linux/drivers/cpufreq/acpi-cpufreq.c
<<
>>
Prefs
   1/*
   2 * acpi-cpufreq.c - ACPI Processor P-States Driver
   3 *
   4 *  Copyright (C) 2001, 2002 Andy Grover <andrew.grover@intel.com>
   5 *  Copyright (C) 2001, 2002 Paul Diefenbaugh <paul.s.diefenbaugh@intel.com>
   6 *  Copyright (C) 2002 - 2004 Dominik Brodowski <linux@brodo.de>
   7 *  Copyright (C) 2006       Denis Sadykov <denis.m.sadykov@intel.com>
   8 *
   9 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  10 *
  11 *  This program is free software; you can redistribute it and/or modify
  12 *  it under the terms of the GNU General Public License as published by
  13 *  the Free Software Foundation; either version 2 of the License, or (at
  14 *  your option) any later version.
  15 *
  16 *  This program is distributed in the hope that it will be useful, but
  17 *  WITHOUT ANY WARRANTY; without even the implied warranty of
  18 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19 *  General Public License for more details.
  20 *
  21 *  You should have received a copy of the GNU General Public License along
  22 *  with this program; if not, write to the Free Software Foundation, Inc.,
  23 *  59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.
  24 *
  25 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  26 */
  27
  28#include <linux/kernel.h>
  29#include <linux/module.h>
  30#include <linux/init.h>
  31#include <linux/smp.h>
  32#include <linux/sched.h>
  33#include <linux/cpufreq.h>
  34#include <linux/compiler.h>
  35#include <linux/dmi.h>
  36#include <linux/slab.h>
  37
  38#include <linux/acpi.h>
  39#include <linux/io.h>
  40#include <linux/delay.h>
  41#include <linux/uaccess.h>
  42
  43#include <acpi/processor.h>
  44
  45#include <asm/msr.h>
  46#include <asm/processor.h>
  47#include <asm/cpufeature.h>
  48#include "mperf.h"
  49
  50MODULE_AUTHOR("Paul Diefenbaugh, Dominik Brodowski");
  51MODULE_DESCRIPTION("ACPI Processor P-States Driver");
  52MODULE_LICENSE("GPL");
  53
  54#define PFX "acpi-cpufreq: "
  55
  56enum {
  57        UNDEFINED_CAPABLE = 0,
  58        SYSTEM_INTEL_MSR_CAPABLE,
  59        SYSTEM_AMD_MSR_CAPABLE,
  60        SYSTEM_IO_CAPABLE,
  61};
  62
  63#define INTEL_MSR_RANGE         (0xffff)
  64#define AMD_MSR_RANGE           (0x7)
  65
  66#define MSR_K7_HWCR_CPB_DIS     (1ULL << 25)
  67
  68struct acpi_cpufreq_data {
  69        struct acpi_processor_performance *acpi_data;
  70        struct cpufreq_frequency_table *freq_table;
  71        unsigned int resume;
  72        unsigned int cpu_feature;
  73        cpumask_var_t freqdomain_cpus;
  74};
  75
  76static DEFINE_PER_CPU(struct acpi_cpufreq_data *, acfreq_data);
  77
  78/* acpi_perf_data is a pointer to percpu data. */
  79static struct acpi_processor_performance __percpu *acpi_perf_data;
  80
  81static struct cpufreq_driver acpi_cpufreq_driver;
  82
  83static unsigned int acpi_pstate_strict;
  84static bool boost_enabled, boost_supported;
  85static struct msr __percpu *msrs;
  86
  87static bool boost_state(unsigned int cpu)
  88{
  89        u32 lo, hi;
  90        u64 msr;
  91
  92        switch (boot_cpu_data.x86_vendor) {
  93        case X86_VENDOR_INTEL:
  94                rdmsr_on_cpu(cpu, MSR_IA32_MISC_ENABLE, &lo, &hi);
  95                msr = lo | ((u64)hi << 32);
  96                return !(msr & MSR_IA32_MISC_ENABLE_TURBO_DISABLE);
  97        case X86_VENDOR_AMD:
  98                rdmsr_on_cpu(cpu, MSR_K7_HWCR, &lo, &hi);
  99                msr = lo | ((u64)hi << 32);
 100                return !(msr & MSR_K7_HWCR_CPB_DIS);
 101        }
 102        return false;
 103}
 104
 105static void boost_set_msrs(bool enable, const struct cpumask *cpumask)
 106{
 107        u32 cpu;
 108        u32 msr_addr;
 109        u64 msr_mask;
 110
 111        switch (boot_cpu_data.x86_vendor) {
 112        case X86_VENDOR_INTEL:
 113                msr_addr = MSR_IA32_MISC_ENABLE;
 114                msr_mask = MSR_IA32_MISC_ENABLE_TURBO_DISABLE;
 115                break;
 116        case X86_VENDOR_AMD:
 117                msr_addr = MSR_K7_HWCR;
 118                msr_mask = MSR_K7_HWCR_CPB_DIS;
 119                break;
 120        default:
 121                return;
 122        }
 123
 124        rdmsr_on_cpus(cpumask, msr_addr, msrs);
 125
 126        for_each_cpu(cpu, cpumask) {
 127                struct msr *reg = per_cpu_ptr(msrs, cpu);
