p4tcc.c revision 142308
1/*-
2 * Copyright (c) 2005 Nate Lawson
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
15 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
16 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
17 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
18 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
19 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
20 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
21 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
22 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27/*
28 * Throttle clock frequency by using the thermal control circuit.  This
29 * operates independently of SpeedStep and ACPI throttling and is supported
30 * on Pentium 4 and later models (feature TM).
31 *
32 * Reference:  Intel Developer's manual v.3 #245472-012
33 *
34 * The original version of this driver was written by Ted Unangst for
35 * OpenBSD and imported by Maxim Sobolev.  It was rewritten by Nate Lawson
36 * for use with the cpufreq framework.
37 */
38
39#include <sys/cdefs.h>
40__FBSDID("$FreeBSD: head/sys/i386/cpufreq/p4tcc.c 142308 2005-02-23 16:42:56Z njl $");
41
42#include <sys/param.h>
43#include <sys/systm.h>
44#include <sys/bus.h>
45#include <sys/cpu.h>
46#include <sys/kernel.h>
47#include <sys/module.h>
48
49#include <machine/md_var.h>
50#include <machine/specialreg.h>
51
52#include "cpufreq_if.h"
53
54struct p4tcc_softc {
55	device_t	dev;
56	int		set_count;
57	int		lowest_val;
58	int		auto_mode;
59};
60
61#define TCC_NUM_SETTINGS	8
62
63#define TCC_ENABLE_ONDEMAND	(1<<4)
64#define TCC_REG_OFFSET		1
65#define TCC_SPEED_PERCENT(x)	((10000 * (x)) / TCC_NUM_SETTINGS)
66
67static void	p4tcc_identify(driver_t *driver, device_t parent);
68static int	p4tcc_probe(device_t dev);
69static int	p4tcc_attach(device_t dev);
70static int	p4tcc_settings(device_t dev, struct cf_setting *sets,
71		    int *count);
72static int	p4tcc_set(device_t dev, const struct cf_setting *set);
73static int	p4tcc_get(device_t dev, struct cf_setting *set);
74static int	p4tcc_type(device_t dev, int *type);
75
76static device_method_t p4tcc_methods[] = {
77	/* Device interface */
78	DEVMETHOD(device_identify,	p4tcc_identify),
79	DEVMETHOD(device_probe,		p4tcc_probe),
80	DEVMETHOD(device_attach,	p4tcc_attach),
81
82	/* cpufreq interface */
83	DEVMETHOD(cpufreq_drv_set,	p4tcc_set),
84	DEVMETHOD(cpufreq_drv_get,	p4tcc_get),
85	DEVMETHOD(cpufreq_drv_type,	p4tcc_type),
86	DEVMETHOD(cpufreq_drv_settings,	p4tcc_settings),
87	{0, 0}
88};
89
90static driver_t p4tcc_driver = {
91	"p4tcc",
92	p4tcc_methods,
93	sizeof(struct p4tcc_softc),
94};
95
96static devclass_t p4tcc_devclass;
97DRIVER_MODULE(p4tcc, cpu, p4tcc_driver, p4tcc_devclass, 0, 0);
98
99static void
100p4tcc_identify(driver_t *driver, device_t parent)
101{
102
103	if ((cpu_feature & (CPUID_ACPI | CPUID_TM)) != (CPUID_ACPI | CPUID_TM))
104		return;
105	if (BUS_ADD_CHILD(parent, 0, "p4tcc", -1) == NULL)
106		device_printf(parent, "add p4tcc child failed\n");
107}
108
109static int
110p4tcc_probe(device_t dev)
111{
112
113	if (resource_disabled("p4tcc", 0))
114		return (ENXIO);
115
116	device_set_desc(dev, "CPU Frequency Thermal Control");
117	return (0);
118}
119
120static int
121p4tcc_attach(device_t dev)
122{
123	struct p4tcc_softc *sc;
124
125	sc = device_get_softc(dev);
126	sc->dev = dev;
127	sc->set_count = TCC_NUM_SETTINGS;
128
129	/*
130	 * On boot, the TCC is usually in Automatic mode where reading the
131	 * current performance level is likely to produce bogus results.
132	 * We record that state here and don't trust the contents of the
133	 * status MSR until we've set it ourselves.
134	 */
135	sc->auto_mode = TRUE;
136
137	switch (cpu_id & 0xf) {
138	case 0x22:
139	case 0x24:
140	case 0x25:
141	case 0x27:
142	case 0x29:
143		/*
144		 * These CPU models hang when set to 12.5%.
145		 * See Errata O50, P44, and Z21.
