xref: /freebsd/sys/dev/amdtemp/amdtemp.c (revision 641a6cfb86023499caafe26a4d821a0b885cf00b)
1 /*-
2  * Copyright (c) 2008, 2009 Rui Paulo <rpaulo@FreeBSD.org>
3  * Copyright (c) 2009 Norikatsu Shigemura <nork@FreeBSD.org>
4  * Copyright (c) 2009-2012 Jung-uk Kim <jkim@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  * Driver for the AMD CPU on-die thermal sensors.
31  * Initially based on the k8temp Linux driver.
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/param.h>
38 #include <sys/bus.h>
39 #include <sys/conf.h>
40 #include <sys/kernel.h>
41 #include <sys/module.h>
42 #include <sys/sysctl.h>
43 #include <sys/systm.h>
44 
45 #include <machine/cpufunc.h>
46 #include <machine/md_var.h>
47 #include <machine/specialreg.h>
48 
49 #include <dev/pci/pcivar.h>
50 #include <x86/pci_cfgreg.h>
51 
52 typedef enum {
53 	CORE0_SENSOR0,
54 	CORE0_SENSOR1,
55 	CORE1_SENSOR0,
56 	CORE1_SENSOR1,
57 	CORE0,
58 	CORE1
59 } amdsensor_t;
60 
61 struct amdtemp_softc {
62 	device_t	sc_dev;
63 	int		sc_ncores;
64 	int		sc_ntemps;
65 	int		sc_flags;
66 #define	AMDTEMP_FLAG_CS_SWAP	0x01	/* ThermSenseCoreSel is inverted. */
67 #define	AMDTEMP_FLAG_CT_10BIT	0x02	/* CurTmp is 10-bit wide. */
68 #define	AMDTEMP_FLAG_ALT_OFFSET	0x04	/* CurTmp starts at -28C. */
69 	int32_t		sc_offset;
70 	int32_t		(*sc_gettemp)(device_t, amdsensor_t);
71 	struct sysctl_oid *sc_sysctl_cpu[MAXCPU];
72 	struct intr_config_hook sc_ich;
73 };
74 
75 #define	VENDORID_AMD		0x1022
76 #define	DEVICEID_AMD_MISC0F	0x1103
77 #define	DEVICEID_AMD_MISC10	0x1203
78 #define	DEVICEID_AMD_MISC11	0x1303
79 #define	DEVICEID_AMD_MISC14	0x1703
80 #define	DEVICEID_AMD_MISC15	0x1603
81 
82 static struct amdtemp_product {
83 	uint16_t	amdtemp_vendorid;
84 	uint16_t	amdtemp_deviceid;
85 } amdtemp_products[] = {
86 	{ VENDORID_AMD,	DEVICEID_AMD_MISC0F },
87 	{ VENDORID_AMD,	DEVICEID_AMD_MISC10 },
88 	{ VENDORID_AMD,	DEVICEID_AMD_MISC11 },
89 	{ VENDORID_AMD,	DEVICEID_AMD_MISC14 },
90 	{ VENDORID_AMD,	DEVICEID_AMD_MISC15 },
91 	{ 0, 0 }
92 };
93 
94 /*
95  * Reported Temperature Control Register
96  */
97 #define	AMDTEMP_REPTMP_CTRL	0xa4
98 
99 /*
100  * Thermaltrip Status Register (Family 0Fh only)
101  */
102 #define	AMDTEMP_THERMTP_STAT	0xe4
103 #define	AMDTEMP_TTSR_SELCORE	0x04
104 #define	AMDTEMP_TTSR_SELSENSOR	0x40
105 
106 /*
107  * DRAM Configuration High Register
108  */
109 #define	AMDTEMP_DRAM_CONF_HIGH	0x94	/* Function 2 */
110 #define	AMDTEMP_DRAM_MODE_DDR3	0x0100
111 
112 /*
113  * CPU Family/Model Register
114  */
115 #define	AMDTEMP_CPUID		0xfc
116 
117 /*
118  * Device methods.
