xref: /freebsd/sys/dev/amdtemp/amdtemp.c (revision 94cba8034ba53725c225c85e35724f0c2b13cea5)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2008, 2009 Rui Paulo <rpaulo@FreeBSD.org>
5  * Copyright (c) 2009 Norikatsu Shigemura <nork@FreeBSD.org>
6  * Copyright (c) 2009-2012 Jung-uk Kim <jkim@FreeBSD.org>
7  * All rights reserved.
8  * Copyright (c) 2017-2020 Conrad Meyer <cem@FreeBSD.org>. All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
21  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22  * DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
23  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
24  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
25  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
27  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
28  * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Driver for the AMD CPU on-die thermal sensors.
34  * Initially based on the k8temp Linux driver.
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include <sys/param.h>
41 #include <sys/bus.h>
42 #include <sys/conf.h>
43 #include <sys/kernel.h>
44 #include <sys/module.h>
45 #include <sys/sysctl.h>
46 #include <sys/systm.h>
47 
48 #include <machine/cpufunc.h>
49 #include <machine/md_var.h>
50 #include <machine/specialreg.h>
51 
52 #include <dev/pci/pcivar.h>
53 #include <x86/pci_cfgreg.h>
54 
55 #include <dev/amdsmn/amdsmn.h>
56 
57 typedef enum {
58 	CORE0_SENSOR0,
59 	CORE0_SENSOR1,
60 	CORE1_SENSOR0,
61 	CORE1_SENSOR1,
62 	CORE0,
63 	CORE1,
64 	CCD1,
65 	CCD_BASE = CCD1,
66 	CCD2,
67 	CCD3,
68 	CCD4,
69 	CCD5,
70 	CCD6,
71 	CCD7,
72 	CCD8,
73 	CCD_MAX = CCD8,
74 	NUM_CCDS = CCD_MAX - CCD_BASE + 1,
75 } amdsensor_t;
76 
77 struct amdtemp_softc {
78 	int		sc_ncores;
79 	int		sc_ntemps;
80 	int		sc_flags;
81 #define	AMDTEMP_FLAG_CS_SWAP	0x01	/* ThermSenseCoreSel is inverted. */
82 #define	AMDTEMP_FLAG_CT_10BIT	0x02	/* CurTmp is 10-bit wide. */
83 #define	AMDTEMP_FLAG_ALT_OFFSET	0x04	/* CurTmp starts at -28C. */
84 	int32_t		sc_offset;
85 	int32_t		(*sc_gettemp)(device_t, amdsensor_t);
86 	struct sysctl_oid *sc_sysctl_cpu[MAXCPU];
87 	struct intr_config_hook sc_ich;
88 	device_t	sc_smn;
89 };
90 
91 /*
92  * N.B. The numbers in macro names below are significant and represent CPU
93  * family and model numbers.  Do not make up fictitious family or model numbers
94  * when adding support for new devices.
95  */
96 #define	VENDORID_AMD		0x1022
97 #define	DEVICEID_AMD_MISC0F	0x1103
98 #define	DEVICEID_AMD_MISC10	0x1203
99 #define	DEVICEID_AMD_MISC11	0x1303
100 #define	DEVICEID_AMD_MISC14	0x1703
101 #define	DEVICEID_AMD_MISC15	0x1603
102 #define	DEVICEID_AMD_MISC15_M10H	0x1403
103 #define	DEVICEID_AMD_MISC15_M30H	0x141d
104 #define	DEVICEID_AMD_MISC15_M60H_ROOT	0x1576
105 #define	DEVICEID_AMD_MISC16	0x1533
106 #define	DEVICEID_AMD_MISC16_M30H	0x1583
107 #define	DEVICEID_AMD_HOSTB17H_ROOT	0x1450
108 #define	DEVICEID_AMD_HOSTB17H_M10H_ROOT	0x15d0
109 #define	DEVICEID_AMD_HOSTB17H_M30H_ROOT	0x1480	/* Also M70h. */
110 
111 static const struct amdtemp_product {
112 	uint16_t	amdtemp_vendorid;
113 	uint16_t	amdtemp_deviceid;
114 	/*
115 	 * 0xFC register is only valid on the D18F3 PCI device; SMN temp
116 	 * drivers do not attach to that device.
