blob: 1a38dd7dfe4e56c9b749cac1e741fe4ad092024c [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,
Julius Wernera474a512012-11-27 14:17:58 +010074 .en_core_tk_irqen = 1,
Len Brown26717172010-03-08 14:07:30 -050075};
76/* intel_idle.max_cstate=0 disables driver */
Len Brown137ecc72013-02-01 21:35:35 -050077static int max_cstate = CPUIDLE_STATE_MAX - 1;
Len Brown26717172010-03-08 14:07:30 -050078
Len Brownc4236282010-05-28 02:22:03 -040079static unsigned int mwait_substates;
Len Brown26717172010-03-08 14:07:30 -050080
Shaohua Li2a2d31c2011-01-10 09:38:12 +080081#define LAPIC_TIMER_ALWAYS_RELIABLE 0xFFFFFFFF
Len Brown26717172010-03-08 14:07:30 -050082/* Reliable LAPIC Timer States, bit 1 for C1 etc. */
Len Brownd13780d2010-07-07 00:12:03 -040083static unsigned int lapic_timer_reliable_states = (1 << 1); /* Default to only C1 */
Len Brown26717172010-03-08 14:07:30 -050084
Andi Kleenb66b8b92012-01-26 00:09:07 +010085struct idle_cpu {
86 struct cpuidle_state *state_table;
87
88 /*
89 * Hardware C-state auto-demotion may not always be optimal.
90 * Indicate which enable bits to clear here.
91 */
92 unsigned long auto_demotion_disable_flags;
Len Brown32e95182013-02-02 01:31:56 -050093 bool disable_promotion_to_c1e;
Andi Kleenb66b8b92012-01-26 00:09:07 +010094};
95
96static const struct idle_cpu *icpu;
Namhyung Kim3265eba2010-08-08 03:10:03 +090097static struct cpuidle_device __percpu *intel_idle_cpuidle_devices;
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +053098static int intel_idle(struct cpuidle_device *dev,
99 struct cpuidle_driver *drv, int index);
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200100static int intel_idle_cpu_init(int cpu);
Len Brown26717172010-03-08 14:07:30 -0500101
102static struct cpuidle_state *cpuidle_state_table;
103
104/*
Len Brown956d0332011-01-12 02:51:20 -0500105 * 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/*
Len Brownb1beab42013-01-31 19:55:37 -0500113 * 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/*
Len Brown26717172010-03-08 14:07:30 -0500123 * 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 */
Len Brown137ecc72013-02-01 21:35:35 -0500127static struct cpuidle_state nehalem_cstates[CPUIDLE_STATE_MAX] = {
Len Browne022e7e2013-02-01 23:37:30 -0500128 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100129 .name = "C1-NHM",
Len Brown26717172010-03-08 14:07:30 -0500130 .desc = "MWAIT 0x00",
Len Brownb1beab42013-01-31 19:55:37 -0500131 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
Len Brown26717172010-03-08 14:07:30 -0500132 .exit_latency = 3,
Len Brown26717172010-03-08 14:07:30 -0500133 .target_residency = 6,
134 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500135 {
Len Brown32e95182013-02-02 01:31:56 -0500136 .name = "C1E-NHM",
137 .desc = "MWAIT 0x01",
138 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
139 .exit_latency = 10,
140 .target_residency = 20,
141 .enter = &intel_idle },
142 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100143 .name = "C3-NHM",
Len Brown26717172010-03-08 14:07:30 -0500144 .desc = "MWAIT 0x10",
Len Brownb1beab42013-01-31 19:55:37 -0500145 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown26717172010-03-08 14:07:30 -0500146 .exit_latency = 20,
Len Brown26717172010-03-08 14:07:30 -0500147 .target_residency = 80,
148 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500149 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100150 .name = "C6-NHM",
Len Brown26717172010-03-08 14:07:30 -0500151 .desc = "MWAIT 0x20",
Len Brownb1beab42013-01-31 19:55:37 -0500152 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown26717172010-03-08 14:07:30 -0500153 .exit_latency = 200,
Len Brown26717172010-03-08 14:07:30 -0500154 .target_residency = 800,
155 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500156 {
157 .enter = NULL }
Len Brown26717172010-03-08 14:07:30 -0500158};
159
Len Brown137ecc72013-02-01 21:35:35 -0500160static struct cpuidle_state snb_cstates[CPUIDLE_STATE_MAX] = {
