xref: /linux/drivers/gpu/drm/amd/amdgpu/aqua_vanjaram.c (revision 4d614ce8ffd757e4c7944bf9b5598b4a250a8a61)
1 /*
2  * Copyright 2022 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 "soc15.h"
25 
26 #include "soc15_common.h"
27 #include "amdgpu_reg_state.h"
28 #include "amdgpu_xcp.h"
29 #include "gfx_v9_4_3.h"
30 #include "gfxhub_v1_2.h"
31 #include "sdma_v4_4_2.h"
32 
33 #define XCP_INST_MASK(num_inst, xcp_id)                                        \
34 	(num_inst ? GENMASK(num_inst - 1, 0) << (xcp_id * num_inst) : 0)
35 
36 #define AMDGPU_XCP_OPS_KFD	(1 << 0)
37 
38 void aqua_vanjaram_doorbell_index_init(struct amdgpu_device *adev)
39 {
40 	int i;
41 
42 	adev->doorbell_index.kiq = AMDGPU_DOORBELL_LAYOUT1_KIQ_START;
43 
44 	adev->doorbell_index.mec_ring0 = AMDGPU_DOORBELL_LAYOUT1_MEC_RING_START;
45 
46 	adev->doorbell_index.userqueue_start = AMDGPU_DOORBELL_LAYOUT1_USERQUEUE_START;
47 	adev->doorbell_index.userqueue_end = AMDGPU_DOORBELL_LAYOUT1_USERQUEUE_END;
48 	adev->doorbell_index.xcc_doorbell_range = AMDGPU_DOORBELL_LAYOUT1_XCC_RANGE;
49 
50 	adev->doorbell_index.sdma_doorbell_range = 20;
51 	for (i = 0; i < adev->sdma.num_instances; i++)
52 		adev->doorbell_index.sdma_engine[i] =
53 			AMDGPU_DOORBELL_LAYOUT1_sDMA_ENGINE_START +
54 			i * (adev->doorbell_index.sdma_doorbell_range >> 1);
55 
56 	adev->doorbell_index.ih = AMDGPU_DOORBELL_LAYOUT1_IH;
57 	adev->doorbell_index.vcn.vcn_ring0_1 = AMDGPU_DOORBELL_LAYOUT1_VCN_START;
58 
59 	adev->doorbell_index.first_non_cp = AMDGPU_DOORBELL_LAYOUT1_FIRST_NON_CP;
60 	adev->doorbell_index.last_non_cp = AMDGPU_DOORBELL_LAYOUT1_LAST_NON_CP;
61 
62 	adev->doorbell_index.max_assignment = AMDGPU_DOORBELL_LAYOUT1_MAX_ASSIGNMENT << 1;
63 }
64 
65 static bool aqua_vanjaram_xcp_vcn_shared(struct amdgpu_device *adev)
66 {
67 	return (adev->xcp_mgr->num_xcps > adev->vcn.num_vcn_inst);
68 }
69 
70 static void aqua_vanjaram_set_xcp_id(struct amdgpu_device *adev,
71 			     uint32_t inst_idx, struct amdgpu_ring *ring)
72 {
73 	int xcp_id;
74 	enum AMDGPU_XCP_IP_BLOCK ip_blk;
75 	uint32_t inst_mask;
76 
77 	ring->xcp_id = AMDGPU_XCP_NO_PARTITION;
78 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE)
79 		adev->gfx.enforce_isolation[0].xcp_id = ring->xcp_id;
80 	if ((adev->xcp_mgr->mode == AMDGPU_XCP_MODE_NONE) ||
81 	    (ring->funcs->type == AMDGPU_RING_TYPE_CPER))
82 		return;
83 
84 	inst_mask = 1 << inst_idx;
85 
86 	switch (ring->funcs->type) {
87 	case AMDGPU_HW_IP_GFX:
88 	case AMDGPU_RING_TYPE_COMPUTE:
89 	case AMDGPU_RING_TYPE_KIQ:
90 		ip_blk = AMDGPU_XCP_GFX;
91 		break;
92 	case AMDGPU_RING_TYPE_SDMA:
93 		ip_blk = AMDGPU_XCP_SDMA;
94 		break;
95 	case AMDGPU_RING_TYPE_VCN_ENC:
96 	case AMDGPU_RING_TYPE_VCN_JPEG:
97 		ip_blk = AMDGPU_XCP_VCN;
98 		break;
99 	default:
100 		DRM_ERROR("Not support ring type %d!", ring->funcs->type);
101 		return;
102 	}
103 
104 	for (xcp_id = 0; xcp_id < adev->xcp_mgr->num_xcps; xcp_id++) {
105 		if (adev->xcp_mgr->xcp[xcp_id].ip[ip_blk].inst_mask & inst_mask) {
106 			ring->xcp_id = xcp_id;
107 			dev_dbg(adev->dev, "ring:%s xcp_id :%u", ring->name,
108 				ring->xcp_id);
109 			if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE)
110 				adev->gfx.enforce_isolation[xcp_id].xcp_id = xcp_id;
111 			break;
112 		}
113 	}
114 }
115 
116 static void aqua_vanjaram_xcp_gpu_sched_update(
117 		struct amdgpu_device *adev,
118 		struct amdgpu_ring *ring,
119 		unsigned int sel_xcp_id)
120 {
121 	unsigned int *num_gpu_sched;
122 
123 	num_gpu_sched = &adev->xcp_mgr->xcp[sel_xcp_id]
124 			.gpu_sched[ring->funcs->type][ring->hw_prio].num_scheds;
125 	adev->xcp_mgr->xcp[sel_xcp_id].gpu_sched[ring->funcs->type][ring->hw_prio]
126 			.sched[(*num_gpu_sched)++] = &ring->sched;
127 	DRM_DEBUG("%s :[%d] gpu_sched[%d][%d] = %d", ring->name,
128 			sel_xcp_id, ring->funcs->type,
129 			ring->hw_prio, *num_gpu_sched);
130 }
131 
132 static int aqua_vanjaram_xcp_sched_list_update(
133 		struct amdgpu_device *adev)
134 {
135 	struct amdgpu_ring *ring;
136 	int i;
137 
138 	for (i = 0; i < MAX_XCP; i++) {
139 		atomic_set(&adev->xcp_mgr->xcp[i].ref_cnt, 0);
140 		memset(adev->xcp_mgr->xcp[i].gpu_sched, 0, sizeof(adev->xcp_mgr->xcp->gpu_sched));
141 	}
142 
143 	if (adev->xcp_mgr->mode == AMDGPU_XCP_MODE_NONE)
144 		return 0;
145 
146 	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
147 		ring = adev->rings[i];
148 		if (!ring || !ring->sched.ready || ring->no_scheduler)
149 			continue;
150 
151 		aqua_vanjaram_xcp_gpu_sched_update(adev, ring, ring->xcp_id);
152 
153 		/* VCN may be shared by two partitions under CPX MODE in certain
154 		 * configs.
