2 * Pentium 4/Xeon CPU on demand clock modulation/speed scaling
3 * (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
4 * (C) 2002 Zwane Mwaikambo <zwane@commfireservices.com>
5 * (C) 2002 Arjan van de Ven <arjanv@redhat.com>
6 * (C) 2002 Tora T. Engstad
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License
11 * as published by the Free Software Foundation; either version
12 * 2 of the License, or (at your option) any later version.
14 * The author(s) of this software shall not be held liable for damages
15 * of any nature resulting due to the use of this software. This
16 * software is provided AS-IS with no warranties.
18 * Date Errata Description
19 * 20020525 N44, O17 12.5% or 25% DC causes lockup
23 #include <linux/config.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/init.h>
27 #include <linux/smp.h>
28 #include <linux/cpufreq.h>
29 #include <linux/slab.h>
30 #include <linux/sched.h>
32 #include <asm/processor.h>
34 #include <asm/timex.h>
36 #include "speedstep-lib.h"
38 #define PFX "cpufreq: "
41 * Duty Cycle (3bits), note DC_DISABLE is not specified in
42 * intel docs i just use it to mean disable
45 DC_RESV, DC_DFLT, DC_25PT, DC_38PT, DC_50PT,
46 DC_64PT, DC_75PT, DC_88PT, DC_DISABLE
52 static int has_N44_O17_errata[NR_CPUS];
53 static unsigned int stock_freq;
54 static struct cpufreq_driver p4clockmod_driver;
55 static unsigned int cpufreq_p4_get(unsigned int cpu);
57 static int cpufreq_p4_setdc(unsigned int cpu, unsigned int newstate)
61 if (!cpu_online(cpu) || (newstate > DC_DISABLE) || (newstate == DC_RESV))
64 rdmsr(MSR_IA32_THERM_STATUS, l, h);
67 printk(KERN_DEBUG PFX "CPU#%d currently thermal throttled\n", cpu);
69 if (has_N44_O17_errata[cpu] && (newstate == DC_25PT || newstate == DC_DFLT))
72 rdmsr(MSR_IA32_THERM_CONTROL, l, h);
73 if (newstate == DC_DISABLE) {
74 /* printk(KERN_INFO PFX "CPU#%d disabling modulation\n", cpu); */
75 wrmsr(MSR_IA32_THERM_CONTROL, l & ~(1<<4), h);
77 /* printk(KERN_INFO PFX "CPU#%d setting duty cycle to %d%%\n",
78 cpu, ((125 * newstate) / 10)); */
79 /* bits 63 - 5 : reserved
80 * bit 4 : enable/disable
81 * bits 3-1 : duty cycle
85 l = l | (1<<4) | ((newstate & 0x7)<<1);
86 wrmsr(MSR_IA32_THERM_CONTROL, l, h);
93 static struct cpufreq_frequency_table p4clockmod_table[] = {
94 {DC_RESV, CPUFREQ_ENTRY_INVALID},
103 {DC_RESV, CPUFREQ_TABLE_END},
107 static int cpufreq_p4_target(struct cpufreq_policy *policy,
108 unsigned int target_freq,
109 unsigned int relation)
111 unsigned int newstate = DC_RESV;
112 struct cpufreq_freqs freqs;
113 cpumask_t cpus_allowed, affected_cpu_map;
116 if (cpufreq_frequency_table_target(policy, &p4clockmod_table[0], target_freq, relation, &newstate))
119 freqs.old = cpufreq_p4_get(policy->cpu);
120 freqs.new = stock_freq * p4clockmod_table[newstate].index / 8;
122 if (freqs.new == freqs.old)
125 /* switch to physical CPU where state is to be changed*/
126 cpus_allowed = current->cpus_allowed;
128 /* only run on CPU to be set, or on its sibling */
130 affected_cpu_map = cpu_sibling_map[policy->cpu];
132 affected_cpu_map = cpumask_of_cpu(policy->cpu);
137 if (cpu_isset(i, affected_cpu_map)) {
139 cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
143 /* run on each logical CPU, see section 13.15.3 of IA32 Intel Architecture Software
144 * Developer's Manual, Volume 3
147 if (cpu_isset(i, affected_cpu_map)) {
148 cpumask_t this_cpu = cpumask_of_cpu(i);
150 set_cpus_allowed(current, this_cpu);
151 BUG_ON(smp_processor_id() != i);
153 cpufreq_p4_setdc(i, p4clockmod_table[newstate].index);
156 set_cpus_allowed(current, cpus_allowed);
160 if (cpu_isset(i, affected_cpu_map)) {
162 cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
170 static int cpufreq_p4_verify(struct cpufreq_policy *policy)
172 return cpufreq_frequency_table_verify(policy, &p4clockmod_table[0]);
176 static unsigned int cpufreq_p4_get_frequency(struct cpuinfo_x86 *c)
178 if ((c->x86 == 0x06) && (c->x86_model == 0x09)) {
179 /* Pentium M (Banias) */
180 printk(KERN_WARNING PFX "Warning: Pentium M detected. "
181 "The speedstep_centrino module offers voltage scaling"
