xref: /linux/drivers/gpu/drm/amd/pm/amdgpu_pm.c (revision 7f4f3b14e8079ecde096bd734af10e30d40c27b7)
1 /*
2  * Copyright 2017 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  * Authors: Rafał Miłecki <zajec5@gmail.com>
23  *          Alex Deucher <alexdeucher@gmail.com>
24  */
25 
26 #include "amdgpu.h"
27 #include "amdgpu_drv.h"
28 #include "amdgpu_pm.h"
29 #include "amdgpu_dpm.h"
30 #include "atom.h"
31 #include <linux/pci.h>
32 #include <linux/hwmon.h>
33 #include <linux/hwmon-sysfs.h>
34 #include <linux/nospec.h>
35 #include <linux/pm_runtime.h>
36 #include <asm/processor.h>
37 
38 #define MAX_NUM_OF_FEATURES_PER_SUBSET		8
39 #define MAX_NUM_OF_SUBSETS			8
40 
41 #define DEVICE_ATTR_IS(_name)		(attr_id == device_attr_id__##_name)
42 
43 struct od_attribute {
44 	struct kobj_attribute	attribute;
45 	struct list_head	entry;
46 };
47 
48 struct od_kobj {
49 	struct kobject		kobj;
50 	struct list_head	entry;
51 	struct list_head	attribute;
52 	void			*priv;
53 };
54 
55 struct od_feature_ops {
56 	umode_t (*is_visible)(struct amdgpu_device *adev);
57 	ssize_t (*show)(struct kobject *kobj, struct kobj_attribute *attr,
58 			char *buf);
59 	ssize_t (*store)(struct kobject *kobj, struct kobj_attribute *attr,
60 			 const char *buf, size_t count);
61 };
62 
63 struct od_feature_item {
64 	const char		*name;
65 	struct od_feature_ops	ops;
66 };
67 
68 struct od_feature_container {
69 	char				*name;
70 	struct od_feature_ops		ops;
71 	struct od_feature_item		sub_feature[MAX_NUM_OF_FEATURES_PER_SUBSET];
72 };
73 
74 struct od_feature_set {
75 	struct od_feature_container	containers[MAX_NUM_OF_SUBSETS];
76 };
77 
78 static const struct hwmon_temp_label {
79 	enum PP_HWMON_TEMP channel;
80 	const char *label;
81 } temp_label[] = {
82 	{PP_TEMP_EDGE, "edge"},
83 	{PP_TEMP_JUNCTION, "junction"},
84 	{PP_TEMP_MEM, "mem"},
85 };
86 
87 const char * const amdgpu_pp_profile_name[] = {
88 	"BOOTUP_DEFAULT",
89 	"3D_FULL_SCREEN",
90 	"POWER_SAVING",
91 	"VIDEO",
92 	"VR",
93 	"COMPUTE",
94 	"CUSTOM",
95 	"WINDOW_3D",
96 	"CAPPED",
97 	"UNCAPPED",
98 };
99 
100 /**
101  * DOC: power_dpm_state
102  *
103  * The power_dpm_state file is a legacy interface and is only provided for
104  * backwards compatibility. The amdgpu driver provides a sysfs API for adjusting
105  * certain power related parameters.  The file power_dpm_state is used for this.
106  * It accepts the following arguments:
107  *
108  * - battery
109  *
110  * - balanced
111  *
112  * - performance
113  *
114  * battery
115  *
116  * On older GPUs, the vbios provided a special power state for battery
117  * operation.  Selecting battery switched to this state.  This is no
118  * longer provided on newer GPUs so the option does nothing in that case.
119  *
120  * balanced
121  *
122  * On older GPUs, the vbios provided a special power state for balanced
123  * operation.  Selecting balanced switched to this state.  This is no
124  * longer provided on newer GPUs so the option does nothing in that case.
125  *
126  * performance
127  *
128  * On older GPUs, the vbios provided a special power state for performance
129  * operation.  Selecting performance switched to this state.  This is no
130  * longer provided on newer GPUs so the option does nothing in that case.
131  *
132  */
133 
134 static ssize_t amdgpu_get_power_dpm_state(struct device *dev,
135 					  struct device_attribute *attr,
136 					  char *buf)
137 {
138 	struct drm_device *ddev = dev_get_drvdata(dev);
139 	struct amdgpu_device *adev = drm_to_adev(ddev);
140 	enum amd_pm_state_type pm;
141 	int ret;
142 
143 	if (amdgpu_in_reset(adev))
144 		return -EPERM;
145 	if (adev->in_suspend && !adev->in_runpm)
146 		return -EPERM;
147 
148 	ret = pm_runtime_get_if_active(ddev->dev);
149 	if (ret <= 0)
150 		return ret ?: -EPERM;
151 
152 	amdgpu_dpm_get_current_power_state(adev, &pm);
153 
154 	pm_runtime_put_autosuspend(ddev->dev);
155 
156 	return sysfs_emit(buf, "%s\n",
157 			  (pm == POWER_STATE_TYPE_BATTERY) ? "battery" :
158 			  (pm == POWER_STATE_TYPE_BALANCED) ? "balanced" : "performance");
159 }
160 
161 static ssize_t amdgpu_set_power_dpm_state(struct device *dev,
162 					  struct device_attribute *attr,
163 					  const char *buf,
164 					  size_t count)
165 {
166 	struct drm_device *ddev = dev_get_drvdata(dev);
167 	struct amdgpu_device *adev = drm_to_adev(ddev);
168 	enum amd_pm_state_type  state;
169 	int ret;
170 
171 	if (amdgpu_in_reset(adev))
172 		return -EPERM;
173 	if (adev->in_suspend && !adev->in_runpm)
174 		return -EPERM;
175 
176 	if (strncmp("battery", buf, strlen("battery")) == 0)
177 		state = POWER_STATE_TYPE_BATTERY;
178 	else if (strncmp("balanced", buf, strlen("balanced")) == 0)
179 		state = POWER_STATE_TYPE_BALANCED;
180 	else if (strncmp("performance", buf, strlen("performance")) == 0)
181 		state = POWER_STATE_TYPE_PERFORMANCE;
182 	else
183 		return -EINVAL;
184 
185 	ret = pm_runtime_resume_and_get(ddev->dev);
186 	if (ret < 0)
187 		return ret;
188 
189 	amdgpu_dpm_set_power_state(adev, state);
190 
191 	pm_runtime_mark_last_busy(ddev->dev);
192 	pm_runtime_put_autosuspend(ddev->dev);
193 
194 	return count;
195 }
196 
197 
198 /**
199  * DOC: power_dpm_force_performance_level
200  *
201  * The amdgpu driver provides a sysfs API for adjusting certain power
202  * related parameters.  The file power_dpm_force_performance_level is
203  * used for this.  It accepts the following arguments:
204  *
205  * - auto
206  *
207  * - low
208  *
209  * - high
210  *
211  * - manual
212  *
213  * - profile_standard
214  *
215  * - profile_min_sclk
216  *
217  * - profile_min_mclk
218  *
219  * - profile_peak
220  *
221  * auto
222  *
223  * When auto is selected, the driver will attempt to dynamically select
224  * the optimal power profile for current conditions in the driver.
225  *
226  * low
227  *
228  * When low is selected, the clocks are forced to the lowest power state.
229  *
230  * high
231  *
232  * When high is selected, the clocks are forced to the highest power state.
233  *
234  * manual
235  *
236  * When manual is selected, the user can manually adjust which power states
237  * are enabled for each clock domain via the sysfs pp_dpm_mclk, pp_dpm_sclk,
238  * and pp_dpm_pcie files and adjust the power state transition heuristics
239  * via the pp_power_profile_mode sysfs file.
240  *
241  * profile_standard
242  * profile_min_sclk
243  * profile_min_mclk
244  * profile_peak
245  *
246  * When the profiling modes are selected, clock and power gating are
247  * disabled and the clocks are set for different profiling cases. This
248  * mode is recommended for profiling specific work loads where you do
249  * not want clock or power gating for clock fluctuation to interfere
250  * with your results. profile_standard sets the clocks to a fixed clock
251  * level which varies from asic to asic.  profile_min_sclk forces the sclk
252  * to the lowest level.  profile_min_mclk forces the mclk to the lowest level.
253  * profile_peak sets all clocks (mclk, sclk, pcie) to the highest levels.
254  *
255  */
256 
257 static ssize_t amdgpu_get_power_dpm_force_performance_level(struct device *dev,
258 							    struct device_attribute *attr,
259 							    char *buf)
260 {
261 	struct drm_device *ddev = dev_get_drvdata(dev);
262 	struct amdgpu_device *adev = drm_to_adev(ddev);
263 	enum amd_dpm_forced_level level = 0xff;
264 	int ret;
265 
266 	if (amdgpu_in_reset(adev))
267 		return -EPERM;
268 	if (adev->in_suspend && !adev->in_runpm)
269 		return -EPERM;
270 
271 	ret = pm_runtime_get_if_active(ddev->dev);
272 	if (ret <= 0)
273 		return ret ?: -EPERM;
274 
275 	level = amdgpu_dpm_get_performance_level(adev);
276 
277 	pm_runtime_put_autosuspend(ddev->dev);
278 
279 	return sysfs_emit(buf, "%s\n",
280 			  (level == AMD_DPM_FORCED_LEVEL_AUTO) ? "auto" :
281 			  (level == AMD_DPM_FORCED_LEVEL_LOW) ? "low" :
282 			  (level == AMD_DPM_FORCED_LEVEL_HIGH) ? "high" :
283 			  (level == AMD_DPM_FORCED_LEVEL_MANUAL) ? "manual" :
284 			  (level == AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD) ? "profile_standard" :
285 			  (level == AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK) ? "profile_min_sclk" :
286 			  (level == AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK) ? "profile_min_mclk" :
287 			  (level == AMD_DPM_FORCED_LEVEL_PROFILE_PEAK) ? "profile_peak" :
288 			  (level == AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM) ? "perf_determinism" :
289 			  "unknown");
290 }
291 
292 static ssize_t amdgpu_set_power_dpm_force_performance_level(struct device *dev,
293 							    struct device_attribute *attr,
294 							    const char *buf,
295 							    size_t count)
296 {
297 	struct drm_device *ddev = dev_get_drvdata(dev);
298 	struct amdgpu_device *adev = drm_to_adev(ddev);
299 	enum amd_dpm_forced_level level;
300 	int ret = 0;
301 
302 	if (amdgpu_in_reset(adev))
303 		return -EPERM;
304 	if (adev->in_suspend && !adev->in_runpm)
305 		return -EPERM;
306 
307 	if (strncmp("low", buf, strlen("low")) == 0) {
308 		level = AMD_DPM_FORCED_LEVEL_LOW;
309 	} else if (strncmp("high", buf, strlen("high")) == 0) {
310 		level = AMD_DPM_FORCED_LEVEL_HIGH;
311 	} else if (strncmp("auto", buf, strlen("auto")) == 0) {
312 		level = AMD_DPM_FORCED_LEVEL_AUTO;
313 	} else if (strncmp("manual", buf, strlen("manual")) == 0) {
314 		level = AMD_DPM_FORCED_LEVEL_MANUAL;
315 	} else if (strncmp("profile_exit", buf, strlen("profile_exit")) == 0) {
316 		level = AMD_DPM_FORCED_LEVEL_PROFILE_EXIT;
317 	} else if (strncmp("profile_standard", buf, strlen("profile_standard")) == 0) {
318 		level = AMD_DPM_FORCED_LEVEL_PROFILE_STANDARD;
319 	} else if (strncmp("profile_min_sclk", buf, strlen("profile_min_sclk")) == 0) {
320 		level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_SCLK;
321 	} else if (strncmp("profile_min_mclk", buf, strlen("profile_min_mclk")) == 0) {
322 		level = AMD_DPM_FORCED_LEVEL_PROFILE_MIN_MCLK;
323 	} else if (strncmp("profile_peak", buf, strlen("profile_peak")) == 0) {
324 		level = AMD_DPM_FORCED_LEVEL_PROFILE_PEAK;
325 	} else if (strncmp("perf_determinism", buf, strlen("perf_determinism")) == 0) {
326 		level = AMD_DPM_FORCED_LEVEL_PERF_DETERMINISM;
327 	}  else {
328 		return -EINVAL;
329 	}
330 
331 	ret = pm_runtime_resume_and_get(ddev->dev);
332 	if (ret < 0)
333 		return ret;
334 
335 	mutex_lock(&adev->pm.stable_pstate_ctx_lock);
336 	if (amdgpu_dpm_force_performance_level(adev, level)) {
337 		pm_runtime_mark_last_busy(ddev->dev);
338 		pm_runtime_put_autosuspend(ddev->dev);
339 		mutex_unlock(&adev->pm.stable_pstate_ctx_lock);
340 		return -EINVAL;
341 	}
342 	/* override whatever a user ctx may have set */
343 	adev->pm.stable_pstate_ctx = NULL;
344 	mutex_unlock(&adev->pm.stable_pstate_ctx_lock);
345 
346 	pm_runtime_mark_last_busy(ddev->dev);
347 	pm_runtime_put_autosuspend(ddev->dev);
348 
349 	return count;
350 }
351 
352 static ssize_t amdgpu_get_pp_num_states(struct device *dev,
353 		struct device_attribute *attr,
354 		char *buf)
355 {
356 	struct drm_device *ddev = dev_get_drvdata(dev);
357 	struct amdgpu_device *adev = drm_to_adev(ddev);
358 	struct pp_states_info data;
359 	uint32_t i;
360 	int buf_len, ret;
361 
362 	if (amdgpu_in_reset(adev))
363 		return -EPERM;
364 	if (adev->in_suspend && !adev->in_runpm)
365 		return -EPERM;
366 
367 	ret = pm_runtime_get_if_active(ddev->dev);
368 	if (ret <= 0)
369 		return ret ?: -EPERM;
370 
371 	if (amdgpu_dpm_get_pp_num_states(adev, &data))
372 		memset(&data, 0, sizeof(data));
373 
374 	pm_runtime_put_autosuspend(ddev->dev);
375 
376 	buf_len = sysfs_emit(buf, "states: %d\n", data.nums);
377 	for (i = 0; i < data.nums; i++)
378 		buf_len += sysfs_emit_at(buf, buf_len, "%d %s\n", i,
379 				(data.states[i] == POWER_STATE_TYPE_INTERNAL_BOOT) ? "boot" :
380 				(data.states[i] == POWER_STATE_TYPE_BATTERY) ? "battery" :
381 				(data.states[i] == POWER_STATE_TYPE_BALANCED) ? "balanced" :
382 				(data.states[i] == POWER_STATE_TYPE_PERFORMANCE) ? "performance" : "default");
383 
384 	return buf_len;
385 }
386 
387 static ssize_t amdgpu_get_pp_cur_state(struct device *dev,
388 		struct device_attribute *attr,
389 		char *buf)
390 {
391 	struct drm_device *ddev = dev_get_drvdata(dev);
392 	struct amdgpu_device *adev = drm_to_adev(ddev);
393 	struct pp_states_info data = {0};
394 	enum amd_pm_state_type pm = 0;
395 	int i = 0, ret = 0;
396 
397 	if (amdgpu_in_reset(adev))
398 		return -EPERM;
399 	if (adev->in_suspend && !adev->in_runpm)
400 		return -EPERM;
401 
402 	ret = pm_runtime_get_if_active(ddev->dev);
403 	if (ret <= 0)
404 		return ret ?: -EPERM;
405 
406 	amdgpu_dpm_get_current_power_state(adev, &pm);
407 
408 	ret = amdgpu_dpm_get_pp_num_states(adev, &data);
409 
410 	pm_runtime_put_autosuspend(ddev->dev);
411 
412 	if (ret)
413 		return ret;
414 
415 	for (i = 0; i < data.nums; i++) {
416 		if (pm == data.states[i])
417 			break;
418 	}
419 
420 	if (i == data.nums)
421 		i = -EINVAL;
422 
423 	return sysfs_emit(buf, "%d\n", i);
424 }
425 
426 static ssize_t amdgpu_get_pp_force_state(struct device *dev,
427 		struct device_attribute *attr,
428 		char *buf)
429 {
430 	struct drm_device *ddev = dev_get_drvdata(dev);
431 	struct amdgpu_device *adev = drm_to_adev(ddev);
432 
433 	if (amdgpu_in_reset(adev))
434 		return -EPERM;
435 	if (adev->in_suspend && !adev->in_runpm)
436 		return -EPERM;
437 
438 	if (adev->pm.pp_force_state_enabled)
439 		return amdgpu_get_pp_cur_state(dev, attr, buf);
440 	else
441 		return sysfs_emit(buf, "\n");
442 }
443 
444 static ssize_t amdgpu_set_pp_force_state(struct device *dev,
445 		struct device_attribute *attr,
446 		const char *buf,
447 		size_t count)
448 {
449 	struct drm_device *ddev = dev_get_drvdata(dev);
450 	struct amdgpu_device *adev = drm_to_adev(ddev);
451 	enum amd_pm_state_type state = 0;
452 	struct pp_states_info data;
453 	unsigned long idx;
454 	int ret;
455 
456 	if (amdgpu_in_reset(adev))
457 		return -EPERM;
458 	if (adev->in_suspend && !adev->in_runpm)
459 		return -EPERM;
460 
461 	adev->pm.pp_force_state_enabled = false;
462 
463 	if (strlen(buf) == 1)
464 		return count;
465 
466 	ret = kstrtoul(buf, 0, &idx);
467 	if (ret || idx >= ARRAY_SIZE(data.states))
468 		return -EINVAL;
469 
470 	idx = array_index_nospec(idx, ARRAY_SIZE(data.states));
471 
472 	ret = pm_runtime_resume_and_get(ddev->dev);
473 	if (ret < 0)
474 		return ret;
475 
476 	ret = amdgpu_dpm_get_pp_num_states(adev, &data);
477 	if (ret)
478 		goto err_out;
479 
480 	state = data.states[idx];
481 
482 	/* only set user selected power states */
483 	if (state != POWER_STATE_TYPE_INTERNAL_BOOT &&
484 	    state != POWER_STATE_TYPE_DEFAULT) {
485 		ret = amdgpu_dpm_dispatch_task(adev,
486 				AMD_PP_TASK_ENABLE_USER_STATE, &state);
487 		if (ret)
488 			goto err_out;
489 
490 		adev->pm.pp_force_state_enabled = true;
491 	}
492 
493 	pm_runtime_mark_last_busy(ddev->dev);
494 	pm_runtime_put_autosuspend(ddev->dev);
495 
496 	return count;
497 
498 err_out:
499 	pm_runtime_mark_last_busy(ddev->dev);
500 	pm_runtime_put_autosuspend(ddev->dev);
501 	return ret;
502 }
503 
504 /**
505  * DOC: pp_table
506  *
507  * The amdgpu driver provides a sysfs API for uploading new powerplay
508  * tables.  The file pp_table is used for this.  Reading the file
509  * will dump the current power play table.  Writing to the file
510  * will attempt to upload a new powerplay table and re-initialize
511  * powerplay using that new table.
512  *
513  */
514 
515 static ssize_t amdgpu_get_pp_table(struct device *dev,
516 		struct device_attribute *attr,
517 		char *buf)
518 {
519 	struct drm_device *ddev = dev_get_drvdata(dev);
520 	struct amdgpu_device *adev = drm_to_adev(ddev);
521 	char *table = NULL;
522 	int size, ret;
523 
524 	if (amdgpu_in_reset(adev))
525 		return -EPERM;
526 	if (adev->in_suspend && !adev->in_runpm)
527 		return -EPERM;
528 
529 	ret = pm_runtime_get_if_active(ddev->dev);
530 	if (ret <= 0)
531 		return ret ?: -EPERM;
532 
533 	size = amdgpu_dpm_get_pp_table(adev, &table);
534 
535 	pm_runtime_put_autosuspend(ddev->dev);
536 
537 	if (size <= 0)
538 		return size;
539 
540 	if (size >= PAGE_SIZE)
541 		size = PAGE_SIZE - 1;
542 
543 	memcpy(buf, table, size);
544 
545 	return size;
546 }
547 
548 static ssize_t amdgpu_set_pp_table(struct device *dev,
549 		struct device_attribute *attr,
550 		const char *buf,
551 		size_t count)
552 {
553 	struct drm_device *ddev = dev_get_drvdata(dev);
554 	struct amdgpu_device *adev = drm_to_adev(ddev);
555 	int ret = 0;
556 
557 	if (amdgpu_in_reset(adev))
558 		return -EPERM;
559 	if (adev->in_suspend && !adev->in_runpm)
560 		return -EPERM;
561 
562 	ret = pm_runtime_resume_and_get(ddev->dev);
563 	if (ret < 0)
564 		return ret;
565 
566 	ret = amdgpu_dpm_set_pp_table(adev, buf, count);
567 
568 	pm_runtime_mark_last_busy(ddev->dev);
569 	pm_runtime_put_autosuspend(ddev->dev);
570 
571 	if (ret)
572 		return ret;
573 
574 	return count;
575 }
576 
577 /**
578  * DOC: pp_od_clk_voltage
579  *
580  * The amdgpu driver provides a sysfs API for adjusting the clocks and voltages
581  * in each power level within a power state.  The pp_od_clk_voltage is used for
582  * this.
583  *
584  * Note that the actual memory controller clock rate are exposed, not
585  * the effective memory clock of the DRAMs. To translate it, use the
586  * following formula:
587  *
588  * Clock conversion (Mhz):
589  *
590  * HBM: effective_memory_clock = memory_controller_clock * 1
591  *
592  * G5: effective_memory_clock = memory_controller_clock * 1
593  *
594  * G6: effective_memory_clock = memory_controller_clock * 2
595  *
596  * DRAM data rate (MT/s):
597  *
598  * HBM: effective_memory_clock * 2 = data_rate
599  *
600  * G5: effective_memory_clock * 4 = data_rate
601  *
602  * G6: effective_memory_clock * 8 = data_rate
603  *
604  * Bandwidth (MB/s):
605  *
606  * data_rate * vram_bit_width / 8 = memory_bandwidth
607  *
608  * Some examples:
609  *
610  * G5 on RX460:
611  *
612  * memory_controller_clock = 1750 Mhz
613  *
614  * effective_memory_clock = 1750 Mhz * 1 = 1750 Mhz
615  *
616  * data rate = 1750 * 4 = 7000 MT/s
617  *
618  * memory_bandwidth = 7000 * 128 bits / 8 = 112000 MB/s
619  *
620  * G6 on RX5700:
621  *
622  * memory_controller_clock = 875 Mhz
623  *
624  * effective_memory_clock = 875 Mhz * 2 = 1750 Mhz
625  *
626  * data rate = 1750 * 8 = 14000 MT/s
627  *
628  * memory_bandwidth = 14000 * 256 bits / 8 = 448000 MB/s
629  *
630  * < For Vega10 and previous ASICs >
631  *
632  * Reading the file will display:
633  *
634  * - a list of engine clock levels and voltages labeled OD_SCLK
635  *
636  * - a list of memory clock levels and voltages labeled OD_MCLK
637  *
638  * - a list of valid ranges for sclk, mclk, and voltage labeled OD_RANGE
639  *
640  * To manually adjust these settings, first select manual using
641  * power_dpm_force_performance_level. Enter a new value for each
642  * level by writing a string that contains "s/m level clock voltage" to
643  * the file.  E.g., "s 1 500 820" will update sclk level 1 to be 500 MHz
644  * at 820 mV; "m 0 350 810" will update mclk level 0 to be 350 MHz at
645  * 810 mV.  When you have edited all of the states as needed, write
646  * "c" (commit) to the file to commit your changes.  If you want to reset to the
647  * default power levels, write "r" (reset) to the file to reset them.
648  *
649  *
650  * < For Vega20 and newer ASICs >
651  *
652  * Reading the file will display:
653  *
654  * - minimum and maximum engine clock labeled OD_SCLK
655  *
656  * - minimum(not available for Vega20 and Navi1x) and maximum memory
657  *   clock labeled OD_MCLK
658  *
659  * - three <frequency, voltage> points labeled OD_VDDC_CURVE.
660  *   They can be used to calibrate the sclk voltage curve. This is
661  *   available for Vega20 and NV1X.
662  *
663  * - voltage offset(in mV) applied on target voltage calculation.
664  *   This is available for Sienna Cichlid, Navy Flounder, Dimgrey
665  *   Cavefish and some later SMU13 ASICs. For these ASICs, the target
666  *   voltage calculation can be illustrated by "voltage = voltage
667  *   calculated from v/f curve + overdrive vddgfx offset"
668  *
669  * - a list of valid ranges for sclk, mclk, voltage curve points
670  *   or voltage offset labeled OD_RANGE
671  *
672  * < For APUs >
673  *
674  * Reading the file will display:
675  *
676  * - minimum and maximum engine clock labeled OD_SCLK
677  *
678  * - a list of valid ranges for sclk labeled OD_RANGE
679  *
680  * < For VanGogh >
681  *
682  * Reading the file will display:
683  *
684  * - minimum and maximum engine clock labeled OD_SCLK
685  * - minimum and maximum core clocks labeled OD_CCLK
686  *
687  * - a list of valid ranges for sclk and cclk labeled OD_RANGE
688  *
689  * To manually adjust these settings:
690  *
691  * - First select manual using power_dpm_force_performance_level
692  *
693  * - For clock frequency setting, enter a new value by writing a
694  *   string that contains "s/m index clock" to the file. The index
695  *   should be 0 if to set minimum clock. And 1 if to set maximum
696  *   clock. E.g., "s 0 500" will update minimum sclk to be 500 MHz.
697  *   "m 1 800" will update maximum mclk to be 800Mhz. For core
698  *   clocks on VanGogh, the string contains "p core index clock".
