blob: fa6964d8681a0d126fcf7c4845896b3f06298f8f [file] [log] [blame]
Len Brown26717172010-03-08 14:07:30 -05001/*
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>
Len Brown26717172010-03-08 14:07:30 -050059#include <trace/events/power.h>
60#include <linux/sched.h>
Shaohua Li2a2d31c2011-01-10 09:38:12 +080061#include <linux/notifier.h>
62#include <linux/cpu.h>
Paul Gortmaker7c52d552011-05-27 12:33:10 -040063#include <linux/module.h>
Andi Kleenb66b8b92012-01-26 00:09:07 +010064#include <asm/cpu_device_id.h>
H. Peter Anvinbc83ccc2010-09-17 15:36:40 -070065#include <asm/mwait.h>
Len Brown14796fc2011-01-18 20:48:27 -050066#include <asm/msr.h>
Len Brown26717172010-03-08 14:07:30 -050067
68#define INTEL_IDLE_VERSION "0.4"
69#define PREFIX "intel_idle: "
70
Len Brown26717172010-03-08 14:07:30 -050071static struct cpuidle_driver intel_idle_driver = {
72 .name = "intel_idle",
73 .owner = THIS_MODULE,
74};
75/* intel_idle.max_cstate=0 disables driver */
Len Brown137ecc72013-02-01 21:35:35 -050076static int max_cstate = CPUIDLE_STATE_MAX - 1;
Len Brown26717172010-03-08 14:07:30 -050077
Len Brownc4236282010-05-28 02:22:03 -040078static unsigned int mwait_substates;
Len Brown26717172010-03-08 14:07:30 -050079
Shaohua Li2a2d31c2011-01-10 09:38:12 +080080#define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF
Len Brown26717172010-03-08 14:07:30 -050081/* Reliable LAPIC Timer States, bit 1 for C1 etc. */
Len Brownd13780d2010-07-07 00:12:03 -040082static unsigned int lapic_timer_reliable_states = (1 << 1); /* Default to only C1 */
Len Brown26717172010-03-08 14:07:30 -050083
Andi Kleenb66b8b92012-01-26 00:09:07 +010084struct 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;
Len Brown32e95182013-02-02 01:31:56 -050092 bool disable_promotion_to_c1e;
Andi Kleenb66b8b92012-01-26 00:09:07 +010093};
94
95static const struct idle_cpu *icpu;
Namhyung Kim3265eba2010-08-08 03:10:03 +090096static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +053097static int intel_idle(struct cpuidle_device *dev,
98 struct cpuidle_driver *drv, int index);
Daniel Lezcano25ac7762012-07-05 15:23:25 +020099static int intel_idle_cpu_init(int cpu);
Len Brown26717172010-03-08 14:07:30 -0500100
101static struct cpuidle_state *cpuidle_state_table;
102
103/*
Len Brown956d0332011-01-12 02:51:20 -0500104 * 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/*
Len Brownb1beab42013-01-31 19:55:37 -0500112 * 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/*
Len Brown26717172010-03-08 14:07:30 -0500122 * 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 */
Len Brown137ecc72013-02-01 21:35:35 -0500126static struct cpuidle_state nehalem_cstates[CPUIDLE_STATE_MAX] = {
Len Browne022e7e2013-02-01 23:37:30 -0500127 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100128 .name = "C1-NHM",
Len Brown26717172010-03-08 14:07:30 -0500129 .desc = "MWAIT 0x00",
Len Brownb1beab42013-01-31 19:55:37 -0500130 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
Len Brown26717172010-03-08 14:07:30 -0500131 .exit_latency = 3,
Len Brown26717172010-03-08 14:07:30 -0500132 .target_residency = 6,
133 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500134 {
Len Brown32e95182013-02-02 01:31:56 -0500135 .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 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100142 .name = "C3-NHM",
Len Brown26717172010-03-08 14:07:30 -0500143 .desc = "MWAIT 0x10",
Len Brownb1beab42013-01-31 19:55:37 -0500144 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown26717172010-03-08 14:07:30 -0500145 .exit_latency = 20,
Len Brown26717172010-03-08 14:07:30 -0500146 .target_residency = 80,
147 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500148 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100149 .name = "C6-NHM",
Len Brown26717172010-03-08 14:07:30 -0500150 .desc = "MWAIT 0x20",
