Merge branches 'acpi_pad', 'acpica', 'apei-bugzilla-43282', 'battery', 'cpuidle-coupl...
[platform/adaptation/renesas_rcar/renesas_kernel.git] / drivers / idle / intel_idle.c
1 /*
2  * intel_idle.c - native hardware idle loop for modern Intel processors
3  *
4  * Copyright (c) 2010, Intel Corporation.
5  * Len Brown <len.brown@intel.com>
6  *
7  * This program is free software; you can redistribute it and/or modify it
8  * under the terms and conditions of the GNU General Public License,
9  * version 2, as published by the Free Software Foundation.
10  *
11  * This program is distributed in the hope it will be useful, but WITHOUT
12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
14  * more details.
15  *
16  * You should have received a copy of the GNU General Public License along with
17  * this program; if not, write to the Free Software Foundation, Inc.,
18  * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19  */
20
21 /*
22  * intel_idle is a cpuidle driver that loads on specific Intel processors
23  * in lieu of the legacy ACPI processor_idle driver.  The intent is to
24  * make Linux more efficient on these processors, as intel_idle knows
25  * more than ACPI, as well as make Linux more immune to ACPI BIOS bugs.
26  */
27
28 /*
29  * Design Assumptions
30  *
31  * All CPUs have same idle states as boot CPU
32  *
33  * Chipset BM_STS (bus master status) bit is a NOP
34  *      for preventing entry into deep C-stats
35  */
36
37 /*
38  * Known limitations
39  *
40  * The driver currently initializes for_each_online_cpu() upon modprobe.
41  * It it unaware of subsequent processors hot-added to the system.
42  * This means that if you boot with maxcpus=n and later online
43  * processors above n, those processors will use C1 only.
44  *
45  * ACPI has a .suspend hack to turn off deep c-statees during suspend
46  * to avoid complications with the lapic timer workaround.
47  * Have not seen issues with suspend, but may need same workaround here.
48  *
49  * There is currently no kernel-based automatic probing/loading mechanism
50  * if the driver is built as a module.
51  */
52
53 /* un-comment DEBUG to enable pr_debug() statements */
54 #define DEBUG
55
56 #include <linux/kernel.h>
57 #include <linux/cpuidle.h>
58 #include <linux/clockchips.h>
59 #include <linux/hrtimer.h>      /* ktime_get_real() */
60 #include <trace/events/power.h>
61 #include <linux/sched.h>
62 #include <linux/notifier.h>
63 #include <linux/cpu.h>
64 #include <linux/module.h>
65 #include <asm/cpu_device_id.h>
66 #include <asm/mwait.h>
67 #include <asm/msr.h>
68
69 #define INTEL_IDLE_VERSION "0.4"
70 #define PREFIX "intel_idle: "
71
72 static struct cpuidle_driver intel_idle_driver = {
73         .name = "intel_idle",
74         .owner = THIS_MODULE,
75 };
76 /* intel_idle.max_cstate=0 disables driver */
77 static int max_cstate = MWAIT_MAX_NUM_CSTATES - 1;
78
79 static unsigned int mwait_substates;
80
81 #define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF
82 /* Reliable LAPIC Timer States, bit 1 for C1 etc.  */
83 static unsigned int lapic_timer_reliable_states = (1 << 1);      /* Default to only C1 */
84
85 struct idle_cpu {
86         struct cpuidle_state *state_table;
87
88         /*
89          * Hardware C-state auto-demotion may not always be optimal.
90          * Indicate which enable bits to clear here.
91          */
92         unsigned long auto_demotion_disable_flags;
93 };
94
95 static const struct idle_cpu *icpu;
96 static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
97 static int intel_idle(struct cpuidle_device *dev,
98                         struct cpuidle_driver *drv, int index);
99
100 static struct cpuidle_state *cpuidle_state_table;
101
102 /*
103  * Set this flag for states where the HW flushes the TLB for us
104  * and so we don't need cross-calls to keep it consistent.
105  * If this flag is set, SW flushes the TLB, so even if the
106  * HW doesn't do the flushing, this flag is safe to use.
107  */
108 #define CPUIDLE_FLAG_TLB_FLUSHED        0x10000
109
110 /*
111  * States are indexed by the cstate number,
112  * which is also the index into the MWAIT hint array.
