Linux Audio

Check our new training course

Loading...
v3.1
  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 <asm/mwait.h>
 65#include <asm/msr.h>
 66
 67#define INTEL_IDLE_VERSION "0.4"
 68#define PREFIX "intel_idle: "
 69
 70static struct cpuidle_driver intel_idle_driver = {
 71	.name = "intel_idle",
 72	.owner = THIS_MODULE,
 73};
 74/* intel_idle.max_cstate=0 disables driver */
 75static int max_cstate = MWAIT_MAX_NUM_CSTATES - 1;
 76
 77static unsigned int mwait_substates;
 78
 79#define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF
 80/* Reliable LAPIC Timer States, bit 1 for C1 etc.  */
 81static unsigned int lapic_timer_reliable_states = (1 << 1);	 /* Default to only C1 */
 82
 83static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
 84static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state);
 85
 86static struct cpuidle_state *cpuidle_state_table;
 
 
 
 
 
 
 
 87
 88/*
 89 * Hardware C-state auto-demotion may not always be optimal.
 90 * Indicate which enable bits to clear here.
 91 */
 92static unsigned long long auto_demotion_disable_flags;
 
 
 93
 94/*
 95 * Set this flag for states where the HW flushes the TLB for us
 96 * and so we don't need cross-calls to keep it consistent.
 97 * If this flag is set, SW flushes the TLB, so even if the
 98 * HW doesn't do the flushing, this flag is safe to use.
 99 */
100#define CPUIDLE_FLAG_TLB_FLUSHED	0x10000
101
102/*
 
 
 
 
 
 
 
 
 
 
103 * States are indexed by the cstate number,
104 * which is also the index into the MWAIT hint array.
105 * Thus C0 is a dummy.
106 */
107static struct cpuidle_state nehalem_cstates[MWAIT_MAX_NUM_CSTATES] = {
108	{ /* MWAIT C0 */ },
109	{ /* MWAIT C1 */
110		.name = "C1-NHM",
111		.desc = "MWAIT 0x00",
112		.driver_data = (void *) 0x00,
113		.flags = CPUIDLE_FLAG_TIME_VALID,
114		.exit_latency = 3,
115		.target_residency = 6,
116		.enter = &intel_idle },
117	{ /* MWAIT C2 */
118		.name = "C3-NHM",
 
 
 
 
 
 
 
 
 
119		.desc = "MWAIT 0x10",
120		.driver_data = (void *) 0x10,
121		.flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
122		.exit_latency = 20,
123		.target_residency = 80,
124		.enter = &intel_idle },
125	{ /* MWAIT C3 */
126		.name = "C6-NHM",
 
127		.desc = "MWAIT 0x20",
128		.driver_data = (void *) 0x20,
129		.flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
130		.exit_latency = 200,
131		.target_residency = 800,
132		.enter = &intel_idle },
 
 
 
133};
134
135static struct cpuidle_state snb_cstates[MWAIT_MAX_NUM_CSTATES] = {
136	{ /* MWAIT C0 */ },
137	{ /* MWAIT C1 */
138		.name = "C1-SNB",
139		.desc = "MWAIT 0x00",
140		.driver_data = (void *) 0x00,
141		.flags = CPUIDLE_FLAG_TIME_VALID,
142		.exit_latency = 1,
143		.target_residency = 1,
144		.enter = &intel_idle },
145	{ /* MWAIT C2 */
146		.name = "C3-SNB",
 
 
 
 
 
 
 
 
147		.desc = "MWAIT 0x10",
148		.driver_data = (void *) 0x10,
149		.flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
150		.exit_latency = 80,
151		.target_residency = 211,
152		.enter = &intel_idle },
153	{ /* MWAIT C3 */
154		.name = "C6-SNB",
 
155		.desc = "MWAIT 0x20",
156		.driver_data = (void *) 0x20,
157		.flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
158		.exit_latency = 104,
159		.target_residency = 345,
160		.enter = &intel_idle },
161	{ /* MWAIT C4 */
162		.name = "C7-SNB",
 
163		.desc = "MWAIT 0x30",
164		.driver_data = (void *) 0x30,
165		.flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
166		.exit_latency = 109,
167		.target_residency = 345,
168		.enter = &intel_idle },
 
 
 
169};
170
171static struct cpuidle_state atom_cstates[MWAIT_MAX_NUM_CSTATES] = {
172	{ /* MWAIT C0 */ },
173	{ /* MWAIT C1 */
174		.name = "C1-ATM",
175		.desc = "MWAIT 0x00",
176		.driver_data = (void *) 0x00,
177		.flags = CPUIDLE_FLAG_TIME_VALID,
178		.exit_latency = 1,
179		.target_residency = 4,
180		.enter = &intel_idle },
181	{ /* MWAIT C2 */
182		.name = "C2-ATM",
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
183		.desc = "MWAIT 0x10",
184		.driver_data = (void *) 0x10,
185		.flags = CPUIDLE_FLAG_TIME_VALID,
186		.exit_latency = 20,
187		.target_residency = 80,
188		.enter = &intel_idle },
189	{ /* MWAIT C3 */ },
190	{ /* MWAIT C4 */
191		.name = "C4-ATM",
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
192		.desc = "MWAIT 0x30",
193		.driver_data = (void *) 0x30,
194		.flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
195		.exit_latency = 100,
196		.target_residency = 400,
197		.enter = &intel_idle },
198	{ /* MWAIT C5 */ },
199	{ /* MWAIT C6 */
200		.name = "C6-ATM",
201		.desc = "MWAIT 0x52",
202		.driver_data = (void *) 0x52,
203		.flags = CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
204		.exit_latency = 140,
205		.target_residency = 560,
206		.enter = &intel_idle },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
207};
208
209/**
210 * intel_idle
211 * @dev: cpuidle_device
212 * @state: cpuidle state
 
213 *
 
214 */
215static int intel_idle(struct cpuidle_device *dev, struct cpuidle_state *state)
 
216{
217	unsigned long ecx = 1; /* break on interrupt flag */
218	unsigned long eax = (unsigned long)cpuidle_get_statedata(state);
 
219	unsigned int cstate;
220	ktime_t kt_before, kt_after;
221	s64 usec_delta;
222	int cpu = smp_processor_id();
223
224	cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
225
226	local_irq_disable();
227
228	/*
229	 * leave_mm() to avoid costly and often unnecessary wakeups
230	 * for flushing the user TLB's associated with the active mm.
231	 */
232	if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
233		leave_mm(cpu);
234
235	if (!(lapic_timer_reliable_states & (1 << (cstate))))
236		clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
237
238	kt_before = ktime_get_real();
239
240	stop_critical_timings();
241	if (!need_resched()) {
242
243		__monitor((void *)&current_thread_info()->flags, 0, 0);
244		smp_mb();
245		if (!need_resched())
246			__mwait(eax, ecx);
247	}
248
249	start_critical_timings();
250
251	kt_after = ktime_get_real();
252	usec_delta = ktime_to_us(ktime_sub(kt_after, kt_before));
253
254	local_irq_enable();
 
255
256	if (!(lapic_timer_reliable_states & (1 << (cstate))))
257		clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
 
 
 
 
 
 
 
 
 
258
259	return usec_delta;
260}
261
262static void __setup_broadcast_timer(void *arg)
263{
264	unsigned long reason = (unsigned long)arg;
265	int cpu = smp_processor_id();
 
 
 
266
267	reason = reason ?
268		CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
 
269
270	clockevents_notify(reason, &cpu);
 
 
271}
272
273static int setup_broadcast_cpuhp_notify(struct notifier_block *n,
274		unsigned long action, void *hcpu)
275{
276	int hotcpu = (unsigned long)hcpu;
277
278	switch (action & 0xf) {
279	case CPU_ONLINE:
280		smp_call_function_single(hotcpu, __setup_broadcast_timer,
281			(void *)true, 1);
282		break;
283	}
284	return NOTIFY_OK;
285}
286
287static struct notifier_block setup_broadcast_notifier = {
288	.notifier_call = setup_broadcast_cpuhp_notify,
 
 
289};
290
291static void auto_demotion_disable(void *dummy)
292{
293	unsigned long long msr_bits;
294
295	rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
296	msr_bits &= ~auto_demotion_disable_flags;
297	wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
298}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
299
300/*
301 * intel_idle_probe()
302 */
303static int intel_idle_probe(void)
304{
305	unsigned int eax, ebx, ecx;
 
