2 * x86_pkg_temp_thermal driver
3 * Copyright (c) 2013, Intel Corporation.
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.
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
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.
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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>
30 #include <linux/thermal.h>
31 #include <linux/debugfs.h>
32 #include <asm/cpu_device_id.h>
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
45 #define PKG_TEMP_THERMAL_NOTIFY_DELAY 5000
46 static int notify_delay_ms = PKG_TEMP_THERMAL_NOTIFY_DELAY;
47 module_param(notify_delay_ms, int, 0644);
48 MODULE_PARM_DESC(notify_delay_ms,
49 "User space notification delay in milli seconds.");
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.
56 #define MAX_NUMBER_OF_TRIPS 2
57 /* Limit number of package temp zones */
58 #define MAX_PKG_TEMP_ZONE_IDS 256
60 struct phy_dev_entry {
61 struct list_head list;
66 u32 start_pkg_therm_low;
67 u32 start_pkg_therm_high;
68 struct thermal_zone_device *tzone;
71 static struct thermal_zone_params pkg_temp_tz_params = {
75 /* List maintaining number of package instances */
76 static LIST_HEAD(phy_dev_list);
77 static DEFINE_MUTEX(phy_dev_list_mutex);
79 /* Interrupt to work function schedule queue */
80 static DEFINE_PER_CPU(struct delayed_work, pkg_temp_thermal_threshold_work);
82 /* To track if the work is already scheduled on a package */
83 static u8 *pkg_work_scheduled;
85 /* Spin lock to prevent races with pkg_work_scheduled */
86 static spinlock_t pkg_work_lock;
87 static u16 max_phy_id;
89 /* Debug counters to show using debugfs */
90 static struct dentry *debugfs;
91 static unsigned int pkg_interrupt_cnt;
92 static unsigned int pkg_work_cnt;
94 static int pkg_temp_debugfs_init(void)
98 debugfs = debugfs_create_dir("pkg_temp_thermal", NULL);
102 d = debugfs_create_u32("pkg_thres_interrupt", S_IRUGO, debugfs,
103 (u32 *)&pkg_interrupt_cnt);
107 d = debugfs_create_u32("pkg_thres_work", S_IRUGO, debugfs,
108 (u32 *)&pkg_work_cnt);
115 debugfs_remove_recursive(debugfs);
119 static struct phy_dev_entry
120 *pkg_temp_thermal_get_phy_entry(unsigned int cpu)
122 u16 phys_proc_id = topology_physical_package_id(cpu);
123 struct phy_dev_entry *phy_ptr;
125 mutex_lock(&phy_dev_list_mutex);
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);
133 mutex_unlock(&phy_dev_list_mutex);
139 * tj-max is is interesting because threshold is set relative to this
142 static int get_tj_max(int cpu, u32 *tj_max)
148 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
152 val = (eax >> 16) & 0xff;
154 *tj_max = val * 1000;
167 static int sys_get_curr_temp(struct thermal_zone_device *tzd, int *temp)
170 struct phy_dev_entry *phy_dev_entry;
172 phy_dev_entry = tzd->devdata;
173 rdmsr_on_cpu(phy_dev_entry->first_cpu, MSR_IA32_PACKAGE_THERM_STATUS,
175 if (eax & 0x80000000) {
176 *temp = phy_dev_entry->tj_max -
177 ((eax >> 16) & 0x7f) * 1000;
178 pr_debug("sys_get_curr_temp %d\n", *temp);
185 static int sys_get_trip_temp(struct thermal_zone_device *tzd,
189 struct phy_dev_entry *phy_dev_entry;
191 unsigned long thres_reg_value;
194 if (trip >= MAX_NUMBER_OF_TRIPS)
197 phy_dev_entry = tzd->devdata;
200 mask = THERM_MASK_THRESHOLD1;
201 shift = THERM_SHIFT_THRESHOLD1;
203 mask = THERM_MASK_THRESHOLD0;
204 shift = THERM_SHIFT_THRESHOLD0;
207 ret = rdmsr_on_cpu(phy_dev_entry->first_cpu,
208 MSR_IA32_PACKAGE_THERM_INTERRUPT, &eax, &edx);
212 thres_reg_value = (eax & mask) >> shift;
214 *temp = phy_dev_entry->tj_max - thres_reg_value * 1000;
217 pr_debug("sys_get_trip_temp %d\n", *temp);
222 static int sys_set_trip_temp(struct thermal_zone_device *tzd, int trip,
226 struct phy_dev_entry *phy_dev_entry;
227 u32 mask, shift, intr;
230 phy_dev_entry = tzd->devdata;
232 if (trip >= MAX_NUMBER_OF_TRIPS || temp >= phy_dev_entry->tj_max)
235 ret = rdmsr_on_cpu(phy_dev_entry->first_cpu,
236 MSR_IA32_PACKAGE_THERM_INTERRUPT,
242 mask = THERM_MASK_THRESHOLD1;
243 shift = THERM_SHIFT_THRESHOLD1;
244 intr = THERM_INT_THRESHOLD1_ENABLE;
246 mask = THERM_MASK_THRESHOLD0;
247 shift = THERM_SHIFT_THRESHOLD0;
248 intr = THERM_INT_THRESHOLD0_ENABLE;
252 * When users space sets a trip temperature == 0, which is indication
253 * that, it is no longer interested in receiving notifications.
