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
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/cpumask.h>
29#include <linux/timex.h>
30
31#include <asm/processor.h>
32#include <asm/msr.h>
33#include <asm/timer.h>
34#include <asm/cpu_device_id.h>
35
36#include "speedstep-lib.h"
37
38#define PFX	"p4-clockmod: "
39
40/*
41 * Duty Cycle (3bits), note DC_DISABLE is not specified in
42 * intel docs i just use it to mean disable
43 */
44enum {
45	DC_RESV, DC_DFLT, DC_25PT, DC_38PT, DC_50PT,
46	DC_64PT, DC_75PT, DC_88PT, DC_DISABLE
47};
48
49#define DC_ENTRIES	8
50
51
52static int has_N44_O17_errata[NR_CPUS];
53static unsigned int stock_freq;
54static struct cpufreq_driver p4clockmod_driver;
55static unsigned int cpufreq_p4_get(unsigned int cpu);
56
57static int cpufreq_p4_setdc(unsigned int cpu, unsigned int newstate)
58{
59	u32 l, h;
60
61	if ((newstate > DC_DISABLE) || (newstate == DC_RESV))
62		return -EINVAL;
63
64	rdmsr_on_cpu(cpu, MSR_IA32_THERM_STATUS, &l, &h);
65
66	if (l & 0x01)
67		pr_debug("CPU#%d currently thermal throttled\n", cpu);
68
69	if (has_N44_O17_errata[cpu] &&
70	    (newstate == DC_25PT || newstate == DC_DFLT))
71		newstate = DC_38PT;
72
73	rdmsr_on_cpu(cpu, MSR_IA32_THERM_CONTROL, &l, &h);
74	if (newstate == DC_DISABLE) {
75		pr_debug("CPU#%d disabling modulation\n", cpu);
76		wrmsr_on_cpu(cpu, MSR_IA32_THERM_CONTROL, l & ~(1<<4), h);
77	} else {
78		pr_debug("CPU#%d setting duty cycle to %d%%\n",
79			cpu, ((125 * newstate) / 10));
80		/* bits 63 - 5	: reserved
81		 * bit  4	: enable/disable
82		 * bits 3-1	: duty cycle
83		 * bit  0	: reserved
84		 */
85		l = (l & ~14);
86		l = l | (1<<4) | ((newstate & 0x7)<<1);
87		wrmsr_on_cpu(cpu, MSR_IA32_THERM_CONTROL, l, h);
88	}
89
90	return 0;
91}
92
93
94static struct cpufreq_frequency_table p4clockmod_table[] = {
95	{0, DC_RESV, CPUFREQ_ENTRY_INVALID},
96	{0, DC_DFLT, 0},
97	{0, DC_25PT, 0},
98	{0, DC_38PT, 0},
99	{0, DC_50PT, 0},
100	{0, DC_64PT, 0},
101	{0, DC_75PT, 0},
102	{0, DC_88PT, 0},
103	{0, DC_DISABLE, 0},
104	{0, DC_RESV, CPUFREQ_TABLE_END},
105};
106
107
108static int cpufreq_p4_target(struct cpufreq_policy *policy, unsigned int index)
109{
110	int i;
111
112	/* run on each logical CPU,
113	 * see section 13.15.3 of IA32 Intel Architecture Software
114	 * Developer's Manual, Volume 3
115	 */
116	for_each_cpu(i, policy->cpus)
117		cpufreq_p4_setdc(i, p4clockmod_table[index].driver_data);
118
119	return 0;
120}
121
122
123static unsigned int cpufreq_p4_get_frequency(struct cpuinfo_x86 *c)
124{
125	if (c->x86 == 0x06) {
126		if (cpu_has(c, X86_FEATURE_EST))
127			printk_once(KERN_WARNING PFX "Warning: EST-capable "
128			       "CPU detected. The acpi-cpufreq module offers "
129			       "voltage scaling in addition to frequency "
130			       "scaling. You should use that instead of "
131			       "p4-clockmod, if possible.\n");
132		switch (c->x86_model) {
133		case 0x0E: /* Core */
134		case 0x0F: /* Core Duo */
135		case 0x16: /* Celeron Core */
136		case 0x1C: /* Atom */
137			p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS;
138			return speedstep_get_frequency(SPEEDSTEP_CPU_PCORE);
139		case 0x0D: /* Pentium M (Dothan) */
140			p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS;
141			/* fall through */
142		case 0x09: /* Pentium M (Banias) */
143			return speedstep_get_frequency(SPEEDSTEP_CPU_PM);
144		}
145	}
146
147	if (c->x86 != 0xF)
148		return 0;
149
150	/* on P-4s, the TSC runs with constant frequency independent whether
151	 * throttling is active or not. */
152	p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS;
153
154	if (speedstep_detect_processor() == SPEEDSTEP_CPU_P4M) {
