xref: /linux/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c (revision cfda8617e22a8bf217a613d0b3ba3a38778443ba)
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 #include "hdp/hdp_4_0_offset.h"
42 
43 #include "soc15_common.h"
44 #include "clearstate_gfx9.h"
45 #include "v9_structs.h"
46 
47 #include "ivsrcid/gfx/irqsrcs_gfx_9_0.h"
48 
49 #include "amdgpu_ras.h"
50 
51 #define GFX9_NUM_GFX_RINGS     1
52 #define GFX9_MEC_HPD_SIZE 4096
53 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L
54 #define RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET 0x00000000L
55 
56 #define mmPWR_MISC_CNTL_STATUS					0x0183
57 #define mmPWR_MISC_CNTL_STATUS_BASE_IDX				0
58 #define PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN__SHIFT	0x0
59 #define PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT		0x1
60 #define PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK		0x00000001L
61 #define PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK		0x00000006L
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_mec2.bin");
112 MODULE_FIRMWARE("amdgpu/arcturus_rlc.bin");
113 
114 MODULE_FIRMWARE("amdgpu/renoir_ce.bin");
115 MODULE_FIRMWARE("amdgpu/renoir_pfp.bin");
116 MODULE_FIRMWARE("amdgpu/renoir_me.bin");
117 MODULE_FIRMWARE("amdgpu/renoir_mec.bin");
118 MODULE_FIRMWARE("amdgpu/renoir_mec2.bin");
119 MODULE_FIRMWARE("amdgpu/renoir_rlc.bin");
120 
121 #define mmTCP_CHAN_STEER_0_ARCT								0x0b03
122 #define mmTCP_CHAN_STEER_0_ARCT_BASE_IDX							0
123 #define mmTCP_CHAN_STEER_1_ARCT								0x0b04
124 #define mmTCP_CHAN_STEER_1_ARCT_BASE_IDX							0
125 #define mmTCP_CHAN_STEER_2_ARCT								0x0b09
126 #define mmTCP_CHAN_STEER_2_ARCT_BASE_IDX							0
127 #define mmTCP_CHAN_STEER_3_ARCT								0x0b0a
128 #define mmTCP_CHAN_STEER_3_ARCT_BASE_IDX							0
129 #define mmTCP_CHAN_STEER_4_ARCT								0x0b0b
130 #define mmTCP_CHAN_STEER_4_ARCT_BASE_IDX							0
131 #define mmTCP_CHAN_STEER_5_ARCT								0x0b0c
132 #define mmTCP_CHAN_STEER_5_ARCT_BASE_IDX							0
133 
134 struct ras_gfx_subblock_reg {
135 	const char *name;
136 	uint32_t hwip;
137 	uint32_t inst;
138 	uint32_t seg;
139 	uint32_t reg_offset;
140 	uint32_t sec_count_mask;
141 	uint32_t sec_count_shift;
142 	uint32_t ded_count_mask;
143 	uint32_t ded_count_shift;
144 };
145 
146 enum ta_ras_gfx_subblock {
147 	/*CPC*/
148 	TA_RAS_BLOCK__GFX_CPC_INDEX_START = 0,
149 	TA_RAS_BLOCK__GFX_CPC_SCRATCH = TA_RAS_BLOCK__GFX_CPC_INDEX_START,
150 	TA_RAS_BLOCK__GFX_CPC_UCODE,
151 	TA_RAS_BLOCK__GFX_DC_STATE_ME1,
152 	TA_RAS_BLOCK__GFX_DC_CSINVOC_ME1,
153 	TA_RAS_BLOCK__GFX_DC_RESTORE_ME1,
154 	TA_RAS_BLOCK__GFX_DC_STATE_ME2,
155 	TA_RAS_BLOCK__GFX_DC_CSINVOC_ME2,
156 	TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
157 	TA_RAS_BLOCK__GFX_CPC_INDEX_END = TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
158 	/* CPF*/
159 	TA_RAS_BLOCK__GFX_CPF_INDEX_START,
160 	TA_RAS_BLOCK__GFX_CPF_ROQ_ME2 = TA_RAS_BLOCK__GFX_CPF_INDEX_START,
161 	TA_RAS_BLOCK__GFX_CPF_ROQ_ME1,
162 	TA_RAS_BLOCK__GFX_CPF_TAG,
163 	TA_RAS_BLOCK__GFX_CPF_INDEX_END = TA_RAS_BLOCK__GFX_CPF_TAG,
164 	/* CPG*/
165 	TA_RAS_BLOCK__GFX_CPG_INDEX_START,
166 	TA_RAS_BLOCK__GFX_CPG_DMA_ROQ = TA_RAS_BLOCK__GFX_CPG_INDEX_START,
167 	TA_RAS_BLOCK__GFX_CPG_DMA_TAG,
168 	TA_RAS_BLOCK__GFX_CPG_TAG,
169 	TA_RAS_BLOCK__GFX_CPG_INDEX_END = TA_RAS_BLOCK__GFX_CPG_TAG,
170 	/* GDS*/
171 	TA_RAS_BLOCK__GFX_GDS_INDEX_START,
172 	TA_RAS_BLOCK__GFX_GDS_MEM = TA_RAS_BLOCK__GFX_GDS_INDEX_START,
173 	TA_RAS_BLOCK__GFX_GDS_INPUT_QUEUE,
174 	TA_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM,
175 	TA_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM,
176 	TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
177 	TA_RAS_BLOCK__GFX_GDS_INDEX_END = TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
178 	/* SPI*/
179 	TA_RAS_BLOCK__GFX_SPI_SR_MEM,
180 	/* SQ*/
181 	TA_RAS_BLOCK__GFX_SQ_INDEX_START,
182 	TA_RAS_BLOCK__GFX_SQ_SGPR = TA_RAS_BLOCK__GFX_SQ_INDEX_START,
183 	TA_RAS_BLOCK__GFX_SQ_LDS_D,
184 	TA_RAS_BLOCK__GFX_SQ_LDS_I,
185 	TA_RAS_BLOCK__GFX_SQ_VGPR, /* VGPR = SP*/
186 	TA_RAS_BLOCK__GFX_SQ_INDEX_END = TA_RAS_BLOCK__GFX_SQ_VGPR,
187 	/* SQC (3 ranges)*/
188 	TA_RAS_BLOCK__GFX_SQC_INDEX_START,
189 	/* SQC range 0*/
190 	TA_RAS_BLOCK__GFX_SQC_INDEX0_START = TA_RAS_BLOCK__GFX_SQC_INDEX_START,
191 	TA_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO =
192 		TA_RAS_BLOCK__GFX_SQC_INDEX0_START,
193 	TA_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF,
194 	TA_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO,
195 	TA_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF,
196 	TA_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO,
197 	TA_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF,
198 	TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
199 	TA_RAS_BLOCK__GFX_SQC_INDEX0_END =
200 		TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
201 	/* SQC range 1*/
202 	TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
203 	TA_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM =
204 		TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
205 	TA_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO,
206 	TA_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO,
207 	TA_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM,
208 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM,
209 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO,
210 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO,
211 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM,
212 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
213 	TA_RAS_BLOCK__GFX_SQC_INDEX1_END =
214 		TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
215 	/* SQC range 2*/
216 	TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
217 	TA_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM =
218 		TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
219 	TA_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO,
220 	TA_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO,
221 	TA_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM,
222 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM,
223 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO,
224 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO,
225 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM,
226 	TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
227 	TA_RAS_BLOCK__GFX_SQC_INDEX2_END =
228 		TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
229 	TA_RAS_BLOCK__GFX_SQC_INDEX_END = TA_RAS_BLOCK__GFX_SQC_INDEX2_END,
230 	/* TA*/
231 	TA_RAS_BLOCK__GFX_TA_INDEX_START,
232 	TA_RAS_BLOCK__GFX_TA_FS_DFIFO = TA_RAS_BLOCK__GFX_TA_INDEX_START,
233 	TA_RAS_BLOCK__GFX_TA_FS_AFIFO,
234 	TA_RAS_BLOCK__GFX_TA_FL_LFIFO,
235 	TA_RAS_BLOCK__GFX_TA_FX_LFIFO,
236 	TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
237 	TA_RAS_BLOCK__GFX_TA_INDEX_END = TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
238 	/* TCA*/
239 	TA_RAS_BLOCK__GFX_TCA_INDEX_START,
240 	TA_RAS_BLOCK__GFX_TCA_HOLE_FIFO = TA_RAS_BLOCK__GFX_TCA_INDEX_START,
241 	TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
242 	TA_RAS_BLOCK__GFX_TCA_INDEX_END = TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
243 	/* TCC (5 sub-ranges)*/
244 	TA_RAS_BLOCK__GFX_TCC_INDEX_START,
245 	/* TCC range 0*/
246 	TA_RAS_BLOCK__GFX_TCC_INDEX0_START = TA_RAS_BLOCK__GFX_TCC_INDEX_START,
247 	TA_RAS_BLOCK__GFX_TCC_CACHE_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX0_START,
248 	TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1,
249 	TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0,
250 	TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1,
251 	TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0,
252 	TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1,
253 	TA_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG,
254 	TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
255 	TA_RAS_BLOCK__GFX_TCC_INDEX0_END = TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
256 	/* TCC range 1*/
257 	TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
258 	TA_RAS_BLOCK__GFX_TCC_IN_USE_DEC = TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
259 	TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
260 	TA_RAS_BLOCK__GFX_TCC_INDEX1_END =
261 		TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
262 	/* TCC range 2*/
263 	TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
264 	TA_RAS_BLOCK__GFX_TCC_RETURN_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
265 	TA_RAS_BLOCK__GFX_TCC_RETURN_CONTROL,
266 	TA_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO,
267 	TA_RAS_BLOCK__GFX_TCC_WRITE_RETURN,
268 	TA_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ,
269 	TA_RAS_BLOCK__GFX_TCC_SRC_FIFO,
270 	TA_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM,
271 	TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
272 	TA_RAS_BLOCK__GFX_TCC_INDEX2_END =
273 		TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
274 	/* TCC range 3*/
275 	TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
276 	TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO = TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
277 	TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
278 	TA_RAS_BLOCK__GFX_TCC_INDEX3_END =
279 		TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
280 	/* TCC range 4*/
281 	TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
282 	TA_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN =
283 		TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
284 	TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
285 	TA_RAS_BLOCK__GFX_TCC_INDEX4_END =
286 		TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
287 	TA_RAS_BLOCK__GFX_TCC_INDEX_END = TA_RAS_BLOCK__GFX_TCC_INDEX4_END,
288 	/* TCI*/
289 	TA_RAS_BLOCK__GFX_TCI_WRITE_RAM,
290 	/* TCP*/
291 	TA_RAS_BLOCK__GFX_TCP_INDEX_START,
292 	TA_RAS_BLOCK__GFX_TCP_CACHE_RAM = TA_RAS_BLOCK__GFX_TCP_INDEX_START,
293 	TA_RAS_BLOCK__GFX_TCP_LFIFO_RAM,
294 	TA_RAS_BLOCK__GFX_TCP_CMD_FIFO,
295 	TA_RAS_BLOCK__GFX_TCP_VM_FIFO,
296 	TA_RAS_BLOCK__GFX_TCP_DB_RAM,
297 	TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0,
298 	TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
299 	TA_RAS_BLOCK__GFX_TCP_INDEX_END = TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
300 	/* TD*/
301 	TA_RAS_BLOCK__GFX_TD_INDEX_START,
302 	TA_RAS_BLOCK__GFX_TD_SS_FIFO_LO = TA_RAS_BLOCK__GFX_TD_INDEX_START,
303 	TA_RAS_BLOCK__GFX_TD_SS_FIFO_HI,
304 	TA_RAS_BLOCK__GFX_TD_CS_FIFO,
305 	TA_RAS_BLOCK__GFX_TD_INDEX_END = TA_RAS_BLOCK__GFX_TD_CS_FIFO,
306 	/* EA (3 sub-ranges)*/
307 	TA_RAS_BLOCK__GFX_EA_INDEX_START,
308 	/* EA range 0*/
309 	TA_RAS_BLOCK__GFX_EA_INDEX0_START = TA_RAS_BLOCK__GFX_EA_INDEX_START,
310 	TA_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM = TA_RAS_BLOCK__GFX_EA_INDEX0_START,
311 	TA_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM,
312 	TA_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM,
313 	TA_RAS_BLOCK__GFX_EA_RRET_TAGMEM,
314 	TA_RAS_BLOCK__GFX_EA_WRET_TAGMEM,
315 	TA_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM,
316 	TA_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM,
317 	TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
318 	TA_RAS_BLOCK__GFX_EA_INDEX0_END = TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
319 	/* EA range 1*/
320 	TA_RAS_BLOCK__GFX_EA_INDEX1_START,
321 	TA_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM = TA_RAS_BLOCK__GFX_EA_INDEX1_START,
322 	TA_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM,
323 	TA_RAS_BLOCK__GFX_EA_IORD_CMDMEM,
324 	TA_RAS_BLOCK__GFX_EA_IOWR_CMDMEM,
325 	TA_RAS_BLOCK__GFX_EA_IOWR_DATAMEM,
326 	TA_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM,
327 	TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
328 	TA_RAS_BLOCK__GFX_EA_INDEX1_END = TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
329 	/* EA range 2*/
330 	TA_RAS_BLOCK__GFX_EA_INDEX2_START,
331 	TA_RAS_BLOCK__GFX_EA_MAM_D0MEM = TA_RAS_BLOCK__GFX_EA_INDEX2_START,
332 	TA_RAS_BLOCK__GFX_EA_MAM_D1MEM,
333 	TA_RAS_BLOCK__GFX_EA_MAM_D2MEM,
334 	TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
335 	TA_RAS_BLOCK__GFX_EA_INDEX2_END = TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
336 	TA_RAS_BLOCK__GFX_EA_INDEX_END = TA_RAS_BLOCK__GFX_EA_INDEX2_END,
337 	/* UTC VM L2 bank*/
338 	TA_RAS_BLOCK__UTC_VML2_BANK_CACHE,
339 	/* UTC VM walker*/
340 	TA_RAS_BLOCK__UTC_VML2_WALKER,
341 	/* UTC ATC L2 2MB cache*/
342 	TA_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK,
343 	/* UTC ATC L2 4KB cache*/
344 	TA_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK,
345 	TA_RAS_BLOCK__GFX_MAX
346 };
347 
348 struct ras_gfx_subblock {
349 	unsigned char *name;
350 	int ta_subblock;
351 	int hw_supported_error_type;
352 	int sw_supported_error_type;
353 };
354 
355 #define AMDGPU_RAS_SUB_BLOCK(subblock, a, b, c, d, e, f, g, h)                             \
356 	[AMDGPU_RAS_BLOCK__##subblock] = {                                     \
357 		#subblock,                                                     \
358 		TA_RAS_BLOCK__##subblock,                                      \
359 		((a) | ((b) << 1) | ((c) << 2) | ((d) << 3)),                  \
360 		(((e) << 1) | ((f) << 3) | (g) | ((h) << 2)),                  \
361 	}
362 
363 static const struct ras_gfx_subblock ras_gfx_subblocks[] = {
364 	AMDGPU_RAS_SUB_BLOCK(GFX_CPC_SCRATCH, 0, 1, 1, 1, 1, 0, 0, 1),
365 	AMDGPU_RAS_SUB_BLOCK(GFX_CPC_UCODE, 0, 1, 1, 1, 1, 0, 0, 1),
366 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME1, 1, 0, 0, 1, 0, 0, 1, 0),
367 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME1, 1, 0, 0, 1, 0, 0, 0, 0),
368 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME1, 1, 0, 0, 1, 0, 0, 0, 0),
369 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
370 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
371 	AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
372 	AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
373 	AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME1, 1, 0, 0, 1, 0, 0, 1, 0),
374 	AMDGPU_RAS_SUB_BLOCK(GFX_CPF_TAG, 0, 1, 1, 1, 1, 0, 0, 1),
375 	AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_ROQ, 1, 0, 0, 1, 0, 0, 1, 0),
376 	AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_TAG, 0, 1, 1, 1, 0, 1, 0, 1),
377 	AMDGPU_RAS_SUB_BLOCK(GFX_CPG_TAG, 0, 1, 1, 1, 1, 1, 0, 1),
378 	AMDGPU_RAS_SUB_BLOCK(GFX_GDS_MEM, 0, 1, 1, 1, 0, 0, 0, 0),
379 	AMDGPU_RAS_SUB_BLOCK(GFX_GDS_INPUT_QUEUE, 1, 0, 0, 1, 0, 0, 0, 0),
380 	AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_CMD_RAM_MEM, 0, 1, 1, 1, 0, 0, 0,
381 			     0),
382 	AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_DATA_RAM_MEM, 1, 0, 0, 1, 0, 0, 0,
383 			     0),
384 	AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PIPE_MEM, 0, 1, 1, 1, 0, 0, 0, 0),
385 	AMDGPU_RAS_SUB_BLOCK(GFX_SPI_SR_MEM, 1, 0, 0, 1, 0, 0, 0, 0),
386 	AMDGPU_RAS_SUB_BLOCK(GFX_SQ_SGPR, 0, 1, 1, 1, 0, 0, 0, 0),
387 	AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_D, 0, 1, 1, 1, 1, 0, 0, 1),
388 	AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_I, 0, 1, 1, 1, 0, 0, 0, 0),
389 	AMDGPU_RAS_SUB_BLOCK(GFX_SQ_VGPR, 0, 1, 1, 1, 0, 0, 0, 0),
390 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 1),
391 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
392 			     0, 0),
393 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,
394 			     0),
395 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
396 			     0, 0),
397 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_UTCL1_LFIFO, 0, 1, 1, 1, 1, 0, 0,
398 			     0),
399 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
400 			     0, 0),
401 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,
402 			     0),
403 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,
404 			     1),
405 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,
406 			     0, 0, 0),
407 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
408 			     0),
409 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
410 			     0),
411 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,
412 			     0),
413 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,
414 			     0),
415 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
416 			     0),
417 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,
418 			     0, 0),
419 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
420 			     0),
421 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,
422 			     0),
423 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,
424 			     0, 0, 0),
425 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
426 			     0),
427 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
428 			     0),
429 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,
430 			     0),
431 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,
432 			     0),
433 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
434 			     0),
435 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,
436 			     0, 0),
437 	AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
438 			     0),
439 	AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_DFIFO, 0, 1, 1, 1, 1, 0, 0, 1),
440 	AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_AFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
441 	AMDGPU_RAS_SUB_BLOCK(GFX_TA_FL_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
442 	AMDGPU_RAS_SUB_BLOCK(GFX_TA_FX_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
443 	AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_CFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
444 	AMDGPU_RAS_SUB_BLOCK(GFX_TCA_HOLE_FIFO, 1, 0, 0, 1, 0, 1, 1, 0),
445 	AMDGPU_RAS_SUB_BLOCK(GFX_TCA_REQ_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
446 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA, 0, 1, 1, 1, 1, 0, 0, 1),
447 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_0_1, 0, 1, 1, 1, 1, 0, 0,
448 			     1),
449 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_0, 0, 1, 1, 1, 1, 0, 0,
450 			     1),
451 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_1, 0, 1, 1, 1, 1, 0, 0,
452 			     1),
453 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_0, 0, 1, 1, 1, 0, 0, 0,
454 			     0),
455 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_1, 0, 1, 1, 1, 0, 0, 0,
456 			     0),
457 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_HIGH_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0),
458 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LOW_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0),
459 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_DEC, 1, 0, 0, 1, 0, 0, 0, 0),
460 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_TRANSFER, 1, 0, 0, 1, 0, 0, 0, 0),
461 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_DATA, 1, 0, 0, 1, 0, 0, 0, 0),
462 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_CONTROL, 1, 0, 0, 1, 0, 0, 0, 0),
463 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_UC_ATOMIC_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
464 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_RETURN, 1, 0, 0, 1, 0, 1, 1, 0),
465 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_CACHE_READ, 1, 0, 0, 1, 0, 0, 0, 0),
466 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO, 0, 1, 1, 1, 0, 0, 0, 0),
467 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 1, 0),
468 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_TAG_PROBE_FIFO, 1, 0, 0, 1, 0, 0, 0,
469 			     0),
470 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
471 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 0,
472 			     0),
473 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRRET_TAG_WRITE_RETURN, 1, 0, 0, 1, 0, 0,
474 			     0, 0),
475 	AMDGPU_RAS_SUB_BLOCK(GFX_TCC_ATOMIC_RETURN_BUFFER, 1, 0, 0, 1, 0, 0, 0,
476 			     0),
477 	AMDGPU_RAS_SUB_BLOCK(GFX_TCI_WRITE_RAM, 1, 0, 0, 1, 0, 0, 0, 0),
478 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CACHE_RAM, 0, 1, 1, 1, 1, 0, 0, 1),
479 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_LFIFO_RAM, 0, 1, 1, 1, 0, 0, 0, 0),
480 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CMD_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
481 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_VM_FIFO, 0, 1, 1, 1, 0, 0, 0, 0),
482 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_DB_RAM, 1, 0, 0, 1, 0, 0, 0, 0),
483 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO0, 0, 1, 1, 1, 0, 0, 0, 0),
484 	AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO1, 0, 1, 1, 1, 0, 0, 0, 0),
485 	AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_LO, 0, 1, 1, 1, 1, 0, 0, 1),
486 	AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_HI, 0, 1, 1, 1, 0, 0, 0, 0),
487 	AMDGPU_RAS_SUB_BLOCK(GFX_TD_CS_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
488 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_CMDMEM, 0, 1, 1, 1, 1, 0, 0, 1),
489 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
490 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0),
491 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_RRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0),
492 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_WRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0),
493 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
494 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
495 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0),
496 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
497 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
498 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_IORD_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0),
499 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0),
500 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_DATAMEM, 1, 0, 0, 1, 0, 0, 0, 0),
501 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
502 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
503 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D0MEM, 1, 0, 0, 1, 0, 0, 0, 0),
504 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D1MEM, 1, 0, 0, 1, 0, 0, 0, 0),
505 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D2MEM, 1, 0, 0, 1, 0, 0, 0, 0),
506 	AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D3MEM, 1, 0, 0, 1, 0, 0, 0, 0),
507 	AMDGPU_RAS_SUB_BLOCK(UTC_VML2_BANK_CACHE, 0, 1, 1, 1, 0, 0, 0, 0),
508 	AMDGPU_RAS_SUB_BLOCK(UTC_VML2_WALKER, 0, 1, 1, 1, 0, 0, 0, 0),
509 	AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_2M_BANK, 1, 0, 0, 1, 0, 0, 0, 0),
510 	AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_4K_BANK, 0, 1, 1, 1, 0, 0, 0, 0),
511 };
512 
513 static const struct soc15_reg_golden golden_settings_gc_9_0[] =
514 {
515 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400),
516 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG3, 0x80000000, 0x80000000),
517 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
518 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
519 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
520 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
521 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
522 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
523 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
524 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87),
525 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f),
526 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
527 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
528 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
529 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
530 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
531 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff),
532 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
533 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
534 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
535 };
536 
537 static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] =
538 {
539 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0x0000f000, 0x00012107),
540 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
541 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
542 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
543 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
544 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
545 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x2a114042),
546 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
547 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0x00008000, 0x00048000),
548 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
549 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
550 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
551 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
552 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
553 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
554 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x01000107),
555 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x00001800, 0x00000800),
556 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080)
557 };
558 
559 static const struct soc15_reg_golden golden_settings_gc_9_0_vg20[] =
560 {
561 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x0f000080, 0x04000080),
562 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
563 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
564 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x22014042),
565 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x22014042),
566 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0x00003e00, 0x00000400),
567 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xff840000, 0x04040000),
568 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00030000),
569 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff010f, 0x01000107),
570 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0x000b0000, 0x000b0000),
571 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01000000, 0x01000000)
572 };
573 
574 static const struct soc15_reg_golden golden_settings_gc_9_1[] =
575 {
576 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
577 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
578 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
579 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
580 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
581 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
582 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
583 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
584 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
585 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
586 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
587 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
588 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
589 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
590 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
591 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
592 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
593 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120),
594 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
595 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000000ff),
596 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080),
597 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
598 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
599 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
600 };
601 
602 static const struct soc15_reg_golden golden_settings_gc_9_1_rv1[] =
603 {
604 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
605 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24000042),
606 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24000042),
607 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04048000),
608 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_MODE_CNTL_1, 0x06000000, 0x06000000),
609 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
610 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x00000800)
611 };
612 
613 static const struct soc15_reg_golden golden_settings_gc_9_1_rv2[] =
614 {
615 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0xff7fffff, 0x04000000),
616 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
617 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000),
618 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x7f0fffff, 0x08000080),
619 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0xff8fffff, 0x08000080),
620 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x7f8fffff, 0x08000080),
621 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x26013041),
622 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x26013041),
623 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x3f8fffff, 0x08000080),
624 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
625 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0xff0fffff, 0x08000080),
626 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0xff0fffff, 0x08000080),
627 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0xff0fffff, 0x08000080),
628 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0xff0fffff, 0x08000080),
629 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0xff0fffff, 0x08000080),
630 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
631 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010),
632 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000),
633 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x3f8fffff, 0x08000080),
634 };
635 
636 static const struct soc15_reg_golden golden_settings_gc_9_1_rn[] =
637 {
638 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
639 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000),
640 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400),
641 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x24000042),
