xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/df_v3_6.c (revision 817396dc9f6ab2481b94071de2e586aae876e89c)
1 /*
2  * Copyright 2018 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  */
23 #include "amdgpu.h"
24 #include "df_v3_6.h"
25 
26 #include "df/df_3_6_default.h"
27 #include "df/df_3_6_offset.h"
28 #include "df/df_3_6_sh_mask.h"
29 
30 static u32 df_v3_6_channel_number[] = {1, 2, 0, 4, 0, 8, 0,
31 				       16, 32, 0, 0, 0, 2, 4, 8};
32 
33 /* init df format attrs */
34 AMDGPU_PMU_ATTR(event,		"config:0-7");
35 AMDGPU_PMU_ATTR(instance,	"config:8-15");
36 AMDGPU_PMU_ATTR(umask,		"config:16-23");
37 
38 /* df format attributes  */
39 static struct attribute *df_v3_6_format_attrs[] = {
40 	&pmu_attr_event.attr,
41 	&pmu_attr_instance.attr,
42 	&pmu_attr_umask.attr,
43 	NULL
44 };
45 
46 /* df format attribute group */
47 static struct attribute_group df_v3_6_format_attr_group = {
48 	.name = "format",
49 	.attrs = df_v3_6_format_attrs,
50 };
51 
52 /* df event attrs */
53 AMDGPU_PMU_ATTR(cake0_pcsout_txdata,
54 		      "event=0x7,instance=0x46,umask=0x2");
55 AMDGPU_PMU_ATTR(cake1_pcsout_txdata,
56 		      "event=0x7,instance=0x47,umask=0x2");
57 AMDGPU_PMU_ATTR(cake0_pcsout_txmeta,
58 		      "event=0x7,instance=0x46,umask=0x4");
59 AMDGPU_PMU_ATTR(cake1_pcsout_txmeta,
60 		      "event=0x7,instance=0x47,umask=0x4");
61 AMDGPU_PMU_ATTR(cake0_ftiinstat_reqalloc,
62 		      "event=0xb,instance=0x46,umask=0x4");
63 AMDGPU_PMU_ATTR(cake1_ftiinstat_reqalloc,
64 		      "event=0xb,instance=0x47,umask=0x4");
65 AMDGPU_PMU_ATTR(cake0_ftiinstat_rspalloc,
66 		      "event=0xb,instance=0x46,umask=0x8");
67 AMDGPU_PMU_ATTR(cake1_ftiinstat_rspalloc,
68 		      "event=0xb,instance=0x47,umask=0x8");
69 
70 /* df event attributes  */
71 static struct attribute *df_v3_6_event_attrs[] = {
72 	&pmu_attr_cake0_pcsout_txdata.attr,
73 	&pmu_attr_cake1_pcsout_txdata.attr,
74 	&pmu_attr_cake0_pcsout_txmeta.attr,
75 	&pmu_attr_cake1_pcsout_txmeta.attr,
76 	&pmu_attr_cake0_ftiinstat_reqalloc.attr,
77 	&pmu_attr_cake1_ftiinstat_reqalloc.attr,
78 	&pmu_attr_cake0_ftiinstat_rspalloc.attr,
79 	&pmu_attr_cake1_ftiinstat_rspalloc.attr,
80 	NULL
81 };
82 
83 /* df event attribute group */
84 static struct attribute_group df_v3_6_event_attr_group = {
85 	.name = "events",
86 	.attrs = df_v3_6_event_attrs
87 };
88 
89 /* df event attr groups  */
90 const struct attribute_group *df_v3_6_attr_groups[] = {
91 		&df_v3_6_format_attr_group,
92 		&df_v3_6_event_attr_group,
93 		NULL
94 };
95 
96 static uint64_t df_v3_6_get_fica(struct amdgpu_device *adev,
97 				 uint32_t ficaa_val)
98 {
99 	unsigned long flags, address, data;
100 	uint32_t ficadl_val, ficadh_val;
101 
102 	address = adev->nbio.funcs->get_pcie_index_offset(adev);
103 	data = adev->nbio.funcs->get_pcie_data_offset(adev);
104 
105 	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
106 	WREG32(address, smnDF_PIE_AON_FabricIndirectConfigAccessAddress3);
107 	WREG32(data, ficaa_val);
108 
109 	WREG32(address, smnDF_PIE_AON_FabricIndirectConfigAccessDataLo3);
110 	ficadl_val = RREG32(data);
111 
112 	WREG32(address, smnDF_PIE_AON_FabricIndirectConfigAccessDataHi3);
113 	ficadh_val = RREG32(data);
114 
115 	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
116 
117 	return (((ficadh_val & 0xFFFFFFFFFFFFFFFF) << 32) | ficadl_val);
118 }
119 
