
150 int tjmax = 100000;
151 int tjmax_ee = 85000;
152 int usemsr_ee = 1;
153 int err;
154 u32 eax, edx;
155 int i;
156 u16 devfn = PCI_DEVFN(0, 0);
157 struct pci_dev *host_bridge = pci_get_domain_bus_and_slot(0, 0, devfn);
158
159 /*
160 * Explicit tjmax table entries override heuristics.
161 * First try PCI host bridge IDs, followed by model ID strings
162 * and model/stepping information.
163 */
164 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL) {
165 for (i = 0; i < ARRAY_SIZE(tjmax_pci_table); i++) {
166 if (host_bridge->device == tjmax_pci_table[i].device) {
167 pci_dev_put(host_bridge);
168 return tjmax_pci_table[i].tjmax;
169 }
170 }
171 }
172 pci_dev_put(host_bridge);
173
174 for (i = 0; i < ARRAY_SIZE(tjmax_table); i++) {
175 if (strstr(c->x86_model_id, tjmax_table[i].id))
176 return tjmax_table[i].tjmax;
177 }
178
179 for (i = 0; i < ARRAY_SIZE(tjmax_model_table); i++) {
180 const struct tjmax_model *tm = &tjmax_model_table[i];
181 if (c->x86_model == tm->model &&
182 (tm->mask == ANY || c->x86_stepping == tm->mask))
183 return tm->tjmax;
184 }
185
186 /* Early chips have no MSR for TjMax */
187
188 if (c->x86_model == 0xf && c->x86_stepping < 4)
189 usemsr_ee = 0;
190
191 if (c->x86_model > 0xe && usemsr_ee) {
192 u8 platform_id;
193
194 /*
195 * Now we can detect the mobile CPU using Intel provided table
196 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
197 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
198 */
199 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
200 if (err) {
201 dev_warn(dev,
202 "Unable to access MSR 0x17, assuming desktop"
203 " CPU\n");
204 usemsr_ee = 0;
205 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
206 /*
207 * Trust bit 28 up to Penryn, I could not find any
208 * documentation on that; if you happen to know
209 * someone at Intel please ask
210 */
211 usemsr_ee = 0;
212 } else {
213 /* Platform ID bits 52:50 (EDX starts at bit 32) */
214 platform_id = (edx >> 18) & 0x7;
215
216 /*
217 * Mobile Penryn CPU seems to be platform ID 7 or 5
218 * (guesswork)
219 */
220 if (c->x86_model == 0x17 &&
221 (platform_id == 5 || platform_id == 7)) {
222 /*
223 * If MSR EE bit is set, set it to 90 degrees C,
224 * otherwise 105 degrees C
225 */
226 tjmax_ee = 90000;
227 tjmax = 105000;
228 }
229 }
230 }
231
232 if (usemsr_ee) {
233 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
234 if (err) {
235 dev_warn(dev,
236 "Unable to access MSR 0xEE, for Tjmax, left"
237 " at default\n");
238 } else if (eax & 0x40000000) {
239 tjmax = tjmax_ee;
240 }
241 } else if (tjmax == 100000) {
242 /*
243 * If we don't use msr EE it means we are desktop CPU
244 * (with exeception of Atom)
245 */
246 dev_warn(dev, "Using relative temperature scale!\n");
247 }
248
249 return tjmax;
250}
251
252static bool cpu_has_tjmax(struct cpuinfo_x86 *c)
253{
254 u8 model = c->x86_model;
255
256 return model > 0xe &&
257 model != 0x1c &&
258 model != 0x26 &&
259 model != 0x27 &&
260 model != 0x35 &&
261 model != 0x36;
262}
263
264static int get_tjmax(struct temp_data *tdata, struct device *dev)
265{
266 struct cpuinfo_x86 *c = &cpu_data(tdata->cpu);
267 int err;
268 u32 eax, edx;
269 u32 val;
270
271 /* use static tjmax once it is set */
272 if (tdata->tjmax)
273 return tdata->tjmax;
274
275 /*
276 * A new feature of current Intel(R) processors, the
277 * IA32_TEMPERATURE_TARGET contains the TjMax value
278 */
279 err = rdmsr_safe_on_cpu(tdata->cpu, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
280 if (err) {
281 if (cpu_has_tjmax(c))
282 dev_warn(dev, "Unable to read TjMax from CPU %u\n", tdata->cpu);
283 } else {
284 val = (eax >> 16) & 0xff;
285 if (val)
286 return val * 1000;
287 }
288
289 if (force_tjmax) {
290 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
291 force_tjmax);
292 tdata->tjmax = force_tjmax * 1000;
293 } else {
294 /*
295 * An assumption is made for early CPUs and unreadable MSR.
296 * NOTE: the calculated value may not be correct.
