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>
22 #include <asm/cpuidle.h>
25 * Expose only those Hardware idle states via the cpuidle framework
26 * that have latency value below POWERNV_THRESHOLD_LATENCY_NS.
28 #define POWERNV_THRESHOLD_LATENCY_NS 200000
30 static struct cpuidle_driver powernv_idle_driver = {
31 .name = "powernv_idle",
35 static int max_idle_state;
36 static struct cpuidle_state *cpuidle_state_table;
38 struct stop_psscr_table {
43 static struct stop_psscr_table stop_psscr_table[CPUIDLE_STATE_MAX];
45 static u64 snooze_timeout;
46 static bool snooze_timeout_en;
48 static int snooze_loop(struct cpuidle_device *dev,
49 struct cpuidle_driver *drv,
55 set_thread_flag(TIF_POLLING_NRFLAG);
57 snooze_exit_time = get_tb() + snooze_timeout;
59 while (!need_resched()) {
62 if (snooze_timeout_en && get_tb() > snooze_exit_time)
68 clear_thread_flag(TIF_POLLING_NRFLAG);
73 static int nap_loop(struct cpuidle_device *dev,
74 struct cpuidle_driver *drv,
83 /* Register for fastsleep only in oneshot mode of broadcast */
84 #ifdef CONFIG_TICK_ONESHOT
85 static int fastsleep_loop(struct cpuidle_device *dev,
86 struct cpuidle_driver *drv,
89 unsigned long old_lpcr = mfspr(SPRN_LPCR);
90 unsigned long new_lpcr;
92 if (unlikely(system_state < SYSTEM_RUNNING))
96 /* Do not exit powersave upon decrementer as we've setup the timer
99 new_lpcr &= ~LPCR_PECE1;
101 mtspr(SPRN_LPCR, new_lpcr);
104 mtspr(SPRN_LPCR, old_lpcr);
110 static int stop_loop(struct cpuidle_device *dev,
111 struct cpuidle_driver *drv,
114 ppc64_runlatch_off();
115 power9_idle_stop(stop_psscr_table[index].val,
116 stop_psscr_table[index].mask);
122 * States for dedicated partition case.
124 static struct cpuidle_state powernv_states[CPUIDLE_STATE_MAX] = {
129 .target_residency = 0,
130 .enter = snooze_loop },
133 static int powernv_cpuidle_cpu_online(unsigned int cpu)
135 struct cpuidle_device *dev = per_cpu(cpuidle_devices, cpu);
137 if (dev && cpuidle_get_driver()) {
138 cpuidle_pause_and_lock();
139 cpuidle_enable_device(dev);
140 cpuidle_resume_and_unlock();
145 static int powernv_cpuidle_cpu_dead(unsigned int cpu)
147 struct cpuidle_device *dev = per_cpu(cpuidle_devices, cpu);
149 if (dev && cpuidle_get_driver()) {
150 cpuidle_pause_and_lock();
151 cpuidle_disable_device(dev);
152 cpuidle_resume_and_unlock();
158 * powernv_cpuidle_driver_init()
160 static int powernv_cpuidle_driver_init(void)
163 struct cpuidle_driver *drv = &powernv_idle_driver;
165 drv->state_count = 0;
167 for (idle_state = 0; idle_state < max_idle_state; ++idle_state) {
168 /* Is the state not enabled? */
169 if (cpuidle_state_table[idle_state].enter == NULL)
172 drv->states[drv->state_count] = /* structure copy */
173 cpuidle_state_table[idle_state];
175 drv->state_count += 1;
179 * On the PowerNV platform cpu_present may be less than cpu_possible in
180 * cases when firmware detects the CPU, but it is not available to the
181 * OS. If CONFIG_HOTPLUG_CPU=n, then such CPUs are not hotplugable at
182 * run time and hence cpu_devices are not created for those CPUs by the
183 * generic topology_init().
