1 // SPDX-License-Identifier: GPL-2.0
2
3 /*
4 * Copyright 2016-2022 HabanaLabs, Ltd.
5 * All Rights Reserved.
6 */
7
8 #include "habanalabs.h"
9
10 #include <linux/pci.h>
11 #include <linux/types.h>
12
clk_max_freq_mhz_show(struct device * dev,struct device_attribute * attr,char * buf)13 static ssize_t clk_max_freq_mhz_show(struct device *dev, struct device_attribute *attr, char *buf)
14 {
15 struct hl_device *hdev = dev_get_drvdata(dev);
16 long value;
17
18 if (!hl_device_operational(hdev, NULL))
19 return -ENODEV;
20
21 value = hl_fw_get_frequency(hdev, hdev->asic_prop.clk_pll_index, false);
22 if (value < 0)
23 return value;
24
25 hdev->asic_prop.max_freq_value = value;
26
27 return sprintf(buf, "%lu\n", (value / 1000 / 1000));
28 }
29
clk_max_freq_mhz_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)30 static ssize_t clk_max_freq_mhz_store(struct device *dev, struct device_attribute *attr,
31 const char *buf, size_t count)
32 {
33 struct hl_device *hdev = dev_get_drvdata(dev);
34 int rc;
35 u64 value;
36
37 if (!hl_device_operational(hdev, NULL)) {
38 count = -ENODEV;
39 goto fail;
40 }
41
42 rc = kstrtoull(buf, 0, &value);
43 if (rc) {
44 count = -EINVAL;
45 goto fail;
46 }
47
48 hdev->asic_prop.max_freq_value = value * 1000 * 1000;
49
50 hl_fw_set_frequency(hdev, hdev->asic_prop.clk_pll_index, hdev->asic_prop.max_freq_value);
51
52 fail:
53 return count;
54 }
55
clk_cur_freq_mhz_show(struct device * dev,struct device_attribute * attr,char * buf)56 static ssize_t clk_cur_freq_mhz_show(struct device *dev, struct device_attribute *attr, char *buf)
57 {
58 struct hl_device *hdev = dev_get_drvdata(dev);
59 long value;
60
61 if (!hl_device_operational(hdev, NULL))
62 return -ENODEV;
63
64 value = hl_fw_get_frequency(hdev, hdev->asic_prop.clk_pll_index, true);
65 if (value < 0)
66 return value;
67
68 return sprintf(buf, "%lu\n", (value / 1000 / 1000));
69 }
70
71 static DEVICE_ATTR_RW(clk_max_freq_mhz);
72 static DEVICE_ATTR_RO(clk_cur_freq_mhz);
73
74 static struct attribute *hl_dev_clk_attrs[] = {
75 &dev_attr_clk_max_freq_mhz.attr,
76 &dev_attr_clk_cur_freq_mhz.attr,
77 NULL,
78 };
79
vrm_ver_show(struct device * dev,struct device_attribute * attr,char * buf)80 static ssize_t vrm_ver_show(struct device *dev, struct device_attribute *attr, char *buf)
81 {
82 struct hl_device *hdev = dev_get_drvdata(dev);
83 struct cpucp_info *cpucp_info;
84 u32 infineon_second_stage_version;
85 u32 infineon_second_stage_first_instance;
86 u32 infineon_second_stage_second_instance;
87 u32 infineon_second_stage_third_instance;
88 u32 mask = 0xff;
89
90 cpucp_info = &hdev->asic_prop.cpucp_info;
91
92 infineon_second_stage_version = le32_to_cpu(cpucp_info->infineon_second_stage_version);
93 infineon_second_stage_first_instance = infineon_second_stage_version & mask;
94 infineon_second_stage_second_instance =
95 (infineon_second_stage_version >> 8) & mask;
96 infineon_second_stage_third_instance =
97 (infineon_second_stage_version >> 16) & mask;
98
99 if (cpucp_info->infineon_version && cpucp_info->infineon_second_stage_version)
100 return sprintf(buf, "%#04x %#04x:%#04x:%#04x\n",
101 le32_to_cpu(cpucp_info->infineon_version),
102 infineon_second_stage_first_instance,
103 infineon_second_stage_second_instance,
104 infineon_second_stage_third_instance);
105 else if (cpucp_info->infineon_second_stage_version)
106 return sprintf(buf, "%#04x:%#04x:%#04x\n",
107 infineon_second_stage_first_instance,
108 infineon_second_stage_second_instance,
109 infineon_second_stage_third_instance);
110 else if (cpucp_info->infineon_version)