Len Browne022e7e2013-02-01 23:37:30 -0500161 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100162 .name = "C1-SNB",
Len Brownd13780d2010-07-07 00:12:03 -0400163 .desc = "MWAIT 0x00",
Len Brownb1beab42013-01-31 19:55:37 -0500164 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
Len Brown32e95182013-02-02 01:31:56 -0500165 .exit_latency = 2,
166 .target_residency = 2,
167 .enter = &intel_idle },
168 {
169 .name = "C1E-SNB",
170 .desc = "MWAIT 0x01",
171 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
172 .exit_latency = 10,
173 .target_residency = 20,
Len Brownd13780d2010-07-07 00:12:03 -0400174 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500175 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100176 .name = "C3-SNB",
Len Brownd13780d2010-07-07 00:12:03 -0400177 .desc = "MWAIT 0x10",
Len Brownb1beab42013-01-31 19:55:37 -0500178 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brownd13780d2010-07-07 00:12:03 -0400179 .exit_latency = 80,
Len Brownddbd5502010-12-13 18:28:22 -0500180 .target_residency = 211,
Len Brownd13780d2010-07-07 00:12:03 -0400181 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500182 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100183 .name = "C6-SNB",
Len Brownd13780d2010-07-07 00:12:03 -0400184 .desc = "MWAIT 0x20",
Len Brownb1beab42013-01-31 19:55:37 -0500185 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brownd13780d2010-07-07 00:12:03 -0400186 .exit_latency = 104,
Len Brownddbd5502010-12-13 18:28:22 -0500187 .target_residency = 345,
Len Brownd13780d2010-07-07 00:12:03 -0400188 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500189 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100190 .name = "C7-SNB",
Len Brownd13780d2010-07-07 00:12:03 -0400191 .desc = "MWAIT 0x30",
Len Brownb1beab42013-01-31 19:55:37 -0500192 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brownd13780d2010-07-07 00:12:03 -0400193 .exit_latency = 109,
Len Brownddbd5502010-12-13 18:28:22 -0500194 .target_residency = 345,
Len Brownd13780d2010-07-07 00:12:03 -0400195 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500196 {
197 .enter = NULL }
Len Brownd13780d2010-07-07 00:12:03 -0400198};
199
Len Brown137ecc72013-02-01 21:35:35 -0500200static struct cpuidle_state ivb_cstates[CPUIDLE_STATE_MAX] = {
Len Browne022e7e2013-02-01 23:37:30 -0500201 {
Len Brown6edab082012-06-01 19:45:32 -0400202 .name = "C1-IVB",
203 .desc = "MWAIT 0x00",
Len Brownb1beab42013-01-31 19:55:37 -0500204 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
Len Brown6edab082012-06-01 19:45:32 -0400205 .exit_latency = 1,
206 .target_residency = 1,
207 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500208 {
Len Brown32e95182013-02-02 01:31:56 -0500209 .name = "C1E-IVB",
210 .desc = "MWAIT 0x01",
211 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
212 .exit_latency = 10,
213 .target_residency = 20,
214 .enter = &intel_idle },
215 {
Len Brown6edab082012-06-01 19:45:32 -0400216 .name = "C3-IVB",
217 .desc = "MWAIT 0x10",
Len Brownb1beab42013-01-31 19:55:37 -0500218 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown6edab082012-06-01 19:45:32 -0400219 .exit_latency = 59,
220 .target_residency = 156,
221 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500222 {
Len Brown6edab082012-06-01 19:45:32 -0400223 .name = "C6-IVB",
224 .desc = "MWAIT 0x20",
Len Brownb1beab42013-01-31 19:55:37 -0500225 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown6edab082012-06-01 19:45:32 -0400226 .exit_latency = 80,
227 .target_residency = 300,
228 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500229 {
Len Brown6edab082012-06-01 19:45:32 -0400230 .name = "C7-IVB",
231 .desc = "MWAIT 0x30",
Len Brownb1beab42013-01-31 19:55:37 -0500232 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown6edab082012-06-01 19:45:32 -0400233 .exit_latency = 87,
234 .target_residency = 300,
235 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500236 {
237 .enter = NULL }
Len Brown6edab082012-06-01 19:45:32 -0400238};
