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 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) & 0xff) * 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");
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 32 /* 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) for_each_cpu(i, cpu_sibling_mask(cpu))
67#else
68#define for_each_sibling(i, cpu) for (i = 0; false; )
69#endif
70
71/*
72 * Per-Core Temperature Data
73 * @last_updated: The time when the current temperature value was updated
74 * earlier (in jiffies).
75 * @cpu_core_id: The CPU Core from which temperature values should be read
76 * This value is passed as "id" field to rdmsr/wrmsr functions.
77 * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
78 * from where the temperature values should be read.
79 * @attr_size: Total number of pre-core attrs displayed in the sysfs.
80 * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
81 * Otherwise, temp_data holds coretemp data.
82 * @valid: If this is 1, the current temperature is valid.
83 */
84struct temp_data {
85 int temp;
86 int ttarget;
87 int tjmax;
88 unsigned long last_updated;
89 unsigned int cpu;
90 u32 cpu_core_id;
91 u32 status_reg;
92 int attr_size;
93 bool is_pkg_data;
94 bool valid;
95 struct sensor_device_attribute sd_attrs[TOTAL_ATTRS];
96 char attr_name[TOTAL_ATTRS][CORETEMP_NAME_LENGTH];
97 struct attribute *attrs[TOTAL_ATTRS + 1];
98 struct attribute_group attr_group;
99 struct mutex update_lock;
100};
101
102/* Platform Data per Physical CPU */
103struct platform_data {
104 struct device *hwmon_dev;
105 u16 phys_proc_id;
106 struct temp_data *core_data[MAX_CORE_DATA];
107 struct device_attribute name_attr;
108};
109
110struct pdev_entry {
111 struct list_head list;
112 struct platform_device *pdev;
113 u16 phys_proc_id;
114};
115
116static LIST_HEAD(pdev_list);
117static DEFINE_MUTEX(pdev_list_mutex);
118
119static ssize_t show_label(struct device *dev,
120 struct device_attribute *devattr, char *buf)
121{
122 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
123 struct platform_data *pdata = dev_get_drvdata(dev);
124 struct temp_data *tdata = pdata->core_data[attr->index];
125
126 if (tdata->is_pkg_data)
127 return sprintf(buf, "Physical id %u\n", pdata->phys_proc_id);
128
129 return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
130}
131
132static ssize_t show_crit_alarm(struct device *dev,
133 struct device_attribute *devattr, char *buf)
134{
135 u32 eax, edx;
136 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
137 struct platform_data *pdata = dev_get_drvdata(dev);
138 struct temp_data *tdata = pdata->core_data[attr->index];
139
140 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
141
142 return sprintf(buf, "%d\n", (eax >> 5) & 1);
143}
144
145static ssize_t show_tjmax(struct device *dev,
146 struct device_attribute *devattr, char *buf)
147{
148 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
149 struct platform_data *pdata = dev_get_drvdata(dev);
150
151 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
152}
153
154static ssize_t show_ttarget(struct device *dev,
155 struct device_attribute *devattr, char *buf)
156{
157 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
158 struct platform_data *pdata = dev_get_drvdata(dev);
159
160 return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
161}
162
163static ssize_t show_temp(struct device *dev,
164 struct device_attribute *devattr, char *buf)
165{
166 u32 eax, edx;
167 struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
168 struct platform_data *pdata = dev_get_drvdata(dev);
169 struct temp_data *tdata = pdata->core_data[attr->index];
170
171 mutex_lock(&tdata->update_lock);
172
173 /* Check whether the time interval has elapsed */
174 if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
175 rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
176 /*
177 * Ignore the valid bit. In all observed cases the register
178 * value is either low or zero if the valid bit is 0.
179 * Return it instead of reporting an error which doesn't
180 * really help at all.