642 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x24000042),
643 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
644 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
645 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
646 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
647 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
648 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120),
649 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGCEA_PROBE_MAP, 0xffffffff, 0x0000cccc),
650 };
651 
652 static const struct soc15_reg_golden golden_settings_gc_9_x_common[] =
653 {
654 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0xffffffff, 0x000001ff),
655 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_INDEX, 0xffffffff, 0x00000000),
656 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_DATA, 0xffffffff, 0x2544c382)
657 };
658 
659 static const struct soc15_reg_golden golden_settings_gc_9_2_1[] =
660 {
661 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
662 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
663 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
664 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
665 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
666 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
667 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
668 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
669 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87),
670 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f),
671 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
672 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
673 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
674 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
675 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
676 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff)
677 };
678 
679 static const struct soc15_reg_golden golden_settings_gc_9_2_1_vg12[] =
680 {
681 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x00000080, 0x04000080),
682 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
683 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
684 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24104041),
685 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24104041),
686 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
687 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff03ff, 0x01000107),
688 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
689 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x76325410),
690 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000),
691 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
692 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
693 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
694 };
695 
696 static const struct soc15_reg_golden golden_settings_gc_9_4_1_arct[] =
697 {
698 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
699 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x10b0000),
700 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_0_ARCT, 0x3fffffff, 0x346f0a4e),
701 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_1_ARCT, 0x3fffffff, 0x1c642ca),
702 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_2_ARCT, 0x3fffffff, 0x26f45098),
703 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_3_ARCT, 0x3fffffff, 0x2ebd9fe3),
704 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_4_ARCT, 0x3fffffff, 0xb90f5b1),
705 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_5_ARCT, 0x3ff, 0x135),
706 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_CONFIG, 0xffffffff, 0x011A0000),
707 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_FIFO_SIZES, 0xffffffff, 0x00000f00),
708 };
709 
710 static const u32 GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[] =
711 {
712 	mmRLC_SRM_INDEX_CNTL_ADDR_0 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
713 	mmRLC_SRM_INDEX_CNTL_ADDR_1 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
714 	mmRLC_SRM_INDEX_CNTL_ADDR_2 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
715 	mmRLC_SRM_INDEX_CNTL_ADDR_3 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
716 	mmRLC_SRM_INDEX_CNTL_ADDR_4 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
717 	mmRLC_SRM_INDEX_CNTL_ADDR_5 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
718 	mmRLC_SRM_INDEX_CNTL_ADDR_6 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
719 	mmRLC_SRM_INDEX_CNTL_ADDR_7 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
720 };
721 
722 static const u32 GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[] =
723 {
724 	mmRLC_SRM_INDEX_CNTL_DATA_0 - mmRLC_SRM_INDEX_CNTL_DATA_0,
725 	mmRLC_SRM_INDEX_CNTL_DATA_1 - mmRLC_SRM_INDEX_CNTL_DATA_0,
726 	mmRLC_SRM_INDEX_CNTL_DATA_2 - mmRLC_SRM_INDEX_CNTL_DATA_0,
727 	mmRLC_SRM_INDEX_CNTL_DATA_3 - mmRLC_SRM_INDEX_CNTL_DATA_0,
728 	mmRLC_SRM_INDEX_CNTL_DATA_4 - mmRLC_SRM_INDEX_CNTL_DATA_0,
729 	mmRLC_SRM_INDEX_CNTL_DATA_5 - mmRLC_SRM_INDEX_CNTL_DATA_0,
730 	mmRLC_SRM_INDEX_CNTL_DATA_6 - mmRLC_SRM_INDEX_CNTL_DATA_0,
731 	mmRLC_SRM_INDEX_CNTL_DATA_7 - mmRLC_SRM_INDEX_CNTL_DATA_0,
732 };
733 
734 #define VEGA10_GB_ADDR_CONFIG_GOLDEN 0x2a114042
735 #define VEGA12_GB_ADDR_CONFIG_GOLDEN 0x24104041
736 #define RAVEN_GB_ADDR_CONFIG_GOLDEN 0x24000042
737 #define RAVEN2_GB_ADDR_CONFIG_GOLDEN 0x26013041
738 
739 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev);
740 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev);
741 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev);
742 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev);
743 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
744                                  struct amdgpu_cu_info *cu_info);
745 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev);
746 static void gfx_v9_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num, u32 instance);
747 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring);
748 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring);
749 static int gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
750 					  void *ras_error_status);
751 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
752 				     void *inject_if);
753 
754 static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev)
755 {
756 	switch (adev->asic_type) {
757 	case CHIP_VEGA10:
758 		soc15_program_register_sequence(adev,
759 						golden_settings_gc_9_0,
760 						ARRAY_SIZE(golden_settings_gc_9_0));
761 		soc15_program_register_sequence(adev,
762 						golden_settings_gc_9_0_vg10,
763 						ARRAY_SIZE(golden_settings_gc_9_0_vg10));
764 		break;
765 	case CHIP_VEGA12:
766 		soc15_program_register_sequence(adev,
767 						golden_settings_gc_9_2_1,
768 						ARRAY_SIZE(golden_settings_gc_9_2_1));
769 		soc15_program_register_sequence(adev,
770 						golden_settings_gc_9_2_1_vg12,
771 						ARRAY_SIZE(golden_settings_gc_9_2_1_vg12));
772 		break;
773 	case CHIP_VEGA20:
774 		soc15_program_register_sequence(adev,
775 						golden_settings_gc_9_0,
776 						ARRAY_SIZE(golden_settings_gc_9_0));
777 		soc15_program_register_sequence(adev,
778 						golden_settings_gc_9_0_vg20,
779 						ARRAY_SIZE(golden_settings_gc_9_0_vg20));
780 		break;
781 	case CHIP_ARCTURUS:
782 		soc15_program_register_sequence(adev,
783 						golden_settings_gc_9_4_1_arct,
784 						ARRAY_SIZE(golden_settings_gc_9_4_1_arct));
785 		break;
786 	case CHIP_RAVEN:
787 		soc15_program_register_sequence(adev, golden_settings_gc_9_1,
788 						ARRAY_SIZE(golden_settings_gc_9_1));
789 		if (adev->rev_id >= 8)
790 			soc15_program_register_sequence(adev,
791 							golden_settings_gc_9_1_rv2,
792 							ARRAY_SIZE(golden_settings_gc_9_1_rv2));
793 		else
794 			soc15_program_register_sequence(adev,
795 							golden_settings_gc_9_1_rv1,
796 							ARRAY_SIZE(golden_settings_gc_9_1_rv1));
797 		break;
798 	 case CHIP_RENOIR:
799 		soc15_program_register_sequence(adev,
800 						golden_settings_gc_9_1_rn,
801 						ARRAY_SIZE(golden_settings_gc_9_1_rn));
802 		return; /* for renoir, don't need common goldensetting */
803 	default:
804 		break;
805 	}
806 
807 	if (adev->asic_type != CHIP_ARCTURUS)
808 		soc15_program_register_sequence(adev, golden_settings_gc_9_x_common,
809 						(const u32)ARRAY_SIZE(golden_settings_gc_9_x_common));
810 }
811 
812 static void gfx_v9_0_scratch_init(struct amdgpu_device *adev)
813 {
814 	adev->gfx.scratch.num_reg = 8;
815 	adev->gfx.scratch.reg_base = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0);
816 	adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
817 }
818 
819 static void gfx_v9_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
820 				       bool wc, uint32_t reg, uint32_t val)
821 {
822 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
823 	amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) |
824 				WRITE_DATA_DST_SEL(0) |
825 				(wc ? WR_CONFIRM : 0));
826 	amdgpu_ring_write(ring, reg);
827 	amdgpu_ring_write(ring, 0);
828 	amdgpu_ring_write(ring, val);
829 }
830 
831 static void gfx_v9_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
832 				  int mem_space, int opt, uint32_t addr0,
833 				  uint32_t addr1, uint32_t ref, uint32_t mask,
834 				  uint32_t inv)
835 {
836 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
837 	amdgpu_ring_write(ring,
838 				 /* memory (1) or register (0) */
839 				 (WAIT_REG_MEM_MEM_SPACE(mem_space) |
840 				 WAIT_REG_MEM_OPERATION(opt) | /* wait */
841 				 WAIT_REG_MEM_FUNCTION(3) |  /* equal */
842 				 WAIT_REG_MEM_ENGINE(eng_sel)));
843 
844 	if (mem_space)
845 		BUG_ON(addr0 & 0x3); /* Dword align */
846 	amdgpu_ring_write(ring, addr0);
847 	amdgpu_ring_write(ring, addr1);
848 	amdgpu_ring_write(ring, ref);
849 	amdgpu_ring_write(ring, mask);
850 	amdgpu_ring_write(ring, inv); /* poll interval */
851 }
852 
853 static int gfx_v9_0_ring_test_ring(struct amdgpu_ring *ring)
854 {
855 	struct amdgpu_device *adev = ring->adev;
856 	uint32_t scratch;
857 	uint32_t tmp = 0;
858 	unsigned i;
859 	int r;
860 
861 	r = amdgpu_gfx_scratch_get(adev, &scratch);
862 	if (r)
863 		return r;
864 
865 	WREG32(scratch, 0xCAFEDEAD);
866 	r = amdgpu_ring_alloc(ring, 3);
867 	if (r)
868 		goto error_free_scratch;
869 
870 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
871 	amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
872 	amdgpu_ring_write(ring, 0xDEADBEEF);
873 	amdgpu_ring_commit(ring);
874 
875 	for (i = 0; i < adev->usec_timeout; i++) {
876 		tmp = RREG32(scratch);
877 		if (tmp == 0xDEADBEEF)
878 			break;
879 		udelay(1);
880 	}
881 
882 	if (i >= adev->usec_timeout)
883 		r = -ETIMEDOUT;
884 
885 error_free_scratch:
886 	amdgpu_gfx_scratch_free(adev, scratch);
887 	return r;
888 }
889 
890 static int gfx_v9_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
891 {
892 	struct amdgpu_device *adev = ring->adev;
893 	struct amdgpu_ib ib;
894 	struct dma_fence *f = NULL;
895 
896 	unsigned index;
897 	uint64_t gpu_addr;
898 	uint32_t tmp;
899 	long r;
900 
901 	r = amdgpu_device_wb_get(adev, &index);
902 	if (r)
903 		return r;
904 
905 	gpu_addr = adev->wb.gpu_addr + (index * 4);
906 	adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
907 	memset(&ib, 0, sizeof(ib));
908 	r = amdgpu_ib_get(adev, NULL, 16, &ib);
909 	if (r)
910 		goto err1;
911 
912 	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
913 	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
914 	ib.ptr[2] = lower_32_bits(gpu_addr);
915 	ib.ptr[3] = upper_32_bits(gpu_addr);
916 	ib.ptr[4] = 0xDEADBEEF;
917 	ib.length_dw = 5;
918 
919 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
920 	if (r)
921 		goto err2;
922 
923 	r = dma_fence_wait_timeout(f, false, timeout);
924 	if (r == 0) {
925 		r = -ETIMEDOUT;
926 		goto err2;
927 	} else if (r < 0) {
928 		goto err2;
929 	}
930 
931 	tmp = adev->wb.wb[index];
932 	if (tmp == 0xDEADBEEF)
933 		r = 0;
934 	else
935 		r = -EINVAL;
936 
937 err2:
938 	amdgpu_ib_free(adev, &ib, NULL);
939 	dma_fence_put(f);
940 err1:
941 	amdgpu_device_wb_free(adev, index);
942 	return r;
943 }
944 
945 
946 static void gfx_v9_0_free_microcode(struct amdgpu_device *adev)
947 {
948 	release_firmware(adev->gfx.pfp_fw);
949 	adev->gfx.pfp_fw = NULL;
950 	release_firmware(adev->gfx.me_fw);
951 	adev->gfx.me_fw = NULL;
952 	release_firmware(adev->gfx.ce_fw);
953 	adev->gfx.ce_fw = NULL;
954 	release_firmware(adev->gfx.rlc_fw);
955 	adev->gfx.rlc_fw = NULL;
956 	release_firmware(adev->gfx.mec_fw);
957 	adev->gfx.mec_fw = NULL;
958 	release_firmware(adev->gfx.mec2_fw);
959 	adev->gfx.mec2_fw = NULL;
960 
961 	kfree(adev->gfx.rlc.register_list_format);
962 }
963 
964 static void gfx_v9_0_init_rlc_ext_microcode(struct amdgpu_device *adev)
965 {
966 	const struct rlc_firmware_header_v2_1 *rlc_hdr;
967 
968 	rlc_hdr = (const struct rlc_firmware_header_v2_1 *)adev->gfx.rlc_fw->data;
969 	adev->gfx.rlc_srlc_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_ucode_ver);
970 	adev->gfx.rlc_srlc_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_feature_ver);
971 	adev->gfx.rlc.save_restore_list_cntl_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_cntl_size_bytes);
972 	adev->gfx.rlc.save_restore_list_cntl = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_cntl_offset_bytes);
973 	adev->gfx.rlc_srlg_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_ucode_ver);
974 	adev->gfx.rlc_srlg_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_feature_ver);
975 	adev->gfx.rlc.save_restore_list_gpm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_gpm_size_bytes);
976 	adev->gfx.rlc.save_restore_list_gpm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_gpm_offset_bytes);
977 	adev->gfx.rlc_srls_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_ucode_ver);
978 	adev->gfx.rlc_srls_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_feature_ver);
979 	adev->gfx.rlc.save_restore_list_srm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_srm_size_bytes);
980 	adev->gfx.rlc.save_restore_list_srm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_srm_offset_bytes);
981 	adev->gfx.rlc.reg_list_format_direct_reg_list_length =
982 			le32_to_cpu(rlc_hdr->reg_list_format_direct_reg_list_length);
983 }
984 
985 static void gfx_v9_0_check_fw_write_wait(struct amdgpu_device *adev)
986 {
987 	adev->gfx.me_fw_write_wait = false;
988 	adev->gfx.mec_fw_write_wait = false;
989 
990 	if ((adev->gfx.mec_fw_version < 0x000001a5) ||
991 	    (adev->gfx.mec_feature_version < 46) ||
992 	    (adev->gfx.pfp_fw_version < 0x000000b7) ||
993 	    (adev->gfx.pfp_feature_version < 46))
994 		DRM_WARN_ONCE("Warning: check cp_fw_version and update it to realize \
995 			      GRBM requires 1-cycle delay in cp firmware\n");
996 
997 	switch (adev->asic_type) {
998 	case CHIP_VEGA10:
999 		if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1000 		    (adev->gfx.me_feature_version >= 42) &&
1001 		    (adev->gfx.pfp_fw_version >=  0x000000b1) &&
1002 		    (adev->gfx.pfp_feature_version >= 42))
1003 			adev->gfx.me_fw_write_wait = true;
1004 
1005 		if ((adev->gfx.mec_fw_version >=  0x00000193) &&
1006 		    (adev->gfx.mec_feature_version >= 42))
1007 			adev->gfx.mec_fw_write_wait = true;
1008 		break;
1009 	case CHIP_VEGA12:
1010 		if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1011 		    (adev->gfx.me_feature_version >= 44) &&
1012 		    (adev->gfx.pfp_fw_version >=  0x000000b2) &&
1013 		    (adev->gfx.pfp_feature_version >= 44))
1014 			adev->gfx.me_fw_write_wait = true;
1015 
1016 		if ((adev->gfx.mec_fw_version >=  0x00000196) &&
1017 		    (adev->gfx.mec_feature_version >= 44))
1018 			adev->gfx.mec_fw_write_wait = true;
1019 		break;
1020 	case CHIP_VEGA20:
1021 		if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1022 		    (adev->gfx.me_feature_version >= 44) &&
1023 		    (adev->gfx.pfp_fw_version >=  0x000000b2) &&
1024 		    (adev->gfx.pfp_feature_version >= 44))
1025 			adev->gfx.me_fw_write_wait = true;
1026 
1027 		if ((adev->gfx.mec_fw_version >=  0x00000197) &&
1028 		    (adev->gfx.mec_feature_version >= 44))
1029 			adev->gfx.mec_fw_write_wait = true;
1030 		break;
1031 	case CHIP_RAVEN:
1032 		if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1033 		    (adev->gfx.me_feature_version >= 42) &&
1034 		    (adev->gfx.pfp_fw_version >=  0x000000b1) &&
1035 		    (adev->gfx.pfp_feature_version >= 42))
1036 			adev->gfx.me_fw_write_wait = true;
1037 
1038 		if ((adev->gfx.mec_fw_version >=  0x00000192) &&
1039 		    (adev->gfx.mec_feature_version >= 42))
1040 			adev->gfx.mec_fw_write_wait = true;
1041 		break;
1042 	default:
1043 		break;
1044 	}
1045 }
1046 
1047 static void gfx_v9_0_check_if_need_gfxoff(struct amdgpu_device *adev)
1048 {
1049 	switch (adev->asic_type) {
1050 	case CHIP_VEGA10:
1051 	case CHIP_VEGA12:
1052 	case CHIP_VEGA20:
1053 		break;
1054 	case CHIP_RAVEN:
1055 		/* Disable GFXOFF on original raven.  There are combinations
1056 		 * of sbios and platforms that are not stable.
1057 		 */
1058 		if (!(adev->rev_id >= 0x8 || adev->pdev->device == 0x15d8))
1059 			adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1060 		else if (!(adev->rev_id >= 0x8 || adev->pdev->device == 0x15d8)
1061 			 &&((adev->gfx.rlc_fw_version != 106 &&
1062 			     adev->gfx.rlc_fw_version < 531) ||
1063 			    (adev->gfx.rlc_fw_version == 53815) ||
1064 			    (adev->gfx.rlc_feature_version < 1) ||
1065 			    !adev->gfx.rlc.is_rlc_v2_1))
1066 			adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1067 
1068 		if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1069 			adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG |
1070 				AMD_PG_SUPPORT_CP |
1071 				AMD_PG_SUPPORT_RLC_SMU_HS;
1072 		break;
1073 	case CHIP_RENOIR:
1074 		if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1075 			adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG |
1076 				AMD_PG_SUPPORT_CP |
1077 				AMD_PG_SUPPORT_RLC_SMU_HS;
1078 		break;
1079 	default:
1080 		break;
1081 	}
1082 }
1083 
1084 static int gfx_v9_0_init_cp_gfx_microcode(struct amdgpu_device *adev,
1085 					  const char *chip_name)
1086 {
1087 	char fw_name[30];
1088 	int err;
1089 	struct amdgpu_firmware_info *info = NULL;
1090 	const struct common_firmware_header *header = NULL;
1091 	const struct gfx_firmware_header_v1_0 *cp_hdr;
1092 
1093 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
1094 	err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1095 	if (err)
1096 		goto out;
1097 	err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
1098 	if (err)
1099 		goto out;
1100 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
1101 	adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1102 	adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1103 
1104 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1105 	err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1106 	if (err)
1107 		goto out;
1108 	err = amdgpu_ucode_validate(adev->gfx.me_fw);
1109 	if (err)
1110 		goto out;
1111 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
1112 	adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1113 	adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1114 
1115 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1116 	err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1117 	if (err)
1118 		goto out;
1119 	err = amdgpu_ucode_validate(adev->gfx.ce_fw);
1120 	if (err)
1121 		goto out;
1122 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
1123 	adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1124 	adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1125 
1126 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1127 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
1128 		info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
1129 		info->fw = adev->gfx.pfp_fw;
1130 		header = (const struct common_firmware_header *)info->fw->data;
1131 		adev->firmware.fw_size +=
1132 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1133 
1134 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
1135 		info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
1136 		info->fw = adev->gfx.me_fw;
1137 		header = (const struct common_firmware_header *)info->fw->data;
1138 		adev->firmware.fw_size +=
1139 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1140 
1141 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE];
1142 		info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
1143 		info->fw = adev->gfx.ce_fw;
1144 		header = (const struct common_firmware_header *)info->fw->data;
1145 		adev->firmware.fw_size +=
1146 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1147 	}
1148 
1149 out:
1150 	if (err) {
1151 		dev_err(adev->dev,
1152 			"gfx9: Failed to load firmware \"%s\"\n",
1153 			fw_name);
1154 		release_firmware(adev->gfx.pfp_fw);
1155 		adev->gfx.pfp_fw = NULL;
1156 		release_firmware(adev->gfx.me_fw);
1157 		adev->gfx.me_fw = NULL;
1158 		release_firmware(adev->gfx.ce_fw);
1159 		adev->gfx.ce_fw = NULL;
1160 	}
1161 	return err;
1162 }
1163 
1164 static int gfx_v9_0_init_rlc_microcode(struct amdgpu_device *adev,
1165 					  const char *chip_name)
1166 {
1167 	char fw_name[30];
1168 	int err;
1169 	struct amdgpu_firmware_info *info = NULL;
1170 	const struct common_firmware_header *header = NULL;
1171 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
1172 	unsigned int *tmp = NULL;
1173 	unsigned int i = 0;
1174 	uint16_t version_major;
1175 	uint16_t version_minor;
1176 	uint32_t smu_version;
1177 
1178 	/*
1179 	 * For Picasso && AM4 SOCKET board, we use picasso_rlc_am4.bin
1180 	 * instead of picasso_rlc.bin.
1181 	 * Judgment method:
1182 	 * PCO AM4: revision >= 0xC8 && revision <= 0xCF
1183 	 *          or revision >= 0xD8 && revision <= 0xDF
1184 	 * otherwise is PCO FP5
1185 	 */
1186 	if (!strcmp(chip_name, "picasso") &&
1187 		(((adev->pdev->revision >= 0xC8) && (adev->pdev->revision <= 0xCF)) ||
1188 		((adev->pdev->revision >= 0xD8) && (adev->pdev->revision <= 0xDF))))
1189 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc_am4.bin", chip_name);
1190 	else if (!strcmp(chip_name, "raven") && (amdgpu_pm_load_smu_firmware(adev, &smu_version) == 0) &&
1191 		(smu_version >= 0x41e2b))
1192 		/**
1193 		*SMC is loaded by SBIOS on APU and it's able to get the SMU version directly.
1194 		*/
1195 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_kicker_rlc.bin", chip_name);
1196 	else
1197 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
1198 	err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
1199 	if (err)
1200 		goto out;
1201 	err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
1202 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1203 
1204 	version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
1205 	version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
1206 	if (version_major == 2 && version_minor == 1)
1207 		adev->gfx.rlc.is_rlc_v2_1 = true;
1208 
1209 	adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
1210 	adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
1211 	adev->gfx.rlc.save_and_restore_offset =
1212 			le32_to_cpu(rlc_hdr->save_and_restore_offset);
1213 	adev->gfx.rlc.clear_state_descriptor_offset =
1214 			le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
1215 	adev->gfx.rlc.avail_scratch_ram_locations =
1216 			le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
1217 	adev->gfx.rlc.reg_restore_list_size =
1218 			le32_to_cpu(rlc_hdr->reg_restore_list_size);
1219 	adev->gfx.rlc.reg_list_format_start =
1220 			le32_to_cpu(rlc_hdr->reg_list_format_start);
1221 	adev->gfx.rlc.reg_list_format_separate_start =
1222 			le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
1223 	adev->gfx.rlc.starting_offsets_start =
1224 			le32_to_cpu(rlc_hdr->starting_offsets_start);
1225 	adev->gfx.rlc.reg_list_format_size_bytes =
1226 			le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
1227 	adev->gfx.rlc.reg_list_size_bytes =
1228 			le32_to_cpu(rlc_hdr->reg_list_size_bytes);
1229 	adev->gfx.rlc.register_list_format =
1230 			kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
1231 				adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
1232 	if (!adev->gfx.rlc.register_list_format) {
1233 		err = -ENOMEM;
1234 		goto out;
1235 	}
1236 
1237 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1238 			le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
1239 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2); i++)
1240 		adev->gfx.rlc.register_list_format[i] =	le32_to_cpu(tmp[i]);
1241 
1242 	adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
1243 
1244 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1245 			le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
1246 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_size_bytes >> 2); i++)
1247 		adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
1248 
1249 	if (adev->gfx.rlc.is_rlc_v2_1)
1250 		gfx_v9_0_init_rlc_ext_microcode(adev);
1251 
1252 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1253 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
1254 		info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
1255 		info->fw = adev->gfx.rlc_fw;
1256 		header = (const struct common_firmware_header *)info->fw->data;
1257 		adev->firmware.fw_size +=
1258 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1259 
1260 		if (adev->gfx.rlc.is_rlc_v2_1 &&
1261 		    adev->gfx.rlc.save_restore_list_cntl_size_bytes &&
1262 		    adev->gfx.rlc.save_restore_list_gpm_size_bytes &&
1263 		    adev->gfx.rlc.save_restore_list_srm_size_bytes) {
1264 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL];
1265 			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL;
1266 			info->fw = adev->gfx.rlc_fw;
1267 			adev->firmware.fw_size +=
1268 				ALIGN(adev->gfx.rlc.save_restore_list_cntl_size_bytes, PAGE_SIZE);
1269 
1270 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM];
1271 			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM;
1272 			info->fw = adev->gfx.rlc_fw;
1273 			adev->firmware.fw_size +=
1274 				ALIGN(adev->gfx.rlc.save_restore_list_gpm_size_bytes, PAGE_SIZE);
1275 
1276 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM];
1277 			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM;
1278 			info->fw = adev->gfx.rlc_fw;
1279 			adev->firmware.fw_size +=
1280 				ALIGN(adev->gfx.rlc.save_restore_list_srm_size_bytes, PAGE_SIZE);
1281 		}
1282 	}
1283 
1284 out:
1285 	if (err) {
1286 		dev_err(adev->dev,
1287 			"gfx9: Failed to load firmware \"%s\"\n",
1288 			fw_name);
1289 		release_firmware(adev->gfx.rlc_fw);
1290 		adev->gfx.rlc_fw = NULL;
1291 	}
1292 	return err;
1293 }
1294 
1295 static int gfx_v9_0_init_cp_compute_microcode(struct amdgpu_device *adev,
1296 					  const char *chip_name)
1297 {
1298 	char fw_name[30];
1299 	int err;
1300 	struct amdgpu_firmware_info *info = NULL;
1301 	const struct common_firmware_header *header = NULL;
1302 	const struct gfx_firmware_header_v1_0 *cp_hdr;
1303 
1304 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1305 	err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1306 	if (err)
1307 		goto out;
1308 	err = amdgpu_ucode_validate(adev->gfx.mec_fw);
1309 	if (err)
1310 		goto out;
1311 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1312 	adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1313 	adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1314 
1315 
1316 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1317 	err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1318 	if (!err) {
1319 		err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
1320 		if (err)
1321 			goto out;
1322 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1323 		adev->gfx.mec2_fw->data;
1324 		adev->gfx.mec2_fw_version =
1325 		le32_to_cpu(cp_hdr->header.ucode_version);
1326 		adev->gfx.mec2_feature_version =
1327 		le32_to_cpu(cp_hdr->ucode_feature_version);
1328 	} else {
1329 		err = 0;
1330 		adev->gfx.mec2_fw = NULL;
1331 	}
1332 
1333 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1334 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
1335 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
1336 		info->fw = adev->gfx.mec_fw;
1337 		header = (const struct common_firmware_header *)info->fw->data;
1338 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)info->fw->data;
1339 		adev->firmware.fw_size +=
1340 			ALIGN(le32_to_cpu(header->ucode_size_bytes) - le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
1341 
1342 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1_JT];
1343 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1_JT;
1344 		info->fw = adev->gfx.mec_fw;
1345 		adev->firmware.fw_size +=
1346 			ALIGN(le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
1347 
1348 		if (adev->gfx.mec2_fw) {
1349 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
1350 			info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
1351 			info->fw = adev->gfx.mec2_fw;
1352 			header = (const struct common_firmware_header *)info->fw->data;
1353 			cp_hdr = (const struct gfx_firmware_header_v1_0 *)info->fw->data;
1354 			adev->firmware.fw_size +=
1355 				ALIGN(le32_to_cpu(header->ucode_size_bytes) - le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
1356 
1357 			/* TODO: Determine if MEC2 JT FW loading can be removed
1358 				 for all GFX V9 asic and above */
1359 			if (adev->asic_type != CHIP_ARCTURUS &&
1360 			    adev->asic_type != CHIP_RENOIR) {
1361 				info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2_JT];
1362 				info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2_JT;
1363 				info->fw = adev->gfx.mec2_fw;
1364 				adev->firmware.fw_size +=
1365 					ALIGN(le32_to_cpu(cp_hdr->jt_size) * 4,
1366 					PAGE_SIZE);
1367 			}
1368 		}
1369 	}
1370 
1371 out:
1372 	gfx_v9_0_check_if_need_gfxoff(adev);
1373 	gfx_v9_0_check_fw_write_wait(adev);
1374 	if (err) {
1375 		dev_err(adev->dev,
1376 			"gfx9: Failed to load firmware \"%s\"\n",
1377 			fw_name);