297 */
298 tdata->tjmax = adjust_tjmax(c, tdata->cpu, dev);
299 }
300 return tdata->tjmax;
301}
302
303static int get_ttarget(struct temp_data *tdata, struct device *dev)
304{
305 u32 eax, edx;
306 int tjmax, ttarget_offset, ret;
307
308 /*
309 * ttarget is valid only if tjmax can be retrieved from
310 * MSR_IA32_TEMPERATURE_TARGET
311 */
312 if (tdata->tjmax)
313 return -ENODEV;
314
315 ret = rdmsr_safe_on_cpu(tdata->cpu, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
316 if (ret)
317 return ret;
318
319 tjmax = (eax >> 16) & 0xff;
320
321 /* Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET. */
322 ttarget_offset = (eax >> 8) & 0xff;
323
324 return (tjmax - ttarget_offset) * 1000;
325}
326
327/* Keep track of how many zone pointers we allocated in init() */
328static int max_zones __read_mostly;
329/* Array of zone pointers. Serialized by cpu hotplug lock */
330static struct platform_device **zone_devices;
331
332static ssize_t show_label(struct device *dev,
333 struct device_attribute *devattr, char *buf)
334{
335 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
336 struct platform_data *pdata = dev_get_drvdata(dev);
337 struct temp_data *tdata = pdata->core_data[attr->index];
338
339 if (tdata->is_pkg_data)
340 return sprintf(buf, "Package id %u\n", pdata->pkg_id);
341
342 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
343}
344
345static ssize_t show_crit_alarm(struct device *dev,
346 struct device_attribute *devattr, char *buf)
347{
348 u32 eax, edx;
349 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
350 struct platform_data *pdata = dev_get_drvdata(dev);
351 struct temp_data *tdata = pdata->core_data[attr->index];
352
353 mutex_lock(&tdata->update_lock);
354 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
355 mutex_unlock(&tdata->update_lock);
356
357 return sprintf(buf, "%d\n", (eax >> 5) & 1);
358}
359
360static ssize_t show_tjmax(struct device *dev,
361 struct device_attribute *devattr, char *buf)
362{
363 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
364 struct platform_data *pdata = dev_get_drvdata(dev);
365 struct temp_data *tdata = pdata->core_data[attr->index];
366 int tjmax;
367
368 mutex_lock(&tdata->update_lock);
369 tjmax = get_tjmax(tdata, dev);
370 mutex_unlock(&tdata->update_lock);
371
372 return sprintf(buf, "%d\n", tjmax);
373}
374
375static ssize_t show_ttarget(struct device *dev,
376 struct device_attribute *devattr, char *buf)
377{
378 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
379 struct platform_data *pdata = dev_get_drvdata(dev);
380 struct temp_data *tdata = pdata->core_data[attr->index];
381 int ttarget;
382
383 mutex_lock(&tdata->update_lock);
384 ttarget = get_ttarget(tdata, dev);
385 mutex_unlock(&tdata->update_lock);
386
387 if (ttarget < 0)
388 return ttarget;
389 return sprintf(buf, "%d\n", ttarget);
390}
391
392static ssize_t show_temp(struct device *dev,
393 struct device_attribute *devattr, char *buf)
394{
395 u32 eax, edx;
396 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
397 struct platform_data *pdata = dev_get_drvdata(dev);
398 struct temp_data *tdata = pdata->core_data[attr->index];
399 int tjmax;
400
401 mutex_lock(&tdata->update_lock);
402
403 tjmax = get_tjmax(tdata, dev);
404 /* Check whether the time interval has elapsed */
405 if (time_after(jiffies, tdata->last_updated + HZ)) {
406 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
407 /*
408 * Ignore the valid bit. In all observed cases the register
409 * value is either low or zero if the valid bit is 0.
410 * Return it instead of reporting an error which doesn't
411 * really help at all.
412 */
413 tdata->temp = tjmax - ((eax >> 16) & 0x7f) * 1000;
414 tdata->last_updated = jiffies;
415 }
416
417 mutex_unlock(&tdata->update_lock);
418 return sprintf(buf, "%d\n", tdata->temp);
419}
420
421static int create_core_attrs(struct temp_data *tdata, struct device *dev,
422 int index)
423{
424 int i;
425 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
426 struct device_attribute *devattr, char *buf) = {
427 show_label, show_crit_alarm, show_temp, show_tjmax,
428 show_ttarget };
429 static const char *const suffixes[TOTAL_ATTRS] = {
430 "label", "crit_alarm", "input", "crit", "max"
431 };
432
433 for (i = 0; i < tdata->attr_size; i++) {
434 /*
435 * We map the attr number to core id of the CPU
436 * The attr number is always core id + 2
437 * The Pkgtemp will always show up as temp1_*, if available
438 */
439 int attr_no = tdata->is_pkg_data ? 1 : tdata->cpu_core_id + 2;
440
441 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH,
442 "temp%d_%s", attr_no, suffixes[i]);
443 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
444 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
445 tdata->sd_attrs[i].dev_attr.attr.mode = 0444;
446 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
447 tdata->sd_attrs[i].index = index;
448 tdata->attrs[i] = &tdata->sd_attrs[i].dev_attr.attr;
449 }
450 tdata->attr_group.attrs = tdata->attrs;
451 return sysfs_create_group(&dev->kobj, &tdata->attr_group);
452}
453
454
455static int chk_ucode_version(unsigned int cpu)
456{
457 struct cpuinfo_x86 *c = &cpu_data(cpu);
458
459 /*
460 * Check if we have problem with errata AE18 of Core processors:
461 * Readings might stop update when processor visited too deep sleep,
462 * fixed for stepping D0 (6EC).