155 		 */
156 		if ((ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC ||
157 		     ring->funcs->type == AMDGPU_RING_TYPE_VCN_JPEG) &&
158 		    aqua_vanjaram_xcp_vcn_shared(adev))
159 			aqua_vanjaram_xcp_gpu_sched_update(adev, ring, ring->xcp_id + 1);
160 	}
161 
162 	return 0;
163 }
164 
165 static int aqua_vanjaram_update_partition_sched_list(struct amdgpu_device *adev)
166 {
167 	int i;
168 
169 	for (i = 0; i < adev->num_rings; i++) {
170 		struct amdgpu_ring *ring = adev->rings[i];
171 
172 		if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ||
173 			ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
174 			aqua_vanjaram_set_xcp_id(adev, ring->xcc_id, ring);
175 		else
176 			aqua_vanjaram_set_xcp_id(adev, ring->me, ring);
177 	}
178 
179 	return aqua_vanjaram_xcp_sched_list_update(adev);
180 }
181 
182 static int aqua_vanjaram_select_scheds(
183 		struct amdgpu_device *adev,
184 		u32 hw_ip,
185 		u32 hw_prio,
186 		struct amdgpu_fpriv *fpriv,
187 		unsigned int *num_scheds,
188 		struct drm_gpu_scheduler ***scheds)
189 {
190 	u32 sel_xcp_id;
191 	int i;
192 
193 	if (fpriv->xcp_id == AMDGPU_XCP_NO_PARTITION) {
194 		u32 least_ref_cnt = ~0;
195 
196 		fpriv->xcp_id = 0;
197 		for (i = 0; i < adev->xcp_mgr->num_xcps; i++) {
198 			u32 total_ref_cnt;
199 
200 			total_ref_cnt = atomic_read(&adev->xcp_mgr->xcp[i].ref_cnt);
201 			if (total_ref_cnt < least_ref_cnt) {
202 				fpriv->xcp_id = i;
203 				least_ref_cnt = total_ref_cnt;
204 			}
205 		}
206 	}
207 	sel_xcp_id = fpriv->xcp_id;
208 
209 	if (adev->xcp_mgr->xcp[sel_xcp_id].gpu_sched[hw_ip][hw_prio].num_scheds) {
210 		*num_scheds = adev->xcp_mgr->xcp[fpriv->xcp_id].gpu_sched[hw_ip][hw_prio].num_scheds;
211 		*scheds = adev->xcp_mgr->xcp[fpriv->xcp_id].gpu_sched[hw_ip][hw_prio].sched;
212 		atomic_inc(&adev->xcp_mgr->xcp[sel_xcp_id].ref_cnt);
213 		DRM_DEBUG("Selected partition #%d", sel_xcp_id);
214 	} else {
215 		DRM_ERROR("Failed to schedule partition #%d.", sel_xcp_id);
216 		return -ENOENT;
217 	}
218 
219 	return 0;
220 }
221 
222 static int8_t aqua_vanjaram_logical_to_dev_inst(struct amdgpu_device *adev,
223 					 enum amd_hw_ip_block_type block,
224 					 int8_t inst)
225 {
226 	int8_t dev_inst;
227 
228 	switch (block) {
229 	case GC_HWIP:
230 	case SDMA0_HWIP:
231 	/* Both JPEG and VCN as JPEG is only alias of VCN */
232 	case VCN_HWIP:
233 		dev_inst = adev->ip_map.dev_inst[block][inst];
234 		break;
235 	default:
236 		/* For rest of the IPs, no look up required.
237 		 * Assume 'logical instance == physical instance' for all configs. */
238 		dev_inst = inst;
239 		break;
240 	}
241 
242 	return dev_inst;
243 }
244 
245 static uint32_t aqua_vanjaram_logical_to_dev_mask(struct amdgpu_device *adev,
246 					 enum amd_hw_ip_block_type block,
247 					 uint32_t mask)
248 {
249 	uint32_t dev_mask = 0;
250 	int8_t log_inst, dev_inst;
251 
252 	while (mask) {
253 		log_inst = ffs(mask) - 1;
254 		dev_inst = aqua_vanjaram_logical_to_dev_inst(adev, block, log_inst);
255 		dev_mask |= (1 << dev_inst);
256 		mask &= ~(1 << log_inst);
257 	}
258 
259 	return dev_mask;
260 }
261 
262 static void aqua_vanjaram_populate_ip_map(struct amdgpu_device *adev,
263 					  enum amd_hw_ip_block_type ip_block,
264 					  uint32_t inst_mask)
265 {
266 	int l = 0, i;
267 
268 	while (inst_mask) {
269 		i = ffs(inst_mask) - 1;
270 		adev->ip_map.dev_inst[ip_block][l++] = i;
271 		inst_mask &= ~(1 << i);
272 	}
273 	for (; l < HWIP_MAX_INSTANCE; l++)
274 		adev->ip_map.dev_inst[ip_block][l] = -1;
275 }
276 
277 void aqua_vanjaram_ip_map_init(struct amdgpu_device *adev)
278 {
279 	u32 ip_map[][2] = {
280 		{ GC_HWIP, adev->gfx.xcc_mask },
281 		{ SDMA0_HWIP, adev->sdma.sdma_mask },
282 		{ VCN_HWIP, adev->vcn.inst_mask },
283 	};
284 	int i;
285 
286 	for (i = 0; i < ARRAY_SIZE(ip_map); ++i)
287 		aqua_vanjaram_populate_ip_map(adev, ip_map[i][0], ip_map[i][1]);
288 
289 	adev->ip_map.logical_to_dev_inst = aqua_vanjaram_logical_to_dev_inst;
290 	adev->ip_map.logical_to_dev_mask = aqua_vanjaram_logical_to_dev_mask;
291 }
292 
293 /* Fixed pattern for smn addressing on different AIDs:
294  *   bit[34]: indicate cross AID access
295  *   bit[33:32]: indicate target AID id
296  * AID id range is 0 ~ 3 as maximum AID number is 4.
