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
58 struct phy_dev_entry
{
59 struct list_head list
;
64 u32 start_pkg_therm_low
;
65 u32 start_pkg_therm_high
;
66 struct thermal_zone_device
*tzone
;
69 /* List maintaining number of package instances */
70 static LIST_HEAD(phy_dev_list
);
71 static DEFINE_MUTEX(phy_dev_list_mutex
);
73 /* Interrupt to work function schedule queue */
74 static DEFINE_PER_CPU(struct delayed_work
, pkg_temp_thermal_threshold_work
);
76 /* To track if the work is already scheduled on a package */
77 static u8
*pkg_work_scheduled
;
79 /* Spin lock to prevent races with pkg_work_scheduled */
80 static spinlock_t pkg_work_lock
;
81 static u16 max_phy_id
;
83 /* Debug counters to show using debugfs */
84 static struct dentry
*debugfs
;
85 static unsigned int pkg_interrupt_cnt
;
86 static unsigned int pkg_work_cnt
;
88 static int pkg_temp_debugfs_init(void)
92 debugfs
= debugfs_create_dir("pkg_temp_thermal", NULL
);
96 d
= debugfs_create_u32("pkg_thres_interrupt", S_IRUGO
, debugfs
,
97 (u32
*)&pkg_interrupt_cnt
);
101 d
= debugfs_create_u32("pkg_thres_work", S_IRUGO
, debugfs
,
102 (u32
*)&pkg_work_cnt
);
109 debugfs_remove_recursive(debugfs
);
113 static struct phy_dev_entry
114 *pkg_temp_thermal_get_phy_entry(unsigned int cpu
)
116 u16 phys_proc_id
= topology_physical_package_id(cpu
);
117 struct phy_dev_entry
*phy_ptr
;
119 mutex_lock(&phy_dev_list_mutex
);
121 list_for_each_entry(phy_ptr
, &phy_dev_list
, list
)
122 if (phy_ptr
->phys_proc_id
== phys_proc_id
) {
123 mutex_unlock(&phy_dev_list_mutex
);
127 mutex_unlock(&phy_dev_list_mutex
);
133 * tj-max is is interesting because threshold is set relative to this
136 static int get_tj_max(int cpu
, u32
*tj_max
)
142 err
= rdmsr_safe_on_cpu(cpu
, MSR_IA32_TEMPERATURE_TARGET
, &eax
, &edx
);
146 val
= (eax
>> 16) & 0xff;
148 *tj_max
= val
* 1000;
161 static int sys_get_curr_temp(struct thermal_zone_device
*tzd
, unsigned long *temp
)
164 struct phy_dev_entry
*phy_dev_entry
;
166 phy_dev_entry
= tzd
->devdata
;
167 rdmsr_on_cpu(phy_dev_entry
->first_cpu
, MSR_IA32_PACKAGE_THERM_STATUS
,
169 if (eax
& 0x80000000) {
170 *temp
= phy_dev_entry
->tj_max
-
171 ((eax
>> 16) & 0x7f) * 1000;
172 pr_debug("sys_get_curr_temp %ld\n", *temp
);
179 static int sys_get_trip_temp(struct thermal_zone_device
*tzd
,
180 int trip
, unsigned long *temp
)
183 struct phy_dev_entry
*phy_dev_entry
;
185 unsigned long thres_reg_value
;
188 if (trip
>= MAX_NUMBER_OF_TRIPS
)
191 phy_dev_entry
= tzd
->devdata
;
194 mask
= THERM_MASK_THRESHOLD1
;
195 shift
= THERM_SHIFT_THRESHOLD1
;
197 mask
= THERM_MASK_THRESHOLD0
;
198 shift
= THERM_SHIFT_THRESHOLD0
;
201 ret
= rdmsr_on_cpu(phy_dev_entry
->first_cpu
,
202 MSR_IA32_PACKAGE_THERM_INTERRUPT
, &eax
, &edx
);
206 thres_reg_value
= (eax
& mask
) >> shift
;
208 *temp
= phy_dev_entry
->tj_max
- thres_reg_value
* 1000;
211 pr_debug("sys_get_trip_temp %ld\n", *temp
);
216 int sys_set_trip_temp(struct thermal_zone_device
*tzd
, int trip
,
220 struct phy_dev_entry
*phy_dev_entry
;
221 u32 mask
, shift
, intr
;
224 phy_dev_entry
= tzd
->devdata
;
226 if (trip
>= MAX_NUMBER_OF_TRIPS
|| temp
>= phy_dev_entry
->tj_max
)
229 ret
= rdmsr_on_cpu(phy_dev_entry
->first_cpu
,
230 MSR_IA32_PACKAGE_THERM_INTERRUPT
,
236 mask
= THERM_MASK_THRESHOLD1
;
237 shift
= THERM_SHIFT_THRESHOLD1
;
238 intr
= THERM_INT_THRESHOLD1_ENABLE
;
240 mask
= THERM_MASK_THRESHOLD0
;
241 shift
= THERM_SHIFT_THRESHOLD0
;
242 intr
= THERM_INT_THRESHOLD0_ENABLE
;
246 * When users space sets a trip temperature == 0, which is indication
247 * that, it is no longer interested in receiving notifications.
