p4-clockmod.c 8.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338
  1. /*
  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
  7. * All Rights Reserved
  8. *
  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.
  13. *
  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.
  17. *
  18. * Date Errata Description
  19. * 20020525 N44, O17 12.5% or 25% DC causes lockup
  20. *
  21. */
  22. #include <linux/config.h>
  23. #include <linux/kernel.h>
  24. #include <linux/module.h>
  25. #include <linux/init.h>
  26. #include <linux/smp.h>
  27. #include <linux/cpufreq.h>
  28. #include <linux/slab.h>
  29. #include <linux/cpumask.h>
  30. #include <linux/sched.h> /* current / set_cpus_allowed() */
  31. #include <asm/processor.h>
  32. #include <asm/msr.h>
  33. #include <asm/timex.h>
  34. #include "speedstep-lib.h"
  35. #define PFX "p4-clockmod: "
  36. #define dprintk(msg...) cpufreq_debug_printk(CPUFREQ_DEBUG_DRIVER, "p4-clockmod", msg)
  37. /*
  38. * Duty Cycle (3bits), note DC_DISABLE is not specified in
  39. * intel docs i just use it to mean disable
  40. */
  41. enum {
  42. DC_RESV, DC_DFLT, DC_25PT, DC_38PT, DC_50PT,
  43. DC_64PT, DC_75PT, DC_88PT, DC_DISABLE
  44. };
  45. #define DC_ENTRIES 8
  46. static int has_N44_O17_errata[NR_CPUS];
  47. static unsigned int stock_freq;
  48. static struct cpufreq_driver p4clockmod_driver;
  49. static unsigned int cpufreq_p4_get(unsigned int cpu);
  50. static int cpufreq_p4_setdc(unsigned int cpu, unsigned int newstate)
  51. {
  52. u32 l, h;
  53. if (!cpu_online(cpu) || (newstate > DC_DISABLE) || (newstate == DC_RESV))
  54. return -EINVAL;
  55. rdmsr(MSR_IA32_THERM_STATUS, l, h);
  56. if (l & 0x01)
  57. dprintk("CPU#%d currently thermal throttled\n", cpu);
  58. if (has_N44_O17_errata[cpu] && (newstate == DC_25PT || newstate == DC_DFLT))
  59. newstate = DC_38PT;
  60. rdmsr(MSR_IA32_THERM_CONTROL, l, h);
  61. if (newstate == DC_DISABLE) {
  62. dprintk("CPU#%d disabling modulation\n", cpu);
  63. wrmsr(MSR_IA32_THERM_CONTROL, l & ~(1<<4), h);
  64. } else {
  65. dprintk("CPU#%d setting duty cycle to %d%%\n",
  66. cpu, ((125 * newstate) / 10));
  67. /* bits 63 - 5 : reserved
  68. * bit 4 : enable/disable
  69. * bits 3-1 : duty cycle
  70. * bit 0 : reserved
  71. */
  72. l = (l & ~14);
  73. l = l | (1<<4) | ((newstate & 0x7)<<1);
  74. wrmsr(MSR_IA32_THERM_CONTROL, l, h);
  75. }
  76. return 0;
  77. }
  78. static struct cpufreq_frequency_table p4clockmod_table[] = {
  79. {DC_RESV, CPUFREQ_ENTRY_INVALID},
  80. {DC_DFLT, 0},
  81. {DC_25PT, 0},
  82. {DC_38PT, 0},
  83. {DC_50PT, 0},
  84. {DC_64PT, 0},
  85. {DC_75PT, 0},
  86. {DC_88PT, 0},
  87. {DC_DISABLE, 0},
  88. {DC_RESV, CPUFREQ_TABLE_END},
  89. };
  90. static int cpufreq_p4_target(struct cpufreq_policy *policy,
  91. unsigned int target_freq,
  92. unsigned int relation)
  93. {
  94. unsigned int newstate = DC_RESV;
  95. struct cpufreq_freqs freqs;
  96. cpumask_t cpus_allowed;
  97. int i;
  98. if (cpufreq_frequency_table_target(policy, &p4clockmod_table[0], target_freq, relation, &newstate))
  99. return -EINVAL;
  100. freqs.old = cpufreq_p4_get(policy->cpu);
  101. freqs.new = stock_freq * p4clockmod_table[newstate].index / 8;