182 " in addition of frequency scaling. You should use "
183 "that instead of p4-clockmod, if possible.\n");
184 return speedstep_get_processor_frequency(SPEEDSTEP_PROCESSOR_PM);
187 if ((c->x86 == 0x06) && (c->x86_model == 0x13)) {
188 /* Pentium M (Dothan) */
189 printk(KERN_WARNING PFX "Warning: Pentium M detected. "
190 "The speedstep_centrino module offers voltage scaling"
191 " in addition of frequency scaling. You should use "
192 "that instead of p4-clockmod, if possible.\n");
193 /* on P-4s, the TSC runs with constant frequency independent wether
194 * throttling is active or not. */
195 p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS;
196 return speedstep_get_processor_frequency(SPEEDSTEP_PROCESSOR_PM);
200 printk(KERN_WARNING PFX "Unknown p4-clockmod-capable CPU. Please send an e-mail to <linux@brodo.de>\n");
204 /* on P-4s, the TSC runs with constant frequency independent wether
205 * throttling is active or not. */
206 p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS;
208 if (speedstep_detect_processor() == SPEEDSTEP_PROCESSOR_P4M) {
209 printk(KERN_WARNING PFX "Warning: Pentium 4-M detected. "
210 "The speedstep-ich or acpi cpufreq modules offer "
211 "voltage scaling in addition of frequency scaling. "
212 "You should use either one instead of p4-clockmod, "
214 return speedstep_get_processor_frequency(SPEEDSTEP_PROCESSOR_P4M);
217 return speedstep_get_processor_frequency(SPEEDSTEP_PROCESSOR_P4D);
222 static int cpufreq_p4_cpu_init(struct cpufreq_policy *policy)
224 struct cpuinfo_x86 *c = &cpu_data[policy->cpu];
228 /* Errata workaround */
229 cpuid = (c->x86 << 8) | (c->x86_model << 4) | c->x86_mask;
235 has_N44_O17_errata[policy->cpu] = 1;
238 /* get max frequency */
239 stock_freq = cpufreq_p4_get_frequency(c);
244 for (i=1; (p4clockmod_table[i].frequency != CPUFREQ_TABLE_END); i++) {
245 if ((i<2) && (has_N44_O17_errata[policy->cpu]))
246 p4clockmod_table[i].frequency = CPUFREQ_ENTRY_INVALID;
248 p4clockmod_table[i].frequency = (stock_freq * i)/8;
250 cpufreq_frequency_table_get_attr(p4clockmod_table, policy->cpu);
252 /* cpuinfo and default policy values */
253 policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
254 policy->cpuinfo.transition_latency = 1000000; /* assumed */
255 policy->cur = stock_freq;
257 return cpufreq_frequency_table_cpuinfo(policy, &p4clockmod_table[0]);
261 static int cpufreq_p4_cpu_exit(struct cpufreq_policy *policy)
263 cpufreq_frequency_table_put_attr(policy->cpu);
267 static unsigned int cpufreq_p4_get(unsigned int cpu)
269 cpumask_t cpus_allowed, affected_cpu_map;
272 cpus_allowed = current->cpus_allowed;
273 affected_cpu_map = cpumask_of_cpu(cpu);
275 set_cpus_allowed(current, affected_cpu_map);
276 BUG_ON(!cpu_isset(smp_processor_id(), affected_cpu_map));
278 rdmsr(MSR_IA32_THERM_CONTROL, l, h);
280 set_cpus_allowed(current, cpus_allowed);
289 return (stock_freq * l / 8);
294 static struct freq_attr* p4clockmod_attr[] = {
295 &cpufreq_freq_attr_scaling_available_freqs,
299 static struct cpufreq_driver p4clockmod_driver = {
300 .verify = cpufreq_p4_verify,
301 .target = cpufreq_p4_target,
302 .init = cpufreq_p4_cpu_init,
303 .exit = cpufreq_p4_cpu_exit,
304 .get = cpufreq_p4_get,
305 .name = "p4-clockmod",
306 .owner = THIS_MODULE,
307 .attr = p4clockmod_attr,
311 static int __init cpufreq_p4_init(void)
313 struct cpuinfo_x86 *c = cpu_data;
316 * THERM_CONTROL is architectural for IA32 now, so
317 * we can rely on the capability checks
319 if (c->x86_vendor != X86_VENDOR_INTEL)
322 if (!test_bit(X86_FEATURE_ACPI, c->x86_capability) ||
323 !test_bit(X86_FEATURE_ACC, c->x86_capability))
326 printk(KERN_INFO PFX "P4/Xeon(TM) CPU On-Demand Clock Modulation available\n");
328 return cpufreq_register_driver(&p4clockmod_driver);
332 static void __exit cpufreq_p4_exit(void)
334 cpufreq_unregister_driver(&p4clockmod_driver);
338 MODULE_AUTHOR ("Zwane Mwaikambo <zwane@commfireservices.com>");
339 MODULE_DESCRIPTION ("cpufreq driver for Pentium(TM) 4/Xeon(TM)");
340 MODULE_LICENSE ("GPL");
342 late_initcall(cpufreq_p4_init);
343 module_exit(cpufreq_p4_exit);