239
Len Brown137ecc72013-02-01 21:35:35 -0500240static struct cpuidle_state hsw_cstates[CPUIDLE_STATE_MAX] = {
Len Browne022e7e2013-02-01 23:37:30 -0500241 {
Len Brown85a4d2d2013-01-31 14:40:49 -0500242 .name = "C1-HSW",
243 .desc = "MWAIT 0x00",
244 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
245 .exit_latency = 2,
246 .target_residency = 2,
247 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500248 {
Len Brown32e95182013-02-02 01:31:56 -0500249 .name = "C1E-HSW",
250 .desc = "MWAIT 0x01",
251 .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
252 .exit_latency = 10,
253 .target_residency = 20,
254 .enter = &intel_idle },
255 {
Len Brown85a4d2d2013-01-31 14:40:49 -0500256 .name = "C3-HSW",
257 .desc = "MWAIT 0x10",
258 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
259 .exit_latency = 33,
260 .target_residency = 100,
261 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500262 {
Len Brown85a4d2d2013-01-31 14:40:49 -0500263 .name = "C6-HSW",
264 .desc = "MWAIT 0x20",
265 .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
266 .exit_latency = 133,
267 .target_residency = 400,
268 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500269 {
Len Brown85a4d2d2013-01-31 14:40:49 -0500270 .name = "C7s-HSW",
271 .desc = "MWAIT 0x32",
272 .flags = MWAIT2flg(0x32) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
273 .exit_latency = 166,
274 .target_residency = 500,
275 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500276 {
277 .enter = NULL }
Len Brown85a4d2d2013-01-31 14:40:49 -0500278};
279
Len Brown137ecc72013-02-01 21:35:35 -0500280static struct cpuidle_state atom_cstates[CPUIDLE_STATE_MAX] = {
Len Browne022e7e2013-02-01 23:37:30 -0500281 {
Len Brown32e95182013-02-02 01:31:56 -0500282 .name = "C1E-ATM",
Len Brown26717172010-03-08 14:07:30 -0500283 .desc = "MWAIT 0x00",
Len Brownb1beab42013-01-31 19:55:37 -0500284 .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
Len Brown32e95182013-02-02 01:31:56 -0500285 .exit_latency = 10,
286 .target_residency = 20,
Len Brown26717172010-03-08 14:07:30 -0500287 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500288 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100289 .name = "C2-ATM",
Len Brown26717172010-03-08 14:07:30 -0500290 .desc = "MWAIT 0x10",
Len Brownb1beab42013-01-31 19:55:37 -0500291 .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID,
Len Brown26717172010-03-08 14:07:30 -0500292 .exit_latency = 20,
Len Brown26717172010-03-08 14:07:30 -0500293 .target_residency = 80,
294 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500295 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100296 .name = "C4-ATM",
Len Brown26717172010-03-08 14:07:30 -0500297 .desc = "MWAIT 0x30",
Len Brownb1beab42013-01-31 19:55:37 -0500298 .flags = MWAIT2flg(0x30) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown26717172010-03-08 14:07:30 -0500299 .exit_latency = 100,
Len Brown26717172010-03-08 14:07:30 -0500300 .target_residency = 400,
301 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500302 {
Thomas Renninger15e123e2011-02-27 22:36:43 +0100303 .name = "C6-ATM",
Len Brown7fcca7d2010-10-05 13:43:14 -0400304 .desc = "MWAIT 0x52",
Len Brownb1beab42013-01-31 19:55:37 -0500305 .flags = MWAIT2flg(0x52) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
Len Brown7fcca7d2010-10-05 13:43:14 -0400306 .exit_latency = 140,
Len Brown7fcca7d2010-10-05 13:43:14 -0400307 .target_residency = 560,
308 .enter = &intel_idle },
Len Browne022e7e2013-02-01 23:37:30 -0500309 {
310 .enter = NULL }
Len Brown26717172010-03-08 14:07:30 -0500311};
312
Len Brown26717172010-03-08 14:07:30 -0500313/**
314 * intel_idle
315 * @dev: cpuidle_device
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530316 * @drv: cpuidle driver
Deepthi Dharware978aa72011-10-28 16:20:09 +0530317 * @index: index of cpuidle state
Len Brown26717172010-03-08 14:07:30 -0500318 *
Yanmin Zhang63ff07b2012-01-10 15:48:21 -0800319 * Must be called under local_irq_disable().
