]> asedeno.scripts.mit.edu Git - linux.git/blob - arch/x86/kernel/cpu/mcheck/therm_throt.c
PM / QoS: Remove global notifiers
[linux.git] / arch / x86 / kernel / cpu / mcheck / therm_throt.c
1 /*
2  * Thermal throttle event support code (such as syslog messaging and rate
3  * limiting) that was factored out from x86_64 (mce_intel.c) and i386 (p4.c).
4  *
5  * This allows consistent reporting of CPU thermal throttle events.
6  *
7  * Maintains a counter in /sys that keeps track of the number of thermal
8  * events, such that the user knows how bad the thermal problem might be
9  * (since the logging to syslog and mcelog is rate limited).
10  *
11  * Author: Dmitriy Zavin (dmitriyz@google.com)
12  *
13  * Credits: Adapted from Zwane Mwaikambo's original code in mce_intel.c.
14  *          Inspired by Ross Biro's and Al Borchers' counter code.
15  */
16 #include <linux/interrupt.h>
17 #include <linux/notifier.h>
18 #include <linux/jiffies.h>
19 #include <linux/kernel.h>
20 #include <linux/percpu.h>
21 #include <linux/export.h>
22 #include <linux/types.h>
23 #include <linux/init.h>
24 #include <linux/smp.h>
25 #include <linux/cpu.h>
26
27 #include <asm/processor.h>
28 #include <asm/apic.h>
29 #include <asm/mce.h>
30 #include <asm/msr.h>
31 #include <asm/trace/irq_vectors.h>
32
33 /* How long to wait between reporting thermal events */
34 #define CHECK_INTERVAL          (300 * HZ)
35
36 #define THERMAL_THROTTLING_EVENT        0
37 #define POWER_LIMIT_EVENT               1
38
39 /*
40  * Current thermal event state:
41  */
42 struct _thermal_state {
43         bool                    new_event;
44         int                     event;
45         u64                     next_check;
46         unsigned long           count;
47         unsigned long           last_count;
48 };
49
50 struct thermal_state {
51         struct _thermal_state core_throttle;
52         struct _thermal_state core_power_limit;
53         struct _thermal_state package_throttle;
54         struct _thermal_state package_power_limit;
55         struct _thermal_state core_thresh0;
56         struct _thermal_state core_thresh1;
57         struct _thermal_state pkg_thresh0;
58         struct _thermal_state pkg_thresh1;
59 };
60
61 /* Callback to handle core threshold interrupts */
62 int (*platform_thermal_notify)(__u64 msr_val);
63 EXPORT_SYMBOL(platform_thermal_notify);
64
65 /* Callback to handle core package threshold_interrupts */
66 int (*platform_thermal_package_notify)(__u64 msr_val);
67 EXPORT_SYMBOL_GPL(platform_thermal_package_notify);
68
69 /* Callback support of rate control, return true, if
70  * callback has rate control */
71 bool (*platform_thermal_package_rate_control)(void);
72 EXPORT_SYMBOL_GPL(platform_thermal_package_rate_control);
73
74
75 static DEFINE_PER_CPU(struct thermal_state, thermal_state);
76
77 static atomic_t therm_throt_en  = ATOMIC_INIT(0);
78
79 static u32 lvtthmr_init __read_mostly;
80
81 #ifdef CONFIG_SYSFS
82 #define define_therm_throt_device_one_ro(_name)                         \
83         static DEVICE_ATTR(_name, 0444,                                 \
84                            therm_throt_device_show_##_name,             \
85                                    NULL)                                \
86
87 #define define_therm_throt_device_show_func(event, name)                \
88                                                                         \
89 static ssize_t therm_throt_device_show_##event##_##name(                \
90                         struct device *dev,                             \
91                         struct device_attribute *attr,                  \
92                         char *buf)                                      \
93 {                                                                       \
94         unsigned int cpu = dev->id;                                     \
95         ssize_t ret;                                                    \
96                                                                         \
97         preempt_disable();      /* CPU hotplug */                       \
98         if (cpu_online(cpu)) {                                          \
99                 ret = sprintf(buf, "%lu\n",                             \
100                               per_cpu(thermal_state, cpu).event.name);  \
101         } else                                                          \
102                 ret = 0;                                                \
103         preempt_enable();                                               \
