2 * cpuidle-powernv - idle state cpuidle driver.
3 * Adapted from drivers/cpuidle/cpuidle-pseries
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>
16 #include <linux/slab.h>
18 #include <asm/machdep.h>
19 #include <asm/firmware.h>
21 #include <asm/runlatch.h>
23 #define MAX_POWERNV_IDLE_STATES 8
25 struct cpuidle_driver powernv_idle_driver = {
26 .name = "powernv_idle",
30 static int max_idle_state;
31 static struct cpuidle_state *cpuidle_state_table;
32 static u64 default_snooze_timeout;
33 static bool snooze_timeout_en;
35 static u64 get_snooze_timeout(struct cpuidle_device *dev,
36 struct cpuidle_driver *drv,
41 if (unlikely(!snooze_timeout_en))
42 return default_snooze_timeout;
44 for (i = index + 1; i < drv->state_count; i++) {
45 struct cpuidle_state *s = &drv->states[i];
46 struct cpuidle_state_usage *su = &dev->states_usage[i];
48 if (s->disabled || su->disable)
51 return s->target_residency * tb_ticks_per_usec;
54 return default_snooze_timeout;
57 static int snooze_loop(struct cpuidle_device *dev,
58 struct cpuidle_driver *drv,
64 set_thread_flag(TIF_POLLING_NRFLAG);
66 snooze_exit_time = get_tb() + get_snooze_timeout(dev, drv, index);
68 while (!need_resched()) {
71 if (snooze_timeout_en && get_tb() > snooze_exit_time)
77 clear_thread_flag(TIF_POLLING_NRFLAG);
82 static int nap_loop(struct cpuidle_device *dev,
83 struct cpuidle_driver *drv,
92 /* Register for fastsleep only in oneshot mode of broadcast */
93 #ifdef CONFIG_TICK_ONESHOT
94 static int fastsleep_loop(struct cpuidle_device *dev,
95 struct cpuidle_driver *drv,
98 unsigned long old_lpcr = mfspr(SPRN_LPCR);
99 unsigned long new_lpcr;
101 if (unlikely(system_state < SYSTEM_RUNNING))
105 /* Do not exit powersave upon decrementer as we've setup the timer
108 new_lpcr &= ~LPCR_PECE1;
110 mtspr(SPRN_LPCR, new_lpcr);
113 mtspr(SPRN_LPCR, old_lpcr);
119 * States for dedicated partition case.
121 static struct cpuidle_state powernv_states[MAX_POWERNV_IDLE_STATES] = {
126 .target_residency = 0,
127 .enter = &snooze_loop },
130 static int powernv_cpuidle_add_cpu_notifier(struct notifier_block *n,
131 unsigned long action, void *hcpu)
133 int hotcpu = (unsigned long)hcpu;
134 struct cpuidle_device *dev =
135 per_cpu(cpuidle_devices, hotcpu);
137 if (dev && cpuidle_get_driver()) {
140 case CPU_ONLINE_FROZEN:
141 cpuidle_pause_and_lock();
142 cpuidle_enable_device(dev);
143 cpuidle_resume_and_unlock();
147 case CPU_DEAD_FROZEN:
148 cpuidle_pause_and_lock();
149 cpuidle_disable_device(dev);
150 cpuidle_resume_and_unlock();
160 static struct notifier_block setup_hotplug_notifier = {
161 .notifier_call = powernv_cpuidle_add_cpu_notifier,
165 * powernv_cpuidle_driver_init()
167 static int powernv_cpuidle_driver_init(void)
170 struct cpuidle_driver *drv = &powernv_idle_driver;
172 drv->state_count = 0;
174 for (idle_state = 0; idle_state < max_idle_state; ++idle_state) {
175 /* Is the state not enabled? */
176 if (cpuidle_state_table[idle_state].enter == NULL)
179 drv->states[drv->state_count] = /* structure copy */
180 cpuidle_state_table[idle_state];
182 drv->state_count += 1;
186 * On the PowerNV platform cpu_present may be less than cpu_possible in
187 * cases when firmware detects the CPU, but it is not available to the
188 * OS. If CONFIG_HOTPLUG_CPU=n, then such CPUs are not hotplugable at
189 * run time and hence cpu_devices are not created for those CPUs by the
190 * generic topology_init().
192 * drv->cpumask defaults to cpu_possible_mask in
193 * __cpuidle_driver_init(). This breaks cpuidle on PowerNV where
194 * cpu_devices are not created for CPUs in cpu_possible_mask that
195 * cannot be hot-added later at run time.