185 * drv->cpumask defaults to cpu_possible_mask in
186 * __cpuidle_driver_init(). This breaks cpuidle on PowerNV where
187 * cpu_devices are not created for CPUs in cpu_possible_mask that
188 * cannot be hot-added later at run time.
190 * Trying cpuidle_register_device() on a CPU without a cpu_device is
191 * incorrect, so pass a correct CPU mask to the generic cpuidle driver.
194 drv->cpumask = (struct cpumask *)cpu_present_mask;
199 static inline void add_powernv_state(int index, const char *name,
201 int (*idle_fn)(struct cpuidle_device *,
202 struct cpuidle_driver *,
204 unsigned int target_residency,
205 unsigned int exit_latency,
206 u64 psscr_val, u64 psscr_mask)
208 strlcpy(powernv_states[index].name, name, CPUIDLE_NAME_LEN);
209 strlcpy(powernv_states[index].desc, name, CPUIDLE_NAME_LEN);
210 powernv_states[index].flags = flags;
211 powernv_states[index].target_residency = target_residency;
212 powernv_states[index].exit_latency = exit_latency;
213 powernv_states[index].enter = idle_fn;
214 stop_psscr_table[index].val = psscr_val;
215 stop_psscr_table[index].mask = psscr_mask;
218 static int powernv_add_idle_states(void)
220 struct device_node *power_mgt;
221 int nr_idle_states = 1; /* Snooze */
223 u32 latency_ns[CPUIDLE_STATE_MAX];
224 u32 residency_ns[CPUIDLE_STATE_MAX];
225 u32 flags[CPUIDLE_STATE_MAX];
226 u64 psscr_val[CPUIDLE_STATE_MAX];
227 u64 psscr_mask[CPUIDLE_STATE_MAX];
228 const char *names[CPUIDLE_STATE_MAX];
229 u32 has_stop_states = 0;
232 /* Currently we have snooze statically defined */
234 power_mgt = of_find_node_by_path("/ibm,opal/power-mgt");
236 pr_warn("opal: PowerMgmt Node not found\n");
240 /* Read values of any property to determine the num of idle states */
241 dt_idle_states = of_property_count_u32_elems(power_mgt, "ibm,cpu-idle-state-flags");
242 if (dt_idle_states < 0) {
243 pr_warn("cpuidle-powernv: no idle states found in the DT\n");
248 * Since snooze is used as first idle state, max idle states allowed is
249 * CPUIDLE_STATE_MAX -1
251 if (dt_idle_states > CPUIDLE_STATE_MAX - 1) {
252 pr_warn("cpuidle-powernv: discovered idle states more than allowed");
253 dt_idle_states = CPUIDLE_STATE_MAX - 1;
256 if (of_property_read_u32_array(power_mgt,
257 "ibm,cpu-idle-state-flags", flags, dt_idle_states)) {
258 pr_warn("cpuidle-powernv : missing ibm,cpu-idle-state-flags in DT\n");
262 if (of_property_read_u32_array(power_mgt,
263 "ibm,cpu-idle-state-latencies-ns", latency_ns,
265 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-latencies-ns in DT\n");
268 if (of_property_read_string_array(power_mgt,
269 "ibm,cpu-idle-state-names", names, dt_idle_states) < 0) {
270 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-names in DT\n");
275 * If the idle states use stop instruction, probe for psscr values
276 * and psscr mask which are necessary to specify required stop level.
278 has_stop_states = (flags[0] &
279 (OPAL_PM_STOP_INST_FAST | OPAL_PM_STOP_INST_DEEP));
280 if (has_stop_states) {
281 if (of_property_read_u64_array(power_mgt,
282 "ibm,cpu-idle-state-psscr", psscr_val, dt_idle_states)) {
283 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr in DT\n");
287 if (of_property_read_u64_array(power_mgt,
288 "ibm,cpu-idle-state-psscr-mask",
289 psscr_mask, dt_idle_states)) {
290 pr_warn("cpuidle-powernv:Missing ibm,cpu-idle-state-psscr-mask in DT\n");
295 rc = of_property_read_u32_array(power_mgt,
296 "ibm,cpu-idle-state-residency-ns", residency_ns, dt_idle_states);
298 for (i = 0; i < dt_idle_states; i++) {
299 unsigned int exit_latency, target_residency;
301 * If an idle state has exit latency beyond
302 * POWERNV_THRESHOLD_LATENCY_NS then don't use it
305 if (latency_ns[i] > POWERNV_THRESHOLD_LATENCY_NS)
308 * Firmware passes residency and latency values in ns.
