xref: /linux/drivers/hwmon/coretemp.c (revision e9e8bcb8178e197d889ec31e79fa1ddc1732c8f9)
1 /*
2  * coretemp.c - Linux kernel module for hardware monitoring
3  *
4  * Copyright (C) 2007 Rudolf Marek <r.marek@assembler.cz>
5  *
6  * Inspired from many hwmon drivers
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; version 2 of the License.
11  *
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License
18  * along with this program; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
20  * 02110-1301 USA.
21  */
22 
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24 
25 #include <linux/module.h>
26 #include <linux/init.h>
27 #include <linux/slab.h>
28 #include <linux/jiffies.h>
29 #include <linux/hwmon.h>
30 #include <linux/sysfs.h>
31 #include <linux/hwmon-sysfs.h>
32 #include <linux/err.h>
33 #include <linux/mutex.h>
34 #include <linux/list.h>
35 #include <linux/platform_device.h>
36 #include <linux/cpu.h>
37 #include <linux/pci.h>
38 #include <linux/smp.h>
39 #include <asm/msr.h>
40 #include <asm/processor.h>
41 
42 #define DRVNAME	"coretemp"
43 
44 #define BASE_SYSFS_ATTR_NO	2	/* Sysfs Base attr no for coretemp */
45 #define NUM_REAL_CORES		16	/* Number of Real cores per cpu */
46 #define CORETEMP_NAME_LENGTH	17	/* String Length of attrs */
47 #define MAX_ATTRS		5	/* Maximum no of per-core attrs */
48 #define MAX_CORE_DATA		(NUM_REAL_CORES + BASE_SYSFS_ATTR_NO)
49 
50 #ifdef CONFIG_SMP
51 #define TO_PHYS_ID(cpu)		cpu_data(cpu).phys_proc_id
52 #define TO_CORE_ID(cpu)		cpu_data(cpu).cpu_core_id
53 #define TO_ATTR_NO(cpu)		(TO_CORE_ID(cpu) + BASE_SYSFS_ATTR_NO)
54 #define for_each_sibling(i, cpu)	for_each_cpu(i, cpu_sibling_mask(cpu))
55 #else
56 #define TO_PHYS_ID(cpu)		(cpu)
57 #define TO_CORE_ID(cpu)		(cpu)
58 #define TO_ATTR_NO(cpu)		(cpu)
59 #define for_each_sibling(i, cpu)	for (i = 0; false; )
60 #endif
61 
62 /*
63  * Per-Core Temperature Data
64  * @last_updated: The time when the current temperature value was updated
65  *		earlier (in jiffies).
66  * @cpu_core_id: The CPU Core from which temperature values should be read
67  *		This value is passed as "id" field to rdmsr/wrmsr functions.
68  * @status_reg: One of IA32_THERM_STATUS or IA32_PACKAGE_THERM_STATUS,
69  *		from where the temperature values should be read.
70  * @is_pkg_data: If this is 1, the temp_data holds pkgtemp data.
71  *		Otherwise, temp_data holds coretemp data.
72  * @valid: If this is 1, the current temperature is valid.
