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>
32 #include <asm/cputhreads.h>
33 #include <asm/firmware.h>
35 #include <asm/smp.h> /* Required for cpu_sibling_mask() in UP configs */
38 #define POWERNV_MAX_PSTATES 256
39 #define PMSR_PSAFE_ENABLE (1UL << 30)
40 #define PMSR_SPR_EM_DISABLE (1UL << 31)
41 #define PMSR_MAX(x) ((x >> 32) & 0xFF)
43 static struct cpufreq_frequency_table powernv_freqs[POWERNV_MAX_PSTATES+1];
44 static bool rebooting, throttled, occ_reset;
50 struct work_struct throttle;
57 * Note: The set of pstates consists of contiguous integers, the
58 * smallest of which is indicated by powernv_pstate_info.min, the
59 * largest of which is indicated by powernv_pstate_info.max.
61 * The nominal pstate is the highest non-turbo pstate in this
62 * platform. This is indicated by powernv_pstate_info.nominal.
64 static struct powernv_pstate_info {
69 } powernv_pstate_info;
72 * Initialize the freq table based on data obtained
73 * from the firmware passed via device-tree
75 static int init_powernv_pstates(void)
77 struct device_node *power_mgt;
78 int i, pstate_min, pstate_max, pstate_nominal, nr_pstates = 0;
79 const __be32 *pstate_ids, *pstate_freqs;
80 u32 len_ids, len_freqs;
82 power_mgt = of_find_node_by_path("/ibm,opal/power-mgt");
84 pr_warn("power-mgt node not found\n");
88 if (of_property_read_u32(power_mgt, "ibm,pstate-min", &pstate_min)) {
89 pr_warn("ibm,pstate-min node not found\n");
93 if (of_property_read_u32(power_mgt, "ibm,pstate-max", &pstate_max)) {
94 pr_warn("ibm,pstate-max node not found\n");
98 if (of_property_read_u32(power_mgt, "ibm,pstate-nominal",
100 pr_warn("ibm,pstate-nominal not found\n");
103 pr_info("cpufreq pstate min %d nominal %d max %d\n", pstate_min,
104 pstate_nominal, pstate_max);
106 pstate_ids = of_get_property(power_mgt, "ibm,pstate-ids", &len_ids);
108 pr_warn("ibm,pstate-ids not found\n");
112 pstate_freqs = of_get_property(power_mgt, "ibm,pstate-frequencies-mhz",
115 pr_warn("ibm,pstate-frequencies-mhz not found\n");
119 if (len_ids != len_freqs) {
120 pr_warn("Entries in ibm,pstate-ids and "
121 "ibm,pstate-frequencies-mhz does not match\n");
124 nr_pstates = min(len_ids, len_freqs) / sizeof(u32);
126 pr_warn("No PStates found\n");
130 pr_debug("NR PStates %d\n", nr_pstates);
131 for (i = 0; i < nr_pstates; i++) {
132 u32 id = be32_to_cpu(pstate_ids[i]);
133 u32 freq = be32_to_cpu(pstate_freqs[i]);
135 pr_debug("PState id %d freq %d MHz\n", id, freq);
136 powernv_freqs[i].frequency = freq * 1000; /* kHz */
137 powernv_freqs[i].driver_data = id;
139 /* End of list marker entry */
140 powernv_freqs[i].frequency = CPUFREQ_TABLE_END;
142 powernv_pstate_info.min = pstate_min;
143 powernv_pstate_info.max = pstate_max;
144 powernv_pstate_info.nominal = pstate_nominal;
145 powernv_pstate_info.nr_pstates = nr_pstates;
150 /* Returns the CPU frequency corresponding to the pstate_id. */
151 static unsigned int pstate_id_to_freq(int pstate_id)
155 i = powernv_pstate_info.max - pstate_id;
156 if (i >= powernv_pstate_info.nr_pstates || i < 0) {
157 pr_warn("PState id %d outside of PState table, "
158 "reporting nominal id %d instead\n",
159 pstate_id, powernv_pstate_info.nominal);
160 i = powernv_pstate_info.max - powernv_pstate_info.nominal;
163 return powernv_freqs[i].frequency;
167 * cpuinfo_nominal_freq_show - Show the nominal CPU frequency as indicated by
170 static ssize_t cpuinfo_nominal_freq_show(struct cpufreq_policy *policy,
173 return sprintf(buf, "%u\n",
174 pstate_id_to_freq(powernv_pstate_info.nominal));
177 struct freq_attr cpufreq_freq_attr_cpuinfo_nominal_freq =
178 __ATTR_RO(cpuinfo_nominal_freq);
180 static struct freq_attr *powernv_cpu_freq_attr[] = {
181 &cpufreq_freq_attr_scaling_available_freqs,
182 &cpufreq_freq_attr_cpuinfo_nominal_freq,
186 /* Helper routines */
188 /* Access helpers to power mgt SPR */
190 static inline unsigned long get_pmspr(unsigned long sprn)
194 return mfspr(SPRN_PMCR);
197 return mfspr(SPRN_PMICR);
200 return mfspr(SPRN_PMSR);
205 static inline void set_pmspr(unsigned long sprn, unsigned long val)
209 mtspr(SPRN_PMCR, val);
213 mtspr(SPRN_PMICR, val);
220 * Use objects of this type to query/update
221 * pstates on a remote CPU via smp_call_function.