Len Brown26717172010-03-08 14:07:30 -0500320 */
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530321static int intel_idle(struct cpuidle_device *dev,
322 struct cpuidle_driver *drv, int index)
Len Brown26717172010-03-08 14:07:30 -0500323{
324 unsigned long ecx = 1; /* break on interrupt flag */
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530325 struct cpuidle_state *state = &drv->states[index];
Len Brownb1beab42013-01-31 19:55:37 -0500326 unsigned long eax = flg2MWAIT(state->flags);
Len Brown26717172010-03-08 14:07:30 -0500327 unsigned int cstate;
Len Brown26717172010-03-08 14:07:30 -0500328 int cpu = smp_processor_id();
329
330 cstate = (((eax) >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
331
Suresh Siddha6110a1f2010-09-30 21:19:07 -0400332 /*
Len Brownc8381cc2010-10-15 20:43:06 -0400333 * leave_mm() to avoid costly and often unnecessary wakeups
334 * for flushing the user TLB's associated with the active mm.
Suresh Siddha6110a1f2010-09-30 21:19:07 -0400335 */
Len Brownc8381cc2010-10-15 20:43:06 -0400336 if (state->flags & CPUIDLE_FLAG_TLB_FLUSHED)
Suresh Siddha6110a1f2010-09-30 21:19:07 -0400337 leave_mm(cpu);
338
Len Brown26717172010-03-08 14:07:30 -0500339 if (!(lapic_timer_reliable_states & (1 << (cstate))))
340 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
341
Len Brown26717172010-03-08 14:07:30 -0500342 stop_critical_timings();
Len Brown26717172010-03-08 14:07:30 -0500343 if (!need_resched()) {
344
345 __monitor((void *)&current_thread_info()->flags, 0, 0);
346 smp_mb();
347 if (!need_resched())
348 __mwait(eax, ecx);
349 }
350
351 start_critical_timings();
352
Len Brown26717172010-03-08 14:07:30 -0500353 if (!(lapic_timer_reliable_states & (1 << (cstate))))
354 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
355
Deepthi Dharware978aa72011-10-28 16:20:09 +0530356 return index;
Len Brown26717172010-03-08 14:07:30 -0500357}
358
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800359static void __setup_broadcast_timer(void *arg)
360{
361 unsigned long reason = (unsigned long)arg;
362 int cpu = smp_processor_id();
363
364 reason = reason ?
365 CLOCK_EVT_NOTIFY_BROADCAST_ON : CLOCK_EVT_NOTIFY_BROADCAST_OFF;
366
367 clockevents_notify(reason, &cpu);
368}
369
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200370static int cpu_hotplug_notify(struct notifier_block *n,
371 unsigned long action, void *hcpu)
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800372{
373 int hotcpu = (unsigned long)hcpu;
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200374 struct cpuidle_device *dev;
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800375
376 switch (action & 0xf) {
377 case CPU_ONLINE:
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200378
379 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE)
380 smp_call_function_single(hotcpu, __setup_broadcast_timer,
381 (void *)true, 1);
382
383 /*
384 * Some systems can hotplug a cpu at runtime after
385 * the kernel has booted, we have to initialize the
386 * driver in this case
387 */
388 dev = per_cpu_ptr(intel_idle_cpuidle_devices, hotcpu);
389 if (!dev->registered)
390 intel_idle_cpu_init(hotcpu);
391
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800392 break;
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800393 }
394 return NOTIFY_OK;
395}
396
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200397static struct notifier_block cpu_hotplug_notifier = {