104                                                                         \
105         return ret;                                                     \
106 }
107
108 define_therm_throt_device_show_func(core_throttle, count);
109 define_therm_throt_device_one_ro(core_throttle_count);
110
111 define_therm_throt_device_show_func(core_power_limit, count);
112 define_therm_throt_device_one_ro(core_power_limit_count);
113
114 define_therm_throt_device_show_func(package_throttle, count);
115 define_therm_throt_device_one_ro(package_throttle_count);
116
117 define_therm_throt_device_show_func(package_power_limit, count);
118 define_therm_throt_device_one_ro(package_power_limit_count);
119
120 static struct attribute *thermal_throttle_attrs[] = {
121         &dev_attr_core_throttle_count.attr,
122         NULL
123 };
124
125 static struct attribute_group thermal_attr_group = {
126         .attrs  = thermal_throttle_attrs,
127         .name   = "thermal_throttle"
128 };
129 #endif /* CONFIG_SYSFS */
130
131 #define CORE_LEVEL      0
132 #define PACKAGE_LEVEL   1
133
134 /***
135  * therm_throt_process - Process thermal throttling event from interrupt
136  * @curr: Whether the condition is current or not (boolean), since the
137  *        thermal interrupt normally gets called both when the thermal
138  *        event begins and once the event has ended.
139  *
140  * This function is called by the thermal interrupt after the
141  * IRQ has been acknowledged.
142  *
143  * It will take care of rate limiting and printing messages to the syslog.
144  *
145  * Returns: 0 : Event should NOT be further logged, i.e. still in
146  *              "timeout" from previous log message.
147  *          1 : Event should be logged further, and a message has been
148  *              printed to the syslog.
149  */
150 static int therm_throt_process(bool new_event, int event, int level)
151 {
152         struct _thermal_state *state;
153         unsigned int this_cpu = smp_processor_id();
154         bool old_event;
155         u64 now;
156         struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
157
158         now = get_jiffies_64();
159         if (level == CORE_LEVEL) {
160                 if (event == THERMAL_THROTTLING_EVENT)
161                         state = &pstate->core_throttle;
162                 else if (event == POWER_LIMIT_EVENT)
163                         state = &pstate->core_power_limit;
164                 else
165                          return 0;
166         } else if (level == PACKAGE_LEVEL) {
167                 if (event == THERMAL_THROTTLING_EVENT)
168                         state = &pstate->package_throttle;
169                 else if (event == POWER_LIMIT_EVENT)
170                         state = &pstate->package_power_limit;
171                 else
172                         return 0;
173         } else
174                 return 0;
175
176         old_event = state->new_event;
177         state->new_event = new_event;
178
179         if (new_event)
180                 state->count++;
181
182         if (time_before64(now, state->next_check) &&
183                         state->count != state->last_count)
184                 return 0;
185
186         state->next_check = now + CHECK_INTERVAL;
187         state->last_count = state->count;
188
189         /* if we just entered the thermal event */
190         if (new_event) {
191                 if (event == THERMAL_THROTTLING_EVENT)
192                         pr_crit("CPU%d: %s temperature above threshold, cpu clock throttled (total events = %lu)\n",
193                                 this_cpu,
194                                 level == CORE_LEVEL ? "Core" : "Package",
195                                 state->count);
196                 return 1;
197         }
198         if (old_event) {
199                 if (event == THERMAL_THROTTLING_EVENT)
200                         pr_info("CPU%d: %s temperature/speed normal\n", this_cpu,
201                                 level == CORE_LEVEL ? "Core" : "Package");
202                 return 1;
203         }
204
205         return 0;
206 }
207
208 static int thresh_event_valid(int level, int event)
209 {
210         struct _thermal_state *state;
211         unsigned int this_cpu = smp_processor_id();
212         struct thermal_state *pstate = &per_cpu(thermal_state, this_cpu);
213         u64 now = get_jiffies_64();
214
215         if (level == PACKAGE_LEVEL)
216                 state = (event == 0) ? &pstate->pkg_thresh0 :
217                                                 &pstate->pkg_thresh1;
