xref: /linux/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c (revision cbac924200b838cfb8d8b1415113d788089dc50b)
1 /*
2  * Copyright 2016 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 
24 #include <linux/delay.h>
25 #include <linux/kernel.h>
26 #include <linux/firmware.h>
27 #include <linux/module.h>
28 #include <linux/pci.h>
29 
30 #include "amdgpu.h"
31 #include "amdgpu_gfx.h"
32 #include "soc15.h"
33 #include "soc15d.h"
34 #include "amdgpu_atomfirmware.h"
35 #include "amdgpu_pm.h"
36 
37 #include "gc/gc_9_0_offset.h"
38 #include "gc/gc_9_0_sh_mask.h"
39 
40 #include "vega10_enum.h"
41 
42 #include "soc15_common.h"
43 #include "clearstate_gfx9.h"
44 #include "v9_structs.h"
45 
46 #include "ivsrcid/gfx/irqsrcs_gfx_9_0.h"
47 
48 #include "amdgpu_ras.h"
49 
50 #include "gfx_v9_4.h"
51 #include "gfx_v9_0.h"
52 #include "gfx_v9_4_2.h"
53 
54 #include "asic_reg/pwr/pwr_10_0_offset.h"
55 #include "asic_reg/pwr/pwr_10_0_sh_mask.h"
56 #include "asic_reg/gc/gc_9_0_default.h"
57 
58 #define GFX9_NUM_GFX_RINGS     1
59 #define GFX9_MEC_HPD_SIZE 4096
60 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L
61 #define RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET 0x00000000L
62 
63 #define mmGCEA_PROBE_MAP                        0x070c
64 #define mmGCEA_PROBE_MAP_BASE_IDX               0
65 
66 MODULE_FIRMWARE("amdgpu/vega10_ce.bin");
67 MODULE_FIRMWARE("amdgpu/vega10_pfp.bin");
68 MODULE_FIRMWARE("amdgpu/vega10_me.bin");
69 MODULE_FIRMWARE("amdgpu/vega10_mec.bin");
70 MODULE_FIRMWARE("amdgpu/vega10_mec2.bin");
71 MODULE_FIRMWARE("amdgpu/vega10_rlc.bin");
72 
73 MODULE_FIRMWARE("amdgpu/vega12_ce.bin");
74 MODULE_FIRMWARE("amdgpu/vega12_pfp.bin");
75 MODULE_FIRMWARE("amdgpu/vega12_me.bin");
76 MODULE_FIRMWARE("amdgpu/vega12_mec.bin");
77 MODULE_FIRMWARE("amdgpu/vega12_mec2.bin");
78 MODULE_FIRMWARE("amdgpu/vega12_rlc.bin");
79 
80 MODULE_FIRMWARE("amdgpu/vega20_ce.bin");
81 MODULE_FIRMWARE("amdgpu/vega20_pfp.bin");
82 MODULE_FIRMWARE("amdgpu/vega20_me.bin");
83 MODULE_FIRMWARE("amdgpu/vega20_mec.bin");
84 MODULE_FIRMWARE("amdgpu/vega20_mec2.bin");
85 MODULE_FIRMWARE("amdgpu/vega20_rlc.bin");
86 
87 MODULE_FIRMWARE("amdgpu/raven_ce.bin");
88 MODULE_FIRMWARE("amdgpu/raven_pfp.bin");
89 MODULE_FIRMWARE("amdgpu/raven_me.bin");
90 MODULE_FIRMWARE("amdgpu/raven_mec.bin");
91 MODULE_FIRMWARE("amdgpu/raven_mec2.bin");
92 MODULE_FIRMWARE("amdgpu/raven_rlc.bin");
93 
94 MODULE_FIRMWARE("amdgpu/picasso_ce.bin");
95 MODULE_FIRMWARE("amdgpu/picasso_pfp.bin");
96 MODULE_FIRMWARE("amdgpu/picasso_me.bin");
97 MODULE_FIRMWARE("amdgpu/picasso_mec.bin");
98 MODULE_FIRMWARE("amdgpu/picasso_mec2.bin");
99 MODULE_FIRMWARE("amdgpu/picasso_rlc.bin");
100 MODULE_FIRMWARE("amdgpu/picasso_rlc_am4.bin");
101 
102 MODULE_FIRMWARE("amdgpu/raven2_ce.bin");
103 MODULE_FIRMWARE("amdgpu/raven2_pfp.bin");
104 MODULE_FIRMWARE("amdgpu/raven2_me.bin");
105 MODULE_FIRMWARE("amdgpu/raven2_mec.bin");
106 MODULE_FIRMWARE("amdgpu/raven2_mec2.bin");
107 MODULE_FIRMWARE("amdgpu/raven2_rlc.bin");
108 MODULE_FIRMWARE("amdgpu/raven_kicker_rlc.bin");
109 
110 MODULE_FIRMWARE("amdgpu/arcturus_mec.bin");
111 MODULE_FIRMWARE("amdgpu/arcturus_rlc.bin");
112 
113 MODULE_FIRMWARE("amdgpu/renoir_ce.bin");
114 MODULE_FIRMWARE("amdgpu/renoir_pfp.bin");
115 MODULE_FIRMWARE("amdgpu/renoir_me.bin");
116 MODULE_FIRMWARE("amdgpu/renoir_mec.bin");
117 MODULE_FIRMWARE("amdgpu/renoir_rlc.bin");
118 
119 MODULE_FIRMWARE("amdgpu/green_sardine_ce.bin");
120 MODULE_FIRMWARE("amdgpu/green_sardine_pfp.bin");
121 MODULE_FIRMWARE("amdgpu/green_sardine_me.bin");
122 MODULE_FIRMWARE("amdgpu/green_sardine_mec.bin");
123 MODULE_FIRMWARE("amdgpu/green_sardine_mec2.bin");
124 MODULE_FIRMWARE("amdgpu/green_sardine_rlc.bin");
125 
126 MODULE_FIRMWARE("amdgpu/aldebaran_mec.bin");
127 MODULE_FIRMWARE("amdgpu/aldebaran_mec2.bin");
128 MODULE_FIRMWARE("amdgpu/aldebaran_rlc.bin");
129 
130 #define mmTCP_CHAN_STEER_0_ARCT								0x0b03
131 #define mmTCP_CHAN_STEER_0_ARCT_BASE_IDX							0
132 #define mmTCP_CHAN_STEER_1_ARCT								0x0b04
133 #define mmTCP_CHAN_STEER_1_ARCT_BASE_IDX							0
134 #define mmTCP_CHAN_STEER_2_ARCT								0x0b09
135 #define mmTCP_CHAN_STEER_2_ARCT_BASE_IDX							0
136 #define mmTCP_CHAN_STEER_3_ARCT								0x0b0a
137 #define mmTCP_CHAN_STEER_3_ARCT_BASE_IDX							0
138 #define mmTCP_CHAN_STEER_4_ARCT								0x0b0b
139 #define mmTCP_CHAN_STEER_4_ARCT_BASE_IDX							0
140 #define mmTCP_CHAN_STEER_5_ARCT								0x0b0c
141 #define mmTCP_CHAN_STEER_5_ARCT_BASE_IDX							0
142 
143 #define mmGOLDEN_TSC_COUNT_UPPER_Renoir                0x0025
144 #define mmGOLDEN_TSC_COUNT_UPPER_Renoir_BASE_IDX       1
145 #define mmGOLDEN_TSC_COUNT_LOWER_Renoir                0x0026
146 #define mmGOLDEN_TSC_COUNT_LOWER_Renoir_BASE_IDX       1
147 
148 enum ta_ras_gfx_subblock {
149 	/*CPC*/
150 	TA_RAS_BLOCK__GFX_CPC_INDEX_START = 0,
151 	TA_RAS_BLOCK__GFX_CPC_SCRATCH = TA_RAS_BLOCK__GFX_CPC_INDEX_START,
152 	TA_RAS_BLOCK__GFX_CPC_UCODE,
153 	TA_RAS_BLOCK__GFX_DC_STATE_ME1,
154 	TA_RAS_BLOCK__GFX_DC_CSINVOC_ME1,
155 	TA_RAS_BLOCK__GFX_DC_RESTORE_ME1,
156 	TA_RAS_BLOCK__GFX_DC_STATE_ME2,
157 	TA_RAS_BLOCK__GFX_DC_CSINVOC_ME2,
158 	TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
159 	TA_RAS_BLOCK__GFX_CPC_INDEX_END = TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
160 	/* CPF*/
161 	TA_RAS_BLOCK__GFX_CPF_INDEX_START,
162 	TA_RAS_BLOCK__GFX_CPF_ROQ_ME2 = TA_RAS_BLOCK__GFX_CPF_INDEX_START,
163 	TA_RAS_BLOCK__GFX_CPF_ROQ_ME1,
164 	TA_RAS_BLOCK__GFX_CPF_TAG,
165 	TA_RAS_BLOCK__GFX_CPF_INDEX_END = TA_RAS_BLOCK__GFX_CPF_TAG,
166 	/* CPG*/
167 	TA_RAS_BLOCK__GFX_CPG_INDEX_START,
168 	TA_RAS_BLOCK__GFX_CPG_DMA_ROQ = TA_RAS_BLOCK__GFX_CPG_INDEX_START,
169 	TA_RAS_BLOCK__GFX_CPG_DMA_TAG,
170 	TA_RAS_BLOCK__GFX_CPG_TAG,
171 	TA_RAS_BLOCK__GFX_CPG_INDEX_END = TA_RAS_BLOCK__GFX_CPG_TAG,
172 	/* GDS*/
173 	TA_RAS_BLOCK__GFX_GDS_INDEX_START,
174 	TA_RAS_BLOCK__GFX_GDS_MEM = TA_RAS_BLOCK__GFX_GDS_INDEX_START,
175 	TA_RAS_BLOCK__GFX_GDS_INPUT_QUEUE,
176 	TA_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM,
177 	TA_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM,
178 	TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
179 	TA_RAS_BLOCK__GFX_GDS_INDEX_END = TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
180 	/* SPI*/
181 	TA_RAS_BLOCK__GFX_SPI_SR_MEM,
182 	/* SQ*/
183 	TA_RAS_BLOCK__GFX_SQ_INDEX_START,
184 	TA_RAS_BLOCK__GFX_SQ_SGPR = TA_RAS_BLOCK__GFX_SQ_INDEX_START,
185 	TA_RAS_BLOCK__GFX_SQ_LDS_D,
186 	TA_RAS_BLOCK__GFX_SQ_LDS_I,
187 	TA_RAS_BLOCK__GFX_SQ_VGPR, /* VGPR = SP*/
188 	TA_RAS_BLOCK__GFX_SQ_INDEX_END = TA_RAS_BLOCK__GFX_SQ_VGPR,
189 	/* SQC (3 ranges)*/
190 	TA_RAS_BLOCK__GFX_SQC_INDEX_START,
191 	/* SQC range 0*/
192 	TA_RAS_BLOCK__GFX_SQC_INDEX0_START = TA_RAS_BLOCK__GFX_SQC_INDEX_START,
193 	TA_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO =
194 		TA_RAS_BLOCK__GFX_SQC_INDEX0_START,
195 	TA_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF,
196 	TA_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO,
197 	TA_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF,
198 	TA_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO,
199 	TA_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF,
200 	TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
201 	TA_RAS_BLOCK__GFX_SQC_INDEX0_END =
202 		TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
203 	/* SQC range 1*/
204 	TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
205 	TA_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM =
206 		TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
207 	TA_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO,
208 	TA_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO,
209 	TA_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM,
210 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM,
211 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO,
212 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO,
213 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM,
214 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
215 	TA_RAS_BLOCK__GFX_SQC_INDEX1_END =
216 		TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
217 	/* SQC range 2*/
218 	TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
219 	TA_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM =
220 		TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
221 	TA_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO,
222 	TA_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO,
223 	TA_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM,
224 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM,
225 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO,
226 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO,
227 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM,
228 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
229 	TA_RAS_BLOCK__GFX_SQC_INDEX2_END =
230 		TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
231 	TA_RAS_BLOCK__GFX_SQC_INDEX_END = TA_RAS_BLOCK__GFX_SQC_INDEX2_END,
232 	/* TA*/
233 	TA_RAS_BLOCK__GFX_TA_INDEX_START,
234 	TA_RAS_BLOCK__GFX_TA_FS_DFIFO = TA_RAS_BLOCK__GFX_TA_INDEX_START,
235 	TA_RAS_BLOCK__GFX_TA_FS_AFIFO,
236 	TA_RAS_BLOCK__GFX_TA_FL_LFIFO,
237 	TA_RAS_BLOCK__GFX_TA_FX_LFIFO,
238 	TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
239 	TA_RAS_BLOCK__GFX_TA_INDEX_END = TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
240 	/* TCA*/
241 	TA_RAS_BLOCK__GFX_TCA_INDEX_START,
242 	TA_RAS_BLOCK__GFX_TCA_HOLE_FIFO = TA_RAS_BLOCK__GFX_TCA_INDEX_START,
243 	TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
244 	TA_RAS_BLOCK__GFX_TCA_INDEX_END = TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
245 	/* TCC (5 sub-ranges)*/
246 	TA_RAS_BLOCK__GFX_TCC_INDEX_START,
247 	/* TCC range 0*/
248 	TA_RAS_BLOCK__GFX_TCC_INDEX0_START = TA_RAS_BLOCK__GFX_TCC_INDEX_START,
249 	TA_RAS_BLOCK__GFX_TCC_CACHE_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX0_START,
250 	TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1,
251 	TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0,
252 	TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1,
253 	TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0,
254 	TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1,
255 	TA_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG,
256 	TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
257 	TA_RAS_BLOCK__GFX_TCC_INDEX0_END = TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
258 	/* TCC range 1*/
259 	TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
260 	TA_RAS_BLOCK__GFX_TCC_IN_USE_DEC = TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
261 	TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
262 	TA_RAS_BLOCK__GFX_TCC_INDEX1_END =
263 		TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
264 	/* TCC range 2*/
265 	TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
266 	TA_RAS_BLOCK__GFX_TCC_RETURN_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
267 	TA_RAS_BLOCK__GFX_TCC_RETURN_CONTROL,
268 	TA_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO,
269 	TA_RAS_BLOCK__GFX_TCC_WRITE_RETURN,
270 	TA_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ,
271 	TA_RAS_BLOCK__GFX_TCC_SRC_FIFO,
272 	TA_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM,
273 	TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
274 	TA_RAS_BLOCK__GFX_TCC_INDEX2_END =
275 		TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
276 	/* TCC range 3*/
277 	TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
278 	TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO = TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
279 	TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
280 	TA_RAS_BLOCK__GFX_TCC_INDEX3_END =
281 		TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
282 	/* TCC range 4*/
283 	TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
284 	TA_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN =
285 		TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
286 	TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
287 	TA_RAS_BLOCK__GFX_TCC_INDEX4_END =
288 		TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
289 	TA_RAS_BLOCK__GFX_TCC_INDEX_END = TA_RAS_BLOCK__GFX_TCC_INDEX4_END,
290 	/* TCI*/
291 	TA_RAS_BLOCK__GFX_TCI_WRITE_RAM,
292 	/* TCP*/
293 	TA_RAS_BLOCK__GFX_TCP_INDEX_START,
294 	TA_RAS_BLOCK__GFX_TCP_CACHE_RAM = TA_RAS_BLOCK__GFX_TCP_INDEX_START,
295 	TA_RAS_BLOCK__GFX_TCP_LFIFO_RAM,
296 	TA_RAS_BLOCK__GFX_TCP_CMD_FIFO,
297 	TA_RAS_BLOCK__GFX_TCP_VM_FIFO,
298 	TA_RAS_BLOCK__GFX_TCP_DB_RAM,
299 	TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0,
300 	TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
301 	TA_RAS_BLOCK__GFX_TCP_INDEX_END = TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
302 	/* TD*/
303 	TA_RAS_BLOCK__GFX_TD_INDEX_START,
304 	TA_RAS_BLOCK__GFX_TD_SS_FIFO_LO = TA_RAS_BLOCK__GFX_TD_INDEX_START,
305 	TA_RAS_BLOCK__GFX_TD_SS_FIFO_HI,
306 	TA_RAS_BLOCK__GFX_TD_CS_FIFO,
307 	TA_RAS_BLOCK__GFX_TD_INDEX_END = TA_RAS_BLOCK__GFX_TD_CS_FIFO,
308 	/* EA (3 sub-ranges)*/
309 	TA_RAS_BLOCK__GFX_EA_INDEX_START,
310 	/* EA range 0*/
311 	TA_RAS_BLOCK__GFX_EA_INDEX0_START = TA_RAS_BLOCK__GFX_EA_INDEX_START,
312 	TA_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM = TA_RAS_BLOCK__GFX_EA_INDEX0_START,
313 	TA_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM,
314 	TA_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM,
315 	TA_RAS_BLOCK__GFX_EA_RRET_TAGMEM,
316 	TA_RAS_BLOCK__GFX_EA_WRET_TAGMEM,
317 	TA_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM,
318 	TA_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM,
319 	TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
320 	TA_RAS_BLOCK__GFX_EA_INDEX0_END = TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
321 	/* EA range 1*/
322 	TA_RAS_BLOCK__GFX_EA_INDEX1_START,
323 	TA_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM = TA_RAS_BLOCK__GFX_EA_INDEX1_START,
324 	TA_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM,
325 	TA_RAS_BLOCK__GFX_EA_IORD_CMDMEM,
326 	TA_RAS_BLOCK__GFX_EA_IOWR_CMDMEM,
327 	TA_RAS_BLOCK__GFX_EA_IOWR_DATAMEM,
328 	TA_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM,
329 	TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
330 	TA_RAS_BLOCK__GFX_EA_INDEX1_END = TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
331 	/* EA range 2*/
332 	TA_RAS_BLOCK__GFX_EA_INDEX2_START,
333 	TA_RAS_BLOCK__GFX_EA_MAM_D0MEM = TA_RAS_BLOCK__GFX_EA_INDEX2_START,
334 	TA_RAS_BLOCK__GFX_EA_MAM_D1MEM,
335 	TA_RAS_BLOCK__GFX_EA_MAM_D2MEM,
336 	TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
337 	TA_RAS_BLOCK__GFX_EA_INDEX2_END = TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
338 	TA_RAS_BLOCK__GFX_EA_INDEX_END = TA_RAS_BLOCK__GFX_EA_INDEX2_END,
339 	/* UTC VM L2 bank*/
340 	TA_RAS_BLOCK__UTC_VML2_BANK_CACHE,
341 	/* UTC VM walker*/
342 	TA_RAS_BLOCK__UTC_VML2_WALKER,
343 	/* UTC ATC L2 2MB cache*/
344 	TA_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK,
345 	/* UTC ATC L2 4KB cache*/
346 	TA_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK,
347 	TA_RAS_BLOCK__GFX_MAX
348 };
349 
350 struct ras_gfx_subblock {
351 	unsigned char *name;
352 	int ta_subblock;
353 	int hw_supported_error_type;
354 	int sw_supported_error_type;
355 };
356 
357 #define AMDGPU_RAS_SUB_BLOCK(subblock, a, b, c, d, e, f, g, h)                             \
358 	[AMDGPU_RAS_BLOCK__##subblock] = {                                     \
359 		#subblock,                                                     \
360 		TA_RAS_BLOCK__##subblock,                                      \
361 		((a) | ((b) << 1) | ((c) << 2) | ((d) << 3)),                  \
362 		(((e) << 1) | ((f) << 3) | (g) | ((h) << 2)),                  \
363 	}
364 
365 static const struct ras_gfx_subblock ras_gfx_subblocks[] = {
366 	AMDGPU_RAS_SUB_BLOCK(GFX_CPC_SCRATCH, 0, 1, 1, 1, 1, 0, 0, 1),
367 	AMDGPU_RAS_SUB_BLOCK(GFX_CPC_UCODE, 0, 1, 1, 1, 1, 0, 0, 1),
368 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME1, 1, 0, 0, 1, 0, 0, 1, 0),
369 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME1, 1, 0, 0, 1, 0, 0, 0, 0),
370 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME1, 1, 0, 0, 1, 0, 0, 0, 0),
371 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
372 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
373 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
374 	AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
375 	AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME1, 1, 0, 0, 1, 0, 0, 1, 0),
376 	AMDGPU_RAS_SUB_BLOCK(GFX_CPF_TAG, 0, 1, 1, 1, 1, 0, 0, 1),
377 	AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_ROQ, 1, 0, 0, 1, 0, 0, 1, 0),
378 	AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_TAG, 0, 1, 1, 1, 0, 1, 0, 1),
379 	AMDGPU_RAS_SUB_BLOCK(GFX_CPG_TAG, 0, 1, 1, 1, 1, 1, 0, 1),
380 	AMDGPU_RAS_SUB_BLOCK(GFX_GDS_MEM, 0, 1, 1, 1, 0, 0, 0, 0),
381 	AMDGPU_RAS_SUB_BLOCK(GFX_GDS_INPUT_QUEUE, 1, 0, 0, 1, 0, 0, 0, 0),
382 	AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_CMD_RAM_MEM, 0, 1, 1, 1, 0, 0, 0,
383 			     0),
384 	AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_DATA_RAM_MEM, 1, 0, 0, 1, 0, 0, 0,
385 			     0),
386 	AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PIPE_MEM, 0, 1, 1, 1, 0, 0, 0, 0),
387 	AMDGPU_RAS_SUB_BLOCK(GFX_SPI_SR_MEM, 1, 0, 0, 1, 0, 0, 0, 0),
388 	AMDGPU_RAS_SUB_BLOCK(GFX_SQ_SGPR, 0, 1, 1, 1, 0, 0, 0, 0),
389 	AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_D, 0, 1, 1, 1, 1, 0, 0, 1),
390 	AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_I, 0, 1, 1, 1, 0, 0, 0, 0),
391 	AMDGPU_RAS_SUB_BLOCK(GFX_SQ_VGPR, 0, 1, 1, 1, 0, 0, 0, 0),
392 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 1),
393 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
394 			     0, 0),
395 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,
396 			     0),
397 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
398 			     0, 0),
399 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_UTCL1_LFIFO, 0, 1, 1, 1, 1, 0, 0,
400 			     0),
401 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
402 			     0, 0),
403 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,
404 			     0),
405 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,
406 			     1),
407 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,
408 			     0, 0, 0),
409 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
410 			     0),
411 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
412 			     0),
413 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,
414 			     0),
415 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,
416 			     0),
417 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
418 			     0),
419 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,
420 			     0, 0),
421 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
422 			     0),
423 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,
424 			     0),
425 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,
426 			     0, 0, 0),
427 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
428 			     0),
429 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
430 			     0),
431 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,
432 			     0),
433 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,
434 			     0),
435 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
436 			     0),
437 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,
438 			     0, 0),
439 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
440 			     0),
441 	AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_DFIFO, 0, 1, 1, 1, 1, 0, 0, 1),
442 	AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_AFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
443 	AMDGPU_RAS_SUB_BLOCK(GFX_TA_FL_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
444 	AMDGPU_RAS_SUB_BLOCK(GFX_TA_FX_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
445 	AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_CFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
446 	AMDGPU_RAS_SUB_BLOCK(GFX_TCA_HOLE_FIFO, 1, 0, 0, 1, 0, 1, 1, 0),
447 	AMDGPU_RAS_SUB_BLOCK(GFX_TCA_REQ_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
448 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA, 0, 1, 1, 1, 1, 0, 0, 1),
449 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_0_1, 0, 1, 1, 1, 1, 0, 0,
450 			     1),
451 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_0, 0, 1, 1, 1, 1, 0, 0,
452 			     1),
453 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_1, 0, 1, 1, 1, 1, 0, 0,
454 			     1),
455 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_0, 0, 1, 1, 1, 0, 0, 0,
456 			     0),
457 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_1, 0, 1, 1, 1, 0, 0, 0,
458 			     0),
459 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_HIGH_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0),
460 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LOW_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0),
461 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_DEC, 1, 0, 0, 1, 0, 0, 0, 0),
462 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_TRANSFER, 1, 0, 0, 1, 0, 0, 0, 0),
463 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_DATA, 1, 0, 0, 1, 0, 0, 0, 0),
464 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_CONTROL, 1, 0, 0, 1, 0, 0, 0, 0),
465 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_UC_ATOMIC_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
466 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_RETURN, 1, 0, 0, 1, 0, 1, 1, 0),
467 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_CACHE_READ, 1, 0, 0, 1, 0, 0, 0, 0),
468 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO, 0, 1, 1, 1, 0, 0, 0, 0),
469 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 1, 0),
470 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_TAG_PROBE_FIFO, 1, 0, 0, 1, 0, 0, 0,
471 			     0),
472 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
473 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 0,
474 			     0),
475 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRRET_TAG_WRITE_RETURN, 1, 0, 0, 1, 0, 0,
476 			     0, 0),
477 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_ATOMIC_RETURN_BUFFER, 1, 0, 0, 1, 0, 0, 0,
478 			     0),
479 	AMDGPU_RAS_SUB_BLOCK(GFX_TCI_WRITE_RAM, 1, 0, 0, 1, 0, 0, 0, 0),
480 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CACHE_RAM, 0, 1, 1, 1, 1, 0, 0, 1),
481 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_LFIFO_RAM, 0, 1, 1, 1, 0, 0, 0, 0),
482 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CMD_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
483 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_VM_FIFO, 0, 1, 1, 1, 0, 0, 0, 0),
484 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_DB_RAM, 1, 0, 0, 1, 0, 0, 0, 0),
485 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO0, 0, 1, 1, 1, 0, 0, 0, 0),
486 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO1, 0, 1, 1, 1, 0, 0, 0, 0),
487 	AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_LO, 0, 1, 1, 1, 1, 0, 0, 1),
488 	AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_HI, 0, 1, 1, 1, 0, 0, 0, 0),
489 	AMDGPU_RAS_SUB_BLOCK(GFX_TD_CS_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
490 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_CMDMEM, 0, 1, 1, 1, 1, 0, 0, 1),
491 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
492 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0),
493 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_RRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0),
494 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_WRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0),
495 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
496 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
497 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0),
498 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
499 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
500 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_IORD_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0),
501 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0),
502 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_DATAMEM, 1, 0, 0, 1, 0, 0, 0, 0),
503 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
504 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
505 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D0MEM, 1, 0, 0, 1, 0, 0, 0, 0),
506 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D1MEM, 1, 0, 0, 1, 0, 0, 0, 0),
507 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D2MEM, 1, 0, 0, 1, 0, 0, 0, 0),
508 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D3MEM, 1, 0, 0, 1, 0, 0, 0, 0),
509 	AMDGPU_RAS_SUB_BLOCK(UTC_VML2_BANK_CACHE, 0, 1, 1, 1, 0, 0, 0, 0),
510 	AMDGPU_RAS_SUB_BLOCK(UTC_VML2_WALKER, 0, 1, 1, 1, 0, 0, 0, 0),
511 	AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_2M_BANK, 1, 0, 0, 1, 0, 0, 0, 0),
512 	AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_4K_BANK, 0, 1, 1, 1, 0, 0, 0, 0),
513 };
514 
515 static const struct soc15_reg_golden golden_settings_gc_9_0[] =
516 {
517 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400),
518 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG3, 0x80000000, 0x80000000),
519 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
520 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
521 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
522 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
523 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
524 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
525 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
526 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x00ffff87),
527 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x00ffff8f),
528 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
529 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
530 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
531 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
532 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
533 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff),
534 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
535 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
536 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
537 };
538 
539 static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] =
540 {
541 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0x0000f000, 0x00012107),
542 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
543 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
544 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
545 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
546 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
547 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x2a114042),
548 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
549 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0x00008000, 0x00048000),
550 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
551 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
552 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
553 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
554 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
555 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
556 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x01000107),
557 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x00001800, 0x00000800),
558 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080)
559 };
560 
561 static const struct soc15_reg_golden golden_settings_gc_9_0_vg20[] =
562 {
563 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x0f000080, 0x04000080),
564 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
565 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
566 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x22014042),
567 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x22014042),
568 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0x00003e00, 0x00000400),
569 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xff840000, 0x04040000),
570 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00030000),
571 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff010f, 0x01000107),
572 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0x000b0000, 0x000b0000),
573 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01000000, 0x01000000)
574 };
575 
576 static const struct soc15_reg_golden golden_settings_gc_9_1[] =
577 {
578 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
579 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
580 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
581 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
582 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
583 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
584 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
585 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
586 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
587 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
588 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
589 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
590 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
591 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
592 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
593 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
594 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
595 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120),
596 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
597 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000000ff),
598 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080),
599 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
600 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
601 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
602 };
603 
604 static const struct soc15_reg_golden golden_settings_gc_9_1_rv1[] =
605 {
606 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
607 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24000042),
608 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24000042),
609 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04048000),
610 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_MODE_CNTL_1, 0x06000000, 0x06000000),
611 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
612 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x00000800)
613 };
614 
615 static const struct soc15_reg_golden golden_settings_gc_9_1_rv2[] =
616 {
617 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0xff7fffff, 0x04000000),
618 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
619 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000),
620 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x7f0fffff, 0x08000080),
621 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0xff8fffff, 0x08000080),
622 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x7f8fffff, 0x08000080),
623 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x26013041),
624 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x26013041),
