David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2 | /* |
| 3 | * Pentium 4/Xeon CPU on demand clock modulation/speed scaling |
| 4 | * (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de> |
| 5 | * (C) 2002 Zwane Mwaikambo <zwane@commfireservices.com> |
| 6 | * (C) 2002 Arjan van de Ven <arjanv@redhat.com> |
| 7 | * (C) 2002 Tora T. Engstad |
| 8 | * All Rights Reserved |
| 9 | * |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 10 | * The author(s) of this software shall not be held liable for damages |
| 11 | * of any nature resulting due to the use of this software. This |
| 12 | * software is provided AS-IS with no warranties. |
| 13 | * |
| 14 | * Date Errata Description |
| 15 | * 20020525 N44, O17 12.5% or 25% DC causes lockup |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 16 | */ |
| 17 | |
| 18 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 19 | |
| 20 | #include <linux/kernel.h> |
| 21 | #include <linux/module.h> |
| 22 | #include <linux/init.h> |
| 23 | #include <linux/smp.h> |
| 24 | #include <linux/cpufreq.h> |
| 25 | #include <linux/cpumask.h> |
| 26 | #include <linux/timex.h> |
| 27 | |
| 28 | #include <asm/processor.h> |
| 29 | #include <asm/msr.h> |
| 30 | #include <asm/timer.h> |
| 31 | #include <asm/cpu_device_id.h> |
| 32 | |
| 33 | #include "speedstep-lib.h" |
| 34 | |
| 35 | /* |
| 36 | * Duty Cycle (3bits), note DC_DISABLE is not specified in |
| 37 | * intel docs i just use it to mean disable |
| 38 | */ |
| 39 | enum { |
| 40 | DC_RESV, DC_DFLT, DC_25PT, DC_38PT, DC_50PT, |
| 41 | DC_64PT, DC_75PT, DC_88PT, DC_DISABLE |
| 42 | }; |
| 43 | |
| 44 | #define DC_ENTRIES 8 |
| 45 | |
| 46 | |
| 47 | static int has_N44_O17_errata[NR_CPUS]; |
| 48 | static unsigned int stock_freq; |
| 49 | static struct cpufreq_driver p4clockmod_driver; |
| 50 | static unsigned int cpufreq_p4_get(unsigned int cpu); |
| 51 | |
| 52 | static int cpufreq_p4_setdc(unsigned int cpu, unsigned int newstate) |
| 53 | { |
| 54 | u32 l, h; |
| 55 | |
| 56 | if ((newstate > DC_DISABLE) || (newstate == DC_RESV)) |
| 57 | return -EINVAL; |
| 58 | |
| 59 | rdmsr_on_cpu(cpu, MSR_IA32_THERM_STATUS, &l, &h); |
| 60 | |
| 61 | if (l & 0x01) |
| 62 | pr_debug("CPU#%d currently thermal throttled\n", cpu); |
| 63 | |
| 64 | if (has_N44_O17_errata[cpu] && |
| 65 | (newstate == DC_25PT || newstate == DC_DFLT)) |
| 66 | newstate = DC_38PT; |
| 67 | |
| 68 | rdmsr_on_cpu(cpu, MSR_IA32_THERM_CONTROL, &l, &h); |
| 69 | if (newstate == DC_DISABLE) { |
| 70 | pr_debug("CPU#%d disabling modulation\n", cpu); |
| 71 | wrmsr_on_cpu(cpu, MSR_IA32_THERM_CONTROL, l & ~(1<<4), h); |
| 72 | } else { |
| 73 | pr_debug("CPU#%d setting duty cycle to %d%%\n", |
| 74 | cpu, ((125 * newstate) / 10)); |
| 75 | /* bits 63 - 5 : reserved |
| 76 | * bit 4 : enable/disable |
| 77 | * bits 3-1 : duty cycle |
| 78 | * bit 0 : reserved |
| 79 | */ |
| 80 | l = (l & ~14); |
| 81 | l = l | (1<<4) | ((newstate & 0x7)<<1); |
| 82 | wrmsr_on_cpu(cpu, MSR_IA32_THERM_CONTROL, l, h); |
| 83 | } |
| 84 | |
| 85 | return 0; |
| 86 | } |
| 87 | |
| 88 | |
| 89 | static struct cpufreq_frequency_table p4clockmod_table[] = { |