218         else
219                 state = (event == 0) ? &pstate->core_thresh0 :
220                                                 &pstate->core_thresh1;
221
222         if (time_before64(now, state->next_check))
223                 return 0;
224
225         state->next_check = now + CHECK_INTERVAL;
226
227         return 1;
228 }
229
230 static bool int_pln_enable;
231 static int __init int_pln_enable_setup(char *s)
232 {
233         int_pln_enable = true;
234
235         return 1;
236 }
237 __setup("int_pln_enable", int_pln_enable_setup);
238
239 #ifdef CONFIG_SYSFS
240 /* Add/Remove thermal_throttle interface for CPU device: */
241 static int thermal_throttle_add_dev(struct device *dev, unsigned int cpu)
242 {
243         int err;
244         struct cpuinfo_x86 *c = &cpu_data(cpu);
245
246         err = sysfs_create_group(&dev->kobj, &thermal_attr_group);
247         if (err)
248                 return err;
249
250         if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
251                 err = sysfs_add_file_to_group(&dev->kobj,
252                                               &dev_attr_core_power_limit_count.attr,
253                                               thermal_attr_group.name);
254         if (cpu_has(c, X86_FEATURE_PTS)) {
255                 err = sysfs_add_file_to_group(&dev->kobj,
256                                               &dev_attr_package_throttle_count.attr,
257                                               thermal_attr_group.name);
258                 if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
259                         err = sysfs_add_file_to_group(&dev->kobj,
260                                         &dev_attr_package_power_limit_count.attr,
261                                         thermal_attr_group.name);
262         }
263
264         return err;
265 }
266
267 static void thermal_throttle_remove_dev(struct device *dev)
268 {
269         sysfs_remove_group(&dev->kobj, &thermal_attr_group);
270 }
271
272 /* Get notified when a cpu comes on/off. Be hotplug friendly. */
273 static int thermal_throttle_online(unsigned int cpu)
274 {
275         struct device *dev = get_cpu_device(cpu);
276
277         return thermal_throttle_add_dev(dev, cpu);
278 }
279
280 static int thermal_throttle_offline(unsigned int cpu)
281 {
282         struct device *dev = get_cpu_device(cpu);
283
284         thermal_throttle_remove_dev(dev);
285         return 0;
286 }
287
288 static __init int thermal_throttle_init_device(void)
289 {
290         int ret;
291
292         if (!atomic_read(&therm_throt_en))
293                 return 0;
294
295         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/therm:online",
296                                 thermal_throttle_online,
297                                 thermal_throttle_offline);
298         return ret < 0 ? ret : 0;
299 }
300 device_initcall(thermal_throttle_init_device);
301
302 #endif /* CONFIG_SYSFS */
303
304 static void notify_package_thresholds(__u64 msr_val)
305 {
306         bool notify_thres_0 = false;
307         bool notify_thres_1 = false;
308
309         if (!platform_thermal_package_notify)
310                 return;
311
312         /* lower threshold check */
313         if (msr_val & THERM_LOG_THRESHOLD0)
314                 notify_thres_0 = true;
315         /* higher threshold check */
316         if (msr_val & THERM_LOG_THRESHOLD1)
317                 notify_thres_1 = true;
318
319         if (!notify_thres_0 && !notify_thres_1)
320                 return;
321
322         if (platform_thermal_package_rate_control &&
323                 platform_thermal_package_rate_control()) {
324                 /* Rate control is implemented in callback */
325                 platform_thermal_package_notify(msr_val);
326                 return;
327         }
328
329         /* lower threshold reached */
330         if (notify_thres_0 && thresh_event_valid(PACKAGE_LEVEL, 0))
331                 platform_thermal_package_notify(msr_val);
332         /* higher threshold reached */
333         if (notify_thres_1 && thresh_event_valid(PACKAGE_LEVEL, 1))
334                 platform_thermal_package_notify(msr_val);
335 }
336
337 static void notify_thresholds(__u64 msr_val)
338 {
339         /* check whether the interrupt handler is defined;
340          * otherwise simply return
341          */
342         if (!platform_thermal_notify)
343                 return;
344
345         /* lower threshold reached */
346         if ((msr_val & THERM_LOG_THRESHOLD0) &&
347                         thresh_event_valid(CORE_LEVEL, 0))
