2 * POWERNV cpufreq driver for the IBM POWER processors
4 * (C) Copyright IBM 2014
6 * Author: Vaidyanathan Srinivasan <svaidy at linux.vnet.ibm.com>
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2, or (at your option)
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
20 #define pr_fmt(fmt) "powernv-cpufreq: " fmt
22 #include <linux/kernel.h>
23 #include <linux/sysfs.h>
24 #include <linux/cpumask.h>
25 #include <linux/module.h>
26 #include <linux/cpufreq.h>
27 #include <linux/smp.h>
29 #include <linux/reboot.h>
30 #include <linux/slab.h>
31 #include <linux/cpu.h>
32 #include <trace/events/power.h>
34 #include <asm/cputhreads.h>
35 #include <asm/firmware.h>
37 #include <asm/smp.h> /* Required for cpu_sibling_mask() in UP configs */
39 #include <linux/timer.h>
41 #define POWERNV_MAX_PSTATES 256
42 #define PMSR_PSAFE_ENABLE (1UL << 30)
43 #define PMSR_SPR_EM_DISABLE (1UL << 31)
44 #define PMSR_MAX(x) ((x >> 32) & 0xFF)
46 #define MAX_RAMP_DOWN_TIME 5120
48 * On an idle system we want the global pstate to ramp-down from max value to
49 * min over a span of ~5 secs. Also we want it to initially ramp-down slowly and
50 * then ramp-down rapidly later on.
52 * This gives a percentage rampdown for time elapsed in milliseconds.
53 * ramp_down_percentage = ((ms * ms) >> 18)
54 * ~= 3.8 * (sec * sec)
56 * At 0 ms ramp_down_percent = 0
57 * At 5120 ms ramp_down_percent = 100
59 #define ramp_down_percent(time) ((time * time) >> 18)
61 /* Interval after which the timer is queued to bring down global pstate */
62 #define GPSTATE_TIMER_INTERVAL 2000
65 * struct global_pstate_info - Per policy data structure to maintain history of
67 * @highest_lpstate_idx: The local pstate index from which we are
69 * @elapsed_time: Time in ms spent in ramping down from
71 * @last_sampled_time: Time from boot in ms when global pstates were
73 * @last_lpstate_idx, Last set value of local pstate and global
74 * last_gpstate_idx pstate in terms of cpufreq table index
75 * @timer: Is used for ramping down if cpu goes idle for
76 * a long time with global pstate held high
77 * @gpstate_lock: A spinlock to maintain synchronization between
78 * routines called by the timer handler and
79 * governer's target_index calls
81 struct global_pstate_info
{
82 int highest_lpstate_idx
;
83 unsigned int elapsed_time
;
84 unsigned int last_sampled_time
;
87 spinlock_t gpstate_lock
;
88 struct timer_list timer
;
91 static struct cpufreq_frequency_table powernv_freqs
[POWERNV_MAX_PSTATES
+1];
92 static bool rebooting
, throttled
, occ_reset
;
94 static const char * const throttle_reason
[] = {
97 "Processor Over Temperature",
98 "Power Supply Failure",
103 enum throttle_reason_type
{
107 POWER_SUPPLY_FAILURE
,
119 struct work_struct throttle
;
121 int throttle_sub_turbo
;
122 int reason
[OCC_MAX_REASON
];
126 static DEFINE_PER_CPU(struct chip
*, chip_info
);
130 * The set of pstates consists of contiguous integers.
131 * powernv_pstate_info stores the index of the frequency table for
132 * max, min and nominal frequencies. It also stores number of
133 * available frequencies.
135 * powernv_pstate_info.nominal indicates the index to the highest
136 * non-turbo frequency.
