Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * k10temp.c - AMD Family 10h/11h/12h/14h/15h/16h/17h
4 * processor hardware monitoring
5 *
6 * Copyright (c) 2009 Clemens Ladisch <clemens@ladisch.de>
7 * Copyright (c) 2020 Guenter Roeck <linux@roeck-us.net>
8 *
9 * Implementation notes:
10 * - CCD register address information as well as the calculation to
11 * convert raw register values is from https://github.com/ocerman/zenpower.
12 * The information is not confirmed from chip datasheets, but experiments
13 * suggest that it provides reasonable temperature values.
14 */
15
16#include <linux/bitops.h>
17#include <linux/err.h>
18#include <linux/hwmon.h>
19#include <linux/init.h>
20#include <linux/module.h>
21#include <linux/pci.h>
22#include <linux/pci_ids.h>
23#include <asm/amd_nb.h>
24#include <asm/processor.h>
25
26MODULE_DESCRIPTION("AMD Family 10h+ CPU core temperature monitor");
27MODULE_AUTHOR("Clemens Ladisch <clemens@ladisch.de>");
28MODULE_LICENSE("GPL");
29
30static bool force;
31module_param(force, bool, 0444);
32MODULE_PARM_DESC(force, "force loading on processors with erratum 319");
33
34/* Provide lock for writing to NB_SMU_IND_ADDR */
35static DEFINE_MUTEX(nb_smu_ind_mutex);
36
37#ifndef PCI_DEVICE_ID_AMD_15H_M70H_NB_F3
38#define PCI_DEVICE_ID_AMD_15H_M70H_NB_F3 0x15b3
39#endif
40
41/* CPUID function 0x80000001, ebx */
42#define CPUID_PKGTYPE_MASK GENMASK(31, 28)
43#define CPUID_PKGTYPE_F 0x00000000
44#define CPUID_PKGTYPE_AM2R2_AM3 0x10000000
45
46/* DRAM controller (PCI function 2) */
47#define REG_DCT0_CONFIG_HIGH 0x094
48#define DDR3_MODE BIT(8)
49
50/* miscellaneous (PCI function 3) */
51#define REG_HARDWARE_THERMAL_CONTROL 0x64
52#define HTC_ENABLE BIT(0)
53
54#define REG_REPORTED_TEMPERATURE 0xa4
55
56#define REG_NORTHBRIDGE_CAPABILITIES 0xe8
57#define NB_CAP_HTC BIT(10)
58
59/*
60 * For F15h M60h and M70h, REG_HARDWARE_THERMAL_CONTROL
61 * and REG_REPORTED_TEMPERATURE have been moved to
62 * D0F0xBC_xD820_0C64 [Hardware Temperature Control]
63 * D0F0xBC_xD820_0CA4 [Reported Temperature Control]
64 */
65#define F15H_M60H_HARDWARE_TEMP_CTRL_OFFSET 0xd8200c64
66#define F15H_M60H_REPORTED_TEMP_CTRL_OFFSET 0xd8200ca4
67
68/* Common for Zen CPU families (Family 17h and 18h and 19h and 1Ah) */
69#define ZEN_REPORTED_TEMP_CTRL_BASE 0x00059800
70
71#define ZEN_CCD_TEMP(offset, x) (ZEN_REPORTED_TEMP_CTRL_BASE + \
72 (offset) + ((x) * 4))
73#define ZEN_CCD_TEMP_VALID BIT(11)
74#define ZEN_CCD_TEMP_MASK GENMASK(10, 0)
75
76#define ZEN_CUR_TEMP_SHIFT 21
77#define ZEN_CUR_TEMP_RANGE_SEL_MASK BIT(19)
78#define ZEN_CUR_TEMP_TJ_SEL_MASK GENMASK(17, 16)
79
80/*
81 * AMD's Industrial processor 3255 supports temperature from -40 deg to 105 deg Celsius.
82 * Use the model name to identify 3255 CPUs and set a flag to display negative temperature.
83 * Do not round off to zero for negative Tctl or Tdie values if the flag is set
84 */
85#define AMD_I3255_STR "3255"
86
87struct k10temp_data {
88 struct pci_dev *pdev;
89 void (*read_htcreg)(struct pci_dev *pdev, u32 *regval);
90 void (*read_tempreg)(struct pci_dev *pdev, u32 *regval);
91 int temp_offset;
92 u32 temp_adjust_mask;
93 u32 show_temp;
94 bool is_zen;
95 u32 ccd_offset;
96 bool disp_negative;
97};
98
99#define TCTL_BIT 0
100#define TDIE_BIT 1
101#define TCCD_BIT(x) ((x) + 2)
102
103#define HAVE_TEMP(d, channel) ((d)->show_temp & BIT(channel))
104#define HAVE_TDIE(d) HAVE_TEMP(d, TDIE_BIT)
105
106struct tctl_offset {
107 u8 model;
108 char const *id;
109 int offset;
110};
111
112static const struct tctl_offset tctl_offset_table[] = {
113 { 0x17, "AMD Ryzen 5 1600X", 20000 },
114 { 0x17, "AMD Ryzen 7 1700X", 20000 },
115 { 0x17, "AMD Ryzen 7 1800X", 20000 },
116 { 0x17, "AMD Ryzen 7 2700X", 10000 },
117 { 0x17, "AMD Ryzen Threadripper 19", 27000 }, /* 19{00,20,50}X */
118 { 0x17, "AMD Ryzen Threadripper 29", 27000 }, /* 29{20,50,70,90}[W]X */
119};
120
121static void read_htcreg_pci(struct pci_dev *pdev, u32 *regval)
122{
123 pci_read_config_dword(pdev, REG_HARDWARE_THERMAL_CONTROL, regval);
124}
125
126static void read_tempreg_pci(struct pci_dev *pdev, u32 *regval)
127{
128 pci_read_config_dword(pdev, REG_REPORTED_TEMPERATURE, regval);
129}
130
131static void amd_nb_index_read(struct pci_dev *pdev, unsigned int devfn,
132 unsigned int base, int offset, u32 *val)
133{
134 mutex_lock(&nb_smu_ind_mutex);