625 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x3f8fffff, 0x08000080),
626 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
627 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0xff0fffff, 0x08000080),
628 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0xff0fffff, 0x08000080),
629 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0xff0fffff, 0x08000080),
630 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0xff0fffff, 0x08000080),
631 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0xff0fffff, 0x08000080),
632 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
633 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010),
634 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000),
635 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x3f8fffff, 0x08000080),
636 };
637 
638 static const struct soc15_reg_golden golden_settings_gc_9_1_rn[] =
639 {
640 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
641 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000),
642 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400),
643 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x24000042),
644 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x24000042),
645 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
646 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
647 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
648 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
649 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
650 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120),
651 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGCEA_PROBE_MAP, 0xffffffff, 0x0000cccc),
652 };
653 
654 static const struct soc15_reg_golden golden_settings_gc_9_x_common[] =
655 {
656 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0xffffffff, 0x000001ff),
657 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_INDEX, 0xffffffff, 0x00000000),
658 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_DATA, 0xffffffff, 0x2544c382)
659 };
660 
661 static const struct soc15_reg_golden golden_settings_gc_9_2_1[] =
662 {
663 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
664 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
665 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
666 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
667 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
668 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
669 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
670 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
671 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87),
672 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f),
673 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
674 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
675 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
676 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
677 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
678 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff)
679 };
680 
681 static const struct soc15_reg_golden golden_settings_gc_9_2_1_vg12[] =
682 {
683 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x00000080, 0x04000080),
684 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
685 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
686 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24104041),
687 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24104041),
688 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
689 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff03ff, 0x01000107),
690 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
691 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x76325410),
692 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000),
693 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
694 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
695 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
696 };
697 
698 static const struct soc15_reg_golden golden_settings_gc_9_4_1_arct[] =
699 {
700 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
701 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x10b0000),
702 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_0_ARCT, 0x3fffffff, 0x346f0a4e),
703 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_1_ARCT, 0x3fffffff, 0x1c642ca),
704 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_2_ARCT, 0x3fffffff, 0x26f45098),
705 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_3_ARCT, 0x3fffffff, 0x2ebd9fe3),
706 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_4_ARCT, 0x3fffffff, 0xb90f5b1),
707 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_5_ARCT, 0x3ff, 0x135),
708 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_CONFIG, 0xffffffff, 0x011A0000),
709 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_FIFO_SIZES, 0xffffffff, 0x00000f00),
710 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_UTCL1_CNTL1, 0x30000000, 0x30000000)
711 };
712 
713 static const struct soc15_reg_rlcg rlcg_access_gc_9_0[] = {
714 	{SOC15_REG_ENTRY(GC, 0, mmGRBM_GFX_INDEX)},
715 	{SOC15_REG_ENTRY(GC, 0, mmSQ_IND_INDEX)},
716 };
717 
718 static const u32 GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[] =
719 {
720 	mmRLC_SRM_INDEX_CNTL_ADDR_0 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
721 	mmRLC_SRM_INDEX_CNTL_ADDR_1 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
722 	mmRLC_SRM_INDEX_CNTL_ADDR_2 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
723 	mmRLC_SRM_INDEX_CNTL_ADDR_3 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
724 	mmRLC_SRM_INDEX_CNTL_ADDR_4 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
725 	mmRLC_SRM_INDEX_CNTL_ADDR_5 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
726 	mmRLC_SRM_INDEX_CNTL_ADDR_6 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
727 	mmRLC_SRM_INDEX_CNTL_ADDR_7 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
728 };
729 
730 static const u32 GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[] =
731 {
732 	mmRLC_SRM_INDEX_CNTL_DATA_0 - mmRLC_SRM_INDEX_CNTL_DATA_0,
733 	mmRLC_SRM_INDEX_CNTL_DATA_1 - mmRLC_SRM_INDEX_CNTL_DATA_0,
734 	mmRLC_SRM_INDEX_CNTL_DATA_2 - mmRLC_SRM_INDEX_CNTL_DATA_0,
735 	mmRLC_SRM_INDEX_CNTL_DATA_3 - mmRLC_SRM_INDEX_CNTL_DATA_0,
736 	mmRLC_SRM_INDEX_CNTL_DATA_4 - mmRLC_SRM_INDEX_CNTL_DATA_0,
737 	mmRLC_SRM_INDEX_CNTL_DATA_5 - mmRLC_SRM_INDEX_CNTL_DATA_0,
738 	mmRLC_SRM_INDEX_CNTL_DATA_6 - mmRLC_SRM_INDEX_CNTL_DATA_0,
739 	mmRLC_SRM_INDEX_CNTL_DATA_7 - mmRLC_SRM_INDEX_CNTL_DATA_0,
740 };
741 
742 #define VEGA10_GB_ADDR_CONFIG_GOLDEN 0x2a114042
743 #define VEGA12_GB_ADDR_CONFIG_GOLDEN 0x24104041
744 #define RAVEN_GB_ADDR_CONFIG_GOLDEN 0x24000042
745 #define RAVEN2_GB_ADDR_CONFIG_GOLDEN 0x26013041
746 
747 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev);
748 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev);
749 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev);
750 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev);
751 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
752 				struct amdgpu_cu_info *cu_info);
753 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev);
754 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring);
755 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring);
756 static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
757 					  void *ras_error_status);
758 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
759 				     void *inject_if);
760 static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev);
761 
762 static void gfx_v9_0_kiq_set_resources(struct amdgpu_ring *kiq_ring,
763 				uint64_t queue_mask)
764 {
765 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
766 	amdgpu_ring_write(kiq_ring,
767 		PACKET3_SET_RESOURCES_VMID_MASK(0) |
768 		/* vmid_mask:0* queue_type:0 (KIQ) */
769 		PACKET3_SET_RESOURCES_QUEUE_TYPE(0));
770 	amdgpu_ring_write(kiq_ring,
771 			lower_32_bits(queue_mask));	/* queue mask lo */
772 	amdgpu_ring_write(kiq_ring,
773 			upper_32_bits(queue_mask));	/* queue mask hi */
774 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
775 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
776 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
777 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
778 }
779 
780 static void gfx_v9_0_kiq_map_queues(struct amdgpu_ring *kiq_ring,
781 				 struct amdgpu_ring *ring)
782 {
783 	struct amdgpu_device *adev = kiq_ring->adev;
784 	uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
785 	uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
786 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
787 
788 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
789 	/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
790 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
791 			 PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
792 			 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
793 			 PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
794 			 PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
795 			 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
796 			 /*queue_type: normal compute queue */
797 			 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) |
798 			 /* alloc format: all_on_one_pipe */
799 			 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) |
800 			 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
801 			 /* num_queues: must be 1 */
802 			 PACKET3_MAP_QUEUES_NUM_QUEUES(1));
803 	amdgpu_ring_write(kiq_ring,
804 			PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
805 	amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
806 	amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
807 	amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
808 	amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
809 }
810 
811 static void gfx_v9_0_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
812 				   struct amdgpu_ring *ring,
813 				   enum amdgpu_unmap_queues_action action,
814 				   u64 gpu_addr, u64 seq)
815 {
816 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
817 
818 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
819 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
820 			  PACKET3_UNMAP_QUEUES_ACTION(action) |
821 			  PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
822 			  PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
823 			  PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
824 	amdgpu_ring_write(kiq_ring,
825 			PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
826 
827 	if (action == PREEMPT_QUEUES_NO_UNMAP) {
828 		amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr));
829 		amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr));
830 		amdgpu_ring_write(kiq_ring, seq);
831 	} else {
832 		amdgpu_ring_write(kiq_ring, 0);
833 		amdgpu_ring_write(kiq_ring, 0);
834 		amdgpu_ring_write(kiq_ring, 0);
835 	}
836 }
837 
838 static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
839 				   struct amdgpu_ring *ring,
840 				   u64 addr,
841 				   u64 seq)
842 {
843 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
844 
845 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
846 	amdgpu_ring_write(kiq_ring,
847 			  PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
848 			  PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
849 			  PACKET3_QUERY_STATUS_COMMAND(2));
850 	/* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
851 	amdgpu_ring_write(kiq_ring,
852 			PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
853 			PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
854 	amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
855 	amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
856 	amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
857 	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
858 }
859 
860 static void gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
861 				uint16_t pasid, uint32_t flush_type,
862 				bool all_hub)
863 {
864 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
865 	amdgpu_ring_write(kiq_ring,
866 			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
867 			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
868 			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
869 			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
870 }
871 
872 static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
873 	.kiq_set_resources = gfx_v9_0_kiq_set_resources,
874 	.kiq_map_queues = gfx_v9_0_kiq_map_queues,
875 	.kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
876 	.kiq_query_status = gfx_v9_0_kiq_query_status,
877 	.kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs,
878 	.set_resources_size = 8,
879 	.map_queues_size = 7,
880 	.unmap_queues_size = 6,
881 	.query_status_size = 7,
882 	.invalidate_tlbs_size = 2,
883 };
884 
885 static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
886 {
887 	adev->gfx.kiq.pmf = &gfx_v9_0_kiq_pm4_funcs;
888 }
889 
890 static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev)
891 {
892 	switch (adev->ip_versions[GC_HWIP][0]) {
893 	case IP_VERSION(9, 0, 1):
894 		soc15_program_register_sequence(adev,
895 						golden_settings_gc_9_0,
896 						ARRAY_SIZE(golden_settings_gc_9_0));
897 		soc15_program_register_sequence(adev,
898 						golden_settings_gc_9_0_vg10,
899 						ARRAY_SIZE(golden_settings_gc_9_0_vg10));
900 		break;
901 	case IP_VERSION(9, 2, 1):
902 		soc15_program_register_sequence(adev,
903 						golden_settings_gc_9_2_1,
904 						ARRAY_SIZE(golden_settings_gc_9_2_1));
905 		soc15_program_register_sequence(adev,
906 						golden_settings_gc_9_2_1_vg12,
907 						ARRAY_SIZE(golden_settings_gc_9_2_1_vg12));
908 		break;
909 	case IP_VERSION(9, 4, 0):
910 		soc15_program_register_sequence(adev,
911 						golden_settings_gc_9_0,
912 						ARRAY_SIZE(golden_settings_gc_9_0));
913 		soc15_program_register_sequence(adev,
914 						golden_settings_gc_9_0_vg20,
915 						ARRAY_SIZE(golden_settings_gc_9_0_vg20));
916 		break;
917 	case IP_VERSION(9, 4, 1):
918 		soc15_program_register_sequence(adev,
919 						golden_settings_gc_9_4_1_arct,
920 						ARRAY_SIZE(golden_settings_gc_9_4_1_arct));
921 		break;
922 	case IP_VERSION(9, 2, 2):
923 	case IP_VERSION(9, 1, 0):
924 		soc15_program_register_sequence(adev, golden_settings_gc_9_1,
925 						ARRAY_SIZE(golden_settings_gc_9_1));
926 		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
927 			soc15_program_register_sequence(adev,
928 							golden_settings_gc_9_1_rv2,
929 							ARRAY_SIZE(golden_settings_gc_9_1_rv2));
930 		else
931 			soc15_program_register_sequence(adev,
932 							golden_settings_gc_9_1_rv1,
933 							ARRAY_SIZE(golden_settings_gc_9_1_rv1));
934 		break;
935 	 case IP_VERSION(9, 3, 0):
936 		soc15_program_register_sequence(adev,
937 						golden_settings_gc_9_1_rn,
938 						ARRAY_SIZE(golden_settings_gc_9_1_rn));
939 		return; /* for renoir, don't need common goldensetting */
940 	case IP_VERSION(9, 4, 2):
941 		gfx_v9_4_2_init_golden_registers(adev,
942 						 adev->smuio.funcs->get_die_id(adev));
943 		break;
944 	default:
945 		break;
946 	}
947 
948 	if ((adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)) &&
949 	    (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 2)))
950 		soc15_program_register_sequence(adev, golden_settings_gc_9_x_common,
951 						(const u32)ARRAY_SIZE(golden_settings_gc_9_x_common));
952 }
953 
954 static void gfx_v9_0_scratch_init(struct amdgpu_device *adev)
955 {
956 	adev->gfx.scratch.num_reg = 8;
957 	adev->gfx.scratch.reg_base = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0);
958 	adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
959 }
960 
961 static void gfx_v9_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
962 				       bool wc, uint32_t reg, uint32_t val)
963 {
964 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
965 	amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) |
966 				WRITE_DATA_DST_SEL(0) |
967 				(wc ? WR_CONFIRM : 0));
968 	amdgpu_ring_write(ring, reg);
969 	amdgpu_ring_write(ring, 0);
970 	amdgpu_ring_write(ring, val);
971 }
972 
973 static void gfx_v9_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
974 				  int mem_space, int opt, uint32_t addr0,
975 				  uint32_t addr1, uint32_t ref, uint32_t mask,
976 				  uint32_t inv)
977 {
978 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
979 	amdgpu_ring_write(ring,
980 				 /* memory (1) or register (0) */
981 				 (WAIT_REG_MEM_MEM_SPACE(mem_space) |
982 				 WAIT_REG_MEM_OPERATION(opt) | /* wait */
983 				 WAIT_REG_MEM_FUNCTION(3) |  /* equal */
984 				 WAIT_REG_MEM_ENGINE(eng_sel)));
985 
986 	if (mem_space)
987 		BUG_ON(addr0 & 0x3); /* Dword align */
988 	amdgpu_ring_write(ring, addr0);
989 	amdgpu_ring_write(ring, addr1);
990 	amdgpu_ring_write(ring, ref);
991 	amdgpu_ring_write(ring, mask);
992 	amdgpu_ring_write(ring, inv); /* poll interval */
993 }
994 
995 static int gfx_v9_0_ring_test_ring(struct amdgpu_ring *ring)
996 {
997 	struct amdgpu_device *adev = ring->adev;
998 	uint32_t scratch;
999 	uint32_t tmp = 0;
1000 	unsigned i;
1001 	int r;
1002 
1003 	r = amdgpu_gfx_scratch_get(adev, &scratch);
1004 	if (r)
1005 		return r;
1006 
1007 	WREG32(scratch, 0xCAFEDEAD);
1008 	r = amdgpu_ring_alloc(ring, 3);
1009 	if (r)
1010 		goto error_free_scratch;
1011 
1012 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
1013 	amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
1014 	amdgpu_ring_write(ring, 0xDEADBEEF);
1015 	amdgpu_ring_commit(ring);
1016 
1017 	for (i = 0; i < adev->usec_timeout; i++) {
1018 		tmp = RREG32(scratch);
1019 		if (tmp == 0xDEADBEEF)
1020 			break;
1021 		udelay(1);
1022 	}
1023 
1024 	if (i >= adev->usec_timeout)
1025 		r = -ETIMEDOUT;
1026 
1027 error_free_scratch:
1028 	amdgpu_gfx_scratch_free(adev, scratch);
1029 	return r;
1030 }
1031 
1032 static int gfx_v9_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1033 {
1034 	struct amdgpu_device *adev = ring->adev;
1035 	struct amdgpu_ib ib;
1036 	struct dma_fence *f = NULL;
1037 
1038 	unsigned index;
1039 	uint64_t gpu_addr;
1040 	uint32_t tmp;
1041 	long r;
1042 
1043 	r = amdgpu_device_wb_get(adev, &index);
1044 	if (r)
1045 		return r;
1046 
1047 	gpu_addr = adev->wb.gpu_addr + (index * 4);
1048 	adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
1049 	memset(&ib, 0, sizeof(ib));
1050 	r = amdgpu_ib_get(adev, NULL, 16,
1051 					AMDGPU_IB_POOL_DIRECT, &ib);
1052 	if (r)
1053 		goto err1;
1054 
1055 	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
1056 	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
1057 	ib.ptr[2] = lower_32_bits(gpu_addr);
1058 	ib.ptr[3] = upper_32_bits(gpu_addr);
1059 	ib.ptr[4] = 0xDEADBEEF;
1060 	ib.length_dw = 5;
1061 
1062 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1063 	if (r)
1064 		goto err2;
1065 
1066 	r = dma_fence_wait_timeout(f, false, timeout);
1067 	if (r == 0) {
1068 		r = -ETIMEDOUT;
1069 		goto err2;
1070 	} else if (r < 0) {
1071 		goto err2;
1072 	}
1073 
1074 	tmp = adev->wb.wb[index];
1075 	if (tmp == 0xDEADBEEF)
1076 		r = 0;
1077 	else
1078 		r = -EINVAL;
1079 
1080 err2:
1081 	amdgpu_ib_free(adev, &ib, NULL);
1082 	dma_fence_put(f);
1083 err1:
1084 	amdgpu_device_wb_free(adev, index);
1085 	return r;
1086 }
1087 
1088 
1089 static void gfx_v9_0_free_microcode(struct amdgpu_device *adev)
1090 {
1091 	release_firmware(adev->gfx.pfp_fw);
1092 	adev->gfx.pfp_fw = NULL;
1093 	release_firmware(adev->gfx.me_fw);
1094 	adev->gfx.me_fw = NULL;
1095 	release_firmware(adev->gfx.ce_fw);
1096 	adev->gfx.ce_fw = NULL;
1097 	release_firmware(adev->gfx.rlc_fw);
1098 	adev->gfx.rlc_fw = NULL;
1099 	release_firmware(adev->gfx.mec_fw);
1100 	adev->gfx.mec_fw = NULL;
1101 	release_firmware(adev->gfx.mec2_fw);
1102 	adev->gfx.mec2_fw = NULL;
1103 
1104 	kfree(adev->gfx.rlc.register_list_format);
1105 }
1106 
1107 static void gfx_v9_0_init_rlc_ext_microcode(struct amdgpu_device *adev)
1108 {
1109 	const struct rlc_firmware_header_v2_1 *rlc_hdr;
1110 
1111 	rlc_hdr = (const struct rlc_firmware_header_v2_1 *)adev->gfx.rlc_fw->data;
1112 	adev->gfx.rlc_srlc_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_ucode_ver);
1113 	adev->gfx.rlc_srlc_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_feature_ver);
1114 	adev->gfx.rlc.save_restore_list_cntl_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_cntl_size_bytes);
1115 	adev->gfx.rlc.save_restore_list_cntl = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_cntl_offset_bytes);
1116 	adev->gfx.rlc_srlg_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_ucode_ver);
1117 	adev->gfx.rlc_srlg_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_feature_ver);
1118 	adev->gfx.rlc.save_restore_list_gpm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_gpm_size_bytes);
1119 	adev->gfx.rlc.save_restore_list_gpm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_gpm_offset_bytes);
1120 	adev->gfx.rlc_srls_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_ucode_ver);
1121 	adev->gfx.rlc_srls_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_feature_ver);
1122 	adev->gfx.rlc.save_restore_list_srm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_srm_size_bytes);
1123 	adev->gfx.rlc.save_restore_list_srm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_srm_offset_bytes);
1124 	adev->gfx.rlc.reg_list_format_direct_reg_list_length =
1125 			le32_to_cpu(rlc_hdr->reg_list_format_direct_reg_list_length);
1126 }
1127 
1128 static void gfx_v9_0_check_fw_write_wait(struct amdgpu_device *adev)
1129 {
1130 	adev->gfx.me_fw_write_wait = false;
1131 	adev->gfx.mec_fw_write_wait = false;
1132 
1133 	if ((adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)) &&
1134 	    ((adev->gfx.mec_fw_version < 0x000001a5) ||
1135 	    (adev->gfx.mec_feature_version < 46) ||
1136 	    (adev->gfx.pfp_fw_version < 0x000000b7) ||
1137 	    (adev->gfx.pfp_feature_version < 46)))
1138 		DRM_WARN_ONCE("CP firmware version too old, please update!");
1139 
1140 	switch (adev->ip_versions[GC_HWIP][0]) {
1141 	case IP_VERSION(9, 0, 1):
1142 		if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1143 		    (adev->gfx.me_feature_version >= 42) &&
1144 		    (adev->gfx.pfp_fw_version >=  0x000000b1) &&
1145 		    (adev->gfx.pfp_feature_version >= 42))
1146 			adev->gfx.me_fw_write_wait = true;
1147 
1148 		if ((adev->gfx.mec_fw_version >=  0x00000193) &&
1149 		    (adev->gfx.mec_feature_version >= 42))
1150 			adev->gfx.mec_fw_write_wait = true;
1151 		break;
1152 	case IP_VERSION(9, 2, 1):
1153 		if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1154 		    (adev->gfx.me_feature_version >= 44) &&
1155 		    (adev->gfx.pfp_fw_version >=  0x000000b2) &&
1156 		    (adev->gfx.pfp_feature_version >= 44))
1157 			adev->gfx.me_fw_write_wait = true;
1158 
1159 		if ((adev->gfx.mec_fw_version >=  0x00000196) &&
1160 		    (adev->gfx.mec_feature_version >= 44))
1161 			adev->gfx.mec_fw_write_wait = true;
1162 		break;
1163 	case IP_VERSION(9, 4, 0):
1164 		if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1165 		    (adev->gfx.me_feature_version >= 44) &&
1166 		    (adev->gfx.pfp_fw_version >=  0x000000b2) &&
1167 		    (adev->gfx.pfp_feature_version >= 44))
1168 			adev->gfx.me_fw_write_wait = true;
1169 
1170 		if ((adev->gfx.mec_fw_version >=  0x00000197) &&
1171 		    (adev->gfx.mec_feature_version >= 44))
1172 			adev->gfx.mec_fw_write_wait = true;
1173 		break;
1174 	case IP_VERSION(9, 1, 0):
1175 	case IP_VERSION(9, 2, 2):
1176 		if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1177 		    (adev->gfx.me_feature_version >= 42) &&
1178 		    (adev->gfx.pfp_fw_version >=  0x000000b1) &&
1179 		    (adev->gfx.pfp_feature_version >= 42))
1180 			adev->gfx.me_fw_write_wait = true;
1181 
1182 		if ((adev->gfx.mec_fw_version >=  0x00000192) &&
1183 		    (adev->gfx.mec_feature_version >= 42))
1184 			adev->gfx.mec_fw_write_wait = true;
1185 		break;
1186 	default:
1187 		adev->gfx.me_fw_write_wait = true;
1188 		adev->gfx.mec_fw_write_wait = true;
1189 		break;
1190 	}
1191 }
1192 
1193 struct amdgpu_gfxoff_quirk {
1194 	u16 chip_vendor;
1195 	u16 chip_device;
1196 	u16 subsys_vendor;
1197 	u16 subsys_device;
1198 	u8 revision;
1199 };
1200 
1201 static const struct amdgpu_gfxoff_quirk amdgpu_gfxoff_quirk_list[] = {
1202 	/* https://bugzilla.kernel.org/show_bug.cgi?id=204689 */
1203 	{ 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc8 },
1204 	/* https://bugzilla.kernel.org/show_bug.cgi?id=207171 */
1205 	{ 0x1002, 0x15dd, 0x103c, 0x83e7, 0xd3 },
1206 	/* GFXOFF is unstable on C6 parts with a VBIOS 113-RAVEN-114 */
1207 	{ 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc6 },
1208 	{ 0, 0, 0, 0, 0 },
1209 };
1210 
1211 static bool gfx_v9_0_should_disable_gfxoff(struct pci_dev *pdev)
1212 {
1213 	const struct amdgpu_gfxoff_quirk *p = amdgpu_gfxoff_quirk_list;
1214 
1215 	while (p && p->chip_device != 0) {
1216 		if (pdev->vendor == p->chip_vendor &&
1217 		    pdev->device == p->chip_device &&
1218 		    pdev->subsystem_vendor == p->subsys_vendor &&
1219 		    pdev->subsystem_device == p->subsys_device &&
1220 		    pdev->revision == p->revision) {
1221 			return true;
1222 		}
1223 		++p;
1224 	}
1225 	return false;
1226 }
1227 
1228 static bool is_raven_kicker(struct amdgpu_device *adev)
1229 {
1230 	if (adev->pm.fw_version >= 0x41e2b)
1231 		return true;
1232 	else
1233 		return false;
1234 }
1235 
1236 static bool check_if_enlarge_doorbell_range(struct amdgpu_device *adev)
1237 {
1238 	if ((adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 3, 0)) &&
1239 	    (adev->gfx.me_fw_version >= 0x000000a5) &&
1240 	    (adev->gfx.me_feature_version >= 52))
1241 		return true;
1242 	else
1243 		return false;
1244 }
1245 
1246 static void gfx_v9_0_check_if_need_gfxoff(struct amdgpu_device *adev)
1247 {
1248 	if (gfx_v9_0_should_disable_gfxoff(adev->pdev))
1249 		adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1250 
1251 	switch (adev->ip_versions[GC_HWIP][0]) {
1252 	case IP_VERSION(9, 0, 1):
1253 	case IP_VERSION(9, 2, 1):
1254 	case IP_VERSION(9, 4, 0):
1255 		break;
1256 	case IP_VERSION(9, 2, 2):
1257 	case IP_VERSION(9, 1, 0):
1258 		if (!((adev->apu_flags & AMD_APU_IS_RAVEN2) ||
1259 		      (adev->apu_flags & AMD_APU_IS_PICASSO)) &&
1260 		    ((!is_raven_kicker(adev) &&
1261 		      adev->gfx.rlc_fw_version < 531) ||
1262 		     (adev->gfx.rlc_feature_version < 1) ||
1263 		     !adev->gfx.rlc.is_rlc_v2_1))
1264 			adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1265 
1266 		if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1267 			adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG |
1268 				AMD_PG_SUPPORT_CP |
1269 				AMD_PG_SUPPORT_RLC_SMU_HS;
1270 		break;
1271 	case IP_VERSION(9, 3, 0):
1272 		if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1273 			adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG |
1274 				AMD_PG_SUPPORT_CP |
1275 				AMD_PG_SUPPORT_RLC_SMU_HS;
1276 		break;
1277 	default:
1278 		break;
1279 	}
1280 }
1281 
1282 static int gfx_v9_0_init_cp_gfx_microcode(struct amdgpu_device *adev,
1283 					  const char *chip_name)
1284 {
1285 	char fw_name[30];
1286 	int err;
1287 	struct amdgpu_firmware_info *info = NULL;
1288 	const struct common_firmware_header *header = NULL;
1289 	const struct gfx_firmware_header_v1_0 *cp_hdr;
1290 
1291 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
1292 	err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1293 	if (err)
1294 		goto out;
1295 	err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
1296 	if (err)
1297 		goto out;
1298 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
1299 	adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1300 	adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1301 
1302 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1303 	err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1304 	if (err)
1305 		goto out;
1306 	err = amdgpu_ucode_validate(adev->gfx.me_fw);
1307 	if (err)
1308 		goto out;
1309 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
1310 	adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1311 	adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1312 
1313 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1314 	err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1315 	if (err)
1316 		goto out;
1317 	err = amdgpu_ucode_validate(adev->gfx.ce_fw);
1318 	if (err)
1319 		goto out;
1320 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
1321 	adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1322 	adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1323 
1324 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1325 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
1326 		info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
1327 		info->fw = adev->gfx.pfp_fw;
1328 		header = (const struct common_firmware_header *)info->fw->data;
1329 		adev->firmware.fw_size +=
1330 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1331 
1332 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
1333 		info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
1334 		info->fw = adev->gfx.me_fw;
1335 		header = (const struct common_firmware_header *)info->fw->data;
1336 		adev->firmware.fw_size +=
1337 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1338 
1339 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE];
1340 		info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
1341 		info->fw = adev->gfx.ce_fw;
1342 		header = (const struct common_firmware_header *)info->fw->data;
1343 		adev->firmware.fw_size +=
1344 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1345 	}
1346 
1347 out:
1348 	if (err) {
1349 		dev_err(adev->dev,
1350 			"gfx9: Failed to load firmware \"%s\"\n",
1351 			fw_name);
1352 		release_firmware(adev->gfx.pfp_fw);
1353 		adev->gfx.pfp_fw = NULL;
1354 		release_firmware(adev->gfx.me_fw);
1355 		adev->gfx.me_fw = NULL;
1356 		release_firmware(adev->gfx.ce_fw);
1357 		adev->gfx.ce_fw = NULL;
1358 	}
1359 	return err;
1360 }
1361 
1362 static int gfx_v9_0_init_rlc_microcode(struct amdgpu_device *adev,
1363 					  const char *chip_name)
1364 {
1365 	char fw_name[30];
1366 	int err;
1367 	struct amdgpu_firmware_info *info = NULL;
1368 	const struct common_firmware_header *header = NULL;
1369 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
1370 	unsigned int *tmp = NULL;
1371 	unsigned int i = 0;
1372 	uint16_t version_major;
1373 	uint16_t version_minor;
1374 	uint32_t smu_version;
1375 
1376 	/*
1377 	 * For Picasso && AM4 SOCKET board, we use picasso_rlc_am4.bin
1378 	 * instead of picasso_rlc.bin.
1379 	 * Judgment method:
1380 	 * PCO AM4: revision >= 0xC8 && revision <= 0xCF
1381 	 *          or revision >= 0xD8 && revision <= 0xDF
1382 	 * otherwise is PCO FP5
1383 	 */
1384 	if (!strcmp(chip_name, "picasso") &&
1385 		(((adev->pdev->revision >= 0xC8) && (adev->pdev->revision <= 0xCF)) ||
1386 		((adev->pdev->revision >= 0xD8) && (adev->pdev->revision <= 0xDF))))
1387 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc_am4.bin", chip_name);
1388 	else if (!strcmp(chip_name, "raven") && (amdgpu_pm_load_smu_firmware(adev, &smu_version) == 0) &&
1389 		(smu_version >= 0x41e2b))
1390 		/**
1391 		*SMC is loaded by SBIOS on APU and it's able to get the SMU version directly.