  102. if (freqs.new == freqs.old)
  103. return 0;
  104. /* notifiers */
  105. for_each_cpu_mask(i, policy->cpus) {
  106. freqs.cpu = i;
  107. cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
  108. }
  109. /* run on each logical CPU, see section 13.15.3 of IA32 Intel Architecture Software
  110. * Developer's Manual, Volume 3
  111. */
  112. cpus_allowed = current->cpus_allowed;
  113. for_each_cpu_mask(i, policy->cpus) {
  114. cpumask_t this_cpu = cpumask_of_cpu(i);
  115. set_cpus_allowed(current, this_cpu);
  116. BUG_ON(smp_processor_id() != i);
  117. cpufreq_p4_setdc(i, p4clockmod_table[newstate].index);
  118. }
  119. set_cpus_allowed(current, cpus_allowed);
  120. /* notifiers */
  121. for_each_cpu_mask(i, policy->cpus) {
  122. freqs.cpu = i;
  123. cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
  124. }
  125. return 0;
  126. }
  127. static int cpufreq_p4_verify(struct cpufreq_policy *policy)
  128. {
  129. return cpufreq_frequency_table_verify(policy, &p4clockmod_table[0]);
  130. }
  131. static unsigned int cpufreq_p4_get_frequency(struct cpuinfo_x86 *c)
  132. {
  133. if ((c->x86 == 0x06) && (c->x86_model == 0x09)) {
  134. /* Pentium M (Banias) */
  135. printk(KERN_WARNING PFX "Warning: Pentium M detected. "
  136. "The speedstep_centrino module offers voltage scaling"
  137. " in addition of frequency scaling. You should use "
  138. "that instead of p4-clockmod, if possible.\n");
  139. return speedstep_get_processor_frequency(SPEEDSTEP_PROCESSOR_PM);
  140. }
  141. if ((c->x86 == 0x06) && (c->x86_model == 0x0D)) {
  142. /* Pentium M (Dothan) */
  143. printk(KERN_WARNING PFX "Warning: Pentium M detected. "
  144. "The speedstep_centrino module offers voltage scaling"
  145. " in addition of frequency scaling. You should use "
  146. "that instead of p4-clockmod, if possible.\n");
  147. /* on P-4s, the TSC runs with constant frequency independent whether
  148. * throttling is active or not. */
  149. p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS;
  150. return speedstep_get_processor_frequency(SPEEDSTEP_PROCESSOR_PM);
  151. }
  152. if (c->x86 != 0xF) {
  153. printk(KERN_WARNING PFX "Unknown p4-clockmod-capable CPU. Please send an e-mail to <linux@brodo.de>\n");
  154. return 0;
  155. }
  156. /* on P-4s, the TSC runs with constant frequency independent whether
  157. * throttling is active or not. */
  158. p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS;
  159. if (speedstep_detect_processor() == SPEEDSTEP_PROCESSOR_P4M) {
  160. printk(KERN_WARNING PFX "Warning: Pentium 4-M detected. "
  161. "The speedstep-ich or acpi cpufreq modules offer "
  162. "voltage scaling in addition of frequency scaling. "
  163. "You should use either one instead of p4-clockmod, "
  164. "if possible.\n");
  165. return speedstep_get_processor_frequency(SPEEDSTEP_PROCESSOR_P4M);
  166. }
  167. return speedstep_get_processor_frequency(SPEEDSTEP_PROCESSOR_P4D);
  168. }
  169. static int cpufreq_p4_cpu_init(struct cpufreq_policy *policy)
  170. {
  171. struct cpuinfo_x86 *c = &cpu_data[policy->cpu];
  172. int cpuid = 0;
  173. unsigned int i;
  174. #ifdef CONFIG_SMP
  175. policy->cpus = cpu_sibling_map[policy->cpu];
  176. #endif
  177. /* Errata workaround */
