drivers/cpufreq/powernv-cpufreq.c

   1 /*
   2  * POWERNV cpufreq driver for the IBM POWER processors
   3  *
   4  * (C) Copyright IBM 2014
   5  *
   6  * Author: Vaidyanathan Srinivasan <svaidy at linux.vnet.ibm.com>
   7  *
   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)
  11  * any later version.
  12  *
  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.
  17  *
  18  */
  19
  20 #define pr_fmt(fmt)     "powernv-cpufreq: " fmt
  21
  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>
  28 #include <linux/of.h>
  29
  30 #include <asm/cputhreads.h>
  31 #include <asm/reg.h>
  32 #include <asm/smp.h> /* Required for cpu_sibling_mask() in UP configs */
  33
  34 #define POWERNV_MAX_PSTATES     256
  35
  36 static struct cpufreq_frequency_table powernv_freqs[POWERNV_MAX_PSTATES+1];
  37
  38 /*
  39  * Note: The set of pstates consists of contiguous integers, the
  40  * smallest of which is indicated by powernv_pstate_info.min, the
  41  * largest of which is indicated by powernv_pstate_info.max.
  42  *
  43  * The nominal pstate is the highest non-turbo pstate in this
  44  * platform. This is indicated by powernv_pstate_info.nominal.
  45  */
  46 static struct powernv_pstate_info {
  47         int min;
  48         int max;
  49         int nominal;
  50         int nr_pstates;
  51 } powernv_pstate_info;
  52
  53 /*
  54  * Initialize the freq table based on data obtained
  55  * from the firmware passed via device-tree
  56  */
  57 static int init_powernv_pstates(void)
  58 {
  59         struct device_node *power_mgt;
  60         int i, pstate_min, pstate_max, pstate_nominal, nr_pstates = 0;
  61         const __be32 *pstate_ids, *pstate_freqs;
  62         u32 len_ids, len_freqs;
  63
  64         power_mgt = of_find_node_by_path("/ibm,opal/power-mgt");
  65         if (!power_mgt) {
  66                 pr_warn("power-mgt node not found\n");
  67                 return -ENODEV;
  68         }
  69
  70         if (of_property_read_u32(power_mgt, "ibm,pstate-min", &pstate_min)) {
  71                 pr_warn("ibm,pstate-min node not found\n");
  72                 return -ENODEV;
  73         }
  74
  75         if (of_property_read_u32(power_mgt, "ibm,pstate-max", &pstate_max)) {
  76                 pr_warn("ibm,pstate-max node not found\n");
  77                 return -ENODEV;
  78         }
  79
  80         if (of_property_read_u32(power_mgt, "ibm,pstate-nominal",
  81                                  &pstate_nominal)) {
  82                 pr_warn("ibm,pstate-nominal not found\n");
  83                 return -ENODEV;
  84         }
  85         pr_info("cpufreq pstate min %d nominal %d max %d\n", pstate_min,
  86                 pstate_nominal, pstate_max);
  87
  88         pstate_ids = of_get_property(power_mgt, "ibm,pstate-ids", &len_ids);
  89         if (!pstate_ids) {
  90                 pr_warn("ibm,pstate-ids not found\n");
  91                 return -ENODEV;
  92         }
  93
  94         pstate_freqs = of_get_property(power_mgt, "ibm,pstate-frequencies-mhz",
  95                                       &len_freqs);
  96         if (!pstate_freqs) {
  97                 pr_warn("ibm,pstate-frequencies-mhz not found\n");
  98                 return -ENODEV;
  99         }
 100
 101         WARN_ON(len_ids != len_freqs);
 102         nr_pstates = min(len_ids, len_freqs) / sizeof(u32);
 103         if (!nr_pstates) {
 104                 pr_warn("No PStates found\n");
 105                 return -ENODEV;
 106         }
 107
 108         pr_debug("NR PStates %d\n", nr_pstates);
 109         for (i = 0; i < nr_pstates; i++) {
 110                 u32 id = be32_to_cpu(pstate_ids[i]);
 111                 u32 freq = be32_to_cpu(pstate_freqs[i]);
 112
 113                 pr_debug("PState id %d freq %d MHz\n", id, freq);
 114                 powernv_freqs[i].frequency = freq * 1000; /* kHz */
