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