]> git.ipfire.org Git - thirdparty/linux.git/blob - drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
projects / thirdparty / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'io_uring-5.7-2020-05-22' of git://git.kernel.dk/linux-block
[thirdparty/linux.git] / drivers / gpu / drm / amd / amdgpu / gfx_v9_0.c
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 #include "gfx_v9_4.h"
52
53 #define GFX9_NUM_GFX_RINGS 1
54 #define GFX9_MEC_HPD_SIZE 4096
55 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L
56 #define RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET 0x00000000L
57
58 #define mmPWR_MISC_CNTL_STATUS 0x0183
59 #define mmPWR_MISC_CNTL_STATUS_BASE_IDX 0
60 #define PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN__SHIFT 0x0
61 #define PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT 0x1
62 #define PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK 0x00000001L
63 #define PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK 0x00000006L
64
65 #define mmGCEA_PROBE_MAP 0x070c
66 #define mmGCEA_PROBE_MAP_BASE_IDX 0
67
68 MODULE_FIRMWARE("amdgpu/vega10_ce.bin");
69 MODULE_FIRMWARE("amdgpu/vega10_pfp.bin");
70 MODULE_FIRMWARE("amdgpu/vega10_me.bin");
71 MODULE_FIRMWARE("amdgpu/vega10_mec.bin");
72 MODULE_FIRMWARE("amdgpu/vega10_mec2.bin");
73 MODULE_FIRMWARE("amdgpu/vega10_rlc.bin");
74
75 MODULE_FIRMWARE("amdgpu/vega12_ce.bin");
76 MODULE_FIRMWARE("amdgpu/vega12_pfp.bin");
77 MODULE_FIRMWARE("amdgpu/vega12_me.bin");
78 MODULE_FIRMWARE("amdgpu/vega12_mec.bin");
79 MODULE_FIRMWARE("amdgpu/vega12_mec2.bin");
80 MODULE_FIRMWARE("amdgpu/vega12_rlc.bin");
81
82 MODULE_FIRMWARE("amdgpu/vega20_ce.bin");
83 MODULE_FIRMWARE("amdgpu/vega20_pfp.bin");
84 MODULE_FIRMWARE("amdgpu/vega20_me.bin");
85 MODULE_FIRMWARE("amdgpu/vega20_mec.bin");
86 MODULE_FIRMWARE("amdgpu/vega20_mec2.bin");
87 MODULE_FIRMWARE("amdgpu/vega20_rlc.bin");
88
89 MODULE_FIRMWARE("amdgpu/raven_ce.bin");
90 MODULE_FIRMWARE("amdgpu/raven_pfp.bin");
91 MODULE_FIRMWARE("amdgpu/raven_me.bin");
92 MODULE_FIRMWARE("amdgpu/raven_mec.bin");
93 MODULE_FIRMWARE("amdgpu/raven_mec2.bin");
94 MODULE_FIRMWARE("amdgpu/raven_rlc.bin");
95
96 MODULE_FIRMWARE("amdgpu/picasso_ce.bin");
97 MODULE_FIRMWARE("amdgpu/picasso_pfp.bin");
98 MODULE_FIRMWARE("amdgpu/picasso_me.bin");
99 MODULE_FIRMWARE("amdgpu/picasso_mec.bin");
100 MODULE_FIRMWARE("amdgpu/picasso_mec2.bin");
101 MODULE_FIRMWARE("amdgpu/picasso_rlc.bin");
102 MODULE_FIRMWARE("amdgpu/picasso_rlc_am4.bin");
103
104 MODULE_FIRMWARE("amdgpu/raven2_ce.bin");
105 MODULE_FIRMWARE("amdgpu/raven2_pfp.bin");
106 MODULE_FIRMWARE("amdgpu/raven2_me.bin");
107 MODULE_FIRMWARE("amdgpu/raven2_mec.bin");
108 MODULE_FIRMWARE("amdgpu/raven2_mec2.bin");
109 MODULE_FIRMWARE("amdgpu/raven2_rlc.bin");
110 MODULE_FIRMWARE("amdgpu/raven_kicker_rlc.bin");
111
112 MODULE_FIRMWARE("amdgpu/arcturus_mec.bin");
113 MODULE_FIRMWARE("amdgpu/arcturus_mec2.bin");
114 MODULE_FIRMWARE("amdgpu/arcturus_rlc.bin");
115
116 MODULE_FIRMWARE("amdgpu/renoir_ce.bin");
117 MODULE_FIRMWARE("amdgpu/renoir_pfp.bin");
118 MODULE_FIRMWARE("amdgpu/renoir_me.bin");
119 MODULE_FIRMWARE("amdgpu/renoir_mec.bin");
120 MODULE_FIRMWARE("amdgpu/renoir_mec2.bin");
121 MODULE_FIRMWARE("amdgpu/renoir_rlc.bin");
122
123 #define mmTCP_CHAN_STEER_0_ARCT 0x0b03
124 #define mmTCP_CHAN_STEER_0_ARCT_BASE_IDX 0
125 #define mmTCP_CHAN_STEER_1_ARCT 0x0b04
126 #define mmTCP_CHAN_STEER_1_ARCT_BASE_IDX 0
127 #define mmTCP_CHAN_STEER_2_ARCT 0x0b09
128 #define mmTCP_CHAN_STEER_2_ARCT_BASE_IDX 0
129 #define mmTCP_CHAN_STEER_3_ARCT 0x0b0a
130 #define mmTCP_CHAN_STEER_3_ARCT_BASE_IDX 0
131 #define mmTCP_CHAN_STEER_4_ARCT 0x0b0b
132 #define mmTCP_CHAN_STEER_4_ARCT_BASE_IDX 0
133 #define mmTCP_CHAN_STEER_5_ARCT 0x0b0c
134 #define mmTCP_CHAN_STEER_5_ARCT_BASE_IDX 0
135
136 enum ta_ras_gfx_subblock {
137 /*CPC*/
138 TA_RAS_BLOCK__GFX_CPC_INDEX_START = 0,
139 TA_RAS_BLOCK__GFX_CPC_SCRATCH = TA_RAS_BLOCK__GFX_CPC_INDEX_START,
140 TA_RAS_BLOCK__GFX_CPC_UCODE,
141 TA_RAS_BLOCK__GFX_DC_STATE_ME1,
142 TA_RAS_BLOCK__GFX_DC_CSINVOC_ME1,
143 TA_RAS_BLOCK__GFX_DC_RESTORE_ME1,
144 TA_RAS_BLOCK__GFX_DC_STATE_ME2,
145 TA_RAS_BLOCK__GFX_DC_CSINVOC_ME2,
146 TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
147 TA_RAS_BLOCK__GFX_CPC_INDEX_END = TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
148 /* CPF*/
149 TA_RAS_BLOCK__GFX_CPF_INDEX_START,
150 TA_RAS_BLOCK__GFX_CPF_ROQ_ME2 = TA_RAS_BLOCK__GFX_CPF_INDEX_START,
151 TA_RAS_BLOCK__GFX_CPF_ROQ_ME1,
152 TA_RAS_BLOCK__GFX_CPF_TAG,
153 TA_RAS_BLOCK__GFX_CPF_INDEX_END = TA_RAS_BLOCK__GFX_CPF_TAG,
154 /* CPG*/
155 TA_RAS_BLOCK__GFX_CPG_INDEX_START,
156 TA_RAS_BLOCK__GFX_CPG_DMA_ROQ = TA_RAS_BLOCK__GFX_CPG_INDEX_START,
157 TA_RAS_BLOCK__GFX_CPG_DMA_TAG,
158 TA_RAS_BLOCK__GFX_CPG_TAG,
159 TA_RAS_BLOCK__GFX_CPG_INDEX_END = TA_RAS_BLOCK__GFX_CPG_TAG,
160 /* GDS*/
161 TA_RAS_BLOCK__GFX_GDS_INDEX_START,
162 TA_RAS_BLOCK__GFX_GDS_MEM = TA_RAS_BLOCK__GFX_GDS_INDEX_START,
163 TA_RAS_BLOCK__GFX_GDS_INPUT_QUEUE,
164 TA_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM,
165 TA_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM,
166 TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