120 static void df_v3_6_set_fica(struct amdgpu_device *adev, uint32_t ficaa_val,
121 			     uint32_t ficadl_val, uint32_t ficadh_val)
122 {
123 	unsigned long flags, address, data;
124 
125 	address = adev->nbio.funcs->get_pcie_index_offset(adev);
126 	data = adev->nbio.funcs->get_pcie_data_offset(adev);
127 
128 	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
129 	WREG32(address, smnDF_PIE_AON_FabricIndirectConfigAccessAddress3);
130 	WREG32(data, ficaa_val);
131 
132 	WREG32(address, smnDF_PIE_AON_FabricIndirectConfigAccessDataLo3);
133 	WREG32(data, ficadl_val);
134 
135 	WREG32(address, smnDF_PIE_AON_FabricIndirectConfigAccessDataHi3);
136 	WREG32(data, ficadh_val);
137 
138 	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
139 }
140 
141 /*
142  * df_v3_6_perfmon_rreg - read perfmon lo and hi
143  *
144  * required to be atomic.  no mmio method provided so subsequent reads for lo
145  * and hi require to preserve df finite state machine
146  */
147 static void df_v3_6_perfmon_rreg(struct amdgpu_device *adev,
148 			    uint32_t lo_addr, uint32_t *lo_val,
149 			    uint32_t hi_addr, uint32_t *hi_val)
150 {
151 	unsigned long flags, address, data;
152 
153 	address = adev->nbio.funcs->get_pcie_index_offset(adev);
154 	data = adev->nbio.funcs->get_pcie_data_offset(adev);
155 
156 	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
157 	WREG32(address, lo_addr);
158 	*lo_val = RREG32(data);
159 	WREG32(address, hi_addr);
160 	*hi_val = RREG32(data);
161 	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
162 }
163 
164 /*
165  * df_v3_6_perfmon_wreg - write to perfmon lo and hi
166  *
167  * required to be atomic.  no mmio method provided so subsequent reads after
168  * data writes cannot occur to preserve data fabrics finite state machine.
169  */
170 static void df_v3_6_perfmon_wreg(struct amdgpu_device *adev, uint32_t lo_addr,
171 			    uint32_t lo_val, uint32_t hi_addr, uint32_t hi_val)
172 {
173 	unsigned long flags, address, data;
174 
175 	address = adev->nbio.funcs->get_pcie_index_offset(adev);
176 	data = adev->nbio.funcs->get_pcie_data_offset(adev);
177 
178 	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
179 	WREG32(address, lo_addr);
180 	WREG32(data, lo_val);
181 	WREG32(address, hi_addr);
182 	WREG32(data, hi_val);
183 	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
184 }
185 
186 /* same as perfmon_wreg but return status on write value check */
187 static int df_v3_6_perfmon_arm_with_status(struct amdgpu_device *adev,
188 					  uint32_t lo_addr, uint32_t lo_val,
189 					  uint32_t hi_addr, uint32_t  hi_val)
190 {
191 	unsigned long flags, address, data;
192 	uint32_t lo_val_rb, hi_val_rb;
193 
194 	address = adev->nbio.funcs->get_pcie_index_offset(adev);
195 	data = adev->nbio.funcs->get_pcie_data_offset(adev);
196 
197 	spin_lock_irqsave(&adev->pcie_idx_lock, flags);
198 	WREG32(address, lo_addr);
199 	WREG32(data, lo_val);
200 	WREG32(address, hi_addr);
201 	WREG32(data, hi_val);
202 
203 	WREG32(address, lo_addr);
204 	lo_val_rb = RREG32(data);
205 	WREG32(address, hi_addr);
206 	hi_val_rb = RREG32(data);
207 	spin_unlock_irqrestore(&adev->pcie_idx_lock, flags);
208 
209 	if (!(lo_val == lo_val_rb && hi_val == hi_val_rb))
210 		return -EBUSY;
211 
212 	return 0;
213 }
214 
215 
216 /*
217  * retry arming counters every 100 usecs within 1 millisecond interval.