Len Brownb1beab42013-01-31 19:55:37 -0500151 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown26717172010-03-08 14:07:30 -0500152 .exit_latency = 200,
Len Brown26717172010-03-08 14:07:30 -0500153 .target_residency = 800,
154 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500155 {
156 .enter = NULL }
Len Brown26717172010-03-08 14:07:30 -0500157};
158
Len Brown137ecc72013-02-01 21:35:35 -0500159static struct cpuidle_state snb_cstates[CPUIDLE_STATE_MAX] = {
Len Browne022e7e2013-02-01 23:37:30 -0500160 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100161 .name = "C1-SNB",
Len Brownd13780d2010-07-07 00:12:03 -0400162 .desc = "MWAIT 0x00",
Len Brownb1beab42013-01-31 19:55:37 -0500163 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
Len Brown32e95182013-02-02 01:31:56 -0500164 .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,
Len Brownd13780d2010-07-07 00:12:03 -0400173 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500174 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100175 .name = "C3-SNB",
Len Brownd13780d2010-07-07 00:12:03 -0400176 .desc = "MWAIT 0x10",
Len Brownb1beab42013-01-31 19:55:37 -0500177 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brownd13780d2010-07-07 00:12:03 -0400178 .exit_latency = 80,
Len Brownddbd5502010-12-13 18:28:22 -0500179 .target_residency = 211,
Len Brownd13780d2010-07-07 00:12:03 -0400180 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500181 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100182 .name = "C6-SNB",
Len Brownd13780d2010-07-07 00:12:03 -0400183 .desc = "MWAIT 0x20",
Len Brownb1beab42013-01-31 19:55:37 -0500184 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brownd13780d2010-07-07 00:12:03 -0400185 .exit_latency = 104,
Len Brownddbd5502010-12-13 18:28:22 -0500186 .target_residency = 345,
Len Brownd13780d2010-07-07 00:12:03 -0400187 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500188 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100189 .name = "C7-SNB",
Len Brownd13780d2010-07-07 00:12:03 -0400190 .desc = "MWAIT 0x30",
Len Brownb1beab42013-01-31 19:55:37 -0500191 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brownd13780d2010-07-07 00:12:03 -0400192 .exit_latency = 109,
Len Brownddbd5502010-12-13 18:28:22 -0500193 .target_residency = 345,
Len Brownd13780d2010-07-07 00:12:03 -0400194 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500195 {
196 .enter = NULL }
Len Brownd13780d2010-07-07 00:12:03 -0400197};
198
Len Brown137ecc72013-02-01 21:35:35 -0500199static struct cpuidle_state ivb_cstates[CPUIDLE_STATE_MAX] = {
Len Browne022e7e2013-02-01 23:37:30 -0500200 {
Len Brown6edab082012-06-01 19:45:32 -0400201 .name = "C1-IVB",
202 .desc = "MWAIT 0x00",
Len Brownb1beab42013-01-31 19:55:37 -0500203 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
Len Brown6edab082012-06-01 19:45:32 -0400204 .exit_latency = 1,
205 .target_residency = 1,
206 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500207 {
Len Brown32e95182013-02-02 01:31:56 -0500208 .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 {
Len Brown6edab082012-06-01 19:45:32 -0400215 .name = "C3-IVB",
216 .desc = "MWAIT 0x10",
Len Brownb1beab42013-01-31 19:55:37 -0500217 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown6edab082012-06-01 19:45:32 -0400218 .exit_latency = 59,
219 .target_residency = 156,
220 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500221 {
Len Brown6edab082012-06-01 19:45:32 -0400222 .name = "C6-IVB",
223 .desc = "MWAIT 0x20",
Len Brownb1beab42013-01-31 19:55:37 -0500224 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown6edab082012-06-01 19:45:32 -0400225 .exit_latency = 80,
226 .target_residency = 300,
227 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500228 {
Len Brown6edab082012-06-01 19:45:32 -0400229 .name = "C7-IVB",