297  */
298 u64 aqua_vanjaram_encode_ext_smn_addressing(int ext_id)
299 {
300 	u64 ext_offset;
301 
302 	/* local routing and bit[34:32] will be zeros */
303 	if (ext_id == 0)
304 		return 0;
305 
306 	/* Initiated from host, accessing to all non-zero aids are cross traffic */
307 	ext_offset = ((u64)(ext_id & 0x3) << 32) | (1ULL << 34);
308 
309 	return ext_offset;
310 }
311 
312 static enum amdgpu_gfx_partition
313 __aqua_vanjaram_calc_xcp_mode(struct amdgpu_xcp_mgr *xcp_mgr)
314 {
315 	struct amdgpu_device *adev = xcp_mgr->adev;
316 	int num_xcc, num_xcc_per_xcp = 0, mode = 0;
317 
318 	num_xcc = NUM_XCC(xcp_mgr->adev->gfx.xcc_mask);
319 	if (adev->gfx.funcs->get_xccs_per_xcp)
320 		num_xcc_per_xcp = adev->gfx.funcs->get_xccs_per_xcp(adev);
321 	if ((num_xcc_per_xcp) && (num_xcc % num_xcc_per_xcp == 0))
322 		mode = num_xcc / num_xcc_per_xcp;
323 
324 	if (num_xcc_per_xcp == 1)
325 		return AMDGPU_CPX_PARTITION_MODE;
326 
327 	switch (mode) {
328 	case 1:
329 		return AMDGPU_SPX_PARTITION_MODE;
330 	case 2:
331 		return AMDGPU_DPX_PARTITION_MODE;
332 	case 3:
333 		return AMDGPU_TPX_PARTITION_MODE;
334 	case 4:
335 		return AMDGPU_QPX_PARTITION_MODE;
336 	default:
337 		return AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE;
338 	}
339 
340 	return AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE;
341 }
342 
343 static int aqua_vanjaram_query_partition_mode(struct amdgpu_xcp_mgr *xcp_mgr)
344 {
345 	enum amdgpu_gfx_partition derv_mode,
346 		mode = AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE;
347 	struct amdgpu_device *adev = xcp_mgr->adev;
348 
349 	derv_mode = __aqua_vanjaram_calc_xcp_mode(xcp_mgr);
350 
351 	if (amdgpu_sriov_vf(adev))
352 		return derv_mode;
353 
354 	if (adev->nbio.funcs->get_compute_partition_mode) {
355 		mode = adev->nbio.funcs->get_compute_partition_mode(adev);
356 		if (mode != derv_mode)
357 			dev_warn(
358 				adev->dev,
359 				"Mismatch in compute partition mode - reported : %d derived : %d",
360 				mode, derv_mode);
361 	}
362 
363 	return mode;
364 }
365 
366 static int __aqua_vanjaram_get_xcc_per_xcp(struct amdgpu_xcp_mgr *xcp_mgr, int mode)
367 {
368 	int num_xcc, num_xcc_per_xcp = 0;
369 
370 	num_xcc = NUM_XCC(xcp_mgr->adev->gfx.xcc_mask);
371 
372 	switch (mode) {
373 	case AMDGPU_SPX_PARTITION_MODE:
374 		num_xcc_per_xcp = num_xcc;
375 		break;
376 	case AMDGPU_DPX_PARTITION_MODE:
377 		num_xcc_per_xcp = num_xcc / 2;
378 		break;
379 	case AMDGPU_TPX_PARTITION_MODE:
380 		num_xcc_per_xcp = num_xcc / 3;
381 		break;
382 	case AMDGPU_QPX_PARTITION_MODE:
383 		num_xcc_per_xcp = num_xcc / 4;
384 		break;
385 	case AMDGPU_CPX_PARTITION_MODE:
386 		num_xcc_per_xcp = 1;
387 		break;
388 	}
389 
390 	return num_xcc_per_xcp;
391 }
392 
393 static int __aqua_vanjaram_get_xcp_ip_info(struct amdgpu_xcp_mgr *xcp_mgr, int xcp_id,
394 				    enum AMDGPU_XCP_IP_BLOCK ip_id,
395 				    struct amdgpu_xcp_ip *ip)
396 {
397 	struct amdgpu_device *adev = xcp_mgr->adev;
398 	int num_sdma, num_vcn, num_shared_vcn, num_xcp;
399 	int num_xcc_xcp, num_sdma_xcp, num_vcn_xcp;
400 
401 	num_sdma = adev->sdma.num_instances;
402 	num_vcn = adev->vcn.num_vcn_inst;
403 	num_shared_vcn = 1;
404 
405 	num_xcc_xcp = adev->gfx.num_xcc_per_xcp;
406 	num_xcp = NUM_XCC(adev->gfx.xcc_mask) / num_xcc_xcp;
407 
408 	switch (xcp_mgr->mode) {
409 	case AMDGPU_SPX_PARTITION_MODE:
410 	case AMDGPU_DPX_PARTITION_MODE:
411 	case AMDGPU_TPX_PARTITION_MODE:
412 	case AMDGPU_QPX_PARTITION_MODE:
413 	case AMDGPU_CPX_PARTITION_MODE:
414 		num_sdma_xcp = DIV_ROUND_UP(num_sdma, num_xcp);
415 		num_vcn_xcp = DIV_ROUND_UP(num_vcn, num_xcp);
416 		break;
417 	default:
418 		return -EINVAL;
419 	}
420 
421 	if (num_vcn && num_xcp > num_vcn)
422 		num_shared_vcn = num_xcp / num_vcn;
423 
424 	switch (ip_id) {
425 	case AMDGPU_XCP_GFXHUB:
426 		ip->inst_mask = XCP_INST_MASK(num_xcc_xcp, xcp_id);
427 		ip->ip_funcs = &gfxhub_v1_2_xcp_funcs;
428 		break;
429 	case AMDGPU_XCP_GFX:
430 		ip->inst_mask = XCP_INST_MASK(num_xcc_xcp, xcp_id);
431 		ip->ip_funcs = &gfx_v9_4_3_xcp_funcs;
432 		break;
433 	case AMDGPU_XCP_SDMA:
434 		ip->inst_mask = XCP_INST_MASK(num_sdma_xcp, xcp_id);
435 		ip->ip_funcs = &sdma_v4_4_2_xcp_funcs;
436 		break;
437 	case AMDGPU_XCP_VCN:
438 		ip->inst_mask =
439 			XCP_INST_MASK(num_vcn_xcp, xcp_id / num_shared_vcn);
440 		/* TODO : Assign IP funcs */
441 		break;
442 	default:
443 		return -EINVAL;
444 	}
445 
446 	ip->ip_id = ip_id;
447 
448 	return 0;
449 }
450 
451 static int aqua_vanjaram_get_xcp_res_info(struct amdgpu_xcp_mgr *xcp_mgr,
452 					  int mode,
453 					  struct amdgpu_xcp_cfg *xcp_cfg)
454 {
455 	struct amdgpu_device *adev = xcp_mgr->adev;
456 	int max_res[AMDGPU_XCP_RES_MAX] = {};
457 	bool res_lt_xcp;
458 	int num_xcp, i;
459 	u16 nps_modes;
460 
461 	if (!(xcp_mgr->supp_xcp_modes & BIT(mode)))
462 		return -EINVAL;
463 
464 	max_res[AMDGPU_XCP_RES_XCC] = NUM_XCC(adev->gfx.xcc_mask);
465 	max_res[AMDGPU_XCP_RES_DMA] = adev->sdma.num_instances;
466 	max_res[AMDGPU_XCP_RES_DEC] = adev->vcn.num_vcn_inst;
467 	max_res[AMDGPU_XCP_RES_JPEG] = adev->jpeg.num_jpeg_inst;
468 
469 	switch (mode) {
470 	case AMDGPU_SPX_PARTITION_MODE:
471 		num_xcp = 1;
472 		nps_modes = BIT(AMDGPU_NPS1_PARTITION_MODE);
473 		break;
474 	case AMDGPU_DPX_PARTITION_MODE:
475 		num_xcp = 2;
476 		nps_modes = BIT(AMDGPU_NPS1_PARTITION_MODE);
477 		break;
478 	case AMDGPU_TPX_PARTITION_MODE:
479 		num_xcp = 3;
480 		nps_modes = BIT(AMDGPU_NPS1_PARTITION_MODE) |
481 			    BIT(AMDGPU_NPS4_PARTITION_MODE);
482 		break;
483 	case AMDGPU_QPX_PARTITION_MODE:
484 		num_xcp = 4;
485 		nps_modes = BIT(AMDGPU_NPS1_PARTITION_MODE) |
486 			    BIT(AMDGPU_NPS4_PARTITION_MODE);
487 		break;
488 	case AMDGPU_CPX_PARTITION_MODE:
489 		num_xcp = NUM_XCC(adev->gfx.xcc_mask);
490 		nps_modes = BIT(AMDGPU_NPS1_PARTITION_MODE) |
491 			    BIT(AMDGPU_NPS4_PARTITION_MODE);
492 		break;
493 	default:
494 		return -EINVAL;
495 	}
496 
497 	xcp_cfg->compatible_nps_modes =
498 		(adev->gmc.supported_nps_modes & nps_modes);
499 	xcp_cfg->num_res = ARRAY_SIZE(max_res);
500 
501 	for (i = 0; i < xcp_cfg->num_res; i++) {
502 		res_lt_xcp = max_res[i] < num_xcp;
503 		xcp_cfg->xcp_res[i].id = i;
504 		xcp_cfg->xcp_res[i].num_inst =
505 			res_lt_xcp ? 1 : max_res[i] / num_xcp;
506 		xcp_cfg->xcp_res[i].num_inst =
507 			i == AMDGPU_XCP_RES_JPEG ?