 128                if (enable)
 129                        reg->q &= ~msr_mask;
 130                else
 131                        reg->q |= msr_mask;
 132        }
 133
 134        wrmsr_on_cpus(cpumask, msr_addr, msrs);
 135}
 136
 137static ssize_t _store_boost(const char *buf, size_t count)
 138{
 139        int ret;
 140        unsigned long val = 0;
 141
 142        if (!boost_supported)
 143                return -EINVAL;
 144
 145        ret = kstrtoul(buf, 10, &val);
 146        if (ret || (val > 1))
 147                return -EINVAL;
 148
 149        if ((val && boost_enabled) || (!val && !boost_enabled))
 150                return count;
 151
 152        get_online_cpus();
 153
 154        boost_set_msrs(val, cpu_online_mask);
 155
 156        put_online_cpus();
 157
 158        boost_enabled = val;
 159        pr_debug("Core Boosting %sabled.\n", val ? "en" : "dis");
 160
 161        return count;
 162}
 163
 164static ssize_t store_global_boost(struct kobject *kobj, struct attribute *attr,
 165                                  const char *buf, size_t count)
 166{
 167        return _store_boost(buf, count);
 168}
 169
 170static ssize_t show_global_boost(struct kobject *kobj,
 171                                 struct attribute *attr, char *buf)
 172{
 173        return sprintf(buf, "%u\n", boost_enabled);
 174}
 175
 176static struct global_attr global_boost = __ATTR(boost, 0644,
 177                                                show_global_boost,
 178                                                store_global_boost);
 179
 180static ssize_t show_freqdomain_cpus(struct cpufreq_policy *policy, char *buf)
 181{
 182        struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 183
 184        return cpufreq_show_cpus(data->freqdomain_cpus, buf);
 185}
 186
 187cpufreq_freq_attr_ro(freqdomain_cpus);
 188
 189#ifdef CONFIG_X86_ACPI_CPUFREQ_CPB
 190static ssize_t store_cpb(struct cpufreq_policy *policy, const char *buf,
 191                         size_t count)
 192{
 193        return _store_boost(buf, count);
 194}
 195
 196static ssize_t show_cpb(struct cpufreq_policy *policy, char *buf)
 197{
 198        return sprintf(buf, "%u\n", boost_enabled);
 199}
 200
 201static struct freq_attr cpb = __ATTR(cpb, 0644, show_cpb, store_cpb);
 202#endif
 203
 204static int check_est_cpu(unsigned int cpuid)
 205{
 206        struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
 207
 208        return cpu_has(cpu, X86_FEATURE_EST);
 209}
 210
 211static int check_amd_hwpstate_cpu(unsigned int cpuid)
 212{
 213        struct cpuinfo_x86 *cpu = &cpu_data(cpuid);
 214
 215        return cpu_has(cpu, X86_FEATURE_HW_PSTATE);
 216}
 217
 218static unsigned extract_io(u32 value, struct acpi_cpufreq_data *data)
 219{
 220        struct acpi_processor_performance *perf;
 221        int i;
 222
 223        perf = data->acpi_data;
 224
 225        for (i = 0; i < perf->state_count; i++) {
 226                if (value == perf->states[i].status)
 227                        return data->freq_table[i].frequency;
 228        }
 229        return 0;
 230}
 231
 232static unsigned extract_msr(u32 msr, struct acpi_cpufreq_data *data)
 233{
 234        int i;
 235        struct acpi_processor_performance *perf;
 236
 237        if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
 238                msr &= AMD_MSR_RANGE;
 239        else
 240                msr &= INTEL_MSR_RANGE;
 241
 242        perf = data->acpi_data;
 243
 244        for (i = 0; data->freq_table[i].frequency != CPUFREQ_TABLE_END; i++) {
 245                if (msr == perf->states[data->freq_table[i].driver_data].status)
 246                        return data->freq_table[i].frequency;
 247        }
 248        return data->freq_table[0].frequency;
 249}
 250
 251static unsigned extract_freq(u32 val, struct acpi_cpufreq_data *data)
 252{
 253        switch (data->cpu_feature) {
 254        case SYSTEM_INTEL_MSR_CAPABLE:
 255        case SYSTEM_AMD_MSR_CAPABLE:
 256                return extract_msr(val, data);
 257        case SYSTEM_IO_CAPABLE:
 258                return extract_io(val, data);
 259        default:
 260                return 0;
 261        }
 262}
 263
 264struct msr_addr {
 265        u32 reg;
 266};
 267
 268struct io_addr {
 269        u16 port;
 270        u8 bit_width;
 271};
 272
 273struct drv_cmd {
 274        unsigned int type;
 275        const struct cpumask *mask;