117 	 */
118 	bool		amdtemp_has_cpuid;
119 } amdtemp_products[] = {
120 	{ VENDORID_AMD,	DEVICEID_AMD_MISC0F, true },
121 	{ VENDORID_AMD,	DEVICEID_AMD_MISC10, true },
122 	{ VENDORID_AMD,	DEVICEID_AMD_MISC11, true },
123 	{ VENDORID_AMD,	DEVICEID_AMD_MISC14, true },
124 	{ VENDORID_AMD,	DEVICEID_AMD_MISC15, true },
125 	{ VENDORID_AMD,	DEVICEID_AMD_MISC15_M10H, true },
126 	{ VENDORID_AMD,	DEVICEID_AMD_MISC15_M30H, true },
127 	{ VENDORID_AMD,	DEVICEID_AMD_MISC15_M60H_ROOT, false },
128 	{ VENDORID_AMD,	DEVICEID_AMD_MISC16, true },
129 	{ VENDORID_AMD,	DEVICEID_AMD_MISC16_M30H, true },
130 	{ VENDORID_AMD,	DEVICEID_AMD_HOSTB17H_ROOT, false },
131 	{ VENDORID_AMD,	DEVICEID_AMD_HOSTB17H_M10H_ROOT, false },
132 	{ VENDORID_AMD,	DEVICEID_AMD_HOSTB17H_M30H_ROOT, false },
133 };
134 
135 /*
136  * Reported Temperature Control Register, family 0Fh-15h (some models), 16h.
137  */
138 #define	AMDTEMP_REPTMP_CTRL	0xa4
139 
140 #define	AMDTEMP_REPTMP10H_CURTMP_MASK	0x7ff
141 #define	AMDTEMP_REPTMP10H_CURTMP_SHIFT	21
142 #define	AMDTEMP_REPTMP10H_TJSEL_MASK	0x3
143 #define	AMDTEMP_REPTMP10H_TJSEL_SHIFT	16
144 
145 /*
146  * Reported Temperature, Family 15h, M60+
147  *
148  * Same register bit definitions as other Family 15h CPUs, but access is
149  * indirect via SMN, like Family 17h.
150  */
151 #define	AMDTEMP_15H_M60H_REPTMP_CTRL	0xd8200ca4
152 
153 /*
154  * Reported Temperature, Family 17h
155  *
156  * According to AMD OSRR for 17H, section 4.2.1, bits 31-21 of this register
157  * provide the current temp.  bit 19, when clear, means the temp is reported in
158  * a range 0.."225C" (probable typo for 255C), and when set changes the range
159  * to -49..206C.
160  */
161 #define	AMDTEMP_17H_CUR_TMP		0x59800
162 #define	AMDTEMP_17H_CUR_TMP_RANGE_SEL	(1u << 19)
163 /*
164  * The following register set was discovered experimentally by Ondrej Čerman
165  * and collaborators, but is not (yet) documented in a PPR/OSRR (other than
166  * the M70H PPR SMN memory map showing [0x59800, +0x314] as allocated to
167  * SMU::THM).  It seems plausible and the Linux sensor folks have adopted it.
168  */
169 #define	AMDTEMP_17H_CCD_TMP_BASE	0x59954
170 #define	AMDTEMP_17H_CCD_TMP_VALID	(1u << 11)
171 
172 /*
173  * AMD temperature range adjustment, in deciKelvins (i.e., 49.0 Celsius).
174  */
175 #define	AMDTEMP_CURTMP_RANGE_ADJUST	490
176 
177 /*
178  * Thermaltrip Status Register (Family 0Fh only)
179  */
180 #define	AMDTEMP_THERMTP_STAT	0xe4
181 #define	AMDTEMP_TTSR_SELCORE	0x04
182 #define	AMDTEMP_TTSR_SELSENSOR	0x40
183 
184 /*
185  * DRAM Configuration High Register
186  */
187 #define	AMDTEMP_DRAM_CONF_HIGH	0x94	/* Function 2 */
188 #define	AMDTEMP_DRAM_MODE_DDR3	0x0100
189 
190 /*
191  * CPU Family/Model Register
192  */
193 #define	AMDTEMP_CPUID		0xfc
194 
195 /*
196  * Device methods.