508 			xcp_cfg->xcp_res[i].num_inst *
509 			adev->jpeg.num_jpeg_rings : xcp_cfg->xcp_res[i].num_inst;
510 		xcp_cfg->xcp_res[i].num_shared =
511 			res_lt_xcp ? num_xcp / max_res[i] : 1;
512 	}
513 
514 	return 0;
515 }
516 
517 static enum amdgpu_gfx_partition
518 __aqua_vanjaram_get_auto_mode(struct amdgpu_xcp_mgr *xcp_mgr)
519 {
520 	struct amdgpu_device *adev = xcp_mgr->adev;
521 	int num_xcc;
522 
523 	num_xcc = NUM_XCC(xcp_mgr->adev->gfx.xcc_mask);
524 
525 	if (adev->gmc.num_mem_partitions == 1)
526 		return AMDGPU_SPX_PARTITION_MODE;
527 
528 	if (adev->gmc.num_mem_partitions == num_xcc)
529 		return AMDGPU_CPX_PARTITION_MODE;
530 
531 	if (adev->gmc.num_mem_partitions == num_xcc / 2)
532 		return (adev->flags & AMD_IS_APU) ? AMDGPU_TPX_PARTITION_MODE :
533 						    AMDGPU_CPX_PARTITION_MODE;
534 
535 	if (adev->gmc.num_mem_partitions == 2 && !(adev->flags & AMD_IS_APU))
536 		return AMDGPU_DPX_PARTITION_MODE;
537 
538 	return AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE;
539 }
540 
541 static bool __aqua_vanjaram_is_valid_mode(struct amdgpu_xcp_mgr *xcp_mgr,
542 					  enum amdgpu_gfx_partition mode)
543 {
544 	struct amdgpu_device *adev = xcp_mgr->adev;
545 	int num_xcc, num_xccs_per_xcp;
546 
547 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
548 	switch (mode) {
549 	case AMDGPU_SPX_PARTITION_MODE:
550 		return adev->gmc.num_mem_partitions == 1 && num_xcc > 0;
551 	case AMDGPU_DPX_PARTITION_MODE:
552 		return adev->gmc.num_mem_partitions <= 2 && (num_xcc % 4) == 0;
553 	case AMDGPU_TPX_PARTITION_MODE:
554 		return (adev->gmc.num_mem_partitions == 1 ||
555 			adev->gmc.num_mem_partitions == 3) &&
556 		       ((num_xcc % 3) == 0);
557 	case AMDGPU_QPX_PARTITION_MODE:
558 		num_xccs_per_xcp = num_xcc / 4;
559 		return (adev->gmc.num_mem_partitions == 1 ||
560 			adev->gmc.num_mem_partitions == 4) &&
561 		       (num_xccs_per_xcp >= 2);
562 	case AMDGPU_CPX_PARTITION_MODE:
563 		/* (num_xcc > 1) because 1 XCC is considered SPX, not CPX.
564 		 * (num_xcc % adev->gmc.num_mem_partitions) == 0 because
565 		 * num_compute_partitions can't be less than num_mem_partitions
566 		 */
567 		return ((num_xcc > 1) &&
568 		       (num_xcc % adev->gmc.num_mem_partitions) == 0);
569 	default:
570 		return false;
571 	}
572 
573 	return false;
574 }
575 
576 static int __aqua_vanjaram_pre_partition_switch(struct amdgpu_xcp_mgr *xcp_mgr, u32 flags)
577 {
578 	/* TODO:
579 	 * Stop user queues and threads, and make sure GPU is empty of work.
580 	 */
581 
582 	if (flags & AMDGPU_XCP_OPS_KFD)
583 		amdgpu_amdkfd_device_fini_sw(xcp_mgr->adev);
584 
585 	return 0;
586 }
587 
588 static int __aqua_vanjaram_post_partition_switch(struct amdgpu_xcp_mgr *xcp_mgr, u32 flags)
589 {
590 	int ret = 0;
591 
592 	if (flags & AMDGPU_XCP_OPS_KFD) {
593 		amdgpu_amdkfd_device_probe(xcp_mgr->adev);
594 		amdgpu_amdkfd_device_init(xcp_mgr->adev);
595 		/* If KFD init failed, return failure */
596 		if (!xcp_mgr->adev->kfd.init_complete)
597 			ret = -EIO;
598 	}
599 
600 	return ret;
601 }
602 
603 static void
604 __aqua_vanjaram_update_supported_modes(struct amdgpu_xcp_mgr *xcp_mgr)
605 {
606 	struct amdgpu_device *adev = xcp_mgr->adev;
607 
608 	xcp_mgr->supp_xcp_modes = 0;
609 
610 	switch (NUM_XCC(adev->gfx.xcc_mask)) {
611 	case 8:
612 		xcp_mgr->supp_xcp_modes = BIT(AMDGPU_SPX_PARTITION_MODE) |
613 					  BIT(AMDGPU_DPX_PARTITION_MODE) |
614 					  BIT(AMDGPU_QPX_PARTITION_MODE) |
615 					  BIT(AMDGPU_CPX_PARTITION_MODE);
616 		break;
617 	case 6:
618 		xcp_mgr->supp_xcp_modes = BIT(AMDGPU_SPX_PARTITION_MODE) |
619 					  BIT(AMDGPU_TPX_PARTITION_MODE) |
620 					  BIT(AMDGPU_CPX_PARTITION_MODE);
621 		break;
622 	case 4:
623 		xcp_mgr->supp_xcp_modes = BIT(AMDGPU_SPX_PARTITION_MODE) |
624 					  BIT(AMDGPU_DPX_PARTITION_MODE) |
625 					  BIT(AMDGPU_CPX_PARTITION_MODE);
626 		break;
627 	/* this seems only existing in emulation phase */
628 	case 2:
629 		xcp_mgr->supp_xcp_modes = BIT(AMDGPU_SPX_PARTITION_MODE) |
630 					  BIT(AMDGPU_CPX_PARTITION_MODE);
631 		break;
632 	case 1:
633 		xcp_mgr->supp_xcp_modes = BIT(AMDGPU_SPX_PARTITION_MODE) |
634 					  BIT(AMDGPU_CPX_PARTITION_MODE);
635 		break;
636 
637 	default:
638 		break;
639 	}
640 }
641 
642 static void __aqua_vanjaram_update_available_partition_mode(struct amdgpu_xcp_mgr *xcp_mgr)
643 {
644 	int mode;
645 
646 	xcp_mgr->avail_xcp_modes = 0;
647 
648 	for_each_inst(mode, xcp_mgr->supp_xcp_modes) {
649 		if (__aqua_vanjaram_is_valid_mode(xcp_mgr, mode))
650 			xcp_mgr->avail_xcp_modes |= BIT(mode);
651 	}
652 }
653 
654 static int aqua_vanjaram_switch_partition_mode(struct amdgpu_xcp_mgr *xcp_mgr,
655 					       int mode, int *num_xcps)
656 {
657 	int num_xcc_per_xcp, num_xcc, ret;
658 	struct amdgpu_device *adev;
659 	u32 flags = 0;
660 
661 	adev = xcp_mgr->adev;
662 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
663 
664 	if (mode == AMDGPU_AUTO_COMPUTE_PARTITION_MODE) {
665 		mode = __aqua_vanjaram_get_auto_mode(xcp_mgr);
666 		if (mode == AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE) {
667 			dev_err(adev->dev,