135 pci_bus_write_config_dword(pdev->bus, devfn,
136 base, offset);
137 pci_bus_read_config_dword(pdev->bus, devfn,
138 base + 4, val);
139 mutex_unlock(&nb_smu_ind_mutex);
140}
141
142static void read_htcreg_nb_f15(struct pci_dev *pdev, u32 *regval)
143{
144 amd_nb_index_read(pdev, PCI_DEVFN(0, 0), 0xb8,
145 F15H_M60H_HARDWARE_TEMP_CTRL_OFFSET, regval);
146}
147
148static void read_tempreg_nb_f15(struct pci_dev *pdev, u32 *regval)
149{
150 amd_nb_index_read(pdev, PCI_DEVFN(0, 0), 0xb8,
151 F15H_M60H_REPORTED_TEMP_CTRL_OFFSET, regval);
152}
153
154static void read_tempreg_nb_zen(struct pci_dev *pdev, u32 *regval)
155{
156 amd_smn_read(amd_pci_dev_to_node_id(pdev),
157 ZEN_REPORTED_TEMP_CTRL_BASE, regval);
158}
159
160static long get_raw_temp(struct k10temp_data *data)
161{
162 u32 regval;
163 long temp;
164
165 data->read_tempreg(data->pdev, ®val);
166 temp = (regval >> ZEN_CUR_TEMP_SHIFT) * 125;
167 if ((regval & data->temp_adjust_mask) ||
168 (regval & ZEN_CUR_TEMP_TJ_SEL_MASK) == ZEN_CUR_TEMP_TJ_SEL_MASK)
169 temp -= 49000;
170 return temp;
171}
172
173static const char *k10temp_temp_label[] = {
174 "Tctl",
175 "Tdie",
176 "Tccd1",
177 "Tccd2",
178 "Tccd3",
179 "Tccd4",
180 "Tccd5",
181 "Tccd6",
182 "Tccd7",
183 "Tccd8",
184 "Tccd9",
185 "Tccd10",
186 "Tccd11",
187 "Tccd12",
188};
189
190static int k10temp_read_labels(struct device *dev,
191 enum hwmon_sensor_types type,
192 u32 attr, int channel, const char **str)
193{
194 switch (type) {
195 case hwmon_temp:
196 *str = k10temp_temp_label[channel];
197 break;
198 default:
199 return -EOPNOTSUPP;
200 }
201 return 0;
202}
203
204static int k10temp_read_temp(struct device *dev, u32 attr, int channel,
205 long *val)
206{
207 struct k10temp_data *data = dev_get_drvdata(dev);
208 u32 regval;
209
210 switch (attr) {
211 case hwmon_temp_input:
212 switch (channel) {
213 case 0: /* Tctl */
214 *val = get_raw_temp(data);
215 if (*val < 0 && !data->disp_negative)
216 *val = 0;
217 break;
218 case 1: /* Tdie */
219 *val = get_raw_temp(data) - data->temp_offset;
220 if (*val < 0 && !data->disp_negative)
221 *val = 0;
222 break;
223 case 2 ... 13: /* Tccd{1-12} */
224 amd_smn_read(amd_pci_dev_to_node_id(data->pdev),
225 ZEN_CCD_TEMP(data->ccd_offset, channel - 2),
226 ®val);
227 *val = (regval & ZEN_CCD_TEMP_MASK) * 125 - 49000;
228 break;
229 default:
230 return -EOPNOTSUPP;
231 }
232 break;
233 case hwmon_temp_max:
234 *val = 70 * 1000;
235 break;
236 case hwmon_temp_crit:
237 data->read_htcreg(data->pdev, ®val);
238 *val = ((regval >> 16) & 0x7f) * 500 + 52000;
239 break;
240 case hwmon_temp_crit_hyst:
241 data->read_htcreg(data->pdev, ®val);
242 *val = (((regval >> 16) & 0x7f)
243 - ((regval >> 24) & 0xf)) * 500 + 52000;
244 break;
245 default:
246 return -EOPNOTSUPP;
247 }
248 return 0;
249}
250
251static int k10temp_read(struct device *dev, enum hwmon_sensor_types type,
252 u32 attr, int channel, long *val)
253{
254 switch (type) {
255 case hwmon_temp:
256 return k10temp_read_temp(dev, attr, channel, val);
257 default:
258 return -EOPNOTSUPP;
259 }
260}
261
262static umode_t k10temp_is_visible(const void *_data,
263 enum hwmon_sensor_types type,
264 u32 attr, int channel)
265{
266 const struct k10temp_data *data = _data;
267 struct pci_dev *pdev = data->pdev;
268 u32 reg;
269
270 switch (type) {
271 case hwmon_temp:
272 switch (attr) {
273 case hwmon_temp_input:
274 if (!HAVE_TEMP(data, channel))
275 return 0;
276 break;
277 case hwmon_temp_max:
278 if (channel || data->is_zen)
279 return 0;
280 break;
281 case hwmon_temp_crit:
282 case hwmon_temp_crit_hyst:
283 if (channel || !data->read_htcreg)
284 return 0;
285
286 pci_read_config_dword(pdev,
287 REG_NORTHBRIDGE_CAPABILITIES,
288 ®);
289 if (!(reg & NB_CAP_HTC))
290 return 0;
291
292 data->read_htcreg(data->pdev, ®);
293 if (!(reg & HTC_ENABLE))
294 return 0;
295 break;
296 case hwmon_temp_label:
297 /* Show temperature labels only on Zen CPUs */
298 if (!data->is_zen || !HAVE_TEMP(data, channel))
299 return 0;
300 break;
301 default:
302 return 0;
303 }
304 break;
305 default:
306 return 0;
307 }
308 return 0444;
309}
310
311static bool has_erratum_319(struct pci_dev *pdev)
312{
313 u32 pkg_type, reg_dram_cfg;
314
315 if (boot_cpu_data.x86 != 0x10)
316 return false;
317
318 /*
319 * Erratum 319: The thermal sensor of Socket F/AM2+ processors
320 * may be unreliable.