258 l |= (phy_dev_entry->tj_max - temp)/1000 << shift;
262 return wrmsr_on_cpu(phy_dev_entry->first_cpu,
263 MSR_IA32_PACKAGE_THERM_INTERRUPT,
267 static int sys_get_trip_type(struct thermal_zone_device *thermal,
268 int trip, enum thermal_trip_type *type)
271 *type = THERMAL_TRIP_PASSIVE;
276 /* Thermal zone callback registry */
277 static 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,
284 static bool pkg_temp_thermal_platform_thermal_rate_control(void)
289 /* Enable threshold interrupt on local package/cpu */
290 static inline void enable_pkg_thres_interrupt(void)
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;
300 l |= THERM_INT_THRESHOLD0_ENABLE;
302 l |= THERM_INT_THRESHOLD1_ENABLE;
303 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
306 /* Disable threshold interrupt on local package/cpu */
307 static inline void disable_pkg_thres_interrupt(void)
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);
316 static void pkg_temp_thermal_threshold_work_fn(struct work_struct *work)
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);
328 spin_lock_irqsave(&pkg_work_lock, flags);
330 if (unlikely(phy_id > max_phy_id)) {
331 spin_unlock_irqrestore(&pkg_work_lock, flags);
334 pkg_work_scheduled[phy_id] = 0;
335 spin_unlock_irqrestore(&pkg_work_lock, flags);
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);
344 if (msr_val & THERM_LOG_THRESHOLD1) {
345 wrmsrl(MSR_IA32_PACKAGE_THERM_STATUS,
346 msr_val & ~THERM_LOG_THRESHOLD1);
350 pr_debug("thermal_zone_device_update\n");
351 thermal_zone_device_update(phdev->tzone,
352 THERMAL_EVENT_UNSPECIFIED);
356 static int pkg_temp_thermal_platform_thermal_notify(__u64 msr_val)
359 int cpu = smp_processor_id();
360 int phy_id = topology_physical_package_id(cpu);
363 * When a package is in interrupted state, all CPU's in that package
364 * are in the same interrupt state. So scheduling on any one CPU in
365 * the package is enough and simply return for others.