119  */
120 static void 	amdtemp_identify(driver_t *driver, device_t parent);
121 static int	amdtemp_probe(device_t dev);
122 static int	amdtemp_attach(device_t dev);
123 static void	amdtemp_intrhook(void *arg);
124 static int	amdtemp_detach(device_t dev);
125 static int 	amdtemp_match(device_t dev);
126 static int32_t	amdtemp_gettemp0f(device_t dev, amdsensor_t sensor);
127 static int32_t	amdtemp_gettemp(device_t dev, amdsensor_t sensor);
128 static int	amdtemp_sysctl(SYSCTL_HANDLER_ARGS);
129 
130 static device_method_t amdtemp_methods[] = {
131 	/* Device interface */
132 	DEVMETHOD(device_identify,	amdtemp_identify),
133 	DEVMETHOD(device_probe,		amdtemp_probe),
134 	DEVMETHOD(device_attach,	amdtemp_attach),
135 	DEVMETHOD(device_detach,	amdtemp_detach),
136 
137 	{0, 0}
138 };
139 
140 static driver_t amdtemp_driver = {
141 	"amdtemp",
142 	amdtemp_methods,
143 	sizeof(struct amdtemp_softc),
144 };
145 
146 static devclass_t amdtemp_devclass;
147 DRIVER_MODULE(amdtemp, hostb, amdtemp_driver, amdtemp_devclass, NULL, NULL);
148 
149 static int
150 amdtemp_match(device_t dev)
151 {
152 	int i;
153 	uint16_t vendor, devid;
154 
155 	vendor = pci_get_vendor(dev);
156 	devid = pci_get_device(dev);
157 
158 	for (i = 0; amdtemp_products[i].amdtemp_vendorid != 0; i++) {
159 		if (vendor == amdtemp_products[i].amdtemp_vendorid &&
160 		    devid == amdtemp_products[i].amdtemp_deviceid)
161 			return (1);
162 	}
163 
164 	return (0);
165 }
166 
167 static void
168 amdtemp_identify(driver_t *driver, device_t parent)
169 {
170 	device_t child;
171 
172 	/* Make sure we're not being doubly invoked. */
173 	if (device_find_child(parent, "amdtemp", -1) != NULL)
174 		return;
175 
176 	if (amdtemp_match(parent)) {
177 		child = device_add_child(parent, "amdtemp", -1);
178 		if (child == NULL)
179 			device_printf(parent, "add amdtemp child failed\n");
180 	}
181 }
182 
183 static int
184 amdtemp_probe(device_t dev)
185 {
186 	uint32_t family, model;
187 
188 	if (resource_disabled("amdtemp", 0))
189 		return (ENXIO);
190 
191 	family = CPUID_TO_FAMILY(cpu_id);
192 	model = CPUID_TO_MODEL(cpu_id);
193 
194 	switch (family) {
195 	case 0x0f:
196 		if ((model == 0x04 && (cpu_id & CPUID_STEPPING) == 0) ||
197 		    (model == 0x05 && (cpu_id & CPUID_STEPPING) <= 1))
198 			return (ENXIO);
199 		break;
200 	case 0x10:
201 	case 0x11:
202 	case 0x12:
203 	case 0x14:
204 	case 0x15:
205 		break;
206 	default:
207 		return (ENXIO);
208 	}
209 	device_set_desc(dev, "AMD CPU On-Die Thermal Sensors");
210 
211 	return (BUS_PROBE_GENERIC);
212 }
213 
214 static int
215 amdtemp_attach(device_t dev)
216 {
217 	char tn[32];
218 	u_int regs[4];
219 	struct amdtemp_softc *sc = device_get_softc(dev);
220 	struct sysctl_ctx_list *sysctlctx;
221 	struct sysctl_oid *sysctlnode;
222 	uint32_t cpuid, family, model;
223 	u_int bid;
224 	int erratum319, unit;
225 
226 	erratum319 = 0;
227 
228 	/*
229 	 * CPUID Register is available from Revision F.
230 	 */
231 	cpuid = cpu_id;
232 	family = CPUID_TO_FAMILY(cpuid);
233 	model = CPUID_TO_MODEL(cpuid);
234 	if (family != 0x0f || model >= 0x40) {
235 		cpuid = pci_read_config(dev, AMDTEMP_CPUID, 4);
236 		family = CPUID_TO_FAMILY(cpuid);
237 		model = CPUID_TO_MODEL(cpuid);
238 	}
239 
240 	switch (family) {
241 	case 0x0f:
242 		/*
243 		 * Thermaltrip Status Register
244 		 *
245 		 * - ThermSenseCoreSel
246 		 *
247 		 * Revision F & G:	0 - Core1, 1 - Core0
248 		 * Other:		0 - Core0, 1 - Core1
249 		 *
250 		 * - CurTmp
251 		 *
252 		 * Revision G:		bits 23-14
253 		 * Other:		bits 23-16
254 		 *
255 		 * XXX According to the BKDG, CurTmp, ThermSenseSel and
256 		 * ThermSenseCoreSel bits were introduced in Revision F
257 		 * but CurTmp seems working fine as early as Revision C.