1392 		*/
1393 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_kicker_rlc.bin", chip_name);
1394 	else
1395 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
1396 	err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
1397 	if (err)
1398 		goto out;
1399 	err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
1400 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1401 
1402 	version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
1403 	version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
1404 	if (version_major == 2 && version_minor == 1)
1405 		adev->gfx.rlc.is_rlc_v2_1 = true;
1406 
1407 	adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
1408 	adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
1409 	adev->gfx.rlc.save_and_restore_offset =
1410 			le32_to_cpu(rlc_hdr->save_and_restore_offset);
1411 	adev->gfx.rlc.clear_state_descriptor_offset =
1412 			le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
1413 	adev->gfx.rlc.avail_scratch_ram_locations =
1414 			le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
1415 	adev->gfx.rlc.reg_restore_list_size =
1416 			le32_to_cpu(rlc_hdr->reg_restore_list_size);
1417 	adev->gfx.rlc.reg_list_format_start =
1418 			le32_to_cpu(rlc_hdr->reg_list_format_start);
1419 	adev->gfx.rlc.reg_list_format_separate_start =
1420 			le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
1421 	adev->gfx.rlc.starting_offsets_start =
1422 			le32_to_cpu(rlc_hdr->starting_offsets_start);
1423 	adev->gfx.rlc.reg_list_format_size_bytes =
1424 			le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
1425 	adev->gfx.rlc.reg_list_size_bytes =
1426 			le32_to_cpu(rlc_hdr->reg_list_size_bytes);
1427 	adev->gfx.rlc.register_list_format =
1428 			kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
1429 				adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
1430 	if (!adev->gfx.rlc.register_list_format) {
1431 		err = -ENOMEM;
1432 		goto out;
1433 	}
1434 
1435 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1436 			le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
1437 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2); i++)
1438 		adev->gfx.rlc.register_list_format[i] =	le32_to_cpu(tmp[i]);
1439 
1440 	adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
1441 
1442 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1443 			le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
1444 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_size_bytes >> 2); i++)
1445 		adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
1446 
1447 	if (adev->gfx.rlc.is_rlc_v2_1)
1448 		gfx_v9_0_init_rlc_ext_microcode(adev);
1449 
1450 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1451 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
1452 		info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
1453 		info->fw = adev->gfx.rlc_fw;
1454 		header = (const struct common_firmware_header *)info->fw->data;
1455 		adev->firmware.fw_size +=
1456 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1457 
1458 		if (adev->gfx.rlc.is_rlc_v2_1 &&
1459 		    adev->gfx.rlc.save_restore_list_cntl_size_bytes &&
1460 		    adev->gfx.rlc.save_restore_list_gpm_size_bytes &&
1461 		    adev->gfx.rlc.save_restore_list_srm_size_bytes) {
1462 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL];
1463 			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL;
1464 			info->fw = adev->gfx.rlc_fw;
1465 			adev->firmware.fw_size +=
1466 				ALIGN(adev->gfx.rlc.save_restore_list_cntl_size_bytes, PAGE_SIZE);
1467 
1468 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM];
1469 			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM;
1470 			info->fw = adev->gfx.rlc_fw;
1471 			adev->firmware.fw_size +=
1472 				ALIGN(adev->gfx.rlc.save_restore_list_gpm_size_bytes, PAGE_SIZE);
1473 
1474 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM];
1475 			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM;
1476 			info->fw = adev->gfx.rlc_fw;
1477 			adev->firmware.fw_size +=
1478 				ALIGN(adev->gfx.rlc.save_restore_list_srm_size_bytes, PAGE_SIZE);
1479 		}
1480 	}
1481 
1482 out:
1483 	if (err) {
1484 		dev_err(adev->dev,
1485 			"gfx9: Failed to load firmware \"%s\"\n",
1486 			fw_name);
1487 		release_firmware(adev->gfx.rlc_fw);
1488 		adev->gfx.rlc_fw = NULL;
1489 	}
1490 	return err;
1491 }
1492 
1493 static bool gfx_v9_0_load_mec2_fw_bin_support(struct amdgpu_device *adev)
1494 {
1495 	if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2) ||
1496 	    adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1) ||
1497 	    adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 3, 0))
1498 		return false;
1499 
1500 	return true;
1501 }
1502 
1503 static int gfx_v9_0_init_cp_compute_microcode(struct amdgpu_device *adev,
1504 					  const char *chip_name)
1505 {
1506 	char fw_name[30];
1507 	int err;
1508 	struct amdgpu_firmware_info *info = NULL;
1509 	const struct common_firmware_header *header = NULL;
1510 	const struct gfx_firmware_header_v1_0 *cp_hdr;
1511 
1512 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1513 	err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1514 	if (err)
1515 		goto out;
1516 	err = amdgpu_ucode_validate(adev->gfx.mec_fw);
1517 	if (err)
1518 		goto out;
1519 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1520 	adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1521 	adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1522 
1523 
1524 	if (gfx_v9_0_load_mec2_fw_bin_support(adev)) {
1525 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1526 		err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1527 		if (!err) {
1528 			err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
1529 			if (err)
1530 				goto out;
1531 			cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1532 			adev->gfx.mec2_fw->data;
1533 			adev->gfx.mec2_fw_version =
1534 			le32_to_cpu(cp_hdr->header.ucode_version);
1535 			adev->gfx.mec2_feature_version =
1536 			le32_to_cpu(cp_hdr->ucode_feature_version);
1537 		} else {
1538 			err = 0;
1539 			adev->gfx.mec2_fw = NULL;
1540 		}
1541 	} else {
1542 		adev->gfx.mec2_fw_version = adev->gfx.mec_fw_version;
1543 		adev->gfx.mec2_feature_version = adev->gfx.mec_feature_version;
1544 	}
1545 
1546 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1547 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
1548 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
1549 		info->fw = adev->gfx.mec_fw;
1550 		header = (const struct common_firmware_header *)info->fw->data;
1551 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)info->fw->data;
1552 		adev->firmware.fw_size +=
1553 			ALIGN(le32_to_cpu(header->ucode_size_bytes) - le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
1554 
1555 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1_JT];
1556 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1_JT;
1557 		info->fw = adev->gfx.mec_fw;
1558 		adev->firmware.fw_size +=
1559 			ALIGN(le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
1560 
1561 		if (adev->gfx.mec2_fw) {
1562 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
1563 			info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
1564 			info->fw = adev->gfx.mec2_fw;
1565 			header = (const struct common_firmware_header *)info->fw->data;
1566 			cp_hdr = (const struct gfx_firmware_header_v1_0 *)info->fw->data;
1567 			adev->firmware.fw_size +=
1568 				ALIGN(le32_to_cpu(header->ucode_size_bytes) - le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
1569 
1570 			/* TODO: Determine if MEC2 JT FW loading can be removed
1571 				 for all GFX V9 asic and above */
1572 			if (gfx_v9_0_load_mec2_fw_bin_support(adev)) {
1573 				info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2_JT];
1574 				info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2_JT;
1575 				info->fw = adev->gfx.mec2_fw;
1576 				adev->firmware.fw_size +=
1577 					ALIGN(le32_to_cpu(cp_hdr->jt_size) * 4,
1578 					PAGE_SIZE);
1579 			}
1580 		}
1581 	}
1582 
1583 out:
1584 	gfx_v9_0_check_if_need_gfxoff(adev);
1585 	gfx_v9_0_check_fw_write_wait(adev);
1586 	if (err) {
1587 		dev_err(adev->dev,
1588 			"gfx9: Failed to load firmware \"%s\"\n",
1589 			fw_name);
1590 		release_firmware(adev->gfx.mec_fw);
1591 		adev->gfx.mec_fw = NULL;
1592 		release_firmware(adev->gfx.mec2_fw);
1593 		adev->gfx.mec2_fw = NULL;
1594 	}
1595 	return err;
1596 }
1597 
1598 static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
1599 {
1600 	const char *chip_name;
1601 	int r;
1602 
1603 	DRM_DEBUG("\n");
1604 
1605 	switch (adev->ip_versions[GC_HWIP][0]) {
1606 	case IP_VERSION(9, 0, 1):
1607 		chip_name = "vega10";
1608 		break;
1609 	case IP_VERSION(9, 2, 1):
1610 		chip_name = "vega12";
1611 		break;
1612 	case IP_VERSION(9, 4, 0):
1613 		chip_name = "vega20";
1614 		break;
1615 	case IP_VERSION(9, 2, 2):
1616 	case IP_VERSION(9, 1, 0):
1617 		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
1618 			chip_name = "raven2";
1619 		else if (adev->apu_flags & AMD_APU_IS_PICASSO)
1620 			chip_name = "picasso";
1621 		else
1622 			chip_name = "raven";
1623 		break;
1624 	case IP_VERSION(9, 4, 1):
1625 		chip_name = "arcturus";
1626 		break;
1627 	case IP_VERSION(9, 3, 0):
1628 		if (adev->apu_flags & AMD_APU_IS_RENOIR)
1629 			chip_name = "renoir";
1630 		else
1631 			chip_name = "green_sardine";
1632 		break;
1633 	case IP_VERSION(9, 4, 2):
1634 		chip_name = "aldebaran";
1635 		break;
1636 	default:
1637 		BUG();
1638 	}
1639 
1640 	/* No CPG in Arcturus */
1641 	if (adev->gfx.num_gfx_rings) {
1642 		r = gfx_v9_0_init_cp_gfx_microcode(adev, chip_name);
1643 		if (r)
1644 			return r;
1645 	}
1646 
1647 	r = gfx_v9_0_init_rlc_microcode(adev, chip_name);
1648 	if (r)
1649 		return r;
1650 
1651 	r = gfx_v9_0_init_cp_compute_microcode(adev, chip_name);
1652 	if (r)
1653 		return r;
1654 
1655 	return r;
1656 }
1657 
1658 static u32 gfx_v9_0_get_csb_size(struct amdgpu_device *adev)
1659 {
1660 	u32 count = 0;
1661 	const struct cs_section_def *sect = NULL;
1662 	const struct cs_extent_def *ext = NULL;
1663 
1664 	/* begin clear state */
1665 	count += 2;
1666 	/* context control state */
1667 	count += 3;
1668 
1669 	for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
1670 		for (ext = sect->section; ext->extent != NULL; ++ext) {
1671 			if (sect->id == SECT_CONTEXT)
1672 				count += 2 + ext->reg_count;
1673 			else
1674 				return 0;
1675 		}
1676 	}
1677 
1678 	/* end clear state */
1679 	count += 2;
1680 	/* clear state */
1681 	count += 2;
1682 
1683 	return count;
1684 }
1685 
1686 static void gfx_v9_0_get_csb_buffer(struct amdgpu_device *adev,
1687 				    volatile u32 *buffer)
1688 {
1689 	u32 count = 0, i;
1690 	const struct cs_section_def *sect = NULL;
1691 	const struct cs_extent_def *ext = NULL;
1692 
1693 	if (adev->gfx.rlc.cs_data == NULL)
1694 		return;
1695 	if (buffer == NULL)
1696 		return;
1697 
1698 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1699 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1700 
1701 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1702 	buffer[count++] = cpu_to_le32(0x80000000);
1703 	buffer[count++] = cpu_to_le32(0x80000000);
1704 
1705 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1706 		for (ext = sect->section; ext->extent != NULL; ++ext) {
1707 			if (sect->id == SECT_CONTEXT) {
1708 				buffer[count++] =
1709 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1710 				buffer[count++] = cpu_to_le32(ext->reg_index -
1711 						PACKET3_SET_CONTEXT_REG_START);
1712 				for (i = 0; i < ext->reg_count; i++)
1713 					buffer[count++] = cpu_to_le32(ext->extent[i]);
1714 			} else {
1715 				return;
1716 			}
1717 		}
1718 	}
1719 
1720 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1721 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1722 
1723 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1724 	buffer[count++] = cpu_to_le32(0);
1725 }
1726 
1727 static void gfx_v9_0_init_always_on_cu_mask(struct amdgpu_device *adev)
1728 {
1729 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
1730 	uint32_t pg_always_on_cu_num = 2;
1731 	uint32_t always_on_cu_num;
1732 	uint32_t i, j, k;
1733 	uint32_t mask, cu_bitmap, counter;
1734 
1735 	if (adev->flags & AMD_IS_APU)
1736 		always_on_cu_num = 4;
1737 	else if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 2, 1))
1738 		always_on_cu_num = 8;
1739 	else
1740 		always_on_cu_num = 12;
1741 
1742 	mutex_lock(&adev->grbm_idx_mutex);
1743 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1744 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1745 			mask = 1;
1746 			cu_bitmap = 0;
1747 			counter = 0;
1748 			gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
1749 
1750 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
1751 				if (cu_info->bitmap[i][j] & mask) {
1752 					if (counter == pg_always_on_cu_num)
1753 						WREG32_SOC15(GC, 0, mmRLC_PG_ALWAYS_ON_CU_MASK, cu_bitmap);
1754 					if (counter < always_on_cu_num)
1755 						cu_bitmap |= mask;
1756 					else
1757 						break;
1758 					counter++;
1759 				}
1760 				mask <<= 1;
1761 			}
1762 
1763 			WREG32_SOC15(GC, 0, mmRLC_LB_ALWAYS_ACTIVE_CU_MASK, cu_bitmap);
1764 			cu_info->ao_cu_bitmap[i][j] = cu_bitmap;
1765 		}
1766 	}
1767 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1768 	mutex_unlock(&adev->grbm_idx_mutex);
1769 }
1770 
1771 static void gfx_v9_0_init_lbpw(struct amdgpu_device *adev)
1772 {
1773 	uint32_t data;
1774 
1775 	/* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1776 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F);
1777 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x0333A5A7);
1778 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077);
1779 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x30 | 0x40 << 8 | 0x02FA << 16));
1780 
1781 	/* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1782 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000);
1783 
1784 	/* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1785 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000500);
1786 
1787 	mutex_lock(&adev->grbm_idx_mutex);
1788 	/* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1789 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1790 	WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff);
1791 
1792 	/* set mmRLC_LB_PARAMS = 0x003F_1006 */
1793 	data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003);
1794 	data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010);
1795 	data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F);
1796 	WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data);
1797 
1798 	/* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1799 	data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7);
1800 	data &= 0x0000FFFF;
1801 	data |= 0x00C00000;
1802 	WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data);
1803 
1804 	/*
1805 	 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xF (4 CUs AON for Raven),
1806 	 * programmed in gfx_v9_0_init_always_on_cu_mask()
1807 	 */
1808 
1809 	/* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1810 	 * but used for RLC_LB_CNTL configuration */
1811 	data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK;
1812 	data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09);
1813 	data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000);
1814 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data);
1815 	mutex_unlock(&adev->grbm_idx_mutex);
1816 
1817 	gfx_v9_0_init_always_on_cu_mask(adev);
1818 }
1819 
1820 static void gfx_v9_4_init_lbpw(struct amdgpu_device *adev)
1821 {
1822 	uint32_t data;
1823 
1824 	/* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1825 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F);
1826 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x033388F8);
1827 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077);
1828 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x10 | 0x27 << 8 | 0x02FA << 16));
1829 
1830 	/* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1831 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000);
1832 
1833 	/* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1834 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000800);
1835 
1836 	mutex_lock(&adev->grbm_idx_mutex);
1837 	/* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1838 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1839 	WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff);
1840 
1841 	/* set mmRLC_LB_PARAMS = 0x003F_1006 */
1842 	data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003);
1843 	data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010);
1844 	data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F);
1845 	WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data);
1846 
1847 	/* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1848 	data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7);
1849 	data &= 0x0000FFFF;
1850 	data |= 0x00C00000;
1851 	WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data);
1852 
1853 	/*
1854 	 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xFFF (12 CUs AON),
1855 	 * programmed in gfx_v9_0_init_always_on_cu_mask()
1856 	 */
1857 
1858 	/* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1859 	 * but used for RLC_LB_CNTL configuration */
1860 	data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK;
1861 	data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09);
1862 	data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000);
1863 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data);
1864 	mutex_unlock(&adev->grbm_idx_mutex);
1865 
1866 	gfx_v9_0_init_always_on_cu_mask(adev);
1867 }
1868 
1869 static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
1870 {
1871 	WREG32_FIELD15(GC, 0, RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0);
1872 }
1873 
1874 static int gfx_v9_0_cp_jump_table_num(struct amdgpu_device *adev)
1875 {
1876 	if (gfx_v9_0_load_mec2_fw_bin_support(adev))
1877 		return 5;
1878 	else
1879 		return 4;
1880 }
1881 
1882 static void gfx_v9_0_init_rlcg_reg_access_ctrl(struct amdgpu_device *adev)
1883 {
1884 	struct amdgpu_rlcg_reg_access_ctrl *reg_access_ctrl;
1885 
1886 	reg_access_ctrl = &adev->gfx.rlc.reg_access_ctrl;
1887 	reg_access_ctrl->scratch_reg0 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0);
1888 	reg_access_ctrl->scratch_reg1 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG1);
1889 	reg_access_ctrl->scratch_reg2 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG2);
1890 	reg_access_ctrl->scratch_reg3 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG3);
1891 	reg_access_ctrl->grbm_cntl = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_CNTL);
1892 	reg_access_ctrl->grbm_idx = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_INDEX);
1893 	reg_access_ctrl->spare_int = SOC15_REG_OFFSET(GC, 0, mmRLC_SPARE_INT);
1894 	adev->gfx.rlc.rlcg_reg_access_supported = true;
1895 }
1896 
1897 static int gfx_v9_0_rlc_init(struct amdgpu_device *adev)
1898 {
1899 	const struct cs_section_def *cs_data;
1900 	int r;
1901 
1902 	adev->gfx.rlc.cs_data = gfx9_cs_data;
1903 
1904 	cs_data = adev->gfx.rlc.cs_data;
1905 
1906 	if (cs_data) {
1907 		/* init clear state block */
1908 		r = amdgpu_gfx_rlc_init_csb(adev);
1909 		if (r)
1910 			return r;
1911 	}
1912 
1913 	if (adev->flags & AMD_IS_APU) {
1914 		/* TODO: double check the cp_table_size for RV */
1915 		adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1916 		r = amdgpu_gfx_rlc_init_cpt(adev);
1917 		if (r)
1918 			return r;
1919 	}
1920 
1921 	switch (adev->ip_versions[GC_HWIP][0]) {
1922 	case IP_VERSION(9, 2, 2):
1923 	case IP_VERSION(9, 1, 0):
1924 		gfx_v9_0_init_lbpw(adev);
1925 		break;
1926 	case IP_VERSION(9, 4, 0):
1927 		gfx_v9_4_init_lbpw(adev);
1928 		break;
1929 	default:
1930 		break;
1931 	}
1932 
1933 	/* init spm vmid with 0xf */
1934 	if (adev->gfx.rlc.funcs->update_spm_vmid)
1935 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
1936 
1937 	return 0;
1938 }
1939 
1940 static void gfx_v9_0_mec_fini(struct amdgpu_device *adev)
1941 {
1942 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1943 	amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL);
1944 }
1945 
1946 static int gfx_v9_0_mec_init(struct amdgpu_device *adev)
1947 {
1948 	int r;
1949 	u32 *hpd;
1950 	const __le32 *fw_data;
1951 	unsigned fw_size;
1952 	u32 *fw;
1953 	size_t mec_hpd_size;
1954 
1955 	const struct gfx_firmware_header_v1_0 *mec_hdr;
1956 
1957 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1958 
1959 	/* take ownership of the relevant compute queues */
1960 	amdgpu_gfx_compute_queue_acquire(adev);
1961 	mec_hpd_size = adev->gfx.num_compute_rings * GFX9_MEC_HPD_SIZE;
1962 	if (mec_hpd_size) {
1963 		r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1964 					      AMDGPU_GEM_DOMAIN_VRAM,
1965 					      &adev->gfx.mec.hpd_eop_obj,
1966 					      &adev->gfx.mec.hpd_eop_gpu_addr,
1967 					      (void **)&hpd);
1968 		if (r) {
1969 			dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1970 			gfx_v9_0_mec_fini(adev);
1971 			return r;
1972 		}
1973 
1974 		memset(hpd, 0, mec_hpd_size);
1975 
1976 		amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1977 		amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1978 	}
1979 
1980 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1981 
1982 	fw_data = (const __le32 *)
1983 		(adev->gfx.mec_fw->data +
1984 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
1985 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes);
1986 
1987 	r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
1988 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
1989 				      &adev->gfx.mec.mec_fw_obj,
1990 				      &adev->gfx.mec.mec_fw_gpu_addr,
1991 				      (void **)&fw);
1992 	if (r) {
1993 		dev_warn(adev->dev, "(%d) create mec firmware bo failed\n", r);
1994 		gfx_v9_0_mec_fini(adev);
1995 		return r;
1996 	}
1997 
1998 	memcpy(fw, fw_data, fw_size);
1999 
2000 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
2001 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
2002 
2003 	return 0;
2004 }
2005 
2006 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
2007 {
2008 	WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX,
2009 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
2010 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
2011 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
2012 		(SQ_IND_INDEX__FORCE_READ_MASK));
2013 	return RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
2014 }
2015 
2016 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
2017 			   uint32_t wave, uint32_t thread,
2018 			   uint32_t regno, uint32_t num, uint32_t *out)
2019 {
2020 	WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX,
2021 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
2022 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
2023 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
2024 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
2025 		(SQ_IND_INDEX__FORCE_READ_MASK) |
2026 		(SQ_IND_INDEX__AUTO_INCR_MASK));
2027 	while (num--)
2028 		*(out++) = RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
2029 }
2030 
2031 static void gfx_v9_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
2032 {
2033 	/* type 1 wave data */
2034 	dst[(*no_fields)++] = 1;
2035 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
2036 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
2037 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
2038 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
2039 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
2040 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
2041 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
2042 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
2043 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
2044 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
2045 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
2046 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
2047 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
2048 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
2049 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE);
2050 }
2051 
2052 static void gfx_v9_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
2053 				     uint32_t wave, uint32_t start,
2054 				     uint32_t size, uint32_t *dst)
2055 {
2056 	wave_read_regs(
2057 		adev, simd, wave, 0,
2058 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
2059 }
2060 
2061 static void gfx_v9_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t simd,
2062 				     uint32_t wave, uint32_t thread,
2063 				     uint32_t start, uint32_t size,
2064 				     uint32_t *dst)
2065 {
2066 	wave_read_regs(
2067 		adev, simd, wave, thread,
2068 		start + SQIND_WAVE_VGPRS_OFFSET, size, dst);
2069 }
2070 
2071 static void gfx_v9_0_select_me_pipe_q(struct amdgpu_device *adev,
2072 				  u32 me, u32 pipe, u32 q, u32 vm)
2073 {
2074 	soc15_grbm_select(adev, me, pipe, q, vm);
2075 }
2076 
2077 static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = {
2078         .get_gpu_clock_counter = &gfx_v9_0_get_gpu_clock_counter,
2079         .select_se_sh = &gfx_v9_0_select_se_sh,
2080         .read_wave_data = &gfx_v9_0_read_wave_data,
2081         .read_wave_sgprs = &gfx_v9_0_read_wave_sgprs,
2082         .read_wave_vgprs = &gfx_v9_0_read_wave_vgprs,
2083         .select_me_pipe_q = &gfx_v9_0_select_me_pipe_q,
2084 };
2085 
2086 const struct amdgpu_ras_block_hw_ops  gfx_v9_0_ras_ops = {
2087 		.ras_error_inject = &gfx_v9_0_ras_error_inject,
2088 		.query_ras_error_count = &gfx_v9_0_query_ras_error_count,
2089 		.reset_ras_error_count = &gfx_v9_0_reset_ras_error_count,
2090 };
2091 
2092 static struct amdgpu_gfx_ras gfx_v9_0_ras = {
2093 	.ras_block = {
2094 		.hw_ops = &gfx_v9_0_ras_ops,
2095 	},
2096 };
2097 
2098 static int gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
2099 {
2100 	u32 gb_addr_config;
2101 	int err;
2102 
2103 	adev->gfx.funcs = &gfx_v9_0_gfx_funcs;
2104 
2105 	switch (adev->ip_versions[GC_HWIP][0]) {
2106 	case IP_VERSION(9, 0, 1):
2107 		adev->gfx.config.max_hw_contexts = 8;
2108 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2109 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2110 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2111 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2112 		gb_addr_config = VEGA10_GB_ADDR_CONFIG_GOLDEN;
2113 		break;
2114 	case IP_VERSION(9, 2, 1):
2115 		adev->gfx.config.max_hw_contexts = 8;
2116 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2117 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2118 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2119 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2120 		gb_addr_config = VEGA12_GB_ADDR_CONFIG_GOLDEN;
2121 		DRM_INFO("fix gfx.config for vega12\n");
2122 		break;
2123 	case IP_VERSION(9, 4, 0):
2124 		adev->gfx.ras = &gfx_v9_0_ras;
2125 		adev->gfx.config.max_hw_contexts = 8;
2126 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2127 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2128 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2129 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2130 		gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2131 		gb_addr_config &= ~0xf3e777ff;
2132 		gb_addr_config |= 0x22014042;
2133 		/* check vbios table if gpu info is not available */
2134 		err = amdgpu_atomfirmware_get_gfx_info(adev);
2135 		if (err)
2136 			return err;
2137 		break;
2138 	case IP_VERSION(9, 2, 2):
2139 	case IP_VERSION(9, 1, 0):
2140 		adev->gfx.config.max_hw_contexts = 8;
2141 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2142 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2143 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2144 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2145 		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
2146 			gb_addr_config = RAVEN2_GB_ADDR_CONFIG_GOLDEN;
2147 		else
2148 			gb_addr_config = RAVEN_GB_ADDR_CONFIG_GOLDEN;
2149 		break;
2150 	case IP_VERSION(9, 4, 1):
2151 		adev->gfx.ras = &gfx_v9_4_ras;
2152 		adev->gfx.config.max_hw_contexts = 8;
2153 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2154 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2155 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2156 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2157 		gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2158 		gb_addr_config &= ~0xf3e777ff;
2159 		gb_addr_config |= 0x22014042;
2160 		break;
2161 	case IP_VERSION(9, 3, 0):
2162 		adev->gfx.config.max_hw_contexts = 8;
2163 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2164 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2165 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x80;
2166 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2167 		gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2168 		gb_addr_config &= ~0xf3e777ff;
2169 		gb_addr_config |= 0x22010042;
2170 		break;
2171 	case IP_VERSION(9, 4, 2):
2172 		adev->gfx.ras = &gfx_v9_4_2_ras;
2173 		adev->gfx.config.max_hw_contexts = 8;
2174 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2175 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2176 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2177 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2178 		gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2179 		gb_addr_config &= ~0xf3e777ff;
2180 		gb_addr_config |= 0x22014042;
2181 		/* check vbios table if gpu info is not available */
2182 		err = amdgpu_atomfirmware_get_gfx_info(adev);
2183 		if (err)
2184 			return err;
2185 		break;
2186 	default:
2187 		BUG();
2188 		break;
2189 	}
2190 
2191 	if (adev->gfx.ras) {
2192 		err = amdgpu_ras_register_ras_block(adev, &adev->gfx.ras->ras_block);
2193 		if (err) {
2194 			DRM_ERROR("Failed to register gfx ras block!\n");
2195 			return err;
2196 		}
2197 
2198 		strcpy(adev->gfx.ras->ras_block.ras_comm.name, "gfx");
2199 		adev->gfx.ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__GFX;
2200 		adev->gfx.ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
2201 		adev->gfx.ras_if = &adev->gfx.ras->ras_block.ras_comm;
2202 
2203 		/* If not define special ras_late_init function, use gfx default ras_late_init */
2204 		if (!adev->gfx.ras->ras_block.ras_late_init)
2205 			adev->gfx.ras->ras_block.ras_late_init = amdgpu_gfx_ras_late_init;
2206 
2207 		/* If not defined special ras_cb function, use default ras_cb */
2208 		if (!adev->gfx.ras->ras_block.ras_cb)
2209 			adev->gfx.ras->ras_block.ras_cb = amdgpu_gfx_process_ras_data_cb;
2210 	}
2211 
2212 	adev->gfx.config.gb_addr_config = gb_addr_config;
2213 
2214 	adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
2215 			REG_GET_FIELD(
2216 					adev->gfx.config.gb_addr_config,
2217 					GB_ADDR_CONFIG,
2218 					NUM_PIPES);
2219 
2220 	adev->gfx.config.max_tile_pipes =
2221 		adev->gfx.config.gb_addr_config_fields.num_pipes;
2222 
2223 	adev->gfx.config.gb_addr_config_fields.num_banks = 1 <<
2224 			REG_GET_FIELD(
2225 					adev->gfx.config.gb_addr_config,
2226 					GB_ADDR_CONFIG,
2227 					NUM_BANKS);
2228 	adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
2229 			REG_GET_FIELD(
2230 					adev->gfx.config.gb_addr_config,
2231 					GB_ADDR_CONFIG,
2232 					MAX_COMPRESSED_FRAGS);
2233 	adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
2234 			REG_GET_FIELD(
2235 					adev->gfx.config.gb_addr_config,
2236 					GB_ADDR_CONFIG,
2237 					NUM_RB_PER_SE);
2238 	adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
2239 			REG_GET_FIELD(
2240 					adev->gfx.config.gb_addr_config,
2241 					GB_ADDR_CONFIG,
2242 					NUM_SHADER_ENGINES);
2243 	adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
2244 			REG_GET_FIELD(
2245 					adev->gfx.config.gb_addr_config,
2246 					GB_ADDR_CONFIG,
2247 					PIPE_INTERLEAVE_SIZE));
2248 
2249 	return 0;
2250 }
2251 
2252 static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
2253 				      int mec, int pipe, int queue)
2254 {
2255 	unsigned irq_type;
2256 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
2257 	unsigned int hw_prio;
2258 
2259 	ring = &adev->gfx.compute_ring[ring_id];
2260 
2261 	/* mec0 is me1 */
2262 	ring->me = mec + 1;
2263 	ring->pipe = pipe;
2264 	ring->queue = queue;
2265 
2266 	ring->ring_obj = NULL;
2267 	ring->use_doorbell = true;
2268 	ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1;
2269 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
2270 				+ (ring_id * GFX9_MEC_HPD_SIZE);
2271 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
2272 
2273 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
2274 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
2275 		+ ring->pipe;
2276 	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
2277 			AMDGPU_GFX_PIPE_PRIO_HIGH : AMDGPU_GFX_PIPE_PRIO_NORMAL;
2278 	/* type-2 packets are deprecated on MEC, use type-3 instead */
2279 	return amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
2280 				hw_prio, NULL);
2281 }
2282 
2283 static int gfx_v9_0_sw_init(void *handle)
2284 {
2285 	int i, j, k, r, ring_id;
2286 	struct amdgpu_ring *ring;
2287 	struct amdgpu_kiq *kiq;
2288 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2289 
2290 	switch (adev->ip_versions[GC_HWIP][0]) {
2291 	case IP_VERSION(9, 0, 1):
2292 	case IP_VERSION(9, 2, 1):
2293 	case IP_VERSION(9, 4, 0):
2294 	case IP_VERSION(9, 2, 2):
2295 	case IP_VERSION(9, 1, 0):
2296 	case IP_VERSION(9, 4, 1):
2297 	case IP_VERSION(9, 3, 0):
2298 	case IP_VERSION(9, 4, 2):
2299 		adev->gfx.mec.num_mec = 2;
2300 		break;
2301 	default:
2302 		adev->gfx.mec.num_mec = 1;
2303 		break;
2304 	}
2305 
2306 	adev->gfx.mec.num_pipe_per_mec = 4;
2307 	adev->gfx.mec.num_queue_per_pipe = 8;
2308 
2309 	/* EOP Event */
2310 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_EOP_INTERRUPT, &adev->gfx.eop_irq);
2311 	if (r)
2312 		return r;
2313 
2314 	/* Privileged reg */
2315 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_REG_FAULT,
2316 			      &adev->gfx.priv_reg_irq);
2317 	if (r)
2318 		return r;
2319 
2320 	/* Privileged inst */
2321 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_INSTR_FAULT,
2322 			      &adev->gfx.priv_inst_irq);
2323 	if (r)
2324 		return r;
2325 
2326 	/* ECC error */
2327 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_ECC_ERROR,
2328 			      &adev->gfx.cp_ecc_error_irq);
2329 	if (r)
2330 		return r;
2331 
2332 	/* FUE error */
2333 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_FUE_ERROR,
2334 			      &adev->gfx.cp_ecc_error_irq);
2335 	if (r)
2336 		return r;
2337 
2338 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
2339 
2340 	gfx_v9_0_scratch_init(adev);
2341 
2342 	r = gfx_v9_0_init_microcode(adev);
2343 	if (r) {
2344 		DRM_ERROR("Failed to load gfx firmware!\n");
2345 		return r;
2346 	}
2347 
2348 	if (adev->gfx.rlc.funcs) {
2349 		if (adev->gfx.rlc.funcs->init) {
2350 			r = adev->gfx.rlc.funcs->init(adev);
2351 			if (r) {
2352 				dev_err(adev->dev, "Failed to init rlc BOs!\n");
2353 				return r;
2354 			}
2355 		}
2356 	}
2357 
2358 	r = gfx_v9_0_mec_init(adev);
2359 	if (r) {
2360 		DRM_ERROR("Failed to init MEC BOs!\n");
2361 		return r;
2362 	}
2363 
2364 	/* set up the gfx ring */
2365 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
2366 		ring = &adev->gfx.gfx_ring[i];
2367 		ring->ring_obj = NULL;
2368 		if (!i)
2369 			sprintf(ring->name, "gfx");
2370 		else
2371 			sprintf(ring->name, "gfx_%d", i);
2372 		ring->use_doorbell = true;
2373 		ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
2374 		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
2375 				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
2376 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
2377 		if (r)
2378 			return r;
2379 	}
2380 
2381 	/* set up the compute queues - allocate horizontally across pipes */
2382 	ring_id = 0;
2383 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
2384 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2385 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2386 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
2387 					continue;
2388 
2389 				r = gfx_v9_0_compute_ring_init(adev,
2390 							       ring_id,
2391 							       i, k, j);
2392 				if (r)
2393 					return r;
2394 
2395 				ring_id++;
2396 			}
2397 		}
2398 	}
2399 
2400 	r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE);
2401 	if (r) {
2402 		DRM_ERROR("Failed to init KIQ BOs!\n");
2403 		return r;
2404 	}
2405 
2406 	kiq = &adev->gfx.kiq;
2407 	r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
2408 	if (r)
2409 		return r;
2410 
2411 	/* create MQD for all compute queues as wel as KIQ for SRIOV case */
2412 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v9_mqd_allocation));
2413 	if (r)
2414 		return r;
2415 
2416 	adev->gfx.ce_ram_size = 0x8000;
2417 
2418 	r = gfx_v9_0_gpu_early_init(adev);
2419 	if (r)
2420 		return r;
2421 
2422 	return 0;
2423 }
2424 
2425 
2426 static int gfx_v9_0_sw_fini(void *handle)
2427 {
2428 	int i;
2429 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2430 
2431 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2432 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2433 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
2434 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2435 
2436 	amdgpu_gfx_mqd_sw_fini(adev);
2437 	amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring);
2438 	amdgpu_gfx_kiq_fini(adev);
2439 
2440 	gfx_v9_0_mec_fini(adev);
2441 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
2442 				&adev->gfx.rlc.clear_state_gpu_addr,
2443 				(void **)&adev->gfx.rlc.cs_ptr);
2444 	if (adev->flags & AMD_IS_APU) {
2445 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2446 				&adev->gfx.rlc.cp_table_gpu_addr,
2447 				(void **)&adev->gfx.rlc.cp_table_ptr);
2448 	}
2449 	gfx_v9_0_free_microcode(adev);
2450 
2451 	return 0;
2452 }
2453 
2454 
2455 static void gfx_v9_0_tiling_mode_table_init(struct amdgpu_device *adev)
2456 {
2457 	/* TODO */
2458 }
2459 
2460 void gfx_v9_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num,
2461 			   u32 instance)
2462 {
2463 	u32 data;
2464 
2465 	if (instance == 0xffffffff)
2466 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
2467 	else
2468 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
2469 
2470 	if (se_num == 0xffffffff)
2471 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
2472 	else
2473 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
2474 
2475 	if (sh_num == 0xffffffff)
2476 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
2477 	else
2478 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
2479 
2480 	WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_INDEX, data);
2481 }
2482 
2483 static u32 gfx_v9_0_get_rb_active_bitmap(struct amdgpu_device *adev)
2484 {
2485 	u32 data, mask;
2486 
2487 	data = RREG32_SOC15(GC, 0, mmCC_RB_BACKEND_DISABLE);
2488 	data |= RREG32_SOC15(GC, 0, mmGC_USER_RB_BACKEND_DISABLE);
2489 
2490 	data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
2491 	data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
2492 
2493 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
2494 					 adev->gfx.config.max_sh_per_se);
2495 
2496 	return (~data) & mask;
2497 }
2498 
2499 static void gfx_v9_0_setup_rb(struct amdgpu_device *adev)
2500 {
2501 	int i, j;
2502 	u32 data;
2503 	u32 active_rbs = 0;
2504 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
2505 					adev->gfx.config.max_sh_per_se;
2506 
2507 	mutex_lock(&adev->grbm_idx_mutex);
2508 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2509 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2510 			gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
2511 			data = gfx_v9_0_get_rb_active_bitmap(adev);
2512 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
2513 					       rb_bitmap_width_per_sh);
2514 		}
2515 	}
2516 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
2517 	mutex_unlock(&adev->grbm_idx_mutex);
2518 
2519 	adev->gfx.config.backend_enable_mask = active_rbs;
2520 	adev->gfx.config.num_rbs = hweight32(active_rbs);
2521 }
2522 
2523 #define DEFAULT_SH_MEM_BASES	(0x6000)
2524 static void gfx_v9_0_init_compute_vmid(struct amdgpu_device *adev)
2525 {
2526 	int i;
2527 	uint32_t sh_mem_config;
2528 	uint32_t sh_mem_bases;
2529 
2530 	/*
2531 	 * Configure apertures:
2532 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
2533 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
2534 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
2535 	 */
2536 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
2537 
2538 	sh_mem_config = SH_MEM_ADDRESS_MODE_64 |
2539 			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
2540 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
2541 
2542 	mutex_lock(&adev->srbm_mutex);
2543 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
2544 		soc15_grbm_select(adev, 0, 0, 0, i);
2545 		/* CP and shaders */
2546 		WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, sh_mem_config);
2547 		WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, sh_mem_bases);
2548 	}
2549 	soc15_grbm_select(adev, 0, 0, 0, 0);
2550 	mutex_unlock(&adev->srbm_mutex);
2551 
2552 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
2553 	   acccess. These should be enabled by FW for target VMIDs. */
2554 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
2555 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * i, 0);
2556 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * i, 0);
2557 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, i, 0);
2558 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, i, 0);
2559 	}
2560 }
2561 
2562 static void gfx_v9_0_init_gds_vmid(struct amdgpu_device *adev)
2563 {
2564 	int vmid;
2565 
2566 	/*
2567 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
2568 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
2569 	 * the driver can enable them for graphics. VMID0 should maintain
2570 	 * access so that HWS firmware can save/restore entries.
