1/*
2 * x86_pkg_temp_thermal driver
3 * Copyright (c) 2013, Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc.
16 *
17 */
18#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
19
20#include <linux/module.h>
21#include <linux/init.h>
22#include <linux/err.h>
23#include <linux/param.h>
24#include <linux/device.h>
25#include <linux/platform_device.h>
26#include <linux/cpu.h>
27#include <linux/smp.h>
28#include <linux/slab.h>
29#include <linux/pm.h>
30#include <linux/thermal.h>
31#include <linux/debugfs.h>
32#include <asm/cpu_device_id.h>
33#include <asm/mce.h>
34
35/*
36* Rate control delay: Idea is to introduce denounce effect
37* This should be long enough to avoid reduce events, when
38* threshold is set to a temperature, which is constantly
39* violated, but at the short enough to take any action.
40* The action can be remove threshold or change it to next
41* interesting setting. Based on experiments, in around
42* every 5 seconds under load will give us a significant
43* temperature change.
44*/
45#define PKG_TEMP_THERMAL_NOTIFY_DELAY	5000
46static int notify_delay_ms = PKG_TEMP_THERMAL_NOTIFY_DELAY;
47module_param(notify_delay_ms, int, 0644);
48MODULE_PARM_DESC(notify_delay_ms,
49	"User space notification delay in milli seconds.");
50
51/* Number of trip points in thermal zone. Currently it can't
52* be more than 2. MSR can allow setting and getting notifications
53* for only 2 thresholds. This define enforces this, if there
54* is some wrong values returned by cpuid for number of thresholds.
55*/
56#define MAX_NUMBER_OF_TRIPS	2
57/* Limit number of package temp zones */
58#define MAX_PKG_TEMP_ZONE_IDS	256
59
60struct phy_dev_entry {
61	struct list_head list;
62	u16 phys_proc_id;
63	u16 first_cpu;
64	u32 tj_max;
65	int ref_cnt;
66	u32 start_pkg_therm_low;
67	u32 start_pkg_therm_high;
68	struct thermal_zone_device *tzone;
69};
70
71static const struct thermal_zone_params pkg_temp_tz_params = {
72	.no_hwmon	= true,
73};
74
75/* List maintaining number of package instances */
76static LIST_HEAD(phy_dev_list);
77static DEFINE_MUTEX(phy_dev_list_mutex);
78
79/* Interrupt to work function schedule queue */
80static DEFINE_PER_CPU(struct delayed_work, pkg_temp_thermal_threshold_work);
81
82/* To track if the work is already scheduled on a package */
83static u8 *pkg_work_scheduled;
84
85/* Spin lock to prevent races with pkg_work_scheduled */
86static spinlock_t pkg_work_lock;
87static u16 max_phy_id;
88
89/* Debug counters to show using debugfs */
90static struct dentry *debugfs;
91static unsigned int pkg_interrupt_cnt;
92static unsigned int pkg_work_cnt;
93
94static int pkg_temp_debugfs_init(void)
95{
96	struct dentry *d;
97
98	debugfs = debugfs_create_dir("pkg_temp_thermal", NULL);
99	if (!debugfs)
100		return -ENOENT;
101
102	d = debugfs_create_u32("pkg_thres_interrupt", S_IRUGO, debugfs,
103				(u32 *)&pkg_interrupt_cnt);
104	if (!d)
105		goto err_out;
106
107	d = debugfs_create_u32("pkg_thres_work", S_IRUGO, debugfs,
108				(u32 *)&pkg_work_cnt);
109	if (!d)
110		goto err_out;
111
112	return 0;
113
114err_out:
115	debugfs_remove_recursive(debugfs);
116	return -ENOENT;
117}
118
119static struct phy_dev_entry
120			*pkg_temp_thermal_get_phy_entry(unsigned int cpu)
121{
122	u16 phys_proc_id = topology_physical_package_id(cpu);
123	struct phy_dev_entry *phy_ptr;
124
125	mutex_lock(&phy_dev_list_mutex);
126
127	list_for_each_entry(phy_ptr, &phy_dev_list, list)
128		if (phy_ptr->phys_proc_id == phys_proc_id) {
129			mutex_unlock(&phy_dev_list_mutex);
130			return phy_ptr;
131		}
132
133	mutex_unlock(&phy_dev_list_mutex);
134
135	return NULL;
136}
137
138/*
139* tj-max is is interesting because threshold is set relative to this
140* temperature.