167 TA_RAS_BLOCK__GFX_GDS_INDEX_END = TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
168 /* SPI*/
169 TA_RAS_BLOCK__GFX_SPI_SR_MEM,
170 /* SQ*/
171 TA_RAS_BLOCK__GFX_SQ_INDEX_START,
172 TA_RAS_BLOCK__GFX_SQ_SGPR = TA_RAS_BLOCK__GFX_SQ_INDEX_START,
173 TA_RAS_BLOCK__GFX_SQ_LDS_D,
174 TA_RAS_BLOCK__GFX_SQ_LDS_I,
175 TA_RAS_BLOCK__GFX_SQ_VGPR, /* VGPR = SP*/
176 TA_RAS_BLOCK__GFX_SQ_INDEX_END = TA_RAS_BLOCK__GFX_SQ_VGPR,
177 /* SQC (3 ranges)*/
178 TA_RAS_BLOCK__GFX_SQC_INDEX_START,
179 /* SQC range 0*/
180 TA_RAS_BLOCK__GFX_SQC_INDEX0_START = TA_RAS_BLOCK__GFX_SQC_INDEX_START,
181 TA_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO =
182 TA_RAS_BLOCK__GFX_SQC_INDEX0_START,
183 TA_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF,
184 TA_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO,
185 TA_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF,
186 TA_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO,
187 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF,
188 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
189 TA_RAS_BLOCK__GFX_SQC_INDEX0_END =
190 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
191 /* SQC range 1*/
192 TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
193 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM =
194 TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
195 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO,
196 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO,
197 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM,
198 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM,
199 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO,
200 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO,
201 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM,
202 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
203 TA_RAS_BLOCK__GFX_SQC_INDEX1_END =
204 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
205 /* SQC range 2*/
206 TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
207 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM =
208 TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
209 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO,
210 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO,
211 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM,
212 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM,
213 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO,
214 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO,
215 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM,
216 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
217 TA_RAS_BLOCK__GFX_SQC_INDEX2_END =
218 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
219 TA_RAS_BLOCK__GFX_SQC_INDEX_END = TA_RAS_BLOCK__GFX_SQC_INDEX2_END,
220 /* TA*/
221 TA_RAS_BLOCK__GFX_TA_INDEX_START,
222 TA_RAS_BLOCK__GFX_TA_FS_DFIFO = TA_RAS_BLOCK__GFX_TA_INDEX_START,
223 TA_RAS_BLOCK__GFX_TA_FS_AFIFO,
224 TA_RAS_BLOCK__GFX_TA_FL_LFIFO,
225 TA_RAS_BLOCK__GFX_TA_FX_LFIFO,
226 TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
227 TA_RAS_BLOCK__GFX_TA_INDEX_END = TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
228 /* TCA*/
229 TA_RAS_BLOCK__GFX_TCA_INDEX_START,
230 TA_RAS_BLOCK__GFX_TCA_HOLE_FIFO = TA_RAS_BLOCK__GFX_TCA_INDEX_START,
231 TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
232 TA_RAS_BLOCK__GFX_TCA_INDEX_END = TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
233 /* TCC (5 sub-ranges)*/
234 TA_RAS_BLOCK__GFX_TCC_INDEX_START,
235 /* TCC range 0*/
236 TA_RAS_BLOCK__GFX_TCC_INDEX0_START = TA_RAS_BLOCK__GFX_TCC_INDEX_START,
237 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX0_START,
238 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1,
239 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0,
240 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1,
241 TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0,
242 TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1,
243 TA_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG,
244 TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
245 TA_RAS_BLOCK__GFX_TCC_INDEX0_END = TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
246 /* TCC range 1*/
247 TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
248 TA_RAS_BLOCK__GFX_TCC_IN_USE_DEC = TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
249 TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
250 TA_RAS_BLOCK__GFX_TCC_INDEX1_END =
251 TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
252 /* TCC range 2*/
253 TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
254 TA_RAS_BLOCK__GFX_TCC_RETURN_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
255 TA_RAS_BLOCK__GFX_TCC_RETURN_CONTROL,
256 TA_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO,
257 TA_RAS_BLOCK__GFX_TCC_WRITE_RETURN,
258 TA_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ,
259 TA_RAS_BLOCK__GFX_TCC_SRC_FIFO,
260 TA_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM,
261 TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
262 TA_RAS_BLOCK__GFX_TCC_INDEX2_END =
263 TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