181 */
182 tdata->temp = tdata->tjmax - ((eax >> 16) & 0x7f) * 1000;
183 tdata->valid = 1;
184 tdata->last_updated = jiffies;
185 }
186
187 mutex_unlock(&tdata->update_lock);
188 return sprintf(buf, "%d\n", tdata->temp);
189}
190
191struct tjmax_pci {
192 unsigned int device;
193 int tjmax;
194};
195
196static const struct tjmax_pci tjmax_pci_table[] = {
197 { 0x0708, 110000 }, /* CE41x0 (Sodaville ) */
198 { 0x0c72, 102000 }, /* Atom S1240 (Centerton) */
199 { 0x0c73, 95000 }, /* Atom S1220 (Centerton) */
200 { 0x0c75, 95000 }, /* Atom S1260 (Centerton) */
201};
202
203struct tjmax {
204 char const *id;
205 int tjmax;
206};
207
208static const struct tjmax tjmax_table[] = {
209 { "CPU 230", 100000 }, /* Model 0x1c, stepping 2 */
210 { "CPU 330", 125000 }, /* Model 0x1c, stepping 2 */
211};
212
213struct tjmax_model {
214 u8 model;
215 u8 mask;
216 int tjmax;
217};
218
219#define ANY 0xff
220
221static const struct tjmax_model tjmax_model_table[] = {
222 { 0x1c, 10, 100000 }, /* D4xx, K4xx, N4xx, D5xx, K5xx, N5xx */
223 { 0x1c, ANY, 90000 }, /* Z5xx, N2xx, possibly others
224 * Note: Also matches 230 and 330,
225 * which are covered by tjmax_table
226 */
227 { 0x26, ANY, 90000 }, /* Atom Tunnel Creek (Exx), Lincroft (Z6xx)
228 * Note: TjMax for E6xxT is 110C, but CPU type
229 * is undetectable by software
230 */
231 { 0x27, ANY, 90000 }, /* Atom Medfield (Z2460) */
232 { 0x35, ANY, 90000 }, /* Atom Clover Trail/Cloverview (Z27x0) */
233 { 0x36, ANY, 100000 }, /* Atom Cedar Trail/Cedarview (N2xxx, D2xxx)
234 * Also matches S12x0 (stepping 9), covered by
235 * PCI table
236 */
237};
238
239static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
240{
241 /* The 100C is default for both mobile and non mobile CPUs */
242
243 int tjmax = 100000;
244 int tjmax_ee = 85000;
245 int usemsr_ee = 1;
246 int err;
247 u32 eax, edx;
248 int i;
249 struct pci_dev *host_bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));
250
251 /*
252 * Explicit tjmax table entries override heuristics.
253 * First try PCI host bridge IDs, followed by model ID strings
254 * and model/stepping information.
255 */
256 if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL) {
257 for (i = 0; i < ARRAY_SIZE(tjmax_pci_table); i++) {
258 if (host_bridge->device == tjmax_pci_table[i].device)
259 return tjmax_pci_table[i].tjmax;
260 }
261 }
262
263 for (i = 0; i < ARRAY_SIZE(tjmax_table); i++) {
264 if (strstr(c->x86_model_id, tjmax_table[i].id))
265 return tjmax_table[i].tjmax;
266 }
267
268 for (i = 0; i < ARRAY_SIZE(tjmax_model_table); i++) {
269 const struct tjmax_model *tm = &tjmax_model_table[i];
270 if (c->x86_model == tm->model &&
271 (tm->mask == ANY || c->x86_mask == tm->mask))
272 return tm->tjmax;
273 }
274
275 /* Early chips have no MSR for TjMax */
276
277 if (c->x86_model == 0xf && c->x86_mask < 4)
278 usemsr_ee = 0;
279
280 if (c->x86_model > 0xe && usemsr_ee) {
281 u8 platform_id;
282
283 /*
284 * Now we can detect the mobile CPU using Intel provided table
285 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
286 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
287 */
288 err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
289 if (err) {
290 dev_warn(dev,
291 "Unable to access MSR 0x17, assuming desktop"
292 " CPU\n");
293 usemsr_ee = 0;
294 } else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
295 /*
296 * Trust bit 28 up to Penryn, I could not find any
297 * documentation on that; if you happen to know
298 * someone at Intel please ask
299 */
300 usemsr_ee = 0;
301 } else {
302 /* Platform ID bits 52:50 (EDX starts at bit 32) */
303 platform_id = (edx >> 18) & 0x7;
304
305 /*
306 * Mobile Penryn CPU seems to be platform ID 7 or 5
307 * (guesswork)
308 */
309 if (c->x86_model == 0x17 &&
310 (platform_id == 5 || platform_id == 7)) {
311 /*
312 * If MSR EE bit is set, set it to 90 degrees C,
313 * otherwise 105 degrees C
314 */
315 tjmax_ee = 90000;
316 tjmax = 105000;
317 }
318 }
319 }
320
321 if (usemsr_ee) {
322 err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
323 if (err) {
324 dev_warn(dev,
325 "Unable to access MSR 0xEE, for Tjmax, left"
326 " at default\n");
327 } else if (eax & 0x40000000) {
328 tjmax = tjmax_ee;
329 }
330 } else if (tjmax == 100000) {
331 /*
332 * If we don't use msr EE it means we are desktop CPU
333 * (with exeception of Atom)
334 */
335 dev_warn(dev, "Using relative temperature scale!\n");
336 }
337
338 return tjmax;
339}
340
341static bool cpu_has_tjmax(struct cpuinfo_x86 *c)
342{
343 u8 model = c->x86_model;
344
345 return model > 0xe &&
346 model != 0x1c &&
347 model != 0x26 &&
348 model != 0x27 &&
349 model != 0x35 &&
350 model != 0x36;
351}
352
353static int get_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
354{
355 int err;
356 u32 eax, edx;
357 u32 val;
358
359 /*
360 * A new feature of current Intel(R) processors, the
361 * IA32_TEMPERATURE_TARGET contains the TjMax value
362 */
363 err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
364 if (err) {
365 if (cpu_has_tjmax(c))
366 dev_warn(dev, "Unable to read TjMax from CPU %u\n", id);
367 } else {
368 val = (eax >> 16) & 0xff;
369 /*
370 * If the TjMax is not plausible, an assumption
371 * will be used
372 */
373 if (val) {
374 dev_dbg(dev, "TjMax is %d degrees C\n", val);
375 return val * 1000;
376 }
377 }
378
379 if (force_tjmax) {
380 dev_notice(dev, "TjMax forced to %d degrees C by user\n",
381 force_tjmax);
382 return force_tjmax * 1000;
383 }
384
385 /*
386 * An assumption is made for early CPUs and unreadable MSR.
