1 // SPDX-License-Identifier: GPL-2.0
3 * cpuidle-powernv - idle state cpuidle driver.
4 * Adapted from drivers/cpuidle/cpuidle-pseries
8 #include <linux/kernel.h>
9 #include <linux/module.h>
10 #include <linux/init.h>
11 #include <linux/moduleparam.h>
12 #include <linux/cpuidle.h>
13 #include <linux/cpu.h>
14 #include <linux/notifier.h>
15 #include <linux/clockchips.h>
17 #include <linux/slab.h>
19 #include <asm/machdep.h>
20 #include <asm/firmware.h>
22 #include <asm/runlatch.h>
23 #include <asm/cpuidle.h>
26 * Expose only those Hardware idle states via the cpuidle framework
27 * that have latency value below POWERNV_THRESHOLD_LATENCY_NS.
29 #define POWERNV_THRESHOLD_LATENCY_NS 200000
31 static struct cpuidle_driver powernv_idle_driver
= {
32 .name
= "powernv_idle",
36 static int max_idle_state __read_mostly
;
37 static struct cpuidle_state
*cpuidle_state_table __read_mostly
;
39 struct stop_psscr_table
{
44 static struct stop_psscr_table stop_psscr_table
[CPUIDLE_STATE_MAX
] __read_mostly
;
46 static u64 default_snooze_timeout __read_mostly
;
47 static bool snooze_timeout_en __read_mostly
;
49 static u64
get_snooze_timeout(struct cpuidle_device
*dev
,
50 struct cpuidle_driver
*drv
,
55 if (unlikely(!snooze_timeout_en
))
56 return default_snooze_timeout
;
58 for (i
= index
+ 1; i
< drv
->state_count
; i
++) {
59 struct cpuidle_state
*s
= &drv
->states
[i
];
60 struct cpuidle_state_usage
*su
= &dev
->states_usage
[i
];
62 if (s
->disabled
|| su
->disable
)
65 return s
->target_residency
* tb_ticks_per_usec
;
68 return default_snooze_timeout
;
71 static int snooze_loop(struct cpuidle_device
*dev
,
72 struct cpuidle_driver
*drv
,
77 set_thread_flag(TIF_POLLING_NRFLAG
);
81 snooze_exit_time
= get_tb() + get_snooze_timeout(dev
, drv
, index
);
84 while (!need_resched()) {
85 if (likely(snooze_timeout_en
) && get_tb() > snooze_exit_time
) {
87 * Task has not woken up but we are exiting the polling
88 * loop anyway. Require a barrier after polling is
89 * cleared to order subsequent test of need_resched().
91 clear_thread_flag(TIF_POLLING_NRFLAG
);
99 clear_thread_flag(TIF_POLLING_NRFLAG
);
106 static int nap_loop(struct cpuidle_device
*dev
,
107 struct cpuidle_driver
*drv
,
110 power7_idle_type(PNV_THREAD_NAP
);
115 /* Register for fastsleep only in oneshot mode of broadcast */
116 #ifdef CONFIG_TICK_ONESHOT
117 static int fastsleep_loop(struct cpuidle_device
*dev
,
118 struct cpuidle_driver
*drv
,
121 unsigned long old_lpcr
= mfspr(SPRN_LPCR
);
122 unsigned long new_lpcr
;
124 if (unlikely(system_state
< SYSTEM_RUNNING
))
128 /* Do not exit powersave upon decrementer as we've setup the timer
131 new_lpcr
&= ~LPCR_PECE1
;
133 mtspr(SPRN_LPCR
, new_lpcr
);
135 power7_idle_type(PNV_THREAD_SLEEP
);
137 mtspr(SPRN_LPCR
, old_lpcr
);
143 static int stop_loop(struct cpuidle_device
*dev
,
144 struct cpuidle_driver
*drv
,
147 power9_idle_type(stop_psscr_table
[index
].val
,
148 stop_psscr_table
[index
].mask
);
153 * States for dedicated partition case.