306
307	if (max_cstate == 0) {
308		pr_debug(PREFIX "disabled\n");
309		return -EPERM;
310	}
311
312	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
 
 
 
 
 
313		return -ENODEV;
 
314
315	if (!boot_cpu_has(X86_FEATURE_MWAIT))
 
316		return -ENODEV;
 
317
318	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
319		return -ENODEV;
320
321	cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
322
323	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
324		!(ecx & CPUID5_ECX_INTERRUPT_BREAK))
 
325			return -ENODEV;
326
327	pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
328
329
330	if (boot_cpu_data.x86 != 6)	/* family 6 */
331		return -ENODEV;
332
333	switch (boot_cpu_data.x86_model) {
 
334
335	case 0x1A:	/* Core i7, Xeon 5500 series */
336	case 0x1E:	/* Core i7 and i5 Processor - Lynnfield Jasper Forest */
337	case 0x1F:	/* Core i7 and i5 Processor - Nehalem */
338	case 0x2E:	/* Nehalem-EX Xeon */
339	case 0x2F:	/* Westmere-EX Xeon */
340	case 0x25:	/* Westmere */
341	case 0x2C:	/* Westmere */
342		cpuidle_state_table = nehalem_cstates;
343		auto_demotion_disable_flags =
344			(NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE);
345		break;
346
347	case 0x1C:	/* 28 - Atom Processor */
348		cpuidle_state_table = atom_cstates;
349		break;
350
351	case 0x26:	/* 38 - Lincroft Atom Processor */
352		cpuidle_state_table = atom_cstates;
353		auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE;
354		break;
355
356	case 0x2A:	/* SNB */
357	case 0x2D:	/* SNB Xeon */
358		cpuidle_state_table = snb_cstates;
359		break;
360
361	default:
362		pr_debug(PREFIX "does not run on family %d model %d\n",
363			boot_cpu_data.x86, boot_cpu_data.x86_model);
364		return -ENODEV;
365	}
366
367	if (boot_cpu_has(X86_FEATURE_ARAT))	/* Always Reliable APIC Timer */
368		lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
369	else {
370		smp_call_function(__setup_broadcast_timer, (void *)true, 1);
371		register_cpu_notifier(&setup_broadcast_notifier);
372	}
373
374	pr_debug(PREFIX "v" INTEL_IDLE_VERSION
375		" model 0x%X\n", boot_cpu_data.x86_model);
376
377	pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
378		lapic_timer_reliable_states);
379	return 0;
380}
381
382/*
383 * intel_idle_cpuidle_devices_uninit()
384 * unregister, free cpuidle_devices
385 */
386static void intel_idle_cpuidle_devices_uninit(void)
387{
388	int i;
389	struct cpuidle_device *dev;
390
391	for_each_online_cpu(i) {
392		dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
393		cpuidle_unregister_device(dev);
394	}
 
395
396	free_percpu(intel_idle_cpuidle_devices);
397	return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
398}
 
399/*
400 * intel_idle_cpuidle_devices_init()
401 * allocate, initialize, register cpuidle_devices
402 */
403static int intel_idle_cpuidle_devices_init(void)
 
 
 
 
404{
405	int i, cstate;
406	struct cpuidle_device *dev;
407
408	intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
409	if (intel_idle_cpuidle_devices == NULL)
410		return -ENOMEM;
411
412	for_each_online_cpu(i) {
413		dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
414
415		dev->state_count = 1;
 
 
 
 
 
 
 
 
 
 
 
416
417		for (cstate = 1; cstate < MWAIT_MAX_NUM_CSTATES; ++cstate) {
418			int num_substates;
 
 
 
 
419
420			if (cstate > max_cstate) {
421				printk(PREFIX "max_cstate %d reached\n",
422					max_cstate);
423				break;
424			}
 
425
426			/* does the state exist in CPUID.MWAIT? */
427			num_substates = (mwait_substates >> ((cstate) * 4))
428						& MWAIT_SUBSTATE_MASK;
429			if (num_substates == 0)
430				continue;
431			/* is the state not enabled? */
432			if (cpuidle_state_table[cstate].enter == NULL) {
433				/* does the driver not know about the state? */
434				if (*cpuidle_state_table[cstate].name == '\0')
435					pr_debug(PREFIX "unaware of model 0x%x"
436						" MWAIT %d please"
437						" contact lenb@kernel.org",
438					boot_cpu_data.x86_model, cstate);
439				continue;
440			}
441
442			if ((cstate > 2) &&
443				!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
444				mark_tsc_unstable("TSC halts in idle"
445					" states deeper than C2");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
446
447			dev->states[dev->state_count] =	/* structure copy */
448				cpuidle_state_table[cstate];
 
449
450			dev->state_count += 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
451		}
452
453		dev->cpu = i;
454		if (cpuidle_register_device(dev)) {
455			pr_debug(PREFIX "cpuidle_register_device %d failed!\n",
456				 i);
457			intel_idle_cpuidle_devices_uninit();
458			return -EIO;
 
 
 
 
 
 
 
 
 
 
459		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
460	}
461	if (auto_demotion_disable_flags)
462		smp_call_function(auto_demotion_disable, NULL, 1);
 
 
 
 
463
464	return 0;
465}
466
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
467
468static int __init intel_idle_init(void)
469{
470	int retval;
471
472	/* Do not load intel_idle at all for now if idle= is passed */
473	if (boot_option_idle_override != IDLE_NO_OVERRIDE)
474		return -ENODEV;
475
476	retval = intel_idle_probe();
477	if (retval)
478		return retval;
479
 
 
 
 
 
480	retval = cpuidle_register_driver(&intel_idle_driver);
481	if (retval) {
482		printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
483			cpuidle_get_driver()->name);
484		return retval;
 
485	}
486
487	retval = intel_idle_cpuidle_devices_init();
488	if (retval) {
489		cpuidle_unregister_driver(&intel_idle_driver);
490		return retval;
491	}
 
 
 
 
 
492
493	return 0;
494}
495
496static void __exit intel_idle_exit(void)
497{
498	intel_idle_cpuidle_devices_uninit();
499	cpuidle_unregister_driver(&intel_idle_driver);
 
 
 
500
501	if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE) {
502		smp_call_function(__setup_broadcast_timer, (void *)false, 1);
503		unregister_cpu_notifier(&setup_broadcast_notifier);
504	}
505
506	return;
507}
 
508
509module_init(intel_idle_init);
510module_exit(intel_idle_exit);
511
 
 
 
512module_param(max_cstate, int, 0444);
513
514MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
515MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
516MODULE_LICENSE("GPL");
v4.17
   1/*
   2 * intel_idle.c - native hardware idle loop for modern Intel processors
   3 *
   4 * Copyright (c) 2013, 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 */
  50
  51/* un-comment DEBUG to enable pr_debug() statements */
  52#define DEBUG
  53
  54#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  55
  56#include <linux/kernel.h>
  57#include <linux/cpuidle.h>
  58#include <linux/tick.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/moduleparam.h>
  64#include <asm/cpu_device_id.h>
  65#include <asm/intel-family.h>
  66#include <asm/mwait.h>
  67#include <asm/msr.h>
  68
  69#define INTEL_IDLE_VERSION "0.4.1"
 
  70
  71static struct cpuidle_driver intel_idle_driver = {
  72	.name = "intel_idle",
  73	.owner = THIS_MODULE,
  74};
  75/* intel_idle.max_cstate=0 disables driver */
  76static int max_cstate = CPUIDLE_STATE_MAX - 1;
  77
  78static unsigned int mwait_substates;
  79
  80#define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF
  81/* Reliable LAPIC Timer States, bit 1 for C1 etc.  */
  82static unsigned int lapic_timer_reliable_states = (1 << 1);	 /* Default to only C1 */
  83
  84struct 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 byt_auto_demotion_disable_flag;
  93	bool disable_promotion_to_c1e;
  94};
  95
  96static const struct idle_cpu *icpu;
  97static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
  98static int intel_idle(struct cpuidle_device *dev,
  99			struct cpuidle_driver *drv, int index);
 100static void intel_idle_s2idle(struct cpuidle_device *dev,
 101			      struct cpuidle_driver *drv, int index);
 102static struct cpuidle_state *cpuidle_state_table;
 103
 104/*
 105 * Set this flag for states where the HW flushes the TLB for us
 106 * and so we don't need cross-calls to keep it consistent.
 107 * If this flag is set, SW flushes the TLB, so even if the
 108 * HW doesn't do the flushing, this flag is safe to use.
 109 */
 110#define CPUIDLE_FLAG_TLB_FLUSHED	0x10000
 111
 112/*
 113 * MWAIT takes an 8-bit "hint" in EAX "suggesting"
 114 * the C-state (top nibble) and sub-state (bottom nibble)
 115 * 0x00 means "MWAIT(C1)", 0x10 means "MWAIT(C2)" etc.
 116 *
 117 * We store the hint at the top of our "flags" for each state.
 118 */
 119#define flg2MWAIT(flags) (((flags) >> 24) & 0xFF)
 120#define MWAIT2flg(eax) ((eax & 0xFF) << 24)
 121
 122/*
 123 * States are indexed by the cstate number,
 124 * which is also the index into the MWAIT hint array.
 125 * Thus C0 is a dummy.
 126 */
 127static struct cpuidle_state nehalem_cstates[] = {
 128	{
 129		.name = "C1",
 
