xref: /freebsd/sys/dev/amdtemp/amdtemp.c (revision 2b15cb3d0922bd70ea592f0da9b4a5b167f4d53f)
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 #define	DEVICEID_AMD_MISC16	0x1533
83 #define	DEVICEID_AMD_MISC17	0x141d
84 
85 static struct amdtemp_product {
86 	uint16_t	amdtemp_vendorid;
87 	uint16_t	amdtemp_deviceid;
88 } amdtemp_products[] = {
89 	{ VENDORID_AMD,	DEVICEID_AMD_MISC0F },
90 	{ VENDORID_AMD,	DEVICEID_AMD_MISC10 },
91 	{ VENDORID_AMD,	DEVICEID_AMD_MISC11 },
92 	{ VENDORID_AMD,	DEVICEID_AMD_MISC12 },
93 	{ VENDORID_AMD,	DEVICEID_AMD_MISC14 },
94 	{ VENDORID_AMD,	DEVICEID_AMD_MISC15 },
95 	{ VENDORID_AMD,	DEVICEID_AMD_MISC16 },
96 	{ VENDORID_AMD,	DEVICEID_AMD_MISC17 },
97 	{ 0, 0 }
98 };
99 
100 /*
101  * Reported Temperature Control Register
102  */
103 #define	AMDTEMP_REPTMP_CTRL	0xa4
104 
105 /*
106  * Thermaltrip Status Register (Family 0Fh only)
107  */
108 #define	AMDTEMP_THERMTP_STAT	0xe4
109 #define	AMDTEMP_TTSR_SELCORE	0x04
110 #define	AMDTEMP_TTSR_SELSENSOR	0x40
111 
112 /*
113  * DRAM Configuration High Register
114  */
115 #define	AMDTEMP_DRAM_CONF_HIGH	0x94	/* Function 2 */
116 #define	AMDTEMP_DRAM_MODE_DDR3	0x0100
117 
118 /*
119  * CPU Family/Model Register
120  */
121 #define	AMDTEMP_CPUID		0xfc
122 
123 /*
124  * Device methods.
125  */
126 static void 	amdtemp_identify(driver_t *driver, device_t parent);
127 static int	amdtemp_probe(device_t dev);
128 static int	amdtemp_attach(device_t dev);
129 static void	amdtemp_intrhook(void *arg);
130 static int	amdtemp_detach(device_t dev);
131 static int 	amdtemp_match(device_t dev);
132 static int32_t	amdtemp_gettemp0f(device_t dev, amdsensor_t sensor);
133 static int32_t	amdtemp_gettemp(device_t dev, amdsensor_t sensor);
134 static int	amdtemp_sysctl(SYSCTL_HANDLER_ARGS);
135 
136 static device_method_t amdtemp_methods[] = {
137 	/* Device interface */
138 	DEVMETHOD(device_identify,	amdtemp_identify),
139 	DEVMETHOD(device_probe,		amdtemp_probe),
140 	DEVMETHOD(device_attach,	amdtemp_attach),
141 	DEVMETHOD(device_detach,	amdtemp_detach),
142 
143 	DEVMETHOD_END
144 };
145 
146 static driver_t amdtemp_driver = {
147 	"amdtemp",
148 	amdtemp_methods,
149 	sizeof(struct amdtemp_softc),
150 };
151 
152 static devclass_t amdtemp_devclass;
153 DRIVER_MODULE(amdtemp, hostb, amdtemp_driver, amdtemp_devclass, NULL, NULL);
154 
155 static int
156 amdtemp_match(device_t dev)
157 {
158 	int i;
159 	uint16_t vendor, devid;
160 
161 	vendor = pci_get_vendor(dev);
162 	devid = pci_get_device(dev);
163 
164 	for (i = 0; amdtemp_products[i].amdtemp_vendorid != 0; i++) {
165 		if (vendor == amdtemp_products[i].amdtemp_vendorid &&
166 		    devid == amdtemp_products[i].amdtemp_deviceid)
167 			return (1);
168 	}
169 
170 	return (0);