258 		 * However, it is not clear whether ThermSenseSel and/or
259 		 * ThermSenseCoreSel work in undocumented cases as well.
260 		 * In fact, the Linux driver suggests it may not work but
261 		 * we just assume it does until we find otherwise.
262 		 *
263 		 * XXX According to Linux, CurTmp starts at -28C on
264 		 * Socket AM2 Revision G processors, which is not
265 		 * documented anywhere.
266 		 */
267 		if (model >= 0x40)
268 			sc->sc_flags |= AMDTEMP_FLAG_CS_SWAP;
269 		if (model >= 0x60 && model != 0xc1) {
270 			do_cpuid(0x80000001, regs);
271 			bid = (regs[1] >> 9) & 0x1f;
272 			switch (model) {
273 			case 0x68: /* Socket S1g1 */
274 			case 0x6c:
275 			case 0x7c:
276 				break;
277 			case 0x6b: /* Socket AM2 and ASB1 (2 cores) */
278 				if (bid != 0x0b && bid != 0x0c)
279 					sc->sc_flags |=
280 					    AMDTEMP_FLAG_ALT_OFFSET;
281 				break;
282 			case 0x6f: /* Socket AM2 and ASB1 (1 core) */
283 			case 0x7f:
284 				if (bid != 0x07 && bid != 0x09 &&
285 				    bid != 0x0c)
286 					sc->sc_flags |=
287 					    AMDTEMP_FLAG_ALT_OFFSET;
288 				break;
289 			default:
290 				sc->sc_flags |= AMDTEMP_FLAG_ALT_OFFSET;
291 			}
292 			sc->sc_flags |= AMDTEMP_FLAG_CT_10BIT;
293 		}
294 
295 		/*
296 		 * There are two sensors per core.
297 		 */
298 		sc->sc_ntemps = 2;
299 
300 		sc->sc_gettemp = amdtemp_gettemp0f;
301 		break;
302 	case 0x10:
303 		/*
304 		 * Erratum 319 Inaccurate Temperature Measurement
305 		 *
306 		 * http://support.amd.com/us/Processor_TechDocs/41322.pdf
307 		 */
308 		do_cpuid(0x80000001, regs);
309 		switch ((regs[1] >> 28) & 0xf) {
310 		case 0:	/* Socket F */
311 			erratum319 = 1;
312 			break;
313 		case 1:	/* Socket AM2+ or AM3 */
314 			if ((pci_cfgregread(pci_get_bus(dev),
315 			    pci_get_slot(dev), 2, AMDTEMP_DRAM_CONF_HIGH, 2) &
316 			    AMDTEMP_DRAM_MODE_DDR3) != 0 || model > 0x04 ||
317 			    (model == 0x04 && (cpuid & CPUID_STEPPING) >= 3))
318 				break;
319 			/* XXX 00100F42h (RB-C2) exists in both formats. */
320 			erratum319 = 1;
321 			break;
322 		}
323 		/* FALLTHROUGH */
324 	case 0x11:
325 	case 0x12:
326 	case 0x14:
327 	case 0x15:
328 		/*
329 		 * There is only one sensor per package.
330 		 */
331 		sc->sc_ntemps = 1;
332 
333 		sc->sc_gettemp = amdtemp_gettemp;
334 		break;
335 	}
336 
337 	/* Find number of cores per package. */
338 	sc->sc_ncores = (amd_feature2 & AMDID2_CMP) != 0 ?
339 	    (cpu_procinfo2 & AMDID_CMP_CORES) + 1 : 1;
340 	if (sc->sc_ncores > MAXCPU)
341 		return (ENXIO);
342 
343 	if (erratum319)
344 		device_printf(dev,
345 		    "Erratum 319: temperature measurement may be inaccurate\n");
346 	if (bootverbose)
347 		device_printf(dev, "Found %d cores and %d sensors.\n",
348 		    sc->sc_ncores,
349 		    sc->sc_ntemps > 1 ? sc->sc_ntemps * sc->sc_ncores : 1);
350 
351 	/*
352 	 * dev.amdtemp.N tree.