252 l
|= (phy_dev_entry
->tj_max
- temp
)/1000 << shift
;
256 return wrmsr_on_cpu(phy_dev_entry
->first_cpu
,
257 MSR_IA32_PACKAGE_THERM_INTERRUPT
,
261 static int sys_get_trip_type(struct thermal_zone_device
*thermal
,
262 int trip
, enum thermal_trip_type
*type
)
265 *type
= THERMAL_TRIP_PASSIVE
;
270 /* Thermal zone callback registry */
271 static struct thermal_zone_device_ops tzone_ops
= {
272 .get_temp
= sys_get_curr_temp
,
273 .get_trip_temp
= sys_get_trip_temp
,
274 .get_trip_type
= sys_get_trip_type
,
275 .set_trip_temp
= sys_set_trip_temp
,
278 static bool pkg_temp_thermal_platform_thermal_rate_control(void)
283 /* Enable threshold interrupt on local package/cpu */
284 static inline void enable_pkg_thres_interrupt(void)
289 rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT
, l
, h
);
290 /* only enable/disable if it had valid threshold value */
291 thres_0
= (l
& THERM_MASK_THRESHOLD0
) >> THERM_SHIFT_THRESHOLD0
;
292 thres_1
= (l
& THERM_MASK_THRESHOLD1
) >> THERM_SHIFT_THRESHOLD1
;
294 l
|= THERM_INT_THRESHOLD0_ENABLE
;
296 l
|= THERM_INT_THRESHOLD1_ENABLE
;
297 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT
, l
, h
);
300 /* Disable threshold interrupt on local package/cpu */
301 static inline void disable_pkg_thres_interrupt(void)
304 rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT
, l
, h
);
305 wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT
,
306 l
& (~THERM_INT_THRESHOLD0_ENABLE
) &
307 (~THERM_INT_THRESHOLD1_ENABLE
), h
);
310 static void pkg_temp_thermal_threshold_work_fn(struct work_struct
*work
)
313 int cpu
= smp_processor_id();
314 int phy_id
= topology_physical_package_id(cpu
);
315 struct phy_dev_entry
*phdev
= pkg_temp_thermal_get_phy_entry(cpu
);
321 spin_lock(&pkg_work_lock
);
323 if (unlikely(phy_id
> max_phy_id
)) {
324 spin_unlock(&pkg_work_lock
);
327 pkg_work_scheduled
[phy_id
] = 0;
328 spin_unlock(&pkg_work_lock
);
330 enable_pkg_thres_interrupt();
331 rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS
, msr_val
);
332 if (msr_val
& THERM_LOG_THRESHOLD0
) {
333 wrmsrl(MSR_IA32_PACKAGE_THERM_STATUS
,
334 msr_val
& ~THERM_LOG_THRESHOLD0
);
337 if (msr_val
& THERM_LOG_THRESHOLD1
) {
338 wrmsrl(MSR_IA32_PACKAGE_THERM_STATUS
,
339 msr_val
& ~THERM_LOG_THRESHOLD1
);
343 pr_debug("thermal_zone_device_update\n");
344 thermal_zone_device_update(phdev
->tzone
);
348 static int pkg_temp_thermal_platform_thermal_notify(__u64 msr_val
)
351 int cpu
= smp_processor_id();
352 int phy_id
= topology_physical_package_id(cpu
);
355 * When a package is in interrupted state, all CPU's in that package
356 * are in the same interrupt state. So scheduling on any one CPU in
357 * the package is enough and simply return for others.