 276        union {
 277                struct msr_addr msr;
 278                struct io_addr io;
 279        } addr;
 280        u32 val;
 281};
 282
 283/* Called via smp_call_function_single(), on the target CPU */
 284static void do_drv_read(void *_cmd)
 285{
 286        struct drv_cmd *cmd = _cmd;
 287        u32 h;
 288
 289        switch (cmd->type) {
 290        case SYSTEM_INTEL_MSR_CAPABLE:
 291        case SYSTEM_AMD_MSR_CAPABLE:
 292                rdmsr(cmd->addr.msr.reg, cmd->val, h);
 293                break;
 294        case SYSTEM_IO_CAPABLE:
 295                acpi_os_read_port((acpi_io_address)cmd->addr.io.port,
 296                                &cmd->val,
 297                                (u32)cmd->addr.io.bit_width);
 298                break;
 299        default:
 300                break;
 301        }
 302}
 303
 304/* Called via smp_call_function_many(), on the target CPUs */
 305static void do_drv_write(void *_cmd)
 306{
 307        struct drv_cmd *cmd = _cmd;
 308        u32 lo, hi;
 309
 310        switch (cmd->type) {
 311        case SYSTEM_INTEL_MSR_CAPABLE:
 312                rdmsr(cmd->addr.msr.reg, lo, hi);
 313                lo = (lo & ~INTEL_MSR_RANGE) | (cmd->val & INTEL_MSR_RANGE);
 314                wrmsr(cmd->addr.msr.reg, lo, hi);
 315                break;
 316        case SYSTEM_AMD_MSR_CAPABLE:
 317                wrmsr(cmd->addr.msr.reg, cmd->val, 0);
 318                break;
 319        case SYSTEM_IO_CAPABLE:
 320                acpi_os_write_port((acpi_io_address)cmd->addr.io.port,
 321                                cmd->val,
 322                                (u32)cmd->addr.io.bit_width);
 323                break;
 324        default:
 325                break;
 326        }
 327}
 328
 329static void drv_read(struct drv_cmd *cmd)
 330{
 331        int err;
 332        cmd->val = 0;
 333
 334        err = smp_call_function_any(cmd->mask, do_drv_read, cmd, 1);
 335        WARN_ON_ONCE(err);      /* smp_call_function_any() was buggy? */
 336}
 337
 338static void drv_write(struct drv_cmd *cmd)
 339{
 340        int this_cpu;
 341
 342        this_cpu = get_cpu();
 343        if (cpumask_test_cpu(this_cpu, cmd->mask))
 344                do_drv_write(cmd);
 345        smp_call_function_many(cmd->mask, do_drv_write, cmd, 1);
 346        put_cpu();
 347}
 348
 349static u32 get_cur_val(const struct cpumask *mask)
 350{
 351        struct acpi_processor_performance *perf;
 352        struct drv_cmd cmd;
 353
 354        if (unlikely(cpumask_empty(mask)))
 355                return 0;
 356
 357        switch (per_cpu(acfreq_data, cpumask_first(mask))->cpu_feature) {
 358        case SYSTEM_INTEL_MSR_CAPABLE:
 359                cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
 360                cmd.addr.msr.reg = MSR_IA32_PERF_CTL;
 361                break;
 362        case SYSTEM_AMD_MSR_CAPABLE:
 363                cmd.type = SYSTEM_AMD_MSR_CAPABLE;
 364                cmd.addr.msr.reg = MSR_AMD_PERF_CTL;
 365                break;
 366        case SYSTEM_IO_CAPABLE:
 367                cmd.type = SYSTEM_IO_CAPABLE;
 368                perf = per_cpu(acfreq_data, cpumask_first(mask))->acpi_data;
 369                cmd.addr.io.port = perf->control_register.address;
 370                cmd.addr.io.bit_width = perf->control_register.bit_width;
 371                break;
 372        default:
 373                return 0;
 374        }
 375
 376        cmd.mask = mask;
 377        drv_read(&cmd);
 378
 379        pr_debug("get_cur_val = %u\n", cmd.val);
 380
 381        return cmd.val;
 382}
 383
 384static unsigned int get_cur_freq_on_cpu(unsigned int cpu)
 385{
 386        struct acpi_cpufreq_data *data = per_cpu(acfreq_data, cpu);
 387        unsigned int freq;
 388        unsigned int cached_freq;
 389
 390        pr_debug("get_cur_freq_on_cpu (%d)\n", cpu);
 391
 392        if (unlikely(data == NULL ||
 393                     data->acpi_data == NULL || data->freq_table == NULL)) {
 394                return 0;
 395        }
 396
 397        cached_freq = data->freq_table[data->acpi_data->state].frequency;
 398        freq = extract_freq(get_cur_val(cpumask_of(cpu)), data);
 399        if (freq != cached_freq) {
 400                /*
 401                 * The dreaded BIOS frequency change behind our back.
 402                 * Force set the frequency on next target call.