264 /* TCC range 3*/
265 TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
266 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO = TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
267 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
268 TA_RAS_BLOCK__GFX_TCC_INDEX3_END =
269 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
270 /* TCC range 4*/
271 TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
272 TA_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN =
273 TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
274 TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
275 TA_RAS_BLOCK__GFX_TCC_INDEX4_END =
276 TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
277 TA_RAS_BLOCK__GFX_TCC_INDEX_END = TA_RAS_BLOCK__GFX_TCC_INDEX4_END,
278 /* TCI*/
279 TA_RAS_BLOCK__GFX_TCI_WRITE_RAM,
280 /* TCP*/
281 TA_RAS_BLOCK__GFX_TCP_INDEX_START,
282 TA_RAS_BLOCK__GFX_TCP_CACHE_RAM = TA_RAS_BLOCK__GFX_TCP_INDEX_START,
283 TA_RAS_BLOCK__GFX_TCP_LFIFO_RAM,
284 TA_RAS_BLOCK__GFX_TCP_CMD_FIFO,
285 TA_RAS_BLOCK__GFX_TCP_VM_FIFO,
286 TA_RAS_BLOCK__GFX_TCP_DB_RAM,
287 TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0,
288 TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
289 TA_RAS_BLOCK__GFX_TCP_INDEX_END = TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
290 /* TD*/
291 TA_RAS_BLOCK__GFX_TD_INDEX_START,
292 TA_RAS_BLOCK__GFX_TD_SS_FIFO_LO = TA_RAS_BLOCK__GFX_TD_INDEX_START,
293 TA_RAS_BLOCK__GFX_TD_SS_FIFO_HI,
294 TA_RAS_BLOCK__GFX_TD_CS_FIFO,
295 TA_RAS_BLOCK__GFX_TD_INDEX_END = TA_RAS_BLOCK__GFX_TD_CS_FIFO,
296 /* EA (3 sub-ranges)*/
297 TA_RAS_BLOCK__GFX_EA_INDEX_START,
298 /* EA range 0*/
299 TA_RAS_BLOCK__GFX_EA_INDEX0_START = TA_RAS_BLOCK__GFX_EA_INDEX_START,
300 TA_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM = TA_RAS_BLOCK__GFX_EA_INDEX0_START,
301 TA_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM,
302 TA_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM,
303 TA_RAS_BLOCK__GFX_EA_RRET_TAGMEM,
304 TA_RAS_BLOCK__GFX_EA_WRET_TAGMEM,
305 TA_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM,
306 TA_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM,
307 TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
308 TA_RAS_BLOCK__GFX_EA_INDEX0_END = TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
309 /* EA range 1*/
310 TA_RAS_BLOCK__GFX_EA_INDEX1_START,
311 TA_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM = TA_RAS_BLOCK__GFX_EA_INDEX1_START,
312 TA_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM,
313 TA_RAS_BLOCK__GFX_EA_IORD_CMDMEM,
314 TA_RAS_BLOCK__GFX_EA_IOWR_CMDMEM,
315 TA_RAS_BLOCK__GFX_EA_IOWR_DATAMEM,
316 TA_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM,
317 TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
318 TA_RAS_BLOCK__GFX_EA_INDEX1_END = TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
319 /* EA range 2*/
320 TA_RAS_BLOCK__GFX_EA_INDEX2_START,
321 TA_RAS_BLOCK__GFX_EA_MAM_D0MEM = TA_RAS_BLOCK__GFX_EA_INDEX2_START,
322 TA_RAS_BLOCK__GFX_EA_MAM_D1MEM,
323 TA_RAS_BLOCK__GFX_EA_MAM_D2MEM,
324 TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
325 TA_RAS_BLOCK__GFX_EA_INDEX2_END = TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
326 TA_RAS_BLOCK__GFX_EA_INDEX_END = TA_RAS_BLOCK__GFX_EA_INDEX2_END,
327 /* UTC VM L2 bank*/
328 TA_RAS_BLOCK__UTC_VML2_BANK_CACHE,
329 /* UTC VM walker*/
330 TA_RAS_BLOCK__UTC_VML2_WALKER,
331 /* UTC ATC L2 2MB cache*/
332 TA_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK,
333 /* UTC ATC L2 4KB cache*/
334 TA_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK,
335 TA_RAS_BLOCK__GFX_MAX
336 };
337
338 struct ras_gfx_subblock {
339 unsigned char *name;
340 int ta_subblock;
341 int hw_supported_error_type;
342 int sw_supported_error_type;
343 };
344
345 #define AMDGPU_RAS_SUB_BLOCK(subblock, a, b, c, d, e, f, g, h) \
346 [AMDGPU_RAS_BLOCK__##subblock] = { \
347 #subblock, \
348 TA_RAS_BLOCK__##subblock, \
349 ((a) | ((b) << 1) | ((c) << 2) | ((d) << 3)), \
350 (((e) << 1) | ((f) << 3) | (g) | ((h) << 2)), \
351 }
352
353 static const struct ras_gfx_subblock ras_gfx_subblocks[] = {
354 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_SCRATCH, 0, 1, 1, 1, 1, 0, 0, 1),
355 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_UCODE, 0, 1, 1, 1, 1, 0, 0, 1),
356 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME1, 1, 0, 0, 1, 0, 0, 1, 0),
357 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME1, 1, 0, 0, 1, 0, 0, 0, 0),
358 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME1, 1, 0, 0, 1, 0, 0, 0, 0),
359 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
360 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
361 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
362 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
363 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME1, 1, 0, 0, 1, 0, 0, 1, 0),
364 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_TAG, 0, 1, 1, 1, 1, 0, 0, 1),
365 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_ROQ, 1, 0, 0, 1, 0, 0, 1, 0),
366 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_TAG, 0, 1, 1, 1, 0, 1, 0, 1),
