xref: /openbmc/linux/drivers/idle/intel_idle.c (revision 2792d42f)
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 <trace/events/power.h>
60 #include <linux/sched.h>
61 #include <linux/notifier.h>
62 #include <linux/cpu.h>
63 #include <linux/module.h>
64 #include <asm/cpu_device_id.h>
65 #include <asm/mwait.h>
66 #include <asm/msr.h>
67 
68 #define INTEL_IDLE_VERSION "0.4"
69 #define PREFIX "intel_idle: "
70 
71 static struct cpuidle_driver intel_idle_driver = {
72 	.name = "intel_idle",
73 	.owner = THIS_MODULE,
74 };
75 /* intel_idle.max_cstate=0 disables driver */
76 static int max_cstate = CPUIDLE_STATE_MAX - 1;
77 
78 static unsigned int mwait_substates;
79 
80 #define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF
81 /* Reliable LAPIC Timer States, bit 1 for C1 etc.  */
82 static unsigned int lapic_timer_reliable_states = (1 << 1);	 /* Default to only C1 */
83 
84 struct idle_cpu {
85 	struct cpuidle_state *state_table;
86 
87 	/*
88 	 * Hardware C-state auto-demotion may not always be optimal.
89 	 * Indicate which enable bits to clear here.
90 	 */
91 	unsigned long auto_demotion_disable_flags;
92 	bool disable_promotion_to_c1e;
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 static int intel_idle_cpu_init(int cpu);
100 
101 static struct cpuidle_state *cpuidle_state_table;
102 
103 /*
104  * Set this flag for states where the HW flushes the TLB for us
105  * and so we don't need cross-calls to keep it consistent.
106  * If this flag is set, SW flushes the TLB, so even if the
107  * HW doesn't do the flushing, this flag is safe to use.
108  */
109 #define CPUIDLE_FLAG_TLB_FLUSHED	0x10000
110 
111 /*
112  * MWAIT takes an 8-bit "hint" in EAX "suggesting"
113  * the C-state (top nibble) and sub-state (bottom nibble)
114  * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc.
115  *
116  * We store the hint at the top of our "flags" for each state.
117  */
118 #define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
119 #define MWAIT2flg(eax) ((eax & 0xFF) << 24)
120 
121 /*
122  * States are indexed by the cstate number,
123  * which is also the index into the MWAIT hint array.
124  * Thus C0 is a dummy.
125  */
126 static struct cpuidle_state nehalem_cstates[CPUIDLE_STATE_MAX] = {
127 	{
128 		.name = "C1-NHM",
129 		.desc = "MWAIT 0x00",
130 		.flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
131 		.exit_latency = 3,
132 		.target_residency = 6,
133 		.enter = &intel_idle },
134 	{
135 		.name = "C1E-NHM",
136 		.desc = "MWAIT 0x01",
137 		.flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
138 		.exit_latency = 10,
139 		.target_residency = 20,
140 		.enter = &intel_idle },
141 	{
142 		.name = "C3-NHM",
143 		.desc = "MWAIT 0x10",
144 		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
145 		.exit_latency = 20,
146 		.target_residency = 80,
147 		.enter = &intel_idle },
148 	{
149 		.name = "C6-NHM",
150 		.desc = "MWAIT 0x20",
151 		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
152 		.exit_latency = 200,
153 		.target_residency = 800,
154 		.enter = &intel_idle },
155 	{
156 		.enter = NULL }
157 };
158 
159 static struct cpuidle_state snb_cstates[CPUIDLE_STATE_MAX] = {
160 	{
161 		.name = "C1-SNB",
162 		.desc = "MWAIT 0x00",
163 		.flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
164 		.exit_latency = 2,
165 		.target_residency = 2,
166 		.enter = &intel_idle },
167 	{
168 		.name = "C1E-SNB",
169 		.desc = "MWAIT 0x01",
170 		.flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
171 		.exit_latency = 10,
172 		.target_residency = 20,
173 		.enter = &intel_idle },
174 	{
175 		.name = "C3-SNB",
176 		.desc = "MWAIT 0x10",
177 		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
178 		.exit_latency = 80,
179 		.target_residency = 211,
180 		.enter = &intel_idle },
181 	{
182 		.name = "C6-SNB",
183 		.desc = "MWAIT 0x20",
184 		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
185 		.exit_latency = 104,
186 		.target_residency = 345,
187 		.enter = &intel_idle },
188 	{
189 		.name = "C7-SNB",
190 		.desc = "MWAIT 0x30",
191 		.flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
192 		.exit_latency = 109,