463 */
464 if (c->x86_model == 0xe && c->x86_stepping < 0xc && c->microcode < 0x39) {
465 pr_err("Errata AE18 not fixed, update BIOS or microcode of the CPU!\n");
466 return -ENODEV;
467 }
468 return 0;
469}
470
471static struct platform_device *coretemp_get_pdev(unsigned int cpu)
472{
473 int id = topology_logical_die_id(cpu);
474
475 if (id >= 0 && id < max_zones)
476 return zone_devices[id];
477 return NULL;
478}
479
480static struct temp_data *init_temp_data(unsigned int cpu, int pkg_flag)
481{
482 struct temp_data *tdata;
483
484 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
485 if (!tdata)
486 return NULL;
487
488 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
489 MSR_IA32_THERM_STATUS;
490 tdata->is_pkg_data = pkg_flag;
491 tdata->cpu = cpu;
492 tdata->cpu_core_id = topology_core_id(cpu);
493 tdata->attr_size = MAX_CORE_ATTRS;
494 mutex_init(&tdata->update_lock);
495 return tdata;
496}
497
498static int create_core_data(struct platform_device *pdev, unsigned int cpu,
499 int pkg_flag)
500{
501 struct temp_data *tdata;
502 struct platform_data *pdata = platform_get_drvdata(pdev);
503 struct cpuinfo_x86 *c = &cpu_data(cpu);
504 u32 eax, edx;
505 int err, index;
506
507 if (!housekeeping_cpu(cpu, HK_TYPE_MISC))
508 return 0;
509
510 /*
511 * Get the index of tdata in pdata->core_data[]
512 * tdata for package: pdata->core_data[1]
513 * tdata for core: pdata->core_data[2] .. pdata->core_data[NUM_REAL_CORES + 1]
514 */
515 if (pkg_flag) {
516 index = PKG_SYSFS_ATTR_NO;
517 } else {
518 index = ida_alloc_max(&pdata->ida, NUM_REAL_CORES - 1, GFP_KERNEL);
519 if (index < 0)
520 return index;
521
522 pdata->cpu_map[index] = topology_core_id(cpu);
523 index += BASE_SYSFS_ATTR_NO;
524 }
525
526 tdata = init_temp_data(cpu, pkg_flag);
527 if (!tdata) {
528 err = -ENOMEM;
529 goto ida_free;
530 }
531
532 /* Test if we can access the status register */
533 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
534 if (err)
535 goto exit_free;
536
537 /* Make sure tdata->tjmax is a valid indicator for dynamic/static tjmax */
538 get_tjmax(tdata, &pdev->dev);
539
540 /*
541 * The target temperature is available on older CPUs but not in the
542 * MSR_IA32_TEMPERATURE_TARGET register. Atoms don't have the register
543 * at all.
544 */
545 if (c->x86_model > 0xe && c->x86_model != 0x1c)
546 if (get_ttarget(tdata, &pdev->dev) >= 0)
547 tdata->attr_size++;
548
549 pdata->core_data[index] = tdata;
550
551 /* Create sysfs interfaces */
552 err = create_core_attrs(tdata, pdata->hwmon_dev, index);
553 if (err)
554 goto exit_free;
555
556 return 0;
557exit_free:
558 pdata->core_data[index] = NULL;
559 kfree(tdata);
560ida_free:
561 if (!pkg_flag)
562 ida_free(&pdata->ida, index - BASE_SYSFS_ATTR_NO);
563 return err;
564}
565
566static void
567coretemp_add_core(struct platform_device *pdev, unsigned int cpu, int pkg_flag)
568{
569 if (create_core_data(pdev, cpu, pkg_flag))
570 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
571}
572
573static void coretemp_remove_core(struct platform_data *pdata, int indx)
574{
575 struct temp_data *tdata = pdata->core_data[indx];
576
577 /* if we errored on add then this is already gone */
578 if (!tdata)
579 return;
580
581 /* Remove the sysfs attributes */
582 sysfs_remove_group(&pdata->hwmon_dev->kobj, &tdata->attr_group);
583
584 kfree(pdata->core_data[indx]);
585 pdata->core_data[indx] = NULL;
586
587 if (indx >= BASE_SYSFS_ATTR_NO)
588 ida_free(&pdata->ida, indx - BASE_SYSFS_ATTR_NO);
589}
590
591static int coretemp_device_add(int zoneid)
592{
593 struct platform_device *pdev;
594 struct platform_data *pdata;
595 int err;
596
597 /* Initialize the per-zone data structures */
598 pdata = kzalloc(sizeof(*pdata), GFP_KERNEL);
599 if (!pdata)
600 return -ENOMEM;
601
602 pdata->pkg_id = zoneid;
603 ida_init(&pdata->ida);
604
605 pdev = platform_device_alloc(DRVNAME, zoneid);
606 if (!pdev) {
607 err = -ENOMEM;
608 goto err_free_pdata;
609 }
610
611 err = platform_device_add(pdev);
612 if (err)
613 goto err_put_dev;
614
615 platform_set_drvdata(pdev, pdata);
616 zone_devices[zoneid] = pdev;
617 return 0;
618
619err_put_dev:
620 platform_device_put(pdev);
621err_free_pdata:
622 kfree(pdata);
623 return err;
624}
625
626static void coretemp_device_remove(int zoneid)
627{
628 struct platform_device *pdev = zone_devices[zoneid];
629 struct platform_data *pdata = platform_get_drvdata(pdev);
630
631 ida_destroy(&pdata->ida);
632 kfree(pdata);
633 platform_device_unregister(pdev);
634}
635
636static int coretemp_cpu_online(unsigned int cpu)
637{
638 struct platform_device *pdev = coretemp_get_pdev(cpu);
639 struct cpuinfo_x86 *c = &cpu_data(cpu);
640 struct platform_data *pdata;
641
642 /*
643 * Don't execute this on resume as the offline callback did
644 * not get executed on suspend.
645 */
646 if (cpuhp_tasks_frozen)
647 return 0;
648
649 /*
650 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
651 * sensors. We check this bit only, all the early CPUs
652 * without thermal sensors will be filtered out.
653 */
654 if (!cpu_has(c, X86_FEATURE_DTHERM))
655 return -ENODEV;
656
657 pdata = platform_get_drvdata(pdev);
658 if (!pdata->hwmon_dev) {
659 struct device *hwmon;
660
661 /* Check the microcode version of the CPU */
662 if (chk_ucode_version(cpu))
663 return -EINVAL;
664
665 /*
666 * Alright, we have DTS support.
667 * We are bringing the _first_ core in this pkg
668 * online. So, initialize per-pkg data structures and
669 * then bring this core online.
670 */
671 hwmon = hwmon_device_register_with_groups(&pdev->dev, DRVNAME,
672 pdata, NULL);
673 if (IS_ERR(hwmon))
674 return PTR_ERR(hwmon);
675 pdata->hwmon_dev = hwmon;
676
677 /*
678 * Check whether pkgtemp support is available.