111 return sprintf(buf, "%#04x\n", le32_to_cpu(cpucp_info->infineon_version));
112
113 return 0;
114 }
115
116 static DEVICE_ATTR_RO(vrm_ver);
117
118 static struct attribute *hl_dev_vrm_attrs[] = {
119 &dev_attr_vrm_ver.attr,
120 NULL,
121 };
122
uboot_ver_show(struct device * dev,struct device_attribute * attr,char * buf)123 static ssize_t uboot_ver_show(struct device *dev, struct device_attribute *attr,
124 char *buf)
125 {
126 struct hl_device *hdev = dev_get_drvdata(dev);
127
128 return sprintf(buf, "%s\n", hdev->asic_prop.uboot_ver);
129 }
130
armcp_kernel_ver_show(struct device * dev,struct device_attribute * attr,char * buf)131 static ssize_t armcp_kernel_ver_show(struct device *dev,
132 struct device_attribute *attr, char *buf)
133 {
134 struct hl_device *hdev = dev_get_drvdata(dev);
135
136 return sprintf(buf, "%s", hdev->asic_prop.cpucp_info.kernel_version);
137 }
138
armcp_ver_show(struct device * dev,struct device_attribute * attr,char * buf)139 static ssize_t armcp_ver_show(struct device *dev, struct device_attribute *attr,
140 char *buf)
141 {
142 struct hl_device *hdev = dev_get_drvdata(dev);
143
144 return sprintf(buf, "%s\n", hdev->asic_prop.cpucp_info.cpucp_version);
145 }
146
cpld_ver_show(struct device * dev,struct device_attribute * attr,char * buf)147 static ssize_t cpld_ver_show(struct device *dev, struct device_attribute *attr,
148 char *buf)
149 {
150 struct hl_device *hdev = dev_get_drvdata(dev);
151
152 return sprintf(buf, "0x%08x%08x\n",
153 le32_to_cpu(hdev->asic_prop.cpucp_info.cpld_timestamp),
154 le32_to_cpu(hdev->asic_prop.cpucp_info.cpld_version));
155 }
156
cpucp_kernel_ver_show(struct device * dev,struct device_attribute * attr,char * buf)157 static ssize_t cpucp_kernel_ver_show(struct device *dev,
158 struct device_attribute *attr, char *buf)
159 {
160 struct hl_device *hdev = dev_get_drvdata(dev);
161
162 return sprintf(buf, "%s", hdev->asic_prop.cpucp_info.kernel_version);
163 }
164
cpucp_ver_show(struct device * dev,struct device_attribute * attr,char * buf)165 static ssize_t cpucp_ver_show(struct device *dev, struct device_attribute *attr,
166 char *buf)
167 {
168 struct hl_device *hdev = dev_get_drvdata(dev);
169
170 return sprintf(buf, "%s\n", hdev->asic_prop.cpucp_info.cpucp_version);
171 }
172
fuse_ver_show(struct device * dev,struct device_attribute * attr,char * buf)173 static ssize_t fuse_ver_show(struct device *dev, struct device_attribute *attr,
174 char *buf)
175 {
176 struct hl_device *hdev = dev_get_drvdata(dev);
177
178 return sprintf(buf, "%s\n", hdev->asic_prop.cpucp_info.fuse_version);
179 }
180
thermal_ver_show(struct device * dev,struct device_attribute * attr,char * buf)181 static ssize_t thermal_ver_show(struct device *dev,
182 struct device_attribute *attr, char *buf)
183 {
184 struct hl_device *hdev = dev_get_drvdata(dev);
185
186 return sprintf(buf, "%s", hdev->asic_prop.cpucp_info.thermal_version);
187 }
188
fw_os_ver_show(struct device * dev,struct device_attribute * attr,char * buf)189 static ssize_t fw_os_ver_show(struct device *dev,
190 struct device_attribute *attr, char *buf)
191 {
192 struct hl_device *hdev = dev_get_drvdata(dev);
193
194 return sprintf(buf, "%s", hdev->asic_prop.cpucp_info.fw_os_version);
195 }
196
preboot_btl_ver_show(struct device * dev,struct device_attribute * attr,char * buf)197 static ssize_t preboot_btl_ver_show(struct device *dev,
198 struct device_attribute *attr, char *buf)
199 {
200 struct hl_device *hdev = dev_get_drvdata(dev);
201
202 return sprintf(buf, "%s\n", hdev->asic_prop.preboot_ver);