141*/
142static int get_tj_max(int cpu, u32 *tj_max)
143{
144	u32 eax, edx;
145	u32 val;
146	int err;
147
148	err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
149	if (err)
150		goto err_ret;
151	else {
152		val = (eax >> 16) & 0xff;
153		if (val)
154			*tj_max = val * 1000;
155		else {
156			err = -EINVAL;
157			goto err_ret;
158		}
159	}
160
161	return 0;
162err_ret:
163	*tj_max = 0;
164	return err;
165}
166
167static int sys_get_curr_temp(struct thermal_zone_device *tzd, unsigned long *temp)
168{
169	u32 eax, edx;
170	struct phy_dev_entry *phy_dev_entry;
171
172	phy_dev_entry = tzd->devdata;
173	rdmsr_on_cpu(phy_dev_entry->first_cpu, MSR_IA32_PACKAGE_THERM_STATUS,
174			&eax, &edx);
175	if (eax & 0x80000000) {
176		*temp = phy_dev_entry->tj_max -
177				((eax >> 16) & 0x7f) * 1000;
178		pr_debug("sys_get_curr_temp %ld\n", *temp);
179		return 0;
180	}
181
182	return -EINVAL;
183}
184
185static int sys_get_trip_temp(struct thermal_zone_device *tzd,
186		int trip, unsigned long *temp)
187{
188	u32 eax, edx;
189	struct phy_dev_entry *phy_dev_entry;
190	u32 mask, shift;
191	unsigned long thres_reg_value;
192	int ret;
193
194	if (trip >= MAX_NUMBER_OF_TRIPS)
195		return -EINVAL;
196
197	phy_dev_entry = tzd->devdata;
198
199	if (trip) {
200		mask = THERM_MASK_THRESHOLD1;
201		shift = THERM_SHIFT_THRESHOLD1;
202	} else {
203		mask = THERM_MASK_THRESHOLD0;
204		shift = THERM_SHIFT_THRESHOLD0;
205	}
206
207	ret = rdmsr_on_cpu(phy_dev_entry->first_cpu,
208				MSR_IA32_PACKAGE_THERM_INTERRUPT, &eax, &edx);
209	if (ret < 0)
210		return -EINVAL;
211
212	thres_reg_value = (eax & mask) >> shift;
213	if (thres_reg_value)
214		*temp = phy_dev_entry->tj_max - thres_reg_value * 1000;
215	else
216		*temp = 0;
217	pr_debug("sys_get_trip_temp %ld\n", *temp);
218
219	return 0;
220}
221
222static int sys_set_trip_temp(struct thermal_zone_device *tzd, int trip,
223							unsigned long temp)
224{
225	u32 l, h;
226	struct phy_dev_entry *phy_dev_entry;
227	u32 mask, shift, intr;
228	int ret;
229
230	phy_dev_entry = tzd->devdata;
231
232	if (trip >= MAX_NUMBER_OF_TRIPS || temp >= phy_dev_entry->tj_max)
233		return -EINVAL;
234
235	ret = rdmsr_on_cpu(phy_dev_entry->first_cpu,
236					MSR_IA32_PACKAGE_THERM_INTERRUPT,
237					&l, &h);
238	if (ret < 0)
239		return -EINVAL;
240
241	if (trip) {
242		mask = THERM_MASK_THRESHOLD1;
243		shift = THERM_SHIFT_THRESHOLD1;
244		intr = THERM_INT_THRESHOLD1_ENABLE;
245	} else {
246		mask = THERM_MASK_THRESHOLD0;
247		shift = THERM_SHIFT_THRESHOLD0;
248		intr = THERM_INT_THRESHOLD0_ENABLE;
249	}
250	l &= ~mask;
251	/*
252	* When users space sets a trip temperature == 0, which is indication
253	* that, it is no longer interested in receiving notifications.
