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;
60 while (!need_resched()) {
61 if (likely(snooze_timeout_en) && get_tb() > snooze_exit_time)
67 clear_thread_flag(TIF_POLLING_NRFLAG);
72 static int nap_loop(struct cpuidle_device *dev,
73 struct cpuidle_driver *drv,
82 /* Register for fastsleep only in oneshot mode of broadcast */
83 #ifdef CONFIG_TICK_ONESHOT
84 static int fastsleep_loop(struct cpuidle_device *dev,
85 struct cpuidle_driver *drv,
88 unsigned long old_lpcr = mfspr(SPRN_LPCR);
89 unsigned long new_lpcr;
91 if (unlikely(system_state < SYSTEM_RUNNING))
95 /* Do not exit powersave upon decrementer as we've setup the timer
98 new_lpcr &= ~LPCR_PECE1;
100 mtspr(SPRN_LPCR, new_lpcr);
103 mtspr(SPRN_LPCR, old_lpcr);
109 static int stop_loop(struct cpuidle_device *dev,
110 struct cpuidle_driver *drv,
113 ppc64_runlatch_off();
114 power9_idle_stop(stop_psscr_table[index].val,
115 stop_psscr_table[index].mask);
121 * States for dedicated partition case.
123 static struct cpuidle_state powernv_states[CPUIDLE_STATE_MAX] = {
128 .target_residency = 0,
129 .enter = snooze_loop },
132 static int powernv_cpuidle_cpu_online(unsigned int cpu)
134 struct cpuidle_device *dev = per_cpu(cpuidle_devices, cpu);
136 if (dev && cpuidle_get_driver()) {
137 cpuidle_pause_and_lock();
138 cpuidle_enable_device(dev);
139 cpuidle_resume_and_unlock();
144 static int powernv_cpuidle_cpu_dead(unsigned int cpu)
146 struct cpuidle_device *dev = per_cpu(cpuidle_devices, cpu);
148 if (dev && cpuidle_get_driver()) {
149 cpuidle_pause_and_lock();
150 cpuidle_disable_device(dev);
151 cpuidle_resume_and_unlock();
157 * powernv_cpuidle_driver_init()
159 static int powernv_cpuidle_driver_init(void)
162 struct cpuidle_driver *drv = &powernv_idle_driver;
164 drv->state_count = 0;
166 for (idle_state = 0; idle_state < max_idle_state; ++idle_state) {
167 /* Is the state not enabled? */
168 if (cpuidle_state_table[idle_state].enter == NULL)
171 drv->states[drv->state_count] = /* structure copy */
172 cpuidle_state_table[idle_state];
174 drv->state_count += 1;
178 * On the PowerNV platform cpu_present may be less than cpu_possible in
179 * cases when firmware detects the CPU, but it is not available to the
180 * OS. If CONFIG_HOTPLUG_CPU=n, then such CPUs are not hotplugable at
181 * run time and hence cpu_devices are not created for those CPUs by the
182 * generic topology_init().
184 * drv->cpumask defaults to cpu_possible_mask in
185 * __cpuidle_driver_init(). This breaks cpuidle on PowerNV where
186 * cpu_devices are not created for CPUs in cpu_possible_mask that
187 * cannot be hot-added later at run time.
189 * Trying cpuidle_register_device() on a CPU without a cpu_device is
190 * incorrect, so pass a correct CPU mask to the generic cpuidle driver.
