xref: /freebsd/sys/dev/amdtemp/amdtemp.c (revision 4abd7edcbde21ba7a089c7d1a0bba8f87ebece06)
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_MISC12	0x1403
80 #define	DEVICEID_AMD_MISC14	0x1703
81 #define	DEVICEID_AMD_MISC15	0x1603
82 
83 static struct amdtemp_product {
84 	uint16_t	amdtemp_vendorid;
85 	uint16_t	amdtemp_deviceid;
86 } amdtemp_products[] = {
87 	{ VENDORID_AMD,	DEVICEID_AMD_MISC0F },
88 	{ VENDORID_AMD,	DEVICEID_AMD_MISC10 },
89 	{ VENDORID_AMD,	DEVICEID_AMD_MISC11 },
90 	{ VENDORID_AMD,	DEVICEID_AMD_MISC12 },
91 	{ VENDORID_AMD,	DEVICEID_AMD_MISC14 },
92 	{ VENDORID_AMD,	DEVICEID_AMD_MISC15 },
93 	{ 0, 0 }
94 };
95 
96 /*
97  * Reported Temperature Control Register
98  */
99 #define	AMDTEMP_REPTMP_CTRL	0xa4
100 
101 /*
102  * Thermaltrip Status Register (Family 0Fh only)
103  */
104 #define	AMDTEMP_THERMTP_STAT	0xe4
105 #define	AMDTEMP_TTSR_SELCORE	0x04
106 #define	AMDTEMP_TTSR_SELSENSOR	0x40
107 
108 /*
109  * DRAM Configuration High Register
110  */
111 #define	AMDTEMP_DRAM_CONF_HIGH	0x94	/* Function 2 */
112 #define	AMDTEMP_DRAM_MODE_DDR3	0x0100
113 
114 /*
115  * CPU Family/Model Register
116  */
117 #define	AMDTEMP_CPUID		0xfc
118 
119 /*
120  * Device methods.
121  */
122 static void 	amdtemp_identify(driver_t *driver, device_t parent);
123 static int	amdtemp_probe(device_t dev);
124 static int	amdtemp_attach(device_t dev);
125 static void	amdtemp_intrhook(void *arg);
126 static int	amdtemp_detach(device_t dev);
127 static int 	amdtemp_match(device_t dev);
128 static int32_t	amdtemp_gettemp0f(device_t dev, amdsensor_t sensor);
129 static int32_t	amdtemp_gettemp(device_t dev, amdsensor_t sensor);
130 static int	amdtemp_sysctl(SYSCTL_HANDLER_ARGS);
131 
132 static device_method_t amdtemp_methods[] = {
133 	/* Device interface */
134 	DEVMETHOD(device_identify,	amdtemp_identify),
135 	DEVMETHOD(device_probe,		amdtemp_probe),
136 	DEVMETHOD(device_attach,	amdtemp_attach),
137 	DEVMETHOD(device_detach,	amdtemp_detach),
138 
139 	DEVMETHOD_END
140 };
141 
142 static driver_t amdtemp_driver = {
143 	"amdtemp",
144 	amdtemp_methods,
145 	sizeof(struct amdtemp_softc),
146 };
147 
148 static devclass_t amdtemp_devclass;
149 DRIVER_MODULE(amdtemp, hostb, amdtemp_driver, amdtemp_devclass, NULL, NULL);
150 
151 static int
152 amdtemp_match(device_t dev)
153 {
154 	int i;
155 	uint16_t vendor, devid;
156 
157 	vendor = pci_get_vendor(dev);
158 	devid = pci_get_device(dev);
159 
160 	for (i = 0; amdtemp_products[i].amdtemp_vendorid != 0; i++) {
161 		if (vendor == amdtemp_products[i].amdtemp_vendorid &&
162 		    devid == amdtemp_products[i].amdtemp_deviceid)
163 			return (1);
164 	}
165 
166 	return (0);
167 }
168 
169 static void
170 amdtemp_identify(driver_t *driver, device_t parent)
171 {
172 	device_t child;
173 
174 	/* Make sure we're not being doubly invoked. */
175 	if (device_find_child(parent, "amdtemp", -1) != NULL)
176 		return;
177 
178 	if (amdtemp_match(parent)) {
179 		child = device_add_child(parent, "amdtemp", -1);
180 		if (child == NULL)
181 			device_printf(parent, "add amdtemp child failed\n");
182 	}
183 }
184 
185 static int