254	*/
255	if (!temp)
256		l &= ~intr;
257	else {
258		l |= (phy_dev_entry->tj_max - temp)/1000 << shift;
259		l |= intr;
260	}
261
262	return wrmsr_on_cpu(phy_dev_entry->first_cpu,
263					MSR_IA32_PACKAGE_THERM_INTERRUPT,
264					l, h);
265}
266
267static int sys_get_trip_type(struct thermal_zone_device *thermal,
268		int trip, enum thermal_trip_type *type)
269{
270
271	*type = THERMAL_TRIP_PASSIVE;
272
273	return 0;
274}
275
276/* Thermal zone callback registry */
277static struct thermal_zone_device_ops tzone_ops = {
278	.get_temp = sys_get_curr_temp,
279	.get_trip_temp = sys_get_trip_temp,
280	.get_trip_type = sys_get_trip_type,
281	.set_trip_temp = sys_set_trip_temp,
282};
283
284static bool pkg_temp_thermal_platform_thermal_rate_control(void)
285{
286	return true;
287}
288
289/* Enable threshold interrupt on local package/cpu */
290static inline void enable_pkg_thres_interrupt(void)
291{
292	u32 l, h;
293	u8 thres_0, thres_1;
294
295	rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
296	/* only enable/disable if it had valid threshold value */
297	thres_0 = (l & THERM_MASK_THRESHOLD0) >> THERM_SHIFT_THRESHOLD0;
298	thres_1 = (l & THERM_MASK_THRESHOLD1) >> THERM_SHIFT_THRESHOLD1;
299	if (thres_0)
300		l |= THERM_INT_THRESHOLD0_ENABLE;
301	if (thres_1)
302		l |= THERM_INT_THRESHOLD1_ENABLE;
303	wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
304}
305
306/* Disable threshold interrupt on local package/cpu */
307static inline void disable_pkg_thres_interrupt(void)
308{
309	u32 l, h;
310	rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
311	wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
312			l & (~THERM_INT_THRESHOLD0_ENABLE) &
313				(~THERM_INT_THRESHOLD1_ENABLE), h);
314}
315
316static void pkg_temp_thermal_threshold_work_fn(struct work_struct *work)
317{
318	__u64 msr_val;
319	int cpu = smp_processor_id();
320	int phy_id = topology_physical_package_id(cpu);
321	struct phy_dev_entry *phdev = pkg_temp_thermal_get_phy_entry(cpu);
322	bool notify = false;
323	unsigned long flags;
324
325	if (!phdev)
326		return;
327
328	spin_lock_irqsave(&pkg_work_lock, flags);
329	++pkg_work_cnt;
330	if (unlikely(phy_id > max_phy_id)) {
331		spin_unlock_irqrestore(&pkg_work_lock, flags);
332		return;
333	}
334	pkg_work_scheduled[phy_id] = 0;
335	spin_unlock_irqrestore(&pkg_work_lock, flags);
336
337	enable_pkg_thres_interrupt();
338	rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
339	if (msr_val & THERM_LOG_THRESHOLD0) {
340		wrmsrl(MSR_IA32_PACKAGE_THERM_STATUS,
341				msr_val & ~THERM_LOG_THRESHOLD0);
342		notify = true;
343	}
344	if (msr_val & THERM_LOG_THRESHOLD1) {
345		wrmsrl(MSR_IA32_PACKAGE_THERM_STATUS,
346				msr_val & ~THERM_LOG_THRESHOLD1);
347		notify = true;
348	}
349	if (notify) {
350		pr_debug("thermal_zone_device_update\n");
351		thermal_zone_device_update(phdev->tzone);
352	}
353}
354
355static int pkg_temp_thermal_platform_thermal_notify(__u64 msr_val)
356{
357	unsigned long flags;
358	int cpu = smp_processor_id();
359	int phy_id = topology_physical_package_id(cpu);
360
361	/*
362	* When a package is in interrupted state, all CPU's in that package
363	* are in the same interrupt state. So scheduling on any one CPU in
364	* the package is enough and simply return for others.