398 .notifier_call = cpu_hotplug_notify,
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800399};
400
Len Brown14796fc2011-01-18 20:48:27 -0500401static void auto_demotion_disable(void *dummy)
402{
403 unsigned long long msr_bits;
404
405 rdmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
Andi Kleenb66b8b92012-01-26 00:09:07 +0100406 msr_bits &= ~(icpu->auto_demotion_disable_flags);
Len Brown14796fc2011-01-18 20:48:27 -0500407 wrmsrl(MSR_NHM_SNB_PKG_CST_CFG_CTL, msr_bits);
408}
Len Brown32e95182013-02-02 01:31:56 -0500409static void c1e_promotion_disable(void *dummy)
410{
411 unsigned long long msr_bits;
412
413 rdmsrl(MSR_IA32_POWER_CTL, msr_bits);
414 msr_bits &= ~0x2;
415 wrmsrl(MSR_IA32_POWER_CTL, msr_bits);
416}
Len Brown14796fc2011-01-18 20:48:27 -0500417
Andi Kleenb66b8b92012-01-26 00:09:07 +0100418static const struct idle_cpu idle_cpu_nehalem = {
419 .state_table = nehalem_cstates,
Andi Kleenb66b8b92012-01-26 00:09:07 +0100420 .auto_demotion_disable_flags = NHM_C1_AUTO_DEMOTE | NHM_C3_AUTO_DEMOTE,
Len Brown32e95182013-02-02 01:31:56 -0500421 .disable_promotion_to_c1e = true,
Andi Kleenb66b8b92012-01-26 00:09:07 +0100422};
423
424static const struct idle_cpu idle_cpu_atom = {
425 .state_table = atom_cstates,
426};
427
428static const struct idle_cpu idle_cpu_lincroft = {
429 .state_table = atom_cstates,
430 .auto_demotion_disable_flags = ATM_LNC_C6_AUTO_DEMOTE,
431};
432
433static const struct idle_cpu idle_cpu_snb = {
434 .state_table = snb_cstates,
Len Brown32e95182013-02-02 01:31:56 -0500435 .disable_promotion_to_c1e = true,
Andi Kleenb66b8b92012-01-26 00:09:07 +0100436};
437
Len Brown6edab082012-06-01 19:45:32 -0400438static const struct idle_cpu idle_cpu_ivb = {
439 .state_table = ivb_cstates,
Len Brown32e95182013-02-02 01:31:56 -0500440 .disable_promotion_to_c1e = true,
Len Brown6edab082012-06-01 19:45:32 -0400441};
442
Len Brown85a4d2d2013-01-31 14:40:49 -0500443static const struct idle_cpu idle_cpu_hsw = {
444 .state_table = hsw_cstates,
Len Brown32e95182013-02-02 01:31:56 -0500445 .disable_promotion_to_c1e = true,
Len Brown85a4d2d2013-01-31 14:40:49 -0500446};
447
Andi Kleenb66b8b92012-01-26 00:09:07 +0100448#define ICPU(model, cpu) \
449 { X86_VENDOR_INTEL, 6, model, X86_FEATURE_MWAIT, (unsigned long)&cpu }
450
451static const struct x86_cpu_id intel_idle_ids[] = {
452 ICPU(0x1a, idle_cpu_nehalem),
453 ICPU(0x1e, idle_cpu_nehalem),
454 ICPU(0x1f, idle_cpu_nehalem),
Ben Hutchings8bf11932012-02-16 04:13:14 +0000455 ICPU(0x25, idle_cpu_nehalem),
456 ICPU(0x2c, idle_cpu_nehalem),
457 ICPU(0x2e, idle_cpu_nehalem),
Andi Kleenb66b8b92012-01-26 00:09:07 +0100458 ICPU(0x1c, idle_cpu_atom),
459 ICPU(0x26, idle_cpu_lincroft),
Ben Hutchings8bf11932012-02-16 04:13:14 +0000460 ICPU(0x2f, idle_cpu_nehalem),
Andi Kleenb66b8b92012-01-26 00:09:07 +0100461 ICPU(0x2a, idle_cpu_snb),
462 ICPU(0x2d, idle_cpu_snb),
Len Brown6edab082012-06-01 19:45:32 -0400463 ICPU(0x3a, idle_cpu_ivb),
Len Brown23795e52012-09-26 22:28:21 -0400464 ICPU(0x3e, idle_cpu_ivb),
Len Brown85a4d2d2013-01-31 14:40:49 -0500465 ICPU(0x3c, idle_cpu_hsw),
466 ICPU(0x3f, idle_cpu_hsw),
467 ICPU(0x45, idle_cpu_hsw),
Len Brown0b158412013-03-15 10:55:31 -0400468 ICPU(0x46, idle_cpu_hsw),
Andi Kleenb66b8b92012-01-26 00:09:07 +0100469 {}
470};
471MODULE_DEVICE_TABLE(x86cpu, intel_idle_ids);
472
Len Brown26717172010-03-08 14:07:30 -0500473/*
474 * intel_idle_probe()
475 */