 115                 powernv_freqs[i].driver_data = id;
 116         }
 117         /* End of list marker entry */
 118         powernv_freqs[i].frequency = CPUFREQ_TABLE_END;
 119
 120         powernv_pstate_info.min = pstate_min;
 121         powernv_pstate_info.max = pstate_max;
 122         powernv_pstate_info.nominal = pstate_nominal;
 123         powernv_pstate_info.nr_pstates = nr_pstates;
 124
 125         return 0;
 126 }
 127
 128 /* Returns the CPU frequency corresponding to the pstate_id. */
 129 static unsigned int pstate_id_to_freq(int pstate_id)
 130 {
 131         int i;
 132
 133         i = powernv_pstate_info.max - pstate_id;
 134         BUG_ON(i >= powernv_pstate_info.nr_pstates || i < 0);
 135
 136         return powernv_freqs[i].frequency;
 137 }
 138
 139 /*
 140  * cpuinfo_nominal_freq_show - Show the nominal CPU frequency as indicated by
 141  * the firmware
 142  */
 143 static ssize_t cpuinfo_nominal_freq_show(struct cpufreq_policy *policy,
 144                                         char *buf)
 145 {
 146         return sprintf(buf, "%u\n",
 147                 pstate_id_to_freq(powernv_pstate_info.nominal));
 148 }
 149
 150 struct freq_attr cpufreq_freq_attr_cpuinfo_nominal_freq =
 151         __ATTR_RO(cpuinfo_nominal_freq);
 152
 153 static struct freq_attr *powernv_cpu_freq_attr[] = {
 154         &cpufreq_freq_attr_scaling_available_freqs,
 155         &cpufreq_freq_attr_cpuinfo_nominal_freq,
 156         NULL,
 157 };
 158
 159 /* Helper routines */
 160
 161 /* Access helpers to power mgt SPR */
 162
 163 static inline unsigned long get_pmspr(unsigned long sprn)
 164 {
 165         switch (sprn) {
 166         case SPRN_PMCR:
 167                 return mfspr(SPRN_PMCR);
 168
 169         case SPRN_PMICR:
 170                 return mfspr(SPRN_PMICR);
 171
 172         case SPRN_PMSR:
 173                 return mfspr(SPRN_PMSR);
 174         }
 175         BUG();
 176 }
 177
 178 static inline void set_pmspr(unsigned long sprn, unsigned long val)
 179 {
 180         switch (sprn) {
 181         case SPRN_PMCR:
 182                 mtspr(SPRN_PMCR, val);
 183                 return;
 184
 185         case SPRN_PMICR:
 186                 mtspr(SPRN_PMICR, val);
 187                 return;
 188         }
 189         BUG();
 190 }
 191
 192 /*
 193  * Use objects of this type to query/update
 194  * pstates on a remote CPU via smp_call_function.
 195  */
 196 struct powernv_smp_call_data {
 197         unsigned int freq;
 198         int pstate_id;
 199 };
 200
 201 /*
 202  * powernv_read_cpu_freq: Reads the current frequency on this CPU.
 203  *
 204  * Called via smp_call_function.
 205  *
 206  * Note: The caller of the smp_call_function should pass an argument of
 207  * the type 'struct powernv_smp_call_data *' along with this function.
 208  *
 209  * The current frequency on this CPU will be returned via
 210  * ((struct powernv_smp_call_data *)arg)->freq;
 211  */
 212 static void powernv_read_cpu_freq(void *arg)
 213 {
 214         unsigned long pmspr_val;
 215         s8 local_pstate_id;
 216         struct powernv_smp_call_data *freq_data = arg;
 217
 218         pmspr_val = get_pmspr(SPRN_PMSR);
 219
 220         /*
 221          * The local pstate id corresponds bits 48..55 in the PMSR.
 222          * Note: Watch out for the sign!
 223          */
 224         local_pstate_id = (pmspr_val >> 48) & 0xFF;
 225         freq_data->pstate_id = local_pstate_id;
 226         freq_data->freq = pstate_id_to_freq(freq_data->pstate_id);
 227
 228         pr_debug("cpu %d pmsr %016lX pstate_id %d frequency %d kHz\n",
 229                 raw_smp_processor_id(), pmspr_val, freq_data->pstate_id,
 230                 freq_data->freq);
 231 }
 232
 233 /*
 234  * powernv_cpufreq_get: Returns the CPU frequency as reported by the
 235  * firmware for CPU 'cpu'. This value is reported through the sysfs
 236  * file cpuinfo_cur_freq.