73  */
74 struct temp_data {
75 	int temp;
76 	int ttarget;
77 	int tjmax;
78 	unsigned long last_updated;
79 	unsigned int cpu;
80 	u32 cpu_core_id;
81 	u32 status_reg;
82 	bool is_pkg_data;
83 	bool valid;
84 	struct sensor_device_attribute sd_attrs[MAX_ATTRS];
85 	char attr_name[MAX_ATTRS][CORETEMP_NAME_LENGTH];
86 	struct mutex update_lock;
87 };
88 
89 /* Platform Data per Physical CPU */
90 struct platform_data {
91 	struct device *hwmon_dev;
92 	u16 phys_proc_id;
93 	struct temp_data *core_data[MAX_CORE_DATA];
94 	struct device_attribute name_attr;
95 };
96 
97 struct pdev_entry {
98 	struct list_head list;
99 	struct platform_device *pdev;
100 	unsigned int cpu;
101 	u16 phys_proc_id;
102 	u16 cpu_core_id;
103 };
104 
105 static LIST_HEAD(pdev_list);
106 static DEFINE_MUTEX(pdev_list_mutex);
107 
108 static ssize_t show_name(struct device *dev,
109 			struct device_attribute *devattr, char *buf)
110 {
111 	return sprintf(buf, "%s\n", DRVNAME);
112 }
113 
114 static ssize_t show_label(struct device *dev,
115 				struct device_attribute *devattr, char *buf)
116 {
117 	struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
118 	struct platform_data *pdata = dev_get_drvdata(dev);
119 	struct temp_data *tdata = pdata->core_data[attr->index];
120 
121 	if (tdata->is_pkg_data)
122 		return sprintf(buf, "Physical id %u\n", pdata->phys_proc_id);
123 
124 	return sprintf(buf, "Core %u\n", tdata->cpu_core_id);
125 }
126 
127 static ssize_t show_crit_alarm(struct device *dev,
128 				struct device_attribute *devattr, char *buf)
129 {
130 	u32 eax, edx;
131 	struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
132 	struct platform_data *pdata = dev_get_drvdata(dev);
133 	struct temp_data *tdata = pdata->core_data[attr->index];
134 
135 	rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
136 
137 	return sprintf(buf, "%d\n", (eax >> 5) & 1);
138 }
139 
140 static ssize_t show_tjmax(struct device *dev,
141 			struct device_attribute *devattr, char *buf)
142 {
143 	struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
144 	struct platform_data *pdata = dev_get_drvdata(dev);
145 
146 	return sprintf(buf, "%d\n", pdata->core_data[attr->index]->tjmax);
147 }
148 
149 static ssize_t show_ttarget(struct device *dev,
150 				struct device_attribute *devattr, char *buf)
151 {
152 	struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
153 	struct platform_data *pdata = dev_get_drvdata(dev);
154 
155 	return sprintf(buf, "%d\n", pdata->core_data[attr->index]->ttarget);
156 }
157 
158 static ssize_t show_temp(struct device *dev,
159 			struct device_attribute *devattr, char *buf)
160 {
161 	u32 eax, edx;
162 	struct sensor_device_attribute *attr = to_sensor_dev_attr(devattr);
163 	struct platform_data *pdata = dev_get_drvdata(dev);
164 	struct temp_data *tdata = pdata->core_data[attr->index];
165 
166 	mutex_lock(&tdata->update_lock);
167 
168 	/* Check whether the time interval has elapsed */
169 	if (!tdata->valid || time_after(jiffies, tdata->last_updated + HZ)) {
170 		rdmsr_on_cpu(tdata->cpu, tdata->status_reg, &eax, &edx);
171 		tdata->valid = 0;
172 		/* Check whether the data is valid */
173 		if (eax & 0x80000000) {
174 			tdata->temp = tdata->tjmax -
175 					((eax >> 16) & 0x7f) * 1000;
176 			tdata->valid = 1;
177 		}
178 		tdata->last_updated = jiffies;
179 	}
180 
181 	mutex_unlock(&tdata->update_lock);
182 	return tdata->valid ? sprintf(buf, "%d\n", tdata->temp) : -EAGAIN;
183 }
184 