155		printk(KERN_WARNING PFX "Warning: Pentium 4-M detected. "
156		       "The speedstep-ich or acpi cpufreq modules offer "
157		       "voltage scaling in addition of frequency scaling. "
158		       "You should use either one instead of p4-clockmod, "
159		       "if possible.\n");
160		return speedstep_get_frequency(SPEEDSTEP_CPU_P4M);
161	}
162
163	return speedstep_get_frequency(SPEEDSTEP_CPU_P4D);
164}
165
166
167
168static int cpufreq_p4_cpu_init(struct cpufreq_policy *policy)
169{
170	struct cpuinfo_x86 *c = &cpu_data(policy->cpu);
171	int cpuid = 0;
172	unsigned int i;
173
174#ifdef CONFIG_SMP
175	cpumask_copy(policy->cpus, topology_sibling_cpumask(policy->cpu));
176#endif
177
178	/* Errata workaround */
179	cpuid = (c->x86 << 8) | (c->x86_model << 4) | c->x86_mask;
180	switch (cpuid) {
181	case 0x0f07:
182	case 0x0f0a:
183	case 0x0f11:
184	case 0x0f12:
185		has_N44_O17_errata[policy->cpu] = 1;
186		pr_debug("has errata -- disabling low frequencies\n");
187	}
188
189	if (speedstep_detect_processor() == SPEEDSTEP_CPU_P4D &&
190	    c->x86_model < 2) {
191		/* switch to maximum frequency and measure result */
192		cpufreq_p4_setdc(policy->cpu, DC_DISABLE);
193		recalibrate_cpu_khz();
194	}
195	/* get max frequency */
196	stock_freq = cpufreq_p4_get_frequency(c);
197	if (!stock_freq)
198		return -EINVAL;
199
200	/* table init */
201	for (i = 1; (p4clockmod_table[i].frequency != CPUFREQ_TABLE_END); i++) {
202		if ((i < 2) && (has_N44_O17_errata[policy->cpu]))
203			p4clockmod_table[i].frequency = CPUFREQ_ENTRY_INVALID;
204		else
205			p4clockmod_table[i].frequency = (stock_freq * i)/8;
206	}
207
208	/* cpuinfo and default policy values */
209
210	/* the transition latency is set to be 1 higher than the maximum
211	 * transition latency of the ondemand governor */
212	policy->cpuinfo.transition_latency = 10000001;
213
214	return cpufreq_table_validate_and_show(policy, &p4clockmod_table[0]);
215}
216
217
218static unsigned int cpufreq_p4_get(unsigned int cpu)
219{
220	u32 l, h;
221
222	rdmsr_on_cpu(cpu, MSR_IA32_THERM_CONTROL, &l, &h);
223
224	if (l & 0x10) {
225		l = l >> 1;
226		l &= 0x7;
227	} else
228		l = DC_DISABLE;
229
230	if (l != DC_DISABLE)
231		return stock_freq * l / 8;
232
233	return stock_freq;
234}
235
236static struct cpufreq_driver p4clockmod_driver = {
237	.verify		= cpufreq_generic_frequency_table_verify,
238	.target_index	= cpufreq_p4_target,
239	.init		= cpufreq_p4_cpu_init,
240	.get		= cpufreq_p4_get,
241	.name		= "p4-clockmod",
242	.attr		= cpufreq_generic_attr,
243};
244
245static const struct x86_cpu_id cpufreq_p4_id[] = {
246	{ X86_VENDOR_INTEL, X86_FAMILY_ANY, X86_MODEL_ANY, X86_FEATURE_ACC },
247	{}
248};
249
250/*
251 * Intentionally no MODULE_DEVICE_TABLE here: this driver should not
252 * be auto loaded.  Please don't add one.
253 */
254
255static int __init cpufreq_p4_init(void)
256{
257	int ret;
258
259	/*
260	 * THERM_CONTROL is architectural for IA32 now, so
261	 * we can rely on the capability checks
262	 */
263	if (!x86_match_cpu(cpufreq_p4_id) || !boot_cpu_has(X86_FEATURE_ACPI))
264		return -ENODEV;
265
266	ret = cpufreq_register_driver(&p4clockmod_driver);
267	if (!ret)
268		printk(KERN_INFO PFX "P4/Xeon(TM) CPU On-Demand Clock "
269				"Modulation available\n");
270
271	return ret;
272}
273
274
275static void __exit cpufreq_p4_exit(void)
276{
277	cpufreq_unregister_driver(&p4clockmod_driver);
278}
279
280
281MODULE_AUTHOR("Zwane Mwaikambo <zwane@commfireservices.com>");
282MODULE_DESCRIPTION("cpufreq driver for Pentium(TM) 4/Xeon(TM)");
283MODULE_LICENSE("GPL");
284
285late_initcall(cpufreq_p4_init);
286module_exit(cpufreq_p4_exit);
287