Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / cpuidle / cpuidle-powernv.c
1 /*
2  *  cpuidle-powernv - idle state cpuidle driver.
3  *  Adapted from drivers/cpuidle/cpuidle-pseries
4  *
5  */
6
7 #include <linux/kernel.h>
8 #include <linux/module.h>
9 #include <linux/init.h>
10 #include <linux/moduleparam.h>
11 #include <linux/cpuidle.h>
12 #include <linux/cpu.h>
13 #include <linux/notifier.h>
14 #include <linux/clockchips.h>
15 #include <linux/of.h>
16 #include <linux/slab.h>
17
18 #include <asm/machdep.h>
19 #include <asm/firmware.h>
20 #include <asm/opal.h>
21 #include <asm/runlatch.h>
22
23 #define MAX_POWERNV_IDLE_STATES 8
24
25 struct cpuidle_driver powernv_idle_driver = {
26         .name             = "powernv_idle",
27         .owner            = THIS_MODULE,
28 };
29
30 static int max_idle_state;
31 static struct cpuidle_state *cpuidle_state_table;
32
33 static int snooze_loop(struct cpuidle_device *dev,
34                         struct cpuidle_driver *drv,
35                         int index)
36 {
37         local_irq_enable();
38         set_thread_flag(TIF_POLLING_NRFLAG);
39
40         ppc64_runlatch_off();
41         while (!need_resched()) {
42                 HMT_low();
43                 HMT_very_low();
44         }
45
46         HMT_medium();
47         ppc64_runlatch_on();
48         clear_thread_flag(TIF_POLLING_NRFLAG);
49         smp_mb();
50         return index;
51 }
52
53 static int nap_loop(struct cpuidle_device *dev,
54                         struct cpuidle_driver *drv,
55                         int index)
56 {
57         ppc64_runlatch_off();
58         power7_idle();
59         ppc64_runlatch_on();
60         return index;
61 }
62
63 /* Register for fastsleep only in oneshot mode of broadcast */
64 #ifdef CONFIG_TICK_ONESHOT
65 static int fastsleep_loop(struct cpuidle_device *dev,
66                                 struct cpuidle_driver *drv,
67                                 int index)
68 {
69         unsigned long old_lpcr = mfspr(SPRN_LPCR);
70         unsigned long new_lpcr;
71
72         if (unlikely(system_state < SYSTEM_RUNNING))
73                 return index;
74
75         new_lpcr = old_lpcr;
76         /* Do not exit powersave upon decrementer as we've setup the timer
77          * offload.
78          */
79         new_lpcr &= ~LPCR_PECE1;
80
81         mtspr(SPRN_LPCR, new_lpcr);
82         power7_sleep();
83
84         mtspr(SPRN_LPCR, old_lpcr);
85
86         return index;
87 }
88 #endif
89 /*
90  * States for dedicated partition case.
91  */
92 static struct cpuidle_state powernv_states[MAX_POWERNV_IDLE_STATES] = {
93         { /* Snooze */
94                 .name = "snooze",
95                 .desc = "snooze",
96                 .exit_latency = 0,
97                 .target_residency = 0,
98                 .enter = &snooze_loop },
99 };
100
101 static int powernv_cpuidle_add_cpu_notifier(struct notifier_block *n,
102                         unsigned long action, void *hcpu)
103 {
104         int hotcpu = (unsigned long)hcpu;
105         struct cpuidle_device *dev =
106                                 per_cpu(cpuidle_devices, hotcpu);
107
108         if (dev && cpuidle_get_driver()) {
109                 switch (action) {
110                 case CPU_ONLINE:
111                 case CPU_ONLINE_FROZEN:
112                         cpuidle_pause_and_lock();
113                         cpuidle_enable_device(dev);
114                         cpuidle_resume_and_unlock();
115                         break;
116
117                 case CPU_DEAD:
118                 case CPU_DEAD_FROZEN:
119                         cpuidle_pause_and_lock();
120                         cpuidle_disable_device(dev);
121                         cpuidle_resume_and_unlock();
122                         break;
123
124                 default:
125                         return NOTIFY_DONE;
126                 }
127         }
128         return NOTIFY_OK;
129 }
130
131 static struct notifier_block setup_hotplug_notifier = {
132         .notifier_call = powernv_cpuidle_add_cpu_notifier,
133 };
134
135 /*
136  * powernv_cpuidle_driver_init()
137  */
138 static int powernv_cpuidle_driver_init(void)
139 {
140         int idle_state;
141         struct cpuidle_driver *drv = &powernv_idle_driver;
142
143         drv->state_count = 0;
144
145         for (idle_state = 0; idle_state < max_idle_state; ++idle_state) {
146                 /* Is the state not enabled? */
147                 if (cpuidle_state_table[idle_state].enter == NULL)
148                         continue;
149
150                 drv->states[drv->state_count] = /* structure copy */
151                         cpuidle_state_table[idle_state];
152
153                 drv->state_count += 1;
154         }
155
156         return 0;
157 }
158
159 static int powernv_add_idle_states(void)
160 {
161         struct device_node *power_mgt;
162         int nr_idle_states = 1; /* Snooze */
163         int dt_idle_states;
164         u32 *latency_ns, *residency_ns, *flags;
165         int i, rc;
166
167         /* Currently we have snooze statically defined */