138 static struct powernv_pstate_info
{
141 unsigned int nominal
;
142 unsigned int nr_pstates
;
143 } powernv_pstate_info
;
145 /* Use following macros for conversions between pstate_id and index */
146 static inline int idx_to_pstate(unsigned int i
)
148 return powernv_freqs
[i
].driver_data
;
151 static inline unsigned int pstate_to_idx(int pstate
)
154 * abs() is deliberately used so that is works with
155 * both monotonically increasing and decreasing
158 return abs(pstate
- idx_to_pstate(powernv_pstate_info
.max
));
161 static inline void reset_gpstates(struct cpufreq_policy
*policy
)
163 struct global_pstate_info
*gpstates
= policy
->driver_data
;
165 gpstates
->highest_lpstate_idx
= 0;
166 gpstates
->elapsed_time
= 0;
167 gpstates
->last_sampled_time
= 0;
168 gpstates
->last_lpstate_idx
= 0;
169 gpstates
->last_gpstate_idx
= 0;
173 * Initialize the freq table based on data obtained
174 * from the firmware passed via device-tree
176 static int init_powernv_pstates(void)
178 struct device_node
*power_mgt
;
179 int i
, nr_pstates
= 0;
180 const __be32
*pstate_ids
, *pstate_freqs
;
181 u32 len_ids
, len_freqs
;
182 u32 pstate_min
, pstate_max
, pstate_nominal
;
184 power_mgt
= of_find_node_by_path("/ibm,opal/power-mgt");
186 pr_warn("power-mgt node not found\n");
190 if (of_property_read_u32(power_mgt
, "ibm,pstate-min", &pstate_min
)) {
191 pr_warn("ibm,pstate-min node not found\n");
195 if (of_property_read_u32(power_mgt
, "ibm,pstate-max", &pstate_max
)) {
196 pr_warn("ibm,pstate-max node not found\n");
200 if (of_property_read_u32(power_mgt
, "ibm,pstate-nominal",
202 pr_warn("ibm,pstate-nominal not found\n");
205 pr_info("cpufreq pstate min %d nominal %d max %d\n", pstate_min
,
206 pstate_nominal
, pstate_max
);
208 pstate_ids
= of_get_property(power_mgt
, "ibm,pstate-ids", &len_ids
);
210 pr_warn("ibm,pstate-ids not found\n");
214 pstate_freqs
= of_get_property(power_mgt
, "ibm,pstate-frequencies-mhz",
217 pr_warn("ibm,pstate-frequencies-mhz not found\n");
221 if (len_ids
!= len_freqs
) {
222 pr_warn("Entries in ibm,pstate-ids and "
223 "ibm,pstate-frequencies-mhz does not match\n");
226 nr_pstates
= min(len_ids
, len_freqs
) / sizeof(u32
);
228 pr_warn("No PStates found\n");
232 powernv_pstate_info
.nr_pstates
= nr_pstates
;
233 pr_debug("NR PStates %d\n", nr_pstates
);
234 for (i
= 0; i
< nr_pstates
; i
++) {
235 u32 id
= be32_to_cpu(pstate_ids
[i
]);
236 u32 freq
= be32_to_cpu(pstate_freqs
[i
]);
238 pr_debug("PState id %d freq %d MHz\n", id
, freq
);
239 powernv_freqs
[i
].frequency
= freq
* 1000; /* kHz */
240 powernv_freqs
[i
].driver_data
= id
;
242 if (id
== pstate_max
)
243 powernv_pstate_info
.max
= i
;
244 else if (id
== pstate_nominal
)
245 powernv_pstate_info
.nominal
= i
;
246 else if (id
== pstate_min
)
247 powernv_pstate_info
.min
= i
;
250 /* End of list marker entry */
251 powernv_freqs
[i
].frequency
= CPUFREQ_TABLE_END
;
255 /* Returns the CPU frequency corresponding to the pstate_id. */
256 static unsigned int pstate_id_to_freq(int pstate_id
)
260 i
= pstate_to_idx(pstate_id
);
261 if (i
>= powernv_pstate_info
.nr_pstates