2571 	 */
2572 	for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
2573 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * vmid, 0);
2574 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * vmid, 0);
2575 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, vmid, 0);
2576 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, vmid, 0);
2577 	}
2578 }
2579 
2580 static void gfx_v9_0_init_sq_config(struct amdgpu_device *adev)
2581 {
2582 	uint32_t tmp;
2583 
2584 	switch (adev->ip_versions[GC_HWIP][0]) {
2585 	case IP_VERSION(9, 4, 1):
2586 		tmp = RREG32_SOC15(GC, 0, mmSQ_CONFIG);
2587 		tmp = REG_SET_FIELD(tmp, SQ_CONFIG,
2588 					DISABLE_BARRIER_WAITCNT, 1);
2589 		WREG32_SOC15(GC, 0, mmSQ_CONFIG, tmp);
2590 		break;
2591 	default:
2592 		break;
2593 	}
2594 }
2595 
2596 static void gfx_v9_0_constants_init(struct amdgpu_device *adev)
2597 {
2598 	u32 tmp;
2599 	int i;
2600 
2601 	WREG32_FIELD15_RLC(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff);
2602 
2603 	gfx_v9_0_tiling_mode_table_init(adev);
2604 
2605 	gfx_v9_0_setup_rb(adev);
2606 	gfx_v9_0_get_cu_info(adev, &adev->gfx.cu_info);
2607 	adev->gfx.config.db_debug2 = RREG32_SOC15(GC, 0, mmDB_DEBUG2);
2608 
2609 	/* XXX SH_MEM regs */
2610 	/* where to put LDS, scratch, GPUVM in FSA64 space */
2611 	mutex_lock(&adev->srbm_mutex);
2612 	for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_0].num_ids; i++) {
2613 		soc15_grbm_select(adev, 0, 0, 0, i);
2614 		/* CP and shaders */
2615 		if (i == 0) {
2616 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
2617 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
2618 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
2619 					    !!adev->gmc.noretry);
2620 			WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp);
2621 			WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, 0);
2622 		} else {
2623 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
2624 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
2625 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
2626 					    !!adev->gmc.noretry);
2627 			WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp);
2628 			tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
2629 				(adev->gmc.private_aperture_start >> 48));
2630 			tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
2631 				(adev->gmc.shared_aperture_start >> 48));
2632 			WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, tmp);
2633 		}
2634 	}
2635 	soc15_grbm_select(adev, 0, 0, 0, 0);
2636 
2637 	mutex_unlock(&adev->srbm_mutex);
2638 
2639 	gfx_v9_0_init_compute_vmid(adev);
2640 	gfx_v9_0_init_gds_vmid(adev);
2641 	gfx_v9_0_init_sq_config(adev);
2642 }
2643 
2644 static void gfx_v9_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
2645 {
2646 	u32 i, j, k;
2647 	u32 mask;
2648 
2649 	mutex_lock(&adev->grbm_idx_mutex);
2650 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2651 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2652 			gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
2653 			for (k = 0; k < adev->usec_timeout; k++) {
2654 				if (RREG32_SOC15(GC, 0, mmRLC_SERDES_CU_MASTER_BUSY) == 0)
2655 					break;
2656 				udelay(1);
2657 			}
2658 			if (k == adev->usec_timeout) {
2659 				gfx_v9_0_select_se_sh(adev, 0xffffffff,
2660 						      0xffffffff, 0xffffffff);
2661 				mutex_unlock(&adev->grbm_idx_mutex);
2662 				DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
2663 					 i, j);
2664 				return;
2665 			}
2666 		}
2667 	}
2668 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
2669 	mutex_unlock(&adev->grbm_idx_mutex);
2670 
2671 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
2672 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
2673 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
2674 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
2675 	for (k = 0; k < adev->usec_timeout; k++) {
2676 		if ((RREG32_SOC15(GC, 0, mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
2677 			break;
2678 		udelay(1);
2679 	}
2680 }
2681 
2682 static void gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
2683 					       bool enable)
2684 {
2685 	u32 tmp;
2686 
2687 	/* These interrupts should be enabled to drive DS clock */
2688 
2689 	tmp= RREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0);
2690 
2691 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
2692 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
2693 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
2694 	if(adev->gfx.num_gfx_rings)
2695 		tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
2696 
2697 	WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp);
2698 }
2699 
2700 static void gfx_v9_0_init_csb(struct amdgpu_device *adev)
2701 {
2702 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
2703 	/* csib */
2704 	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_HI),
2705 			adev->gfx.rlc.clear_state_gpu_addr >> 32);
2706 	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_LO),
2707 			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
2708 	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_LENGTH),
2709 			adev->gfx.rlc.clear_state_size);
2710 }
2711 
2712 static void gfx_v9_1_parse_ind_reg_list(int *register_list_format,
2713 				int indirect_offset,
2714 				int list_size,
2715 				int *unique_indirect_regs,
2716 				int unique_indirect_reg_count,
2717 				int *indirect_start_offsets,
2718 				int *indirect_start_offsets_count,
2719 				int max_start_offsets_count)
2720 {
2721 	int idx;
2722 
2723 	for (; indirect_offset < list_size; indirect_offset++) {
2724 		WARN_ON(*indirect_start_offsets_count >= max_start_offsets_count);
2725 		indirect_start_offsets[*indirect_start_offsets_count] = indirect_offset;
2726 		*indirect_start_offsets_count = *indirect_start_offsets_count + 1;
2727 
2728 		while (register_list_format[indirect_offset] != 0xFFFFFFFF) {
2729 			indirect_offset += 2;
2730 
2731 			/* look for the matching indice */
2732 			for (idx = 0; idx < unique_indirect_reg_count; idx++) {
2733 				if (unique_indirect_regs[idx] ==
2734 					register_list_format[indirect_offset] ||
2735 					!unique_indirect_regs[idx])
2736 					break;
2737 			}
2738 
2739 			BUG_ON(idx >= unique_indirect_reg_count);
2740 
2741 			if (!unique_indirect_regs[idx])
2742 				unique_indirect_regs[idx] = register_list_format[indirect_offset];
2743 
2744 			indirect_offset++;
2745 		}
2746 	}
2747 }
2748 
2749 static int gfx_v9_1_init_rlc_save_restore_list(struct amdgpu_device *adev)
2750 {
2751 	int unique_indirect_regs[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2752 	int unique_indirect_reg_count = 0;
2753 
2754 	int indirect_start_offsets[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2755 	int indirect_start_offsets_count = 0;
2756 
2757 	int list_size = 0;
2758 	int i = 0, j = 0;
2759 	u32 tmp = 0;
2760 
2761 	u32 *register_list_format =
2762 		kmemdup(adev->gfx.rlc.register_list_format,
2763 			adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
2764 	if (!register_list_format)
2765 		return -ENOMEM;
2766 
2767 	/* setup unique_indirect_regs array and indirect_start_offsets array */
2768 	unique_indirect_reg_count = ARRAY_SIZE(unique_indirect_regs);
2769 	gfx_v9_1_parse_ind_reg_list(register_list_format,
2770 				    adev->gfx.rlc.reg_list_format_direct_reg_list_length,
2771 				    adev->gfx.rlc.reg_list_format_size_bytes >> 2,
2772 				    unique_indirect_regs,
2773 				    unique_indirect_reg_count,
2774 				    indirect_start_offsets,
2775 				    &indirect_start_offsets_count,
2776 				    ARRAY_SIZE(indirect_start_offsets));
2777 
2778 	/* enable auto inc in case it is disabled */
2779 	tmp = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL));
2780 	tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK;
2781 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL), tmp);
2782 
2783 	/* write register_restore table to offset 0x0 using RLC_SRM_ARAM_ADDR/DATA */
2784 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_ADDR),
2785 		RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET);
2786 	for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
2787 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_DATA),
2788 			adev->gfx.rlc.register_restore[i]);
2789 
2790 	/* load indirect register */
2791 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2792 		adev->gfx.rlc.reg_list_format_start);
2793 
2794 	/* direct register portion */
2795 	for (i = 0; i < adev->gfx.rlc.reg_list_format_direct_reg_list_length; i++)
2796 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
2797 			register_list_format[i]);
2798 
2799 	/* indirect register portion */
2800 	while (i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2)) {
2801 		if (register_list_format[i] == 0xFFFFFFFF) {
2802 			WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2803 			continue;
2804 		}
2805 
2806 		WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2807 		WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2808 
2809 		for (j = 0; j < unique_indirect_reg_count; j++) {
2810 			if (register_list_format[i] == unique_indirect_regs[j]) {
2811 				WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, j);
2812 				break;
2813 			}
2814 		}
2815 
2816 		BUG_ON(j >= unique_indirect_reg_count);
2817 
2818 		i++;
2819 	}
2820 
2821 	/* set save/restore list size */
2822 	list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
2823 	list_size = list_size >> 1;
2824 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2825 		adev->gfx.rlc.reg_restore_list_size);
2826 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), list_size);
2827 
2828 	/* write the starting offsets to RLC scratch ram */
2829 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2830 		adev->gfx.rlc.starting_offsets_start);
2831 	for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
2832 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
2833 		       indirect_start_offsets[i]);
2834 
2835 	/* load unique indirect regs*/
2836 	for (i = 0; i < ARRAY_SIZE(unique_indirect_regs); i++) {
2837 		if (unique_indirect_regs[i] != 0) {
2838 			WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_ADDR_0)
2839 			       + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i],
2840 			       unique_indirect_regs[i] & 0x3FFFF);
2841 
2842 			WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_DATA_0)
2843 			       + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i],
2844 			       unique_indirect_regs[i] >> 20);
2845 		}
2846 	}
2847 
2848 	kfree(register_list_format);
2849 	return 0;
2850 }
2851 
2852 static void gfx_v9_0_enable_save_restore_machine(struct amdgpu_device *adev)
2853 {
2854 	WREG32_FIELD15(GC, 0, RLC_SRM_CNTL, SRM_ENABLE, 1);
2855 }
2856 
2857 static void pwr_10_0_gfxip_control_over_cgpg(struct amdgpu_device *adev,
2858 					     bool enable)
2859 {
2860 	uint32_t data = 0;
2861 	uint32_t default_data = 0;
2862 
2863 	default_data = data = RREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS));
2864 	if (enable) {
2865 		/* enable GFXIP control over CGPG */
2866 		data |= PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
2867 		if(default_data != data)
2868 			WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2869 
2870 		/* update status */
2871 		data &= ~PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK;
2872 		data |= (2 << PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT);
2873 		if(default_data != data)
2874 			WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2875 	} else {
2876 		/* restore GFXIP control over GCPG */
2877 		data &= ~PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
2878 		if(default_data != data)
2879 			WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2880 	}
2881 }
2882 
2883 static void gfx_v9_0_init_gfx_power_gating(struct amdgpu_device *adev)
2884 {
2885 	uint32_t data = 0;
2886 
2887 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
2888 			      AMD_PG_SUPPORT_GFX_SMG |
2889 			      AMD_PG_SUPPORT_GFX_DMG)) {
2890 		/* init IDLE_POLL_COUNT = 60 */
2891 		data = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL));
2892 		data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
2893 		data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
2894 		WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL), data);
2895 
2896 		/* init RLC PG Delay */
2897 		data = 0;
2898 		data |= (0x10 << RLC_PG_DELAY__POWER_UP_DELAY__SHIFT);
2899 		data |= (0x10 << RLC_PG_DELAY__POWER_DOWN_DELAY__SHIFT);
2900 		data |= (0x10 << RLC_PG_DELAY__CMD_PROPAGATE_DELAY__SHIFT);
2901 		data |= (0x40 << RLC_PG_DELAY__MEM_SLEEP_DELAY__SHIFT);
2902 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY), data);
2903 
2904 		data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2));
2905 		data &= ~RLC_PG_DELAY_2__SERDES_CMD_DELAY_MASK;
2906 		data |= (0x4 << RLC_PG_DELAY_2__SERDES_CMD_DELAY__SHIFT);
2907 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2), data);
2908 
2909 		data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3));
2910 		data &= ~RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG_MASK;
2911 		data |= (0xff << RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG__SHIFT);
2912 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3), data);
2913 
2914 		data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL));
2915 		data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
2916 
2917 		/* program GRBM_REG_SAVE_GFX_IDLE_THRESHOLD to 0x55f0 */
2918 		data |= (0x55f0 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
2919 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL), data);
2920 		if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 3, 0))
2921 			pwr_10_0_gfxip_control_over_cgpg(adev, true);
2922 	}
2923 }
2924 
2925 static void gfx_v9_0_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
2926 						bool enable)
2927 {
2928 	uint32_t data = 0;
2929 	uint32_t default_data = 0;
2930 
2931 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2932 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2933 			     SMU_CLK_SLOWDOWN_ON_PU_ENABLE,
2934 			     enable ? 1 : 0);
2935 	if (default_data != data)
2936 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2937 }
2938 
2939 static void gfx_v9_0_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
2940 						bool enable)
2941 {
2942 	uint32_t data = 0;
2943 	uint32_t default_data = 0;
2944 
2945 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2946 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2947 			     SMU_CLK_SLOWDOWN_ON_PD_ENABLE,
2948 			     enable ? 1 : 0);
2949 	if(default_data != data)
2950 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2951 }
2952 
2953 static void gfx_v9_0_enable_cp_power_gating(struct amdgpu_device *adev,
2954 					bool enable)
2955 {
2956 	uint32_t data = 0;
2957 	uint32_t default_data = 0;
2958 
2959 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2960 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2961 			     CP_PG_DISABLE,
2962 			     enable ? 0 : 1);
2963 	if(default_data != data)
2964 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2965 }
2966 
2967 static void gfx_v9_0_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
2968 						bool enable)
2969 {
2970 	uint32_t data, default_data;
2971 
2972 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2973 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2974 			     GFX_POWER_GATING_ENABLE,
2975 			     enable ? 1 : 0);
2976 	if(default_data != data)
2977 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2978 }
2979 
2980 static void gfx_v9_0_enable_gfx_pipeline_powergating(struct amdgpu_device *adev,
2981 						bool enable)
2982 {
2983 	uint32_t data, default_data;
2984 
2985 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2986 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2987 			     GFX_PIPELINE_PG_ENABLE,
2988 			     enable ? 1 : 0);
2989 	if(default_data != data)
2990 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2991 
2992 	if (!enable)
2993 		/* read any GFX register to wake up GFX */
2994 		data = RREG32(SOC15_REG_OFFSET(GC, 0, mmDB_RENDER_CONTROL));
2995 }
2996 
2997 static void gfx_v9_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
2998 						       bool enable)
2999 {
3000 	uint32_t data, default_data;
3001 
3002 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
3003 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
3004 			     STATIC_PER_CU_PG_ENABLE,
3005 			     enable ? 1 : 0);
3006 	if(default_data != data)
3007 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
3008 }
3009 
3010 static void gfx_v9_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
3011 						bool enable)
3012 {
3013 	uint32_t data, default_data;
3014 
3015 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
3016 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
3017 			     DYN_PER_CU_PG_ENABLE,
3018 			     enable ? 1 : 0);
3019 	if(default_data != data)
3020 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
3021 }
3022 
3023 static void gfx_v9_0_init_pg(struct amdgpu_device *adev)
3024 {
3025 	gfx_v9_0_init_csb(adev);
3026 
3027 	/*
3028 	 * Rlc save restore list is workable since v2_1.
3029 	 * And it's needed by gfxoff feature.
3030 	 */
3031 	if (adev->gfx.rlc.is_rlc_v2_1) {
3032 		if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 2, 1) ||
3033 		    (adev->apu_flags & AMD_APU_IS_RAVEN2))
3034 			gfx_v9_1_init_rlc_save_restore_list(adev);
3035 		gfx_v9_0_enable_save_restore_machine(adev);
3036 	}
3037 
3038 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
3039 			      AMD_PG_SUPPORT_GFX_SMG |
3040 			      AMD_PG_SUPPORT_GFX_DMG |
3041 			      AMD_PG_SUPPORT_CP |
3042 			      AMD_PG_SUPPORT_GDS |
3043 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
3044 		WREG32_SOC15(GC, 0, mmRLC_JUMP_TABLE_RESTORE,
3045 			     adev->gfx.rlc.cp_table_gpu_addr >> 8);
3046 		gfx_v9_0_init_gfx_power_gating(adev);
3047 	}
3048 }
3049 
3050 static void gfx_v9_0_rlc_stop(struct amdgpu_device *adev)
3051 {
3052 	WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 0);
3053 	gfx_v9_0_enable_gui_idle_interrupt(adev, false);
3054 	gfx_v9_0_wait_for_rlc_serdes(adev);
3055 }
3056 
3057 static void gfx_v9_0_rlc_reset(struct amdgpu_device *adev)
3058 {
3059 	WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
3060 	udelay(50);
3061 	WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
3062 	udelay(50);
3063 }
3064 
3065 static void gfx_v9_0_rlc_start(struct amdgpu_device *adev)
3066 {
3067 #ifdef AMDGPU_RLC_DEBUG_RETRY
3068 	u32 rlc_ucode_ver;
3069 #endif
3070 
3071 	WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1);
3072 	udelay(50);
3073 
3074 	/* carrizo do enable cp interrupt after cp inited */
3075 	if (!(adev->flags & AMD_IS_APU)) {
3076 		gfx_v9_0_enable_gui_idle_interrupt(adev, true);
3077 		udelay(50);
3078 	}
3079 
3080 #ifdef AMDGPU_RLC_DEBUG_RETRY
3081 	/* RLC_GPM_GENERAL_6 : RLC Ucode version */
3082 	rlc_ucode_ver = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_6);
3083 	if(rlc_ucode_ver == 0x108) {
3084 		DRM_INFO("Using rlc debug ucode. mmRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n",
3085 				rlc_ucode_ver, adev->gfx.rlc_fw_version);
3086 		/* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles,
3087 		 * default is 0x9C4 to create a 100us interval */
3088 		WREG32_SOC15(GC, 0, mmRLC_GPM_TIMER_INT_3, 0x9C4);
3089 		/* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr
3090 		 * to disable the page fault retry interrupts, default is
3091 		 * 0x100 (256) */
3092 		WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_12, 0x100);
3093 	}
3094 #endif
3095 }
3096 
3097 static int gfx_v9_0_rlc_load_microcode(struct amdgpu_device *adev)
3098 {
3099 	const struct rlc_firmware_header_v2_0 *hdr;
3100 	const __le32 *fw_data;
3101 	unsigned i, fw_size;
3102 
3103 	if (!adev->gfx.rlc_fw)
3104 		return -EINVAL;
3105 
3106 	hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
3107 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
3108 
3109 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
3110 			   le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3111 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3112 
3113 	WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR,
3114 			RLCG_UCODE_LOADING_START_ADDRESS);
3115 	for (i = 0; i < fw_size; i++)
3116 		WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3117 	WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3118 
3119 	return 0;
3120 }
3121 
3122 static int gfx_v9_0_rlc_resume(struct amdgpu_device *adev)
3123 {
3124 	int r;
3125 
3126 	if (amdgpu_sriov_vf(adev)) {
3127 		gfx_v9_0_init_csb(adev);
3128 		return 0;
3129 	}
3130 
3131 	adev->gfx.rlc.funcs->stop(adev);
3132 
3133 	/* disable CG */
3134 	WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, 0);
3135 
3136 	gfx_v9_0_init_pg(adev);
3137 
3138 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
3139 		/* legacy rlc firmware loading */
3140 		r = gfx_v9_0_rlc_load_microcode(adev);
3141 		if (r)
3142 			return r;
3143 	}
3144 
3145 	switch (adev->ip_versions[GC_HWIP][0]) {
3146 	case IP_VERSION(9, 2, 2):
3147 	case IP_VERSION(9, 1, 0):
3148 		if (amdgpu_lbpw == 0)
3149 			gfx_v9_0_enable_lbpw(adev, false);
3150 		else
3151 			gfx_v9_0_enable_lbpw(adev, true);
3152 		break;
3153 	case IP_VERSION(9, 4, 0):
3154 		if (amdgpu_lbpw > 0)
3155 			gfx_v9_0_enable_lbpw(adev, true);
3156 		else
3157 			gfx_v9_0_enable_lbpw(adev, false);
3158 		break;
3159 	default:
3160 		break;
3161 	}
3162 
3163 	adev->gfx.rlc.funcs->start(adev);
3164 
3165 	return 0;
3166 }
3167 
3168 static void gfx_v9_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
3169 {
3170 	u32 tmp = RREG32_SOC15(GC, 0, mmCP_ME_CNTL);
3171 
3172 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1);
3173 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1);
3174 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, enable ? 0 : 1);
3175 	WREG32_SOC15_RLC(GC, 0, mmCP_ME_CNTL, tmp);
3176 	udelay(50);
3177 }
3178 
3179 static int gfx_v9_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
3180 {
3181 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
3182 	const struct gfx_firmware_header_v1_0 *ce_hdr;
3183 	const struct gfx_firmware_header_v1_0 *me_hdr;
3184 	const __le32 *fw_data;
3185 	unsigned i, fw_size;
3186 
3187 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
3188 		return -EINVAL;
3189 
3190 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)
3191 		adev->gfx.pfp_fw->data;
3192 	ce_hdr = (const struct gfx_firmware_header_v1_0 *)
3193 		adev->gfx.ce_fw->data;
3194 	me_hdr = (const struct gfx_firmware_header_v1_0 *)
3195 		adev->gfx.me_fw->data;
3196 
3197 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
3198 	amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
3199 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
3200 
3201 	gfx_v9_0_cp_gfx_enable(adev, false);
3202 
3203 	/* PFP */
3204 	fw_data = (const __le32 *)
3205 		(adev->gfx.pfp_fw->data +
3206 		 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
3207 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
3208 	WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, 0);
3209 	for (i = 0; i < fw_size; i++)
3210 		WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
3211 	WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
3212 
3213 	/* CE */
3214 	fw_data = (const __le32 *)
3215 		(adev->gfx.ce_fw->data +
3216 		 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
3217 	fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
3218 	WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, 0);
3219 	for (i = 0; i < fw_size; i++)
3220 		WREG32_SOC15(GC, 0, mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
3221 	WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
3222 
3223 	/* ME */
3224 	fw_data = (const __le32 *)
3225 		(adev->gfx.me_fw->data +
3226 		 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
3227 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
3228 	WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, 0);
3229 	for (i = 0; i < fw_size; i++)
3230 		WREG32_SOC15(GC, 0, mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
3231 	WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
3232 
3233 	return 0;
3234 }
3235 
3236 static int gfx_v9_0_cp_gfx_start(struct amdgpu_device *adev)
3237 {
3238 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
3239 	const struct cs_section_def *sect = NULL;
3240 	const struct cs_extent_def *ext = NULL;
3241 	int r, i, tmp;
3242 
3243 	/* init the CP */
3244 	WREG32_SOC15(GC, 0, mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
3245 	WREG32_SOC15(GC, 0, mmCP_DEVICE_ID, 1);
3246 
3247 	gfx_v9_0_cp_gfx_enable(adev, true);
3248 
3249 	r = amdgpu_ring_alloc(ring, gfx_v9_0_get_csb_size(adev) + 4 + 3);
3250 	if (r) {
3251 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
3252 		return r;
3253 	}
3254 
3255 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3256 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3257 
3258 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3259 	amdgpu_ring_write(ring, 0x80000000);
3260 	amdgpu_ring_write(ring, 0x80000000);
3261 
3262 	for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
3263 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3264 			if (sect->id == SECT_CONTEXT) {
3265 				amdgpu_ring_write(ring,
3266 				       PACKET3(PACKET3_SET_CONTEXT_REG,
3267 					       ext->reg_count));
3268 				amdgpu_ring_write(ring,
3269 				       ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
3270 				for (i = 0; i < ext->reg_count; i++)
3271 					amdgpu_ring_write(ring, ext->extent[i]);
3272 			}
3273 		}
3274 	}
3275 
3276 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3277 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
3278 
3279 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3280 	amdgpu_ring_write(ring, 0);
3281 
3282 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
3283 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
3284 	amdgpu_ring_write(ring, 0x8000);
3285 	amdgpu_ring_write(ring, 0x8000);
3286 
3287 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG,1));
3288 	tmp = (PACKET3_SET_UCONFIG_REG_INDEX_TYPE |
3289 		(SOC15_REG_OFFSET(GC, 0, mmVGT_INDEX_TYPE) - PACKET3_SET_UCONFIG_REG_START));
3290 	amdgpu_ring_write(ring, tmp);
3291 	amdgpu_ring_write(ring, 0);
3292 
3293 	amdgpu_ring_commit(ring);
3294 
3295 	return 0;
3296 }
3297 
3298 static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev)
3299 {
3300 	struct amdgpu_ring *ring;
3301 	u32 tmp;
3302 	u32 rb_bufsz;
3303 	u64 rb_addr, rptr_addr, wptr_gpu_addr;
3304 
3305 	/* Set the write pointer delay */
3306 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_DELAY, 0);
3307 
3308 	/* set the RB to use vmid 0 */
3309 	WREG32_SOC15(GC, 0, mmCP_RB_VMID, 0);
3310 
3311 	/* Set ring buffer size */
3312 	ring = &adev->gfx.gfx_ring[0];
3313 	rb_bufsz = order_base_2(ring->ring_size / 8);
3314 	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
3315 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
3316 #ifdef __BIG_ENDIAN
3317 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
3318 #endif
3319 	WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3320 
3321 	/* Initialize the ring buffer's write pointers */
3322 	ring->wptr = 0;
3323 	WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
3324 	WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
3325 
3326 	/* set the wb address wether it's enabled or not */
3327 	rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
3328 	WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
3329 	WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3330 
3331 	wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
3332 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
3333 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
3334 
3335 	mdelay(1);
3336 	WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3337 
3338 	rb_addr = ring->gpu_addr >> 8;
3339 	WREG32_SOC15(GC, 0, mmCP_RB0_BASE, rb_addr);
3340 	WREG32_SOC15(GC, 0, mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
3341 
3342 	tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL);
3343 	if (ring->use_doorbell) {
3344 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3345 				    DOORBELL_OFFSET, ring->doorbell_index);
3346 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3347 				    DOORBELL_EN, 1);
3348 	} else {
3349 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
3350 	}
3351 	WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, tmp);
3352 
3353 	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
3354 			DOORBELL_RANGE_LOWER, ring->doorbell_index);
3355 	WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
3356 
3357 	WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER,
3358 		       CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
3359 
3360 
3361 	/* start the ring */
3362 	gfx_v9_0_cp_gfx_start(adev);
3363 	ring->sched.ready = true;
3364 
3365 	return 0;
3366 }
3367 
3368 static void gfx_v9_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
3369 {
3370 	if (enable) {
3371 		WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL, 0);
3372 	} else {
3373 		WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL,
3374 			(CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
3375 		adev->gfx.kiq.ring.sched.ready = false;
3376 	}
3377 	udelay(50);
3378 }
3379 
3380 static int gfx_v9_0_cp_compute_load_microcode(struct amdgpu_device *adev)
3381 {
3382 	const struct gfx_firmware_header_v1_0 *mec_hdr;
3383 	const __le32 *fw_data;
3384 	unsigned i;
3385 	u32 tmp;
3386 
3387 	if (!adev->gfx.mec_fw)
3388 		return -EINVAL;
3389 
3390 	gfx_v9_0_cp_compute_enable(adev, false);
3391 
3392 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3393 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3394 
3395 	fw_data = (const __le32 *)
3396 		(adev->gfx.mec_fw->data +
3397 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
3398 	tmp = 0;
3399 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
3400 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
3401 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL, tmp);
3402 
3403 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO,
3404 		adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000);
3405 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI,
3406 		upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
3407 
3408 	/* MEC1 */
3409 	WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3410 			 mec_hdr->jt_offset);
3411 	for (i = 0; i < mec_hdr->jt_size; i++)
3412 		WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_DATA,
3413 			le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
3414 
3415 	WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3416 			adev->gfx.mec_fw_version);
3417 	/* Todo : Loading MEC2 firmware is only necessary if MEC2 should run different microcode than MEC1. */
3418 
3419 	return 0;
3420 }
3421 
3422 /* KIQ functions */
3423 static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)
3424 {
3425 	uint32_t tmp;
3426 	struct amdgpu_device *adev = ring->adev;
3427 
3428 	/* tell RLC which is KIQ queue */
3429 	tmp = RREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS);
3430 	tmp &= 0xffffff00;
3431 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
3432 	WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3433 	tmp |= 0x80;
3434 	WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3435 }
3436 
3437 static void gfx_v9_0_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd)
3438 {
3439 	struct amdgpu_device *adev = ring->adev;
3440 
3441 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
3442 		if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) {
3443 			mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
3444 			mqd->cp_hqd_queue_priority =
3445 				AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
3446 		}
3447 	}
3448 }
3449 
3450 static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring)
3451 {
3452 	struct amdgpu_device *adev = ring->adev;
3453 	struct v9_mqd *mqd = ring->mqd_ptr;
3454 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
3455 	uint32_t tmp;
3456 
3457 	mqd->header = 0xC0310800;
3458 	mqd->compute_pipelinestat_enable = 0x00000001;
3459 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
3460 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
3461 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
3462 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
3463 	mqd->compute_static_thread_mgmt_se4 = 0xffffffff;
3464 	mqd->compute_static_thread_mgmt_se5 = 0xffffffff;
3465 	mqd->compute_static_thread_mgmt_se6 = 0xffffffff;
3466 	mqd->compute_static_thread_mgmt_se7 = 0xffffffff;
3467 	mqd->compute_misc_reserved = 0x00000003;
3468 
3469 	mqd->dynamic_cu_mask_addr_lo =
3470 		lower_32_bits(ring->mqd_gpu_addr
3471 			      + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3472 	mqd->dynamic_cu_mask_addr_hi =
3473 		upper_32_bits(ring->mqd_gpu_addr
3474 			      + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3475 
3476 	eop_base_addr = ring->eop_gpu_addr >> 8;
3477 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
3478 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
3479 
3480 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3481 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL);
3482 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
3483 			(order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1));
3484 
3485 	mqd->cp_hqd_eop_control = tmp;
3486 
3487 	/* enable doorbell? */
3488 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
3489 
3490 	if (ring->use_doorbell) {
3491 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3492 				    DOORBELL_OFFSET, ring->doorbell_index);
3493 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3494 				    DOORBELL_EN, 1);
3495 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3496 				    DOORBELL_SOURCE, 0);
3497 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3498 				    DOORBELL_HIT, 0);
3499 	} else {
3500 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3501 					 DOORBELL_EN, 0);
3502 	}
3503 
3504 	mqd->cp_hqd_pq_doorbell_control = tmp;
3505 
3506 	/* disable the queue if it's active */
3507 	ring->wptr = 0;
3508 	mqd->cp_hqd_dequeue_request = 0;
3509 	mqd->cp_hqd_pq_rptr = 0;
3510 	mqd->cp_hqd_pq_wptr_lo = 0;
3511 	mqd->cp_hqd_pq_wptr_hi = 0;
3512 
3513 	/* set the pointer to the MQD */
3514 	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
3515 	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
3516 
3517 	/* set MQD vmid to 0 */
3518 	tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL);
3519 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
3520 	mqd->cp_mqd_control = tmp;
3521 
3522 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3523 	hqd_gpu_addr = ring->gpu_addr >> 8;
3524 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
3525 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3526 
3527 	/* set up the HQD, this is similar to CP_RB0_CNTL */
3528 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL);
3529 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
3530 			    (order_base_2(ring->ring_size / 4) - 1));
3531 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
3532 			((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8));
3533 #ifdef __BIG_ENDIAN
3534 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
3535 #endif
3536 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
3537 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
3538 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
3539 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
3540 	mqd->cp_hqd_pq_control = tmp;
3541 
3542 	/* set the wb address whether it's enabled or not */
3543 	wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
3544 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
3545 	mqd->cp_hqd_pq_rptr_report_addr_hi =
3546 		upper_32_bits(wb_gpu_addr) & 0xffff;
3547 
3548 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3549 	wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
3550 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3551 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3552 
3553 	tmp = 0;
3554 	/* enable the doorbell if requested */
3555 	if (ring->use_doorbell) {
3556 		tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
3557 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3558 				DOORBELL_OFFSET, ring->doorbell_index);
3559 
3560 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3561 					 DOORBELL_EN, 1);
3562 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3563 					 DOORBELL_SOURCE, 0);
3564 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3565 					 DOORBELL_HIT, 0);
3566 	}
3567 
3568 	mqd->cp_hqd_pq_doorbell_control = tmp;
3569 
3570 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3571 	ring->wptr = 0;
3572 	mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR);
3573 
3574 	/* set the vmid for the queue */
3575 	mqd->cp_hqd_vmid = 0;
3576 
3577 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE);
3578 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
3579 	mqd->cp_hqd_persistent_state = tmp;
3580 
3581 	/* set MIN_IB_AVAIL_SIZE */
3582 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL);
3583 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
3584 	mqd->cp_hqd_ib_control = tmp;
3585 
3586 	/* set static priority for a queue/ring */
3587 	gfx_v9_0_mqd_set_priority(ring, mqd);
3588 	mqd->cp_hqd_quantum = RREG32_SOC15(GC, 0, mmCP_HQD_QUANTUM);
3589 
3590 	/* map_queues packet doesn't need activate the queue,
3591 	 * so only kiq need set this field.