193 		.target_residency = 345,
194 		.enter = &intel_idle },
195 	{
196 		.enter = NULL }
197 };
198 
199 static struct cpuidle_state ivb_cstates[CPUIDLE_STATE_MAX] = {
200 	{
201 		.name = "C1-IVB",
202 		.desc = "MWAIT 0x00",
203 		.flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
204 		.exit_latency = 1,
205 		.target_residency = 1,
206 		.enter = &intel_idle },
207 	{
208 		.name = "C1E-IVB",
209 		.desc = "MWAIT 0x01",
210 		.flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
211 		.exit_latency = 10,
212 		.target_residency = 20,
213 		.enter = &intel_idle },
214 	{
215 		.name = "C3-IVB",
216 		.desc = "MWAIT 0x10",
217 		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
218 		.exit_latency = 59,
219 		.target_residency = 156,
220 		.enter = &intel_idle },
221 	{
222 		.name = "C6-IVB",
223 		.desc = "MWAIT 0x20",
224 		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
225 		.exit_latency = 80,
226 		.target_residency = 300,
227 		.enter = &intel_idle },
228 	{
229 		.name = "C7-IVB",
230 		.desc = "MWAIT 0x30",
231 		.flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
232 		.exit_latency = 87,
233 		.target_residency = 300,
234 		.enter = &intel_idle },
235 	{
236 		.enter = NULL }
237 };
238 
239 static struct cpuidle_state hsw_cstates[CPUIDLE_STATE_MAX] = {
240 	{
241 		.name = "C1-HSW",
242 		.desc = "MWAIT 0x00",
243 		.flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
244 		.exit_latency = 2,
245 		.target_residency = 2,
246 		.enter = &intel_idle },
247 	{
248 		.name = "C1E-HSW",
249 		.desc = "MWAIT 0x01",
250 		.flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
251 		.exit_latency = 10,
252 		.target_residency = 20,
253 		.enter = &intel_idle },
254 	{
255 		.name = "C3-HSW",
256 		.desc = "MWAIT 0x10",
257 		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
258 		.exit_latency = 33,
259 		.target_residency = 100,
260 		.enter = &intel_idle },
261 	{
262 		.name = "C6-HSW",
263 		.desc = "MWAIT 0x20",
264 		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
265 		.exit_latency = 133,
266 		.target_residency = 400,
267 		.enter = &intel_idle },
268 	{
269 		.name = "C7s-HSW",
270 		.desc = "MWAIT 0x32",
271 		.flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
272 		.exit_latency = 166,
273 		.target_residency = 500,
274 		.enter = &intel_idle },
275 	{
276 		.name = "C8-HSW",
277 		.desc = "MWAIT 0x40",
278 		.flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
279 		.exit_latency = 300,
280 		.target_residency = 900,
281 		.enter = &intel_idle },
282 	{
283 		.name = "C9-HSW",
284 		.desc = "MWAIT 0x50",
285 		.flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
286 		.exit_latency = 600,
287 		.target_residency = 1800,
288 		.enter = &intel_idle },
289 	{
290 		.name = "C10-HSW",
291 		.desc = "MWAIT 0x60",
292 		.flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
293 		.exit_latency = 2600,
294 		.target_residency = 7700,
295 		.enter = &intel_idle },
296 	{
297 		.enter = NULL }
298 };
299 
300 static struct cpuidle_state atom_cstates[CPUIDLE_STATE_MAX] = {
301 	{
302 		.name = "C1E-ATM",
303 		.desc = "MWAIT 0x00",
304 		.flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
305 		.exit_latency = 10,
306 		.target_residency = 20,
307 		.enter = &intel_idle },
308 	{
309 		.name = "C2-ATM",
310 		.desc = "MWAIT 0x10",
311 		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID,
312 		.exit_latency = 20,
313 		.target_residency = 80,
314 		.enter = &intel_idle },
315 	{
316 		.name = "C4-ATM",
317 		.desc = "MWAIT 0x30",
318 		.flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
319 		.exit_latency = 100,
320 		.target_residency = 400,
321 		.enter = &intel_idle },
322 	{
323 		.name = "C6-ATM",
324 		.desc = "MWAIT 0x52",
325 		.flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
326 		.exit_latency = 140,
327 		.target_residency = 560,
328 		.enter = &intel_idle },
329 	{
330 		.enter = NULL }
331 };
332 
333 /**
334  * intel_idle
335  * @dev: cpuidle_device
336  * @drv: cpuidle driver
337  * @index: index of cpuidle state
338  *
339  * Must be called under local_irq_disable().