171 }
172 
173 static void
174 amdtemp_identify(driver_t *driver, device_t parent)
175 {
176 	device_t child;
177 
178 	/* Make sure we're not being doubly invoked. */
179 	if (device_find_child(parent, "amdtemp", -1) != NULL)
180 		return;
181 
182 	if (amdtemp_match(parent)) {
183 		child = device_add_child(parent, "amdtemp", -1);
184 		if (child == NULL)
185 			device_printf(parent, "add amdtemp child failed\n");
186 	}
187 }
188 
189 static int
190 amdtemp_probe(device_t dev)
191 {
192 	uint32_t family, model;
193 
194 	if (resource_disabled("amdtemp", 0))
195 		return (ENXIO);
196 
197 	family = CPUID_TO_FAMILY(cpu_id);
198 	model = CPUID_TO_MODEL(cpu_id);
199 
200 	switch (family) {
201 	case 0x0f:
202 		if ((model == 0x04 && (cpu_id & CPUID_STEPPING) == 0) ||
203 		    (model == 0x05 && (cpu_id & CPUID_STEPPING) <= 1))
204 			return (ENXIO);
205 		break;
206 	case 0x10:
207 	case 0x11:
208 	case 0x12:
209 	case 0x14:
210 	case 0x15:
211 	case 0x16:
212 		break;
213 	default:
214 		return (ENXIO);
215 	}
216 	device_set_desc(dev, "AMD CPU On-Die Thermal Sensors");
217 
218 	return (BUS_PROBE_GENERIC);
219 }
220 
221 static int
222 amdtemp_attach(device_t dev)
223 {
224 	char tn[32];
225 	u_int regs[4];
226 	struct amdtemp_softc *sc = device_get_softc(dev);
227 	struct sysctl_ctx_list *sysctlctx;
228 	struct sysctl_oid *sysctlnode;
229 	uint32_t cpuid, family, model;
230 	u_int bid;
231 	int erratum319, unit;
232 
233 	erratum319 = 0;
234 
235 	/*
236 	 * CPUID Register is available from Revision F.
237 	 */
238 	cpuid = cpu_id;
239 	family = CPUID_TO_FAMILY(cpuid);
240 	model = CPUID_TO_MODEL(cpuid);
241 	if (family != 0x0f || model >= 0x40) {
242 		cpuid = pci_read_config(dev, AMDTEMP_CPUID, 4);
243 		family = CPUID_TO_FAMILY(cpuid);
244 		model = CPUID_TO_MODEL(cpuid);
245 	}
246 
247 	switch (family) {
248 	case 0x0f:
249 		/*
250 		 * Thermaltrip Status Register
251 		 *
252 		 * - ThermSenseCoreSel
253 		 *
254 		 * Revision F & G:	0 - Core1, 1 - Core0
255 		 * Other:		0 - Core0, 1 - Core1
256 		 *
257 		 * - CurTmp
258 		 *
259 		 * Revision G:		bits 23-14
260 		 * Other:		bits 23-16
261 		 *
262 		 * XXX According to the BKDG, CurTmp, ThermSenseSel and
263 		 * ThermSenseCoreSel bits were introduced in Revision F
264 		 * but CurTmp seems working fine as early as Revision C.
265 		 * However, it is not clear whether ThermSenseSel and/or
266 		 * ThermSenseCoreSel work in undocumented cases as well.
267 		 * In fact, the Linux driver suggests it may not work but
268 		 * we just assume it does until we find otherwise.
269 		 *
270 		 * XXX According to Linux, CurTmp starts at -28C on
271 		 * Socket AM2 Revision G processors, which is not
272 		 * documented anywhere.