 130		.desc = "MWAIT 0x00",
 131		.flags = MWAIT2flg(0x00),
 
 132		.exit_latency = 3,
 133		.target_residency = 6,
 134		.enter = &intel_idle,
 135		.enter_s2idle = intel_idle_s2idle, },
 136	{
 137		.name = "C1E",
 138		.desc = "MWAIT 0x01",
 139		.flags = MWAIT2flg(0x01),
 140		.exit_latency = 10,
 141		.target_residency = 20,
 142		.enter = &intel_idle,
 143		.enter_s2idle = intel_idle_s2idle, },
 144	{
 145		.name = "C3",
 146		.desc = "MWAIT 0x10",
 147		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
 
 148		.exit_latency = 20,
 149		.target_residency = 80,
 150		.enter = &intel_idle,
 151		.enter_s2idle = intel_idle_s2idle, },
 152	{
 153		.name = "C6",
 154		.desc = "MWAIT 0x20",
 155		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 
 156		.exit_latency = 200,
 157		.target_residency = 800,
 158		.enter = &intel_idle,
 159		.enter_s2idle = intel_idle_s2idle, },
 160	{
 161		.enter = NULL }
 162};
 163
 164static struct cpuidle_state snb_cstates[] = {
 165	{
 166		.name = "C1",
 
 167		.desc = "MWAIT 0x00",
 168		.flags = MWAIT2flg(0x00),
 169		.exit_latency = 2,
 170		.target_residency = 2,
 171		.enter = &intel_idle,
 172		.enter_s2idle = intel_idle_s2idle, },
 173	{
 174		.name = "C1E",
 175		.desc = "MWAIT 0x01",
 176		.flags = MWAIT2flg(0x01),
 177		.exit_latency = 10,
 178		.target_residency = 20,
 179		.enter = &intel_idle,
 180		.enter_s2idle = intel_idle_s2idle, },
 181	{
 182		.name = "C3",
 183		.desc = "MWAIT 0x10",
 184		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
 
 185		.exit_latency = 80,
 186		.target_residency = 211,
 187		.enter = &intel_idle,
 188		.enter_s2idle = intel_idle_s2idle, },
 189	{
 190		.name = "C6",
 191		.desc = "MWAIT 0x20",
 192		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 
 193		.exit_latency = 104,
 194		.target_residency = 345,
 195		.enter = &intel_idle,
 196		.enter_s2idle = intel_idle_s2idle, },
 197	{
 198		.name = "C7",
 199		.desc = "MWAIT 0x30",
 200		.flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
 
 201		.exit_latency = 109,
 202		.target_residency = 345,
 203		.enter = &intel_idle,
 204		.enter_s2idle = intel_idle_s2idle, },
 205	{
 206		.enter = NULL }
 207};
 208
 209static struct cpuidle_state byt_cstates[] = {
 210	{
 211		.name = "C1",
 