340  */
341 static int intel_idle(struct cpuidle_device *dev,
342 		struct cpuidle_driver *drv, int index)
343 {
344 	unsigned long ecx = 1; /* break on interrupt flag */
345 	struct cpuidle_state *state = &drv->states[index];
346 	unsigned long eax = flg2MWAIT(state->flags);
347 	unsigned int cstate;
348 	int cpu = smp_processor_id();
349 
350 	cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
351 
352 	/*
353 	 * leave_mm() to avoid costly and often unnecessary wakeups
354 	 * for flushing the user TLB's associated with the active mm.
355 	 */
356 	if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
357 		leave_mm(cpu);
358 
359 	if (!(lapic_timer_reliable_states & (1 << (cstate))))
360 		clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
361 
362 	if (!need_resched()) {
363 
364 		__monitor((void *)&current_thread_info()->flags, 0, 0);
365 		smp_mb();
366 		if (!need_resched())
367 			__mwait(eax, ecx);
368 	}
369 
370 	if (!(lapic_timer_reliable_states & (1 << (cstate))))
371 		clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
372 
373 	return index;
374 }
375 
376 static void __setup_broadcast_timer(void *arg)
377 {
378 	unsigned long reason = (unsigned long)arg;
379 	int cpu = smp_processor_id();
380 
381 	reason = reason ?
382 		CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
383 
384 	clockevents_notify(reason, &cpu);
385 }
386 
387 static int cpu_hotplug_notify(struct notifier_block *n,
388 			      unsigned long action, void *hcpu)
389 {
390 	int hotcpu = (unsigned long)hcpu;
391 	struct cpuidle_device *dev;
392 
393 	switch (action & 0xf) {
394 	case CPU_ONLINE:
395 
396 		if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE)
397 			smp_call_function_single(hotcpu, __setup_broadcast_timer,
398 						 (void *)true, 1);
399 
400 		/*
401 		 * Some systems can hotplug a cpu at runtime after
402 		 * the kernel has booted, we have to initialize the
403 		 * driver in this case
404 		 */
405 		dev = per_cpu_ptr(intel_idle_cpuidle_devices, hotcpu);
406 		if (!dev->registered)
407 			intel_idle_cpu_init(hotcpu);
408 
409 		break;
410 	}
411 	return NOTIFY_OK;
412 }
413 
414 static struct notifier_block cpu_hotplug_notifier = {
415 	.notifier_call = cpu_hotplug_notify,
416 };
417 
418 static void auto_demotion_disable(void *dummy)
419 {
420 	unsigned long long msr_bits;
421 
422 	rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
423 	msr_bits &= ~(icpu->auto_demotion_disable_flags);
424 	wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
425 }
426 static void c1e_promotion_disable(void *dummy)
427 {
428 	unsigned long long msr_bits;
429 
430 	rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
431 	msr_bits &= ~0x2;
432 	wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
433 }
434 
435 static const struct idle_cpu idle_cpu_nehalem = {
436 	.state_table = nehalem_cstates,
437 	.auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
438 	.disable_promotion_to_c1e = true,
439 };
440 
441 static const struct idle_cpu idle_cpu_atom = {
442 	.state_table = atom_cstates,
443 };
444 
445 static const struct idle_cpu idle_cpu_lincroft = {
446 	.state_table = atom_cstates,
447 	.auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
448 };
449 
450 static const struct idle_cpu idle_cpu_snb = {
451 	.state_table = snb_cstates,
452 	.disable_promotion_to_c1e = true,
453 };
454 
455 static const struct idle_cpu idle_cpu_ivb = {
456 	.state_table = ivb_cstates,
457 	.disable_promotion_to_c1e = true,
458 };
459 
460 static const struct idle_cpu idle_cpu_hsw = {
461 	.state_table = hsw_cstates,
462 	.disable_promotion_to_c1e = true,
463 };
464 
465 #define ICPU(model, cpu) \
466 	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_MWAIT, (unsigned long)&cpu }