230 .desc = "MWAIT 0x30",
Len Brownb1beab42013-01-31 19:55:37 -0500231 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown6edab082012-06-01 19:45:32 -0400232 .exit_latency = 87,
233 .target_residency = 300,
234 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500235 {
236 .enter = NULL }
Len Brown6edab082012-06-01 19:45:32 -0400237};
238
Len Brown137ecc72013-02-01 21:35:35 -0500239static struct cpuidle_state hsw_cstates[CPUIDLE_STATE_MAX] = {
Len Browne022e7e2013-02-01 23:37:30 -0500240 {
Len Brown85a4d2d2013-01-31 14:40:49 -0500241 .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 },
Len Browne022e7e2013-02-01 23:37:30 -0500247 {
Len Brown32e95182013-02-02 01:31:56 -0500248 .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 {
Len Brown85a4d2d2013-01-31 14:40:49 -0500255 .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 },
Len Browne022e7e2013-02-01 23:37:30 -0500261 {
Len Brown85a4d2d2013-01-31 14:40:49 -0500262 .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 },
Len Browne022e7e2013-02-01 23:37:30 -0500268 {
Len Brown85a4d2d2013-01-31 14:40:49 -0500269 .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 },
Len Browne022e7e2013-02-01 23:37:30 -0500275 {
Len Brown86239ce2013-02-27 13:18:50 -0500276 .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 {
Len Browne022e7e2013-02-01 23:37:30 -0500297 .enter = NULL }
Len Brown85a4d2d2013-01-31 14:40:49 -0500298};
299
Len Brown137ecc72013-02-01 21:35:35 -0500300static struct cpuidle_state atom_cstates[CPUIDLE_STATE_MAX] = {
Len Browne022e7e2013-02-01 23:37:30 -0500301 {
Len Brown32e95182013-02-02 01:31:56 -0500302 .name = "C1E-ATM",
Len Brown26717172010-03-08 14:07:30 -0500303 .desc = "MWAIT 0x00",
Len Brownb1beab42013-01-31 19:55:37 -0500304 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
Len Brown32e95182013-02-02 01:31:56 -0500305 .exit_latency = 10,
306 .target_residency = 20,
Len Brown26717172010-03-08 14:07:30 -0500307 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500308 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100309 .name = "C2-ATM",
Len Brown26717172010-03-08 14:07:30 -0500310 .desc = "MWAIT 0x10",
Len Brownb1beab42013-01-31 19:55:37 -0500311 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID,
Len Brown26717172010-03-08 14:07:30 -0500312 .exit_latency = 20,
Len Brown26717172010-03-08 14:07:30 -0500313 .target_residency = 80,
314 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500315 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100316 .name = "C4-ATM",
Len Brown26717172010-03-08 14:07:30 -0500317 .desc = "MWAIT 0x30",
Len Brownb1beab42013-01-31 19:55:37 -0500318 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown26717172010-03-08 14:07:30 -0500319 .exit_latency = 100,
Len Brown26717172010-03-08 14:07:30 -0500320 .target_residency = 400,
321 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500322 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100323 .name = "C6-ATM",
Len Brown7fcca7d2010-10-05 13:43:14 -0400324 .desc = "MWAIT 0x52",
Len Brownb1beab42013-01-31 19:55:37 -0500325 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown7fcca7d2010-10-05 13:43:14 -0400326 .exit_latency = 140,
Len Brown7fcca7d2010-10-05 13:43:14 -0400327 .target_residency = 560,
328 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500329 {
330 .enter = NULL }
Len Brown26717172010-03-08 14:07:30 -0500331};
332
Len Brown26717172010-03-08 14:07:30 -0500333/**
334 * intel_idle
335 * @dev: cpuidle_device
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530336 * @drv: cpuidle driver
Deepthi Dharware978aa72011-10-28 16:20:09 +0530337 * @index: index of cpuidle state
Len Brown26717172010-03-08 14:07:30 -0500338 *
Yanmin Zhang63ff07b2012-01-10 15:48:21 -0800339 * Must be called under local_irq_disable().
