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 */
40 #define POWERNV_MAX_PSTATES 256
41 #define PMSR_PSAFE_ENABLE (1UL << 30)
42 #define PMSR_SPR_EM_DISABLE (1UL << 31)
43 #define PMSR_MAX(x) ((x >> 32) & 0xFF)
45 static struct cpufreq_frequency_table powernv_freqs[POWERNV_MAX_PSTATES+1];
46 static bool rebooting, throttled, occ_reset;
47 static unsigned int *core_to_chip_map;
49 static const char * const throttle_reason[] = {
52 "Processor Over Temperature",
53 "Power Supply Failure",
64 struct work_struct throttle;
70 * Note: The set of pstates consists of contiguous integers, the
71 * smallest of which is indicated by powernv_pstate_info.min, the
72 * largest of which is indicated by powernv_pstate_info.max.
74 * The nominal pstate is the highest non-turbo pstate in this
75 * platform. This is indicated by powernv_pstate_info.nominal.
77 static struct powernv_pstate_info {
82 } powernv_pstate_info;
85 * Initialize the freq table based on data obtained
86 * from the firmware passed via device-tree
88 static int init_powernv_pstates(void)
90 struct device_node *power_mgt;
91 int i, pstate_min, pstate_max, pstate_nominal, nr_pstates = 0;
92 const __be32 *pstate_ids, *pstate_freqs;
93 u32 len_ids, len_freqs;
95 power_mgt = of_find_node_by_path("/ibm,opal/power-mgt");
97 pr_warn("power-mgt node not found\n");
101 if (of_property_read_u32(power_mgt, "ibm,pstate-min", &pstate_min)) {
102 pr_warn("ibm,pstate-min node not found\n");
106 if (of_property_read_u32(power_mgt, "ibm,pstate-max", &pstate_max)) {
107 pr_warn("ibm,pstate-max node not found\n");
111 if (of_property_read_u32(power_mgt, "ibm,pstate-nominal",
113 pr_warn("ibm,pstate-nominal not found\n");
116 pr_info("cpufreq pstate min %d nominal %d max %d\n", pstate_min,
117 pstate_nominal, pstate_max);
119 pstate_ids = of_get_property(power_mgt, "ibm,pstate-ids", &len_ids);
121 pr_warn("ibm,pstate-ids not found\n");
125 pstate_freqs = of_get_property(power_mgt, "ibm,pstate-frequencies-mhz",
128 pr_warn("ibm,pstate-frequencies-mhz not found\n");
132 if (len_ids != len_freqs) {
133 pr_warn("Entries in ibm,pstate-ids and "
134 "ibm,pstate-frequencies-mhz does not match\n");
137 nr_pstates = min(len_ids, len_freqs) / sizeof(u32);
139 pr_warn("No PStates found\n");
143 pr_debug("NR PStates %d\n", nr_pstates);
144 for (i = 0; i < nr_pstates; i++) {
145 u32 id = be32_to_cpu(pstate_ids[i]);
146 u32 freq = be32_to_cpu(pstate_freqs[i]);
148 pr_debug("PState id %d freq %d MHz\n", id, freq);
149 powernv_freqs[i].frequency = freq * 1000; /* kHz */
150 powernv_freqs[i].driver_data = id;
152 /* End of list marker entry */
153 powernv_freqs[i].frequency = CPUFREQ_TABLE_END;
155 powernv_pstate_info.min = pstate_min;
156 powernv_pstate_info.max = pstate_max;
157 powernv_pstate_info.nominal = pstate_nominal;
158 powernv_pstate_info.nr_pstates = nr_pstates;
163 /* Returns the CPU frequency corresponding to the pstate_id. */
164 static unsigned int pstate_id_to_freq(int pstate_id)
168 i = powernv_pstate_info.max - pstate_id;
169 if (i >= powernv_pstate_info.nr_pstates || i < 0) {
170 pr_warn("PState id %d outside of PState table, "
171 "reporting nominal id %d instead\n",
172 pstate_id, powernv_pstate_info.nominal);
173 i = powernv_pstate_info.max - powernv_pstate_info.nominal;
176 return powernv_freqs[i].frequency;
180 * cpuinfo_nominal_freq_show - Show the nominal CPU frequency as indicated by
183 static ssize_t cpuinfo_nominal_freq_show(struct cpufreq_policy *policy,
186 return sprintf(buf, "%u\n",
187 pstate_id_to_freq(powernv_pstate_info.nominal));
190 struct freq_attr cpufreq_freq_attr_cpuinfo_nominal_freq =
191 __ATTR_RO(cpuinfo_nominal_freq);
193 static struct freq_attr *powernv_cpu_freq_attr[] = {
194 &cpufreq_freq_attr_scaling_available_freqs,
195 &cpufreq_freq_attr_cpuinfo_nominal_freq,
199 /* Helper routines */
201 /* Access helpers to power mgt SPR */
203 static inline unsigned long get_pmspr(unsigned long sprn)
207 return mfspr(SPRN_PMCR);
210 return mfspr(SPRN_PMICR);
213 return mfspr(SPRN_PMSR);
218 static inline void set_pmspr(unsigned long sprn, unsigned long val)
222 mtspr(SPRN_PMCR, val);
226 mtspr(SPRN_PMICR, val);
233 * Use objects of this type to query/update
234 * pstates on a remote CPU via smp_call_function.