 403                 */
 404                data->resume = 1;
 405        }
 406
 407        pr_debug("cur freq = %u\n", freq);
 408
 409        return freq;
 410}
 411
 412static unsigned int check_freqs(const struct cpumask *mask, unsigned int freq,
 413                                struct acpi_cpufreq_data *data)
 414{
 415        unsigned int cur_freq;
 416        unsigned int i;
 417
 418        for (i = 0; i < 100; i++) {
 419                cur_freq = extract_freq(get_cur_val(mask), data);
 420                if (cur_freq == freq)
 421                        return 1;
 422                udelay(10);
 423        }
 424        return 0;
 425}
 426
 427static int acpi_cpufreq_target(struct cpufreq_policy *policy,
 428                               unsigned int target_freq, unsigned int relation)
 429{
 430        struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 431        struct acpi_processor_performance *perf;
 432        struct cpufreq_freqs freqs;
 433        struct drv_cmd cmd;
 434        unsigned int next_state = 0; /* Index into freq_table */
 435        unsigned int next_perf_state = 0; /* Index into perf table */
 436        int result = 0;
 437
 438        pr_debug("acpi_cpufreq_target %d (%d)\n", target_freq, policy->cpu);
 439
 440        if (unlikely(data == NULL ||
 441             data->acpi_data == NULL || data->freq_table == NULL)) {
 442                return -ENODEV;
 443        }
 444
 445        perf = data->acpi_data;
 446        result = cpufreq_frequency_table_target(policy,
 447                                                data->freq_table,
 448                                                target_freq,
 449                                                relation, &next_state);
 450        if (unlikely(result)) {
 451                result = -ENODEV;
 452                goto out;
 453        }
 454
 455        next_perf_state = data->freq_table[next_state].driver_data;
 456        if (perf->state == next_perf_state) {
 457                if (unlikely(data->resume)) {
 458                        pr_debug("Called after resume, resetting to P%d\n",
 459                                next_perf_state);
 460                        data->resume = 0;
 461                } else {
 462                        pr_debug("Already at target state (P%d)\n",
 463                                next_perf_state);
 464                        goto out;
 465                }
 466        }
 467
 468        switch (data->cpu_feature) {
 469        case SYSTEM_INTEL_MSR_CAPABLE:
 470                cmd.type = SYSTEM_INTEL_MSR_CAPABLE;
 471                cmd.addr.msr.reg = MSR_IA32_PERF_CTL;
 472                cmd.val = (u32) perf->states[next_perf_state].control;
 473                break;
 474        case SYSTEM_AMD_MSR_CAPABLE:
 475                cmd.type = SYSTEM_AMD_MSR_CAPABLE;
 476                cmd.addr.msr.reg = MSR_AMD_PERF_CTL;
 477                cmd.val = (u32) perf->states[next_perf_state].control;
 478                break;
 479        case SYSTEM_IO_CAPABLE:
 480                cmd.type = SYSTEM_IO_CAPABLE;
 481                cmd.addr.io.port = perf->control_register.address;
 482                cmd.addr.io.bit_width = perf->control_register.bit_width;
 483                cmd.val = (u32) perf->states[next_perf_state].control;
 484                break;
 485        default:
 486                result = -ENODEV;
 487                goto out;
 488        }
 489
 490        /* cpufreq holds the hotplug lock, so we are safe from here on */
 491        if (policy->shared_type != CPUFREQ_SHARED_TYPE_ANY)
 492                cmd.mask = policy->cpus;
 493        else
 494                cmd.mask = cpumask_of(policy->cpu);
 495
 496        freqs.old = perf->states[perf->state].core_frequency * 1000;
 497        freqs.new = data->freq_table[next_state].frequency;
 498        cpufreq_notify_transition(policy, &freqs, CPUFREQ_PRECHANGE);
 499
 500        drv_write(&cmd);
 501
 502        if (acpi_pstate_strict) {
 503                if (!check_freqs(cmd.mask, freqs.new, data)) {
 504                        pr_debug("acpi_cpufreq_target failed (%d)\n",
 505                                policy->cpu);
 506                        result = -EAGAIN;
 507                        freqs.new = freqs.old;
 508                }
 509        }
 510
 511        cpufreq_notify_transition(policy, &freqs, CPUFREQ_POSTCHANGE);
 512
 513        if (!result)
 514                perf->state = next_perf_state;
 515
 516out:
 517        return result;
 518}
 519
 520static int acpi_cpufreq_verify(struct cpufreq_policy *policy)
 521{
 522        struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 523
 524        pr_debug("acpi_cpufreq_verify\n");
 525
 526        return cpufreq_frequency_table_verify(policy, data->freq_table);
 527}
 528
 529static unsigned long
 530acpi_cpufreq_guess_freq(struct acpi_cpufreq_data *data, unsigned int cpu)
 531{
 532        struct acpi_processor_performance *perf = data->acpi_data;
 533
 534        if (cpu_khz) {
 535                /* search the closest match to cpu_khz */
 536                unsigned int i;
 537                unsigned long freq;
 538                unsigned long freqn = perf->states[0].core_frequency * 1000;
 539
 540                for (i = 0; i < (perf->state_count-1); i++) {
 541                        freq = freqn;
 542                        freqn = perf->states[i+1].core_frequency * 1000;
 543                        if ((2 * cpu_khz) > (freqn + freq)) {
 544                                perf->state = i;
 545                                return freq;
 546                        }
 547                }
 548                perf->state = perf->state_count-1;
 549                return freqn;
 550        } else {
 551                /* assume CPU is at P0... */
 552                perf->state = 0;
 553                return perf->states[0].core_frequency * 1000;
 554        }
 555}
 556
 557static void free_acpi_perf_data(void)
 558{
 559        unsigned int i;
 560
 561        /* Freeing a NULL pointer is OK, and alloc_percpu zeroes. */
 562        for_each_possible_cpu(i)
 563                free_cpumask_var(per_cpu_ptr(acpi_perf_data, i)
 564                                 ->shared_cpu_map);
 565        free_percpu(acpi_perf_data);
 566}
 567
 568static int boost_notify(struct notifier_block *nb, unsigned long action,
 569                      void *hcpu)
 570{
 571        unsigned cpu = (long)hcpu;
 572        const struct cpumask *cpumask;
 573
 574        cpumask = get_cpu_mask(cpu);
 575
 576        /*
 577         * Clear the boost-disable bit on the CPU_DOWN path so that
 578         * this cpu cannot block the remaining ones from boosting. On
 579         * the CPU_UP path we simply keep the boost-disable flag in
 580         * sync with the current global state.