321 */
322 pkg_type = cpuid_ebx(0x80000001) & CPUID_PKGTYPE_MASK;
323 if (pkg_type == CPUID_PKGTYPE_F)
324 return true;
325 if (pkg_type != CPUID_PKGTYPE_AM2R2_AM3)
326 return false;
327
328 /* DDR3 memory implies socket AM3, which is good */
329 pci_bus_read_config_dword(pdev->bus,
330 PCI_DEVFN(PCI_SLOT(pdev->devfn), 2),
331 REG_DCT0_CONFIG_HIGH, ®_dram_cfg);
332 if (reg_dram_cfg & DDR3_MODE)
333 return false;
334
335 /*
336 * Unfortunately it is possible to run a socket AM3 CPU with DDR2
337 * memory. We blacklist all the cores which do exist in socket AM2+
338 * format. It still isn't perfect, as RB-C2 cores exist in both AM2+
339 * and AM3 formats, but that's the best we can do.
340 */
341 return boot_cpu_data.x86_model < 4 ||
342 (boot_cpu_data.x86_model == 4 && boot_cpu_data.x86_stepping <= 2);
343}
344
345static const struct hwmon_channel_info * const k10temp_info[] = {
346 HWMON_CHANNEL_INFO(temp,
347 HWMON_T_INPUT | HWMON_T_MAX |
348 HWMON_T_CRIT | HWMON_T_CRIT_HYST |
349 HWMON_T_LABEL,
350 HWMON_T_INPUT | HWMON_T_LABEL,
351 HWMON_T_INPUT | HWMON_T_LABEL,
352 HWMON_T_INPUT | HWMON_T_LABEL,
353 HWMON_T_INPUT | HWMON_T_LABEL,
354 HWMON_T_INPUT | HWMON_T_LABEL,
355 HWMON_T_INPUT | HWMON_T_LABEL,
356 HWMON_T_INPUT | HWMON_T_LABEL,
357 HWMON_T_INPUT | HWMON_T_LABEL,
358 HWMON_T_INPUT | HWMON_T_LABEL,
359 HWMON_T_INPUT | HWMON_T_LABEL,
360 HWMON_T_INPUT | HWMON_T_LABEL,
361 HWMON_T_INPUT | HWMON_T_LABEL,
362 HWMON_T_INPUT | HWMON_T_LABEL),
363 NULL
364};
365
366static const struct hwmon_ops k10temp_hwmon_ops = {
367 .is_visible = k10temp_is_visible,
368 .read = k10temp_read,
369 .read_string = k10temp_read_labels,
370};
371
372static const struct hwmon_chip_info k10temp_chip_info = {
373 .ops = &k10temp_hwmon_ops,
374 .info = k10temp_info,
375};
376
377static void k10temp_get_ccd_support(struct pci_dev *pdev,
378 struct k10temp_data *data, int limit)
379{
380 u32 regval;
381 int i;
382
383 for (i = 0; i < limit; i++) {
384 amd_smn_read(amd_pci_dev_to_node_id(pdev),
385 ZEN_CCD_TEMP(data->ccd_offset, i), ®val);
386 if (regval & ZEN_CCD_TEMP_VALID)
387 data->show_temp |= BIT(TCCD_BIT(i));
388 }
389}
390
391static int k10temp_probe(struct pci_dev *pdev, const struct pci_device_id *id)
392{
393 int unreliable = has_erratum_319(pdev);
394 struct device *dev = &pdev->dev;
395 struct k10temp_data *data;
396 struct device *hwmon_dev;
397 int i;
398
399 if (unreliable) {
400 if (!force) {
401 dev_err(dev,
402 "unreliable CPU thermal sensor; monitoring disabled\n");
403 return -ENODEV;
404 }
405 dev_warn(dev,
406 "unreliable CPU thermal sensor; check erratum 319\n");
407 }
408
409 data = devm_kzalloc(dev, sizeof(*data), GFP_KERNEL);
410 if (!data)
411 return -ENOMEM;
412
413 data->pdev = pdev;
414 data->show_temp |= BIT(TCTL_BIT); /* Always show Tctl */
415
416 if (boot_cpu_data.x86 == 0x17 &&
417 strstr(boot_cpu_data.x86_model_id, AMD_I3255_STR)) {
418 data->disp_negative = true;
419 }
420
421 if (boot_cpu_data.x86 == 0x15 &&
422 ((boot_cpu_data.x86_model & 0xf0) == 0x60 ||
423 (boot_cpu_data.x86_model & 0xf0) == 0x70)) {
424 data->read_htcreg = read_htcreg_nb_f15;
425 data->read_tempreg = read_tempreg_nb_f15;
426 } else if (boot_cpu_data.x86 == 0x17 || boot_cpu_data.x86 == 0x18) {
427 data->temp_adjust_mask = ZEN_CUR_TEMP_RANGE_SEL_MASK;
428 data->read_tempreg = read_tempreg_nb_zen;
429 data->is_zen = true;
430
431 switch (boot_cpu_data.x86_model) {
432 case 0x1: /* Zen */
433 case 0x8: /* Zen+ */
434 case 0x11: /* Zen APU */
435 case 0x18: /* Zen+ APU */
436 data->ccd_offset = 0x154;
437 k10temp_get_ccd_support(pdev, data, 4);
438 break;
439 case 0x31: /* Zen2 Threadripper */
440 case 0x60: /* Renoir */
441 case 0x68: /* Lucienne */
442 case 0x71: /* Zen2 */
443 data->ccd_offset = 0x154;
444 k10temp_get_ccd_support(pdev, data, 8);
445 break;
446 case 0xa0 ... 0xaf:
447 data->ccd_offset = 0x300;
448 k10temp_get_ccd_support(pdev, data, 8);
449 break;
450 }
451 } else if (boot_cpu_data.x86 == 0x19) {
452 data->temp_adjust_mask = ZEN_CUR_TEMP_RANGE_SEL_MASK;
453 data->read_tempreg = read_tempreg_nb_zen;
454 data->is_zen = true;
455
456 switch (boot_cpu_data.x86_model) {
457 case 0x0 ... 0x1: /* Zen3 SP3/TR */
458 case 0x8: /* Zen3 TR Chagall */
459 case 0x21: /* Zen3 Ryzen Desktop */
460 case 0x50 ... 0x5f: /* Green Sardine */
461 data->ccd_offset = 0x154;
462 k10temp_get_ccd_support(pdev, data, 8);
