xref: /freebsd/sys/dev/coretemp/coretemp.c (revision a90b9d0159070121c221b966469c3e36d912bf82)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2007, 2008 Rui Paulo <rpaulo@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
18  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
19  * DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
20  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
21  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
22  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
24  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
25  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26  * POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 /*
30  * Device driver for Intel's On Die thermal sensor via MSR.
31  * First introduced in Intel's Core line of processors.
32  */
33 
34 #include <sys/param.h>
35 #include <sys/bus.h>
36 #include <sys/conf.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/module.h>
40 #include <sys/mutex.h>
41 #include <sys/proc.h>	/* for curthread */
42 #include <sys/smp.h>
43 #include <sys/sysctl.h>
44 #include <sys/systm.h>
45 
46 #include <machine/specialreg.h>
47 #include <machine/cpufunc.h>
48 #include <machine/cputypes.h>
49 #include <machine/md_var.h>
50 
51 #define	TZ_ZEROC			2731
52 
53 #define	THERM_CRITICAL_STATUS_LOG       0x20
54 #define	THERM_CRITICAL_STATUS           0x10
55 #define	THERM_STATUS_LOG		0x02
56 #define	THERM_STATUS			0x01
57 #define	THERM_STATUS_TEMP_SHIFT		16
58 #define	THERM_STATUS_TEMP_MASK		0x7f
59 #define	THERM_STATUS_RES_SHIFT		27
60 #define	THERM_STATUS_RES_MASK		0x0f
61 #define	THERM_STATUS_VALID_SHIFT	31
62 #define	THERM_STATUS_VALID_MASK		0x01
63 
64 struct coretemp_softc {
65 	device_t	sc_dev;
66 	int		sc_tjmax;
67 	unsigned int	sc_throttle_log;
68 };
69 
70 /*
71  * Device methods.
72  */
73 static void	coretemp_identify(driver_t *driver, device_t parent);
74 static int	coretemp_probe(device_t dev);
75 static int	coretemp_attach(device_t dev);
76 static int	coretemp_detach(device_t dev);
77 
78 static uint64_t	coretemp_get_thermal_msr(int cpu);
79 static void	coretemp_clear_thermal_msr(int cpu);
80 static int	coretemp_get_val_sysctl(SYSCTL_HANDLER_ARGS);
81 static int	coretemp_throttle_log_sysctl(SYSCTL_HANDLER_ARGS);
82 
83 static device_method_t coretemp_methods[] = {
84 	/* Device interface */
85 	DEVMETHOD(device_identify,	coretemp_identify),
86 	DEVMETHOD(device_probe,		coretemp_probe),
87 	DEVMETHOD(device_attach,	coretemp_attach),
88 	DEVMETHOD(device_detach,	coretemp_detach),
89 
90 	DEVMETHOD_END
91 };
92 
93 static driver_t coretemp_driver = {
94 	"coretemp",
95 	coretemp_methods,
96 	sizeof(struct coretemp_softc),
97 };
98 
99 enum therm_info {
100 	CORETEMP_TEMP,
101 	CORETEMP_DELTA,
102 	CORETEMP_RESOLUTION,
103 	CORETEMP_TJMAX,
104 };
105 
106 DRIVER_MODULE(coretemp, cpu, coretemp_driver, NULL, NULL);
107 
108 static void
109 coretemp_identify(driver_t *driver, device_t parent)
110 {
111 	device_t child;
112 	u_int regs[4];
113 
114 	/* Make sure we're not being doubly invoked. */
115 	if (device_find_child(parent, "coretemp", -1) != NULL)
116 		return;
117 
118 	/* Check that CPUID 0x06 is supported and the vendor is Intel.*/
119 	if (cpu_high < 6 || cpu_vendor_id != CPU_VENDOR_INTEL)
120 		return;
121 	/*
122 	 * CPUID 0x06 returns 1 if the processor has on-die thermal
123 	 * sensors. EBX[0:3] contains the number of sensors.