197  */
198 static void 	amdtemp_identify(driver_t *driver, device_t parent);
199 static int	amdtemp_probe(device_t dev);
200 static int	amdtemp_attach(device_t dev);
201 static void	amdtemp_intrhook(void *arg);
202 static int	amdtemp_detach(device_t dev);
203 static int32_t	amdtemp_gettemp0f(device_t dev, amdsensor_t sensor);
204 static int32_t	amdtemp_gettemp(device_t dev, amdsensor_t sensor);
205 static int32_t	amdtemp_gettemp15hm60h(device_t dev, amdsensor_t sensor);
206 static int32_t	amdtemp_gettemp17h(device_t dev, amdsensor_t sensor);
207 static void	amdtemp_probe_ccd_sensors17h(device_t dev, uint32_t model);
208 static int	amdtemp_sysctl(SYSCTL_HANDLER_ARGS);
209 
210 static device_method_t amdtemp_methods[] = {
211 	/* Device interface */
212 	DEVMETHOD(device_identify,	amdtemp_identify),
213 	DEVMETHOD(device_probe,		amdtemp_probe),
214 	DEVMETHOD(device_attach,	amdtemp_attach),
215 	DEVMETHOD(device_detach,	amdtemp_detach),
216 
217 	DEVMETHOD_END
218 };
219 
220 static driver_t amdtemp_driver = {
221 	"amdtemp",
222 	amdtemp_methods,
223 	sizeof(struct amdtemp_softc),
224 };
225 
226 static devclass_t amdtemp_devclass;
227 DRIVER_MODULE(amdtemp, hostb, amdtemp_driver, amdtemp_devclass, NULL, NULL);
228 MODULE_VERSION(amdtemp, 1);
229 MODULE_DEPEND(amdtemp, amdsmn, 1, 1, 1);
230 MODULE_PNP_INFO("U16:vendor;U16:device", pci, amdtemp, amdtemp_products,
231     nitems(amdtemp_products));
232 
233 static bool
234 amdtemp_match(device_t dev, const struct amdtemp_product **product_out)
235 {
236 	int i;
237 	uint16_t vendor, devid;
238 
239 	vendor = pci_get_vendor(dev);
240 	devid = pci_get_device(dev);
241 
242 	for (i = 0; i < nitems(amdtemp_products); i++) {
243 		if (vendor == amdtemp_products[i].amdtemp_vendorid &&
244 		    devid == amdtemp_products[i].amdtemp_deviceid) {
245 			if (product_out != NULL)
246 				*product_out = &amdtemp_products[i];
247 			return (true);
248 		}
249 	}
250 	return (false);
251 }
252 
253 static void
254 amdtemp_identify(driver_t *driver, device_t parent)
255 {
256 	device_t child;
257 
258 	/* Make sure we're not being doubly invoked. */
259 	if (device_find_child(parent, "amdtemp", -1) != NULL)
260 		return;
261 
262 	if (amdtemp_match(parent, NULL)) {
263 		child = device_add_child(parent, "amdtemp", -1);
264 		if (child == NULL)
265 			device_printf(parent, "add amdtemp child failed\n");
266 	}
267 }
268 
269 static int
270 amdtemp_probe(device_t dev)
271 {
272 	uint32_t family, model;
273 
274 	if (resource_disabled("amdtemp", 0))
275 		return (ENXIO);
276 	if (!amdtemp_match(device_get_parent(dev), NULL))
277 		return (ENXIO);
278 
279 	family = CPUID_TO_FAMILY(cpu_id);
280 	model = CPUID_TO_MODEL(cpu_id);
281 
282 	switch (family) {
283 	case 0x0f:
284 		if ((model == 0x04 && (cpu_id & CPUID_STEPPING) == 0) ||
285 		    (model == 0x05 && (cpu_id & CPUID_STEPPING) <= 1))
286 			return (ENXIO);
287 		break;
288 	case 0x10:
289 	case 0x11:
290 	case 0x12:
291 	case 0x14:
292 	case 0x15:
293 	case 0x16:
294 	case 0x17:
295 		break;
296 	default:
297 		return (ENXIO);
298 	}
299 	device_set_desc(dev, "AMD CPU On-Die Thermal Sensors");
300 
301 	return (BUS_PROBE_GENERIC);
302 }
303 
304 static int
305 amdtemp_attach(device_t dev)
306 {
307 	char tn[32];
308 	u_int regs[4];
309 	const struct amdtemp_product *product;
310 	struct amdtemp_softc *sc;
311 	struct sysctl_ctx_list *sysctlctx;
312 	struct sysctl_oid *sysctlnode;
313 	uint32_t cpuid, family, model;
314 	u_int bid;
315 	int erratum319, unit;
316 	bool needsmn;
317 
318 	sc = device_get_softc(dev);
319 	erratum319 = 0;
320 	needsmn = false;
321 
322 	if (!amdtemp_match(device_get_parent(dev), &product))
323 		return (ENXIO);
324 
325 	cpuid = cpu_id;
326 	family = CPUID_TO_FAMILY(cpuid);
327 	model = CPUID_TO_MODEL(cpuid);
328 
329 	/*
330 	 * This checks for the byzantine condition of running a heterogenous
331 	 * revision multi-socket system where the attach thread is potentially
332 	 * probing a remote socket's PCI device.