463 break;
464 case 0x40 ... 0x4f: /* Yellow Carp */
465 data->ccd_offset = 0x300;
466 k10temp_get_ccd_support(pdev, data, 8);
467 break;
468 case 0x60 ... 0x6f:
469 case 0x70 ... 0x7f:
470 data->ccd_offset = 0x308;
471 k10temp_get_ccd_support(pdev, data, 8);
472 break;
473 case 0x10 ... 0x1f:
474 case 0xa0 ... 0xaf:
475 data->ccd_offset = 0x300;
476 k10temp_get_ccd_support(pdev, data, 12);
477 break;
478 }
479 } else if (boot_cpu_data.x86 == 0x1a) {
480 data->temp_adjust_mask = ZEN_CUR_TEMP_RANGE_SEL_MASK;
481 data->read_tempreg = read_tempreg_nb_zen;
482 data->is_zen = true;
483 } else {
484 data->read_htcreg = read_htcreg_pci;
485 data->read_tempreg = read_tempreg_pci;
486 }
487
488 for (i = 0; i < ARRAY_SIZE(tctl_offset_table); i++) {
489 const struct tctl_offset *entry = &tctl_offset_table[i];
490
491 if (boot_cpu_data.x86 == entry->model &&
492 strstr(boot_cpu_data.x86_model_id, entry->id)) {
493 data->show_temp |= BIT(TDIE_BIT); /* show Tdie */
494 data->temp_offset = entry->offset;
495 break;
496 }
497 }
498
499 hwmon_dev = devm_hwmon_device_register_with_info(dev, "k10temp", data,
500 &k10temp_chip_info,
501 NULL);
502 return PTR_ERR_OR_ZERO(hwmon_dev);
503}
504
505static const struct pci_device_id k10temp_id_table[] = {
506 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_10H_NB_MISC) },
507 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_11H_NB_MISC) },
508 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_CNB17H_F3) },
509 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_NB_F3) },
510 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M10H_F3) },
511 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M30H_NB_F3) },
512 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M60H_NB_F3) },
513 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M70H_NB_F3) },
514 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_16H_NB_F3) },
515 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_16H_M30H_NB_F3) },
516 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_DF_F3) },
517 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M10H_DF_F3) },
518 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M30H_DF_F3) },
519 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M60H_DF_F3) },
520 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M70H_DF_F3) },
521 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_MA0H_DF_F3) },
522 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_DF_F3) },
523 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M10H_DF_F3) },
524 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M40H_DF_F3) },
525 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M50H_DF_F3) },
526 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M60H_DF_F3) },
527 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M70H_DF_F3) },
528 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_19H_M78H_DF_F3) },
529 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_1AH_M00H_DF_F3) },
530 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_1AH_M20H_DF_F3) },
531 { PCI_VDEVICE(HYGON, PCI_DEVICE_ID_AMD_17H_DF_F3) },
532 {}
533};
534MODULE_DEVICE_TABLE(pci, k10temp_id_table);
535
536static struct pci_driver k10temp_driver = {
537 .name = "k10temp",
538 .id_table = k10temp_id_table,
539 .probe = k10temp_probe,
540};
541
542module_pci_driver(k10temp_driver);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * k10temp.c - AMD Family 10h/11h/12h/14h/15h/16h/17h
4 * processor hardware monitoring
5 *
6 * Copyright (c) 2009 Clemens Ladisch <clemens@ladisch.de>
7 * Copyright (c) 2020 Guenter Roeck <linux@roeck-us.net>
8 *
9 * Implementation notes:
10 * - CCD register address information as well as the calculation to
11 * convert raw register values is from https://github.com/ocerman/zenpower.
12 * The information is not confirmed from chip datasheets, but experiments
13 * suggest that it provides reasonable temperature values.
14 * - Register addresses to read chip voltage and current are also from
15 * https://github.com/ocerman/zenpower, and not confirmed from chip
16 * datasheets. Current calibration is board specific and not typically
17 * shared by board vendors. For this reason, current values are
18 * normalized to report 1A/LSB for core current and and 0.25A/LSB for SoC
19 * current. Reported values can be adjusted using the sensors configuration
20 * file.