218  * if retry fails after time out, return error.
219  */
220 #define ARM_RETRY_USEC_TIMEOUT	1000
221 #define ARM_RETRY_USEC_INTERVAL	100
222 static int df_v3_6_perfmon_arm_with_retry(struct amdgpu_device *adev,
223 					  uint32_t lo_addr, uint32_t lo_val,
224 					  uint32_t hi_addr, uint32_t  hi_val)
225 {
226 	int countdown = ARM_RETRY_USEC_TIMEOUT;
227 
228 	while (countdown) {
229 
230 		if (!df_v3_6_perfmon_arm_with_status(adev, lo_addr, lo_val,
231 						     hi_addr, hi_val))
232 			break;
233 
234 		countdown -= ARM_RETRY_USEC_INTERVAL;
235 		udelay(ARM_RETRY_USEC_INTERVAL);
236 	}
237 
238 	return countdown > 0 ? 0 : -ETIME;
239 }
240 
241 /* get the number of df counters available */
242 static ssize_t df_v3_6_get_df_cntr_avail(struct device *dev,
243 		struct device_attribute *attr,
244 		char *buf)
245 {
246 	struct amdgpu_device *adev;
247 	struct drm_device *ddev;
248 	int i, count;
249 
250 	ddev = dev_get_drvdata(dev);
251 	adev = ddev->dev_private;
252 	count = 0;
253 
254 	for (i = 0; i < DF_V3_6_MAX_COUNTERS; i++) {
255 		if (adev->df_perfmon_config_assign_mask[i] == 0)
256 			count++;
257 	}
258 
259 	return snprintf(buf, PAGE_SIZE,	"%i\n", count);
260 }
261 
262 /* device attr for available perfmon counters */
263 static DEVICE_ATTR(df_cntr_avail, S_IRUGO, df_v3_6_get_df_cntr_avail, NULL);
264 
265 /* init perfmons */
266 static void df_v3_6_sw_init(struct amdgpu_device *adev)
267 {
268 	int i, ret;
269 
270 	ret = device_create_file(adev->dev, &dev_attr_df_cntr_avail);
271 	if (ret)
272 		DRM_ERROR("failed to create file for available df counters\n");
273 
274 	for (i = 0; i < AMDGPU_MAX_DF_PERFMONS; i++)
275 		adev->df_perfmon_config_assign_mask[i] = 0;
276 }
277 
278 static void df_v3_6_sw_fini(struct amdgpu_device *adev)
279 {
280 
281 	device_remove_file(adev->dev, &dev_attr_df_cntr_avail);
282 
283 }
284 
285 static void df_v3_6_enable_broadcast_mode(struct amdgpu_device *adev,
286 					  bool enable)
287 {
288 	u32 tmp;
289 
290 	if (enable) {
291 		tmp = RREG32_SOC15(DF, 0, mmFabricConfigAccessControl);
292 		tmp &= ~FabricConfigAccessControl__CfgRegInstAccEn_MASK;
293 		WREG32_SOC15(DF, 0, mmFabricConfigAccessControl, tmp);
294 	} else
295 		WREG32_SOC15(DF, 0, mmFabricConfigAccessControl,
296 			     mmFabricConfigAccessControl_DEFAULT);
297 }
298 
299 static u32 df_v3_6_get_fb_channel_number(struct amdgpu_device *adev)
300 {
301 	u32 tmp;
302 
303 	tmp = RREG32_SOC15(DF, 0, mmDF_CS_UMC_AON0_DramBaseAddress0);
304 	tmp &= DF_CS_UMC_AON0_DramBaseAddress0__IntLvNumChan_MASK;
305 	tmp >>= DF_CS_UMC_AON0_DramBaseAddress0__IntLvNumChan__SHIFT;