186 amdtemp_probe(device_t dev)
187 {
188 	uint32_t family, model;
189 
190 	if (resource_disabled("amdtemp", 0))
191 		return (ENXIO);
192 
193 	family = CPUID_TO_FAMILY(cpu_id);
194 	model = CPUID_TO_MODEL(cpu_id);
195 
196 	switch (family) {
197 	case 0x0f:
198 		if ((model == 0x04 && (cpu_id & CPUID_STEPPING) == 0) ||
199 		    (model == 0x05 && (cpu_id & CPUID_STEPPING) <= 1))
200 			return (ENXIO);
201 		break;
202 	case 0x10:
203 	case 0x11:
204 	case 0x12:
205 	case 0x14:
206 	case 0x15:
207 		break;
208 	default:
209 		return (ENXIO);
210 	}
211 	device_set_desc(dev, "AMD CPU On-Die Thermal Sensors");
212 
213 	return (BUS_PROBE_GENERIC);
214 }
215 
216 static int
217 amdtemp_attach(device_t dev)
218 {
219 	char tn[32];
220 	u_int regs[4];
221 	struct amdtemp_softc *sc = device_get_softc(dev);
222 	struct sysctl_ctx_list *sysctlctx;
223 	struct sysctl_oid *sysctlnode;
224 	uint32_t cpuid, family, model;
225 	u_int bid;
226 	int erratum319, unit;
227 
228 	erratum319 = 0;
229 
230 	/*
231 	 * CPUID Register is available from Revision F.
232 	 */
233 	cpuid = cpu_id;
234 	family = CPUID_TO_FAMILY(cpuid);
235 	model = CPUID_TO_MODEL(cpuid);
236 	if (family != 0x0f || model >= 0x40) {
237 		cpuid = pci_read_config(dev, AMDTEMP_CPUID, 4);
238 		family = CPUID_TO_FAMILY(cpuid);
239 		model = CPUID_TO_MODEL(cpuid);
240 	}
241 
242 	switch (family) {
243 	case 0x0f:
244 		/*
245 		 * Thermaltrip Status Register
246 		 *
247 		 * - ThermSenseCoreSel
248 		 *
249 		 * Revision F & G:	0 - Core1, 1 - Core0
250 		 * Other:		0 - Core0, 1 - Core1
251 		 *
252 		 * - CurTmp
253 		 *
254 		 * Revision G:		bits 23-14
255 		 * Other:		bits 23-16
256 		 *
257 		 * XXX According to the BKDG, CurTmp, ThermSenseSel and
258 		 * ThermSenseCoreSel bits were introduced in Revision F
259 		 * but CurTmp seems working fine as early as Revision C.
260 		 * However, it is not clear whether ThermSenseSel and/or
261 		 * ThermSenseCoreSel work in undocumented cases as well.
262 		 * In fact, the Linux driver suggests it may not work but
263 		 * we just assume it does until we find otherwise.
264 		 *
265 		 * XXX According to Linux, CurTmp starts at -28C on
266 		 * Socket AM2 Revision G processors, which is not
267 		 * documented anywhere.
268 		 */
269 		if (model >= 0x40)
270 			sc->sc_flags |= AMDTEMP_FLAG_CS_SWAP;
271 		if (model >= 0x60 && model != 0xc1) {
272 			do_cpuid(0x80000001, regs);
273 			bid = (regs[1] >> 9) & 0x1f;
274 			switch (model) {
275 			case 0x68: /* Socket S1g1 */
276 			case 0x6c:
277 			case 0x7c:
278 				break;
279 			case 0x6b: /* Socket AM2 and ASB1 (2 cores) */
280 				if (bid != 0x0b && bid != 0x0c)
281 					sc->sc_flags |=
282 					    AMDTEMP_FLAG_ALT_OFFSET;
283 				break;
284 			case 0x6f: /* Socket AM2 and ASB1 (1 core) */
285 			case 0x7f:
286 				if (bid != 0x07 && bid != 0x09 &&
287 				    bid != 0x0c)
288 					sc->sc_flags |=
289 					    AMDTEMP_FLAG_ALT_OFFSET;
290 				break;
291 			default:
292 				sc->sc_flags |= AMDTEMP_FLAG_ALT_OFFSET;
293 			}
294 			sc->sc_flags |= AMDTEMP_FLAG_CT_10BIT;
295 		}
296 
297 		/*
298 		 * There are two sensors per core.