3592 	 */
3593 	if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
3594 		mqd->cp_hqd_active = 1;
3595 
3596 	return 0;
3597 }
3598 
3599 static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring)
3600 {
3601 	struct amdgpu_device *adev = ring->adev;
3602 	struct v9_mqd *mqd = ring->mqd_ptr;
3603 	int j;
3604 
3605 	/* disable wptr polling */
3606 	WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3607 
3608 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR,
3609 	       mqd->cp_hqd_eop_base_addr_lo);
3610 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI,
3611 	       mqd->cp_hqd_eop_base_addr_hi);
3612 
3613 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3614 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_CONTROL,
3615 	       mqd->cp_hqd_eop_control);
3616 
3617 	/* enable doorbell? */
3618 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3619 	       mqd->cp_hqd_pq_doorbell_control);
3620 
3621 	/* disable the queue if it's active */
3622 	if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3623 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3624 		for (j = 0; j < adev->usec_timeout; j++) {
3625 			if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3626 				break;
3627 			udelay(1);
3628 		}
3629 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3630 		       mqd->cp_hqd_dequeue_request);
3631 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR,
3632 		       mqd->cp_hqd_pq_rptr);
3633 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3634 		       mqd->cp_hqd_pq_wptr_lo);
3635 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3636 		       mqd->cp_hqd_pq_wptr_hi);
3637 	}
3638 
3639 	/* set the pointer to the MQD */
3640 	WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR,
3641 	       mqd->cp_mqd_base_addr_lo);
3642 	WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR_HI,
3643 	       mqd->cp_mqd_base_addr_hi);
3644 
3645 	/* set MQD vmid to 0 */
3646 	WREG32_SOC15_RLC(GC, 0, mmCP_MQD_CONTROL,
3647 	       mqd->cp_mqd_control);
3648 
3649 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3650 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE,
3651 	       mqd->cp_hqd_pq_base_lo);
3652 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE_HI,
3653 	       mqd->cp_hqd_pq_base_hi);
3654 
3655 	/* set up the HQD, this is similar to CP_RB0_CNTL */
3656 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_CONTROL,
3657 	       mqd->cp_hqd_pq_control);
3658 
3659 	/* set the wb address whether it's enabled or not */
3660 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR,
3661 				mqd->cp_hqd_pq_rptr_report_addr_lo);
3662 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3663 				mqd->cp_hqd_pq_rptr_report_addr_hi);
3664 
3665 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3666 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR,
3667 	       mqd->cp_hqd_pq_wptr_poll_addr_lo);
3668 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
3669 	       mqd->cp_hqd_pq_wptr_poll_addr_hi);
3670 
3671 	/* enable the doorbell if requested */
3672 	if (ring->use_doorbell) {
3673 		WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER,
3674 					(adev->doorbell_index.kiq * 2) << 2);
3675 		/* If GC has entered CGPG, ringing doorbell > first page
3676 		 * doesn't wakeup GC. Enlarge CP_MEC_DOORBELL_RANGE_UPPER to
3677 		 * workaround this issue. And this change has to align with firmware
3678 		 * update.
3679 		 */
3680 		if (check_if_enlarge_doorbell_range(adev))
3681 			WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
3682 					(adev->doorbell.size - 4));
3683 		else
3684 			WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
3685 					(adev->doorbell_index.userqueue_end * 2) << 2);
3686 	}
3687 
3688 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3689 	       mqd->cp_hqd_pq_doorbell_control);
3690 
3691 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3692 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3693 	       mqd->cp_hqd_pq_wptr_lo);
3694 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3695 	       mqd->cp_hqd_pq_wptr_hi);
3696 
3697 	/* set the vmid for the queue */
3698 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_VMID, mqd->cp_hqd_vmid);
3699 
3700 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE,
3701 	       mqd->cp_hqd_persistent_state);
3702 
3703 	/* activate the queue */
3704 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE,
3705 	       mqd->cp_hqd_active);
3706 
3707 	if (ring->use_doorbell)
3708 		WREG32_FIELD15(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
3709 
3710 	return 0;
3711 }
3712 
3713 static int gfx_v9_0_kiq_fini_register(struct amdgpu_ring *ring)
3714 {
3715 	struct amdgpu_device *adev = ring->adev;
3716 	int j;
3717 
3718 	/* disable the queue if it's active */
3719 	if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3720 
3721 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3722 
3723 		for (j = 0; j < adev->usec_timeout; j++) {
3724 			if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3725 				break;
3726 			udelay(1);
3727 		}
3728 
3729 		if (j == AMDGPU_MAX_USEC_TIMEOUT) {
3730 			DRM_DEBUG("KIQ dequeue request failed.\n");
3731 
3732 			/* Manual disable if dequeue request times out */
3733 			WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE, 0);
3734 		}
3735 
3736 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3737 		      0);
3738 	}
3739 
3740 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IQ_TIMER, 0);
3741 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IB_CONTROL, 0);
3742 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE, 0);
3743 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000);
3744 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0);
3745 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR, 0);
3746 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 0);
3747 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 0);
3748 
3749 	return 0;
3750 }
3751 
3752 static int gfx_v9_0_kiq_init_queue(struct amdgpu_ring *ring)
3753 {
3754 	struct amdgpu_device *adev = ring->adev;
3755 	struct v9_mqd *mqd = ring->mqd_ptr;
3756 	int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
3757 	struct v9_mqd *tmp_mqd;
3758 
3759 	gfx_v9_0_kiq_setting(ring);
3760 
3761 	/* GPU could be in bad state during probe, driver trigger the reset
3762 	 * after load the SMU, in this case , the mqd is not be initialized.
3763 	 * driver need to re-init the mqd.
3764 	 * check mqd->cp_hqd_pq_control since this value should not be 0
3765 	 */
3766 	tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx];
3767 	if (amdgpu_in_reset(adev) && tmp_mqd->cp_hqd_pq_control){
3768 		/* for GPU_RESET case , reset MQD to a clean status */
3769 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3770 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
3771 
3772 		/* reset ring buffer */
3773 		ring->wptr = 0;
3774 		amdgpu_ring_clear_ring(ring);
3775 
3776 		mutex_lock(&adev->srbm_mutex);
3777 		soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3778 		gfx_v9_0_kiq_init_register(ring);
3779 		soc15_grbm_select(adev, 0, 0, 0, 0);
3780 		mutex_unlock(&adev->srbm_mutex);
3781 	} else {
3782 		memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3783 		((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3784 		((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3785 		mutex_lock(&adev->srbm_mutex);
3786 		soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3787 		gfx_v9_0_mqd_init(ring);
3788 		gfx_v9_0_kiq_init_register(ring);
3789 		soc15_grbm_select(adev, 0, 0, 0, 0);
3790 		mutex_unlock(&adev->srbm_mutex);
3791 
3792 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3793 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
3794 	}
3795 
3796 	return 0;
3797 }
3798 
3799 static int gfx_v9_0_kcq_init_queue(struct amdgpu_ring *ring)
3800 {
3801 	struct amdgpu_device *adev = ring->adev;
3802 	struct v9_mqd *mqd = ring->mqd_ptr;
3803 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
3804 	struct v9_mqd *tmp_mqd;
3805 
3806 	/* Same as above kiq init, driver need to re-init the mqd if mqd->cp_hqd_pq_control
3807 	 * is not be initialized before
3808 	 */
3809 	tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx];
3810 
3811 	if (!tmp_mqd->cp_hqd_pq_control ||
3812 	    (!amdgpu_in_reset(adev) && !adev->in_suspend)) {
3813 		memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3814 		((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3815 		((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3816 		mutex_lock(&adev->srbm_mutex);
3817 		soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3818 		gfx_v9_0_mqd_init(ring);
3819 		soc15_grbm_select(adev, 0, 0, 0, 0);
3820 		mutex_unlock(&adev->srbm_mutex);
3821 
3822 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3823 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
3824 	} else if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
3825 		/* reset MQD to a clean status */
3826 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3827 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
3828 
3829 		/* reset ring buffer */
3830 		ring->wptr = 0;
3831 		atomic64_set((atomic64_t *)&adev->wb.wb[ring->wptr_offs], 0);
3832 		amdgpu_ring_clear_ring(ring);
3833 	} else {
3834 		amdgpu_ring_clear_ring(ring);
3835 	}
3836 
3837 	return 0;
3838 }
3839 
3840 static int gfx_v9_0_kiq_resume(struct amdgpu_device *adev)
3841 {
3842 	struct amdgpu_ring *ring;
3843 	int r;
3844 
3845 	ring = &adev->gfx.kiq.ring;
3846 
3847 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
3848 	if (unlikely(r != 0))
3849 		return r;
3850 
3851 	r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3852 	if (unlikely(r != 0))
3853 		return r;
3854 
3855 	gfx_v9_0_kiq_init_queue(ring);
3856 	amdgpu_bo_kunmap(ring->mqd_obj);
3857 	ring->mqd_ptr = NULL;
3858 	amdgpu_bo_unreserve(ring->mqd_obj);
3859 	ring->sched.ready = true;
3860 	return 0;
3861 }
3862 
3863 static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev)
3864 {
3865 	struct amdgpu_ring *ring = NULL;
3866 	int r = 0, i;
3867 
3868 	gfx_v9_0_cp_compute_enable(adev, true);
3869 
3870 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3871 		ring = &adev->gfx.compute_ring[i];
3872 
3873 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
3874 		if (unlikely(r != 0))
3875 			goto done;
3876 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3877 		if (!r) {
3878 			r = gfx_v9_0_kcq_init_queue(ring);
3879 			amdgpu_bo_kunmap(ring->mqd_obj);
3880 			ring->mqd_ptr = NULL;
3881 		}
3882 		amdgpu_bo_unreserve(ring->mqd_obj);
3883 		if (r)
3884 			goto done;
3885 	}
3886 
3887 	r = amdgpu_gfx_enable_kcq(adev);
3888 done:
3889 	return r;
3890 }
3891 
3892 static int gfx_v9_0_cp_resume(struct amdgpu_device *adev)
3893 {
3894 	int r, i;
3895 	struct amdgpu_ring *ring;
3896 
3897 	if (!(adev->flags & AMD_IS_APU))
3898 		gfx_v9_0_enable_gui_idle_interrupt(adev, false);
3899 
3900 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
3901 		if (adev->gfx.num_gfx_rings) {
3902 			/* legacy firmware loading */
3903 			r = gfx_v9_0_cp_gfx_load_microcode(adev);
3904 			if (r)
3905 				return r;
3906 		}
3907 
3908 		r = gfx_v9_0_cp_compute_load_microcode(adev);
3909 		if (r)
3910 			return r;
3911 	}
3912 
3913 	r = gfx_v9_0_kiq_resume(adev);
3914 	if (r)
3915 		return r;
3916 
3917 	if (adev->gfx.num_gfx_rings) {
3918 		r = gfx_v9_0_cp_gfx_resume(adev);
3919 		if (r)
3920 			return r;
3921 	}
3922 
3923 	r = gfx_v9_0_kcq_resume(adev);
3924 	if (r)
3925 		return r;
3926 
3927 	if (adev->gfx.num_gfx_rings) {
3928 		ring = &adev->gfx.gfx_ring[0];
3929 		r = amdgpu_ring_test_helper(ring);
3930 		if (r)
3931 			return r;
3932 	}
3933 
3934 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3935 		ring = &adev->gfx.compute_ring[i];
3936 		amdgpu_ring_test_helper(ring);
3937 	}
3938 
3939 	gfx_v9_0_enable_gui_idle_interrupt(adev, true);
3940 
3941 	return 0;
3942 }
3943 
3944 static void gfx_v9_0_init_tcp_config(struct amdgpu_device *adev)
3945 {
3946 	u32 tmp;
3947 
3948 	if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1) &&
3949 	    adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 2))
3950 		return;
3951 
3952 	tmp = RREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG);
3953 	tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE64KHASH,
3954 				adev->df.hash_status.hash_64k);
3955 	tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE2MHASH,
3956 				adev->df.hash_status.hash_2m);
3957 	tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE1GHASH,
3958 				adev->df.hash_status.hash_1g);
3959 	WREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG, tmp);
3960 }
3961 
3962 static void gfx_v9_0_cp_enable(struct amdgpu_device *adev, bool enable)
3963 {
3964 	if (adev->gfx.num_gfx_rings)
3965 		gfx_v9_0_cp_gfx_enable(adev, enable);
3966 	gfx_v9_0_cp_compute_enable(adev, enable);
3967 }
3968 
3969 static int gfx_v9_0_hw_init(void *handle)
3970 {
3971 	int r;
3972 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3973 
3974 	if (!amdgpu_sriov_vf(adev))
3975 		gfx_v9_0_init_golden_registers(adev);
3976 
3977 	gfx_v9_0_constants_init(adev);
3978 
3979 	gfx_v9_0_init_tcp_config(adev);
3980 
3981 	r = adev->gfx.rlc.funcs->resume(adev);
3982 	if (r)
3983 		return r;
3984 
3985 	r = gfx_v9_0_cp_resume(adev);
3986 	if (r)
3987 		return r;
3988 
3989 	if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
3990 		gfx_v9_4_2_set_power_brake_sequence(adev);
3991 
3992 	return r;
3993 }
3994 
3995 static int gfx_v9_0_hw_fini(void *handle)
3996 {
3997 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3998 
3999 	amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
4000 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4001 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4002 
4003 	/* DF freeze and kcq disable will fail */
4004 	if (!amdgpu_ras_intr_triggered())
4005 		/* disable KCQ to avoid CPC touch memory not valid anymore */
4006 		amdgpu_gfx_disable_kcq(adev);
4007 
4008 	if (amdgpu_sriov_vf(adev)) {
4009 		gfx_v9_0_cp_gfx_enable(adev, false);
4010 		/* must disable polling for SRIOV when hw finished, otherwise
4011 		 * CPC engine may still keep fetching WB address which is already
4012 		 * invalid after sw finished and trigger DMAR reading error in
4013 		 * hypervisor side.
4014 		 */
4015 		WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
4016 		return 0;
4017 	}
4018 
4019 	/* Use deinitialize sequence from CAIL when unbinding device from driver,
4020 	 * otherwise KIQ is hanging when binding back
4021 	 */
4022 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
4023 		mutex_lock(&adev->srbm_mutex);
4024 		soc15_grbm_select(adev, adev->gfx.kiq.ring.me,
4025 				adev->gfx.kiq.ring.pipe,
4026 				adev->gfx.kiq.ring.queue, 0);
4027 		gfx_v9_0_kiq_fini_register(&adev->gfx.kiq.ring);
4028 		soc15_grbm_select(adev, 0, 0, 0, 0);
4029 		mutex_unlock(&adev->srbm_mutex);
4030 	}
4031 
4032 	gfx_v9_0_cp_enable(adev, false);
4033 
4034 	/* Skip stopping RLC with A+A reset or when RLC controls GFX clock */
4035 	if ((adev->gmc.xgmi.connected_to_cpu && amdgpu_in_reset(adev)) ||
4036 	    (adev->ip_versions[GC_HWIP][0] >= IP_VERSION(9, 4, 2))) {
4037 		dev_dbg(adev->dev, "Skipping RLC halt\n");
4038 		return 0;
4039 	}
4040 
4041 	adev->gfx.rlc.funcs->stop(adev);
4042 	return 0;
4043 }
4044 
4045 static int gfx_v9_0_suspend(void *handle)
4046 {
4047 	return gfx_v9_0_hw_fini(handle);
4048 }
4049 
4050 static int gfx_v9_0_resume(void *handle)
4051 {
4052 	return gfx_v9_0_hw_init(handle);
4053 }
4054 
4055 static bool gfx_v9_0_is_idle(void *handle)
4056 {
4057 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4058 
4059 	if (REG_GET_FIELD(RREG32_SOC15(GC, 0, mmGRBM_STATUS),
4060 				GRBM_STATUS, GUI_ACTIVE))
4061 		return false;
4062 	else
4063 		return true;
4064 }
4065 
4066 static int gfx_v9_0_wait_for_idle(void *handle)
4067 {
4068 	unsigned i;
4069 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4070 
4071 	for (i = 0; i < adev->usec_timeout; i++) {
4072 		if (gfx_v9_0_is_idle(handle))
4073 			return 0;
4074 		udelay(1);
4075 	}
4076 	return -ETIMEDOUT;
4077 }
4078 
4079 static int gfx_v9_0_soft_reset(void *handle)
4080 {
4081 	u32 grbm_soft_reset = 0;
4082 	u32 tmp;
4083 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4084 
4085 	/* GRBM_STATUS */
4086 	tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS);
4087 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4088 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4089 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4090 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4091 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4092 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) {
4093 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4094 						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4095 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4096 						GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
4097 	}
4098 
4099 	if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4100 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4101 						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4102 	}
4103 
4104 	/* GRBM_STATUS2 */
4105 	tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2);
4106 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
4107 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4108 						GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4109 
4110 
4111 	if (grbm_soft_reset) {
4112 		/* stop the rlc */
4113 		adev->gfx.rlc.funcs->stop(adev);
4114 
4115 		if (adev->gfx.num_gfx_rings)
4116 			/* Disable GFX parsing/prefetching */
4117 			gfx_v9_0_cp_gfx_enable(adev, false);
4118 
4119 		/* Disable MEC parsing/prefetching */
4120 		gfx_v9_0_cp_compute_enable(adev, false);
4121 
4122 		if (grbm_soft_reset) {
4123 			tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
4124 			tmp |= grbm_soft_reset;
4125 			dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4126 			WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
4127 			tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
4128 
4129 			udelay(50);
4130 
4131 			tmp &= ~grbm_soft_reset;
4132 			WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
4133 			tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
4134 		}
4135 
4136 		/* Wait a little for things to settle down */
4137 		udelay(50);
4138 	}
4139 	return 0;
4140 }
4141 
4142 static uint64_t gfx_v9_0_kiq_read_clock(struct amdgpu_device *adev)
4143 {
4144 	signed long r, cnt = 0;
4145 	unsigned long flags;
4146 	uint32_t seq, reg_val_offs = 0;
4147 	uint64_t value = 0;
4148 	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
4149 	struct amdgpu_ring *ring = &kiq->ring;
4150 
4151 	BUG_ON(!ring->funcs->emit_rreg);
4152 
4153 	spin_lock_irqsave(&kiq->ring_lock, flags);
4154 	if (amdgpu_device_wb_get(adev, &reg_val_offs)) {
4155 		pr_err("critical bug! too many kiq readers\n");
4156 		goto failed_unlock;
4157 	}
4158 	amdgpu_ring_alloc(ring, 32);
4159 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
4160 	amdgpu_ring_write(ring, 9 |	/* src: register*/
4161 				(5 << 8) |	/* dst: memory */
4162 				(1 << 16) |	/* count sel */
4163 				(1 << 20));	/* write confirm */
4164 	amdgpu_ring_write(ring, 0);
4165 	amdgpu_ring_write(ring, 0);
4166 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
4167 				reg_val_offs * 4));
4168 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
4169 				reg_val_offs * 4));
4170 	r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT);
4171 	if (r)
4172 		goto failed_undo;
4173 
4174 	amdgpu_ring_commit(ring);
4175 	spin_unlock_irqrestore(&kiq->ring_lock, flags);
4176 
4177 	r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
4178 
4179 	/* don't wait anymore for gpu reset case because this way may
4180 	 * block gpu_recover() routine forever, e.g. this virt_kiq_rreg
4181 	 * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will
4182 	 * never return if we keep waiting in virt_kiq_rreg, which cause
4183 	 * gpu_recover() hang there.
4184 	 *
4185 	 * also don't wait anymore for IRQ context
4186 	 * */
4187 	if (r < 1 && (amdgpu_in_reset(adev)))
4188 		goto failed_kiq_read;
4189 
4190 	might_sleep();
4191 	while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) {
4192 		msleep(MAX_KIQ_REG_BAILOUT_INTERVAL);
4193 		r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
4194 	}
4195 
4196 	if (cnt > MAX_KIQ_REG_TRY)
4197 		goto failed_kiq_read;
4198 
4199 	mb();
4200 	value = (uint64_t)adev->wb.wb[reg_val_offs] |
4201 		(uint64_t)adev->wb.wb[reg_val_offs + 1 ] << 32ULL;
4202 	amdgpu_device_wb_free(adev, reg_val_offs);
4203 	return value;
4204 
4205 failed_undo:
4206 	amdgpu_ring_undo(ring);
4207 failed_unlock:
4208 	spin_unlock_irqrestore(&kiq->ring_lock, flags);
4209 failed_kiq_read:
4210 	if (reg_val_offs)
4211 		amdgpu_device_wb_free(adev, reg_val_offs);
4212 	pr_err("failed to read gpu clock\n");
4213 	return ~0;
4214 }
4215 
4216 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4217 {
4218 	uint64_t clock, clock_lo, clock_hi, hi_check;
4219 
4220 	switch (adev->ip_versions[GC_HWIP][0]) {
4221 	case IP_VERSION(9, 3, 0):
4222 		preempt_disable();
4223 		clock_hi = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir);
4224 		clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir);
4225 		hi_check = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir);
4226 		/* The SMUIO TSC clock frequency is 100MHz, which sets 32-bit carry over
4227 		 * roughly every 42 seconds.
4228 		 */
4229 		if (hi_check != clock_hi) {
4230 			clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir);
4231 			clock_hi = hi_check;
4232 		}
4233 		preempt_enable();
4234 		clock = clock_lo | (clock_hi << 32ULL);
4235 		break;
4236 	default:
4237 		amdgpu_gfx_off_ctrl(adev, false);
4238 		mutex_lock(&adev->gfx.gpu_clock_mutex);
4239 		if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 0, 1) && amdgpu_sriov_runtime(adev)) {
4240 			clock = gfx_v9_0_kiq_read_clock(adev);
4241 		} else {
4242 			WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4243 			clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB) |
4244 				((uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4245 		}
4246 		mutex_unlock(&adev->gfx.gpu_clock_mutex);
4247 		amdgpu_gfx_off_ctrl(adev, true);
4248 		break;
4249 	}
4250 	return clock;
4251 }
4252 
4253 static void gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4254 					  uint32_t vmid,
4255 					  uint32_t gds_base, uint32_t gds_size,
4256 					  uint32_t gws_base, uint32_t gws_size,
4257 					  uint32_t oa_base, uint32_t oa_size)
4258 {
4259 	struct amdgpu_device *adev = ring->adev;
4260 
4261 	/* GDS Base */
4262 	gfx_v9_0_write_data_to_reg(ring, 0, false,
4263 				   SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_BASE) + 2 * vmid,
4264 				   gds_base);
4265 
4266 	/* GDS Size */
4267 	gfx_v9_0_write_data_to_reg(ring, 0, false,
4268 				   SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE) + 2 * vmid,
4269 				   gds_size);
4270 
4271 	/* GWS */
4272 	gfx_v9_0_write_data_to_reg(ring, 0, false,
4273 				   SOC15_REG_OFFSET(GC, 0, mmGDS_GWS_VMID0) + vmid,
4274 				   gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4275 
4276 	/* OA */
4277 	gfx_v9_0_write_data_to_reg(ring, 0, false,
4278 				   SOC15_REG_OFFSET(GC, 0, mmGDS_OA_VMID0) + vmid,
4279 				   (1 << (oa_size + oa_base)) - (1 << oa_base));
4280 }
4281 
4282 static const u32 vgpr_init_compute_shader[] =
4283 {
4284 	0xb07c0000, 0xbe8000ff,
4285 	0x000000f8, 0xbf110800,
4286 	0x7e000280, 0x7e020280,
4287 	0x7e040280, 0x7e060280,
4288 	0x7e080280, 0x7e0a0280,
4289 	0x7e0c0280, 0x7e0e0280,
4290 	0x80808800, 0xbe803200,
4291 	0xbf84fff5, 0xbf9c0000,
4292 	0xd28c0001, 0x0001007f,
4293 	0xd28d0001, 0x0002027e,
4294 	0x10020288, 0xb8810904,
4295 	0xb7814000, 0xd1196a01,
4296 	0x00000301, 0xbe800087,
4297 	0xbefc00c1, 0xd89c4000,
4298 	0x00020201, 0xd89cc080,
4299 	0x00040401, 0x320202ff,
4300 	0x00000800, 0x80808100,
4301 	0xbf84fff8, 0x7e020280,
4302 	0xbf810000, 0x00000000,
4303 };
4304 
4305 static const u32 sgpr_init_compute_shader[] =
4306 {
4307 	0xb07c0000, 0xbe8000ff,
4308 	0x0000005f, 0xbee50080,
4309 	0xbe812c65, 0xbe822c65,
4310 	0xbe832c65, 0xbe842c65,
4311 	0xbe852c65, 0xb77c0005,
4312 	0x80808500, 0xbf84fff8,
4313 	0xbe800080, 0xbf810000,
4314 };
4315 
4316 static const u32 vgpr_init_compute_shader_arcturus[] = {
4317 	0xd3d94000, 0x18000080, 0xd3d94001, 0x18000080, 0xd3d94002, 0x18000080,
4318 	0xd3d94003, 0x18000080, 0xd3d94004, 0x18000080, 0xd3d94005, 0x18000080,
4319 	0xd3d94006, 0x18000080, 0xd3d94007, 0x18000080, 0xd3d94008, 0x18000080,
4320 	0xd3d94009, 0x18000080, 0xd3d9400a, 0x18000080, 0xd3d9400b, 0x18000080,
4321 	0xd3d9400c, 0x18000080, 0xd3d9400d, 0x18000080, 0xd3d9400e, 0x18000080,
4322 	0xd3d9400f, 0x18000080, 0xd3d94010, 0x18000080, 0xd3d94011, 0x18000080,
4323 	0xd3d94012, 0x18000080, 0xd3d94013, 0x18000080, 0xd3d94014, 0x18000080,
4324 	0xd3d94015, 0x18000080, 0xd3d94016, 0x18000080, 0xd3d94017, 0x18000080,
4325 	0xd3d94018, 0x18000080, 0xd3d94019, 0x18000080, 0xd3d9401a, 0x18000080,
4326 	0xd3d9401b, 0x18000080, 0xd3d9401c, 0x18000080, 0xd3d9401d, 0x18000080,
4327 	0xd3d9401e, 0x18000080, 0xd3d9401f, 0x18000080, 0xd3d94020, 0x18000080,
4328 	0xd3d94021, 0x18000080, 0xd3d94022, 0x18000080, 0xd3d94023, 0x18000080,
4329 	0xd3d94024, 0x18000080, 0xd3d94025, 0x18000080, 0xd3d94026, 0x18000080,
4330 	0xd3d94027, 0x18000080, 0xd3d94028, 0x18000080, 0xd3d94029, 0x18000080,
4331 	0xd3d9402a, 0x18000080, 0xd3d9402b, 0x18000080, 0xd3d9402c, 0x18000080,
4332 	0xd3d9402d, 0x18000080, 0xd3d9402e, 0x18000080, 0xd3d9402f, 0x18000080,
4333 	0xd3d94030, 0x18000080, 0xd3d94031, 0x18000080, 0xd3d94032, 0x18000080,
4334 	0xd3d94033, 0x18000080, 0xd3d94034, 0x18000080, 0xd3d94035, 0x18000080,
4335 	0xd3d94036, 0x18000080, 0xd3d94037, 0x18000080, 0xd3d94038, 0x18000080,
4336 	0xd3d94039, 0x18000080, 0xd3d9403a, 0x18000080, 0xd3d9403b, 0x18000080,
4337 	0xd3d9403c, 0x18000080, 0xd3d9403d, 0x18000080, 0xd3d9403e, 0x18000080,
4338 	0xd3d9403f, 0x18000080, 0xd3d94040, 0x18000080, 0xd3d94041, 0x18000080,
4339 	0xd3d94042, 0x18000080, 0xd3d94043, 0x18000080, 0xd3d94044, 0x18000080,
4340 	0xd3d94045, 0x18000080, 0xd3d94046, 0x18000080, 0xd3d94047, 0x18000080,
4341 	0xd3d94048, 0x18000080, 0xd3d94049, 0x18000080, 0xd3d9404a, 0x18000080,
4342 	0xd3d9404b, 0x18000080, 0xd3d9404c, 0x18000080, 0xd3d9404d, 0x18000080,
4343 	0xd3d9404e, 0x18000080, 0xd3d9404f, 0x18000080, 0xd3d94050, 0x18000080,
4344 	0xd3d94051, 0x18000080, 0xd3d94052, 0x18000080, 0xd3d94053, 0x18000080,
4345 	0xd3d94054, 0x18000080, 0xd3d94055, 0x18000080, 0xd3d94056, 0x18000080,
4346 	0xd3d94057, 0x18000080, 0xd3d94058, 0x18000080, 0xd3d94059, 0x18000080,
4347 	0xd3d9405a, 0x18000080, 0xd3d9405b, 0x18000080, 0xd3d9405c, 0x18000080,
4348 	0xd3d9405d, 0x18000080, 0xd3d9405e, 0x18000080, 0xd3d9405f, 0x18000080,
4349 	0xd3d94060, 0x18000080, 0xd3d94061, 0x18000080, 0xd3d94062, 0x18000080,
4350 	0xd3d94063, 0x18000080, 0xd3d94064, 0x18000080, 0xd3d94065, 0x18000080,
4351 	0xd3d94066, 0x18000080, 0xd3d94067, 0x18000080, 0xd3d94068, 0x18000080,
4352 	0xd3d94069, 0x18000080, 0xd3d9406a, 0x18000080, 0xd3d9406b, 0x18000080,
4353 	0xd3d9406c, 0x18000080, 0xd3d9406d, 0x18000080, 0xd3d9406e, 0x18000080,
4354 	0xd3d9406f, 0x18000080, 0xd3d94070, 0x18000080, 0xd3d94071, 0x18000080,
4355 	0xd3d94072, 0x18000080, 0xd3d94073, 0x18000080, 0xd3d94074, 0x18000080,
4356 	0xd3d94075, 0x18000080, 0xd3d94076, 0x18000080, 0xd3d94077, 0x18000080,
4357 	0xd3d94078, 0x18000080, 0xd3d94079, 0x18000080, 0xd3d9407a, 0x18000080,
4358 	0xd3d9407b, 0x18000080, 0xd3d9407c, 0x18000080, 0xd3d9407d, 0x18000080,
4359 	0xd3d9407e, 0x18000080, 0xd3d9407f, 0x18000080, 0xd3d94080, 0x18000080,
4360 	0xd3d94081, 0x18000080, 0xd3d94082, 0x18000080, 0xd3d94083, 0x18000080,
4361 	0xd3d94084, 0x18000080, 0xd3d94085, 0x18000080, 0xd3d94086, 0x18000080,
4362 	0xd3d94087, 0x18000080, 0xd3d94088, 0x18000080, 0xd3d94089, 0x18000080,
4363 	0xd3d9408a, 0x18000080, 0xd3d9408b, 0x18000080, 0xd3d9408c, 0x18000080,
4364 	0xd3d9408d, 0x18000080, 0xd3d9408e, 0x18000080, 0xd3d9408f, 0x18000080,
4365 	0xd3d94090, 0x18000080, 0xd3d94091, 0x18000080, 0xd3d94092, 0x18000080,
4366 	0xd3d94093, 0x18000080, 0xd3d94094, 0x18000080, 0xd3d94095, 0x18000080,
4367 	0xd3d94096, 0x18000080, 0xd3d94097, 0x18000080, 0xd3d94098, 0x18000080,
4368 	0xd3d94099, 0x18000080, 0xd3d9409a, 0x18000080, 0xd3d9409b, 0x18000080,
4369 	0xd3d9409c, 0x18000080, 0xd3d9409d, 0x18000080, 0xd3d9409e, 0x18000080,
4370 	0xd3d9409f, 0x18000080, 0xd3d940a0, 0x18000080, 0xd3d940a1, 0x18000080,
4371 	0xd3d940a2, 0x18000080, 0xd3d940a3, 0x18000080, 0xd3d940a4, 0x18000080,
4372 	0xd3d940a5, 0x18000080, 0xd3d940a6, 0x18000080, 0xd3d940a7, 0x18000080,
4373 	0xd3d940a8, 0x18000080, 0xd3d940a9, 0x18000080, 0xd3d940aa, 0x18000080,
4374 	0xd3d940ab, 0x18000080, 0xd3d940ac, 0x18000080, 0xd3d940ad, 0x18000080,
4375 	0xd3d940ae, 0x18000080, 0xd3d940af, 0x18000080, 0xd3d940b0, 0x18000080,
4376 	0xd3d940b1, 0x18000080, 0xd3d940b2, 0x18000080, 0xd3d940b3, 0x18000080,
4377 	0xd3d940b4, 0x18000080, 0xd3d940b5, 0x18000080, 0xd3d940b6, 0x18000080,
4378 	0xd3d940b7, 0x18000080, 0xd3d940b8, 0x18000080, 0xd3d940b9, 0x18000080,
4379 	0xd3d940ba, 0x18000080, 0xd3d940bb, 0x18000080, 0xd3d940bc, 0x18000080,
4380 	0xd3d940bd, 0x18000080, 0xd3d940be, 0x18000080, 0xd3d940bf, 0x18000080,
4381 	0xd3d940c0, 0x18000080, 0xd3d940c1, 0x18000080, 0xd3d940c2, 0x18000080,
4382 	0xd3d940c3, 0x18000080, 0xd3d940c4, 0x18000080, 0xd3d940c5, 0x18000080,
4383 	0xd3d940c6, 0x18000080, 0xd3d940c7, 0x18000080, 0xd3d940c8, 0x18000080,
4384 	0xd3d940c9, 0x18000080, 0xd3d940ca, 0x18000080, 0xd3d940cb, 0x18000080,
4385 	0xd3d940cc, 0x18000080, 0xd3d940cd, 0x18000080, 0xd3d940ce, 0x18000080,
4386 	0xd3d940cf, 0x18000080, 0xd3d940d0, 0x18000080, 0xd3d940d1, 0x18000080,
4387 	0xd3d940d2, 0x18000080, 0xd3d940d3, 0x18000080, 0xd3d940d4, 0x18000080,
4388 	0xd3d940d5, 0x18000080, 0xd3d940d6, 0x18000080, 0xd3d940d7, 0x18000080,
4389 	0xd3d940d8, 0x18000080, 0xd3d940d9, 0x18000080, 0xd3d940da, 0x18000080,
4390 	0xd3d940db, 0x18000080, 0xd3d940dc, 0x18000080, 0xd3d940dd, 0x18000080,
4391 	0xd3d940de, 0x18000080, 0xd3d940df, 0x18000080, 0xd3d940e0, 0x18000080,
4392 	0xd3d940e1, 0x18000080, 0xd3d940e2, 0x18000080, 0xd3d940e3, 0x18000080,
4393 	0xd3d940e4, 0x18000080, 0xd3d940e5, 0x18000080, 0xd3d940e6, 0x18000080,
4394 	0xd3d940e7, 0x18000080, 0xd3d940e8, 0x18000080, 0xd3d940e9, 0x18000080,
4395 	0xd3d940ea, 0x18000080, 0xd3d940eb, 0x18000080, 0xd3d940ec, 0x18000080,
4396 	0xd3d940ed, 0x18000080, 0xd3d940ee, 0x18000080, 0xd3d940ef, 0x18000080,
4397 	0xd3d940f0, 0x18000080, 0xd3d940f1, 0x18000080, 0xd3d940f2, 0x18000080,
4398 	0xd3d940f3, 0x18000080, 0xd3d940f4, 0x18000080, 0xd3d940f5, 0x18000080,
4399 	0xd3d940f6, 0x18000080, 0xd3d940f7, 0x18000080, 0xd3d940f8, 0x18000080,
4400 	0xd3d940f9, 0x18000080, 0xd3d940fa, 0x18000080, 0xd3d940fb, 0x18000080,
4401 	0xd3d940fc, 0x18000080, 0xd3d940fd, 0x18000080, 0xd3d940fe, 0x18000080,
4402 	0xd3d940ff, 0x18000080, 0xb07c0000, 0xbe8a00ff, 0x000000f8, 0xbf11080a,
4403 	0x7e000280, 0x7e020280, 0x7e040280, 0x7e060280, 0x7e080280, 0x7e0a0280,
4404 	0x7e0c0280, 0x7e0e0280, 0x808a880a, 0xbe80320a, 0xbf84fff5, 0xbf9c0000,
4405 	0xd28c0001, 0x0001007f, 0xd28d0001, 0x0002027e, 0x10020288, 0xb88b0904,
4406 	0xb78b4000, 0xd1196a01, 0x00001701, 0xbe8a0087, 0xbefc00c1, 0xd89c4000,
4407 	0x00020201, 0xd89cc080, 0x00040401, 0x320202ff, 0x00000800, 0x808a810a,
4408 	0xbf84fff8, 0xbf810000,
4409 };
4410 
4411 /* When below register arrays changed, please update gpr_reg_size,
4412   and sec_ded_counter_reg_size in function gfx_v9_0_do_edc_gpr_workarounds,
4413   to cover all gfx9 ASICs */
4414 static const struct soc15_reg_entry vgpr_init_regs[] = {
4415    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4416    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4417    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 },
4418    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4419    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x3f },
4420    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 },  /* 64KB LDS */
4421    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
4422    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
4423    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
4424    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
4425    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff },
4426    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff },
4427    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff },
4428    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff },
4429 };
4430 
4431 static const struct soc15_reg_entry vgpr_init_regs_arcturus[] = {
4432    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4433    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4434    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 },
4435    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4436    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0xbf },
4437    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 },  /* 64KB LDS */
4438    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
4439    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
4440    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
4441    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
4442    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff },
4443    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff },
4444    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff },
4445    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff },
4446 };
4447 
4448 static const struct soc15_reg_entry sgpr1_init_regs[] = {
4449    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4450    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4451    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 },
4452    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4453    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */
4454    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
4455    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x000000ff },
4456    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x000000ff },
4457    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x000000ff },
4458    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x000000ff },
4459    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x000000ff },
4460    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x000000ff },