185 static int adjust_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
186 {
187 	/* The 100C is default for both mobile and non mobile CPUs */
188 
189 	int tjmax = 100000;
190 	int tjmax_ee = 85000;
191 	int usemsr_ee = 1;
192 	int err;
193 	u32 eax, edx;
194 	struct pci_dev *host_bridge;
195 
196 	/* Early chips have no MSR for TjMax */
197 
198 	if (c->x86_model == 0xf && c->x86_mask < 4)
199 		usemsr_ee = 0;
200 
201 	/* Atom CPUs */
202 
203 	if (c->x86_model == 0x1c) {
204 		usemsr_ee = 0;
205 
206 		host_bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 0));
207 
208 		if (host_bridge && host_bridge->vendor == PCI_VENDOR_ID_INTEL
209 		    && (host_bridge->device == 0xa000	/* NM10 based nettop */
210 		    || host_bridge->device == 0xa010))	/* NM10 based netbook */
211 			tjmax = 100000;
212 		else
213 			tjmax = 90000;
214 
215 		pci_dev_put(host_bridge);
216 	}
217 
218 	if (c->x86_model > 0xe && usemsr_ee) {
219 		u8 platform_id;
220 
221 		/*
222 		 * Now we can detect the mobile CPU using Intel provided table
223 		 * http://softwarecommunity.intel.com/Wiki/Mobility/720.htm
224 		 * For Core2 cores, check MSR 0x17, bit 28 1 = Mobile CPU
225 		 */
226 		err = rdmsr_safe_on_cpu(id, 0x17, &eax, &edx);
227 		if (err) {
228 			dev_warn(dev,
229 				 "Unable to access MSR 0x17, assuming desktop"
230 				 " CPU\n");
231 			usemsr_ee = 0;
232 		} else if (c->x86_model < 0x17 && !(eax & 0x10000000)) {
233 			/*
234 			 * Trust bit 28 up to Penryn, I could not find any
235 			 * documentation on that; if you happen to know
236 			 * someone at Intel please ask
237 			 */
238 			usemsr_ee = 0;
239 		} else {
240 			/* Platform ID bits 52:50 (EDX starts at bit 32) */
241 			platform_id = (edx >> 18) & 0x7;
242 
243 			/*
244 			 * Mobile Penryn CPU seems to be platform ID 7 or 5
245 			 * (guesswork)
246 			 */
247 			if (c->x86_model == 0x17 &&
248 			    (platform_id == 5 || platform_id == 7)) {
249 				/*
250 				 * If MSR EE bit is set, set it to 90 degrees C,
251 				 * otherwise 105 degrees C
252 				 */
253 				tjmax_ee = 90000;
254 				tjmax = 105000;
255 			}
256 		}
257 	}
258 
259 	if (usemsr_ee) {
260 		err = rdmsr_safe_on_cpu(id, 0xee, &eax, &edx);
261 		if (err) {
262 			dev_warn(dev,
263 				 "Unable to access MSR 0xEE, for Tjmax, left"
264 				 " at default\n");
265 		} else if (eax & 0x40000000) {
266 			tjmax = tjmax_ee;
267 		}
268 	} else if (tjmax == 100000) {
269 		/*
270 		 * If we don't use msr EE it means we are desktop CPU
271 		 * (with exeception of Atom)
272 		 */
273 		dev_warn(dev, "Using relative temperature scale!\n");
274 	}
275 
276 	return tjmax;
277 }
278 
279 static int get_tjmax(struct cpuinfo_x86 *c, u32 id, struct device *dev)
280 {
281 	/* The 100C is default for both mobile and non mobile CPUs */
282 	int err;
283 	u32 eax, edx;
284 	u32 val;
285 
286 	/*
287 	 * A new feature of current Intel(R) processors, the
288 	 * IA32_TEMPERATURE_TARGET contains the TjMax value
289 	 */
290 	err = rdmsr_safe_on_cpu(id, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
291 	if (err) {
292 		dev_warn(dev, "Unable to read TjMax from CPU.\n");
293 	} else {
294 		val = (eax >> 16) & 0xff;
295 		/*
296 		 * If the TjMax is not plausible, an assumption
297 		 * will be used
298 		 */
299 		if (val) {
300 			dev_info(dev, "TjMax is %d C.\n", val);
301 			return val * 1000;
302 		}
303 	}
304 
305 	/*
306 	 * An assumption is made for early CPUs and unreadable MSR.
307 	 * NOTE: the calculated value may not be correct.