 212		.desc = "MWAIT 0x00",
 213		.flags = MWAIT2flg(0x00),
 
 214		.exit_latency = 1,
 215		.target_residency = 1,
 216		.enter = &intel_idle,
 217		.enter_s2idle = intel_idle_s2idle, },
 218	{
 219		.name = "C6N",
 220		.desc = "MWAIT 0x58",
 221		.flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
 222		.exit_latency = 300,
 223		.target_residency = 275,
 224		.enter = &intel_idle,
 225		.enter_s2idle = intel_idle_s2idle, },
 226	{
 227		.name = "C6S",
 228		.desc = "MWAIT 0x52",
 229		.flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
 230		.exit_latency = 500,
 231		.target_residency = 560,
 232		.enter = &intel_idle,
 233		.enter_s2idle = intel_idle_s2idle, },
 234	{
 235		.name = "C7",
 236		.desc = "MWAIT 0x60",
 237		.flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
 238		.exit_latency = 1200,
 239		.target_residency = 4000,
 240		.enter = &intel_idle,
 241		.enter_s2idle = intel_idle_s2idle, },
 242	{
 243		.name = "C7S",
 244		.desc = "MWAIT 0x64",
 245		.flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
 246		.exit_latency = 10000,
 247		.target_residency = 20000,
 248		.enter = &intel_idle,
 249		.enter_s2idle = intel_idle_s2idle, },
 250	{
 251		.enter = NULL }
 252};
 253
 254static struct cpuidle_state cht_cstates[] = {
 255	{
 256		.name = "C1",
 257		.desc = "MWAIT 0x00",
 258		.flags = MWAIT2flg(0x00),
 259		.exit_latency = 1,
 260		.target_residency = 1,
 261		.enter = &intel_idle,
 262		.enter_s2idle = intel_idle_s2idle, },
 263	{
 264		.name = "C6N",
 265		.desc = "MWAIT 0x58",
 266		.flags = MWAIT2flg(0x58) | CPUIDLE_FLAG_TLB_FLUSHED,
 267		.exit_latency = 80,
 268		.target_residency = 275,
 269		.enter = &intel_idle,
 270		.enter_s2idle = intel_idle_s2idle, },
 271	{
 272		.name = "C6S",
 273		.desc = "MWAIT 0x52",
 274		.flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
 275		.exit_latency = 200,
 276		.target_residency = 560,
 277		.enter = &intel_idle,
 278		.enter_s2idle = intel_idle_s2idle, },
 279	{
 280		.name = "C7",
 281		.desc = "MWAIT 0x60",
 282		.flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
 283		.exit_latency = 1200,
 284		.target_residency = 4000,
 285		.enter = &intel_idle,
 286		.enter_s2idle = intel_idle_s2idle, },
 287	{
 288		.name = "C7S",
 289		.desc = "MWAIT 0x64",
 290		.flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
 291		.exit_latency = 10000,
 292		.target_residency = 20000,
 293		.enter = &intel_idle,
 294		.enter_s2idle = intel_idle_s2idle, },
 295	{
 296		.enter = NULL }
 297};
 298
 299static struct cpuidle_state ivb_cstates[] = {
 300	{
 301		.name = "C1",
 302		.desc = "MWAIT 0x00",
 303		.flags = MWAIT2flg(0x00),
 304		.exit_latency = 1,
 305		.target_residency = 1,
 306		.enter = &intel_idle,
 307		.enter_s2idle = intel_idle_s2idle, },
 308	{
 309		.name = "C1E",
 310		.desc = "MWAIT 0x01",
 311		.flags = MWAIT2flg(0x01),
 312		.exit_latency = 10,
 313		.target_residency = 20,
 314		.enter = &intel_idle,
 315		.enter_s2idle = intel_idle_s2idle, },
 316	{
 317		.name = "C3",
 318		.desc = "MWAIT 0x10",
 319		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
 320		.exit_latency = 59,
 321		.target_residency = 156,
 322		.enter = &intel_idle,
 323		.enter_s2idle = intel_idle_s2idle, },
 324	{
 325		.name = "C6",
 326		.desc = "MWAIT 0x20",
 327		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 328		.exit_latency = 80,
 329		.target_residency = 300,
 330		.enter = &intel_idle,
 331		.enter_s2idle = intel_idle_s2idle, },
 332	{
 333		.name = "C7",
 334		.desc = "MWAIT 0x30",
 335		.flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
 336		.exit_latency = 87,
 337		.target_residency = 300,
 338		.enter = &intel_idle,
 339		.enter_s2idle = intel_idle_s2idle, },
 340	{
 341		.enter = NULL }
 342};
 343
 344static struct cpuidle_state ivt_cstates[] = {
 345	{
 346		.name = "C1",
 347		.desc = "MWAIT 0x00",
 348		.flags = MWAIT2flg(0x00),
 349		.exit_latency = 1,
 350		.target_residency = 1,
 351		.enter = &intel_idle,
 352		.enter_s2idle = intel_idle_s2idle, },
 353	{
 354		.name = "C1E",
 355		.desc = "MWAIT 0x01",
 356		.flags = MWAIT2flg(0x01),
 357		.exit_latency = 10,
 358		.target_residency = 80,
 359		.enter = &intel_idle,
 360		.enter_s2idle = intel_idle_s2idle, },
 361	{
 362		.name = "C3",
 363		.desc = "MWAIT 0x10",
 364		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
 365		.exit_latency = 59,
 366		.target_residency = 156,
 367		.enter = &intel_idle,
 368		.enter_s2idle = intel_idle_s2idle, },
 369	{
 370		.name = "C6",
 371		.desc = "MWAIT 0x20",
 372		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 373		.exit_latency = 82,
 374		.target_residency = 300,
 375		.enter = &intel_idle,
 376		.enter_s2idle = intel_idle_s2idle, },
 377	{
 378		.enter = NULL }
 379};
 380
 381static struct cpuidle_state ivt_cstates_4s[] = {
 382	{
 383		.name = "C1",
 384		.desc = "MWAIT 0x00",
 385		.flags = MWAIT2flg(0x00),
 386		.exit_latency = 1,
 387		.target_residency = 1,
 388		.enter = &intel_idle,
 389		.enter_s2idle = intel_idle_s2idle, },
 390	{
 391		.name = "C1E",
 392		.desc = "MWAIT 0x01",
 393		.flags = MWAIT2flg(0x01),
 394		.exit_latency = 10,
 395		.target_residency = 250,
 396		.enter = &intel_idle,
 397		.enter_s2idle = intel_idle_s2idle, },
 398	{
 399		.name = "C3",
 400		.desc = "MWAIT 0x10",
 401		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
 402		.exit_latency = 59,
 403		.target_residency = 300,
 404		.enter = &intel_idle,
 405		.enter_s2idle = intel_idle_s2idle, },
 406	{
 407		.name = "C6",
 408		.desc = "MWAIT 0x20",
 409		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 410		.exit_latency = 84,
 411		.target_residency = 400,
 412		.enter = &intel_idle,
 413		.enter_s2idle = intel_idle_s2idle, },
 414	{
 415		.enter = NULL }
 416};
 417
 418static struct cpuidle_state ivt_cstates_8s[] = {
 419	{
 420		.name = "C1",
 421		.desc = "MWAIT 0x00",
 422		.flags = MWAIT2flg(0x00),
 423		.exit_latency = 1,
 424		.target_residency = 1,
 425		.enter = &intel_idle,
 426		.enter_s2idle = intel_idle_s2idle, },
 427	{
 428		.name = "C1E",
 429		.desc = "MWAIT 0x01",
 430		.flags = MWAIT2flg(0x01),
 431		.exit_latency = 10,
 432		.target_residency = 500,
 433		.enter = &intel_idle,
 434		.enter_s2idle = intel_idle_s2idle, },
 435	{
 436		.name = "C3",
 437		.desc = "MWAIT 0x10",
 438		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
 439		.exit_latency = 59,
 440		.target_residency = 600,
 441		.enter = &intel_idle,
 442		.enter_s2idle = intel_idle_s2idle, },
 443	{
 444		.name = "C6",
 445		.desc = "MWAIT 0x20",
 446		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 447		.exit_latency = 88,
 448		.target_residency = 700,
 449		.enter = &intel_idle,
 450		.enter_s2idle = intel_idle_s2idle, },
 451	{
 452		.enter = NULL }
 453};
 454
 455static struct cpuidle_state hsw_cstates[] = {
 456	{
 457		.name = "C1",
 458		.desc = "MWAIT 0x00",
 459		.flags = MWAIT2flg(0x00),
 460		.exit_latency = 2,
 461		.target_residency = 2,
 462		.enter = &intel_idle,
 463		.enter_s2idle = intel_idle_s2idle, },
 464	{
 465		.name = "C1E",
 466		.desc = "MWAIT 0x01",
 467		.flags = MWAIT2flg(0x01),
 468		.exit_latency = 10,
 469		.target_residency = 20,
 470		.enter = &intel_idle,
 471		.enter_s2idle = intel_idle_s2idle, },
 472	{
 473		.name = "C3",
 474		.desc = "MWAIT 0x10",
 475		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
 476		.exit_latency = 33,
 477		.target_residency = 100,
 478		.enter = &intel_idle,
 479		.enter_s2idle = intel_idle_s2idle, },
 480	{
 481		.name = "C6",
 482		.desc = "MWAIT 0x20",
 483		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 484		.exit_latency = 133,
 485		.target_residency = 400,
 486		.enter = &intel_idle,
 487		.enter_s2idle = intel_idle_s2idle, },
 488	{
 489		.name = "C7s",
 490		.desc = "MWAIT 0x32",
 491		.flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
 492		.exit_latency = 166,
 493		.target_residency = 500,
 494		.enter = &intel_idle,
 495		.enter_s2idle = intel_idle_s2idle, },
 496	{
 497		.name = "C8",
 498		.desc = "MWAIT 0x40",
 499		.flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
 500		.exit_latency = 300,
 501		.target_residency = 900,
 502		.enter = &intel_idle,
 503		.enter_s2idle = intel_idle_s2idle, },
 504	{
 505		.name = "C9",
 506		.desc = "MWAIT 0x50",
 507		.flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
 508		.exit_latency = 600,
 509		.target_residency = 1800,
 510		.enter = &intel_idle,
 511		.enter_s2idle = intel_idle_s2idle, },
 512	{
 513		.name = "C10",
 514		.desc = "MWAIT 0x60",
 515		.flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
 516		.exit_latency = 2600,
 517		.target_residency = 7700,
 518		.enter = &intel_idle,
 519		.enter_s2idle = intel_idle_s2idle, },
 520	{
 521		.enter = NULL }
 522};
 523static struct cpuidle_state bdw_cstates[] = {
 524	{
 525		.name = "C1",
 526		.desc = "MWAIT 0x00",
 527		.flags = MWAIT2flg(0x00),
 528		.exit_latency = 2,
 529		.target_residency = 2,
 530		.enter = &intel_idle,
 531		.enter_s2idle = intel_idle_s2idle, },
 532	{
 533		.name = "C1E",
 534		.desc = "MWAIT 0x01",
 535		.flags = MWAIT2flg(0x01),
 536		.exit_latency = 10,
 537		.target_residency = 20,
 538		.enter = &intel_idle,
 539		.enter_s2idle = intel_idle_s2idle, },
 540	{
 541		.name = "C3",
 542		.desc = "MWAIT 0x10",
 543		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
 544		.exit_latency = 40,
 545		.target_residency = 100,
 546		.enter = &intel_idle,
 547		.enter_s2idle = intel_idle_s2idle, },
 548	{
 549		.name = "C6",
 550		.desc = "MWAIT 0x20",
 551		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 552		.exit_latency = 133,
 553		.target_residency = 400,
 554		.enter = &intel_idle,
 555		.enter_s2idle = intel_idle_s2idle, },
 556	{
 557		.name = "C7s",
 558		.desc = "MWAIT 0x32",
 559		.flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TLB_FLUSHED,
 560		.exit_latency = 166,
 561		.target_residency = 500,
 562		.enter = &intel_idle,
 563		.enter_s2idle = intel_idle_s2idle, },
 564	{
 565		.name = "C8",
 566		.desc = "MWAIT 0x40",
 567		.flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
 568		.exit_latency = 300,
 569		.target_residency = 900,
 570		.enter = &intel_idle,
 571		.enter_s2idle = intel_idle_s2idle, },
 572	{
 573		.name = "C9",
 574		.desc = "MWAIT 0x50",
 575		.flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
 576		.exit_latency = 600,
 577		.target_residency = 1800,
 578		.enter = &intel_idle,
 579		.enter_s2idle = intel_idle_s2idle, },
 580	{
 581		.name = "C10",
 582		.desc = "MWAIT 0x60",
 583		.flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
 584		.exit_latency = 2600,
 585		.target_residency = 7700,
 586		.enter = &intel_idle,
 587		.enter_s2idle = intel_idle_s2idle, },
 588	{
 589		.enter = NULL }
 590};
 591
 592static struct cpuidle_state skl_cstates[] = {
 593	{
 594		.name = "C1",
 595		.desc = "MWAIT 0x00",
 596		.flags = MWAIT2flg(0x00),
 597		.exit_latency = 2,
 598		.target_residency = 2,
 599		.enter = &intel_idle,
 600		.enter_s2idle = intel_idle_s2idle, },
 601	{
 602		.name = "C1E",
 603		.desc = "MWAIT 0x01",
 604		.flags = MWAIT2flg(0x01),
 605		.exit_latency = 10,
 606		.target_residency = 20,
 607		.enter = &intel_idle,
 608		.enter_s2idle = intel_idle_s2idle, },
 609	{
 610		.name = "C3",
 611		.desc = "MWAIT 0x10",
 612		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
 613		.exit_latency = 70,
 614		.target_residency = 100,
 615		.enter = &intel_idle,
 616		.enter_s2idle = intel_idle_s2idle, },
 617	{
 618		.name = "C6",
 619		.desc = "MWAIT 0x20",
 620		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 621		.exit_latency = 85,
 622		.target_residency = 200,
 623		.enter = &intel_idle,
 624		.enter_s2idle = intel_idle_s2idle, },
 625	{
 626		.name = "C7s",
 627		.desc = "MWAIT 0x33",
 628		.flags = MWAIT2flg(0x33) | CPUIDLE_FLAG_TLB_FLUSHED,
 629		.exit_latency = 124,
 630		.target_residency = 800,
 631		.enter = &intel_idle,
 632		.enter_s2idle = intel_idle_s2idle, },
 633	{
 634		.name = "C8",
 635		.desc = "MWAIT 0x40",
 636		.flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
 637		.exit_latency = 200,
 638		.target_residency = 800,
 639		.enter = &intel_idle,
 640		.enter_s2idle = intel_idle_s2idle, },
 641	{
 642		.name = "C9",
 643		.desc = "MWAIT 0x50",
 644		.flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
 645		.exit_latency = 480,
 646		.target_residency = 5000,
 647		.enter = &intel_idle,
 648		.enter_s2idle = intel_idle_s2idle, },
 649	{
 650		.name = "C10",
 651		.desc = "MWAIT 0x60",
 652		.flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
 653		.exit_latency = 890,
 654		.target_residency = 5000,
 655		.enter = &intel_idle,
 656		.enter_s2idle = intel_idle_s2idle, },
 657	{
 658		.enter = NULL }
 659};
 660
 661static struct cpuidle_state skx_cstates[] = {
 662	{
 663		.name = "C1",
 664		.desc = "MWAIT 0x00",
 665		.flags = MWAIT2flg(0x00),
 666		.exit_latency = 2,
 667		.target_residency = 2,
 668		.enter = &intel_idle,
 669		.enter_s2idle = intel_idle_s2idle, },
 670	{
 671		.name = "C1E",
 672		.desc = "MWAIT 0x01",
 673		.flags = MWAIT2flg(0x01),
 674		.exit_latency = 10,
 675		.target_residency = 20,
 676		.enter = &intel_idle,
 677		.enter_s2idle = intel_idle_s2idle, },
 678	{
 679		.name = "C6",
 680		.desc = "MWAIT 0x20",
 681		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 682		.exit_latency = 133,
 683		.target_residency = 600,
 684		.enter = &intel_idle,
 685		.enter_s2idle = intel_idle_s2idle, },
 686	{
 687		.enter = NULL }
 688};
 689
 690static struct cpuidle_state atom_cstates[] = {
 691	{
 692		.name = "C1E",
 693		.desc = "MWAIT 0x00",
 694		.flags = MWAIT2flg(0x00),
 695		.exit_latency = 10,
 696		.target_residency = 20,
 697		.enter = &intel_idle,
 698		.enter_s2idle = intel_idle_s2idle, },
 699	{
 700		.name = "C2",
 701		.desc = "MWAIT 0x10",
 702		.flags = MWAIT2flg(0x10),
 