1378 		release_firmware(adev->gfx.mec_fw);
1379 		adev->gfx.mec_fw = NULL;
1380 		release_firmware(adev->gfx.mec2_fw);
1381 		adev->gfx.mec2_fw = NULL;
1382 	}
1383 	return err;
1384 }
1385 
1386 static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
1387 {
1388 	const char *chip_name;
1389 	int r;
1390 
1391 	DRM_DEBUG("\n");
1392 
1393 	switch (adev->asic_type) {
1394 	case CHIP_VEGA10:
1395 		chip_name = "vega10";
1396 		break;
1397 	case CHIP_VEGA12:
1398 		chip_name = "vega12";
1399 		break;
1400 	case CHIP_VEGA20:
1401 		chip_name = "vega20";
1402 		break;
1403 	case CHIP_RAVEN:
1404 		if (adev->rev_id >= 8)
1405 			chip_name = "raven2";
1406 		else if (adev->pdev->device == 0x15d8)
1407 			chip_name = "picasso";
1408 		else
1409 			chip_name = "raven";
1410 		break;
1411 	case CHIP_ARCTURUS:
1412 		chip_name = "arcturus";
1413 		break;
1414 	case CHIP_RENOIR:
1415 		chip_name = "renoir";
1416 		break;
1417 	default:
1418 		BUG();
1419 	}
1420 
1421 	/* No CPG in Arcturus */
1422 	if (adev->asic_type != CHIP_ARCTURUS) {
1423 		r = gfx_v9_0_init_cp_gfx_microcode(adev, chip_name);
1424 		if (r)
1425 			return r;
1426 	}
1427 
1428 	r = gfx_v9_0_init_rlc_microcode(adev, chip_name);
1429 	if (r)
1430 		return r;
1431 
1432 	r = gfx_v9_0_init_cp_compute_microcode(adev, chip_name);
1433 	if (r)
1434 		return r;
1435 
1436 	return r;
1437 }
1438 
1439 static u32 gfx_v9_0_get_csb_size(struct amdgpu_device *adev)
1440 {
1441 	u32 count = 0;
1442 	const struct cs_section_def *sect = NULL;
1443 	const struct cs_extent_def *ext = NULL;
1444 
1445 	/* begin clear state */
1446 	count += 2;
1447 	/* context control state */
1448 	count += 3;
1449 
1450 	for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
1451 		for (ext = sect->section; ext->extent != NULL; ++ext) {
1452 			if (sect->id == SECT_CONTEXT)
1453 				count += 2 + ext->reg_count;
1454 			else
1455 				return 0;
1456 		}
1457 	}
1458 
1459 	/* end clear state */
1460 	count += 2;
1461 	/* clear state */
1462 	count += 2;
1463 
1464 	return count;
1465 }
1466 
1467 static void gfx_v9_0_get_csb_buffer(struct amdgpu_device *adev,
1468 				    volatile u32 *buffer)
1469 {
1470 	u32 count = 0, i;
1471 	const struct cs_section_def *sect = NULL;
1472 	const struct cs_extent_def *ext = NULL;
1473 
1474 	if (adev->gfx.rlc.cs_data == NULL)
1475 		return;
1476 	if (buffer == NULL)
1477 		return;
1478 
1479 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1480 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1481 
1482 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1483 	buffer[count++] = cpu_to_le32(0x80000000);
1484 	buffer[count++] = cpu_to_le32(0x80000000);
1485 
1486 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1487 		for (ext = sect->section; ext->extent != NULL; ++ext) {
1488 			if (sect->id == SECT_CONTEXT) {
1489 				buffer[count++] =
1490 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1491 				buffer[count++] = cpu_to_le32(ext->reg_index -
1492 						PACKET3_SET_CONTEXT_REG_START);
1493 				for (i = 0; i < ext->reg_count; i++)
1494 					buffer[count++] = cpu_to_le32(ext->extent[i]);
1495 			} else {
1496 				return;
1497 			}
1498 		}
1499 	}
1500 
1501 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1502 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1503 
1504 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1505 	buffer[count++] = cpu_to_le32(0);
1506 }
1507 
1508 static void gfx_v9_0_init_always_on_cu_mask(struct amdgpu_device *adev)
1509 {
1510 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
1511 	uint32_t pg_always_on_cu_num = 2;
1512 	uint32_t always_on_cu_num;
1513 	uint32_t i, j, k;
1514 	uint32_t mask, cu_bitmap, counter;
1515 
1516 	if (adev->flags & AMD_IS_APU)
1517 		always_on_cu_num = 4;
1518 	else if (adev->asic_type == CHIP_VEGA12)
1519 		always_on_cu_num = 8;
1520 	else
1521 		always_on_cu_num = 12;
1522 
1523 	mutex_lock(&adev->grbm_idx_mutex);
1524 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1525 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1526 			mask = 1;
1527 			cu_bitmap = 0;
1528 			counter = 0;
1529 			gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
1530 
1531 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
1532 				if (cu_info->bitmap[i][j] & mask) {
1533 					if (counter == pg_always_on_cu_num)
1534 						WREG32_SOC15(GC, 0, mmRLC_PG_ALWAYS_ON_CU_MASK, cu_bitmap);
1535 					if (counter < always_on_cu_num)
1536 						cu_bitmap |= mask;
1537 					else
1538 						break;
1539 					counter++;
1540 				}
1541 				mask <<= 1;
1542 			}
1543 
1544 			WREG32_SOC15(GC, 0, mmRLC_LB_ALWAYS_ACTIVE_CU_MASK, cu_bitmap);
1545 			cu_info->ao_cu_bitmap[i][j] = cu_bitmap;
1546 		}
1547 	}
1548 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1549 	mutex_unlock(&adev->grbm_idx_mutex);
1550 }
1551 
1552 static void gfx_v9_0_init_lbpw(struct amdgpu_device *adev)
1553 {
1554 	uint32_t data;
1555 
1556 	/* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1557 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F);
1558 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x0333A5A7);
1559 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077);
1560 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x30 | 0x40 << 8 | 0x02FA << 16));
1561 
1562 	/* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1563 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000);
1564 
1565 	/* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1566 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000500);
1567 
1568 	mutex_lock(&adev->grbm_idx_mutex);
1569 	/* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1570 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1571 	WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff);
1572 
1573 	/* set mmRLC_LB_PARAMS = 0x003F_1006 */
1574 	data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003);
1575 	data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010);
1576 	data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F);
1577 	WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data);
1578 
1579 	/* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1580 	data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7);
1581 	data &= 0x0000FFFF;
1582 	data |= 0x00C00000;
1583 	WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data);
1584 
1585 	/*
1586 	 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xF (4 CUs AON for Raven),
1587 	 * programmed in gfx_v9_0_init_always_on_cu_mask()
1588 	 */
1589 
1590 	/* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1591 	 * but used for RLC_LB_CNTL configuration */
1592 	data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK;
1593 	data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09);
1594 	data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000);
1595 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data);
1596 	mutex_unlock(&adev->grbm_idx_mutex);
1597 
1598 	gfx_v9_0_init_always_on_cu_mask(adev);
1599 }
1600 
1601 static void gfx_v9_4_init_lbpw(struct amdgpu_device *adev)
1602 {
1603 	uint32_t data;
1604 
1605 	/* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1606 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F);
1607 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x033388F8);
1608 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077);
1609 	WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x10 | 0x27 << 8 | 0x02FA << 16));
1610 
1611 	/* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1612 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000);
1613 
1614 	/* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1615 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000800);
1616 
1617 	mutex_lock(&adev->grbm_idx_mutex);
1618 	/* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1619 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1620 	WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff);
1621 
1622 	/* set mmRLC_LB_PARAMS = 0x003F_1006 */
1623 	data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003);
1624 	data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010);
1625 	data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F);
1626 	WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data);
1627 
1628 	/* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1629 	data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7);
1630 	data &= 0x0000FFFF;
1631 	data |= 0x00C00000;
1632 	WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data);
1633 
1634 	/*
1635 	 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xFFF (12 CUs AON),
1636 	 * programmed in gfx_v9_0_init_always_on_cu_mask()
1637 	 */
1638 
1639 	/* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1640 	 * but used for RLC_LB_CNTL configuration */
1641 	data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK;
1642 	data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09);
1643 	data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000);
1644 	WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data);
1645 	mutex_unlock(&adev->grbm_idx_mutex);
1646 
1647 	gfx_v9_0_init_always_on_cu_mask(adev);
1648 }
1649 
1650 static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
1651 {
1652 	WREG32_FIELD15(GC, 0, RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0);
1653 }
1654 
1655 static int gfx_v9_0_cp_jump_table_num(struct amdgpu_device *adev)
1656 {
1657 	return 5;
1658 }
1659 
1660 static int gfx_v9_0_rlc_init(struct amdgpu_device *adev)
1661 {
1662 	const struct cs_section_def *cs_data;
1663 	int r;
1664 
1665 	adev->gfx.rlc.cs_data = gfx9_cs_data;
1666 
1667 	cs_data = adev->gfx.rlc.cs_data;
1668 
1669 	if (cs_data) {
1670 		/* init clear state block */
1671 		r = amdgpu_gfx_rlc_init_csb(adev);
1672 		if (r)
1673 			return r;
1674 	}
1675 
1676 	if (adev->asic_type == CHIP_RAVEN || adev->asic_type == CHIP_RENOIR) {
1677 		/* TODO: double check the cp_table_size for RV */
1678 		adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1679 		r = amdgpu_gfx_rlc_init_cpt(adev);
1680 		if (r)
1681 			return r;
1682 	}
1683 
1684 	switch (adev->asic_type) {
1685 	case CHIP_RAVEN:
1686 		gfx_v9_0_init_lbpw(adev);
1687 		break;
1688 	case CHIP_VEGA20:
1689 		gfx_v9_4_init_lbpw(adev);
1690 		break;
1691 	default:
1692 		break;
1693 	}
1694 
1695 	return 0;
1696 }
1697 
1698 static void gfx_v9_0_mec_fini(struct amdgpu_device *adev)
1699 {
1700 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1701 	amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL);
1702 }
1703 
1704 static int gfx_v9_0_mec_init(struct amdgpu_device *adev)
1705 {
1706 	int r;
1707 	u32 *hpd;
1708 	const __le32 *fw_data;
1709 	unsigned fw_size;
1710 	u32 *fw;
1711 	size_t mec_hpd_size;
1712 
1713 	const struct gfx_firmware_header_v1_0 *mec_hdr;
1714 
1715 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1716 
1717 	/* take ownership of the relevant compute queues */
1718 	amdgpu_gfx_compute_queue_acquire(adev);
1719 	mec_hpd_size = adev->gfx.num_compute_rings * GFX9_MEC_HPD_SIZE;
1720 
1721 	r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1722 				      AMDGPU_GEM_DOMAIN_VRAM,
1723 				      &adev->gfx.mec.hpd_eop_obj,
1724 				      &adev->gfx.mec.hpd_eop_gpu_addr,
1725 				      (void **)&hpd);
1726 	if (r) {
1727 		dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1728 		gfx_v9_0_mec_fini(adev);
1729 		return r;
1730 	}
1731 
1732 	memset(hpd, 0, adev->gfx.mec.hpd_eop_obj->tbo.mem.size);
1733 
1734 	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1735 	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1736 
1737 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1738 
1739 	fw_data = (const __le32 *)
1740 		(adev->gfx.mec_fw->data +
1741 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
1742 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
1743 
1744 	r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
1745 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
1746 				      &adev->gfx.mec.mec_fw_obj,
1747 				      &adev->gfx.mec.mec_fw_gpu_addr,
1748 				      (void **)&fw);
1749 	if (r) {
1750 		dev_warn(adev->dev, "(%d) create mec firmware bo failed\n", r);
1751 		gfx_v9_0_mec_fini(adev);
1752 		return r;
1753 	}
1754 
1755 	memcpy(fw, fw_data, fw_size);
1756 
1757 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
1758 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
1759 
1760 	return 0;
1761 }
1762 
1763 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
1764 {
1765 	WREG32_SOC15(GC, 0, mmSQ_IND_INDEX,
1766 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1767 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
1768 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
1769 		(SQ_IND_INDEX__FORCE_READ_MASK));
1770 	return RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
1771 }
1772 
1773 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
1774 			   uint32_t wave, uint32_t thread,
1775 			   uint32_t regno, uint32_t num, uint32_t *out)
1776 {
1777 	WREG32_SOC15(GC, 0, mmSQ_IND_INDEX,
1778 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1779 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
1780 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
1781 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
1782 		(SQ_IND_INDEX__FORCE_READ_MASK) |
1783 		(SQ_IND_INDEX__AUTO_INCR_MASK));
1784 	while (num--)
1785 		*(out++) = RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
1786 }
1787 
1788 static void gfx_v9_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
1789 {
1790 	/* type 1 wave data */
1791 	dst[(*no_fields)++] = 1;
1792 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
1793 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
1794 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
1795 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
1796 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
1797 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
1798 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
1799 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
1800 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
1801 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
1802 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
1803 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
1804 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
1805 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
1806 }
1807 
1808 static void gfx_v9_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
1809 				     uint32_t wave, uint32_t start,
1810 				     uint32_t size, uint32_t *dst)
1811 {
1812 	wave_read_regs(
1813 		adev, simd, wave, 0,
1814 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
1815 }
1816 
1817 static void gfx_v9_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t simd,
1818 				     uint32_t wave, uint32_t thread,
1819 				     uint32_t start, uint32_t size,
1820 				     uint32_t *dst)
1821 {
1822 	wave_read_regs(
1823 		adev, simd, wave, thread,
1824 		start + SQIND_WAVE_VGPRS_OFFSET, size, dst);
1825 }
1826 
1827 static void gfx_v9_0_select_me_pipe_q(struct amdgpu_device *adev,
1828 				  u32 me, u32 pipe, u32 q, u32 vm)
1829 {
1830 	soc15_grbm_select(adev, me, pipe, q, vm);
1831 }
1832 
1833 static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = {
1834 	.get_gpu_clock_counter = &gfx_v9_0_get_gpu_clock_counter,
1835 	.select_se_sh = &gfx_v9_0_select_se_sh,
1836 	.read_wave_data = &gfx_v9_0_read_wave_data,
1837 	.read_wave_sgprs = &gfx_v9_0_read_wave_sgprs,
1838 	.read_wave_vgprs = &gfx_v9_0_read_wave_vgprs,
1839 	.select_me_pipe_q = &gfx_v9_0_select_me_pipe_q,
1840 	.ras_error_inject = &gfx_v9_0_ras_error_inject,
1841 	.query_ras_error_count = &gfx_v9_0_query_ras_error_count
1842 };
1843 
1844 static int gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
1845 {
1846 	u32 gb_addr_config;
1847 	int err;
1848 
1849 	adev->gfx.funcs = &gfx_v9_0_gfx_funcs;
1850 
1851 	switch (adev->asic_type) {
1852 	case CHIP_VEGA10:
1853 		adev->gfx.config.max_hw_contexts = 8;
1854 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1855 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1856 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1857 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1858 		gb_addr_config = VEGA10_GB_ADDR_CONFIG_GOLDEN;
1859 		break;
1860 	case CHIP_VEGA12:
1861 		adev->gfx.config.max_hw_contexts = 8;
1862 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1863 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1864 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1865 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1866 		gb_addr_config = VEGA12_GB_ADDR_CONFIG_GOLDEN;
1867 		DRM_INFO("fix gfx.config for vega12\n");
1868 		break;
1869 	case CHIP_VEGA20:
1870 		adev->gfx.config.max_hw_contexts = 8;
1871 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1872 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1873 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1874 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1875 		gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
1876 		gb_addr_config &= ~0xf3e777ff;
1877 		gb_addr_config |= 0x22014042;
1878 		/* check vbios table if gpu info is not available */
1879 		err = amdgpu_atomfirmware_get_gfx_info(adev);
1880 		if (err)
1881 			return err;
1882 		break;
1883 	case CHIP_RAVEN:
1884 		adev->gfx.config.max_hw_contexts = 8;
1885 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1886 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1887 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1888 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1889 		if (adev->rev_id >= 8)
1890 			gb_addr_config = RAVEN2_GB_ADDR_CONFIG_GOLDEN;
1891 		else
1892 			gb_addr_config = RAVEN_GB_ADDR_CONFIG_GOLDEN;
1893 		break;
1894 	case CHIP_ARCTURUS:
1895 		adev->gfx.config.max_hw_contexts = 8;
1896 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1897 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1898 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1899 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1900 		gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
1901 		gb_addr_config &= ~0xf3e777ff;
1902 		gb_addr_config |= 0x22014042;
1903 		break;
1904 	case CHIP_RENOIR:
1905 		adev->gfx.config.max_hw_contexts = 8;
1906 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1907 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1908 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x80;
1909 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1910 		gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
1911 		gb_addr_config &= ~0xf3e777ff;
1912 		gb_addr_config |= 0x22010042;
1913 		break;
1914 	default:
1915 		BUG();
1916 		break;
1917 	}
1918 
1919 	adev->gfx.config.gb_addr_config = gb_addr_config;
1920 
1921 	adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
1922 			REG_GET_FIELD(
1923 					adev->gfx.config.gb_addr_config,
1924 					GB_ADDR_CONFIG,
1925 					NUM_PIPES);
1926 
1927 	adev->gfx.config.max_tile_pipes =
1928 		adev->gfx.config.gb_addr_config_fields.num_pipes;
1929 
1930 	adev->gfx.config.gb_addr_config_fields.num_banks = 1 <<
1931 			REG_GET_FIELD(
1932 					adev->gfx.config.gb_addr_config,
1933 					GB_ADDR_CONFIG,
1934 					NUM_BANKS);
1935 	adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
1936 			REG_GET_FIELD(
1937 					adev->gfx.config.gb_addr_config,
1938 					GB_ADDR_CONFIG,
1939 					MAX_COMPRESSED_FRAGS);
1940 	adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
1941 			REG_GET_FIELD(
1942 					adev->gfx.config.gb_addr_config,
1943 					GB_ADDR_CONFIG,
1944 					NUM_RB_PER_SE);
1945 	adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
1946 			REG_GET_FIELD(
1947 					adev->gfx.config.gb_addr_config,
1948 					GB_ADDR_CONFIG,
1949 					NUM_SHADER_ENGINES);
1950 	adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
1951 			REG_GET_FIELD(
1952 					adev->gfx.config.gb_addr_config,
1953 					GB_ADDR_CONFIG,
1954 					PIPE_INTERLEAVE_SIZE));
1955 
1956 	return 0;
1957 }
1958 
1959 static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1960 				      int mec, int pipe, int queue)
1961 {
1962 	int r;
1963 	unsigned irq_type;
1964 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
1965 
1966 	ring = &adev->gfx.compute_ring[ring_id];
1967 
1968 	/* mec0 is me1 */
1969 	ring->me = mec + 1;
1970 	ring->pipe = pipe;
1971 	ring->queue = queue;
1972 
1973 	ring->ring_obj = NULL;
1974 	ring->use_doorbell = true;
1975 	ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1;
1976 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1977 				+ (ring_id * GFX9_MEC_HPD_SIZE);
1978 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1979 
1980 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1981 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1982 		+ ring->pipe;
1983 
1984 	/* type-2 packets are deprecated on MEC, use type-3 instead */
1985 	r = amdgpu_ring_init(adev, ring, 1024,
1986 			     &adev->gfx.eop_irq, irq_type);
1987 	if (r)
1988 		return r;
1989 
1990 
1991 	return 0;
1992 }
1993 
1994 static int gfx_v9_0_sw_init(void *handle)
1995 {
1996 	int i, j, k, r, ring_id;
1997 	struct amdgpu_ring *ring;
1998 	struct amdgpu_kiq *kiq;
1999 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2000 
2001 	switch (adev->asic_type) {
2002 	case CHIP_VEGA10:
2003 	case CHIP_VEGA12:
2004 	case CHIP_VEGA20:
2005 	case CHIP_RAVEN:
2006 	case CHIP_ARCTURUS:
2007 	case CHIP_RENOIR:
2008 		adev->gfx.mec.num_mec = 2;
2009 		break;
2010 	default:
2011 		adev->gfx.mec.num_mec = 1;
2012 		break;
2013 	}
2014 
2015 	adev->gfx.mec.num_pipe_per_mec = 4;
2016 	adev->gfx.mec.num_queue_per_pipe = 8;
2017 
2018 	/* EOP Event */
2019 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_EOP_INTERRUPT, &adev->gfx.eop_irq);
2020 	if (r)
2021 		return r;
2022 
2023 	/* Privileged reg */
2024 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_REG_FAULT,
2025 			      &adev->gfx.priv_reg_irq);
2026 	if (r)
2027 		return r;
2028 
2029 	/* Privileged inst */
2030 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_INSTR_FAULT,
2031 			      &adev->gfx.priv_inst_irq);
2032 	if (r)
2033 		return r;
2034 
2035 	/* ECC error */
2036 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_ECC_ERROR,
2037 			      &adev->gfx.cp_ecc_error_irq);
2038 	if (r)
2039 		return r;
2040 
2041 	/* FUE error */
2042 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_FUE_ERROR,
2043 			      &adev->gfx.cp_ecc_error_irq);
2044 	if (r)
2045 		return r;
2046 
2047 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
2048 
2049 	gfx_v9_0_scratch_init(adev);
2050 
2051 	r = gfx_v9_0_init_microcode(adev);
2052 	if (r) {
2053 		DRM_ERROR("Failed to load gfx firmware!\n");
2054 		return r;
2055 	}
2056 
2057 	r = adev->gfx.rlc.funcs->init(adev);
2058 	if (r) {
2059 		DRM_ERROR("Failed to init rlc BOs!\n");
2060 		return r;
2061 	}
2062 
2063 	r = gfx_v9_0_mec_init(adev);
2064 	if (r) {
2065 		DRM_ERROR("Failed to init MEC BOs!\n");
2066 		return r;
2067 	}
2068 
2069 	/* set up the gfx ring */
2070 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
2071 		ring = &adev->gfx.gfx_ring[i];
2072 		ring->ring_obj = NULL;
2073 		if (!i)
2074 			sprintf(ring->name, "gfx");
2075 		else
2076 			sprintf(ring->name, "gfx_%d", i);
2077 		ring->use_doorbell = true;
2078 		ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
2079 		r = amdgpu_ring_init(adev, ring, 1024,
2080 				     &adev->gfx.eop_irq, AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP);
2081 		if (r)
2082 			return r;
2083 	}
2084 
2085 	/* set up the compute queues - allocate horizontally across pipes */
2086 	ring_id = 0;
2087 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
2088 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2089 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2090 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
2091 					continue;
2092 
2093 				r = gfx_v9_0_compute_ring_init(adev,
2094 							       ring_id,
2095 							       i, k, j);
2096 				if (r)
2097 					return r;
2098 
2099 				ring_id++;
2100 			}
2101 		}
2102 	}
2103 
2104 	r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE);
2105 	if (r) {
2106 		DRM_ERROR("Failed to init KIQ BOs!\n");
2107 		return r;
2108 	}
2109 
2110 	kiq = &adev->gfx.kiq;
2111 	r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
2112 	if (r)
2113 		return r;
2114 
2115 	/* create MQD for all compute queues as wel as KIQ for SRIOV case */
2116 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v9_mqd_allocation));
2117 	if (r)
2118 		return r;
2119 
2120 	adev->gfx.ce_ram_size = 0x8000;
2121 
2122 	r = gfx_v9_0_gpu_early_init(adev);
2123 	if (r)
2124 		return r;
2125 
2126 	return 0;
2127 }
2128 
2129 
2130 static int gfx_v9_0_sw_fini(void *handle)
2131 {
2132 	int i;
2133 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2134 
2135 	amdgpu_gfx_ras_fini(adev);
2136 
2137 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2138 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2139 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
2140 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2141 
2142 	amdgpu_gfx_mqd_sw_fini(adev);
2143 	amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring);
2144 	amdgpu_gfx_kiq_fini(adev);
2145 
2146 	gfx_v9_0_mec_fini(adev);
2147 	amdgpu_bo_unref(&adev->gfx.rlc.clear_state_obj);
2148 	if (adev->asic_type == CHIP_RAVEN || adev->asic_type == CHIP_RENOIR) {
2149 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2150 				&adev->gfx.rlc.cp_table_gpu_addr,
2151 				(void **)&adev->gfx.rlc.cp_table_ptr);
2152 	}
2153 	gfx_v9_0_free_microcode(adev);
2154 
2155 	return 0;
2156 }
2157 
2158 
2159 static void gfx_v9_0_tiling_mode_table_init(struct amdgpu_device *adev)
2160 {
2161 	/* TODO */
2162 }
2163 
2164 static void gfx_v9_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num, u32 instance)
2165 {
2166 	u32 data;
2167 
2168 	if (instance == 0xffffffff)
2169 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
2170 	else
2171 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
2172 
2173 	if (se_num == 0xffffffff)
2174 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
2175 	else
2176 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
2177 
2178 	if (sh_num == 0xffffffff)
2179 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
2180 	else
2181 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
2182 
2183 	WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_INDEX, data);
2184 }
2185 
2186 static u32 gfx_v9_0_get_rb_active_bitmap(struct amdgpu_device *adev)
2187 {
2188 	u32 data, mask;
2189 
2190 	data = RREG32_SOC15(GC, 0, mmCC_RB_BACKEND_DISABLE);
2191 	data |= RREG32_SOC15(GC, 0, mmGC_USER_RB_BACKEND_DISABLE);
2192 
2193 	data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
2194 	data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
2195 
2196 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
2197 					 adev->gfx.config.max_sh_per_se);
2198 
2199 	return (~data) & mask;
2200 }
2201 
2202 static void gfx_v9_0_setup_rb(struct amdgpu_device *adev)
2203 {
2204 	int i, j;
2205 	u32 data;
2206 	u32 active_rbs = 0;
2207 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
2208 					adev->gfx.config.max_sh_per_se;
2209 
2210 	mutex_lock(&adev->grbm_idx_mutex);
2211 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2212 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2213 			gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
2214 			data = gfx_v9_0_get_rb_active_bitmap(adev);
2215 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
2216 					       rb_bitmap_width_per_sh);
2217 		}
2218 	}
2219 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
2220 	mutex_unlock(&adev->grbm_idx_mutex);
2221 
2222 	adev->gfx.config.backend_enable_mask = active_rbs;
2223 	adev->gfx.config.num_rbs = hweight32(active_rbs);
2224 }
2225 
2226 #define DEFAULT_SH_MEM_BASES	(0x6000)
2227 #define FIRST_COMPUTE_VMID	(8)
2228 #define LAST_COMPUTE_VMID	(16)
2229 static void gfx_v9_0_init_compute_vmid(struct amdgpu_device *adev)
2230 {
2231 	int i;
2232 	uint32_t sh_mem_config;
2233 	uint32_t sh_mem_bases;
2234 
2235 	/*
2236 	 * Configure apertures:
2237 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
2238 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
2239 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
2240 	 */
2241 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
2242 
2243 	sh_mem_config = SH_MEM_ADDRESS_MODE_64 |
2244 			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
2245 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
2246 
2247 	mutex_lock(&adev->srbm_mutex);
2248 	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
2249 		soc15_grbm_select(adev, 0, 0, 0, i);
2250 		/* CP and shaders */
2251 		WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, sh_mem_config);
2252 		WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, sh_mem_bases);
2253 	}
2254 	soc15_grbm_select(adev, 0, 0, 0, 0);
2255 	mutex_unlock(&adev->srbm_mutex);
2256 
2257 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
2258 	   acccess. These should be enabled by FW for target VMIDs. */
2259 	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
2260 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * i, 0);
2261 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * i, 0);
2262 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, i, 0);
2263 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, i, 0);
2264 	}
2265 }
2266 
2267 static void gfx_v9_0_init_gds_vmid(struct amdgpu_device *adev)
2268 {
2269 	int vmid;
2270 
2271 	/*
2272 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
2273 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
2274 	 * the driver can enable them for graphics. VMID0 should maintain
2275 	 * access so that HWS firmware can save/restore entries.
