]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - drivers/hwmon/drivetemp.c
scsi: lpfc: Fix queue failures when recovering from PCI parity error
[mirror_ubuntu-jammy-kernel.git] / drivers / hwmon / drivetemp.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Hwmon client for disk and solid state drives with temperature sensors
4 * Copyright (C) 2019 Zodiac Inflight Innovations
5 *
6 * With input from:
7 * Hwmon client for S.M.A.R.T. hard disk drives with temperature sensors.
8 * (C) 2018 Linus Walleij
9 *
10 * hwmon: Driver for SCSI/ATA temperature sensors
11 * by Constantin Baranov <const@mimas.ru>, submitted September 2009
12 *
13 * This drive supports reporting the temperature of SATA drives. It can be
14 * easily extended to report the temperature of SCSI drives.
15 *
16 * The primary means to read drive temperatures and temperature limits
17 * for ATA drives is the SCT Command Transport feature set as specified in
18 * ATA8-ACS.
19 * It can be used to read the current drive temperature, temperature limits,
20 * and historic minimum and maximum temperatures. The SCT Command Transport
21 * feature set is documented in "AT Attachment 8 - ATA/ATAPI Command Set
22 * (ATA8-ACS)".
23 *
24 * If the SCT Command Transport feature set is not available, drive temperatures
25 * may be readable through SMART attributes. Since SMART attributes are not well
26 * defined, this method is only used as fallback mechanism.
27 *
28 * There are three SMART attributes which may report drive temperatures.
29 * Those are defined as follows (from
30 * http://www.cropel.com/library/smart-attribute-list.aspx).
31 *
32 * 190 Temperature Temperature, monitored by a sensor somewhere inside
33 * the drive. Raw value typicaly holds the actual
34 * temperature (hexadecimal) in its rightmost two digits.
35 *
36 * 194 Temperature Temperature, monitored by a sensor somewhere inside
37 * the drive. Raw value typicaly holds the actual
38 * temperature (hexadecimal) in its rightmost two digits.
39 *
40 * 231 Temperature Temperature, monitored by a sensor somewhere inside
41 * the drive. Raw value typicaly holds the actual
42 * temperature (hexadecimal) in its rightmost two digits.
43 *
44 * Wikipedia defines attributes a bit differently.
45 *
46 * 190 Temperature Value is equal to (100-temp. °C), allowing manufacturer
47 * Difference or to set a minimum threshold which corresponds to a
48 * Airflow maximum temperature. This also follows the convention of
49 * Temperature 100 being a best-case value and lower values being
50 * undesirable. However, some older drives may instead
51 * report raw Temperature (identical to 0xC2) or
52 * Temperature minus 50 here.
53 * 194 Temperature or Indicates the device temperature, if the appropriate
54 * Temperature sensor is fitted. Lowest byte of the raw value contains
55 * Celsius the exact temperature value (Celsius degrees).
56 * 231 Life Left Indicates the approximate SSD life left, in terms of
57 * (SSDs) or program/erase cycles or available reserved blocks.
58 * Temperature A normalized value of 100 represents a new drive, with
59 * a threshold value at 10 indicating a need for
60 * replacement. A value of 0 may mean that the drive is
61 * operating in read-only mode to allow data recovery.
62 * Previously (pre-2010) occasionally used for Drive
63 * Temperature (more typically reported at 0xC2).
64 *
65 * Common denominator is that the first raw byte reports the temperature
66 * in degrees C on almost all drives. Some drives may report a fractional
67 * temperature in the second raw byte.
68 *
69 * Known exceptions (from libatasmart):
70 * - SAMSUNG SV0412H and SAMSUNG SV1204H) report the temperature in 10th
71 * degrees C in the first two raw bytes.
72 * - A few Maxtor drives report an unknown or bad value in attribute 194.
73 * - Certain Apple SSD drives report an unknown value in attribute 190.
74 * Only certain firmware versions are affected.
75 *
76 * Those exceptions affect older ATA drives and are currently ignored.