Len Brown26717172010-03-08 14:07:30 -0500340 */
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530341static int intel_idle(struct cpuidle_device *dev,
342 struct cpuidle_driver *drv, int index)
Len Brown26717172010-03-08 14:07:30 -0500343{
344 unsigned long ecx = 1; /* break on interrupt flag */
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530345 struct cpuidle_state *state = &drv->states[index];
Len Brownb1beab42013-01-31 19:55:37 -0500346 unsigned long eax = flg2MWAIT(state->flags);
Len Brown26717172010-03-08 14:07:30 -0500347 unsigned int cstate;
Len Brown26717172010-03-08 14:07:30 -0500348 int cpu = smp_processor_id();
349
350 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
351
Suresh Siddha6110a1f2010-09-30 21:19:07 -0400352 /*
Len Brownc8381cc2010-10-15 20:43:06 -0400353 * leave_mm() to avoid costly and often unnecessary wakeups
354 * for flushing the user TLB's associated with the active mm.
Suresh Siddha6110a1f2010-09-30 21:19:07 -0400355 */
Len Brownc8381cc2010-10-15 20:43:06 -0400356 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
Suresh Siddha6110a1f2010-09-30 21:19:07 -0400357 leave_mm(cpu);
358
Len Brown26717172010-03-08 14:07:30 -0500359 if (!(lapic_timer_reliable_states & (1 << (cstate))))
360 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
361
Len Brown26717172010-03-08 14:07:30 -0500362 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
Len Brown26717172010-03-08 14:07:30 -0500370 if (!(lapic_timer_reliable_states & (1 << (cstate))))
371 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
372
Deepthi Dharware978aa72011-10-28 16:20:09 +0530373 return index;
Len Brown26717172010-03-08 14:07:30 -0500374}
375
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800376static 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
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200387static int cpu_hotplug_notify(struct notifier_block *n,
388 unsigned long action, void *hcpu)
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800389{
390 int hotcpu = (unsigned long)hcpu;
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200391 struct cpuidle_device *dev;
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800392
393 switch (action & 0xf) {
394 case CPU_ONLINE:
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200395
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
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800409 break;
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800410 }
411 return NOTIFY_OK;
412}
413
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200414static struct notifier_block cpu_hotplug_notifier = {
415 .notifier_call = cpu_hotplug_notify,
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800416};
417
Len Brown14796fc2011-01-18 20:48:27 -0500418static 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);
Andi Kleenb66b8b92012-01-26 00:09:07 +0100423 msr_bits &= ~(icpu->auto_demotion_disable_flags);
Len Brown14796fc2011-01-18 20:48:27 -0500424 wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
425}
Len Brown32e95182013-02-02 01:31:56 -0500426static 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}
Len Brown14796fc2011-01-18 20:48:27 -0500434
Andi Kleenb66b8b92012-01-26 00:09:07 +0100435static const struct idle_cpu idle_cpu_nehalem = {
436 .state_table = nehalem_cstates,
Andi Kleenb66b8b92012-01-26 00:09:07 +0100437 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
Len Brown32e95182013-02-02 01:31:56 -0500438 .disable_promotion_to_c1e = true,
Andi Kleenb66b8b92012-01-26 00:09:07 +0100439};