193 drv->cpumask = (struct cpumask *)cpu_present_mask;
198 static inline void add_powernv_state(int index, const char *name,
200 int (*idle_fn)(struct cpuidle_device *,
201 struct cpuidle_driver *,
203 unsigned int target_residency,
204 unsigned int exit_latency,
205 u64 psscr_val, u64 psscr_mask)
207 strlcpy(powernv_states[index].name, name, CPUIDLE_NAME_LEN);
208 strlcpy(powernv_states[index].desc, name, CPUIDLE_NAME_LEN);
209 powernv_states[index].flags = flags;
210 powernv_states[index].target_residency = target_residency;
211 powernv_states[index].exit_latency = exit_latency;
212 powernv_states[index].enter = idle_fn;
213 stop_psscr_table[index].val = psscr_val;
214 stop_psscr_table[index].mask = psscr_mask;
218 * Returns 0 if prop1_len == prop2_len. Else returns -1
220 static inline int validate_dt_prop_sizes(const char *prop1, int prop1_len,
221 const char *prop2, int prop2_len)
223 if (prop1_len == prop2_len)
226 pr_warn("cpuidle-powernv: array sizes don't match for %s and %s\n",
231 static int powernv_add_idle_states(void)
233 struct device_node *power_mgt;
234 int nr_idle_states = 1; /* Snooze */
235 int dt_idle_states, count;
236 u32 latency_ns[CPUIDLE_STATE_MAX];
237 u32 residency_ns[CPUIDLE_STATE_MAX];
238 u32 flags[CPUIDLE_STATE_MAX];
239 u64 psscr_val[CPUIDLE_STATE_MAX];
240 u64 psscr_mask[CPUIDLE_STATE_MAX];
241 const char *names[CPUIDLE_STATE_MAX];
242 u32 has_stop_states = 0;
245 /* Currently we have snooze statically defined */
247 power_mgt = of_find_node_by_path("/ibm,opal/power-mgt");
249 pr_warn("opal: PowerMgmt Node not found\n");
253 /* Read values of any property to determine the num of idle states */
254 dt_idle_states = of_property_count_u32_elems(power_mgt, "ibm,cpu-idle-state-flags");
255 if (dt_idle_states < 0) {
256 pr_warn("cpuidle-powernv: no idle states found in the DT\n");
260 count = of_property_count_u32_elems(power_mgt,
261 "ibm,cpu-idle-state-latencies-ns");
263 if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags", dt_idle_states,
264 "ibm,cpu-idle-state-latencies-ns",
268 count = of_property_count_strings(power_mgt,
269 "ibm,cpu-idle-state-names");
270 if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags", dt_idle_states,
271 "ibm,cpu-idle-state-names",
276 * Since snooze is used as first idle state, max idle states allowed is
277 * CPUIDLE_STATE_MAX -1
279 if (dt_idle_states > CPUIDLE_STATE_MAX - 1) {
280 pr_warn("cpuidle-powernv: discovered idle states more than allowed");
281 dt_idle_states = CPUIDLE_STATE_MAX - 1;
284 if (of_property_read_u32_array(power_mgt,
285 "ibm,cpu-idle-state-flags", flags, dt_idle_states)) {
286 pr_warn("cpuidle-powernv : missing ibm,cpu-idle-state-flags in DT\n");
290 if (of_property_read_u32_array(power_mgt,
291 "ibm,cpu-idle-state-latencies-ns", latency_ns,
293 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-latencies-ns in DT\n");
296 if (of_property_read_string_array(power_mgt,
297 "ibm,cpu-idle-state-names", names, dt_idle_states) < 0) {
298 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-names in DT\n");
303 * If the idle states use stop instruction, probe for psscr values
304 * and psscr mask which are necessary to specify required stop level.
306 has_stop_states = (flags[0] &
307 (OPAL_PM_STOP_INST_FAST | OPAL_PM_STOP_INST_DEEP));
308 if (has_stop_states) {
309 count = of_property_count_u64_elems(power_mgt,
310 "ibm,cpu-idle-state-psscr");
311 if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags",
313 "ibm,cpu-idle-state-psscr",
317 count = of_property_count_u64_elems(power_mgt,
318 "ibm,cpu-idle-state-psscr-mask");
319 if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags",
321 "ibm,cpu-idle-state-psscr-mask",
325 if (of_property_read_u64_array(power_mgt,
326 "ibm,cpu-idle-state-psscr", psscr_val, dt_idle_states)) {
327 pr_warn("cpuidle-powernv: missing ibm,cpu-idle-state-psscr in DT\n");
331 if (of_property_read_u64_array(power_mgt,
332 "ibm,cpu-idle-state-psscr-mask",
333 psscr_mask, dt_idle_states)) {
334 pr_warn("cpuidle-powernv:Missing ibm,cpu-idle-state-psscr-mask in DT\n");
339 count = of_property_count_u32_elems(power_mgt,
340 "ibm,cpu-idle-state-residency-ns");
344 } else if (validate_dt_prop_sizes("ibm,cpu-idle-state-flags",
346 "ibm,cpu-idle-state-residency-ns",
350 rc = of_property_read_u32_array(power_mgt,
351 "ibm,cpu-idle-state-residency-ns",
352 residency_ns, dt_idle_states);
355 for (i = 0; i < dt_idle_states; i++) {
356 unsigned int exit_latency, target_residency;
358 * If an idle state has exit latency beyond
359 * POWERNV_THRESHOLD_LATENCY_NS then don't use it
362 if (latency_ns[i] > POWERNV_THRESHOLD_LATENCY_NS)
365 * Firmware passes residency and latency values in ns.