|| i
< 0) {
262 pr_warn("PState id %d outside of PState table, "
263 "reporting nominal id %d instead\n",
264 pstate_id
, idx_to_pstate(powernv_pstate_info
.nominal
));
265 i
= powernv_pstate_info
.nominal
;
268 return powernv_freqs
[i
].frequency
;
272 * cpuinfo_nominal_freq_show - Show the nominal CPU frequency as indicated by
275 static ssize_t
cpuinfo_nominal_freq_show(struct cpufreq_policy
*policy
,
278 return sprintf(buf
, "%u\n",
279 powernv_freqs
[powernv_pstate_info
.nominal
].frequency
);
282 struct freq_attr cpufreq_freq_attr_cpuinfo_nominal_freq
=
283 __ATTR_RO(cpuinfo_nominal_freq
);
285 static struct freq_attr
*powernv_cpu_freq_attr
[] = {
286 &cpufreq_freq_attr_scaling_available_freqs
,
287 &cpufreq_freq_attr_cpuinfo_nominal_freq
,
291 #define throttle_attr(name, member) \
292 static ssize_t name##_show(struct cpufreq_policy *policy, char *buf) \
294 struct chip *chip = per_cpu(chip_info, policy->cpu); \
296 return sprintf(buf, "%u\n", chip->member); \
299 static struct freq_attr throttle_attr_##name = __ATTR_RO(name) \
301 throttle_attr(unthrottle, reason[NO_THROTTLE]);
302 throttle_attr(powercap
, reason
[POWERCAP
]);
303 throttle_attr(overtemp
, reason
[CPU_OVERTEMP
]);
304 throttle_attr(supply_fault
, reason
[POWER_SUPPLY_FAILURE
]);
305 throttle_attr(overcurrent
, reason
[OVERCURRENT
]);
306 throttle_attr(occ_reset
, reason
[OCC_RESET_THROTTLE
]);
307 throttle_attr(turbo_stat
, throttle_turbo
);
308 throttle_attr(sub_turbo_stat
, throttle_sub_turbo
);
310 static struct attribute
*throttle_attrs
[] = {
311 &throttle_attr_unthrottle
.attr
,
312 &throttle_attr_powercap
.attr
,
313 &throttle_attr_overtemp
.attr
,
314 &throttle_attr_supply_fault
.attr
,
315 &throttle_attr_overcurrent
.attr
,
316 &throttle_attr_occ_reset
.attr
,
317 &throttle_attr_turbo_stat
.attr
,
318 &throttle_attr_sub_turbo_stat
.attr
,
322 static const struct attribute_group throttle_attr_grp
= {
323 .name
= "throttle_stats",
324 .attrs
= throttle_attrs
,
327 /* Helper routines */
329 /* Access helpers to power mgt SPR */
331 static inline unsigned long get_pmspr(unsigned long sprn
)
335 return mfspr(SPRN_PMCR
);
338 return mfspr(SPRN_PMICR
);
341 return mfspr(SPRN_PMSR
);
346 static inline void set_pmspr(unsigned long sprn
, unsigned long val
)
350 mtspr(SPRN_PMCR
, val
);
354 mtspr(SPRN_PMICR
, val
);
361 * Use objects of this type to query/update
362 * pstates on a remote CPU via smp_call_function.
364 struct powernv_smp_call_data
{
371 * powernv_read_cpu_freq: Reads the current frequency on this CPU.
373 * Called via smp_call_function.
375 * Note: The caller of the smp_call_function should pass an argument of
376 * the type 'struct powernv_smp_call_data *' along with this function.
378 * The current frequency on this CPU will be returned via
379 * ((struct powernv_smp_call_data *)arg)->freq;
381 static void powernv_read_cpu_freq(void *arg
)
383 unsigned long pmspr_val
;
385 struct powernv_smp_call_data
*freq_data
= arg
;
387 pmspr_val
= get_pmspr(SPRN_PMSR
);
390 * The local pstate id corresponds bits 48..55 in the PMSR.
391 * Note: Watch out for the sign!