306 
307 	return tmp;
308 }
309 
310 static u32 df_v3_6_get_hbm_channel_number(struct amdgpu_device *adev)
311 {
312 	int fb_channel_number;
313 
314 	fb_channel_number = adev->df_funcs->get_fb_channel_number(adev);
315 	if (fb_channel_number >= ARRAY_SIZE(df_v3_6_channel_number))
316 		fb_channel_number = 0;
317 
318 	return df_v3_6_channel_number[fb_channel_number];
319 }
320 
321 static void df_v3_6_update_medium_grain_clock_gating(struct amdgpu_device *adev,
322 						     bool enable)
323 {
324 	u32 tmp;
325 
326 	if (adev->cg_flags & AMD_CG_SUPPORT_DF_MGCG) {
327 		/* Put DF on broadcast mode */
328 		adev->df_funcs->enable_broadcast_mode(adev, true);
329 
330 		if (enable) {
331 			tmp = RREG32_SOC15(DF, 0,
332 					mmDF_PIE_AON0_DfGlobalClkGater);
333 			tmp &= ~DF_PIE_AON0_DfGlobalClkGater__MGCGMode_MASK;
334 			tmp |= DF_V3_6_MGCG_ENABLE_15_CYCLE_DELAY;
335 			WREG32_SOC15(DF, 0,
336 					mmDF_PIE_AON0_DfGlobalClkGater, tmp);
337 		} else {
338 			tmp = RREG32_SOC15(DF, 0,
339 					mmDF_PIE_AON0_DfGlobalClkGater);
340 			tmp &= ~DF_PIE_AON0_DfGlobalClkGater__MGCGMode_MASK;
341 			tmp |= DF_V3_6_MGCG_DISABLE;
342 			WREG32_SOC15(DF, 0,
343 					mmDF_PIE_AON0_DfGlobalClkGater, tmp);
344 		}
345 
346 		/* Exit broadcast mode */
347 		adev->df_funcs->enable_broadcast_mode(adev, false);
348 	}
349 }
350 
351 static void df_v3_6_get_clockgating_state(struct amdgpu_device *adev,
352 					  u32 *flags)
353 {
354 	u32 tmp;
355 
356 	/* AMD_CG_SUPPORT_DF_MGCG */
357 	tmp = RREG32_SOC15(DF, 0, mmDF_PIE_AON0_DfGlobalClkGater);
358 	if (tmp & DF_V3_6_MGCG_ENABLE_15_CYCLE_DELAY)
359 		*flags |= AMD_CG_SUPPORT_DF_MGCG;
360 }
361 
362 /* get assigned df perfmon ctr as int */
363 static int df_v3_6_pmc_config_2_cntr(struct amdgpu_device *adev,
364 				      uint64_t config)
365 {
366 	int i;
367 
368 	for (i = 0; i < DF_V3_6_MAX_COUNTERS; i++) {
369 		if ((config & 0x0FFFFFFUL) ==
370 					adev->df_perfmon_config_assign_mask[i])
371 			return i;
372 	}
373 
374 	return -EINVAL;
375 }
376 
377 /* get address based on counter assignment */
378 static void df_v3_6_pmc_get_addr(struct amdgpu_device *adev,
379 				 uint64_t config,
380 				 int is_ctrl,
381 				 uint32_t *lo_base_addr,
382 				 uint32_t *hi_base_addr)
383 {
384 	int target_cntr = df_v3_6_pmc_config_2_cntr(adev, config);
385 
386 	if (target_cntr < 0)
387 		return;
388 
389 	switch (target_cntr) {
390 
391 	case 0:
392 		*lo_base_addr = is_ctrl ? smnPerfMonCtlLo4 : smnPerfMonCtrLo4;
393 		*hi_base_addr = is_ctrl ? smnPerfMonCtlHi4 : smnPerfMonCtrHi4;
394 		break;