476static int intel_idle_probe(void)
477{
Len Brownc4236282010-05-28 02:22:03 -0400478 unsigned int eax, ebx, ecx;
Andi Kleenb66b8b92012-01-26 00:09:07 +0100479 const struct x86_cpu_id *id;
Len Brown26717172010-03-08 14:07:30 -0500480
481 if (max_cstate == 0) {
482 pr_debug(PREFIX "disabled\n");
483 return -EPERM;
484 }
485
Andi Kleenb66b8b92012-01-26 00:09:07 +0100486 id = x86_match_cpu(intel_idle_ids);
487 if (!id) {
488 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
489 boot_cpu_data.x86 == 6)
490 pr_debug(PREFIX "does not run on family %d model %d\n",
491 boot_cpu_data.x86, boot_cpu_data.x86_model);
Len Brown26717172010-03-08 14:07:30 -0500492 return -ENODEV;
Andi Kleenb66b8b92012-01-26 00:09:07 +0100493 }
Len Brown26717172010-03-08 14:07:30 -0500494
495 if (boot_cpu_data.cpuid_level < CPUID_MWAIT_LEAF)
496 return -ENODEV;
497
Len Brownc4236282010-05-28 02:22:03 -0400498 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &mwait_substates);
Len Brown26717172010-03-08 14:07:30 -0500499
500 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) ||
Thomas Renninger5c2a9f02011-12-04 22:17:29 +0100501 !(ecx & CPUID5_ECX_INTERRUPT_BREAK) ||
502 !mwait_substates)
Len Brown26717172010-03-08 14:07:30 -0500503 return -ENODEV;
Len Brown26717172010-03-08 14:07:30 -0500504
Len Brownc4236282010-05-28 02:22:03 -0400505 pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
Len Brown26717172010-03-08 14:07:30 -0500506
Andi Kleenb66b8b92012-01-26 00:09:07 +0100507 icpu = (const struct idle_cpu *)id->driver_data;
508 cpuidle_state_table = icpu->state_table;
Len Brown26717172010-03-08 14:07:30 -0500509
Len Brown56b9aea2010-12-02 01:19:32 -0500510 if (boot_cpu_has(X86_FEATURE_ARAT)) /* Always Reliable APIC Timer */
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800511 lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200512 else
Shaohua Li39a74fd2012-01-10 15:48:19 -0800513 on_each_cpu(__setup_broadcast_timer, (void *)true, 1);
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200514
Len Brown26717172010-03-08 14:07:30 -0500515 pr_debug(PREFIX "v" INTEL_IDLE_VERSION
516 " model 0x%X\n", boot_cpu_data.x86_model);
517
518 pr_debug(PREFIX "lapic_timer_reliable_states 0x%x\n",
519 lapic_timer_reliable_states);
520 return 0;
521}
522
523/*
524 * intel_idle_cpuidle_devices_uninit()
525 * unregister, free cpuidle_devices
526 */
527static void intel_idle_cpuidle_devices_uninit(void)
528{
529 int i;
530 struct cpuidle_device *dev;
531
532 for_each_online_cpu(i) {
533 dev = per_cpu_ptr(intel_idle_cpuidle_devices, i);
534 cpuidle_unregister_device(dev);
535 }
536
537 free_percpu(intel_idle_cpuidle_devices);
538 return;
539}
540/*
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530541 * intel_idle_cpuidle_driver_init()
542 * allocate, initialize cpuidle_states
543 */
544static int intel_idle_cpuidle_driver_init(void)
545{
546 int cstate;
547 struct cpuidle_driver *drv = &intel_idle_driver;
548
549 drv->state_count = 1;
550
Len Browne022e7e2013-02-01 23:37:30 -0500551 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
552 int num_substates, mwait_hint, mwait_cstate, mwait_substate;
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530553
Len Browne022e7e2013-02-01 23:37:30 -0500554 if (cpuidle_state_table[cstate].enter == NULL)
555 break;
556
557 if (cstate + 1 > max_cstate) {
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530558 printk(PREFIX "max_cstate %d reached\n",