467 
468 static const struct x86_cpu_id intel_idle_ids[] = {
469 	ICPU(0x1a, idle_cpu_nehalem),
470 	ICPU(0x1e, idle_cpu_nehalem),
471 	ICPU(0x1f, idle_cpu_nehalem),
472 	ICPU(0x25, idle_cpu_nehalem),
473 	ICPU(0x2c, idle_cpu_nehalem),
474 	ICPU(0x2e, idle_cpu_nehalem),
475 	ICPU(0x1c, idle_cpu_atom),
476 	ICPU(0x26, idle_cpu_lincroft),
477 	ICPU(0x2f, idle_cpu_nehalem),
478 	ICPU(0x2a, idle_cpu_snb),
479 	ICPU(0x2d, idle_cpu_snb),
480 	ICPU(0x3a, idle_cpu_ivb),
481 	ICPU(0x3e, idle_cpu_ivb),
482 	ICPU(0x3c, idle_cpu_hsw),
483 	ICPU(0x3f, idle_cpu_hsw),
484 	ICPU(0x45, idle_cpu_hsw),
485 	ICPU(0x46, idle_cpu_hsw),
486 	{}
487 };
488 MODULE_DEVICE_TABLE(x86cpu, intel_idle_ids);
489 
490 /*
491  * intel_idle_probe()
492  */
493 static int intel_idle_probe(void)
494 {
495 	unsigned int eax, ebx, ecx;
496 	const struct x86_cpu_id *id;
497 
498 	if (max_cstate == 0) {
499 		pr_debug(PREFIX "disabled\n");
500 		return -EPERM;
501 	}
502 
503 	id = x86_match_cpu(intel_idle_ids);
504 	if (!id) {
505 		if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
506 		    boot_cpu_data.x86 == 6)
507 			pr_debug(PREFIX "does not run on family %d model %d\n",
508 				boot_cpu_data.x86, boot_cpu_data.x86_model);
509 		return -ENODEV;
510 	}
511 
512 	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
513 		return -ENODEV;
514 
515 	cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
516 
517 	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
518 	    !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
519 	    !mwait_substates)
520 			return -ENODEV;
521 
522 	pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
523 
524 	icpu = (const struct idle_cpu *)id->driver_data;
525 	cpuidle_state_table = icpu->state_table;
526 
527 	if (boot_cpu_has(X86_FEATURE_ARAT))	/* Always Reliable APIC Timer */
528 		lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
529 	else
530 		on_each_cpu(__setup_broadcast_timer, (void *)true, 1);
531 
532 	pr_debug(PREFIX "v" INTEL_IDLE_VERSION
533 		" model 0x%X\n", boot_cpu_data.x86_model);
534 
535 	pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
536 		lapic_timer_reliable_states);
537 	return 0;
538 }
539 
540 /*
541  * intel_idle_cpuidle_devices_uninit()
542  * unregister, free cpuidle_devices
543  */
544 static void intel_idle_cpuidle_devices_uninit(void)
545 {
546 	int i;
547 	struct cpuidle_device *dev;
548 
549 	for_each_online_cpu(i) {
550 		dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
551 		cpuidle_unregister_device(dev);
552 	}
553 
554 	free_percpu(intel_idle_cpuidle_devices);
555 	return;
556 }
557 /*
558  * intel_idle_cpuidle_driver_init()
559  * allocate, initialize cpuidle_states
560  */
561 static int intel_idle_cpuidle_driver_init(void)
562 {
563 	int cstate;
564 	struct cpuidle_driver *drv = &intel_idle_driver;
565 
566 	drv->state_count = 1;
567 
568 	for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
569 		int num_substates, mwait_hint, mwait_cstate, mwait_substate;
570 
571 		if (cpuidle_state_table[cstate].enter == NULL)
572 			break;
573 
574 		if (cstate + 1 > max_cstate) {
575 			printk(PREFIX "max_cstate %d reached\n",
576 				max_cstate);
577 			break;
578 		}
579 
580 		mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
581 		mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint);
582 		mwait_substate = MWAIT_HINT2SUBSTATE(mwait_hint);
583 
584 		/* does the state exist in CPUID.MWAIT? */
585 		num_substates = (mwait_substates >> ((mwait_cstate + 1) * 4))
586 					& MWAIT_SUBSTATE_MASK;
587 
588 		/* if sub-state in table is not enumerated by CPUID */
589 		if ((mwait_substate + 1) > num_substates)
590 			continue;
591 
592 		if (((mwait_cstate + 1) > 2) &&