146		 */
147		sc->set_count -= 1;
148		break;
149	case 0x07:	/* errata N44 and P18 */
150	case 0x0a:
151	case 0x12:
152	case 0x13:
153		/*
154		 * These CPU models hang when set to 12.5% or 25%.
155		 * See Errata N44 and P18l.
156		 */
157		sc->set_count -= 2;
158		break;
159	}
160	sc->lowest_val = TCC_NUM_SETTINGS - sc->set_count + 1;
161
162	cpufreq_register(dev);
163	return (0);
164}
165
166static int
167p4tcc_settings(device_t dev, struct cf_setting *sets, int *count)
168{
169	struct p4tcc_softc *sc;
170	int i, val;
171
172	sc = device_get_softc(dev);
173	if (sets == NULL || count == NULL)
174		return (EINVAL);
175	if (*count < sc->set_count)
176		return (E2BIG);
177
178	/* Return a list of valid settings for this driver. */
179	memset(sets, CPUFREQ_VAL_UNKNOWN, sizeof(*sets) * sc->set_count);
180	val = TCC_NUM_SETTINGS;
181	for (i = 0; i < sc->set_count; i++, val--) {
182		sets[i].freq = TCC_SPEED_PERCENT(val);
183		sets[i].dev = dev;
184	}
185	*count = sc->set_count;
186
187	return (0);
188}
189
190static int
191p4tcc_set(device_t dev, const struct cf_setting *set)
192{
193	struct p4tcc_softc *sc;
194	uint64_t mask, msr;
195	int val;
196
197	if (set == NULL)
198		return (EINVAL);
199	sc = device_get_softc(dev);
200
201	/*
202	 * Validate requested state converts to a setting that is an integer
203	 * from [sc->lowest_val .. TCC_NUM_SETTINGS].
204	 */
205	val = set->freq * TCC_NUM_SETTINGS / 10000;
206	if (val * 10000 != set->freq * TCC_NUM_SETTINGS ||
207	    val < sc->lowest_val || val > TCC_NUM_SETTINGS)
208		return (EINVAL);
209
210	/*
211	 * Read the current register and mask off the old setting and
212	 * On-Demand bit.  If the new val is < 100%, set it and the On-Demand
213	 * bit, otherwise just return to Automatic mode.
214	 */
215	msr = rdmsr(MSR_THERM_CONTROL);
216	mask = (TCC_NUM_SETTINGS - 1) << TCC_REG_OFFSET;
217	msr &= ~(mask | TCC_ENABLE_ONDEMAND);
218	if (val < TCC_NUM_SETTINGS)
219		msr |= (val << TCC_REG_OFFSET) | TCC_ENABLE_ONDEMAND;
220	wrmsr(MSR_THERM_CONTROL, msr);
221
222	/*
223	 * Record whether we're now in Automatic or On-Demand mode.  We have
224	 * to cache this since there is no reliable way to check if TCC is in
225	 * Automatic mode (i.e., at 100% or possibly 50%).  Reading bit 4 of
226	 * the ACPI Thermal Monitor Control Register produces 0 no matter
227	 * what the current mode.
228	 */
229	if (msr & TCC_ENABLE_ONDEMAND)
230		sc->auto_mode = TRUE;
231	else
232		sc->auto_mode = FALSE;
233
234	return (0);
235}
236
237static int
238p4tcc_get(device_t dev, struct cf_setting *set)
239{
240	struct p4tcc_softc *sc;
241	uint64_t msr;
242	int val;
243
244	if (set == NULL)
245		return (EINVAL);
246	sc = device_get_softc(dev);
247
248	/*
249	 * Read the current register and extract the current setting.  If
250	 * in automatic mode, assume we're at TCC_NUM_SETTINGS (100%).
251	 *
252	 * XXX This is not completely reliable since at high temperatures
253	 * the CPU may be automatically throttling to 50% but it's the best
254	 * we can do.
255	 */
256	if (!sc->auto_mode) {
257		msr = rdmsr(MSR_THERM_CONTROL);
258		val = (msr >> TCC_REG_OFFSET) & (TCC_NUM_SETTINGS - 1);
259	} else
260		val = TCC_NUM_SETTINGS;
261
262	memset(set, CPUFREQ_VAL_UNKNOWN, sizeof(*set));
263	set->freq = TCC_SPEED_PERCENT(val);
264	set->dev = dev;
265
266	return (0);
267}
268
269static int
270p4tcc_type(device_t dev, int *type)
271{
272
273	if (type == NULL)
274		return (EINVAL);
275
276	*type = CPUFREQ_TYPE_RELATIVE;
277	return (0);
278}
279