395 	case 1:
396 		*lo_base_addr = is_ctrl ? smnPerfMonCtlLo5 : smnPerfMonCtrLo5;
397 		*hi_base_addr = is_ctrl ? smnPerfMonCtlHi5 : smnPerfMonCtrHi5;
398 		break;
399 	case 2:
400 		*lo_base_addr = is_ctrl ? smnPerfMonCtlLo6 : smnPerfMonCtrLo6;
401 		*hi_base_addr = is_ctrl ? smnPerfMonCtlHi6 : smnPerfMonCtrHi6;
402 		break;
403 	case 3:
404 		*lo_base_addr = is_ctrl ? smnPerfMonCtlLo7 : smnPerfMonCtrLo7;
405 		*hi_base_addr = is_ctrl ? smnPerfMonCtlHi7 : smnPerfMonCtrHi7;
406 		break;
407 
408 	}
409 
410 }
411 
412 /* get read counter address */
413 static void df_v3_6_pmc_get_read_settings(struct amdgpu_device *adev,
414 					  uint64_t config,
415 					  uint32_t *lo_base_addr,
416 					  uint32_t *hi_base_addr)
417 {
418 	df_v3_6_pmc_get_addr(adev, config, 0, lo_base_addr, hi_base_addr);
419 }
420 
421 /* get control counter settings i.e. address and values to set */
422 static int df_v3_6_pmc_get_ctrl_settings(struct amdgpu_device *adev,
423 					  uint64_t config,
424 					  uint32_t *lo_base_addr,
425 					  uint32_t *hi_base_addr,
426 					  uint32_t *lo_val,
427 					  uint32_t *hi_val)
428 {
429 
430 	uint32_t eventsel, instance, unitmask;
431 	uint32_t instance_10, instance_5432, instance_76;
432 
433 	df_v3_6_pmc_get_addr(adev, config, 1, lo_base_addr, hi_base_addr);
434 
435 	if ((*lo_base_addr == 0) || (*hi_base_addr == 0)) {
436 		DRM_ERROR("[DF PMC] addressing not retrieved! Lo: %x, Hi: %x",
437 				*lo_base_addr, *hi_base_addr);
438 		return -ENXIO;
439 	}
440 
441 	eventsel = DF_V3_6_GET_EVENT(config) & 0x3f;
442 	unitmask = DF_V3_6_GET_UNITMASK(config) & 0xf;
443 	instance = DF_V3_6_GET_INSTANCE(config);
444 
445 	instance_10 = instance & 0x3;
446 	instance_5432 = (instance >> 2) & 0xf;
447 	instance_76 = (instance >> 6) & 0x3;
448 
449 	*lo_val = (unitmask << 8) | (instance_10 << 6) | eventsel | (1 << 22);
450 	*hi_val = (instance_76 << 29) | instance_5432;
451 
452 	DRM_DEBUG_DRIVER("config=%llx addr=%08x:%08x val=%08x:%08x",
453 		config, *lo_base_addr, *hi_base_addr, *lo_val, *hi_val);
454 
455 	return 0;
456 }
457 
458 /* add df performance counters for read */
459 static int df_v3_6_pmc_add_cntr(struct amdgpu_device *adev,
460 				   uint64_t config)
461 {
462 	int i, target_cntr;
463 
464 	target_cntr = df_v3_6_pmc_config_2_cntr(adev, config);
465 
466 	if (target_cntr >= 0)
467 		return 0;
468 
469 	for (i = 0; i < DF_V3_6_MAX_COUNTERS; i++) {
470 		if (adev->df_perfmon_config_assign_mask[i] == 0U) {
471 			adev->df_perfmon_config_assign_mask[i] =
472 							config & 0x0FFFFFFUL;
473 			return 0;
474 		}
475 	}
476 