236 struct powernv_smp_call_data {
242 * powernv_read_cpu_freq: Reads the current frequency on this CPU.
244 * Called via smp_call_function.
246 * Note: The caller of the smp_call_function should pass an argument of
247 * the type 'struct powernv_smp_call_data *' along with this function.
249 * The current frequency on this CPU will be returned via
250 * ((struct powernv_smp_call_data *)arg)->freq;
252 static void powernv_read_cpu_freq(void *arg)
254 unsigned long pmspr_val;
256 struct powernv_smp_call_data *freq_data = arg;
258 pmspr_val = get_pmspr(SPRN_PMSR);
261 * The local pstate id corresponds bits 48..55 in the PMSR.
262 * Note: Watch out for the sign!
264 local_pstate_id = (pmspr_val >> 48) & 0xFF;
265 freq_data->pstate_id = local_pstate_id;
266 freq_data->freq = pstate_id_to_freq(freq_data->pstate_id);
268 pr_debug("cpu %d pmsr %016lX pstate_id %d frequency %d kHz\n",
269 raw_smp_processor_id(), pmspr_val, freq_data->pstate_id,
274 * powernv_cpufreq_get: Returns the CPU frequency as reported by the
275 * firmware for CPU 'cpu'. This value is reported through the sysfs
276 * file cpuinfo_cur_freq.
278 static unsigned int powernv_cpufreq_get(unsigned int cpu)
280 struct powernv_smp_call_data freq_data;
282 smp_call_function_any(cpu_sibling_mask(cpu), powernv_read_cpu_freq,
285 return freq_data.freq;
289 * set_pstate: Sets the pstate on this CPU.
291 * This is called via an smp_call_function.
293 * The caller must ensure that freq_data is of the type
294 * (struct powernv_smp_call_data *) and the pstate_id which needs to be set
295 * on this CPU should be present in freq_data->pstate_id.
297 static void set_pstate(void *freq_data)
300 unsigned long pstate_ul =
301 ((struct powernv_smp_call_data *) freq_data)->pstate_id;
303 val = get_pmspr(SPRN_PMCR);
304 val = val & 0x0000FFFFFFFFFFFFULL;
306 pstate_ul = pstate_ul & 0xFF;
308 /* Set both global(bits 56..63) and local(bits 48..55) PStates */
309 val = val | (pstate_ul << 56) | (pstate_ul << 48);
311 pr_debug("Setting cpu %d pmcr to %016lX\n",
312 raw_smp_processor_id(), val);
313 set_pmspr(SPRN_PMCR, val);
317 * get_nominal_index: Returns the index corresponding to the nominal
318 * pstate in the cpufreq table
320 static inline unsigned int get_nominal_index(void)
322 return powernv_pstate_info.max - powernv_pstate_info.nominal;
325 static void powernv_cpufreq_throttle_check(void *data)
327 unsigned int cpu = smp_processor_id();
328 unsigned int chip_id = core_to_chip_map[cpu_core_index_of_thread(cpu)];
332 pmsr = get_pmspr(SPRN_PMSR);
334 for (i = 0; i < nr_chips; i++)
335 if (chips[i].id == chip_id)
338 /* Check for Pmax Capping */
339 pmsr_pmax = (s8)PMSR_MAX(pmsr);
340 if (pmsr_pmax != powernv_pstate_info.max) {
341 if (chips[i].throttled)
343 chips[i].throttled = true;
344 if (pmsr_pmax < powernv_pstate_info.nominal)
345 pr_warn_once("CPU %d on Chip %u has Pmax reduced below nominal frequency (%d < %d)\n",
346 cpu, chips[i].id, pmsr_pmax,
347 powernv_pstate_info.nominal);
348 trace_powernv_throttle(chips[i].id,
349 throttle_reason[chips[i].throttle_reason],