113  * Thus C0 is a dummy.
114  */
115 static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
116         { /* MWAIT C0 */ },
117         { /* MWAIT C1 */
118                 .name = "C1-NHM",
119                 .desc = "MWAIT 0x00",
120                 .flags = CPUIDLE_FLAG_TIME_VALID,
121                 .exit_latency = 3,
122                 .target_residency = 6,
123                 .enter = &intel_idle },
124         { /* MWAIT C2 */
125                 .name = "C3-NHM",
126                 .desc = "MWAIT 0x10",
127                 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
128                 .exit_latency = 20,
129                 .target_residency = 80,
130                 .enter = &intel_idle },
131         { /* MWAIT C3 */
132                 .name = "C6-NHM",
133                 .desc = "MWAIT 0x20",
134                 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
135                 .exit_latency = 200,
136                 .target_residency = 800,
137                 .enter = &intel_idle },
138 };
139
140 static struct cpuidle_state snb_cstates[MWAIT_MAX_NUM_CSTATES] = {
141         { /* MWAIT C0 */ },
142         { /* MWAIT C1 */
143                 .name = "C1-SNB",
144                 .desc = "MWAIT 0x00",
145                 .flags = CPUIDLE_FLAG_TIME_VALID,
146                 .exit_latency = 1,
147                 .target_residency = 1,
148                 .enter = &intel_idle },
149         { /* MWAIT C2 */
150                 .name = "C3-SNB",
151                 .desc = "MWAIT 0x10",
152                 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
153                 .exit_latency = 80,
154                 .target_residency = 211,
155                 .enter = &intel_idle },
156         { /* MWAIT C3 */
157                 .name = "C6-SNB",
158                 .desc = "MWAIT 0x20",
159                 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
160                 .exit_latency = 104,
161                 .target_residency = 345,
162                 .enter = &intel_idle },
163         { /* MWAIT C4 */
164                 .name = "C7-SNB",
165                 .desc = "MWAIT 0x30",
166                 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
167                 .exit_latency = 109,
168                 .target_residency = 345,
169                 .enter = &intel_idle },
170 };
171
172 static struct cpuidle_state ivb_cstates[MWAIT_MAX_NUM_CSTATES] = {
173         { /* MWAIT C0 */ },
174         { /* MWAIT C1 */
175                 .name = "C1-IVB",
176                 .desc = "MWAIT 0x00",
177                 .flags = CPUIDLE_FLAG_TIME_VALID,
178                 .exit_latency = 1,
179                 .target_residency = 1,
180                 .enter = &intel_idle },
181         { /* MWAIT C2 */
182                 .name = "C3-IVB",
183                 .desc = "MWAIT 0x10",
184                 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
185                 .exit_latency = 59,
186                 .target_residency = 156,
187                 .enter = &intel_idle },
188         { /* MWAIT C3 */
189                 .name = "C6-IVB",
190                 .desc = "MWAIT 0x20",
191                 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
192                 .exit_latency = 80,
193                 .target_residency = 300,
194                 .enter = &intel_idle },
195         { /* MWAIT C4 */
196                 .name = "C7-IVB",
197                 .desc = "MWAIT 0x30",
198                 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
199                 .exit_latency = 87,
200                 .target_residency = 300,
201                 .enter = &intel_idle },
202 };
203
204 static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
205         { /* MWAIT C0 */ },
206         { /* MWAIT C1 */
207                 .name = "C1-ATM",
208                 .desc = "MWAIT 0x00",
209                 .flags = CPUIDLE_FLAG_TIME_VALID,
210                 .exit_latency = 1,
211                 .target_residency = 4,