440
441static const struct idle_cpu idle_cpu_atom = {
442 .state_table = atom_cstates,
443};
444
445static const struct idle_cpu idle_cpu_lincroft = {
446 .state_table = atom_cstates,
447 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
448};
449
450static const struct idle_cpu idle_cpu_snb = {
451 .state_table = snb_cstates,
Len Brown32e95182013-02-02 01:31:56 -0500452 .disable_promotion_to_c1e = true,
Andi Kleenb66b8b92012-01-26 00:09:07 +0100453};
454
Len Brown6edab082012-06-01 19:45:32 -0400455static const struct idle_cpu idle_cpu_ivb = {
456 .state_table = ivb_cstates,
Len Brown32e95182013-02-02 01:31:56 -0500457 .disable_promotion_to_c1e = true,
Len Brown6edab082012-06-01 19:45:32 -0400458};
459
Len Brown85a4d2d2013-01-31 14:40:49 -0500460static const struct idle_cpu idle_cpu_hsw = {
461 .state_table = hsw_cstates,
Len Brown32e95182013-02-02 01:31:56 -0500462 .disable_promotion_to_c1e = true,
Len Brown85a4d2d2013-01-31 14:40:49 -0500463};
464
Andi Kleenb66b8b92012-01-26 00:09:07 +0100465#define ICPU(model, cpu) \
466 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_MWAIT, (unsigned long)&cpu }
467
468static 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),
Ben Hutchings8bf11932012-02-16 04:13:14 +0000472 ICPU(0x25, idle_cpu_nehalem),
473 ICPU(0x2c, idle_cpu_nehalem),
474 ICPU(0x2e, idle_cpu_nehalem),
Andi Kleenb66b8b92012-01-26 00:09:07 +0100475 ICPU(0x1c, idle_cpu_atom),
476 ICPU(0x26, idle_cpu_lincroft),
Ben Hutchings8bf11932012-02-16 04:13:14 +0000477 ICPU(0x2f, idle_cpu_nehalem),
Andi Kleenb66b8b92012-01-26 00:09:07 +0100478 ICPU(0x2a, idle_cpu_snb),
479 ICPU(0x2d, idle_cpu_snb),
Len Brown6edab082012-06-01 19:45:32 -0400480 ICPU(0x3a, idle_cpu_ivb),
Len Brown23795e52012-09-26 22:28:21 -0400481 ICPU(0x3e, idle_cpu_ivb),
Len Brown85a4d2d2013-01-31 14:40:49 -0500482 ICPU(0x3c, idle_cpu_hsw),
483 ICPU(0x3f, idle_cpu_hsw),
484 ICPU(0x45, idle_cpu_hsw),
Len Brown0b158412013-03-15 10:55:31 -0400485 ICPU(0x46, idle_cpu_hsw),
Andi Kleenb66b8b92012-01-26 00:09:07 +0100486 {}
487};
488MODULE_DEVICE_TABLE(x86cpu, intel_idle_ids);
489
Len Brown26717172010-03-08 14:07:30 -0500490/*
491 * intel_idle_probe()
492 */
493static int intel_idle_probe(void)
494{
Len Brownc4236282010-05-28 02:22:03 -0400495 unsigned int eax, ebx, ecx;
Andi Kleenb66b8b92012-01-26 00:09:07 +0100496 const struct x86_cpu_id *id;
Len Brown26717172010-03-08 14:07:30 -0500497
498 if (max_cstate == 0) {
499 pr_debug(PREFIX "disabled\n");
500 return -EPERM;
501 }
502
Andi Kleenb66b8b92012-01-26 00:09:07 +0100503 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);
Len Brown26717172010-03-08 14:07:30 -0500509 return -ENODEV;
Andi Kleenb66b8b92012-01-26 00:09:07 +0100510 }
Len Brown26717172010-03-08 14:07:30 -0500511
512 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
513 return -ENODEV;
514
Len Brownc4236282010-05-28 02:22:03 -0400515 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
Len Brown26717172010-03-08 14:07:30 -0500516
517 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
Thomas Renninger5c2a9f02011-12-04 22:17:29 +0100518 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
519 !mwait_substates)
Len Brown26717172010-03-08 14:07:30 -0500520 return -ENODEV;
Len Brown26717172010-03-08 14:07:30 -0500521
Len Brownc4236282010-05-28 02:22:03 -0400522 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
Len Brown26717172010-03-08 14:07:30 -0500523
Andi Kleenb66b8b92012-01-26 00:09:07 +0100524 icpu = (const struct idle_cpu *)id->driver_data;
525 cpuidle_state_table = icpu->state_table;