351 } else if (chips[i].throttled) {
352 chips[i].throttled = false;
353 trace_powernv_throttle(chips[i].id,
354 throttle_reason[chips[i].throttle_reason],
358 /* Check if Psafe_mode_active is set in PMSR. */
360 if (pmsr & PMSR_PSAFE_ENABLE) {
362 pr_info("Pstate set to safe frequency\n");
365 /* Check if SPR_EM_DISABLE is set in PMSR */
366 if (pmsr & PMSR_SPR_EM_DISABLE) {
368 pr_info("Frequency Control disabled from OS\n");
372 pr_info("PMSR = %16lx\n", pmsr);
373 pr_warn("CPU Frequency could be throttled\n");
378 * powernv_cpufreq_target_index: Sets the frequency corresponding to
379 * the cpufreq table entry indexed by new_index on the cpus in the
382 static int powernv_cpufreq_target_index(struct cpufreq_policy *policy,
383 unsigned int new_index)
385 struct powernv_smp_call_data freq_data;
387 if (unlikely(rebooting) && new_index != get_nominal_index())
391 powernv_cpufreq_throttle_check(NULL);
393 freq_data.pstate_id = powernv_freqs[new_index].driver_data;
396 * Use smp_call_function to send IPI and execute the
397 * mtspr on target CPU. We could do that without IPI
398 * if current CPU is within policy->cpus (core)
400 smp_call_function_any(policy->cpus, set_pstate, &freq_data, 1);
405 static int powernv_cpufreq_cpu_init(struct cpufreq_policy *policy)
409 base = cpu_first_thread_sibling(policy->cpu);
411 for (i = 0; i < threads_per_core; i++)
412 cpumask_set_cpu(base + i, policy->cpus);
414 return cpufreq_table_validate_and_show(policy, powernv_freqs);
417 static int powernv_cpufreq_reboot_notifier(struct notifier_block *nb,
418 unsigned long action, void *unused)
421 struct cpufreq_policy cpu_policy;
424 for_each_online_cpu(cpu) {
425 cpufreq_get_policy(&cpu_policy, cpu);
426 powernv_cpufreq_target_index(&cpu_policy, get_nominal_index());
432 static struct notifier_block powernv_cpufreq_reboot_nb = {
433 .notifier_call = powernv_cpufreq_reboot_notifier,
436 void powernv_cpufreq_work_fn(struct work_struct *work)
438 struct chip *chip = container_of(work, struct chip, throttle);
443 cpumask_and(&mask, &chip->mask, cpu_online_mask);
444 smp_call_function_any(&mask,
445 powernv_cpufreq_throttle_check, NULL, 0);
450 chip->restore = false;
451 for_each_cpu(cpu, &mask) {
453 struct cpufreq_policy policy;
455 cpufreq_get_policy(&policy, cpu);
456 cpufreq_frequency_table_target(&policy, policy.freq_table,
458 CPUFREQ_RELATION_C, &index);
459 powernv_cpufreq_target_index(&policy, index);
460 cpumask_andnot(&mask, &mask, policy.cpus);
466 static int powernv_cpufreq_occ_msg(struct notifier_block *nb,
467 unsigned long msg_type, void *_msg)
469 struct opal_msg *msg = _msg;
470 struct opal_occ_msg omsg;
473 if (msg_type != OPAL_MSG_OCC)
476 omsg.type = be64_to_cpu(msg->params[0]);
481 pr_info("OCC (On Chip Controller - enforces hard thermal/power limits) Resetting\n");
483 * powernv_cpufreq_throttle_check() is called in
484 * target() callback which can detect the throttle state
485 * for governors like ondemand.
486 * But static governors will not call target() often thus
487 * report throttling here.