273 		 */
274 		if (model >= 0x40)
275 			sc->sc_flags |= AMDTEMP_FLAG_CS_SWAP;
276 		if (model >= 0x60 && model != 0xc1) {
277 			do_cpuid(0x80000001, regs);
278 			bid = (regs[1] >> 9) & 0x1f;
279 			switch (model) {
280 			case 0x68: /* Socket S1g1 */
281 			case 0x6c:
282 			case 0x7c:
283 				break;
284 			case 0x6b: /* Socket AM2 and ASB1 (2 cores) */
285 				if (bid != 0x0b && bid != 0x0c)
286 					sc->sc_flags |=
287 					    AMDTEMP_FLAG_ALT_OFFSET;
288 				break;
289 			case 0x6f: /* Socket AM2 and ASB1 (1 core) */
290 			case 0x7f:
291 				if (bid != 0x07 && bid != 0x09 &&
292 				    bid != 0x0c)
293 					sc->sc_flags |=
294 					    AMDTEMP_FLAG_ALT_OFFSET;
295 				break;
296 			default:
297 				sc->sc_flags |= AMDTEMP_FLAG_ALT_OFFSET;
298 			}
299 			sc->sc_flags |= AMDTEMP_FLAG_CT_10BIT;
300 		}
301 
302 		/*
303 		 * There are two sensors per core.
304 		 */
305 		sc->sc_ntemps = 2;
306 
307 		sc->sc_gettemp = amdtemp_gettemp0f;
308 		break;
309 	case 0x10:
310 		/*
311 		 * Erratum 319 Inaccurate Temperature Measurement
312 		 *
313 		 * http://support.amd.com/us/Processor_TechDocs/41322.pdf
314 		 */
315 		do_cpuid(0x80000001, regs);
316 		switch ((regs[1] >> 28) & 0xf) {
317 		case 0:	/* Socket F */
318 			erratum319 = 1;
319 			break;
320 		case 1:	/* Socket AM2+ or AM3 */
321 			if ((pci_cfgregread(pci_get_bus(dev),
322 			    pci_get_slot(dev), 2, AMDTEMP_DRAM_CONF_HIGH, 2) &
323 			    AMDTEMP_DRAM_MODE_DDR3) != 0 || model > 0x04 ||
324 			    (model == 0x04 && (cpuid & CPUID_STEPPING) >= 3))
325 				break;
326 			/* XXX 00100F42h (RB-C2) exists in both formats. */
327 			erratum319 = 1;
328 			break;
329 		}
330 		/* FALLTHROUGH */
331 	case 0x11:
332 	case 0x12:
333 	case 0x14:
334 	case 0x15:
335 	case 0x16:
336 		/*
337 		 * There is only one sensor per package.
338 		 */
339 		sc->sc_ntemps = 1;
340 
341 		sc->sc_gettemp = amdtemp_gettemp;
342 		break;
343 	}
344 
345 	/* Find number of cores per package. */
346 	sc->sc_ncores = (amd_feature2 & AMDID2_CMP) != 0 ?
347 	    (cpu_procinfo2 & AMDID_CMP_CORES) + 1 : 1;
348 	if (sc->sc_ncores > MAXCPU)
349 		return (ENXIO);
350 
351 	if (erratum319)
352 		device_printf(dev,
353 		    "Erratum 319: temperature measurement may be inaccurate\n");
354 	if (bootverbose)
355 		device_printf(dev, "Found %d cores and %d sensors.\n",
356 		    sc->sc_ncores,
357 		    sc->sc_ntemps > 1 ? sc->sc_ntemps * sc->sc_ncores : 1);
358 
359 	/*
360 	 * dev.amdtemp.N tree.