203 }
204
soft_reset_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)205 static ssize_t soft_reset_store(struct device *dev,
206 struct device_attribute *attr, const char *buf,
207 size_t count)
208 {
209 struct hl_device *hdev = dev_get_drvdata(dev);
210 long value;
211 int rc;
212
213 rc = kstrtoul(buf, 0, &value);
214
215 if (rc) {
216 count = -EINVAL;
217 goto out;
218 }
219
220 if (!hdev->asic_prop.allow_inference_soft_reset) {
221 dev_err(hdev->dev, "Device does not support inference soft-reset\n");
222 goto out;
223 }
224
225 dev_warn(hdev->dev, "Inference Soft-Reset requested through sysfs\n");
226
227 hl_device_reset(hdev, 0);
228
229 out:
230 return count;
231 }
232
hard_reset_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)233 static ssize_t hard_reset_store(struct device *dev,
234 struct device_attribute *attr,
235 const char *buf, size_t count)
236 {
237 struct hl_device *hdev = dev_get_drvdata(dev);
238 long value;
239 int rc;
240
241 rc = kstrtoul(buf, 0, &value);
242
243 if (rc) {
244 count = -EINVAL;
245 goto out;
246 }
247
248 dev_warn(hdev->dev, "Hard-Reset requested through sysfs\n");
249
250 hl_device_reset(hdev, HL_DRV_RESET_HARD);
251
252 out:
253 return count;
254 }
255
device_type_show(struct device * dev,struct device_attribute * attr,char * buf)256 static ssize_t device_type_show(struct device *dev,
257 struct device_attribute *attr, char *buf)
258 {
259 struct hl_device *hdev = dev_get_drvdata(dev);
260 char *str;
261
262 switch (hdev->asic_type) {
263 case ASIC_GOYA:
264 str = "GOYA";
265 break;
266 case ASIC_GAUDI:
267 str = "GAUDI";
268 break;
269 case ASIC_GAUDI_SEC:
270 str = "GAUDI SEC";
271 break;
272 case ASIC_GAUDI2:
273 str = "GAUDI2";
274 break;
275 case ASIC_GAUDI2B:
276 str = "GAUDI2B";
277 break;
278 case ASIC_GAUDI2C:
279 str = "GAUDI2C";
280 break;
281 case ASIC_GAUDI2D:
282 str = "GAUDI2D";
283 break;
284 default:
285 dev_err(hdev->dev, "Unrecognized ASIC type %d\n",
286 hdev->asic_type);
287 return -EINVAL;
288 }
289
290 return sprintf(buf, "%s\n", str);
291 }
292
pci_addr_show(struct device * dev,struct device_attribute * attr,char * buf)293 static ssize_t pci_addr_show(struct device *dev, struct device_attribute *attr,
294 char *buf)
295 {
296 struct hl_device *hdev = dev_get_drvdata(dev);
297
298 return sprintf(buf, "%04x:%02x:%02x.%x\n",
299 pci_domain_nr(hdev->pdev->bus),
300 hdev->pdev->bus->number,
301 PCI_SLOT(hdev->pdev->devfn),
302 PCI_FUNC(hdev->pdev->devfn));
303 }
304
status_show(struct device * dev,struct device_attribute * attr,char * buf)305 static ssize_t status_show(struct device *dev, struct device_attribute *attr,
306 char *buf)
307 {
308 struct hl_device *hdev = dev_get_drvdata(dev);
309 char str[HL_STR_MAX];
310
311 strscpy(str, hdev->status[hl_device_status(hdev)], HL_STR_MAX);
312
313 /* use uppercase for backward compatibility */
314 str[0] = 'A' + (str[0] - 'a');
315
316 return sprintf(buf, "%s\n", str);
317 }
318
soft_reset_cnt_show(struct device * dev,struct device_attribute * attr,char * buf)319 static ssize_t soft_reset_cnt_show(struct device *dev,
320 struct device_attribute *attr, char *buf)
321 {
322 struct hl_device *hdev = dev_get_drvdata(dev);
323
324 return sprintf(buf, "%d\n", hdev->reset_info.compute_reset_cnt);
325 }
326
hard_reset_cnt_show(struct device * dev,struct device_attribute * attr,char * buf)327 static ssize_t hard_reset_cnt_show(struct device *dev,
328 struct device_attribute *attr, char *buf)
329 {
330 struct hl_device *hdev = dev_get_drvdata(dev);
331