367 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_TAG, 0, 1, 1, 1, 1, 1, 0, 1),
368 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_MEM, 0, 1, 1, 1, 0, 0, 0, 0),
369 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_INPUT_QUEUE, 1, 0, 0, 1, 0, 0, 0, 0),
370 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_CMD_RAM_MEM, 0, 1, 1, 1, 0, 0, 0,
371 0),
372 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_DATA_RAM_MEM, 1, 0, 0, 1, 0, 0, 0,
373 0),
374 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PIPE_MEM, 0, 1, 1, 1, 0, 0, 0, 0),
375 AMDGPU_RAS_SUB_BLOCK(GFX_SPI_SR_MEM, 1, 0, 0, 1, 0, 0, 0, 0),
376 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_SGPR, 0, 1, 1, 1, 0, 0, 0, 0),
377 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_D, 0, 1, 1, 1, 1, 0, 0, 1),
378 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_I, 0, 1, 1, 1, 0, 0, 0, 0),
379 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_VGPR, 0, 1, 1, 1, 0, 0, 0, 0),
380 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 1),
381 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
382 0, 0),
383 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,
384 0),
385 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
386 0, 0),
387 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_UTCL1_LFIFO, 0, 1, 1, 1, 1, 0, 0,
388 0),
389 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
390 0, 0),
391 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,
392 0),
393 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,
394 1),
395 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,
396 0, 0, 0),
397 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
398 0),
399 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
400 0),
401 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,
402 0),
403 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,
404 0),
405 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
406 0),
407 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,
408 0, 0),
409 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
410 0),
411 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,
412 0),
413 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,
414 0, 0, 0),
415 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
416 0),
417 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
418 0),
419 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,
420 0),
421 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,
422 0),
423 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
424 0),
425 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,
426 0, 0),
427 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
428 0),
429 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_DFIFO, 0, 1, 1, 1, 1, 0, 0, 1),
430 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_AFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
431 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FL_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
432 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FX_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
433 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_CFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
434 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_HOLE_FIFO, 1, 0, 0, 1, 0, 1, 1, 0),
435 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_REQ_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
436 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA, 0, 1, 1, 1, 1, 0, 0, 1),
437 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_0_1, 0, 1, 1, 1, 1, 0, 0,
438 1),
439 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_0, 0, 1, 1, 1, 1, 0, 0,
440 1),
441 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_1, 0, 1, 1, 1, 1, 0, 0,
442 1),
443 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_0, 0, 1, 1, 1, 0, 0, 0,
444 0),
445 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_1, 0, 1, 1, 1, 0, 0, 0,
446 0),
447 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_HIGH_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0),
448 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LOW_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0),
449 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_DEC, 1, 0, 0, 1, 0, 0, 0, 0),
450 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_TRANSFER, 1, 0, 0, 1, 0, 0, 0, 0),
451 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_DATA, 1, 0, 0, 1, 0, 0, 0, 0),
452 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_CONTROL, 1, 0, 0, 1, 0, 0, 0, 0),
453 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_UC_ATOMIC_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
454 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_RETURN, 1, 0, 0, 1, 0, 1, 1, 0),
455 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_CACHE_READ, 1, 0, 0, 1, 0, 0, 0, 0),
456 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO, 0, 1, 1, 1, 0, 0, 0, 0),
457 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 1, 0),
458 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_TAG_PROBE_FIFO, 1, 0, 0, 1, 0, 0, 0,