491 pr_warn("CPU frequency is throttled for duration\n");
496 pr_info("OCC Loading, CPU frequency is throttled until OCC is started\n");
499 omsg.chip = be64_to_cpu(msg->params[1]);
500 omsg.throttle_status = be64_to_cpu(msg->params[2]);
505 pr_info("OCC Active, CPU frequency is no longer throttled\n");
507 for (i = 0; i < nr_chips; i++) {
508 chips[i].restore = true;
509 schedule_work(&chips[i].throttle);
515 for (i = 0; i < nr_chips; i++)
516 if (chips[i].id == omsg.chip)
519 if (omsg.throttle_status >= 0 &&
520 omsg.throttle_status <= OCC_MAX_THROTTLE_STATUS)
521 chips[i].throttle_reason = omsg.throttle_status;
523 if (!omsg.throttle_status)
524 chips[i].restore = true;
526 schedule_work(&chips[i].throttle);
531 static struct notifier_block powernv_cpufreq_opal_nb = {
532 .notifier_call = powernv_cpufreq_occ_msg,
537 static void powernv_cpufreq_stop_cpu(struct cpufreq_policy *policy)
539 struct powernv_smp_call_data freq_data;
541 freq_data.pstate_id = powernv_pstate_info.min;
542 smp_call_function_single(policy->cpu, set_pstate, &freq_data, 1);
545 static struct cpufreq_driver powernv_cpufreq_driver = {
546 .name = "powernv-cpufreq",
547 .flags = CPUFREQ_CONST_LOOPS,
548 .init = powernv_cpufreq_cpu_init,
549 .verify = cpufreq_generic_frequency_table_verify,
550 .target_index = powernv_cpufreq_target_index,
551 .get = powernv_cpufreq_get,
552 .stop_cpu = powernv_cpufreq_stop_cpu,
553 .attr = powernv_cpu_freq_attr,
556 static int init_chip_info(void)
558 unsigned int chip[256];
560 unsigned int prev_chip_id = UINT_MAX;
564 core_to_chip_map = kcalloc(cpu_nr_cores(), sizeof(unsigned int),
566 if (!core_to_chip_map)
569 cpumask_copy(&cpu_mask, cpu_possible_mask);
570 for_each_cpu(cpu, &cpu_mask) {
571 unsigned int id = cpu_to_chip_id(cpu);
573 if (prev_chip_id != id) {
575 chip[nr_chips++] = id;
577 core_to_chip_map[cpu_core_index_of_thread(cpu)] = id;
578 cpumask_andnot(&cpu_mask, &cpu_mask, cpu_sibling_mask(cpu));
581 chips = kcalloc(nr_chips, sizeof(struct chip), GFP_KERNEL);
585 for (i = 0; i < nr_chips; i++) {
586 chips[i].id = chip[i];
587 cpumask_copy(&chips[i].mask, cpumask_of_node(chip[i]));
588 INIT_WORK(&chips[i].throttle, powernv_cpufreq_work_fn);
593 kfree(core_to_chip_map);
598 static inline void clean_chip_info(void)
601 kfree(core_to_chip_map);
604 static inline void unregister_all_notifiers(void)
606 opal_message_notifier_unregister(OPAL_MSG_OCC,
607 &powernv_cpufreq_opal_nb);
608 unregister_reboot_notifier(&powernv_cpufreq_reboot_nb);
611 static int __init powernv_cpufreq_init(void)
615 /* Don't probe on pseries (guest) platforms */
616 if (!firmware_has_feature(FW_FEATURE_OPAL))
619 /* Discover pstates from device tree and init */
620 rc = init_powernv_pstates();
624 /* Populate chip info */
625 rc = init_chip_info();
629 register_reboot_notifier(&powernv_cpufreq_reboot_nb);
630 opal_message_notifier_register(OPAL_MSG_OCC, &powernv_cpufreq_opal_nb);
632 rc = cpufreq_register_driver(&powernv_cpufreq_driver);
636 pr_info("Failed to register the cpufreq driver (%d)\n", rc);
637 unregister_all_notifiers();
640 pr_info("Platform driver disabled. System does not support PState control\n");
643 module_init(powernv_cpufreq_init);
645 static void __exit powernv_cpufreq_exit(void)
647 cpufreq_unregister_driver(&powernv_cpufreq_driver);
648 unregister_all_notifiers();
651 module_exit(powernv_cpufreq_exit);
653 MODULE_LICENSE("GPL");
654 MODULE_AUTHOR("Vaidyanathan Srinivasan <svaidy at linux.vnet.ibm.com>");