359 spin_lock_irqsave(&pkg_work_lock
, flags
);
361 if (unlikely(phy_id
> max_phy_id
) || unlikely(!pkg_work_scheduled
) ||
362 pkg_work_scheduled
[phy_id
]) {
363 disable_pkg_thres_interrupt();
364 spin_unlock_irqrestore(&pkg_work_lock
, flags
);
367 pkg_work_scheduled
[phy_id
] = 1;
368 spin_unlock_irqrestore(&pkg_work_lock
, flags
);
370 disable_pkg_thres_interrupt();
371 schedule_delayed_work_on(cpu
,
372 &per_cpu(pkg_temp_thermal_threshold_work
, cpu
),
373 msecs_to_jiffies(notify_delay_ms
));
377 static int find_siblings_cpu(int cpu
)
380 int id
= topology_physical_package_id(cpu
);
382 for_each_online_cpu(i
)
383 if (i
!= cpu
&& topology_physical_package_id(i
) == id
)
389 static int pkg_temp_thermal_device_add(unsigned int cpu
)
393 struct phy_dev_entry
*phy_dev_entry
;
396 u32 eax
, ebx
, ecx
, edx
;
398 cpuid(6, &eax
, &ebx
, &ecx
, &edx
);
399 thres_count
= ebx
& 0x07;
403 thres_count
= clamp_val(thres_count
, 0, MAX_NUMBER_OF_TRIPS
);
405 err
= get_tj_max(cpu
, &tj_max
);
409 mutex_lock(&phy_dev_list_mutex
);
411 phy_dev_entry
= kzalloc(sizeof(*phy_dev_entry
), GFP_KERNEL
);
412 if (!phy_dev_entry
) {
417 spin_lock(&pkg_work_lock
);
418 if (topology_physical_package_id(cpu
) > max_phy_id
)
419 max_phy_id
= topology_physical_package_id(cpu
);
420 pkg_work_scheduled
= krealloc(pkg_work_scheduled
,
421 (max_phy_id
+1) * sizeof(u8
), GFP_ATOMIC
);
422 if (!pkg_work_scheduled
) {
423 spin_unlock(&pkg_work_lock
);
427 pkg_work_scheduled
[topology_physical_package_id(cpu
)] = 0;
428 spin_unlock(&pkg_work_lock
);
430 phy_dev_entry
->phys_proc_id
= topology_physical_package_id(cpu
);
431 phy_dev_entry
->first_cpu
= cpu
;
432 phy_dev_entry
->tj_max
= tj_max
;
433 phy_dev_entry
->ref_cnt
= 1;
434 snprintf(buffer
, sizeof(buffer
), "pkg-temp-%d\n",
435 phy_dev_entry
->phys_proc_id
);
436 phy_dev_entry
->tzone
= thermal_zone_device_register(buffer
,
438 (thres_count
== MAX_NUMBER_OF_TRIPS
) ?
440 phy_dev_entry
, &tzone_ops
, NULL
, 0, 0);
441 if (IS_ERR(phy_dev_entry
->tzone
)) {
442 err
= PTR_ERR(phy_dev_entry
->tzone
);
445 /* Store MSR value for package thermal interrupt, to restore at exit */
446 rdmsr_on_cpu(cpu
, MSR_IA32_PACKAGE_THERM_INTERRUPT
,
447 &phy_dev_entry
->start_pkg_therm_low
,
448 &phy_dev_entry
->start_pkg_therm_high
);
450 list_add_tail(&phy_dev_entry
->list
, &phy_dev_list
);
451 pr_debug("pkg_temp_thermal_device_add :phy_id %d cpu %d\n",
452 phy_dev_entry
->phys_proc_id
, cpu
);
454 mutex_unlock(&phy_dev_list_mutex
);
459 kfree(phy_dev_entry
);
461 mutex_unlock(&phy_dev_list_mutex
);
467 static int pkg_temp_thermal_device_remove(unsigned int cpu
)
469 struct phy_dev_entry
*n
;
470 u16 phys_proc_id
= topology_physical_package_id(cpu
);
471 struct phy_dev_entry
*phdev
=
472 pkg_temp_thermal_get_phy_entry(cpu
);
477 mutex_lock(&phy_dev_list_mutex
);
478 /* If we are loosing the first cpu for this package, we need change */
479 if (phdev
->first_cpu
== cpu
) {
480 phdev
->first_cpu
= find_siblings_cpu(cpu
);
481 pr_debug("thermal_device_remove: first cpu switched %d\n",
485 * It is possible that no siblings left as this was the last cpu
486 * going offline. We don't need to worry about this assignment
487 * as the phydev entry will be removed in this case and
488 * thermal zone is removed.