367 spin_lock_irqsave(&pkg_work_lock, flags);
369 if (unlikely(phy_id > max_phy_id) || unlikely(!pkg_work_scheduled) ||
370 pkg_work_scheduled[phy_id]) {
371 disable_pkg_thres_interrupt();
372 spin_unlock_irqrestore(&pkg_work_lock, flags);
375 pkg_work_scheduled[phy_id] = 1;
376 spin_unlock_irqrestore(&pkg_work_lock, flags);
378 disable_pkg_thres_interrupt();
379 schedule_delayed_work_on(cpu,
380 &per_cpu(pkg_temp_thermal_threshold_work, cpu),
381 msecs_to_jiffies(notify_delay_ms));
385 static int find_siblings_cpu(int cpu)
388 int id = topology_physical_package_id(cpu);
390 for_each_online_cpu(i)
391 if (i != cpu && topology_physical_package_id(i) == id)
397 static int pkg_temp_thermal_device_add(unsigned int cpu)
401 struct phy_dev_entry *phy_dev_entry;
403 u32 eax, ebx, ecx, edx;
407 cpuid(6, &eax, &ebx, &ecx, &edx);
408 thres_count = ebx & 0x07;
412 if (topology_physical_package_id(cpu) > MAX_PKG_TEMP_ZONE_IDS)
415 thres_count = clamp_val(thres_count, 0, MAX_NUMBER_OF_TRIPS);
417 err = get_tj_max(cpu, &tj_max);
421 mutex_lock(&phy_dev_list_mutex);
423 phy_dev_entry = kzalloc(sizeof(*phy_dev_entry), GFP_KERNEL);
424 if (!phy_dev_entry) {
429 spin_lock_irqsave(&pkg_work_lock, flags);
430 if (topology_physical_package_id(cpu) > max_phy_id)
431 max_phy_id = topology_physical_package_id(cpu);
432 temp = krealloc(pkg_work_scheduled,
433 (max_phy_id+1) * sizeof(u8), GFP_ATOMIC);
435 spin_unlock_irqrestore(&pkg_work_lock, flags);
439 pkg_work_scheduled = temp;
440 pkg_work_scheduled[topology_physical_package_id(cpu)] = 0;
441 spin_unlock_irqrestore(&pkg_work_lock, flags);
443 phy_dev_entry->phys_proc_id = topology_physical_package_id(cpu);
444 phy_dev_entry->first_cpu = cpu;
445 phy_dev_entry->tj_max = tj_max;
446 phy_dev_entry->ref_cnt = 1;
447 phy_dev_entry->tzone = thermal_zone_device_register("x86_pkg_temp",
449 (thres_count == MAX_NUMBER_OF_TRIPS) ?
451 phy_dev_entry, &tzone_ops, &pkg_temp_tz_params, 0, 0);
452 if (IS_ERR(phy_dev_entry->tzone)) {
453 err = PTR_ERR(phy_dev_entry->tzone);
456 /* Store MSR value for package thermal interrupt, to restore at exit */
457 rdmsr_on_cpu(cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT,
458 &phy_dev_entry->start_pkg_therm_low,
459 &phy_dev_entry->start_pkg_therm_high);
461 list_add_tail(&phy_dev_entry->list, &phy_dev_list);
462 pr_debug("pkg_temp_thermal_device_add :phy_id %d cpu %d\n",
463 phy_dev_entry->phys_proc_id, cpu);
465 mutex_unlock(&phy_dev_list_mutex);
470 kfree(phy_dev_entry);
472 mutex_unlock(&phy_dev_list_mutex);
478 static int pkg_temp_thermal_device_remove(unsigned int cpu)
480 struct phy_dev_entry *n;
481 u16 phys_proc_id = topology_physical_package_id(cpu);
482 struct phy_dev_entry *phdev =
483 pkg_temp_thermal_get_phy_entry(cpu);
488 mutex_lock(&phy_dev_list_mutex);
489 /* If we are loosing the first cpu for this package, we need change */
490 if (phdev->first_cpu == cpu) {
491 phdev->first_cpu = find_siblings_cpu(cpu);
492 pr_debug("thermal_device_remove: first cpu switched %d\n",
496 * It is possible that no siblings left as this was the last cpu
497 * going offline. We don't need to worry about this assignment
498 * as the phydev entry will be removed in this case and
499 * thermal zone is removed.