2276 	 */
2277 	for (vmid = 1; vmid < 16; vmid++) {
2278 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * vmid, 0);
2279 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * vmid, 0);
2280 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, vmid, 0);
2281 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, vmid, 0);
2282 	}
2283 }
2284 
2285 static void gfx_v9_0_constants_init(struct amdgpu_device *adev)
2286 {
2287 	u32 tmp;
2288 	int i;
2289 
2290 	WREG32_FIELD15_RLC(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff);
2291 
2292 	gfx_v9_0_tiling_mode_table_init(adev);
2293 
2294 	gfx_v9_0_setup_rb(adev);
2295 	gfx_v9_0_get_cu_info(adev, &adev->gfx.cu_info);
2296 	adev->gfx.config.db_debug2 = RREG32_SOC15(GC, 0, mmDB_DEBUG2);
2297 
2298 	/* XXX SH_MEM regs */
2299 	/* where to put LDS, scratch, GPUVM in FSA64 space */
2300 	mutex_lock(&adev->srbm_mutex);
2301 	for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_0].num_ids; i++) {
2302 		soc15_grbm_select(adev, 0, 0, 0, i);
2303 		/* CP and shaders */
2304 		if (i == 0) {
2305 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
2306 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
2307 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
2308 					    !!amdgpu_noretry);
2309 			WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp);
2310 			WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, 0);
2311 		} else {
2312 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
2313 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
2314 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
2315 					    !!amdgpu_noretry);
2316 			WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp);
2317 			tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
2318 				(adev->gmc.private_aperture_start >> 48));
2319 			tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
2320 				(adev->gmc.shared_aperture_start >> 48));
2321 			WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, tmp);
2322 		}
2323 	}
2324 	soc15_grbm_select(adev, 0, 0, 0, 0);
2325 
2326 	mutex_unlock(&adev->srbm_mutex);
2327 
2328 	gfx_v9_0_init_compute_vmid(adev);
2329 	gfx_v9_0_init_gds_vmid(adev);
2330 }
2331 
2332 static void gfx_v9_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
2333 {
2334 	u32 i, j, k;
2335 	u32 mask;
2336 
2337 	mutex_lock(&adev->grbm_idx_mutex);
2338 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2339 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2340 			gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
2341 			for (k = 0; k < adev->usec_timeout; k++) {
2342 				if (RREG32_SOC15(GC, 0, mmRLC_SERDES_CU_MASTER_BUSY) == 0)
2343 					break;
2344 				udelay(1);
2345 			}
2346 			if (k == adev->usec_timeout) {
2347 				gfx_v9_0_select_se_sh(adev, 0xffffffff,
2348 						      0xffffffff, 0xffffffff);
2349 				mutex_unlock(&adev->grbm_idx_mutex);
2350 				DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
2351 					 i, j);
2352 				return;
2353 			}
2354 		}
2355 	}
2356 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
2357 	mutex_unlock(&adev->grbm_idx_mutex);
2358 
2359 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
2360 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
2361 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
2362 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
2363 	for (k = 0; k < adev->usec_timeout; k++) {
2364 		if ((RREG32_SOC15(GC, 0, mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
2365 			break;
2366 		udelay(1);
2367 	}
2368 }
2369 
2370 static void gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
2371 					       bool enable)
2372 {
2373 	u32 tmp = RREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0);
2374 
2375 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
2376 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
2377 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
2378 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
2379 
2380 	WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp);
2381 }
2382 
2383 static void gfx_v9_0_init_csb(struct amdgpu_device *adev)
2384 {
2385 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
2386 	/* csib */
2387 	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_HI),
2388 			adev->gfx.rlc.clear_state_gpu_addr >> 32);
2389 	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_LO),
2390 			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
2391 	WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_LENGTH),
2392 			adev->gfx.rlc.clear_state_size);
2393 }
2394 
2395 static void gfx_v9_1_parse_ind_reg_list(int *register_list_format,
2396 				int indirect_offset,
2397 				int list_size,
2398 				int *unique_indirect_regs,
2399 				int unique_indirect_reg_count,
2400 				int *indirect_start_offsets,
2401 				int *indirect_start_offsets_count,
2402 				int max_start_offsets_count)
2403 {
2404 	int idx;
2405 
2406 	for (; indirect_offset < list_size; indirect_offset++) {
2407 		WARN_ON(*indirect_start_offsets_count >= max_start_offsets_count);
2408 		indirect_start_offsets[*indirect_start_offsets_count] = indirect_offset;
2409 		*indirect_start_offsets_count = *indirect_start_offsets_count + 1;
2410 
2411 		while (register_list_format[indirect_offset] != 0xFFFFFFFF) {
2412 			indirect_offset += 2;
2413 
2414 			/* look for the matching indice */
2415 			for (idx = 0; idx < unique_indirect_reg_count; idx++) {
2416 				if (unique_indirect_regs[idx] ==
2417 					register_list_format[indirect_offset] ||
2418 					!unique_indirect_regs[idx])
2419 					break;
2420 			}
2421 
2422 			BUG_ON(idx >= unique_indirect_reg_count);
2423 
2424 			if (!unique_indirect_regs[idx])
2425 				unique_indirect_regs[idx] = register_list_format[indirect_offset];
2426 
2427 			indirect_offset++;
2428 		}
2429 	}
2430 }
2431 
2432 static int gfx_v9_1_init_rlc_save_restore_list(struct amdgpu_device *adev)
2433 {
2434 	int unique_indirect_regs[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2435 	int unique_indirect_reg_count = 0;
2436 
2437 	int indirect_start_offsets[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2438 	int indirect_start_offsets_count = 0;
2439 
2440 	int list_size = 0;
2441 	int i = 0, j = 0;
2442 	u32 tmp = 0;
2443 
2444 	u32 *register_list_format =
2445 		kmemdup(adev->gfx.rlc.register_list_format,
2446 			adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
2447 	if (!register_list_format)
2448 		return -ENOMEM;
2449 
2450 	/* setup unique_indirect_regs array and indirect_start_offsets array */
2451 	unique_indirect_reg_count = ARRAY_SIZE(unique_indirect_regs);
2452 	gfx_v9_1_parse_ind_reg_list(register_list_format,
2453 				    adev->gfx.rlc.reg_list_format_direct_reg_list_length,
2454 				    adev->gfx.rlc.reg_list_format_size_bytes >> 2,
2455 				    unique_indirect_regs,
2456 				    unique_indirect_reg_count,
2457 				    indirect_start_offsets,
2458 				    &indirect_start_offsets_count,
2459 				    ARRAY_SIZE(indirect_start_offsets));
2460 
2461 	/* enable auto inc in case it is disabled */
2462 	tmp = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL));
2463 	tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK;
2464 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL), tmp);
2465 
2466 	/* write register_restore table to offset 0x0 using RLC_SRM_ARAM_ADDR/DATA */
2467 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_ADDR),
2468 		RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET);
2469 	for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
2470 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_DATA),
2471 			adev->gfx.rlc.register_restore[i]);
2472 
2473 	/* load indirect register */
2474 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2475 		adev->gfx.rlc.reg_list_format_start);
2476 
2477 	/* direct register portion */
2478 	for (i = 0; i < adev->gfx.rlc.reg_list_format_direct_reg_list_length; i++)
2479 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
2480 			register_list_format[i]);
2481 
2482 	/* indirect register portion */
2483 	while (i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2)) {
2484 		if (register_list_format[i] == 0xFFFFFFFF) {
2485 			WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2486 			continue;
2487 		}
2488 
2489 		WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2490 		WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2491 
2492 		for (j = 0; j < unique_indirect_reg_count; j++) {
2493 			if (register_list_format[i] == unique_indirect_regs[j]) {
2494 				WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, j);
2495 				break;
2496 			}
2497 		}
2498 
2499 		BUG_ON(j >= unique_indirect_reg_count);
2500 
2501 		i++;
2502 	}
2503 
2504 	/* set save/restore list size */
2505 	list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
2506 	list_size = list_size >> 1;
2507 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2508 		adev->gfx.rlc.reg_restore_list_size);
2509 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), list_size);
2510 
2511 	/* write the starting offsets to RLC scratch ram */
2512 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2513 		adev->gfx.rlc.starting_offsets_start);
2514 	for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
2515 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
2516 		       indirect_start_offsets[i]);
2517 
2518 	/* load unique indirect regs*/
2519 	for (i = 0; i < ARRAY_SIZE(unique_indirect_regs); i++) {
2520 		if (unique_indirect_regs[i] != 0) {
2521 			WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_ADDR_0)
2522 			       + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i],
2523 			       unique_indirect_regs[i] & 0x3FFFF);
2524 
2525 			WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_DATA_0)
2526 			       + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i],
2527 			       unique_indirect_regs[i] >> 20);
2528 		}
2529 	}
2530 
2531 	kfree(register_list_format);
2532 	return 0;
2533 }
2534 
2535 static void gfx_v9_0_enable_save_restore_machine(struct amdgpu_device *adev)
2536 {
2537 	WREG32_FIELD15(GC, 0, RLC_SRM_CNTL, SRM_ENABLE, 1);
2538 }
2539 
2540 static void pwr_10_0_gfxip_control_over_cgpg(struct amdgpu_device *adev,
2541 					     bool enable)
2542 {
2543 	uint32_t data = 0;
2544 	uint32_t default_data = 0;
2545 
2546 	default_data = data = RREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS));
2547 	if (enable == true) {
2548 		/* enable GFXIP control over CGPG */
2549 		data |= PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
2550 		if(default_data != data)
2551 			WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2552 
2553 		/* update status */
2554 		data &= ~PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK;
2555 		data |= (2 << PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT);
2556 		if(default_data != data)
2557 			WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2558 	} else {
2559 		/* restore GFXIP control over GCPG */
2560 		data &= ~PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
2561 		if(default_data != data)
2562 			WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2563 	}
2564 }
2565 
2566 static void gfx_v9_0_init_gfx_power_gating(struct amdgpu_device *adev)
2567 {
2568 	uint32_t data = 0;
2569 
2570 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
2571 			      AMD_PG_SUPPORT_GFX_SMG |
2572 			      AMD_PG_SUPPORT_GFX_DMG)) {
2573 		/* init IDLE_POLL_COUNT = 60 */
2574 		data = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL));
2575 		data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
2576 		data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
2577 		WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL), data);
2578 
2579 		/* init RLC PG Delay */
2580 		data = 0;
2581 		data |= (0x10 << RLC_PG_DELAY__POWER_UP_DELAY__SHIFT);
2582 		data |= (0x10 << RLC_PG_DELAY__POWER_DOWN_DELAY__SHIFT);
2583 		data |= (0x10 << RLC_PG_DELAY__CMD_PROPAGATE_DELAY__SHIFT);
2584 		data |= (0x40 << RLC_PG_DELAY__MEM_SLEEP_DELAY__SHIFT);
2585 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY), data);
2586 
2587 		data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2));
2588 		data &= ~RLC_PG_DELAY_2__SERDES_CMD_DELAY_MASK;
2589 		data |= (0x4 << RLC_PG_DELAY_2__SERDES_CMD_DELAY__SHIFT);
2590 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2), data);
2591 
2592 		data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3));
2593 		data &= ~RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG_MASK;
2594 		data |= (0xff << RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG__SHIFT);
2595 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3), data);
2596 
2597 		data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL));
2598 		data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
2599 
2600 		/* program GRBM_REG_SAVE_GFX_IDLE_THRESHOLD to 0x55f0 */
2601 		data |= (0x55f0 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
2602 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL), data);
2603 
2604 		pwr_10_0_gfxip_control_over_cgpg(adev, true);
2605 	}
2606 }
2607 
2608 static void gfx_v9_0_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
2609 						bool enable)
2610 {
2611 	uint32_t data = 0;
2612 	uint32_t default_data = 0;
2613 
2614 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2615 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2616 			     SMU_CLK_SLOWDOWN_ON_PU_ENABLE,
2617 			     enable ? 1 : 0);
2618 	if (default_data != data)
2619 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2620 }
2621 
2622 static void gfx_v9_0_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
2623 						bool enable)
2624 {
2625 	uint32_t data = 0;
2626 	uint32_t default_data = 0;
2627 
2628 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2629 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2630 			     SMU_CLK_SLOWDOWN_ON_PD_ENABLE,
2631 			     enable ? 1 : 0);
2632 	if(default_data != data)
2633 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2634 }
2635 
2636 static void gfx_v9_0_enable_cp_power_gating(struct amdgpu_device *adev,
2637 					bool enable)
2638 {
2639 	uint32_t data = 0;
2640 	uint32_t default_data = 0;
2641 
2642 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2643 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2644 			     CP_PG_DISABLE,
2645 			     enable ? 0 : 1);
2646 	if(default_data != data)
2647 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2648 }
2649 
2650 static void gfx_v9_0_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
2651 						bool enable)
2652 {
2653 	uint32_t data, default_data;
2654 
2655 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2656 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2657 			     GFX_POWER_GATING_ENABLE,
2658 			     enable ? 1 : 0);
2659 	if(default_data != data)
2660 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2661 }
2662 
2663 static void gfx_v9_0_enable_gfx_pipeline_powergating(struct amdgpu_device *adev,
2664 						bool enable)
2665 {
2666 	uint32_t data, default_data;
2667 
2668 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2669 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2670 			     GFX_PIPELINE_PG_ENABLE,
2671 			     enable ? 1 : 0);
2672 	if(default_data != data)
2673 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2674 
2675 	if (!enable)
2676 		/* read any GFX register to wake up GFX */
2677 		data = RREG32(SOC15_REG_OFFSET(GC, 0, mmDB_RENDER_CONTROL));
2678 }
2679 
2680 static void gfx_v9_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
2681 						       bool enable)
2682 {
2683 	uint32_t data, default_data;
2684 
2685 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2686 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2687 			     STATIC_PER_CU_PG_ENABLE,
2688 			     enable ? 1 : 0);
2689 	if(default_data != data)
2690 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2691 }
2692 
2693 static void gfx_v9_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
2694 						bool enable)
2695 {
2696 	uint32_t data, default_data;
2697 
2698 	default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2699 	data = REG_SET_FIELD(data, RLC_PG_CNTL,
2700 			     DYN_PER_CU_PG_ENABLE,
2701 			     enable ? 1 : 0);
2702 	if(default_data != data)
2703 		WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2704 }
2705 
2706 static void gfx_v9_0_init_pg(struct amdgpu_device *adev)
2707 {
2708 	gfx_v9_0_init_csb(adev);
2709 
2710 	/*
2711 	 * Rlc save restore list is workable since v2_1.
2712 	 * And it's needed by gfxoff feature.
2713 	 */
2714 	if (adev->gfx.rlc.is_rlc_v2_1) {
2715 		if (adev->asic_type == CHIP_VEGA12 ||
2716 		    (adev->asic_type == CHIP_RAVEN &&
2717 		     adev->rev_id >= 8))
2718 			gfx_v9_1_init_rlc_save_restore_list(adev);
2719 		gfx_v9_0_enable_save_restore_machine(adev);
2720 	}
2721 
2722 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
2723 			      AMD_PG_SUPPORT_GFX_SMG |
2724 			      AMD_PG_SUPPORT_GFX_DMG |
2725 			      AMD_PG_SUPPORT_CP |
2726 			      AMD_PG_SUPPORT_GDS |
2727 			      AMD_PG_SUPPORT_RLC_SMU_HS)) {
2728 		WREG32(mmRLC_JUMP_TABLE_RESTORE,
2729 		       adev->gfx.rlc.cp_table_gpu_addr >> 8);
2730 		gfx_v9_0_init_gfx_power_gating(adev);
2731 	}
2732 }
2733 
2734 void gfx_v9_0_rlc_stop(struct amdgpu_device *adev)
2735 {
2736 	WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 0);
2737 	gfx_v9_0_enable_gui_idle_interrupt(adev, false);
2738 	gfx_v9_0_wait_for_rlc_serdes(adev);
2739 }
2740 
2741 static void gfx_v9_0_rlc_reset(struct amdgpu_device *adev)
2742 {
2743 	WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
2744 	udelay(50);
2745 	WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
2746 	udelay(50);
2747 }
2748 
2749 static void gfx_v9_0_rlc_start(struct amdgpu_device *adev)
2750 {
2751 #ifdef AMDGPU_RLC_DEBUG_RETRY
2752 	u32 rlc_ucode_ver;
2753 #endif
2754 
2755 	WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1);
2756 	udelay(50);
2757 
2758 	/* carrizo do enable cp interrupt after cp inited */
2759 	if (!(adev->flags & AMD_IS_APU)) {
2760 		gfx_v9_0_enable_gui_idle_interrupt(adev, true);
2761 		udelay(50);
2762 	}
2763 
2764 #ifdef AMDGPU_RLC_DEBUG_RETRY
2765 	/* RLC_GPM_GENERAL_6 : RLC Ucode version */
2766 	rlc_ucode_ver = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_6);
2767 	if(rlc_ucode_ver == 0x108) {
2768 		DRM_INFO("Using rlc debug ucode. mmRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n",
2769 				rlc_ucode_ver, adev->gfx.rlc_fw_version);
2770 		/* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles,
2771 		 * default is 0x9C4 to create a 100us interval */
2772 		WREG32_SOC15(GC, 0, mmRLC_GPM_TIMER_INT_3, 0x9C4);
2773 		/* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr
2774 		 * to disable the page fault retry interrupts, default is
2775 		 * 0x100 (256) */
2776 		WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_12, 0x100);
2777 	}
2778 #endif
2779 }
2780 
2781 static int gfx_v9_0_rlc_load_microcode(struct amdgpu_device *adev)
2782 {
2783 	const struct rlc_firmware_header_v2_0 *hdr;
2784 	const __le32 *fw_data;
2785 	unsigned i, fw_size;
2786 
2787 	if (!adev->gfx.rlc_fw)
2788 		return -EINVAL;
2789 
2790 	hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
2791 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
2792 
2793 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
2794 			   le32_to_cpu(hdr->header.ucode_array_offset_bytes));
2795 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
2796 
2797 	WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR,
2798 			RLCG_UCODE_LOADING_START_ADDRESS);
2799 	for (i = 0; i < fw_size; i++)
2800 		WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
2801 	WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
2802 
2803 	return 0;
2804 }
2805 
2806 static int gfx_v9_0_rlc_resume(struct amdgpu_device *adev)
2807 {
2808 	int r;
2809 
2810 	if (amdgpu_sriov_vf(adev)) {
2811 		gfx_v9_0_init_csb(adev);
2812 		return 0;
2813 	}
2814 
2815 	adev->gfx.rlc.funcs->stop(adev);
2816 
2817 	/* disable CG */
2818 	WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, 0);
2819 
2820 	gfx_v9_0_init_pg(adev);
2821 
2822 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
2823 		/* legacy rlc firmware loading */
2824 		r = gfx_v9_0_rlc_load_microcode(adev);
2825 		if (r)
2826 			return r;
2827 	}
2828 
2829 	switch (adev->asic_type) {
2830 	case CHIP_RAVEN:
2831 		if (amdgpu_lbpw == 0)
2832 			gfx_v9_0_enable_lbpw(adev, false);
2833 		else
2834 			gfx_v9_0_enable_lbpw(adev, true);
2835 		break;
2836 	case CHIP_VEGA20:
2837 		if (amdgpu_lbpw > 0)
2838 			gfx_v9_0_enable_lbpw(adev, true);
2839 		else
2840 			gfx_v9_0_enable_lbpw(adev, false);
2841 		break;
2842 	default:
2843 		break;
2844 	}
2845 
2846 	adev->gfx.rlc.funcs->start(adev);
2847 
2848 	return 0;
2849 }
2850 
2851 static void gfx_v9_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2852 {
2853 	int i;
2854 	u32 tmp = RREG32_SOC15(GC, 0, mmCP_ME_CNTL);
2855 
2856 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1);
2857 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1);
2858 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, enable ? 0 : 1);
2859 	if (!enable) {
2860 		for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2861 			adev->gfx.gfx_ring[i].sched.ready = false;
2862 	}
2863 	WREG32_SOC15_RLC(GC, 0, mmCP_ME_CNTL, tmp);
2864 	udelay(50);
2865 }
2866 
2867 static int gfx_v9_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2868 {
2869 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2870 	const struct gfx_firmware_header_v1_0 *ce_hdr;
2871 	const struct gfx_firmware_header_v1_0 *me_hdr;
2872 	const __le32 *fw_data;
2873 	unsigned i, fw_size;
2874 
2875 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2876 		return -EINVAL;
2877 
2878 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)
2879 		adev->gfx.pfp_fw->data;
2880 	ce_hdr = (const struct gfx_firmware_header_v1_0 *)
2881 		adev->gfx.ce_fw->data;
2882 	me_hdr = (const struct gfx_firmware_header_v1_0 *)
2883 		adev->gfx.me_fw->data;
2884 
2885 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2886 	amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2887 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2888 
2889 	gfx_v9_0_cp_gfx_enable(adev, false);
2890 
2891 	/* PFP */
2892 	fw_data = (const __le32 *)
2893 		(adev->gfx.pfp_fw->data +
2894 		 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2895 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2896 	WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, 0);
2897 	for (i = 0; i < fw_size; i++)
2898 		WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2899 	WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2900 
2901 	/* CE */
2902 	fw_data = (const __le32 *)
2903 		(adev->gfx.ce_fw->data +
2904 		 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2905 	fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
2906 	WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, 0);
2907 	for (i = 0; i < fw_size; i++)
2908 		WREG32_SOC15(GC, 0, mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
2909 	WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
2910 
2911 	/* ME */
2912 	fw_data = (const __le32 *)
2913 		(adev->gfx.me_fw->data +
2914 		 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2915 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
2916 	WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, 0);
2917 	for (i = 0; i < fw_size; i++)
2918 		WREG32_SOC15(GC, 0, mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
2919 	WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
2920 
2921 	return 0;
2922 }
2923 
2924 static int gfx_v9_0_cp_gfx_start(struct amdgpu_device *adev)
2925 {
2926 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
2927 	const struct cs_section_def *sect = NULL;
2928 	const struct cs_extent_def *ext = NULL;
2929 	int r, i, tmp;
2930 
2931 	/* init the CP */
2932 	WREG32_SOC15(GC, 0, mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
2933 	WREG32_SOC15(GC, 0, mmCP_DEVICE_ID, 1);
2934 
2935 	gfx_v9_0_cp_gfx_enable(adev, true);
2936 
2937 	r = amdgpu_ring_alloc(ring, gfx_v9_0_get_csb_size(adev) + 4 + 3);
2938 	if (r) {
2939 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2940 		return r;
2941 	}
2942 
2943 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2944 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2945 
2946 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2947 	amdgpu_ring_write(ring, 0x80000000);
2948 	amdgpu_ring_write(ring, 0x80000000);
2949 
2950 	for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
2951 		for (ext = sect->section; ext->extent != NULL; ++ext) {
2952 			if (sect->id == SECT_CONTEXT) {
2953 				amdgpu_ring_write(ring,
2954 				       PACKET3(PACKET3_SET_CONTEXT_REG,
2955 					       ext->reg_count));
2956 				amdgpu_ring_write(ring,
2957 				       ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
2958 				for (i = 0; i < ext->reg_count; i++)
2959 					amdgpu_ring_write(ring, ext->extent[i]);
2960 			}
2961 		}
2962 	}
2963 
2964 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2965 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2966 
2967 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2968 	amdgpu_ring_write(ring, 0);
2969 
2970 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2971 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2972 	amdgpu_ring_write(ring, 0x8000);
2973 	amdgpu_ring_write(ring, 0x8000);
2974 
2975 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG,1));
2976 	tmp = (PACKET3_SET_UCONFIG_REG_INDEX_TYPE |
2977 		(SOC15_REG_OFFSET(GC, 0, mmVGT_INDEX_TYPE) - PACKET3_SET_UCONFIG_REG_START));
2978 	amdgpu_ring_write(ring, tmp);
2979 	amdgpu_ring_write(ring, 0);
2980 
2981 	amdgpu_ring_commit(ring);
2982 
2983 	return 0;
2984 }
2985 
2986 static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev)
2987 {
2988 	struct amdgpu_ring *ring;
2989 	u32 tmp;
2990 	u32 rb_bufsz;
2991 	u64 rb_addr, rptr_addr, wptr_gpu_addr;
2992 
2993 	/* Set the write pointer delay */
2994 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_DELAY, 0);
2995 
2996 	/* set the RB to use vmid 0 */
2997 	WREG32_SOC15(GC, 0, mmCP_RB_VMID, 0);
2998 
2999 	/* Set ring buffer size */
3000 	ring = &adev->gfx.gfx_ring[0];
3001 	rb_bufsz = order_base_2(ring->ring_size / 8);
3002 	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
3003 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
3004 #ifdef __BIG_ENDIAN
3005 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
3006 #endif
3007 	WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3008 
3009 	/* Initialize the ring buffer's write pointers */
3010 	ring->wptr = 0;
3011 	WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
3012 	WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
3013 
3014 	/* set the wb address wether it's enabled or not */
3015 	rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
3016 	WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
3017 	WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3018 
3019 	wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
3020 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
3021 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
3022 
3023 	mdelay(1);
3024 	WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3025 
3026 	rb_addr = ring->gpu_addr >> 8;
3027 	WREG32_SOC15(GC, 0, mmCP_RB0_BASE, rb_addr);
3028 	WREG32_SOC15(GC, 0, mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
3029 
3030 	tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL);
3031 	if (ring->use_doorbell) {
3032 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3033 				    DOORBELL_OFFSET, ring->doorbell_index);
3034 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3035 				    DOORBELL_EN, 1);
3036 	} else {
3037 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
3038 	}
3039 	WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, tmp);
3040 
3041 	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
3042 			DOORBELL_RANGE_LOWER, ring->doorbell_index);
3043 	WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
3044 
3045 	WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER,
3046 		       CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
3047 
3048 
3049 	/* start the ring */
3050 	gfx_v9_0_cp_gfx_start(adev);
3051 	ring->sched.ready = true;
3052 
3053 	return 0;
3054 }
3055 
3056 static void gfx_v9_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
3057 {
3058 	int i;
3059 
3060 	if (enable) {
3061 		WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL, 0);
3062 	} else {
3063 		WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL,
3064 			(CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
3065 		for (i = 0; i < adev->gfx.num_compute_rings; i++)
3066 			adev->gfx.compute_ring[i].sched.ready = false;
3067 		adev->gfx.kiq.ring.sched.ready = false;
3068 	}
3069 	udelay(50);
3070 }
3071 
3072 static int gfx_v9_0_cp_compute_load_microcode(struct amdgpu_device *adev)
3073 {
3074 	const struct gfx_firmware_header_v1_0 *mec_hdr;
3075 	const __le32 *fw_data;
3076 	unsigned i;
3077 	u32 tmp;
3078 
3079 	if (!adev->gfx.mec_fw)
3080 		return -EINVAL;
3081 
3082 	gfx_v9_0_cp_compute_enable(adev, false);
3083 
3084 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3085 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3086 
3087 	fw_data = (const __le32 *)
3088 		(adev->gfx.mec_fw->data +
3089 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
3090 	tmp = 0;
3091 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
3092 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
3093 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL, tmp);
3094 
3095 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO,
3096 		adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000);
3097 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI,
3098 		upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
3099 
3100 	/* MEC1 */
3101 	WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3102 			 mec_hdr->jt_offset);
3103 	for (i = 0; i < mec_hdr->jt_size; i++)
3104 		WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_DATA,
3105 			le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
3106 
3107 	WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3108 			adev->gfx.mec_fw_version);
3109 	/* Todo : Loading MEC2 firmware is only necessary if MEC2 should run different microcode than MEC1. */
3110 
3111 	return 0;
3112 }
3113 
3114 /* KIQ functions */
3115 static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)
3116 {
3117 	uint32_t tmp;
3118 	struct amdgpu_device *adev = ring->adev;
3119 
3120 	/* tell RLC which is KIQ queue */
3121 	tmp = RREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS);
3122 	tmp &= 0xffffff00;
3123 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
3124 	WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3125 	tmp |= 0x80;
3126 	WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3127 }
3128 
3129 static int gfx_v9_0_kiq_kcq_enable(struct amdgpu_device *adev)
3130 {
3131 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
3132 	uint64_t queue_mask = 0;
3133 	int r, i;
3134 
3135 	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
3136 		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
3137 			continue;
3138 
3139 		/* This situation may be hit in the future if a new HW
3140 		 * generation exposes more than 64 queues. If so, the
3141 		 * definition of queue_mask needs updating */
3142 		if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
3143 			DRM_ERROR("Invalid KCQ enabled: %d\n", i);
3144 			break;
3145 		}
3146 
3147 		queue_mask |= (1ull << i);
3148 	}
3149 
3150 	r = amdgpu_ring_alloc(kiq_ring, (7 * adev->gfx.num_compute_rings) + 8);
3151 	if (r) {
3152 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
3153 		return r;
3154 	}
3155 
3156 	/* set resources */
3157 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
3158 	amdgpu_ring_write(kiq_ring, PACKET3_SET_RESOURCES_VMID_MASK(0) |
3159 			  PACKET3_SET_RESOURCES_QUEUE_TYPE(0));	/* vmid_mask:0 queue_type:0 (KIQ) */
3160 	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
3161 	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
3162 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
3163 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
3164 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
3165 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
3166 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3167 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
3168 		uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
3169 		uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
3170 
3171 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
3172 		/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
3173 		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
3174 				  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
3175 				  PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
3176 				  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
3177 				  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
3178 				  PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
3179 				  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
3180 				  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
3181 				  PACKET3_MAP_QUEUES_ENGINE_SEL(0) | /* engine_sel: compute */
3182 				  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
3183 		amdgpu_ring_write(kiq_ring, PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
3184 		amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
3185 		amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
3186 		amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
3187 		amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
3188 	}
3189 
3190 	r = amdgpu_ring_test_helper(kiq_ring);
3191 	if (r)
3192 		DRM_ERROR("KCQ enable failed\n");
3193 
3194 	return r;
3195 }
3196 
3197 static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring)