212                 .enter = &intel_idle },
213         { /* MWAIT C2 */
214                 .name = "C2-ATM",
215                 .desc = "MWAIT 0x10",
216                 .flags = CPUIDLE_FLAG_TIME_VALID,
217                 .exit_latency = 20,
218                 .target_residency = 80,
219                 .enter = &intel_idle },
220         { /* MWAIT C3 */ },
221         { /* MWAIT C4 */
222                 .name = "C4-ATM",
223                 .desc = "MWAIT 0x30",
224                 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
225                 .exit_latency = 100,
226                 .target_residency = 400,
227                 .enter = &intel_idle },
228         { /* MWAIT C5 */ },
229         { /* MWAIT C6 */
230                 .name = "C6-ATM",
231                 .desc = "MWAIT 0x52",
232                 .flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
233                 .exit_latency = 140,
234                 .target_residency = 560,
235                 .enter = &intel_idle },
236 };
237
238 static long get_driver_data(int cstate)
239 {
240         int driver_data;
241         switch (cstate) {
242
243         case 1: /* MWAIT C1 */
244                 driver_data = 0x00;
245                 break;
246         case 2: /* MWAIT C2 */
247                 driver_data = 0x10;
248                 break;
249         case 3: /* MWAIT C3 */
250                 driver_data = 0x20;
251                 break;
252         case 4: /* MWAIT C4 */
253                 driver_data = 0x30;
254                 break;
255         case 5: /* MWAIT C5 */
256                 driver_data = 0x40;
257                 break;
258         case 6: /* MWAIT C6 */
259                 driver_data = 0x52;
260                 break;
261         default:
262                 driver_data = 0x00;
263         }
264         return driver_data;
265 }
266
267 /**
268  * intel_idle
269  * @dev: cpuidle_device
270  * @drv: cpuidle driver
271  * @index: index of cpuidle state
272  *
273  * Must be called under local_irq_disable().
274  */
275 static int intel_idle(struct cpuidle_device *dev,
276                 struct cpuidle_driver *drv, int index)
277 {
278         unsigned long ecx = 1; /* break on interrupt flag */
279         struct cpuidle_state *state = &drv->states[index];
280         struct cpuidle_state_usage *state_usage = &dev->states_usage[index];
281         unsigned long eax = (unsigned long)cpuidle_get_statedata(state_usage);
282         unsigned int cstate;
283         ktime_t kt_before, kt_after;
284         s64 usec_delta;
285         int cpu = smp_processor_id();
286
287         cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
288
289         /*
290          * leave_mm() to avoid costly and often unnecessary wakeups
291          * for flushing the user TLB's associated with the active mm.
292          */
293         if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
294                 leave_mm(cpu);
295
296         if (!(lapic_timer_reliable_states & (1 << (cstate))))
297                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
298
299         kt_before = ktime_get_real();
300
301         stop_critical_timings();
302         if (!need_resched()) {
303
304                 __monitor((void *)&current_thread_info()->flags, 0, 0);
305                 smp_mb();
306                 if (!need_resched())
307                         __mwait(eax, ecx);
308         }
309
310         start_critical_timings();
311
312         kt_after = ktime_get_real();
313         usec_delta = ktime_to_us(ktime_sub(kt_after, kt_before));
314
315         local_irq_enable();
316
317         if (!(lapic_timer_reliable_states & (1 << (cstate))))
318                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
319
320         /* Update cpuidle counters */
321         dev->last_residency = (int)usec_delta;
322
323         return index;
324 }
325
326 static void __setup_broadcast_timer(void *arg)
327 {
328         unsigned long reason = (unsigned long)arg;
329         int cpu = smp_processor_id();
330
331         reason = reason ?