477 	return -ENOSPC;
478 }
479 
480 #define DEFERRED_ARM_MASK	(1 << 31)
481 static int df_v3_6_pmc_set_deferred(struct amdgpu_device *adev,
482 				    uint64_t config, bool is_deferred)
483 {
484 	int target_cntr;
485 
486 	target_cntr = df_v3_6_pmc_config_2_cntr(adev, config);
487 
488 	if (target_cntr < 0)
489 		return -EINVAL;
490 
491 	if (is_deferred)
492 		adev->df_perfmon_config_assign_mask[target_cntr] |=
493 							DEFERRED_ARM_MASK;
494 	else
495 		adev->df_perfmon_config_assign_mask[target_cntr] &=
496 							~DEFERRED_ARM_MASK;
497 
498 	return 0;
499 }
500 
501 static bool df_v3_6_pmc_is_deferred(struct amdgpu_device *adev,
502 				    uint64_t config)
503 {
504 	int target_cntr;
505 
506 	target_cntr = df_v3_6_pmc_config_2_cntr(adev, config);
507 
508 	/*
509 	 * we never get target_cntr < 0 since this funciton is only called in
510 	 * pmc_count for now but we should check anyways.
511 	 */
512 	return (target_cntr >= 0 &&
513 			(adev->df_perfmon_config_assign_mask[target_cntr]
514 			& DEFERRED_ARM_MASK));
515 
516 }
517 
518 /* release performance counter */
519 static void df_v3_6_pmc_release_cntr(struct amdgpu_device *adev,
520 				     uint64_t config)
521 {
522 	int target_cntr = df_v3_6_pmc_config_2_cntr(adev, config);
523 
524 	if (target_cntr >= 0)
525 		adev->df_perfmon_config_assign_mask[target_cntr] = 0ULL;
526 }
527 
528 
529 static void df_v3_6_reset_perfmon_cntr(struct amdgpu_device *adev,
530 					 uint64_t config)
531 {
532 	uint32_t lo_base_addr, hi_base_addr;
533 
534 	df_v3_6_pmc_get_read_settings(adev, config, &lo_base_addr,
535 				      &hi_base_addr);
536 
537 	if ((lo_base_addr == 0) || (hi_base_addr == 0))
538 		return;
539 
540 	df_v3_6_perfmon_wreg(adev, lo_base_addr, 0, hi_base_addr, 0);
541 }
542 
543 static int df_v3_6_pmc_start(struct amdgpu_device *adev, uint64_t config,
544 			     int is_enable)
545 {
546 	uint32_t lo_base_addr, hi_base_addr, lo_val, hi_val;
547 	int err = 0, ret = 0;
548 
549 	switch (adev->asic_type) {
550 	case CHIP_VEGA20:
551 		if (is_enable)
552 			return df_v3_6_pmc_add_cntr(adev, config);
553 
554 		df_v3_6_reset_perfmon_cntr(adev, config);
555 
556 		ret = df_v3_6_pmc_get_ctrl_settings(adev,
557 					config,
558 					&lo_base_addr,
559 					&hi_base_addr,
560 					&lo_val,
561 					&hi_val);
562 
563 		if (ret)
564 			return ret;
565 
566 		err = df_v3_6_perfmon_arm_with_retry(adev,
567 						     lo_base_addr,
568 						     lo_val,
569 						     hi_base_addr,
570 						     hi_val);
571 
572 		if (err)
573 			ret = df_v3_6_pmc_set_deferred(adev, config, true);