699  *   E.g., "p 2 0 800" would set the minimum core clock on core
700  *   2 to 800Mhz.
701  *
702  *   For sclk voltage curve supported by Vega20 and NV1X, enter the new
703  *   values by writing a string that contains "vc point clock voltage"
704  *   to the file. The points are indexed by 0, 1 and 2. E.g., "vc 0 300
705  *   600" will update point1 with clock set as 300Mhz and voltage as 600mV.
706  *   "vc 2 1000 1000" will update point3 with clock set as 1000Mhz and
707  *   voltage 1000mV.
708  *
709  *   For voltage offset supported by Sienna Cichlid, Navy Flounder, Dimgrey
710  *   Cavefish and some later SMU13 ASICs, enter the new value by writing a
711  *   string that contains "vo offset". E.g., "vo -10" will update the extra
712  *   voltage offset applied to the whole v/f curve line as -10mv.
713  *
714  * - When you have edited all of the states as needed, write "c" (commit)
715  *   to the file to commit your changes
716  *
717  * - If you want to reset to the default power levels, write "r" (reset)
718  *   to the file to reset them
719  *
720  */
721 
722 static ssize_t amdgpu_set_pp_od_clk_voltage(struct device *dev,
723 		struct device_attribute *attr,
724 		const char *buf,
725 		size_t count)
726 {
727 	struct drm_device *ddev = dev_get_drvdata(dev);
728 	struct amdgpu_device *adev = drm_to_adev(ddev);
729 	int ret;
730 	uint32_t parameter_size = 0;
731 	long parameter[64];
732 	char buf_cpy[128];
733 	char *tmp_str;
734 	char *sub_str;
735 	const char delimiter[3] = {' ', '\n', '\0'};
736 	uint32_t type;
737 
738 	if (amdgpu_in_reset(adev))
739 		return -EPERM;
740 	if (adev->in_suspend && !adev->in_runpm)
741 		return -EPERM;
742 
743 	if (count > 127 || count == 0)
744 		return -EINVAL;
745 
746 	if (*buf == 's')
747 		type = PP_OD_EDIT_SCLK_VDDC_TABLE;
748 	else if (*buf == 'p')
749 		type = PP_OD_EDIT_CCLK_VDDC_TABLE;
750 	else if (*buf == 'm')
751 		type = PP_OD_EDIT_MCLK_VDDC_TABLE;
752 	else if (*buf == 'r')
753 		type = PP_OD_RESTORE_DEFAULT_TABLE;
754 	else if (*buf == 'c')
755 		type = PP_OD_COMMIT_DPM_TABLE;
756 	else if (!strncmp(buf, "vc", 2))
757 		type = PP_OD_EDIT_VDDC_CURVE;
758 	else if (!strncmp(buf, "vo", 2))
759 		type = PP_OD_EDIT_VDDGFX_OFFSET;
760 	else
761 		return -EINVAL;
762 
763 	memcpy(buf_cpy, buf, count);
764 	buf_cpy[count] = 0;
765 
766 	tmp_str = buf_cpy;
767 
768 	if ((type == PP_OD_EDIT_VDDC_CURVE) ||
769 	     (type == PP_OD_EDIT_VDDGFX_OFFSET))
770 		tmp_str++;
771 	while (isspace(*++tmp_str));
772 
773 	while ((sub_str = strsep(&tmp_str, delimiter)) != NULL) {
774 		if (strlen(sub_str) == 0)
775 			continue;
776 		ret = kstrtol(sub_str, 0, &parameter[parameter_size]);
777 		if (ret)
778 			return -EINVAL;
779 		parameter_size++;
780 
781 		if (!tmp_str)
782 			break;
783 
784 		while (isspace(*tmp_str))
785 			tmp_str++;
786 	}
787 
788 	ret = pm_runtime_resume_and_get(ddev->dev);
789 	if (ret < 0)
790 		return ret;
791 
792 	if (amdgpu_dpm_set_fine_grain_clk_vol(adev,
793 					      type,
794 					      parameter,
795 					      parameter_size))
796 		goto err_out;
797 
798 	if (amdgpu_dpm_odn_edit_dpm_table(adev, type,
799 					  parameter, parameter_size))
800 		goto err_out;
801 
802 	if (type == PP_OD_COMMIT_DPM_TABLE) {
803 		if (amdgpu_dpm_dispatch_task(adev,
804 					     AMD_PP_TASK_READJUST_POWER_STATE,
805 					     NULL))
806 			goto err_out;
807 	}
808 
809 	pm_runtime_mark_last_busy(ddev->dev);
810 	pm_runtime_put_autosuspend(ddev->dev);
811 
812 	return count;
813 
814 err_out:
815 	pm_runtime_mark_last_busy(ddev->dev);
816 	pm_runtime_put_autosuspend(ddev->dev);
817 	return -EINVAL;
818 }
819 
820 static ssize_t amdgpu_get_pp_od_clk_voltage(struct device *dev,
821 		struct device_attribute *attr,
822 		char *buf)
823 {
824 	struct drm_device *ddev = dev_get_drvdata(dev);
825 	struct amdgpu_device *adev = drm_to_adev(ddev);
826 	int size = 0;
827 	int ret;
828 	enum pp_clock_type od_clocks[6] = {
829 		OD_SCLK,
830 		OD_MCLK,
831 		OD_VDDC_CURVE,
832 		OD_RANGE,
833 		OD_VDDGFX_OFFSET,
834 		OD_CCLK,
835 	};
836 	uint clk_index;
837 
838 	if (amdgpu_in_reset(adev))
839 		return -EPERM;
840 	if (adev->in_suspend && !adev->in_runpm)
841 		return -EPERM;
842 
843 	ret = pm_runtime_get_if_active(ddev->dev);
844 	if (ret <= 0)
845 		return ret ?: -EPERM;
846 
847 	for (clk_index = 0 ; clk_index < 6 ; clk_index++) {
848 		ret = amdgpu_dpm_emit_clock_levels(adev, od_clocks[clk_index], buf, &size);
849 		if (ret)
850 			break;
851 	}
852 	if (ret == -ENOENT) {
853 		size = amdgpu_dpm_print_clock_levels(adev, OD_SCLK, buf);
854 		size += amdgpu_dpm_print_clock_levels(adev, OD_MCLK, buf + size);
855 		size += amdgpu_dpm_print_clock_levels(adev, OD_VDDC_CURVE, buf + size);
856 		size += amdgpu_dpm_print_clock_levels(adev, OD_VDDGFX_OFFSET, buf + size);
857 		size += amdgpu_dpm_print_clock_levels(adev, OD_RANGE, buf + size);
858 		size += amdgpu_dpm_print_clock_levels(adev, OD_CCLK, buf + size);
859 	}
860 
861 	if (size == 0)
862 		size = sysfs_emit(buf, "\n");
863 
864 	pm_runtime_put_autosuspend(ddev->dev);
865 
866 	return size;
867 }
868 
869 /**
870  * DOC: pp_features
871  *
872  * The amdgpu driver provides a sysfs API for adjusting what powerplay
873  * features to be enabled. The file pp_features is used for this. And
874  * this is only available for Vega10 and later dGPUs.
875  *
876  * Reading back the file will show you the followings:
877  * - Current ppfeature masks
878  * - List of the all supported powerplay features with their naming,
879  *   bitmasks and enablement status('Y'/'N' means "enabled"/"disabled").
880  *
881  * To manually enable or disable a specific feature, just set or clear
882  * the corresponding bit from original ppfeature masks and input the
883  * new ppfeature masks.
884  */
885 static ssize_t amdgpu_set_pp_features(struct device *dev,
886 				      struct device_attribute *attr,
887 				      const char *buf,
888 				      size_t count)
889 {
890 	struct drm_device *ddev = dev_get_drvdata(dev);
891 	struct amdgpu_device *adev = drm_to_adev(ddev);
892 	uint64_t featuremask;
893 	int ret;
894 
895 	if (amdgpu_in_reset(adev))
896 		return -EPERM;
897 	if (adev->in_suspend && !adev->in_runpm)
898 		return -EPERM;
899 
900 	ret = kstrtou64(buf, 0, &featuremask);
901 	if (ret)
902 		return -EINVAL;
903 
904 	ret = pm_runtime_resume_and_get(ddev->dev);
905 	if (ret < 0)
906 		return ret;
907 
908 	ret = amdgpu_dpm_set_ppfeature_status(adev, featuremask);
909 
910 	pm_runtime_mark_last_busy(ddev->dev);
911 	pm_runtime_put_autosuspend(ddev->dev);
912 
913 	if (ret)
914 		return -EINVAL;
915 
916 	return count;
917 }
918 
919 static ssize_t amdgpu_get_pp_features(struct device *dev,
920 				      struct device_attribute *attr,
921 				      char *buf)
922 {
923 	struct drm_device *ddev = dev_get_drvdata(dev);
924 	struct amdgpu_device *adev = drm_to_adev(ddev);
925 	ssize_t size;
926 	int ret;
927 
928 	if (amdgpu_in_reset(adev))
929 		return -EPERM;
930 	if (adev->in_suspend && !adev->in_runpm)
931 		return -EPERM;
932 
933 	ret = pm_runtime_get_if_active(ddev->dev);
934 	if (ret <= 0)
935 		return ret ?: -EPERM;
936 
937 	size = amdgpu_dpm_get_ppfeature_status(adev, buf);
938 	if (size <= 0)
939 		size = sysfs_emit(buf, "\n");
940 
941 	pm_runtime_put_autosuspend(ddev->dev);
942 
943 	return size;
944 }
945 
946 /**
947  * DOC: pp_dpm_sclk pp_dpm_mclk pp_dpm_socclk pp_dpm_fclk pp_dpm_dcefclk pp_dpm_pcie
948  *
949  * The amdgpu driver provides a sysfs API for adjusting what power levels
950  * are enabled for a given power state.  The files pp_dpm_sclk, pp_dpm_mclk,
951  * pp_dpm_socclk, pp_dpm_fclk, pp_dpm_dcefclk and pp_dpm_pcie are used for
952  * this.
953  *
954  * pp_dpm_socclk and pp_dpm_dcefclk interfaces are only available for
955  * Vega10 and later ASICs.
956  * pp_dpm_fclk interface is only available for Vega20 and later ASICs.
957  *
958  * Reading back the files will show you the available power levels within
959  * the power state and the clock information for those levels. If deep sleep is
960  * applied to a clock, the level will be denoted by a special level 'S:'
961  * E.g., ::
962  *
963  *  S: 19Mhz *
964  *  0: 615Mhz
965  *  1: 800Mhz
966  *  2: 888Mhz
967  *  3: 1000Mhz
968  *
969  *
970  * To manually adjust these states, first select manual using
971  * power_dpm_force_performance_level.
972  * Secondly, enter a new value for each level by inputing a string that
973  * contains " echo xx xx xx > pp_dpm_sclk/mclk/pcie"
974  * E.g.,
975  *
976  * .. code-block:: bash
977  *
978  *	echo "4 5 6" > pp_dpm_sclk
979  *
980  * will enable sclk levels 4, 5, and 6.
981  *
982  * NOTE: change to the dcefclk max dpm level is not supported now
983  */
984 
985 static ssize_t amdgpu_get_pp_dpm_clock(struct device *dev,
986 		enum pp_clock_type type,
987 		char *buf)
988 {
989 	struct drm_device *ddev = dev_get_drvdata(dev);
990 	struct amdgpu_device *adev = drm_to_adev(ddev);
991 	int size = 0;
992 	int ret = 0;
993 
994 	if (amdgpu_in_reset(adev))
995 		return -EPERM;
996 	if (adev->in_suspend && !adev->in_runpm)
997 		return -EPERM;
998 
999 	ret = pm_runtime_get_if_active(ddev->dev);
1000 	if (ret <= 0)
1001 		return ret ?: -EPERM;
1002 
1003 	ret = amdgpu_dpm_emit_clock_levels(adev, type, buf, &size);
1004 	if (ret == -ENOENT)
1005 		size = amdgpu_dpm_print_clock_levels(adev, type, buf);
1006 
1007 	if (size == 0)
1008 		size = sysfs_emit(buf, "\n");
1009 
1010 	pm_runtime_put_autosuspend(ddev->dev);
1011 
1012 	return size;
1013 }
1014 
1015 /*
1016  * Worst case: 32 bits individually specified, in octal at 12 characters
1017  * per line (+1 for \n).
1018  */
1019 #define AMDGPU_MASK_BUF_MAX	(32 * 13)
1020 
1021 static ssize_t amdgpu_read_mask(const char *buf, size_t count, uint32_t *mask)
1022 {
1023 	int ret;
1024 	unsigned long level;
1025 	char *sub_str = NULL;
1026 	char *tmp;
1027 	char buf_cpy[AMDGPU_MASK_BUF_MAX + 1];
1028 	const char delimiter[3] = {' ', '\n', '\0'};
1029 	size_t bytes;
1030 
1031 	*mask = 0;
1032 
1033 	bytes = min(count, sizeof(buf_cpy) - 1);
1034 	memcpy(buf_cpy, buf, bytes);
1035 	buf_cpy[bytes] = '\0';
1036 	tmp = buf_cpy;
1037 	while ((sub_str = strsep(&tmp, delimiter)) != NULL) {
1038 		if (strlen(sub_str)) {
1039 			ret = kstrtoul(sub_str, 0, &level);
1040 			if (ret || level > 31)
1041 				return -EINVAL;
1042 			*mask |= 1 << level;
1043 		} else
1044 			break;
1045 	}
1046 
1047 	return 0;
1048 }
1049 
1050 static ssize_t amdgpu_set_pp_dpm_clock(struct device *dev,
1051 		enum pp_clock_type type,
1052 		const char *buf,
1053 		size_t count)
1054 {
1055 	struct drm_device *ddev = dev_get_drvdata(dev);
1056 	struct amdgpu_device *adev = drm_to_adev(ddev);
1057 	int ret;
1058 	uint32_t mask = 0;
1059 
1060 	if (amdgpu_in_reset(adev))
1061 		return -EPERM;
1062 	if (adev->in_suspend && !adev->in_runpm)
1063 		return -EPERM;
1064 
1065 	ret = amdgpu_read_mask(buf, count, &mask);
1066 	if (ret)
1067 		return ret;
1068 
1069 	ret = pm_runtime_resume_and_get(ddev->dev);
1070 	if (ret < 0)
1071 		return ret;
1072 
1073 	ret = amdgpu_dpm_force_clock_level(adev, type, mask);
1074 
1075 	pm_runtime_mark_last_busy(ddev->dev);
1076 	pm_runtime_put_autosuspend(ddev->dev);
1077 
1078 	if (ret)
1079 		return -EINVAL;
1080 
1081 	return count;
1082 }
1083 
1084 static ssize_t amdgpu_get_pp_dpm_sclk(struct device *dev,
1085 		struct device_attribute *attr,
1086 		char *buf)
1087 {
1088 	return amdgpu_get_pp_dpm_clock(dev, PP_SCLK, buf);
1089 }
1090 
1091 static ssize_t amdgpu_set_pp_dpm_sclk(struct device *dev,
1092 		struct device_attribute *attr,
1093 		const char *buf,
1094 		size_t count)
1095 {
1096 	return amdgpu_set_pp_dpm_clock(dev, PP_SCLK, buf, count);
1097 }
1098 
1099 static ssize_t amdgpu_get_pp_dpm_mclk(struct device *dev,
1100 		struct device_attribute *attr,
1101 		char *buf)
1102 {
1103 	return amdgpu_get_pp_dpm_clock(dev, PP_MCLK, buf);
1104 }
1105 
1106 static ssize_t amdgpu_set_pp_dpm_mclk(struct device *dev,
1107 		struct device_attribute *attr,
1108 		const char *buf,
1109 		size_t count)
1110 {
1111 	return amdgpu_set_pp_dpm_clock(dev, PP_MCLK, buf, count);
1112 }
1113 
1114 static ssize_t amdgpu_get_pp_dpm_socclk(struct device *dev,
1115 		struct device_attribute *attr,
1116 		char *buf)
1117 {
1118 	return amdgpu_get_pp_dpm_clock(dev, PP_SOCCLK, buf);
1119 }
1120 
1121 static ssize_t amdgpu_set_pp_dpm_socclk(struct device *dev,
1122 		struct device_attribute *attr,
1123 		const char *buf,
1124 		size_t count)
1125 {
1126 	return amdgpu_set_pp_dpm_clock(dev, PP_SOCCLK, buf, count);
1127 }
1128 
1129 static ssize_t amdgpu_get_pp_dpm_fclk(struct device *dev,
1130 		struct device_attribute *attr,
1131 		char *buf)
1132 {
1133 	return amdgpu_get_pp_dpm_clock(dev, PP_FCLK, buf);
1134 }
1135 
1136 static ssize_t amdgpu_set_pp_dpm_fclk(struct device *dev,
1137 		struct device_attribute *attr,
1138 		const char *buf,
1139 		size_t count)
1140 {
1141 	return amdgpu_set_pp_dpm_clock(dev, PP_FCLK, buf, count);
1142 }
1143 
1144 static ssize_t amdgpu_get_pp_dpm_vclk(struct device *dev,
1145 		struct device_attribute *attr,
1146 		char *buf)
1147 {
1148 	return amdgpu_get_pp_dpm_clock(dev, PP_VCLK, buf);
1149 }
1150 
1151 static ssize_t amdgpu_set_pp_dpm_vclk(struct device *dev,
1152 		struct device_attribute *attr,
1153 		const char *buf,
1154 		size_t count)
1155 {
1156 	return amdgpu_set_pp_dpm_clock(dev, PP_VCLK, buf, count);
1157 }
1158 
1159 static ssize_t amdgpu_get_pp_dpm_vclk1(struct device *dev,
1160 		struct device_attribute *attr,
1161 		char *buf)
1162 {
1163 	return amdgpu_get_pp_dpm_clock(dev, PP_VCLK1, buf);
1164 }
1165 
1166 static ssize_t amdgpu_set_pp_dpm_vclk1(struct device *dev,
1167 		struct device_attribute *attr,
1168 		const char *buf,
1169 		size_t count)
1170 {
1171 	return amdgpu_set_pp_dpm_clock(dev, PP_VCLK1, buf, count);
1172 }
1173 
1174 static ssize_t amdgpu_get_pp_dpm_dclk(struct device *dev,
1175 		struct device_attribute *attr,
1176 		char *buf)
1177 {
1178 	return amdgpu_get_pp_dpm_clock(dev, PP_DCLK, buf);
1179 }
1180 
1181 static ssize_t amdgpu_set_pp_dpm_dclk(struct device *dev,
1182 		struct device_attribute *attr,
1183 		const char *buf,
1184 		size_t count)
1185 {
1186 	return amdgpu_set_pp_dpm_clock(dev, PP_DCLK, buf, count);
1187 }
1188 
1189 static ssize_t amdgpu_get_pp_dpm_dclk1(struct device *dev,
1190 		struct device_attribute *attr,
1191 		char *buf)
1192 {
1193 	return amdgpu_get_pp_dpm_clock(dev, PP_DCLK1, buf);
1194 }
1195 
1196 static ssize_t amdgpu_set_pp_dpm_dclk1(struct device *dev,
1197 		struct device_attribute *attr,
1198 		const char *buf,
1199 		size_t count)
1200 {
1201 	return amdgpu_set_pp_dpm_clock(dev, PP_DCLK1, buf, count);
1202 }
1203 
1204 static ssize_t amdgpu_get_pp_dpm_dcefclk(struct device *dev,
1205 		struct device_attribute *attr,
1206 		char *buf)
1207 {
1208 	return amdgpu_get_pp_dpm_clock(dev, PP_DCEFCLK, buf);
1209 }
1210 
1211 static ssize_t amdgpu_set_pp_dpm_dcefclk(struct device *dev,
1212 		struct device_attribute *attr,
1213 		const char *buf,
1214 		size_t count)
1215 {
1216 	return amdgpu_set_pp_dpm_clock(dev, PP_DCEFCLK, buf, count);
1217 }
1218 
1219 static ssize_t amdgpu_get_pp_dpm_pcie(struct device *dev,
1220 		struct device_attribute *attr,
1221 		char *buf)
1222 {
1223 	return amdgpu_get_pp_dpm_clock(dev, PP_PCIE, buf);
1224 }
1225 
1226 static ssize_t amdgpu_set_pp_dpm_pcie(struct device *dev,
1227 		struct device_attribute *attr,
1228 		const char *buf,
1229 		size_t count)
1230 {
1231 	return amdgpu_set_pp_dpm_clock(dev, PP_PCIE, buf, count);
1232 }
1233 
1234 static ssize_t amdgpu_get_pp_sclk_od(struct device *dev,
1235 		struct device_attribute *attr,
1236 		char *buf)
1237 {
1238 	struct drm_device *ddev = dev_get_drvdata(dev);
1239 	struct amdgpu_device *adev = drm_to_adev(ddev);
1240 	uint32_t value = 0;
1241 	int ret;
1242 
1243 	if (amdgpu_in_reset(adev))
1244 		return -EPERM;
1245 	if (adev->in_suspend && !adev->in_runpm)
1246 		return -EPERM;
1247 
1248 	ret = pm_runtime_get_if_active(ddev->dev);
1249 	if (ret <= 0)
1250 		return ret ?: -EPERM;
1251 
1252 	value = amdgpu_dpm_get_sclk_od(adev);
1253 
1254 	pm_runtime_put_autosuspend(ddev->dev);
1255 
1256 	return sysfs_emit(buf, "%d\n", value);
1257 }
1258 
1259 static ssize_t amdgpu_set_pp_sclk_od(struct device *dev,
1260 		struct device_attribute *attr,
1261 		const char *buf,
1262 		size_t count)
1263 {
1264 	struct drm_device *ddev = dev_get_drvdata(dev);
1265 	struct amdgpu_device *adev = drm_to_adev(ddev);
1266 	int ret;
1267 	long int value;
1268 
1269 	if (amdgpu_in_reset(adev))
1270 		return -EPERM;
1271 	if (adev->in_suspend && !adev->in_runpm)
1272 		return -EPERM;
1273 
1274 	ret = kstrtol(buf, 0, &value);
1275 
1276 	if (ret)
1277 		return -EINVAL;
1278 
1279 	ret = pm_runtime_resume_and_get(ddev->dev);
1280 	if (ret < 0)
1281 		return ret;
1282 
1283 	amdgpu_dpm_set_sclk_od(adev, (uint32_t)value);
1284 
1285 	pm_runtime_mark_last_busy(ddev->dev);
1286 	pm_runtime_put_autosuspend(ddev->dev);
1287 
1288 	return count;
1289 }
1290 
1291 static ssize_t amdgpu_get_pp_mclk_od(struct device *dev,
1292 		struct device_attribute *attr,
1293 		char *buf)
1294 {
1295 	struct drm_device *ddev = dev_get_drvdata(dev);
1296 	struct amdgpu_device *adev = drm_to_adev(ddev);
1297 	uint32_t value = 0;
1298 	int ret;
1299 
1300 	if (amdgpu_in_reset(adev))
1301 		return -EPERM;
1302 	if (adev->in_suspend && !adev->in_runpm)
1303 		return -EPERM;
1304 
1305 	ret = pm_runtime_get_if_active(ddev->dev);
1306 	if (ret <= 0)
1307 		return ret ?: -EPERM;
1308 
1309 	value = amdgpu_dpm_get_mclk_od(adev);
1310 
1311 	pm_runtime_put_autosuspend(ddev->dev);
1312 
1313 	return sysfs_emit(buf, "%d\n", value);
1314 }
1315 
1316 static ssize_t amdgpu_set_pp_mclk_od(struct device *dev,
1317 		struct device_attribute *attr,
1318 		const char *buf,
1319 		size_t count)
1320 {
1321 	struct drm_device *ddev = dev_get_drvdata(dev);
1322 	struct amdgpu_device *adev = drm_to_adev(ddev);
1323 	int ret;
1324 	long int value;
1325 
1326 	if (amdgpu_in_reset(adev))
1327 		return -EPERM;
1328 	if (adev->in_suspend && !adev->in_runpm)
1329 		return -EPERM;
1330 
1331 	ret = kstrtol(buf, 0, &value);
1332 
1333 	if (ret)
1334 		return -EINVAL;
1335 
1336 	ret = pm_runtime_resume_and_get(ddev->dev);
1337 	if (ret < 0)
1338 		return ret;
1339 
1340 	amdgpu_dpm_set_mclk_od(adev, (uint32_t)value);
1341 
1342 	pm_runtime_mark_last_busy(ddev->dev);
1343 	pm_runtime_put_autosuspend(ddev->dev);
1344 
1345 	return count;
1346 }
1347 
1348 /**
1349  * DOC: pp_power_profile_mode
1350  *
1351  * The amdgpu driver provides a sysfs API for adjusting the heuristics
1352  * related to switching between power levels in a power state.  The file
1353  * pp_power_profile_mode is used for this.
1354  *
1355  * Reading this file outputs a list of all of the predefined power profiles
1356  * and the relevant heuristics settings for that profile.
1357  *
1358  * To select a profile or create a custom profile, first select manual using
1359  * power_dpm_force_performance_level.  Writing the number of a predefined
1360  * profile to pp_power_profile_mode will enable those heuristics.  To
1361  * create a custom set of heuristics, write a string of numbers to the file
1362  * starting with the number of the custom profile along with a setting
1363  * for each heuristic parameter.  Due to differences across asic families
1364  * the heuristic parameters vary from family to family.