502 pr_debug("thermal_device_remove: pkg: %d cpu %d ref_cnt %d\n",
503 phys_proc_id, cpu, phdev->ref_cnt);
505 list_for_each_entry_safe(phdev, n, &phy_dev_list, list) {
506 if (phdev->phys_proc_id == phys_proc_id) {
507 thermal_zone_device_unregister(phdev->tzone);
508 list_del(&phdev->list);
513 mutex_unlock(&phy_dev_list_mutex);
518 static int get_core_online(unsigned int cpu)
520 struct cpuinfo_x86 *c = &cpu_data(cpu);
521 struct phy_dev_entry *phdev = pkg_temp_thermal_get_phy_entry(cpu);
523 /* Check if there is already an instance for this package */
525 if (!cpu_has(c, X86_FEATURE_DTHERM) ||
526 !cpu_has(c, X86_FEATURE_PTS))
528 if (pkg_temp_thermal_device_add(cpu))
531 mutex_lock(&phy_dev_list_mutex);
533 pr_debug("get_core_online: cpu %d ref_cnt %d\n",
534 cpu, phdev->ref_cnt);
535 mutex_unlock(&phy_dev_list_mutex);
537 INIT_DELAYED_WORK(&per_cpu(pkg_temp_thermal_threshold_work, cpu),
538 pkg_temp_thermal_threshold_work_fn);
540 pr_debug("get_core_online: cpu %d successful\n", cpu);
545 static void put_core_offline(unsigned int cpu)
547 if (!pkg_temp_thermal_device_remove(cpu))
548 cancel_delayed_work_sync(
549 &per_cpu(pkg_temp_thermal_threshold_work, cpu));
551 pr_debug("put_core_offline: cpu %d\n", cpu);
554 static int pkg_temp_thermal_cpu_callback(struct notifier_block *nfb,
555 unsigned long action, void *hcpu)
557 unsigned int cpu = (unsigned long) hcpu;
559 switch (action & ~CPU_TASKS_FROZEN) {
561 case CPU_DOWN_FAILED:
562 get_core_online(cpu);
564 case CPU_DOWN_PREPARE:
565 put_core_offline(cpu);
571 static struct notifier_block pkg_temp_thermal_notifier __refdata = {
572 .notifier_call = pkg_temp_thermal_cpu_callback,
575 static const struct x86_cpu_id __initconst pkg_temp_thermal_ids[] = {
576 { X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_PTS },
579 MODULE_DEVICE_TABLE(x86cpu, pkg_temp_thermal_ids);
581 static int __init pkg_temp_thermal_init(void)
585 if (!x86_match_cpu(pkg_temp_thermal_ids))
588 spin_lock_init(&pkg_work_lock);
589 platform_thermal_package_notify =
590 pkg_temp_thermal_platform_thermal_notify;
591 platform_thermal_package_rate_control =
592 pkg_temp_thermal_platform_thermal_rate_control;
594 cpu_notifier_register_begin();
595 for_each_online_cpu(i)
596 if (get_core_online(i))
598 __register_hotcpu_notifier(&pkg_temp_thermal_notifier);
599 cpu_notifier_register_done();
601 pkg_temp_debugfs_init(); /* Don't care if fails */
606 for_each_online_cpu(i)
608 cpu_notifier_register_done();
609 kfree(pkg_work_scheduled);
610 platform_thermal_package_notify = NULL;
611 platform_thermal_package_rate_control = NULL;
616 static void __exit pkg_temp_thermal_exit(void)
618 struct phy_dev_entry *phdev, *n;
621 cpu_notifier_register_begin();
622 __unregister_hotcpu_notifier(&pkg_temp_thermal_notifier);
623 mutex_lock(&phy_dev_list_mutex);
624 list_for_each_entry_safe(phdev, n, &phy_dev_list, list) {
625 /* Retore old MSR value for package thermal interrupt */
626 wrmsr_on_cpu(phdev->first_cpu,
627 MSR_IA32_PACKAGE_THERM_INTERRUPT,
628 phdev->start_pkg_therm_low,
629 phdev->start_pkg_therm_high);
630 thermal_zone_device_unregister(phdev->tzone);
631 list_del(&phdev->list);
634 mutex_unlock(&phy_dev_list_mutex);
635 platform_thermal_package_notify = NULL;
636 platform_thermal_package_rate_control = NULL;
637 for_each_online_cpu(i)
638 cancel_delayed_work_sync(
639 &per_cpu(pkg_temp_thermal_threshold_work, i));
640 cpu_notifier_register_done();
642 kfree(pkg_work_scheduled);
644 debugfs_remove_recursive(debugfs);
647 module_init(pkg_temp_thermal_init)
648 module_exit(pkg_temp_thermal_exit)
650 MODULE_DESCRIPTION("X86 PKG TEMP Thermal Driver");
651 MODULE_AUTHOR("Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>");
652 MODULE_LICENSE("GPL v2");