124 	 */
125 	do_cpuid(0x06, regs);
126 	if ((regs[0] & 0x1) != 1)
127 		return;
128 
129 	/*
130 	 * We add a child for each CPU since settings must be performed
131 	 * on each CPU in the SMP case.
132 	 */
133 	child = device_add_child(parent, "coretemp", device_get_unit(parent));
134 	if (child == NULL)
135 		device_printf(parent, "add coretemp child failed\n");
136 }
137 
138 static int
139 coretemp_probe(device_t dev)
140 {
141 	if (resource_disabled("coretemp", 0))
142 		return (ENXIO);
143 
144 	device_set_desc(dev, "CPU On-Die Thermal Sensors");
145 
146 	if (!bootverbose && device_get_unit(dev) != 0)
147 		device_quiet(dev);
148 
149 	return (BUS_PROBE_GENERIC);
150 }
151 
152 static int
153 coretemp_attach(device_t dev)
154 {
155 	struct coretemp_softc *sc = device_get_softc(dev);
156 	device_t pdev;
157 	uint64_t msr;
158 	int cpu_model, cpu_stepping;
159 	int ret, tjtarget;
160 	struct sysctl_oid *oid;
161 	struct sysctl_ctx_list *ctx;
162 
163 	sc->sc_dev = dev;
164 	pdev = device_get_parent(dev);
165 	cpu_model = CPUID_TO_MODEL(cpu_id);
166 	cpu_stepping = CPUID_TO_STEPPING(cpu_id);
167 
168 	/*
169 	 * Some CPUs, namely the PIII, don't have thermal sensors, but
170 	 * report them when the CPUID check is performed in
171 	 * coretemp_identify(). This leads to a later GPF when the sensor
172 	 * is queried via a MSR, so we stop here.
173 	 */
174 	if (cpu_model < 0xe)
175 		return (ENXIO);
176 
177 #if 0 /*
178        * XXXrpaulo: I have this CPU model and when it returns from C3
179        * coretemp continues to function properly.
180        */
181 
182 	/*
183 	 * Check for errata AE18.
184 	 * "Processor Digital Thermal Sensor (DTS) Readout stops
185 	 *  updating upon returning from C3/C4 state."
186 	 *
187 	 * Adapted from the Linux coretemp driver.
188 	 */
189 	if (cpu_model == 0xe && cpu_stepping < 0xc) {
190 		msr = rdmsr(MSR_BIOS_SIGN);
191 		msr = msr >> 32;
192 		if (msr < 0x39) {
193 			device_printf(dev, "not supported (Intel errata "
194 			    "AE18), try updating your BIOS\n");
195 			return (ENXIO);
196 		}
197 	}
198 #endif
199 
200 	/*
201 	 * Use 100C as the initial value.
202 	 */
203 	sc->sc_tjmax = 100;
204 
205 	if ((cpu_model == 0xf && cpu_stepping >= 2) || cpu_model == 0xe) {
206 		/*
207 		 * On some Core 2 CPUs, there's an undocumented MSR that
208 		 * can tell us if Tj(max) is 100 or 85.
209 		 *
210 		 * The if-clause for CPUs having the MSR_IA32_EXT_CONFIG was adapted
211 		 * from the Linux coretemp driver.
212 		 */
213 		msr = rdmsr(MSR_IA32_EXT_CONFIG);
214 		if (msr & (1 << 30))
215 			sc->sc_tjmax = 85;
216 	} else if (cpu_model == 0x17) {
217 		switch (cpu_stepping) {
218 		case 0x6:	/* Mobile Core 2 Duo */
219 			sc->sc_tjmax = 105;
220 			break;
221 		default:	/* Unknown stepping */
222 			break;
223 		}
224 	} else if (cpu_model == 0x1c) {
225 		switch (cpu_stepping) {
226 		case 0xa:	/* 45nm Atom D400, N400 and D500 series */
227 			sc->sc_tjmax = 100;
228 			break;
229 		default:
230 			sc->sc_tjmax = 90;
231 			break;
232 		}
233 	} else {
234 		/*
235 		 * Attempt to get Tj(max) from MSR IA32_TEMPERATURE_TARGET.