 581         */
 582
 583        switch (action) {
 584        case CPU_UP_PREPARE:
 585        case CPU_UP_PREPARE_FROZEN:
 586                boost_set_msrs(boost_enabled, cpumask);
 587                break;
 588
 589        case CPU_DOWN_PREPARE:
 590        case CPU_DOWN_PREPARE_FROZEN:
 591                boost_set_msrs(1, cpumask);
 592                break;
 593
 594        default:
 595                break;
 596        }
 597
 598        return NOTIFY_OK;
 599}
 600
 601
 602static struct notifier_block boost_nb = {
 603        .notifier_call          = boost_notify,
 604};
 605
 606/*
 607 * acpi_cpufreq_early_init - initialize ACPI P-States library
 608 *
 609 * Initialize the ACPI P-States library (drivers/acpi/processor_perflib.c)
 610 * in order to determine correct frequency and voltage pairings. We can
 611 * do _PDC and _PSD and find out the processor dependency for the
 612 * actual init that will happen later...
 613 */
 614static int __init acpi_cpufreq_early_init(void)
 615{
 616        unsigned int i;
 617        pr_debug("acpi_cpufreq_early_init\n");
 618
 619        acpi_perf_data = alloc_percpu(struct acpi_processor_performance);
 620        if (!acpi_perf_data) {
 621                pr_debug("Memory allocation error for acpi_perf_data.\n");
 622                return -ENOMEM;
 623        }
 624        for_each_possible_cpu(i) {
 625                if (!zalloc_cpumask_var_node(
 626                        &per_cpu_ptr(acpi_perf_data, i)->shared_cpu_map,
 627                        GFP_KERNEL, cpu_to_node(i))) {
 628
 629                        /* Freeing a NULL pointer is OK: alloc_percpu zeroes. */
 630                        free_acpi_perf_data();
 631                        return -ENOMEM;
 632                }
 633        }
 634
 635        /* Do initialization in ACPI core */
 636        acpi_processor_preregister_performance(acpi_perf_data);
 637        return 0;
 638}
 639
 640#ifdef CONFIG_SMP
 641/*
 642 * Some BIOSes do SW_ANY coordination internally, either set it up in hw
 643 * or do it in BIOS firmware and won't inform about it to OS. If not
 644 * detected, this has a side effect of making CPU run at a different speed
 645 * than OS intended it to run at. Detect it and handle it cleanly.