459 0),
460 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
461 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 0,
462 0),
463 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRRET_TAG_WRITE_RETURN, 1, 0, 0, 1, 0, 0,
464 0, 0),
465 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_ATOMIC_RETURN_BUFFER, 1, 0, 0, 1, 0, 0, 0,
466 0),
467 AMDGPU_RAS_SUB_BLOCK(GFX_TCI_WRITE_RAM, 1, 0, 0, 1, 0, 0, 0, 0),
468 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CACHE_RAM, 0, 1, 1, 1, 1, 0, 0, 1),
469 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_LFIFO_RAM, 0, 1, 1, 1, 0, 0, 0, 0),
470 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CMD_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
471 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_VM_FIFO, 0, 1, 1, 1, 0, 0, 0, 0),
472 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_DB_RAM, 1, 0, 0, 1, 0, 0, 0, 0),
473 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO0, 0, 1, 1, 1, 0, 0, 0, 0),
474 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO1, 0, 1, 1, 1, 0, 0, 0, 0),
475 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_LO, 0, 1, 1, 1, 1, 0, 0, 1),
476 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_HI, 0, 1, 1, 1, 0, 0, 0, 0),
477 AMDGPU_RAS_SUB_BLOCK(GFX_TD_CS_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
478 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_CMDMEM, 0, 1, 1, 1, 1, 0, 0, 1),
479 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
480 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0),
481 AMDGPU_RAS_SUB_BLOCK(GFX_EA_RRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0),
482 AMDGPU_RAS_SUB_BLOCK(GFX_EA_WRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0),
483 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
484 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
485 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0),
486 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
487 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
488 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IORD_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0),
489 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0),
490 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_DATAMEM, 1, 0, 0, 1, 0, 0, 0, 0),
491 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
492 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
493 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D0MEM, 1, 0, 0, 1, 0, 0, 0, 0),
494 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D1MEM, 1, 0, 0, 1, 0, 0, 0, 0),
495 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D2MEM, 1, 0, 0, 1, 0, 0, 0, 0),
496 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D3MEM, 1, 0, 0, 1, 0, 0, 0, 0),
497 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_BANK_CACHE, 0, 1, 1, 1, 0, 0, 0, 0),
498 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_WALKER, 0, 1, 1, 1, 0, 0, 0, 0),
499 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_2M_BANK, 1, 0, 0, 1, 0, 0, 0, 0),
500 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_4K_BANK, 0, 1, 1, 1, 0, 0, 0, 0),
501 };
502
503 static const struct soc15_reg_golden golden_settings_gc_9_0[] =
504 {
505 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400),
506 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG3, 0x80000000, 0x80000000),
507 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
508 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
509 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
510 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
511 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
512 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
513 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
514 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87),
515 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f),
516 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
517 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
518 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
519 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
520 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
521 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff),
522 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
523 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
524 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
525 };
526
527 static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] =
528 {
529 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0x0000f000, 0x00012107),
530 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
531 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
532 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
533 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
534 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
535 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x2a114042),
536 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
537 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0x00008000, 0x00048000),
538 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
