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;
48 static const char * const throttle_reason[] = {
51 "Processor Over Temperature",
52 "Power Supply Failure",
57 enum throttle_reason_type {
73 struct work_struct throttle;
75 int throttle_sub_turbo;
76 int reason[OCC_MAX_REASON];
80 static DEFINE_PER_CPU(struct chip *, chip_info);
83 * Note: The set of pstates consists of contiguous integers, the
84 * smallest of which is indicated by powernv_pstate_info.min, the
85 * largest of which is indicated by powernv_pstate_info.max.
87 * The nominal pstate is the highest non-turbo pstate in this
88 * platform. This is indicated by powernv_pstate_info.nominal.
90 static struct powernv_pstate_info {
95 } powernv_pstate_info;
98 * Initialize the freq table based on data obtained
99 * from the firmware passed via device-tree
101 static int init_powernv_pstates(void)
103 struct device_node *power_mgt;
104 int i, pstate_min, pstate_max, pstate_nominal, nr_pstates = 0;
105 const __be32 *pstate_ids, *pstate_freqs;
106 u32 len_ids, len_freqs;
108 power_mgt = of_find_node_by_path("/ibm,opal/power-mgt");
110 pr_warn("power-mgt node not found\n");
114 if (of_property_read_u32(power_mgt, "ibm,pstate-min", &pstate_min)) {
115 pr_warn("ibm,pstate-min node not found\n");
119 if (of_property_read_u32(power_mgt, "ibm,pstate-max", &pstate_max)) {
120 pr_warn("ibm,pstate-max node not found\n");
124 if (of_property_read_u32(power_mgt, "ibm,pstate-nominal",
126 pr_warn("ibm,pstate-nominal not found\n");
129 pr_info("cpufreq pstate min %d nominal %d max %d\n", pstate_min,
130 pstate_nominal, pstate_max);
132 pstate_ids = of_get_property(power_mgt, "ibm,pstate-ids", &len_ids);
134 pr_warn("ibm,pstate-ids not found\n");
138 pstate_freqs = of_get_property(power_mgt, "ibm,pstate-frequencies-mhz",
141 pr_warn("ibm,pstate-frequencies-mhz not found\n");
145 if (len_ids != len_freqs) {
146 pr_warn("Entries in ibm,pstate-ids and "
147 "ibm,pstate-frequencies-mhz does not match\n");
150 nr_pstates = min(len_ids, len_freqs) / sizeof(u32);
152 pr_warn("No PStates found\n");
156 pr_debug("NR PStates %d\n", nr_pstates);
157 for (i = 0; i < nr_pstates; i++) {
158 u32 id = be32_to_cpu(pstate_ids[i]);
159 u32 freq = be32_to_cpu(pstate_freqs[i]);
161 pr_debug("PState id %d freq %d MHz\n", id, freq);
162 powernv_freqs[i].frequency = freq * 1000; /* kHz */
163 powernv_freqs[i].driver_data = id;
165 /* End of list marker entry */
166 powernv_freqs[i].frequency = CPUFREQ_TABLE_END;
168 powernv_pstate_info.min = pstate_min;
169 powernv_pstate_info.max = pstate_max;
170 powernv_pstate_info.nominal = pstate_nominal;
171 powernv_pstate_info.nr_pstates = nr_pstates;
176 /* Returns the CPU frequency corresponding to the pstate_id. */
177 static unsigned int pstate_id_to_freq(int pstate_id)
181 i = powernv_pstate_info.max - pstate_id;
182 if (i >= powernv_pstate_info.nr_pstates || i < 0) {
183 pr_warn("PState id %d outside of PState table, "
184 "reporting nominal id %d instead\n",
185 pstate_id, powernv_pstate_info.nominal);
186 i = powernv_pstate_info.max - powernv_pstate_info.nominal;
189 return powernv_freqs[i].frequency;
193 * cpuinfo_nominal_freq_show - Show the nominal CPU frequency as indicated by
196 static ssize_t cpuinfo_nominal_freq_show(struct cpufreq_policy *policy,
199 return sprintf(buf, "%u\n",
200 pstate_id_to_freq(powernv_pstate_info.nominal));
203 struct freq_attr cpufreq_freq_attr_cpuinfo_nominal_freq =