308 	 */
309 	return adjust_tjmax(c, id, dev);
310 }
311 
312 static void __devinit get_ucode_rev_on_cpu(void *edx)
313 {
314 	u32 eax;
315 
316 	wrmsr(MSR_IA32_UCODE_REV, 0, 0);
317 	sync_core();
318 	rdmsr(MSR_IA32_UCODE_REV, eax, *(u32 *)edx);
319 }
320 
321 static int get_pkg_tjmax(unsigned int cpu, struct device *dev)
322 {
323 	int err;
324 	u32 eax, edx, val;
325 
326 	err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
327 	if (!err) {
328 		val = (eax >> 16) & 0xff;
329 		if (val)
330 			return val * 1000;
331 	}
332 	dev_warn(dev, "Unable to read Pkg-TjMax from CPU:%u\n", cpu);
333 	return 100000; /* Default TjMax: 100 degree celsius */
334 }
335 
336 static int create_name_attr(struct platform_data *pdata, struct device *dev)
337 {
338 	sysfs_attr_init(&pdata->name_attr.attr);
339 	pdata->name_attr.attr.name = "name";
340 	pdata->name_attr.attr.mode = S_IRUGO;
341 	pdata->name_attr.show = show_name;
342 	return device_create_file(dev, &pdata->name_attr);
343 }
344 
345 static int create_core_attrs(struct temp_data *tdata, struct device *dev,
346 				int attr_no)
347 {
348 	int err, i;
349 	static ssize_t (*rd_ptr[MAX_ATTRS]) (struct device *dev,
350 			struct device_attribute *devattr, char *buf) = {
351 			show_label, show_crit_alarm, show_ttarget,
352 			show_temp, show_tjmax };
353 	static const char *names[MAX_ATTRS] = {
354 					"temp%d_label", "temp%d_crit_alarm",
355 					"temp%d_max", "temp%d_input",
356 					"temp%d_crit" };
357 
358 	for (i = 0; i < MAX_ATTRS; i++) {
359 		snprintf(tdata->attr_name[i], CORETEMP_NAME_LENGTH, names[i],
360 			attr_no);
361 		sysfs_attr_init(&tdata->sd_attrs[i].dev_attr.attr);
362 		tdata->sd_attrs[i].dev_attr.attr.name = tdata->attr_name[i];
363 		tdata->sd_attrs[i].dev_attr.attr.mode = S_IRUGO;
364 		tdata->sd_attrs[i].dev_attr.show = rd_ptr[i];
365 		tdata->sd_attrs[i].dev_attr.store = NULL;
366 		tdata->sd_attrs[i].index = attr_no;
367 		err = device_create_file(dev, &tdata->sd_attrs[i].dev_attr);
368 		if (err)
369 			goto exit_free;
370 	}
371 	return 0;
372 
373 exit_free:
374 	while (--i >= 0)
375 		device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
376 	return err;
377 }
378 
379 static void update_ttarget(__u8 cpu_model, struct temp_data *tdata,
380 				struct device *dev)
381 {
382 	int err;
383 	u32 eax, edx;
384 
385 	/*
386 	 * Initialize ttarget value. Eventually this will be
387 	 * initialized with the value from MSR_IA32_THERM_INTERRUPT
388 	 * register. If IA32_TEMPERATURE_TARGET is supported, this
389 	 * value will be over written below.
390 	 * To Do: Patch to initialize ttarget from MSR_IA32_THERM_INTERRUPT
391 	 */
392 	tdata->ttarget = tdata->tjmax - 20000;
393 
394 	/*
395 	 * Read the still undocumented IA32_TEMPERATURE_TARGET. It exists
396 	 * on older CPUs but not in this register,
397 	 * Atoms don't have it either.
398 	 */
399 	if (cpu_model > 0xe && cpu_model != 0x1c) {
400 		err = rdmsr_safe_on_cpu(tdata->cpu,
401 				MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
402 		if (err) {
403 			dev_warn(dev,
404 			"Unable to read IA32_TEMPERATURE_TARGET MSR\n");
405 		} else {
406 			tdata->ttarget = tdata->tjmax -
407 					((eax >> 8) & 0xff) * 1000;
408 		}
409 	}
410 }
411 
412 static int __devinit chk_ucode_version(struct platform_device *pdev)
413 {
414 	struct cpuinfo_x86 *c = &cpu_data(pdev->id);
415 	int err;
416 	u32 edx;
417 
418 	/*
419 	 * Check if we have problem with errata AE18 of Core processors:
420 	 * Readings might stop update when processor visited too deep sleep,
421 	 * fixed for stepping D0 (6EC).