333 	 *
334 	 * Currently, such scenarios are unsupported on models using the SMN
335 	 * (because on those models, amdtemp(4) attaches to a different PCI
336 	 * device than the one that contains AMDTEMP_CPUID).
337 	 *
338 	 * The ancient 0x0F family of devices only supports this register from
339 	 * models 40h+.
340 	 */
341 	if (product->amdtemp_has_cpuid && (family > 0x0f ||
342 	    (family == 0x0f && model >= 0x40))) {
343 		cpuid = pci_read_config(device_get_parent(dev), AMDTEMP_CPUID,
344 		    4);
345 		family = CPUID_TO_FAMILY(cpuid);
346 		model = CPUID_TO_MODEL(cpuid);
347 	}
348 
349 	switch (family) {
350 	case 0x0f:
351 		/*
352 		 * Thermaltrip Status Register
353 		 *
354 		 * - ThermSenseCoreSel
355 		 *
356 		 * Revision F & G:	0 - Core1, 1 - Core0
357 		 * Other:		0 - Core0, 1 - Core1
358 		 *
359 		 * - CurTmp
360 		 *
361 		 * Revision G:		bits 23-14
362 		 * Other:		bits 23-16
363 		 *
364 		 * XXX According to the BKDG, CurTmp, ThermSenseSel and
365 		 * ThermSenseCoreSel bits were introduced in Revision F
366 		 * but CurTmp seems working fine as early as Revision C.
367 		 * However, it is not clear whether ThermSenseSel and/or
368 		 * ThermSenseCoreSel work in undocumented cases as well.
369 		 * In fact, the Linux driver suggests it may not work but
370 		 * we just assume it does until we find otherwise.
371 		 *
372 		 * XXX According to Linux, CurTmp starts at -28C on
373 		 * Socket AM2 Revision G processors, which is not
374 		 * documented anywhere.
375 		 */
376 		if (model >= 0x40)
377 			sc->sc_flags |= AMDTEMP_FLAG_CS_SWAP;
378 		if (model >= 0x60 && model != 0xc1) {
379 			do_cpuid(0x80000001, regs);
380 			bid = (regs[1] >> 9) & 0x1f;
381 			switch (model) {
382 			case 0x68: /* Socket S1g1 */
383 			case 0x6c:
384 			case 0x7c:
385 				break;
386 			case 0x6b: /* Socket AM2 and ASB1 (2 cores) */
387 				if (bid != 0x0b && bid != 0x0c)
388 					sc->sc_flags |=
389 					    AMDTEMP_FLAG_ALT_OFFSET;
390 				break;
391 			case 0x6f: /* Socket AM2 and ASB1 (1 core) */
392 			case 0x7f:
393 				if (bid != 0x07 && bid != 0x09 &&
394 				    bid != 0x0c)
395 					sc->sc_flags |=
396 					    AMDTEMP_FLAG_ALT_OFFSET;
397 				break;
398 			default:
399 				sc->sc_flags |= AMDTEMP_FLAG_ALT_OFFSET;
400 			}
401 			sc->sc_flags |= AMDTEMP_FLAG_CT_10BIT;
402 		}
403 
404 		/*
405 		 * There are two sensors per core.