361 	 */
362 	unit = device_get_unit(dev);
363 	snprintf(tn, sizeof(tn), "dev.amdtemp.%d.sensor_offset", unit);
364 	TUNABLE_INT_FETCH(tn, &sc->sc_offset);
365 
366 	sysctlctx = device_get_sysctl_ctx(dev);
367 	SYSCTL_ADD_INT(sysctlctx,
368 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO,
369 	    "sensor_offset", CTLFLAG_RW, &sc->sc_offset, 0,
370 	    "Temperature sensor offset");
371 	sysctlnode = SYSCTL_ADD_NODE(sysctlctx,
372 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO,
373 	    "core0", CTLFLAG_RD, 0, "Core 0");
374 
375 	SYSCTL_ADD_PROC(sysctlctx,
376 	    SYSCTL_CHILDREN(sysctlnode),
377 	    OID_AUTO, "sensor0", CTLTYPE_INT | CTLFLAG_RD,
378 	    dev, CORE0_SENSOR0, amdtemp_sysctl, "IK",
379 	    "Core 0 / Sensor 0 temperature");
380 
381 	if (sc->sc_ntemps > 1) {
382 		SYSCTL_ADD_PROC(sysctlctx,
383 		    SYSCTL_CHILDREN(sysctlnode),
384 		    OID_AUTO, "sensor1", CTLTYPE_INT | CTLFLAG_RD,
385 		    dev, CORE0_SENSOR1, amdtemp_sysctl, "IK",
386 		    "Core 0 / Sensor 1 temperature");
387 
388 		if (sc->sc_ncores > 1) {
389 			sysctlnode = SYSCTL_ADD_NODE(sysctlctx,
390 			    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
391 			    OID_AUTO, "core1", CTLFLAG_RD, 0, "Core 1");
392 
393 			SYSCTL_ADD_PROC(sysctlctx,
394 			    SYSCTL_CHILDREN(sysctlnode),
395 			    OID_AUTO, "sensor0", CTLTYPE_INT | CTLFLAG_RD,
396 			    dev, CORE1_SENSOR0, amdtemp_sysctl, "IK",
397 			    "Core 1 / Sensor 0 temperature");
398 
399 			SYSCTL_ADD_PROC(sysctlctx,
400 			    SYSCTL_CHILDREN(sysctlnode),
401 			    OID_AUTO, "sensor1", CTLTYPE_INT | CTLFLAG_RD,
402 			    dev, CORE1_SENSOR1, amdtemp_sysctl, "IK",
403 			    "Core 1 / Sensor 1 temperature");
404 		}
405 	}
406 
407 	/*
408 	 * Try to create dev.cpu sysctl entries and setup intrhook function.
409 	 * This is needed because the cpu driver may be loaded late on boot,
410 	 * after us.
411 	 */
412 	amdtemp_intrhook(dev);
413 	sc->sc_ich.ich_func = amdtemp_intrhook;
414 	sc->sc_ich.ich_arg = dev;
415 	if (config_intrhook_establish(&sc->sc_ich) != 0) {
416 		device_printf(dev, "config_intrhook_establish failed!\n");
417 		return (ENXIO);
418 	}
419 
420 	return (0);
421 }
422 
423 void
424 amdtemp_intrhook(void *arg)
425 {
426 	struct amdtemp_softc *sc;
427 	struct sysctl_ctx_list *sysctlctx;
428 	device_t dev = (device_t)arg;
429 	device_t acpi, cpu, nexus;
430 	amdsensor_t sensor;
431 	int i;
432 
433 	sc = device_get_softc(dev);
434 
435 	/*
436 	 * dev.cpu.N.temperature.
437 	 */
438 	nexus = device_find_child(root_bus, "nexus", 0);
439 	acpi = device_find_child(nexus, "acpi", 0);
440 
441 	for (i = 0; i < sc->sc_ncores; i++) {
442 		if (sc->sc_sysctl_cpu[i] != NULL)
443 			continue;
444 		cpu = device_find_child(acpi, "cpu",
445 		    device_get_unit(dev) * sc->sc_ncores + i);
446 		if (cpu != NULL) {
447 			sysctlctx = device_get_sysctl_ctx(cpu);
448 
449 			sensor = sc->sc_ntemps > 1 ?