668 				"Invalid config, no compatible compute partition mode found, available memory partitions: %d",
669 				adev->gmc.num_mem_partitions);
670 			return -EINVAL;
671 		}
672 	} else if (!__aqua_vanjaram_is_valid_mode(xcp_mgr, mode)) {
673 		dev_err(adev->dev,
674 			"Invalid compute partition mode requested, requested: %s, available memory partitions: %d",
675 			amdgpu_gfx_compute_mode_desc(mode), adev->gmc.num_mem_partitions);
676 		return -EINVAL;
677 	}
678 
679 	if (adev->kfd.init_complete && !amdgpu_in_reset(adev))
680 		flags |= AMDGPU_XCP_OPS_KFD;
681 
682 	if (flags & AMDGPU_XCP_OPS_KFD) {
683 		ret = amdgpu_amdkfd_check_and_lock_kfd(adev);
684 		if (ret)
685 			goto out;
686 	}
687 
688 	ret = __aqua_vanjaram_pre_partition_switch(xcp_mgr, flags);
689 	if (ret)
690 		goto unlock;
691 
692 	num_xcc_per_xcp = __aqua_vanjaram_get_xcc_per_xcp(xcp_mgr, mode);
693 	if (adev->gfx.funcs->switch_partition_mode)
694 		adev->gfx.funcs->switch_partition_mode(xcp_mgr->adev,
695 						       num_xcc_per_xcp);
696 
697 	/* Init info about new xcps */
698 	*num_xcps = num_xcc / num_xcc_per_xcp;
699 	amdgpu_xcp_init(xcp_mgr, *num_xcps, mode);
700 
701 	ret = __aqua_vanjaram_post_partition_switch(xcp_mgr, flags);
702 	if (!ret)
703 		__aqua_vanjaram_update_available_partition_mode(xcp_mgr);
704 unlock:
705 	if (flags & AMDGPU_XCP_OPS_KFD)
706 		amdgpu_amdkfd_unlock_kfd(adev);
707 out:
708 	return ret;
709 }
710 
711 static int __aqua_vanjaram_get_xcp_mem_id(struct amdgpu_device *adev,
712 					  int xcc_id, uint8_t *mem_id)
713 {
714 	/* memory/spatial modes validation check is already done */
715 	*mem_id = xcc_id / adev->gfx.num_xcc_per_xcp;
716 	*mem_id /= adev->xcp_mgr->num_xcp_per_mem_partition;
717 
718 	return 0;
719 }
720 
721 static int aqua_vanjaram_get_xcp_mem_id(struct amdgpu_xcp_mgr *xcp_mgr,
722 					struct amdgpu_xcp *xcp, uint8_t *mem_id)
723 {
724 	struct amdgpu_numa_info numa_info;
725 	struct amdgpu_device *adev;
726 	uint32_t xcc_mask;
727 	int r, i, xcc_id;
728 
729 	adev = xcp_mgr->adev;
730 	/* TODO: BIOS is not returning the right info now
731 	 * Check on this later
732 	 */
733 	/*
734 	if (adev->gmc.gmc_funcs->query_mem_partition_mode)
735 		mode = adev->gmc.gmc_funcs->query_mem_partition_mode(adev);
736 	*/
737 	if (adev->gmc.num_mem_partitions == 1) {
738 		/* Only one range */
739 		*mem_id = 0;
740 		return 0;
741 	}
742 
743 	r = amdgpu_xcp_get_inst_details(xcp, AMDGPU_XCP_GFX, &xcc_mask);
744 	if (r || !xcc_mask)
745 		return -EINVAL;
746 
747 	xcc_id = ffs(xcc_mask) - 1;
748 	if (!adev->gmc.is_app_apu)
749 		return __aqua_vanjaram_get_xcp_mem_id(adev, xcc_id, mem_id);
750 
751 	r = amdgpu_acpi_get_mem_info(adev, xcc_id, &numa_info);
752 
753 	if (r)
754 		return r;
755 
756 	r = -EINVAL;
757 	for (i = 0; i < adev->gmc.num_mem_partitions; ++i) {
758 		if (adev->gmc.mem_partitions[i].numa.node == numa_info.nid) {
759 			*mem_id = i;
760 			r = 0;
761 			break;
762 		}
763 	}
764 
765 	return r;
766 }
767 
768 static int aqua_vanjaram_get_xcp_ip_details(struct amdgpu_xcp_mgr *xcp_mgr, int xcp_id,
769 				     enum AMDGPU_XCP_IP_BLOCK ip_id,
770 				     struct amdgpu_xcp_ip *ip)
771 {
772 	if (!ip)
773 		return -EINVAL;
774 
775 	return __aqua_vanjaram_get_xcp_ip_info(xcp_mgr, xcp_id, ip_id, ip);
776 }
777 
778 struct amdgpu_xcp_mgr_funcs aqua_vanjaram_xcp_funcs = {
779 	.switch_partition_mode = &aqua_vanjaram_switch_partition_mode,
780 	.query_partition_mode = &aqua_vanjaram_query_partition_mode,
781 	.get_ip_details = &aqua_vanjaram_get_xcp_ip_details,
782 	.get_xcp_res_info = &aqua_vanjaram_get_xcp_res_info,
783 	.get_xcp_mem_id = &aqua_vanjaram_get_xcp_mem_id,
784 	.select_scheds = &aqua_vanjaram_select_scheds,
785 	.update_partition_sched_list =
786 		&aqua_vanjaram_update_partition_sched_list
787 };
788 
789 static int aqua_vanjaram_xcp_mgr_init(struct amdgpu_device *adev)
790 {
791 	int ret;
792 
793 	if (amdgpu_sriov_vf(adev))
794 		aqua_vanjaram_xcp_funcs.switch_partition_mode = NULL;
795 
796 	ret = amdgpu_xcp_mgr_init(adev, AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE, 1,
797 				  &aqua_vanjaram_xcp_funcs);
798 	if (ret)
799 		return ret;
800 
801 	__aqua_vanjaram_update_supported_modes(adev->xcp_mgr);
802 	/* TODO: Default memory node affinity init */
803 
804 	return ret;
805 }
806 
807 int aqua_vanjaram_init_soc_config(struct amdgpu_device *adev)
808 {
809 	u32 mask, avail_inst, inst_mask = adev->sdma.sdma_mask;
810 	int ret, i;
811 
812 	/* generally 1 AID supports 4 instances */
813 	adev->sdma.num_inst_per_aid = 4;
814 	adev->sdma.num_instances = NUM_SDMA(adev->sdma.sdma_mask);
815 
816 	adev->aid_mask = i = 1;
817 	inst_mask >>= adev->sdma.num_inst_per_aid;
818 
819 	for (mask = (1 << adev->sdma.num_inst_per_aid) - 1; inst_mask;
820 	     inst_mask >>= adev->sdma.num_inst_per_aid, ++i) {
821 		avail_inst = inst_mask & mask;
822 		if (avail_inst == mask || avail_inst == 0x3 ||
823 		    avail_inst == 0xc)
824 			adev->aid_mask |= (1 << i);
825 	}
826 
827 	/* Harvest config is not used for aqua vanjaram. VCN and JPEGs will be
828 	 * addressed based on logical instance ids.