3198 {
3199 	struct amdgpu_device *adev = ring->adev;
3200 	struct v9_mqd *mqd = ring->mqd_ptr;
3201 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
3202 	uint32_t tmp;
3203 
3204 	mqd->header = 0xC0310800;
3205 	mqd->compute_pipelinestat_enable = 0x00000001;
3206 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
3207 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
3208 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
3209 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
3210 	mqd->compute_static_thread_mgmt_se4 = 0xffffffff;
3211 	mqd->compute_static_thread_mgmt_se5 = 0xffffffff;
3212 	mqd->compute_static_thread_mgmt_se6 = 0xffffffff;
3213 	mqd->compute_static_thread_mgmt_se7 = 0xffffffff;
3214 	mqd->compute_misc_reserved = 0x00000003;
3215 
3216 	mqd->dynamic_cu_mask_addr_lo =
3217 		lower_32_bits(ring->mqd_gpu_addr
3218 			      + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3219 	mqd->dynamic_cu_mask_addr_hi =
3220 		upper_32_bits(ring->mqd_gpu_addr
3221 			      + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3222 
3223 	eop_base_addr = ring->eop_gpu_addr >> 8;
3224 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
3225 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
3226 
3227 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3228 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL);
3229 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
3230 			(order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1));
3231 
3232 	mqd->cp_hqd_eop_control = tmp;
3233 
3234 	/* enable doorbell? */
3235 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
3236 
3237 	if (ring->use_doorbell) {
3238 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3239 				    DOORBELL_OFFSET, ring->doorbell_index);
3240 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3241 				    DOORBELL_EN, 1);
3242 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3243 				    DOORBELL_SOURCE, 0);
3244 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3245 				    DOORBELL_HIT, 0);
3246 	} else {
3247 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3248 					 DOORBELL_EN, 0);
3249 	}
3250 
3251 	mqd->cp_hqd_pq_doorbell_control = tmp;
3252 
3253 	/* disable the queue if it's active */
3254 	ring->wptr = 0;
3255 	mqd->cp_hqd_dequeue_request = 0;
3256 	mqd->cp_hqd_pq_rptr = 0;
3257 	mqd->cp_hqd_pq_wptr_lo = 0;
3258 	mqd->cp_hqd_pq_wptr_hi = 0;
3259 
3260 	/* set the pointer to the MQD */
3261 	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
3262 	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
3263 
3264 	/* set MQD vmid to 0 */
3265 	tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL);
3266 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
3267 	mqd->cp_mqd_control = tmp;
3268 
3269 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3270 	hqd_gpu_addr = ring->gpu_addr >> 8;
3271 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
3272 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3273 
3274 	/* set up the HQD, this is similar to CP_RB0_CNTL */
3275 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL);
3276 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
3277 			    (order_base_2(ring->ring_size / 4) - 1));
3278 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
3279 			((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8));
3280 #ifdef __BIG_ENDIAN
3281 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
3282 #endif
3283 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
3284 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
3285 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
3286 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
3287 	mqd->cp_hqd_pq_control = tmp;
3288 
3289 	/* set the wb address whether it's enabled or not */
3290 	wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
3291 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
3292 	mqd->cp_hqd_pq_rptr_report_addr_hi =
3293 		upper_32_bits(wb_gpu_addr) & 0xffff;
3294 
3295 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3296 	wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
3297 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3298 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3299 
3300 	tmp = 0;
3301 	/* enable the doorbell if requested */
3302 	if (ring->use_doorbell) {
3303 		tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
3304 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3305 				DOORBELL_OFFSET, ring->doorbell_index);
3306 
3307 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3308 					 DOORBELL_EN, 1);
3309 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3310 					 DOORBELL_SOURCE, 0);
3311 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3312 					 DOORBELL_HIT, 0);
3313 	}
3314 
3315 	mqd->cp_hqd_pq_doorbell_control = tmp;
3316 
3317 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3318 	ring->wptr = 0;
3319 	mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR);
3320 
3321 	/* set the vmid for the queue */
3322 	mqd->cp_hqd_vmid = 0;
3323 
3324 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE);
3325 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
3326 	mqd->cp_hqd_persistent_state = tmp;
3327 
3328 	/* set MIN_IB_AVAIL_SIZE */
3329 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL);
3330 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
3331 	mqd->cp_hqd_ib_control = tmp;
3332 
3333 	/* activate the queue */
3334 	mqd->cp_hqd_active = 1;
3335 
3336 	return 0;
3337 }
3338 
3339 static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring)
3340 {
3341 	struct amdgpu_device *adev = ring->adev;
3342 	struct v9_mqd *mqd = ring->mqd_ptr;
3343 	int j;
3344 
3345 	/* disable wptr polling */
3346 	WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3347 
3348 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR,
3349 	       mqd->cp_hqd_eop_base_addr_lo);
3350 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI,
3351 	       mqd->cp_hqd_eop_base_addr_hi);
3352 
3353 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3354 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_CONTROL,
3355 	       mqd->cp_hqd_eop_control);
3356 
3357 	/* enable doorbell? */
3358 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3359 	       mqd->cp_hqd_pq_doorbell_control);
3360 
3361 	/* disable the queue if it's active */
3362 	if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3363 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3364 		for (j = 0; j < adev->usec_timeout; j++) {
3365 			if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3366 				break;
3367 			udelay(1);
3368 		}
3369 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3370 		       mqd->cp_hqd_dequeue_request);
3371 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR,
3372 		       mqd->cp_hqd_pq_rptr);
3373 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3374 		       mqd->cp_hqd_pq_wptr_lo);
3375 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3376 		       mqd->cp_hqd_pq_wptr_hi);
3377 	}
3378 
3379 	/* set the pointer to the MQD */
3380 	WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR,
3381 	       mqd->cp_mqd_base_addr_lo);
3382 	WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR_HI,
3383 	       mqd->cp_mqd_base_addr_hi);
3384 
3385 	/* set MQD vmid to 0 */
3386 	WREG32_SOC15_RLC(GC, 0, mmCP_MQD_CONTROL,
3387 	       mqd->cp_mqd_control);
3388 
3389 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3390 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE,
3391 	       mqd->cp_hqd_pq_base_lo);
3392 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE_HI,
3393 	       mqd->cp_hqd_pq_base_hi);
3394 
3395 	/* set up the HQD, this is similar to CP_RB0_CNTL */
3396 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_CONTROL,
3397 	       mqd->cp_hqd_pq_control);
3398 
3399 	/* set the wb address whether it's enabled or not */
3400 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR,
3401 				mqd->cp_hqd_pq_rptr_report_addr_lo);
3402 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3403 				mqd->cp_hqd_pq_rptr_report_addr_hi);
3404 
3405 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3406 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR,
3407 	       mqd->cp_hqd_pq_wptr_poll_addr_lo);
3408 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
3409 	       mqd->cp_hqd_pq_wptr_poll_addr_hi);
3410 
3411 	/* enable the doorbell if requested */
3412 	if (ring->use_doorbell) {
3413 		WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER,
3414 					(adev->doorbell_index.kiq * 2) << 2);
3415 		WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
3416 					(adev->doorbell_index.userqueue_end * 2) << 2);
3417 	}
3418 
3419 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3420 	       mqd->cp_hqd_pq_doorbell_control);
3421 
3422 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3423 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3424 	       mqd->cp_hqd_pq_wptr_lo);
3425 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3426 	       mqd->cp_hqd_pq_wptr_hi);
3427 
3428 	/* set the vmid for the queue */
3429 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_VMID, mqd->cp_hqd_vmid);
3430 
3431 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE,
3432 	       mqd->cp_hqd_persistent_state);
3433 
3434 	/* activate the queue */
3435 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE,
3436 	       mqd->cp_hqd_active);
3437 
3438 	if (ring->use_doorbell)
3439 		WREG32_FIELD15(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
3440 
3441 	return 0;
3442 }
3443 
3444 static int gfx_v9_0_kiq_fini_register(struct amdgpu_ring *ring)
3445 {
3446 	struct amdgpu_device *adev = ring->adev;
3447 	int j;
3448 
3449 	/* disable the queue if it's active */
3450 	if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3451 
3452 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3453 
3454 		for (j = 0; j < adev->usec_timeout; j++) {
3455 			if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3456 				break;
3457 			udelay(1);
3458 		}
3459 
3460 		if (j == AMDGPU_MAX_USEC_TIMEOUT) {
3461 			DRM_DEBUG("KIQ dequeue request failed.\n");
3462 
3463 			/* Manual disable if dequeue request times out */
3464 			WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE, 0);
3465 		}
3466 
3467 		WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3468 		      0);
3469 	}
3470 
3471 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IQ_TIMER, 0);
3472 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IB_CONTROL, 0);
3473 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE, 0);
3474 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000);
3475 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0);
3476 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR, 0);
3477 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 0);
3478 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 0);
3479 
3480 	return 0;
3481 }
3482 
3483 static int gfx_v9_0_kiq_init_queue(struct amdgpu_ring *ring)
3484 {
3485 	struct amdgpu_device *adev = ring->adev;
3486 	struct v9_mqd *mqd = ring->mqd_ptr;
3487 	int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
3488 
3489 	gfx_v9_0_kiq_setting(ring);
3490 
3491 	if (adev->in_gpu_reset) { /* for GPU_RESET case */
3492 		/* reset MQD to a clean status */
3493 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3494 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
3495 
3496 		/* reset ring buffer */
3497 		ring->wptr = 0;
3498 		amdgpu_ring_clear_ring(ring);
3499 
3500 		mutex_lock(&adev->srbm_mutex);
3501 		soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3502 		gfx_v9_0_kiq_init_register(ring);
3503 		soc15_grbm_select(adev, 0, 0, 0, 0);
3504 		mutex_unlock(&adev->srbm_mutex);
3505 	} else {
3506 		memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3507 		((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3508 		((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3509 		mutex_lock(&adev->srbm_mutex);
3510 		soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3511 		gfx_v9_0_mqd_init(ring);
3512 		gfx_v9_0_kiq_init_register(ring);
3513 		soc15_grbm_select(adev, 0, 0, 0, 0);
3514 		mutex_unlock(&adev->srbm_mutex);
3515 
3516 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3517 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
3518 	}
3519 
3520 	return 0;
3521 }
3522 
3523 static int gfx_v9_0_kcq_init_queue(struct amdgpu_ring *ring)
3524 {
3525 	struct amdgpu_device *adev = ring->adev;
3526 	struct v9_mqd *mqd = ring->mqd_ptr;
3527 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
3528 
3529 	if (!adev->in_gpu_reset && !adev->in_suspend) {
3530 		memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3531 		((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3532 		((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3533 		mutex_lock(&adev->srbm_mutex);
3534 		soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3535 		gfx_v9_0_mqd_init(ring);
3536 		soc15_grbm_select(adev, 0, 0, 0, 0);
3537 		mutex_unlock(&adev->srbm_mutex);
3538 
3539 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3540 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
3541 	} else if (adev->in_gpu_reset) { /* for GPU_RESET case */
3542 		/* reset MQD to a clean status */
3543 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3544 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
3545 
3546 		/* reset ring buffer */
3547 		ring->wptr = 0;
3548 		amdgpu_ring_clear_ring(ring);
3549 	} else {
3550 		amdgpu_ring_clear_ring(ring);
3551 	}
3552 
3553 	return 0;
3554 }
3555 
3556 static int gfx_v9_0_kiq_resume(struct amdgpu_device *adev)
3557 {
3558 	struct amdgpu_ring *ring;
3559 	int r;
3560 
3561 	ring = &adev->gfx.kiq.ring;
3562 
3563 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
3564 	if (unlikely(r != 0))
3565 		return r;
3566 
3567 	r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3568 	if (unlikely(r != 0))
3569 		return r;
3570 
3571 	gfx_v9_0_kiq_init_queue(ring);
3572 	amdgpu_bo_kunmap(ring->mqd_obj);
3573 	ring->mqd_ptr = NULL;
3574 	amdgpu_bo_unreserve(ring->mqd_obj);
3575 	ring->sched.ready = true;
3576 	return 0;
3577 }
3578 
3579 static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev)
3580 {
3581 	struct amdgpu_ring *ring = NULL;
3582 	int r = 0, i;
3583 
3584 	gfx_v9_0_cp_compute_enable(adev, true);
3585 
3586 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3587 		ring = &adev->gfx.compute_ring[i];
3588 
3589 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
3590 		if (unlikely(r != 0))
3591 			goto done;
3592 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3593 		if (!r) {
3594 			r = gfx_v9_0_kcq_init_queue(ring);
3595 			amdgpu_bo_kunmap(ring->mqd_obj);
3596 			ring->mqd_ptr = NULL;
3597 		}
3598 		amdgpu_bo_unreserve(ring->mqd_obj);
3599 		if (r)
3600 			goto done;
3601 	}
3602 
3603 	r = gfx_v9_0_kiq_kcq_enable(adev);
3604 done:
3605 	return r;
3606 }
3607 
3608 static int gfx_v9_0_cp_resume(struct amdgpu_device *adev)
3609 {
3610 	int r, i;
3611 	struct amdgpu_ring *ring;
3612 
3613 	if (!(adev->flags & AMD_IS_APU))
3614 		gfx_v9_0_enable_gui_idle_interrupt(adev, false);
3615 
3616 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
3617 		if (adev->asic_type != CHIP_ARCTURUS) {
3618 			/* legacy firmware loading */
3619 			r = gfx_v9_0_cp_gfx_load_microcode(adev);
3620 			if (r)
3621 				return r;
3622 		}
3623 
3624 		r = gfx_v9_0_cp_compute_load_microcode(adev);
3625 		if (r)
3626 			return r;
3627 	}
3628 
3629 	r = gfx_v9_0_kiq_resume(adev);
3630 	if (r)
3631 		return r;
3632 
3633 	if (adev->asic_type != CHIP_ARCTURUS) {
3634 		r = gfx_v9_0_cp_gfx_resume(adev);
3635 		if (r)
3636 			return r;
3637 	}
3638 
3639 	r = gfx_v9_0_kcq_resume(adev);
3640 	if (r)
3641 		return r;
3642 
3643 	if (adev->asic_type != CHIP_ARCTURUS) {
3644 		ring = &adev->gfx.gfx_ring[0];
3645 		r = amdgpu_ring_test_helper(ring);
3646 		if (r)
3647 			return r;
3648 	}
3649 
3650 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3651 		ring = &adev->gfx.compute_ring[i];
3652 		amdgpu_ring_test_helper(ring);
3653 	}
3654 
3655 	gfx_v9_0_enable_gui_idle_interrupt(adev, true);
3656 
3657 	return 0;
3658 }
3659 
3660 static void gfx_v9_0_cp_enable(struct amdgpu_device *adev, bool enable)
3661 {
3662 	if (adev->asic_type != CHIP_ARCTURUS)
3663 		gfx_v9_0_cp_gfx_enable(adev, enable);
3664 	gfx_v9_0_cp_compute_enable(adev, enable);
3665 }
3666 
3667 static int gfx_v9_0_hw_init(void *handle)
3668 {
3669 	int r;
3670 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3671 
3672 	if (!amdgpu_sriov_vf(adev))
3673 		gfx_v9_0_init_golden_registers(adev);
3674 
3675 	gfx_v9_0_constants_init(adev);
3676 
3677 	r = adev->gfx.rlc.funcs->resume(adev);
3678 	if (r)
3679 		return r;
3680 
3681 	r = gfx_v9_0_cp_resume(adev);
3682 	if (r)
3683 		return r;
3684 
3685 	return r;
3686 }
3687 
3688 static int gfx_v9_0_kcq_disable(struct amdgpu_device *adev)
3689 {
3690 	int r, i;
3691 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
3692 
3693 	r = amdgpu_ring_alloc(kiq_ring, 6 * adev->gfx.num_compute_rings);
3694 	if (r)
3695 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
3696 
3697 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3698 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
3699 
3700 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
3701 		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
3702 						PACKET3_UNMAP_QUEUES_ACTION(1) | /* RESET_QUEUES */
3703 						PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
3704 						PACKET3_UNMAP_QUEUES_ENGINE_SEL(0) |
3705 						PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
3706 		amdgpu_ring_write(kiq_ring, PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
3707 		amdgpu_ring_write(kiq_ring, 0);
3708 		amdgpu_ring_write(kiq_ring, 0);
3709 		amdgpu_ring_write(kiq_ring, 0);
3710 	}
3711 	r = amdgpu_ring_test_helper(kiq_ring);
3712 	if (r)
3713 		DRM_ERROR("KCQ disable failed\n");
3714 
3715 	return r;
3716 }
3717 
3718 static int gfx_v9_0_hw_fini(void *handle)
3719 {
3720 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3721 
3722 	amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
3723 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
3724 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
3725 
3726 	/* DF freeze and kcq disable will fail */
3727 	if (!amdgpu_ras_intr_triggered())
3728 		/* disable KCQ to avoid CPC touch memory not valid anymore */
3729 		gfx_v9_0_kcq_disable(adev);
3730 
3731 	if (amdgpu_sriov_vf(adev)) {
3732 		gfx_v9_0_cp_gfx_enable(adev, false);
3733 		/* must disable polling for SRIOV when hw finished, otherwise
3734 		 * CPC engine may still keep fetching WB address which is already
3735 		 * invalid after sw finished and trigger DMAR reading error in
3736 		 * hypervisor side.
3737 		 */
3738 		WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3739 		return 0;
3740 	}
3741 
3742 	/* Use deinitialize sequence from CAIL when unbinding device from driver,
3743 	 * otherwise KIQ is hanging when binding back
3744 	 */
3745 	if (!adev->in_gpu_reset && !adev->in_suspend) {
3746 		mutex_lock(&adev->srbm_mutex);
3747 		soc15_grbm_select(adev, adev->gfx.kiq.ring.me,
3748 				adev->gfx.kiq.ring.pipe,
3749 				adev->gfx.kiq.ring.queue, 0);
3750 		gfx_v9_0_kiq_fini_register(&adev->gfx.kiq.ring);
3751 		soc15_grbm_select(adev, 0, 0, 0, 0);
3752 		mutex_unlock(&adev->srbm_mutex);
3753 	}
3754 
3755 	gfx_v9_0_cp_enable(adev, false);
3756 	adev->gfx.rlc.funcs->stop(adev);
3757 
3758 	return 0;
3759 }
3760 
3761 static int gfx_v9_0_suspend(void *handle)
3762 {
3763 	return gfx_v9_0_hw_fini(handle);
3764 }
3765 
3766 static int gfx_v9_0_resume(void *handle)
3767 {
3768 	return gfx_v9_0_hw_init(handle);
3769 }
3770 
3771 static bool gfx_v9_0_is_idle(void *handle)
3772 {
3773 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3774 
3775 	if (REG_GET_FIELD(RREG32_SOC15(GC, 0, mmGRBM_STATUS),
3776 				GRBM_STATUS, GUI_ACTIVE))
3777 		return false;
3778 	else
3779 		return true;
3780 }
3781 
3782 static int gfx_v9_0_wait_for_idle(void *handle)
3783 {
3784 	unsigned i;
3785 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3786 
3787 	for (i = 0; i < adev->usec_timeout; i++) {
3788 		if (gfx_v9_0_is_idle(handle))
3789 			return 0;
3790 		udelay(1);
3791 	}
3792 	return -ETIMEDOUT;
3793 }
3794 
3795 static int gfx_v9_0_soft_reset(void *handle)
3796 {
3797 	u32 grbm_soft_reset = 0;
3798 	u32 tmp;
3799 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3800 
3801 	/* GRBM_STATUS */
3802 	tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS);
3803 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
3804 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
3805 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
3806 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
3807 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
3808 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) {
3809 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3810 						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
3811 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3812 						GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
3813 	}
3814 
3815 	if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
3816 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3817 						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
3818 	}
3819 
3820 	/* GRBM_STATUS2 */
3821 	tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2);
3822 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
3823 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3824 						GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
3825 
3826 
3827 	if (grbm_soft_reset) {
3828 		/* stop the rlc */
3829 		adev->gfx.rlc.funcs->stop(adev);
3830 
3831 		if (adev->asic_type != CHIP_ARCTURUS)
3832 			/* Disable GFX parsing/prefetching */
3833 			gfx_v9_0_cp_gfx_enable(adev, false);
3834 
3835 		/* Disable MEC parsing/prefetching */
3836 		gfx_v9_0_cp_compute_enable(adev, false);
3837 
3838 		if (grbm_soft_reset) {
3839 			tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3840 			tmp |= grbm_soft_reset;
3841 			dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
3842 			WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
3843 			tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3844 
3845 			udelay(50);
3846 
3847 			tmp &= ~grbm_soft_reset;
3848 			WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
3849 			tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3850 		}
3851 
3852 		/* Wait a little for things to settle down */
3853 		udelay(50);
3854 	}
3855 	return 0;
3856 }
3857 
3858 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev)
3859 {
3860 	uint64_t clock;
3861 
3862 	mutex_lock(&adev->gfx.gpu_clock_mutex);
3863 	if (adev->asic_type == CHIP_VEGA10 && amdgpu_sriov_runtime(adev)) {
3864 		uint32_t tmp, lsb, msb, i = 0;
3865 		do {
3866 			if (i != 0)
3867 				udelay(1);
3868 			tmp = RREG32_SOC15(GC, 0, mmRLC_REFCLOCK_TIMESTAMP_MSB);
3869 			lsb = RREG32_SOC15(GC, 0, mmRLC_REFCLOCK_TIMESTAMP_LSB);
3870 			msb = RREG32_SOC15(GC, 0, mmRLC_REFCLOCK_TIMESTAMP_MSB);
3871 			i++;
3872 		} while (unlikely(tmp != msb) && (i < adev->usec_timeout));
3873 		clock = (uint64_t)lsb | ((uint64_t)msb << 32ULL);
3874 	} else {
3875 		WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
3876 		clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB) |
3877 			((uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
3878 	}
3879 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
3880 	return clock;
3881 }
3882 
3883 static void gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
3884 					  uint32_t vmid,
3885 					  uint32_t gds_base, uint32_t gds_size,
3886 					  uint32_t gws_base, uint32_t gws_size,
3887 					  uint32_t oa_base, uint32_t oa_size)
3888 {
3889 	struct amdgpu_device *adev = ring->adev;
3890 
3891 	/* GDS Base */
3892 	gfx_v9_0_write_data_to_reg(ring, 0, false,
3893 				   SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_BASE) + 2 * vmid,
3894 				   gds_base);
3895 
3896 	/* GDS Size */
3897 	gfx_v9_0_write_data_to_reg(ring, 0, false,
3898 				   SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE) + 2 * vmid,
3899 				   gds_size);
3900 
3901 	/* GWS */
3902 	gfx_v9_0_write_data_to_reg(ring, 0, false,
3903 				   SOC15_REG_OFFSET(GC, 0, mmGDS_GWS_VMID0) + vmid,
3904 				   gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
3905 
3906 	/* OA */
3907 	gfx_v9_0_write_data_to_reg(ring, 0, false,
3908 				   SOC15_REG_OFFSET(GC, 0, mmGDS_OA_VMID0) + vmid,
3909 				   (1 << (oa_size + oa_base)) - (1 << oa_base));
3910 }
3911 
3912 static const u32 vgpr_init_compute_shader[] =
3913 {
3914 	0xb07c0000, 0xbe8000ff,
3915 	0x000000f8, 0xbf110800,
3916 	0x7e000280, 0x7e020280,
3917 	0x7e040280, 0x7e060280,
3918 	0x7e080280, 0x7e0a0280,
3919 	0x7e0c0280, 0x7e0e0280,
3920 	0x80808800, 0xbe803200,
3921 	0xbf84fff5, 0xbf9c0000,
3922 	0xd28c0001, 0x0001007f,
3923 	0xd28d0001, 0x0002027e,
3924 	0x10020288, 0xb8810904,
3925 	0xb7814000, 0xd1196a01,
3926 	0x00000301, 0xbe800087,
3927 	0xbefc00c1, 0xd89c4000,
3928 	0x00020201, 0xd89cc080,
3929 	0x00040401, 0x320202ff,
3930 	0x00000800, 0x80808100,
3931 	0xbf84fff8, 0x7e020280,
3932 	0xbf810000, 0x00000000,
3933 };
3934 
3935 static const u32 sgpr_init_compute_shader[] =
3936 {
3937 	0xb07c0000, 0xbe8000ff,
3938 	0x0000005f, 0xbee50080,
3939 	0xbe812c65, 0xbe822c65,
3940 	0xbe832c65, 0xbe842c65,
3941 	0xbe852c65, 0xb77c0005,
3942 	0x80808500, 0xbf84fff8,
3943 	0xbe800080, 0xbf810000,
3944 };
3945 
3946 static const struct soc15_reg_entry vgpr_init_regs[] = {
3947    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
3948    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
3949    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
3950    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
3951    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x1000000 }, /* CU_GROUP_COUNT=1 */
3952    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 256*2 },
3953    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 1 },
3954    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
3955    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x100007f }, /* VGPRS=15 (256 logical VGPRs, SGPRS=1 (16 SGPRs, BULKY=1 */
3956    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 },  /* 64KB LDS */
3957 };
3958 
3959 static const struct soc15_reg_entry sgpr_init_regs[] = {
3960    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
3961    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
3962    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
3963    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
3964    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x1000000 }, /* CU_GROUP_COUNT=1 */
3965    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 256*2 },
3966    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 1 },
3967    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
3968    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x340 }, /* SGPRS=13 (112 GPRS) */
3969    { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
3970 };
3971 
3972 static const struct soc15_reg_entry sec_ded_counter_registers[] = {
3973    { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT), 0, 1, 1},
3974    { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT), 0, 1, 1},
3975    { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 0, 1, 1},
3976    { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT), 0, 1, 1},
3977    { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 0, 1, 1},
3978    { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT), 0, 1, 1},
3979    { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT), 0, 1, 1},
3980    { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT), 0, 1, 1},
3981    { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT), 0, 1, 1},
3982    { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 0, 1, 1},
3983    { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_GRBM_CNT), 0, 1, 1},
3984    { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_DED), 0, 1, 1},
3985    { SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT), 0, 4, 1},
3986    { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 0, 4, 6},
3987    { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_DED_CNT), 0, 4, 16},
3988    { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_INFO), 0, 4, 16},
3989    { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_SEC_CNT), 0, 4, 16},
3990    { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 0, 1, 16},
3991    { SOC15_REG_ENTRY(GC, 0, mmTCP_ATC_EDC_GATCL1_CNT), 0, 4, 16},
3992    { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT), 0, 4, 16},
3993    { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 0, 4, 16},
3994    { SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 0, 4, 16},
3995    { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 0, 4, 6},
3996    { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 0, 4, 16},
3997    { SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 0, 4, 16},
3998    { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 0, 1, 1},
3999    { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 0, 1, 1},
4000    { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 0, 1, 32},
4001    { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 0, 1, 32},
4002    { SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT), 0, 1, 72},
4003    { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 0, 1, 16},
4004    { SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 0, 1, 2},
4005    { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 0, 4, 6},
4006 };
4007 
4008 static int gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device *adev)
4009 {
4010 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4011 	int i, r;
4012 
4013 	/* only support when RAS is enabled */
4014 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4015 		return 0;
4016 
4017 	r = amdgpu_ring_alloc(ring, 7);
4018 	if (r) {
4019 		DRM_ERROR("amdgpu: GDS workarounds failed to lock ring %s (%d).\n",
4020 			ring->name, r);
4021 		return r;
4022 	}
4023 
4024 	WREG32_SOC15(GC, 0, mmGDS_VMID0_BASE, 0x00000000);
4025 	WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, adev->gds.gds_size);
4026 
4027 	amdgpu_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5));
4028 	amdgpu_ring_write(ring, (PACKET3_DMA_DATA_CP_SYNC |
4029 				PACKET3_DMA_DATA_DST_SEL(1) |
4030 				PACKET3_DMA_DATA_SRC_SEL(2) |
4031 				PACKET3_DMA_DATA_ENGINE(0)));
4032 	amdgpu_ring_write(ring, 0);
4033 	amdgpu_ring_write(ring, 0);
4034 	amdgpu_ring_write(ring, 0);
4035 	amdgpu_ring_write(ring, 0);
4036 	amdgpu_ring_write(ring, PACKET3_DMA_DATA_CMD_RAW_WAIT |
4037 				adev->gds.gds_size);
4038 
4039 	amdgpu_ring_commit(ring);
4040 
4041 	for (i = 0; i < adev->usec_timeout; i++) {
4042 		if (ring->wptr == gfx_v9_0_ring_get_rptr_compute(ring))
4043 			break;
4044 		udelay(1);
4045 	}
4046 
4047 	if (i >= adev->usec_timeout)
4048 		r = -ETIMEDOUT;
4049 
4050 	WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, 0x00000000);
4051 
4052 	return r;
4053 }
4054 
4055 static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
4056 {
4057 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4058 	struct amdgpu_ib ib;
4059 	struct dma_fence *f = NULL;
4060 	int r, i, j, k;
4061 	unsigned total_size, vgpr_offset, sgpr_offset;
4062 	u64 gpu_addr;
4063 
4064 	/* only support when RAS is enabled */
4065 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4066 		return 0;
4067 
4068 	/* bail if the compute ring is not ready */
4069 	if (!ring->sched.ready)
4070 		return 0;
4071 
4072 	total_size =
4073 		((ARRAY_SIZE(vgpr_init_regs) * 3) + 4 + 5 + 2) * 4;
4074 	total_size +=
4075 		((ARRAY_SIZE(sgpr_init_regs) * 3) + 4 + 5 + 2) * 4;
4076 	total_size = ALIGN(total_size, 256);
4077 	vgpr_offset = total_size;
4078 	total_size += ALIGN(sizeof(vgpr_init_compute_shader), 256);
4079 	sgpr_offset = total_size;
4080 	total_size += sizeof(sgpr_init_compute_shader);
4081 
4082 	/* allocate an indirect buffer to put the commands in */
4083 	memset(&ib, 0, sizeof(ib));
4084 	r = amdgpu_ib_get(adev, NULL, total_size, &ib);
4085 	if (r) {
4086 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
4087 		return r;
4088 	}
4089 
4090 	/* load the compute shaders */
4091 	for (i = 0; i < ARRAY_SIZE(vgpr_init_compute_shader); i++)
4092 		ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_compute_shader[i];
4093 
4094 	for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
4095 		ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
4096 
4097 	/* init the ib length to 0 */
4098 	ib.length_dw = 0;
4099 
4100 	/* VGPR */
4101 	/* write the register state for the compute dispatch */
4102 	for (i = 0; i < ARRAY_SIZE(vgpr_init_regs); i++) {
4103 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4104 		ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(vgpr_init_regs[i])
4105 								- PACKET3_SET_SH_REG_START;
4106 		ib.ptr[ib.length_dw++] = vgpr_init_regs[i].reg_value;
4107 	}
4108 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4109 	gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
4110 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4111 	ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4112 							- PACKET3_SET_SH_REG_START;
4113 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4114 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4115 
4116 	/* write dispatch packet */
4117 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4118 	ib.ptr[ib.length_dw++] = 128; /* x */
4119 	ib.ptr[ib.length_dw++] = 1; /* y */
4120 	ib.ptr[ib.length_dw++] = 1; /* z */
4121 	ib.ptr[ib.length_dw++] =
4122 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4123 
4124 	/* write CS partial flush packet */
