1 // SPDX-License-Identifier: GPL-2.0-only
3 * x86_pkg_temp_thermal driver
4 * Copyright (c) 2013, Intel Corporation.
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 #include <linux/module.h>
9 #include <linux/init.h>
10 #include <linux/err.h>
11 #include <linux/param.h>
12 #include <linux/device.h>
13 #include <linux/platform_device.h>
14 #include <linux/cpu.h>
15 #include <linux/smp.h>
16 #include <linux/slab.h>
18 #include <linux/thermal.h>
19 #include <linux/debugfs.h>
20 #include <asm/cpu_device_id.h>
24 * Rate control delay: Idea is to introduce denounce effect
25 * This should be long enough to avoid reduce events, when
26 * threshold is set to a temperature, which is constantly
27 * violated, but at the short enough to take any action.
28 * The action can be remove threshold or change it to next
29 * interesting setting. Based on experiments, in around
30 * every 5 seconds under load will give us a significant
33 #define PKG_TEMP_THERMAL_NOTIFY_DELAY 5000
34 static int notify_delay_ms = PKG_TEMP_THERMAL_NOTIFY_DELAY;
35 module_param(notify_delay_ms, int, 0644);
36 MODULE_PARM_DESC(notify_delay_ms,
37 "User space notification delay in milli seconds.");
39 /* Number of trip points in thermal zone. Currently it can't
40 * be more than 2. MSR can allow setting and getting notifications
41 * for only 2 thresholds. This define enforces this, if there
42 * is some wrong values returned by cpuid for number of thresholds.
44 #define MAX_NUMBER_OF_TRIPS 2
50 u32 msr_pkg_therm_low;
51 u32 msr_pkg_therm_high;
52 struct delayed_work work;
53 struct thermal_zone_device *tzone;
54 struct cpumask cpumask;
57 static struct thermal_zone_params pkg_temp_tz_params = {
61 /* Keep track of how many zone pointers we allocated in init() */
62 static int max_id __read_mostly;
63 /* Array of zone pointers */
64 static struct zone_device **zones;
65 /* Serializes interrupt notification, work and hotplug */
66 static DEFINE_RAW_SPINLOCK(pkg_temp_lock);
67 /* Protects zone operation in the work function against hotplug removal */
68 static DEFINE_MUTEX(thermal_zone_mutex);
70 /* The dynamically assigned cpu hotplug state for module_exit() */
71 static enum cpuhp_state pkg_thermal_hp_state __read_mostly;
73 /* Debug counters to show using debugfs */
74 static struct dentry *debugfs;
75 static unsigned int pkg_interrupt_cnt;
76 static unsigned int pkg_work_cnt;
78 static void pkg_temp_debugfs_init(void)
80 debugfs = debugfs_create_dir("pkg_temp_thermal", NULL);
82 debugfs_create_u32("pkg_thres_interrupt", S_IRUGO, debugfs,
84 debugfs_create_u32("pkg_thres_work", S_IRUGO, debugfs,
91 * - cpu hotplug: Read serialized by cpu hotplug lock
92 * Write must hold pkg_temp_lock
94 * - Other callsites: Must hold pkg_temp_lock
96 static struct zone_device *pkg_temp_thermal_get_dev(unsigned int cpu)
98 int id = topology_logical_die_id(cpu);
100 if (id >= 0 && id < max_id)
106 * tj-max is is interesting because threshold is set relative to this
109 static int get_tj_max(int cpu, u32 *tj_max)
114 err = rdmsr_safe_on_cpu(cpu, MSR_IA32_TEMPERATURE_TARGET, &eax, &edx);
118 val = (eax >> 16) & 0xff;
119 *tj_max = val * 1000;
121 return val ? 0 : -EINVAL;
124 static int sys_get_curr_temp(struct thermal_zone_device *tzd, int *temp)
126 struct zone_device *zonedev = tzd->devdata;
129 rdmsr_on_cpu(zonedev->cpu, MSR_IA32_PACKAGE_THERM_STATUS,
131 if (eax & 0x80000000) {
132 *temp = zonedev->tj_max - ((eax >> 16) & 0x7f) * 1000;
133 pr_debug("sys_get_curr_temp %d\n", *temp);