309 * cpuidle expects it in us.
311 exit_latency = latency_ns[i] / 1000;
313 target_residency = residency_ns[i] / 1000;
315 target_residency = 0;
317 if (has_stop_states) {
318 int err = validate_psscr_val_mask(&psscr_val[i],
322 report_invalid_psscr_val(psscr_val[i], err);
328 * For nap and fastsleep, use default target_residency
329 * values if f/w does not expose it.
331 if (flags[i] & OPAL_PM_NAP_ENABLED) {
333 target_residency = 100;
335 add_powernv_state(nr_idle_states, "Nap",
336 CPUIDLE_FLAG_NONE, nap_loop,
337 target_residency, exit_latency, 0, 0);
338 } else if ((flags[i] & OPAL_PM_STOP_INST_FAST) &&
339 !(flags[i] & OPAL_PM_TIMEBASE_STOP)) {
340 add_powernv_state(nr_idle_states, names[i],
341 CPUIDLE_FLAG_NONE, stop_loop,
342 target_residency, exit_latency,
343 psscr_val[i], psscr_mask[i]);
347 * All cpuidle states with CPUIDLE_FLAG_TIMER_STOP set must come
348 * within this config dependency check.
350 #ifdef CONFIG_TICK_ONESHOT
351 if (flags[i] & OPAL_PM_SLEEP_ENABLED ||
352 flags[i] & OPAL_PM_SLEEP_ENABLED_ER1) {
354 target_residency = 300000;
355 /* Add FASTSLEEP state */
356 add_powernv_state(nr_idle_states, "FastSleep",
357 CPUIDLE_FLAG_TIMER_STOP,
359 target_residency, exit_latency, 0, 0);
360 } else if ((flags[i] & OPAL_PM_STOP_INST_DEEP) &&
361 (flags[i] & OPAL_PM_TIMEBASE_STOP)) {
362 add_powernv_state(nr_idle_states, names[i],
363 CPUIDLE_FLAG_TIMER_STOP, stop_loop,
364 target_residency, exit_latency,
365 psscr_val[i], psscr_mask[i]);
371 return nr_idle_states;
375 * powernv_idle_probe()
376 * Choose state table for shared versus dedicated partition
378 static int powernv_idle_probe(void)
380 if (cpuidle_disable != IDLE_NO_OVERRIDE)
383 if (firmware_has_feature(FW_FEATURE_OPAL)) {
384 cpuidle_state_table = powernv_states;
385 /* Device tree can indicate more idle states */
386 max_idle_state = powernv_add_idle_states();
387 if (max_idle_state > 1) {
388 snooze_timeout_en = true;
389 snooze_timeout = powernv_states[1].target_residency *
398 static int __init powernv_processor_idle_init(void)
402 retval = powernv_idle_probe();
406 powernv_cpuidle_driver_init();
407 retval = cpuidle_register(&powernv_idle_driver, NULL);
409 printk(KERN_DEBUG "Registration of powernv driver failed.\n");
413 retval = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
414 "cpuidle/powernv:online",
415 powernv_cpuidle_cpu_online, NULL);
417 retval = cpuhp_setup_state_nocalls(CPUHP_CPUIDLE_DEAD,
418 "cpuidle/powernv:dead", NULL,
419 powernv_cpuidle_cpu_dead);
421 printk(KERN_DEBUG "powernv_idle_driver registered\n");
425 device_initcall(powernv_processor_idle_init);