299 		 */
300 		sc->sc_ntemps = 2;
301 
302 		sc->sc_gettemp = amdtemp_gettemp0f;
303 		break;
304 	case 0x10:
305 		/*
306 		 * Erratum 319 Inaccurate Temperature Measurement
307 		 *
308 		 * http://support.amd.com/us/Processor_TechDocs/41322.pdf
309 		 */
310 		do_cpuid(0x80000001, regs);
311 		switch ((regs[1] >> 28) & 0xf) {
312 		case 0:	/* Socket F */
313 			erratum319 = 1;
314 			break;
315 		case 1:	/* Socket AM2+ or AM3 */
316 			if ((pci_cfgregread(pci_get_bus(dev),
317 			    pci_get_slot(dev), 2, AMDTEMP_DRAM_CONF_HIGH, 2) &
318 			    AMDTEMP_DRAM_MODE_DDR3) != 0 || model > 0x04 ||
319 			    (model == 0x04 && (cpuid & CPUID_STEPPING) >= 3))
320 				break;
321 			/* XXX 00100F42h (RB-C2) exists in both formats. */
322 			erratum319 = 1;
323 			break;
324 		}
325 		/* FALLTHROUGH */
326 	case 0x11:
327 	case 0x12:
328 	case 0x14:
329 	case 0x15:
330 		/*
331 		 * There is only one sensor per package.
332 		 */
333 		sc->sc_ntemps = 1;
334 
335 		sc->sc_gettemp = amdtemp_gettemp;
336 		break;
337 	}
338 
339 	/* Find number of cores per package. */
340 	sc->sc_ncores = (amd_feature2 & AMDID2_CMP) != 0 ?
341 	    (cpu_procinfo2 & AMDID_CMP_CORES) + 1 : 1;
342 	if (sc->sc_ncores > MAXCPU)
343 		return (ENXIO);
344 
345 	if (erratum319)
346 		device_printf(dev,
347 		    "Erratum 319: temperature measurement may be inaccurate\n");
348 	if (bootverbose)
349 		device_printf(dev, "Found %d cores and %d sensors.\n",
350 		    sc->sc_ncores,
351 		    sc->sc_ntemps > 1 ? sc->sc_ntemps * sc->sc_ncores : 1);
352 
353 	/*
354 	 * dev.amdtemp.N tree.