491 pr_debug("thermal_device_remove: pkg: %d cpu %d ref_cnt %d\n",
492 phys_proc_id
, cpu
, phdev
->ref_cnt
);
494 list_for_each_entry_safe(phdev
, n
, &phy_dev_list
, list
) {
495 if (phdev
->phys_proc_id
== phys_proc_id
) {
496 thermal_zone_device_unregister(phdev
->tzone
);
497 list_del(&phdev
->list
);
502 mutex_unlock(&phy_dev_list_mutex
);
507 static int get_core_online(unsigned int cpu
)
509 struct cpuinfo_x86
*c
= &cpu_data(cpu
);
510 struct phy_dev_entry
*phdev
= pkg_temp_thermal_get_phy_entry(cpu
);
512 /* Check if there is already an instance for this package */
514 if (!cpu_has(c
, X86_FEATURE_DTHERM
) &&
515 !cpu_has(c
, X86_FEATURE_PTS
))
517 if (pkg_temp_thermal_device_add(cpu
))
520 mutex_lock(&phy_dev_list_mutex
);
522 pr_debug("get_core_online: cpu %d ref_cnt %d\n",
523 cpu
, phdev
->ref_cnt
);
524 mutex_unlock(&phy_dev_list_mutex
);
526 INIT_DELAYED_WORK(&per_cpu(pkg_temp_thermal_threshold_work
, cpu
),
527 pkg_temp_thermal_threshold_work_fn
);
529 pr_debug("get_core_online: cpu %d successful\n", cpu
);
534 static void put_core_offline(unsigned int cpu
)
536 if (!pkg_temp_thermal_device_remove(cpu
))
537 cancel_delayed_work_sync(
538 &per_cpu(pkg_temp_thermal_threshold_work
, cpu
));
540 pr_debug("put_core_offline: cpu %d\n", cpu
);
543 static int pkg_temp_thermal_cpu_callback(struct notifier_block
*nfb
,
544 unsigned long action
, void *hcpu
)
546 unsigned int cpu
= (unsigned long) hcpu
;
550 case CPU_DOWN_FAILED
:
551 get_core_online(cpu
);
553 case CPU_DOWN_PREPARE
:
554 put_core_offline(cpu
);
560 static struct notifier_block pkg_temp_thermal_notifier __refdata
= {
561 .notifier_call
= pkg_temp_thermal_cpu_callback
,
564 static const struct x86_cpu_id __initconst pkg_temp_thermal_ids
[] = {
565 { X86_VENDOR_INTEL
, X86_FAMILY_ANY
, X86_MODEL_ANY
, X86_FEATURE_DTHERM
},
568 MODULE_DEVICE_TABLE(x86cpu
, pkg_temp_thermal_ids
);
570 static int __init
pkg_temp_thermal_init(void)
574 if (!x86_match_cpu(pkg_temp_thermal_ids
))
577 spin_lock_init(&pkg_work_lock
);
578 platform_thermal_package_notify
=
579 pkg_temp_thermal_platform_thermal_notify
;
580 platform_thermal_package_rate_control
=
581 pkg_temp_thermal_platform_thermal_rate_control
;
584 for_each_online_cpu(i
)
585 if (get_core_online(i
))
587 register_hotcpu_notifier(&pkg_temp_thermal_notifier
);
590 pkg_temp_debugfs_init(); /* Don't care if fails */
596 for_each_online_cpu(i
)
599 kfree(pkg_work_scheduled
);
600 platform_thermal_package_notify
= NULL
;
601 platform_thermal_package_rate_control
= NULL
;
606 static void __exit
pkg_temp_thermal_exit(void)
608 struct phy_dev_entry
*phdev
, *n
;
612 unregister_hotcpu_notifier(&pkg_temp_thermal_notifier
);
613 mutex_lock(&phy_dev_list_mutex
);
614 list_for_each_entry_safe(phdev
, n
, &phy_dev_list
, list
) {
615 /* Retore old MSR value for package thermal interrupt */
616 wrmsr_on_cpu(phdev
->first_cpu
,
617 MSR_IA32_PACKAGE_THERM_INTERRUPT
,
618 phdev
->start_pkg_therm_low
,
619 phdev
->start_pkg_therm_high
);
620 thermal_zone_device_unregister(phdev
->tzone
);
621 list_del(&phdev
->list
);
624 mutex_unlock(&phy_dev_list_mutex
);
625 platform_thermal_package_notify
= NULL
;
626 platform_thermal_package_rate_control
= NULL
;
627 for_each_online_cpu(i
)
628 cancel_delayed_work_sync(
629 &per_cpu(pkg_temp_thermal_threshold_work
, i
));
632 kfree(pkg_work_scheduled
);
634 debugfs_remove_recursive(debugfs
);
637 module_init(pkg_temp_thermal_init
)
638 module_exit(pkg_temp_thermal_exit
)
640 MODULE_DESCRIPTION("X86 PKG TEMP Thermal Driver");
641 MODULE_AUTHOR("Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>");
642 MODULE_LICENSE("GPL v2");