139 static int sys_get_trip_temp(struct thermal_zone_device *tzd,
142 struct zone_device *zonedev = tzd->devdata;
143 unsigned long thres_reg_value;
144 u32 mask, shift, eax, edx;
147 if (trip >= MAX_NUMBER_OF_TRIPS)
151 mask = THERM_MASK_THRESHOLD1;
152 shift = THERM_SHIFT_THRESHOLD1;
154 mask = THERM_MASK_THRESHOLD0;
155 shift = THERM_SHIFT_THRESHOLD0;
158 ret = rdmsr_on_cpu(zonedev->cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT,
163 thres_reg_value = (eax & mask) >> shift;
165 *temp = zonedev->tj_max - thres_reg_value * 1000;
167 *temp = THERMAL_TEMP_INVALID;
168 pr_debug("sys_get_trip_temp %d\n", *temp);
174 sys_set_trip_temp(struct thermal_zone_device *tzd, int trip, int temp)
176 struct zone_device *zonedev = tzd->devdata;
177 u32 l, h, mask, shift, intr;
180 if (trip >= MAX_NUMBER_OF_TRIPS || temp >= zonedev->tj_max)
183 ret = rdmsr_on_cpu(zonedev->cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT,
189 mask = THERM_MASK_THRESHOLD1;
190 shift = THERM_SHIFT_THRESHOLD1;
191 intr = THERM_INT_THRESHOLD1_ENABLE;
193 mask = THERM_MASK_THRESHOLD0;
194 shift = THERM_SHIFT_THRESHOLD0;
195 intr = THERM_INT_THRESHOLD0_ENABLE;
199 * When users space sets a trip temperature == 0, which is indication
200 * that, it is no longer interested in receiving notifications.
205 l |= (zonedev->tj_max - temp)/1000 << shift;
209 return wrmsr_on_cpu(zonedev->cpu, MSR_IA32_PACKAGE_THERM_INTERRUPT,
213 static int sys_get_trip_type(struct thermal_zone_device *thermal, int trip,
214 enum thermal_trip_type *type)
216 *type = THERMAL_TRIP_PASSIVE;
220 /* Thermal zone callback registry */
221 static struct thermal_zone_device_ops tzone_ops = {
222 .get_temp = sys_get_curr_temp,
223 .get_trip_temp = sys_get_trip_temp,
224 .get_trip_type = sys_get_trip_type,
225 .set_trip_temp = sys_set_trip_temp,
228 static bool pkg_thermal_rate_control(void)
233 /* Enable threshold interrupt on local package/cpu */
234 static inline void enable_pkg_thres_interrupt(void)
239 rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
240 /* only enable/disable if it had valid threshold value */
241 thres_0 = (l & THERM_MASK_THRESHOLD0) >> THERM_SHIFT_THRESHOLD0;
242 thres_1 = (l & THERM_MASK_THRESHOLD1) >> THERM_SHIFT_THRESHOLD1;
244 l |= THERM_INT_THRESHOLD0_ENABLE;
246 l |= THERM_INT_THRESHOLD1_ENABLE;
247 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
250 /* Disable threshold interrupt on local package/cpu */
251 static inline void disable_pkg_thres_interrupt(void)
255 rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
257 l &= ~(THERM_INT_THRESHOLD0_ENABLE | THERM_INT_THRESHOLD1_ENABLE);
258 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
261 static void pkg_temp_thermal_threshold_work_fn(struct work_struct *work)
263 struct thermal_zone_device *tzone = NULL;
264 int cpu = smp_processor_id();
265 struct zone_device *zonedev;
268 mutex_lock(&thermal_zone_mutex);
269 raw_spin_lock_irq(&pkg_temp_lock);
272 zonedev = pkg_temp_thermal_get_dev(cpu);
274 raw_spin_unlock_irq(&pkg_temp_lock);
275 mutex_unlock(&thermal_zone_mutex);
278 zonedev->work_scheduled = false;
280 rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
281 wr_val = msr_val & ~(THERM_LOG_THRESHOLD0 | THERM_LOG_THRESHOLD1);
282 if (wr_val != msr_val) {
283 wrmsrl(MSR_IA32_PACKAGE_THERM_STATUS, wr_val);
284 tzone = zonedev->tzone;
287 enable_pkg_thres_interrupt();
288 raw_spin_unlock_irq(&pkg_temp_lock);
291 * If tzone is not NULL, then thermal_zone_mutex will prevent the
292 * concurrent removal in the cpu offline callback.