406 		 */
407 		sc->sc_ntemps = 2;
408 
409 		sc->sc_gettemp = amdtemp_gettemp0f;
410 		break;
411 	case 0x10:
412 		/*
413 		 * Erratum 319 Inaccurate Temperature Measurement
414 		 *
415 		 * http://support.amd.com/us/Processor_TechDocs/41322.pdf
416 		 */
417 		do_cpuid(0x80000001, regs);
418 		switch ((regs[1] >> 28) & 0xf) {
419 		case 0:	/* Socket F */
420 			erratum319 = 1;
421 			break;
422 		case 1:	/* Socket AM2+ or AM3 */
423 			if ((pci_cfgregread(pci_get_bus(dev),
424 			    pci_get_slot(dev), 2, AMDTEMP_DRAM_CONF_HIGH, 2) &
425 			    AMDTEMP_DRAM_MODE_DDR3) != 0 || model > 0x04 ||
426 			    (model == 0x04 && (cpuid & CPUID_STEPPING) >= 3))
427 				break;
428 			/* XXX 00100F42h (RB-C2) exists in both formats. */
429 			erratum319 = 1;
430 			break;
431 		}
432 		/* FALLTHROUGH */
433 	case 0x11:
434 	case 0x12:
435 	case 0x14:
436 	case 0x15:
437 	case 0x16:
438 		sc->sc_ntemps = 1;
439 		/*
440 		 * Some later (60h+) models of family 15h use a similar SMN
441 		 * network as family 17h.  (However, the register index differs
442 		 * from 17h and the decoding matches other 10h-15h models,
443 		 * which differ from 17h.)
444 		 */
445 		if (family == 0x15 && model >= 0x60) {
446 			sc->sc_gettemp = amdtemp_gettemp15hm60h;
447 			needsmn = true;
448 		} else
449 			sc->sc_gettemp = amdtemp_gettemp;
450 		break;
451 	case 0x17:
452 		sc->sc_ntemps = 1;
453 		sc->sc_gettemp = amdtemp_gettemp17h;
454 		needsmn = true;
455 		break;
456 	default:
457 		device_printf(dev, "Bogus family 0x%x\n", family);
458 		return (ENXIO);
459 	}
460 
461 	if (needsmn) {
462 		sc->sc_smn = device_find_child(
463 		    device_get_parent(dev), "amdsmn", -1);
464 		if (sc->sc_smn == NULL) {
465 			if (bootverbose)
466 				device_printf(dev, "No SMN device found\n");
467 			return (ENXIO);
468 		}
469 	}
470 
471 	/* Find number of cores per package. */
472 	sc->sc_ncores = (amd_feature2 & AMDID2_CMP) != 0 ?
473 	    (cpu_procinfo2 & AMDID_CMP_CORES) + 1 : 1;
474 	if (sc->sc_ncores > MAXCPU)
475 		return (ENXIO);
476 
477 	if (erratum319)
478 		device_printf(dev,
479 		    "Erratum 319: temperature measurement may be inaccurate\n");
480 	if (bootverbose)
481 		device_printf(dev, "Found %d cores and %d sensors.\n",
482 		    sc->sc_ncores,
483 		    sc->sc_ntemps > 1 ? sc->sc_ntemps * sc->sc_ncores : 1);
484 
485 	/*
486 	 * dev.amdtemp.N tree.
487 	 */
488 	unit = device_get_unit(dev);
489 	snprintf(tn, sizeof(tn), "dev.amdtemp.%d.sensor_offset", unit);
490 	TUNABLE_INT_FETCH(tn, &sc->sc_offset);
491 
492 	sysctlctx = device_get_sysctl_ctx(dev);
493 	SYSCTL_ADD_INT(sysctlctx,
494 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO,
495 	    "sensor_offset", CTLFLAG_RW, &sc->sc_offset, 0,
496 	    "Temperature sensor offset");
497 	sysctlnode = SYSCTL_ADD_NODE(sysctlctx,
498 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO,
499 	    "core0", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Core 0");
500 
501 	SYSCTL_ADD_PROC(sysctlctx,
502 	    SYSCTL_CHILDREN(sysctlnode),
503 	    OID_AUTO, "sensor0",
504 	    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
505 	    dev, CORE0_SENSOR0, amdtemp_sysctl, "IK",
506 	    "Core 0 / Sensor 0 temperature");
507 
508 	if (family == 0x17)
509 		amdtemp_probe_ccd_sensors17h(dev, model);
510 	else if (sc->sc_ntemps > 1) {
511 		SYSCTL_ADD_PROC(sysctlctx,
512 		    SYSCTL_CHILDREN(sysctlnode),
513 		    OID_AUTO, "sensor1",
514 		    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
515 		    dev, CORE0_SENSOR1, amdtemp_sysctl, "IK",
516 		    "Core 0 / Sensor 1 temperature");
517 
518 		if (sc->sc_ncores > 1) {
519 			sysctlnode = SYSCTL_ADD_NODE(sysctlctx,
520 			    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
521 			    OID_AUTO, "core1", CTLFLAG_RD | CTLFLAG_MPSAFE,
522 			    0, "Core 1");
523 
524 			SYSCTL_ADD_PROC(sysctlctx,
525 			    SYSCTL_CHILDREN(sysctlnode),
526 			    OID_AUTO, "sensor0",
527 			    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
528 			    dev, CORE1_SENSOR0, amdtemp_sysctl, "IK",
529 			    "Core 1 / Sensor 0 temperature");
530 
531 			SYSCTL_ADD_PROC(sysctlctx,
532 			    SYSCTL_CHILDREN(sysctlnode),
533 			    OID_AUTO, "sensor1",
534 			    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
535 			    dev, CORE1_SENSOR1, amdtemp_sysctl, "IK",
536 			    "Core 1 / Sensor 1 temperature");
537 		}
538 	}
539 
540 	/*
541 	 * Try to create dev.cpu sysctl entries and setup intrhook function.