387 * NOTE: the calculated value may not be correct.
388 */
389 return adjust_tjmax(c, id, dev);
390}
391
392static int create_core_attrs(struct temp_data *tdata, struct device *dev,
393 int attr_no)
394{
395 int i;
396 static ssize_t (*const rd_ptr[TOTAL_ATTRS]) (struct device *dev,
397 struct device_attribute *devattr, char *buf) = {
398 show_label, show_crit_alarm, show_temp, show_tjmax,
399 show_ttarget };
400 static const char *const names[TOTAL_ATTRS] = {
401 "temp%d_label", "temp%d_crit_alarm",
402 "temp%d_input", "temp%d_crit",
403 "temp%d_max" };
404
405 for (i = 0; i < tdata->attr_size; i++) {
406 snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH, names[i],
407 attr_no);
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 .owner = THIS_MODULE,
600 .name = DRVNAME,
601 },
602 .probe = coretemp_probe,
603 .remove = coretemp_remove,
604};
605
606static int coretemp_device_add(unsigned int cpu)
607{
608 int err;
609 struct platform_device *pdev;
610 struct pdev_entry *pdev_entry;
611
612 mutex_lock(&pdev_list_mutex);
613
614 pdev = platform_device_alloc(DRVNAME, TO_PHYS_ID(cpu));
615 if (!pdev) {
616 err = -ENOMEM;
617 pr_err("Device allocation failed\n");
618 goto exit;
619 }
620
621 pdev_entry = kzalloc(sizeof(struct pdev_entry), GFP_KERNEL);
622 if (!pdev_entry) {
623 err = -ENOMEM;
624 goto exit_device_put;
625 }
626
627 err = platform_device_add(pdev);
628 if (err) {
629 pr_err("Device addition failed (%d)\n", err);
630 goto exit_device_free;
631 }
632
633 pdev_entry->pdev = pdev;
634 pdev_entry->phys_proc_id = pdev->id;
635
636 list_add_tail(&pdev_entry->list, &pdev_list);
637 mutex_unlock(&pdev_list_mutex);
638
639 return 0;
640
641exit_device_free:
642 kfree(pdev_entry);
643exit_device_put:
644 platform_device_put(pdev);
645exit:
646 mutex_unlock(&pdev_list_mutex);
647 return err;
648}
649
650static void coretemp_device_remove(unsigned int cpu)
651{
652 struct pdev_entry *p, *n;
653 u16 phys_proc_id = TO_PHYS_ID(cpu);
654
655 mutex_lock(&pdev_list_mutex);
656 list_for_each_entry_safe(p, n, &pdev_list, list) {
657 if (p->phys_proc_id != phys_proc_id)
658 continue;
659 platform_device_unregister(p->pdev);
660 list_del(&p->list);
661 kfree(p);
662 }
663 mutex_unlock(&pdev_list_mutex);
664}
665
666static bool is_any_core_online(struct platform_data *pdata)
667{
668 int i;
669
670 /* Find online cores, except pkgtemp data */
671 for (i = MAX_CORE_DATA - 1; i >= 0; --i) {
672 if (pdata->core_data[i] &&
673 !pdata->core_data[i]->is_pkg_data) {
674 return true;
675 }
676 }
677 return false;
678}
679
680static void get_core_online(unsigned int cpu)
681{
682 struct cpuinfo_x86 *c = &cpu_data(cpu);
683 struct platform_device *pdev = coretemp_get_pdev(cpu);
684 int err;
685
686 /*
687 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
688 * sensors. We check this bit only, all the early CPUs
689 * without thermal sensors will be filtered out.
690 */
691 if (!cpu_has(c, X86_FEATURE_DTHERM))
692 return;
693
694 if (!pdev) {
695 /* Check the microcode version of the CPU */
696 if (chk_ucode_version(cpu))
697 return;
698
699 /*
700 * Alright, we have DTS support.