4125 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4126 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4127 
4128 	/* SGPR */
4129 	/* write the register state for the compute dispatch */
4130 	for (i = 0; i < ARRAY_SIZE(sgpr_init_regs); i++) {
4131 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4132 		ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr_init_regs[i])
4133 								- PACKET3_SET_SH_REG_START;
4134 		ib.ptr[ib.length_dw++] = sgpr_init_regs[i].reg_value;
4135 	}
4136 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4137 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4138 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4139 	ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4140 							- PACKET3_SET_SH_REG_START;
4141 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4142 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4143 
4144 	/* write dispatch packet */
4145 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4146 	ib.ptr[ib.length_dw++] = 128; /* x */
4147 	ib.ptr[ib.length_dw++] = 1; /* y */
4148 	ib.ptr[ib.length_dw++] = 1; /* z */
4149 	ib.ptr[ib.length_dw++] =
4150 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4151 
4152 	/* write CS partial flush packet */
4153 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4154 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4155 
4156 	/* shedule the ib on the ring */
4157 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
4158 	if (r) {
4159 		DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
4160 		goto fail;
4161 	}
4162 
4163 	/* wait for the GPU to finish processing the IB */
4164 	r = dma_fence_wait(f, false);
4165 	if (r) {
4166 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
4167 		goto fail;
4168 	}
4169 
4170 	/* read back registers to clear the counters */
4171 	mutex_lock(&adev->grbm_idx_mutex);
4172 	for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++) {
4173 		for (j = 0; j < sec_ded_counter_registers[i].se_num; j++) {
4174 			for (k = 0; k < sec_ded_counter_registers[i].instance; k++) {
4175 				gfx_v9_0_select_se_sh(adev, j, 0x0, k);
4176 				RREG32(SOC15_REG_ENTRY_OFFSET(sec_ded_counter_registers[i]));
4177 			}
4178 		}
4179 	}
4180 	WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, 0xe0000000);
4181 	mutex_unlock(&adev->grbm_idx_mutex);
4182 
4183 fail:
4184 	amdgpu_ib_free(adev, &ib, NULL);
4185 	dma_fence_put(f);
4186 
4187 	return r;
4188 }
4189 
4190 static int gfx_v9_0_early_init(void *handle)
4191 {
4192 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4193 
4194 	if (adev->asic_type == CHIP_ARCTURUS)
4195 		adev->gfx.num_gfx_rings = 0;
4196 	else
4197 		adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS;
4198 	adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
4199 	gfx_v9_0_set_ring_funcs(adev);
4200 	gfx_v9_0_set_irq_funcs(adev);
4201 	gfx_v9_0_set_gds_init(adev);
4202 	gfx_v9_0_set_rlc_funcs(adev);
4203 
4204 	return 0;
4205 }
4206 
4207 static int gfx_v9_0_ecc_late_init(void *handle)
4208 {
4209 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4210 	int r;
4211 
4212 	r = amdgpu_gfx_ras_late_init(adev);
4213 	if (r)
4214 		return r;
4215 
4216 	r = gfx_v9_0_do_edc_gds_workarounds(adev);
4217 	if (r)
4218 		return r;
4219 
4220 	/* requires IBs so do in late init after IB pool is initialized */
4221 	r = gfx_v9_0_do_edc_gpr_workarounds(adev);
4222 	if (r)
4223 		return r;
4224 
4225 	return 0;
4226 }
4227 
4228 static int gfx_v9_0_late_init(void *handle)
4229 {
4230 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4231 	int r;
4232 
4233 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4234 	if (r)
4235 		return r;
4236 
4237 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4238 	if (r)
4239 		return r;
4240 
4241 	r = gfx_v9_0_ecc_late_init(handle);
4242 	if (r)
4243 		return r;
4244 
4245 	return 0;
4246 }
4247 
4248 static bool gfx_v9_0_is_rlc_enabled(struct amdgpu_device *adev)
4249 {
4250 	uint32_t rlc_setting;
4251 
4252 	/* if RLC is not enabled, do nothing */
4253 	rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL);
4254 	if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
4255 		return false;
4256 
4257 	return true;
4258 }
4259 
4260 static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev)
4261 {
4262 	uint32_t data;
4263 	unsigned i;
4264 
4265 	data = RLC_SAFE_MODE__CMD_MASK;
4266 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
4267 	WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4268 
4269 	/* wait for RLC_SAFE_MODE */
4270 	for (i = 0; i < adev->usec_timeout; i++) {
4271 		if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
4272 			break;
4273 		udelay(1);
4274 	}
4275 }
4276 
4277 static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev)
4278 {
4279 	uint32_t data;
4280 
4281 	data = RLC_SAFE_MODE__CMD_MASK;
4282 	WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4283 }
4284 
4285 static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev,
4286 						bool enable)
4287 {
4288 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4289 
4290 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
4291 		gfx_v9_0_enable_gfx_cg_power_gating(adev, true);
4292 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4293 			gfx_v9_0_enable_gfx_pipeline_powergating(adev, true);
4294 	} else {
4295 		gfx_v9_0_enable_gfx_cg_power_gating(adev, false);
4296 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4297 			gfx_v9_0_enable_gfx_pipeline_powergating(adev, false);
4298 	}
4299 
4300 	amdgpu_gfx_rlc_exit_safe_mode(adev);
4301 }
4302 
4303 static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev,
4304 						bool enable)
4305 {
4306 	/* TODO: double check if we need to perform under safe mode */
4307 	/* gfx_v9_0_enter_rlc_safe_mode(adev); */
4308 
4309 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
4310 		gfx_v9_0_enable_gfx_static_mg_power_gating(adev, true);
4311 	else
4312 		gfx_v9_0_enable_gfx_static_mg_power_gating(adev, false);
4313 
4314 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
4315 		gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, true);
4316 	else
4317 		gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, false);
4318 
4319 	/* gfx_v9_0_exit_rlc_safe_mode(adev); */
4320 }
4321 
4322 static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
4323 						      bool enable)
4324 {
4325 	uint32_t data, def;
4326 
4327 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4328 
4329 	/* It is disabled by HW by default */
4330 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
4331 		/* 1 - RLC_CGTT_MGCG_OVERRIDE */
4332 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4333 
4334 		if (adev->asic_type != CHIP_VEGA12)
4335 			data &= ~RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
4336 
4337 		data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4338 			  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4339 			  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4340 
4341 		/* only for Vega10 & Raven1 */
4342 		data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK;
4343 
4344 		if (def != data)
4345 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4346 
4347 		/* MGLS is a global flag to control all MGLS in GFX */
4348 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
4349 			/* 2 - RLC memory Light sleep */
4350 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
4351 				def = data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4352 				data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4353 				if (def != data)
4354 					WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4355 			}
4356 			/* 3 - CP memory Light sleep */
4357 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
4358 				def = data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4359 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4360 				if (def != data)
4361 					WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4362 			}
4363 		}
4364 	} else {
4365 		/* 1 - MGCG_OVERRIDE */
4366 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4367 
4368 		if (adev->asic_type != CHIP_VEGA12)
4369 			data |= RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
4370 
4371 		data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4372 			 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4373 			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4374 			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4375 
4376 		if (def != data)
4377 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4378 
4379 		/* 2 - disable MGLS in RLC */
4380 		data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4381 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
4382 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4383 			WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4384 		}
4385 
4386 		/* 3 - disable MGLS in CP */
4387 		data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4388 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
4389 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4390 			WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4391 		}
4392 	}
4393 
4394 	amdgpu_gfx_rlc_exit_safe_mode(adev);
4395 }
4396 
4397 static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev,
4398 					   bool enable)
4399 {
4400 	uint32_t data, def;
4401 
4402 	if (adev->asic_type == CHIP_ARCTURUS)
4403 		return;
4404 
4405 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4406 
4407 	/* Enable 3D CGCG/CGLS */
4408 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)) {
4409 		/* write cmd to clear cgcg/cgls ov */
4410 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4411 		/* unset CGCG override */
4412 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK;
4413 		/* update CGCG and CGLS override bits */
4414 		if (def != data)
4415 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4416 
4417 		/* enable 3Dcgcg FSM(0x0000363f) */
4418 		def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4419 
4420 		data = (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4421 			RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
4422 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
4423 			data |= (0x000F << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4424 				RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
4425 		if (def != data)
4426 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
4427 
4428 		/* set IDLE_POLL_COUNT(0x00900100) */
4429 		def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
4430 		data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4431 			(0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4432 		if (def != data)
4433 			WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
4434 	} else {
4435 		/* Disable CGCG/CGLS */
4436 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4437 		/* disable cgcg, cgls should be disabled */
4438 		data &= ~(RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK |
4439 			  RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK);
4440 		/* disable cgcg and cgls in FSM */
4441 		if (def != data)
4442 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
4443 	}
4444 
4445 	amdgpu_gfx_rlc_exit_safe_mode(adev);
4446 }
4447 
4448 static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
4449 						      bool enable)
4450 {
4451 	uint32_t def, data;
4452 
4453 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4454 
4455 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
4456 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4457 		/* unset CGCG override */
4458 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
4459 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4460 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4461 		else
4462 			data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4463 		/* update CGCG and CGLS override bits */
4464 		if (def != data)
4465 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4466 
4467 		/* enable cgcg FSM(0x0000363F) */
4468 		def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4469 
4470 		if (adev->asic_type == CHIP_ARCTURUS)
4471 			data = (0x2000 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4472 				RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4473 		else
4474 			data = (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4475 				RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4476 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4477 			data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4478 				RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
4479 		if (def != data)
4480 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
4481 
4482 		/* set IDLE_POLL_COUNT(0x00900100) */
4483 		def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
4484 		data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4485 			(0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4486 		if (def != data)
4487 			WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
4488 	} else {
4489 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4490 		/* reset CGCG/CGLS bits */
4491 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
4492 		/* disable cgcg and cgls in FSM */
4493 		if (def != data)
4494 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
4495 	}
4496 
4497 	amdgpu_gfx_rlc_exit_safe_mode(adev);
4498 }
4499 
4500 static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev,
4501 					    bool enable)
4502 {
4503 	if (enable) {
4504 		/* CGCG/CGLS should be enabled after MGCG/MGLS
4505 		 * ===  MGCG + MGLS ===
4506 		 */
4507 		gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
4508 		/* ===  CGCG /CGLS for GFX 3D Only === */
4509 		gfx_v9_0_update_3d_clock_gating(adev, enable);
4510 		/* ===  CGCG + CGLS === */
4511 		gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
4512 	} else {
4513 		/* CGCG/CGLS should be disabled before MGCG/MGLS
4514 		 * ===  CGCG + CGLS ===
4515 		 */
4516 		gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
4517 		/* ===  CGCG /CGLS for GFX 3D Only === */
4518 		gfx_v9_0_update_3d_clock_gating(adev, enable);
4519 		/* ===  MGCG + MGLS === */
4520 		gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
4521 	}
4522 	return 0;
4523 }
4524 
4525 static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = {
4526 	.is_rlc_enabled = gfx_v9_0_is_rlc_enabled,
4527 	.set_safe_mode = gfx_v9_0_set_safe_mode,
4528 	.unset_safe_mode = gfx_v9_0_unset_safe_mode,
4529 	.init = gfx_v9_0_rlc_init,
4530 	.get_csb_size = gfx_v9_0_get_csb_size,
4531 	.get_csb_buffer = gfx_v9_0_get_csb_buffer,
4532 	.get_cp_table_num = gfx_v9_0_cp_jump_table_num,
4533 	.resume = gfx_v9_0_rlc_resume,
4534 	.stop = gfx_v9_0_rlc_stop,
4535 	.reset = gfx_v9_0_rlc_reset,
4536 	.start = gfx_v9_0_rlc_start
4537 };
4538 
4539 static int gfx_v9_0_set_powergating_state(void *handle,
4540 					  enum amd_powergating_state state)
4541 {
4542 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4543 	bool enable = (state == AMD_PG_STATE_GATE) ? true : false;
4544 
4545 	switch (adev->asic_type) {
4546 	case CHIP_RAVEN:
4547 	case CHIP_RENOIR:
4548 		if (!enable) {
4549 			amdgpu_gfx_off_ctrl(adev, false);
4550 			cancel_delayed_work_sync(&adev->gfx.gfx_off_delay_work);
4551 		}
4552 		if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
4553 			gfx_v9_0_enable_sck_slow_down_on_power_up(adev, true);
4554 			gfx_v9_0_enable_sck_slow_down_on_power_down(adev, true);
4555 		} else {
4556 			gfx_v9_0_enable_sck_slow_down_on_power_up(adev, false);
4557 			gfx_v9_0_enable_sck_slow_down_on_power_down(adev, false);
4558 		}
4559 
4560 		if (adev->pg_flags & AMD_PG_SUPPORT_CP)
4561 			gfx_v9_0_enable_cp_power_gating(adev, true);
4562 		else
4563 			gfx_v9_0_enable_cp_power_gating(adev, false);
4564 
4565 		/* update gfx cgpg state */
4566 		gfx_v9_0_update_gfx_cg_power_gating(adev, enable);
4567 
4568 		/* update mgcg state */
4569 		gfx_v9_0_update_gfx_mg_power_gating(adev, enable);
4570 
4571 		if (enable)
4572 			amdgpu_gfx_off_ctrl(adev, true);
4573 		break;
4574 	case CHIP_VEGA12:
4575 		if (!enable) {
4576 			amdgpu_gfx_off_ctrl(adev, false);
4577 			cancel_delayed_work_sync(&adev->gfx.gfx_off_delay_work);
4578 		} else {
4579 			amdgpu_gfx_off_ctrl(adev, true);
4580 		}
4581 		break;
4582 	default:
4583 		break;
4584 	}
4585 
4586 	return 0;
4587 }
4588 
4589 static int gfx_v9_0_set_clockgating_state(void *handle,
4590 					  enum amd_clockgating_state state)
4591 {
4592 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4593 
4594 	if (amdgpu_sriov_vf(adev))
4595 		return 0;
4596 
4597 	switch (adev->asic_type) {
4598 	case CHIP_VEGA10:
4599 	case CHIP_VEGA12:
4600 	case CHIP_VEGA20:
4601 	case CHIP_RAVEN:
4602 	case CHIP_ARCTURUS:
4603 	case CHIP_RENOIR:
4604 		gfx_v9_0_update_gfx_clock_gating(adev,
4605 						 state == AMD_CG_STATE_GATE ? true : false);
4606 		break;
4607 	default:
4608 		break;
4609 	}
4610 	return 0;
4611 }
4612 
4613 static void gfx_v9_0_get_clockgating_state(void *handle, u32 *flags)
4614 {
4615 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4616 	int data;
4617 
4618 	if (amdgpu_sriov_vf(adev))
4619 		*flags = 0;
4620 
4621 	/* AMD_CG_SUPPORT_GFX_MGCG */
4622 	data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4623 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
4624 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
4625 
4626 	/* AMD_CG_SUPPORT_GFX_CGCG */
4627 	data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4628 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
4629 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
4630 
4631 	/* AMD_CG_SUPPORT_GFX_CGLS */
4632 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
4633 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
4634 
4635 	/* AMD_CG_SUPPORT_GFX_RLC_LS */
4636 	data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4637 	if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
4638 		*flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
4639 
4640 	/* AMD_CG_SUPPORT_GFX_CP_LS */
4641 	data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4642 	if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
4643 		*flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
4644 
4645 	if (adev->asic_type != CHIP_ARCTURUS) {
4646 		/* AMD_CG_SUPPORT_GFX_3D_CGCG */
4647 		data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4648 		if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK)
4649 			*flags |= AMD_CG_SUPPORT_GFX_3D_CGCG;
4650 
4651 		/* AMD_CG_SUPPORT_GFX_3D_CGLS */
4652 		if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK)
4653 			*flags |= AMD_CG_SUPPORT_GFX_3D_CGLS;
4654 	}
4655 }
4656 
4657 static u64 gfx_v9_0_ring_get_rptr_gfx(struct amdgpu_ring *ring)
4658 {
4659 	return ring->adev->wb.wb[ring->rptr_offs]; /* gfx9 is 32bit rptr*/
4660 }
4661 
4662 static u64 gfx_v9_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
4663 {
4664 	struct amdgpu_device *adev = ring->adev;
4665 	u64 wptr;
4666 
4667 	/* XXX check if swapping is necessary on BE */
4668 	if (ring->use_doorbell) {
4669 		wptr = atomic64_read((atomic64_t *)&adev->wb.wb[ring->wptr_offs]);
4670 	} else {
4671 		wptr = RREG32_SOC15(GC, 0, mmCP_RB0_WPTR);
4672 		wptr += (u64)RREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI) << 32;
4673 	}
4674 
4675 	return wptr;
4676 }
4677 
4678 static void gfx_v9_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
4679 {
4680 	struct amdgpu_device *adev = ring->adev;
4681 
4682 	if (ring->use_doorbell) {
4683 		/* XXX check if swapping is necessary on BE */
4684 		atomic64_set((atomic64_t*)&adev->wb.wb[ring->wptr_offs], ring->wptr);
4685 		WDOORBELL64(ring->doorbell_index, ring->wptr);
4686 	} else {
4687 		WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
4688 		WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
4689 	}
4690 }
4691 
4692 static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
4693 {
4694 	struct amdgpu_device *adev = ring->adev;
4695 	u32 ref_and_mask, reg_mem_engine;
4696 	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
4697 
4698 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
4699 		switch (ring->me) {
4700 		case 1:
4701 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
4702 			break;
4703 		case 2:
4704 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
4705 			break;
4706 		default:
4707 			return;
4708 		}
4709 		reg_mem_engine = 0;
4710 	} else {
4711 		ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
4712 		reg_mem_engine = 1; /* pfp */
4713 	}
4714 
4715 	gfx_v9_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
4716 			      adev->nbio.funcs->get_hdp_flush_req_offset(adev),
4717 			      adev->nbio.funcs->get_hdp_flush_done_offset(adev),
4718 			      ref_and_mask, ref_and_mask, 0x20);
4719 }
4720 
4721 static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
4722 					struct amdgpu_job *job,
4723 					struct amdgpu_ib *ib,
4724 					uint32_t flags)
4725 {
4726 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
4727 	u32 header, control = 0;
4728 
4729 	if (ib->flags & AMDGPU_IB_FLAG_CE)
4730 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
4731 	else
4732 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
4733 
4734 	control |= ib->length_dw | (vmid << 24);
4735 
4736 	if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
4737 		control |= INDIRECT_BUFFER_PRE_ENB(1);
4738 
4739 		if (!(ib->flags & AMDGPU_IB_FLAG_CE))
4740 			gfx_v9_0_ring_emit_de_meta(ring);
4741 	}
4742 
4743 	amdgpu_ring_write(ring, header);
4744 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
4745 	amdgpu_ring_write(ring,
4746 #ifdef __BIG_ENDIAN
4747 		(2 << 0) |
4748 #endif
4749 		lower_32_bits(ib->gpu_addr));
4750 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
4751 	amdgpu_ring_write(ring, control);
4752 }
4753 
4754 static void gfx_v9_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
4755 					  struct amdgpu_job *job,
4756 					  struct amdgpu_ib *ib,
4757 					  uint32_t flags)
4758 {
4759 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
4760 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
4761 
4762 	/* Currently, there is a high possibility to get wave ID mismatch
4763 	 * between ME and GDS, leading to a hw deadlock, because ME generates
4764 	 * different wave IDs than the GDS expects. This situation happens
4765 	 * randomly when at least 5 compute pipes use GDS ordered append.
4766 	 * The wave IDs generated by ME are also wrong after suspend/resume.
4767 	 * Those are probably bugs somewhere else in the kernel driver.
4768 	 *
4769 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
4770 	 * GDS to 0 for this ring (me/pipe).
4771 	 */
4772 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
4773 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
4774 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID);
4775 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
4776 	}
4777 
4778 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
4779 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
4780 	amdgpu_ring_write(ring,
4781 #ifdef __BIG_ENDIAN
4782 				(2 << 0) |
4783 #endif
4784 				lower_32_bits(ib->gpu_addr));
4785 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
4786 	amdgpu_ring_write(ring, control);
4787 }
4788 
4789 static void gfx_v9_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
4790 				     u64 seq, unsigned flags)
4791 {
4792 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
4793 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
4794 	bool writeback = flags & AMDGPU_FENCE_FLAG_TC_WB_ONLY;
4795 
4796 	/* RELEASE_MEM - flush caches, send int */
4797 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
4798 	amdgpu_ring_write(ring, ((writeback ? (EOP_TC_WB_ACTION_EN |
4799 					       EOP_TC_NC_ACTION_EN) :
4800 					      (EOP_TCL1_ACTION_EN |
4801 					       EOP_TC_ACTION_EN |
4802 					       EOP_TC_WB_ACTION_EN |
4803 					       EOP_TC_MD_ACTION_EN)) |
4804 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
4805 				 EVENT_INDEX(5)));
4806 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
4807 
4808 	/*
4809 	 * the address should be Qword aligned if 64bit write, Dword
4810 	 * aligned if only send 32bit data low (discard data high)
4811 	 */
4812 	if (write64bit)
4813 		BUG_ON(addr & 0x7);
4814 	else
4815 		BUG_ON(addr & 0x3);
4816 	amdgpu_ring_write(ring, lower_32_bits(addr));
4817 	amdgpu_ring_write(ring, upper_32_bits(addr));
4818 	amdgpu_ring_write(ring, lower_32_bits(seq));
4819 	amdgpu_ring_write(ring, upper_32_bits(seq));
4820 	amdgpu_ring_write(ring, 0);
4821 }
4822 
4823 static void gfx_v9_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
4824 {
4825 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
4826 	uint32_t seq = ring->fence_drv.sync_seq;
4827 	uint64_t addr = ring->fence_drv.gpu_addr;
4828 
4829 	gfx_v9_0_wait_reg_mem(ring, usepfp, 1, 0,
4830 			      lower_32_bits(addr), upper_32_bits(addr),
4831 			      seq, 0xffffffff, 4);
4832 }
4833 
4834 static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
4835 					unsigned vmid, uint64_t pd_addr)
4836 {
4837 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
4838 
4839 	/* compute doesn't have PFP */
4840 	if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
4841 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
4842 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
4843 		amdgpu_ring_write(ring, 0x0);
4844 	}
4845 }
4846 
4847 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring)
4848 {
4849 	return ring->adev->wb.wb[ring->rptr_offs]; /* gfx9 hardware is 32bit rptr */
4850 }
4851 
4852 static u64 gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
4853 {
4854 	u64 wptr;
4855 
4856 	/* XXX check if swapping is necessary on BE */
4857 	if (ring->use_doorbell)
4858 		wptr = atomic64_read((atomic64_t *)&ring->adev->wb.wb[ring->wptr_offs]);
4859 	else
4860 		BUG();
4861 	return wptr;
4862 }
4863 
4864 static void gfx_v9_0_ring_set_pipe_percent(struct amdgpu_ring *ring,
4865 					   bool acquire)
4866 {
4867 	struct amdgpu_device *adev = ring->adev;
4868 	int pipe_num, tmp, reg;
4869 	int pipe_percent = acquire ? SPI_WCL_PIPE_PERCENT_GFX__VALUE_MASK : 0x1;
4870 
4871 	pipe_num = ring->me * adev->gfx.mec.num_pipe_per_mec + ring->pipe;
4872 
4873 	/* first me only has 2 entries, GFX and HP3D */
4874 	if (ring->me > 0)
4875 		pipe_num -= 2;
4876 
4877 	reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_GFX) + pipe_num;
4878 	tmp = RREG32(reg);
4879 	tmp = REG_SET_FIELD(tmp, SPI_WCL_PIPE_PERCENT_GFX, VALUE, pipe_percent);
4880 	WREG32(reg, tmp);
4881 }
4882 
4883 static void gfx_v9_0_pipe_reserve_resources(struct amdgpu_device *adev,
4884 					    struct amdgpu_ring *ring,
4885 					    bool acquire)
4886 {
4887 	int i, pipe;
4888 	bool reserve;
4889 	struct amdgpu_ring *iring;
4890 
4891 	mutex_lock(&adev->gfx.pipe_reserve_mutex);
4892 	pipe = amdgpu_gfx_mec_queue_to_bit(adev, ring->me, ring->pipe, 0);
4893 	if (acquire)
4894 		set_bit(pipe, adev->gfx.pipe_reserve_bitmap);
4895 	else
4896 		clear_bit(pipe, adev->gfx.pipe_reserve_bitmap);
4897 
4898 	if (!bitmap_weight(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES)) {
4899 		/* Clear all reservations - everyone reacquires all resources */
4900 		for (i = 0; i < adev->gfx.num_gfx_rings; ++i)
4901 			gfx_v9_0_ring_set_pipe_percent(&adev->gfx.gfx_ring[i],
4902 						       true);
4903 
4904 		for (i = 0; i < adev->gfx.num_compute_rings; ++i)
4905 			gfx_v9_0_ring_set_pipe_percent(&adev->gfx.compute_ring[i],
4906 						       true);
4907 	} else {
4908 		/* Lower all pipes without a current reservation */
4909 		for (i = 0; i < adev->gfx.num_gfx_rings; ++i) {
4910 			iring = &adev->gfx.gfx_ring[i];
4911 			pipe = amdgpu_gfx_mec_queue_to_bit(adev,
4912 							   iring->me,
4913 							   iring->pipe,
4914 							   0);
4915 			reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap);
4916 			gfx_v9_0_ring_set_pipe_percent(iring, reserve);
4917 		}
4918 
4919 		for (i = 0; i < adev->gfx.num_compute_rings; ++i) {
4920 			iring = &adev->gfx.compute_ring[i];
4921 			pipe = amdgpu_gfx_mec_queue_to_bit(adev,
4922 							   iring->me,
4923 							   iring->pipe,
4924 							   0);
4925 			reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap);
4926 			gfx_v9_0_ring_set_pipe_percent(iring, reserve);
4927 		}
4928 	}
4929 
4930 	mutex_unlock(&adev->gfx.pipe_reserve_mutex);
4931 }
4932 
4933 static void gfx_v9_0_hqd_set_priority(struct amdgpu_device *adev,
4934 				      struct amdgpu_ring *ring,
4935 				      bool acquire)
4936 {
4937 	uint32_t pipe_priority = acquire ? 0x2 : 0x0;
4938 	uint32_t queue_priority = acquire ? 0xf : 0x0;
4939 
4940 	mutex_lock(&adev->srbm_mutex);
4941 	soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4942 
4943 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PIPE_PRIORITY, pipe_priority);
4944 	WREG32_SOC15_RLC(GC, 0, mmCP_HQD_QUEUE_PRIORITY, queue_priority);
4945 
4946 	soc15_grbm_select(adev, 0, 0, 0, 0);
4947 	mutex_unlock(&adev->srbm_mutex);
4948 }
4949 
4950 static void gfx_v9_0_ring_set_priority_compute(struct amdgpu_ring *ring,
4951 					       enum drm_sched_priority priority)
4952 {
4953 	struct amdgpu_device *adev = ring->adev;
4954 	bool acquire = priority == DRM_SCHED_PRIORITY_HIGH_HW;
4955 
4956 	if (ring->funcs->type != AMDGPU_RING_TYPE_COMPUTE)
4957 		return;
4958 
4959 	gfx_v9_0_hqd_set_priority(adev, ring, acquire);
4960 	gfx_v9_0_pipe_reserve_resources(adev, ring, acquire);
4961 }
4962 
4963 static void gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
4964 {
4965 	struct amdgpu_device *adev = ring->adev;
4966 
4967 	/* XXX check if swapping is necessary on BE */
4968 	if (ring->use_doorbell) {
4969 		atomic64_set((atomic64_t*)&adev->wb.wb[ring->wptr_offs], ring->wptr);
4970 		WDOORBELL64(ring->doorbell_index, ring->wptr);
4971 	} else{
4972 		BUG(); /* only DOORBELL method supported on gfx9 now */
4973 	}
4974 }
4975 
4976 static void gfx_v9_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
4977 					 u64 seq, unsigned int flags)
4978 {
4979 	struct amdgpu_device *adev = ring->adev;
4980 
4981 	/* we only allocate 32bit for each seq wb address */
4982 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
4983 
4984 	/* write fence seq to the "addr" */
4985 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4986 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4987 				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
4988 	amdgpu_ring_write(ring, lower_32_bits(addr));
4989 	amdgpu_ring_write(ring, upper_32_bits(addr));
4990 	amdgpu_ring_write(ring, lower_32_bits(seq));
4991 
4992 	if (flags & AMDGPU_FENCE_FLAG_INT) {
4993 		/* set register to trigger INT */
4994 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4995 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4996 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
4997 		amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, mmCPC_INT_STATUS));
4998 		amdgpu_ring_write(ring, 0);
4999 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
5000 	}
5001 }
5002 
5003 static void gfx_v9_ring_emit_sb(struct amdgpu_ring *ring)
5004 {
5005 	amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
5006 	amdgpu_ring_write(ring, 0);
5007 }
5008 
5009 static void gfx_v9_0_ring_emit_ce_meta(struct amdgpu_ring *ring)
5010 {
5011 	struct v9_ce_ib_state ce_payload = {0};
5012 	uint64_t csa_addr;
5013 	int cnt;
5014 
5015 	cnt = (sizeof(ce_payload) >> 2) + 4 - 2;
5016 	csa_addr = amdgpu_csa_vaddr(ring->adev);
5017 
5018 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5019 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
5020 				 WRITE_DATA_DST_SEL(8) |
5021 				 WR_CONFIRM) |
5022 				 WRITE_DATA_CACHE_POLICY(0));
5023 	amdgpu_ring_write(ring, lower_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, ce_payload)));
5024 	amdgpu_ring_write(ring, upper_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, ce_payload)));
5025 	amdgpu_ring_write_multiple(ring, (void *)&ce_payload, sizeof(ce_payload) >> 2);
5026 }
5027 
5028 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring)
5029 {
5030 	struct v9_de_ib_state de_payload = {0};
5031 	uint64_t csa_addr, gds_addr;
5032 	int cnt;
5033 
5034 	csa_addr = amdgpu_csa_vaddr(ring->adev);
5035 	gds_addr = csa_addr + 4096;
5036 	de_payload.gds_backup_addrlo = lower_32_bits(gds_addr);
5037 	de_payload.gds_backup_addrhi = upper_32_bits(gds_addr);
5038 
5039 	cnt = (sizeof(de_payload) >> 2) + 4 - 2;
5040 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5041 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
5042 				 WRITE_DATA_DST_SEL(8) |
5043 				 WR_CONFIRM) |
5044 				 WRITE_DATA_CACHE_POLICY(0));
5045 	amdgpu_ring_write(ring, lower_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, de_payload)));
5046 	amdgpu_ring_write(ring, upper_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, de_payload)));
5047 	amdgpu_ring_write_multiple(ring, (void *)&de_payload, sizeof(de_payload) >> 2);
5048 }
5049 
5050 static void gfx_v9_0_ring_emit_tmz(struct amdgpu_ring *ring, bool start)
5051 {
5052 	amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0));
5053 	amdgpu_ring_write(ring, FRAME_CMD(start ? 0 : 1)); /* frame_end */
5054 }
5055 
5056 static void gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
5057 {
5058 	uint32_t dw2 = 0;
5059 
5060 	if (amdgpu_sriov_vf(ring->adev))
5061 		gfx_v9_0_ring_emit_ce_meta(ring);
5062 
5063 	gfx_v9_0_ring_emit_tmz(ring, true);
5064 
5065 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
5066 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
5067 		/* set load_global_config & load_global_uconfig */
5068 		dw2 |= 0x8001;
5069 		/* set load_cs_sh_regs */
5070 		dw2 |= 0x01000000;
5071 		/* set load_per_context_state & load_gfx_sh_regs for GFX */
5072 		dw2 |= 0x10002;
5073 
5074 		/* set load_ce_ram if preamble presented */
5075 		if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
5076 			dw2 |= 0x10000000;
5077 	} else {
5078 		/* still load_ce_ram if this is the first time preamble presented
5079 		 * although there is no context switch happens.