422 	 */
423 	if (c->x86_model == 0xe && c->x86_mask < 0xc) {
424 		/* check for microcode update */
425 		err = smp_call_function_single(pdev->id, get_ucode_rev_on_cpu,
426 					       &edx, 1);
427 		if (err) {
428 			dev_err(&pdev->dev,
429 				"Cannot determine microcode revision of "
430 				"CPU#%u (%d)!\n", pdev->id, err);
431 			return -ENODEV;
432 		} else if (edx < 0x39) {
433 			dev_err(&pdev->dev,
434 				"Errata AE18 not fixed, update BIOS or "
435 				"microcode of the CPU!\n");
436 			return -ENODEV;
437 		}
438 	}
439 	return 0;
440 }
441 
442 static struct platform_device *coretemp_get_pdev(unsigned int cpu)
443 {
444 	u16 phys_proc_id = TO_PHYS_ID(cpu);
445 	struct pdev_entry *p;
446 
447 	mutex_lock(&pdev_list_mutex);
448 
449 	list_for_each_entry(p, &pdev_list, list)
450 		if (p->phys_proc_id == phys_proc_id) {
451 			mutex_unlock(&pdev_list_mutex);
452 			return p->pdev;
453 		}
454 
455 	mutex_unlock(&pdev_list_mutex);
456 	return NULL;
457 }
458 
459 static struct temp_data *init_temp_data(unsigned int cpu, int pkg_flag)
460 {
461 	struct temp_data *tdata;
462 
463 	tdata = kzalloc(sizeof(struct temp_data), GFP_KERNEL);
464 	if (!tdata)
465 		return NULL;
466 
467 	tdata->status_reg = pkg_flag ? MSR_IA32_PACKAGE_THERM_STATUS :
468 							MSR_IA32_THERM_STATUS;
469 	tdata->is_pkg_data = pkg_flag;
470 	tdata->cpu = cpu;
471 	tdata->cpu_core_id = TO_CORE_ID(cpu);
472 	mutex_init(&tdata->update_lock);
473 	return tdata;
474 }
475 
476 static int create_core_data(struct platform_data *pdata,
477 				struct platform_device *pdev,
478 				unsigned int cpu, int pkg_flag)
479 {
480 	struct temp_data *tdata;
481 	struct cpuinfo_x86 *c = &cpu_data(cpu);
482 	u32 eax, edx;
483 	int err, attr_no;
484 
485 	/*
486 	 * Find attr number for sysfs:
487 	 * We map the attr number to core id of the CPU
488 	 * The attr number is always core id + 2
489 	 * The Pkgtemp will always show up as temp1_*, if available
490 	 */
491 	attr_no = pkg_flag ? 1 : TO_ATTR_NO(cpu);
492 
493 	if (attr_no > MAX_CORE_DATA - 1)
494 		return -ERANGE;
495 
496 	/*
497 	 * Provide a single set of attributes for all HT siblings of a core
498 	 * to avoid duplicate sensors (the processor ID and core ID of all
499 	 * HT siblings of a core are the same).
500 	 * Skip if a HT sibling of this core is already registered.
501 	 * This is not an error.