355 	 */
356 	unit = device_get_unit(dev);
357 	snprintf(tn, sizeof(tn), "dev.amdtemp.%d.sensor_offset", unit);
358 	TUNABLE_INT_FETCH(tn, &sc->sc_offset);
359 
360 	sysctlctx = device_get_sysctl_ctx(dev);
361 	SYSCTL_ADD_INT(sysctlctx,
362 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO,
363 	    "sensor_offset", CTLFLAG_RW, &sc->sc_offset, 0,
364 	    "Temperature sensor offset");
365 	sysctlnode = SYSCTL_ADD_NODE(sysctlctx,
366 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO,
367 	    "core0", CTLFLAG_RD, 0, "Core 0");
368 
369 	SYSCTL_ADD_PROC(sysctlctx,
370 	    SYSCTL_CHILDREN(sysctlnode),
371 	    OID_AUTO, "sensor0", CTLTYPE_INT | CTLFLAG_RD,
372 	    dev, CORE0_SENSOR0, amdtemp_sysctl, "IK",
373 	    "Core 0 / Sensor 0 temperature");
374 
375 	if (sc->sc_ntemps > 1) {
376 		SYSCTL_ADD_PROC(sysctlctx,
377 		    SYSCTL_CHILDREN(sysctlnode),
378 		    OID_AUTO, "sensor1", CTLTYPE_INT | CTLFLAG_RD,
379 		    dev, CORE0_SENSOR1, amdtemp_sysctl, "IK",
380 		    "Core 0 / Sensor 1 temperature");
381 
382 		if (sc->sc_ncores > 1) {
383 			sysctlnode = SYSCTL_ADD_NODE(sysctlctx,
384 			    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
385 			    OID_AUTO, "core1", CTLFLAG_RD, 0, "Core 1");
386 
387 			SYSCTL_ADD_PROC(sysctlctx,
388 			    SYSCTL_CHILDREN(sysctlnode),
389 			    OID_AUTO, "sensor0", CTLTYPE_INT | CTLFLAG_RD,
390 			    dev, CORE1_SENSOR0, amdtemp_sysctl, "IK",
391 			    "Core 1 / Sensor 0 temperature");
392 
393 			SYSCTL_ADD_PROC(sysctlctx,
394 			    SYSCTL_CHILDREN(sysctlnode),
395 			    OID_AUTO, "sensor1", CTLTYPE_INT | CTLFLAG_RD,
396 			    dev, CORE1_SENSOR1, amdtemp_sysctl, "IK",
397 			    "Core 1 / Sensor 1 temperature");
398 		}
399 	}
400 
401 	/*
402 	 * Try to create dev.cpu sysctl entries and setup intrhook function.
403 	 * This is needed because the cpu driver may be loaded late on boot,
404 	 * after us.
405 	 */
406 	amdtemp_intrhook(dev);
407 	sc->sc_ich.ich_func = amdtemp_intrhook;
408 	sc->sc_ich.ich_arg = dev;
409 	if (config_intrhook_establish(&sc->sc_ich) != 0) {
410 		device_printf(dev, "config_intrhook_establish failed!\n");
411 		return (ENXIO);
412 	}
413 
414 	return (0);
415 }
416 
417 void
418 amdtemp_intrhook(void *arg)
419 {
420 	struct amdtemp_softc *sc;
421 	struct sysctl_ctx_list *sysctlctx;
422 	device_t dev = (device_t)arg;
423 	device_t acpi, cpu, nexus;
424 	amdsensor_t sensor;
425 	int i;
426 
427 	sc = device_get_softc(dev);
428 
429 	/*
430 	 * dev.cpu.N.temperature.
431 	 */
432 	nexus = device_find_child(root_bus, "nexus", 0);
433 	acpi = device_find_child(nexus, "acpi", 0);
434 
435 	for (i = 0; i < sc->sc_ncores; i++) {
436 		if (sc->sc_sysctl_cpu[i] != NULL)
437 			continue;
438 		cpu = device_find_child(acpi, "cpu",
439 		    device_get_unit(dev) * sc->sc_ncores + i);
440 		if (cpu != NULL) {
441 			sysctlctx = device_get_sysctl_ctx(cpu);
442 
443 			sensor = sc->sc_ntemps > 1 ?