829 	 */
830 	adev->vcn.harvest_config = 0;
831 	adev->vcn.num_inst_per_aid = 1;
832 	adev->vcn.num_vcn_inst = hweight32(adev->vcn.inst_mask);
833 	adev->jpeg.harvest_config = 0;
834 	adev->jpeg.num_inst_per_aid = 1;
835 	adev->jpeg.num_jpeg_inst = hweight32(adev->jpeg.inst_mask);
836 
837 	ret = aqua_vanjaram_xcp_mgr_init(adev);
838 	if (ret)
839 		return ret;
840 
841 	aqua_vanjaram_ip_map_init(adev);
842 
843 	return 0;
844 }
845 
846 static void aqua_read_smn(struct amdgpu_device *adev,
847 			  struct amdgpu_smn_reg_data *regdata,
848 			  uint64_t smn_addr)
849 {
850 	regdata->addr = smn_addr;
851 	regdata->value = RREG32_PCIE(smn_addr);
852 }
853 
854 struct aqua_reg_list {
855 	uint64_t start_addr;
856 	uint32_t num_regs;
857 	uint32_t incrx;
858 };
859 
860 #define DW_ADDR_INCR	4
861 
862 static void aqua_read_smn_ext(struct amdgpu_device *adev,
863 			      struct amdgpu_smn_reg_data *regdata,
864 			      uint64_t smn_addr, int i)
865 {
866 	regdata->addr =
867 		smn_addr + adev->asic_funcs->encode_ext_smn_addressing(i);
868 	regdata->value = RREG32_PCIE_EXT(regdata->addr);
869 }
870 
871 #define smnreg_0x1A340218	0x1A340218
872 #define smnreg_0x1A3402E4	0x1A3402E4
873 #define smnreg_0x1A340294	0x1A340294
874 #define smreg_0x1A380088	0x1A380088
875 
876 #define NUM_PCIE_SMN_REGS	14
877 
878 static struct aqua_reg_list pcie_reg_addrs[] = {
879 	{ smnreg_0x1A340218, 1, 0 },
880 	{ smnreg_0x1A3402E4, 1, 0 },
881 	{ smnreg_0x1A340294, 6, DW_ADDR_INCR },
882 	{ smreg_0x1A380088, 6, DW_ADDR_INCR },
883 };
884 
885 static ssize_t aqua_vanjaram_read_pcie_state(struct amdgpu_device *adev,
886 					     void *buf, size_t max_size)
887 {
888 	struct amdgpu_reg_state_pcie_v1_0 *pcie_reg_state;
889 	uint32_t start_addr, incrx, num_regs, szbuf;
890 	struct amdgpu_regs_pcie_v1_0 *pcie_regs;
891 	struct amdgpu_smn_reg_data *reg_data;
892 	struct pci_dev *us_pdev, *ds_pdev;
893 	int aer_cap, r, n;
894 
895 	if (!buf || !max_size)
896 		return -EINVAL;
897 
898 	pcie_reg_state = (struct amdgpu_reg_state_pcie_v1_0 *)buf;
899 
900 	szbuf = sizeof(*pcie_reg_state) +
901 		amdgpu_reginst_size(1, sizeof(*pcie_regs), NUM_PCIE_SMN_REGS);
902 	/* Only one instance of pcie regs */
903 	if (max_size < szbuf)
904 		return -EOVERFLOW;
905 
906 	pcie_regs = (struct amdgpu_regs_pcie_v1_0 *)((uint8_t *)buf +
907 						     sizeof(*pcie_reg_state));
908 	pcie_regs->inst_header.instance = 0;
909 	pcie_regs->inst_header.state = AMDGPU_INST_S_OK;
910 	pcie_regs->inst_header.num_smn_regs = NUM_PCIE_SMN_REGS;
911 
912 	reg_data = pcie_regs->smn_reg_values;
913 
914 	for (r = 0; r < ARRAY_SIZE(pcie_reg_addrs); r++) {
915 		start_addr = pcie_reg_addrs[r].start_addr;
916 		incrx = pcie_reg_addrs[r].incrx;
917 		num_regs = pcie_reg_addrs[r].num_regs;
918 		for (n = 0; n < num_regs; n++) {
919 			aqua_read_smn(adev, reg_data, start_addr + n * incrx);
920 			++reg_data;
921 		}
922 	}
923 
924 	ds_pdev = pci_upstream_bridge(adev->pdev);
925 	us_pdev = pci_upstream_bridge(ds_pdev);
926 
927 	pcie_capability_read_word(us_pdev, PCI_EXP_DEVSTA,
928 				  &pcie_regs->device_status);
929 	pcie_capability_read_word(us_pdev, PCI_EXP_LNKSTA,
930 				  &pcie_regs->link_status);
931 
932 	aer_cap = pci_find_ext_capability(us_pdev, PCI_EXT_CAP_ID_ERR);
933 	if (aer_cap) {
934 		pci_read_config_dword(us_pdev, aer_cap + PCI_ERR_COR_STATUS,
935 				      &pcie_regs->pcie_corr_err_status);
936 		pci_read_config_dword(us_pdev, aer_cap + PCI_ERR_UNCOR_STATUS,
937 				      &pcie_regs->pcie_uncorr_err_status);
938 	}
939 
940 	pci_read_config_dword(us_pdev, PCI_PRIMARY_BUS,
941 			      &pcie_regs->sub_bus_number_latency);
942 
943 	pcie_reg_state->common_header.structure_size = szbuf;
944 	pcie_reg_state->common_header.format_revision = 1;
945 	pcie_reg_state->common_header.content_revision = 0;
946 	pcie_reg_state->common_header.state_type = AMDGPU_REG_STATE_TYPE_PCIE;
947 	pcie_reg_state->common_header.num_instances = 1;
948 
949 	return pcie_reg_state->common_header.structure_size;
950 }
951 
952 #define smnreg_0x11A00050	0x11A00050
953 #define smnreg_0x11A00180	0x11A00180
954 #define smnreg_0x11A00070	0x11A00070
955 #define smnreg_0x11A00200	0x11A00200
956 #define smnreg_0x11A0020C	0x11A0020C
957 #define smnreg_0x11A00210	0x11A00210
958 #define smnreg_0x11A00108	0x11A00108
959 
960 #define XGMI_LINK_REG(smnreg, l) ((smnreg) | (l << 20))
961 
962 #define NUM_XGMI_SMN_REGS 25
963 
964 static struct aqua_reg_list xgmi_reg_addrs[] = {
965 	{ smnreg_0x11A00050, 1, 0 },
966 	{ smnreg_0x11A00180, 16, DW_ADDR_INCR },
967 	{ smnreg_0x11A00070, 4, DW_ADDR_INCR },
968 	{ smnreg_0x11A00200, 1, 0 },
969 	{ smnreg_0x11A0020C, 1, 0 },
970 	{ smnreg_0x11A00210, 1, 0 },
971 	{ smnreg_0x11A00108, 1, 0 },
972 };
973 