 703		.exit_latency = 20,
 704		.target_residency = 80,
 705		.enter = &intel_idle,
 706		.enter_s2idle = intel_idle_s2idle, },
 707	{
 708		.name = "C4",
 709		.desc = "MWAIT 0x30",
 710		.flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
 711		.exit_latency = 100,
 712		.target_residency = 400,
 713		.enter = &intel_idle,
 714		.enter_s2idle = intel_idle_s2idle, },
 715	{
 716		.name = "C6",
 717		.desc = "MWAIT 0x52",
 718		.flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
 719		.exit_latency = 140,
 720		.target_residency = 560,
 721		.enter = &intel_idle,
 722		.enter_s2idle = intel_idle_s2idle, },
 723	{
 724		.enter = NULL }
 725};
 726static struct cpuidle_state tangier_cstates[] = {
 727	{
 728		.name = "C1",
 729		.desc = "MWAIT 0x00",
 730		.flags = MWAIT2flg(0x00),
 731		.exit_latency = 1,
 732		.target_residency = 4,
 733		.enter = &intel_idle,
 734		.enter_s2idle = intel_idle_s2idle, },
 735	{
 736		.name = "C4",
 737		.desc = "MWAIT 0x30",
 738		.flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TLB_FLUSHED,
 
 739		.exit_latency = 100,
 740		.target_residency = 400,
 741		.enter = &intel_idle,
 742		.enter_s2idle = intel_idle_s2idle, },
 743	{
 744		.name = "C6",
 745		.desc = "MWAIT 0x52",
 746		.flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TLB_FLUSHED,
 