1365  *
1366  */
1367 
1368 static ssize_t amdgpu_get_pp_power_profile_mode(struct device *dev,
1369 		struct device_attribute *attr,
1370 		char *buf)
1371 {
1372 	struct drm_device *ddev = dev_get_drvdata(dev);
1373 	struct amdgpu_device *adev = drm_to_adev(ddev);
1374 	ssize_t size;
1375 	int ret;
1376 
1377 	if (amdgpu_in_reset(adev))
1378 		return -EPERM;
1379 	if (adev->in_suspend && !adev->in_runpm)
1380 		return -EPERM;
1381 
1382 	ret = pm_runtime_get_if_active(ddev->dev);
1383 	if (ret <= 0)
1384 		return ret ?: -EPERM;
1385 
1386 	size = amdgpu_dpm_get_power_profile_mode(adev, buf);
1387 	if (size <= 0)
1388 		size = sysfs_emit(buf, "\n");
1389 
1390 	pm_runtime_put_autosuspend(ddev->dev);
1391 
1392 	return size;
1393 }
1394 
1395 
1396 static ssize_t amdgpu_set_pp_power_profile_mode(struct device *dev,
1397 		struct device_attribute *attr,
1398 		const char *buf,
1399 		size_t count)
1400 {
1401 	int ret;
1402 	struct drm_device *ddev = dev_get_drvdata(dev);
1403 	struct amdgpu_device *adev = drm_to_adev(ddev);
1404 	uint32_t parameter_size = 0;
1405 	long parameter[64];
1406 	char *sub_str, buf_cpy[128];
1407 	char *tmp_str;
1408 	uint32_t i = 0;
1409 	char tmp[2];
1410 	long int profile_mode = 0;
1411 	const char delimiter[3] = {' ', '\n', '\0'};
1412 
1413 	if (amdgpu_in_reset(adev))
1414 		return -EPERM;
1415 	if (adev->in_suspend && !adev->in_runpm)
1416 		return -EPERM;
1417 
1418 	tmp[0] = *(buf);
1419 	tmp[1] = '\0';
1420 	ret = kstrtol(tmp, 0, &profile_mode);
1421 	if (ret)
1422 		return -EINVAL;
1423 
1424 	if (profile_mode == PP_SMC_POWER_PROFILE_CUSTOM) {
1425 		if (count < 2 || count > 127)
1426 			return -EINVAL;
1427 		while (isspace(*++buf))
1428 			i++;
1429 		memcpy(buf_cpy, buf, count-i);
1430 		tmp_str = buf_cpy;
1431 		while ((sub_str = strsep(&tmp_str, delimiter)) != NULL) {
1432 			if (strlen(sub_str) == 0)
1433 				continue;
1434 			ret = kstrtol(sub_str, 0, &parameter[parameter_size]);
1435 			if (ret)
1436 				return -EINVAL;
1437 			parameter_size++;
1438 			while (isspace(*tmp_str))
1439 				tmp_str++;
1440 		}
1441 	}
1442 	parameter[parameter_size] = profile_mode;
1443 
1444 	ret = pm_runtime_resume_and_get(ddev->dev);
1445 	if (ret < 0)
1446 		return ret;
1447 
1448 	ret = amdgpu_dpm_set_power_profile_mode(adev, parameter, parameter_size);
1449 
1450 	pm_runtime_mark_last_busy(ddev->dev);
1451 	pm_runtime_put_autosuspend(ddev->dev);
1452 
1453 	if (!ret)
1454 		return count;
1455 
1456 	return -EINVAL;
1457 }
1458 
1459 static int amdgpu_hwmon_get_sensor_generic(struct amdgpu_device *adev,
1460 					   enum amd_pp_sensors sensor,
1461 					   void *query)
1462 {
1463 	int r, size = sizeof(uint32_t);
1464 
1465 	if (amdgpu_in_reset(adev))
1466 		return -EPERM;
1467 	if (adev->in_suspend && !adev->in_runpm)
1468 		return -EPERM;
1469 
1470 	r = pm_runtime_get_if_active(adev->dev);
1471 	if (r <= 0)
1472 		return r ?: -EPERM;
1473 
1474 	/* get the sensor value */
1475 	r = amdgpu_dpm_read_sensor(adev, sensor, query, &size);
1476 
1477 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
1478 
1479 	return r;
1480 }
1481 
1482 /**
1483  * DOC: gpu_busy_percent
1484  *
1485  * The amdgpu driver provides a sysfs API for reading how busy the GPU
1486  * is as a percentage.  The file gpu_busy_percent is used for this.
1487  * The SMU firmware computes a percentage of load based on the
1488  * aggregate activity level in the IP cores.
1489  */
1490 static ssize_t amdgpu_get_gpu_busy_percent(struct device *dev,
1491 					   struct device_attribute *attr,
1492 					   char *buf)
1493 {
1494 	struct drm_device *ddev = dev_get_drvdata(dev);
1495 	struct amdgpu_device *adev = drm_to_adev(ddev);
1496 	unsigned int value;
1497 	int r;
1498 
1499 	r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_GPU_LOAD, &value);
1500 	if (r)
1501 		return r;
1502 
1503 	return sysfs_emit(buf, "%d\n", value);
1504 }
1505 
1506 /**
1507  * DOC: mem_busy_percent
1508  *
1509  * The amdgpu driver provides a sysfs API for reading how busy the VRAM
1510  * is as a percentage.  The file mem_busy_percent is used for this.
1511  * The SMU firmware computes a percentage of load based on the
1512  * aggregate activity level in the IP cores.
1513  */
1514 static ssize_t amdgpu_get_mem_busy_percent(struct device *dev,
1515 					   struct device_attribute *attr,
1516 					   char *buf)
1517 {
1518 	struct drm_device *ddev = dev_get_drvdata(dev);
1519 	struct amdgpu_device *adev = drm_to_adev(ddev);
1520 	unsigned int value;
1521 	int r;
1522 
1523 	r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_MEM_LOAD, &value);
1524 	if (r)
1525 		return r;
1526 
1527 	return sysfs_emit(buf, "%d\n", value);
1528 }
1529 
1530 /**
1531  * DOC: vcn_busy_percent
1532  *
1533  * The amdgpu driver provides a sysfs API for reading how busy the VCN
1534  * is as a percentage.  The file vcn_busy_percent is used for this.
1535  * The SMU firmware computes a percentage of load based on the
1536  * aggregate activity level in the IP cores.
1537  */
1538 static ssize_t amdgpu_get_vcn_busy_percent(struct device *dev,
1539 						  struct device_attribute *attr,
1540 						  char *buf)
1541 {
1542 	struct drm_device *ddev = dev_get_drvdata(dev);
1543 	struct amdgpu_device *adev = drm_to_adev(ddev);
1544 	unsigned int value;
1545 	int r;
1546 
1547 	r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_VCN_LOAD, &value);
1548 	if (r)
1549 		return r;
1550 
1551 	return sysfs_emit(buf, "%d\n", value);
1552 }
1553 
1554 /**
1555  * DOC: pcie_bw
1556  *
1557  * The amdgpu driver provides a sysfs API for estimating how much data
1558  * has been received and sent by the GPU in the last second through PCIe.
1559  * The file pcie_bw is used for this.
1560  * The Perf counters count the number of received and sent messages and return
1561  * those values, as well as the maximum payload size of a PCIe packet (mps).
1562  * Note that it is not possible to easily and quickly obtain the size of each
1563  * packet transmitted, so we output the max payload size (mps) to allow for
1564  * quick estimation of the PCIe bandwidth usage
1565  */
1566 static ssize_t amdgpu_get_pcie_bw(struct device *dev,
1567 		struct device_attribute *attr,
1568 		char *buf)
1569 {
1570 	struct drm_device *ddev = dev_get_drvdata(dev);
1571 	struct amdgpu_device *adev = drm_to_adev(ddev);
1572 	uint64_t count0 = 0, count1 = 0;
1573 	int ret;
1574 
1575 	if (amdgpu_in_reset(adev))
1576 		return -EPERM;
1577 	if (adev->in_suspend && !adev->in_runpm)
1578 		return -EPERM;
1579 
1580 	if (adev->flags & AMD_IS_APU)
1581 		return -ENODATA;
1582 
1583 	if (!adev->asic_funcs->get_pcie_usage)
1584 		return -ENODATA;
1585 
1586 	ret = pm_runtime_get_if_active(ddev->dev);
1587 	if (ret <= 0)
1588 		return ret ?: -EPERM;
1589 
1590 	amdgpu_asic_get_pcie_usage(adev, &count0, &count1);
1591 
1592 	pm_runtime_put_autosuspend(ddev->dev);
1593 
1594 	return sysfs_emit(buf, "%llu %llu %i\n",
1595 			  count0, count1, pcie_get_mps(adev->pdev));
1596 }
1597 
1598 /**
1599  * DOC: unique_id
1600  *
1601  * The amdgpu driver provides a sysfs API for providing a unique ID for the GPU
1602  * The file unique_id is used for this.
1603  * This will provide a Unique ID that will persist from machine to machine
1604  *
1605  * NOTE: This will only work for GFX9 and newer. This file will be absent
1606  * on unsupported ASICs (GFX8 and older)
1607  */
1608 static ssize_t amdgpu_get_unique_id(struct device *dev,
1609 		struct device_attribute *attr,
1610 		char *buf)
1611 {
1612 	struct drm_device *ddev = dev_get_drvdata(dev);
1613 	struct amdgpu_device *adev = drm_to_adev(ddev);
1614 
1615 	if (amdgpu_in_reset(adev))
1616 		return -EPERM;
1617 	if (adev->in_suspend && !adev->in_runpm)
1618 		return -EPERM;
1619 
1620 	if (adev->unique_id)
1621 		return sysfs_emit(buf, "%016llx\n", adev->unique_id);
1622 
1623 	return 0;
1624 }
1625 
1626 /**
1627  * DOC: thermal_throttling_logging
1628  *
1629  * Thermal throttling pulls down the clock frequency and thus the performance.
1630  * It's an useful mechanism to protect the chip from overheating. Since it
1631  * impacts performance, the user controls whether it is enabled and if so,
1632  * the log frequency.
1633  *
1634  * Reading back the file shows you the status(enabled or disabled) and
1635  * the interval(in seconds) between each thermal logging.
1636  *
1637  * Writing an integer to the file, sets a new logging interval, in seconds.
1638  * The value should be between 1 and 3600. If the value is less than 1,
1639  * thermal logging is disabled. Values greater than 3600 are ignored.
1640  */
1641 static ssize_t amdgpu_get_thermal_throttling_logging(struct device *dev,
1642 						     struct device_attribute *attr,
1643 						     char *buf)
1644 {
1645 	struct drm_device *ddev = dev_get_drvdata(dev);
1646 	struct amdgpu_device *adev = drm_to_adev(ddev);
1647 
1648 	return sysfs_emit(buf, "%s: thermal throttling logging %s, with interval %d seconds\n",
1649 			  adev_to_drm(adev)->unique,
1650 			  atomic_read(&adev->throttling_logging_enabled) ? "enabled" : "disabled",
1651 			  adev->throttling_logging_rs.interval / HZ + 1);
1652 }
1653 
1654 static ssize_t amdgpu_set_thermal_throttling_logging(struct device *dev,
1655 						     struct device_attribute *attr,
1656 						     const char *buf,
1657 						     size_t count)
1658 {
1659 	struct drm_device *ddev = dev_get_drvdata(dev);
1660 	struct amdgpu_device *adev = drm_to_adev(ddev);
1661 	long throttling_logging_interval;
1662 	unsigned long flags;
1663 	int ret = 0;
1664 
1665 	ret = kstrtol(buf, 0, &throttling_logging_interval);
1666 	if (ret)
1667 		return ret;
1668 
1669 	if (throttling_logging_interval > 3600)
1670 		return -EINVAL;
1671 
1672 	if (throttling_logging_interval > 0) {
1673 		raw_spin_lock_irqsave(&adev->throttling_logging_rs.lock, flags);
1674 		/*
1675 		 * Reset the ratelimit timer internals.
1676 		 * This can effectively restart the timer.
1677 		 */
1678 		adev->throttling_logging_rs.interval =
1679 			(throttling_logging_interval - 1) * HZ;
1680 		adev->throttling_logging_rs.begin = 0;
1681 		adev->throttling_logging_rs.printed = 0;
1682 		adev->throttling_logging_rs.missed = 0;
1683 		raw_spin_unlock_irqrestore(&adev->throttling_logging_rs.lock, flags);
1684 
1685 		atomic_set(&adev->throttling_logging_enabled, 1);
1686 	} else {
1687 		atomic_set(&adev->throttling_logging_enabled, 0);
1688 	}
1689 
1690 	return count;
1691 }
1692 
1693 /**
1694  * DOC: apu_thermal_cap
1695  *
1696  * The amdgpu driver provides a sysfs API for retrieving/updating thermal
1697  * limit temperature in millidegrees Celsius
1698  *
1699  * Reading back the file shows you core limit value
1700  *
1701  * Writing an integer to the file, sets a new thermal limit. The value
1702  * should be between 0 and 100. If the value is less than 0 or greater
1703  * than 100, then the write request will be ignored.
1704  */
1705 static ssize_t amdgpu_get_apu_thermal_cap(struct device *dev,
1706 					 struct device_attribute *attr,
1707 					 char *buf)
1708 {
1709 	int ret, size;
1710 	u32 limit;
1711 	struct drm_device *ddev = dev_get_drvdata(dev);
1712 	struct amdgpu_device *adev = drm_to_adev(ddev);
1713 
1714 	ret = pm_runtime_get_if_active(ddev->dev);
1715 	if (ret <= 0)
1716 		return ret ?: -EPERM;
1717 
1718 	ret = amdgpu_dpm_get_apu_thermal_limit(adev, &limit);
1719 	if (!ret)
1720 		size = sysfs_emit(buf, "%u\n", limit);
1721 	else
1722 		size = sysfs_emit(buf, "failed to get thermal limit\n");
1723 
1724 	pm_runtime_put_autosuspend(ddev->dev);
1725 
1726 	return size;
1727 }
1728 
1729 static ssize_t amdgpu_set_apu_thermal_cap(struct device *dev,
1730 					 struct device_attribute *attr,
1731 					 const char *buf,
1732 					 size_t count)
1733 {
1734 	int ret;
1735 	u32 value;
1736 	struct drm_device *ddev = dev_get_drvdata(dev);
1737 	struct amdgpu_device *adev = drm_to_adev(ddev);
1738 
1739 	ret = kstrtou32(buf, 10, &value);
1740 	if (ret)
1741 		return ret;
1742 
1743 	if (value > 100) {
1744 		dev_err(dev, "Invalid argument !\n");
1745 		return -EINVAL;
1746 	}
1747 
1748 	ret = pm_runtime_resume_and_get(ddev->dev);
1749 	if (ret < 0)
1750 		return ret;
1751 
1752 	ret = amdgpu_dpm_set_apu_thermal_limit(adev, value);
1753 	if (ret) {
1754 		pm_runtime_mark_last_busy(ddev->dev);
1755 		pm_runtime_put_autosuspend(ddev->dev);
1756 		dev_err(dev, "failed to update thermal limit\n");
1757 		return ret;
1758 	}
1759 
1760 	pm_runtime_mark_last_busy(ddev->dev);
1761 	pm_runtime_put_autosuspend(ddev->dev);
1762 
1763 	return count;
1764 }
1765 
1766 static int amdgpu_pm_metrics_attr_update(struct amdgpu_device *adev,
1767 					 struct amdgpu_device_attr *attr,
1768 					 uint32_t mask,
1769 					 enum amdgpu_device_attr_states *states)
1770 {
1771 	if (amdgpu_dpm_get_pm_metrics(adev, NULL, 0) == -EOPNOTSUPP)
1772 		*states = ATTR_STATE_UNSUPPORTED;
1773 
1774 	return 0;
1775 }
1776 
1777 static ssize_t amdgpu_get_pm_metrics(struct device *dev,
1778 				     struct device_attribute *attr, char *buf)
1779 {
1780 	struct drm_device *ddev = dev_get_drvdata(dev);
1781 	struct amdgpu_device *adev = drm_to_adev(ddev);
1782 	ssize_t size = 0;
1783 	int ret;
1784 
1785 	if (amdgpu_in_reset(adev))
1786 		return -EPERM;
1787 	if (adev->in_suspend && !adev->in_runpm)
1788 		return -EPERM;
1789 
1790 	ret = pm_runtime_get_if_active(ddev->dev);
1791 	if (ret <= 0)
1792 		return ret ?: -EPERM;
1793 
1794 	size = amdgpu_dpm_get_pm_metrics(adev, buf, PAGE_SIZE);
1795 
1796 	pm_runtime_put_autosuspend(ddev->dev);
1797 
1798 	return size;
1799 }
1800 
1801 /**
1802  * DOC: gpu_metrics
1803  *
1804  * The amdgpu driver provides a sysfs API for retrieving current gpu
1805  * metrics data. The file gpu_metrics is used for this. Reading the
1806  * file will dump all the current gpu metrics data.
1807  *
1808  * These data include temperature, frequency, engines utilization,
1809  * power consume, throttler status, fan speed and cpu core statistics(
1810  * available for APU only). That's it will give a snapshot of all sensors
1811  * at the same time.
1812  */
1813 static ssize_t amdgpu_get_gpu_metrics(struct device *dev,
1814 				      struct device_attribute *attr,
1815 				      char *buf)
1816 {
1817 	struct drm_device *ddev = dev_get_drvdata(dev);
1818 	struct amdgpu_device *adev = drm_to_adev(ddev);
1819 	void *gpu_metrics;
1820 	ssize_t size = 0;
1821 	int ret;
1822 
1823 	if (amdgpu_in_reset(adev))
1824 		return -EPERM;
1825 	if (adev->in_suspend && !adev->in_runpm)
1826 		return -EPERM;
1827 
1828 	ret = pm_runtime_get_if_active(ddev->dev);
1829 	if (ret <= 0)
1830 		return ret ?: -EPERM;
1831 
1832 	size = amdgpu_dpm_get_gpu_metrics(adev, &gpu_metrics);
1833 	if (size <= 0)
1834 		goto out;
1835 
1836 	if (size >= PAGE_SIZE)
1837 		size = PAGE_SIZE - 1;
1838 
1839 	memcpy(buf, gpu_metrics, size);
1840 
1841 out:
1842 	pm_runtime_put_autosuspend(ddev->dev);
1843 
1844 	return size;
1845 }
1846 
1847 static int amdgpu_show_powershift_percent(struct device *dev,
1848 					char *buf, enum amd_pp_sensors sensor)
1849 {
1850 	struct drm_device *ddev = dev_get_drvdata(dev);
1851 	struct amdgpu_device *adev = drm_to_adev(ddev);
1852 	uint32_t ss_power;
1853 	int r = 0, i;
1854 
1855 	r = amdgpu_hwmon_get_sensor_generic(adev, sensor, (void *)&ss_power);
1856 	if (r == -EOPNOTSUPP) {
1857 		/* sensor not available on dGPU, try to read from APU */
1858 		adev = NULL;
1859 		mutex_lock(&mgpu_info.mutex);
1860 		for (i = 0; i < mgpu_info.num_gpu; i++) {
1861 			if (mgpu_info.gpu_ins[i].adev->flags & AMD_IS_APU) {
1862 				adev = mgpu_info.gpu_ins[i].adev;
1863 				break;
1864 			}
1865 		}
1866 		mutex_unlock(&mgpu_info.mutex);
1867 		if (adev)
1868 			r = amdgpu_hwmon_get_sensor_generic(adev, sensor, (void *)&ss_power);
1869 	}
1870 
1871 	if (r)
1872 		return r;
1873 
1874 	return sysfs_emit(buf, "%u%%\n", ss_power);
1875 }
1876 
1877 /**
1878  * DOC: smartshift_apu_power
1879  *
1880  * The amdgpu driver provides a sysfs API for reporting APU power
1881  * shift in percentage if platform supports smartshift. Value 0 means that
1882  * there is no powershift and values between [1-100] means that the power
1883  * is shifted to APU, the percentage of boost is with respect to APU power
1884  * limit on the platform.
1885  */
1886 
1887 static ssize_t amdgpu_get_smartshift_apu_power(struct device *dev, struct device_attribute *attr,
1888 					       char *buf)
1889 {
1890 	return amdgpu_show_powershift_percent(dev, buf, AMDGPU_PP_SENSOR_SS_APU_SHARE);
1891 }
1892 
1893 /**
1894  * DOC: smartshift_dgpu_power
1895  *
1896  * The amdgpu driver provides a sysfs API for reporting dGPU power
1897  * shift in percentage if platform supports smartshift. Value 0 means that
1898  * there is no powershift and values between [1-100] means that the power is
1899  * shifted to dGPU, the percentage of boost is with respect to dGPU power
1900  * limit on the platform.
1901  */
1902 
1903 static ssize_t amdgpu_get_smartshift_dgpu_power(struct device *dev, struct device_attribute *attr,
1904 						char *buf)
1905 {
1906 	return amdgpu_show_powershift_percent(dev, buf, AMDGPU_PP_SENSOR_SS_DGPU_SHARE);
1907 }
1908 
1909 /**
1910  * DOC: smartshift_bias
1911  *
1912  * The amdgpu driver provides a sysfs API for reporting the
1913  * smartshift(SS2.0) bias level. The value ranges from -100 to 100
1914  * and the default is 0. -100 sets maximum preference to APU
1915  * and 100 sets max perference to dGPU.
1916  */
1917 
1918 static ssize_t amdgpu_get_smartshift_bias(struct device *dev,
1919 					  struct device_attribute *attr,
1920 					  char *buf)
1921 {
1922 	int r = 0;
1923 
1924 	r = sysfs_emit(buf, "%d\n", amdgpu_smartshift_bias);
1925 
1926 	return r;
1927 }
1928 
1929 static ssize_t amdgpu_set_smartshift_bias(struct device *dev,
1930 					  struct device_attribute *attr,
1931 					  const char *buf, size_t count)
1932 {
1933 	struct drm_device *ddev = dev_get_drvdata(dev);
1934 	struct amdgpu_device *adev = drm_to_adev(ddev);
1935 	int r = 0;
1936 	int bias = 0;
1937 
1938 	if (amdgpu_in_reset(adev))
1939 		return -EPERM;
1940 	if (adev->in_suspend && !adev->in_runpm)
1941 		return -EPERM;
1942 
1943 	r = pm_runtime_resume_and_get(ddev->dev);
1944 	if (r < 0)
1945 		return r;
1946 
1947 	r = kstrtoint(buf, 10, &bias);
1948 	if (r)
1949 		goto out;
1950 
1951 	if (bias > AMDGPU_SMARTSHIFT_MAX_BIAS)
1952 		bias = AMDGPU_SMARTSHIFT_MAX_BIAS;
1953 	else if (bias < AMDGPU_SMARTSHIFT_MIN_BIAS)
1954 		bias = AMDGPU_SMARTSHIFT_MIN_BIAS;
1955 
1956 	amdgpu_smartshift_bias = bias;
1957 	r = count;
1958 
1959 	/* TODO: update bias level with SMU message */
1960 
1961 out:
1962 	pm_runtime_mark_last_busy(ddev->dev);
1963 	pm_runtime_put_autosuspend(ddev->dev);
1964 	return r;
1965 }
1966 
1967 static int ss_power_attr_update(struct amdgpu_device *adev, struct amdgpu_device_attr *attr,
1968 				uint32_t mask, enum amdgpu_device_attr_states *states)
1969 {
1970 	if (!amdgpu_device_supports_smart_shift(adev_to_drm(adev)))
1971 		*states = ATTR_STATE_UNSUPPORTED;
1972 
1973 	return 0;
1974 }
1975 
1976 static int ss_bias_attr_update(struct amdgpu_device *adev, struct amdgpu_device_attr *attr,
1977 			       uint32_t mask, enum amdgpu_device_attr_states *states)
1978 {
1979 	uint32_t ss_power;
1980 
1981 	if (!amdgpu_device_supports_smart_shift(adev_to_drm(adev)))
1982 		*states = ATTR_STATE_UNSUPPORTED;
1983 	else if (amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_SS_APU_SHARE,
1984 		 (void *)&ss_power))
1985 		*states = ATTR_STATE_UNSUPPORTED;
1986 	else if (amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_SS_DGPU_SHARE,
1987 		 (void *)&ss_power))
1988 		*states = ATTR_STATE_UNSUPPORTED;
1989 
1990 	return 0;
1991 }
1992 
1993 static int pp_od_clk_voltage_attr_update(struct amdgpu_device *adev, struct amdgpu_device_attr *attr,
1994 					 uint32_t mask, enum amdgpu_device_attr_states *states)
1995 {
1996 	uint32_t gc_ver = amdgpu_ip_version(adev, GC_HWIP, 0);
1997 
1998 	*states = ATTR_STATE_SUPPORTED;
1999 
2000 	if (!amdgpu_dpm_is_overdrive_supported(adev)) {
2001 		*states = ATTR_STATE_UNSUPPORTED;
2002 		return 0;
2003 	}
2004 
2005 	/* Enable pp_od_clk_voltage node for gc 9.4.3 SRIOV/BM support */
2006 	if (gc_ver == IP_VERSION(9, 4, 3) ||
2007 	    gc_ver == IP_VERSION(9, 4, 4)) {
2008 		if (amdgpu_sriov_vf(adev) && !amdgpu_sriov_is_pp_one_vf(adev))
2009 			*states = ATTR_STATE_UNSUPPORTED;
2010 		return 0;
2011 	}
2012 
2013 	if (!(attr->flags & mask))
2014 		*states = ATTR_STATE_UNSUPPORTED;
2015 
2016 	return 0;
2017 }
2018 
2019 static int pp_dpm_dcefclk_attr_update(struct amdgpu_device *adev, struct amdgpu_device_attr *attr,
2020 				      uint32_t mask, enum amdgpu_device_attr_states *states)
2021 {
2022 	struct device_attribute *dev_attr = &attr->dev_attr;
2023 	uint32_t gc_ver;
2024 
2025 	*states = ATTR_STATE_SUPPORTED;
2026 
2027 	if (!(attr->flags & mask)) {
2028 		*states = ATTR_STATE_UNSUPPORTED;
2029 		return 0;
2030 	}
2031 
2032 	gc_ver = amdgpu_ip_version(adev, GC_HWIP, 0);
2033 	/* dcefclk node is not available on gfx 11.0.3 sriov */
2034 	if ((gc_ver == IP_VERSION(11, 0, 3) && amdgpu_sriov_is_pp_one_vf(adev)) ||
2035 	    gc_ver < IP_VERSION(9, 0, 0) ||
2036 	    !amdgpu_device_has_display_hardware(adev))
2037 		*states = ATTR_STATE_UNSUPPORTED;
2038 
2039 	/* SMU MP1 does not support dcefclk level setting,
2040 	 * setting should not be allowed from VF if not in one VF mode.