502 	 */
503 	if (pdata->core_data[attr_no] != NULL)
504 		return 0;
505 
506 	tdata = init_temp_data(cpu, pkg_flag);
507 	if (!tdata)
508 		return -ENOMEM;
509 
510 	/* Test if we can access the status register */
511 	err = rdmsr_safe_on_cpu(cpu, tdata->status_reg, &eax, &edx);
512 	if (err)
513 		goto exit_free;
514 
515 	/* We can access status register. Get Critical Temperature */
516 	if (pkg_flag)
517 		tdata->tjmax = get_pkg_tjmax(pdev->id, &pdev->dev);
518 	else
519 		tdata->tjmax = get_tjmax(c, cpu, &pdev->dev);
520 
521 	update_ttarget(c->x86_model, tdata, &pdev->dev);
522 	pdata->core_data[attr_no] = tdata;
523 
524 	/* Create sysfs interfaces */
525 	err = create_core_attrs(tdata, &pdev->dev, attr_no);
526 	if (err)
527 		goto exit_free;
528 
529 	return 0;
530 exit_free:
531 	kfree(tdata);
532 	return err;
533 }
534 
535 static void coretemp_add_core(unsigned int cpu, int pkg_flag)
536 {
537 	struct platform_data *pdata;
538 	struct platform_device *pdev = coretemp_get_pdev(cpu);
539 	int err;
540 
541 	if (!pdev)
542 		return;
543 
544 	pdata = platform_get_drvdata(pdev);
545 
546 	err = create_core_data(pdata, pdev, cpu, pkg_flag);
547 	if (err)
548 		dev_err(&pdev->dev, "Adding Core %u failed\n", cpu);
549 }
550 
551 static void coretemp_remove_core(struct platform_data *pdata,
552 				struct device *dev, int indx)
553 {
554 	int i;
555 	struct temp_data *tdata = pdata->core_data[indx];
556 
557 	/* Remove the sysfs attributes */
558 	for (i = 0; i < MAX_ATTRS; i++)
559 		device_remove_file(dev, &tdata->sd_attrs[i].dev_attr);
560 
561 	kfree(pdata->core_data[indx]);
562 	pdata->core_data[indx] = NULL;
563 }
564 
565 static int __devinit coretemp_probe(struct platform_device *pdev)
566 {
567 	struct platform_data *pdata;
568 	int err;
569 
570 	/* Check the microcode version of the CPU */
571 	err = chk_ucode_version(pdev);
572 	if (err)
573 		return err;
574 
575 	/* Initialize the per-package data structures */
576 	pdata = kzalloc(sizeof(struct platform_data), GFP_KERNEL);
577 	if (!pdata)
578 		return -ENOMEM;
579 
580 	err = create_name_attr(pdata, &pdev->dev);
581 	if (err)
582 		goto exit_free;
583 
584 	pdata->phys_proc_id = TO_PHYS_ID(pdev->id);
585 	platform_set_drvdata(pdev, pdata);
586 
587 	pdata->hwmon_dev = hwmon_device_register(&pdev->dev);
588 	if (IS_ERR(pdata->hwmon_dev)) {
589 		err = PTR_ERR(pdata->hwmon_dev);
590 		dev_err(&pdev->dev, "Class registration failed (%d)\n", err);
591 		goto exit_name;
592 	}
593 	return 0;
594 
595 exit_name:
596 	device_remove_file(&pdev->dev, &pdata->name_attr);
597 	platform_set_drvdata(pdev, NULL);
598 exit_free:
599 	kfree(pdata);
600 	return err;
601 }
602 
603 static int __devexit coretemp_remove(struct platform_device *pdev)
604 {
605 	struct platform_data *pdata = platform_get_drvdata(pdev);
606 	int i;
607 
608 	for (i = MAX_CORE_DATA - 1; i >= 0; --i)
609 		if (pdata->core_data[i])
610 			coretemp_remove_core(pdata, &pdev->dev, i);
611 
612 	device_remove_file(&pdev->dev, &pdata->name_attr);
613 	hwmon_device_unregister(pdata->hwmon_dev);
614 	platform_set_drvdata(pdev, NULL);
615 	kfree(pdata);