21 */
22
23#include <linux/bitops.h>
24#include <linux/debugfs.h>
25#include <linux/err.h>
26#include <linux/hwmon.h>
27#include <linux/init.h>
28#include <linux/module.h>
29#include <linux/pci.h>
30#include <linux/pci_ids.h>
31#include <asm/amd_nb.h>
32#include <asm/processor.h>
33
34MODULE_DESCRIPTION("AMD Family 10h+ CPU core temperature monitor");
35MODULE_AUTHOR("Clemens Ladisch <clemens@ladisch.de>");
36MODULE_LICENSE("GPL");
37
38static bool force;
39module_param(force, bool, 0444);
40MODULE_PARM_DESC(force, "force loading on processors with erratum 319");
41
42/* Provide lock for writing to NB_SMU_IND_ADDR */
43static DEFINE_MUTEX(nb_smu_ind_mutex);
44
45#ifndef PCI_DEVICE_ID_AMD_15H_M70H_NB_F3
46#define PCI_DEVICE_ID_AMD_15H_M70H_NB_F3 0x15b3
47#endif
48
49/* CPUID function 0x80000001, ebx */
50#define CPUID_PKGTYPE_MASK GENMASK(31, 28)
51#define CPUID_PKGTYPE_F 0x00000000
52#define CPUID_PKGTYPE_AM2R2_AM3 0x10000000
53
54/* DRAM controller (PCI function 2) */
55#define REG_DCT0_CONFIG_HIGH 0x094
56#define DDR3_MODE BIT(8)
57
58/* miscellaneous (PCI function 3) */
59#define REG_HARDWARE_THERMAL_CONTROL 0x64
60#define HTC_ENABLE BIT(0)
61
62#define REG_REPORTED_TEMPERATURE 0xa4
63
64#define REG_NORTHBRIDGE_CAPABILITIES 0xe8
65#define NB_CAP_HTC BIT(10)
66
67/*
68 * For F15h M60h and M70h, REG_HARDWARE_THERMAL_CONTROL
69 * and REG_REPORTED_TEMPERATURE have been moved to
70 * D0F0xBC_xD820_0C64 [Hardware Temperature Control]
71 * D0F0xBC_xD820_0CA4 [Reported Temperature Control]
72 */
73#define F15H_M60H_HARDWARE_TEMP_CTRL_OFFSET 0xd8200c64
74#define F15H_M60H_REPORTED_TEMP_CTRL_OFFSET 0xd8200ca4
75
76/* F17h M01h Access througn SMN */
77#define F17H_M01H_REPORTED_TEMP_CTRL_OFFSET 0x00059800
78
79#define F17H_M70H_CCD_TEMP(x) (0x00059954 + ((x) * 4))
80#define F17H_M70H_CCD_TEMP_VALID BIT(11)
81#define F17H_M70H_CCD_TEMP_MASK GENMASK(10, 0)
82
83#define F17H_M01H_SVI 0x0005A000
84#define F17H_M01H_SVI_TEL_PLANE0 (F17H_M01H_SVI + 0xc)
85#define F17H_M01H_SVI_TEL_PLANE1 (F17H_M01H_SVI + 0x10)
86
87#define CUR_TEMP_SHIFT 21
88#define CUR_TEMP_RANGE_SEL_MASK BIT(19)
89
90#define CFACTOR_ICORE 1000000 /* 1A / LSB */
91#define CFACTOR_ISOC 250000 /* 0.25A / LSB */
92
93struct k10temp_data {
94 struct pci_dev *pdev;
95 void (*read_htcreg)(struct pci_dev *pdev, u32 *regval);
96 void (*read_tempreg)(struct pci_dev *pdev, u32 *regval);
97 int temp_offset;
98 u32 temp_adjust_mask;
99 u32 show_temp;
100 u32 svi_addr[2];
101 bool is_zen;
102 bool show_current;
103 int cfactor[2];
104};
105
106#define TCTL_BIT 0
107#define TDIE_BIT 1
108#define TCCD_BIT(x) ((x) + 2)
109
110#define HAVE_TEMP(d, channel) ((d)->show_temp & BIT(channel))
111#define HAVE_TDIE(d) HAVE_TEMP(d, TDIE_BIT)
112
113struct tctl_offset {
114 u8 model;
115 char const *id;
116 int offset;
117};
118
119static const struct tctl_offset tctl_offset_table[] = {
120 { 0x17, "AMD Ryzen 5 1600X", 20000 },
121 { 0x17, "AMD Ryzen 7 1700X", 20000 },
122 { 0x17, "AMD Ryzen 7 1800X", 20000 },
123 { 0x17, "AMD Ryzen 7 2700X", 10000 },
124 { 0x17, "AMD Ryzen Threadripper 19", 27000 }, /* 19{00,20,50}X */
125 { 0x17, "AMD Ryzen Threadripper 29", 27000 }, /* 29{20,50,70,90}[W]X */
126};
127
128static bool is_threadripper(void)
129{
130 return strstr(boot_cpu_data.x86_model_id, "Threadripper");
131}
132
133static bool is_epyc(void)
134{
135 return strstr(boot_cpu_data.x86_model_id, "EPYC");
136}
137
138static void read_htcreg_pci(struct pci_dev *pdev, u32 *regval)
139{
140 pci_read_config_dword(pdev, REG_HARDWARE_THERMAL_CONTROL, regval);
141}
142
143static void read_tempreg_pci(struct pci_dev *pdev, u32 *regval)
144{
145 pci_read_config_dword(pdev, REG_REPORTED_TEMPERATURE, regval);
146}
147
148static void amd_nb_index_read(struct pci_dev *pdev, unsigned int devfn,
149 unsigned int base, int offset, u32 *val)
150{
151 mutex_lock(&nb_smu_ind_mutex);
152 pci_bus_write_config_dword(pdev->bus, devfn,
153 base, offset);
154 pci_bus_read_config_dword(pdev->bus, devfn,
155 base + 4, val);
156 mutex_unlock(&nb_smu_ind_mutex);
157}
158
159static void read_htcreg_nb_f15(struct pci_dev *pdev, u32 *regval)
160{
161 amd_nb_index_read(pdev, PCI_DEVFN(0, 0), 0xb8,
162 F15H_M60H_HARDWARE_TEMP_CTRL_OFFSET, regval);
163}
164
165static void read_tempreg_nb_f15(struct pci_dev *pdev, u32 *regval)
166{
167 amd_nb_index_read(pdev, PCI_DEVFN(0, 0), 0xb8,
168 F15H_M60H_REPORTED_TEMP_CTRL_OFFSET, regval);
169}
170
171static void read_tempreg_nb_f17(struct pci_dev *pdev, u32 *regval)
172{
173 amd_smn_read(amd_pci_dev_to_node_id(pdev),
174 F17H_M01H_REPORTED_TEMP_CTRL_OFFSET, regval);
175}
176
177static long get_raw_temp(struct k10temp_data *data)
178{
179 u32 regval;
180 long temp;
181
182 data->read_tempreg(data->pdev, ®val);