559 max_cstate);
560 break;
561 }
562
Len Browne022e7e2013-02-01 23:37:30 -0500563 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
564 mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint);
565 mwait_substate = MWAIT_HINT2SUBSTATE(mwait_hint);
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530566
Len Browne022e7e2013-02-01 23:37:30 -0500567 /* does the state exist in CPUID.MWAIT? */
568 num_substates = (mwait_substates >> ((mwait_cstate + 1) * 4))
569 & MWAIT_SUBSTATE_MASK;
570
571 /* if sub-state in table is not enumerated by CPUID */
572 if ((mwait_substate + 1) > num_substates)
573 continue;
574
575 if (((mwait_cstate + 1) > 2) &&
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530576 !boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
577 mark_tsc_unstable("TSC halts in idle"
578 " states deeper than C2");
579
580 drv->states[drv->state_count] = /* structure copy */
581 cpuidle_state_table[cstate];
582
583 drv->state_count += 1;
584 }
585
Andi Kleenb66b8b92012-01-26 00:09:07 +0100586 if (icpu->auto_demotion_disable_flags)
Shaohua Li39a74fd2012-01-10 15:48:19 -0800587 on_each_cpu(auto_demotion_disable, NULL, 1);
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530588
Len Brown32e95182013-02-02 01:31:56 -0500589 if (icpu->disable_promotion_to_c1e) /* each-cpu is redundant */
590 on_each_cpu(c1e_promotion_disable, NULL, 1);
591
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530592 return 0;
593}
594
595
596/*
Thomas Renninger65b7f832012-01-17 22:40:08 +0100597 * intel_idle_cpu_init()
Len Brown26717172010-03-08 14:07:30 -0500598 * allocate, initialize, register cpuidle_devices
Thomas Renninger65b7f832012-01-17 22:40:08 +0100599 * @cpu: cpu/core to initialize
Len Brown26717172010-03-08 14:07:30 -0500600 */
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200601static int intel_idle_cpu_init(int cpu)
Len Brown26717172010-03-08 14:07:30 -0500602{
Thomas Renninger65b7f832012-01-17 22:40:08 +0100603 int cstate;
Len Brown26717172010-03-08 14:07:30 -0500604 struct cpuidle_device *dev;
605
Thomas Renninger65b7f832012-01-17 22:40:08 +0100606 dev = per_cpu_ptr(intel_idle_cpuidle_devices, cpu);
Len Brown26717172010-03-08 14:07:30 -0500607
Thomas Renninger65b7f832012-01-17 22:40:08 +0100608 dev->state_count = 1;
Len Brown26717172010-03-08 14:07:30 -0500609
Len Browne022e7e2013-02-01 23:37:30 -0500610 for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {
611 int num_substates, mwait_hint, mwait_cstate, mwait_substate;
Len Brown26717172010-03-08 14:07:30 -0500612
Len Browne022e7e2013-02-01 23:37:30 -0500613 if (cpuidle_state_table[cstate].enter == NULL)
614 continue;
615
616 if (cstate + 1 > max_cstate) {
Marcos Paulo de Souzadc716e92012-03-21 16:33:43 -0700617 printk(PREFIX "max_cstate %d reached\n", max_cstate);
Thomas Renninger65b7f832012-01-17 22:40:08 +0100618 break;
619 }
Len Brown26717172010-03-08 14:07:30 -0500620
Len Browne022e7e2013-02-01 23:37:30 -0500621 mwait_hint = flg2MWAIT(cpuidle_state_table[cstate].flags);
622 mwait_cstate = MWAIT_HINT2CSTATE(mwait_hint);
623 mwait_substate = MWAIT_HINT2SUBSTATE(mwait_hint);
624
Thomas Renninger65b7f832012-01-17 22:40:08 +0100625 /* does the state exist in CPUID.MWAIT? */
Len Browne022e7e2013-02-01 23:37:30 -0500626 num_substates = (mwait_substates >> ((mwait_cstate + 1) * 4))
627 & MWAIT_SUBSTATE_MASK;
628
629 /* if sub-state in table is not enumerated by CPUID */
630 if ((mwait_substate + 1) > num_substates)