539 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
540 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
541 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
542 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
543 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
544 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x01000107),
545 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x00001800, 0x00000800),
546 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080)
547 };
548
549 static const struct soc15_reg_golden golden_settings_gc_9_0_vg20[] =
550 {
551 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x0f000080, 0x04000080),
552 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
553 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
554 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x22014042),
555 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x22014042),
556 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0x00003e00, 0x00000400),
557 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xff840000, 0x04040000),
558 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00030000),
559 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff010f, 0x01000107),
560 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0x000b0000, 0x000b0000),
561 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01000000, 0x01000000)
562 };
563
564 static const struct soc15_reg_golden golden_settings_gc_9_1[] =
565 {
566 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
567 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
568 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
569 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
570 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
571 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
572 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
573 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
574 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
575 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
576 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
577 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
578 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
579 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
580 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
581 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
582 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
583 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120),
584 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
585 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000000ff),
586 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080),
587 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
588 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
589 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
590 };
591
592 static const struct soc15_reg_golden golden_settings_gc_9_1_rv1[] =
593 {
594 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
595 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24000042),
596 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24000042),
597 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04048000),
598 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_MODE_CNTL_1, 0x06000000, 0x06000000),
599 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
600 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x00000800)
601 };
602
603 static const struct soc15_reg_golden golden_settings_gc_9_1_rv2[] =
604 {
605 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0xff7fffff, 0x04000000),
606 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
607 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000),
608 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x7f0fffff, 0x08000080),
609 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0xff8fffff, 0x08000080),
610 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x7f8fffff, 0x08000080),
611 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x26013041),
612 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x26013041),
613 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x3f8fffff, 0x08000080),
614 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
615 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0xff0fffff, 0x08000080),
616 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0xff0fffff, 0x08000080),
617 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0xff0fffff, 0x08000080),
618 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0xff0fffff, 0x08000080),
619 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0xff0fffff, 0x08000080),
620 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
621 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010),
622 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000),
623 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x3f8fffff, 0x08000080),
624 };
625
626 static const struct soc15_reg_golden golden_settings_gc_9_1_rn[] =
627 {
628 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
629 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000),