183 temp = (regval >> CUR_TEMP_SHIFT) * 125;
184 if (regval & data->temp_adjust_mask)
185 temp -= 49000;
186 return temp;
187}
188
189static const char *k10temp_temp_label[] = {
190 "Tctl",
191 "Tdie",
192 "Tccd1",
193 "Tccd2",
194 "Tccd3",
195 "Tccd4",
196 "Tccd5",
197 "Tccd6",
198 "Tccd7",
199 "Tccd8",
200};
201
202static const char *k10temp_in_label[] = {
203 "Vcore",
204 "Vsoc",
205};
206
207static const char *k10temp_curr_label[] = {
208 "Icore",
209 "Isoc",
210};
211
212static int k10temp_read_labels(struct device *dev,
213 enum hwmon_sensor_types type,
214 u32 attr, int channel, const char **str)
215{
216 switch (type) {
217 case hwmon_temp:
218 *str = k10temp_temp_label[channel];
219 break;
220 case hwmon_in:
221 *str = k10temp_in_label[channel];
222 break;
223 case hwmon_curr:
224 *str = k10temp_curr_label[channel];
225 break;
226 default:
227 return -EOPNOTSUPP;
228 }
229 return 0;
230}
231
232static int k10temp_read_curr(struct device *dev, u32 attr, int channel,
233 long *val)
234{
235 struct k10temp_data *data = dev_get_drvdata(dev);
236 u32 regval;
237
238 switch (attr) {
239 case hwmon_curr_input:
240 amd_smn_read(amd_pci_dev_to_node_id(data->pdev),
241 data->svi_addr[channel], ®val);
242 *val = DIV_ROUND_CLOSEST(data->cfactor[channel] *
243 (regval & 0xff),
244 1000);
245 break;
246 default:
247 return -EOPNOTSUPP;
248 }
249 return 0;
250}
251
252static int k10temp_read_in(struct device *dev, u32 attr, int channel, long *val)
253{
254 struct k10temp_data *data = dev_get_drvdata(dev);
255 u32 regval;
256
257 switch (attr) {
258 case hwmon_in_input:
259 amd_smn_read(amd_pci_dev_to_node_id(data->pdev),
260 data->svi_addr[channel], ®val);
261 regval = (regval >> 16) & 0xff;
262 *val = DIV_ROUND_CLOSEST(155000 - regval * 625, 100);
263 break;
264 default:
265 return -EOPNOTSUPP;
266 }
267 return 0;
268}
269
270static int k10temp_read_temp(struct device *dev, u32 attr, int channel,
271 long *val)
272{
273 struct k10temp_data *data = dev_get_drvdata(dev);
274 u32 regval;
275
276 switch (attr) {
277 case hwmon_temp_input:
278 switch (channel) {
279 case 0: /* Tctl */
280 *val = get_raw_temp(data);
281 if (*val < 0)
282 *val = 0;
283 break;
284 case 1: /* Tdie */
285 *val = get_raw_temp(data) - data->temp_offset;
286 if (*val < 0)
287 *val = 0;
288 break;
289 case 2 ... 9: /* Tccd{1-8} */
290 amd_smn_read(amd_pci_dev_to_node_id(data->pdev),
291 F17H_M70H_CCD_TEMP(channel - 2), ®val);
292 *val = (regval & F17H_M70H_CCD_TEMP_MASK) * 125 - 49000;
293 break;
294 default:
295 return -EOPNOTSUPP;
296 }
297 break;
298 case hwmon_temp_max:
299 *val = 70 * 1000;
300 break;
301 case hwmon_temp_crit:
302 data->read_htcreg(data->pdev, ®val);
303 *val = ((regval >> 16) & 0x7f) * 500 + 52000;
304 break;
305 case hwmon_temp_crit_hyst:
306 data->read_htcreg(data->pdev, ®val);
307 *val = (((regval >> 16) & 0x7f)
308 - ((regval >> 24) & 0xf)) * 500 + 52000;
309 break;
310 default:
311 return -EOPNOTSUPP;
312 }
313 return 0;
314}
315
316static int k10temp_read(struct device *dev, enum hwmon_sensor_types type,
317 u32 attr, int channel, long *val)
318{
319 switch (type) {
320 case hwmon_temp:
321 return k10temp_read_temp(dev, attr, channel, val);
322 case hwmon_in:
323 return k10temp_read_in(dev, attr, channel, val);
324 case hwmon_curr:
325 return k10temp_read_curr(dev, attr, channel, val);
326 default:
327 return -EOPNOTSUPP;
328 }
329}
330
331static umode_t k10temp_is_visible(const void *_data,
332 enum hwmon_sensor_types type,
333 u32 attr, int channel)
334{
335 const struct k10temp_data *data = _data;
336 struct pci_dev *pdev = data->pdev;
337 u32 reg;
338
339 switch (type) {
340 case hwmon_temp:
341 switch (attr) {
342 case hwmon_temp_input:
343 if (!HAVE_TEMP(data, channel))
344 return 0;
345 break;
346 case hwmon_temp_max:
347 if (channel || data->is_zen)
348 return 0;
349 break;
350 case hwmon_temp_crit:
351 case hwmon_temp_crit_hyst:
352 if (channel || !data->read_htcreg)
353 return 0;
354
355 pci_read_config_dword(pdev,
356 REG_NORTHBRIDGE_CAPABILITIES,
357 ®);
358 if (!(reg & NB_CAP_HTC))
359 return 0;
360
361 data->read_htcreg(data->pdev, ®);
362 if (!(reg & HTC_ENABLE))
363 return 0;
364 break;
365 case hwmon_temp_label:
366 /* Show temperature labels only on Zen CPUs */
367 if (!data->is_zen || !HAVE_TEMP(data, channel))
368 return 0;
369 break;
370 default:
371 return 0;
372 }
373 break;
374 case hwmon_in:
375 case hwmon_curr:
376 if (!data->show_current)
377 return 0;
378 break;
379 default:
380 return 0;
381 }
382 return 0444;
383}
384
385static bool has_erratum_319(struct pci_dev *pdev)
386{
387 u32 pkg_type, reg_dram_cfg;
388
389 if (boot_cpu_data.x86 != 0x10)
390 return false;
391
392 /*
393 * Erratum 319: The thermal sensor of Socket F/AM2+ processors
394 * may be unreliable.