  178. cpuid = (c->x86 << 8) | (c->x86_model << 4) | c->x86_mask;
  179. switch (cpuid) {
  180. case 0x0f07:
  181. case 0x0f0a:
  182. case 0x0f11:
  183. case 0x0f12:
  184. has_N44_O17_errata[policy->cpu] = 1;
  185. dprintk("has errata -- disabling low frequencies\n");
  186. }
  187. /* get max frequency */
  188. stock_freq = cpufreq_p4_get_frequency(c);
  189. if (!stock_freq)
  190. return -EINVAL;
  191. /* table init */
  192. for (i=1; (p4clockmod_table[i].frequency != CPUFREQ_TABLE_END); i++) {
  193. if ((i<2) && (has_N44_O17_errata[policy->cpu]))
  194. p4clockmod_table[i].frequency = CPUFREQ_ENTRY_INVALID;
  195. else
  196. p4clockmod_table[i].frequency = (stock_freq * i)/8;
  197. }
  198. cpufreq_frequency_table_get_attr(p4clockmod_table, policy->cpu);
  199. /* cpuinfo and default policy values */
  200. policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
  201. policy->cpuinfo.transition_latency = 1000000; /* assumed */
  202. policy->cur = stock_freq;
  203. return cpufreq_frequency_table_cpuinfo(policy, &p4clockmod_table[0]);
  204. }
  205. static int cpufreq_p4_cpu_exit(struct cpufreq_policy *policy)
  206. {
  207. cpufreq_frequency_table_put_attr(policy->cpu);
  208. return 0;
  209. }
  210. static unsigned int cpufreq_p4_get(unsigned int cpu)
  211. {
  212. cpumask_t cpus_allowed;
  213. u32 l, h;
  214. cpus_allowed = current->cpus_allowed;
  215. set_cpus_allowed(current, cpumask_of_cpu(cpu));
  216. BUG_ON(smp_processor_id() != cpu);
  217. rdmsr(MSR_IA32_THERM_CONTROL, l, h);
  218. set_cpus_allowed(current, cpus_allowed);
  219. if (l & 0x10) {
  220. l = l >> 1;
  221. l &= 0x7;
  222. } else
  223. l = DC_DISABLE;
  224. if (l != DC_DISABLE)
  225. return (stock_freq * l / 8);
  226. return stock_freq;
  227. }
  228. static struct freq_attr* p4clockmod_attr[] = {
  229. &cpufreq_freq_attr_scaling_available_freqs,
  230. NULL,
  231. };
  232. static struct cpufreq_driver p4clockmod_driver = {
  233. .verify = cpufreq_p4_verify,
  234. .target = cpufreq_p4_target,
  235. .init = cpufreq_p4_cpu_init,
  236. .exit = cpufreq_p4_cpu_exit,
  237. .get = cpufreq_p4_get,
  238. .name = "p4-clockmod",
  239. .owner = THIS_MODULE,
  240. .attr = p4clockmod_attr,
  241. };
  242. static int __init cpufreq_p4_init(void)
  243. {
  244. struct cpuinfo_x86 *c = cpu_data;
  245. int ret;
  246. /*
  247. * THERM_CONTROL is architectural for IA32 now, so
  248. * we can rely on the capability checks
  249. */
  250. if (c->x86_vendor != X86_VENDOR_INTEL)
  251. return -ENODEV;
  252. if (!test_bit(X86_FEATURE_ACPI, c->x86_capability) ||
  253. !test_bit(X86_FEATURE_ACC, c->x86_capability))
  254. return -ENODEV;
  255. ret = cpufreq_register_driver(&p4clockmod_driver);
  256. if (!ret)
  257. printk(KERN_INFO PFX "P4/Xeon(TM) CPU On-Demand Clock Modulation available\n");
  258. return (ret);
  259. }
  260. static void __exit cpufreq_p4_exit(void)
  261. {
  262. cpufreq_unregister_driver(&p4clockmod_driver);
  263. }
  264. MODULE_AUTHOR ("Zwane Mwaikambo <zwane@commfireservices.com>");
  265. MODULE_DESCRIPTION ("cpufreq driver for Pentium(TM) 4/Xeon(TM)");
  266. MODULE_LICENSE ("GPL");
  267. late_initcall(cpufreq_p4_init);
  268. module_exit(cpufreq_p4_exit);