393 local_pstate_id
= (pmspr_val
>> 48) & 0xFF;
394 freq_data
->pstate_id
= local_pstate_id
;
395 freq_data
->freq
= pstate_id_to_freq(freq_data
->pstate_id
);
397 pr_debug("cpu %d pmsr %016lX pstate_id %d frequency %d kHz\n",
398 raw_smp_processor_id(), pmspr_val
, freq_data
->pstate_id
,
403 * powernv_cpufreq_get: Returns the CPU frequency as reported by the
404 * firmware for CPU 'cpu'. This value is reported through the sysfs
405 * file cpuinfo_cur_freq.
407 static unsigned int powernv_cpufreq_get(unsigned int cpu
)
409 struct powernv_smp_call_data freq_data
;
411 smp_call_function_any(cpu_sibling_mask(cpu
), powernv_read_cpu_freq
,
414 return freq_data
.freq
;
418 * set_pstate: Sets the pstate on this CPU.
420 * This is called via an smp_call_function.
422 * The caller must ensure that freq_data is of the type
423 * (struct powernv_smp_call_data *) and the pstate_id which needs to be set
424 * on this CPU should be present in freq_data->pstate_id.
426 static void set_pstate(void *data
)
429 struct powernv_smp_call_data
*freq_data
= data
;
430 unsigned long pstate_ul
= freq_data
->pstate_id
;
431 unsigned long gpstate_ul
= freq_data
->gpstate_id
;
433 val
= get_pmspr(SPRN_PMCR
);
434 val
= val
& 0x0000FFFFFFFFFFFFULL
;
436 pstate_ul
= pstate_ul
& 0xFF;
437 gpstate_ul
= gpstate_ul
& 0xFF;
439 /* Set both global(bits 56..63) and local(bits 48..55) PStates */
440 val
= val
| (gpstate_ul
<< 56) | (pstate_ul
<< 48);
442 pr_debug("Setting cpu %d pmcr to %016lX\n",
443 raw_smp_processor_id(), val
);
444 set_pmspr(SPRN_PMCR
, val
);
448 * get_nominal_index: Returns the index corresponding to the nominal
449 * pstate in the cpufreq table
451 static inline unsigned int get_nominal_index(void)
453 return powernv_pstate_info
.nominal
;
456 static void powernv_cpufreq_throttle_check(void *data
)
459 unsigned int cpu
= smp_processor_id();
462 unsigned int pmsr_pmax_idx
;
464 pmsr
= get_pmspr(SPRN_PMSR
);
465 chip
= this_cpu_read(chip_info
);
467 /* Check for Pmax Capping */
468 pmsr_pmax
= (s8
)PMSR_MAX(pmsr
);
469 pmsr_pmax_idx
= pstate_to_idx(pmsr_pmax
);
470 if (pmsr_pmax_idx
!= powernv_pstate_info
.max
) {
473 chip
->throttled
= true;
474 if (pmsr_pmax_idx
> powernv_pstate_info
.nominal
) {
475 pr_warn_once("CPU %d on Chip %u has Pmax(%d) reduced below nominal frequency(%d)\n",
476 cpu
, chip
->id
, pmsr_pmax
,
477 idx_to_pstate(powernv_pstate_info
.nominal
));
478 chip
->throttle_sub_turbo
++;
480 chip
->throttle_turbo
++;
482 trace_powernv_throttle(chip
->id
,
483 throttle_reason
[chip
->throttle_reason
],
485 } else if (chip
->throttled
) {
486 chip
->throttled
= false;
487 trace_powernv_throttle(chip
->id
,
488 throttle_reason
[chip
->throttle_reason
],
492 /* Check if Psafe_mode_active is set in PMSR. */
494 if (pmsr
& PMSR_PSAFE_ENABLE
) {
496 pr_info("Pstate set to safe frequency\n");
499 /* Check if SPR_EM_DISABLE is set in PMSR */
500 if (pmsr
& PMSR_SPR_EM_DISABLE
) {
502 pr_info("Frequency Control disabled from OS\n");
506 pr_info("PMSR = %16lx\n", pmsr
);
507 pr_warn("CPU Frequency could be throttled\n");
512 * calc_global_pstate - Calculate global pstate
513 * @elapsed_time: Elapsed time in milliseconds
514 * @local_pstate_idx: New local pstate
515 * @highest_lpstate_idx: pstate from which its ramping down
517 * Finds the appropriate global pstate based on the pstate from which its
518 * ramping down and the time elapsed in ramping down. It follows a quadratic
519 * equation which ensures that it reaches ramping down to pmin in 5sec.