679 * If so, add interfaces for pkgtemp.
680 */
681 if (cpu_has(c, X86_FEATURE_PTS))
682 coretemp_add_core(pdev, cpu, 1);
683 }
684
685 /*
686 * Check whether a thread sibling is already online. If not add the
687 * interface for this CPU core.
688 */
689 if (!cpumask_intersects(&pdata->cpumask, topology_sibling_cpumask(cpu)))
690 coretemp_add_core(pdev, cpu, 0);
691
692 cpumask_set_cpu(cpu, &pdata->cpumask);
693 return 0;
694}
695
696static int coretemp_cpu_offline(unsigned int cpu)
697{
698 struct platform_device *pdev = coretemp_get_pdev(cpu);
699 struct platform_data *pd;
700 struct temp_data *tdata;
701 int i, indx = -1, target;
702
703 /* No need to tear down any interfaces for suspend */
704 if (cpuhp_tasks_frozen)
705 return 0;
706
707 /* If the physical CPU device does not exist, just return */
708 pd = platform_get_drvdata(pdev);
709 if (!pd->hwmon_dev)
710 return 0;
711
712 for (i = 0; i < NUM_REAL_CORES; i++) {
713 if (pd->cpu_map[i] == topology_core_id(cpu)) {
714 indx = i + BASE_SYSFS_ATTR_NO;
715 break;
716 }
717 }
718
719 /* Too many cores and this core is not populated, just return */
720 if (indx < 0)
721 return 0;
722
723 tdata = pd->core_data[indx];
724
725 cpumask_clear_cpu(cpu, &pd->cpumask);
726
727 /*
728 * If this is the last thread sibling, remove the CPU core
729 * interface, If there is still a sibling online, transfer the
730 * target cpu of that core interface to it.
731 */
732 target = cpumask_any_and(&pd->cpumask, topology_sibling_cpumask(cpu));
733 if (target >= nr_cpu_ids) {
734 coretemp_remove_core(pd, indx);
735 } else if (tdata && tdata->cpu == cpu) {
736 mutex_lock(&tdata->update_lock);
737 tdata->cpu = target;
738 mutex_unlock(&tdata->update_lock);
739 }
740
741 /*
742 * If all cores in this pkg are offline, remove the interface.
743 */
744 tdata = pd->core_data[PKG_SYSFS_ATTR_NO];
745 if (cpumask_empty(&pd->cpumask)) {
746 if (tdata)
747 coretemp_remove_core(pd, PKG_SYSFS_ATTR_NO);
748 hwmon_device_unregister(pd->hwmon_dev);
749 pd->hwmon_dev = NULL;
750 return 0;
751 }
752
753 /*
754 * Check whether this core is the target for the package
755 * interface. We need to assign it to some other cpu.
756 */
757 if (tdata && tdata->cpu == cpu) {
758 target = cpumask_first(&pd->cpumask);
759 mutex_lock(&tdata->update_lock);
760 tdata->cpu = target;
761 mutex_unlock(&tdata->update_lock);
762 }
763 return 0;
764}
765static const struct x86_cpu_id __initconst coretemp_ids[] = {
766 X86_MATCH_VENDOR_FEATURE(INTEL, X86_FEATURE_DTHERM, NULL),
767 {}
768};
769MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
770
771static enum cpuhp_state coretemp_hp_online;
772
773static int __init coretemp_init(void)
774{
775 int i, err;
776
777 /*
778 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
779 * sensors. We check this bit only, all the early CPUs
780 * without thermal sensors will be filtered out.
781 */
782 if (!x86_match_cpu(coretemp_ids))
783 return -ENODEV;
784
785 max_zones = topology_max_packages() * topology_max_die_per_package();
786 zone_devices = kcalloc(max_zones, sizeof(struct platform_device *),
787 GFP_KERNEL);
788 if (!zone_devices)
789 return -ENOMEM;
790
791 for (i = 0; i < max_zones; i++) {
792 err = coretemp_device_add(i);
793 if (err)
794 goto outzone;
795 }
796
797 err = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hwmon/coretemp:online",
798 coretemp_cpu_online, coretemp_cpu_offline);
799 if (err < 0)
800 goto outzone;
801 coretemp_hp_online = err;
802 return 0;
803
804outzone:
805 while (i--)
806 coretemp_device_remove(i);
807 kfree(zone_devices);
808 return err;
809}
810module_init(coretemp_init)
811
812static void __exit coretemp_exit(void)
813{
814 int i;
815
816 cpuhp_remove_state(coretemp_hp_online);
817 for (i = 0; i < max_zones; i++)
818 coretemp_device_remove(i);
819 kfree(zone_devices);
820}
821module_exit(coretemp_exit)
822
823MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
824MODULE_DESCRIPTION("Intel Core temperature monitor");
825MODULE_LICENSE("GPL");
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/smp.h>
38#include <linux/moduleparam.h>
39#include <linux/pci.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 128 /* Number of Real cores per cpu */
56#define CORETEMP_NAME_LENGTH 19 /* 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) \
67 for_each_cpu(i, topology_sibling_cpumask(cpu))
68#else
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 attribute *attrs[TOTAL_ATTRS + 1];
99 struct attribute_group attr_group;
100 struct mutex update_lock;
101};
102
103/* Platform Data per Physical CPU */
104struct platform_data {
105 struct device *hwmon_dev;