77 * Also, the second raw byte (possibly reporting the fractional temperature)
78 * is currently ignored.
79 *
80 * Many drives also report temperature limits in additional SMART data raw
81 * bytes. The format of those is not well defined and varies widely.
82 * The driver does not currently attempt to report those limits.
83 *
84 * According to data in smartmontools, attribute 231 is rarely used to report
85 * drive temperatures. At the same time, several drives report SSD life left
86 * in attribute 231, but do not support temperature sensors. For this reason,
87 * attribute 231 is currently ignored.
88 *
89 * Following above definitions, temperatures are reported as follows.
90 * If SCT Command Transport is supported, it is used to read the
91 * temperature and, if available, temperature limits.
92 * - Otherwise, if SMART attribute 194 is supported, it is used to read
93 * the temperature.
94 * - Otherwise, if SMART attribute 190 is supported, it is used to read
95 * the temperature.
96 */
97
98 #include <linux/ata.h>
99 #include <linux/bits.h>
100 #include <linux/device.h>
101 #include <linux/hwmon.h>
102 #include <linux/kernel.h>
103 #include <linux/list.h>
104 #include <linux/module.h>
105 #include <linux/mutex.h>
106 #include <scsi/scsi_cmnd.h>
107 #include <scsi/scsi_device.h>
108 #include <scsi/scsi_driver.h>
109 #include <scsi/scsi_proto.h>
110
111 struct drivetemp_data {
112 struct list_head list; /* list of instantiated devices */
113 struct mutex lock; /* protect data buffer accesses */
114 struct scsi_device *sdev; /* SCSI device */
115 struct device *dev; /* instantiating device */
116 struct device *hwdev; /* hardware monitoring device */
117 u8 smartdata[ATA_SECT_SIZE]; /* local buffer */
118 int (*get_temp)(struct drivetemp_data *st, u32 attr, long *val);
119 bool have_temp_lowest; /* lowest temp in SCT status */
120 bool have_temp_highest; /* highest temp in SCT status */
121 bool have_temp_min; /* have min temp */
122 bool have_temp_max; /* have max temp */
123 bool have_temp_lcrit; /* have lower critical limit */
124 bool have_temp_crit; /* have critical limit */
125 int temp_min; /* min temp */
126 int temp_max; /* max temp */
127 int temp_lcrit; /* lower critical limit */
128 int temp_crit; /* critical limit */
129 };
130
131 static LIST_HEAD(drivetemp_devlist);
132
133 #define ATA_MAX_SMART_ATTRS 30
134 #define SMART_TEMP_PROP_190 190
135 #define SMART_TEMP_PROP_194 194
136
137 #define SCT_STATUS_REQ_ADDR 0xe0
138 #define SCT_STATUS_VERSION_LOW 0 /* log byte offsets */
139 #define SCT_STATUS_VERSION_HIGH 1
140 #define SCT_STATUS_TEMP 200
141 #define SCT_STATUS_TEMP_LOWEST 201
142 #define SCT_STATUS_TEMP_HIGHEST 202
143 #define SCT_READ_LOG_ADDR 0xe1
144 #define SMART_READ_LOG 0xd5
145 #define SMART_WRITE_LOG 0xd6
146
147 #define INVALID_TEMP 0x80
148
149 #define temp_is_valid(temp) ((temp) != INVALID_TEMP)
150 #define temp_from_sct(temp) (((s8)(temp)) * 1000)
151
152 static inline bool ata_id_smart_supported(u16 *id)
153 {
154 return id[ATA_ID_COMMAND_SET_1] & BIT(0);
155 }
156
157 static inline bool ata_id_smart_enabled(u16 *id)
158 {
159 return id[ATA_ID_CFS_ENABLE_1] & BIT(0);
160 }
161
162 static int drivetemp_scsi_command(struct drivetemp_data *st,
163 u8 ata_command, u8 feature,
164 u8 lba_low, u8 lba_mid, u8 lba_high)
165 {
166 u8 scsi_cmd[MAX_COMMAND_SIZE];
167 int data_dir;
168
169 memset(scsi_cmd, 0, sizeof(scsi_cmd));
170 scsi_cmd[0] = ATA_16;
171 if (ata_command == ATA_CMD_SMART && feature == SMART_WRITE_LOG) {
172 scsi_cmd[1] = (5 << 1); /* PIO Data-out */
173 /*
174 * No off.line or cc, write to dev, block count in sector count
175 * field.
