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