236 		 *
237 		 * This method is described in Intel white paper "CPU
238 		 * Monitoring With DTS/PECI". (#322683)
239 		 */
240 		ret = rdmsr_safe(MSR_IA32_TEMPERATURE_TARGET, &msr);
241 		if (ret == 0) {
242 			tjtarget = (msr >> 16) & 0xff;
243 
244 			/*
245 			 * On earlier generation of processors, the value
246 			 * obtained from IA32_TEMPERATURE_TARGET register is
247 			 * an offset that needs to be summed with a model
248 			 * specific base.  It is however not clear what
249 			 * these numbers are, with the publicly available
250 			 * documents from Intel.
251 			 *
252 			 * For now, we consider [70, 110]C range, as
253 			 * described in #322683, as "reasonable" and accept
254 			 * these values whenever the MSR is available for
255 			 * read, regardless the CPU model.
256 			 */
257 			if (tjtarget >= 70 && tjtarget <= 110)
258 				sc->sc_tjmax = tjtarget;
259 			else
260 				device_printf(dev, "Tj(target) value %d "
261 				    "does not seem right.\n", tjtarget);
262 		} else
263 			device_printf(dev, "Can not get Tj(target) "
264 			    "from your CPU, using 100C.\n");
265 	}
266 
267 	if (bootverbose)
268 		device_printf(dev, "Setting TjMax=%d\n", sc->sc_tjmax);
269 
270 	ctx = device_get_sysctl_ctx(dev);
271 
272 	oid = SYSCTL_ADD_NODE(ctx,
273 	    SYSCTL_CHILDREN(device_get_sysctl_tree(pdev)), OID_AUTO,
274 	    "coretemp", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
275 	    "Per-CPU thermal information");
276 
277 	/*
278 	 * Add the MIBs to dev.cpu.N and dev.cpu.N.coretemp.
279 	 */
280 	SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(pdev)),
281 	    OID_AUTO, "temperature", CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
282 	    dev, CORETEMP_TEMP, coretemp_get_val_sysctl, "IK",
283 	    "Current temperature");
284 	SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO, "delta",
285 	    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, dev, CORETEMP_DELTA,
286 	    coretemp_get_val_sysctl, "I",
287 	    "Delta between TCC activation and current temperature");
288 	SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO, "resolution",
289 	    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, dev, CORETEMP_RESOLUTION,
290 	    coretemp_get_val_sysctl, "I",
291 	    "Resolution of CPU thermal sensor");
292 	SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO, "tjmax",
293 	    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, dev, CORETEMP_TJMAX,
294 	    coretemp_get_val_sysctl, "IK",
295 	    "TCC activation temperature");
296 	SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO,
297 	    "throttle_log", CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, dev, 0,
298 	    coretemp_throttle_log_sysctl, "I",
299 	    "Set to 1 if the thermal sensor has tripped");
300 
301 	return (0);
302 }
303 
304 static int
305 coretemp_detach(device_t dev)
306 {
307 	return (0);
308 }
309 
310 struct coretemp_args {
311 	u_int		msr;
312 	uint64_t	val;
313 };
314 
315 /*
316  * The digital temperature reading is located at bit 16
317  * of MSR_THERM_STATUS.
318  *
319  * There is a bit on that MSR that indicates whether the
320  * temperature is valid or not.
321  *
322  * The temperature is computed by subtracting the temperature
323  * reading by Tj(max).