365	*/
366	spin_lock_irqsave(&pkg_work_lock, flags);
367	++pkg_interrupt_cnt;
368	if (unlikely(phy_id > max_phy_id) || unlikely(!pkg_work_scheduled) ||
369			pkg_work_scheduled[phy_id]) {
370		disable_pkg_thres_interrupt();
371		spin_unlock_irqrestore(&pkg_work_lock, flags);
372		return -EINVAL;
373	}
374	pkg_work_scheduled[phy_id] = 1;
375	spin_unlock_irqrestore(&pkg_work_lock, flags);
376
377	disable_pkg_thres_interrupt();
378	schedule_delayed_work_on(cpu,
379				&per_cpu(pkg_temp_thermal_threshold_work, cpu),
380				msecs_to_jiffies(notify_delay_ms));
381	return 0;
382}
383
384static int find_siblings_cpu(int cpu)
385{
386	int i;
387	int id = topology_physical_package_id(cpu);
388
389	for_each_online_cpu(i)
390		if (i != cpu && topology_physical_package_id(i) == id)
391			return i;
392
393	return 0;
394}
395
396static int pkg_temp_thermal_device_add(unsigned int cpu)
397{
398	int err;
399	u32 tj_max;
400	struct phy_dev_entry *phy_dev_entry;
401	int thres_count;
402	u32 eax, ebx, ecx, edx;
403	u8 *temp;
404	unsigned long flags;
405
406	cpuid(6, &eax, &ebx, &ecx, &edx);
407	thres_count = ebx & 0x07;
408	if (!thres_count)
409		return -ENODEV;
410
411	if (topology_physical_package_id(cpu) > MAX_PKG_TEMP_ZONE_IDS)
412		return -ENODEV;
413
414	thres_count = clamp_val(thres_count, 0, MAX_NUMBER_OF_TRIPS);
415
416	err = get_tj_max(cpu, &tj_max);
417	if (err)
418		goto err_ret;
419
420	mutex_lock(&phy_dev_list_mutex);
421
422	phy_dev_entry = kzalloc(sizeof(*phy_dev_entry), GFP_KERNEL);
423	if (!phy_dev_entry) {
424		err = -ENOMEM;
425		goto err_ret_unlock;
426	}
427
428	spin_lock_irqsave(&pkg_work_lock, flags);
429	if (topology_physical_package_id(cpu) > max_phy_id)
430		max_phy_id = topology_physical_package_id(cpu);
431	temp = krealloc(pkg_work_scheduled,
432			(max_phy_id+1) * sizeof(u8), GFP_ATOMIC);
433	if (!temp) {
434		spin_unlock_irqrestore(&pkg_work_lock, flags);
435		err = -ENOMEM;
436		goto err_ret_free;
437	}
438	pkg_work_scheduled = temp;
439	pkg_work_scheduled[topology_physical_package_id(cpu)] = 0;
440	spin_unlock_irqrestore(&pkg_work_lock, flags);
441
442	phy_dev_entry->phys_proc_id = topology_physical_package_id(cpu);
443	phy_dev_entry->first_cpu = cpu;
444	phy_dev_entry->tj_max = tj_max;
445	phy_dev_entry->ref_cnt = 1;
446	phy_dev_entry->tzone = thermal_zone_device_register("x86_pkg_temp",
447			thres_count,
448			(thres_count == MAX_NUMBER_OF_TRIPS) ?