 237  */
 238 unsigned int powernv_cpufreq_get(unsigned int cpu)
 239 {
 240         struct powernv_smp_call_data freq_data;
 241
 242         smp_call_function_any(cpu_sibling_mask(cpu), powernv_read_cpu_freq,
 243                         &freq_data, 1);
 244
 245         return freq_data.freq;
 246 }
 247
 248 /*
 249  * set_pstate: Sets the pstate on this CPU.
 250  *
 251  * This is called via an smp_call_function.
 252  *
 253  * The caller must ensure that freq_data is of the type
 254  * (struct powernv_smp_call_data *) and the pstate_id which needs to be set
 255  * on this CPU should be present in freq_data->pstate_id.
 256  */
 257 static void set_pstate(void *freq_data)
 258 {
 259         unsigned long val;
 260         unsigned long pstate_ul =
 261                 ((struct powernv_smp_call_data *) freq_data)->pstate_id;
 262
 263         val = get_pmspr(SPRN_PMCR);
 264         val = val & 0x0000FFFFFFFFFFFFULL;
 265
 266         pstate_ul = pstate_ul & 0xFF;
 267
 268         /* Set both global(bits 56..63) and local(bits 48..55) PStates */
 269         val = val | (pstate_ul << 56) | (pstate_ul << 48);
 270
 271         pr_debug("Setting cpu %d pmcr to %016lX\n",
 272                         raw_smp_processor_id(), val);
 273         set_pmspr(SPRN_PMCR, val);
 274 }
 275
 276 /*
 277  * powernv_cpufreq_target_index: Sets the frequency corresponding to
 278  * the cpufreq table entry indexed by new_index on the cpus in the
 279  * mask policy->cpus
 280  */
 281 static int powernv_cpufreq_target_index(struct cpufreq_policy *policy,
 282                                         unsigned int new_index)
 283 {
 284         struct powernv_smp_call_data freq_data;
 285
 286         freq_data.pstate_id = powernv_freqs[new_index].driver_data;
 287
 288         /*
 289          * Use smp_call_function to send IPI and execute the
 290          * mtspr on target CPU.  We could do that without IPI
 291          * if current CPU is within policy->cpus (core)
 292          */
 293         smp_call_function_any(policy->cpus, set_pstate, &freq_data, 1);
 294
 295         return 0;
 296 }
 297
 298 static int powernv_cpufreq_cpu_init(struct cpufreq_policy *policy)
 299 {
 300         int base, i;
 301
 302         base = cpu_first_thread_sibling(policy->cpu);
 303
 304         for (i = 0; i < threads_per_core; i++)
 305                 cpumask_set_cpu(base + i, policy->cpus);
 306
 307         return cpufreq_table_validate_and_show(policy, powernv_freqs);
 308 }
 309
 310 static struct cpufreq_driver powernv_cpufreq_driver = {
 311         .name           = "powernv-cpufreq",
 312         .flags          = CPUFREQ_CONST_LOOPS,
 313         .init           = powernv_cpufreq_cpu_init,
 314         .verify         = cpufreq_generic_frequency_table_verify,
 315         .target_index   = powernv_cpufreq_target_index,
 316         .get            = powernv_cpufreq_get,
 317         .attr           = powernv_cpu_freq_attr,
 318 };
 319
 320 static int __init powernv_cpufreq_init(void)
 321 {
 322         int rc = 0;
 323
 324         /* Discover pstates from device tree and init */
 325         rc = init_powernv_pstates();
 326         if (rc) {
 327                 pr_info("powernv-cpufreq disabled. System does not support PState control\n");
 328                 return rc;
 329         }
 330
 331         return cpufreq_register_driver(&powernv_cpufreq_driver);
 332 }
 333 module_init(powernv_cpufreq_init);
 334
 335 static void __exit powernv_cpufreq_exit(void)
 336 {
 337         cpufreq_unregister_driver(&powernv_cpufreq_driver);
 338 }
 339 module_exit(powernv_cpufreq_exit);
 340
 341 MODULE_LICENSE("GPL");
 342 MODULE_AUTHOR("Vaidyanathan Srinivasan <svaidy at linux.vnet.ibm.com>");