223 struct powernv_smp_call_data {
229 * powernv_read_cpu_freq: Reads the current frequency on this CPU.
231 * Called via smp_call_function.
233 * Note: The caller of the smp_call_function should pass an argument of
234 * the type 'struct powernv_smp_call_data *' along with this function.
236 * The current frequency on this CPU will be returned via
237 * ((struct powernv_smp_call_data *)arg)->freq;
239 static void powernv_read_cpu_freq(void *arg)
241 unsigned long pmspr_val;
243 struct powernv_smp_call_data *freq_data = arg;
245 pmspr_val = get_pmspr(SPRN_PMSR);
248 * The local pstate id corresponds bits 48..55 in the PMSR.
249 * Note: Watch out for the sign!
251 local_pstate_id = (pmspr_val >> 48) & 0xFF;
252 freq_data->pstate_id = local_pstate_id;
253 freq_data->freq = pstate_id_to_freq(freq_data->pstate_id);
255 pr_debug("cpu %d pmsr %016lX pstate_id %d frequency %d kHz\n",
256 raw_smp_processor_id(), pmspr_val, freq_data->pstate_id,
261 * powernv_cpufreq_get: Returns the CPU frequency as reported by the
262 * firmware for CPU 'cpu'. This value is reported through the sysfs
263 * file cpuinfo_cur_freq.
265 static unsigned int powernv_cpufreq_get(unsigned int cpu)
267 struct powernv_smp_call_data freq_data;
269 smp_call_function_any(cpu_sibling_mask(cpu), powernv_read_cpu_freq,
272 return freq_data.freq;
276 * set_pstate: Sets the pstate on this CPU.
278 * This is called via an smp_call_function.
280 * The caller must ensure that freq_data is of the type
281 * (struct powernv_smp_call_data *) and the pstate_id which needs to be set
282 * on this CPU should be present in freq_data->pstate_id.
284 static void set_pstate(void *freq_data)
287 unsigned long pstate_ul =
288 ((struct powernv_smp_call_data *) freq_data)->pstate_id;
290 val = get_pmspr(SPRN_PMCR);
291 val = val & 0x0000FFFFFFFFFFFFULL;
293 pstate_ul = pstate_ul & 0xFF;
295 /* Set both global(bits 56..63) and local(bits 48..55) PStates */
296 val = val | (pstate_ul << 56) | (pstate_ul << 48);
298 pr_debug("Setting cpu %d pmcr to %016lX\n",
299 raw_smp_processor_id(), val);
300 set_pmspr(SPRN_PMCR, val);
304 * get_nominal_index: Returns the index corresponding to the nominal
305 * pstate in the cpufreq table
307 static inline unsigned int get_nominal_index(void)
309 return powernv_pstate_info.max - powernv_pstate_info.nominal;
312 static void powernv_cpufreq_throttle_check(void *data)
314 unsigned int cpu = smp_processor_id();
318 pmsr = get_pmspr(SPRN_PMSR);
320 for (i = 0; i < nr_chips; i++)
321 if (chips[i].id == cpu_to_chip_id(cpu))
324 /* Check for Pmax Capping */
325 pmsr_pmax = (s8)PMSR_MAX(pmsr);
326 if (pmsr_pmax != powernv_pstate_info.max) {
327 if (chips[i].throttled)
329 chips[i].throttled = true;
330 if (pmsr_pmax < powernv_pstate_info.nominal)
331 pr_crit("CPU %d on Chip %u has Pmax reduced below nominal frequency (%d < %d)\n",
332 cpu, chips[i].id, pmsr_pmax,
333 powernv_pstate_info.nominal);
335 pr_info("CPU %d on Chip %u has Pmax reduced below turbo frequency (%d < %d)\n",
336 cpu, chips[i].id, pmsr_pmax,
337 powernv_pstate_info.max);
338 } else if (chips[i].throttled) {