176 */
177 scsi_cmd[2] = 0x06;
178 data_dir = DMA_TO_DEVICE;
179 } else {
180 scsi_cmd[1] = (4 << 1); /* PIO Data-in */
181 /*
182 * No off.line or cc, read from dev, block count in sector count
183 * field.
184 */
185 scsi_cmd[2] = 0x0e;
186 data_dir = DMA_FROM_DEVICE;
187 }
188 scsi_cmd[4] = feature;
189 scsi_cmd[6] = 1; /* 1 sector */
190 scsi_cmd[8] = lba_low;
191 scsi_cmd[10] = lba_mid;
192 scsi_cmd[12] = lba_high;
193 scsi_cmd[14] = ata_command;
194
195 return scsi_execute_req(st->sdev, scsi_cmd, data_dir,
196 st->smartdata, ATA_SECT_SIZE, NULL, HZ, 5,
197 NULL);
198 }
199
200 static int drivetemp_ata_command(struct drivetemp_data *st, u8 feature,
201 u8 select)
202 {
203 return drivetemp_scsi_command(st, ATA_CMD_SMART, feature, select,
204 ATA_SMART_LBAM_PASS, ATA_SMART_LBAH_PASS);
205 }
206
207 static int drivetemp_get_smarttemp(struct drivetemp_data *st, u32 attr,
208 long *temp)
209 {
210 u8 *buf = st->smartdata;
211 bool have_temp = false;
212 u8 temp_raw;
213 u8 csum;
214 int err;
215 int i;
216
217 err = drivetemp_ata_command(st, ATA_SMART_READ_VALUES, 0);
218 if (err)
219 return err;
220
221 /* Checksum the read value table */
222 csum = 0;
223 for (i = 0; i < ATA_SECT_SIZE; i++)
224 csum += buf[i];
225 if (csum) {
226 dev_dbg(&st->sdev->sdev_gendev,
227 "checksum error reading SMART values\n");
228 return -EIO;
229 }
230
231 for (i = 0; i < ATA_MAX_SMART_ATTRS; i++) {
232 u8 *attr = buf + i * 12;
233 int id = attr[2];
234
235 if (!id)
236 continue;
237
238 if (id == SMART_TEMP_PROP_190) {
239 temp_raw = attr[7];
240 have_temp = true;
241 }
242 if (id == SMART_TEMP_PROP_194) {
243 temp_raw = attr[7];
244 have_temp = true;
245 break;
246 }
247 }
248
249 if (have_temp) {
250 *temp = temp_raw * 1000;
251 return 0;
252 }
253
254 return -ENXIO;
255 }
256
257 static int drivetemp_get_scttemp(struct drivetemp_data *st, u32 attr, long *val)
258 {
259 u8 *buf = st->smartdata;
260 int err;
261
262 err = drivetemp_ata_command(st, SMART_READ_LOG, SCT_STATUS_REQ_ADDR);
263 if (err)
264 return err;
265 switch (attr) {
266 case hwmon_temp_input:
267 if (!temp_is_valid(buf[SCT_STATUS_TEMP]))
268 return -ENODATA;
269 *val = temp_from_sct(buf[SCT_STATUS_TEMP]);
270 break;
271 case hwmon_temp_lowest:
272 if (!temp_is_valid(buf[SCT_STATUS_TEMP_LOWEST]))
273 return -ENODATA;
274 *val = temp_from_sct(buf[SCT_STATUS_TEMP_LOWEST]);
275 break;
276 case hwmon_temp_highest:
277 if (!temp_is_valid(buf[SCT_STATUS_TEMP_HIGHEST]))
278 return -ENODATA;
279 *val = temp_from_sct(buf[SCT_STATUS_TEMP_HIGHEST]);
280 break;
281 default:
282 err = -EINVAL;
283 break;
284 }
285 return err;
286 }
287
288 static const char * const sct_avoid_models[] = {
289 /*
290 * These drives will have WRITE FPDMA QUEUED command timeouts and sometimes just
291 * freeze until power-cycled under heavy write loads when their temperature is
292 * getting polled in SCT mode. The SMART mode seems to be fine, though.
