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#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");