332 return sprintf(buf, "%d\n", hdev->reset_info.hard_reset_cnt);
333 }
334
max_power_show(struct device * dev,struct device_attribute * attr,char * buf)335 static ssize_t max_power_show(struct device *dev, struct device_attribute *attr,
336 char *buf)
337 {
338 struct hl_device *hdev = dev_get_drvdata(dev);
339 long val;
340
341 if (!hl_device_operational(hdev, NULL))
342 return -ENODEV;
343
344 val = hl_fw_get_max_power(hdev);
345 if (val < 0)
346 return val;
347
348 return sprintf(buf, "%lu\n", val);
349 }
350
max_power_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)351 static ssize_t max_power_store(struct device *dev,
352 struct device_attribute *attr, const char *buf, size_t count)
353 {
354 struct hl_device *hdev = dev_get_drvdata(dev);
355 unsigned long value;
356 int rc;
357
358 if (!hl_device_operational(hdev, NULL)) {
359 count = -ENODEV;
360 goto out;
361 }
362
363 rc = kstrtoul(buf, 0, &value);
364
365 if (rc) {
366 count = -EINVAL;
367 goto out;
368 }
369
370 hdev->max_power = value;
371 hl_fw_set_max_power(hdev);
372
373 out:
374 return count;
375 }
376
eeprom_read_handler(struct file * filp,struct kobject * kobj,const struct bin_attribute * attr,char * buf,loff_t offset,size_t max_size)377 static ssize_t eeprom_read_handler(struct file *filp, struct kobject *kobj,
378 const struct bin_attribute *attr, char *buf, loff_t offset,
379 size_t max_size)
380 {
381 struct device *dev = kobj_to_dev(kobj);
382 struct hl_device *hdev = dev_get_drvdata(dev);
383 char *data;
384 int rc;
385
386 if (!hl_device_operational(hdev, NULL))
387 return -ENODEV;
388
389 if (!max_size)
390 return -EINVAL;
391
392 data = kzalloc(max_size, GFP_KERNEL);
393 if (!data)
394 return -ENOMEM;
395
396 rc = hdev->asic_funcs->get_eeprom_data(hdev, data, max_size);
397 if (rc)
398 goto out;
399
400 memcpy(buf, data, max_size);
401
402 out:
403 kfree(data);
404
405 return max_size;
406 }
407
security_enabled_show(struct device * dev,struct device_attribute * attr,char * buf)408 static ssize_t security_enabled_show(struct device *dev,
409 struct device_attribute *attr, char *buf)
410 {
411 struct hl_device *hdev = dev_get_drvdata(dev);
412
413 return sprintf(buf, "%d\n", hdev->asic_prop.fw_security_enabled);
414 }
415
module_id_show(struct device * dev,struct device_attribute * attr,char * buf)416 static ssize_t module_id_show(struct device *dev,
417 struct device_attribute *attr, char *buf)
418 {
419 struct hl_device *hdev = dev_get_drvdata(dev);
420
421 return sprintf(buf, "%u\n", le32_to_cpu(hdev->asic_prop.cpucp_info.card_location));
422 }
423
parent_device_show(struct device * dev,struct device_attribute * attr,char * buf)424 static ssize_t parent_device_show(struct device *dev, struct device_attribute *attr, char *buf)
425 {
426 struct hl_device *hdev = dev_get_drvdata(dev);
427
428 return sprintf(buf, "%s\n", HL_DEV_NAME(hdev));
429 }
430
431 static DEVICE_ATTR_RO(armcp_kernel_ver);
432 static DEVICE_ATTR_RO(armcp_ver);
433 static DEVICE_ATTR_RO(cpld_ver);
434 static DEVICE_ATTR_RO(cpucp_kernel_ver);
435 static DEVICE_ATTR_RO(cpucp_ver);
436 static DEVICE_ATTR_RO(device_type);
437 static DEVICE_ATTR_RO(fuse_ver);
438 static DEVICE_ATTR_WO(hard_reset);
439 static DEVICE_ATTR_RO(hard_reset_cnt);
440 static DEVICE_ATTR_RW(max_power);
441 static DEVICE_ATTR_RO(pci_addr);
442 static DEVICE_ATTR_RO(preboot_btl_ver);
443 static DEVICE_ATTR_WO(soft_reset);
444 static DEVICE_ATTR_RO(soft_reset_cnt);
445 static DEVICE_ATTR_RO(status);
446 static DEVICE_ATTR_RO(thermal_ver);
447 static DEVICE_ATTR_RO(uboot_ver);