366 * cpuidle expects it in us.
368 exit_latency = latency_ns[i] / 1000;
370 target_residency = residency_ns[i] / 1000;
372 target_residency = 0;
374 if (has_stop_states) {
375 int err = validate_psscr_val_mask(&psscr_val[i],
379 report_invalid_psscr_val(psscr_val[i], err);
385 * For nap and fastsleep, use default target_residency
386 * values if f/w does not expose it.
388 if (flags[i] & OPAL_PM_NAP_ENABLED) {
390 target_residency = 100;
392 add_powernv_state(nr_idle_states, "Nap",
393 CPUIDLE_FLAG_NONE, nap_loop,
394 target_residency, exit_latency, 0, 0);
395 } else if ((flags[i] & OPAL_PM_STOP_INST_FAST) &&
396 !(flags[i] & OPAL_PM_TIMEBASE_STOP)) {
397 add_powernv_state(nr_idle_states, names[i],
398 CPUIDLE_FLAG_NONE, stop_loop,
399 target_residency, exit_latency,
400 psscr_val[i], psscr_mask[i]);
404 * All cpuidle states with CPUIDLE_FLAG_TIMER_STOP set must come
405 * within this config dependency check.
407 #ifdef CONFIG_TICK_ONESHOT
408 if (flags[i] & OPAL_PM_SLEEP_ENABLED ||
409 flags[i] & OPAL_PM_SLEEP_ENABLED_ER1) {
411 target_residency = 300000;
412 /* Add FASTSLEEP state */
413 add_powernv_state(nr_idle_states, "FastSleep",
414 CPUIDLE_FLAG_TIMER_STOP,
416 target_residency, exit_latency, 0, 0);
417 } else if ((flags[i] & OPAL_PM_STOP_INST_DEEP) &&
418 (flags[i] & OPAL_PM_TIMEBASE_STOP)) {
419 add_powernv_state(nr_idle_states, names[i],
420 CPUIDLE_FLAG_TIMER_STOP, stop_loop,
421 target_residency, exit_latency,
422 psscr_val[i], psscr_mask[i]);
428 return nr_idle_states;
432 * powernv_idle_probe()
433 * Choose state table for shared versus dedicated partition
435 static int powernv_idle_probe(void)
437 if (cpuidle_disable != IDLE_NO_OVERRIDE)
440 if (firmware_has_feature(FW_FEATURE_OPAL)) {
441 cpuidle_state_table = powernv_states;
442 /* Device tree can indicate more idle states */
443 max_idle_state = powernv_add_idle_states();
444 if (max_idle_state > 1) {
445 snooze_timeout_en = true;
446 snooze_timeout = powernv_states[1].target_residency *
455 static int __init powernv_processor_idle_init(void)
459 retval = powernv_idle_probe();
463 powernv_cpuidle_driver_init();
464 retval = cpuidle_register(&powernv_idle_driver, NULL);
466 printk(KERN_DEBUG "Registration of powernv driver failed.\n");
470 retval = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
471 "cpuidle/powernv:online",
472 powernv_cpuidle_cpu_online, NULL);
474 retval = cpuhp_setup_state_nocalls(CPUHP_CPUIDLE_DEAD,
475 "cpuidle/powernv:dead", NULL,
476 powernv_cpuidle_cpu_dead);
478 printk(KERN_DEBUG "powernv_idle_driver registered\n");
482 device_initcall(powernv_processor_idle_init);