155 static struct cpuidle_state powernv_states
[CPUIDLE_STATE_MAX
] = {
160 .target_residency
= 0,
161 .enter
= snooze_loop
},
164 static int powernv_cpuidle_cpu_online(unsigned int cpu
)
166 struct cpuidle_device
*dev
= per_cpu(cpuidle_devices
, cpu
);
168 if (dev
&& cpuidle_get_driver()) {
169 cpuidle_pause_and_lock();
170 cpuidle_enable_device(dev
);
171 cpuidle_resume_and_unlock();
176 static int powernv_cpuidle_cpu_dead(unsigned int cpu
)
178 struct cpuidle_device
*dev
= per_cpu(cpuidle_devices
, cpu
);
180 if (dev
&& cpuidle_get_driver()) {
181 cpuidle_pause_and_lock();
182 cpuidle_disable_device(dev
);
183 cpuidle_resume_and_unlock();
189 * powernv_cpuidle_driver_init()
191 static int powernv_cpuidle_driver_init(void)
194 struct cpuidle_driver
*drv
= &powernv_idle_driver
;
196 drv
->state_count
= 0;
198 for (idle_state
= 0; idle_state
< max_idle_state
; ++idle_state
) {
199 /* Is the state not enabled? */
200 if (cpuidle_state_table
[idle_state
].enter
== NULL
)
203 drv
->states
[drv
->state_count
] = /* structure copy */
204 cpuidle_state_table
[idle_state
];
206 drv
->state_count
+= 1;
210 * On the PowerNV platform cpu_present may be less than cpu_possible in
211 * cases when firmware detects the CPU, but it is not available to the
212 * OS. If CONFIG_HOTPLUG_CPU=n, then such CPUs are not hotplugable at
213 * run time and hence cpu_devices are not created for those CPUs by the
214 * generic topology_init().
216 * drv->cpumask defaults to cpu_possible_mask in
217 * __cpuidle_driver_init(). This breaks cpuidle on PowerNV where
218 * cpu_devices are not created for CPUs in cpu_possible_mask that
219 * cannot be hot-added later at run time.
221 * Trying cpuidle_register_device() on a CPU without a cpu_device is
222 * incorrect, so pass a correct CPU mask to the generic cpuidle driver.
225 drv
->cpumask
= (struct cpumask
*)cpu_present_mask
;
230 static inline void add_powernv_state(int index
, const char *name
,
232 int (*idle_fn
)(struct cpuidle_device
*,
233 struct cpuidle_driver
*,
235 unsigned int target_residency
,
236 unsigned int exit_latency
,
237 u64 psscr_val
, u64 psscr_mask
)
239 strlcpy(powernv_states
[index
].name
, name
, CPUIDLE_NAME_LEN
);
240 strlcpy(powernv_states
[index
].desc
, name
, CPUIDLE_NAME_LEN
);
241 powernv_states
[index
].flags
= flags
;
242 powernv_states
[index
].target_residency
= target_residency
;
243 powernv_states
[index
].exit_latency
= exit_latency
;
244 powernv_states
[index
].enter
= idle_fn
;
245 /* For power8 and below psscr_* will be 0 */
246 stop_psscr_table
[index
].val
= psscr_val
;
247 stop_psscr_table
[index
].mask
= psscr_mask
;
251 * Returns 0 if prop1_len == prop2_len. Else returns -1
253 static inline int validate_dt_prop_sizes(const char *prop1
, int prop1_len
,
254 const char *prop2
, int prop2_len
)
256 if (prop1_len
== prop2_len
)
259 pr_warn("cpuidle-powernv: array sizes don't match for %s and %s\n",
264 extern u32
pnv_get_supported_cpuidle_states(void);
265 static int powernv_add_idle_states(void)
267 int nr_idle_states
= 1; /* Snooze */
269 u32 has_stop_states
= 0;
271 u32 supported_flags
= pnv_get_supported_cpuidle_states();
274 /* Currently we have snooze statically defined */
275 if (nr_pnv_idle_states
<= 0) {
276 pr_warn("cpuidle-powernv : Only Snooze is available\n");
280 /* TODO: Count only states which are eligible for cpuidle */
281 dt_idle_states
= nr_pnv_idle_states
;
284 * Since snooze is used as first idle state, max idle states allowed is
285 * CPUIDLE_STATE_MAX -1
287 if (nr_pnv_idle_states
> CPUIDLE_STATE_MAX
- 1) {
288 pr_warn("cpuidle-powernv: discovered idle states more than allowed");
289 dt_idle_states
= CPUIDLE_STATE_MAX
- 1;
293 * If the idle states use stop instruction, probe for psscr values
294 * and psscr mask which are necessary to specify required stop level.