2041 	 */
2042 	if (gc_ver >= IP_VERSION(10, 0, 0) ||
2043 	    (amdgpu_sriov_vf(adev) && !amdgpu_sriov_is_pp_one_vf(adev))) {
2044 		dev_attr->attr.mode &= ~S_IWUGO;
2045 		dev_attr->store = NULL;
2046 	}
2047 
2048 	return 0;
2049 }
2050 
2051 static int pp_dpm_clk_default_attr_update(struct amdgpu_device *adev, struct amdgpu_device_attr *attr,
2052 					  uint32_t mask, enum amdgpu_device_attr_states *states)
2053 {
2054 	struct device_attribute *dev_attr = &attr->dev_attr;
2055 	enum amdgpu_device_attr_id attr_id = attr->attr_id;
2056 	uint32_t mp1_ver = amdgpu_ip_version(adev, MP1_HWIP, 0);
2057 	uint32_t gc_ver = amdgpu_ip_version(adev, GC_HWIP, 0);
2058 
2059 	*states = ATTR_STATE_SUPPORTED;
2060 
2061 	if (!(attr->flags & mask)) {
2062 		*states = ATTR_STATE_UNSUPPORTED;
2063 		return 0;
2064 	}
2065 
2066 	if (DEVICE_ATTR_IS(pp_dpm_socclk)) {
2067 		if (gc_ver < IP_VERSION(9, 0, 0))
2068 			*states = ATTR_STATE_UNSUPPORTED;
2069 	} else if (DEVICE_ATTR_IS(pp_dpm_fclk)) {
2070 		if (mp1_ver < IP_VERSION(10, 0, 0))
2071 			*states = ATTR_STATE_UNSUPPORTED;
2072 	} else if (DEVICE_ATTR_IS(pp_dpm_vclk)) {
2073 		if (!(gc_ver == IP_VERSION(10, 3, 1) ||
2074 		      gc_ver == IP_VERSION(10, 3, 3) ||
2075 		      gc_ver == IP_VERSION(10, 3, 6) ||
2076 		      gc_ver == IP_VERSION(10, 3, 7) ||
2077 		      gc_ver == IP_VERSION(10, 3, 0) ||
2078 		      gc_ver == IP_VERSION(10, 1, 2) ||
2079 		      gc_ver == IP_VERSION(11, 0, 0) ||
2080 		      gc_ver == IP_VERSION(11, 0, 1) ||
2081 		      gc_ver == IP_VERSION(11, 0, 4) ||
2082 		      gc_ver == IP_VERSION(11, 5, 0) ||
2083 		      gc_ver == IP_VERSION(11, 0, 2) ||
2084 		      gc_ver == IP_VERSION(11, 0, 3) ||
2085 		      gc_ver == IP_VERSION(9, 4, 3) ||
2086 		      gc_ver == IP_VERSION(9, 4, 4)))
2087 			*states = ATTR_STATE_UNSUPPORTED;
2088 	} else if (DEVICE_ATTR_IS(pp_dpm_vclk1)) {
2089 		if (!((gc_ver == IP_VERSION(10, 3, 1) ||
2090 		       gc_ver == IP_VERSION(10, 3, 0) ||
2091 		       gc_ver == IP_VERSION(11, 0, 2) ||
2092 		       gc_ver == IP_VERSION(11, 0, 3)) && adev->vcn.num_vcn_inst >= 2))
2093 			*states = ATTR_STATE_UNSUPPORTED;
2094 	} else if (DEVICE_ATTR_IS(pp_dpm_dclk)) {
2095 		if (!(gc_ver == IP_VERSION(10, 3, 1) ||
2096 		      gc_ver == IP_VERSION(10, 3, 3) ||
2097 		      gc_ver == IP_VERSION(10, 3, 6) ||
2098 		      gc_ver == IP_VERSION(10, 3, 7) ||
2099 		      gc_ver == IP_VERSION(10, 3, 0) ||
2100 		      gc_ver == IP_VERSION(10, 1, 2) ||
2101 		      gc_ver == IP_VERSION(11, 0, 0) ||
2102 		      gc_ver == IP_VERSION(11, 0, 1) ||
2103 		      gc_ver == IP_VERSION(11, 0, 4) ||
2104 		      gc_ver == IP_VERSION(11, 5, 0) ||
2105 		      gc_ver == IP_VERSION(11, 0, 2) ||
2106 		      gc_ver == IP_VERSION(11, 0, 3) ||
2107 		      gc_ver == IP_VERSION(9, 4, 3) ||
2108 		      gc_ver == IP_VERSION(9, 4, 4)))
2109 			*states = ATTR_STATE_UNSUPPORTED;
2110 	} else if (DEVICE_ATTR_IS(pp_dpm_dclk1)) {
2111 		if (!((gc_ver == IP_VERSION(10, 3, 1) ||
2112 		       gc_ver == IP_VERSION(10, 3, 0) ||
2113 		       gc_ver == IP_VERSION(11, 0, 2) ||
2114 		       gc_ver == IP_VERSION(11, 0, 3)) && adev->vcn.num_vcn_inst >= 2))
2115 			*states = ATTR_STATE_UNSUPPORTED;
2116 	} else if (DEVICE_ATTR_IS(pp_dpm_pcie)) {
2117 		if (gc_ver == IP_VERSION(9, 4, 2) ||
2118 		    gc_ver == IP_VERSION(9, 4, 3) ||
2119 		    gc_ver == IP_VERSION(9, 4, 4))
2120 			*states = ATTR_STATE_UNSUPPORTED;
2121 	}
2122 
2123 	switch (gc_ver) {
2124 	case IP_VERSION(9, 4, 1):
2125 	case IP_VERSION(9, 4, 2):
2126 		/* the Mi series card does not support standalone mclk/socclk/fclk level setting */
2127 		if (DEVICE_ATTR_IS(pp_dpm_mclk) ||
2128 		    DEVICE_ATTR_IS(pp_dpm_socclk) ||
2129 		    DEVICE_ATTR_IS(pp_dpm_fclk)) {
2130 			dev_attr->attr.mode &= ~S_IWUGO;
2131 			dev_attr->store = NULL;
2132 		}
2133 		break;
2134 	default:
2135 		break;
2136 	}
2137 
2138 	/* setting should not be allowed from VF if not in one VF mode */
2139 	if (amdgpu_sriov_vf(adev) && amdgpu_sriov_is_pp_one_vf(adev)) {
2140 		dev_attr->attr.mode &= ~S_IWUGO;
2141 		dev_attr->store = NULL;
2142 	}
2143 
2144 	return 0;
2145 }
2146 
2147 /* pm policy attributes */
2148 struct amdgpu_pm_policy_attr {
2149 	struct device_attribute dev_attr;
2150 	enum pp_pm_policy id;
2151 };
2152 
2153 /**
2154  * DOC: pm_policy
2155  *
2156  * Certain SOCs can support different power policies to optimize application
2157  * performance. However, this policy is provided only at SOC level and not at a
2158  * per-process level. This is useful especially when entire SOC is utilized for
2159  * dedicated workload.
2160  *
2161  * The amdgpu driver provides a sysfs API for selecting the policy. Presently,
2162  * only two types of policies are supported through this interface.
2163  *
2164  *  Pstate Policy Selection - This is to select different Pstate profiles which
2165  *  decides clock/throttling preferences.
2166  *
2167  *  XGMI PLPD Policy Selection - When multiple devices are connected over XGMI,
2168  *  this helps to select policy to be applied for per link power down.
2169  *
2170  * The list of available policies and policy levels vary between SOCs. They can
2171  * be viewed under pm_policy node directory. If SOC doesn't support any policy,
2172  * this node won't be available. The different policies supported will be
2173  * available as separate nodes under pm_policy.
2174  *
2175  *	cat /sys/bus/pci/devices/.../pm_policy/<policy_type>
2176  *
2177  * Reading the policy file shows the different levels supported. The level which
2178  * is applied presently is denoted by * (asterisk). E.g.,
2179  *
2180  * .. code-block:: console
2181  *
2182  *	cat /sys/bus/pci/devices/.../pm_policy/soc_pstate
2183  *	0 : soc_pstate_default
2184  *	1 : soc_pstate_0
2185  *	2 : soc_pstate_1*
2186  *	3 : soc_pstate_2
2187  *
2188  *	cat /sys/bus/pci/devices/.../pm_policy/xgmi_plpd
2189  *	0 : plpd_disallow
2190  *	1 : plpd_default
2191  *	2 : plpd_optimized*
2192  *
2193  * To apply a specific policy
2194  *
2195  * "echo  <level> > /sys/bus/pci/devices/.../pm_policy/<policy_type>"
2196  *
2197  * For the levels listed in the example above, to select "plpd_optimized" for
2198  * XGMI and "soc_pstate_2" for soc pstate policy -
2199  *
2200  * .. code-block:: console
2201  *
2202  *	echo "2" > /sys/bus/pci/devices/.../pm_policy/xgmi_plpd
2203  *	echo "3" > /sys/bus/pci/devices/.../pm_policy/soc_pstate
2204  *
2205  */
2206 static ssize_t amdgpu_get_pm_policy_attr(struct device *dev,
2207 					 struct device_attribute *attr,
2208 					 char *buf)
2209 {
2210 	struct drm_device *ddev = dev_get_drvdata(dev);
2211 	struct amdgpu_device *adev = drm_to_adev(ddev);
2212 	struct amdgpu_pm_policy_attr *policy_attr;
2213 
2214 	policy_attr =
2215 		container_of(attr, struct amdgpu_pm_policy_attr, dev_attr);
2216 
2217 	if (amdgpu_in_reset(adev))
2218 		return -EPERM;
2219 	if (adev->in_suspend && !adev->in_runpm)
2220 		return -EPERM;
2221 
2222 	return amdgpu_dpm_get_pm_policy_info(adev, policy_attr->id, buf);
2223 }
2224 
2225 static ssize_t amdgpu_set_pm_policy_attr(struct device *dev,
2226 					 struct device_attribute *attr,
2227 					 const char *buf, size_t count)
2228 {
2229 	struct drm_device *ddev = dev_get_drvdata(dev);
2230 	struct amdgpu_device *adev = drm_to_adev(ddev);
2231 	struct amdgpu_pm_policy_attr *policy_attr;
2232 	int ret, num_params = 0;
2233 	char delimiter[] = " \n\t";
2234 	char tmp_buf[128];
2235 	char *tmp, *param;
2236 	long val;
2237 
2238 	if (amdgpu_in_reset(adev))
2239 		return -EPERM;
2240 	if (adev->in_suspend && !adev->in_runpm)
2241 		return -EPERM;
2242 
2243 	count = min(count, sizeof(tmp_buf));
2244 	memcpy(tmp_buf, buf, count);
2245 	tmp_buf[count - 1] = '\0';
2246 	tmp = tmp_buf;
2247 
2248 	tmp = skip_spaces(tmp);
2249 	while ((param = strsep(&tmp, delimiter))) {
2250 		if (!strlen(param)) {
2251 			tmp = skip_spaces(tmp);
2252 			continue;
2253 		}
2254 		ret = kstrtol(param, 0, &val);
2255 		if (ret)
2256 			return -EINVAL;
2257 		num_params++;
2258 		if (num_params > 1)
2259 			return -EINVAL;
2260 	}
2261 
2262 	if (num_params != 1)
2263 		return -EINVAL;
2264 
2265 	policy_attr =
2266 		container_of(attr, struct amdgpu_pm_policy_attr, dev_attr);
2267 
2268 	ret = pm_runtime_resume_and_get(ddev->dev);
2269 	if (ret < 0)
2270 		return ret;
2271 
2272 	ret = amdgpu_dpm_set_pm_policy(adev, policy_attr->id, val);
2273 
2274 	pm_runtime_mark_last_busy(ddev->dev);
2275 	pm_runtime_put_autosuspend(ddev->dev);
2276 
2277 	if (ret)
2278 		return ret;
2279 
2280 	return count;
2281 }
2282 
2283 #define AMDGPU_PM_POLICY_ATTR(_name, _id)                                  \
2284 	static struct amdgpu_pm_policy_attr pm_policy_attr_##_name = {     \
2285 		.dev_attr = __ATTR(_name, 0644, amdgpu_get_pm_policy_attr, \
2286 				   amdgpu_set_pm_policy_attr),             \
2287 		.id = PP_PM_POLICY_##_id,                                  \
2288 	};
2289 
2290 #define AMDGPU_PM_POLICY_ATTR_VAR(_name) pm_policy_attr_##_name.dev_attr.attr
2291 
2292 AMDGPU_PM_POLICY_ATTR(soc_pstate, SOC_PSTATE)
2293 AMDGPU_PM_POLICY_ATTR(xgmi_plpd, XGMI_PLPD)
2294 
2295 static struct attribute *pm_policy_attrs[] = {
2296 	&AMDGPU_PM_POLICY_ATTR_VAR(soc_pstate),
2297 	&AMDGPU_PM_POLICY_ATTR_VAR(xgmi_plpd),
2298 	NULL
2299 };
2300 
2301 static umode_t amdgpu_pm_policy_attr_visible(struct kobject *kobj,
2302 					     struct attribute *attr, int n)
2303 {
2304 	struct device *dev = kobj_to_dev(kobj);
2305 	struct drm_device *ddev = dev_get_drvdata(dev);
2306 	struct amdgpu_device *adev = drm_to_adev(ddev);
2307 	struct amdgpu_pm_policy_attr *policy_attr;
2308 
2309 	policy_attr =
2310 		container_of(attr, struct amdgpu_pm_policy_attr, dev_attr.attr);
2311 
2312 	if (amdgpu_dpm_get_pm_policy_info(adev, policy_attr->id, NULL) ==
2313 	    -ENOENT)
2314 		return 0;
2315 
2316 	return attr->mode;
2317 }
2318 
2319 const struct attribute_group amdgpu_pm_policy_attr_group = {
2320 	.name = "pm_policy",
2321 	.attrs = pm_policy_attrs,
2322 	.is_visible = amdgpu_pm_policy_attr_visible,
2323 };
2324 
2325 static struct amdgpu_device_attr amdgpu_device_attrs[] = {
2326 	AMDGPU_DEVICE_ATTR_RW(power_dpm_state,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2327 	AMDGPU_DEVICE_ATTR_RW(power_dpm_force_performance_level,	ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2328 	AMDGPU_DEVICE_ATTR_RO(pp_num_states,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2329 	AMDGPU_DEVICE_ATTR_RO(pp_cur_state,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2330 	AMDGPU_DEVICE_ATTR_RW(pp_force_state,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2331 	AMDGPU_DEVICE_ATTR_RW(pp_table,					ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2332 	AMDGPU_DEVICE_ATTR_RW(pp_dpm_sclk,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF,
2333 			      .attr_update = pp_dpm_clk_default_attr_update),
2334 	AMDGPU_DEVICE_ATTR_RW(pp_dpm_mclk,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF,
2335 			      .attr_update = pp_dpm_clk_default_attr_update),
2336 	AMDGPU_DEVICE_ATTR_RW(pp_dpm_socclk,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF,
2337 			      .attr_update = pp_dpm_clk_default_attr_update),
2338 	AMDGPU_DEVICE_ATTR_RW(pp_dpm_fclk,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF,
2339 			      .attr_update = pp_dpm_clk_default_attr_update),
2340 	AMDGPU_DEVICE_ATTR_RW(pp_dpm_vclk,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF,
2341 			      .attr_update = pp_dpm_clk_default_attr_update),
2342 	AMDGPU_DEVICE_ATTR_RW(pp_dpm_vclk1,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF,
2343 			      .attr_update = pp_dpm_clk_default_attr_update),
2344 	AMDGPU_DEVICE_ATTR_RW(pp_dpm_dclk,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF,
2345 			      .attr_update = pp_dpm_clk_default_attr_update),
2346 	AMDGPU_DEVICE_ATTR_RW(pp_dpm_dclk1,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF,
2347 			      .attr_update = pp_dpm_clk_default_attr_update),
2348 	AMDGPU_DEVICE_ATTR_RW(pp_dpm_dcefclk,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF,
2349 			      .attr_update = pp_dpm_dcefclk_attr_update),
2350 	AMDGPU_DEVICE_ATTR_RW(pp_dpm_pcie,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF,
2351 			      .attr_update = pp_dpm_clk_default_attr_update),
2352 	AMDGPU_DEVICE_ATTR_RW(pp_sclk_od,				ATTR_FLAG_BASIC),
2353 	AMDGPU_DEVICE_ATTR_RW(pp_mclk_od,				ATTR_FLAG_BASIC),
2354 	AMDGPU_DEVICE_ATTR_RW(pp_power_profile_mode,			ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2355 	AMDGPU_DEVICE_ATTR_RW(pp_od_clk_voltage,			ATTR_FLAG_BASIC,
2356 			      .attr_update = pp_od_clk_voltage_attr_update),
2357 	AMDGPU_DEVICE_ATTR_RO(gpu_busy_percent,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2358 	AMDGPU_DEVICE_ATTR_RO(mem_busy_percent,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2359 	AMDGPU_DEVICE_ATTR_RO(vcn_busy_percent,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2360 	AMDGPU_DEVICE_ATTR_RO(pcie_bw,					ATTR_FLAG_BASIC),
2361 	AMDGPU_DEVICE_ATTR_RW(pp_features,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2362 	AMDGPU_DEVICE_ATTR_RO(unique_id,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2363 	AMDGPU_DEVICE_ATTR_RW(thermal_throttling_logging,		ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2364 	AMDGPU_DEVICE_ATTR_RW(apu_thermal_cap,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2365 	AMDGPU_DEVICE_ATTR_RO(gpu_metrics,				ATTR_FLAG_BASIC|ATTR_FLAG_ONEVF),
2366 	AMDGPU_DEVICE_ATTR_RO(smartshift_apu_power,			ATTR_FLAG_BASIC,
2367 			      .attr_update = ss_power_attr_update),
2368 	AMDGPU_DEVICE_ATTR_RO(smartshift_dgpu_power,			ATTR_FLAG_BASIC,
2369 			      .attr_update = ss_power_attr_update),
2370 	AMDGPU_DEVICE_ATTR_RW(smartshift_bias,				ATTR_FLAG_BASIC,
2371 			      .attr_update = ss_bias_attr_update),
2372 	AMDGPU_DEVICE_ATTR_RO(pm_metrics,				ATTR_FLAG_BASIC,
2373 			      .attr_update = amdgpu_pm_metrics_attr_update),
2374 };
2375 
2376 static int default_attr_update(struct amdgpu_device *adev, struct amdgpu_device_attr *attr,
2377 			       uint32_t mask, enum amdgpu_device_attr_states *states)
2378 {
2379 	struct device_attribute *dev_attr = &attr->dev_attr;
2380 	enum amdgpu_device_attr_id attr_id = attr->attr_id;
2381 	uint32_t gc_ver = amdgpu_ip_version(adev, GC_HWIP, 0);
2382 
2383 	if (!(attr->flags & mask)) {
2384 		*states = ATTR_STATE_UNSUPPORTED;
2385 		return 0;
2386 	}
2387 
2388 	if (DEVICE_ATTR_IS(mem_busy_percent)) {
2389 		if ((adev->flags & AMD_IS_APU &&
2390 		     gc_ver != IP_VERSION(9, 4, 3)) ||
2391 		    gc_ver == IP_VERSION(9, 0, 1))
2392 			*states = ATTR_STATE_UNSUPPORTED;
2393 	} else if (DEVICE_ATTR_IS(vcn_busy_percent)) {
2394 		if (!(gc_ver == IP_VERSION(10, 3, 1) ||
2395 			  gc_ver == IP_VERSION(10, 3, 3) ||
2396 			  gc_ver == IP_VERSION(10, 3, 6) ||
2397 			  gc_ver == IP_VERSION(10, 3, 7) ||
2398 			  gc_ver == IP_VERSION(11, 0, 1) ||
2399 			  gc_ver == IP_VERSION(11, 0, 4) ||
2400 			  gc_ver == IP_VERSION(11, 5, 0)))
2401 			*states = ATTR_STATE_UNSUPPORTED;
2402 	} else if (DEVICE_ATTR_IS(pcie_bw)) {
2403 		/* PCIe Perf counters won't work on APU nodes */
2404 		if (adev->flags & AMD_IS_APU ||
2405 		    !adev->asic_funcs->get_pcie_usage)
2406 			*states = ATTR_STATE_UNSUPPORTED;
2407 	} else if (DEVICE_ATTR_IS(unique_id)) {
2408 		switch (gc_ver) {
2409 		case IP_VERSION(9, 0, 1):
2410 		case IP_VERSION(9, 4, 0):
2411 		case IP_VERSION(9, 4, 1):
2412 		case IP_VERSION(9, 4, 2):
2413 		case IP_VERSION(9, 4, 3):
2414 		case IP_VERSION(9, 4, 4):
2415 		case IP_VERSION(10, 3, 0):
2416 		case IP_VERSION(11, 0, 0):
2417 		case IP_VERSION(11, 0, 1):
2418 		case IP_VERSION(11, 0, 2):
2419 		case IP_VERSION(11, 0, 3):
2420 			*states = ATTR_STATE_SUPPORTED;
2421 			break;
2422 		default:
2423 			*states = ATTR_STATE_UNSUPPORTED;
2424 		}
2425 	} else if (DEVICE_ATTR_IS(pp_features)) {
2426 		if ((adev->flags & AMD_IS_APU &&
2427 		     gc_ver != IP_VERSION(9, 4, 3)) ||
2428 		    gc_ver < IP_VERSION(9, 0, 0))
2429 			*states = ATTR_STATE_UNSUPPORTED;
2430 	} else if (DEVICE_ATTR_IS(gpu_metrics)) {
2431 		if (gc_ver < IP_VERSION(9, 1, 0))
2432 			*states = ATTR_STATE_UNSUPPORTED;
2433 	} else if (DEVICE_ATTR_IS(pp_power_profile_mode)) {
2434 		if (amdgpu_dpm_get_power_profile_mode(adev, NULL) == -EOPNOTSUPP)
2435 			*states = ATTR_STATE_UNSUPPORTED;
2436 		else if ((gc_ver == IP_VERSION(10, 3, 0) ||
2437 			  gc_ver == IP_VERSION(11, 0, 3)) && amdgpu_sriov_vf(adev))
2438 			*states = ATTR_STATE_UNSUPPORTED;
2439 	} else if (DEVICE_ATTR_IS(pp_mclk_od)) {
2440 		if (amdgpu_dpm_get_mclk_od(adev) == -EOPNOTSUPP)
2441 			*states = ATTR_STATE_UNSUPPORTED;
2442 	} else if (DEVICE_ATTR_IS(pp_sclk_od)) {
2443 		if (amdgpu_dpm_get_sclk_od(adev) == -EOPNOTSUPP)
2444 			*states = ATTR_STATE_UNSUPPORTED;
2445 	} else if (DEVICE_ATTR_IS(apu_thermal_cap)) {
2446 		u32 limit;
2447 
2448 		if (amdgpu_dpm_get_apu_thermal_limit(adev, &limit) ==
2449 		    -EOPNOTSUPP)
2450 			*states = ATTR_STATE_UNSUPPORTED;
2451 	}
2452 
2453 	switch (gc_ver) {
2454 	case IP_VERSION(10, 3, 0):
2455 		if (DEVICE_ATTR_IS(power_dpm_force_performance_level) &&
2456 		    amdgpu_sriov_vf(adev)) {
2457 			dev_attr->attr.mode &= ~0222;
2458 			dev_attr->store = NULL;
2459 		}
2460 		break;
2461 	default:
2462 		break;
2463 	}
2464 
2465 	return 0;
2466 }
2467 
2468 
2469 static int amdgpu_device_attr_create(struct amdgpu_device *adev,
2470 				     struct amdgpu_device_attr *attr,
2471 				     uint32_t mask, struct list_head *attr_list)
2472 {
2473 	int ret = 0;
2474 	enum amdgpu_device_attr_states attr_states = ATTR_STATE_SUPPORTED;
2475 	struct amdgpu_device_attr_entry *attr_entry;
2476 	struct device_attribute *dev_attr;
2477 	const char *name;
2478 
2479 	int (*attr_update)(struct amdgpu_device *adev, struct amdgpu_device_attr *attr,
2480 			   uint32_t mask, enum amdgpu_device_attr_states *states) = default_attr_update;
2481 
2482 	if (!attr)
2483 		return -EINVAL;
2484 
2485 	dev_attr = &attr->dev_attr;
2486 	name = dev_attr->attr.name;
2487 
2488 	attr_update = attr->attr_update ? attr->attr_update : default_attr_update;
2489 
2490 	ret = attr_update(adev, attr, mask, &attr_states);
2491 	if (ret) {
2492 		dev_err(adev->dev, "failed to update device file %s, ret = %d\n",
2493 			name, ret);
2494 		return ret;
2495 	}
2496 
2497 	if (attr_states == ATTR_STATE_UNSUPPORTED)
2498 		return 0;
2499 
2500 	ret = device_create_file(adev->dev, dev_attr);
2501 	if (ret) {
2502 		dev_err(adev->dev, "failed to create device file %s, ret = %d\n",
2503 			name, ret);
2504 	}
2505 
2506 	attr_entry = kmalloc(sizeof(*attr_entry), GFP_KERNEL);
2507 	if (!attr_entry)
2508 		return -ENOMEM;
2509 
2510 	attr_entry->attr = attr;
2511 	INIT_LIST_HEAD(&attr_entry->entry);
2512 
2513 	list_add_tail(&attr_entry->entry, attr_list);
2514 
2515 	return ret;
2516 }
2517 
2518 static void amdgpu_device_attr_remove(struct amdgpu_device *adev, struct amdgpu_device_attr *attr)