395 */
396 pkg_type = cpuid_ebx(0x80000001) & CPUID_PKGTYPE_MASK;
397 if (pkg_type == CPUID_PKGTYPE_F)
398 return true;
399 if (pkg_type != CPUID_PKGTYPE_AM2R2_AM3)
400 return false;
401
402 /* DDR3 memory implies socket AM3, which is good */
403 pci_bus_read_config_dword(pdev->bus,
404 PCI_DEVFN(PCI_SLOT(pdev->devfn), 2),
405 REG_DCT0_CONFIG_HIGH, ®_dram_cfg);
406 if (reg_dram_cfg & DDR3_MODE)
407 return false;
408
409 /*
410 * Unfortunately it is possible to run a socket AM3 CPU with DDR2
411 * memory. We blacklist all the cores which do exist in socket AM2+
412 * format. It still isn't perfect, as RB-C2 cores exist in both AM2+
413 * and AM3 formats, but that's the best we can do.
414 */
415 return boot_cpu_data.x86_model < 4 ||
416 (boot_cpu_data.x86_model == 4 && boot_cpu_data.x86_stepping <= 2);
417}
418
419#ifdef CONFIG_DEBUG_FS
420
421static void k10temp_smn_regs_show(struct seq_file *s, struct pci_dev *pdev,
422 u32 addr, int count)
423{
424 u32 reg;
425 int i;
426
427 for (i = 0; i < count; i++) {
428 if (!(i & 3))
429 seq_printf(s, "0x%06x: ", addr + i * 4);
430 amd_smn_read(amd_pci_dev_to_node_id(pdev), addr + i * 4, ®);
431 seq_printf(s, "%08x ", reg);
432 if ((i & 3) == 3)
433 seq_puts(s, "\n");
434 }
435}
436
437static int svi_show(struct seq_file *s, void *unused)
438{
439 struct k10temp_data *data = s->private;
440
441 k10temp_smn_regs_show(s, data->pdev, F17H_M01H_SVI, 32);
442 return 0;
443}
444DEFINE_SHOW_ATTRIBUTE(svi);
445
446static int thm_show(struct seq_file *s, void *unused)
447{
448 struct k10temp_data *data = s->private;
449
450 k10temp_smn_regs_show(s, data->pdev,
451 F17H_M01H_REPORTED_TEMP_CTRL_OFFSET, 256);
452 return 0;
453}
454DEFINE_SHOW_ATTRIBUTE(thm);
455
456static void k10temp_debugfs_cleanup(void *ddir)
457{
458 debugfs_remove_recursive(ddir);
459}
460
461static void k10temp_init_debugfs(struct k10temp_data *data)
462{
463 struct dentry *debugfs;
464 char name[32];
465
466 /* Only show debugfs data for Family 17h/18h CPUs */
467 if (!data->is_zen)
468 return;
469
470 scnprintf(name, sizeof(name), "k10temp-%s", pci_name(data->pdev));
471
472 debugfs = debugfs_create_dir(name, NULL);
473 if (debugfs) {
474 debugfs_create_file("svi", 0444, debugfs, data, &svi_fops);
475 debugfs_create_file("thm", 0444, debugfs, data, &thm_fops);
476 devm_add_action_or_reset(&data->pdev->dev,
477 k10temp_debugfs_cleanup, debugfs);
478 }
479}
480
481#else
482
483static void k10temp_init_debugfs(struct k10temp_data *data)
484{
485}
486
487#endif
488
489static const struct hwmon_channel_info *k10temp_info[] = {
490 HWMON_CHANNEL_INFO(temp,
491 HWMON_T_INPUT | HWMON_T_MAX |
492 HWMON_T_CRIT | HWMON_T_CRIT_HYST |
493 HWMON_T_LABEL,
494 HWMON_T_INPUT | HWMON_T_LABEL,
495 HWMON_T_INPUT | HWMON_T_LABEL,
496 HWMON_T_INPUT | HWMON_T_LABEL,
497 HWMON_T_INPUT | HWMON_T_LABEL,
498 HWMON_T_INPUT | HWMON_T_LABEL,
499 HWMON_T_INPUT | HWMON_T_LABEL,
500 HWMON_T_INPUT | HWMON_T_LABEL,
501 HWMON_T_INPUT | HWMON_T_LABEL,
502 HWMON_T_INPUT | HWMON_T_LABEL),
503 HWMON_CHANNEL_INFO(in,
504 HWMON_I_INPUT | HWMON_I_LABEL,
505 HWMON_I_INPUT | HWMON_I_LABEL),
506 HWMON_CHANNEL_INFO(curr,
507 HWMON_C_INPUT | HWMON_C_LABEL,
508 HWMON_C_INPUT | HWMON_C_LABEL),
509 NULL
510};
511
512static const struct hwmon_ops k10temp_hwmon_ops = {
513 .is_visible = k10temp_is_visible,
514 .read = k10temp_read,
515 .read_string = k10temp_read_labels,
516};
517
518static const struct hwmon_chip_info k10temp_chip_info = {
519 .ops = &k10temp_hwmon_ops,
520 .info = k10temp_info,
521};
522
523static void k10temp_get_ccd_support(struct pci_dev *pdev,
524 struct k10temp_data *data, int limit)
525{
526 u32 regval;
527 int i;
528
529 for (i = 0; i < limit; i++) {
530 amd_smn_read(amd_pci_dev_to_node_id(pdev),
531 F17H_M70H_CCD_TEMP(i), ®val);
532 if (regval & F17H_M70H_CCD_TEMP_VALID)
533 data->show_temp |= BIT(TCCD_BIT(i));
534 }
535}
536
537static int k10temp_probe(struct pci_dev *pdev, const struct pci_device_id *id)