616 	return 0;
617 }
618 
619 static struct platform_driver coretemp_driver = {
620 	.driver = {
621 		.owner = THIS_MODULE,
622 		.name = DRVNAME,
623 	},
624 	.probe = coretemp_probe,
625 	.remove = __devexit_p(coretemp_remove),
626 };
627 
628 static int __cpuinit coretemp_device_add(unsigned int cpu)
629 {
630 	int err;
631 	struct platform_device *pdev;
632 	struct pdev_entry *pdev_entry;
633 
634 	mutex_lock(&pdev_list_mutex);
635 
636 	pdev = platform_device_alloc(DRVNAME, cpu);
637 	if (!pdev) {
638 		err = -ENOMEM;
639 		pr_err("Device allocation failed\n");
640 		goto exit;
641 	}
642 
643 	pdev_entry = kzalloc(sizeof(struct pdev_entry), GFP_KERNEL);
644 	if (!pdev_entry) {
645 		err = -ENOMEM;
646 		goto exit_device_put;
647 	}
648 
649 	err = platform_device_add(pdev);
650 	if (err) {
651 		pr_err("Device addition failed (%d)\n", err);
652 		goto exit_device_free;
653 	}
654 
655 	pdev_entry->pdev = pdev;
656 	pdev_entry->cpu = cpu;
657 	pdev_entry->phys_proc_id = TO_PHYS_ID(cpu);
658 	pdev_entry->cpu_core_id = TO_CORE_ID(cpu);
659 
660 	list_add_tail(&pdev_entry->list, &pdev_list);
661 	mutex_unlock(&pdev_list_mutex);
662 
663 	return 0;
664 
665 exit_device_free:
666 	kfree(pdev_entry);
667 exit_device_put:
668 	platform_device_put(pdev);
669 exit:
670 	mutex_unlock(&pdev_list_mutex);
671 	return err;
672 }
673 
674 static void coretemp_device_remove(unsigned int cpu)
675 {
676 	struct pdev_entry *p, *n;
677 	u16 phys_proc_id = TO_PHYS_ID(cpu);
678 
679 	mutex_lock(&pdev_list_mutex);
680 	list_for_each_entry_safe(p, n, &pdev_list, list) {
681 		if (p->phys_proc_id != phys_proc_id)
682 			continue;
683 		platform_device_unregister(p->pdev);
684 		list_del(&p->list);
685 		kfree(p);
686 	}
687 	mutex_unlock(&pdev_list_mutex);
688 }
689 
690 static bool is_any_core_online(struct platform_data *pdata)
691 {
692 	int i;
693 
694 	/* Find online cores, except pkgtemp data */
695 	for (i = MAX_CORE_DATA - 1; i >= 0; --i) {
696 		if (pdata->core_data[i] &&
697 			!pdata->core_data[i]->is_pkg_data) {
698 			return true;
699 		}
700 	}
701 	return false;
702 }
703 
704 static void __cpuinit get_core_online(unsigned int cpu)
705 {
706 	struct cpuinfo_x86 *c = &cpu_data(cpu);
707 	struct platform_device *pdev = coretemp_get_pdev(cpu);
708 	int err;
709 
710 	/*
711 	 * CPUID.06H.EAX[0] indicates whether the CPU has thermal
712 	 * sensors. We check this bit only, all the early CPUs
713 	 * without thermal sensors will be filtered out.
714 	 */
715 	if (!cpu_has(c, X86_FEATURE_DTS))
716 		return;
717 
718 	if (!pdev) {
719 		/*
720 		 * Alright, we have DTS support.
721 		 * We are bringing the _first_ core in this pkg
722 		 * online. So, initialize per-pkg data structures and
723 		 * then bring this core online.
724 		 */
725 		err = coretemp_device_add(cpu);
726 		if (err)
727 			return;
728 		/*
729 		 * Check whether pkgtemp support is available.
730 		 * If so, add interfaces for pkgtemp.
731 		 */
732 		if (cpu_has(c, X86_FEATURE_PTS))
733 			coretemp_add_core(cpu, 1);
734 	}
735 	/*
736 	 * Physical CPU device already exists.
737 	 * So, just add interfaces for this core.