2519 {
2520 	struct device_attribute *dev_attr = &attr->dev_attr;
2521 
2522 	device_remove_file(adev->dev, dev_attr);
2523 }
2524 
2525 static void amdgpu_device_attr_remove_groups(struct amdgpu_device *adev,
2526 					     struct list_head *attr_list);
2527 
2528 static int amdgpu_device_attr_create_groups(struct amdgpu_device *adev,
2529 					    struct amdgpu_device_attr *attrs,
2530 					    uint32_t counts,
2531 					    uint32_t mask,
2532 					    struct list_head *attr_list)
2533 {
2534 	int ret = 0;
2535 	uint32_t i = 0;
2536 
2537 	for (i = 0; i < counts; i++) {
2538 		ret = amdgpu_device_attr_create(adev, &attrs[i], mask, attr_list);
2539 		if (ret)
2540 			goto failed;
2541 	}
2542 
2543 	return 0;
2544 
2545 failed:
2546 	amdgpu_device_attr_remove_groups(adev, attr_list);
2547 
2548 	return ret;
2549 }
2550 
2551 static void amdgpu_device_attr_remove_groups(struct amdgpu_device *adev,
2552 					     struct list_head *attr_list)
2553 {
2554 	struct amdgpu_device_attr_entry *entry, *entry_tmp;
2555 
2556 	if (list_empty(attr_list))
2557 		return ;
2558 
2559 	list_for_each_entry_safe(entry, entry_tmp, attr_list, entry) {
2560 		amdgpu_device_attr_remove(adev, entry->attr);
2561 		list_del(&entry->entry);
2562 		kfree(entry);
2563 	}
2564 }
2565 
2566 static ssize_t amdgpu_hwmon_show_temp(struct device *dev,
2567 				      struct device_attribute *attr,
2568 				      char *buf)
2569 {
2570 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2571 	int channel = to_sensor_dev_attr(attr)->index;
2572 	int r, temp = 0;
2573 
2574 	if (channel >= PP_TEMP_MAX)
2575 		return -EINVAL;
2576 
2577 	switch (channel) {
2578 	case PP_TEMP_JUNCTION:
2579 		/* get current junction temperature */
2580 		r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_HOTSPOT_TEMP,
2581 					   (void *)&temp);
2582 		break;
2583 	case PP_TEMP_EDGE:
2584 		/* get current edge temperature */
2585 		r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_EDGE_TEMP,
2586 					   (void *)&temp);
2587 		break;
2588 	case PP_TEMP_MEM:
2589 		/* get current memory temperature */
2590 		r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_MEM_TEMP,
2591 					   (void *)&temp);
2592 		break;
2593 	default:
2594 		r = -EINVAL;
2595 		break;
2596 	}
2597 
2598 	if (r)
2599 		return r;
2600 
2601 	return sysfs_emit(buf, "%d\n", temp);
2602 }
2603 
2604 static ssize_t amdgpu_hwmon_show_temp_thresh(struct device *dev,
2605 					     struct device_attribute *attr,
2606 					     char *buf)
2607 {
2608 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2609 	int hyst = to_sensor_dev_attr(attr)->index;
2610 	int temp;
2611 
2612 	if (hyst)
2613 		temp = adev->pm.dpm.thermal.min_temp;
2614 	else
2615 		temp = adev->pm.dpm.thermal.max_temp;
2616 
2617 	return sysfs_emit(buf, "%d\n", temp);
2618 }
2619 
2620 static ssize_t amdgpu_hwmon_show_hotspot_temp_thresh(struct device *dev,
2621 					     struct device_attribute *attr,
2622 					     char *buf)
2623 {
2624 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2625 	int hyst = to_sensor_dev_attr(attr)->index;
2626 	int temp;
2627 
2628 	if (hyst)
2629 		temp = adev->pm.dpm.thermal.min_hotspot_temp;
2630 	else
2631 		temp = adev->pm.dpm.thermal.max_hotspot_crit_temp;
2632 
2633 	return sysfs_emit(buf, "%d\n", temp);
2634 }
2635 
2636 static ssize_t amdgpu_hwmon_show_mem_temp_thresh(struct device *dev,
2637 					     struct device_attribute *attr,
2638 					     char *buf)
2639 {
2640 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2641 	int hyst = to_sensor_dev_attr(attr)->index;
2642 	int temp;
2643 
2644 	if (hyst)
2645 		temp = adev->pm.dpm.thermal.min_mem_temp;
2646 	else
2647 		temp = adev->pm.dpm.thermal.max_mem_crit_temp;
2648 
2649 	return sysfs_emit(buf, "%d\n", temp);
2650 }
2651 
2652 static ssize_t amdgpu_hwmon_show_temp_label(struct device *dev,
2653 					     struct device_attribute *attr,
2654 					     char *buf)
2655 {
2656 	int channel = to_sensor_dev_attr(attr)->index;
2657 
2658 	if (channel >= PP_TEMP_MAX)
2659 		return -EINVAL;
2660 
2661 	return sysfs_emit(buf, "%s\n", temp_label[channel].label);
2662 }
2663 
2664 static ssize_t amdgpu_hwmon_show_temp_emergency(struct device *dev,
2665 					     struct device_attribute *attr,
2666 					     char *buf)
2667 {
2668 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2669 	int channel = to_sensor_dev_attr(attr)->index;
2670 	int temp = 0;
2671 
2672 	if (channel >= PP_TEMP_MAX)
2673 		return -EINVAL;
2674 
2675 	switch (channel) {
2676 	case PP_TEMP_JUNCTION:
2677 		temp = adev->pm.dpm.thermal.max_hotspot_emergency_temp;
2678 		break;
2679 	case PP_TEMP_EDGE:
2680 		temp = adev->pm.dpm.thermal.max_edge_emergency_temp;
2681 		break;
2682 	case PP_TEMP_MEM:
2683 		temp = adev->pm.dpm.thermal.max_mem_emergency_temp;
2684 		break;
2685 	}
2686 
2687 	return sysfs_emit(buf, "%d\n", temp);
2688 }
2689 
2690 static ssize_t amdgpu_hwmon_get_pwm1_enable(struct device *dev,
2691 					    struct device_attribute *attr,
2692 					    char *buf)
2693 {
2694 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2695 	u32 pwm_mode = 0;
2696 	int ret;
2697 
2698 	if (amdgpu_in_reset(adev))
2699 		return -EPERM;
2700 	if (adev->in_suspend && !adev->in_runpm)
2701 		return -EPERM;
2702 
2703 	ret = pm_runtime_get_if_active(adev->dev);
2704 	if (ret <= 0)
2705 		return ret ?: -EPERM;
2706 
2707 	ret = amdgpu_dpm_get_fan_control_mode(adev, &pwm_mode);
2708 
2709 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
2710 
2711 	if (ret)
2712 		return -EINVAL;
2713 
2714 	return sysfs_emit(buf, "%u\n", pwm_mode);
2715 }
2716 
2717 static ssize_t amdgpu_hwmon_set_pwm1_enable(struct device *dev,
2718 					    struct device_attribute *attr,
2719 					    const char *buf,
2720 					    size_t count)
2721 {
2722 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2723 	int err, ret;
2724 	u32 pwm_mode;
2725 	int value;
2726 
2727 	if (amdgpu_in_reset(adev))
2728 		return -EPERM;
2729 	if (adev->in_suspend && !adev->in_runpm)
2730 		return -EPERM;
2731 
2732 	err = kstrtoint(buf, 10, &value);
2733 	if (err)
2734 		return err;
2735 
2736 	if (value == 0)
2737 		pwm_mode = AMD_FAN_CTRL_NONE;
2738 	else if (value == 1)
2739 		pwm_mode = AMD_FAN_CTRL_MANUAL;
2740 	else if (value == 2)
2741 		pwm_mode = AMD_FAN_CTRL_AUTO;
2742 	else
2743 		return -EINVAL;
2744 
2745 	ret = pm_runtime_resume_and_get(adev->dev);
2746 	if (ret < 0)
2747 		return ret;
2748 
2749 	ret = amdgpu_dpm_set_fan_control_mode(adev, pwm_mode);
2750 
2751 	pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
2752 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
2753 
2754 	if (ret)
2755 		return -EINVAL;
2756 
2757 	return count;
2758 }
2759 
2760 static ssize_t amdgpu_hwmon_get_pwm1_min(struct device *dev,
2761 					 struct device_attribute *attr,
2762 					 char *buf)
2763 {
2764 	return sysfs_emit(buf, "%i\n", 0);
2765 }
2766 
2767 static ssize_t amdgpu_hwmon_get_pwm1_max(struct device *dev,
2768 					 struct device_attribute *attr,
2769 					 char *buf)
2770 {
2771 	return sysfs_emit(buf, "%i\n", 255);
2772 }
2773 
2774 static ssize_t amdgpu_hwmon_set_pwm1(struct device *dev,
2775 				     struct device_attribute *attr,
2776 				     const char *buf, size_t count)
2777 {
2778 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2779 	int err;
2780 	u32 value;
2781 	u32 pwm_mode;
2782 
2783 	if (amdgpu_in_reset(adev))
2784 		return -EPERM;
2785 	if (adev->in_suspend && !adev->in_runpm)
2786 		return -EPERM;
2787 
2788 	err = kstrtou32(buf, 10, &value);
2789 	if (err)
2790 		return err;
2791 
2792 	err = pm_runtime_resume_and_get(adev->dev);
2793 	if (err < 0)
2794 		return err;
2795 
2796 	err = amdgpu_dpm_get_fan_control_mode(adev, &pwm_mode);
2797 	if (err)
2798 		goto out;
2799 
2800 	if (pwm_mode != AMD_FAN_CTRL_MANUAL) {
2801 		pr_info("manual fan speed control should be enabled first\n");
2802 		err = -EINVAL;
2803 		goto out;
2804 	}
2805 
2806 	err = amdgpu_dpm_set_fan_speed_pwm(adev, value);
2807 
2808 out:
2809 	pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
2810 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
2811 
2812 	if (err)
2813 		return err;
2814 
2815 	return count;
2816 }
2817 
2818 static ssize_t amdgpu_hwmon_get_pwm1(struct device *dev,
2819 				     struct device_attribute *attr,
2820 				     char *buf)
2821 {
2822 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2823 	int err;
2824 	u32 speed = 0;
2825 
2826 	if (amdgpu_in_reset(adev))
2827 		return -EPERM;
2828 	if (adev->in_suspend && !adev->in_runpm)
2829 		return -EPERM;
2830 
2831 	err = pm_runtime_get_if_active(adev->dev);
2832 	if (err <= 0)
2833 		return err ?: -EPERM;
2834 
2835 	err = amdgpu_dpm_get_fan_speed_pwm(adev, &speed);
2836 
2837 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
2838 
2839 	if (err)
2840 		return err;
2841 
2842 	return sysfs_emit(buf, "%i\n", speed);
2843 }
2844 
2845 static ssize_t amdgpu_hwmon_get_fan1_input(struct device *dev,
2846 					   struct device_attribute *attr,
2847 					   char *buf)
2848 {
2849 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2850 	int err;
2851 	u32 speed = 0;
2852 
2853 	if (amdgpu_in_reset(adev))
2854 		return -EPERM;
2855 	if (adev->in_suspend && !adev->in_runpm)
2856 		return -EPERM;
2857 
2858 	err = pm_runtime_get_if_active(adev->dev);
2859 	if (err <= 0)
2860 		return err ?: -EPERM;
2861 
2862 	err = amdgpu_dpm_get_fan_speed_rpm(adev, &speed);
2863 
2864 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
2865 
2866 	if (err)
2867 		return err;
2868 
2869 	return sysfs_emit(buf, "%i\n", speed);
2870 }
2871 
2872 static ssize_t amdgpu_hwmon_get_fan1_min(struct device *dev,
2873 					 struct device_attribute *attr,
2874 					 char *buf)
2875 {
2876 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2877 	u32 min_rpm = 0;
2878 	int r;
2879 
2880 	r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_MIN_FAN_RPM,
2881 				   (void *)&min_rpm);
2882 
2883 	if (r)
2884 		return r;
2885 
2886 	return sysfs_emit(buf, "%d\n", min_rpm);
2887 }
2888 
2889 static ssize_t amdgpu_hwmon_get_fan1_max(struct device *dev,
2890 					 struct device_attribute *attr,
2891 					 char *buf)
2892 {
2893 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2894 	u32 max_rpm = 0;
2895 	int r;
2896 
2897 	r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_MAX_FAN_RPM,
2898 				   (void *)&max_rpm);
2899 
2900 	if (r)
2901 		return r;
2902 
2903 	return sysfs_emit(buf, "%d\n", max_rpm);
2904 }
2905 
2906 static ssize_t amdgpu_hwmon_get_fan1_target(struct device *dev,
2907 					   struct device_attribute *attr,
2908 					   char *buf)
2909 {
2910 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2911 	int err;
2912 	u32 rpm = 0;
2913 
2914 	if (amdgpu_in_reset(adev))
2915 		return -EPERM;
2916 	if (adev->in_suspend && !adev->in_runpm)
2917 		return -EPERM;
2918 
2919 	err = pm_runtime_get_if_active(adev->dev);
2920 	if (err <= 0)
2921 		return err ?: -EPERM;
2922 
2923 	err = amdgpu_dpm_get_fan_speed_rpm(adev, &rpm);
2924 
2925 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
2926 
2927 	if (err)
2928 		return err;
2929 
2930 	return sysfs_emit(buf, "%i\n", rpm);
2931 }
2932 
2933 static ssize_t amdgpu_hwmon_set_fan1_target(struct device *dev,
2934 				     struct device_attribute *attr,
2935 				     const char *buf, size_t count)
2936 {
2937 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2938 	int err;
2939 	u32 value;
2940 	u32 pwm_mode;
2941 
2942 	if (amdgpu_in_reset(adev))
2943 		return -EPERM;
2944 	if (adev->in_suspend && !adev->in_runpm)
2945 		return -EPERM;
2946 
2947 	err = kstrtou32(buf, 10, &value);
2948 	if (err)
2949 		return err;
2950 
2951 	err = pm_runtime_resume_and_get(adev->dev);
2952 	if (err < 0)
2953 		return err;
2954 
2955 	err = amdgpu_dpm_get_fan_control_mode(adev, &pwm_mode);
2956 	if (err)
2957 		goto out;
2958 
2959 	if (pwm_mode != AMD_FAN_CTRL_MANUAL) {
2960 		err = -ENODATA;
2961 		goto out;
2962 	}
2963 
2964 	err = amdgpu_dpm_set_fan_speed_rpm(adev, value);
2965 
2966 out:
2967 	pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
2968 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
2969 
2970 	if (err)
2971 		return err;
2972 
2973 	return count;
2974 }
2975 
2976 static ssize_t amdgpu_hwmon_get_fan1_enable(struct device *dev,
2977 					    struct device_attribute *attr,
2978 					    char *buf)
2979 {
2980 	struct amdgpu_device *adev = dev_get_drvdata(dev);
2981 	u32 pwm_mode = 0;
2982 	int ret;
2983 
2984 	if (amdgpu_in_reset(adev))
2985 		return -EPERM;
2986 	if (adev->in_suspend && !adev->in_runpm)
2987 		return -EPERM;
2988 
2989 	ret = pm_runtime_get_if_active(adev->dev);
2990 	if (ret <= 0)
2991 		return ret ?: -EPERM;
2992 
2993 	ret = amdgpu_dpm_get_fan_control_mode(adev, &pwm_mode);
2994 
2995 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
2996 
2997 	if (ret)
2998 		return -EINVAL;
2999 
3000 	return sysfs_emit(buf, "%i\n", pwm_mode == AMD_FAN_CTRL_AUTO ? 0 : 1);
3001 }
3002 
3003 static ssize_t amdgpu_hwmon_set_fan1_enable(struct device *dev,
3004 					    struct device_attribute *attr,
3005 					    const char *buf,
3006 					    size_t count)
3007 {
3008 	struct amdgpu_device *adev = dev_get_drvdata(dev);
3009 	int err;
3010 	int value;
3011 	u32 pwm_mode;
3012 
3013 	if (amdgpu_in_reset(adev))
3014 		return -EPERM;
3015 	if (adev->in_suspend && !adev->in_runpm)
3016 		return -EPERM;
3017 
3018 	err = kstrtoint(buf, 10, &value);
3019 	if (err)
3020 		return err;
3021 
3022 	if (value == 0)
3023 		pwm_mode = AMD_FAN_CTRL_AUTO;
3024 	else if (value == 1)
3025 		pwm_mode = AMD_FAN_CTRL_MANUAL;
3026 	else
3027 		return -EINVAL;
3028 
3029 	err = pm_runtime_resume_and_get(adev->dev);
3030 	if (err < 0)
3031 		return err;
3032 
3033 	err = amdgpu_dpm_set_fan_control_mode(adev, pwm_mode);
3034 
3035 	pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
3036 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
3037 
3038 	if (err)
3039 		return -EINVAL;
3040 
3041 	return count;
3042 }
3043 
3044 static ssize_t amdgpu_hwmon_show_vddgfx(struct device *dev,
3045 					struct device_attribute *attr,
3046 					char *buf)
3047 {
3048 	struct amdgpu_device *adev = dev_get_drvdata(dev);
3049 	u32 vddgfx;
3050 	int r;
3051 
3052 	/* get the voltage */
3053 	r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_VDDGFX,
3054 				   (void *)&vddgfx);
3055 	if (r)
3056 		return r;
3057 
3058 	return sysfs_emit(buf, "%d\n", vddgfx);
3059 }
3060 
3061 static ssize_t amdgpu_hwmon_show_vddgfx_label(struct device *dev,
3062 					      struct device_attribute *attr,
3063 					      char *buf)
3064 {
3065 	return sysfs_emit(buf, "vddgfx\n");
3066 }
3067 
3068 static ssize_t amdgpu_hwmon_show_vddnb(struct device *dev,
3069 				       struct device_attribute *attr,
3070 				       char *buf)
3071 {
3072 	struct amdgpu_device *adev = dev_get_drvdata(dev);
3073 	u32 vddnb;
3074 	int r;
3075 
3076 	/* only APUs have vddnb */
3077 	if  (!(adev->flags & AMD_IS_APU))
3078 		return -EINVAL;
3079 
3080 	/* get the voltage */
3081 	r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_VDDNB,
3082 				   (void *)&vddnb);
3083 	if (r)
3084 		return r;
3085 
3086 	return sysfs_emit(buf, "%d\n", vddnb);
3087 }
3088 
3089 static ssize_t amdgpu_hwmon_show_vddnb_label(struct device *dev,
3090 					      struct device_attribute *attr,
3091 					      char *buf)
3092 {
3093 	return sysfs_emit(buf, "vddnb\n");
3094 }
3095 
3096 static int amdgpu_hwmon_get_power(struct device *dev,
3097 				  enum amd_pp_sensors sensor)
3098 {
3099 	struct amdgpu_device *adev = dev_get_drvdata(dev);
3100 	unsigned int uw;
3101 	u32 query = 0;
3102 	int r;
3103 
3104 	r = amdgpu_hwmon_get_sensor_generic(adev, sensor, (void *)&query);
3105 	if (r)
3106 		return r;
3107 
3108 	/* convert to microwatts */
3109 	uw = (query >> 8) * 1000000 + (query & 0xff) * 1000;
3110 
3111 	return uw;
3112 }
3113 
3114 static ssize_t amdgpu_hwmon_show_power_avg(struct device *dev,
3115 					   struct device_attribute *attr,
3116 					   char *buf)
3117 {
3118 	ssize_t val;
3119 
3120 	val = amdgpu_hwmon_get_power(dev, AMDGPU_PP_SENSOR_GPU_AVG_POWER);
3121 	if (val < 0)
3122 		return val;
3123 
3124 	return sysfs_emit(buf, "%zd\n", val);
3125 }
3126 
3127 static ssize_t amdgpu_hwmon_show_power_input(struct device *dev,
3128 					     struct device_attribute *attr,
3129 					     char *buf)
3130 {
3131 	ssize_t val;
3132 
3133 	val = amdgpu_hwmon_get_power(dev, AMDGPU_PP_SENSOR_GPU_INPUT_POWER);
3134 	if (val < 0)
3135 		return val;
3136 
3137 	return sysfs_emit(buf, "%zd\n", val);
3138 }
3139 
3140 static ssize_t amdgpu_hwmon_show_power_cap_generic(struct device *dev,
3141 					struct device_attribute *attr,
3142 					char *buf,
3143 					enum pp_power_limit_level pp_limit_level)
3144 {
3145 	struct amdgpu_device *adev = dev_get_drvdata(dev);
3146 	enum pp_power_type power_type = to_sensor_dev_attr(attr)->index;
3147 	uint32_t limit;
3148 	ssize_t size;
3149 	int r;
3150 
3151 	if (amdgpu_in_reset(adev))
3152 		return -EPERM;
3153 	if (adev->in_suspend && !adev->in_runpm)
3154 		return -EPERM;
3155 
3156 	r = pm_runtime_get_if_active(adev->dev);
3157 	if (r <= 0)
3158 		return r ?: -EPERM;
3159 
3160 	r = amdgpu_dpm_get_power_limit(adev, &limit,
3161 				      pp_limit_level, power_type);
3162 
3163 	if (!r)
3164 		size = sysfs_emit(buf, "%u\n", limit * 1000000);
3165 	else
3166 		size = sysfs_emit(buf, "\n");
3167 
3168 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
3169 
3170 	return size;
3171 }
3172 
3173 static ssize_t amdgpu_hwmon_show_power_cap_min(struct device *dev,
3174 					 struct device_attribute *attr,
3175 					 char *buf)
3176 {
3177 	return amdgpu_hwmon_show_power_cap_generic(dev, attr, buf, PP_PWR_LIMIT_MIN);
3178 }
3179 
3180 static ssize_t amdgpu_hwmon_show_power_cap_max(struct device *dev,
3181 					 struct device_attribute *attr,
3182 					 char *buf)
3183 {
3184 	return amdgpu_hwmon_show_power_cap_generic(dev, attr, buf, PP_PWR_LIMIT_MAX);
3185 
3186 }
3187 
3188 static ssize_t amdgpu_hwmon_show_power_cap(struct device *dev,
3189 					 struct device_attribute *attr,
3190 					 char *buf)
3191 {
3192 	return amdgpu_hwmon_show_power_cap_generic(dev, attr, buf, PP_PWR_LIMIT_CURRENT);
3193 
3194 }
3195 
3196 static ssize_t amdgpu_hwmon_show_power_cap_default(struct device *dev,
3197 					 struct device_attribute *attr,
3198 					 char *buf)
3199 {
3200 	return amdgpu_hwmon_show_power_cap_generic(dev, attr, buf, PP_PWR_LIMIT_DEFAULT);
3201 
3202 }
3203 
3204 static ssize_t amdgpu_hwmon_show_power_label(struct device *dev,
3205 					 struct device_attribute *attr,
3206 					 char *buf)
3207 {
3208 	struct amdgpu_device *adev = dev_get_drvdata(dev);
3209 	uint32_t gc_ver = amdgpu_ip_version(adev, GC_HWIP, 0);
3210 
3211 	if (gc_ver == IP_VERSION(10, 3, 1))
3212 		return sysfs_emit(buf, "%s\n",
3213 				  to_sensor_dev_attr(attr)->index == PP_PWR_TYPE_FAST ?