542 	 * This is needed because the cpu driver may be loaded late on boot,
543 	 * after us.
544 	 */
545 	amdtemp_intrhook(dev);
546 	sc->sc_ich.ich_func = amdtemp_intrhook;
547 	sc->sc_ich.ich_arg = dev;
548 	if (config_intrhook_establish(&sc->sc_ich) != 0) {
549 		device_printf(dev, "config_intrhook_establish failed!\n");
550 		return (ENXIO);
551 	}
552 
553 	return (0);
554 }
555 
556 void
557 amdtemp_intrhook(void *arg)
558 {
559 	struct amdtemp_softc *sc;
560 	struct sysctl_ctx_list *sysctlctx;
561 	device_t dev = (device_t)arg;
562 	device_t acpi, cpu, nexus;
563 	amdsensor_t sensor;
564 	int i;
565 
566 	sc = device_get_softc(dev);
567 
568 	/*
569 	 * dev.cpu.N.temperature.
570 	 */
571 	nexus = device_find_child(root_bus, "nexus", 0);
572 	acpi = device_find_child(nexus, "acpi", 0);
573 
574 	for (i = 0; i < sc->sc_ncores; i++) {
575 		if (sc->sc_sysctl_cpu[i] != NULL)
576 			continue;
577 		cpu = device_find_child(acpi, "cpu",
578 		    device_get_unit(dev) * sc->sc_ncores + i);
579 		if (cpu != NULL) {
580 			sysctlctx = device_get_sysctl_ctx(cpu);
581 
582 			sensor = sc->sc_ntemps > 1 ?
583 			    (i == 0 ? CORE0 : CORE1) : CORE0_SENSOR0;
584 			sc->sc_sysctl_cpu[i] = SYSCTL_ADD_PROC(sysctlctx,
585 			    SYSCTL_CHILDREN(device_get_sysctl_tree(cpu)),
586 			    OID_AUTO, "temperature",
587 			    CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
588 			    dev, sensor, amdtemp_sysctl, "IK",
589 			    "Current temparature");
590 		}
591 	}
592 	if (sc->sc_ich.ich_arg != NULL)
593 		config_intrhook_disestablish(&sc->sc_ich);
594 }
595 
596 int
597 amdtemp_detach(device_t dev)
598 {
599 	struct amdtemp_softc *sc = device_get_softc(dev);
600 	int i;
601 
602 	for (i = 0; i < sc->sc_ncores; i++)
603 		if (sc->sc_sysctl_cpu[i] != NULL)
604 			sysctl_remove_oid(sc->sc_sysctl_cpu[i], 1, 0);
605 
606 	/* NewBus removes the dev.amdtemp.N tree by itself. */
607 
608 	return (0);
609 }
610 
611 static int
612 amdtemp_sysctl(SYSCTL_HANDLER_ARGS)
613 {
614 	device_t dev = (device_t)arg1;
615 	struct amdtemp_softc *sc = device_get_softc(dev);
616 	amdsensor_t sensor = (amdsensor_t)arg2;
617 	int32_t auxtemp[2], temp;
618 	int error;
619 
620 	switch (sensor) {
621 	case CORE0:
622 		auxtemp[0] = sc->sc_gettemp(dev, CORE0_SENSOR0);
623 		auxtemp[1] = sc->sc_gettemp(dev, CORE0_SENSOR1);
624 		temp = imax(auxtemp[0], auxtemp[1]);
625 		break;
626 	case CORE1:
627 		auxtemp[0] = sc->sc_gettemp(dev, CORE1_SENSOR0);
628 		auxtemp[1] = sc->sc_gettemp(dev, CORE1_SENSOR1);
629 		temp = imax(auxtemp[0], auxtemp[1]);
630 		break;
631 	default:
632 		temp = sc->sc_gettemp(dev, sensor);
633 		break;
634 	}
635 	error = sysctl_handle_int(oidp, &temp, 0, req);
636 
637 	return (error);
638 }
639 
640 #define	AMDTEMP_ZERO_C_TO_K	2731
641 
642 static int32_t
643 amdtemp_gettemp0f(device_t dev, amdsensor_t sensor)
644 {
645 	struct amdtemp_softc *sc = device_get_softc(dev);
646 	uint32_t mask, offset, temp;
647 
648 	/* Set Sensor/Core selector. */
649 	temp = pci_read_config(dev, AMDTEMP_THERMTP_STAT, 1);
650 	temp &= ~(AMDTEMP_TTSR_SELCORE | AMDTEMP_TTSR_SELSENSOR);
651 	switch (sensor) {
652 	case CORE0_SENSOR1:
653 		temp |= AMDTEMP_TTSR_SELSENSOR;
654 		/* FALLTHROUGH */
655 	case CORE0_SENSOR0:
656 	case CORE0:
657 		if ((sc->sc_flags & AMDTEMP_FLAG_CS_SWAP) != 0)
658 			temp |= AMDTEMP_TTSR_SELCORE;
659 		break;
660 	case CORE1_SENSOR1:
661 		temp |= AMDTEMP_TTSR_SELSENSOR;
662 		/* FALLTHROUGH */
663 	case CORE1_SENSOR0:
664 	case CORE1:
665 		if ((sc->sc_flags & AMDTEMP_FLAG_CS_SWAP) == 0)
666 			temp |= AMDTEMP_TTSR_SELCORE;
667 		break;
668 	default:
669 		__assert_unreachable();
670 	}
671 	pci_write_config(dev, AMDTEMP_THERMTP_STAT, temp, 1);
672 
673 	mask = (sc->sc_flags & AMDTEMP_FLAG_CT_10BIT) != 0 ? 0x3ff : 0x3fc;
674 	offset = (sc->sc_flags & AMDTEMP_FLAG_ALT_OFFSET) != 0 ? 28 : 49;
675 	temp = pci_read_config(dev, AMDTEMP_THERMTP_STAT, 4);
676 	temp = ((temp >> 14) & mask) * 5 / 2;
677 	temp += AMDTEMP_ZERO_C_TO_K + (sc->sc_offset - offset) * 10;
678 
679 	return (temp);
680 }
681 
682 static uint32_t
683 amdtemp_decode_fam10h_to_17h(int32_t sc_offset, uint32_t val, bool minus49)
684 {
685 	uint32_t temp;
686 
687 	/* Convert raw register subfield units (0.125C) to units of 0.1C. */
688 	temp = (val & AMDTEMP_REPTMP10H_CURTMP_MASK) * 5 / 4;
689 
690 	if (minus49)
691 		temp -= AMDTEMP_CURTMP_RANGE_ADJUST;
692 
693 	temp += AMDTEMP_ZERO_C_TO_K + sc_offset * 10;
694 	return (temp);
695 }
696 
697 static uint32_t
698 amdtemp_decode_fam10h_to_16h(int32_t sc_offset, uint32_t val)
699 {
700 	bool minus49;
701 
702 	/*
703 	 * On Family 15h and higher, if CurTmpTjSel is 11b, the range is
704 	 * adjusted down by 49.0 degrees Celsius.  (This adjustment is not
705 	 * documented in BKDGs prior to family 15h model 00h.)