204 __ATTR_RO(cpuinfo_nominal_freq);
206 static struct freq_attr *powernv_cpu_freq_attr[] = {
207 &cpufreq_freq_attr_scaling_available_freqs,
208 &cpufreq_freq_attr_cpuinfo_nominal_freq,
212 #define throttle_attr(name, member) \
213 static ssize_t name##_show(struct cpufreq_policy *policy, char *buf) \
215 struct chip *chip = per_cpu(chip_info, policy->cpu); \
217 return sprintf(buf, "%u\n", chip->member); \
220 static struct freq_attr throttle_attr_##name = __ATTR_RO(name) \
222 throttle_attr(unthrottle, reason[NO_THROTTLE]);
223 throttle_attr(powercap, reason[POWERCAP]);
224 throttle_attr(overtemp, reason[CPU_OVERTEMP]);
225 throttle_attr(supply_fault, reason[POWER_SUPPLY_FAILURE]);
226 throttle_attr(overcurrent, reason[OVERCURRENT]);
227 throttle_attr(occ_reset, reason[OCC_RESET_THROTTLE]);
228 throttle_attr(turbo_stat, throttle_turbo);
229 throttle_attr(sub_turbo_stat, throttle_sub_turbo);
231 static struct attribute *throttle_attrs[] = {
232 &throttle_attr_unthrottle.attr,
233 &throttle_attr_powercap.attr,
234 &throttle_attr_overtemp.attr,
235 &throttle_attr_supply_fault.attr,
236 &throttle_attr_overcurrent.attr,
237 &throttle_attr_occ_reset.attr,
238 &throttle_attr_turbo_stat.attr,
239 &throttle_attr_sub_turbo_stat.attr,
243 static const struct attribute_group throttle_attr_grp = {
244 .name = "throttle_stats",
245 .attrs = throttle_attrs,
248 /* Helper routines */
250 /* Access helpers to power mgt SPR */
252 static inline unsigned long get_pmspr(unsigned long sprn)
256 return mfspr(SPRN_PMCR);
259 return mfspr(SPRN_PMICR);
262 return mfspr(SPRN_PMSR);
267 static inline void set_pmspr(unsigned long sprn, unsigned long val)
271 mtspr(SPRN_PMCR, val);
275 mtspr(SPRN_PMICR, val);
282 * Use objects of this type to query/update
283 * pstates on a remote CPU via smp_call_function.
285 struct powernv_smp_call_data {
291 * powernv_read_cpu_freq: Reads the current frequency on this CPU.
293 * Called via smp_call_function.
295 * Note: The caller of the smp_call_function should pass an argument of
296 * the type 'struct powernv_smp_call_data *' along with this function.
298 * The current frequency on this CPU will be returned via
299 * ((struct powernv_smp_call_data *)arg)->freq;
301 static void powernv_read_cpu_freq(void *arg)
303 unsigned long pmspr_val;
305 struct powernv_smp_call_data *freq_data = arg;
307 pmspr_val = get_pmspr(SPRN_PMSR);
310 * The local pstate id corresponds bits 48..55 in the PMSR.
311 * Note: Watch out for the sign!
313 local_pstate_id = (pmspr_val >> 48) & 0xFF;
314 freq_data->pstate_id = local_pstate_id;
315 freq_data->freq = pstate_id_to_freq(freq_data->pstate_id);
317 pr_debug("cpu %d pmsr %016lX pstate_id %d frequency %d kHz\n",
318 raw_smp_processor_id(), pmspr_val, freq_data->pstate_id,
323 * powernv_cpufreq_get: Returns the CPU frequency as reported by the
324 * firmware for CPU 'cpu'. This value is reported through the sysfs
325 * file cpuinfo_cur_freq.
327 static unsigned int powernv_cpufreq_get(unsigned int cpu)
329 struct powernv_smp_call_data freq_data;
331 smp_call_function_any(cpu_sibling_mask(cpu), powernv_read_cpu_freq,
334 return freq_data.freq;
338 * set_pstate: Sets the pstate on this CPU.
340 * This is called via an smp_call_function.
342 * The caller must ensure that freq_data is of the type
343 * (struct powernv_smp_call_data *) and the pstate_id which needs to be set
344 * on this CPU should be present in freq_data->pstate_id.