3214 				  "fastPPT" : "slowPPT");
3215 	else
3216 		return sysfs_emit(buf, "PPT\n");
3217 }
3218 
3219 static ssize_t amdgpu_hwmon_set_power_cap(struct device *dev,
3220 		struct device_attribute *attr,
3221 		const char *buf,
3222 		size_t count)
3223 {
3224 	struct amdgpu_device *adev = dev_get_drvdata(dev);
3225 	int limit_type = to_sensor_dev_attr(attr)->index;
3226 	int err;
3227 	u32 value;
3228 
3229 	if (amdgpu_in_reset(adev))
3230 		return -EPERM;
3231 	if (adev->in_suspend && !adev->in_runpm)
3232 		return -EPERM;
3233 
3234 	if (amdgpu_sriov_vf(adev))
3235 		return -EINVAL;
3236 
3237 	err = kstrtou32(buf, 10, &value);
3238 	if (err)
3239 		return err;
3240 
3241 	value = value / 1000000; /* convert to Watt */
3242 	value |= limit_type << 24;
3243 
3244 	err = pm_runtime_resume_and_get(adev->dev);
3245 	if (err < 0)
3246 		return err;
3247 
3248 	err = amdgpu_dpm_set_power_limit(adev, value);
3249 
3250 	pm_runtime_mark_last_busy(adev_to_drm(adev)->dev);
3251 	pm_runtime_put_autosuspend(adev_to_drm(adev)->dev);
3252 
3253 	if (err)
3254 		return err;
3255 
3256 	return count;
3257 }
3258 
3259 static ssize_t amdgpu_hwmon_show_sclk(struct device *dev,
3260 				      struct device_attribute *attr,
3261 				      char *buf)
3262 {
3263 	struct amdgpu_device *adev = dev_get_drvdata(dev);
3264 	uint32_t sclk;
3265 	int r;
3266 
3267 	/* get the sclk */
3268 	r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_GFX_SCLK,
3269 				   (void *)&sclk);
3270 	if (r)
3271 		return r;
3272 
3273 	return sysfs_emit(buf, "%u\n", sclk * 10 * 1000);
3274 }
3275 
3276 static ssize_t amdgpu_hwmon_show_sclk_label(struct device *dev,
3277 					    struct device_attribute *attr,
3278 					    char *buf)
3279 {
3280 	return sysfs_emit(buf, "sclk\n");
3281 }
3282 
3283 static ssize_t amdgpu_hwmon_show_mclk(struct device *dev,
3284 				      struct device_attribute *attr,
3285 				      char *buf)
3286 {
3287 	struct amdgpu_device *adev = dev_get_drvdata(dev);
3288 	uint32_t mclk;
3289 	int r;
3290 
3291 	/* get the sclk */
3292 	r = amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_GFX_MCLK,
3293 				   (void *)&mclk);
3294 	if (r)
3295 		return r;
3296 
3297 	return sysfs_emit(buf, "%u\n", mclk * 10 * 1000);
3298 }
3299 
3300 static ssize_t amdgpu_hwmon_show_mclk_label(struct device *dev,
3301 					    struct device_attribute *attr,
3302 					    char *buf)
3303 {
3304 	return sysfs_emit(buf, "mclk\n");
3305 }
3306 
3307 /**
3308  * DOC: hwmon
3309  *
3310  * The amdgpu driver exposes the following sensor interfaces:
3311  *
3312  * - GPU temperature (via the on-die sensor)
3313  *
3314  * - GPU voltage
3315  *
3316  * - Northbridge voltage (APUs only)
3317  *
3318  * - GPU power
3319  *
3320  * - GPU fan
3321  *
3322  * - GPU gfx/compute engine clock
3323  *
3324  * - GPU memory clock (dGPU only)
3325  *
3326  * hwmon interfaces for GPU temperature:
3327  *
3328  * - temp[1-3]_input: the on die GPU temperature in millidegrees Celsius
3329  *   - temp2_input and temp3_input are supported on SOC15 dGPUs only
3330  *
3331  * - temp[1-3]_label: temperature channel label
3332  *   - temp2_label and temp3_label are supported on SOC15 dGPUs only
3333  *
3334  * - temp[1-3]_crit: temperature critical max value in millidegrees Celsius
3335  *   - temp2_crit and temp3_crit are supported on SOC15 dGPUs only
3336  *
3337  * - temp[1-3]_crit_hyst: temperature hysteresis for critical limit in millidegrees Celsius
3338  *   - temp2_crit_hyst and temp3_crit_hyst are supported on SOC15 dGPUs only
3339  *
3340  * - temp[1-3]_emergency: temperature emergency max value(asic shutdown) in millidegrees Celsius
3341  *   - these are supported on SOC15 dGPUs only
3342  *
3343  * hwmon interfaces for GPU voltage:
3344  *
3345  * - in0_input: the voltage on the GPU in millivolts
3346  *
3347  * - in1_input: the voltage on the Northbridge in millivolts
3348  *
3349  * hwmon interfaces for GPU power:
3350  *
3351  * - power1_average: average power used by the SoC in microWatts.  On APUs this includes the CPU.
3352  *
3353  * - power1_input: instantaneous power used by the SoC in microWatts.  On APUs this includes the CPU.
3354  *
3355  * - power1_cap_min: minimum cap supported in microWatts
3356  *
3357  * - power1_cap_max: maximum cap supported in microWatts
3358  *
3359  * - power1_cap: selected power cap in microWatts
3360  *
3361  * hwmon interfaces for GPU fan:
3362  *
3363  * - pwm1: pulse width modulation fan level (0-255)
3364  *
3365  * - pwm1_enable: pulse width modulation fan control method (0: no fan speed control, 1: manual fan speed control using pwm interface, 2: automatic fan speed control)
3366  *
3367  * - pwm1_min: pulse width modulation fan control minimum level (0)
3368  *
3369  * - pwm1_max: pulse width modulation fan control maximum level (255)
3370  *
3371  * - fan1_min: a minimum value Unit: revolution/min (RPM)
3372  *
3373  * - fan1_max: a maximum value Unit: revolution/max (RPM)
3374  *
3375  * - fan1_input: fan speed in RPM
3376  *
3377  * - fan[1-\*]_target: Desired fan speed Unit: revolution/min (RPM)
3378  *
3379  * - fan[1-\*]_enable: Enable or disable the sensors.1: Enable 0: Disable
3380  *
3381  * NOTE: DO NOT set the fan speed via "pwm1" and "fan[1-\*]_target" interfaces at the same time.
3382  *       That will get the former one overridden.
3383  *
3384  * hwmon interfaces for GPU clocks:
3385  *
3386  * - freq1_input: the gfx/compute clock in hertz
3387  *
3388  * - freq2_input: the memory clock in hertz
3389  *
3390  * You can use hwmon tools like sensors to view this information on your system.
3391  *
3392  */
3393 
3394 static SENSOR_DEVICE_ATTR(temp1_input, S_IRUGO, amdgpu_hwmon_show_temp, NULL, PP_TEMP_EDGE);
3395 static SENSOR_DEVICE_ATTR(temp1_crit, S_IRUGO, amdgpu_hwmon_show_temp_thresh, NULL, 0);
3396 static SENSOR_DEVICE_ATTR(temp1_crit_hyst, S_IRUGO, amdgpu_hwmon_show_temp_thresh, NULL, 1);
3397 static SENSOR_DEVICE_ATTR(temp1_emergency, S_IRUGO, amdgpu_hwmon_show_temp_emergency, NULL, PP_TEMP_EDGE);
3398 static SENSOR_DEVICE_ATTR(temp2_input, S_IRUGO, amdgpu_hwmon_show_temp, NULL, PP_TEMP_JUNCTION);
3399 static SENSOR_DEVICE_ATTR(temp2_crit, S_IRUGO, amdgpu_hwmon_show_hotspot_temp_thresh, NULL, 0);
3400 static SENSOR_DEVICE_ATTR(temp2_crit_hyst, S_IRUGO, amdgpu_hwmon_show_hotspot_temp_thresh, NULL, 1);
3401 static SENSOR_DEVICE_ATTR(temp2_emergency, S_IRUGO, amdgpu_hwmon_show_temp_emergency, NULL, PP_TEMP_JUNCTION);
3402 static SENSOR_DEVICE_ATTR(temp3_input, S_IRUGO, amdgpu_hwmon_show_temp, NULL, PP_TEMP_MEM);
3403 static SENSOR_DEVICE_ATTR(temp3_crit, S_IRUGO, amdgpu_hwmon_show_mem_temp_thresh, NULL, 0);
3404 static SENSOR_DEVICE_ATTR(temp3_crit_hyst, S_IRUGO, amdgpu_hwmon_show_mem_temp_thresh, NULL, 1);
3405 static SENSOR_DEVICE_ATTR(temp3_emergency, S_IRUGO, amdgpu_hwmon_show_temp_emergency, NULL, PP_TEMP_MEM);
3406 static SENSOR_DEVICE_ATTR(temp1_label, S_IRUGO, amdgpu_hwmon_show_temp_label, NULL, PP_TEMP_EDGE);
3407 static SENSOR_DEVICE_ATTR(temp2_label, S_IRUGO, amdgpu_hwmon_show_temp_label, NULL, PP_TEMP_JUNCTION);
3408 static SENSOR_DEVICE_ATTR(temp3_label, S_IRUGO, amdgpu_hwmon_show_temp_label, NULL, PP_TEMP_MEM);
3409 static SENSOR_DEVICE_ATTR(pwm1, S_IRUGO | S_IWUSR, amdgpu_hwmon_get_pwm1, amdgpu_hwmon_set_pwm1, 0);
3410 static SENSOR_DEVICE_ATTR(pwm1_enable, S_IRUGO | S_IWUSR, amdgpu_hwmon_get_pwm1_enable, amdgpu_hwmon_set_pwm1_enable, 0);
3411 static SENSOR_DEVICE_ATTR(pwm1_min, S_IRUGO, amdgpu_hwmon_get_pwm1_min, NULL, 0);
3412 static SENSOR_DEVICE_ATTR(pwm1_max, S_IRUGO, amdgpu_hwmon_get_pwm1_max, NULL, 0);
3413 static SENSOR_DEVICE_ATTR(fan1_input, S_IRUGO, amdgpu_hwmon_get_fan1_input, NULL, 0);
3414 static SENSOR_DEVICE_ATTR(fan1_min, S_IRUGO, amdgpu_hwmon_get_fan1_min, NULL, 0);
3415 static SENSOR_DEVICE_ATTR(fan1_max, S_IRUGO, amdgpu_hwmon_get_fan1_max, NULL, 0);
3416 static SENSOR_DEVICE_ATTR(fan1_target, S_IRUGO | S_IWUSR, amdgpu_hwmon_get_fan1_target, amdgpu_hwmon_set_fan1_target, 0);
3417 static SENSOR_DEVICE_ATTR(fan1_enable, S_IRUGO | S_IWUSR, amdgpu_hwmon_get_fan1_enable, amdgpu_hwmon_set_fan1_enable, 0);
3418 static SENSOR_DEVICE_ATTR(in0_input, S_IRUGO, amdgpu_hwmon_show_vddgfx, NULL, 0);
3419 static SENSOR_DEVICE_ATTR(in0_label, S_IRUGO, amdgpu_hwmon_show_vddgfx_label, NULL, 0);
3420 static SENSOR_DEVICE_ATTR(in1_input, S_IRUGO, amdgpu_hwmon_show_vddnb, NULL, 0);
3421 static SENSOR_DEVICE_ATTR(in1_label, S_IRUGO, amdgpu_hwmon_show_vddnb_label, NULL, 0);
3422 static SENSOR_DEVICE_ATTR(power1_average, S_IRUGO, amdgpu_hwmon_show_power_avg, NULL, 0);
3423 static SENSOR_DEVICE_ATTR(power1_input, S_IRUGO, amdgpu_hwmon_show_power_input, NULL, 0);
3424 static SENSOR_DEVICE_ATTR(power1_cap_max, S_IRUGO, amdgpu_hwmon_show_power_cap_max, NULL, 0);
3425 static SENSOR_DEVICE_ATTR(power1_cap_min, S_IRUGO, amdgpu_hwmon_show_power_cap_min, NULL, 0);
3426 static SENSOR_DEVICE_ATTR(power1_cap, S_IRUGO | S_IWUSR, amdgpu_hwmon_show_power_cap, amdgpu_hwmon_set_power_cap, 0);
3427 static SENSOR_DEVICE_ATTR(power1_cap_default, S_IRUGO, amdgpu_hwmon_show_power_cap_default, NULL, 0);
3428 static SENSOR_DEVICE_ATTR(power1_label, S_IRUGO, amdgpu_hwmon_show_power_label, NULL, 0);
3429 static SENSOR_DEVICE_ATTR(power2_average, S_IRUGO, amdgpu_hwmon_show_power_avg, NULL, 1);
3430 static SENSOR_DEVICE_ATTR(power2_cap_max, S_IRUGO, amdgpu_hwmon_show_power_cap_max, NULL, 1);
3431 static SENSOR_DEVICE_ATTR(power2_cap_min, S_IRUGO, amdgpu_hwmon_show_power_cap_min, NULL, 1);
3432 static SENSOR_DEVICE_ATTR(power2_cap, S_IRUGO | S_IWUSR, amdgpu_hwmon_show_power_cap, amdgpu_hwmon_set_power_cap, 1);
3433 static SENSOR_DEVICE_ATTR(power2_cap_default, S_IRUGO, amdgpu_hwmon_show_power_cap_default, NULL, 1);
3434 static SENSOR_DEVICE_ATTR(power2_label, S_IRUGO, amdgpu_hwmon_show_power_label, NULL, 1);
3435 static SENSOR_DEVICE_ATTR(freq1_input, S_IRUGO, amdgpu_hwmon_show_sclk, NULL, 0);
3436 static SENSOR_DEVICE_ATTR(freq1_label, S_IRUGO, amdgpu_hwmon_show_sclk_label, NULL, 0);
3437 static SENSOR_DEVICE_ATTR(freq2_input, S_IRUGO, amdgpu_hwmon_show_mclk, NULL, 0);
3438 static SENSOR_DEVICE_ATTR(freq2_label, S_IRUGO, amdgpu_hwmon_show_mclk_label, NULL, 0);
3439 
3440 static struct attribute *hwmon_attributes[] = {
3441 	&sensor_dev_attr_temp1_input.dev_attr.attr,
3442 	&sensor_dev_attr_temp1_crit.dev_attr.attr,
3443 	&sensor_dev_attr_temp1_crit_hyst.dev_attr.attr,
3444 	&sensor_dev_attr_temp2_input.dev_attr.attr,
3445 	&sensor_dev_attr_temp2_crit.dev_attr.attr,
3446 	&sensor_dev_attr_temp2_crit_hyst.dev_attr.attr,
3447 	&sensor_dev_attr_temp3_input.dev_attr.attr,
3448 	&sensor_dev_attr_temp3_crit.dev_attr.attr,
3449 	&sensor_dev_attr_temp3_crit_hyst.dev_attr.attr,
3450 	&sensor_dev_attr_temp1_emergency.dev_attr.attr,
3451 	&sensor_dev_attr_temp2_emergency.dev_attr.attr,
3452 	&sensor_dev_attr_temp3_emergency.dev_attr.attr,
3453 	&sensor_dev_attr_temp1_label.dev_attr.attr,
3454 	&sensor_dev_attr_temp2_label.dev_attr.attr,
3455 	&sensor_dev_attr_temp3_label.dev_attr.attr,
3456 	&sensor_dev_attr_pwm1.dev_attr.attr,
3457 	&sensor_dev_attr_pwm1_enable.dev_attr.attr,
3458 	&sensor_dev_attr_pwm1_min.dev_attr.attr,
3459 	&sensor_dev_attr_pwm1_max.dev_attr.attr,
3460 	&sensor_dev_attr_fan1_input.dev_attr.attr,
3461 	&sensor_dev_attr_fan1_min.dev_attr.attr,
3462 	&sensor_dev_attr_fan1_max.dev_attr.attr,
3463 	&sensor_dev_attr_fan1_target.dev_attr.attr,
3464 	&sensor_dev_attr_fan1_enable.dev_attr.attr,
3465 	&sensor_dev_attr_in0_input.dev_attr.attr,
3466 	&sensor_dev_attr_in0_label.dev_attr.attr,
3467 	&sensor_dev_attr_in1_input.dev_attr.attr,
3468 	&sensor_dev_attr_in1_label.dev_attr.attr,
3469 	&sensor_dev_attr_power1_average.dev_attr.attr,
3470 	&sensor_dev_attr_power1_input.dev_attr.attr,
3471 	&sensor_dev_attr_power1_cap_max.dev_attr.attr,
3472 	&sensor_dev_attr_power1_cap_min.dev_attr.attr,
3473 	&sensor_dev_attr_power1_cap.dev_attr.attr,
3474 	&sensor_dev_attr_power1_cap_default.dev_attr.attr,
3475 	&sensor_dev_attr_power1_label.dev_attr.attr,
3476 	&sensor_dev_attr_power2_average.dev_attr.attr,
3477 	&sensor_dev_attr_power2_cap_max.dev_attr.attr,
3478 	&sensor_dev_attr_power2_cap_min.dev_attr.attr,
3479 	&sensor_dev_attr_power2_cap.dev_attr.attr,
3480 	&sensor_dev_attr_power2_cap_default.dev_attr.attr,
3481 	&sensor_dev_attr_power2_label.dev_attr.attr,
3482 	&sensor_dev_attr_freq1_input.dev_attr.attr,
3483 	&sensor_dev_attr_freq1_label.dev_attr.attr,
3484 	&sensor_dev_attr_freq2_input.dev_attr.attr,
3485 	&sensor_dev_attr_freq2_label.dev_attr.attr,
3486 	NULL
3487 };
3488 
3489 static umode_t hwmon_attributes_visible(struct kobject *kobj,
3490 					struct attribute *attr, int index)
3491 {
3492 	struct device *dev = kobj_to_dev(kobj);
3493 	struct amdgpu_device *adev = dev_get_drvdata(dev);
3494 	umode_t effective_mode = attr->mode;
3495 	uint32_t gc_ver = amdgpu_ip_version(adev, GC_HWIP, 0);
3496 	uint32_t tmp;
3497 
3498 	/* under pp one vf mode manage of hwmon attributes is not supported */
3499 	if (amdgpu_sriov_is_pp_one_vf(adev))
3500 		effective_mode &= ~S_IWUSR;
3501 
3502 	/* Skip fan attributes if fan is not present */
3503 	if (adev->pm.no_fan && (attr == &sensor_dev_attr_pwm1.dev_attr.attr ||
3504 	    attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr ||
3505 	    attr == &sensor_dev_attr_pwm1_max.dev_attr.attr ||
3506 	    attr == &sensor_dev_attr_pwm1_min.dev_attr.attr ||
3507 	    attr == &sensor_dev_attr_fan1_input.dev_attr.attr ||
3508 	    attr == &sensor_dev_attr_fan1_min.dev_attr.attr ||
3509 	    attr == &sensor_dev_attr_fan1_max.dev_attr.attr ||
3510 	    attr == &sensor_dev_attr_fan1_target.dev_attr.attr ||
3511 	    attr == &sensor_dev_attr_fan1_enable.dev_attr.attr))
3512 		return 0;
3513 
3514 	/* Skip fan attributes on APU */
3515 	if ((adev->flags & AMD_IS_APU) &&
3516 	    (attr == &sensor_dev_attr_pwm1.dev_attr.attr ||
3517 	     attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr ||
3518 	     attr == &sensor_dev_attr_pwm1_max.dev_attr.attr ||
3519 	     attr == &sensor_dev_attr_pwm1_min.dev_attr.attr ||
3520 	     attr == &sensor_dev_attr_fan1_input.dev_attr.attr ||
3521 	     attr == &sensor_dev_attr_fan1_min.dev_attr.attr ||
3522 	     attr == &sensor_dev_attr_fan1_max.dev_attr.attr ||
3523 	     attr == &sensor_dev_attr_fan1_target.dev_attr.attr ||
3524 	     attr == &sensor_dev_attr_fan1_enable.dev_attr.attr))
3525 		return 0;
3526 
3527 	/* Skip crit temp on APU */
3528 	if ((((adev->flags & AMD_IS_APU) && (adev->family >= AMDGPU_FAMILY_CZ)) ||
3529 	    (gc_ver == IP_VERSION(9, 4, 3) || gc_ver == IP_VERSION(9, 4, 4))) &&
3530 	    (attr == &sensor_dev_attr_temp1_crit.dev_attr.attr ||
3531 	     attr == &sensor_dev_attr_temp1_crit_hyst.dev_attr.attr))
3532 		return 0;
3533 
3534 	/* Skip limit attributes if DPM is not enabled */
3535 	if (!adev->pm.dpm_enabled &&
3536 	    (attr == &sensor_dev_attr_temp1_crit.dev_attr.attr ||
3537 	     attr == &sensor_dev_attr_temp1_crit_hyst.dev_attr.attr ||
3538 	     attr == &sensor_dev_attr_pwm1.dev_attr.attr ||
3539 	     attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr ||
3540 	     attr == &sensor_dev_attr_pwm1_max.dev_attr.attr ||
3541 	     attr == &sensor_dev_attr_pwm1_min.dev_attr.attr ||
3542 	     attr == &sensor_dev_attr_fan1_input.dev_attr.attr ||
3543 	     attr == &sensor_dev_attr_fan1_min.dev_attr.attr ||
3544 	     attr == &sensor_dev_attr_fan1_max.dev_attr.attr ||
3545 	     attr == &sensor_dev_attr_fan1_target.dev_attr.attr ||
3546 	     attr == &sensor_dev_attr_fan1_enable.dev_attr.attr))
3547 		return 0;
3548 
3549 	/* mask fan attributes if we have no bindings for this asic to expose */
3550 	if (((amdgpu_dpm_get_fan_speed_pwm(adev, NULL) == -EOPNOTSUPP) &&
3551 	      attr == &sensor_dev_attr_pwm1.dev_attr.attr) || /* can't query fan */
3552 	    ((amdgpu_dpm_get_fan_control_mode(adev, NULL) == -EOPNOTSUPP) &&
3553 	     attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr)) /* can't query state */
3554 		effective_mode &= ~S_IRUGO;
3555 
3556 	if (((amdgpu_dpm_set_fan_speed_pwm(adev, U32_MAX) == -EOPNOTSUPP) &&
3557 	      attr == &sensor_dev_attr_pwm1.dev_attr.attr) || /* can't manage fan */
3558 	      ((amdgpu_dpm_set_fan_control_mode(adev, U32_MAX) == -EOPNOTSUPP) &&
3559 	      attr == &sensor_dev_attr_pwm1_enable.dev_attr.attr)) /* can't manage state */
3560 		effective_mode &= ~S_IWUSR;
3561 
3562 	/* not implemented yet for APUs other than GC 10.3.1 (vangogh) and 9.4.3 */
3563 	if (((adev->family == AMDGPU_FAMILY_SI) ||
3564 	     ((adev->flags & AMD_IS_APU) && (gc_ver != IP_VERSION(10, 3, 1)) &&
3565 	      (gc_ver != IP_VERSION(9, 4, 3) && gc_ver != IP_VERSION(9, 4, 4)))) &&
3566 	    (attr == &sensor_dev_attr_power1_cap_max.dev_attr.attr ||
3567 	     attr == &sensor_dev_attr_power1_cap_min.dev_attr.attr ||
3568 	     attr == &sensor_dev_attr_power1_cap.dev_attr.attr ||
3569 	     attr == &sensor_dev_attr_power1_cap_default.dev_attr.attr))
3570 		return 0;
3571 
3572 	/* not implemented yet for APUs having < GC 9.3.0 (Renoir) */
3573 	if (((adev->family == AMDGPU_FAMILY_SI) ||
3574 	     ((adev->flags & AMD_IS_APU) && (gc_ver < IP_VERSION(9, 3, 0)))) &&
3575 	    (attr == &sensor_dev_attr_power1_average.dev_attr.attr))
3576 		return 0;
3577 
3578 	/* not all products support both average and instantaneous */
3579 	if (attr == &sensor_dev_attr_power1_average.dev_attr.attr &&
3580 	    amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_GPU_AVG_POWER, (void *)&tmp) == -EOPNOTSUPP)
3581 		return 0;
3582 	if (attr == &sensor_dev_attr_power1_input.dev_attr.attr &&
3583 	    amdgpu_hwmon_get_sensor_generic(adev, AMDGPU_PP_SENSOR_GPU_INPUT_POWER, (void *)&tmp) == -EOPNOTSUPP)
3584 		return 0;
3585 
3586 	/* hide max/min values if we can't both query and manage the fan */
3587 	if (((amdgpu_dpm_set_fan_speed_pwm(adev, U32_MAX) == -EOPNOTSUPP) &&
3588 	      (amdgpu_dpm_get_fan_speed_pwm(adev, NULL) == -EOPNOTSUPP) &&
3589 	      (amdgpu_dpm_set_fan_speed_rpm(adev, U32_MAX) == -EOPNOTSUPP) &&
3590 	      (amdgpu_dpm_get_fan_speed_rpm(adev, NULL) == -EOPNOTSUPP)) &&
3591 	    (attr == &sensor_dev_attr_pwm1_max.dev_attr.attr ||
3592 	     attr == &sensor_dev_attr_pwm1_min.dev_attr.attr))
3593 		return 0;
3594 
3595 	if ((amdgpu_dpm_set_fan_speed_rpm(adev, U32_MAX) == -EOPNOTSUPP) &&
3596 	     (amdgpu_dpm_get_fan_speed_rpm(adev, NULL) == -EOPNOTSUPP) &&
3597 	     (attr == &sensor_dev_attr_fan1_max.dev_attr.attr ||
3598 	     attr == &sensor_dev_attr_fan1_min.dev_attr.attr))
3599 		return 0;
3600 
3601 	if ((adev->family == AMDGPU_FAMILY_SI ||	/* not implemented yet */
3602 	     adev->family == AMDGPU_FAMILY_KV ||	/* not implemented yet */
3603 	     (gc_ver == IP_VERSION(9, 4, 3) ||
3604 	      gc_ver == IP_VERSION(9, 4, 4))) &&
3605 	    (attr == &sensor_dev_attr_in0_input.dev_attr.attr ||
3606 	     attr == &sensor_dev_attr_in0_label.dev_attr.attr))
3607 		return 0;
3608 
3609 	/* only APUs other than gc 9,4,3 have vddnb */
3610 	if ((!(adev->flags & AMD_IS_APU) ||
3611 	     (gc_ver == IP_VERSION(9, 4, 3) ||
3612 	      gc_ver == IP_VERSION(9, 4, 4))) &&
3613 	    (attr == &sensor_dev_attr_in1_input.dev_attr.attr ||
3614 	     attr == &sensor_dev_attr_in1_label.dev_attr.attr))
3615 		return 0;
3616 
3617 	/* no mclk on APUs other than gc 9,4,3*/
3618 	if (((adev->flags & AMD_IS_APU) && (gc_ver != IP_VERSION(9, 4, 3))) &&
3619 	    (attr == &sensor_dev_attr_freq2_input.dev_attr.attr ||
3620 	     attr == &sensor_dev_attr_freq2_label.dev_attr.attr))
3621 		return 0;
3622 
3623 	if (((adev->flags & AMD_IS_APU) || gc_ver < IP_VERSION(9, 0, 0)) &&
3624 	    (gc_ver != IP_VERSION(9, 4, 3) && gc_ver != IP_VERSION(9, 4, 4)) &&
3625 	    (attr == &sensor_dev_attr_temp2_input.dev_attr.attr ||
3626 	     attr == &sensor_dev_attr_temp2_label.dev_attr.attr ||
3627 	     attr == &sensor_dev_attr_temp2_crit.dev_attr.attr ||
3628 	     attr == &sensor_dev_attr_temp3_input.dev_attr.attr ||
3629 	     attr == &sensor_dev_attr_temp3_label.dev_attr.attr ||
3630 	     attr == &sensor_dev_attr_temp3_crit.dev_attr.attr))
3631 		return 0;
3632 
3633 	/* hotspot temperature for gc 9,4,3*/
3634 	if (gc_ver == IP_VERSION(9, 4, 3) ||
3635 	    gc_ver == IP_VERSION(9, 4, 4)) {
3636 		if (attr == &sensor_dev_attr_temp1_input.dev_attr.attr ||
3637 		    attr == &sensor_dev_attr_temp1_emergency.dev_attr.attr ||
3638 		    attr == &sensor_dev_attr_temp1_label.dev_attr.attr)
3639 			return 0;
3640 
3641 		if (attr == &sensor_dev_attr_temp2_emergency.dev_attr.attr ||
3642 		    attr == &sensor_dev_attr_temp3_emergency.dev_attr.attr)
3643 			return attr->mode;
3644 	}
3645 
3646 	/* only SOC15 dGPUs support hotspot and mem temperatures */
3647 	if (((adev->flags & AMD_IS_APU) || gc_ver < IP_VERSION(9, 0, 0)) &&
3648 	    (attr == &sensor_dev_attr_temp2_crit_hyst.dev_attr.attr ||
3649 	     attr == &sensor_dev_attr_temp3_crit_hyst.dev_attr.attr ||
3650 	     attr == &sensor_dev_attr_temp1_emergency.dev_attr.attr ||
3651 	     attr == &sensor_dev_attr_temp2_emergency.dev_attr.attr ||
3652 	     attr == &sensor_dev_attr_temp3_emergency.dev_attr.attr))
3653 		return 0;
3654 
3655 	/* only Vangogh has fast PPT limit and power labels */
3656 	if (!(gc_ver == IP_VERSION(10, 3, 1)) &&
3657 	    (attr == &sensor_dev_attr_power2_average.dev_attr.attr ||
3658 	     attr == &sensor_dev_attr_power2_cap_max.dev_attr.attr ||
3659 	     attr == &sensor_dev_attr_power2_cap_min.dev_attr.attr ||
3660 	     attr == &sensor_dev_attr_power2_cap.dev_attr.attr ||
3661 	     attr == &sensor_dev_attr_power2_cap_default.dev_attr.attr ||
3662 	     attr == &sensor_dev_attr_power2_label.dev_attr.attr))
3663 		return 0;
3664 
3665 	return effective_mode;
3666 }
3667 
3668 static const struct attribute_group hwmon_attrgroup = {
3669 	.attrs = hwmon_attributes,
3670 	.is_visible = hwmon_attributes_visible,
3671 };
3672 