444 			    (i == 0 ? CORE0 : CORE1) : CORE0_SENSOR0;
445 			sc->sc_sysctl_cpu[i] = SYSCTL_ADD_PROC(sysctlctx,
446 			    SYSCTL_CHILDREN(device_get_sysctl_tree(cpu)),
447 			    OID_AUTO, "temperature", CTLTYPE_INT | CTLFLAG_RD,
448 			    dev, sensor, amdtemp_sysctl, "IK",
449 			    "Current temparature");
450 		}
451 	}
452 	if (sc->sc_ich.ich_arg != NULL)
453 		config_intrhook_disestablish(&sc->sc_ich);
454 }
455 
456 int
457 amdtemp_detach(device_t dev)
458 {
459 	struct amdtemp_softc *sc = device_get_softc(dev);
460 	int i;
461 
462 	for (i = 0; i < sc->sc_ncores; i++)
463 		if (sc->sc_sysctl_cpu[i] != NULL)
464 			sysctl_remove_oid(sc->sc_sysctl_cpu[i], 1, 0);
465 
466 	/* NewBus removes the dev.amdtemp.N tree by itself. */
467 
468 	return (0);
469 }
470 
471 static int
472 amdtemp_sysctl(SYSCTL_HANDLER_ARGS)
473 {
474 	device_t dev = (device_t)arg1;
475 	struct amdtemp_softc *sc = device_get_softc(dev);
476 	amdsensor_t sensor = (amdsensor_t)arg2;
477 	int32_t auxtemp[2], temp;
478 	int error;
479 
480 	switch (sensor) {
481 	case CORE0:
482 		auxtemp[0] = sc->sc_gettemp(dev, CORE0_SENSOR0);
483 		auxtemp[1] = sc->sc_gettemp(dev, CORE0_SENSOR1);
484 		temp = imax(auxtemp[0], auxtemp[1]);
485 		break;
486 	case CORE1:
487 		auxtemp[0] = sc->sc_gettemp(dev, CORE1_SENSOR0);
488 		auxtemp[1] = sc->sc_gettemp(dev, CORE1_SENSOR1);
489 		temp = imax(auxtemp[0], auxtemp[1]);
490 		break;
491 	default:
492 		temp = sc->sc_gettemp(dev, sensor);
493 		break;
494 	}
495 	error = sysctl_handle_int(oidp, &temp, 0, req);
496 
497 	return (error);
498 }
499 
500 #define	AMDTEMP_ZERO_C_TO_K	2732
501 
502 static int32_t
503 amdtemp_gettemp0f(device_t dev, amdsensor_t sensor)
504 {
505 	struct amdtemp_softc *sc = device_get_softc(dev);
506 	uint32_t mask, offset, temp;
507 
508 	/* Set Sensor/Core selector. */
509 	temp = pci_read_config(dev, AMDTEMP_THERMTP_STAT, 1);
510 	temp &= ~(AMDTEMP_TTSR_SELCORE | AMDTEMP_TTSR_SELSENSOR);
511 	switch (sensor) {
512 	case CORE0_SENSOR1:
513 		temp |= AMDTEMP_TTSR_SELSENSOR;
514 		/* FALLTHROUGH */
515 	case CORE0_SENSOR0:
516 	case CORE0:
517 		if ((sc->sc_flags & AMDTEMP_FLAG_CS_SWAP) != 0)
518 			temp |= AMDTEMP_TTSR_SELCORE;
519 		break;
520 	case CORE1_SENSOR1:
521 		temp |= AMDTEMP_TTSR_SELSENSOR;
522 		/* FALLTHROUGH */
523 	case CORE1_SENSOR0:
524 	case CORE1:
525 		if ((sc->sc_flags & AMDTEMP_FLAG_CS_SWAP) == 0)
526 			temp |= AMDTEMP_TTSR_SELCORE;
527 		break;
528 	}
529 	pci_write_config(dev, AMDTEMP_THERMTP_STAT, temp, 1);
530 
531 	mask = (sc->sc_flags & AMDTEMP_FLAG_CT_10BIT) != 0 ? 0x3ff : 0x3fc;
532 	offset = (sc->sc_flags & AMDTEMP_FLAG_ALT_OFFSET) != 0 ? 28 : 49;
533 	temp = pci_read_config(dev, AMDTEMP_THERMTP_STAT, 4);
534 	temp = ((temp >> 14) & mask) * 5 / 2;
535 	temp += AMDTEMP_ZERO_C_TO_K + (sc->sc_offset - offset) * 10;
536 
537 	return (temp);
538 }
539 
540 static int32_t
541 amdtemp_gettemp(device_t dev, amdsensor_t sensor)
542 {
543 	struct amdtemp_softc *sc = device_get_softc(dev);
544 	uint32_t temp;
545 
546 	temp = pci_read_config(dev, AMDTEMP_REPTMP_CTRL, 4);
547 	temp = ((temp >> 21) & 0x7ff) * 5 / 4;
548 	temp += AMDTEMP_ZERO_C_TO_K + sc->sc_offset * 10;
549 
550 	return (temp);
551 }
552