| 90 | {0, DC_RESV, CPUFREQ_ENTRY_INVALID}, |
| 91 | {0, DC_DFLT, 0}, |
| 92 | {0, DC_25PT, 0}, |
| 93 | {0, DC_38PT, 0}, |
| 94 | {0, DC_50PT, 0}, |
| 95 | {0, DC_64PT, 0}, |
| 96 | {0, DC_75PT, 0}, |
| 97 | {0, DC_88PT, 0}, |
| 98 | {0, DC_DISABLE, 0}, |
| 99 | {0, DC_RESV, CPUFREQ_TABLE_END}, |
| 100 | }; |
| 101 | |
| 102 | |
| 103 | static int cpufreq_p4_target(struct cpufreq_policy *policy, unsigned int index) |
| 104 | { |
| 105 | int i; |
| 106 | |
| 107 | /* run on each logical CPU, |
| 108 | * see section 13.15.3 of IA32 Intel Architecture Software |
| 109 | * Developer's Manual, Volume 3 |
| 110 | */ |
| 111 | for_each_cpu(i, policy->cpus) |
| 112 | cpufreq_p4_setdc(i, p4clockmod_table[index].driver_data); |
| 113 | |
| 114 | return 0; |
| 115 | } |
| 116 | |
| 117 | |
| 118 | static unsigned int cpufreq_p4_get_frequency(struct cpuinfo_x86 *c) |
| 119 | { |
| 120 | if (c->x86 == 0x06) { |
| 121 | if (cpu_has(c, X86_FEATURE_EST)) |
| 122 | pr_warn_once("Warning: EST-capable CPU detected. The acpi-cpufreq module offers voltage scaling in addition to frequency scaling. You should use that instead of p4-clockmod, if possible.\n"); |
| 123 | switch (c->x86_model) { |
| 124 | case 0x0E: /* Core */ |
| 125 | case 0x0F: /* Core Duo */ |
| 126 | case 0x16: /* Celeron Core */ |
| 127 | case 0x1C: /* Atom */ |
| 128 | p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS; |
| 129 | return speedstep_get_frequency(SPEEDSTEP_CPU_PCORE); |
| 130 | case 0x0D: /* Pentium M (Dothan) */ |
| 131 | p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS; |
| 132 | /* fall through */ |
| 133 | case 0x09: /* Pentium M (Banias) */ |
| 134 | return speedstep_get_frequency(SPEEDSTEP_CPU_PM); |
| 135 | } |
| 136 | } |
| 137 | |
| 138 | if (c->x86 != 0xF) |
| 139 | return 0; |
| 140 | |
| 141 | /* on P-4s, the TSC runs with constant frequency independent whether |
| 142 | * throttling is active or not. */ |
| 143 | p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS; |
| 144 | |
| 145 | if (speedstep_detect_processor() == SPEEDSTEP_CPU_P4M) { |
| 146 | pr_warn("Warning: Pentium 4-M detected. The speedstep-ich or acpi cpufreq modules offer voltage scaling in addition of frequency scaling. You should use either one instead of p4-clockmod, if possible.\n"); |
| 147 | return speedstep_get_frequency(SPEEDSTEP_CPU_P4M); |
| 148 | } |
| 149 | |
| 150 | return speedstep_get_frequency(SPEEDSTEP_CPU_P4D); |
| 151 | } |
| 152 | |
| 153 | |
| 154 | |
| 155 | static int cpufreq_p4_cpu_init(struct cpufreq_policy *policy) |
| 156 | { |
| 157 | struct cpuinfo_x86 *c = &cpu_data(policy->cpu); |
| 158 | int cpuid = 0; |
| 159 | unsigned int i; |
| 160 | |
| 161 | #ifdef CONFIG_SMP |
| 162 | cpumask_copy(policy->cpus, topology_sibling_cpumask(policy->cpu)); |
| 163 | #endif |
| 164 | |
| 165 | /* Errata workaround */ |
| 166 | cpuid = (c->x86 << 8) | (c->x86_model << 4) | c->x86_stepping; |
| 167 | switch (cpuid) { |
| 168 | case 0x0f07: |
| 169 | case 0x0f0a: |
| 170 | case 0x0f11: |
| 171 | case 0x0f12: |
| 172 | has_N44_O17_errata[policy->cpu] = 1; |
| 173 | pr_debug("has errata -- disabling low frequencies\n"); |
| 174 | } |
| 175 | |