5080 		 */
5081 		if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
5082 			dw2 |= 0x10000000;
5083 	}
5084 
5085 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
5086 	amdgpu_ring_write(ring, dw2);
5087 	amdgpu_ring_write(ring, 0);
5088 }
5089 
5090 static unsigned gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
5091 {
5092 	unsigned ret;
5093 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
5094 	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
5095 	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
5096 	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
5097 	ret = ring->wptr & ring->buf_mask;
5098 	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
5099 	return ret;
5100 }
5101 
5102 static void gfx_v9_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
5103 {
5104 	unsigned cur;
5105 	BUG_ON(offset > ring->buf_mask);
5106 	BUG_ON(ring->ring[offset] != 0x55aa55aa);
5107 
5108 	cur = (ring->wptr & ring->buf_mask) - 1;
5109 	if (likely(cur > offset))
5110 		ring->ring[offset] = cur - offset;
5111 	else
5112 		ring->ring[offset] = (ring->ring_size>>2) - offset + cur;
5113 }
5114 
5115 static void gfx_v9_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg)
5116 {
5117 	struct amdgpu_device *adev = ring->adev;
5118 
5119 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
5120 	amdgpu_ring_write(ring, 0 |	/* src: register*/
5121 				(5 << 8) |	/* dst: memory */
5122 				(1 << 20));	/* write confirm */
5123 	amdgpu_ring_write(ring, reg);
5124 	amdgpu_ring_write(ring, 0);
5125 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
5126 				adev->virt.reg_val_offs * 4));
5127 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
5128 				adev->virt.reg_val_offs * 4));
5129 }
5130 
5131 static void gfx_v9_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
5132 				    uint32_t val)
5133 {
5134 	uint32_t cmd = 0;
5135 
5136 	switch (ring->funcs->type) {
5137 	case AMDGPU_RING_TYPE_GFX:
5138 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
5139 		break;
5140 	case AMDGPU_RING_TYPE_KIQ:
5141 		cmd = (1 << 16); /* no inc addr */
5142 		break;
5143 	default:
5144 		cmd = WR_CONFIRM;
5145 		break;
5146 	}
5147 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5148 	amdgpu_ring_write(ring, cmd);
5149 	amdgpu_ring_write(ring, reg);
5150 	amdgpu_ring_write(ring, 0);
5151 	amdgpu_ring_write(ring, val);
5152 }
5153 
5154 static void gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
5155 					uint32_t val, uint32_t mask)
5156 {
5157 	gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
5158 }
5159 
5160 static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
5161 						  uint32_t reg0, uint32_t reg1,
5162 						  uint32_t ref, uint32_t mask)
5163 {
5164 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5165 	struct amdgpu_device *adev = ring->adev;
5166 	bool fw_version_ok = (ring->funcs->type == AMDGPU_RING_TYPE_GFX) ?
5167 		adev->gfx.me_fw_write_wait : adev->gfx.mec_fw_write_wait;
5168 
5169 	if (fw_version_ok)
5170 		gfx_v9_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1,
5171 				      ref, mask, 0x20);
5172 	else
5173 		amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1,
5174 							   ref, mask);
5175 }
5176 
5177 static void gfx_v9_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
5178 {
5179 	struct amdgpu_device *adev = ring->adev;
5180 	uint32_t value = 0;
5181 
5182 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
5183 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
5184 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
5185 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
5186 	WREG32_SOC15(GC, 0, mmSQ_CMD, value);
5187 }
5188 
5189 static void gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
5190 						 enum amdgpu_interrupt_state state)
5191 {
5192 	switch (state) {
5193 	case AMDGPU_IRQ_STATE_DISABLE:
5194 	case AMDGPU_IRQ_STATE_ENABLE:
5195 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5196 			       TIME_STAMP_INT_ENABLE,
5197 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5198 		break;
5199 	default:
5200 		break;
5201 	}
5202 }
5203 
5204 static void gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
5205 						     int me, int pipe,
5206 						     enum amdgpu_interrupt_state state)
5207 {
5208 	u32 mec_int_cntl, mec_int_cntl_reg;
5209 
5210 	/*
5211 	 * amdgpu controls only the first MEC. That's why this function only
5212 	 * handles the setting of interrupts for this specific MEC. All other
5213 	 * pipes' interrupts are set by amdkfd.
5214 	 */
5215 
5216 	if (me == 1) {
5217 		switch (pipe) {
5218 		case 0:
5219 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
5220 			break;
5221 		case 1:
5222 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL);
5223 			break;
5224 		case 2:
5225 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL);
5226 			break;
5227 		case 3:
5228 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL);
5229 			break;
5230 		default:
5231 			DRM_DEBUG("invalid pipe %d\n", pipe);
5232 			return;
5233 		}
5234 	} else {
5235 		DRM_DEBUG("invalid me %d\n", me);
5236 		return;
5237 	}
5238 
5239 	switch (state) {
5240 	case AMDGPU_IRQ_STATE_DISABLE:
5241 		mec_int_cntl = RREG32(mec_int_cntl_reg);
5242 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5243 					     TIME_STAMP_INT_ENABLE, 0);
5244 		WREG32(mec_int_cntl_reg, mec_int_cntl);
5245 		break;
5246 	case AMDGPU_IRQ_STATE_ENABLE:
5247 		mec_int_cntl = RREG32(mec_int_cntl_reg);
5248 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5249 					     TIME_STAMP_INT_ENABLE, 1);
5250 		WREG32(mec_int_cntl_reg, mec_int_cntl);
5251 		break;
5252 	default:
5253 		break;
5254 	}
5255 }
5256 
5257 static int gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
5258 					     struct amdgpu_irq_src *source,
5259 					     unsigned type,
5260 					     enum amdgpu_interrupt_state state)
5261 {
5262 	switch (state) {
5263 	case AMDGPU_IRQ_STATE_DISABLE:
5264 	case AMDGPU_IRQ_STATE_ENABLE:
5265 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5266 			       PRIV_REG_INT_ENABLE,
5267 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5268 		break;
5269 	default:
5270 		break;
5271 	}
5272 
5273 	return 0;
5274 }
5275 
5276 static int gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
5277 					      struct amdgpu_irq_src *source,
5278 					      unsigned type,
5279 					      enum amdgpu_interrupt_state state)
5280 {
5281 	switch (state) {
5282 	case AMDGPU_IRQ_STATE_DISABLE:
5283 	case AMDGPU_IRQ_STATE_ENABLE:
5284 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5285 			       PRIV_INSTR_INT_ENABLE,
5286 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5287 	default:
5288 		break;
5289 	}
5290 
5291 	return 0;
5292 }
5293 
5294 #define ENABLE_ECC_ON_ME_PIPE(me, pipe)				\
5295 	WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
5296 			CP_ECC_ERROR_INT_ENABLE, 1)
5297 
5298 #define DISABLE_ECC_ON_ME_PIPE(me, pipe)			\
5299 	WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
5300 			CP_ECC_ERROR_INT_ENABLE, 0)
5301 
5302 static int gfx_v9_0_set_cp_ecc_error_state(struct amdgpu_device *adev,
5303 					      struct amdgpu_irq_src *source,
5304 					      unsigned type,
5305 					      enum amdgpu_interrupt_state state)
5306 {
5307 	switch (state) {
5308 	case AMDGPU_IRQ_STATE_DISABLE:
5309 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5310 				CP_ECC_ERROR_INT_ENABLE, 0);
5311 		DISABLE_ECC_ON_ME_PIPE(1, 0);
5312 		DISABLE_ECC_ON_ME_PIPE(1, 1);
5313 		DISABLE_ECC_ON_ME_PIPE(1, 2);
5314 		DISABLE_ECC_ON_ME_PIPE(1, 3);
5315 		break;
5316 
5317 	case AMDGPU_IRQ_STATE_ENABLE:
5318 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5319 				CP_ECC_ERROR_INT_ENABLE, 1);
5320 		ENABLE_ECC_ON_ME_PIPE(1, 0);
5321 		ENABLE_ECC_ON_ME_PIPE(1, 1);
5322 		ENABLE_ECC_ON_ME_PIPE(1, 2);
5323 		ENABLE_ECC_ON_ME_PIPE(1, 3);
5324 		break;
5325 	default:
5326 		break;
5327 	}
5328 
5329 	return 0;
5330 }
5331 
5332 
5333 static int gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device *adev,
5334 					    struct amdgpu_irq_src *src,
5335 					    unsigned type,
5336 					    enum amdgpu_interrupt_state state)
5337 {
5338 	switch (type) {
5339 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
5340 		gfx_v9_0_set_gfx_eop_interrupt_state(adev, state);
5341 		break;
5342 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
5343 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
5344 		break;
5345 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
5346 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
5347 		break;
5348 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
5349 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
5350 		break;
5351 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
5352 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
5353 		break;
5354 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
5355 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
5356 		break;
5357 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
5358 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
5359 		break;
5360 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
5361 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
5362 		break;
5363 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
5364 		gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
5365 		break;
5366 	default:
5367 		break;
5368 	}
5369 	return 0;
5370 }
5371 
5372 static int gfx_v9_0_eop_irq(struct amdgpu_device *adev,
5373 			    struct amdgpu_irq_src *source,
5374 			    struct amdgpu_iv_entry *entry)
5375 {
5376 	int i;
5377 	u8 me_id, pipe_id, queue_id;
5378 	struct amdgpu_ring *ring;
5379 
5380 	DRM_DEBUG("IH: CP EOP\n");
5381 	me_id = (entry->ring_id & 0x0c) >> 2;
5382 	pipe_id = (entry->ring_id & 0x03) >> 0;
5383 	queue_id = (entry->ring_id & 0x70) >> 4;
5384 
5385 	switch (me_id) {
5386 	case 0:
5387 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
5388 		break;
5389 	case 1:
5390 	case 2:
5391 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5392 			ring = &adev->gfx.compute_ring[i];
5393 			/* Per-queue interrupt is supported for MEC starting from VI.
5394 			  * The interrupt can only be enabled/disabled per pipe instead of per queue.
5395 			  */
5396 			if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
5397 				amdgpu_fence_process(ring);
5398 		}
5399 		break;
5400 	}
5401 	return 0;
5402 }
5403 
5404 static void gfx_v9_0_fault(struct amdgpu_device *adev,
5405 			   struct amdgpu_iv_entry *entry)
5406 {
5407 	u8 me_id, pipe_id, queue_id;
5408 	struct amdgpu_ring *ring;
5409 	int i;
5410 
5411 	me_id = (entry->ring_id & 0x0c) >> 2;
5412 	pipe_id = (entry->ring_id & 0x03) >> 0;
5413 	queue_id = (entry->ring_id & 0x70) >> 4;
5414 
5415 	switch (me_id) {
5416 	case 0:
5417 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
5418 		break;
5419 	case 1:
5420 	case 2:
5421 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5422 			ring = &adev->gfx.compute_ring[i];
5423 			if (ring->me == me_id && ring->pipe == pipe_id &&
5424 			    ring->queue == queue_id)
5425 				drm_sched_fault(&ring->sched);
5426 		}
5427 		break;
5428 	}
5429 }
5430 
5431 static int gfx_v9_0_priv_reg_irq(struct amdgpu_device *adev,
5432 				 struct amdgpu_irq_src *source,
5433 				 struct amdgpu_iv_entry *entry)
5434 {
5435 	DRM_ERROR("Illegal register access in command stream\n");
5436 	gfx_v9_0_fault(adev, entry);
5437 	return 0;
5438 }
5439 
5440 static int gfx_v9_0_priv_inst_irq(struct amdgpu_device *adev,
5441 				  struct amdgpu_irq_src *source,
5442 				  struct amdgpu_iv_entry *entry)
5443 {
5444 	DRM_ERROR("Illegal instruction in command stream\n");
5445 	gfx_v9_0_fault(adev, entry);
5446 	return 0;
5447 }
5448 
5449 
5450 static const struct ras_gfx_subblock_reg ras_subblock_regs[] = {
5451 	{ "CPC_SCRATCH", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT),
5452 	  SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, SEC_COUNT),
5453 	  SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, DED_COUNT)
5454 	},
5455 	{ "CPC_UCODE", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT),
5456 	  SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, SEC_COUNT),
5457 	  SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, DED_COUNT)
5458 	},
5459 	{ "CPF_ROQ_ME1", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
5460 	  SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME1),
5461 	  0, 0
5462 	},
5463 	{ "CPF_ROQ_ME2", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
5464 	  SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME2),
5465 	  0, 0
5466 	},
5467 	{ "CPF_TAG", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT),
5468 	  SOC15_REG_FIELD(CPF_EDC_TAG_CNT, SEC_COUNT),
5469 	  SOC15_REG_FIELD(CPF_EDC_TAG_CNT, DED_COUNT)
5470 	},
5471 	{ "CPG_DMA_ROQ", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
5472 	  SOC15_REG_FIELD(CPG_EDC_DMA_CNT, ROQ_COUNT),
5473 	  0, 0
5474 	},
5475 	{ "CPG_DMA_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
5476 	  SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_SEC_COUNT),
5477 	  SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_DED_COUNT)
5478 	},
5479 	{ "CPG_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT),
5480 	  SOC15_REG_FIELD(CPG_EDC_TAG_CNT, SEC_COUNT),
5481 	  SOC15_REG_FIELD(CPG_EDC_TAG_CNT, DED_COUNT)
5482 	},
5483 	{ "DC_CSINVOC", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT),
5484 	  SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, COUNT_ME1),
5485 	  0, 0
5486 	},
5487 	{ "DC_RESTORE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT),
5488 	  SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, COUNT_ME1),
5489 	  0, 0
5490 	},
5491 	{ "DC_STATE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT),
5492 	  SOC15_REG_FIELD(DC_EDC_STATE_CNT, COUNT_ME1),
5493 	  0, 0
5494 	},
5495 	{ "GDS_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
5496 	  SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_SEC),
5497 	  SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_DED)
5498 	},
5499 	{ "GDS_INPUT_QUEUE", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
5500 	  SOC15_REG_FIELD(GDS_EDC_CNT, GDS_INPUT_QUEUE_SED),
5501 	  0, 0
5502 	},
5503 	{ "GDS_ME0_CS_PIPE_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
5504 	  SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_SEC),
5505 	  SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_DED)
5506 	},
5507 	{ "GDS_OA_PHY_PHY_CMD_RAM_MEM",
5508 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
5509 	  SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_SEC),
5510 	  SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_DED)
5511 	},
5512 	{ "GDS_OA_PHY_PHY_DATA_RAM_MEM",
5513 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
5514 	  SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_DATA_RAM_MEM_SED),
5515 	  0, 0
5516 	},
5517 	{ "GDS_OA_PIPE_ME1_PIPE0_PIPE_MEM",
5518 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
5519 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_SEC),
5520 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_DED)
5521 	},
5522 	{ "GDS_OA_PIPE_ME1_PIPE1_PIPE_MEM",
5523 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
5524 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_SEC),
5525 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_DED)
5526 	},
5527 	{ "GDS_OA_PIPE_ME1_PIPE2_PIPE_MEM",
5528 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
5529 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_SEC),
5530 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_DED)
5531 	},
5532 	{ "GDS_OA_PIPE_ME1_PIPE3_PIPE_MEM",
5533 	  SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
5534 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_SEC),
5535 	  SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_DED)
5536 	},
5537 	{ "SPI_SR_MEM", SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT),
5538 	  SOC15_REG_FIELD(SPI_EDC_CNT, SPI_SR_MEM_SED_COUNT),
5539 	  0, 0
5540 	},
5541 	{ "TA_FS_DFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
5542 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_SEC_COUNT),
5543 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_DED_COUNT)
5544 	},
5545 	{ "TA_FS_AFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
5546 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_SED_COUNT),
5547 	  0, 0
5548 	},
5549 	{ "TA_FL_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
5550 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FL_LFIFO_SED_COUNT),
5551 	  0, 0
5552 	},
5553 	{ "TA_FX_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
5554 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FX_LFIFO_SED_COUNT),
5555 	  0, 0
5556 	},
5557 	{ "TA_FS_CFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
5558 	  SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_CFIFO_SED_COUNT),
5559 	  0, 0
5560 	},
5561 	{ "TCA_HOLE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
5562 	  SOC15_REG_FIELD(TCA_EDC_CNT, HOLE_FIFO_SED_COUNT),
5563 	  0, 0
5564 	},
5565 	{ "TCA_REQ_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
5566 	  SOC15_REG_FIELD(TCA_EDC_CNT, REQ_FIFO_SED_COUNT),
5567 	  0, 0
5568 	},
5569 	{ "TCC_CACHE_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5570 	  SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_SEC_COUNT),
5571 	  SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_DED_COUNT)
5572 	},
5573 	{ "TCC_CACHE_DIRTY", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5574 	  SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_SEC_COUNT),
5575 	  SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_DED_COUNT)
5576 	},
5577 	{ "TCC_HIGH_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5578 	  SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_SEC_COUNT),
5579 	  SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_DED_COUNT)
5580 	},
5581 	{ "TCC_LOW_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5582 	  SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_SEC_COUNT),
5583 	  SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_DED_COUNT)
5584 	},
5585 	{ "TCC_SRC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5586 	  SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_SEC_COUNT),
5587 	  SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_DED_COUNT)
5588 	},
5589 	{ "TCC_IN_USE_DEC", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5590 	  SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_DEC_SED_COUNT),
5591 	  0, 0
5592 	},
5593 	{ "TCC_IN_USE_TRANSFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5594 	  SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_TRANSFER_SED_COUNT),
5595 	  0, 0
5596 	},
5597 	{ "TCC_LATENCY_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5598 	  SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_SED_COUNT),
5599 	  0, 0
5600 	},
5601 	{ "TCC_RETURN_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5602 	  SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_DATA_SED_COUNT),
5603 	  0, 0
5604 	},
5605 	{ "TCC_RETURN_CONTROL", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5606 	  SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_CONTROL_SED_COUNT),
5607 	  0, 0
5608 	},
5609 	{ "TCC_UC_ATOMIC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5610 	  SOC15_REG_FIELD(TCC_EDC_CNT, UC_ATOMIC_FIFO_SED_COUNT),
5611 	  0, 0
5612 	},
5613 	{ "TCC_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
5614 	  SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_RETURN_SED_COUNT),
5615 	  0, 0
5616 	},
5617 	{ "TCC_WRITE_CACHE_READ", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
5618 	  SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_CACHE_READ_SED_COUNT),
5619 	  0, 0
5620 	},
5621 	{ "TCC_SRC_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
5622 	  SOC15_REG_FIELD(TCC_EDC_CNT2, SRC_FIFO_NEXT_RAM_SED_COUNT),
5623 	  0, 0
5624 	},
5625 	{ "TCC_LATENCY_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
5626 	  SOC15_REG_FIELD(TCC_EDC_CNT2, LATENCY_FIFO_NEXT_RAM_SED_COUNT),
5627 	  0, 0
5628 	},
5629 	{ "TCC_CACHE_TAG_PROBE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
5630 	  SOC15_REG_FIELD(TCC_EDC_CNT2, CACHE_TAG_PROBE_FIFO_SED_COUNT),
5631 	  0, 0
5632 	},
5633 	{ "TCC_WRRET_TAG_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
5634 	  SOC15_REG_FIELD(TCC_EDC_CNT2, WRRET_TAG_WRITE_RETURN_SED_COUNT),
5635 	  0, 0
5636 	},
5637 	{ "TCC_ATOMIC_RETURN_BUFFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
5638 	  SOC15_REG_FIELD(TCC_EDC_CNT2, ATOMIC_RETURN_BUFFER_SED_COUNT),
5639 	  0, 0
5640 	},
5641 	{ "TCI_WRITE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT),
5642 	  SOC15_REG_FIELD(TCI_EDC_CNT, WRITE_RAM_SED_COUNT),
5643 	  0, 0
5644 	},
5645 	{ "TCP_CACHE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
5646 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_SEC_COUNT),
5647 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_DED_COUNT)
5648 	},
5649 	{ "TCP_LFIFO_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
5650 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_SEC_COUNT),
5651 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_DED_COUNT)
5652 	},
5653 	{ "TCP_CMD_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
5654 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CMD_FIFO_SED_COUNT),
5655 	  0, 0
5656 	},
5657 	{ "TCP_VM_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
5658 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, VM_FIFO_SEC_COUNT),
5659 	  0, 0
5660 	},
5661 	{ "TCP_DB_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
5662 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, DB_RAM_SED_COUNT),
5663 	  0, 0
5664 	},
5665 	{ "TCP_UTCL1_LFIFO0", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
5666 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_SEC_COUNT),
5667 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_DED_COUNT)
5668 	},
5669 	{ "TCP_UTCL1_LFIFO1", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
5670 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_SEC_COUNT),
5671 	  SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_DED_COUNT)
5672 	},
5673 	{ "TD_SS_FIFO_LO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
5674 	  SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_SEC_COUNT),
5675 	  SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_DED_COUNT)
5676 	},
5677 	{ "TD_SS_FIFO_HI", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
5678 	  SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_SEC_COUNT),
5679 	  SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_DED_COUNT)
5680 	},
5681 	{ "TD_CS_FIFO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
5682 	  SOC15_REG_FIELD(TD_EDC_CNT, CS_FIFO_SED_COUNT),
5683 	  0, 0
5684 	},
5685 	{ "SQ_LDS_D", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
5686 	  SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_SEC_COUNT),
5687 	  SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_DED_COUNT)
5688 	},
5689 	{ "SQ_LDS_I", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
5690 	  SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_SEC_COUNT),
5691 	  SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_DED_COUNT)
5692 	},
5693 	{ "SQ_SGPR", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
5694 	  SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_SEC_COUNT),
5695 	  SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_DED_COUNT)
5696 	},
5697 	{ "SQ_VGPR0", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
5698 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_SEC_COUNT),
5699 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_DED_COUNT)
5700 	},
5701 	{ "SQ_VGPR1", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
5702 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_SEC_COUNT),
5703 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_DED_COUNT)
5704 	},
5705 	{ "SQ_VGPR2", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
5706 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_SEC_COUNT),
5707 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_DED_COUNT)
5708 	},
5709 	{ "SQ_VGPR3", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
5710 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_SEC_COUNT),
5711 	  SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_DED_COUNT)
5712 	},
5713 	{ "SQC_DATA_CU0_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
5714 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_SEC_COUNT),
5715 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_DED_COUNT)
5716 	},
5717 	{ "SQC_DATA_CU0_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
5718 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_SEC_COUNT),
5719 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_DED_COUNT)
5720 	},
5721 	{ "SQC_DATA_CU1_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
5722 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_SEC_COUNT),
5723 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_DED_COUNT)
5724 	},
5725 	{ "SQC_DATA_CU1_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
5726 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_SEC_COUNT),
5727 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_DED_COUNT)
5728 	},
5729 	{ "SQC_DATA_CU2_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
5730 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_SEC_COUNT),
5731 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_DED_COUNT)
5732 	},
5733 	{ "SQC_DATA_CU2_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
5734 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_SEC_COUNT),
5735 	  SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_DED_COUNT)
5736 	},
5737 	{ "SQC_INST_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
5738 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_SEC_COUNT),
5739 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_DED_COUNT)
5740 	},
5741 	{ "SQC_INST_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
5742 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_SEC_COUNT),
5743 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_DED_COUNT)
5744 	},
5745 	{ "SQC_DATA_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
5746 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_SEC_COUNT),
5747 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_DED_COUNT)
5748 	},
5749 	{ "SQC_DATA_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
5750 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_SEC_COUNT),
5751 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_DED_COUNT)
5752 	},
5753 	{ "SQC_INST_BANKA_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
5754 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_UTCL1_MISS_FIFO_SED_COUNT),
5755 	  0, 0
5756 	},
5757 	{ "SQC_INST_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
5758 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_MISS_FIFO_SED_COUNT),
5759 	  0, 0
5760 	},
5761 	{ "SQC_DATA_BANKA_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
5762 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_HIT_FIFO_SED_COUNT),
5763 	  0, 0
5764 	},
5765 	{ "SQC_DATA_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
5766 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_MISS_FIFO_SED_COUNT),
5767 	  0, 0
5768 	},
5769 	{ "SQC_DATA_BANKA_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
5770 	  SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_DIRTY_BIT_RAM_SED_COUNT),
5771 	  0, 0
5772 	},
5773 	{ "SQC_INST_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
5774 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_SEC_COUNT),
5775 	  SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_DED_COUNT)
5776 	},
5777 	{ "SQC_INST_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
5778 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_SEC_COUNT),
5779 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_DED_COUNT)
5780 	},
5781 	{ "SQC_INST_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
5782 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_SEC_COUNT),
5783 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_DED_COUNT)