197 * Trying cpuidle_register_device() on a CPU without a cpu_device is
198 * incorrect, so pass a correct CPU mask to the generic cpuidle driver.
201 drv->cpumask = (struct cpumask *)cpu_present_mask;
206 static int powernv_add_idle_states(void)
208 struct device_node *power_mgt;
209 int nr_idle_states = 1; /* Snooze */
211 u32 *latency_ns, *residency_ns, *flags;
214 /* Currently we have snooze statically defined */
216 power_mgt = of_find_node_by_path("/ibm,opal/power-mgt");
218 pr_warn("opal: PowerMgmt Node not found\n");
222 /* Read values of any property to determine the num of idle states */
223 dt_idle_states = of_property_count_u32_elems(power_mgt, "ibm,cpu-idle-state-flags");
224 if (dt_idle_states < 0) {
225 pr_warn("cpuidle-powernv: no idle states found in the DT\n");
229 flags = kzalloc(sizeof(*flags) * dt_idle_states, GFP_KERNEL);
230 if (of_property_read_u32_array(power_mgt,
231 "ibm,cpu-idle-state-flags", flags, dt_idle_states)) {
232 pr_warn("cpuidle-powernv : missing ibm,cpu-idle-state-flags in DT\n");
236 latency_ns = kzalloc(sizeof(*latency_ns) * dt_idle_states, GFP_KERNEL);
237 rc = of_property_read_u32_array(power_mgt,
238 "ibm,cpu-idle-state-latencies-ns", latency_ns, dt_idle_states);
240 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-latencies-ns in DT\n");
241 goto out_free_latency;
244 residency_ns = kzalloc(sizeof(*residency_ns) * dt_idle_states, GFP_KERNEL);
245 rc = of_property_read_u32_array(power_mgt,
246 "ibm,cpu-idle-state-residency-ns", residency_ns, dt_idle_states);
248 for (i = 0; i < dt_idle_states; i++) {
251 * Cpuidle accepts exit_latency and target_residency in us.
252 * Use default target_residency values if f/w does not expose it.
254 if (flags[i] & OPAL_PM_NAP_ENABLED) {
256 strcpy(powernv_states[nr_idle_states].name, "Nap");
257 strcpy(powernv_states[nr_idle_states].desc, "Nap");
258 powernv_states[nr_idle_states].flags = 0;
259 powernv_states[nr_idle_states].target_residency = 100;
260 powernv_states[nr_idle_states].enter = &nap_loop;
264 * All cpuidle states with CPUIDLE_FLAG_TIMER_STOP set must come
265 * within this config dependency check.
267 #ifdef CONFIG_TICK_ONESHOT
268 if (flags[i] & OPAL_PM_SLEEP_ENABLED ||
269 flags[i] & OPAL_PM_SLEEP_ENABLED_ER1) {
270 /* Add FASTSLEEP state */
271 strcpy(powernv_states[nr_idle_states].name, "FastSleep");
272 strcpy(powernv_states[nr_idle_states].desc, "FastSleep");
273 powernv_states[nr_idle_states].flags = CPUIDLE_FLAG_TIMER_STOP;
274 powernv_states[nr_idle_states].target_residency = 300000;
275 powernv_states[nr_idle_states].enter = &fastsleep_loop;
278 powernv_states[nr_idle_states].exit_latency =
279 ((unsigned int)latency_ns[i]) / 1000;
282 powernv_states[nr_idle_states].target_residency =
283 ((unsigned int)residency_ns[i]) / 1000;
295 return nr_idle_states;
299 * powernv_idle_probe()
300 * Choose state table for shared versus dedicated partition
302 static int powernv_idle_probe(void)
304 if (cpuidle_disable != IDLE_NO_OVERRIDE)
307 if (firmware_has_feature(FW_FEATURE_OPAL)) {
308 cpuidle_state_table = powernv_states;
309 /* Device tree can indicate more idle states */
310 max_idle_state = powernv_add_idle_states();
311 default_snooze_timeout = TICK_USEC * tb_ticks_per_usec;
312 if (max_idle_state > 1)
313 snooze_timeout_en = true;
320 static int __init powernv_processor_idle_init(void)
324 retval = powernv_idle_probe();
328 powernv_cpuidle_driver_init();
329 retval = cpuidle_register(&powernv_idle_driver, NULL);
331 printk(KERN_DEBUG "Registration of powernv driver failed.\n");
335 register_cpu_notifier(&setup_hotplug_notifier);
336 printk(KERN_DEBUG "powernv_idle_driver registered\n");
340 device_initcall(powernv_processor_idle_init);