353 	 */
354 	unit = device_get_unit(dev);
355 	snprintf(tn, sizeof(tn), "dev.amdtemp.%d.sensor_offset", unit);
356 	TUNABLE_INT_FETCH(tn, &sc->sc_offset);
357 
358 	sysctlctx = device_get_sysctl_ctx(dev);
359 	SYSCTL_ADD_INT(sysctlctx,
360 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO,
361 	    "sensor_offset", CTLFLAG_RW, &sc->sc_offset, 0,
362 	    "Temperature sensor offset");
363 	sysctlnode = SYSCTL_ADD_NODE(sysctlctx,
364 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO,
365 	    "core0", CTLFLAG_RD, 0, "Core 0");
366 
367 	SYSCTL_ADD_PROC(sysctlctx,
368 	    SYSCTL_CHILDREN(sysctlnode),
369 	    OID_AUTO, "sensor0", CTLTYPE_INT | CTLFLAG_RD,
370 	    dev, CORE0_SENSOR0, amdtemp_sysctl, "IK",
371 	    "Core 0 / Sensor 0 temperature");
372 
373 	if (sc->sc_ntemps > 1) {
374 		SYSCTL_ADD_PROC(sysctlctx,
375 		    SYSCTL_CHILDREN(sysctlnode),
376 		    OID_AUTO, "sensor1", CTLTYPE_INT | CTLFLAG_RD,
377 		    dev, CORE0_SENSOR1, amdtemp_sysctl, "IK",
378 		    "Core 0 / Sensor 1 temperature");
379 
380 		if (sc->sc_ncores > 1) {
381 			sysctlnode = SYSCTL_ADD_NODE(sysctlctx,
382 			    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
383 			    OID_AUTO, "core1", CTLFLAG_RD, 0, "Core 1");
384 
385 			SYSCTL_ADD_PROC(sysctlctx,
386 			    SYSCTL_CHILDREN(sysctlnode),
387 			    OID_AUTO, "sensor0", CTLTYPE_INT | CTLFLAG_RD,
388 			    dev, CORE1_SENSOR0, amdtemp_sysctl, "IK",
389 			    "Core 1 / Sensor 0 temperature");
390 
391 			SYSCTL_ADD_PROC(sysctlctx,
392 			    SYSCTL_CHILDREN(sysctlnode),
393 			    OID_AUTO, "sensor1", CTLTYPE_INT | CTLFLAG_RD,
394 			    dev, CORE1_SENSOR1, amdtemp_sysctl, "IK",
395 			    "Core 1 / Sensor 1 temperature");
396 		}
397 	}
398 
399 	/*
400 	 * Try to create dev.cpu sysctl entries and setup intrhook function.
401 	 * This is needed because the cpu driver may be loaded late on boot,
402 	 * after us.
403 	 */
404 	amdtemp_intrhook(dev);
405 	sc->sc_ich.ich_func = amdtemp_intrhook;
406 	sc->sc_ich.ich_arg = dev;
407 	if (config_intrhook_establish(&sc->sc_ich) != 0) {
408 		device_printf(dev, "config_intrhook_establish failed!\n");
409 		return (ENXIO);
410 	}
411 
412 	return (0);
413 }
414 
415 void
416 amdtemp_intrhook(void *arg)
417 {
418 	struct amdtemp_softc *sc;
419 	struct sysctl_ctx_list *sysctlctx;
420 	device_t dev = (device_t)arg;
421 	device_t acpi, cpu, nexus;
422 	amdsensor_t sensor;
423 	int i;
424 
425 	sc = device_get_softc(dev);
426 
427 	/*
428 	 * dev.cpu.N.temperature.
429 	 */
430 	nexus = device_find_child(root_bus, "nexus", 0);
431 	acpi = device_find_child(nexus, "acpi", 0);
432 
433 	for (i = 0; i < sc->sc_ncores; i++) {
434 		if (sc->sc_sysctl_cpu[i] != NULL)
435 			continue;
436 		cpu = device_find_child(acpi, "cpu",
437 		    device_get_unit(dev) * sc->sc_ncores + i);
438 		if (cpu != NULL) {
439 			sysctlctx = device_get_sysctl_ctx(cpu);
440 
441 			sensor = sc->sc_ntemps > 1 ?