339 chips[i].throttled = false;
340 pr_info("CPU %d on Chip %u has Pmax restored to %d\n", cpu,
341 chips[i].id, pmsr_pmax);
344 /* Check if Psafe_mode_active is set in PMSR. */
346 if (pmsr & PMSR_PSAFE_ENABLE) {
348 pr_info("Pstate set to safe frequency\n");
351 /* Check if SPR_EM_DISABLE is set in PMSR */
352 if (pmsr & PMSR_SPR_EM_DISABLE) {
354 pr_info("Frequency Control disabled from OS\n");
358 pr_info("PMSR = %16lx\n", pmsr);
359 pr_crit("CPU Frequency could be throttled\n");
364 * powernv_cpufreq_target_index: Sets the frequency corresponding to
365 * the cpufreq table entry indexed by new_index on the cpus in the
368 static int powernv_cpufreq_target_index(struct cpufreq_policy *policy,
369 unsigned int new_index)
371 struct powernv_smp_call_data freq_data;
373 if (unlikely(rebooting) && new_index != get_nominal_index())
377 powernv_cpufreq_throttle_check(NULL);
379 freq_data.pstate_id = powernv_freqs[new_index].driver_data;
382 * Use smp_call_function to send IPI and execute the
383 * mtspr on target CPU. We could do that without IPI
384 * if current CPU is within policy->cpus (core)
386 smp_call_function_any(policy->cpus, set_pstate, &freq_data, 1);
391 static int powernv_cpufreq_cpu_init(struct cpufreq_policy *policy)
395 base = cpu_first_thread_sibling(policy->cpu);
397 for (i = 0; i < threads_per_core; i++)
398 cpumask_set_cpu(base + i, policy->cpus);
400 return cpufreq_table_validate_and_show(policy, powernv_freqs);
403 static int powernv_cpufreq_reboot_notifier(struct notifier_block *nb,
404 unsigned long action, void *unused)
407 struct cpufreq_policy cpu_policy;
410 for_each_online_cpu(cpu) {
411 cpufreq_get_policy(&cpu_policy, cpu);
412 powernv_cpufreq_target_index(&cpu_policy, get_nominal_index());
418 static struct notifier_block powernv_cpufreq_reboot_nb = {
419 .notifier_call = powernv_cpufreq_reboot_notifier,
422 void powernv_cpufreq_work_fn(struct work_struct *work)
424 struct chip *chip = container_of(work, struct chip, throttle);
428 smp_call_function_any(&chip->mask,
429 powernv_cpufreq_throttle_check, NULL, 0);
434 chip->restore = false;
435 cpumask_copy(mask, &chip->mask);
436 for_each_cpu_and(cpu, mask, cpu_online_mask) {
438 struct cpufreq_policy policy;
440 cpufreq_get_policy(&policy, cpu);
441 cpufreq_frequency_table_target(&policy, policy.freq_table,
443 CPUFREQ_RELATION_C, &index);
444 powernv_cpufreq_target_index(&policy, index);
445 for_each_cpu(tcpu, policy.cpus)
446 cpumask_clear_cpu(tcpu, mask);
450 static char throttle_reason[][30] = {
453 "Processor Over Temperature",
454 "Power Supply Failure",
459 static int powernv_cpufreq_occ_msg(struct notifier_block *nb,
460 unsigned long msg_type, void *_msg)
462 struct opal_msg *msg = _msg;
463 struct opal_occ_msg omsg;
466 if (msg_type != OPAL_MSG_OCC)
469 omsg.type = be64_to_cpu(msg->params[0]);
474 pr_info("OCC (On Chip Controller - enforces hard thermal/power limits) Resetting\n");
476 * powernv_cpufreq_throttle_check() is called in
477 * target() callback which can detect the throttle state
478 * for governors like ondemand.