Len Brown26717172010-03-08 14:07:30 -0500526
Len Brown56b9aea2010-12-02 01:19:32 -0500527 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800528 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200529 else
Shaohua Li39a74fd2012-01-10 15:48:19 -0800530 on_each_cpu(__setup_broadcast_timer, (void *)true, 1);
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200531
Len Brown26717172010-03-08 14:07:30 -0500532 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 */
544static 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/*
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530558 * intel_idle_cpuidle_driver_init()
559 * allocate, initialize cpuidle_states
560 */
561static 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
Len Browne022e7e2013-02-01 23:37:30 -0500568 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
569 int num_substates, mwait_hint, mwait_cstate, mwait_substate;
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530570
Len Browne022e7e2013-02-01 23:37:30 -0500571 if (cpuidle_state_table[cstate].enter == NULL)
572 break;
573
574 if (cstate + 1 > max_cstate) {
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530575 printk(PREFIX "max_cstate %d reached\n",
576 max_cstate);
577 break;
578 }
579
Len Browne022e7e2013-02-01 23:37:30 -0500580 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
581 mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint);
582 mwait_substate = MWAIT_HINT2SUBSTATE(mwait_hint);
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530583
Len Browne022e7e2013-02-01 23:37:30 -0500584 /* 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) &&
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530593 !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
Andi Kleenb66b8b92012-01-26 00:09:07 +0100603 if (icpu->auto_demotion_disable_flags)
Shaohua Li39a74fd2012-01-10 15:48:19 -0800604 on_each_cpu(auto_demotion_disable, NULL, 1);
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530605
Len Brown32e95182013-02-02 01:31:56 -0500606 if (icpu->disable_promotion_to_c1e) /* each-cpu is redundant */
607 on_each_cpu(c1e_promotion_disable, NULL, 1);
608
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530609 return 0;
610}
611
612
613/*
Thomas Renninger65b7f832012-01-17 22:40:08 +0100614 * intel_idle_cpu_init()
Len Brown26717172010-03-08 14:07:30 -0500615 * allocate, initialize, register cpuidle_devices
Thomas Renninger65b7f832012-01-17 22:40:08 +0100616 * @cpu: cpu/core to initialize
Len Brown26717172010-03-08 14:07:30 -0500617 */
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200618static int intel_idle_cpu_init(int cpu)
Len Brown26717172010-03-08 14:07:30 -0500619{
Thomas Renninger65b7f832012-01-17 22:40:08 +0100620 int cstate;
Len Brown26717172010-03-08 14:07:30 -0500621 struct cpuidle_device *dev;
622
Thomas Renninger65b7f832012-01-17 22:40:08 +0100623 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
Len Brown26717172010-03-08 14:07:30 -0500624
Thomas Renninger65b7f832012-01-17 22:40:08 +0100625 dev->state_count = 1;
Len Brown26717172010-03-08 14:07:30 -0500626
Len Browne022e7e2013-02-01 23:37:30 -0500627 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
628 int num_substates, mwait_hint, mwait_cstate, mwait_substate;
Len Brown26717172010-03-08 14:07:30 -0500629
Len Browne022e7e2013-02-01 23:37:30 -0500630 if (cpuidle_state_table[cstate].enter == NULL)
631 continue;
632
633 if (cstate + 1 > max_cstate) {
Marcos Paulo de Souzadc716e92012-03-21 16:33:43 -0700634 printk(PREFIX "max_cstate %d reached\n", max_cstate);