295 thermal_zone_device_update(tzone, THERMAL_EVENT_UNSPECIFIED);
297 mutex_unlock(&thermal_zone_mutex);
300 static void pkg_thermal_schedule_work(int cpu, struct delayed_work *work)
302 unsigned long ms = msecs_to_jiffies(notify_delay_ms);
304 schedule_delayed_work_on(cpu, work, ms);
307 static int pkg_thermal_notify(u64 msr_val)
309 int cpu = smp_processor_id();
310 struct zone_device *zonedev;
313 raw_spin_lock_irqsave(&pkg_temp_lock, flags);
316 disable_pkg_thres_interrupt();
318 /* Work is per package, so scheduling it once is enough. */
319 zonedev = pkg_temp_thermal_get_dev(cpu);
320 if (zonedev && !zonedev->work_scheduled) {
321 zonedev->work_scheduled = true;
322 pkg_thermal_schedule_work(zonedev->cpu, &zonedev->work);
325 raw_spin_unlock_irqrestore(&pkg_temp_lock, flags);
329 static int pkg_temp_thermal_device_add(unsigned int cpu)
331 int id = topology_logical_die_id(cpu);
332 u32 tj_max, eax, ebx, ecx, edx;
333 struct zone_device *zonedev;
334 int thres_count, err;
339 cpuid(6, &eax, &ebx, &ecx, &edx);
340 thres_count = ebx & 0x07;
344 thres_count = clamp_val(thres_count, 0, MAX_NUMBER_OF_TRIPS);
346 err = get_tj_max(cpu, &tj_max);
350 zonedev = kzalloc(sizeof(*zonedev), GFP_KERNEL);
354 INIT_DELAYED_WORK(&zonedev->work, pkg_temp_thermal_threshold_work_fn);
356 zonedev->tj_max = tj_max;
357 zonedev->tzone = thermal_zone_device_register("x86_pkg_temp",
359 (thres_count == MAX_NUMBER_OF_TRIPS) ? 0x03 : 0x01,
360 zonedev, &tzone_ops, &pkg_temp_tz_params, 0, 0);
361 if (IS_ERR(zonedev->tzone)) {
362 err = PTR_ERR(zonedev->tzone);
366 err = thermal_zone_device_enable(zonedev->tzone);
368 thermal_zone_device_unregister(zonedev->tzone);
372 /* Store MSR value for package thermal interrupt, to restore at exit */
373 rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, zonedev->msr_pkg_therm_low,
374 zonedev->msr_pkg_therm_high);
376 cpumask_set_cpu(cpu, &zonedev->cpumask);
377 raw_spin_lock_irq(&pkg_temp_lock);
379 raw_spin_unlock_irq(&pkg_temp_lock);
383 static int pkg_thermal_cpu_offline(unsigned int cpu)
385 struct zone_device *zonedev = pkg_temp_thermal_get_dev(cpu);
386 bool lastcpu, was_target;
392 target = cpumask_any_but(&zonedev->cpumask, cpu);
393 cpumask_clear_cpu(cpu, &zonedev->cpumask);
394 lastcpu = target >= nr_cpu_ids;
396 * Remove the sysfs files, if this is the last cpu in the package
397 * before doing further cleanups.