Thomas Renninger65b7f832012-01-17 22:40:08 +0100631 continue;
Len Brown26717172010-03-08 14:07:30 -0500632
Marcos Paulo de Souzadc716e92012-03-21 16:33:43 -0700633 dev->state_count += 1;
634 }
635
Thomas Renninger65b7f832012-01-17 22:40:08 +0100636 dev->cpu = cpu;
Len Brown26717172010-03-08 14:07:30 -0500637
Thomas Renninger65b7f832012-01-17 22:40:08 +0100638 if (cpuidle_register_device(dev)) {
639 pr_debug(PREFIX "cpuidle_register_device %d failed!\n", cpu);
640 intel_idle_cpuidle_devices_uninit();
641 return -EIO;
Len Brown26717172010-03-08 14:07:30 -0500642 }
643
Andi Kleenb66b8b92012-01-26 00:09:07 +0100644 if (icpu->auto_demotion_disable_flags)
Thomas Renninger65b7f832012-01-17 22:40:08 +0100645 smp_call_function_single(cpu, auto_demotion_disable, NULL, 1);
646
Len Brown26717172010-03-08 14:07:30 -0500647 return 0;
648}
Len Brown26717172010-03-08 14:07:30 -0500649
650static int __init intel_idle_init(void)
651{
Thomas Renninger65b7f832012-01-17 22:40:08 +0100652 int retval, i;
Len Brown26717172010-03-08 14:07:30 -0500653
Thomas Renningerd1896042010-11-03 17:06:14 +0100654 /* Do not load intel_idle at all for now if idle= is passed */
655 if (boot_option_idle_override != IDLE_NO_OVERRIDE)
656 return -ENODEV;
657
Len Brown26717172010-03-08 14:07:30 -0500658 retval = intel_idle_probe();
659 if (retval)
660 return retval;
661
Deepthi Dharwar46bcfad2011-10-28 16:20:42 +0530662 intel_idle_cpuidle_driver_init();
Len Brown26717172010-03-08 14:07:30 -0500663 retval = cpuidle_register_driver(&intel_idle_driver);
664 if (retval) {
Konrad Rzeszutek Wilk3735d522012-08-16 22:06:55 +0200665 struct cpuidle_driver *drv = cpuidle_get_driver();
Len Brown26717172010-03-08 14:07:30 -0500666 printk(KERN_DEBUG PREFIX "intel_idle yielding to %s",
Konrad Rzeszutek Wilk3735d522012-08-16 22:06:55 +0200667 drv ? drv->name : "none");
Len Brown26717172010-03-08 14:07:30 -0500668 return retval;
669 }
670
Thomas Renninger65b7f832012-01-17 22:40:08 +0100671 intel_idle_cpuidle_devices = alloc_percpu(struct cpuidle_device);
672 if (intel_idle_cpuidle_devices == NULL)
673 return -ENOMEM;
674
675 for_each_online_cpu(i) {
676 retval = intel_idle_cpu_init(i);
677 if (retval) {
678 cpuidle_unregister_driver(&intel_idle_driver);
679 return retval;
680 }
Len Brown26717172010-03-08 14:07:30 -0500681 }
Konrad Rzeszutek Wilk6f8c2e72013-01-16 23:40:01 +0100682 register_cpu_notifier(&cpu_hotplug_notifier);
Len Brown26717172010-03-08 14:07:30 -0500683
684 return 0;
685}
686
687static void __exit intel_idle_exit(void)
688{
689 intel_idle_cpuidle_devices_uninit();
690 cpuidle_unregister_driver(&intel_idle_driver);
691
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200692
693 if (lapic_timer_reliable_states != LAPIC_TIMER_ALWAYS_RELIABLE)
Shaohua Li39a74fd2012-01-10 15:48:19 -0800694 on_each_cpu(__setup_broadcast_timer, (void *)false, 1);
Daniel Lezcano25ac7762012-07-05 15:23:25 +0200695 unregister_cpu_notifier(&cpu_hotplug_notifier);
Shaohua Li2a2d31c2011-01-10 09:38:12 +0800696
Len Brown26717172010-03-08 14:07:30 -0500697 return;
698}
699
700module_init(intel_idle_init);
701module_exit(intel_idle_exit);
702
Len Brown26717172010-03-08 14:07:30 -0500703module_param(max_cstate, int, 0444);
Len Brown26717172010-03-08 14:07:30 -0500704
705MODULE_AUTHOR("Len Brown <len.brown@intel.com>");
706MODULE_DESCRIPTION("Cpuidle driver for Intel Hardware v" INTEL_IDLE_VERSION);
707MODULE_LICENSE("GPL");