106 u16 phys_proc_id;
107 struct temp_data *core_data[MAX_CORE_DATA];
108 struct device_attribute name_attr;
109};
110
111struct pdev_entry {
112 struct list_head list;
113 struct platform_device *pdev;
114 u16 phys_proc_id;
115};
116
117static LIST_HEAD(pdev_list);
118static DEFINE_MUTEX(pdev_list_mutex);
119
120static ssize_t show_label(struct device *dev,
121 struct device_attribute *devattr, char *buf)
122{
123 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
124 struct platform_data *pdata = dev_get_drvdata(dev);
125 struct temp_data *tdata = pdata->core_data[attr->index];
126
127 if (tdata->is_pkg_data)
128 return sprintf(buf, "Physical id %u\n", pdata->phys_proc_id);
129
130 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
131}
132
133static ssize_t show_crit_alarm(struct device *dev,
134 struct device_attribute *devattr, char *buf)
135{
136 u32 eax, edx;
137 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
138 struct platform_data *pdata = dev_get_drvdata(dev);
139 struct temp_data *tdata = pdata->core_data[attr->index];
140
141 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
142
143 return sprintf(buf, "%d\n", (eax >> 5) & 1);
144}
145
146static ssize_t show_tjmax(struct device *dev,
147 struct device_attribute *devattr, char *buf)
148{
149 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
150 struct platform_data *pdata = dev_get_drvdata(dev);
151
152 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
153}
154
155static ssize_t show_ttarget(struct device *dev,
156 struct device_attribute *devattr, char *buf)
157{
158 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
159 struct platform_data *pdata = dev_get_drvdata(dev);
160
161 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
162}
163
164static ssize_t show_temp(struct device *dev,
165 struct device_attribute *devattr, char *buf)
166{
167 u32 eax, edx;
168 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
169 struct platform_data *pdata = dev_get_drvdata(dev);
170 struct temp_data *tdata = pdata->core_data[attr->index];
171
172 mutex_lock(&tdata->update_lock);
173
174 /* Check whether the time interval has elapsed */
175 if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
176 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
177 /*
178 * Ignore the valid bit. In all observed cases the register
179 * value is either low or zero if the valid bit is 0.
180 * Return it instead of reporting an error which doesn't
181 * really help at all.
182 */
183 tdata->temp = tdata->tjmax - ((eax >> 16) & 0x7f) * 1000;
184 tdata->valid = 1;
185 tdata->last_updated = jiffies;
186 }
187
188 mutex_unlock(&tdata->update_lock);
189 return sprintf(buf, "%d\n", tdata->temp);
190}
191
192struct tjmax_pci {
193 unsigned int device;
194 int tjmax;
195};
196
197static const struct tjmax_pci tjmax_pci_table[] = {
198 { 0x0708, 110000 }, /* CE41x0 (Sodaville ) */
199 { 0x0c72, 102000 }, /* Atom S1240 (Centerton) */
200 { 0x0c73, 95000 }, /* Atom S1220 (Centerton) */
201 { 0x0c75, 95000 }, /* Atom S1260 (Centerton) */
202};
203
204struct tjmax {
205 char const *id;
206 int tjmax;
207};
208
209static const struct tjmax tjmax_table[] = {
210 { "CPU 230", 100000 }, /* Model 0x1c, stepping 2 */
211 { "CPU 330", 125000 }, /* Model 0x1c, stepping 2 */
212};
213
214struct tjmax_model {
215 u8 model;
216 u8 mask;
217 int tjmax;
218};
219
220#define ANY 0xff
221
222static const struct tjmax_model tjmax_model_table[] = {
223 { 0x1c, 10, 100000 }, /* D4xx, K4xx, N4xx, D5xx, K5xx, N5xx */
224 { 0x1c, ANY, 90000 }, /* Z5xx, N2xx, possibly others
225 * Note: Also matches 230 and 330,
226 * which are covered by tjmax_table
227 */
228 { 0x26, ANY, 90000 }, /* Atom Tunnel Creek (Exx), Lincroft (Z6xx)
229 * Note: TjMax for E6xxT is 110C, but CPU type
230 * is undetectable by software
231 */
232 { 0x27, ANY, 90000 }, /* Atom Medfield (Z2460) */
233 { 0x35, ANY, 90000 }, /* Atom Clover Trail/Cloverview (Z27x0) */
234 { 0x36, ANY, 100000 }, /* Atom Cedar Trail/Cedarview (N2xxx, D2xxx)
235 * Also matches S12x0 (stepping 9), covered by
236 * PCI table
237 */
238};
239
240static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
241{
242 /* The 100C is default for both mobile and non mobile CPUs */
243
244 int tjmax = 100000;
245 int tjmax_ee = 85000;
246 int usemsr_ee = 1;
247 int err;
248 u32 eax, edx;
249 int i;
250 struct pci_dev *host_bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));
251
252 /*
253 * Explicit tjmax table entries override heuristics.
254 * First try PCI host bridge IDs, followed by model ID strings
255 * and model/stepping information.