293 *
294 * While only the 3 TB model (DT01ACA3) was actually caught exhibiting the
295 * problem let's play safe here to avoid data corruption and ban the whole
296 * DT01ACAx family.
297
298 * The models from this array are prefix-matched.
299 */
300 "TOSHIBA DT01ACA",
301 };
302
303 static bool drivetemp_sct_avoid(struct drivetemp_data *st)
304 {
305 struct scsi_device *sdev = st->sdev;
306 unsigned int ctr;
307
308 if (!sdev->model)
309 return false;
310
311 /*
312 * The "model" field contains just the raw SCSI INQUIRY response
313 * "product identification" field, which has a width of 16 bytes.
314 * This field is space-filled, but is NOT NULL-terminated.
315 */
316 for (ctr = 0; ctr < ARRAY_SIZE(sct_avoid_models); ctr++)
317 if (!strncmp(sdev->model, sct_avoid_models[ctr],
318 strlen(sct_avoid_models[ctr])))
319 return true;
320
321 return false;
322 }
323
324 static int drivetemp_identify_sata(struct drivetemp_data *st)
325 {
326 struct scsi_device *sdev = st->sdev;
327 u8 *buf = st->smartdata;
328 struct scsi_vpd *vpd;
329 bool is_ata, is_sata;
330 bool have_sct_data_table;
331 bool have_sct_temp;
332 bool have_smart;
333 bool have_sct;
334 u16 *ata_id;
335 u16 version;
336 long temp;
337 int err;
338
339 /* SCSI-ATA Translation present? */
340 rcu_read_lock();
341 vpd = rcu_dereference(sdev->vpd_pg89);
342
343 /*
344 * Verify that ATA IDENTIFY DEVICE data is included in ATA Information
345 * VPD and that the drive implements the SATA protocol.
346 */
347 if (!vpd || vpd->len < 572 || vpd->data[56] != ATA_CMD_ID_ATA ||
348 vpd->data[36] != 0x34) {
349 rcu_read_unlock();
350 return -ENODEV;
351 }
352 ata_id = (u16 *)&vpd->data[60];
353 is_ata = ata_id_is_ata(ata_id);
354 is_sata = ata_id_is_sata(ata_id);
355 have_sct = ata_id_sct_supported(ata_id);
356 have_sct_data_table = ata_id_sct_data_tables(ata_id);
357 have_smart = ata_id_smart_supported(ata_id) &&
358 ata_id_smart_enabled(ata_id);
359
360 rcu_read_unlock();
361
362 /* bail out if this is not a SATA device */
363 if (!is_ata || !is_sata)
364 return -ENODEV;
365
366 if (have_sct && drivetemp_sct_avoid(st)) {
367 dev_notice(&sdev->sdev_gendev,
368 "will avoid using SCT for temperature monitoring\n");
369 have_sct = false;
370 }
371
372 if (!have_sct)
373 goto skip_sct;
374
375 err = drivetemp_ata_command(st, SMART_READ_LOG, SCT_STATUS_REQ_ADDR);
376 if (err)
377 goto skip_sct;
378
379 version = (buf[SCT_STATUS_VERSION_HIGH] << 8) |
380 buf[SCT_STATUS_VERSION_LOW];
381 if (version != 2 && version != 3)
382 goto skip_sct;
383
384 have_sct_temp = temp_is_valid(buf[SCT_STATUS_TEMP]);
385 if (!have_sct_temp)
386 goto skip_sct;
387
388 st->have_temp_lowest = temp_is_valid(buf[SCT_STATUS_TEMP_LOWEST]);
389 st->have_temp_highest = temp_is_valid(buf[SCT_STATUS_TEMP_HIGHEST]);
390
391 if (!have_sct_data_table)