 646 */
 647static int bios_with_sw_any_bug;
 648
 649static int sw_any_bug_found(const struct dmi_system_id *d)
 650{
 651        bios_with_sw_any_bug = 1;
 652        return 0;
 653}
 654
 655static const struct dmi_system_id sw_any_bug_dmi_table[] = {
 656        {
 657                .callback = sw_any_bug_found,
 658                .ident = "Supermicro Server X6DLP",
 659                .matches = {
 660                        DMI_MATCH(DMI_SYS_VENDOR, "Supermicro"),
 661                        DMI_MATCH(DMI_BIOS_VERSION, "080010"),
 662                        DMI_MATCH(DMI_PRODUCT_NAME, "X6DLP"),
 663                },
 664        },
 665        { }
 666};
 667
 668static int acpi_cpufreq_blacklist(struct cpuinfo_x86 *c)
 669{
 670        /* Intel Xeon Processor 7100 Series Specification Update
 671         * http://www.intel.com/Assets/PDF/specupdate/314554.pdf
 672         * AL30: A Machine Check Exception (MCE) Occurring during an
 673         * Enhanced Intel SpeedStep Technology Ratio Change May Cause
 674         * Both Processor Cores to Lock Up. */
 675        if (c->x86_vendor == X86_VENDOR_INTEL) {
 676                if ((c->x86 == 15) &&
 677                    (c->x86_model == 6) &&
 678                    (c->x86_mask == 8)) {
 679                        printk(KERN_INFO "acpi-cpufreq: Intel(R) "
 680                            "Xeon(R) 7100 Errata AL30, processors may "
 681                            "lock up on frequency changes: disabling "
 682                            "acpi-cpufreq.\n");
 683                        return -ENODEV;
 684                    }
 685                }
 686        return 0;
 687}
 688#endif
 689
 690static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
 691{
 692        unsigned int i;
 693        unsigned int valid_states = 0;
 694        unsigned int cpu = policy->cpu;
 695        struct acpi_cpufreq_data *data;
 696        unsigned int result = 0;
 697        struct cpuinfo_x86 *c = &cpu_data(policy->cpu);
 698        struct acpi_processor_performance *perf;
 699#ifdef CONFIG_SMP
 700        static int blacklisted;
 701#endif
 702
 703        pr_debug("acpi_cpufreq_cpu_init\n");
 704
 705#ifdef CONFIG_SMP
 706        if (blacklisted)
 707                return blacklisted;
 708        blacklisted = acpi_cpufreq_blacklist(c);
 709        if (blacklisted)
 710                return blacklisted;
 711#endif
 712
 713        data = kzalloc(sizeof(struct acpi_cpufreq_data), GFP_KERNEL);
 714        if (!data)
 715                return -ENOMEM;
 716
 717        if (!zalloc_cpumask_var(&data->freqdomain_cpus, GFP_KERNEL)) {
 718                result = -ENOMEM;
 719                goto err_free;
 720        }
 721
 722        data->acpi_data = per_cpu_ptr(acpi_perf_data, cpu);
 723        per_cpu(acfreq_data, cpu) = data;
 724
 725        if (cpu_has(c, X86_FEATURE_CONSTANT_TSC))
 726                acpi_cpufreq_driver.flags |= CPUFREQ_CONST_LOOPS;
 727
 728        result = acpi_processor_register_performance(data->acpi_data, cpu);
 729        if (result)
 730                goto err_free_mask;
 731
 732        perf = data->acpi_data;
 733        policy->shared_type = perf->shared_type;
 734
 735        /*
 736         * Will let policy->cpus know about dependency only when software
 737         * coordination is required.
 738         */
 739        if (policy->shared_type == CPUFREQ_SHARED_TYPE_ALL ||
 740            policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) {
 741                cpumask_copy(policy->cpus, perf->shared_cpu_map);
 742        }
 743        cpumask_copy(data->freqdomain_cpus, perf->shared_cpu_map);
 744
 745#ifdef CONFIG_SMP
 746        dmi_check_system(sw_any_bug_dmi_table);
 747        if (bios_with_sw_any_bug && !policy_is_shared(policy)) {
 748                policy->shared_type = CPUFREQ_SHARED_TYPE_ALL;
 749                cpumask_copy(policy->cpus, cpu_core_mask(cpu));
 750        }
 751
 752        if (check_amd_hwpstate_cpu(cpu) && !acpi_pstate_strict) {
 753                cpumask_clear(policy->cpus);
 754                cpumask_set_cpu(cpu, policy->cpus);
 755                cpumask_copy(data->freqdomain_cpus, cpu_sibling_mask(cpu));
 756                policy->shared_type = CPUFREQ_SHARED_TYPE_HW;
 757                pr_info_once(PFX "overriding BIOS provided _PSD data\n");
 758        }
 759#endif
 760
 761        /* capability check */
 762        if (perf->state_count <= 1) {
 763                pr_debug("No P-States\n");
 764                result = -ENODEV;
 765                goto err_unreg;
 766        }
 767
 768        if (perf->control_register.space_id != perf->status_register.space_id) {
 769                result = -ENODEV;
 770                goto err_unreg;
 771        }
 772
 773        switch (perf->control_register.space_id) {
 774        case ACPI_ADR_SPACE_SYSTEM_IO:
 775                if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
 776                    boot_cpu_data.x86 == 0xf) {
 777                        pr_debug("AMD K8 systems must use native drivers.\n");
 778                        result = -ENODEV;
 779                        goto err_unreg;
 780                }
 781                pr_debug("SYSTEM IO addr space\n");