296 has_stop_states
= (pnv_idle_states
[0].flags
&
297 (OPAL_PM_STOP_INST_FAST
| OPAL_PM_STOP_INST_DEEP
));
299 for (i
= 0; i
< dt_idle_states
; i
++) {
300 unsigned int exit_latency
, target_residency
;
301 bool stops_timebase
= false;
302 struct pnv_idle_states_t
*state
= &pnv_idle_states
[i
];
305 * Skip the platform idle state whose flag isn't in
306 * the supported_cpuidle_states flag mask.
308 if ((state
->flags
& supported_flags
) != state
->flags
)
311 * If an idle state has exit latency beyond
312 * POWERNV_THRESHOLD_LATENCY_NS then don't use it
315 if (state
->latency_ns
> POWERNV_THRESHOLD_LATENCY_NS
)
318 * Firmware passes residency and latency values in ns.
319 * cpuidle expects it in us.
321 exit_latency
= DIV_ROUND_UP(state
->latency_ns
, 1000);
322 target_residency
= DIV_ROUND_UP(state
->residency_ns
, 1000);
324 if (has_stop_states
&& !(state
->valid
))
327 if (state
->flags
& OPAL_PM_TIMEBASE_STOP
)
328 stops_timebase
= true;
330 if (state
->flags
& OPAL_PM_NAP_ENABLED
) {
332 add_powernv_state(nr_idle_states
, "Nap",
333 CPUIDLE_FLAG_NONE
, nap_loop
,
334 target_residency
, exit_latency
, 0, 0);
335 } else if (has_stop_states
&& !stops_timebase
) {
336 add_powernv_state(nr_idle_states
, state
->name
,
337 CPUIDLE_FLAG_NONE
, stop_loop
,
338 target_residency
, exit_latency
,
344 * All cpuidle states with CPUIDLE_FLAG_TIMER_STOP set must come
345 * within this config dependency check.
347 #ifdef CONFIG_TICK_ONESHOT
348 else if (state
->flags
& OPAL_PM_SLEEP_ENABLED
||
349 state
->flags
& OPAL_PM_SLEEP_ENABLED_ER1
) {
350 /* Add FASTSLEEP state */
351 add_powernv_state(nr_idle_states
, "FastSleep",
352 CPUIDLE_FLAG_TIMER_STOP
,
354 target_residency
, exit_latency
, 0, 0);
355 } else if (has_stop_states
&& stops_timebase
) {
356 add_powernv_state(nr_idle_states
, state
->name
,
357 CPUIDLE_FLAG_TIMER_STOP
, stop_loop
,
358 target_residency
, exit_latency
,
368 return nr_idle_states
;
372 * powernv_idle_probe()
373 * Choose state table for shared versus dedicated partition
375 static int powernv_idle_probe(void)
377 if (cpuidle_disable
!= IDLE_NO_OVERRIDE
)
380 if (firmware_has_feature(FW_FEATURE_OPAL
)) {
381 cpuidle_state_table
= powernv_states
;
382 /* Device tree can indicate more idle states */
383 max_idle_state
= powernv_add_idle_states();
384 default_snooze_timeout
= TICK_USEC
* tb_ticks_per_usec
;
385 if (max_idle_state
> 1)
386 snooze_timeout_en
= true;
393 static int __init
powernv_processor_idle_init(void)
397 retval
= powernv_idle_probe();
401 powernv_cpuidle_driver_init();
402 retval
= cpuidle_register(&powernv_idle_driver
, NULL
);
404 printk(KERN_DEBUG
"Registration of powernv driver failed.\n");
408 retval
= cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN
,
409 "cpuidle/powernv:online",
410 powernv_cpuidle_cpu_online
, NULL
);
412 retval
= cpuhp_setup_state_nocalls(CPUHP_CPUIDLE_DEAD
,
413 "cpuidle/powernv:dead", NULL
,
414 powernv_cpuidle_cpu_dead
);
416 printk(KERN_DEBUG
"powernv_idle_driver registered\n");
420 device_initcall(powernv_processor_idle_init
);