738 	 */
739 	coretemp_add_core(cpu, 0);
740 }
741 
742 static void __cpuinit put_core_offline(unsigned int cpu)
743 {
744 	int i, indx;
745 	struct platform_data *pdata;
746 	struct platform_device *pdev = coretemp_get_pdev(cpu);
747 
748 	/* If the physical CPU device does not exist, just return */
749 	if (!pdev)
750 		return;
751 
752 	pdata = platform_get_drvdata(pdev);
753 
754 	indx = TO_ATTR_NO(cpu);
755 
756 	if (pdata->core_data[indx] && pdata->core_data[indx]->cpu == cpu)
757 		coretemp_remove_core(pdata, &pdev->dev, indx);
758 
759 	/*
760 	 * If a HT sibling of a core is taken offline, but another HT sibling
761 	 * of the same core is still online, register the alternate sibling.
762 	 * This ensures that exactly one set of attributes is provided as long
763 	 * as at least one HT sibling of a core is online.
764 	 */
765 	for_each_sibling(i, cpu) {
766 		if (i != cpu) {
767 			get_core_online(i);
768 			/*
769 			 * Display temperature sensor data for one HT sibling
770 			 * per core only, so abort the loop after one such
771 			 * sibling has been found.
772 			 */
773 			break;
774 		}
775 	}
776 	/*
777 	 * If all cores in this pkg are offline, remove the device.
778 	 * coretemp_device_remove calls unregister_platform_device,
779 	 * which in turn calls coretemp_remove. This removes the
780 	 * pkgtemp entry and does other clean ups.
781 	 */
782 	if (!is_any_core_online(pdata))
783 		coretemp_device_remove(cpu);
784 }
785 
786 static int __cpuinit coretemp_cpu_callback(struct notifier_block *nfb,
787 				 unsigned long action, void *hcpu)
788 {
789 	unsigned int cpu = (unsigned long) hcpu;
790 
791 	switch (action) {
792 	case CPU_ONLINE:
793 	case CPU_DOWN_FAILED:
794 		get_core_online(cpu);
795 		break;
796 	case CPU_DOWN_PREPARE:
797 		put_core_offline(cpu);
798 		break;
799 	}
800 	return NOTIFY_OK;
801 }
802 
803 static struct notifier_block coretemp_cpu_notifier __refdata = {
804 	.notifier_call = coretemp_cpu_callback,
805 };
806 
807 static int __init coretemp_init(void)
808 {
809 	int i, err = -ENODEV;
810 
811 	/* quick check if we run Intel */
812 	if (cpu_data(0).x86_vendor != X86_VENDOR_INTEL)
813 		goto exit;
814 
815 	err = platform_driver_register(&coretemp_driver);
816 	if (err)
817 		goto exit;
818 
819 	for_each_online_cpu(i)
820 		get_core_online(i);
821 
822 #ifndef CONFIG_HOTPLUG_CPU
823 	if (list_empty(&pdev_list)) {
824 		err = -ENODEV;
825 		goto exit_driver_unreg;
826 	}
827 #endif
828 
829 	register_hotcpu_notifier(&coretemp_cpu_notifier);
830 	return 0;
831 
832 #ifndef CONFIG_HOTPLUG_CPU
833 exit_driver_unreg:
834 	platform_driver_unregister(&coretemp_driver);
835 #endif
836 exit:
837 	return err;
838 }
839 
840 static void __exit coretemp_exit(void)
841 {
842 	struct pdev_entry *p, *n;
843 
844 	unregister_hotcpu_notifier(&coretemp_cpu_notifier);
845 	mutex_lock(&pdev_list_mutex);
846 	list_for_each_entry_safe(p, n, &pdev_list, list) {
847 		platform_device_unregister(p->pdev);
848 		list_del(&p->list);
849 		kfree(p);
850 	}
851 	mutex_unlock(&pdev_list_mutex);
852 	platform_driver_unregister(&coretemp_driver);
853 }
854 
855 MODULE_AUTHOR("Rudolf Marek <r.marek@assembler.cz>");
856 MODULE_DESCRIPTION("Intel Core temperature monitor");
857 MODULE_LICENSE("GPL");
858 
859 module_init(coretemp_init)
860 module_exit(coretemp_exit)
861