168
169         power_mgt = of_find_node_by_path("/ibm,opal/power-mgt");
170         if (!power_mgt) {
171                 pr_warn("opal: PowerMgmt Node not found\n");
172                 goto out;
173         }
174
175         /* Read values of any property to determine the num of idle states */
176         dt_idle_states = of_property_count_u32_elems(power_mgt, "ibm,cpu-idle-state-flags");
177         if (dt_idle_states < 0) {
178                 pr_warn("cpuidle-powernv: no idle states found in the DT\n");
179                 goto out;
180         }
181
182         flags = kzalloc(sizeof(*flags) * dt_idle_states, GFP_KERNEL);
183         if (of_property_read_u32_array(power_mgt,
184                         "ibm,cpu-idle-state-flags", flags, dt_idle_states)) {
185                 pr_warn("cpuidle-powernv : missing ibm,cpu-idle-state-flags in DT\n");
186                 goto out_free_flags;
187         }
188
189         latency_ns = kzalloc(sizeof(*latency_ns) * dt_idle_states, GFP_KERNEL);
190         rc = of_property_read_u32_array(power_mgt,
191                 "ibm,cpu-idle-state-latencies-ns", latency_ns, dt_idle_states);
192         if (rc) {
193                 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-latencies-ns in DT\n");
194                 goto out_free_latency;
195         }
196
197         residency_ns = kzalloc(sizeof(*residency_ns) * dt_idle_states, GFP_KERNEL);
198         rc = of_property_read_u32_array(power_mgt,
199                 "ibm,cpu-idle-state-residency-ns", residency_ns, dt_idle_states);
200
201         for (i = 0; i < dt_idle_states; i++) {
202
203                 /*
204                  * Cpuidle accepts exit_latency and target_residency in us.
205                  * Use default target_residency values if f/w does not expose it.
206                  */
207                 if (flags[i] & OPAL_PM_NAP_ENABLED) {
208                         /* Add NAP state */
209                         strcpy(powernv_states[nr_idle_states].name, "Nap");
210                         strcpy(powernv_states[nr_idle_states].desc, "Nap");
211                         powernv_states[nr_idle_states].flags = 0;
212                         powernv_states[nr_idle_states].target_residency = 100;
213                         powernv_states[nr_idle_states].enter = &nap_loop;
214                 }
215
216                 /*
217                  * All cpuidle states with CPUIDLE_FLAG_TIMER_STOP set must come
218                  * within this config dependency check.
219                  */
220 #ifdef CONFIG_TICK_ONESHOT
221                 if (flags[i] & OPAL_PM_SLEEP_ENABLED ||
222                         flags[i] & OPAL_PM_SLEEP_ENABLED_ER1) {
223                         /* Add FASTSLEEP state */
224                         strcpy(powernv_states[nr_idle_states].name, "FastSleep");
225                         strcpy(powernv_states[nr_idle_states].desc, "FastSleep");
226                         powernv_states[nr_idle_states].flags = CPUIDLE_FLAG_TIMER_STOP;
227                         powernv_states[nr_idle_states].target_residency = 300000;
228                         powernv_states[nr_idle_states].enter = &fastsleep_loop;
229                 }
230 #endif
231                 powernv_states[nr_idle_states].exit_latency =
232                                 ((unsigned int)latency_ns[i]) / 1000;
233
234                 if (!rc) {
235                         powernv_states[nr_idle_states].target_residency =
236                                 ((unsigned int)residency_ns[i]) / 1000;
237                 }
238
239                 nr_idle_states++;
240         }
241
242         kfree(residency_ns);
243 out_free_latency:
244         kfree(latency_ns);
245 out_free_flags:
246         kfree(flags);
247 out:
248         return nr_idle_states;
249 }
250
251 /*
252  * powernv_idle_probe()
253  * Choose state table for shared versus dedicated partition
254  */
255 static int powernv_idle_probe(void)
256 {
257         if (cpuidle_disable != IDLE_NO_OVERRIDE)
258                 return -ENODEV;
259
260         if (firmware_has_feature(FW_FEATURE_OPALv3)) {
261                 cpuidle_state_table = powernv_states;
262                 /* Device tree can indicate more idle states */
263                 max_idle_state = powernv_add_idle_states();
264         } else
265                 return -ENODEV;
266
267         return 0;
268 }
269
270 static int __init powernv_processor_idle_init(void)
271 {
272         int retval;
273
274         retval = powernv_idle_probe();
275         if (retval)
276                 return retval;
277
278         powernv_cpuidle_driver_init();
279         retval = cpuidle_register(&powernv_idle_driver, NULL);
280         if (retval) {
281                 printk(KERN_DEBUG "Registration of powernv driver failed.\n");
282                 return retval;
283         }
284
285         register_cpu_notifier(&setup_hotplug_notifier);
286         printk(KERN_DEBUG "powernv_idle_driver registered\n");
287         return 0;
288 }
289
290 device_initcall(powernv_processor_idle_init);