521 static inline int calc_global_pstate(unsigned int elapsed_time
,
522 int highest_lpstate_idx
,
523 int local_pstate_idx
)
528 * Using ramp_down_percent we get the percentage of rampdown
529 * that we are expecting to be dropping. Difference between
530 * highest_lpstate_idx and powernv_pstate_info.min will give a absolute
531 * number of how many pstates we will drop eventually by the end of
532 * 5 seconds, then just scale it get the number pstates to be dropped.
534 index_diff
= ((int)ramp_down_percent(elapsed_time
) *
535 (powernv_pstate_info
.min
- highest_lpstate_idx
)) / 100;
537 /* Ensure that global pstate is >= to local pstate */
538 if (highest_lpstate_idx
+ index_diff
>= local_pstate_idx
)
539 return local_pstate_idx
;
541 return highest_lpstate_idx
+ index_diff
;
544 static inline void queue_gpstate_timer(struct global_pstate_info
*gpstates
)
546 unsigned int timer_interval
;
549 * Setting up timer to fire after GPSTATE_TIMER_INTERVAL ms, But
550 * if it exceeds MAX_RAMP_DOWN_TIME ms for ramp down time.
551 * Set timer such that it fires exactly at MAX_RAMP_DOWN_TIME
552 * seconds of ramp down time.
554 if ((gpstates
->elapsed_time
+ GPSTATE_TIMER_INTERVAL
)
555 > MAX_RAMP_DOWN_TIME
)
556 timer_interval
= MAX_RAMP_DOWN_TIME
- gpstates
->elapsed_time
;
558 timer_interval
= GPSTATE_TIMER_INTERVAL
;
560 mod_timer(&gpstates
->timer
, jiffies
+ msecs_to_jiffies(timer_interval
));
564 * gpstate_timer_handler
566 * @data: pointer to cpufreq_policy on which timer was queued
568 * This handler brings down the global pstate closer to the local pstate
569 * according quadratic equation. Queues a new timer if it is still not equal
572 void gpstate_timer_handler(unsigned long data
)
574 struct cpufreq_policy
*policy
= (struct cpufreq_policy
*)data
;
575 struct global_pstate_info
*gpstates
= policy
->driver_data
;
577 unsigned int time_diff
= jiffies_to_msecs(jiffies
)
578 - gpstates
->last_sampled_time
;
579 struct powernv_smp_call_data freq_data
;
581 if (!spin_trylock(&gpstates
->gpstate_lock
))
584 gpstates
->last_sampled_time
+= time_diff
;
585 gpstates
->elapsed_time
+= time_diff
;
586 freq_data
.pstate_id
= idx_to_pstate(gpstates
->last_lpstate_idx
);
588 if ((gpstates
->last_gpstate_idx
== gpstates
->last_lpstate_idx
) ||
589 (gpstates
->elapsed_time
> MAX_RAMP_DOWN_TIME
)) {
590 gpstate_idx
= pstate_to_idx(freq_data
.pstate_id
);
591 reset_gpstates(policy
);
592 gpstates
->highest_lpstate_idx
= gpstate_idx
;
594 gpstate_idx
= calc_global_pstate(gpstates
->elapsed_time
,
595 gpstates
->highest_lpstate_idx
,
596 freq_data
.pstate_id
);
600 * If local pstate is equal to global pstate, rampdown is over
601 * So timer is not required to be queued.