538{
539 int unreliable = has_erratum_319(pdev);
540 struct device *dev = &pdev->dev;
541 struct k10temp_data *data;
542 struct device *hwmon_dev;
543 int i;
544
545 if (unreliable) {
546 if (!force) {
547 dev_err(dev,
548 "unreliable CPU thermal sensor; monitoring disabled\n");
549 return -ENODEV;
550 }
551 dev_warn(dev,
552 "unreliable CPU thermal sensor; check erratum 319\n");
553 }
554
555 data = devm_kzalloc(dev, sizeof(*data), GFP_KERNEL);
556 if (!data)
557 return -ENOMEM;
558
559 data->pdev = pdev;
560 data->show_temp |= BIT(TCTL_BIT); /* Always show Tctl */
561
562 if (boot_cpu_data.x86 == 0x15 &&
563 ((boot_cpu_data.x86_model & 0xf0) == 0x60 ||
564 (boot_cpu_data.x86_model & 0xf0) == 0x70)) {
565 data->read_htcreg = read_htcreg_nb_f15;
566 data->read_tempreg = read_tempreg_nb_f15;
567 } else if (boot_cpu_data.x86 == 0x17 || boot_cpu_data.x86 == 0x18) {
568 data->temp_adjust_mask = CUR_TEMP_RANGE_SEL_MASK;
569 data->read_tempreg = read_tempreg_nb_f17;
570 data->show_temp |= BIT(TDIE_BIT); /* show Tdie */
571 data->is_zen = true;
572
573 switch (boot_cpu_data.x86_model) {
574 case 0x1: /* Zen */
575 case 0x8: /* Zen+ */
576 case 0x11: /* Zen APU */
577 case 0x18: /* Zen+ APU */
578 data->show_current = !is_threadripper() && !is_epyc();
579 data->svi_addr[0] = F17H_M01H_SVI_TEL_PLANE0;
580 data->svi_addr[1] = F17H_M01H_SVI_TEL_PLANE1;
581 data->cfactor[0] = CFACTOR_ICORE;
582 data->cfactor[1] = CFACTOR_ISOC;
583 k10temp_get_ccd_support(pdev, data, 4);
584 break;
585 case 0x31: /* Zen2 Threadripper */
586 case 0x71: /* Zen2 */
587 data->show_current = !is_threadripper() && !is_epyc();
588 data->cfactor[0] = CFACTOR_ICORE;
589 data->cfactor[1] = CFACTOR_ISOC;
590 data->svi_addr[0] = F17H_M01H_SVI_TEL_PLANE1;
591 data->svi_addr[1] = F17H_M01H_SVI_TEL_PLANE0;
592 k10temp_get_ccd_support(pdev, data, 8);
593 break;
594 }
595 } else {
596 data->read_htcreg = read_htcreg_pci;
597 data->read_tempreg = read_tempreg_pci;
598 }
599
600 for (i = 0; i < ARRAY_SIZE(tctl_offset_table); i++) {
601 const struct tctl_offset *entry = &tctl_offset_table[i];
602
603 if (boot_cpu_data.x86 == entry->model &&
604 strstr(boot_cpu_data.x86_model_id, entry->id)) {
605 data->temp_offset = entry->offset;
606 break;
607 }
608 }
609
610 hwmon_dev = devm_hwmon_device_register_with_info(dev, "k10temp", data,
611 &k10temp_chip_info,
612 NULL);
613 if (IS_ERR(hwmon_dev))
614 return PTR_ERR(hwmon_dev);
615
616 k10temp_init_debugfs(data);
617
618 return 0;
619}
620
621static const struct pci_device_id k10temp_id_table[] = {
622 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_10H_NB_MISC) },
623 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_11H_NB_MISC) },
624 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_CNB17H_F3) },
625 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_NB_F3) },
626 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M10H_F3) },
627 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M30H_NB_F3) },
628 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M60H_NB_F3) },
629 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_15H_M70H_NB_F3) },
630 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_16H_NB_F3) },
631 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_16H_M30H_NB_F3) },
632 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_DF_F3) },
633 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M10H_DF_F3) },
634 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M30H_DF_F3) },
635 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M60H_DF_F3) },
636 { PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_17H_M70H_DF_F3) },
637 { PCI_VDEVICE(HYGON, PCI_DEVICE_ID_AMD_17H_DF_F3) },
638 {}
639};
640MODULE_DEVICE_TABLE(pci, k10temp_id_table);
641
642static struct pci_driver k10temp_driver = {
643 .name = "k10temp",
644 .id_table = k10temp_id_table,
645 .probe = k10temp_probe,
646};
647
648module_pci_driver(k10temp_driver);