392 goto skip_sct_data;
393
394 /* Request and read temperature history table */
395 memset(buf, '\0', sizeof(st->smartdata));
396 buf[0] = 5; /* data table command */
397 buf[2] = 1; /* read table */
398 buf[4] = 2; /* temperature history table */
399
400 err = drivetemp_ata_command(st, SMART_WRITE_LOG, SCT_STATUS_REQ_ADDR);
401 if (err)
402 goto skip_sct_data;
403
404 err = drivetemp_ata_command(st, SMART_READ_LOG, SCT_READ_LOG_ADDR);
405 if (err)
406 goto skip_sct_data;
407
408 /*
409 * Temperature limits per AT Attachment 8 -
410 * ATA/ATAPI Command Set (ATA8-ACS)
411 */
412 st->have_temp_max = temp_is_valid(buf[6]);
413 st->have_temp_crit = temp_is_valid(buf[7]);
414 st->have_temp_min = temp_is_valid(buf[8]);
415 st->have_temp_lcrit = temp_is_valid(buf[9]);
416
417 st->temp_max = temp_from_sct(buf[6]);
418 st->temp_crit = temp_from_sct(buf[7]);
419 st->temp_min = temp_from_sct(buf[8]);
420 st->temp_lcrit = temp_from_sct(buf[9]);
421
422 skip_sct_data:
423 if (have_sct_temp) {
424 st->get_temp = drivetemp_get_scttemp;
425 return 0;
426 }
427 skip_sct:
428 if (!have_smart)
429 return -ENODEV;
430 st->get_temp = drivetemp_get_smarttemp;
431 return drivetemp_get_smarttemp(st, hwmon_temp_input, &temp);
432 }
433
434 static int drivetemp_identify(struct drivetemp_data *st)
435 {
436 struct scsi_device *sdev = st->sdev;
437
438 /* Bail out immediately if there is no inquiry data */
439 if (!sdev->inquiry || sdev->inquiry_len < 16)
440 return -ENODEV;
441
442 /* Disk device? */
443 if (sdev->type != TYPE_DISK && sdev->type != TYPE_ZBC)
444 return -ENODEV;
445
446 return drivetemp_identify_sata(st);
447 }
448
449 static int drivetemp_read(struct device *dev, enum hwmon_sensor_types type,
450 u32 attr, int channel, long *val)
451 {
452 struct drivetemp_data *st = dev_get_drvdata(dev);
453 int err = 0;
454
455 if (type != hwmon_temp)
456 return -EINVAL;
457
458 switch (attr) {
459 case hwmon_temp_input:
460 case hwmon_temp_lowest:
461 case hwmon_temp_highest:
462 mutex_lock(&st->lock);
463 err = st->get_temp(st, attr, val);
464 mutex_unlock(&st->lock);
465 break;
466 case hwmon_temp_lcrit:
467 *val = st->temp_lcrit;
468 break;
469 case hwmon_temp_min:
470 *val = st->temp_min;
471 break;
472 case hwmon_temp_max:
473 *val = st->temp_max;
474 break;
475 case hwmon_temp_crit:
476 *val = st->temp_crit;
477 break;
478 default:
479 err = -EINVAL;
480 break;
481 }
482 return err;
483 }
484
485 static umode_t drivetemp_is_visible(const void *data,
486 enum hwmon_sensor_types type,
487 u32 attr, int channel)
488 {
489 const struct drivetemp_data *st = data;
490
491 switch (type) {
492 case hwmon_temp:
493 switch (attr) {
494 case hwmon_temp_input:
495 return 0444;
496 case hwmon_temp_lowest:
497 if (st->have_temp_lowest)
498 return 0444;
499 break;
500 case hwmon_temp_highest:
501 if (st->have_temp_highest)
502 return 0444;
503 break;
504 case hwmon_temp_min:
505 if (st->have_temp_min)
506 return 0444;
507 break;
508 case hwmon_temp_max:
509 if (st->have_temp_max)
510 return 0444;
511 break;
512 case hwmon_temp_lcrit:
513 if (st->have_temp_lcrit)
514 return 0444;
515 break;
516 case hwmon_temp_crit:
517 if (st->have_temp_crit)
518 return 0444;
519 break;
520 default:
521 break;
522 }
523 break;
524 default:
525 break;
526 }
527 return 0;
528 }
529
530 static const struct hwmon_channel_info *drivetemp_info[] = {
531 HWMON_CHANNEL_INFO(chip,
532 HWMON_C_REGISTER_TZ),
533 HWMON_CHANNEL_INFO(temp, HWMON_T_INPUT |
534 HWMON_T_LOWEST | HWMON_T_HIGHEST |
535 HWMON_T_MIN | HWMON_T_MAX |
536 HWMON_T_LCRIT | HWMON_T_CRIT),
537 NULL
538 };
539
540 static const struct hwmon_ops drivetemp_ops = {
541 .is_visible = drivetemp_is_visible,
542 .read = drivetemp_read,
543 };
544
545 static const struct hwmon_chip_info drivetemp_chip_info = {
546 .ops = &drivetemp_ops,
547 .info = drivetemp_info,
548 };
549
550 /*
551 * The device argument points to sdev->sdev_dev. Its parent is
552 * sdev->sdev_gendev, which we can use to get the scsi_device pointer.
553 */
554 static int drivetemp_add(struct device *dev, struct class_interface *intf)
555 {
556 struct scsi_device *sdev = to_scsi_device(dev->parent);
557 struct drivetemp_data *st;
558 int err;
559
560 st = kzalloc(sizeof(*st), GFP_KERNEL);
561 if (!st)
562 return -ENOMEM;
563
564 st->sdev = sdev;
565 st->dev = dev;
566 mutex_init(&st->lock);
567
568 if (drivetemp_identify(st)) {
569 err = -ENODEV;
570 goto abort;
571 }
572
573 st->hwdev = hwmon_device_register_with_info(dev->parent, "drivetemp",
574 st, &drivetemp_chip_info,
575 NULL);
576 if (IS_ERR(st->hwdev)) {
577 err = PTR_ERR(st->hwdev);
578 goto abort;
579 }
580
581 list_add(&st->list, &drivetemp_devlist);
582 return 0;
583
584 abort:
585 kfree(st);
586 return err;
587 }
588
589 static void drivetemp_remove(struct device *dev, struct class_interface *intf)
590 {
591 struct drivetemp_data *st, *tmp;
592
593 list_for_each_entry_safe(st, tmp, &drivetemp_devlist, list) {
594 if (st->dev == dev) {
595 list_del(&st->list);
596 hwmon_device_unregister(st->hwdev);
597 kfree(st);
598 break;
599 }
600 }
601 }
602
603 static struct class_interface drivetemp_interface = {
604 .add_dev = drivetemp_add,
605 .remove_dev = drivetemp_remove,
606 };
607
608 static int __init drivetemp_init(void)
609 {
610 return scsi_register_interface(&drivetemp_interface);
611 }
612
613 static void __exit drivetemp_exit(void)
614 {
615 scsi_unregister_interface(&drivetemp_interface);
616 }
617
618 module_init(drivetemp_init);
619 module_exit(drivetemp_exit);
620
621 MODULE_AUTHOR("Guenter Roeck <linus@roeck-us.net>");
622 MODULE_DESCRIPTION("Hard drive temperature monitor");
623 MODULE_LICENSE("GPL");