603 if (gpstate_idx
!= gpstates
->last_lpstate_idx
)
604 queue_gpstate_timer(gpstates
);
606 freq_data
.gpstate_id
= idx_to_pstate(gpstate_idx
);
607 gpstates
->last_gpstate_idx
= pstate_to_idx(freq_data
.gpstate_id
);
608 gpstates
->last_lpstate_idx
= pstate_to_idx(freq_data
.pstate_id
);
610 spin_unlock(&gpstates
->gpstate_lock
);
612 /* Timer may get migrated to a different cpu on cpu hot unplug */
613 smp_call_function_any(policy
->cpus
, set_pstate
, &freq_data
, 1);
617 * powernv_cpufreq_target_index: Sets the frequency corresponding to
618 * the cpufreq table entry indexed by new_index on the cpus in the
621 static int powernv_cpufreq_target_index(struct cpufreq_policy
*policy
,
622 unsigned int new_index
)
624 struct powernv_smp_call_data freq_data
;
625 unsigned int cur_msec
, gpstate_idx
;
626 struct global_pstate_info
*gpstates
= policy
->driver_data
;
628 if (unlikely(rebooting
) && new_index
!= get_nominal_index())
632 powernv_cpufreq_throttle_check(NULL
);
634 cur_msec
= jiffies_to_msecs(get_jiffies_64());
636 spin_lock(&gpstates
->gpstate_lock
);
637 freq_data
.pstate_id
= idx_to_pstate(new_index
);
639 if (!gpstates
->last_sampled_time
) {
640 gpstate_idx
= new_index
;
641 gpstates
->highest_lpstate_idx
= new_index
;
645 if (gpstates
->last_gpstate_idx
< new_index
) {
646 gpstates
->elapsed_time
+= cur_msec
-
647 gpstates
->last_sampled_time
;
650 * If its has been ramping down for more than MAX_RAMP_DOWN_TIME
651 * we should be resetting all global pstate related data. Set it
652 * equal to local pstate to start fresh.
654 if (gpstates
->elapsed_time
> MAX_RAMP_DOWN_TIME
) {
655 reset_gpstates(policy
);
656 gpstates
->highest_lpstate_idx
= new_index
;
657 gpstate_idx
= new_index
;
659 /* Elaspsed_time is less than 5 seconds, continue to rampdown */
660 gpstate_idx
= calc_global_pstate(gpstates
->elapsed_time
,
661 gpstates
->highest_lpstate_idx
,
665 reset_gpstates(policy
);
666 gpstates
->highest_lpstate_idx
= new_index
;
667 gpstate_idx
= new_index
;
671 * If local pstate is equal to global pstate, rampdown is over
672 * So timer is not required to be queued.
674 if (gpstate_idx
!= new_index
)
675 queue_gpstate_timer(gpstates
);
677 del_timer_sync(&gpstates
->timer
);
680 freq_data
.gpstate_id
= idx_to_pstate(gpstate_idx
);
681 gpstates
->last_sampled_time
= cur_msec
;
682 gpstates
->last_gpstate_idx
= gpstate_idx
;
683 gpstates
->last_lpstate_idx
= new_index
;
685 spin_unlock(&gpstates
->gpstate_lock
);
688 * Use smp_call_function to send IPI and execute the
689 * mtspr on target CPU. We could do that without IPI
690 * if current CPU is within policy->cpus (core)
692 smp_call_function_any(policy
->cpus
, set_pstate
, &freq_data
, 1);