 747		.exit_latency = 140,
 748		.target_residency = 560,
 749		.enter = &intel_idle,
 750		.enter_s2idle = intel_idle_s2idle, },
 751	{
 752		.name = "C7",
 753		.desc = "MWAIT 0x60",
 754		.flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
 755		.exit_latency = 1200,
 756		.target_residency = 4000,
 757		.enter = &intel_idle,
 758		.enter_s2idle = intel_idle_s2idle, },
 759	{
 760		.name = "C9",
 761		.desc = "MWAIT 0x64",
 762		.flags = MWAIT2flg(0x64) | CPUIDLE_FLAG_TLB_FLUSHED,
 763		.exit_latency = 10000,
 764		.target_residency = 20000,
 765		.enter = &intel_idle,
 766		.enter_s2idle = intel_idle_s2idle, },
 767	{
 768		.enter = NULL }
 769};
 770static struct cpuidle_state avn_cstates[] = {
 771	{
 772		.name = "C1",
 773		.desc = "MWAIT 0x00",
 774		.flags = MWAIT2flg(0x00),
 775		.exit_latency = 2,
 776		.target_residency = 2,
 777		.enter = &intel_idle,
 778		.enter_s2idle = intel_idle_s2idle, },
 779	{
 780		.name = "C6",
 781		.desc = "MWAIT 0x51",
 782		.flags = MWAIT2flg(0x51) | CPUIDLE_FLAG_TLB_FLUSHED,
 783		.exit_latency = 15,
 784		.target_residency = 45,
 785		.enter = &intel_idle,
 786		.enter_s2idle = intel_idle_s2idle, },
 787	{
 788		.enter = NULL }
 789};
 790static struct cpuidle_state knl_cstates[] = {
 791	{
 792		.name = "C1",
 793		.desc = "MWAIT 0x00",
 794		.flags = MWAIT2flg(0x00),
 795		.exit_latency = 1,
 796		.target_residency = 2,
 797		.enter = &intel_idle,
 798		.enter_s2idle = intel_idle_s2idle },
 799	{
 800		.name = "C6",
 801		.desc = "MWAIT 0x10",
 802		.flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TLB_FLUSHED,
 803		.exit_latency = 120,
 804		.target_residency = 500,
 805		.enter = &intel_idle,
 806		.enter_s2idle = intel_idle_s2idle },
 807	{
 808		.enter = NULL }
 809};
 810
 811static struct cpuidle_state bxt_cstates[] = {
 812	{
 813		.name = "C1",
 814		.desc = "MWAIT 0x00",
 815		.flags = MWAIT2flg(0x00),
 816		.exit_latency = 2,
 817		.target_residency = 2,
 818		.enter = &intel_idle,
 819		.enter_s2idle = intel_idle_s2idle, },
 820	{
 821		.name = "C1E",
 822		.desc = "MWAIT 0x01",
 823		.flags = MWAIT2flg(0x01),
 824		.exit_latency = 10,
 825		.target_residency = 20,
 826		.enter = &intel_idle,
 827		.enter_s2idle = intel_idle_s2idle, },
 828	{
 829		.name = "C6",
 830		.desc = "MWAIT 0x20",
 831		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 832		.exit_latency = 133,
 833		.target_residency = 133,
 834		.enter = &intel_idle,
 835		.enter_s2idle = intel_idle_s2idle, },
 836	{
 837		.name = "C7s",
 838		.desc = "MWAIT 0x31",
 839		.flags = MWAIT2flg(0x31) | CPUIDLE_FLAG_TLB_FLUSHED,
 840		.exit_latency = 155,
 841		.target_residency = 155,
 842		.enter = &intel_idle,
 843		.enter_s2idle = intel_idle_s2idle, },
 844	{
 845		.name = "C8",
 846		.desc = "MWAIT 0x40",
 847		.flags = MWAIT2flg(0x40) | CPUIDLE_FLAG_TLB_FLUSHED,
 848		.exit_latency = 1000,
 849		.target_residency = 1000,
 850		.enter = &intel_idle,
 851		.enter_s2idle = intel_idle_s2idle, },
 852	{
 853		.name = "C9",
 854		.desc = "MWAIT 0x50",
 855		.flags = MWAIT2flg(0x50) | CPUIDLE_FLAG_TLB_FLUSHED,
 856		.exit_latency = 2000,
 857		.target_residency = 2000,
 858		.enter = &intel_idle,
 859		.enter_s2idle = intel_idle_s2idle, },
 860	{
 861		.name = "C10",
 862		.desc = "MWAIT 0x60",
 863		.flags = MWAIT2flg(0x60) | CPUIDLE_FLAG_TLB_FLUSHED,
 864		.exit_latency = 10000,
 865		.target_residency = 10000,
 866		.enter = &intel_idle,
 867		.enter_s2idle = intel_idle_s2idle, },
 868	{
 869		.enter = NULL }
 870};
 871
 872static struct cpuidle_state dnv_cstates[] = {
 873	{
 874		.name = "C1",
 875		.desc = "MWAIT 0x00",
 876		.flags = MWAIT2flg(0x00),
 877		.exit_latency = 2,
 878		.target_residency = 2,
 879		.enter = &intel_idle,
 880		.enter_s2idle = intel_idle_s2idle, },
 881	{
 882		.name = "C1E",
 883		.desc = "MWAIT 0x01",
 884		.flags = MWAIT2flg(0x01),
 885		.exit_latency = 10,
 886		.target_residency = 20,
 887		.enter = &intel_idle,
 888		.enter_s2idle = intel_idle_s2idle, },
 889	{
 890		.name = "C6",
 891		.desc = "MWAIT 0x20",
 892		.flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TLB_FLUSHED,
 893		.exit_latency = 50,
 894		.target_residency = 500,
 895		.enter = &intel_idle,
 896		.enter_s2idle = intel_idle_s2idle, },
 897	{
 898		.enter = NULL }
 899};
 900
 901/**
 902 * intel_idle
 903 * @dev: cpuidle_device
 904 * @drv: cpuidle driver
 905 * @index: index of cpuidle state
 906 *
 907 * Must be called under local_irq_disable().
 908 */
 909static __cpuidle int intel_idle(struct cpuidle_device *dev,
 910				struct cpuidle_driver *drv, int index)
 911{
 912	unsigned long ecx = 1; /* break on interrupt flag */
 913	struct cpuidle_state *state = &drv->states[index];
 914	unsigned long eax = flg2MWAIT(state->flags);
 915	unsigned int cstate;
 916	bool uninitialized_var(tick);
 
 917	int cpu = smp_processor_id();
 918
 
 
 
 
 919	/*
 920	 * leave_mm() to avoid costly and often unnecessary wakeups
 921	 * for flushing the user TLB's associated with the active mm.
 922	 */
 923	if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
 924		leave_mm(cpu);
 925
 926	if (!static_cpu_has(X86_FEATURE_ARAT)) {
 927		cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) &
 928				MWAIT_CSTATE_MASK) + 1;
 929		tick = false;
 930		if (!(lapic_timer_reliable_states & (1 << (cstate)))) {
 931			tick = true;
 932			tick_broadcast_enter();
 933		}
 
 
 
 
 934	}
 935
 936	mwait_idle_with_hints(eax, ecx);
 937
 938	if (!static_cpu_has(X86_FEATURE_ARAT) && tick)
 939		tick_broadcast_exit();
 940
 941	return index;
 942}
 943
 944/**
 945 * intel_idle_s2idle - simplified "enter" callback routine for suspend-to-idle
 946 * @dev: cpuidle_device
 947 * @drv: cpuidle driver
 948 * @index: state index
 949 */
 950static void intel_idle_s2idle(struct cpuidle_device *dev,
 951			     struct cpuidle_driver *drv, int index)
 952{
 953	unsigned long ecx = 1; /* break on interrupt flag */
 954	unsigned long eax = flg2MWAIT(drv->states[index].flags);
 955
 956	mwait_idle_with_hints(eax, ecx);
 957}
 958
 959static void __setup_broadcast_timer(bool on)
 960{
 961	if (on)
 962		tick_broadcast_enable();
 963	else
 964		tick_broadcast_disable();
 965}
 966
 967static void auto_demotion_disable(void)
 968{
 969	unsigned long long msr_bits;
 970
 971	rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
 972	msr_bits &= ~(icpu->auto_demotion_disable_flags);
 973	wrmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr_bits);
 974}
 975static void c1e_promotion_disable(void)
 
 
 976{
 977	unsigned long long msr_bits;
 978
 979	rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
 980	msr_bits &= ~0x2;
 981	wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
 
 
 