332                 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
333
334         clockevents_notify(reason, &cpu);
335 }
336
337 static int setup_broadcast_cpuhp_notify(struct notifier_block *n,
338                 unsigned long action, void *hcpu)
339 {
340         int hotcpu = (unsigned long)hcpu;
341
342         switch (action & 0xf) {
343         case CPU_ONLINE:
344                 smp_call_function_single(hotcpu, __setup_broadcast_timer,
345                         (void *)true, 1);
346                 break;
347         }
348         return NOTIFY_OK;
349 }
350
351 static struct notifier_block setup_broadcast_notifier = {
352         .notifier_call = setup_broadcast_cpuhp_notify,
353 };
354
355 static void auto_demotion_disable(void *dummy)
356 {
357         unsigned long long msr_bits;
358
359         rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
360         msr_bits &= ~(icpu->auto_demotion_disable_flags);
361         wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
362 }
363
364 static const struct idle_cpu idle_cpu_nehalem = {
365         .state_table = nehalem_cstates,
366         .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
367 };
368
369 static const struct idle_cpu idle_cpu_atom = {
370         .state_table = atom_cstates,
371 };
372
373 static const struct idle_cpu idle_cpu_lincroft = {
374         .state_table = atom_cstates,
375         .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
376 };
377
378 static const struct idle_cpu idle_cpu_snb = {
379         .state_table = snb_cstates,
380 };
381
382 static const struct idle_cpu idle_cpu_ivb = {
383         .state_table = ivb_cstates,
384 };
385
386 #define ICPU(model, cpu) \
387         { X86_VENDOR_INTEL, 6, model, X86_FEATURE_MWAIT, (unsigned long)&cpu }
388
389 static const struct x86_cpu_id intel_idle_ids[] = {
390         ICPU(0x1a, idle_cpu_nehalem),
391         ICPU(0x1e, idle_cpu_nehalem),
392         ICPU(0x1f, idle_cpu_nehalem),
393         ICPU(0x25, idle_cpu_nehalem),
394         ICPU(0x2c, idle_cpu_nehalem),
395         ICPU(0x2e, idle_cpu_nehalem),
396         ICPU(0x1c, idle_cpu_atom),
397         ICPU(0x26, idle_cpu_lincroft),
398         ICPU(0x2f, idle_cpu_nehalem),
399         ICPU(0x2a, idle_cpu_snb),
400         ICPU(0x2d, idle_cpu_snb),
401         ICPU(0x3a, idle_cpu_ivb),
402         {}
403 };
404 MODULE_DEVICE_TABLE(x86cpu, intel_idle_ids);
405
406 /*
407  * intel_idle_probe()
408  */
409 static int intel_idle_probe(void)
410 {
411         unsigned int eax, ebx, ecx;
412         const struct x86_cpu_id *id;
413
414         if (max_cstate == 0) {
415                 pr_debug(PREFIX "disabled\n");
416                 return -EPERM;
417         }
418
419         id = x86_match_cpu(intel_idle_ids);
420         if (!id) {
421                 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
422                     boot_cpu_data.x86 == 6)
423                         pr_debug(PREFIX "does not run on family %d model %d\n",
424                                 boot_cpu_data.x86, boot_cpu_data.x86_model);
425                 return -ENODEV;
426         }
427
428         if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
429                 return -ENODEV;
430
431         cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
432
433         if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
434             !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
435             !mwait_substates)
436                         return -ENODEV;
437
438         pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
439
440         icpu = (const struct idle_cpu *)id->driver_data;
441         cpuidle_state_table = icpu->state_table;
442
443         if (boot_cpu_has(X86_FEATURE_ARAT))     /* Always Reliable APIC Timer */
444                 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
445         else {
446                 on_each_cpu(__setup_broadcast_timer, (void *)true, 1);
447                 register_cpu_notifier(&setup_broadcast_notifier);
448         }
449
450         pr_debug(PREFIX "v" INTEL_IDLE_VERSION
451                 " model 0x%X\n", boot_cpu_data.x86_model);
452
453         pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
454                 lapic_timer_reliable_states);
455         return 0;
456 }
457
458 /*
459  * intel_idle_cpuidle_devices_uninit()
460  * unregister, free cpuidle_devices
461  */
462 static void intel_idle_cpuidle_devices_uninit(void)
463 {
464         int i;
465         struct cpuidle_device *dev;
466
467         for_each_online_cpu(i) {
468                 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
469                 cpuidle_unregister_device(dev);
470         }
471
472         free_percpu(intel_idle_cpuidle_devices);
473         return;
474 }
475 /*
476  * intel_idle_cpuidle_driver_init()
477  * allocate, initialize cpuidle_states
478  */
479 static int intel_idle_cpuidle_driver_init(void)
480 {
481         int cstate;
482         struct cpuidle_driver *drv = &intel_idle_driver;
483
484         drv->state_count = 1;
485
486         for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
487                 int num_substates;
488
489                 if (cstate > max_cstate) {