593 			!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
594 			mark_tsc_unstable("TSC halts in idle"
595 					" states deeper than C2");
596 
597 		drv->states[drv->state_count] =	/* structure copy */
598 			cpuidle_state_table[cstate];
599 
600 		drv->state_count += 1;
601 	}
602 
603 	if (icpu->auto_demotion_disable_flags)
604 		on_each_cpu(auto_demotion_disable, NULL, 1);
605 
606 	if (icpu->disable_promotion_to_c1e)	/* each-cpu is redundant */
607 		on_each_cpu(c1e_promotion_disable, NULL, 1);
608 
609 	return 0;
610 }
611 
612 
613 /*
614  * intel_idle_cpu_init()
615  * allocate, initialize, register cpuidle_devices
616  * @cpu: cpu/core to initialize
617  */
618 static int intel_idle_cpu_init(int cpu)
619 {
620 	int cstate;
621 	struct cpuidle_device *dev;
622 
623 	dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
624 
625 	dev->state_count = 1;
626 
627 	for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
628 		int num_substates, mwait_hint, mwait_cstate, mwait_substate;
629 
630 		if (cpuidle_state_table[cstate].enter == NULL)
631 			continue;
632 
633 		if (cstate + 1 > max_cstate) {
634 			printk(PREFIX "max_cstate %d reached\n", max_cstate);
635 			break;
636 		}
637 
638 		mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
639 		mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint);
640 		mwait_substate = MWAIT_HINT2SUBSTATE(mwait_hint);
641 
642 		/* does the state exist in CPUID.MWAIT? */
643 		num_substates = (mwait_substates >> ((mwait_cstate + 1) * 4))
644 					& MWAIT_SUBSTATE_MASK;
645 
646 		/* if sub-state in table is not enumerated by CPUID */
647 		if ((mwait_substate + 1) > num_substates)
648 			continue;
649 
650 		dev->state_count += 1;
651 	}
652 
653 	dev->cpu = cpu;
654 
655 	if (cpuidle_register_device(dev)) {
656 		pr_debug(PREFIX "cpuidle_register_device %d failed!\n", cpu);
657 		intel_idle_cpuidle_devices_uninit();
658 		return -EIO;
659 	}
660 
661 	if (icpu->auto_demotion_disable_flags)
662 		smp_call_function_single(cpu, auto_demotion_disable, NULL, 1);
663 
664 	return 0;
665 }
666 
667 static int __init intel_idle_init(void)
668 {
669 	int retval, i;
670 
671 	/* Do not load intel_idle at all for now if idle= is passed */
672 	if (boot_option_idle_override != IDLE_NO_OVERRIDE)
673 		return -ENODEV;
674 
675 	retval = intel_idle_probe();
676 	if (retval)
677 		return retval;
678 
679 	intel_idle_cpuidle_driver_init();
680 	retval = cpuidle_register_driver(&intel_idle_driver);
681 	if (retval) {
682 		struct cpuidle_driver *drv = cpuidle_get_driver();
683 		printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
684 			drv ? drv->name : "none");
685 		return retval;
686 	}
687 
688 	intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
689 	if (intel_idle_cpuidle_devices == NULL)
690 		return -ENOMEM;
691 
692 	for_each_online_cpu(i) {
693 		retval = intel_idle_cpu_init(i);
694 		if (retval) {
695 			cpuidle_unregister_driver(&intel_idle_driver);
696 			return retval;
697 		}
698 	}
699 	register_cpu_notifier(&cpu_hotplug_notifier);
700 
701 	return 0;
702 }
703 
704 static void __exit intel_idle_exit(void)
705 {
706 	intel_idle_cpuidle_devices_uninit();
707 	cpuidle_unregister_driver(&intel_idle_driver);
708 
709 
710 	if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE)
711 		on_each_cpu(__setup_broadcast_timer, (void *)false, 1);
712 	unregister_cpu_notifier(&cpu_hotplug_notifier);
713 
714 	return;
715 }
716 
717 module_init(intel_idle_init);
718 module_exit(intel_idle_exit);
719 
720 module_param(max_cstate, int, 0444);
721 
722 MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
723 MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
724 MODULE_LICENSE("GPL");
725