5784 	},
5785 	{ "SQC_DATA_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
5786 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_SEC_COUNT),
5787 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_DED_COUNT)
5788 	},
5789 	{ "SQC_DATA_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
5790 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_SEC_COUNT),
5791 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_DED_COUNT)
5792 	},
5793 	{ "SQC_INST_BANKB_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
5794 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_UTCL1_MISS_FIFO_SED_COUNT),
5795 	  0, 0
5796 	},
5797 	{ "SQC_INST_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
5798 	  SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_MISS_FIFO_SED_COUNT),
5799 	  0, 0
5800 	},
5801 	{ "SQC_DATA_BANKB_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
5802 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_HIT_FIFO_SED_COUNT),
5803 	  0, 0
5804 	},
5805 	{ "SQC_DATA_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
5806 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_MISS_FIFO_SED_COUNT),
5807 	  0, 0
5808 	},
5809 	{ "SQC_DATA_BANKB_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
5810 	  SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_DIRTY_BIT_RAM_SED_COUNT),
5811 	  0, 0
5812 	},
5813 	{ "EA_DRAMRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
5814 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
5815 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT)
5816 	},
5817 	{ "EA_DRAMWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
5818 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
5819 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT)
5820 	},
5821 	{ "EA_DRAMWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
5822 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
5823 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT)
5824 	},
5825 	{ "EA_RRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
5826 	  SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
5827 	  SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_DED_COUNT)
5828 	},
5829 	{ "EA_WRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
5830 	  SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
5831 	  SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_DED_COUNT)
5832 	},
5833 	{ "EA_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
5834 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
5835 	  0, 0
5836 	},
5837 	{ "EA_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
5838 	  SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
5839 	  0, 0
5840 	},
5841 	{ "EA_IORD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
5842 	  SOC15_REG_FIELD(GCEA_EDC_CNT, IORD_CMDMEM_SED_COUNT),
5843 	  0, 0
5844 	},
5845 	{ "EA_IOWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
5846 	  SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
5847 	  0, 0
5848 	},
5849 	{ "EA_IOWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
5850 	  SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_DATAMEM_SED_COUNT),
5851 	  0, 0
5852 	},
5853 	{ "GMIRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
5854 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
5855 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT)
5856 	},
5857 	{ "GMIWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
5858 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
5859 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT)
5860 	},
5861 	{ "GMIWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
5862 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
5863 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT)
5864 	},
5865 	{ "GMIRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
5866 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
5867 	  0, 0
5868 	},
5869 	{ "GMIWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
5870 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
5871 	  0, 0
5872 	},
5873 	{ "MAM_D0MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
5874 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D0MEM_SED_COUNT),
5875 	  0, 0
5876 	},
5877 	{ "MAM_D1MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
5878 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D1MEM_SED_COUNT),
5879 	  0, 0
5880 	},
5881 	{ "MAM_D2MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
5882 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D2MEM_SED_COUNT),
5883 	  0, 0
5884 	},
5885 	{ "MAM_D3MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
5886 	  SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D3MEM_SED_COUNT),
5887 	  0, 0
5888 	}
5889 };
5890 
5891 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
5892 				     void *inject_if)
5893 {
5894 	struct ras_inject_if *info = (struct ras_inject_if *)inject_if;
5895 	int ret;
5896 	struct ta_ras_trigger_error_input block_info = { 0 };
5897 
5898 	if (adev->asic_type != CHIP_VEGA20)
5899 		return -EINVAL;
5900 
5901 	if (info->head.sub_block_index >= ARRAY_SIZE(ras_gfx_subblocks))
5902 		return -EINVAL;
5903 
5904 	if (!ras_gfx_subblocks[info->head.sub_block_index].name)
5905 		return -EPERM;
5906 
5907 	if (!(ras_gfx_subblocks[info->head.sub_block_index].hw_supported_error_type &
5908 	      info->head.type)) {
5909 		DRM_ERROR("GFX Subblock %s, hardware do not support type 0x%x\n",
5910 			ras_gfx_subblocks[info->head.sub_block_index].name,
5911 			info->head.type);
5912 		return -EPERM;
5913 	}
5914 
5915 	if (!(ras_gfx_subblocks[info->head.sub_block_index].sw_supported_error_type &
5916 	      info->head.type)) {
5917 		DRM_ERROR("GFX Subblock %s, driver do not support type 0x%x\n",
5918 			ras_gfx_subblocks[info->head.sub_block_index].name,
5919 			info->head.type);
5920 		return -EPERM;
5921 	}
5922 
5923 	block_info.block_id = amdgpu_ras_block_to_ta(info->head.block);
5924 	block_info.sub_block_index =
5925 		ras_gfx_subblocks[info->head.sub_block_index].ta_subblock;
5926 	block_info.inject_error_type = amdgpu_ras_error_to_ta(info->head.type);
5927 	block_info.address = info->address;
5928 	block_info.value = info->value;
5929 
5930 	mutex_lock(&adev->grbm_idx_mutex);
5931 	ret = psp_ras_trigger_error(&adev->psp, &block_info);
5932 	mutex_unlock(&adev->grbm_idx_mutex);
5933 
5934 	return ret;
5935 }
5936 
5937 static const char *vml2_mems[] = {
5938 	"UTC_VML2_BANK_CACHE_0_BIGK_MEM0",
5939 	"UTC_VML2_BANK_CACHE_0_BIGK_MEM1",
5940 	"UTC_VML2_BANK_CACHE_0_4K_MEM0",
5941 	"UTC_VML2_BANK_CACHE_0_4K_MEM1",
5942 	"UTC_VML2_BANK_CACHE_1_BIGK_MEM0",
5943 	"UTC_VML2_BANK_CACHE_1_BIGK_MEM1",
5944 	"UTC_VML2_BANK_CACHE_1_4K_MEM0",
5945 	"UTC_VML2_BANK_CACHE_1_4K_MEM1",
5946 	"UTC_VML2_BANK_CACHE_2_BIGK_MEM0",
5947 	"UTC_VML2_BANK_CACHE_2_BIGK_MEM1",
5948 	"UTC_VML2_BANK_CACHE_2_4K_MEM0",
5949 	"UTC_VML2_BANK_CACHE_2_4K_MEM1",
5950 	"UTC_VML2_BANK_CACHE_3_BIGK_MEM0",
5951 	"UTC_VML2_BANK_CACHE_3_BIGK_MEM1",
5952 	"UTC_VML2_BANK_CACHE_3_4K_MEM0",
5953 	"UTC_VML2_BANK_CACHE_3_4K_MEM1",
5954 };
5955 
5956 static const char *vml2_walker_mems[] = {
5957 	"UTC_VML2_CACHE_PDE0_MEM0",
5958 	"UTC_VML2_CACHE_PDE0_MEM1",
5959 	"UTC_VML2_CACHE_PDE1_MEM0",
5960 	"UTC_VML2_CACHE_PDE1_MEM1",
5961 	"UTC_VML2_CACHE_PDE2_MEM0",
5962 	"UTC_VML2_CACHE_PDE2_MEM1",
5963 	"UTC_VML2_RDIF_LOG_FIFO",
5964 };
5965 
5966 static const char *atc_l2_cache_2m_mems[] = {
5967 	"UTC_ATCL2_CACHE_2M_BANK0_WAY0_MEM",
5968 	"UTC_ATCL2_CACHE_2M_BANK0_WAY1_MEM",
5969 	"UTC_ATCL2_CACHE_2M_BANK1_WAY0_MEM",
5970 	"UTC_ATCL2_CACHE_2M_BANK1_WAY1_MEM",
5971 };
5972 
5973 static const char *atc_l2_cache_4k_mems[] = {
5974 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM0",
5975 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM1",
5976 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM2",
5977 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM3",
5978 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM4",
5979 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM5",
5980 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM6",
5981 	"UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM7",
5982 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM0",
5983 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM1",
5984 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM2",
5985 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM3",
5986 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM4",
5987 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM5",
5988 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM6",
5989 	"UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM7",
5990 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM0",
5991 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM1",
5992 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM2",
5993 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM3",
5994 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM4",
5995 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM5",
5996 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM6",
5997 	"UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM7",
5998 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM0",
5999 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM1",
6000 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM2",
6001 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM3",
6002 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM4",
6003 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM5",
6004 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM6",
6005 	"UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM7",
6006 };
6007 
6008 static int gfx_v9_0_query_utc_edc_status(struct amdgpu_device *adev,
6009 					 struct ras_err_data *err_data)
6010 {
6011 	uint32_t i, data;
6012 	uint32_t sec_count, ded_count;
6013 
6014 	WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6015 	WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0);
6016 	WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6017 	WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0);
6018 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6019 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0);
6020 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6021 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0);
6022 
6023 	for (i = 0; i < 16; i++) {
6024 		WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i);
6025 		data = RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT);
6026 
6027 		sec_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, SEC_COUNT);
6028 		if (sec_count) {
6029 			DRM_INFO("Instance[%d]: SubBlock %s, SEC %d\n", i,
6030 				 vml2_mems[i], sec_count);
6031 			err_data->ce_count += sec_count;
6032 		}
6033 
6034 		ded_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, DED_COUNT);
6035 		if (ded_count) {
6036 			DRM_INFO("Instance[%d]: SubBlock %s, DED %d\n", i,
6037 				 vml2_mems[i], ded_count);
6038 			err_data->ue_count += ded_count;
6039 		}
6040 	}
6041 
6042 	for (i = 0; i < 7; i++) {
6043 		WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i);
6044 		data = RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT);
6045 
6046 		sec_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6047 						SEC_COUNT);
6048 		if (sec_count) {
6049 			DRM_INFO("Instance[%d]: SubBlock %s, SEC %d\n", i,
6050 				 vml2_walker_mems[i], sec_count);
6051 			err_data->ce_count += sec_count;
6052 		}
6053 
6054 		ded_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6055 						DED_COUNT);
6056 		if (ded_count) {
6057 			DRM_INFO("Instance[%d]: SubBlock %s, DED %d\n", i,
6058 				 vml2_walker_mems[i], ded_count);
6059 			err_data->ue_count += ded_count;
6060 		}
6061 	}
6062 
6063 	for (i = 0; i < 4; i++) {
6064 		WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i);
6065 		data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT);
6066 
6067 		sec_count = (data & 0x00006000L) >> 0xd;
6068 		if (sec_count) {
6069 			DRM_INFO("Instance[%d]: SubBlock %s, SEC %d\n", i,
6070 				 atc_l2_cache_2m_mems[i], sec_count);
6071 			err_data->ce_count += sec_count;
6072 		}
6073 	}
6074 
6075 	for (i = 0; i < 32; i++) {
6076 		WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i);
6077 		data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT);
6078 
6079 		sec_count = (data & 0x00006000L) >> 0xd;
6080 		if (sec_count) {
6081 			DRM_INFO("Instance[%d]: SubBlock %s, SEC %d\n", i,
6082 				 atc_l2_cache_4k_mems[i], sec_count);
6083 			err_data->ce_count += sec_count;
6084 		}
6085 
6086 		ded_count = (data & 0x00018000L) >> 0xf;
6087 		if (ded_count) {
6088 			DRM_INFO("Instance[%d]: SubBlock %s, DED %d\n", i,
6089 				 atc_l2_cache_4k_mems[i], ded_count);
6090 			err_data->ue_count += ded_count;
6091 		}
6092 	}
6093 
6094 	WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6095 	WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6096 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6097 	WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6098 
6099 	return 0;
6100 }
6101 
6102 static int __get_ras_error_count(const struct soc15_reg_entry *reg,
6103 	uint32_t se_id, uint32_t inst_id, uint32_t value,
6104 	uint32_t *sec_count, uint32_t *ded_count)
6105 {
6106 	uint32_t i;
6107 	uint32_t sec_cnt, ded_cnt;
6108 
6109 	for (i = 0; i < ARRAY_SIZE(ras_subblock_regs); i++) {
6110 		if(ras_subblock_regs[i].reg_offset != reg->reg_offset ||
6111 			ras_subblock_regs[i].seg != reg->seg ||
6112 			ras_subblock_regs[i].inst != reg->inst)
6113 			continue;
6114 
6115 		sec_cnt = (value &
6116 				ras_subblock_regs[i].sec_count_mask) >>
6117 				ras_subblock_regs[i].sec_count_shift;
6118 		if (sec_cnt) {
6119 			DRM_INFO("GFX SubBlock %s, Instance[%d][%d], SEC %d\n",
6120 				ras_subblock_regs[i].name,
6121 				se_id, inst_id,
6122 				sec_cnt);
6123 			*sec_count += sec_cnt;
6124 		}
6125 
6126 		ded_cnt = (value &
6127 				ras_subblock_regs[i].ded_count_mask) >>
6128 				ras_subblock_regs[i].ded_count_shift;
6129 		if (ded_cnt) {
6130 			DRM_INFO("GFX SubBlock %s, Instance[%d][%d], DED %d\n",
6131 				ras_subblock_regs[i].name,
6132 				se_id, inst_id,
6133 				ded_cnt);
6134 			*ded_count += ded_cnt;
6135 		}
6136 	}
6137 
6138 	return 0;
6139 }
6140 
6141 static int gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
6142 					  void *ras_error_status)
6143 {
6144 	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
6145 	uint32_t sec_count = 0, ded_count = 0;
6146 	uint32_t i, j, k;
6147 	uint32_t reg_value;
6148 
6149 	if (adev->asic_type != CHIP_VEGA20)
6150 		return -EINVAL;
6151 
6152 	err_data->ue_count = 0;
6153 	err_data->ce_count = 0;
6154 
6155 	mutex_lock(&adev->grbm_idx_mutex);
6156 
6157 	for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++) {
6158 		for (j = 0; j < sec_ded_counter_registers[i].se_num; j++) {
6159 			for (k = 0; k < sec_ded_counter_registers[i].instance; k++) {
6160 				gfx_v9_0_select_se_sh(adev, j, 0, k);
6161 				reg_value =
6162 					RREG32(SOC15_REG_ENTRY_OFFSET(sec_ded_counter_registers[i]));
6163 				if (reg_value)
6164 					__get_ras_error_count(&sec_ded_counter_registers[i],
6165 							j, k, reg_value,
6166 							&sec_count, &ded_count);
6167 			}
6168 		}
6169 	}
6170 
6171 	err_data->ce_count += sec_count;
6172 	err_data->ue_count += ded_count;
6173 
6174 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
6175 	mutex_unlock(&adev->grbm_idx_mutex);
6176 
6177 	gfx_v9_0_query_utc_edc_status(adev, err_data);
6178 
6179 	return 0;
6180 }
6181 
6182 static const struct amd_ip_funcs gfx_v9_0_ip_funcs = {
6183 	.name = "gfx_v9_0",
6184 	.early_init = gfx_v9_0_early_init,
6185 	.late_init = gfx_v9_0_late_init,
6186 	.sw_init = gfx_v9_0_sw_init,
6187 	.sw_fini = gfx_v9_0_sw_fini,
6188 	.hw_init = gfx_v9_0_hw_init,
6189 	.hw_fini = gfx_v9_0_hw_fini,
6190 	.suspend = gfx_v9_0_suspend,
6191 	.resume = gfx_v9_0_resume,
6192 	.is_idle = gfx_v9_0_is_idle,
6193 	.wait_for_idle = gfx_v9_0_wait_for_idle,
6194 	.soft_reset = gfx_v9_0_soft_reset,
6195 	.set_clockgating_state = gfx_v9_0_set_clockgating_state,
6196 	.set_powergating_state = gfx_v9_0_set_powergating_state,
6197 	.get_clockgating_state = gfx_v9_0_get_clockgating_state,
6198 };
6199 
6200 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
6201 	.type = AMDGPU_RING_TYPE_GFX,
6202 	.align_mask = 0xff,
6203 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6204 	.support_64bit_ptrs = true,
6205 	.vmhub = AMDGPU_GFXHUB_0,
6206 	.get_rptr = gfx_v9_0_ring_get_rptr_gfx,
6207 	.get_wptr = gfx_v9_0_ring_get_wptr_gfx,
6208 	.set_wptr = gfx_v9_0_ring_set_wptr_gfx,
6209 	.emit_frame_size = /* totally 242 maximum if 16 IBs */
6210 		5 +  /* COND_EXEC */
6211 		7 +  /* PIPELINE_SYNC */
6212 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6213 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6214 		2 + /* VM_FLUSH */
6215 		8 +  /* FENCE for VM_FLUSH */
6216 		20 + /* GDS switch */
6217 		4 + /* double SWITCH_BUFFER,
6218 		       the first COND_EXEC jump to the place just
6219 			   prior to this double SWITCH_BUFFER  */
6220 		5 + /* COND_EXEC */
6221 		7 +	 /*	HDP_flush */
6222 		4 +	 /*	VGT_flush */
6223 		14 + /*	CE_META */
6224 		31 + /*	DE_META */
6225 		3 + /* CNTX_CTRL */
6226 		5 + /* HDP_INVL */
6227 		8 + 8 + /* FENCE x2 */
6228 		2, /* SWITCH_BUFFER */
6229 	.emit_ib_size =	4, /* gfx_v9_0_ring_emit_ib_gfx */
6230 	.emit_ib = gfx_v9_0_ring_emit_ib_gfx,
6231 	.emit_fence = gfx_v9_0_ring_emit_fence,
6232 	.emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
6233 	.emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
6234 	.emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
6235 	.emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
6236 	.test_ring = gfx_v9_0_ring_test_ring,
6237 	.test_ib = gfx_v9_0_ring_test_ib,
6238 	.insert_nop = amdgpu_ring_insert_nop,
6239 	.pad_ib = amdgpu_ring_generic_pad_ib,
6240 	.emit_switch_buffer = gfx_v9_ring_emit_sb,
6241 	.emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
6242 	.init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
6243 	.patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec,
6244 	.emit_tmz = gfx_v9_0_ring_emit_tmz,
6245 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
6246 	.emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6247 	.emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6248 	.soft_recovery = gfx_v9_0_ring_soft_recovery,
6249 };
6250 
6251 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
6252 	.type = AMDGPU_RING_TYPE_COMPUTE,
6253 	.align_mask = 0xff,
6254 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6255 	.support_64bit_ptrs = true,
6256 	.vmhub = AMDGPU_GFXHUB_0,
6257 	.get_rptr = gfx_v9_0_ring_get_rptr_compute,
6258 	.get_wptr = gfx_v9_0_ring_get_wptr_compute,
6259 	.set_wptr = gfx_v9_0_ring_set_wptr_compute,
6260 	.emit_frame_size =
6261 		20 + /* gfx_v9_0_ring_emit_gds_switch */
6262 		7 + /* gfx_v9_0_ring_emit_hdp_flush */
6263 		5 + /* hdp invalidate */
6264 		7 + /* gfx_v9_0_ring_emit_pipeline_sync */
6265 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6266 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6267 		2 + /* gfx_v9_0_ring_emit_vm_flush */
6268 		8 + 8 + 8, /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
6269 	.emit_ib_size =	7, /* gfx_v9_0_ring_emit_ib_compute */
6270 	.emit_ib = gfx_v9_0_ring_emit_ib_compute,
6271 	.emit_fence = gfx_v9_0_ring_emit_fence,
6272 	.emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
6273 	.emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
6274 	.emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
6275 	.emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
6276 	.test_ring = gfx_v9_0_ring_test_ring,
6277 	.test_ib = gfx_v9_0_ring_test_ib,
6278 	.insert_nop = amdgpu_ring_insert_nop,
6279 	.pad_ib = amdgpu_ring_generic_pad_ib,
6280 	.set_priority = gfx_v9_0_ring_set_priority_compute,
6281 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
6282 	.emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6283 	.emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6284 };
6285 
6286 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
6287 	.type = AMDGPU_RING_TYPE_KIQ,
6288 	.align_mask = 0xff,
6289 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6290 	.support_64bit_ptrs = true,
6291 	.vmhub = AMDGPU_GFXHUB_0,
6292 	.get_rptr = gfx_v9_0_ring_get_rptr_compute,
6293 	.get_wptr = gfx_v9_0_ring_get_wptr_compute,
6294 	.set_wptr = gfx_v9_0_ring_set_wptr_compute,
6295 	.emit_frame_size =
6296 		20 + /* gfx_v9_0_ring_emit_gds_switch */
6297 		7 + /* gfx_v9_0_ring_emit_hdp_flush */
6298 		5 + /* hdp invalidate */
6299 		7 + /* gfx_v9_0_ring_emit_pipeline_sync */
6300 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6301 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6302 		2 + /* gfx_v9_0_ring_emit_vm_flush */
6303 		8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */
6304 	.emit_ib_size =	7, /* gfx_v9_0_ring_emit_ib_compute */
6305 	.emit_fence = gfx_v9_0_ring_emit_fence_kiq,
6306 	.test_ring = gfx_v9_0_ring_test_ring,
6307 	.insert_nop = amdgpu_ring_insert_nop,
6308 	.pad_ib = amdgpu_ring_generic_pad_ib,
6309 	.emit_rreg = gfx_v9_0_ring_emit_rreg,
6310 	.emit_wreg = gfx_v9_0_ring_emit_wreg,
6311 	.emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6312 	.emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6313 };
6314 
6315 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev)
6316 {
6317 	int i;
6318 
6319 	adev->gfx.kiq.ring.funcs = &gfx_v9_0_ring_funcs_kiq;
6320 
6321 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
6322 		adev->gfx.gfx_ring[i].funcs = &gfx_v9_0_ring_funcs_gfx;
6323 
6324 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
6325 		adev->gfx.compute_ring[i].funcs = &gfx_v9_0_ring_funcs_compute;
6326 }
6327 
6328 static const struct amdgpu_irq_src_funcs gfx_v9_0_eop_irq_funcs = {
6329 	.set = gfx_v9_0_set_eop_interrupt_state,
6330 	.process = gfx_v9_0_eop_irq,
6331 };
6332 
6333 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_reg_irq_funcs = {
6334 	.set = gfx_v9_0_set_priv_reg_fault_state,
6335 	.process = gfx_v9_0_priv_reg_irq,
6336 };
6337 
6338 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_inst_irq_funcs = {
6339 	.set = gfx_v9_0_set_priv_inst_fault_state,
6340 	.process = gfx_v9_0_priv_inst_irq,
6341 };
6342 
6343 static const struct amdgpu_irq_src_funcs gfx_v9_0_cp_ecc_error_irq_funcs = {
6344 	.set = gfx_v9_0_set_cp_ecc_error_state,
6345 	.process = amdgpu_gfx_cp_ecc_error_irq,
6346 };
6347 
6348 
6349 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev)
6350 {
6351 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
6352 	adev->gfx.eop_irq.funcs = &gfx_v9_0_eop_irq_funcs;
6353 
6354 	adev->gfx.priv_reg_irq.num_types = 1;
6355 	adev->gfx.priv_reg_irq.funcs = &gfx_v9_0_priv_reg_irq_funcs;
6356 
6357 	adev->gfx.priv_inst_irq.num_types = 1;
6358 	adev->gfx.priv_inst_irq.funcs = &gfx_v9_0_priv_inst_irq_funcs;
6359 
6360 	adev->gfx.cp_ecc_error_irq.num_types = 2; /*C5 ECC error and C9 FUE error*/
6361 	adev->gfx.cp_ecc_error_irq.funcs = &gfx_v9_0_cp_ecc_error_irq_funcs;
6362 }
6363 
6364 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev)
6365 {
6366 	switch (adev->asic_type) {
6367 	case CHIP_VEGA10:
6368 	case CHIP_VEGA12:
6369 	case CHIP_VEGA20:
6370 	case CHIP_RAVEN:
6371 	case CHIP_ARCTURUS:
6372 	case CHIP_RENOIR:
6373 		adev->gfx.rlc.funcs = &gfx_v9_0_rlc_funcs;
6374 		break;
6375 	default:
6376 		break;
6377 	}
6378 }
6379 
6380 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev)
6381 {
6382 	/* init asci gds info */
6383 	switch (adev->asic_type) {
6384 	case CHIP_VEGA10:
6385 	case CHIP_VEGA12:
6386 	case CHIP_VEGA20:
6387 		adev->gds.gds_size = 0x10000;
6388 		break;
6389 	case CHIP_RAVEN:
6390 	case CHIP_ARCTURUS:
6391 		adev->gds.gds_size = 0x1000;
6392 		break;
6393 	default:
6394 		adev->gds.gds_size = 0x10000;
6395 		break;
6396 	}
6397 
6398 	switch (adev->asic_type) {
6399 	case CHIP_VEGA10:
6400 	case CHIP_VEGA20:
6401 		adev->gds.gds_compute_max_wave_id = 0x7ff;
6402 		break;
6403 	case CHIP_VEGA12:
6404 		adev->gds.gds_compute_max_wave_id = 0x27f;
6405 		break;
6406 	case CHIP_RAVEN:
6407 		if (adev->rev_id >= 0x8)
6408 			adev->gds.gds_compute_max_wave_id = 0x77; /* raven2 */
6409 		else
6410 			adev->gds.gds_compute_max_wave_id = 0x15f; /* raven1 */
6411 		break;
6412 	case CHIP_ARCTURUS:
6413 		adev->gds.gds_compute_max_wave_id = 0xfff;
6414 		break;
6415 	default:
6416 		/* this really depends on the chip */
6417 		adev->gds.gds_compute_max_wave_id = 0x7ff;
6418 		break;
6419 	}
6420 
6421 	adev->gds.gws_size = 64;
6422 	adev->gds.oa_size = 16;
6423 }
6424 
6425 static void gfx_v9_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
6426 						 u32 bitmap)
6427 {
6428 	u32 data;
6429 
6430 	if (!bitmap)
6431 		return;
6432 
6433 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
6434 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
6435 
6436 	WREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG, data);
6437 }
6438 
6439 static u32 gfx_v9_0_get_cu_active_bitmap(struct amdgpu_device *adev)
6440 {
6441 	u32 data, mask;
6442 
6443 	data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
6444 	data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
6445 
6446 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
6447 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
6448 
6449 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
6450 
6451 	return (~data) & mask;
6452 }
6453 
6454 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
6455 				 struct amdgpu_cu_info *cu_info)
6456 {
6457 	int i, j, k, counter, active_cu_number = 0;
6458 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
6459 	unsigned disable_masks[4 * 4];
6460 
6461 	if (!adev || !cu_info)
6462 		return -EINVAL;
6463 
6464 	/*
6465 	 * 16 comes from bitmap array size 4*4, and it can cover all gfx9 ASICs
6466 	 */
6467 	if (adev->gfx.config.max_shader_engines *
6468 		adev->gfx.config.max_sh_per_se > 16)
6469 		return -EINVAL;
6470 
6471 	amdgpu_gfx_parse_disable_cu(disable_masks,
6472 				    adev->gfx.config.max_shader_engines,
6473 				    adev->gfx.config.max_sh_per_se);
6474 
6475 	mutex_lock(&adev->grbm_idx_mutex);
6476 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
6477 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
6478 			mask = 1;
6479 			ao_bitmap = 0;
6480 			counter = 0;
6481 			gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
6482 			gfx_v9_0_set_user_cu_inactive_bitmap(
6483 				adev, disable_masks[i * adev->gfx.config.max_sh_per_se + j]);
6484 			bitmap = gfx_v9_0_get_cu_active_bitmap(adev);
6485 
6486 			/*
6487 			 * The bitmap(and ao_cu_bitmap) in cu_info structure is
6488 			 * 4x4 size array, and it's usually suitable for Vega
6489 			 * ASICs which has 4*2 SE/SH layout.
6490 			 * But for Arcturus, SE/SH layout is changed to 8*1.
6491 			 * To mostly reduce the impact, we make it compatible
6492 			 * with current bitmap array as below:
6493 			 *    SE4,SH0 --> bitmap[0][1]
6494 			 *    SE5,SH0 --> bitmap[1][1]
6495 			 *    SE6,SH0 --> bitmap[2][1]
6496 			 *    SE7,SH0 --> bitmap[3][1]
6497 			 */
6498 			cu_info->bitmap[i % 4][j + i / 4] = bitmap;
6499 
6500 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
6501 				if (bitmap & mask) {
6502 					if (counter < adev->gfx.config.max_cu_per_sh)
6503 						ao_bitmap |= mask;
6504 					counter ++;
6505 				}
6506 				mask <<= 1;
6507 			}
6508 			active_cu_number += counter;
6509 			if (i < 2 && j < 2)
6510 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
6511 			cu_info->ao_cu_bitmap[i % 4][j + i / 4] = ao_bitmap;
6512 		}
6513 	}
6514 	gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
6515 	mutex_unlock(&adev->grbm_idx_mutex);
6516 
6517 	cu_info->number = active_cu_number;
6518 	cu_info->ao_cu_mask = ao_cu_mask;
6519 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
6520 
6521 	return 0;
6522 }
6523 
6524 const struct amdgpu_ip_block_version gfx_v9_0_ip_block =
6525 {
6526 	.type = AMD_IP_BLOCK_TYPE_GFX,
6527 	.major = 9,
6528 	.minor = 0,
6529 	.rev = 0,
6530 	.funcs = &gfx_v9_0_ip_funcs,
6531 };
6532