 
 982}
 983
 984static const struct idle_cpu idle_cpu_nehalem = {
 985	.state_table = nehalem_cstates,
 986	.auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
 987	.disable_promotion_to_c1e = true,
 988};
 989
 990static const struct idle_cpu idle_cpu_atom = {
 991	.state_table = atom_cstates,
 992};
 993
 994static const struct idle_cpu idle_cpu_tangier = {
 995	.state_table = tangier_cstates,
 996};
 997
 998static const struct idle_cpu idle_cpu_lincroft = {
 999	.state_table = atom_cstates,
1000	.auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
1001};
1002
1003static const struct idle_cpu idle_cpu_snb = {
1004	.state_table = snb_cstates,
1005	.disable_promotion_to_c1e = true,
1006};
1007
1008static const struct idle_cpu idle_cpu_byt = {
1009	.state_table = byt_cstates,
1010	.disable_promotion_to_c1e = true,
1011	.byt_auto_demotion_disable_flag = true,
1012};
1013
1014static const struct idle_cpu idle_cpu_cht = {
1015	.state_table = cht_cstates,
1016	.disable_promotion_to_c1e = true,
1017	.byt_auto_demotion_disable_flag = true,
1018};
1019
1020static const struct idle_cpu idle_cpu_ivb = {
1021	.state_table = ivb_cstates,
1022	.disable_promotion_to_c1e = true,
1023};
1024
1025static const struct idle_cpu idle_cpu_ivt = {
1026	.state_table = ivt_cstates,
1027	.disable_promotion_to_c1e = true,
1028};
1029
1030static const struct idle_cpu idle_cpu_hsw = {
1031	.state_table = hsw_cstates,
1032	.disable_promotion_to_c1e = true,
1033};
1034
1035static const struct idle_cpu idle_cpu_bdw = {
1036	.state_table = bdw_cstates,
1037	.disable_promotion_to_c1e = true,
1038};
1039
1040static const struct idle_cpu idle_cpu_skl = {
1041	.state_table = skl_cstates,
1042	.disable_promotion_to_c1e = true,
1043};
1044
1045static const struct idle_cpu idle_cpu_skx = {
1046	.state_table = skx_cstates,
1047	.disable_promotion_to_c1e = true,
1048};
1049
1050static const struct idle_cpu idle_cpu_avn = {
1051	.state_table = avn_cstates,
1052	.disable_promotion_to_c1e = true,
1053};
1054
1055static const struct idle_cpu idle_cpu_knl = {
1056	.state_table = knl_cstates,
1057};
1058
1059static const struct idle_cpu idle_cpu_bxt = {
1060	.state_table = bxt_cstates,
1061	.disable_promotion_to_c1e = true,
1062};
1063
1064static const struct idle_cpu idle_cpu_dnv = {
1065	.state_table = dnv_cstates,
1066	.disable_promotion_to_c1e = true,
1067};
1068
1069#define ICPU(model, cpu) \
1070	{ X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY, (unsigned long)&cpu }
1071
1072static const struct x86_cpu_id intel_idle_ids[] __initconst = {
1073	ICPU(INTEL_FAM6_NEHALEM_EP,		idle_cpu_nehalem),
1074	ICPU(INTEL_FAM6_NEHALEM,		idle_cpu_nehalem),
1075	ICPU(INTEL_FAM6_NEHALEM_G,		idle_cpu_nehalem),
1076	ICPU(INTEL_FAM6_WESTMERE,		idle_cpu_nehalem),
1077	ICPU(INTEL_FAM6_WESTMERE_EP,		idle_cpu_nehalem),
1078	ICPU(INTEL_FAM6_NEHALEM_EX,		idle_cpu_nehalem),
1079	ICPU(INTEL_FAM6_ATOM_PINEVIEW,		idle_cpu_atom),
1080	ICPU(INTEL_FAM6_ATOM_LINCROFT,		idle_cpu_lincroft),
1081	ICPU(INTEL_FAM6_WESTMERE_EX,		idle_cpu_nehalem),
1082	ICPU(INTEL_FAM6_SANDYBRIDGE,		idle_cpu_snb),
1083	ICPU(INTEL_FAM6_SANDYBRIDGE_X,		idle_cpu_snb),
1084	ICPU(INTEL_FAM6_ATOM_CEDARVIEW,		idle_cpu_atom),
1085	ICPU(INTEL_FAM6_ATOM_SILVERMONT1,	idle_cpu_byt),
1086	ICPU(INTEL_FAM6_ATOM_MERRIFIELD,	idle_cpu_tangier),
1087	ICPU(INTEL_FAM6_ATOM_AIRMONT,		idle_cpu_cht),
1088	ICPU(INTEL_FAM6_IVYBRIDGE,		idle_cpu_ivb),
1089	ICPU(INTEL_FAM6_IVYBRIDGE_X,		idle_cpu_ivt),
1090	ICPU(INTEL_FAM6_HASWELL_CORE,		idle_cpu_hsw),
1091	ICPU(INTEL_FAM6_HASWELL_X,		idle_cpu_hsw),
1092	ICPU(INTEL_FAM6_HASWELL_ULT,		idle_cpu_hsw),
1093	ICPU(INTEL_FAM6_HASWELL_GT3E,		idle_cpu_hsw),
1094	ICPU(INTEL_FAM6_ATOM_SILVERMONT2,	idle_cpu_avn),
1095	ICPU(INTEL_FAM6_BROADWELL_CORE,		idle_cpu_bdw),
1096	ICPU(INTEL_FAM6_BROADWELL_GT3E,		idle_cpu_bdw),
1097	ICPU(INTEL_FAM6_BROADWELL_X,		idle_cpu_bdw),
1098	ICPU(INTEL_FAM6_BROADWELL_XEON_D,	idle_cpu_bdw),
1099	ICPU(INTEL_FAM6_SKYLAKE_MOBILE,		idle_cpu_skl),
1100	ICPU(INTEL_FAM6_SKYLAKE_DESKTOP,	idle_cpu_skl),
1101	ICPU(INTEL_FAM6_KABYLAKE_MOBILE,	idle_cpu_skl),
1102	ICPU(INTEL_FAM6_KABYLAKE_DESKTOP,	idle_cpu_skl),
1103	ICPU(INTEL_FAM6_SKYLAKE_X,		idle_cpu_skx),
1104	ICPU(INTEL_FAM6_XEON_PHI_KNL,		idle_cpu_knl),
1105	ICPU(INTEL_FAM6_XEON_PHI_KNM,		idle_cpu_knl),
1106	ICPU(INTEL_FAM6_ATOM_GOLDMONT,		idle_cpu_bxt),
1107	ICPU(INTEL_FAM6_ATOM_GEMINI_LAKE,	idle_cpu_bxt),
1108	ICPU(INTEL_FAM6_ATOM_DENVERTON,		idle_cpu_dnv),
1109	{}
1110};
1111
1112/*
1113 * intel_idle_probe()
1114 */
1115static int __init intel_idle_probe(void)
1116{
1117	unsigned int eax, ebx, ecx;
1118	const struct x86_cpu_id *id;
1119
1120	if (max_cstate == 0) {
1121		pr_debug("disabled\n");
1122		return -EPERM;
1123	}
1124
1125	id = x86_match_cpu(intel_idle_ids);
1126	if (!id) {
1127		if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
1128		    boot_cpu_data.x86 == 6)
1129			pr_debug("does not run on family %d model %d\n",
1130				 boot_cpu_data.x86, boot_cpu_data.x86_model);
1131		return -ENODEV;
1132	}
1133
1134	if (!boot_cpu_has(X86_FEATURE_MWAIT)) {
1135		pr_debug("Please enable MWAIT in BIOS SETUP\n");
1136		return -ENODEV;
1137	}
1138
1139	if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
1140		return -ENODEV;
1141
1142	cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
1143
1144	if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
1145	    !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
1146	    !mwait_substates)
1147			return -ENODEV;
1148
1149	pr_debug("MWAIT substates: 0x%x\n", mwait_substates);
 
 
 
 
1150
1151	icpu = (const struct idle_cpu *)id->driver_data;
1152	cpuidle_state_table = icpu->state_table;
1153
1154	pr_debug("v" INTEL_IDLE_VERSION " model 0x%X\n",
1155		 boot_cpu_data.x86_model);
 
 
 
 
 
 
 
 
 
 
 
 
 
1156
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1157	return 0;
1158}
1159
1160/*
1161 * intel_idle_cpuidle_devices_uninit()
1162 * Unregisters the cpuidle devices.
1163 */
1164static void intel_idle_cpuidle_devices_uninit(void)
1165{
1166	int i;
1167	struct cpuidle_device *dev;
1168
1169	for_each_online_cpu(i) {
1170		dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
1171		cpuidle_unregister_device(dev);
1172	}
1173}
1174
1175/*
1176 * ivt_idle_state_table_update(void)
1177 *
1178 * Tune IVT multi-socket targets
1179 * Assumption: num_sockets == (max_package_num + 1)
1180 */
1181static void ivt_idle_state_table_update(void)
1182{
1183	/* IVT uses a different table for 1-2, 3-4, and > 4 sockets */
1184	int cpu, package_num, num_sockets = 1;
1185
1186	for_each_online_cpu(cpu) {
1187		package_num = topology_physical_package_id(cpu);
1188		if (package_num + 1 > num_sockets) {
1189			num_sockets = package_num + 1;
1190
1191			if (num_sockets > 4) {
1192				cpuidle_state_table = ivt_cstates_8s;
1193				return;
1194			}
1195		}
1196	}
1197
1198	if (num_sockets > 2)
1199		cpuidle_state_table = ivt_cstates_4s;
1200
1201	/* else, 1 and 2 socket systems use default ivt_cstates */
1202}
1203
1204/*
1205 * Translate IRTL (Interrupt Response Time Limit) MSR to usec
 
1206 */
1207
1208static unsigned int irtl_ns_units[] = {
1209	1, 32, 1024, 32768, 1048576, 33554432, 0, 0 };
1210
1211static unsigned long long irtl_2_usec(unsigned long long irtl)
1212{
1213	unsigned long long ns;
 
1214
1215	if (!irtl)
1216		return 0;
 
1217
1218	ns = irtl_ns_units[(irtl >> 10) & 0x7];
 
1219
1220	return div64_u64((irtl & 0x3FF) * ns, 1000);
1221}
1222/*
1223 * bxt_idle_state_table_update(void)
1224 *
1225 * On BXT, we trust the IRTL to show the definitive maximum latency
1226 * We use the same value for target_residency.
1227 */
1228static void bxt_idle_state_table_update(void)
1229{
1230	unsigned long long msr;
1231	unsigned int usec;
1232
1233	rdmsrl(MSR_PKGC6_IRTL, msr);
1234	usec = irtl_2_usec(msr);
1235	if (usec) {
1236		bxt_cstates[2].exit_latency = usec;
1237		bxt_cstates[2].target_residency = usec;
1238	}
1239
1240	rdmsrl(MSR_PKGC7_IRTL, msr);
1241	usec = irtl_2_usec(msr);
1242	if (usec) {
1243		bxt_cstates[3].exit_latency = usec;
1244		bxt_cstates[3].target_residency = usec;
1245	}
1246
1247	rdmsrl(MSR_PKGC8_IRTL, msr);
1248	usec = irtl_2_usec(msr);
1249	if (usec) {
1250		bxt_cstates[4].exit_latency = usec;
1251		bxt_cstates[4].target_residency = usec;
1252	}
 