256 */
257 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL) {
258 for (i = 0; i < ARRAY_SIZE(tjmax_pci_table); i++) {
259 if (host_bridge->device == tjmax_pci_table[i].device)
260 return tjmax_pci_table[i].tjmax;
261 }
262 }
263
264 for (i = 0; i < ARRAY_SIZE(tjmax_table); i++) {
265 if (strstr(c->x86_model_id, tjmax_table[i].id))
266 return tjmax_table[i].tjmax;
267 }
268
269 for (i = 0; i < ARRAY_SIZE(tjmax_model_table); i++) {
270 const struct tjmax_model *tm = &tjmax_model_table[i];
271 if (c->x86_model == tm->model &&
272 (tm->mask == ANY || c->x86_mask == tm->mask))
273 return tm->tjmax;
274 }
275
276 /* Early chips have no MSR for TjMax */
277
278 if (c->x86_model == 0xf && c->x86_mask < 4)
279 usemsr_ee = 0;
280
281 if (c->x86_model > 0xe && usemsr_ee) {
282 u8 platform_id;
283
284 /*
285 * Now we can detect the mobile CPU using Intel provided table
286 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
287 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
288 */
289 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
290 if (err) {
291 dev_warn(dev,
292 "Unable to access MSR 0x17, assuming desktop"
293 " CPU\n");
294 usemsr_ee = 0;
295 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
296 /*
297 * Trust bit 28 up to Penryn, I could not find any
298 * documentation on that; if you happen to know
299 * someone at Intel please ask
300 */
301 usemsr_ee = 0;
302 } else {
303 /* Platform ID bits 52:50 (EDX starts at bit 32) */
304 platform_id = (edx >> 18) & 0x7;
305
306 /*
307 * Mobile Penryn CPU seems to be platform ID 7 or 5
308 * (guesswork)
309 */
310 if (c->x86_model == 0x17 &&
311 (platform_id == 5 || platform_id == 7)) {
312 /*
313 * If MSR EE bit is set, set it to 90 degrees C,
314 * otherwise 105 degrees C
315 */
316 tjmax_ee = 90000;
317 tjmax = 105000;
318 }
319 }
320 }
321
322 if (usemsr_ee) {
323 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
324 if (err) {
325 dev_warn(dev,
326 "Unable to access MSR 0xEE, for Tjmax, left"
327 " at default\n");
328 } else if (eax & 0x40000000) {
329 tjmax = tjmax_ee;
330 }
331 } else if (tjmax == 100000) {
332 /*
333 * If we don't use msr EE it means we are desktop CPU
334 * (with exeception of Atom)
335 */
336 dev_warn(dev, "Using relative temperature scale!\n");
337 }
338
339 return tjmax;
340}
341
342static bool cpu_has_tjmax(struct cpuinfo_x86 *c)
343{
344 u8 model = c->x86_model;
345
346 return model > 0xe &&
347 model != 0x1c &&
348 model != 0x26 &&
349 model != 0x27 &&
350 model != 0x35 &&
351 model != 0x36;
352}
353
354static int get_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
355{
356 int err;
357 u32 eax, edx;
358 u32 val;
359
360 /*
361 * A new feature of current Intel(R) processors, the
362 * IA32_TEMPERATURE_TARGET contains the TjMax value
363 */
364 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
365 if (err) {
366 if (cpu_has_tjmax(c))
367 dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
368 } else {
369 val = (eax >> 16) & 0xff;
370 /*
371 * If the TjMax is not plausible, an assumption
372 * will be used
373 */
374 if (val) {
375 dev_dbg(dev, "TjMax is %d degrees C\n", val);
376 return val * 1000;
377 }
378 }
379
380 if (force_tjmax) {
381 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
382 force_tjmax);
383 return force_tjmax * 1000;
384 }
385
386 /*
387 * An assumption is made for early CPUs and unreadable MSR.
388 * NOTE: the calculated value may not be correct.
389 */
390 return adjust_tjmax(c, id, dev);
391}
392
393static int create_core_attrs(struct temp_data *tdata, struct device *dev,
394 int attr_no)
395{
396 int i;
397 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
398 struct device_attribute *devattr, char *buf) = {
399 show_label, show_crit_alarm, show_temp, show_tjmax,
400 show_ttarget };
401 static const char *const suffixes[TOTAL_ATTRS] = {
402 "label", "crit_alarm", "input", "crit", "max"
403 };
404
405 for (i = 0; i < tdata->attr_size; i++) {
406 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH,
407 "temp%d_%s", attr_no, suffixes[i]);
408 sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
409 tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
410 tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
411 tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
412 tdata->sd_attrs[i].index = attr_no;
413 tdata->attrs[i] = &tdata->sd_attrs[i].dev_attr.attr;
414 }
415 tdata->attr_group.attrs = tdata->attrs;
416 return sysfs_create_group(&dev->kobj, &tdata->attr_group);
417}
418
419
420static int chk_ucode_version(unsigned int cpu)
421{
422 struct cpuinfo_x86 *c = &cpu_data(cpu);
423
424 /*
425 * Check if we have problem with errata AE18 of Core processors:
426 * Readings might stop update when processor visited too deep sleep,
427 * fixed for stepping D0 (6EC).
428 */
429 if (c->x86_model == 0xe && c->x86_mask < 0xc && c->microcode < 0x39) {
430 pr_err("Errata AE18 not fixed, update BIOS or microcode of the CPU!\n");
431 return -ENODEV;
432 }
433 return 0;
434}
435
436static struct platform_device *coretemp_get_pdev(unsigned int cpu)
437{
438 u16 phys_proc_id = TO_PHYS_ID(cpu);
439 struct pdev_entry *p;
440
441 mutex_lock(&pdev_list_mutex);
442
443 list_for_each_entry(p, &pdev_list, list)
444 if (p->phys_proc_id == phys_proc_id) {
445 mutex_unlock(&pdev_list_mutex);
446 return p->pdev;
447 }
448
449 mutex_unlock(&pdev_list_mutex);
450 return NULL;
451}
452
453static struct temp_data *init_temp_data(unsigned int cpu, int pkg_flag)
454{
455 struct temp_data *tdata;
456
457 tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
458 if (!tdata)
459 return NULL;
460
461 tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
462 MSR_IA32_THERM_STATUS;
463 tdata->is_pkg_data = pkg_flag;
464 tdata->cpu = cpu;
465 tdata->cpu_core_id = TO_CORE_ID(cpu);
466 tdata->attr_size = MAX_CORE_ATTRS;
467 mutex_init(&tdata->update_lock);
468 return tdata;
469}
470
471static int create_core_data(struct platform_device *pdev, unsigned int cpu,
472 int pkg_flag)
473{
474 struct temp_data *tdata;
475 struct platform_data *pdata = platform_get_drvdata(pdev);
476 struct cpuinfo_x86 *c = &cpu_data(cpu);
477 u32 eax, edx;
478 int err, attr_no;
479
480 /*
481 * Find attr number for sysfs:
482 * We map the attr number to core id of the CPU
483 * The attr number is always core id + 2
484 * The Pkgtemp will always show up as temp1_*, if available
485 */
486 attr_no = pkg_flag ? 1 : TO_ATTR_NO(cpu);
487
488 if (attr_no > MAX_CORE_DATA - 1)
489 return -ERANGE;
490
491 /*
492 * Provide a single set of attributes for all HT siblings of a core
493 * to avoid duplicate sensors (the processor ID and core ID of all
494 * HT siblings of a core are the same).