348                 platform_thermal_notify(msr_val);
349         /* higher threshold reached */
350         if ((msr_val & THERM_LOG_THRESHOLD1) &&
351                         thresh_event_valid(CORE_LEVEL, 1))
352                 platform_thermal_notify(msr_val);
353 }
354
355 /* Thermal transition interrupt handler */
356 static void intel_thermal_interrupt(void)
357 {
358         __u64 msr_val;
359
360         if (static_cpu_has(X86_FEATURE_HWP))
361                 wrmsrl_safe(MSR_HWP_STATUS, 0);
362
363         rdmsrl(MSR_IA32_THERM_STATUS, msr_val);
364
365         /* Check for violation of core thermal thresholds*/
366         notify_thresholds(msr_val);
367
368         if (therm_throt_process(msr_val & THERM_STATUS_PROCHOT,
369                                 THERMAL_THROTTLING_EVENT,
370                                 CORE_LEVEL) != 0)
371                 mce_log_therm_throt_event(msr_val);
372
373         if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
374                 therm_throt_process(msr_val & THERM_STATUS_POWER_LIMIT,
375                                         POWER_LIMIT_EVENT,
376                                         CORE_LEVEL);
377
378         if (this_cpu_has(X86_FEATURE_PTS)) {
379                 rdmsrl(MSR_IA32_PACKAGE_THERM_STATUS, msr_val);
380                 /* check violations of package thermal thresholds */
381                 notify_package_thresholds(msr_val);
382                 therm_throt_process(msr_val & PACKAGE_THERM_STATUS_PROCHOT,
383                                         THERMAL_THROTTLING_EVENT,
384                                         PACKAGE_LEVEL);
385                 if (this_cpu_has(X86_FEATURE_PLN) && int_pln_enable)
386                         therm_throt_process(msr_val &
387                                         PACKAGE_THERM_STATUS_POWER_LIMIT,
388                                         POWER_LIMIT_EVENT,
389                                         PACKAGE_LEVEL);
390         }
391 }
392
393 static void unexpected_thermal_interrupt(void)
394 {
395         pr_err("CPU%d: Unexpected LVT thermal interrupt!\n",
396                 smp_processor_id());
397 }
398
399 static void (*smp_thermal_vector)(void) = unexpected_thermal_interrupt;
400
401 static inline void __smp_thermal_interrupt(void)
402 {
403         inc_irq_stat(irq_thermal_count);
404         smp_thermal_vector();
405 }
406
407 asmlinkage __visible void smp_thermal_interrupt(struct pt_regs *regs)
408 {
409         entering_irq();
410         __smp_thermal_interrupt();
411         exiting_ack_irq();
412 }
413
414 asmlinkage __visible void smp_trace_thermal_interrupt(struct pt_regs *regs)
415 {
416         entering_irq();
417         trace_thermal_apic_entry(THERMAL_APIC_VECTOR);
418         __smp_thermal_interrupt();
419         trace_thermal_apic_exit(THERMAL_APIC_VECTOR);
420         exiting_ack_irq();
421 }
422
423 /* Thermal monitoring depends on APIC, ACPI and clock modulation */
424 static int intel_thermal_supported(struct cpuinfo_x86 *c)
425 {
426         if (!boot_cpu_has(X86_FEATURE_APIC))
427                 return 0;
428         if (!cpu_has(c, X86_FEATURE_ACPI) || !cpu_has(c, X86_FEATURE_ACC))
429                 return 0;
430         return 1;
431 }
432
433 void __init mcheck_intel_therm_init(void)
434 {
435         /*
436          * This function is only called on boot CPU. Save the init thermal
437          * LVT value on BSP and use that value to restore APs' thermal LVT
438          * entry BIOS programmed later
439          */
440         if (intel_thermal_supported(&boot_cpu_data))
441                 lvtthmr_init = apic_read(APIC_LVTTHMR);
442 }
443
444 void intel_init_thermal(struct cpuinfo_x86 *c)
445 {
446         unsigned int cpu = smp_processor_id();
447         int tm2 = 0;
448         u32 l, h;
449
450         if (!intel_thermal_supported(c))
451                 return;
452
453         /*
454          * First check if its enabled already, in which case there might
455          * be some SMM goo which handles it, so we can't even put a handler
456          * since it might be delivered via SMI already:
457          */
458         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
459
460         h = lvtthmr_init;
461         /*
462          * The initial value of thermal LVT entries on all APs always reads
463          * 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI
464          * sequence to them and LVT registers are reset to 0s except for
465          * the mask bits which are set to 1s when APs receive INIT IPI.