479 * But static governors will not call target() often thus
480 * report throttling here.
484 pr_crit("CPU frequency is throttled for duration\n");
489 pr_info("OCC Loading, CPU frequency is throttled until OCC is started\n");
492 omsg.chip = be64_to_cpu(msg->params[1]);
493 omsg.throttle_status = be64_to_cpu(msg->params[2]);
498 pr_info("OCC Active, CPU frequency is no longer throttled\n");
500 for (i = 0; i < nr_chips; i++) {
501 chips[i].restore = true;
502 schedule_work(&chips[i].throttle);
508 if (omsg.throttle_status &&
509 omsg.throttle_status <= OCC_MAX_THROTTLE_STATUS)
510 pr_info("OCC: Chip %u Pmax reduced due to %s\n",
511 (unsigned int)omsg.chip,
512 throttle_reason[omsg.throttle_status]);
513 else if (!omsg.throttle_status)
514 pr_info("OCC: Chip %u %s\n", (unsigned int)omsg.chip,
515 throttle_reason[omsg.throttle_status]);
519 for (i = 0; i < nr_chips; i++)
520 if (chips[i].id == omsg.chip) {
521 if (!omsg.throttle_status)
522 chips[i].restore = true;
523 schedule_work(&chips[i].throttle);
529 static struct notifier_block powernv_cpufreq_opal_nb = {
530 .notifier_call = powernv_cpufreq_occ_msg,
535 static void powernv_cpufreq_stop_cpu(struct cpufreq_policy *policy)
537 struct powernv_smp_call_data freq_data;
539 freq_data.pstate_id = powernv_pstate_info.min;
540 smp_call_function_single(policy->cpu, set_pstate, &freq_data, 1);
543 static struct cpufreq_driver powernv_cpufreq_driver = {
544 .name = "powernv-cpufreq",
545 .flags = CPUFREQ_CONST_LOOPS,
546 .init = powernv_cpufreq_cpu_init,
547 .verify = cpufreq_generic_frequency_table_verify,
548 .target_index = powernv_cpufreq_target_index,
549 .get = powernv_cpufreq_get,
550 .stop_cpu = powernv_cpufreq_stop_cpu,
551 .attr = powernv_cpu_freq_attr,
554 static int init_chip_info(void)
556 unsigned int chip[256];
558 unsigned int prev_chip_id = UINT_MAX;
560 for_each_possible_cpu(cpu) {
561 unsigned int id = cpu_to_chip_id(cpu);
563 if (prev_chip_id != id) {
565 chip[nr_chips++] = id;
569 chips = kmalloc_array(nr_chips, sizeof(struct chip), GFP_KERNEL);
573 for (i = 0; i < nr_chips; i++) {
574 chips[i].id = chip[i];
575 chips[i].throttled = false;
576 cpumask_copy(&chips[i].mask, cpumask_of_node(chip[i]));
577 INIT_WORK(&chips[i].throttle, powernv_cpufreq_work_fn);
578 chips[i].restore = false;
584 static int __init powernv_cpufreq_init(void)
588 /* Don't probe on pseries (guest) platforms */
589 if (!firmware_has_feature(FW_FEATURE_OPALv3))
592 /* Discover pstates from device tree and init */
593 rc = init_powernv_pstates();
595 pr_info("powernv-cpufreq disabled. System does not support PState control\n");
599 /* Populate chip info */
600 rc = init_chip_info();
604 register_reboot_notifier(&powernv_cpufreq_reboot_nb);
605 opal_message_notifier_register(OPAL_MSG_OCC, &powernv_cpufreq_opal_nb);
606 return cpufreq_register_driver(&powernv_cpufreq_driver);
608 module_init(powernv_cpufreq_init);
610 static void __exit powernv_cpufreq_exit(void)
612 unregister_reboot_notifier(&powernv_cpufreq_reboot_nb);
613 opal_message_notifier_unregister(OPAL_MSG_OCC,
614 &powernv_cpufreq_opal_nb);
615 cpufreq_unregister_driver(&powernv_cpufreq_driver);
617 module_exit(powernv_cpufreq_exit);
619 MODULE_LICENSE("GPL");
620 MODULE_AUTHOR("Vaidyanathan Srinivasan <svaidy at linux.vnet.ibm.com>");