495 * Skip if a HT sibling of this core is already registered.
496 * This is not an error.
497 */
498 if (pdata->core_data[attr_no] != NULL)
499 return 0;
500
501 tdata = init_temp_data(cpu, pkg_flag);
502 if (!tdata)
503 return -ENOMEM;
504
505 /* Test if we can access the status register */
506 err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
507 if (err)
508 goto exit_free;
509
510 /* We can access status register. Get Critical Temperature */
511 tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
512
513 /*
514 * Read the still undocumented bits 8:15 of IA32_TEMPERATURE_TARGET.
515 * The target temperature is available on older CPUs but not in this
516 * register. Atoms don't have the register at all.
517 */
518 if (c->x86_model > 0xe && c->x86_model != 0x1c) {
519 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET,
520 &eax, &edx);
521 if (!err) {
522 tdata->ttarget
523 = tdata->tjmax - ((eax >> 8) & 0xff) * 1000;
524 tdata->attr_size++;
525 }
526 }
527
528 pdata->core_data[attr_no] = tdata;
529
530 /* Create sysfs interfaces */
531 err = create_core_attrs(tdata, pdata->hwmon_dev, attr_no);
532 if (err)
533 goto exit_free;
534
535 return 0;
536exit_free:
537 pdata->core_data[attr_no] = NULL;
538 kfree(tdata);
539 return err;
540}
541
542static void coretemp_add_core(unsigned int cpu, int pkg_flag)
543{
544 struct platform_device *pdev = coretemp_get_pdev(cpu);
545 int err;
546
547 if (!pdev)
548 return;
549
550 err = create_core_data(pdev, cpu, pkg_flag);
551 if (err)
552 dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
553}
554
555static void coretemp_remove_core(struct platform_data *pdata,
556 int indx)
557{
558 struct temp_data *tdata = pdata->core_data[indx];
559
560 /* Remove the sysfs attributes */
561 sysfs_remove_group(&pdata->hwmon_dev->kobj, &tdata->attr_group);
562
563 kfree(pdata->core_data[indx]);
564 pdata->core_data[indx] = NULL;
565}
566
567static int coretemp_probe(struct platform_device *pdev)
568{
569 struct device *dev = &pdev->dev;
570 struct platform_data *pdata;
571
572 /* Initialize the per-package data structures */
573 pdata = devm_kzalloc(dev, sizeof(struct platform_data), GFP_KERNEL);
574 if (!pdata)
575 return -ENOMEM;
576
577 pdata->phys_proc_id = pdev->id;
578 platform_set_drvdata(pdev, pdata);
579
580 pdata->hwmon_dev = devm_hwmon_device_register_with_groups(dev, DRVNAME,
581 pdata, NULL);
582 return PTR_ERR_OR_ZERO(pdata->hwmon_dev);
583}
584
585static int coretemp_remove(struct platform_device *pdev)
586{
587 struct platform_data *pdata = platform_get_drvdata(pdev);
588 int i;
589
590 for (i = MAX_CORE_DATA - 1; i >= 0; --i)
591 if (pdata->core_data[i])
592 coretemp_remove_core(pdata, i);
593
594 return 0;
595}
596
597static struct platform_driver coretemp_driver = {
598 .driver = {
599 .name = DRVNAME,
600 },
601 .probe = coretemp_probe,
602 .remove = coretemp_remove,
603};
604
605static int coretemp_device_add(unsigned int cpu)
606{
607 int err;
608 struct platform_device *pdev;
609 struct pdev_entry *pdev_entry;
610
611 mutex_lock(&pdev_list_mutex);
612
613 pdev = platform_device_alloc(DRVNAME, TO_PHYS_ID(cpu));
614 if (!pdev) {
615 err = -ENOMEM;
616 pr_err("Device allocation failed\n");
617 goto exit;
618 }
619
620 pdev_entry = kzalloc(sizeof(struct pdev_entry), GFP_KERNEL);
621 if (!pdev_entry) {
622 err = -ENOMEM;
623 goto exit_device_put;
624 }
625
626 err = platform_device_add(pdev);
627 if (err) {
628 pr_err("Device addition failed (%d)\n", err);
629 goto exit_device_free;
630 }
631
632 pdev_entry->pdev = pdev;
633 pdev_entry->phys_proc_id = pdev->id;
634
635 list_add_tail(&pdev_entry->list, &pdev_list);
636 mutex_unlock(&pdev_list_mutex);
637
638 return 0;
639
640exit_device_free:
641 kfree(pdev_entry);
642exit_device_put:
643 platform_device_put(pdev);
644exit:
645 mutex_unlock(&pdev_list_mutex);
646 return err;
647}
648
649static void coretemp_device_remove(unsigned int cpu)
650{
651 struct pdev_entry *p, *n;
652 u16 phys_proc_id = TO_PHYS_ID(cpu);
653
654 mutex_lock(&pdev_list_mutex);
655 list_for_each_entry_safe(p, n, &pdev_list, list) {
656 if (p->phys_proc_id != phys_proc_id)
657 continue;
658 platform_device_unregister(p->pdev);
659 list_del(&p->list);
660 kfree(p);
661 }
662 mutex_unlock(&pdev_list_mutex);
663}
664
665static bool is_any_core_online(struct platform_data *pdata)
666{
667 int i;
668
669 /* Find online cores, except pkgtemp data */
670 for (i = MAX_CORE_DATA - 1; i >= 0; --i) {
671 if (pdata->core_data[i] &&
672 !pdata->core_data[i]->is_pkg_data) {
673 return true;
674 }
675 }
676 return false;
677}
678
679static void get_core_online(unsigned int cpu)
680{
681 struct cpuinfo_x86 *c = &cpu_data(cpu);
682 struct platform_device *pdev = coretemp_get_pdev(cpu);
683 int err;
684
685 /*
686 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
687 * sensors. We check this bit only, all the early CPUs
688 * without thermal sensors will be filtered out.