450 			    (i == 0 ? CORE0 : CORE1) : CORE0_SENSOR0;
451 			sc->sc_sysctl_cpu[i] = SYSCTL_ADD_PROC(sysctlctx,
452 			    SYSCTL_CHILDREN(device_get_sysctl_tree(cpu)),
453 			    OID_AUTO, "temperature", CTLTYPE_INT | CTLFLAG_RD,
454 			    dev, sensor, amdtemp_sysctl, "IK",
455 			    "Current temparature");
456 		}
457 	}
458 	if (sc->sc_ich.ich_arg != NULL)
459 		config_intrhook_disestablish(&sc->sc_ich);
460 }
461 
462 int
463 amdtemp_detach(device_t dev)
464 {
465 	struct amdtemp_softc *sc = device_get_softc(dev);
466 	int i;
467 
468 	for (i = 0; i < sc->sc_ncores; i++)
469 		if (sc->sc_sysctl_cpu[i] != NULL)
470 			sysctl_remove_oid(sc->sc_sysctl_cpu[i], 1, 0);
471 
472 	/* NewBus removes the dev.amdtemp.N tree by itself. */
473 
474 	return (0);
475 }
476 
477 static int
478 amdtemp_sysctl(SYSCTL_HANDLER_ARGS)
479 {
480 	device_t dev = (device_t)arg1;
481 	struct amdtemp_softc *sc = device_get_softc(dev);
482 	amdsensor_t sensor = (amdsensor_t)arg2;
483 	int32_t auxtemp[2], temp;
484 	int error;
485 
486 	switch (sensor) {
487 	case CORE0:
488 		auxtemp[0] = sc->sc_gettemp(dev, CORE0_SENSOR0);
489 		auxtemp[1] = sc->sc_gettemp(dev, CORE0_SENSOR1);
490 		temp = imax(auxtemp[0], auxtemp[1]);
491 		break;
492 	case CORE1:
493 		auxtemp[0] = sc->sc_gettemp(dev, CORE1_SENSOR0);
494 		auxtemp[1] = sc->sc_gettemp(dev, CORE1_SENSOR1);
495 		temp = imax(auxtemp[0], auxtemp[1]);
496 		break;
497 	default:
498 		temp = sc->sc_gettemp(dev, sensor);
499 		break;
500 	}
501 	error = sysctl_handle_int(oidp, &temp, 0, req);
502 
503 	return (error);
504 }
505 
506 #define	AMDTEMP_ZERO_C_TO_K	2732
507 
508 static int32_t
509 amdtemp_gettemp0f(device_t dev, amdsensor_t sensor)
510 {
511 	struct amdtemp_softc *sc = device_get_softc(dev);
512 	uint32_t mask, offset, temp;
513 
514 	/* Set Sensor/Core selector. */
515 	temp = pci_read_config(dev, AMDTEMP_THERMTP_STAT, 1);
516 	temp &= ~(AMDTEMP_TTSR_SELCORE | AMDTEMP_TTSR_SELSENSOR);
517 	switch (sensor) {
518 	case CORE0_SENSOR1:
519 		temp |= AMDTEMP_TTSR_SELSENSOR;
520 		/* FALLTHROUGH */
521 	case CORE0_SENSOR0:
522 	case CORE0:
523 		if ((sc->sc_flags & AMDTEMP_FLAG_CS_SWAP) != 0)
524 			temp |= AMDTEMP_TTSR_SELCORE;
525 		break;
526 	case CORE1_SENSOR1:
527 		temp |= AMDTEMP_TTSR_SELSENSOR;
528 		/* FALLTHROUGH */
529 	case CORE1_SENSOR0:
530 	case CORE1:
531 		if ((sc->sc_flags & AMDTEMP_FLAG_CS_SWAP) == 0)
532 			temp |= AMDTEMP_TTSR_SELCORE;
533 		break;
534 	}
535 	pci_write_config(dev, AMDTEMP_THERMTP_STAT, temp, 1);
536 
537 	mask = (sc->sc_flags & AMDTEMP_FLAG_CT_10BIT) != 0 ? 0x3ff : 0x3fc;
538 	offset = (sc->sc_flags & AMDTEMP_FLAG_ALT_OFFSET) != 0 ? 28 : 49;
539 	temp = pci_read_config(dev, AMDTEMP_THERMTP_STAT, 4);
540 	temp = ((temp >> 14) & mask) * 5 / 2;
541 	temp += AMDTEMP_ZERO_C_TO_K + (sc->sc_offset - offset) * 10;
542 
543 	return (temp);
544 }
545 
546 static int32_t
547 amdtemp_gettemp(device_t dev, amdsensor_t sensor)
548 {
549 	struct amdtemp_softc *sc = device_get_softc(dev);
550 	uint32_t temp;
551 
552 	temp = pci_read_config(dev, AMDTEMP_REPTMP_CTRL, 4);
553 	temp = ((temp >> 21) & 0x7ff) * 5 / 4;
554 	temp += AMDTEMP_ZERO_C_TO_K + sc->sc_offset * 10;
555 
556 	return (temp);
557 }
558