630 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400),
631 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x24000042),
632 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x24000042),
633 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
634 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
635 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
636 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
637 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
638 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120),
639 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGCEA_PROBE_MAP, 0xffffffff, 0x0000cccc),
640 };
641
642 static const struct soc15_reg_golden golden_settings_gc_9_x_common[] =
643 {
644 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0xffffffff, 0x000001ff),
645 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_INDEX, 0xffffffff, 0x00000000),
646 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_DATA, 0xffffffff, 0x2544c382)
647 };
648
649 static const struct soc15_reg_golden golden_settings_gc_9_2_1[] =
650 {
651 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
652 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
653 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
654 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
655 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
656 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
657 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
658 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
659 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87),
660 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f),
661 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
662 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
663 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
664 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
665 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
666 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff)
667 };
668
669 static const struct soc15_reg_golden golden_settings_gc_9_2_1_vg12[] =
670 {
671 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x00000080, 0x04000080),
672 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
673 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
674 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24104041),
675 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24104041),
676 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
677 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff03ff, 0x01000107),
678 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
679 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x76325410),
680 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000),
681 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
682 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
683 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
684 };
685
686 static const struct soc15_reg_golden golden_settings_gc_9_4_1_arct[] =
687 {
688 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
689 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x10b0000),
690 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_0_ARCT, 0x3fffffff, 0x346f0a4e),
691 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_1_ARCT, 0x3fffffff, 0x1c642ca),
692 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_2_ARCT, 0x3fffffff, 0x26f45098),
693 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_3_ARCT, 0x3fffffff, 0x2ebd9fe3),
694 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_4_ARCT, 0x3fffffff, 0xb90f5b1),
695 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_5_ARCT, 0x3ff, 0x135),
696 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_CONFIG, 0xffffffff, 0x011A0000),
697 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_FIFO_SIZES, 0xffffffff, 0x00000f00),
698 };
699
700 static const struct soc15_reg_rlcg rlcg_access_gc_9_0[] = {
701 {SOC15_REG_ENTRY(GC, 0, mmGRBM_GFX_INDEX)},
702 {SOC15_REG_ENTRY(GC, 0, mmSQ_IND_INDEX)},
703 };
704
705 static const u32 GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[] =
706 {
707 mmRLC_SRM_INDEX_CNTL_ADDR_0 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
708 mmRLC_SRM_INDEX_CNTL_ADDR_1 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
709 mmRLC_SRM_INDEX_CNTL_ADDR_2 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
710 mmRLC_SRM_INDEX_CNTL_ADDR_3 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
711 mmRLC_SRM_INDEX_CNTL_ADDR_4 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
712 mmRLC_SRM_INDEX_CNTL_ADDR_5 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
713 mmRLC_SRM_INDEX_CNTL_ADDR_6 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
714 mmRLC_SRM_INDEX_CNTL_ADDR_7 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
715 };
716
717 static const u32 GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[] =
718 {
719 mmRLC_SRM_INDEX_CNTL_DATA_0 - mmRLC_SRM_INDEX_CNTL_DATA_0,
720 mmRLC_SRM_INDEX_CNTL_DATA_1 - mmRLC_SRM_INDEX_CNTL_DATA_0,
721 mmRLC_SRM_INDEX_CNTL_DATA_2 - mmRLC_SRM_INDEX_CNTL_DATA_0,