974 static ssize_t aqua_vanjaram_read_xgmi_state(struct amdgpu_device *adev,
975 					     void *buf, size_t max_size)
976 {
977 	struct amdgpu_reg_state_xgmi_v1_0 *xgmi_reg_state;
978 	uint32_t start_addr, incrx, num_regs, szbuf;
979 	struct amdgpu_regs_xgmi_v1_0 *xgmi_regs;
980 	struct amdgpu_smn_reg_data *reg_data;
981 	const int max_xgmi_instances = 8;
982 	int inst = 0, i, j, r, n;
983 	const int xgmi_inst = 2;
984 	void *p;
985 
986 	if (!buf || !max_size)
987 		return -EINVAL;
988 
989 	xgmi_reg_state = (struct amdgpu_reg_state_xgmi_v1_0 *)buf;
990 
991 	szbuf = sizeof(*xgmi_reg_state) +
992 		amdgpu_reginst_size(max_xgmi_instances, sizeof(*xgmi_regs),
993 				    NUM_XGMI_SMN_REGS);
994 	/* Only one instance of pcie regs */
995 	if (max_size < szbuf)
996 		return -EOVERFLOW;
997 
998 	p = &xgmi_reg_state->xgmi_state_regs[0];
999 	for_each_inst(i, adev->aid_mask) {
1000 		for (j = 0; j < xgmi_inst; ++j) {
1001 			xgmi_regs = (struct amdgpu_regs_xgmi_v1_0 *)p;
1002 			xgmi_regs->inst_header.instance = inst++;
1003 
1004 			xgmi_regs->inst_header.state = AMDGPU_INST_S_OK;
1005 			xgmi_regs->inst_header.num_smn_regs = NUM_XGMI_SMN_REGS;
1006 
1007 			reg_data = xgmi_regs->smn_reg_values;
1008 
1009 			for (r = 0; r < ARRAY_SIZE(xgmi_reg_addrs); r++) {
1010 				start_addr = xgmi_reg_addrs[r].start_addr;
1011 				incrx = xgmi_reg_addrs[r].incrx;
1012 				num_regs = xgmi_reg_addrs[r].num_regs;
1013 
1014 				for (n = 0; n < num_regs; n++) {
1015 					aqua_read_smn_ext(
1016 						adev, reg_data,
1017 						XGMI_LINK_REG(start_addr, j) +
1018 							n * incrx,
1019 						i);
1020 					++reg_data;
1021 				}
1022 			}
1023 			p = reg_data;
1024 		}
1025 	}
1026 
1027 	xgmi_reg_state->common_header.structure_size = szbuf;
1028 	xgmi_reg_state->common_header.format_revision = 1;
1029 	xgmi_reg_state->common_header.content_revision = 0;
1030 	xgmi_reg_state->common_header.state_type = AMDGPU_REG_STATE_TYPE_XGMI;
1031 	xgmi_reg_state->common_header.num_instances = max_xgmi_instances;
1032 
1033 	return xgmi_reg_state->common_header.structure_size;
1034 }
1035 
1036 #define smnreg_0x11C00070	0x11C00070
1037 #define smnreg_0x11C00210	0x11C00210
1038 
1039 static struct aqua_reg_list wafl_reg_addrs[] = {
1040 	{ smnreg_0x11C00070, 4, DW_ADDR_INCR },
1041 	{ smnreg_0x11C00210, 1, 0 },
1042 };
1043 
1044 #define WAFL_LINK_REG(smnreg, l) ((smnreg) | (l << 20))
1045 
1046 #define NUM_WAFL_SMN_REGS 5
1047 
1048 static ssize_t aqua_vanjaram_read_wafl_state(struct amdgpu_device *adev,
1049 					     void *buf, size_t max_size)
1050 {
1051 	struct amdgpu_reg_state_wafl_v1_0 *wafl_reg_state;
1052 	uint32_t start_addr, incrx, num_regs, szbuf;
1053 	struct amdgpu_regs_wafl_v1_0 *wafl_regs;
1054 	struct amdgpu_smn_reg_data *reg_data;
1055 	const int max_wafl_instances = 8;
1056 	int inst = 0, i, j, r, n;
1057 	const int wafl_inst = 2;
1058 	void *p;
1059 
1060 	if (!buf || !max_size)
1061 		return -EINVAL;
1062 
1063 	wafl_reg_state = (struct amdgpu_reg_state_wafl_v1_0 *)buf;
1064 
1065 	szbuf = sizeof(*wafl_reg_state) +
1066 		amdgpu_reginst_size(max_wafl_instances, sizeof(*wafl_regs),
1067 				    NUM_WAFL_SMN_REGS);
1068 
1069 	if (max_size < szbuf)
1070 		return -EOVERFLOW;
1071 
1072 	p = &wafl_reg_state->wafl_state_regs[0];
1073 	for_each_inst(i, adev->aid_mask) {
1074 		for (j = 0; j < wafl_inst; ++j) {
1075 			wafl_regs = (struct amdgpu_regs_wafl_v1_0 *)p;
1076 			wafl_regs->inst_header.instance = inst++;
1077 
1078 			wafl_regs->inst_header.state = AMDGPU_INST_S_OK;
1079 			wafl_regs->inst_header.num_smn_regs = NUM_WAFL_SMN_REGS;
1080 
1081 			reg_data = wafl_regs->smn_reg_values;
1082 
1083 			for (r = 0; r < ARRAY_SIZE(wafl_reg_addrs); r++) {
1084 				start_addr = wafl_reg_addrs[r].start_addr;
1085 				incrx = wafl_reg_addrs[r].incrx;
1086 				num_regs = wafl_reg_addrs[r].num_regs;
1087 				for (n = 0; n < num_regs; n++) {
1088 					aqua_read_smn_ext(
1089 						adev, reg_data,
1090 						WAFL_LINK_REG(start_addr, j) +
1091 							n * incrx,
1092 						i);
1093 					++reg_data;
1094 				}
1095 			}
1096 			p = reg_data;
1097 		}
1098 	}
1099 
1100 	wafl_reg_state->common_header.structure_size = szbuf;
1101 	wafl_reg_state->common_header.format_revision = 1;
1102 	wafl_reg_state->common_header.content_revision = 0;
1103 	wafl_reg_state->common_header.state_type = AMDGPU_REG_STATE_TYPE_WAFL;
1104 	wafl_reg_state->common_header.num_instances = max_wafl_instances;
1105 
1106 	return wafl_reg_state->common_header.structure_size;
1107 }
1108 
1109 #define smnreg_0x1B311060 0x1B311060
1110 #define smnreg_0x1B411060 0x1B411060
1111 #define smnreg_0x1B511060 0x1B511060
1112 #define smnreg_0x1B611060 0x1B611060
1113 
1114 #define smnreg_0x1C307120 0x1C307120
1115 #define smnreg_0x1C317120 0x1C317120
1116 
1117 #define smnreg_0x1C320830 0x1C320830