490                         printk(PREFIX "max_cstate %d reached\n",
491                                 max_cstate);
492                         break;
493                 }
494
495                 /* does the state exist in CPUID.MWAIT? */
496                 num_substates = (mwait_substates >> ((cstate) * 4))
497                                         & MWAIT_SUBSTATE_MASK;
498                 if (num_substates == 0)
499                         continue;
500                 /* is the state not enabled? */
501                 if (cpuidle_state_table[cstate].enter == NULL) {
502                         /* does the driver not know about the state? */
503                         if (*cpuidle_state_table[cstate].name == '\0')
504                                 pr_debug(PREFIX "unaware of model 0x%x"
505                                         " MWAIT %d please"
506                                         " contact lenb@kernel.org",
507                                 boot_cpu_data.x86_model, cstate);
508                         continue;
509                 }
510
511                 if ((cstate > 2) &&
512                         !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
513                         mark_tsc_unstable("TSC halts in idle"
514                                         " states deeper than C2");
515
516                 drv->states[drv->state_count] = /* structure copy */
517                         cpuidle_state_table[cstate];
518
519                 drv->state_count += 1;
520         }
521
522         if (icpu->auto_demotion_disable_flags)
523                 on_each_cpu(auto_demotion_disable, NULL, 1);
524
525         return 0;
526 }
527
528
529 /*
530  * intel_idle_cpu_init()
531  * allocate, initialize, register cpuidle_devices
532  * @cpu: cpu/core to initialize
533  */
534 int intel_idle_cpu_init(int cpu)
535 {
536         int cstate;
537         struct cpuidle_device *dev;
538
539         dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
540
541         dev->state_count = 1;
542
543         for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
544                 int num_substates;
545
546                 if (cstate > max_cstate) {
547                         printk(PREFIX "max_cstate %d reached\n", max_cstate);
548                         break;
549                 }
550
551                 /* does the state exist in CPUID.MWAIT? */
552                 num_substates = (mwait_substates >> ((cstate) * 4))
553                         & MWAIT_SUBSTATE_MASK;
554                 if (num_substates == 0)
555                         continue;
556                 /* is the state not enabled? */
557                 if (cpuidle_state_table[cstate].enter == NULL)
558                         continue;
559
560                 dev->states_usage[dev->state_count].driver_data =
561                         (void *)get_driver_data(cstate);
562
563                 dev->state_count += 1;
564         }
565
566         dev->cpu = cpu;
567
568         if (cpuidle_register_device(dev)) {
569                 pr_debug(PREFIX "cpuidle_register_device %d failed!\n", cpu);
570                 intel_idle_cpuidle_devices_uninit();
571                 return -EIO;
572         }
573
574         if (icpu->auto_demotion_disable_flags)
575                 smp_call_function_single(cpu, auto_demotion_disable, NULL, 1);
576
577         return 0;
578 }
579 EXPORT_SYMBOL_GPL(intel_idle_cpu_init);
580
581 static int __init intel_idle_init(void)
582 {
583         int retval, i;
584
585         /* Do not load intel_idle at all for now if idle= is passed */
586         if (boot_option_idle_override != IDLE_NO_OVERRIDE)
587                 return -ENODEV;
588
589         retval = intel_idle_probe();
590         if (retval)
591                 return retval;
592
593         intel_idle_cpuidle_driver_init();
594         retval = cpuidle_register_driver(&intel_idle_driver);
595         if (retval) {
596                 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
597                         cpuidle_get_driver()->name);
598                 return retval;
599         }
600
601         intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
602         if (intel_idle_cpuidle_devices == NULL)
603                 return -ENOMEM;
604
605         for_each_online_cpu(i) {
606                 retval = intel_idle_cpu_init(i);
607                 if (retval) {
608                         cpuidle_unregister_driver(&intel_idle_driver);
609                         return retval;
610                 }
611         }
612
613         return 0;
614 }
615
616 static void __exit intel_idle_exit(void)
617 {
618         intel_idle_cpuidle_devices_uninit();
619         cpuidle_unregister_driver(&intel_idle_driver);
620
621         if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE) {
622                 on_each_cpu(__setup_broadcast_timer, (void *)false, 1);
623                 unregister_cpu_notifier(&setup_broadcast_notifier);
624         }
625
626         return;
627 }
628
629 module_init(intel_idle_init);
630 module_exit(intel_idle_exit);
631
632 module_param(max_cstate, int, 0444);
633
634 MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
635 MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
636 MODULE_LICENSE("GPL");