689 */
690 if (!cpu_has(c, X86_FEATURE_DTHERM))
691 return;
692
693 if (!pdev) {
694 /* Check the microcode version of the CPU */
695 if (chk_ucode_version(cpu))
696 return;
697
698 /*
699 * Alright, we have DTS support.
700 * We are bringing the _first_ core in this pkg
701 * online. So, initialize per-pkg data structures and
702 * then bring this core online.
703 */
704 err = coretemp_device_add(cpu);
705 if (err)
706 return;
707 /*
708 * Check whether pkgtemp support is available.
709 * If so, add interfaces for pkgtemp.
710 */
711 if (cpu_has(c, X86_FEATURE_PTS))
712 coretemp_add_core(cpu, 1);
713 }
714 /*
715 * Physical CPU device already exists.
716 * So, just add interfaces for this core.
717 */
718 coretemp_add_core(cpu, 0);
719}
720
721static void put_core_offline(unsigned int cpu)
722{
723 int i, indx;
724 struct platform_data *pdata;
725 struct platform_device *pdev = coretemp_get_pdev(cpu);
726
727 /* If the physical CPU device does not exist, just return */
728 if (!pdev)
729 return;
730
731 pdata = platform_get_drvdata(pdev);
732
733 indx = TO_ATTR_NO(cpu);
734
735 /* The core id is too big, just return */
736 if (indx > MAX_CORE_DATA - 1)
737 return;
738
739 if (pdata->core_data[indx] && pdata->core_data[indx]->cpu == cpu)
740 coretemp_remove_core(pdata, indx);
741
742 /*
743 * If a HT sibling of a core is taken offline, but another HT sibling
744 * of the same core is still online, register the alternate sibling.
745 * This ensures that exactly one set of attributes is provided as long
746 * as at least one HT sibling of a core is online.
747 */
748 for_each_sibling(i, cpu) {
749 if (i != cpu) {
750 get_core_online(i);
751 /*
752 * Display temperature sensor data for one HT sibling
753 * per core only, so abort the loop after one such
754 * sibling has been found.
755 */
756 break;
757 }
758 }
759 /*
760 * If all cores in this pkg are offline, remove the device.
761 * coretemp_device_remove calls unregister_platform_device,
762 * which in turn calls coretemp_remove. This removes the
763 * pkgtemp entry and does other clean ups.
764 */
765 if (!is_any_core_online(pdata))
766 coretemp_device_remove(cpu);
767}
768
769static int coretemp_cpu_callback(struct notifier_block *nfb,
770 unsigned long action, void *hcpu)
771{
772 unsigned int cpu = (unsigned long) hcpu;
773
774 switch (action) {
775 case CPU_ONLINE:
776 case CPU_DOWN_FAILED:
777 get_core_online(cpu);
778 break;
779 case CPU_DOWN_PREPARE:
780 put_core_offline(cpu);
781 break;
782 }
783 return NOTIFY_OK;
784}
785
786static struct notifier_block coretemp_cpu_notifier __refdata = {
787 .notifier_call = coretemp_cpu_callback,
788};
789
790static const struct x86_cpu_id __initconst coretemp_ids[] = {
791 { X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_DTHERM },
792 {}
793};
794MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
795
796static int __init coretemp_init(void)
797{
798 int i, err;
799
800 /*
801 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
802 * sensors. We check this bit only, all the early CPUs
803 * without thermal sensors will be filtered out.
804 */
805 if (!x86_match_cpu(coretemp_ids))
806 return -ENODEV;
807
808 err = platform_driver_register(&coretemp_driver);
809 if (err)
810 goto exit;
811
812 cpu_notifier_register_begin();
813 for_each_online_cpu(i)
814 get_core_online(i);
815
816#ifndef CONFIG_HOTPLUG_CPU
817 if (list_empty(&pdev_list)) {
818 cpu_notifier_register_done();
819 err = -ENODEV;
820 goto exit_driver_unreg;
821 }
822#endif
823
824 __register_hotcpu_notifier(&coretemp_cpu_notifier);
825 cpu_notifier_register_done();
826 return 0;
827
828#ifndef CONFIG_HOTPLUG_CPU
829exit_driver_unreg:
830 platform_driver_unregister(&coretemp_driver);
831#endif
832exit:
833 return err;
834}
835
836static void __exit coretemp_exit(void)
837{
838 struct pdev_entry *p, *n;
839
840 cpu_notifier_register_begin();
841 __unregister_hotcpu_notifier(&coretemp_cpu_notifier);
842 mutex_lock(&pdev_list_mutex);
843 list_for_each_entry_safe(p, n, &pdev_list, list) {
844 platform_device_unregister(p->pdev);
845 list_del(&p->list);
846 kfree(p);
847 }
848 mutex_unlock(&pdev_list_mutex);
849 cpu_notifier_register_done();
850 platform_driver_unregister(&coretemp_driver);
851}
852
853MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
854MODULE_DESCRIPTION("Intel Core temperature monitor");
855MODULE_LICENSE("GPL");
856
857module_init(coretemp_init)
858module_exit(coretemp_exit)