696 static int powernv_cpufreq_cpu_init(struct cpufreq_policy
*policy
)
699 struct kernfs_node
*kn
;
700 struct global_pstate_info
*gpstates
;
702 base
= cpu_first_thread_sibling(policy
->cpu
);
704 for (i
= 0; i
< threads_per_core
; i
++)
705 cpumask_set_cpu(base
+ i
, policy
->cpus
);
707 kn
= kernfs_find_and_get(policy
->kobj
.sd
, throttle_attr_grp
.name
);
711 ret
= sysfs_create_group(&policy
->kobj
, &throttle_attr_grp
);
713 pr_info("Failed to create throttle stats directory for cpu %d\n",
721 gpstates
= kzalloc(sizeof(*gpstates
), GFP_KERNEL
);
725 policy
->driver_data
= gpstates
;
727 /* initialize timer */
728 init_timer_pinned_deferrable(&gpstates
->timer
);
729 gpstates
->timer
.data
= (unsigned long)policy
;
730 gpstates
->timer
.function
= gpstate_timer_handler
;
731 gpstates
->timer
.expires
= jiffies
+
732 msecs_to_jiffies(GPSTATE_TIMER_INTERVAL
);
733 spin_lock_init(&gpstates
->gpstate_lock
);
734 ret
= cpufreq_table_validate_and_show(policy
, powernv_freqs
);
737 kfree(policy
->driver_data
);
742 static int powernv_cpufreq_cpu_exit(struct cpufreq_policy
*policy
)
744 /* timer is deleted in cpufreq_cpu_stop() */
745 kfree(policy
->driver_data
);
750 static int powernv_cpufreq_reboot_notifier(struct notifier_block
*nb
,
751 unsigned long action
, void *unused
)
754 struct cpufreq_policy cpu_policy
;
757 for_each_online_cpu(cpu
) {
758 cpufreq_get_policy(&cpu_policy
, cpu
);
759 powernv_cpufreq_target_index(&cpu_policy
, get_nominal_index());
765 static struct notifier_block powernv_cpufreq_reboot_nb
= {
766 .notifier_call
= powernv_cpufreq_reboot_notifier
,
769 void powernv_cpufreq_work_fn(struct work_struct
*work
)
771 struct chip
*chip
= container_of(work
, struct chip
, throttle
);
776 cpumask_and(&mask
, &chip
->mask
, cpu_online_mask
);
777 smp_call_function_any(&mask
,
778 powernv_cpufreq_throttle_check
, NULL
, 0);
783 chip
->restore
= false;
784 for_each_cpu(cpu
, &mask
) {
786 struct cpufreq_policy policy
;
788 cpufreq_get_policy(&policy
, cpu
);
789 index
= cpufreq_table_find_index_c(&policy
, policy
.cur
);
790 powernv_cpufreq_target_index(&policy
, index
);
791 cpumask_andnot(&mask
, &mask
, policy
.cpus
);
797 static int powernv_cpufreq_occ_msg(struct notifier_block
*nb
,
798 unsigned long msg_type
, void *_msg
)
800 struct opal_msg
*msg
= _msg
;
801 struct opal_occ_msg omsg
;
804 if (msg_type
!= OPAL_MSG_OCC
)
807 omsg
.type
= be64_to_cpu(msg
->params
[0]);
812 pr_info("OCC (On Chip Controller - enforces hard thermal/power limits) Resetting\n");
814 * powernv_cpufreq_throttle_check() is called in
815 * target() callback which can detect the throttle state
816 * for governors like ondemand.
817 * But static governors will not call target() often thus
818 * report throttling here.