4461    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x000000ff },
4462    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x000000ff },
4463 };
4464 
4465 static const struct soc15_reg_entry sgpr2_init_regs[] = {
4466    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4467    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4468    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 },
4469    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4470    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */
4471    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
4472    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x0000ff00 },
4473    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x0000ff00 },
4474    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x0000ff00 },
4475    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x0000ff00 },
4476    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x0000ff00 },
4477    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x0000ff00 },
4478    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x0000ff00 },
4479    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x0000ff00 },
4480 };
4481 
4482 static const struct soc15_reg_entry gfx_v9_0_edc_counter_regs[] = {
4483    { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT), 0, 1, 1},
4484    { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT), 0, 1, 1},
4485    { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 0, 1, 1},
4486    { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT), 0, 1, 1},
4487    { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 0, 1, 1},
4488    { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT), 0, 1, 1},
4489    { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT), 0, 1, 1},
4490    { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT), 0, 1, 1},
4491    { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT), 0, 1, 1},
4492    { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 0, 1, 1},
4493    { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_GRBM_CNT), 0, 1, 1},
4494    { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_DED), 0, 1, 1},
4495    { SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT), 0, 4, 1},
4496    { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 0, 4, 6},
4497    { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_DED_CNT), 0, 4, 16},
4498    { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_INFO), 0, 4, 16},
4499    { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_SEC_CNT), 0, 4, 16},
4500    { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 0, 1, 16},
4501    { SOC15_REG_ENTRY(GC, 0, mmTCP_ATC_EDC_GATCL1_CNT), 0, 4, 16},
4502    { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT), 0, 4, 16},
4503    { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 0, 4, 16},
4504    { SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 0, 4, 16},
4505    { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 0, 4, 6},
4506    { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 0, 4, 16},
4507    { SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 0, 4, 16},
4508    { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 0, 1, 1},
4509    { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 0, 1, 1},
4510    { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 0, 1, 32},
4511    { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 0, 1, 32},
4512    { SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT), 0, 1, 72},
4513    { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 0, 1, 16},
4514    { SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 0, 1, 2},
4515    { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 0, 4, 6},
4516 };
4517 
4518 static int gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device *adev)
4519 {
4520 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4521 	int i, r;
4522 
4523 	/* only support when RAS is enabled */
4524 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4525 		return 0;
4526 
4527 	r = amdgpu_ring_alloc(ring, 7);
4528 	if (r) {
4529 		DRM_ERROR("amdgpu: GDS workarounds failed to lock ring %s (%d).\n",
4530 			ring->name, r);
4531 		return r;
4532 	}
4533 
4534 	WREG32_SOC15(GC, 0, mmGDS_VMID0_BASE, 0x00000000);
4535 	WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, adev->gds.gds_size);
4536 
4537 	amdgpu_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5));
4538 	amdgpu_ring_write(ring, (PACKET3_DMA_DATA_CP_SYNC |
4539 				PACKET3_DMA_DATA_DST_SEL(1) |
4540 				PACKET3_DMA_DATA_SRC_SEL(2) |
4541 				PACKET3_DMA_DATA_ENGINE(0)));
4542 	amdgpu_ring_write(ring, 0);
4543 	amdgpu_ring_write(ring, 0);
4544 	amdgpu_ring_write(ring, 0);
4545 	amdgpu_ring_write(ring, 0);
4546 	amdgpu_ring_write(ring, PACKET3_DMA_DATA_CMD_RAW_WAIT |
4547 				adev->gds.gds_size);
4548 
4549 	amdgpu_ring_commit(ring);
4550 
4551 	for (i = 0; i < adev->usec_timeout; i++) {
4552 		if (ring->wptr == gfx_v9_0_ring_get_rptr_compute(ring))
4553 			break;
4554 		udelay(1);
4555 	}
4556 
4557 	if (i >= adev->usec_timeout)
4558 		r = -ETIMEDOUT;
4559 
4560 	WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, 0x00000000);
4561 
4562 	return r;
4563 }
4564 
4565 static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
4566 {
4567 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4568 	struct amdgpu_ib ib;
4569 	struct dma_fence *f = NULL;
4570 	int r, i;
4571 	unsigned total_size, vgpr_offset, sgpr_offset;
4572 	u64 gpu_addr;
4573 
4574 	int compute_dim_x = adev->gfx.config.max_shader_engines *
4575 						adev->gfx.config.max_cu_per_sh *
4576 						adev->gfx.config.max_sh_per_se;
4577 	int sgpr_work_group_size = 5;
4578 	int gpr_reg_size = adev->gfx.config.max_shader_engines + 6;
4579 	int vgpr_init_shader_size;
4580 	const u32 *vgpr_init_shader_ptr;
4581 	const struct soc15_reg_entry *vgpr_init_regs_ptr;
4582 
4583 	/* only support when RAS is enabled */
4584 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4585 		return 0;
4586 
4587 	/* bail if the compute ring is not ready */
4588 	if (!ring->sched.ready)
4589 		return 0;
4590 
4591 	if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1)) {
4592 		vgpr_init_shader_ptr = vgpr_init_compute_shader_arcturus;
4593 		vgpr_init_shader_size = sizeof(vgpr_init_compute_shader_arcturus);
4594 		vgpr_init_regs_ptr = vgpr_init_regs_arcturus;
4595 	} else {
4596 		vgpr_init_shader_ptr = vgpr_init_compute_shader;
4597 		vgpr_init_shader_size = sizeof(vgpr_init_compute_shader);
4598 		vgpr_init_regs_ptr = vgpr_init_regs;
4599 	}
4600 
4601 	total_size =
4602 		(gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* VGPRS */
4603 	total_size +=
4604 		(gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS1 */
4605 	total_size +=
4606 		(gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS2 */
4607 	total_size = ALIGN(total_size, 256);
4608 	vgpr_offset = total_size;
4609 	total_size += ALIGN(vgpr_init_shader_size, 256);
4610 	sgpr_offset = total_size;
4611 	total_size += sizeof(sgpr_init_compute_shader);
4612 
4613 	/* allocate an indirect buffer to put the commands in */
4614 	memset(&ib, 0, sizeof(ib));
4615 	r = amdgpu_ib_get(adev, NULL, total_size,
4616 					AMDGPU_IB_POOL_DIRECT, &ib);
4617 	if (r) {
4618 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
4619 		return r;
4620 	}
4621 
4622 	/* load the compute shaders */
4623 	for (i = 0; i < vgpr_init_shader_size/sizeof(u32); i++)
4624 		ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_shader_ptr[i];
4625 
4626 	for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
4627 		ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
4628 
4629 	/* init the ib length to 0 */
4630 	ib.length_dw = 0;
4631 
4632 	/* VGPR */
4633 	/* write the register state for the compute dispatch */
4634 	for (i = 0; i < gpr_reg_size; i++) {
4635 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4636 		ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(vgpr_init_regs_ptr[i])
4637 								- PACKET3_SET_SH_REG_START;
4638 		ib.ptr[ib.length_dw++] = vgpr_init_regs_ptr[i].reg_value;
4639 	}
4640 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4641 	gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
4642 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4643 	ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4644 							- PACKET3_SET_SH_REG_START;
4645 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4646 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4647 
4648 	/* write dispatch packet */
4649 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4650 	ib.ptr[ib.length_dw++] = compute_dim_x * 2; /* x */
4651 	ib.ptr[ib.length_dw++] = 1; /* y */
4652 	ib.ptr[ib.length_dw++] = 1; /* z */
4653 	ib.ptr[ib.length_dw++] =
4654 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4655 
4656 	/* write CS partial flush packet */
4657 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4658 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4659 
4660 	/* SGPR1 */
4661 	/* write the register state for the compute dispatch */
4662 	for (i = 0; i < gpr_reg_size; i++) {
4663 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4664 		ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr1_init_regs[i])
4665 								- PACKET3_SET_SH_REG_START;
4666 		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i].reg_value;
4667 	}
4668 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4669 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4670 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4671 	ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4672 							- PACKET3_SET_SH_REG_START;
4673 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4674 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4675 
4676 	/* write dispatch packet */
4677 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4678 	ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */
4679 	ib.ptr[ib.length_dw++] = 1; /* y */
4680 	ib.ptr[ib.length_dw++] = 1; /* z */
4681 	ib.ptr[ib.length_dw++] =
4682 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4683 
4684 	/* write CS partial flush packet */
4685 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4686 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4687 
4688 	/* SGPR2 */
4689 	/* write the register state for the compute dispatch */
4690 	for (i = 0; i < gpr_reg_size; i++) {
4691 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4692 		ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr2_init_regs[i])
4693 								- PACKET3_SET_SH_REG_START;
4694 		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i].reg_value;
4695 	}
4696 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4697 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4698 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4699 	ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4700 							- PACKET3_SET_SH_REG_START;
4701 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4702 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4703 
4704 	/* write dispatch packet */
4705 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4706 	ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */
4707 	ib.ptr[ib.length_dw++] = 1; /* y */
4708 	ib.ptr[ib.length_dw++] = 1; /* z */
4709 	ib.ptr[ib.length_dw++] =
4710 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4711 
4712 	/* write CS partial flush packet */
4713 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4714 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4715 
4716 	/* shedule the ib on the ring */
4717 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
4718 	if (r) {
4719 		DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
4720 		goto fail;
4721 	}
4722 
4723 	/* wait for the GPU to finish processing the IB */
4724 	r = dma_fence_wait(f, false);
4725 	if (r) {
4726 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
4727 		goto fail;
4728 	}
4729 
4730 fail:
4731 	amdgpu_ib_free(adev, &ib, NULL);
4732 	dma_fence_put(f);
4733 
4734 	return r;
4735 }
4736 
4737 static int gfx_v9_0_early_init(void *handle)
4738 {
4739 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4740 
4741 	if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1) ||
4742 	    adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
4743 		adev->gfx.num_gfx_rings = 0;
4744 	else
4745 		adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS;
4746 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
4747 					  AMDGPU_MAX_COMPUTE_RINGS);
4748 	gfx_v9_0_set_kiq_pm4_funcs(adev);
4749 	gfx_v9_0_set_ring_funcs(adev);
4750 	gfx_v9_0_set_irq_funcs(adev);
4751 	gfx_v9_0_set_gds_init(adev);
4752 	gfx_v9_0_set_rlc_funcs(adev);
4753 
4754 	/* init rlcg reg access ctrl */
4755 	gfx_v9_0_init_rlcg_reg_access_ctrl(adev);
4756 
4757 	return 0;
4758 }
4759 
4760 static int gfx_v9_0_ecc_late_init(void *handle)
4761 {
4762 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4763 	int r;
4764 
4765 	/*
4766 	 * Temp workaround to fix the issue that CP firmware fails to
4767 	 * update read pointer when CPDMA is writing clearing operation
4768 	 * to GDS in suspend/resume sequence on several cards. So just
4769 	 * limit this operation in cold boot sequence.
4770 	 */
4771 	if ((!adev->in_suspend) &&
4772 	    (adev->gds.gds_size)) {
4773 		r = gfx_v9_0_do_edc_gds_workarounds(adev);
4774 		if (r)
4775 			return r;
4776 	}
4777 
4778 	/* requires IBs so do in late init after IB pool is initialized */
4779 	if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
4780 		r = gfx_v9_4_2_do_edc_gpr_workarounds(adev);
4781 	else
4782 		r = gfx_v9_0_do_edc_gpr_workarounds(adev);
4783 
4784 	if (r)
4785 		return r;
4786 
4787 	if (adev->gfx.ras &&
4788 	    adev->gfx.ras->enable_watchdog_timer)
4789 		adev->gfx.ras->enable_watchdog_timer(adev);
4790 
4791 	return 0;
4792 }
4793 
4794 static int gfx_v9_0_late_init(void *handle)
4795 {
4796 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4797 	int r;
4798 
4799 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4800 	if (r)
4801 		return r;
4802 
4803 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4804 	if (r)
4805 		return r;
4806 
4807 	r = gfx_v9_0_ecc_late_init(handle);
4808 	if (r)
4809 		return r;
4810 
4811 	return 0;
4812 }
4813 
4814 static bool gfx_v9_0_is_rlc_enabled(struct amdgpu_device *adev)
4815 {
4816 	uint32_t rlc_setting;
4817 
4818 	/* if RLC is not enabled, do nothing */
4819 	rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL);
4820 	if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
4821 		return false;
4822 
4823 	return true;
4824 }
4825 
4826 static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev)
4827 {
4828 	uint32_t data;
4829 	unsigned i;
4830 
4831 	data = RLC_SAFE_MODE__CMD_MASK;
4832 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
4833 	WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4834 
4835 	/* wait for RLC_SAFE_MODE */
4836 	for (i = 0; i < adev->usec_timeout; i++) {
4837 		if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
4838 			break;
4839 		udelay(1);
4840 	}
4841 }
4842 
4843 static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev)
4844 {
4845 	uint32_t data;
4846 
4847 	data = RLC_SAFE_MODE__CMD_MASK;
4848 	WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4849 }
4850 
4851 static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev,
4852 						bool enable)
4853 {
4854 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4855 
4856 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
4857 		gfx_v9_0_enable_gfx_cg_power_gating(adev, true);
4858 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4859 			gfx_v9_0_enable_gfx_pipeline_powergating(adev, true);
4860 	} else {
4861 		gfx_v9_0_enable_gfx_cg_power_gating(adev, false);
4862 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4863 			gfx_v9_0_enable_gfx_pipeline_powergating(adev, false);
4864 	}
4865 
4866 	amdgpu_gfx_rlc_exit_safe_mode(adev);
4867 }
4868 
4869 static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev,
4870 						bool enable)
4871 {
4872 	/* TODO: double check if we need to perform under safe mode */
4873 	/* gfx_v9_0_enter_rlc_safe_mode(adev); */
4874 
4875 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
4876 		gfx_v9_0_enable_gfx_static_mg_power_gating(adev, true);
4877 	else
4878 		gfx_v9_0_enable_gfx_static_mg_power_gating(adev, false);
4879 
4880 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
4881 		gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, true);
4882 	else
4883 		gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, false);
4884 
4885 	/* gfx_v9_0_exit_rlc_safe_mode(adev); */
4886 }
4887 
4888 static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
4889 						      bool enable)
4890 {
4891 	uint32_t data, def;
4892 
4893 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4894 
4895 	/* It is disabled by HW by default */
4896 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
4897 		/* 1 - RLC_CGTT_MGCG_OVERRIDE */
4898 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4899 
4900 		if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1))
4901 			data &= ~RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
4902 
4903 		data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4904 			  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4905 			  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4906 
4907 		/* only for Vega10 & Raven1 */
4908 		data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK;
4909 
4910 		if (def != data)
4911 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4912 
4913 		/* MGLS is a global flag to control all MGLS in GFX */
4914 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
4915 			/* 2 - RLC memory Light sleep */
4916 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
4917 				def = data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4918 				data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4919 				if (def != data)
4920 					WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4921 			}
4922 			/* 3 - CP memory Light sleep */
4923 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
4924 				def = data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4925 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4926 				if (def != data)
4927 					WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4928 			}
4929 		}
4930 	} else {
4931 		/* 1 - MGCG_OVERRIDE */
4932 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4933 
4934 		if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1))
4935 			data |= RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
4936 
4937 		data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4938 			 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4939 			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4940 			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4941 
4942 		if (def != data)
4943 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4944 
4945 		/* 2 - disable MGLS in RLC */
4946 		data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4947 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
4948 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4949 			WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4950 		}
4951 
4952 		/* 3 - disable MGLS in CP */
4953 		data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4954 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
4955 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4956 			WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4957 		}
4958 	}
4959 
4960 	amdgpu_gfx_rlc_exit_safe_mode(adev);
4961 }
4962 
4963 static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev,
4964 					   bool enable)
4965 {
4966 	uint32_t data, def;
4967 
4968 	if (!adev->gfx.num_gfx_rings)
4969 		return;
4970 
4971 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4972 
4973 	/* Enable 3D CGCG/CGLS */
4974 	if (enable) {
4975 		/* write cmd to clear cgcg/cgls ov */
4976 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4977 		/* unset CGCG override */
4978 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK;
4979 		/* update CGCG and CGLS override bits */
4980 		if (def != data)
4981 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4982 
4983 		/* enable 3Dcgcg FSM(0x0000363f) */
4984 		def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4985 
4986 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)
4987 			data = (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4988 				RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
4989 		else
4990 			data = 0x0 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT;
4991 
4992 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
4993 			data |= (0x000F << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4994 				RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
4995 		if (def != data)
4996 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
4997 
4998 		/* set IDLE_POLL_COUNT(0x00900100) */
4999 		def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
5000 		data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
5001 			(0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
5002 		if (def != data)
5003 			WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
5004 	} else {
5005 		/* Disable CGCG/CGLS */
5006 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
5007 		/* disable cgcg, cgls should be disabled */
5008 		data &= ~(RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK |
5009 			  RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK);
5010 		/* disable cgcg and cgls in FSM */
5011 		if (def != data)
5012 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
5013 	}
5014 
5015 	amdgpu_gfx_rlc_exit_safe_mode(adev);
5016 }
5017 
5018 static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
5019 						      bool enable)
5020 {
5021 	uint32_t def, data;
5022 
5023 	amdgpu_gfx_rlc_enter_safe_mode(adev);
5024 
5025 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
5026 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
5027 		/* unset CGCG override */
5028 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
5029 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
5030 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
5031 		else
5032 			data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
5033 		/* update CGCG and CGLS override bits */
5034 		if (def != data)
5035 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
5036 
5037 		/* enable cgcg FSM(0x0000363F) */
5038 		def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
5039 
5040 		if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1))
5041 			data = (0x2000 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
5042 				RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5043 		else
5044 			data = (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
5045 				RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5046 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
5047 			data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
5048 				RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5049 		if (def != data)
5050 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
5051 
5052 		/* set IDLE_POLL_COUNT(0x00900100) */
5053 		def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
5054 		data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
5055 			(0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
5056 		if (def != data)
5057 			WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
5058 	} else {
5059 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
5060 		/* reset CGCG/CGLS bits */
5061 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
5062 		/* disable cgcg and cgls in FSM */
5063 		if (def != data)
5064 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
5065 	}
5066 
5067 	amdgpu_gfx_rlc_exit_safe_mode(adev);
5068 }
5069 
5070 static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev,
5071 					    bool enable)
5072 {
5073 	if (enable) {
5074 		/* CGCG/CGLS should be enabled after MGCG/MGLS
5075 		 * ===  MGCG + MGLS ===
5076 		 */
5077 		gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
5078 		/* ===  CGCG /CGLS for GFX 3D Only === */
5079 		gfx_v9_0_update_3d_clock_gating(adev, enable);
5080 		/* ===  CGCG + CGLS === */
5081 		gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
5082 	} else {
5083 		/* CGCG/CGLS should be disabled before MGCG/MGLS
5084 		 * ===  CGCG + CGLS ===
5085 		 */
5086 		gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
5087 		/* ===  CGCG /CGLS for GFX 3D Only === */
5088 		gfx_v9_0_update_3d_clock_gating(adev, enable);
5089 		/* ===  MGCG + MGLS === */
5090 		gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
5091 	}
5092 	return 0;
5093 }
5094 
5095 static void gfx_v9_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
5096 {
5097 	u32 reg, data;
5098 
5099 	amdgpu_gfx_off_ctrl(adev, false);
5100 
5101 	reg = SOC15_REG_OFFSET(GC, 0, mmRLC_SPM_MC_CNTL);
5102 	if (amdgpu_sriov_is_pp_one_vf(adev))
5103 		data = RREG32_NO_KIQ(reg);
5104 	else
5105 		data = RREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL);
5106 
5107 	data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK;
5108 	data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT;
5109 
5110 	if (amdgpu_sriov_is_pp_one_vf(adev))
5111 		WREG32_SOC15_NO_KIQ(GC, 0, mmRLC_SPM_MC_CNTL, data);
5112 	else
5113 		WREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL, data);
5114 
5115 	amdgpu_gfx_off_ctrl(adev, true);
5116 }
5117 
5118 static bool gfx_v9_0_check_rlcg_range(struct amdgpu_device *adev,
5119 					uint32_t offset,
5120 					struct soc15_reg_rlcg *entries, int arr_size)
5121 {
5122 	int i;
5123 	uint32_t reg;
5124 
5125 	if (!entries)
5126 		return false;
5127 
5128 	for (i = 0; i < arr_size; i++) {
5129 		const struct soc15_reg_rlcg *entry;
5130 
5131 		entry = &entries[i];
5132 		reg = adev->reg_offset[entry->hwip][entry->instance][entry->segment] + entry->reg;
5133 		if (offset == reg)
5134 			return true;
5135 	}
5136 
5137 	return false;
5138 }
5139 
5140 static bool gfx_v9_0_is_rlcg_access_range(struct amdgpu_device *adev, u32 offset)
5141 {
5142 	return gfx_v9_0_check_rlcg_range(adev, offset,
5143 					(void *)rlcg_access_gc_9_0,
5144 					ARRAY_SIZE(rlcg_access_gc_9_0));
5145 }
5146 
5147 static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = {
5148 	.is_rlc_enabled = gfx_v9_0_is_rlc_enabled,
5149 	.set_safe_mode = gfx_v9_0_set_safe_mode,
5150 	.unset_safe_mode = gfx_v9_0_unset_safe_mode,
5151 	.init = gfx_v9_0_rlc_init,
5152 	.get_csb_size = gfx_v9_0_get_csb_size,
5153 	.get_csb_buffer = gfx_v9_0_get_csb_buffer,
5154 	.get_cp_table_num = gfx_v9_0_cp_jump_table_num,
5155 	.resume = gfx_v9_0_rlc_resume,
5156 	.stop = gfx_v9_0_rlc_stop,
5157 	.reset = gfx_v9_0_rlc_reset,
5158 	.start = gfx_v9_0_rlc_start,
5159 	.update_spm_vmid = gfx_v9_0_update_spm_vmid,
5160 	.is_rlcg_access_range = gfx_v9_0_is_rlcg_access_range,
5161 };
5162 
5163 static int gfx_v9_0_set_powergating_state(void *handle,
5164 					  enum amd_powergating_state state)
5165 {
5166 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5167 	bool enable = (state == AMD_PG_STATE_GATE);
5168 
5169 	switch (adev->ip_versions[GC_HWIP][0]) {
5170 	case IP_VERSION(9, 2, 2):
5171 	case IP_VERSION(9, 1, 0):
5172 	case IP_VERSION(9, 3, 0):
5173 		if (!enable)
5174 			amdgpu_gfx_off_ctrl(adev, false);
5175 
5176 		if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
5177 			gfx_v9_0_enable_sck_slow_down_on_power_up(adev, true);
5178 			gfx_v9_0_enable_sck_slow_down_on_power_down(adev, true);
5179 		} else {
5180 			gfx_v9_0_enable_sck_slow_down_on_power_up(adev, false);
5181 			gfx_v9_0_enable_sck_slow_down_on_power_down(adev, false);
5182 		}
5183 
5184 		if (adev->pg_flags & AMD_PG_SUPPORT_CP)
5185 			gfx_v9_0_enable_cp_power_gating(adev, true);
5186 		else
5187 			gfx_v9_0_enable_cp_power_gating(adev, false);
5188 
5189 		/* update gfx cgpg state */
5190 		gfx_v9_0_update_gfx_cg_power_gating(adev, enable);
5191 
5192 		/* update mgcg state */
5193 		gfx_v9_0_update_gfx_mg_power_gating(adev, enable);
5194 
5195 		if (enable)
5196 			amdgpu_gfx_off_ctrl(adev, true);
5197 		break;
5198 	case IP_VERSION(9, 2, 1):
5199 		amdgpu_gfx_off_ctrl(adev, enable);
5200 		break;
5201 	default:
5202 		break;
5203 	}
5204 
5205 	return 0;
5206 }
5207 
5208 static int gfx_v9_0_set_clockgating_state(void *handle,
5209 					  enum amd_clockgating_state state)
5210 {
5211 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5212 
5213 	if (amdgpu_sriov_vf(adev))
5214 		return 0;
5215 
5216 	switch (adev->ip_versions[GC_HWIP][0]) {
5217 	case IP_VERSION(9, 0, 1):
5218 	case IP_VERSION(9, 2, 1):
5219 	case IP_VERSION(9, 4, 0):
5220 	case IP_VERSION(9, 2, 2):
5221 	case IP_VERSION(9, 1, 0):
5222 	case IP_VERSION(9, 4, 1):
5223 	case IP_VERSION(9, 3, 0):
5224 	case IP_VERSION(9, 4, 2):
5225 		gfx_v9_0_update_gfx_clock_gating(adev,
5226 						 state == AMD_CG_STATE_GATE);
5227 		break;
5228 	default:
5229 		break;
5230 	}
5231 	return 0;
5232 }
5233 
5234 static void gfx_v9_0_get_clockgating_state(void *handle, u32 *flags)
5235 {
5236 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5237 	int data;
5238 
5239 	if (amdgpu_sriov_vf(adev))
5240 		*flags = 0;
5241 
5242 	/* AMD_CG_SUPPORT_GFX_MGCG */
5243 	data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE));
5244 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
5245 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
5246 
5247 	/* AMD_CG_SUPPORT_GFX_CGCG */
5248 	data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL));
5249 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5250 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
5251 
5252 	/* AMD_CG_SUPPORT_GFX_CGLS */
5253 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5254 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
5255 
5256 	/* AMD_CG_SUPPORT_GFX_RLC_LS */
5257 	data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_MEM_SLP_CNTL));
5258 	if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5259 		*flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5260 
5261 	/* AMD_CG_SUPPORT_GFX_CP_LS */
5262 	data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmCP_MEM_SLP_CNTL));
5263 	if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5264 		*flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5265 
5266 	if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)) {
5267 		/* AMD_CG_SUPPORT_GFX_3D_CGCG */
5268 		data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D));
5269 		if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK)
5270 			*flags |= AMD_CG_SUPPORT_GFX_3D_CGCG;
5271 
5272 		/* AMD_CG_SUPPORT_GFX_3D_CGLS */
5273 		if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK)
5274 			*flags |= AMD_CG_SUPPORT_GFX_3D_CGLS;
5275 	}
5276 }
5277 
5278 static u64 gfx_v9_0_ring_get_rptr_gfx(struct amdgpu_ring *ring)
5279 {
5280 	return ring->adev->wb.wb[ring->rptr_offs]; /* gfx9 is 32bit rptr*/
5281 }
5282 
5283 static u64 gfx_v9_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
5284 {
5285 	struct amdgpu_device *adev = ring->adev;
5286 	u64 wptr;
5287 
5288 	/* XXX check if swapping is necessary on BE */
5289 	if (ring->use_doorbell) {
5290 		wptr = atomic64_read((atomic64_t *)&adev->wb.wb[ring->wptr_offs]);
5291 	} else {
5292 		wptr = RREG32_SOC15(GC, 0, mmCP_RB0_WPTR);
5293 		wptr += (u64)RREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI) << 32;
5294 	}
5295 
5296 	return wptr;
5297 }
5298 
5299 static void gfx_v9_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
5300 {
5301 	struct amdgpu_device *adev = ring->adev;
5302 
5303 	if (ring->use_doorbell) {
5304 		/* XXX check if swapping is necessary on BE */
5305 		atomic64_set((atomic64_t *)&adev->wb.wb[ring->wptr_offs], ring->wptr);
5306 		WDOORBELL64(ring->doorbell_index, ring->wptr);
5307 	} else {
5308 		WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
5309 		WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
5310 	}
5311 }
5312 
5313 static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
5314 {
5315 	struct amdgpu_device *adev = ring->adev;
5316 	u32 ref_and_mask, reg_mem_engine;
5317 	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
5318 
5319 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
5320 		switch (ring->me) {
5321 		case 1:
5322 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
5323 			break;
5324 		case 2:
5325 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
5326 			break;
5327 		default:
5328 			return;
5329 		}
5330 		reg_mem_engine = 0;
5331 	} else {
5332 		ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
5333 		reg_mem_engine = 1; /* pfp */
5334 	}
5335 
5336 	gfx_v9_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
5337 			      adev->nbio.funcs->get_hdp_flush_req_offset(adev),
5338 			      adev->nbio.funcs->get_hdp_flush_done_offset(adev),
5339 			      ref_and_mask, ref_and_mask, 0x20);
5340 }
5341 
5342 static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
5343 					struct amdgpu_job *job,
5344 					struct amdgpu_ib *ib,
5345 					uint32_t flags)
5346 {
5347 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5348 	u32 header, control = 0;
5349 
5350 	if (ib->flags & AMDGPU_IB_FLAG_CE)
5351 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
5352 	else
5353 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
5354 
5355 	control |= ib->length_dw | (vmid << 24);
5356 
5357 	if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
5358 		control |= INDIRECT_BUFFER_PRE_ENB(1);
5359 
5360 		if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid)
5361 			gfx_v9_0_ring_emit_de_meta(ring);
5362 	}
5363 
5364 	amdgpu_ring_write(ring, header);
5365 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5366 	amdgpu_ring_write(ring,
5367 #ifdef __BIG_ENDIAN
5368 		(2 << 0) |
5369 #endif
5370 		lower_32_bits(ib->gpu_addr));
5371 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5372 	amdgpu_ring_write(ring, control);
5373 }
5374 
5375 static void gfx_v9_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
5376 					  struct amdgpu_job *job,
5377 					  struct amdgpu_ib *ib,
5378 					  uint32_t flags)
5379 {
5380 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5381 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
5382 
5383 	/* Currently, there is a high possibility to get wave ID mismatch
5384 	 * between ME and GDS, leading to a hw deadlock, because ME generates
5385 	 * different wave IDs than the GDS expects. This situation happens
5386 	 * randomly when at least 5 compute pipes use GDS ordered append.