 782                data->cpu_feature = SYSTEM_IO_CAPABLE;
 783                break;
 784        case ACPI_ADR_SPACE_FIXED_HARDWARE:
 785                pr_debug("HARDWARE addr space\n");
 786                if (check_est_cpu(cpu)) {
 787                        data->cpu_feature = SYSTEM_INTEL_MSR_CAPABLE;
 788                        break;
 789                }
 790                if (check_amd_hwpstate_cpu(cpu)) {
 791                        data->cpu_feature = SYSTEM_AMD_MSR_CAPABLE;
 792                        break;
 793                }
 794                result = -ENODEV;
 795                goto err_unreg;
 796        default:
 797                pr_debug("Unknown addr space %d\n",
 798                        (u32) (perf->control_register.space_id));
 799                result = -ENODEV;
 800                goto err_unreg;
 801        }
 802
 803        data->freq_table = kmalloc(sizeof(struct cpufreq_frequency_table) *
 804                    (perf->state_count+1), GFP_KERNEL);
 805        if (!data->freq_table) {
 806                result = -ENOMEM;
 807                goto err_unreg;
 808        }
 809
 810        /* detect transition latency */
 811        policy->cpuinfo.transition_latency = 0;
 812        for (i = 0; i < perf->state_count; i++) {
 813                if ((perf->states[i].transition_latency * 1000) >
 814                    policy->cpuinfo.transition_latency)
 815                        policy->cpuinfo.transition_latency =
 816                            perf->states[i].transition_latency * 1000;
 817        }
 818
 819        /* Check for high latency (>20uS) from buggy BIOSes, like on T42 */
 820        if (perf->control_register.space_id == ACPI_ADR_SPACE_FIXED_HARDWARE &&
 821            policy->cpuinfo.transition_latency > 20 * 1000) {
 822                policy->cpuinfo.transition_latency = 20 * 1000;
 823                printk_once(KERN_INFO
 824                            "P-state transition latency capped at 20 uS\n");
 825        }
 826
 827        /* table init */
 828        for (i = 0; i < perf->state_count; i++) {
 829                if (i > 0 && perf->states[i].core_frequency >=
 830                    data->freq_table[valid_states-1].frequency / 1000)
 831                        continue;
 832
 833                data->freq_table[valid_states].driver_data = i;
 834                data->freq_table[valid_states].frequency =
 835                    perf->states[i].core_frequency * 1000;
 836                valid_states++;
 837        }
 838        data->freq_table[valid_states].frequency = CPUFREQ_TABLE_END;
 839        perf->state = 0;
 840
 841        result = cpufreq_frequency_table_cpuinfo(policy, data->freq_table);
 842        if (result)
 843                goto err_freqfree;
 844
 845        if (perf->states[0].core_frequency * 1000 != policy->cpuinfo.max_freq)
 846                printk(KERN_WARNING FW_WARN "P-state 0 is not max freq\n");
 847
 848        switch (perf->control_register.space_id) {
 849        case ACPI_ADR_SPACE_SYSTEM_IO:
 850                /* Current speed is unknown and not detectable by IO port */
 851                policy->cur = acpi_cpufreq_guess_freq(data, policy->cpu);
 852                break;
 853        case ACPI_ADR_SPACE_FIXED_HARDWARE:
 854                acpi_cpufreq_driver.get = get_cur_freq_on_cpu;
 855                policy->cur = get_cur_freq_on_cpu(cpu);
 856                break;
 857        default:
 858                break;
 859        }
 860
 861        /* notify BIOS that we exist */
 862        acpi_processor_notify_smm(THIS_MODULE);
 863
 864        /* Check for APERF/MPERF support in hardware */
 865        if (boot_cpu_has(X86_FEATURE_APERFMPERF))
 866                acpi_cpufreq_driver.getavg = cpufreq_get_measured_perf;
 867
 868        pr_debug("CPU%u - ACPI performance management activated.\n", cpu);
 869        for (i = 0; i < perf->state_count; i++)
 870                pr_debug("     %cP%d: %d MHz, %d mW, %d uS\n",
 871                        (i == perf->state ? '*' : ' '), i,
 872                        (u32) perf->states[i].core_frequency,
 873                        (u32) perf->states[i].power,
 874                        (u32) perf->states[i].transition_latency);
 875
 876        cpufreq_frequency_table_get_attr(data->freq_table, policy->cpu);
 877
 878        /*
 879         * the first call to ->target() should result in us actually
 880         * writing something to the appropriate registers.
 881         */
 882        data->resume = 1;
 883
 884        return result;
 885
 886err_freqfree:
 887        kfree(data->freq_table);
 888err_unreg:
 889        acpi_processor_unregister_performance(perf, cpu);
 890err_free_mask:
 891        free_cpumask_var(data->freqdomain_cpus);
 892err_free:
 893        kfree(data);
 894        per_cpu(acfreq_data, cpu) = NULL;
 895
 896        return result;
 897}
 898
 899static int acpi_cpufreq_cpu_exit(struct cpufreq_policy *policy)
 900{
 901        struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 902
 903        pr_debug("acpi_cpufreq_cpu_exit\n");
 904
 905        if (data) {
 906                cpufreq_frequency_table_put_attr(policy->cpu);
 907                per_cpu(acfreq_data, policy->cpu) = NULL;
 908                acpi_processor_unregister_performance(data->acpi_data,