706 	 */
707 	minus49 = (CPUID_TO_FAMILY(cpu_id) >= 0x15 &&
708 	    ((val >> AMDTEMP_REPTMP10H_TJSEL_SHIFT) &
709 	    AMDTEMP_REPTMP10H_TJSEL_MASK) == 0x3);
710 
711 	return (amdtemp_decode_fam10h_to_17h(sc_offset,
712 	    val >> AMDTEMP_REPTMP10H_CURTMP_SHIFT, minus49));
713 }
714 
715 static uint32_t
716 amdtemp_decode_fam17h_tctl(int32_t sc_offset, uint32_t val)
717 {
718 	bool minus49;
719 
720 	minus49 = ((val & AMDTEMP_17H_CUR_TMP_RANGE_SEL) != 0);
721 	return (amdtemp_decode_fam10h_to_17h(sc_offset,
722 	    val >> AMDTEMP_REPTMP10H_CURTMP_SHIFT, minus49));
723 }
724 
725 static int32_t
726 amdtemp_gettemp(device_t dev, amdsensor_t sensor)
727 {
728 	struct amdtemp_softc *sc = device_get_softc(dev);
729 	uint32_t temp;
730 
731 	temp = pci_read_config(dev, AMDTEMP_REPTMP_CTRL, 4);
732 	return (amdtemp_decode_fam10h_to_16h(sc->sc_offset, temp));
733 }
734 
735 static int32_t
736 amdtemp_gettemp15hm60h(device_t dev, amdsensor_t sensor)
737 {
738 	struct amdtemp_softc *sc = device_get_softc(dev);
739 	uint32_t val;
740 	int error;
741 
742 	error = amdsmn_read(sc->sc_smn, AMDTEMP_15H_M60H_REPTMP_CTRL, &val);
743 	KASSERT(error == 0, ("amdsmn_read"));
744 	return (amdtemp_decode_fam10h_to_16h(sc->sc_offset, val));
745 }
746 
747 static int32_t
748 amdtemp_gettemp17h(device_t dev, amdsensor_t sensor)
749 {
750 	struct amdtemp_softc *sc = device_get_softc(dev);
751 	uint32_t val;
752 	int error;
753 
754 	switch (sensor) {
755 	case CORE0_SENSOR0:
756 		/* Tctl */
757 		error = amdsmn_read(sc->sc_smn, AMDTEMP_17H_CUR_TMP, &val);
758 		KASSERT(error == 0, ("amdsmn_read"));
759 		return (amdtemp_decode_fam17h_tctl(sc->sc_offset, val));
760 	case CCD_BASE ... CCD_MAX:
761 		/* Tccd<N> */
762 		error = amdsmn_read(sc->sc_smn, AMDTEMP_17H_CCD_TMP_BASE +
763 		    (((int)sensor - CCD_BASE) * sizeof(val)), &val);
764 		KASSERT(error == 0, ("amdsmn_read2"));
765 		KASSERT((val & AMDTEMP_17H_CCD_TMP_VALID) != 0,
766 		    ("sensor %d: not valid", (int)sensor));
767 		return (amdtemp_decode_fam10h_to_17h(sc->sc_offset, val, true));
768 	default:
769 		__assert_unreachable();
770 	}
771 }
772 
773 static void
774 amdtemp_probe_ccd_sensors17h(device_t dev, uint32_t model)
775 {
776 	char sensor_name[16], sensor_descr[32];
777 	struct amdtemp_softc *sc;
778 	uint32_t maxreg, i, val;
779 	int error;
780 
781 	switch (model) {
782 	case 0x00 ... 0x1f: /* Zen1, Zen+ */
783 		maxreg = 4;
784 		break;
785 	case 0x30 ... 0x3f: /* Zen2 TR/Epyc */
786 	case 0x70 ... 0x7f: /* Zen2 Ryzen */
787 		maxreg = 8;
788 		_Static_assert((int)NUM_CCDS >= 8, "");
789 		break;
790 	default:
791 		device_printf(dev,
792 		    "Unrecognized Family 17h Model: %02xh\n", model);
793 		return;
794 	}
795 
796 	sc = device_get_softc(dev);
797 	for (i = 0; i < maxreg; i++) {
798 		error = amdsmn_read(sc->sc_smn, AMDTEMP_17H_CCD_TMP_BASE +
799 		    (i * sizeof(val)), &val);
800 		if (error != 0)
801 			continue;
802 		if ((val & AMDTEMP_17H_CCD_TMP_VALID) == 0)
803 			continue;
804 
805 		snprintf(sensor_name, sizeof(sensor_name), "ccd%u", i);
806 		snprintf(sensor_descr, sizeof(sensor_descr),
807 		    "CCD %u temperature (Tccd%u)", i, i);
808 
809 		SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev),
810 		    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO,
811 		    sensor_name, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
812 		    dev, CCD_BASE + i, amdtemp_sysctl, "IK", sensor_descr);
813 	}
814 }
815