574 
575 		break;
576 	default:
577 		break;
578 	}
579 
580 	return ret;
581 }
582 
583 static int df_v3_6_pmc_stop(struct amdgpu_device *adev, uint64_t config,
584 			    int is_disable)
585 {
586 	uint32_t lo_base_addr, hi_base_addr, lo_val, hi_val;
587 	int ret = 0;
588 
589 	switch (adev->asic_type) {
590 	case CHIP_VEGA20:
591 		ret = df_v3_6_pmc_get_ctrl_settings(adev,
592 			config,
593 			&lo_base_addr,
594 			&hi_base_addr,
595 			&lo_val,
596 			&hi_val);
597 
598 		if (ret)
599 			return ret;
600 
601 		df_v3_6_reset_perfmon_cntr(adev, config);
602 
603 		if (is_disable)
604 			df_v3_6_pmc_release_cntr(adev, config);
605 
606 		break;
607 	default:
608 		break;
609 	}
610 
611 	return ret;
612 }
613 
614 static void df_v3_6_pmc_get_count(struct amdgpu_device *adev,
615 				  uint64_t config,
616 				  uint64_t *count)
617 {
618 	uint32_t lo_base_addr, hi_base_addr, lo_val = 0, hi_val = 0;
619 	*count = 0;
620 
621 	switch (adev->asic_type) {
622 	case CHIP_VEGA20:
623 		df_v3_6_pmc_get_read_settings(adev, config, &lo_base_addr,
624 				      &hi_base_addr);
625 
626 		if ((lo_base_addr == 0) || (hi_base_addr == 0))
627 			return;
628 
629 		/* rearm the counter or throw away count value on failure */
630 		if (df_v3_6_pmc_is_deferred(adev, config)) {
631 			int rearm_err = df_v3_6_perfmon_arm_with_status(adev,
632 							lo_base_addr, lo_val,
633 							hi_base_addr, hi_val);
634 
635 			if (rearm_err)
636 				return;
637 
638 			df_v3_6_pmc_set_deferred(adev, config, false);
639 		}
640 
641 		df_v3_6_perfmon_rreg(adev, lo_base_addr, &lo_val,
642 				hi_base_addr, &hi_val);
643 
644 		*count  = ((hi_val | 0ULL) << 32) | (lo_val | 0ULL);
645 
646 		if (*count >= DF_V3_6_PERFMON_OVERFLOW)
647 			*count = 0;
648 
649 		DRM_DEBUG_DRIVER("config=%llx addr=%08x:%08x val=%08x:%08x",
650 			 config, lo_base_addr, hi_base_addr, lo_val, hi_val);
651 
652 		break;
653 	default:
654 		break;
655 	}
656 }
657 
658 const struct amdgpu_df_funcs df_v3_6_funcs = {
659 	.sw_init = df_v3_6_sw_init,
660 	.sw_fini = df_v3_6_sw_fini,
661 	.enable_broadcast_mode = df_v3_6_enable_broadcast_mode,
662 	.get_fb_channel_number = df_v3_6_get_fb_channel_number,
663 	.get_hbm_channel_number = df_v3_6_get_hbm_channel_number,
664 	.update_medium_grain_clock_gating =
665 			df_v3_6_update_medium_grain_clock_gating,
666 	.get_clockgating_state = df_v3_6_get_clockgating_state,
667 	.pmc_start = df_v3_6_pmc_start,
668 	.pmc_stop = df_v3_6_pmc_stop,
669 	.pmc_get_count = df_v3_6_pmc_get_count,
670 	.get_fica = df_v3_6_get_fica,
671 	.set_fica = df_v3_6_set_fica
672 };
673