 909                                                      policy->cpu);
 910                free_cpumask_var(data->freqdomain_cpus);
 911                kfree(data->freq_table);
 912                kfree(data);
 913        }
 914
 915        return 0;
 916}
 917
 918static int acpi_cpufreq_resume(struct cpufreq_policy *policy)
 919{
 920        struct acpi_cpufreq_data *data = per_cpu(acfreq_data, policy->cpu);
 921
 922        pr_debug("acpi_cpufreq_resume\n");
 923
 924        data->resume = 1;
 925
 926        return 0;
 927}
 928
 929static struct freq_attr *acpi_cpufreq_attr[] = {
 930        &cpufreq_freq_attr_scaling_available_freqs,
 931        &freqdomain_cpus,
 932        NULL,   /* this is a placeholder for cpb, do not remove */
 933        NULL,
 934};
 935
 936static struct cpufreq_driver acpi_cpufreq_driver = {
 937        .verify         = acpi_cpufreq_verify,
 938        .target         = acpi_cpufreq_target,
 939        .bios_limit     = acpi_processor_get_bios_limit,
 940        .init           = acpi_cpufreq_cpu_init,
 941        .exit           = acpi_cpufreq_cpu_exit,
 942        .resume         = acpi_cpufreq_resume,
 943        .name           = "acpi-cpufreq",
 944        .owner          = THIS_MODULE,
 945        .attr           = acpi_cpufreq_attr,
 946};
 947
 948static void __init acpi_cpufreq_boost_init(void)
 949{
 950        if (boot_cpu_has(X86_FEATURE_CPB) || boot_cpu_has(X86_FEATURE_IDA)) {
 951                msrs = msrs_alloc();
 952
 953                if (!msrs)
 954                        return;
 955
 956                boost_supported = true;
 957                boost_enabled = boost_state(0);
 958
 959                get_online_cpus();
 960
 961                /* Force all MSRs to the same value */
 962                boost_set_msrs(boost_enabled, cpu_online_mask);
 963
 964                register_cpu_notifier(&boost_nb);
 965
 966                put_online_cpus();
 967        } else
 968                global_boost.attr.mode = 0444;
 969
 970        /* We create the boost file in any case, though for systems without
 971         * hardware support it will be read-only and hardwired to return 0.
 972         */
 973        if (cpufreq_sysfs_create_file(&(global_boost.attr)))
 974                pr_warn(PFX "could not register global boost sysfs file\n");
 975        else
 976                pr_debug("registered global boost sysfs file\n");
 977}
 978
 979static void __exit acpi_cpufreq_boost_exit(void)
 980{
 981        cpufreq_sysfs_remove_file(&(global_boost.attr));
 982
 983        if (msrs) {
 984                unregister_cpu_notifier(&boost_nb);
 985
 986                msrs_free(msrs);
 987                msrs = NULL;
 988        }
 989}
 990
 991static int __init acpi_cpufreq_init(void)
 992{
 993        int ret;
 994
 995        if (acpi_disabled)
 996                return 0;
 997
 998        pr_debug("acpi_cpufreq_init\n");
 999
1000        ret = acpi_cpufreq_early_init();
1001        if (ret)
1002                return ret;
1003
1004#ifdef CONFIG_X86_ACPI_CPUFREQ_CPB
1005        /* this is a sysfs file with a strange name and an even stranger
1006         * semantic - per CPU instantiation, but system global effect.
1007         * Lets enable it only on AMD CPUs for compatibility reasons and
1008         * only if configured. This is considered legacy code, which
1009         * will probably be removed at some point in the future.
1010         */
1011        if (check_amd_hwpstate_cpu(0)) {
1012                struct freq_attr **iter;
1013
1014                pr_debug("adding sysfs entry for cpb\n");
1015
1016                for (iter = acpi_cpufreq_attr; *iter != NULL; iter++)
1017                        ;
1018
1019                /* make sure there is a terminator behind it */
1020                if (iter[1] == NULL)
1021                        *iter = &cpb;
1022        }
1023#endif
1024
1025        ret = cpufreq_register_driver(&acpi_cpufreq_driver);
1026        if (ret)
1027                free_acpi_perf_data();
1028        else
1029                acpi_cpufreq_boost_init();
1030
1031        return ret;
1032}
1033
1034static void __exit acpi_cpufreq_exit(void)
1035{
1036        pr_debug("acpi_cpufreq_exit\n");
1037
1038        acpi_cpufreq_boost_exit();
1039
1040        cpufreq_unregister_driver(&acpi_cpufreq_driver);
1041
1042        free_acpi_perf_data();
1043}
1044
1045module_param(acpi_pstate_strict, uint, 0644);
1046MODULE_PARM_DESC(acpi_pstate_strict,
1047        "value 0 or non-zero. non-zero -> strict ACPI checks are "
1048        "performed during frequency changes.");
1049
1050late_initcall(acpi_cpufreq_init);
1051module_exit(acpi_cpufreq_exit);
1052
1053static const struct x86_cpu_id acpi_cpufreq_ids[] = {
1054        X86_FEATURE_MATCH(X86_FEATURE_ACPI),
1055        X86_FEATURE_MATCH(X86_FEATURE_HW_PSTATE),
1056        {}
1057};
1058MODULE_DEVICE_TABLE(x86cpu, acpi_cpufreq_ids);
1059
1060static const struct acpi_device_id processor_device_ids[] = {
1061        {ACPI_PROCESSOR_OBJECT_HID, },
1062        {ACPI_PROCESSOR_DEVICE_HID, },
1063        {},
1064};
1065MODULE_DEVICE_TABLE(acpi, processor_device_ids);
1066
1067MODULE_ALIAS("acpi");
1068
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.