3673 static const struct attribute_group *hwmon_groups[] = {
3674 	&hwmon_attrgroup,
3675 	NULL
3676 };
3677 
3678 static int amdgpu_retrieve_od_settings(struct amdgpu_device *adev,
3679 				       enum pp_clock_type od_type,
3680 				       char *buf)
3681 {
3682 	int size = 0;
3683 	int ret;
3684 
3685 	if (amdgpu_in_reset(adev))
3686 		return -EPERM;
3687 	if (adev->in_suspend && !adev->in_runpm)
3688 		return -EPERM;
3689 
3690 	ret = pm_runtime_get_if_active(adev->dev);
3691 	if (ret <= 0)
3692 		return ret ?: -EPERM;
3693 
3694 	size = amdgpu_dpm_print_clock_levels(adev, od_type, buf);
3695 	if (size == 0)
3696 		size = sysfs_emit(buf, "\n");
3697 
3698 	pm_runtime_put_autosuspend(adev->dev);
3699 
3700 	return size;
3701 }
3702 
3703 static int parse_input_od_command_lines(const char *buf,
3704 					size_t count,
3705 					u32 *type,
3706 					long *params,
3707 					uint32_t *num_of_params)
3708 {
3709 	const char delimiter[3] = {' ', '\n', '\0'};
3710 	uint32_t parameter_size = 0;
3711 	char buf_cpy[128] = {0};
3712 	char *tmp_str, *sub_str;
3713 	int ret;
3714 
3715 	if (count > sizeof(buf_cpy) - 1)
3716 		return -EINVAL;
3717 
3718 	memcpy(buf_cpy, buf, count);
3719 	tmp_str = buf_cpy;
3720 
3721 	/* skip heading spaces */
3722 	while (isspace(*tmp_str))
3723 		tmp_str++;
3724 
3725 	switch (*tmp_str) {
3726 	case 'c':
3727 		*type = PP_OD_COMMIT_DPM_TABLE;
3728 		return 0;
3729 	case 'r':
3730 		params[parameter_size] = *type;
3731 		*num_of_params = 1;
3732 		*type = PP_OD_RESTORE_DEFAULT_TABLE;
3733 		return 0;
3734 	default:
3735 		break;
3736 	}
3737 
3738 	while ((sub_str = strsep(&tmp_str, delimiter)) != NULL) {
3739 		if (strlen(sub_str) == 0)
3740 			continue;
3741 
3742 		ret = kstrtol(sub_str, 0, &params[parameter_size]);
3743 		if (ret)
3744 			return -EINVAL;
3745 		parameter_size++;
3746 
3747 		while (isspace(*tmp_str))
3748 			tmp_str++;
3749 	}
3750 
3751 	*num_of_params = parameter_size;
3752 
3753 	return 0;
3754 }
3755 
3756 static int
3757 amdgpu_distribute_custom_od_settings(struct amdgpu_device *adev,
3758 				     enum PP_OD_DPM_TABLE_COMMAND cmd_type,
3759 				     const char *in_buf,
3760 				     size_t count)
3761 {
3762 	uint32_t parameter_size = 0;
3763 	long parameter[64];
3764 	int ret;
3765 
3766 	if (amdgpu_in_reset(adev))
3767 		return -EPERM;
3768 	if (adev->in_suspend && !adev->in_runpm)
3769 		return -EPERM;
3770 
3771 	ret = parse_input_od_command_lines(in_buf,
3772 					   count,
3773 					   &cmd_type,
3774 					   parameter,
3775 					   &parameter_size);
3776 	if (ret)
3777 		return ret;
3778 
3779 	ret = pm_runtime_resume_and_get(adev->dev);
3780 	if (ret < 0)
3781 		return ret;
3782 
3783 	ret = amdgpu_dpm_odn_edit_dpm_table(adev,
3784 					    cmd_type,
3785 					    parameter,
3786 					    parameter_size);
3787 	if (ret)
3788 		goto err_out;
3789 
3790 	if (cmd_type == PP_OD_COMMIT_DPM_TABLE) {
3791 		ret = amdgpu_dpm_dispatch_task(adev,
3792 					       AMD_PP_TASK_READJUST_POWER_STATE,
3793 					       NULL);
3794 		if (ret)
3795 			goto err_out;
3796 	}
3797 
3798 	pm_runtime_mark_last_busy(adev->dev);
3799 	pm_runtime_put_autosuspend(adev->dev);
3800 
3801 	return count;
3802 
3803 err_out:
3804 	pm_runtime_mark_last_busy(adev->dev);
3805 	pm_runtime_put_autosuspend(adev->dev);
3806 
3807 	return ret;
3808 }
3809 
3810 /**
3811  * DOC: fan_curve
3812  *
3813  * The amdgpu driver provides a sysfs API for checking and adjusting the fan
3814  * control curve line.
3815  *
3816  * Reading back the file shows you the current settings(temperature in Celsius
3817  * degree and fan speed in pwm) applied to every anchor point of the curve line
3818  * and their permitted ranges if changable.
3819  *
3820  * Writing a desired string(with the format like "anchor_point_index temperature
3821  * fan_speed_in_pwm") to the file, change the settings for the specific anchor
3822  * point accordingly.
3823  *
3824  * When you have finished the editing, write "c" (commit) to the file to commit
3825  * your changes.
3826  *
3827  * If you want to reset to the default value, write "r" (reset) to the file to
3828  * reset them
3829  *
3830  * There are two fan control modes supported: auto and manual. With auto mode,
3831  * PMFW handles the fan speed control(how fan speed reacts to ASIC temperature).
3832  * While with manual mode, users can set their own fan curve line as what
3833  * described here. Normally the ASIC is booted up with auto mode. Any
3834  * settings via this interface will switch the fan control to manual mode
3835  * implicitly.
3836  */
3837 static ssize_t fan_curve_show(struct kobject *kobj,
3838 			      struct kobj_attribute *attr,
3839 			      char *buf)
3840 {
3841 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
3842 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
3843 
3844 	return (ssize_t)amdgpu_retrieve_od_settings(adev, OD_FAN_CURVE, buf);
3845 }
3846 
3847 static ssize_t fan_curve_store(struct kobject *kobj,
3848 			       struct kobj_attribute *attr,
3849 			       const char *buf,
3850 			       size_t count)
3851 {
3852 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
3853 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
3854 
3855 	return (ssize_t)amdgpu_distribute_custom_od_settings(adev,
3856 							     PP_OD_EDIT_FAN_CURVE,
3857 							     buf,
3858 							     count);
3859 }
3860 
3861 static umode_t fan_curve_visible(struct amdgpu_device *adev)
3862 {
3863 	umode_t umode = 0000;
3864 
3865 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_FAN_CURVE_RETRIEVE)
3866 		umode |= S_IRUSR | S_IRGRP | S_IROTH;
3867 
3868 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_FAN_CURVE_SET)
3869 		umode |= S_IWUSR;
3870 
3871 	return umode;
3872 }
3873 
3874 /**
3875  * DOC: acoustic_limit_rpm_threshold
3876  *
3877  * The amdgpu driver provides a sysfs API for checking and adjusting the
3878  * acoustic limit in RPM for fan control.
3879  *
3880  * Reading back the file shows you the current setting and the permitted
3881  * ranges if changable.
3882  *
3883  * Writing an integer to the file, change the setting accordingly.
3884  *
3885  * When you have finished the editing, write "c" (commit) to the file to commit
3886  * your changes.
3887  *
3888  * If you want to reset to the default value, write "r" (reset) to the file to
3889  * reset them
3890  *
3891  * This setting works under auto fan control mode only. It adjusts the PMFW's
3892  * behavior about the maximum speed in RPM the fan can spin. Setting via this
3893  * interface will switch the fan control to auto mode implicitly.
3894  */
3895 static ssize_t acoustic_limit_threshold_show(struct kobject *kobj,
3896 					     struct kobj_attribute *attr,
3897 					     char *buf)
3898 {
3899 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
3900 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
3901 
3902 	return (ssize_t)amdgpu_retrieve_od_settings(adev, OD_ACOUSTIC_LIMIT, buf);
3903 }
3904 
3905 static ssize_t acoustic_limit_threshold_store(struct kobject *kobj,
3906 					      struct kobj_attribute *attr,
3907 					      const char *buf,
3908 					      size_t count)
3909 {
3910 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
3911 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
3912 
3913 	return (ssize_t)amdgpu_distribute_custom_od_settings(adev,
3914 							     PP_OD_EDIT_ACOUSTIC_LIMIT,
3915 							     buf,
3916 							     count);
3917 }
3918 
3919 static umode_t acoustic_limit_threshold_visible(struct amdgpu_device *adev)
3920 {
3921 	umode_t umode = 0000;
3922 
3923 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_ACOUSTIC_LIMIT_THRESHOLD_RETRIEVE)
3924 		umode |= S_IRUSR | S_IRGRP | S_IROTH;
3925 
3926 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_ACOUSTIC_LIMIT_THRESHOLD_SET)
3927 		umode |= S_IWUSR;
3928 
3929 	return umode;
3930 }
3931 
3932 /**
3933  * DOC: acoustic_target_rpm_threshold
3934  *
3935  * The amdgpu driver provides a sysfs API for checking and adjusting the
3936  * acoustic target in RPM for fan control.
3937  *
3938  * Reading back the file shows you the current setting and the permitted
3939  * ranges if changable.
3940  *
3941  * Writing an integer to the file, change the setting accordingly.
3942  *
3943  * When you have finished the editing, write "c" (commit) to the file to commit
3944  * your changes.
3945  *
3946  * If you want to reset to the default value, write "r" (reset) to the file to
3947  * reset them
3948  *
3949  * This setting works under auto fan control mode only. It can co-exist with
3950  * other settings which can work also under auto mode. It adjusts the PMFW's
3951  * behavior about the maximum speed in RPM the fan can spin when ASIC
3952  * temperature is not greater than target temperature. Setting via this
3953  * interface will switch the fan control to auto mode implicitly.
3954  */
3955 static ssize_t acoustic_target_threshold_show(struct kobject *kobj,
3956 					      struct kobj_attribute *attr,
3957 					      char *buf)
3958 {
3959 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
3960 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
3961 
3962 	return (ssize_t)amdgpu_retrieve_od_settings(adev, OD_ACOUSTIC_TARGET, buf);
3963 }
3964 
3965 static ssize_t acoustic_target_threshold_store(struct kobject *kobj,
3966 					       struct kobj_attribute *attr,
3967 					       const char *buf,
3968 					       size_t count)
3969 {
3970 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
3971 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
3972 
3973 	return (ssize_t)amdgpu_distribute_custom_od_settings(adev,
3974 							     PP_OD_EDIT_ACOUSTIC_TARGET,
3975 							     buf,
3976 							     count);
3977 }
3978 
3979 static umode_t acoustic_target_threshold_visible(struct amdgpu_device *adev)
3980 {
3981 	umode_t umode = 0000;
3982 
3983 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_ACOUSTIC_TARGET_THRESHOLD_RETRIEVE)
3984 		umode |= S_IRUSR | S_IRGRP | S_IROTH;
3985 
3986 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_ACOUSTIC_TARGET_THRESHOLD_SET)
3987 		umode |= S_IWUSR;
3988 
3989 	return umode;
3990 }
3991 
3992 /**
3993  * DOC: fan_target_temperature
3994  *
3995  * The amdgpu driver provides a sysfs API for checking and adjusting the
3996  * target tempeature in Celsius degree for fan control.
3997  *
3998  * Reading back the file shows you the current setting and the permitted
3999  * ranges if changable.
4000  *
4001  * Writing an integer to the file, change the setting accordingly.
4002  *
4003  * When you have finished the editing, write "c" (commit) to the file to commit
4004  * your changes.
4005  *
4006  * If you want to reset to the default value, write "r" (reset) to the file to
4007  * reset them
4008  *
4009  * This setting works under auto fan control mode only. It can co-exist with
4010  * other settings which can work also under auto mode. Paring with the
4011  * acoustic_target_rpm_threshold setting, they define the maximum speed in
4012  * RPM the fan can spin when ASIC temperature is not greater than target
4013  * temperature. Setting via this interface will switch the fan control to
4014  * auto mode implicitly.
4015  */
4016 static ssize_t fan_target_temperature_show(struct kobject *kobj,
4017 					   struct kobj_attribute *attr,
4018 					   char *buf)
4019 {
4020 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
4021 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
4022 
4023 	return (ssize_t)amdgpu_retrieve_od_settings(adev, OD_FAN_TARGET_TEMPERATURE, buf);
4024 }
4025 
4026 static ssize_t fan_target_temperature_store(struct kobject *kobj,
4027 					    struct kobj_attribute *attr,
4028 					    const char *buf,
4029 					    size_t count)
4030 {
4031 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
4032 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
4033 
4034 	return (ssize_t)amdgpu_distribute_custom_od_settings(adev,
4035 							     PP_OD_EDIT_FAN_TARGET_TEMPERATURE,
4036 							     buf,
4037 							     count);
4038 }
4039 
4040 static umode_t fan_target_temperature_visible(struct amdgpu_device *adev)
4041 {
4042 	umode_t umode = 0000;
4043 
4044 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_FAN_TARGET_TEMPERATURE_RETRIEVE)
4045 		umode |= S_IRUSR | S_IRGRP | S_IROTH;
4046 
4047 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_FAN_TARGET_TEMPERATURE_SET)
4048 		umode |= S_IWUSR;
4049 
4050 	return umode;
4051 }
4052 
4053 /**
4054  * DOC: fan_minimum_pwm
4055  *
4056  * The amdgpu driver provides a sysfs API for checking and adjusting the
4057  * minimum fan speed in PWM.
4058  *
4059  * Reading back the file shows you the current setting and the permitted
4060  * ranges if changable.
4061  *
4062  * Writing an integer to the file, change the setting accordingly.
4063  *
4064  * When you have finished the editing, write "c" (commit) to the file to commit
4065  * your changes.
4066  *
4067  * If you want to reset to the default value, write "r" (reset) to the file to
4068  * reset them
4069  *
4070  * This setting works under auto fan control mode only. It can co-exist with
4071  * other settings which can work also under auto mode. It adjusts the PMFW's
4072  * behavior about the minimum fan speed in PWM the fan should spin. Setting
4073  * via this interface will switch the fan control to auto mode implicitly.
4074  */
4075 static ssize_t fan_minimum_pwm_show(struct kobject *kobj,
4076 				    struct kobj_attribute *attr,
4077 				    char *buf)
4078 {
4079 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
4080 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
4081 
4082 	return (ssize_t)amdgpu_retrieve_od_settings(adev, OD_FAN_MINIMUM_PWM, buf);
4083 }
4084 
4085 static ssize_t fan_minimum_pwm_store(struct kobject *kobj,
4086 				     struct kobj_attribute *attr,
4087 				     const char *buf,
4088 				     size_t count)
4089 {
4090 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
4091 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
4092 
4093 	return (ssize_t)amdgpu_distribute_custom_od_settings(adev,
4094 							     PP_OD_EDIT_FAN_MINIMUM_PWM,
4095 							     buf,
4096 							     count);
4097 }
4098 
4099 static umode_t fan_minimum_pwm_visible(struct amdgpu_device *adev)
4100 {
4101 	umode_t umode = 0000;
4102 
4103 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_FAN_MINIMUM_PWM_RETRIEVE)
4104 		umode |= S_IRUSR | S_IRGRP | S_IROTH;
4105 
4106 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_FAN_MINIMUM_PWM_SET)
4107 		umode |= S_IWUSR;
4108 
4109 	return umode;
4110 }
4111 
4112 /**
4113  * DOC: fan_zero_rpm_enable
4114  *
4115  * The amdgpu driver provides a sysfs API for checking and adjusting the
4116  * zero RPM feature.
4117  *
4118  * Reading back the file shows you the current setting and the permitted
4119  * ranges if changable.
4120  *
4121  * Writing an integer to the file, change the setting accordingly.
4122  *
4123  * When you have finished the editing, write "c" (commit) to the file to commit
4124  * your changes.
4125  *
4126  * If you want to reset to the default value, write "r" (reset) to the file to
4127  * reset them.
4128  */
4129 static ssize_t fan_zero_rpm_enable_show(struct kobject *kobj,
4130 					   struct kobj_attribute *attr,
4131 					   char *buf)
4132 {
4133 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
4134 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
4135 
4136 	return (ssize_t)amdgpu_retrieve_od_settings(adev, OD_FAN_ZERO_RPM_ENABLE, buf);
4137 }
4138 
4139 static ssize_t fan_zero_rpm_enable_store(struct kobject *kobj,
4140 					    struct kobj_attribute *attr,
4141 					    const char *buf,
4142 					    size_t count)
4143 {
4144 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
4145 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
4146 
4147 	return (ssize_t)amdgpu_distribute_custom_od_settings(adev,
4148 							     PP_OD_EDIT_FAN_ZERO_RPM_ENABLE,
4149 							     buf,
4150 							     count);
4151 }
4152 
4153 static umode_t fan_zero_rpm_enable_visible(struct amdgpu_device *adev)
4154 {
4155 	umode_t umode = 0000;
4156 
4157 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_FAN_ZERO_RPM_ENABLE_RETRIEVE)
4158 		umode |= S_IRUSR | S_IRGRP | S_IROTH;
4159 
4160 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_FAN_ZERO_RPM_ENABLE_SET)
4161 		umode |= S_IWUSR;
4162 
4163 	return umode;
4164 }
4165 
4166 /**
4167  * DOC: fan_zero_rpm_stop_temperature
4168  *
4169  * The amdgpu driver provides a sysfs API for checking and adjusting the
4170  * zero RPM stop temperature feature.
4171  *
4172  * Reading back the file shows you the current setting and the permitted
4173  * ranges if changable.
4174  *
4175  * Writing an integer to the file, change the setting accordingly.
4176  *
4177  * When you have finished the editing, write "c" (commit) to the file to commit
4178  * your changes.
4179  *
4180  * If you want to reset to the default value, write "r" (reset) to the file to
4181  * reset them.
4182  *
4183  * This setting works only if the Zero RPM setting is enabled. It adjusts the
4184  * temperature below which the fan can stop.
4185  */
4186 static ssize_t fan_zero_rpm_stop_temp_show(struct kobject *kobj,
4187 					   struct kobj_attribute *attr,
4188 					   char *buf)
4189 {
4190 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
4191 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
4192 
4193 	return (ssize_t)amdgpu_retrieve_od_settings(adev, OD_FAN_ZERO_RPM_STOP_TEMP, buf);
4194 }
4195 
4196 static ssize_t fan_zero_rpm_stop_temp_store(struct kobject *kobj,
4197 					    struct kobj_attribute *attr,
4198 					    const char *buf,
4199 					    size_t count)
4200 {
4201 	struct od_kobj *container = container_of(kobj, struct od_kobj, kobj);
4202 	struct amdgpu_device *adev = (struct amdgpu_device *)container->priv;
4203 
4204 	return (ssize_t)amdgpu_distribute_custom_od_settings(adev,
4205 							     PP_OD_EDIT_FAN_ZERO_RPM_STOP_TEMP,
4206 							     buf,
4207 							     count);
4208 }
4209 
4210 static umode_t fan_zero_rpm_stop_temp_visible(struct amdgpu_device *adev)
4211 {
4212 	umode_t umode = 0000;
4213 
4214 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_FAN_ZERO_RPM_STOP_TEMP_RETRIEVE)
4215 		umode |= S_IRUSR | S_IRGRP | S_IROTH;
4216 
4217 	if (adev->pm.od_feature_mask & OD_OPS_SUPPORT_FAN_ZERO_RPM_STOP_TEMP_SET)
4218 		umode |= S_IWUSR;
4219 
4220 	return umode;
4221 }
4222 
4223 static struct od_feature_set amdgpu_od_set = {
4224 	.containers = {
4225 		[0] = {
4226 			.name = "fan_ctrl",
4227 			.sub_feature = {
4228 				[0] = {
4229 					.name = "fan_curve",
4230 					.ops = {
4231 						.is_visible = fan_curve_visible,
4232 						.show = fan_curve_show,
4233 						.store = fan_curve_store,
4234 					},
4235 				},
4236 				[1] = {
4237 					.name = "acoustic_limit_rpm_threshold",
4238 					.ops = {
4239 						.is_visible = acoustic_limit_threshold_visible,
4240 						.show = acoustic_limit_threshold_show,
4241 						.store = acoustic_limit_threshold_store,
4242 					},
4243 				},
4244 				[2] = {
4245 					.name = "acoustic_target_rpm_threshold",
4246 					.ops = {
4247 						.is_visible = acoustic_target_threshold_visible,
4248 						.show = acoustic_target_threshold_show,
4249 						.store = acoustic_target_threshold_store,
4250 					},
4251 				},
4252 				[3] = {
4253 					.name = "fan_target_temperature",
4254 					.ops = {
4255 						.is_visible = fan_target_temperature_visible,
4256 						.show = fan_target_temperature_show,
4257 						.store = fan_target_temperature_store,
4258 					},
4259 				},
4260 				[4] = {
4261 					.name = "fan_minimum_pwm",
4262 					.ops = {
4263 						.is_visible = fan_minimum_pwm_visible,
4264 						.show = fan_minimum_pwm_show,
4265 						.store = fan_minimum_pwm_store,
4266 					},
4267 				},
4268 				[5] = {
4269 					.name = "fan_zero_rpm_enable",
4270 					.ops = {
4271 						.is_visible = fan_zero_rpm_enable_visible,
4272 						.show = fan_zero_rpm_enable_show,
4273 						.store = fan_zero_rpm_enable_store,
4274 					},
4275 				},
4276 				[6] = {
4277 					.name = "fan_zero_rpm_stop_temperature",
4278 					.ops = {
4279 						.is_visible = fan_zero_rpm_stop_temp_visible,
4280 						.show = fan_zero_rpm_stop_temp_show,
4281 						.store = fan_zero_rpm_stop_temp_store,
4282 					},
4283 				},
4284 			},
4285 		},
4286 	},
4287 };
4288 
4289 static void od_kobj_release(struct kobject *kobj)
4290 {
4291 	struct od_kobj *od_kobj = container_of(kobj, struct od_kobj, kobj);
4292 
4293 	kfree(od_kobj);
4294 }
4295 
4296 static const struct kobj_type od_ktype = {
4297 	.release	= od_kobj_release,
4298 	.sysfs_ops	= &kobj_sysfs_ops,
4299 };
4300 
4301 static void amdgpu_od_set_fini(struct amdgpu_device *adev)
4302 {
4303 	struct od_kobj *container, *container_next;
4304 	struct od_attribute *attribute, *attribute_next;
4305 
4306 	if (list_empty(&adev->pm.od_kobj_list))
4307 		return;
4308 
4309 	list_for_each_entry_safe(container, container_next,
4310 				 &adev->pm.od_kobj_list, entry) {
4311 		list_del(&container->entry);
4312 
4313 		list_for_each_entry_safe(attribute, attribute_next,
4314 					 &container->attribute, entry) {
4315 			list_del(&attribute->entry);
4316 			sysfs_remove_file(&container->kobj,
4317 					  &attribute->attribute.attr);
4318 			kfree(attribute);
4319 		}
4320 
4321 		kobject_put(&container->kobj);
4322 	}
4323 }
4324 
4325 static bool amdgpu_is_od_feature_supported(struct amdgpu_device *adev,
4326 					   struct od_feature_ops *feature_ops)
4327 {
4328 	umode_t mode;
4329 
4330 	if (!feature_ops->is_visible)
4331 		return false;
4332 
4333 	/*
4334 	 * If the feature has no user read and write mode set,
4335 	 * we can assume the feature is actually not supported.(?)