Thomas Renninger65b7f832012-01-17 22:40:08 +0100635 break;
636 }
Len Brown26717172010-03-08 14:07:30 -0500637
Len Browne022e7e2013-02-01 23:37:30 -0500638 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
639 mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint);
640 mwait_substate = MWAIT_HINT2SUBSTATE(mwait_hint);
641
Thomas Renninger65b7f832012-01-17 22:40:08 +0100642 /* does the state exist in CPUID.MWAIT? */
Len Browne022e7e2013-02-01 23:37:30 -0500643 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)
Thomas Renninger65b7f832012-01-17 22:40:08 +0100648 continue;
Len Brown26717172010-03-08 14:07:30 -0500649
Marcos Paulo de Souzadc716e92012-03-21 16:33:43 -0700650 dev->state_count += 1;
651 }
652
Thomas Renninger65b7f832012-01-17 22:40:08 +0100653 dev->cpu = cpu;
Len Brown26717172010-03-08 14:07:30 -0500654
Thomas Renninger65b7f832012-01-17 22:40:08 +0100655 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;
Len Brown26717172010-03-08 14:07:30 -0500659 }
660
Andi Kleenb66b8b92012-01-26 00:09:07 +0100661 if (icpu->auto_demotion_disable_flags)
Thomas Renninger65b7f832012-01-17 22:40:08 +0100662 smp_call_function_single(cpu, auto_demotion_disable, NULL, 1);
663
Len Brown26717172010-03-08 14:07:30 -0500664 return 0;
665}
Len Brown26717172010-03-08 14:07:30 -0500666
667static int __init intel_idle_init(void)
668{
Thomas Renninger65b7f832012-01-17 22:40:08 +0100669 int retval, i;
Len Brown26717172010-03-08 14:07:30 -0500670
Thomas Renningerd1896042010-11-03 17:06:14 +0100671 /* 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
Len Brown26717172010-03-08 14:07:30 -0500675 retval = intel_idle_probe();
676 if (retval)
677 return retval;
678
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530679 intel_idle_cpuidle_driver_init();
Len Brown26717172010-03-08 14:07:30 -0500680 retval = cpuidle_register_driver(&intel_idle_driver);
681 if (retval) {
Konrad Rzeszutek Wilk3735d522012-08-16 22:06:55 +0200682 struct cpuidle_driver *drv = cpuidle_get_driver();
Len Brown26717172010-03-08 14:07:30 -0500683 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
Konrad Rzeszutek Wilk3735d522012-08-16 22:06:55 +0200684 drv ? drv->name : "none");
Len Brown26717172010-03-08 14:07:30 -0500685 return retval;
686 }
687
Thomas Renninger65b7f832012-01-17 22:40:08 +0100688 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 }
Len Brown26717172010-03-08 14:07:30 -0500698 }
Konrad Rzeszutek Wilk6f8c2e72013-01-16 23:40:01 +0100699 register_cpu_notifier(&cpu_hotplug_notifier);
Len Brown26717172010-03-08 14:07:30 -0500700
701 return 0;
702}
703
704static void __exit intel_idle_exit(void)
705{
706 intel_idle_cpuidle_devices_uninit();
707 cpuidle_unregister_driver(&intel_idle_driver);
708
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200709
710 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE)
Shaohua Li39a74fd2012-01-10 15:48:19 -0800711 on_each_cpu(__setup_broadcast_timer, (void *)false, 1);
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200712 unregister_cpu_notifier(&cpu_hotplug_notifier);
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800713
Len Brown26717172010-03-08 14:07:30 -0500714 return;
715}
716
717module_init(intel_idle_init);
718module_exit(intel_idle_exit);
719
Len Brown26717172010-03-08 14:07:30 -0500720module_param(max_cstate, int, 0444);
Len Brown26717172010-03-08 14:07:30 -0500721
722MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
723MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
724MODULE_LICENSE("GPL");