5387 	 * The wave IDs generated by ME are also wrong after suspend/resume.
5388 	 * Those are probably bugs somewhere else in the kernel driver.
5389 	 *
5390 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
5391 	 * GDS to 0 for this ring (me/pipe).
5392 	 */
5393 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
5394 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
5395 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID);
5396 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
5397 	}
5398 
5399 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
5400 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5401 	amdgpu_ring_write(ring,
5402 #ifdef __BIG_ENDIAN
5403 				(2 << 0) |
5404 #endif
5405 				lower_32_bits(ib->gpu_addr));
5406 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5407 	amdgpu_ring_write(ring, control);
5408 }
5409 
5410 static void gfx_v9_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
5411 				     u64 seq, unsigned flags)
5412 {
5413 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
5414 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
5415 	bool writeback = flags & AMDGPU_FENCE_FLAG_TC_WB_ONLY;
5416 
5417 	/* RELEASE_MEM - flush caches, send int */
5418 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
5419 	amdgpu_ring_write(ring, ((writeback ? (EOP_TC_WB_ACTION_EN |
5420 					       EOP_TC_NC_ACTION_EN) :
5421 					      (EOP_TCL1_ACTION_EN |
5422 					       EOP_TC_ACTION_EN |
5423 					       EOP_TC_WB_ACTION_EN |
5424 					       EOP_TC_MD_ACTION_EN)) |
5425 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
5426 				 EVENT_INDEX(5)));
5427 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
5428 
5429 	/*
5430 	 * the address should be Qword aligned if 64bit write, Dword
5431 	 * aligned if only send 32bit data low (discard data high)
5432 	 */
5433 	if (write64bit)
5434 		BUG_ON(addr & 0x7);
5435 	else
5436 		BUG_ON(addr & 0x3);
5437 	amdgpu_ring_write(ring, lower_32_bits(addr));
5438 	amdgpu_ring_write(ring, upper_32_bits(addr));
5439 	amdgpu_ring_write(ring, lower_32_bits(seq));
5440 	amdgpu_ring_write(ring, upper_32_bits(seq));
5441 	amdgpu_ring_write(ring, 0);
5442 }
5443 
5444 static void gfx_v9_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
5445 {
5446 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5447 	uint32_t seq = ring->fence_drv.sync_seq;
5448 	uint64_t addr = ring->fence_drv.gpu_addr;
5449 
5450 	gfx_v9_0_wait_reg_mem(ring, usepfp, 1, 0,
5451 			      lower_32_bits(addr), upper_32_bits(addr),
5452 			      seq, 0xffffffff, 4);
5453 }
5454 
5455 static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
5456 					unsigned vmid, uint64_t pd_addr)
5457 {
5458 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
5459 
5460 	/* compute doesn't have PFP */
5461 	if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
5462 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
5463 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
5464 		amdgpu_ring_write(ring, 0x0);
5465 	}
5466 }
5467 
5468 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring)
5469 {
5470 	return ring->adev->wb.wb[ring->rptr_offs]; /* gfx9 hardware is 32bit rptr */
5471 }
5472 
5473 static u64 gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
5474 {
5475 	u64 wptr;
5476 
5477 	/* XXX check if swapping is necessary on BE */
5478 	if (ring->use_doorbell)
5479 		wptr = atomic64_read((atomic64_t *)&ring->adev->wb.wb[ring->wptr_offs]);
5480 	else
5481 		BUG();
5482 	return wptr;
5483 }
5484 
5485 static void gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
5486 {
5487 	struct amdgpu_device *adev = ring->adev;
5488 
5489 	/* XXX check if swapping is necessary on BE */
5490 	if (ring->use_doorbell) {
5491 		atomic64_set((atomic64_t *)&adev->wb.wb[ring->wptr_offs], ring->wptr);
5492 		WDOORBELL64(ring->doorbell_index, ring->wptr);
5493 	} else{
5494 		BUG(); /* only DOORBELL method supported on gfx9 now */
5495 	}
5496 }
5497 
5498 static void gfx_v9_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
5499 					 u64 seq, unsigned int flags)
5500 {
5501 	struct amdgpu_device *adev = ring->adev;
5502 
5503 	/* we only allocate 32bit for each seq wb address */
5504 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
5505 
5506 	/* write fence seq to the "addr" */
5507 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5508 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5509 				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
5510 	amdgpu_ring_write(ring, lower_32_bits(addr));
5511 	amdgpu_ring_write(ring, upper_32_bits(addr));
5512 	amdgpu_ring_write(ring, lower_32_bits(seq));
5513 
5514 	if (flags & AMDGPU_FENCE_FLAG_INT) {
5515 		/* set register to trigger INT */
5516 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5517 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5518 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
5519 		amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, mmCPC_INT_STATUS));
5520 		amdgpu_ring_write(ring, 0);
5521 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
5522 	}
5523 }
5524 
5525 static void gfx_v9_ring_emit_sb(struct amdgpu_ring *ring)
5526 {
5527 	amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
5528 	amdgpu_ring_write(ring, 0);
5529 }
5530 
5531 static void gfx_v9_0_ring_emit_ce_meta(struct amdgpu_ring *ring)
5532 {
5533 	struct v9_ce_ib_state ce_payload = {0};
5534 	uint64_t csa_addr;
5535 	int cnt;
5536 
5537 	cnt = (sizeof(ce_payload) >> 2) + 4 - 2;
5538 	csa_addr = amdgpu_csa_vaddr(ring->adev);
5539 
5540 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5541 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
5542 				 WRITE_DATA_DST_SEL(8) |
5543 				 WR_CONFIRM) |
5544 				 WRITE_DATA_CACHE_POLICY(0));
5545 	amdgpu_ring_write(ring, lower_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, ce_payload)));
5546 	amdgpu_ring_write(ring, upper_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, ce_payload)));
5547 	amdgpu_ring_write_multiple(ring, (void *)&ce_payload, sizeof(ce_payload) >> 2);
5548 }
5549 
5550 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring)
5551 {
5552 	struct v9_de_ib_state de_payload = {0};
5553 	uint64_t csa_addr, gds_addr;
5554 	int cnt;
5555 
5556 	csa_addr = amdgpu_csa_vaddr(ring->adev);
5557 	gds_addr = csa_addr + 4096;
5558 	de_payload.gds_backup_addrlo = lower_32_bits(gds_addr);
5559 	de_payload.gds_backup_addrhi = upper_32_bits(gds_addr);
5560 
5561 	cnt = (sizeof(de_payload) >> 2) + 4 - 2;
5562 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5563 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
5564 				 WRITE_DATA_DST_SEL(8) |
5565 				 WR_CONFIRM) |
5566 				 WRITE_DATA_CACHE_POLICY(0));
5567 	amdgpu_ring_write(ring, lower_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, de_payload)));
5568 	amdgpu_ring_write(ring, upper_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, de_payload)));
5569 	amdgpu_ring_write_multiple(ring, (void *)&de_payload, sizeof(de_payload) >> 2);
5570 }
5571 
5572 static void gfx_v9_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start,
5573 				   bool secure)
5574 {
5575 	uint32_t v = secure ? FRAME_TMZ : 0;
5576 
5577 	amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0));
5578 	amdgpu_ring_write(ring, v | FRAME_CMD(start ? 0 : 1));
5579 }
5580 
5581 static void gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
5582 {
5583 	uint32_t dw2 = 0;
5584 
5585 	if (amdgpu_sriov_vf(ring->adev))
5586 		gfx_v9_0_ring_emit_ce_meta(ring);
5587 
5588 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
5589 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
5590 		/* set load_global_config & load_global_uconfig */
5591 		dw2 |= 0x8001;
5592 		/* set load_cs_sh_regs */
5593 		dw2 |= 0x01000000;
5594 		/* set load_per_context_state & load_gfx_sh_regs for GFX */
5595 		dw2 |= 0x10002;
5596 
5597 		/* set load_ce_ram if preamble presented */
5598 		if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
5599 			dw2 |= 0x10000000;
5600 	} else {
5601 		/* still load_ce_ram if this is the first time preamble presented
5602 		 * although there is no context switch happens.
5603 		 */
5604 		if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
5605 			dw2 |= 0x10000000;
5606 	}
5607 
5608 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
5609 	amdgpu_ring_write(ring, dw2);
5610 	amdgpu_ring_write(ring, 0);
5611 }
5612 
5613 static unsigned gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
5614 {
5615 	unsigned ret;
5616 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
5617 	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
5618 	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
5619 	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
5620 	ret = ring->wptr & ring->buf_mask;
5621 	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
5622 	return ret;
5623 }
5624 
5625 static void gfx_v9_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
5626 {
5627 	unsigned cur;
5628 	BUG_ON(offset > ring->buf_mask);
5629 	BUG_ON(ring->ring[offset] != 0x55aa55aa);
5630 
5631 	cur = (ring->wptr & ring->buf_mask) - 1;
5632 	if (likely(cur > offset))
5633 		ring->ring[offset] = cur - offset;
5634 	else
5635 		ring->ring[offset] = (ring->ring_size>>2) - offset + cur;
5636 }
5637 
5638 static void gfx_v9_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
5639 				    uint32_t reg_val_offs)
5640 {
5641 	struct amdgpu_device *adev = ring->adev;
5642 
5643 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
5644 	amdgpu_ring_write(ring, 0 |	/* src: register*/
5645 				(5 << 8) |	/* dst: memory */
5646 				(1 << 20));	/* write confirm */
5647 	amdgpu_ring_write(ring, reg);
5648 	amdgpu_ring_write(ring, 0);
5649 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
5650 				reg_val_offs * 4));
5651 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
5652 				reg_val_offs * 4));
5653 }
5654 
5655 static void gfx_v9_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
5656 				    uint32_t val)
5657 {
5658 	uint32_t cmd = 0;
5659 
5660 	switch (ring->funcs->type) {
5661 	case AMDGPU_RING_TYPE_GFX:
5662 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
5663 		break;
5664 	case AMDGPU_RING_TYPE_KIQ:
5665 		cmd = (1 << 16); /* no inc addr */
5666 		break;
5667 	default:
5668 		cmd = WR_CONFIRM;
5669 		break;
5670 	}
5671 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5672 	amdgpu_ring_write(ring, cmd);
5673 	amdgpu_ring_write(ring, reg);
5674 	amdgpu_ring_write(ring, 0);
5675 	amdgpu_ring_write(ring, val);
5676 }
5677 
5678 static void gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
5679 					uint32_t val, uint32_t mask)
5680 {
5681 	gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
5682 }
5683 
5684 static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
5685 						  uint32_t reg0, uint32_t reg1,
5686 						  uint32_t ref, uint32_t mask)
5687 {
5688 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5689 	struct amdgpu_device *adev = ring->adev;
5690 	bool fw_version_ok = (ring->funcs->type == AMDGPU_RING_TYPE_GFX) ?
5691 		adev->gfx.me_fw_write_wait : adev->gfx.mec_fw_write_wait;
5692 
5693 	if (fw_version_ok)
5694 		gfx_v9_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1,
5695 				      ref, mask, 0x20);
5696 	else
5697 		amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1,
5698 							   ref, mask);
5699 }
5700 
5701 static void gfx_v9_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
5702 {
5703 	struct amdgpu_device *adev = ring->adev;
5704 	uint32_t value = 0;
5705 
5706 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
5707 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
5708 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
5709 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
5710 	WREG32_SOC15(GC, 0, mmSQ_CMD, value);
5711 }
5712 
5713 static void gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
5714 						 enum amdgpu_interrupt_state state)
5715 {
5716 	switch (state) {
5717 	case AMDGPU_IRQ_STATE_DISABLE:
5718 	case AMDGPU_IRQ_STATE_ENABLE:
5719 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5720 			       TIME_STAMP_INT_ENABLE,
5721 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5722 		break;
5723 	default:
5724 		break;
5725 	}
5726 }
5727 
5728 static void gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
5729 						     int me, int pipe,
5730 						     enum amdgpu_interrupt_state state)
5731 {
5732 	u32 mec_int_cntl, mec_int_cntl_reg;
5733 
5734 	/*
5735 	 * amdgpu controls only the first MEC. That's why this function only
5736 	 * handles the setting of interrupts for this specific MEC. All other
5737 	 * pipes' interrupts are set by amdkfd.
5738 	 */
5739 
5740 	if (me == 1) {
5741 		switch (pipe) {
5742 		case 0:
5743 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
5744 			break;
5745 		case 1:
5746 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL);
5747 			break;
5748 		case 2:
5749 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL);
5750 			break;
5751 		case 3:
5752 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL);
5753 			break;
5754 		default:
5755 			DRM_DEBUG("invalid pipe %d\n", pipe);
5756 			return;
5757 		}
5758 	} else {
5759 		DRM_DEBUG("invalid me %d\n", me);
5760 		return;
5761 	}
5762 
5763 	switch (state) {
5764 	case AMDGPU_IRQ_STATE_DISABLE:
5765 		mec_int_cntl = RREG32_SOC15_IP(GC,mec_int_cntl_reg);
5766 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5767 					     TIME_STAMP_INT_ENABLE, 0);
5768 		WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5769 		break;
5770 	case AMDGPU_IRQ_STATE_ENABLE:
5771 		mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg);
5772 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5773 					     TIME_STAMP_INT_ENABLE, 1);
5774 		WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5775 		break;
5776 	default:
5777 		break;
5778 	}
5779 }
5780 
5781 static int gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
5782 					     struct amdgpu_irq_src *source,
5783 					     unsigned type,
5784 					     enum amdgpu_interrupt_state state)
5785 {
5786 	switch (state) {
5787 	case AMDGPU_IRQ_STATE_DISABLE:
5788 	case AMDGPU_IRQ_STATE_ENABLE:
5789 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5790 			       PRIV_REG_INT_ENABLE,
5791 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5792 		break;
5793 	default:
5794 		break;
5795 	}
5796 
5797 	return 0;
5798 }
5799 
5800 static int gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
5801 					      struct amdgpu_irq_src *source,
5802 					      unsigned type,
5803 					      enum amdgpu_interrupt_state state)
5804 {
5805 	switch (state) {
5806 	case AMDGPU_IRQ_STATE_DISABLE:
5807 	case AMDGPU_IRQ_STATE_ENABLE:
5808 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5809 			       PRIV_INSTR_INT_ENABLE,
5810 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5811 		break;
5812 	default:
5813 		break;
5814 	}
5815 
5816 	return 0;
5817 }
5818 
5819 #define ENABLE_ECC_ON_ME_PIPE(me, pipe)				\
5820 	WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
5821 			CP_ECC_ERROR_INT_ENABLE, 1)
5822 
5823 #define DISABLE_ECC_ON_ME_PIPE(me, pipe)			\
5824 	WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
5825 			CP_ECC_ERROR_INT_ENABLE, 0)
5826 
5827 static int gfx_v9_0_set_cp_ecc_error_state(struct amdgpu_device *adev,
5828 					      struct amdgpu_irq_src *source,
5829 					      unsigned type,
5830 					      enum amdgpu_interrupt_state state)
5831 {
5832 	switch (state) {
5833 	case AMDGPU_IRQ_STATE_DISABLE:
5834 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5835 				CP_ECC_ERROR_INT_ENABLE, 0);
5836 		DISABLE_ECC_ON_ME_PIPE(1, 0);
5837 		DISABLE_ECC_ON_ME_PIPE(1, 1);
5838 		DISABLE_ECC_ON_ME_PIPE(1, 2);
5839 		DISABLE_ECC_ON_ME_PIPE(1, 3);
5840 		break;
5841 
5842 	case AMDGPU_IRQ_STATE_ENABLE:
5843 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5844 				CP_ECC_ERROR_INT_ENABLE, 1);
5845 		ENABLE_ECC_ON_ME_PIPE(1, 0);
5846 		ENABLE_ECC_ON_ME_PIPE(1, 1);
5847 		ENABLE_ECC_ON_ME_PIPE(1, 2);
5848 		ENABLE_ECC_ON_ME_PIPE(1, 3);
5849 		break;
5850 	default:
5851 		break;
5852 	}
5853 
5854 	return 0;
5855 }
5856 
5857 
5858 static int gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device *adev,
5859 					    struct amdgpu_irq_src *src,
5860 					    unsigned type,
5861 					    enum amdgpu_interrupt_state state)
5862 {
5863 	switch (type) {
5864 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
5865 		gfx_v9_0_set_gfx_eop_interrupt_state(adev, state);
5866 		break;
5867 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
5868 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
5869 		break;
5870 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
5871 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
5872 		break;
5873 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
5874 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
5875 		break;
5876 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
5877 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
5878 		break;
5879 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
5880 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
5881 		break;
5882 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
5883 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
5884 		break;
5885 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
5886 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
5887 		break;
5888 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
5889 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
5890 		break;
5891 	default:
5892 		break;
5893 	}
5894 	return 0;
5895 }
5896 
5897 static int gfx_v9_0_eop_irq(struct amdgpu_device *adev,
5898 			    struct amdgpu_irq_src *source,
5899 			    struct amdgpu_iv_entry *entry)
5900 {
5901 	int i;
5902 	u8 me_id, pipe_id, queue_id;
5903 	struct amdgpu_ring *ring;
5904 
5905 	DRM_DEBUG("IH: CP EOP\n");
5906 	me_id = (entry->ring_id & 0x0c) >> 2;
5907 	pipe_id = (entry->ring_id & 0x03) >> 0;
5908 	queue_id = (entry->ring_id & 0x70) >> 4;
5909 
5910 	switch (me_id) {
5911 	case 0:
5912 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
5913 		break;
5914 	case 1:
5915 	case 2:
5916 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5917 			ring = &adev->gfx.compute_ring[i];
5918 			/* Per-queue interrupt is supported for MEC starting from VI.
5919 			  * The interrupt can only be enabled/disabled per pipe instead of per queue.
5920 			  */
5921 			if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
5922 				amdgpu_fence_process(ring);
5923 		}
5924 		break;
5925 	}
5926 	return 0;
5927 }
5928 
5929 static void gfx_v9_0_fault(struct amdgpu_device *adev,
5930 			   struct amdgpu_iv_entry *entry)
5931 {
5932 	u8 me_id, pipe_id, queue_id;
5933 	struct amdgpu_ring *ring;
5934 	int i;
5935 
5936 	me_id = (entry->ring_id & 0x0c) >> 2;
5937 	pipe_id = (entry->ring_id & 0x03) >> 0;
5938 	queue_id = (entry->ring_id & 0x70) >> 4;
5939 
5940 	switch (me_id) {
5941 	case 0:
5942 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
5943 		break;
5944 	case 1:
5945 	case 2:
5946 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5947 			ring = &adev->gfx.compute_ring[i];
5948 			if (ring->me == me_id && ring->pipe == pipe_id &&
5949 			    ring->queue == queue_id)
5950 				drm_sched_fault(&ring->sched);
5951 		}
5952 		break;
5953 	}
5954 }
5955 
5956 static int gfx_v9_0_priv_reg_irq(struct amdgpu_device *adev,
5957 				 struct amdgpu_irq_src *source,
5958 				 struct amdgpu_iv_entry *entry)
5959 {
5960 	DRM_ERROR("Illegal register access in command stream\n");
5961 	gfx_v9_0_fault(adev, entry);
5962 	return 0;
5963 }
5964 
5965 static int gfx_v9_0_priv_inst_irq(struct amdgpu_device *adev,
5966 				  struct amdgpu_irq_src *source,
5967 				  struct amdgpu_iv_entry *entry)
5968 {
5969 	DRM_ERROR("Illegal instruction in command stream\n");
5970 	gfx_v9_0_fault(adev, entry);
5971 	return 0;
5972 }
5973 
5974 
5975 static const struct soc15_ras_field_entry gfx_v9_0_ras_fields[] = {
5976 	{ "CPC_SCRATCH", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT),
5977 	  SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, SEC_COUNT),
5978 	  SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, DED_COUNT)
5979 	},
5980 	{ "CPC_UCODE", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT),
5981 	  SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, SEC_COUNT),
5982 	  SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, DED_COUNT)
5983 	},
5984 	{ "CPF_ROQ_ME1", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
5985 	  SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME1),
5986 	  0, 0
5987 	},
5988 	{ "CPF_ROQ_ME2", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
5989 	  SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME2),
5990 	  0, 0
5991 	},
5992 	{ "CPF_TAG", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT),
5993 	  SOC15_REG_FIELD(CPF_EDC_TAG_CNT, SEC_COUNT),
5994 	  SOC15_REG_FIELD(CPF_EDC_TAG_CNT, DED_COUNT)
5995 	},
5996 	{ "CPG_DMA_ROQ", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
5997 	  SOC15_REG_FIELD(CPG_EDC_DMA_CNT, ROQ_COUNT),
5998 	  0, 0
5999 	},
6000 	{ "CPG_DMA_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
6001 	  SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_SEC_COUNT),
6002 	  SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_DED_COUNT)
6003 	},
6004 	{ "CPG_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT),
6005 	  SOC15_REG_FIELD(CPG_EDC_TAG_CNT, SEC_COUNT),
6006 	  SOC15_REG_FIELD(CPG_EDC_TAG_CNT, DED_COUNT)
6007 	},
6008 	{ "DC_CSINVOC", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT),
6009 	  SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, COUNT_ME1),
6010 	  0, 0
6011 	},
6012 	{ "DC_RESTORE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT),
6013 	  SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, COUNT_ME1),
6014 	  0, 0
6015 	},
6016 	{ "DC_STATE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT),
6017 	  SOC15_REG_FIELD(DC_EDC_STATE_CNT, COUNT_ME1),
6018 	  0, 0
6019 	},
6020 	{ "GDS_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
6021 	  SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_SEC),
6022 	  SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_DED)
6023 	},
6024 	{ "GDS_INPUT_QUEUE", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
6025 	  SOC15_REG_FIELD(GDS_EDC_CNT, GDS_INPUT_QUEUE_SED),
6026 	  0, 0
6027 	},
6028 	{ "GDS_ME0_CS_PIPE_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6029 	  SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_SEC),
6030 	  SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_DED)
6031 	},
6032 	{ "GDS_OA_PHY_PHY_CMD_RAM_MEM",
6033 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6034 	  SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_SEC),
6035 	  SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_DED)
6036 	},
6037 	{ "GDS_OA_PHY_PHY_DATA_RAM_MEM",
6038 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6039 	  SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_DATA_RAM_MEM_SED),
6040 	  0, 0
6041 	},
6042 	{ "GDS_OA_PIPE_ME1_PIPE0_PIPE_MEM",
6043 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6044 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_SEC),
6045 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_DED)
6046 	},
6047 	{ "GDS_OA_PIPE_ME1_PIPE1_PIPE_MEM",
6048 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6049 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_SEC),
6050 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_DED)
6051 	},
6052 	{ "GDS_OA_PIPE_ME1_PIPE2_PIPE_MEM",
6053 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6054 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_SEC),
6055 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_DED)
6056 	},
6057 	{ "GDS_OA_PIPE_ME1_PIPE3_PIPE_MEM",
6058 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6059 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_SEC),
6060 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_DED)
6061 	},
6062 	{ "SPI_SR_MEM", SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT),
6063 	  SOC15_REG_FIELD(SPI_EDC_CNT, SPI_SR_MEM_SED_COUNT),
6064 	  0, 0
6065 	},
6066 	{ "TA_FS_DFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6067 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_SEC_COUNT),
6068 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_DED_COUNT)
6069 	},
6070 	{ "TA_FS_AFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6071 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_SED_COUNT),
6072 	  0, 0
6073 	},
6074 	{ "TA_FL_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6075 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FL_LFIFO_SED_COUNT),
6076 	  0, 0
6077 	},
6078 	{ "TA_FX_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6079 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FX_LFIFO_SED_COUNT),
6080 	  0, 0
6081 	},
6082 	{ "TA_FS_CFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6083 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_CFIFO_SED_COUNT),
6084 	  0, 0
6085 	},
6086 	{ "TCA_HOLE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
6087 	  SOC15_REG_FIELD(TCA_EDC_CNT, HOLE_FIFO_SED_COUNT),
6088 	  0, 0
6089 	},
6090 	{ "TCA_REQ_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
6091 	  SOC15_REG_FIELD(TCA_EDC_CNT, REQ_FIFO_SED_COUNT),
6092 	  0, 0
6093 	},
6094 	{ "TCC_CACHE_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6095 	  SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_SEC_COUNT),
6096 	  SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_DED_COUNT)
6097 	},
6098 	{ "TCC_CACHE_DIRTY", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6099 	  SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_SEC_COUNT),
6100 	  SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_DED_COUNT)
6101 	},
6102 	{ "TCC_HIGH_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6103 	  SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_SEC_COUNT),
6104 	  SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_DED_COUNT)
6105 	},
6106 	{ "TCC_LOW_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6107 	  SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_SEC_COUNT),
6108 	  SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_DED_COUNT)
6109 	},
6110 	{ "TCC_SRC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6111 	  SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_SEC_COUNT),
6112 	  SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_DED_COUNT)
6113 	},
6114 	{ "TCC_IN_USE_DEC", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6115 	  SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_DEC_SED_COUNT),
6116 	  0, 0
6117 	},
6118 	{ "TCC_IN_USE_TRANSFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6119 	  SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_TRANSFER_SED_COUNT),
6120 	  0, 0
6121 	},
6122 	{ "TCC_LATENCY_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6123 	  SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_SED_COUNT),
6124 	  0, 0
6125 	},
6126 	{ "TCC_RETURN_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6127 	  SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_DATA_SED_COUNT),
6128 	  0, 0
6129 	},
6130 	{ "TCC_RETURN_CONTROL", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6131 	  SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_CONTROL_SED_COUNT),
6132 	  0, 0
6133 	},
6134 	{ "TCC_UC_ATOMIC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6135 	  SOC15_REG_FIELD(TCC_EDC_CNT, UC_ATOMIC_FIFO_SED_COUNT),
6136 	  0, 0
6137 	},
6138 	{ "TCC_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6139 	  SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_RETURN_SED_COUNT),
6140 	  0, 0
6141 	},
6142 	{ "TCC_WRITE_CACHE_READ", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6143 	  SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_CACHE_READ_SED_COUNT),
6144 	  0, 0
6145 	},
6146 	{ "TCC_SRC_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6147 	  SOC15_REG_FIELD(TCC_EDC_CNT2, SRC_FIFO_NEXT_RAM_SED_COUNT),
6148 	  0, 0
6149 	},
6150 	{ "TCC_LATENCY_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6151 	  SOC15_REG_FIELD(TCC_EDC_CNT2, LATENCY_FIFO_NEXT_RAM_SED_COUNT),
6152 	  0, 0
6153 	},
6154 	{ "TCC_CACHE_TAG_PROBE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6155 	  SOC15_REG_FIELD(TCC_EDC_CNT2, CACHE_TAG_PROBE_FIFO_SED_COUNT),
6156 	  0, 0
6157 	},
6158 	{ "TCC_WRRET_TAG_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6159 	  SOC15_REG_FIELD(TCC_EDC_CNT2, WRRET_TAG_WRITE_RETURN_SED_COUNT),
6160 	  0, 0
6161 	},
6162 	{ "TCC_ATOMIC_RETURN_BUFFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6163 	  SOC15_REG_FIELD(TCC_EDC_CNT2, ATOMIC_RETURN_BUFFER_SED_COUNT),
6164 	  0, 0
6165 	},
6166 	{ "TCI_WRITE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT),
6167 	  SOC15_REG_FIELD(TCI_EDC_CNT, WRITE_RAM_SED_COUNT),
6168 	  0, 0
6169 	},
6170 	{ "TCP_CACHE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6171 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_SEC_COUNT),
6172 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_DED_COUNT)
6173 	},
6174 	{ "TCP_LFIFO_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6175 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_SEC_COUNT),
6176 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_DED_COUNT)
6177 	},
6178 	{ "TCP_CMD_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6179 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CMD_FIFO_SED_COUNT),
6180 	  0, 0
6181 	},
6182 	{ "TCP_VM_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6183 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, VM_FIFO_SEC_COUNT),
6184 	  0, 0
6185 	},
6186 	{ "TCP_DB_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6187 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, DB_RAM_SED_COUNT),
6188 	  0, 0
6189 	},
6190 	{ "TCP_UTCL1_LFIFO0", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6191 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_SEC_COUNT),
6192 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_DED_COUNT)
6193 	},
6194 	{ "TCP_UTCL1_LFIFO1", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6195 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_SEC_COUNT),
6196 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_DED_COUNT)
6197 	},
6198 	{ "TD_SS_FIFO_LO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6199 	  SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_SEC_COUNT),
6200 	  SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_DED_COUNT)
6201 	},
6202 	{ "TD_SS_FIFO_HI", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6203 	  SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_SEC_COUNT),
6204 	  SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_DED_COUNT)
6205 	},
6206 	{ "TD_CS_FIFO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6207 	  SOC15_REG_FIELD(TD_EDC_CNT, CS_FIFO_SED_COUNT),
6208 	  0, 0
6209 	},
6210 	{ "SQ_LDS_D", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6211 	  SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_SEC_COUNT),
6212 	  SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_DED_COUNT)
6213 	},
6214 	{ "SQ_LDS_I", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6215 	  SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_SEC_COUNT),
6216 	  SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_DED_COUNT)
6217 	},
6218 	{ "SQ_SGPR", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6219 	  SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_SEC_COUNT),
6220 	  SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_DED_COUNT)
6221 	},
6222 	{ "SQ_VGPR0", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6223 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_SEC_COUNT),
6224 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_DED_COUNT)
6225 	},
6226 	{ "SQ_VGPR1", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6227 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_SEC_COUNT),
6228 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_DED_COUNT)
6229 	},
6230 	{ "SQ_VGPR2", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6231 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_SEC_COUNT),
6232 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_DED_COUNT)
6233 	},
6234 	{ "SQ_VGPR3", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6235 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_SEC_COUNT),
6236 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_DED_COUNT)
6237 	},
6238 	{ "SQC_DATA_CU0_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6239 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_SEC_COUNT),
6240 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_DED_COUNT)
6241 	},
6242 	{ "SQC_DATA_CU0_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6243 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_SEC_COUNT),
6244 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_DED_COUNT)
6245 	},
6246 	{ "SQC_DATA_CU1_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6247 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_SEC_COUNT),
6248 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_DED_COUNT)
6249 	},
6250 	{ "SQC_DATA_CU1_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6251 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_SEC_COUNT),
6252 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_DED_COUNT)
6253 	},
6254 	{ "SQC_DATA_CU2_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6255 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_SEC_COUNT),
6256 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_DED_COUNT)
6257 	},
6258 	{ "SQC_DATA_CU2_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6259 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_SEC_COUNT),
6260 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_DED_COUNT)
6261 	},
6262 	{ "SQC_INST_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6263 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_SEC_COUNT),
6264 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_DED_COUNT)
6265 	},
6266 	{ "SQC_INST_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6267 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_SEC_COUNT),
6268 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_DED_COUNT)
6269 	},
6270 	{ "SQC_DATA_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6271 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_SEC_COUNT),
6272 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_DED_COUNT)
6273 	},
6274 	{ "SQC_DATA_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6275 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_SEC_COUNT),
6276 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_DED_COUNT)
6277 	},
6278 	{ "SQC_INST_BANKA_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6279 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_UTCL1_MISS_FIFO_SED_COUNT),
6280 	  0, 0
6281 	},
6282 	{ "SQC_INST_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6283 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_MISS_FIFO_SED_COUNT),
6284 	  0, 0
6285 	},
6286 	{ "SQC_DATA_BANKA_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6287 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_HIT_FIFO_SED_COUNT),
6288 	  0, 0
6289 	},
6290 	{ "SQC_DATA_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6291 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_MISS_FIFO_SED_COUNT),
6292 	  0, 0
6293 	},