400 struct thermal_zone_device *tzone = zonedev->tzone;
403 * We must protect against a work function calling
404 * thermal_zone_update, after/while unregister. We null out
405 * the pointer under the zone mutex, so the worker function
408 mutex_lock(&thermal_zone_mutex);
409 zonedev->tzone = NULL;
410 mutex_unlock(&thermal_zone_mutex);
412 thermal_zone_device_unregister(tzone);
415 /* Protect against work and interrupts */
416 raw_spin_lock_irq(&pkg_temp_lock);
419 * Check whether this cpu was the current target and store the new
420 * one. When we drop the lock, then the interrupt notify function
421 * will see the new target.
423 was_target = zonedev->cpu == cpu;
424 zonedev->cpu = target;
427 * If this is the last CPU in the package remove the package
428 * reference from the array and restore the interrupt MSR. When we
429 * drop the lock neither the interrupt notify function nor the
430 * worker will see the package anymore.
433 zones[topology_logical_die_id(cpu)] = NULL;
434 /* After this point nothing touches the MSR anymore. */
435 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
436 zonedev->msr_pkg_therm_low, zonedev->msr_pkg_therm_high);
440 * Check whether there is work scheduled and whether the work is
441 * targeted at the outgoing CPU.
443 if (zonedev->work_scheduled && was_target) {
445 * To cancel the work we need to drop the lock, otherwise
446 * we might deadlock if the work needs to be flushed.
448 raw_spin_unlock_irq(&pkg_temp_lock);
449 cancel_delayed_work_sync(&zonedev->work);
450 raw_spin_lock_irq(&pkg_temp_lock);
452 * If this is not the last cpu in the package and the work
453 * did not run after we dropped the lock above, then we
454 * need to reschedule the work, otherwise the interrupt
455 * stays disabled forever.
457 if (!lastcpu && zonedev->work_scheduled)
458 pkg_thermal_schedule_work(target, &zonedev->work);
461 raw_spin_unlock_irq(&pkg_temp_lock);
463 /* Final cleanup if this is the last cpu */
469 static int pkg_thermal_cpu_online(unsigned int cpu)
471 struct zone_device *zonedev = pkg_temp_thermal_get_dev(cpu);
472 struct cpuinfo_x86 *c = &cpu_data(cpu);
475 if (!cpu_has(c, X86_FEATURE_DTHERM) || !cpu_has(c, X86_FEATURE_PTS))
478 /* If the package exists, nothing to do */
480 cpumask_set_cpu(cpu, &zonedev->cpumask);
483 return pkg_temp_thermal_device_add(cpu);
486 static const struct x86_cpu_id __initconst pkg_temp_thermal_ids[] = {
487 X86_MATCH_VENDOR_FEATURE(INTEL, X86_FEATURE_PTS, NULL),
490 MODULE_DEVICE_TABLE(x86cpu, pkg_temp_thermal_ids);
492 static int __init pkg_temp_thermal_init(void)
496 if (!x86_match_cpu(pkg_temp_thermal_ids))
499 max_id = topology_max_packages() * topology_max_die_per_package();
500 zones = kcalloc(max_id, sizeof(struct zone_device *),
505 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "thermal/x86_pkg:online",
506 pkg_thermal_cpu_online, pkg_thermal_cpu_offline);
510 /* Store the state for module exit */
511 pkg_thermal_hp_state = ret;
513 platform_thermal_package_notify = pkg_thermal_notify;
514 platform_thermal_package_rate_control = pkg_thermal_rate_control;
516 /* Don't care if it fails */
517 pkg_temp_debugfs_init();
524 module_init(pkg_temp_thermal_init)
526 static void __exit pkg_temp_thermal_exit(void)
528 platform_thermal_package_notify = NULL;
529 platform_thermal_package_rate_control = NULL;
531 cpuhp_remove_state(pkg_thermal_hp_state);
532 debugfs_remove_recursive(debugfs);
535 module_exit(pkg_temp_thermal_exit)
537 MODULE_DESCRIPTION("X86 PKG TEMP Thermal Driver");
538 MODULE_AUTHOR("Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>");
539 MODULE_LICENSE("GPL v2");