346 static void set_pstate(void *freq_data)
349 unsigned long pstate_ul =
350 ((struct powernv_smp_call_data *) freq_data)->pstate_id;
352 val = get_pmspr(SPRN_PMCR);
353 val = val & 0x0000FFFFFFFFFFFFULL;
355 pstate_ul = pstate_ul & 0xFF;
357 /* Set both global(bits 56..63) and local(bits 48..55) PStates */
358 val = val | (pstate_ul << 56) | (pstate_ul << 48);
360 pr_debug("Setting cpu %d pmcr to %016lX\n",
361 raw_smp_processor_id(), val);
362 set_pmspr(SPRN_PMCR, val);
366 * get_nominal_index: Returns the index corresponding to the nominal
367 * pstate in the cpufreq table
369 static inline unsigned int get_nominal_index(void)
371 return powernv_pstate_info.max - powernv_pstate_info.nominal;
374 static void powernv_cpufreq_throttle_check(void *data)
377 unsigned int cpu = smp_processor_id();
381 pmsr = get_pmspr(SPRN_PMSR);
382 chip = this_cpu_read(chip_info);
384 /* Check for Pmax Capping */
385 pmsr_pmax = (s8)PMSR_MAX(pmsr);
386 if (pmsr_pmax != powernv_pstate_info.max) {
389 chip->throttled = true;
390 if (pmsr_pmax < powernv_pstate_info.nominal) {
391 pr_warn_once("CPU %d on Chip %u has Pmax reduced below nominal frequency (%d < %d)\n",
392 cpu, chip->id, pmsr_pmax,
393 powernv_pstate_info.nominal);
394 chip->throttle_sub_turbo++;
396 chip->throttle_turbo++;
398 trace_powernv_throttle(chip->id,
399 throttle_reason[chip->throttle_reason],
401 } else if (chip->throttled) {
402 chip->throttled = false;
403 trace_powernv_throttle(chip->id,
404 throttle_reason[chip->throttle_reason],
408 /* Check if Psafe_mode_active is set in PMSR. */
410 if (pmsr & PMSR_PSAFE_ENABLE) {
412 pr_info("Pstate set to safe frequency\n");
415 /* Check if SPR_EM_DISABLE is set in PMSR */
416 if (pmsr & PMSR_SPR_EM_DISABLE) {
418 pr_info("Frequency Control disabled from OS\n");
422 pr_info("PMSR = %16lx\n", pmsr);
423 pr_warn("CPU Frequency could be throttled\n");
428 * powernv_cpufreq_target_index: Sets the frequency corresponding to
429 * the cpufreq table entry indexed by new_index on the cpus in the
432 static int powernv_cpufreq_target_index(struct cpufreq_policy *policy,
433 unsigned int new_index)
435 struct powernv_smp_call_data freq_data;
437 if (unlikely(rebooting) && new_index != get_nominal_index())
441 powernv_cpufreq_throttle_check(NULL);
443 freq_data.pstate_id = powernv_freqs[new_index].driver_data;
446 * Use smp_call_function to send IPI and execute the
447 * mtspr on target CPU. We could do that without IPI
448 * if current CPU is within policy->cpus (core)
450 smp_call_function_any(policy->cpus, set_pstate, &freq_data, 1);
455 static int powernv_cpufreq_cpu_init(struct cpufreq_policy *policy)
459 base = cpu_first_thread_sibling(policy->cpu);
461 for (i = 0; i < threads_per_core; i++)
462 cpumask_set_cpu(base + i, policy->cpus);
464 if (!policy->driver_data) {
467 ret = sysfs_create_group(&policy->kobj, &throttle_attr_grp);
469 pr_info("Failed to create throttle stats directory for cpu %d\n",
474 * policy->driver_data is used as a flag for one-time
475 * creation of throttle sysfs files.
477 policy->driver_data = policy;
479 return cpufreq_table_validate_and_show(policy, powernv_freqs);
482 static int powernv_cpufreq_reboot_notifier(struct notifier_block *nb,
483 unsigned long action, void *unused)
486 struct cpufreq_policy cpu_policy;
489 for_each_online_cpu(cpu) {
490 cpufreq_get_policy(&cpu_policy, cpu);
491 powernv_cpufreq_target_index(&cpu_policy, get_nominal_index());
497 static struct notifier_block powernv_cpufreq_reboot_nb = {
498 .notifier_call = powernv_cpufreq_reboot_notifier,
501 void powernv_cpufreq_work_fn(struct work_struct *work)
503 struct chip *chip = container_of(work, struct chip, throttle);
508 cpumask_and(&mask, &chip->mask, cpu_online_mask);
509 smp_call_function_any(&mask,
510 powernv_cpufreq_throttle_check, NULL, 0);
515 chip->restore = false;
516 for_each_cpu(cpu, &mask) {
518 struct cpufreq_policy policy;
520 cpufreq_get_policy(&policy, cpu);
521 cpufreq_frequency_table_target(&policy, policy.freq_table,
523 CPUFREQ_RELATION_C, &index);
524 powernv_cpufreq_target_index(&policy, index);
525 cpumask_andnot(&mask, &mask, policy.cpus);
531 static int powernv_cpufreq_occ_msg(struct notifier_block *nb,
532 unsigned long msg_type, void *_msg)
534 struct opal_msg *msg = _msg;
535 struct opal_occ_msg omsg;
538 if (msg_type != OPAL_MSG_OCC)
541 omsg.type = be64_to_cpu(msg->params[0]);
546 pr_info("OCC (On Chip Controller - enforces hard thermal/power limits) Resetting\n");
548 * powernv_cpufreq_throttle_check() is called in
549 * target() callback which can detect the throttle state
550 * for governors like ondemand.