4336 	 * And the revelant sysfs interface should not be exposed.
4337 	 */
4338 	mode = feature_ops->is_visible(adev);
4339 	if (mode & (S_IRUSR | S_IWUSR))
4340 		return true;
4341 
4342 	return false;
4343 }
4344 
4345 static bool amdgpu_od_is_self_contained(struct amdgpu_device *adev,
4346 					struct od_feature_container *container)
4347 {
4348 	int i;
4349 
4350 	/*
4351 	 * If there is no valid entry within the container, the container
4352 	 * is recognized as a self contained container. And the valid entry
4353 	 * here means it has a valid naming and it is visible/supported by
4354 	 * the ASIC.
4355 	 */
4356 	for (i = 0; i < ARRAY_SIZE(container->sub_feature); i++) {
4357 		if (container->sub_feature[i].name &&
4358 		    amdgpu_is_od_feature_supported(adev,
4359 			&container->sub_feature[i].ops))
4360 			return false;
4361 	}
4362 
4363 	return true;
4364 }
4365 
4366 static int amdgpu_od_set_init(struct amdgpu_device *adev)
4367 {
4368 	struct od_kobj *top_set, *sub_set;
4369 	struct od_attribute *attribute;
4370 	struct od_feature_container *container;
4371 	struct od_feature_item *feature;
4372 	int i, j;
4373 	int ret;
4374 
4375 	/* Setup the top `gpu_od` directory which holds all other OD interfaces */
4376 	top_set = kzalloc(sizeof(*top_set), GFP_KERNEL);
4377 	if (!top_set)
4378 		return -ENOMEM;
4379 	list_add(&top_set->entry, &adev->pm.od_kobj_list);
4380 
4381 	ret = kobject_init_and_add(&top_set->kobj,
4382 				   &od_ktype,
4383 				   &adev->dev->kobj,
4384 				   "%s",
4385 				   "gpu_od");
4386 	if (ret)
4387 		goto err_out;
4388 	INIT_LIST_HEAD(&top_set->attribute);
4389 	top_set->priv = adev;
4390 
4391 	for (i = 0; i < ARRAY_SIZE(amdgpu_od_set.containers); i++) {
4392 		container = &amdgpu_od_set.containers[i];
4393 
4394 		if (!container->name)
4395 			continue;
4396 
4397 		/*
4398 		 * If there is valid entries within the container, the container
4399 		 * will be presented as a sub directory and all its holding entries
4400 		 * will be presented as plain files under it.
4401 		 * While if there is no valid entry within the container, the container
4402 		 * itself will be presented as a plain file under top `gpu_od` directory.
4403 		 */
4404 		if (amdgpu_od_is_self_contained(adev, container)) {
4405 			if (!amdgpu_is_od_feature_supported(adev,
4406 			     &container->ops))
4407 				continue;
4408 
4409 			/*
4410 			 * The container is presented as a plain file under top `gpu_od`
4411 			 * directory.
4412 			 */
4413 			attribute = kzalloc(sizeof(*attribute), GFP_KERNEL);
4414 			if (!attribute) {
4415 				ret = -ENOMEM;
4416 				goto err_out;
4417 			}
4418 			list_add(&attribute->entry, &top_set->attribute);
4419 
4420 			attribute->attribute.attr.mode =
4421 					container->ops.is_visible(adev);
4422 			attribute->attribute.attr.name = container->name;
4423 			attribute->attribute.show =
4424 					container->ops.show;
4425 			attribute->attribute.store =
4426 					container->ops.store;
4427 			ret = sysfs_create_file(&top_set->kobj,
4428 						&attribute->attribute.attr);
4429 			if (ret)
4430 				goto err_out;
4431 		} else {
4432 			/* The container is presented as a sub directory. */
4433 			sub_set = kzalloc(sizeof(*sub_set), GFP_KERNEL);
4434 			if (!sub_set) {
4435 				ret = -ENOMEM;
4436 				goto err_out;
4437 			}
4438 			list_add(&sub_set->entry, &adev->pm.od_kobj_list);
4439 
4440 			ret = kobject_init_and_add(&sub_set->kobj,
4441 						   &od_ktype,
4442 						   &top_set->kobj,
4443 						   "%s",
4444 						   container->name);
4445 			if (ret)
4446 				goto err_out;
4447 			INIT_LIST_HEAD(&sub_set->attribute);
4448 			sub_set->priv = adev;
4449 
4450 			for (j = 0; j < ARRAY_SIZE(container->sub_feature); j++) {
4451 				feature = &container->sub_feature[j];
4452 				if (!feature->name)
4453 					continue;
4454 
4455 				if (!amdgpu_is_od_feature_supported(adev,
4456 				     &feature->ops))
4457 					continue;
4458 
4459 				/*
4460 				 * With the container presented as a sub directory, the entry within
4461 				 * it is presented as a plain file under the sub directory.
4462 				 */
4463 				attribute = kzalloc(sizeof(*attribute), GFP_KERNEL);
4464 				if (!attribute) {
4465 					ret = -ENOMEM;
4466 					goto err_out;
4467 				}
4468 				list_add(&attribute->entry, &sub_set->attribute);
4469 
4470 				attribute->attribute.attr.mode =
4471 						feature->ops.is_visible(adev);
4472 				attribute->attribute.attr.name = feature->name;
4473 				attribute->attribute.show =
4474 						feature->ops.show;
4475 				attribute->attribute.store =
4476 						feature->ops.store;
4477 				ret = sysfs_create_file(&sub_set->kobj,
4478 							&attribute->attribute.attr);
4479 				if (ret)
4480 					goto err_out;
4481 			}
4482 		}
4483 	}
4484 
4485 	/*
4486 	 * If gpu_od is the only member in the list, that means gpu_od is an
4487 	 * empty directory, so remove it.
4488 	 */
4489 	if (list_is_singular(&adev->pm.od_kobj_list))
4490 		goto err_out;
4491 
4492 	return 0;
4493 
4494 err_out:
4495 	amdgpu_od_set_fini(adev);
4496 
4497 	return ret;
4498 }
4499 
4500 int amdgpu_pm_sysfs_init(struct amdgpu_device *adev)
4501 {
4502 	enum amdgpu_sriov_vf_mode mode;
4503 	uint32_t mask = 0;
4504 	int ret;
4505 
4506 	if (adev->pm.sysfs_initialized)
4507 		return 0;
4508 
4509 	INIT_LIST_HEAD(&adev->pm.pm_attr_list);
4510 
4511 	if (adev->pm.dpm_enabled == 0)
4512 		return 0;
4513 
4514 	mode = amdgpu_virt_get_sriov_vf_mode(adev);
4515 
4516 	/* under multi-vf mode, the hwmon attributes are all not supported */
4517 	if (mode != SRIOV_VF_MODE_MULTI_VF) {
4518 		adev->pm.int_hwmon_dev = hwmon_device_register_with_groups(adev->dev,
4519 									DRIVER_NAME, adev,
4520 									hwmon_groups);
4521 		if (IS_ERR(adev->pm.int_hwmon_dev)) {
4522 			ret = PTR_ERR(adev->pm.int_hwmon_dev);
4523 			dev_err(adev->dev, "Unable to register hwmon device: %d\n", ret);
4524 			return ret;
4525 		}
4526 	}
4527 
4528 	switch (mode) {
4529 	case SRIOV_VF_MODE_ONE_VF:
4530 		mask = ATTR_FLAG_ONEVF;
4531 		break;
4532 	case SRIOV_VF_MODE_MULTI_VF:
4533 		mask = 0;
4534 		break;
4535 	case SRIOV_VF_MODE_BARE_METAL:
4536 	default:
4537 		mask = ATTR_FLAG_MASK_ALL;
4538 		break;
4539 	}
4540 
4541 	ret = amdgpu_device_attr_create_groups(adev,
4542 					       amdgpu_device_attrs,
4543 					       ARRAY_SIZE(amdgpu_device_attrs),
4544 					       mask,
4545 					       &adev->pm.pm_attr_list);
4546 	if (ret)
4547 		goto err_out0;
4548 
4549 	if (amdgpu_dpm_is_overdrive_supported(adev)) {
4550 		ret = amdgpu_od_set_init(adev);
4551 		if (ret)
4552 			goto err_out1;
4553 	} else if (adev->pm.pp_feature & PP_OVERDRIVE_MASK) {
4554 		dev_info(adev->dev, "overdrive feature is not supported\n");
4555 	}
4556 
4557 	if (amdgpu_dpm_get_pm_policy_info(adev, PP_PM_POLICY_NONE, NULL) !=
4558 	    -EOPNOTSUPP) {
4559 		ret = devm_device_add_group(adev->dev,
4560 					    &amdgpu_pm_policy_attr_group);
4561 		if (ret)
4562 			goto err_out0;
4563 	}
4564 
4565 	adev->pm.sysfs_initialized = true;
4566 
4567 	return 0;
4568 
4569 err_out1:
4570 	amdgpu_device_attr_remove_groups(adev, &adev->pm.pm_attr_list);
4571 err_out0:
4572 	if (adev->pm.int_hwmon_dev)
4573 		hwmon_device_unregister(adev->pm.int_hwmon_dev);
4574 
4575 	return ret;
4576 }
4577 
4578 void amdgpu_pm_sysfs_fini(struct amdgpu_device *adev)
4579 {
4580 	amdgpu_od_set_fini(adev);
4581 
4582 	if (adev->pm.int_hwmon_dev)
4583 		hwmon_device_unregister(adev->pm.int_hwmon_dev);
4584 
4585 	amdgpu_device_attr_remove_groups(adev, &adev->pm.pm_attr_list);
4586 }
4587 
4588 /*
4589  * Debugfs info
4590  */
4591 #if defined(CONFIG_DEBUG_FS)
4592 
4593 static void amdgpu_debugfs_prints_cpu_info(struct seq_file *m,
4594 					   struct amdgpu_device *adev)
4595 {
4596 	uint16_t *p_val;
4597 	uint32_t size;
4598 	int i;
4599 	uint32_t num_cpu_cores = amdgpu_dpm_get_num_cpu_cores(adev);
4600 
4601 	if (amdgpu_dpm_is_cclk_dpm_supported(adev)) {
4602 		p_val = kcalloc(num_cpu_cores, sizeof(uint16_t),
4603 				GFP_KERNEL);
4604 
4605 		if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_CPU_CLK,
4606 					    (void *)p_val, &size)) {
4607 			for (i = 0; i < num_cpu_cores; i++)
4608 				seq_printf(m, "\t%u MHz (CPU%d)\n",
4609 					   *(p_val + i), i);
4610 		}
4611 
4612 		kfree(p_val);
4613 	}
4614 }
4615 
4616 static int amdgpu_debugfs_pm_info_pp(struct seq_file *m, struct amdgpu_device *adev)
4617 {
4618 	uint32_t mp1_ver = amdgpu_ip_version(adev, MP1_HWIP, 0);
4619 	uint32_t gc_ver = amdgpu_ip_version(adev, GC_HWIP, 0);
4620 	uint32_t value;
4621 	uint64_t value64 = 0;
4622 	uint32_t query = 0;
4623 	int size;
4624 
4625 	/* GPU Clocks */
4626 	size = sizeof(value);
4627 	seq_printf(m, "GFX Clocks and Power:\n");
4628 
4629 	amdgpu_debugfs_prints_cpu_info(m, adev);
4630 
4631 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GFX_MCLK, (void *)&value, &size))
4632 		seq_printf(m, "\t%u MHz (MCLK)\n", value/100);
4633 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GFX_SCLK, (void *)&value, &size))
4634 		seq_printf(m, "\t%u MHz (SCLK)\n", value/100);
4635 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_STABLE_PSTATE_SCLK, (void *)&value, &size))
4636 		seq_printf(m, "\t%u MHz (PSTATE_SCLK)\n", value/100);
4637 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_STABLE_PSTATE_MCLK, (void *)&value, &size))
4638 		seq_printf(m, "\t%u MHz (PSTATE_MCLK)\n", value/100);
4639 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_VDDGFX, (void *)&value, &size))
4640 		seq_printf(m, "\t%u mV (VDDGFX)\n", value);
4641 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_VDDNB, (void *)&value, &size))
4642 		seq_printf(m, "\t%u mV (VDDNB)\n", value);
4643 	size = sizeof(uint32_t);
4644 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GPU_AVG_POWER, (void *)&query, &size)) {
4645 		if (adev->flags & AMD_IS_APU)
4646 			seq_printf(m, "\t%u.%02u W (average SoC including CPU)\n", query >> 8, query & 0xff);
4647 		else
4648 			seq_printf(m, "\t%u.%02u W (average SoC)\n", query >> 8, query & 0xff);
4649 	}
4650 	size = sizeof(uint32_t);
4651 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GPU_INPUT_POWER, (void *)&query, &size)) {
4652 		if (adev->flags & AMD_IS_APU)
4653 			seq_printf(m, "\t%u.%02u W (current SoC including CPU)\n", query >> 8, query & 0xff);
4654 		else
4655 			seq_printf(m, "\t%u.%02u W (current SoC)\n", query >> 8, query & 0xff);
4656 	}
4657 	size = sizeof(value);
4658 	seq_printf(m, "\n");
4659 
4660 	/* GPU Temp */
4661 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GPU_TEMP, (void *)&value, &size))
4662 		seq_printf(m, "GPU Temperature: %u C\n", value/1000);
4663 
4664 	/* GPU Load */
4665 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_GPU_LOAD, (void *)&value, &size))
4666 		seq_printf(m, "GPU Load: %u %%\n", value);
4667 	/* MEM Load */
4668 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_MEM_LOAD, (void *)&value, &size))
4669 		seq_printf(m, "MEM Load: %u %%\n", value);
4670 	/* VCN Load */
4671 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_VCN_LOAD, (void *)&value, &size))
4672 		seq_printf(m, "VCN Load: %u %%\n", value);
4673 
4674 	seq_printf(m, "\n");
4675 
4676 	/* SMC feature mask */
4677 	if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_ENABLED_SMC_FEATURES_MASK, (void *)&value64, &size))
4678 		seq_printf(m, "SMC Feature Mask: 0x%016llx\n", value64);
4679 
4680 	/* ASICs greater than CHIP_VEGA20 supports these sensors */
4681 	if (gc_ver != IP_VERSION(9, 4, 0) && mp1_ver > IP_VERSION(9, 0, 0)) {
4682 		/* VCN clocks */
4683 		if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_VCN_POWER_STATE, (void *)&value, &size)) {
4684 			if (!value) {
4685 				seq_printf(m, "VCN: Powered down\n");
4686 			} else {
4687 				seq_printf(m, "VCN: Powered up\n");
4688 				if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_UVD_DCLK, (void *)&value, &size))
4689 					seq_printf(m, "\t%u MHz (DCLK)\n", value/100);
4690 				if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_UVD_VCLK, (void *)&value, &size))
4691 					seq_printf(m, "\t%u MHz (VCLK)\n", value/100);
4692 			}
4693 		}
4694 		seq_printf(m, "\n");
4695 	} else {
4696 		/* UVD clocks */
4697 		if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_UVD_POWER, (void *)&value, &size)) {
4698 			if (!value) {
4699 				seq_printf(m, "UVD: Powered down\n");
4700 			} else {
4701 				seq_printf(m, "UVD: Powered up\n");
4702 				if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_UVD_DCLK, (void *)&value, &size))
4703 					seq_printf(m, "\t%u MHz (DCLK)\n", value/100);
4704 				if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_UVD_VCLK, (void *)&value, &size))
4705 					seq_printf(m, "\t%u MHz (VCLK)\n", value/100);
4706 			}
4707 		}
4708 		seq_printf(m, "\n");
4709 
4710 		/* VCE clocks */
4711 		if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_VCE_POWER, (void *)&value, &size)) {
4712 			if (!value) {
4713 				seq_printf(m, "VCE: Powered down\n");
4714 			} else {
4715 				seq_printf(m, "VCE: Powered up\n");
4716 				if (!amdgpu_dpm_read_sensor(adev, AMDGPU_PP_SENSOR_VCE_ECCLK, (void *)&value, &size))
4717 					seq_printf(m, "\t%u MHz (ECCLK)\n", value/100);
4718 			}
4719 		}
4720 	}
4721 
4722 	return 0;
4723 }
4724 
4725 static const struct cg_flag_name clocks[] = {
4726 	{AMD_CG_SUPPORT_GFX_FGCG, "Graphics Fine Grain Clock Gating"},
4727 	{AMD_CG_SUPPORT_GFX_MGCG, "Graphics Medium Grain Clock Gating"},
4728 	{AMD_CG_SUPPORT_GFX_MGLS, "Graphics Medium Grain memory Light Sleep"},
4729 	{AMD_CG_SUPPORT_GFX_CGCG, "Graphics Coarse Grain Clock Gating"},
4730 	{AMD_CG_SUPPORT_GFX_CGLS, "Graphics Coarse Grain memory Light Sleep"},
4731 	{AMD_CG_SUPPORT_GFX_CGTS, "Graphics Coarse Grain Tree Shader Clock Gating"},
4732 	{AMD_CG_SUPPORT_GFX_CGTS_LS, "Graphics Coarse Grain Tree Shader Light Sleep"},
4733 	{AMD_CG_SUPPORT_GFX_CP_LS, "Graphics Command Processor Light Sleep"},
4734 	{AMD_CG_SUPPORT_GFX_RLC_LS, "Graphics Run List Controller Light Sleep"},
4735 	{AMD_CG_SUPPORT_GFX_3D_CGCG, "Graphics 3D Coarse Grain Clock Gating"},
4736 	{AMD_CG_SUPPORT_GFX_3D_CGLS, "Graphics 3D Coarse Grain memory Light Sleep"},
4737 	{AMD_CG_SUPPORT_MC_LS, "Memory Controller Light Sleep"},
4738 	{AMD_CG_SUPPORT_MC_MGCG, "Memory Controller Medium Grain Clock Gating"},
4739 	{AMD_CG_SUPPORT_SDMA_LS, "System Direct Memory Access Light Sleep"},
4740 	{AMD_CG_SUPPORT_SDMA_MGCG, "System Direct Memory Access Medium Grain Clock Gating"},
4741 	{AMD_CG_SUPPORT_BIF_MGCG, "Bus Interface Medium Grain Clock Gating"},
4742 	{AMD_CG_SUPPORT_BIF_LS, "Bus Interface Light Sleep"},
4743 	{AMD_CG_SUPPORT_UVD_MGCG, "Unified Video Decoder Medium Grain Clock Gating"},
4744 	{AMD_CG_SUPPORT_VCE_MGCG, "Video Compression Engine Medium Grain Clock Gating"},
4745 	{AMD_CG_SUPPORT_HDP_LS, "Host Data Path Light Sleep"},
4746 	{AMD_CG_SUPPORT_HDP_MGCG, "Host Data Path Medium Grain Clock Gating"},
4747 	{AMD_CG_SUPPORT_DRM_MGCG, "Digital Right Management Medium Grain Clock Gating"},
4748 	{AMD_CG_SUPPORT_DRM_LS, "Digital Right Management Light Sleep"},
4749 	{AMD_CG_SUPPORT_ROM_MGCG, "Rom Medium Grain Clock Gating"},
4750 	{AMD_CG_SUPPORT_DF_MGCG, "Data Fabric Medium Grain Clock Gating"},
4751 	{AMD_CG_SUPPORT_VCN_MGCG, "VCN Medium Grain Clock Gating"},
4752 	{AMD_CG_SUPPORT_HDP_DS, "Host Data Path Deep Sleep"},
4753 	{AMD_CG_SUPPORT_HDP_SD, "Host Data Path Shutdown"},
4754 	{AMD_CG_SUPPORT_IH_CG, "Interrupt Handler Clock Gating"},
4755 	{AMD_CG_SUPPORT_JPEG_MGCG, "JPEG Medium Grain Clock Gating"},
4756 	{AMD_CG_SUPPORT_REPEATER_FGCG, "Repeater Fine Grain Clock Gating"},
4757 	{AMD_CG_SUPPORT_GFX_PERF_CLK, "Perfmon Clock Gating"},
4758 	{AMD_CG_SUPPORT_ATHUB_MGCG, "Address Translation Hub Medium Grain Clock Gating"},
4759 	{AMD_CG_SUPPORT_ATHUB_LS, "Address Translation Hub Light Sleep"},
4760 	{0, NULL},
4761 };
4762 
4763 static void amdgpu_parse_cg_state(struct seq_file *m, u64 flags)
4764 {
4765 	int i;
4766 
4767 	for (i = 0; clocks[i].flag; i++)
4768 		seq_printf(m, "\t%s: %s\n", clocks[i].name,
4769 			   (flags & clocks[i].flag) ? "On" : "Off");
4770 }
4771 
4772 static int amdgpu_debugfs_pm_info_show(struct seq_file *m, void *unused)
4773 {
4774 	struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
4775 	struct drm_device *dev = adev_to_drm(adev);
4776 	u64 flags = 0;
4777 	int r;
4778 
4779 	if (amdgpu_in_reset(adev))
4780 		return -EPERM;
4781 	if (adev->in_suspend && !adev->in_runpm)
4782 		return -EPERM;
4783 
4784 	r = pm_runtime_resume_and_get(dev->dev);
4785 	if (r < 0)
4786 		return r;
4787 
4788 	if (amdgpu_dpm_debugfs_print_current_performance_level(adev, m)) {
4789 		r = amdgpu_debugfs_pm_info_pp(m, adev);
4790 		if (r)
4791 			goto out;
4792 	}
4793 
4794 	amdgpu_device_ip_get_clockgating_state(adev, &flags);
4795 
4796 	seq_printf(m, "Clock Gating Flags Mask: 0x%llx\n", flags);
4797 	amdgpu_parse_cg_state(m, flags);
4798 	seq_printf(m, "\n");
4799 
4800 out:
4801 	pm_runtime_put_autosuspend(dev->dev);
4802 
4803 	return r;
4804 }
4805 
4806 DEFINE_SHOW_ATTRIBUTE(amdgpu_debugfs_pm_info);
4807 
4808 /*
4809  * amdgpu_pm_priv_buffer_read - Read memory region allocated to FW
4810  *
4811  * Reads debug memory region allocated to PMFW
4812  */
4813 static ssize_t amdgpu_pm_prv_buffer_read(struct file *f, char __user *buf,
4814 					 size_t size, loff_t *pos)
4815 {
4816 	struct amdgpu_device *adev = file_inode(f)->i_private;
4817 	size_t smu_prv_buf_size;
4818 	void *smu_prv_buf;
4819 	int ret = 0;
4820 
4821 	if (amdgpu_in_reset(adev))
4822 		return -EPERM;
4823 	if (adev->in_suspend && !adev->in_runpm)
4824 		return -EPERM;
4825 
4826 	ret = amdgpu_dpm_get_smu_prv_buf_details(adev, &smu_prv_buf, &smu_prv_buf_size);
4827 	if (ret)
4828 		return ret;
4829 
4830 	if (!smu_prv_buf || !smu_prv_buf_size)
4831 		return -EINVAL;
4832 
4833 	return simple_read_from_buffer(buf, size, pos, smu_prv_buf,
4834 				       smu_prv_buf_size);
4835 }
4836 
4837 static const struct file_operations amdgpu_debugfs_pm_prv_buffer_fops = {
4838 	.owner = THIS_MODULE,
4839 	.open = simple_open,
4840 	.read = amdgpu_pm_prv_buffer_read,
4841 	.llseek = default_llseek,
4842 };
4843 
4844 #endif
4845 
4846 void amdgpu_debugfs_pm_init(struct amdgpu_device *adev)
4847 {
4848 #if defined(CONFIG_DEBUG_FS)
4849 	struct drm_minor *minor = adev_to_drm(adev)->primary;
4850 	struct dentry *root = minor->debugfs_root;
4851 
4852 	if (!adev->pm.dpm_enabled)
4853 		return;
4854 
4855 	debugfs_create_file("amdgpu_pm_info", 0444, root, adev,
4856 			    &amdgpu_debugfs_pm_info_fops);
4857 
4858 	if (adev->pm.smu_prv_buffer_size > 0)
4859 		debugfs_create_file_size("amdgpu_pm_prv_buffer", 0444, root,
4860 					 adev,
4861 					 &amdgpu_debugfs_pm_prv_buffer_fops,
4862 					 adev->pm.smu_prv_buffer_size);
4863 
4864 	amdgpu_dpm_stb_debug_fs_init(adev);
4865 #endif
4866 }
4867