| 176 | if (speedstep_detect_processor() == SPEEDSTEP_CPU_P4D && |
| 177 | c->x86_model < 2) { |
| 178 | /* switch to maximum frequency and measure result */ |
| 179 | cpufreq_p4_setdc(policy->cpu, DC_DISABLE); |
| 180 | recalibrate_cpu_khz(); |
| 181 | } |
| 182 | /* get max frequency */ |
| 183 | stock_freq = cpufreq_p4_get_frequency(c); |
| 184 | if (!stock_freq) |
| 185 | return -EINVAL; |
| 186 | |
| 187 | /* table init */ |
| 188 | for (i = 1; (p4clockmod_table[i].frequency != CPUFREQ_TABLE_END); i++) { |
| 189 | if ((i < 2) && (has_N44_O17_errata[policy->cpu])) |
| 190 | p4clockmod_table[i].frequency = CPUFREQ_ENTRY_INVALID; |
| 191 | else |
| 192 | p4clockmod_table[i].frequency = (stock_freq * i)/8; |
| 193 | } |
| 194 | |
| 195 | /* cpuinfo and default policy values */ |
| 196 | |
| 197 | /* the transition latency is set to be 1 higher than the maximum |
| 198 | * transition latency of the ondemand governor */ |
| 199 | policy->cpuinfo.transition_latency = 10000001; |
| 200 | policy->freq_table = &p4clockmod_table[0]; |
| 201 | |
| 202 | return 0; |
| 203 | } |
| 204 | |
| 205 | |
| 206 | static unsigned int cpufreq_p4_get(unsigned int cpu) |
| 207 | { |
| 208 | u32 l, h; |
| 209 | |
| 210 | rdmsr_on_cpu(cpu, MSR_IA32_THERM_CONTROL, &l, &h); |
| 211 | |
| 212 | if (l & 0x10) { |
| 213 | l = l >> 1; |
| 214 | l &= 0x7; |
| 215 | } else |
| 216 | l = DC_DISABLE; |
| 217 | |
| 218 | if (l != DC_DISABLE) |
| 219 | return stock_freq * l / 8; |
| 220 | |
| 221 | return stock_freq; |
| 222 | } |
| 223 | |
| 224 | static struct cpufreq_driver p4clockmod_driver = { |
| 225 | .verify = cpufreq_generic_frequency_table_verify, |
| 226 | .target_index = cpufreq_p4_target, |
| 227 | .init = cpufreq_p4_cpu_init, |
| 228 | .get = cpufreq_p4_get, |
| 229 | .name = "p4-clockmod", |
| 230 | .attr = cpufreq_generic_attr, |
| 231 | }; |
| 232 | |
| 233 | static const struct x86_cpu_id cpufreq_p4_id[] = { |
| 234 | { X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_ACC }, |
| 235 | {} |
| 236 | }; |
| 237 | |
| 238 | /* |
| 239 | * Intentionally no MODULE_DEVICE_TABLE here: this driver should not |
| 240 | * be auto loaded. Please don't add one. |
| 241 | */ |
| 242 | |
| 243 | static int __init cpufreq_p4_init(void) |
| 244 | { |
| 245 | int ret; |
| 246 | |
| 247 | /* |
| 248 | * THERM_CONTROL is architectural for IA32 now, so |
| 249 | * we can rely on the capability checks |
| 250 | */ |
| 251 | if (!x86_match_cpu(cpufreq_p4_id) || !boot_cpu_has(X86_FEATURE_ACPI)) |
| 252 | return -ENODEV; |
| 253 | |
| 254 | ret = cpufreq_register_driver(&p4clockmod_driver); |
| 255 | if (!ret) |
| 256 | pr_info("P4/Xeon(TM) CPU On-Demand Clock Modulation available\n"); |
| 257 | |
| 258 | return ret; |
| 259 | } |
| 260 | |
| 261 | |
| 262 | static void __exit cpufreq_p4_exit(void) |
| 263 | { |
| 264 | cpufreq_unregister_driver(&p4clockmod_driver); |
| 265 | } |
| 266 | |
| 267 | |
| 268 | MODULE_AUTHOR("Zwane Mwaikambo <zwane@commfireservices.com>"); |
| 269 | MODULE_DESCRIPTION("cpufreq driver for Pentium(TM) 4/Xeon(TM)"); |
| 270 | MODULE_LICENSE("GPL"); |
| 271 | |
| 272 | late_initcall(cpufreq_p4_init); |
| 273 | module_exit(cpufreq_p4_exit); |