551 * But static governors will not call target() often thus
552 * report throttling here.
556 pr_warn("CPU frequency is throttled for duration\n");
561 pr_info("OCC Loading, CPU frequency is throttled until OCC is started\n");
564 omsg.chip = be64_to_cpu(msg->params[1]);
565 omsg.throttle_status = be64_to_cpu(msg->params[2]);
570 pr_info("OCC Active, CPU frequency is no longer throttled\n");
572 for (i = 0; i < nr_chips; i++) {
573 chips[i].restore = true;
574 schedule_work(&chips[i].throttle);
580 for (i = 0; i < nr_chips; i++)
581 if (chips[i].id == omsg.chip)
584 if (omsg.throttle_status >= 0 &&
585 omsg.throttle_status <= OCC_MAX_THROTTLE_STATUS) {
586 chips[i].throttle_reason = omsg.throttle_status;
587 chips[i].reason[omsg.throttle_status]++;
590 if (!omsg.throttle_status)
591 chips[i].restore = true;
593 schedule_work(&chips[i].throttle);
598 static struct notifier_block powernv_cpufreq_opal_nb = {
599 .notifier_call = powernv_cpufreq_occ_msg,
604 static void powernv_cpufreq_stop_cpu(struct cpufreq_policy *policy)
606 struct powernv_smp_call_data freq_data;
608 freq_data.pstate_id = powernv_pstate_info.min;
609 smp_call_function_single(policy->cpu, set_pstate, &freq_data, 1);
612 static struct cpufreq_driver powernv_cpufreq_driver = {
613 .name = "powernv-cpufreq",
614 .flags = CPUFREQ_CONST_LOOPS,
615 .init = powernv_cpufreq_cpu_init,
616 .verify = cpufreq_generic_frequency_table_verify,
617 .target_index = powernv_cpufreq_target_index,
618 .get = powernv_cpufreq_get,
619 .stop_cpu = powernv_cpufreq_stop_cpu,
620 .attr = powernv_cpu_freq_attr,
623 static int init_chip_info(void)
625 unsigned int chip[256];
627 unsigned int prev_chip_id = UINT_MAX;
629 for_each_possible_cpu(cpu) {
630 unsigned int id = cpu_to_chip_id(cpu);
632 if (prev_chip_id != id) {
634 chip[nr_chips++] = id;
638 chips = kcalloc(nr_chips, sizeof(struct chip), GFP_KERNEL);
642 for (i = 0; i < nr_chips; i++) {
643 chips[i].id = chip[i];
644 cpumask_copy(&chips[i].mask, cpumask_of_node(chip[i]));
645 INIT_WORK(&chips[i].throttle, powernv_cpufreq_work_fn);
646 for_each_cpu(cpu, &chips[i].mask)
647 per_cpu(chip_info, cpu) = &chips[i];
653 static inline void clean_chip_info(void)
658 static inline void unregister_all_notifiers(void)
660 opal_message_notifier_unregister(OPAL_MSG_OCC,
661 &powernv_cpufreq_opal_nb);
662 unregister_reboot_notifier(&powernv_cpufreq_reboot_nb);
665 static int __init powernv_cpufreq_init(void)
669 /* Don't probe on pseries (guest) platforms */
670 if (!firmware_has_feature(FW_FEATURE_OPAL))
673 /* Discover pstates from device tree and init */
674 rc = init_powernv_pstates();
678 /* Populate chip info */
679 rc = init_chip_info();
683 register_reboot_notifier(&powernv_cpufreq_reboot_nb);
684 opal_message_notifier_register(OPAL_MSG_OCC, &powernv_cpufreq_opal_nb);
686 rc = cpufreq_register_driver(&powernv_cpufreq_driver);
690 pr_info("Failed to register the cpufreq driver (%d)\n", rc);
691 unregister_all_notifiers();
694 pr_info("Platform driver disabled. System does not support PState control\n");
697 module_init(powernv_cpufreq_init);
699 static void __exit powernv_cpufreq_exit(void)
701 cpufreq_unregister_driver(&powernv_cpufreq_driver);
702 unregister_all_notifiers();
705 module_exit(powernv_cpufreq_exit);
707 MODULE_LICENSE("GPL");
708 MODULE_AUTHOR("Vaidyanathan Srinivasan <svaidy at linux.vnet.ibm.com>");