442 			    (i == 0 ? CORE0 : CORE1) : CORE0_SENSOR0;
443 			sc->sc_sysctl_cpu[i] = SYSCTL_ADD_PROC(sysctlctx,
444 			    SYSCTL_CHILDREN(device_get_sysctl_tree(cpu)),
445 			    OID_AUTO, "temperature", CTLTYPE_INT | CTLFLAG_RD,
446 			    dev, sensor, amdtemp_sysctl, "IK",
447 			    "Current temparature");
448 		}
449 	}
450 	if (sc->sc_ich.ich_arg != NULL)
451 		config_intrhook_disestablish(&sc->sc_ich);
452 }
453 
454 int
455 amdtemp_detach(device_t dev)
456 {
457 	struct amdtemp_softc *sc = device_get_softc(dev);
458 	int i;
459 
460 	for (i = 0; i < sc->sc_ncores; i++)
461 		if (sc->sc_sysctl_cpu[i] != NULL)
462 			sysctl_remove_oid(sc->sc_sysctl_cpu[i], 1, 0);
463 
464 	/* NewBus removes the dev.amdtemp.N tree by itself. */
465 
466 	return (0);
467 }
468 
469 static int
470 amdtemp_sysctl(SYSCTL_HANDLER_ARGS)
471 {
472 	device_t dev = (device_t)arg1;
473 	struct amdtemp_softc *sc = device_get_softc(dev);
474 	amdsensor_t sensor = (amdsensor_t)arg2;
475 	int32_t auxtemp[2], temp;
476 	int error;
477 
478 	switch (sensor) {
479 	case CORE0:
480 		auxtemp[0] = sc->sc_gettemp(dev, CORE0_SENSOR0);
481 		auxtemp[1] = sc->sc_gettemp(dev, CORE0_SENSOR1);
482 		temp = imax(auxtemp[0], auxtemp[1]);
483 		break;
484 	case CORE1:
485 		auxtemp[0] = sc->sc_gettemp(dev, CORE1_SENSOR0);
486 		auxtemp[1] = sc->sc_gettemp(dev, CORE1_SENSOR1);
487 		temp = imax(auxtemp[0], auxtemp[1]);
488 		break;
489 	default:
490 		temp = sc->sc_gettemp(dev, sensor);
491 		break;
492 	}
493 	error = sysctl_handle_int(oidp, &temp, 0, req);
494 
495 	return (error);
496 }
497 
498 #define	AMDTEMP_ZERO_C_TO_K	2732
499 
500 static int32_t
501 amdtemp_gettemp0f(device_t dev, amdsensor_t sensor)
502 {
503 	struct amdtemp_softc *sc = device_get_softc(dev);
504 	uint32_t mask, offset, temp;
505 
506 	/* Set Sensor/Core selector. */
507 	temp = pci_read_config(dev, AMDTEMP_THERMTP_STAT, 1);
508 	temp &= ~(AMDTEMP_TTSR_SELCORE | AMDTEMP_TTSR_SELSENSOR);
509 	switch (sensor) {
510 	case CORE0_SENSOR1:
511 		temp |= AMDTEMP_TTSR_SELSENSOR;
512 		/* FALLTHROUGH */
513 	case CORE0_SENSOR0:
514 	case CORE0:
515 		if ((sc->sc_flags & AMDTEMP_FLAG_CS_SWAP) != 0)
516 			temp |= AMDTEMP_TTSR_SELCORE;
517 		break;
518 	case CORE1_SENSOR1:
519 		temp |= AMDTEMP_TTSR_SELSENSOR;
520 		/* FALLTHROUGH */
521 	case CORE1_SENSOR0:
522 	case CORE1:
523 		if ((sc->sc_flags & AMDTEMP_FLAG_CS_SWAP) == 0)
524 			temp |= AMDTEMP_TTSR_SELCORE;
525 		break;
526 	}
527 	pci_write_config(dev, AMDTEMP_THERMTP_STAT, temp, 1);
528 
529 	mask = (sc->sc_flags & AMDTEMP_FLAG_CT_10BIT) != 0 ? 0x3ff : 0x3fc;
530 	offset = (sc->sc_flags & AMDTEMP_FLAG_ALT_OFFSET) != 0 ? 28 : 49;
531 	temp = pci_read_config(dev, AMDTEMP_THERMTP_STAT, 4);
532 	temp = ((temp >> 14) & mask) * 5 / 2;
533 	temp += AMDTEMP_ZERO_C_TO_K + (sc->sc_offset - offset) * 10;
534 
535 	return (temp);
536 }
537 
538 static int32_t
539 amdtemp_gettemp(device_t dev, amdsensor_t sensor)
540 {
541 	struct amdtemp_softc *sc = device_get_softc(dev);
542 	uint32_t temp;
543 
544 	temp = pci_read_config(dev, AMDTEMP_REPTMP_CTRL, 4);
545 	temp = ((temp >> 21) & 0x7ff) * 5 / 4;
546 	temp += AMDTEMP_ZERO_C_TO_K + sc->sc_offset * 10;
547 
548 	return (temp);
549 }
550