449				0x03 : 0x01,
450			phy_dev_entry, &tzone_ops, &pkg_temp_tz_params, 0, 0);
451	if (IS_ERR(phy_dev_entry->tzone)) {
452		err = PTR_ERR(phy_dev_entry->tzone);
453		goto err_ret_free;
454	}
455	/* Store MSR value for package thermal interrupt, to restore at exit */
456	rdmsr_on_cpu(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT,
457				&phy_dev_entry->start_pkg_therm_low,
458				&phy_dev_entry->start_pkg_therm_high);
459
460	list_add_tail(&phy_dev_entry->list, &phy_dev_list);
461	pr_debug("pkg_temp_thermal_device_add :phy_id %d cpu %d\n",
462			phy_dev_entry->phys_proc_id, cpu);
463
464	mutex_unlock(&phy_dev_list_mutex);
465
466	return 0;
467
468err_ret_free:
469	kfree(phy_dev_entry);
470err_ret_unlock:
471	mutex_unlock(&phy_dev_list_mutex);
472
473err_ret:
474	return err;
475}
476
477static int pkg_temp_thermal_device_remove(unsigned int cpu)
478{
479	struct phy_dev_entry *n;
480	u16 phys_proc_id = topology_physical_package_id(cpu);
481	struct phy_dev_entry *phdev =
482			pkg_temp_thermal_get_phy_entry(cpu);
483
484	if (!phdev)
485		return -ENODEV;
486
487	mutex_lock(&phy_dev_list_mutex);
488	/* If we are loosing the first cpu for this package, we need change */
489	if (phdev->first_cpu == cpu) {
490		phdev->first_cpu = find_siblings_cpu(cpu);
491		pr_debug("thermal_device_remove: first cpu switched %d\n",
492					phdev->first_cpu);
493	}
494	/*
495	* It is possible that no siblings left as this was the last cpu
496	* going offline. We don't need to worry about this assignment
497	* as the phydev entry will be removed in this case and
498	* thermal zone is removed.
499	*/
500	--phdev->ref_cnt;
501	pr_debug("thermal_device_remove: pkg: %d cpu %d ref_cnt %d\n",
502					phys_proc_id, cpu, phdev->ref_cnt);
503	if (!phdev->ref_cnt)
504		list_for_each_entry_safe(phdev, n, &phy_dev_list, list) {
505			if (phdev->phys_proc_id == phys_proc_id) {
506				thermal_zone_device_unregister(phdev->tzone);
507				list_del(&phdev->list);
508				kfree(phdev);
509				break;
510			}
511		}
512	mutex_unlock(&phy_dev_list_mutex);
513
514	return 0;
515}
516
517static int get_core_online(unsigned int cpu)
518{
519	struct cpuinfo_x86 *c = &cpu_data(cpu);
520	struct phy_dev_entry *phdev = pkg_temp_thermal_get_phy_entry(cpu);
521
522	/* Check if there is already an instance for this package */
523	if (!phdev) {
524		if (!cpu_has(c, X86_FEATURE_DTHERM) ||
525					!cpu_has(c, X86_FEATURE_PTS))
526			return -ENODEV;
527		if (pkg_temp_thermal_device_add(cpu))
528			return -ENODEV;
529	} else {
530		mutex_lock(&phy_dev_list_mutex);
531		++phdev->ref_cnt;
532		pr_debug("get_core_online: cpu %d ref_cnt %d\n",
533						cpu, phdev->ref_cnt);
534		mutex_unlock(&phy_dev_list_mutex);
535	}
536	INIT_DELAYED_WORK(&per_cpu(pkg_temp_thermal_threshold_work, cpu),
537			pkg_temp_thermal_threshold_work_fn);
538
539	pr_debug("get_core_online: cpu %d successful\n", cpu);
540
541	return 0;
542}
543
544static void put_core_offline(unsigned int cpu)
545{
546	if (!pkg_temp_thermal_device_remove(cpu))