1118 #define smnreg_0x1C380830 0x1C380830
1119 #define smnreg_0x1C3D0830 0x1C3D0830
1120 #define smnreg_0x1C420830 0x1C420830
1121 
1122 #define smnreg_0x1C320100 0x1C320100
1123 #define smnreg_0x1C380100 0x1C380100
1124 #define smnreg_0x1C3D0100 0x1C3D0100
1125 #define smnreg_0x1C420100 0x1C420100
1126 
1127 #define smnreg_0x1B310500 0x1B310500
1128 #define smnreg_0x1C300400 0x1C300400
1129 
1130 #define USR_CAKE_INCR 0x11000
1131 #define USR_LINK_INCR 0x100000
1132 #define USR_CP_INCR 0x10000
1133 
1134 #define NUM_USR_SMN_REGS	20
1135 
1136 struct aqua_reg_list usr_reg_addrs[] = {
1137 	{ smnreg_0x1B311060, 4, DW_ADDR_INCR },
1138 	{ smnreg_0x1B411060, 4, DW_ADDR_INCR },
1139 	{ smnreg_0x1B511060, 4, DW_ADDR_INCR },
1140 	{ smnreg_0x1B611060, 4, DW_ADDR_INCR },
1141 	{ smnreg_0x1C307120, 2, DW_ADDR_INCR },
1142 	{ smnreg_0x1C317120, 2, DW_ADDR_INCR },
1143 };
1144 
1145 #define NUM_USR1_SMN_REGS	46
1146 struct aqua_reg_list usr1_reg_addrs[] = {
1147 	{ smnreg_0x1C320830, 6, USR_CAKE_INCR },
1148 	{ smnreg_0x1C380830, 5, USR_CAKE_INCR },
1149 	{ smnreg_0x1C3D0830, 5, USR_CAKE_INCR },
1150 	{ smnreg_0x1C420830, 4, USR_CAKE_INCR },
1151 	{ smnreg_0x1C320100, 6, USR_CAKE_INCR },
1152 	{ smnreg_0x1C380100, 5, USR_CAKE_INCR },
1153 	{ smnreg_0x1C3D0100, 5, USR_CAKE_INCR },
1154 	{ smnreg_0x1C420100, 4, USR_CAKE_INCR },
1155 	{ smnreg_0x1B310500, 4, USR_LINK_INCR },
1156 	{ smnreg_0x1C300400, 2, USR_CP_INCR },
1157 };
1158 
1159 static ssize_t aqua_vanjaram_read_usr_state(struct amdgpu_device *adev,
1160 					    void *buf, size_t max_size,
1161 					    int reg_state)
1162 {
1163 	uint32_t start_addr, incrx, num_regs, szbuf, num_smn;
1164 	struct amdgpu_reg_state_usr_v1_0 *usr_reg_state;
1165 	struct amdgpu_regs_usr_v1_0 *usr_regs;
1166 	struct amdgpu_smn_reg_data *reg_data;
1167 	const int max_usr_instances = 4;
1168 	struct aqua_reg_list *reg_addrs;
1169 	int inst = 0, i, n, r, arr_size;
1170 	void *p;
1171 
1172 	if (!buf || !max_size)
1173 		return -EINVAL;
1174 
1175 	switch (reg_state) {
1176 	case AMDGPU_REG_STATE_TYPE_USR:
1177 		arr_size = ARRAY_SIZE(usr_reg_addrs);
1178 		reg_addrs = usr_reg_addrs;
1179 		num_smn = NUM_USR_SMN_REGS;
1180 		break;
1181 	case AMDGPU_REG_STATE_TYPE_USR_1:
1182 		arr_size = ARRAY_SIZE(usr1_reg_addrs);
1183 		reg_addrs = usr1_reg_addrs;
1184 		num_smn = NUM_USR1_SMN_REGS;
1185 		break;
1186 	default:
1187 		return -EINVAL;
1188 	}
1189 
1190 	usr_reg_state = (struct amdgpu_reg_state_usr_v1_0 *)buf;
1191 
1192 	szbuf = sizeof(*usr_reg_state) + amdgpu_reginst_size(max_usr_instances,
1193 							     sizeof(*usr_regs),
1194 							     num_smn);
1195 	if (max_size < szbuf)
1196 		return -EOVERFLOW;
1197 
1198 	p = &usr_reg_state->usr_state_regs[0];
1199 	for_each_inst(i, adev->aid_mask) {
1200 		usr_regs = (struct amdgpu_regs_usr_v1_0 *)p;
1201 		usr_regs->inst_header.instance = inst++;
1202 		usr_regs->inst_header.state = AMDGPU_INST_S_OK;
1203 		usr_regs->inst_header.num_smn_regs = num_smn;
1204 		reg_data = usr_regs->smn_reg_values;
1205 
1206 		for (r = 0; r < arr_size; r++) {
1207 			start_addr = reg_addrs[r].start_addr;
1208 			incrx = reg_addrs[r].incrx;
1209 			num_regs = reg_addrs[r].num_regs;
1210 			for (n = 0; n < num_regs; n++) {
1211 				aqua_read_smn_ext(adev, reg_data,
1212 						  start_addr + n * incrx, i);
1213 				reg_data++;
1214 			}
1215 		}
1216 		p = reg_data;
1217 	}
1218 
1219 	usr_reg_state->common_header.structure_size = szbuf;
1220 	usr_reg_state->common_header.format_revision = 1;
1221 	usr_reg_state->common_header.content_revision = 0;
1222 	usr_reg_state->common_header.state_type = AMDGPU_REG_STATE_TYPE_USR;
1223 	usr_reg_state->common_header.num_instances = max_usr_instances;
1224 
1225 	return usr_reg_state->common_header.structure_size;
1226 }
1227 
1228 ssize_t aqua_vanjaram_get_reg_state(struct amdgpu_device *adev,
1229 				    enum amdgpu_reg_state reg_state, void *buf,
1230 				    size_t max_size)
1231 {
1232 	ssize_t size;
1233 
1234 	switch (reg_state) {
1235 	case AMDGPU_REG_STATE_TYPE_PCIE:
1236 		size = aqua_vanjaram_read_pcie_state(adev, buf, max_size);
1237 		break;
1238 	case AMDGPU_REG_STATE_TYPE_XGMI:
1239 		size = aqua_vanjaram_read_xgmi_state(adev, buf, max_size);
1240 		break;
1241 	case AMDGPU_REG_STATE_TYPE_WAFL:
1242 		size = aqua_vanjaram_read_wafl_state(adev, buf, max_size);
1243 		break;
1244 	case AMDGPU_REG_STATE_TYPE_USR:
1245 		size = aqua_vanjaram_read_usr_state(adev, buf, max_size,
1246 						    AMDGPU_REG_STATE_TYPE_USR);
1247 		break;
1248 	case AMDGPU_REG_STATE_TYPE_USR_1:
1249 		size = aqua_vanjaram_read_usr_state(
1250 			adev, buf, max_size, AMDGPU_REG_STATE_TYPE_USR_1);
1251 		break;
1252 	default:
1253 		return -EINVAL;
1254 	}
1255 
1256 	return size;
1257 }
1258