466          * If BIOS takes over the thermal interrupt and sets its interrupt
467          * delivery mode to SMI (not fixed), it restores the value that the
468          * BIOS has programmed on AP based on BSP's info we saved since BIOS
469          * is always setting the same value for all threads/cores.
470          */
471         if ((h & APIC_DM_FIXED_MASK) != APIC_DM_FIXED)
472                 apic_write(APIC_LVTTHMR, lvtthmr_init);
473
474
475         if ((l & MSR_IA32_MISC_ENABLE_TM1) && (h & APIC_DM_SMI)) {
476                 if (system_state == SYSTEM_BOOTING)
477                         pr_debug("CPU%d: Thermal monitoring handled by SMI\n", cpu);
478                 return;
479         }
480
481         /* early Pentium M models use different method for enabling TM2 */
482         if (cpu_has(c, X86_FEATURE_TM2)) {
483                 if (c->x86 == 6 && (c->x86_model == 9 || c->x86_model == 13)) {
484                         rdmsr(MSR_THERM2_CTL, l, h);
485                         if (l & MSR_THERM2_CTL_TM_SELECT)
486                                 tm2 = 1;
487                 } else if (l & MSR_IA32_MISC_ENABLE_TM2)
488                         tm2 = 1;
489         }
490
491         /* We'll mask the thermal vector in the lapic till we're ready: */
492         h = THERMAL_APIC_VECTOR | APIC_DM_FIXED | APIC_LVT_MASKED;
493         apic_write(APIC_LVTTHMR, h);
494
495         rdmsr(MSR_IA32_THERM_INTERRUPT, l, h);
496         if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
497                 wrmsr(MSR_IA32_THERM_INTERRUPT,
498                         (l | (THERM_INT_LOW_ENABLE
499                         | THERM_INT_HIGH_ENABLE)) & ~THERM_INT_PLN_ENABLE, h);
500         else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
501                 wrmsr(MSR_IA32_THERM_INTERRUPT,
502                         l | (THERM_INT_LOW_ENABLE
503                         | THERM_INT_HIGH_ENABLE | THERM_INT_PLN_ENABLE), h);
504         else
505                 wrmsr(MSR_IA32_THERM_INTERRUPT,
506                       l | (THERM_INT_LOW_ENABLE | THERM_INT_HIGH_ENABLE), h);
507
508         if (cpu_has(c, X86_FEATURE_PTS)) {
509                 rdmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT, l, h);
510                 if (cpu_has(c, X86_FEATURE_PLN) && !int_pln_enable)
511                         wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
512                                 (l | (PACKAGE_THERM_INT_LOW_ENABLE
513                                 | PACKAGE_THERM_INT_HIGH_ENABLE))
514                                 & ~PACKAGE_THERM_INT_PLN_ENABLE, h);
515                 else if (cpu_has(c, X86_FEATURE_PLN) && int_pln_enable)
516                         wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
517                                 l | (PACKAGE_THERM_INT_LOW_ENABLE
518                                 | PACKAGE_THERM_INT_HIGH_ENABLE
519                                 | PACKAGE_THERM_INT_PLN_ENABLE), h);
520                 else
521                         wrmsr(MSR_IA32_PACKAGE_THERM_INTERRUPT,
522                               l | (PACKAGE_THERM_INT_LOW_ENABLE
523                                 | PACKAGE_THERM_INT_HIGH_ENABLE), h);
524         }
525
526         smp_thermal_vector = intel_thermal_interrupt;
527
528         rdmsr(MSR_IA32_MISC_ENABLE, l, h);
529         wrmsr(MSR_IA32_MISC_ENABLE, l | MSR_IA32_MISC_ENABLE_TM1, h);
530
531         /* Unmask the thermal vector: */
532         l = apic_read(APIC_LVTTHMR);
533         apic_write(APIC_LVTTHMR, l & ~APIC_LVT_MASKED);
534
535         pr_info_once("CPU0: Thermal monitoring enabled (%s)\n",
536                       tm2 ? "TM2" : "TM1");
537
538         /* enable thermal throttle processing */
539         atomic_set(&therm_throt_en, 1);
540 }