448 static DEVICE_ATTR_RO(fw_os_ver);
449 static DEVICE_ATTR_RO(security_enabled);
450 static DEVICE_ATTR_RO(module_id);
451 static DEVICE_ATTR_RO(parent_device);
452
453 static const struct bin_attribute bin_attr_eeprom = {
454 .attr = {.name = "eeprom", .mode = (0444)},
455 .size = PAGE_SIZE,
456 .read = eeprom_read_handler
457 };
458
459 static struct attribute *hl_dev_attrs[] = {
460 &dev_attr_armcp_kernel_ver.attr,
461 &dev_attr_armcp_ver.attr,
462 &dev_attr_cpld_ver.attr,
463 &dev_attr_cpucp_kernel_ver.attr,
464 &dev_attr_cpucp_ver.attr,
465 &dev_attr_device_type.attr,
466 &dev_attr_fuse_ver.attr,
467 &dev_attr_hard_reset.attr,
468 &dev_attr_hard_reset_cnt.attr,
469 &dev_attr_max_power.attr,
470 &dev_attr_pci_addr.attr,
471 &dev_attr_preboot_btl_ver.attr,
472 &dev_attr_status.attr,
473 &dev_attr_thermal_ver.attr,
474 &dev_attr_uboot_ver.attr,
475 &dev_attr_fw_os_ver.attr,
476 &dev_attr_security_enabled.attr,
477 &dev_attr_module_id.attr,
478 &dev_attr_parent_device.attr,
479 NULL,
480 };
481
482 static const struct bin_attribute *const hl_dev_bin_attrs[] = {
483 &bin_attr_eeprom,
484 NULL
485 };
486
487 static struct attribute_group hl_dev_attr_group = {
488 .attrs = hl_dev_attrs,
489 .bin_attrs = hl_dev_bin_attrs,
490 };
491
492 static struct attribute_group hl_dev_clks_attr_group;
493 static struct attribute_group hl_dev_vrm_attr_group;
494
495 static const struct attribute_group *hl_dev_attr_groups[] = {
496 &hl_dev_attr_group,
497 &hl_dev_clks_attr_group,
498 &hl_dev_vrm_attr_group,
499 NULL,
500 };
501
502 static struct attribute *hl_dev_inference_attrs[] = {
503 &dev_attr_soft_reset.attr,
504 &dev_attr_soft_reset_cnt.attr,
505 NULL,
506 };
507
508 static struct attribute_group hl_dev_inference_attr_group = {
509 .attrs = hl_dev_inference_attrs,
510 };
511
512 static const struct attribute_group *hl_dev_inference_attr_groups[] = {
513 &hl_dev_inference_attr_group,
514 NULL,
515 };
516
hl_sysfs_add_dev_clk_attr(struct hl_device * hdev,struct attribute_group * dev_clk_attr_grp)517 void hl_sysfs_add_dev_clk_attr(struct hl_device *hdev, struct attribute_group *dev_clk_attr_grp)
518 {
519 dev_clk_attr_grp->attrs = hl_dev_clk_attrs;
520 }
521
hl_sysfs_add_dev_vrm_attr(struct hl_device * hdev,struct attribute_group * dev_vrm_attr_grp)522 void hl_sysfs_add_dev_vrm_attr(struct hl_device *hdev, struct attribute_group *dev_vrm_attr_grp)
523 {
524 dev_vrm_attr_grp->attrs = hl_dev_vrm_attrs;
525 }
526
hl_sysfs_init(struct hl_device * hdev)527 int hl_sysfs_init(struct hl_device *hdev)
528 {
529 int rc;
530
531 hdev->max_power = hdev->asic_prop.max_power_default;
532
533 hdev->asic_funcs->add_device_attr(hdev, &hl_dev_clks_attr_group, &hl_dev_vrm_attr_group);
534
535 rc = device_add_groups(hdev->dev, hl_dev_attr_groups);
536 if (rc) {
537 dev_err(hdev->dev,
538 "Failed to add groups to device, error %d\n", rc);
539 return rc;
540 }
541
542 if (!hdev->asic_prop.allow_inference_soft_reset)
543 return 0;
544
545 rc = device_add_groups(hdev->dev, hl_dev_inference_attr_groups);
546 if (rc) {
547 dev_err(hdev->dev,
548 "Failed to add groups to device, error %d\n", rc);
549 goto remove_groups;
550 }
551
552 return 0;
553
554 remove_groups:
555 device_remove_groups(hdev->dev, hl_dev_attr_groups);
556 return rc;
557 }
558
hl_sysfs_fini(struct hl_device * hdev)559 void hl_sysfs_fini(struct hl_device *hdev)
560 {
561 device_remove_groups(hdev->dev, hl_dev_attr_groups);
562
563 if (!hdev->asic_prop.allow_inference_soft_reset)
564 return;
565
566 device_remove_groups(hdev->dev, hl_dev_inference_attr_groups);
567 }
568