822 pr_warn("CPU frequency is throttled for duration\n");
827 pr_info("OCC Loading, CPU frequency is throttled until OCC is started\n");
830 omsg
.chip
= be64_to_cpu(msg
->params
[1]);
831 omsg
.throttle_status
= be64_to_cpu(msg
->params
[2]);
836 pr_info("OCC Active, CPU frequency is no longer throttled\n");
838 for (i
= 0; i
< nr_chips
; i
++) {
839 chips
[i
].restore
= true;
840 schedule_work(&chips
[i
].throttle
);
846 for (i
= 0; i
< nr_chips
; i
++)
847 if (chips
[i
].id
== omsg
.chip
)
850 if (omsg
.throttle_status
>= 0 &&
851 omsg
.throttle_status
<= OCC_MAX_THROTTLE_STATUS
) {
852 chips
[i
].throttle_reason
= omsg
.throttle_status
;
853 chips
[i
].reason
[omsg
.throttle_status
]++;
856 if (!omsg
.throttle_status
)
857 chips
[i
].restore
= true;
859 schedule_work(&chips
[i
].throttle
);
864 static struct notifier_block powernv_cpufreq_opal_nb
= {
865 .notifier_call
= powernv_cpufreq_occ_msg
,
870 static void powernv_cpufreq_stop_cpu(struct cpufreq_policy
*policy
)
872 struct powernv_smp_call_data freq_data
;
873 struct global_pstate_info
*gpstates
= policy
->driver_data
;
875 freq_data
.pstate_id
= idx_to_pstate(powernv_pstate_info
.min
);
876 freq_data
.gpstate_id
= idx_to_pstate(powernv_pstate_info
.min
);
877 smp_call_function_single(policy
->cpu
, set_pstate
, &freq_data
, 1);
878 del_timer_sync(&gpstates
->timer
);
881 static struct cpufreq_driver powernv_cpufreq_driver
= {
882 .name
= "powernv-cpufreq",
883 .flags
= CPUFREQ_CONST_LOOPS
,
884 .init
= powernv_cpufreq_cpu_init
,
885 .exit
= powernv_cpufreq_cpu_exit
,
886 .verify
= cpufreq_generic_frequency_table_verify
,
887 .target_index
= powernv_cpufreq_target_index
,
888 .get
= powernv_cpufreq_get
,
889 .stop_cpu
= powernv_cpufreq_stop_cpu
,
890 .attr
= powernv_cpu_freq_attr
,
893 static int init_chip_info(void)
895 unsigned int chip
[256];
897 unsigned int prev_chip_id
= UINT_MAX
;
899 for_each_possible_cpu(cpu
) {
900 unsigned int id
= cpu_to_chip_id(cpu
);
902 if (prev_chip_id
!= id
) {
904 chip
[nr_chips
++] = id
;
908 chips
= kcalloc(nr_chips
, sizeof(struct chip
), GFP_KERNEL
);
912 for (i
= 0; i
< nr_chips
; i
++) {
913 chips
[i
].id
= chip
[i
];
914 cpumask_copy(&chips
[i
].mask
, cpumask_of_node(chip
[i
]));
915 INIT_WORK(&chips
[i
].throttle
, powernv_cpufreq_work_fn
);
916 for_each_cpu(cpu
, &chips
[i
].mask
)
917 per_cpu(chip_info
, cpu
) = &chips
[i
];
923 static inline void clean_chip_info(void)
928 static inline void unregister_all_notifiers(void)
930 opal_message_notifier_unregister(OPAL_MSG_OCC
,
931 &powernv_cpufreq_opal_nb
);
932 unregister_reboot_notifier(&powernv_cpufreq_reboot_nb
);
935 static int __init
powernv_cpufreq_init(void)
939 /* Don't probe on pseries (guest) platforms */
940 if (!firmware_has_feature(FW_FEATURE_OPAL
))
943 /* Discover pstates from device tree and init */
944 rc
= init_powernv_pstates();
948 /* Populate chip info */
949 rc
= init_chip_info();
953 register_reboot_notifier(&powernv_cpufreq_reboot_nb
);
954 opal_message_notifier_register(OPAL_MSG_OCC
, &powernv_cpufreq_opal_nb
);
956 rc
= cpufreq_register_driver(&powernv_cpufreq_driver
);
960 pr_info("Failed to register the cpufreq driver (%d)\n", rc
);
961 unregister_all_notifiers();
964 pr_info("Platform driver disabled. System does not support PState control\n");
967 module_init(powernv_cpufreq_init
);
969 static void __exit
powernv_cpufreq_exit(void)
971 cpufreq_unregister_driver(&powernv_cpufreq_driver
);
972 unregister_all_notifiers();
975 module_exit(powernv_cpufreq_exit
);
977 MODULE_LICENSE("GPL");
978 MODULE_AUTHOR("Vaidyanathan Srinivasan <svaidy at linux.vnet.ibm.com>");