324  */
325 static uint64_t
326 coretemp_get_thermal_msr(int cpu)
327 {
328 	uint64_t res;
329 
330 	x86_msr_op(MSR_THERM_STATUS, MSR_OP_RENDEZVOUS_ONE | MSR_OP_READ |
331 	    MSR_OP_CPUID(cpu), 0, &res);
332 	return (res);
333 }
334 
335 static void
336 coretemp_clear_thermal_msr(int cpu)
337 {
338 	x86_msr_op(MSR_THERM_STATUS, MSR_OP_RENDEZVOUS_ONE | MSR_OP_WRITE |
339 	    MSR_OP_CPUID(cpu), 0, NULL);
340 }
341 
342 static int
343 coretemp_get_val_sysctl(SYSCTL_HANDLER_ARGS)
344 {
345 	device_t dev;
346 	uint64_t msr;
347 	int val, tmp;
348 	struct coretemp_softc *sc;
349 	enum therm_info type;
350 	char stemp[16];
351 
352 	dev = (device_t) arg1;
353 	msr = coretemp_get_thermal_msr(device_get_unit(dev));
354 	sc = device_get_softc(dev);
355 	type = arg2;
356 
357 	if (((msr >> THERM_STATUS_VALID_SHIFT) & THERM_STATUS_VALID_MASK) != 1) {
358 		val = -1;
359 	} else {
360 		switch (type) {
361 		case CORETEMP_TEMP:
362 			tmp = (msr >> THERM_STATUS_TEMP_SHIFT) &
363 			    THERM_STATUS_TEMP_MASK;
364 			val = (sc->sc_tjmax - tmp) * 10 + TZ_ZEROC;
365 			break;
366 		case CORETEMP_DELTA:
367 			val = (msr >> THERM_STATUS_TEMP_SHIFT) &
368 			    THERM_STATUS_TEMP_MASK;
369 			break;
370 		case CORETEMP_RESOLUTION:
371 			val = (msr >> THERM_STATUS_RES_SHIFT) &
372 			    THERM_STATUS_RES_MASK;
373 			break;
374 		case CORETEMP_TJMAX:
375 			val = sc->sc_tjmax * 10 + TZ_ZEROC;
376 			break;
377 		}
378 	}
379 
380 	if (msr & THERM_STATUS_LOG) {
381 		coretemp_clear_thermal_msr(device_get_unit(dev));
382 		sc->sc_throttle_log = 1;
383 
384 		/*
385 		 * Check for Critical Temperature Status and Critical
386 		 * Temperature Log.  It doesn't really matter if the
387 		 * current temperature is invalid because the "Critical
388 		 * Temperature Log" bit will tell us if the Critical
389 		 * Temperature has * been reached in past. It's not
390 		 * directly related to the current temperature.
391 		 *
392 		 * If we reach a critical level, allow devctl(4)
393 		 * to catch this and shutdown the system.
394 		 */
395 		if (msr & THERM_CRITICAL_STATUS) {
396 			tmp = (msr >> THERM_STATUS_TEMP_SHIFT) &
397 			    THERM_STATUS_TEMP_MASK;
398 			tmp = (sc->sc_tjmax - tmp) * 10 + TZ_ZEROC;
399 			device_printf(dev, "critical temperature detected, "
400 			    "suggest system shutdown\n");
401 			snprintf(stemp, sizeof(stemp), "%d", tmp);
402 			devctl_notify("coretemp", "Thermal", stemp,
403 			    "notify=0xcc");
404 		}
405 	}
406 
407 	return (sysctl_handle_int(oidp, &val, 0, req));
408 }
409 
410 static int
411 coretemp_throttle_log_sysctl(SYSCTL_HANDLER_ARGS)
412 {
413 	device_t dev;
414 	uint64_t msr;
415 	int error, val;
416 	struct coretemp_softc *sc;
417 
418 	dev = (device_t) arg1;
419 	msr = coretemp_get_thermal_msr(device_get_unit(dev));
420 	sc = device_get_softc(dev);
421 
422 	if (msr & THERM_STATUS_LOG) {
423 		coretemp_clear_thermal_msr(device_get_unit(dev));
424 		sc->sc_throttle_log = 1;
425 	}
426 
427 	val = sc->sc_throttle_log;
428 
429 	error = sysctl_handle_int(oidp, &val, 0, req);
430 
431 	if (error || !req->newptr)
432 		return (error);
433 	else if (val != 0)
434 		return (EINVAL);
435 
436 	coretemp_clear_thermal_msr(device_get_unit(dev));
437 	sc->sc_throttle_log = 0;
438 
439 	return (0);
440 }
441