6294 	{ "SQC_DATA_BANKA_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6295 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_DIRTY_BIT_RAM_SED_COUNT),
6296 	  0, 0
6297 	},
6298 	{ "SQC_INST_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6299 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_SEC_COUNT),
6300 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_DED_COUNT)
6301 	},
6302 	{ "SQC_INST_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6303 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_SEC_COUNT),
6304 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_DED_COUNT)
6305 	},
6306 	{ "SQC_INST_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6307 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_SEC_COUNT),
6308 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_DED_COUNT)
6309 	},
6310 	{ "SQC_DATA_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6311 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_SEC_COUNT),
6312 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_DED_COUNT)
6313 	},
6314 	{ "SQC_DATA_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6315 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_SEC_COUNT),
6316 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_DED_COUNT)
6317 	},
6318 	{ "SQC_INST_BANKB_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6319 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_UTCL1_MISS_FIFO_SED_COUNT),
6320 	  0, 0
6321 	},
6322 	{ "SQC_INST_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6323 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_MISS_FIFO_SED_COUNT),
6324 	  0, 0
6325 	},
6326 	{ "SQC_DATA_BANKB_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6327 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_HIT_FIFO_SED_COUNT),
6328 	  0, 0
6329 	},
6330 	{ "SQC_DATA_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6331 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_MISS_FIFO_SED_COUNT),
6332 	  0, 0
6333 	},
6334 	{ "SQC_DATA_BANKB_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6335 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_DIRTY_BIT_RAM_SED_COUNT),
6336 	  0, 0
6337 	},
6338 	{ "EA_DRAMRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6339 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
6340 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT)
6341 	},
6342 	{ "EA_DRAMWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6343 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
6344 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT)
6345 	},
6346 	{ "EA_DRAMWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6347 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
6348 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT)
6349 	},
6350 	{ "EA_RRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6351 	  SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
6352 	  SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_DED_COUNT)
6353 	},
6354 	{ "EA_WRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6355 	  SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
6356 	  SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_DED_COUNT)
6357 	},
6358 	{ "EA_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6359 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
6360 	  0, 0
6361 	},
6362 	{ "EA_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6363 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
6364 	  0, 0
6365 	},
6366 	{ "EA_IORD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6367 	  SOC15_REG_FIELD(GCEA_EDC_CNT, IORD_CMDMEM_SED_COUNT),
6368 	  0, 0
6369 	},
6370 	{ "EA_IOWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6371 	  SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
6372 	  0, 0
6373 	},
6374 	{ "EA_IOWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6375 	  SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_DATAMEM_SED_COUNT),
6376 	  0, 0
6377 	},
6378 	{ "GMIRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6379 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
6380 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT)
6381 	},
6382 	{ "GMIWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6383 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
6384 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT)
6385 	},
6386 	{ "GMIWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6387 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
6388 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT)
6389 	},
6390 	{ "GMIRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6391 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
6392 	  0, 0
6393 	},
6394 	{ "GMIWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6395 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
6396 	  0, 0
6397 	},
6398 	{ "MAM_D0MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6399 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D0MEM_SED_COUNT),
6400 	  0, 0
6401 	},
6402 	{ "MAM_D1MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6403 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D1MEM_SED_COUNT),
6404 	  0, 0
6405 	},
6406 	{ "MAM_D2MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6407 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D2MEM_SED_COUNT),
6408 	  0, 0
6409 	},
6410 	{ "MAM_D3MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6411 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D3MEM_SED_COUNT),
6412 	  0, 0
6413 	}
6414 };
6415 
6416 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
6417 				     void *inject_if)
6418 {
6419 	struct ras_inject_if *info = (struct ras_inject_if *)inject_if;
6420 	int ret;
6421 	struct ta_ras_trigger_error_input block_info = { 0 };
6422 
6423 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6424 		return -EINVAL;
6425 
6426 	if (info->head.sub_block_index >= ARRAY_SIZE(ras_gfx_subblocks))
6427 		return -EINVAL;
6428 
6429 	if (!ras_gfx_subblocks[info->head.sub_block_index].name)
6430 		return -EPERM;
6431 
6432 	if (!(ras_gfx_subblocks[info->head.sub_block_index].hw_supported_error_type &
6433 	      info->head.type)) {
6434 		DRM_ERROR("GFX Subblock %s, hardware do not support type 0x%x\n",
6435 			ras_gfx_subblocks[info->head.sub_block_index].name,
6436 			info->head.type);
6437 		return -EPERM;
6438 	}
6439 
6440 	if (!(ras_gfx_subblocks[info->head.sub_block_index].sw_supported_error_type &
6441 	      info->head.type)) {
6442 		DRM_ERROR("GFX Subblock %s, driver do not support type 0x%x\n",
6443 			ras_gfx_subblocks[info->head.sub_block_index].name,
6444 			info->head.type);
6445 		return -EPERM;
6446 	}
6447 
6448 	block_info.block_id = amdgpu_ras_block_to_ta(info->head.block);
6449 	block_info.sub_block_index =
6450 		ras_gfx_subblocks[info->head.sub_block_index].ta_subblock;
6451 	block_info.inject_error_type = amdgpu_ras_error_to_ta(info->head.type);
6452 	block_info.address = info->address;
6453 	block_info.value = info->value;
6454 
6455 	mutex_lock(&adev->grbm_idx_mutex);
6456 	ret = psp_ras_trigger_error(&adev->psp, &block_info);
6457 	mutex_unlock(&adev->grbm_idx_mutex);
6458 
6459 	return ret;
6460 }
6461 
6462 static const char *vml2_mems[] = {
6463 	"UTC_VML2_BANK_CACHE_0_BIGK_MEM0",
6464 	"UTC_VML2_BANK_CACHE_0_BIGK_MEM1",
6465 	"UTC_VML2_BANK_CACHE_0_4K_MEM0",
6466 	"UTC_VML2_BANK_CACHE_0_4K_MEM1",
6467 	"UTC_VML2_BANK_CACHE_1_BIGK_MEM0",
6468 	"UTC_VML2_BANK_CACHE_1_BIGK_MEM1",
6469 	"UTC_VML2_BANK_CACHE_1_4K_MEM0",
6470 	"UTC_VML2_BANK_CACHE_1_4K_MEM1",
6471 	"UTC_VML2_BANK_CACHE_2_BIGK_MEM0",
6472 	"UTC_VML2_BANK_CACHE_2_BIGK_MEM1",
6473 	"UTC_VML2_BANK_CACHE_2_4K_MEM0",
6474 	"UTC_VML2_BANK_CACHE_2_4K_MEM1",
6475 	"UTC_VML2_BANK_CACHE_3_BIGK_MEM0",
6476 	"UTC_VML2_BANK_CACHE_3_BIGK_MEM1",
6477 	"UTC_VML2_BANK_CACHE_3_4K_MEM0",
6478 	"UTC_VML2_BANK_CACHE_3_4K_MEM1",
6479 };
6480 
6481 static const char *vml2_walker_mems[] = {
6482 	"UTC_VML2_CACHE_PDE0_MEM0",
6483 	"UTC_VML2_CACHE_PDE0_MEM1",
6484 	"UTC_VML2_CACHE_PDE1_MEM0",
6485 	"UTC_VML2_CACHE_PDE1_MEM1",
6486 	"UTC_VML2_CACHE_PDE2_MEM0",
6487 	"UTC_VML2_CACHE_PDE2_MEM1",
6488 	"UTC_VML2_RDIF_LOG_FIFO",
6489 };
6490 
6491 static const char *atc_l2_cache_2m_mems[] = {
6492 	"UTC_ATCL2_CACHE_2M_BANK0_WAY0_MEM",
6493 	"UTC_ATCL2_CACHE_2M_BANK0_WAY1_MEM",
6494 	"UTC_ATCL2_CACHE_2M_BANK1_WAY0_MEM",
6495 	"UTC_ATCL2_CACHE_2M_BANK1_WAY1_MEM",
6496 };
6497 
6498 static const char *atc_l2_cache_4k_mems[] = {
6499 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM0",
6500 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM1",
6501 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM2",
6502 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM3",
6503 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM4",
6504 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM5",
6505 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM6",
6506 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM7",
6507 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM0",
6508 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM1",
6509 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM2",
6510 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM3",
6511 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM4",
6512 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM5",
6513 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM6",
6514 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM7",
6515 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM0",
6516 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM1",
6517 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM2",
6518 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM3",
6519 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM4",
6520 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM5",
6521 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM6",
6522 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM7",
6523 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM0",
6524 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM1",
6525 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM2",
6526 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM3",
6527 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM4",
6528 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM5",
6529 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM6",
6530 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM7",
6531 };
6532 
6533 static int gfx_v9_0_query_utc_edc_status(struct amdgpu_device *adev,
6534 					 struct ras_err_data *err_data)
6535 {
6536 	uint32_t i, data;
6537 	uint32_t sec_count, ded_count;
6538 
6539 	WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6540 	WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0);
6541 	WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6542 	WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0);
6543 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6544 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0);
6545 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6546 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0);
6547 
6548 	for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) {
6549 		WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i);
6550 		data = RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT);
6551 
6552 		sec_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, SEC_COUNT);
6553 		if (sec_count) {
6554 			dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6555 				"SEC %d\n", i, vml2_mems[i], sec_count);
6556 			err_data->ce_count += sec_count;
6557 		}
6558 
6559 		ded_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, DED_COUNT);
6560 		if (ded_count) {
6561 			dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6562 				"DED %d\n", i, vml2_mems[i], ded_count);
6563 			err_data->ue_count += ded_count;
6564 		}
6565 	}
6566 
6567 	for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) {
6568 		WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i);
6569 		data = RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT);
6570 
6571 		sec_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6572 						SEC_COUNT);
6573 		if (sec_count) {
6574 			dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6575 				"SEC %d\n", i, vml2_walker_mems[i], sec_count);
6576 			err_data->ce_count += sec_count;
6577 		}
6578 
6579 		ded_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6580 						DED_COUNT);
6581 		if (ded_count) {
6582 			dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6583 				"DED %d\n", i, vml2_walker_mems[i], ded_count);
6584 			err_data->ue_count += ded_count;
6585 		}
6586 	}
6587 
6588 	for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) {
6589 		WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i);
6590 		data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT);
6591 
6592 		sec_count = (data & 0x00006000L) >> 0xd;
6593 		if (sec_count) {
6594 			dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6595 				"SEC %d\n", i, atc_l2_cache_2m_mems[i],
6596 				sec_count);
6597 			err_data->ce_count += sec_count;
6598 		}
6599 	}
6600 
6601 	for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) {
6602 		WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i);
6603 		data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT);
6604 
6605 		sec_count = (data & 0x00006000L) >> 0xd;
6606 		if (sec_count) {
6607 			dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6608 				"SEC %d\n", i, atc_l2_cache_4k_mems[i],
6609 				sec_count);
6610 			err_data->ce_count += sec_count;
6611 		}
6612 
6613 		ded_count = (data & 0x00018000L) >> 0xf;
6614 		if (ded_count) {
6615 			dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6616 				"DED %d\n", i, atc_l2_cache_4k_mems[i],
6617 				ded_count);
6618 			err_data->ue_count += ded_count;
6619 		}
6620 	}
6621 
6622 	WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6623 	WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6624 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6625 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6626 
6627 	return 0;
6628 }
6629 
6630 static int gfx_v9_0_ras_error_count(struct amdgpu_device *adev,
6631 	const struct soc15_reg_entry *reg,
6632 	uint32_t se_id, uint32_t inst_id, uint32_t value,
6633 	uint32_t *sec_count, uint32_t *ded_count)
6634 {
6635 	uint32_t i;
6636 	uint32_t sec_cnt, ded_cnt;
6637 
6638 	for (i = 0; i < ARRAY_SIZE(gfx_v9_0_ras_fields); i++) {
6639 		if(gfx_v9_0_ras_fields[i].reg_offset != reg->reg_offset ||
6640 			gfx_v9_0_ras_fields[i].seg != reg->seg ||
6641 			gfx_v9_0_ras_fields[i].inst != reg->inst)
6642 			continue;
6643 
6644 		sec_cnt = (value &
6645 				gfx_v9_0_ras_fields[i].sec_count_mask) >>
6646 				gfx_v9_0_ras_fields[i].sec_count_shift;
6647 		if (sec_cnt) {
6648 			dev_info(adev->dev, "GFX SubBlock %s, "
6649 				"Instance[%d][%d], SEC %d\n",
6650 				gfx_v9_0_ras_fields[i].name,
6651 				se_id, inst_id,
6652 				sec_cnt);
6653 			*sec_count += sec_cnt;
6654 		}
6655 
6656 		ded_cnt = (value &
6657 				gfx_v9_0_ras_fields[i].ded_count_mask) >>
6658 				gfx_v9_0_ras_fields[i].ded_count_shift;
6659 		if (ded_cnt) {
6660 			dev_info(adev->dev, "GFX SubBlock %s, "
6661 				"Instance[%d][%d], DED %d\n",
6662 				gfx_v9_0_ras_fields[i].name,
6663 				se_id, inst_id,
6664 				ded_cnt);
6665 			*ded_count += ded_cnt;
6666 		}
6667 	}
6668 
6669 	return 0;
6670 }
6671 
6672 static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev)
6673 {
6674 	int i, j, k;
6675 
6676 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6677 		return;
6678 
6679 	/* read back registers to clear the counters */
6680 	mutex_lock(&adev->grbm_idx_mutex);
6681 	for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) {
6682 		for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) {
6683 			for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) {
6684 				gfx_v9_0_select_se_sh(adev, j, 0x0, k);
6685 				RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i]));
6686 			}
6687 		}
6688 	}
6689 	WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, 0xe0000000);
6690 	mutex_unlock(&adev->grbm_idx_mutex);
6691 
6692 	WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6693 	WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0);
6694 	WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6695 	WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0);
6696 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6697 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0);
6698 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6699 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0);
6700 
6701 	for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) {
6702 		WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i);
6703 		RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT);
6704 	}
6705 
6706 	for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) {
6707 		WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i);
6708 		RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT);
6709 	}
6710 
6711 	for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) {
6712 		WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i);
6713 		RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT);
6714 	}
6715 
6716 	for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) {
6717 		WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i);
6718 		RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT);
6719 	}
6720 
6721 	WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6722 	WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6723 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6724 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6725 }
6726 
6727 static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
6728 					  void *ras_error_status)
6729 {
6730 	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
6731 	uint32_t sec_count = 0, ded_count = 0;
6732 	uint32_t i, j, k;
6733 	uint32_t reg_value;
6734 
6735 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6736 		return;
6737 
6738 	err_data->ue_count = 0;
6739 	err_data->ce_count = 0;
6740 
6741 	mutex_lock(&adev->grbm_idx_mutex);
6742 
6743 	for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) {
6744 		for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) {
6745 			for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) {
6746 				gfx_v9_0_select_se_sh(adev, j, 0, k);
6747 				reg_value =
6748 					RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i]));
6749 				if (reg_value)
6750 					gfx_v9_0_ras_error_count(adev,
6751 						&gfx_v9_0_edc_counter_regs[i],
6752 						j, k, reg_value,
6753 						&sec_count, &ded_count);
6754 			}
6755 		}
6756 	}
6757 
6758 	err_data->ce_count += sec_count;
6759 	err_data->ue_count += ded_count;
6760 
6761 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
6762 	mutex_unlock(&adev->grbm_idx_mutex);
6763 
6764 	gfx_v9_0_query_utc_edc_status(adev, err_data);
6765 }
6766 
6767 static void gfx_v9_0_emit_mem_sync(struct amdgpu_ring *ring)
6768 {
6769 	const unsigned int cp_coher_cntl =
6770 			PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_ICACHE_ACTION_ENA(1) |
6771 			PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_KCACHE_ACTION_ENA(1) |
6772 			PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_ACTION_ENA(1) |
6773 			PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TCL1_ACTION_ENA(1) |
6774 			PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_WB_ACTION_ENA(1);
6775 
6776 	/* ACQUIRE_MEM -make one or more surfaces valid for use by the subsequent operations */
6777 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
6778 	amdgpu_ring_write(ring, cp_coher_cntl); /* CP_COHER_CNTL */
6779 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
6780 	amdgpu_ring_write(ring, 0xffffff);  /* CP_COHER_SIZE_HI */
6781 	amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */
6782 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE_HI */
6783 	amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */
6784 }
6785 
6786 static void gfx_v9_0_emit_wave_limit_cs(struct amdgpu_ring *ring,
6787 					uint32_t pipe, bool enable)
6788 {
6789 	struct amdgpu_device *adev = ring->adev;
6790 	uint32_t val;
6791 	uint32_t wcl_cs_reg;
6792 
6793 	/* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */
6794 	val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS0_DEFAULT;
6795 
6796 	switch (pipe) {
6797 	case 0:
6798 		wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS0);
6799 		break;
6800 	case 1:
6801 		wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS1);
6802 		break;
6803 	case 2:
6804 		wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS2);
6805 		break;
6806 	case 3:
6807 		wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS3);
6808 		break;
6809 	default:
6810 		DRM_DEBUG("invalid pipe %d\n", pipe);
6811 		return;
6812 	}
6813 
6814 	amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val);
6815 
6816 }
6817 static void gfx_v9_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable)
6818 {
6819 	struct amdgpu_device *adev = ring->adev;
6820 	uint32_t val;
6821 	int i;
6822 
6823 
6824 	/* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit
6825 	 * number of gfx waves. Setting 5 bit will make sure gfx only gets
6826 	 * around 25% of gpu resources.
6827 	 */
6828 	val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT;
6829 	amdgpu_ring_emit_wreg(ring,
6830 			      SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_GFX),
6831 			      val);
6832 
6833 	/* Restrict waves for normal/low priority compute queues as well
6834 	 * to get best QoS for high priority compute jobs.
6835 	 *
6836 	 * amdgpu controls only 1st ME(0-3 CS pipes).
6837 	 */
6838 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
6839 		if (i != ring->pipe)
6840 			gfx_v9_0_emit_wave_limit_cs(ring, i, enable);
6841 
6842 	}
6843 }
6844 
6845 static const struct amd_ip_funcs gfx_v9_0_ip_funcs = {
6846 	.name = "gfx_v9_0",
6847 	.early_init = gfx_v9_0_early_init,
6848 	.late_init = gfx_v9_0_late_init,
6849 	.sw_init = gfx_v9_0_sw_init,
6850 	.sw_fini = gfx_v9_0_sw_fini,
6851 	.hw_init = gfx_v9_0_hw_init,
6852 	.hw_fini = gfx_v9_0_hw_fini,
6853 	.suspend = gfx_v9_0_suspend,
6854 	.resume = gfx_v9_0_resume,
6855 	.is_idle = gfx_v9_0_is_idle,
6856 	.wait_for_idle = gfx_v9_0_wait_for_idle,
6857 	.soft_reset = gfx_v9_0_soft_reset,
6858 	.set_clockgating_state = gfx_v9_0_set_clockgating_state,
6859 	.set_powergating_state = gfx_v9_0_set_powergating_state,
6860 	.get_clockgating_state = gfx_v9_0_get_clockgating_state,
6861 };
6862 
6863 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
6864 	.type = AMDGPU_RING_TYPE_GFX,
6865 	.align_mask = 0xff,
6866 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6867 	.support_64bit_ptrs = true,
6868 	.secure_submission_supported = true,
6869 	.vmhub = AMDGPU_GFXHUB_0,
6870 	.get_rptr = gfx_v9_0_ring_get_rptr_gfx,
6871 	.get_wptr = gfx_v9_0_ring_get_wptr_gfx,
6872 	.set_wptr = gfx_v9_0_ring_set_wptr_gfx,
6873 	.emit_frame_size = /* totally 242 maximum if 16 IBs */
6874 		5 +  /* COND_EXEC */
6875 		7 +  /* PIPELINE_SYNC */
6876 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6877 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6878 		2 + /* VM_FLUSH */
6879 		8 +  /* FENCE for VM_FLUSH */
6880 		20 + /* GDS switch */
6881 		4 + /* double SWITCH_BUFFER,
6882 		       the first COND_EXEC jump to the place just
6883 			   prior to this double SWITCH_BUFFER  */
6884 		5 + /* COND_EXEC */
6885 		7 +	 /*	HDP_flush */
6886 		4 +	 /*	VGT_flush */
6887 		14 + /*	CE_META */
6888 		31 + /*	DE_META */
6889 		3 + /* CNTX_CTRL */
6890 		5 + /* HDP_INVL */
6891 		8 + 8 + /* FENCE x2 */
6892 		2 + /* SWITCH_BUFFER */
6893 		7, /* gfx_v9_0_emit_mem_sync */
6894 	.emit_ib_size =	4, /* gfx_v9_0_ring_emit_ib_gfx */
6895 	.emit_ib = gfx_v9_0_ring_emit_ib_gfx,
6896 	.emit_fence = gfx_v9_0_ring_emit_fence,
6897 	.emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
6898 	.emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
6899 	.emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
6900 	.emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
6901 	.test_ring = gfx_v9_0_ring_test_ring,
6902 	.test_ib = gfx_v9_0_ring_test_ib,
6903 	.insert_nop = amdgpu_ring_insert_nop,
6904 	.pad_ib = amdgpu_ring_generic_pad_ib,
6905 	.emit_switch_buffer = gfx_v9_ring_emit_sb,
6906 	.emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
6907 	.init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
6908 	.patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec,
6909 	.emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl,
6910 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
6911 	.emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6912 	.emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6913 	.soft_recovery = gfx_v9_0_ring_soft_recovery,
6914 	.emit_mem_sync = gfx_v9_0_emit_mem_sync,
6915 };
6916 
6917 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
6918 	.type = AMDGPU_RING_TYPE_COMPUTE,
6919 	.align_mask = 0xff,
6920 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6921 	.support_64bit_ptrs = true,
6922 	.vmhub = AMDGPU_GFXHUB_0,
6923 	.get_rptr = gfx_v9_0_ring_get_rptr_compute,
6924 	.get_wptr = gfx_v9_0_ring_get_wptr_compute,
6925 	.set_wptr = gfx_v9_0_ring_set_wptr_compute,
6926 	.emit_frame_size =
6927 		20 + /* gfx_v9_0_ring_emit_gds_switch */
6928 		7 + /* gfx_v9_0_ring_emit_hdp_flush */
6929 		5 + /* hdp invalidate */
6930 		7 + /* gfx_v9_0_ring_emit_pipeline_sync */
6931 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6932 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6933 		2 + /* gfx_v9_0_ring_emit_vm_flush */
6934 		8 + 8 + 8 + /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
6935 		7 + /* gfx_v9_0_emit_mem_sync */
6936 		5 + /* gfx_v9_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */
6937 		15, /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */
6938 	.emit_ib_size =	7, /* gfx_v9_0_ring_emit_ib_compute */
6939 	.emit_ib = gfx_v9_0_ring_emit_ib_compute,
6940 	.emit_fence = gfx_v9_0_ring_emit_fence,
6941 	.emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
6942 	.emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
6943 	.emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
6944 	.emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
6945 	.test_ring = gfx_v9_0_ring_test_ring,
6946 	.test_ib = gfx_v9_0_ring_test_ib,
6947 	.insert_nop = amdgpu_ring_insert_nop,
6948 	.pad_ib = amdgpu_ring_generic_pad_ib,
6949 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
6950 	.emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6951 	.emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6952 	.emit_mem_sync = gfx_v9_0_emit_mem_sync,
6953 	.emit_wave_limit = gfx_v9_0_emit_wave_limit,
6954 };
6955 
6956 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
6957 	.type = AMDGPU_RING_TYPE_KIQ,
6958 	.align_mask = 0xff,
6959 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6960 	.support_64bit_ptrs = true,
6961 	.vmhub = AMDGPU_GFXHUB_0,
6962 	.get_rptr = gfx_v9_0_ring_get_rptr_compute,
6963 	.get_wptr = gfx_v9_0_ring_get_wptr_compute,
6964 	.set_wptr = gfx_v9_0_ring_set_wptr_compute,
6965 	.emit_frame_size =
6966 		20 + /* gfx_v9_0_ring_emit_gds_switch */
6967 		7 + /* gfx_v9_0_ring_emit_hdp_flush */
6968 		5 + /* hdp invalidate */
6969 		7 + /* gfx_v9_0_ring_emit_pipeline_sync */
6970 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6971 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6972 		2 + /* gfx_v9_0_ring_emit_vm_flush */
6973 		8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */
6974 	.emit_ib_size =	7, /* gfx_v9_0_ring_emit_ib_compute */
6975 	.emit_fence = gfx_v9_0_ring_emit_fence_kiq,
6976 	.test_ring = gfx_v9_0_ring_test_ring,
6977 	.insert_nop = amdgpu_ring_insert_nop,
6978 	.pad_ib = amdgpu_ring_generic_pad_ib,
6979 	.emit_rreg = gfx_v9_0_ring_emit_rreg,
6980 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
6981 	.emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6982 	.emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6983 };
6984 
6985 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev)
6986 {
6987 	int i;
6988 
6989 	adev->gfx.kiq.ring.funcs = &gfx_v9_0_ring_funcs_kiq;
6990 
6991 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
6992 		adev->gfx.gfx_ring[i].funcs = &gfx_v9_0_ring_funcs_gfx;
6993 
6994 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
6995 		adev->gfx.compute_ring[i].funcs = &gfx_v9_0_ring_funcs_compute;
6996 }
6997 
6998 static const struct amdgpu_irq_src_funcs gfx_v9_0_eop_irq_funcs = {
6999 	.set = gfx_v9_0_set_eop_interrupt_state,
7000 	.process = gfx_v9_0_eop_irq,
7001 };
7002 
7003 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_reg_irq_funcs = {
7004 	.set = gfx_v9_0_set_priv_reg_fault_state,
7005 	.process = gfx_v9_0_priv_reg_irq,
7006 };
7007 
7008 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_inst_irq_funcs = {
7009 	.set = gfx_v9_0_set_priv_inst_fault_state,
7010 	.process = gfx_v9_0_priv_inst_irq,
7011 };
7012 
7013 static const struct amdgpu_irq_src_funcs gfx_v9_0_cp_ecc_error_irq_funcs = {
7014 	.set = gfx_v9_0_set_cp_ecc_error_state,
7015 	.process = amdgpu_gfx_cp_ecc_error_irq,
7016 };
7017 
7018 
7019 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev)
7020 {
7021 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
7022 	adev->gfx.eop_irq.funcs = &gfx_v9_0_eop_irq_funcs;
7023 
7024 	adev->gfx.priv_reg_irq.num_types = 1;
7025 	adev->gfx.priv_reg_irq.funcs = &gfx_v9_0_priv_reg_irq_funcs;
7026 
7027 	adev->gfx.priv_inst_irq.num_types = 1;
7028 	adev->gfx.priv_inst_irq.funcs = &gfx_v9_0_priv_inst_irq_funcs;
7029 
7030 	adev->gfx.cp_ecc_error_irq.num_types = 2; /*C5 ECC error and C9 FUE error*/
7031 	adev->gfx.cp_ecc_error_irq.funcs = &gfx_v9_0_cp_ecc_error_irq_funcs;
7032 }
7033 
7034 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev)
7035 {
7036 	switch (adev->ip_versions[GC_HWIP][0]) {
7037 	case IP_VERSION(9, 0, 1):
7038 	case IP_VERSION(9, 2, 1):
7039 	case IP_VERSION(9, 4, 0):
7040 	case IP_VERSION(9, 2, 2):
7041 	case IP_VERSION(9, 1, 0):
7042 	case IP_VERSION(9, 4, 1):
7043 	case IP_VERSION(9, 3, 0):
7044 	case IP_VERSION(9, 4, 2):
7045 		adev->gfx.rlc.funcs = &gfx_v9_0_rlc_funcs;
7046 		break;
7047 	default:
7048 		break;
7049 	}
7050 }
7051 
7052 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev)
7053 {
7054 	/* init asci gds info */
7055 	switch (adev->ip_versions[GC_HWIP][0]) {
7056 	case IP_VERSION(9, 0, 1):
7057 	case IP_VERSION(9, 2, 1):
7058 	case IP_VERSION(9, 4, 0):
7059 		adev->gds.gds_size = 0x10000;
7060 		break;
7061 	case IP_VERSION(9, 2, 2):
7062 	case IP_VERSION(9, 1, 0):
7063 	case IP_VERSION(9, 4, 1):
7064 		adev->gds.gds_size = 0x1000;
7065 		break;
7066 	case IP_VERSION(9, 4, 2):
7067 		/* aldebaran removed all the GDS internal memory,
7068 		 * only support GWS opcode in kernel, like barrier
7069 		 * semaphore.etc */
7070 		adev->gds.gds_size = 0;
7071 		break;
7072 	default:
7073 		adev->gds.gds_size = 0x10000;
7074 		break;
7075 	}
7076 
7077 	switch (adev->ip_versions[GC_HWIP][0]) {
7078 	case IP_VERSION(9, 0, 1):
7079 	case IP_VERSION(9, 4, 0):
7080 		adev->gds.gds_compute_max_wave_id = 0x7ff;
7081 		break;
7082 	case IP_VERSION(9, 2, 1):
7083 		adev->gds.gds_compute_max_wave_id = 0x27f;
7084 		break;
7085 	case IP_VERSION(9, 2, 2):
7086 	case IP_VERSION(9, 1, 0):
7087 		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
7088 			adev->gds.gds_compute_max_wave_id = 0x77; /* raven2 */
7089 		else
7090 			adev->gds.gds_compute_max_wave_id = 0x15f; /* raven1 */
7091 		break;
7092 	case IP_VERSION(9, 4, 1):
7093 		adev->gds.gds_compute_max_wave_id = 0xfff;
7094 		break;
7095 	case IP_VERSION(9, 4, 2):
7096 		/* deprecated for Aldebaran, no usage at all */
7097 		adev->gds.gds_compute_max_wave_id = 0;
7098 		break;
7099 	default:
7100 		/* this really depends on the chip */
7101 		adev->gds.gds_compute_max_wave_id = 0x7ff;
7102 		break;
7103 	}
7104 
7105 	adev->gds.gws_size = 64;
7106 	adev->gds.oa_size = 16;
7107 }
7108 
7109 static void gfx_v9_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
7110 						 u32 bitmap)
7111 {
7112 	u32 data;
7113 
7114 	if (!bitmap)
7115 		return;
7116 
7117 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7118 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7119 
7120 	WREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG, data);
7121 }
7122 
7123 static u32 gfx_v9_0_get_cu_active_bitmap(struct amdgpu_device *adev)
7124 {
7125 	u32 data, mask;
7126 
7127 	data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
7128 	data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
7129 
7130 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7131 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7132 
7133 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
7134 
7135 	return (~data) & mask;
7136 }
7137 
7138 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
7139 				 struct amdgpu_cu_info *cu_info)
7140 {
7141 	int i, j, k, counter, active_cu_number = 0;
7142 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
7143 	unsigned disable_masks[4 * 4];
7144 
7145 	if (!adev || !cu_info)
7146 		return -EINVAL;
7147 
7148 	/*
7149 	 * 16 comes from bitmap array size 4*4, and it can cover all gfx9 ASICs
7150 	 */
7151 	if (adev->gfx.config.max_shader_engines *
7152 		adev->gfx.config.max_sh_per_se > 16)
7153 		return -EINVAL;
7154 
7155 	amdgpu_gfx_parse_disable_cu(disable_masks,
7156 				    adev->gfx.config.max_shader_engines,
7157 				    adev->gfx.config.max_sh_per_se);
7158 
7159 	mutex_lock(&adev->grbm_idx_mutex);
7160 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
7161 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
7162 			mask = 1;
7163 			ao_bitmap = 0;
7164 			counter = 0;
7165 			gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
7166 			gfx_v9_0_set_user_cu_inactive_bitmap(
7167 				adev, disable_masks[i * adev->gfx.config.max_sh_per_se + j]);
7168 			bitmap = gfx_v9_0_get_cu_active_bitmap(adev);
7169 
7170 			/*
7171 			 * The bitmap(and ao_cu_bitmap) in cu_info structure is
7172 			 * 4x4 size array, and it's usually suitable for Vega
7173 			 * ASICs which has 4*2 SE/SH layout.
7174 			 * But for Arcturus, SE/SH layout is changed to 8*1.
7175 			 * To mostly reduce the impact, we make it compatible
7176 			 * with current bitmap array as below:
7177 			 *    SE4,SH0 --> bitmap[0][1]
7178 			 *    SE5,SH0 --> bitmap[1][1]
7179 			 *    SE6,SH0 --> bitmap[2][1]
7180 			 *    SE7,SH0 --> bitmap[3][1]
7181 			 */
7182 			cu_info->bitmap[i % 4][j + i / 4] = bitmap;
7183 
7184 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
7185 				if (bitmap & mask) {
7186 					if (counter < adev->gfx.config.max_cu_per_sh)
7187 						ao_bitmap |= mask;
7188 					counter ++;
7189 				}
7190 				mask <<= 1;
7191 			}
7192 			active_cu_number += counter;
7193 			if (i < 2 && j < 2)
7194 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
7195 			cu_info->ao_cu_bitmap[i % 4][j + i / 4] = ao_bitmap;
7196 		}
7197 	}
7198 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
7199 	mutex_unlock(&adev->grbm_idx_mutex);
7200 
7201 	cu_info->number = active_cu_number;
7202 	cu_info->ao_cu_mask = ao_cu_mask;
7203 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
7204 
7205 	return 0;
7206 }
7207 
7208 const struct amdgpu_ip_block_version gfx_v9_0_ip_block =
7209 {
7210 	.type = AMD_IP_BLOCK_TYPE_GFX,
7211 	.major = 9,
7212 	.minor = 0,
7213 	.rev = 0,
7214 	.funcs = &gfx_v9_0_ip_funcs,
7215 };
7216