701 * We are bringing the _first_ core in this pkg
702 * online. So, initialize per-pkg data structures and
703 * then bring this core online.
704 */
705 err = coretemp_device_add(cpu);
706 if (err)
707 return;
708 /*
709 * Check whether pkgtemp support is available.
710 * If so, add interfaces for pkgtemp.
711 */
712 if (cpu_has(c, X86_FEATURE_PTS))
713 coretemp_add_core(cpu, 1);
714 }
715 /*
716 * Physical CPU device already exists.
717 * So, just add interfaces for this core.
718 */
719 coretemp_add_core(cpu, 0);
720}
721
722static void put_core_offline(unsigned int cpu)
723{
724 int i, indx;
725 struct platform_data *pdata;
726 struct platform_device *pdev = coretemp_get_pdev(cpu);
727
728 /* If the physical CPU device does not exist, just return */
729 if (!pdev)
730 return;
731
732 pdata = platform_get_drvdata(pdev);
733
734 indx = TO_ATTR_NO(cpu);
735
736 /* The core id is too big, just return */
737 if (indx > MAX_CORE_DATA - 1)
738 return;
739
740 if (pdata->core_data[indx] && pdata->core_data[indx]->cpu == cpu)
741 coretemp_remove_core(pdata, indx);
742
743 /*
744 * If a HT sibling of a core is taken offline, but another HT sibling
745 * of the same core is still online, register the alternate sibling.
746 * This ensures that exactly one set of attributes is provided as long
747 * as at least one HT sibling of a core is online.
748 */
749 for_each_sibling(i, cpu) {
750 if (i != cpu) {
751 get_core_online(i);
752 /*
753 * Display temperature sensor data for one HT sibling
754 * per core only, so abort the loop after one such
755 * sibling has been found.
756 */
757 break;
758 }
759 }
760 /*
761 * If all cores in this pkg are offline, remove the device.
762 * coretemp_device_remove calls unregister_platform_device,
763 * which in turn calls coretemp_remove. This removes the
764 * pkgtemp entry and does other clean ups.
765 */
766 if (!is_any_core_online(pdata))
767 coretemp_device_remove(cpu);
768}
769
770static int coretemp_cpu_callback(struct notifier_block *nfb,
771 unsigned long action, void *hcpu)
772{
773 unsigned int cpu = (unsigned long) hcpu;
774
775 switch (action) {
776 case CPU_ONLINE:
777 case CPU_DOWN_FAILED:
778 get_core_online(cpu);
779 break;
780 case CPU_DOWN_PREPARE:
781 put_core_offline(cpu);
782 break;
783 }
784 return NOTIFY_OK;
785}
786
787static struct notifier_block coretemp_cpu_notifier __refdata = {
788 .notifier_call = coretemp_cpu_callback,
789};
790
791static const struct x86_cpu_id __initconst coretemp_ids[] = {
792 { X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_DTHERM },
793 {}
794};
795MODULE_DEVICE_TABLE(x86cpu, coretemp_ids);
796
797static int __init coretemp_init(void)
798{
799 int i, err;
800
801 /*
802 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
803 * sensors. We check this bit only, all the early CPUs
804 * without thermal sensors will be filtered out.
805 */
806 if (!x86_match_cpu(coretemp_ids))
807 return -ENODEV;
808
809 err = platform_driver_register(&coretemp_driver);
810 if (err)
811 goto exit;
812
813 cpu_notifier_register_begin();
814 for_each_online_cpu(i)
815 get_core_online(i);
816
817#ifndef CONFIG_HOTPLUG_CPU
818 if (list_empty(&pdev_list)) {
819 cpu_notifier_register_done();
820 err = -ENODEV;
821 goto exit_driver_unreg;
822 }
823#endif
824
825 __register_hotcpu_notifier(&coretemp_cpu_notifier);
826 cpu_notifier_register_done();
827 return 0;
828
829#ifndef CONFIG_HOTPLUG_CPU
830exit_driver_unreg:
831 platform_driver_unregister(&coretemp_driver);
832#endif
833exit:
834 return err;
835}
836
837static void __exit coretemp_exit(void)
838{
839 struct pdev_entry *p, *n;
840
841 cpu_notifier_register_begin();
842 __unregister_hotcpu_notifier(&coretemp_cpu_notifier);
843 mutex_lock(&pdev_list_mutex);
844 list_for_each_entry_safe(p, n, &pdev_list, list) {
845 platform_device_unregister(p->pdev);
846 list_del(&p->list);
847 kfree(p);
848 }
849 mutex_unlock(&pdev_list_mutex);
850 cpu_notifier_register_done();
851 platform_driver_unregister(&coretemp_driver);
852}
853
854MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
855MODULE_DESCRIPTION("Intel Core temperature monitor");
856MODULE_LICENSE("GPL");
857
858module_init(coretemp_init)
859module_exit(coretemp_exit)