 
 
 
 
 
 
 
 
1253
1254	rdmsrl(MSR_PKGC9_IRTL, msr);
1255	usec = irtl_2_usec(msr);
1256	if (usec) {
1257		bxt_cstates[5].exit_latency = usec;
1258		bxt_cstates[5].target_residency = usec;
1259	}
1260
1261	rdmsrl(MSR_PKGC10_IRTL, msr);
1262	usec = irtl_2_usec(msr);
1263	if (usec) {
1264		bxt_cstates[6].exit_latency = usec;
1265		bxt_cstates[6].target_residency = usec;
1266	}
1267
1268}
1269/*
1270 * sklh_idle_state_table_update(void)
1271 *
1272 * On SKL-H (model 0x5e) disable C8 and C9 if:
1273 * C10 is enabled and SGX disabled
1274 */
1275static void sklh_idle_state_table_update(void)
1276{
1277	unsigned long long msr;
1278	unsigned int eax, ebx, ecx, edx;
1279
1280
1281	/* if PC10 disabled via cmdline intel_idle.max_cstate=7 or shallower */
1282	if (max_cstate <= 7)
1283		return;
1284
1285	/* if PC10 not present in CPUID.MWAIT.EDX */
1286	if ((mwait_substates & (0xF << 28)) == 0)
1287		return;
1288
1289	rdmsrl(MSR_PKG_CST_CONFIG_CONTROL, msr);
1290
1291	/* PC10 is not enabled in PKG C-state limit */
1292	if ((msr & 0xF) != 8)
1293		return;
1294
1295	ecx = 0;
1296	cpuid(7, &eax, &ebx, &ecx, &edx);
1297
1298	/* if SGX is present */
1299	if (ebx & (1 << 2)) {
1300
1301		rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1302
1303		/* if SGX is enabled */
1304		if (msr & (1 << 18))
1305			return;
1306	}
1307
1308	skl_cstates[5].disabled = 1;	/* C8-SKL */
1309	skl_cstates[6].disabled = 1;	/* C9-SKL */
1310}
1311/*
1312 * intel_idle_state_table_update()
1313 *
1314 * Update the default state_table for this CPU-id
1315 */
1316
1317static void intel_idle_state_table_update(void)
1318{
1319	switch (boot_cpu_data.x86_model) {
1320
1321	case INTEL_FAM6_IVYBRIDGE_X:
1322		ivt_idle_state_table_update();
1323		break;
1324	case INTEL_FAM6_ATOM_GOLDMONT:
1325	case INTEL_FAM6_ATOM_GEMINI_LAKE:
1326		bxt_idle_state_table_update();
1327		break;
1328	case INTEL_FAM6_SKYLAKE_DESKTOP:
1329		sklh_idle_state_table_update();
1330		break;
1331	}
1332}
1333
1334/*
1335 * intel_idle_cpuidle_driver_init()
1336 * allocate, initialize cpuidle_states
1337 */
1338static void __init intel_idle_cpuidle_driver_init(void)
1339{
1340	int cstate;
1341	struct cpuidle_driver *drv = &intel_idle_driver;
1342
1343	intel_idle_state_table_update();
1344
1345	cpuidle_poll_state_init(drv);
1346	drv->state_count = 1;
1347
1348	for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
1349		int num_substates, mwait_hint, mwait_cstate;
1350
1351		if ((cpuidle_state_table[cstate].enter == NULL) &&
1352		    (cpuidle_state_table[cstate].enter_s2idle == NULL))
1353			break;
1354
1355		if (cstate + 1 > max_cstate) {
1356			pr_info("max_cstate %d reached\n", max_cstate);
1357			break;
1358		}
1359
1360		mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
1361		mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint);
1362
1363		/* number of sub-states for this state in CPUID.MWAIT */
1364		num_substates = (mwait_substates >> ((mwait_cstate + 1) * 4))
1365					& MWAIT_SUBSTATE_MASK;
1366
1367		/* if NO sub-states for this state in CPUID, skip it */
1368		if (num_substates == 0)
1369			continue;
1370
1371		/* if state marked as disabled, skip it */
1372		if (cpuidle_state_table[cstate].disabled != 0) {
1373			pr_debug("state %s is disabled\n",
1374				 cpuidle_state_table[cstate].name);
1375			continue;
1376		}
1377
1378
1379		if (((mwait_cstate + 1) > 2) &&
1380			!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
1381			mark_tsc_unstable("TSC halts in idle"
1382					" states deeper than C2");
1383
1384		drv->states[drv->state_count] =	/* structure copy */
1385			cpuidle_state_table[cstate];
1386
1387		drv->state_count += 1;
1388	}
1389
1390	if (icpu->byt_auto_demotion_disable_flag) {
1391		wrmsrl(MSR_CC6_DEMOTION_POLICY_CONFIG, 0);
1392		wrmsrl(MSR_MC6_DEMOTION_POLICY_CONFIG, 0);
1393	}
1394}
1395
1396
1397/*
1398 * intel_idle_cpu_init()
1399 * allocate, initialize, register cpuidle_devices
1400 * @cpu: cpu/core to initialize
1401 */
1402static int intel_idle_cpu_init(unsigned int cpu)
1403{
1404	struct cpuidle_device *dev;
1405
1406	dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
1407	dev->cpu = cpu;
1408
1409	if (cpuidle_register_device(dev)) {
1410		pr_debug("cpuidle_register_device %d failed!\n", cpu);
1411		return -EIO;
1412	}
1413
1414	if (icpu->auto_demotion_disable_flags)
1415		auto_demotion_disable();
1416
1417	if (icpu->disable_promotion_to_c1e)
1418		c1e_promotion_disable();
1419
1420	return 0;
1421}
1422
1423static int intel_idle_cpu_online(unsigned int cpu)
1424{
1425	struct cpuidle_device *dev;
1426
1427	if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE)
1428		__setup_broadcast_timer(true);
1429
1430	/*
1431	 * Some systems can hotplug a cpu at runtime after
1432	 * the kernel has booted, we have to initialize the
1433	 * driver in this case
1434	 */
1435	dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
1436	if (!dev->registered)
1437		return intel_idle_cpu_init(cpu);
1438
1439	return 0;
1440}
1441
1442static int __init intel_idle_init(void)
1443{
1444	int retval;
1445
1446	/* Do not load intel_idle at all for now if idle= is passed */
1447	if (boot_option_idle_override != IDLE_NO_OVERRIDE)
1448		return -ENODEV;
1449
1450	retval = intel_idle_probe();
1451	if (retval)
1452		return retval;
1453
1454	intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
1455	if (intel_idle_cpuidle_devices == NULL)
1456		return -ENOMEM;
1457
1458	intel_idle_cpuidle_driver_init();
1459	retval = cpuidle_register_driver(&intel_idle_driver);
1460	if (retval) {
1461		struct cpuidle_driver *drv = cpuidle_get_driver();
1462		printk(KERN_DEBUG pr_fmt("intel_idle yielding to %s\n"),
1463		       drv ? drv->name : "none");
1464		goto init_driver_fail;
1465	}
1466
1467	if (boot_cpu_has(X86_FEATURE_ARAT))	/* Always Reliable APIC Timer */
1468		lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
1469
1470	retval = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "idle/intel:online",
1471				   intel_idle_cpu_online, NULL);
1472	if (retval < 0)
1473		goto hp_setup_fail;
1474
1475	pr_debug("lapic_timer_reliable_states 0x%x\n",
1476		 lapic_timer_reliable_states);
1477
1478	return 0;
 
1479
1480hp_setup_fail:
 
1481	intel_idle_cpuidle_devices_uninit();
1482	cpuidle_unregister_driver(&intel_idle_driver);
1483init_driver_fail:
1484	free_percpu(intel_idle_cpuidle_devices);
1485	return retval;
1486
 
 
 
 
 
 
1487}
1488device_initcall(intel_idle_init);
1489
1490/*
1491 * We are not really modular, but we used to support that.  Meaning we also
1492 * support "intel_idle.max_cstate=..." at boot and also a read-only export of
1493 * it at /sys/module/intel_idle/parameters/max_cstate -- so using module_param
1494 * is the easiest way (currently) to continue doing that.
1495 */
1496module_param(max_cstate, int, 0444);