547		cancel_delayed_work_sync(
548			&per_cpu(pkg_temp_thermal_threshold_work, cpu));
549
550	pr_debug("put_core_offline: cpu %d\n", cpu);
551}
552
553static int pkg_temp_thermal_cpu_callback(struct notifier_block *nfb,
554				 unsigned long action, void *hcpu)
555{
556	unsigned int cpu = (unsigned long) hcpu;
557
558	switch (action) {
559	case CPU_ONLINE:
560	case CPU_DOWN_FAILED:
561		get_core_online(cpu);
562		break;
563	case CPU_DOWN_PREPARE:
564		put_core_offline(cpu);
565		break;
566	}
567	return NOTIFY_OK;
568}
569
570static struct notifier_block pkg_temp_thermal_notifier __refdata = {
571	.notifier_call = pkg_temp_thermal_cpu_callback,
572};
573
574static const struct x86_cpu_id __initconst pkg_temp_thermal_ids[] = {
575	{ X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_PTS },
576	{}
577};
578MODULE_DEVICE_TABLE(x86cpu, pkg_temp_thermal_ids);
579
580static int __init pkg_temp_thermal_init(void)
581{
582	int i;
583
584	if (!x86_match_cpu(pkg_temp_thermal_ids))
585		return -ENODEV;
586
587	spin_lock_init(&pkg_work_lock);
588	platform_thermal_package_notify =
589			pkg_temp_thermal_platform_thermal_notify;
590	platform_thermal_package_rate_control =
591			pkg_temp_thermal_platform_thermal_rate_control;
592
593	cpu_notifier_register_begin();
594	for_each_online_cpu(i)
595		if (get_core_online(i))
596			goto err_ret;
597	__register_hotcpu_notifier(&pkg_temp_thermal_notifier);
598	cpu_notifier_register_done();
599
600	pkg_temp_debugfs_init(); /* Don't care if fails */
601
602	return 0;
603
604err_ret:
605	for_each_online_cpu(i)
606		put_core_offline(i);
607	cpu_notifier_register_done();
608	kfree(pkg_work_scheduled);
609	platform_thermal_package_notify = NULL;
610	platform_thermal_package_rate_control = NULL;
611
612	return -ENODEV;
613}
614
615static void __exit pkg_temp_thermal_exit(void)
616{
617	struct phy_dev_entry *phdev, *n;
618	int i;
619
620	cpu_notifier_register_begin();
621	__unregister_hotcpu_notifier(&pkg_temp_thermal_notifier);
622	mutex_lock(&phy_dev_list_mutex);
623	list_for_each_entry_safe(phdev, n, &phy_dev_list, list) {
624		/* Retore old MSR value for package thermal interrupt */
625		wrmsr_on_cpu(phdev->first_cpu,
626			MSR_IA32_PACKAGE_THERM_INTERRUPT,
627			phdev->start_pkg_therm_low,
628			phdev->start_pkg_therm_high);
629		thermal_zone_device_unregister(phdev->tzone);
630		list_del(&phdev->list);
631		kfree(phdev);
632	}
633	mutex_unlock(&phy_dev_list_mutex);
634	platform_thermal_package_notify = NULL;
635	platform_thermal_package_rate_control = NULL;
636	for_each_online_cpu(i)
637		cancel_delayed_work_sync(
638			&per_cpu(pkg_temp_thermal_threshold_work, i));
639	cpu_notifier_register_done();
640
641	kfree(pkg_work_scheduled);
642
643	debugfs_remove_recursive(debugfs);
644}
645
646module_init(pkg_temp_thermal_init)
647module_exit(pkg_temp_thermal_exit)
648
649MODULE_DESCRIPTION("X86 PKG TEMP Thermal Driver");
650MODULE_AUTHOR("Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>");
651MODULE_LICENSE("GPL v2");
652