722 mmRLC_SRM_INDEX_CNTL_DATA_3 - mmRLC_SRM_INDEX_CNTL_DATA_0,
723 mmRLC_SRM_INDEX_CNTL_DATA_4 - mmRLC_SRM_INDEX_CNTL_DATA_0,
724 mmRLC_SRM_INDEX_CNTL_DATA_5 - mmRLC_SRM_INDEX_CNTL_DATA_0,
725 mmRLC_SRM_INDEX_CNTL_DATA_6 - mmRLC_SRM_INDEX_CNTL_DATA_0,
726 mmRLC_SRM_INDEX_CNTL_DATA_7 - mmRLC_SRM_INDEX_CNTL_DATA_0,
727 };
728
729 void gfx_v9_0_rlcg_wreg(struct amdgpu_device *adev, u32 offset, u32 v)
730 {
731 static void *scratch_reg0;
732 static void *scratch_reg1;
733 static void *scratch_reg2;
734 static void *scratch_reg3;
735 static void *spare_int;
736 static uint32_t grbm_cntl;
737 static uint32_t grbm_idx;
738
739 scratch_reg0 = adev->rmmio + (adev->reg_offset[GC_HWIP][0][mmSCRATCH_REG0_BASE_IDX] + mmSCRATCH_REG0)*4;
740 scratch_reg1 = adev->rmmio + (adev->reg_offset[GC_HWIP][0][mmSCRATCH_REG1_BASE_IDX] + mmSCRATCH_REG1)*4;
741 scratch_reg2 = adev->rmmio + (adev->reg_offset[GC_HWIP][0][mmSCRATCH_REG1_BASE_IDX] + mmSCRATCH_REG2)*4;
742 scratch_reg3 = adev->rmmio + (adev->reg_offset[GC_HWIP][0][mmSCRATCH_REG1_BASE_IDX] + mmSCRATCH_REG3)*4;
743 spare_int = adev->rmmio + (adev->reg_offset[GC_HWIP][0][mmRLC_SPARE_INT_BASE_IDX] + mmRLC_SPARE_INT)*4;
744
745 grbm_cntl = adev->reg_offset[GC_HWIP][0][mmGRBM_GFX_CNTL_BASE_IDX] + mmGRBM_GFX_CNTL;
746 grbm_idx = adev->reg_offset[GC_HWIP][0][mmGRBM_GFX_INDEX_BASE_IDX] + mmGRBM_GFX_INDEX;
747
748 if (amdgpu_sriov_runtime(adev)) {
749 pr_err("shouldn't call rlcg write register during runtime\n");
750 return;
751 }
752
753 if (offset == grbm_cntl || offset == grbm_idx) {
754 if (offset == grbm_cntl)
755 writel(v, scratch_reg2);
756 else if (offset == grbm_idx)
757 writel(v, scratch_reg3);
758
759 writel(v, ((void __iomem *)adev->rmmio) + (offset * 4));
760 } else {
761 uint32_t i = 0;
762 uint32_t retries = 50000;
763
764 writel(v, scratch_reg0);
765 writel(offset | 0x80000000, scratch_reg1);
766 writel(1, spare_int);
767 for (i = 0; i < retries; i++) {
768 u32 tmp;
769
770 tmp = readl(scratch_reg1);
771 if (!(tmp & 0x80000000))
772 break;
773
774 udelay(10);
775 }
776 if (i >= retries)
777 pr_err("timeout: rlcg program reg:0x%05x failed !\n", offset);
778 }
779
780 }
781
782 #define VEGA10_GB_ADDR_CONFIG_GOLDEN 0x2a114042
783 #define VEGA12_GB_ADDR_CONFIG_GOLDEN 0x24104041
784 #define RAVEN_GB_ADDR_CONFIG_GOLDEN 0x24000042
785 #define RAVEN2_GB_ADDR_CONFIG_GOLDEN 0x26013041
786
787 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev);
788 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev);
789 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev);
790 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev);
791 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
792 struct amdgpu_cu_info *cu_info);
793 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev);
794 static void gfx_v9_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num, u32 instance);
795 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring);
796 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring);
797 static int gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
798 void *ras_error_status);
799 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
800 void *inject_if);
801 static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev);
802
803 static void gfx_v9_0_kiq_set_resources(struct amdgpu_ring *kiq_ring,
804 uint64_t queue_mask)
805 {
806 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
807 amdgpu_ring_write(kiq_ring,
808 PACKET3_SET_RESOURCES_VMID_MASK(0) |
809 /* vmid_mask:0* queue_type:0 (KIQ) */
810 PACKET3_SET_RESOURCES_QUEUE_TYPE(0));
811 amdgpu_ring_write(kiq_ring,
812 lower_32_bits(queue_mask)); /* queue mask lo */
813 amdgpu_ring_write(kiq_ring,
814 upper_32_bits(queue_mask)); /* queue mask hi */
815 amdgpu_ring_write(kiq_ring, 0); /* gws mask lo */
816 amdgpu_ring_write(kiq_ring, 0); /* gws mask hi */
817 amdgpu_ring_write(kiq_ring, 0); /* oac mask */
818 amdgpu_ring_write(kiq_ring, 0); /* gds heap base:0, gds heap size:0 */
819 }
820
821 static void gfx_v9_0_kiq_map_queues(struct amdgpu_ring *kiq_ring,
822 struct amdgpu_ring *ring)
823 {
824 struct amdgpu_device *adev = kiq_ring->adev;
825 uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
826 uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
827 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
828
829 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
830 /* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
831 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
832 PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
833 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
834 PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
835 PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
836 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
837 /*queue_type: normal compute queue */
838 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) |
839 /* alloc format: all_on_one_pipe */
840 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) |
841 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
842 /* num_queues: must be 1 */
843 PACKET3_MAP_QUEUES_NUM_QUEUES(1));
844 amdgpu_ring_write(kiq_ring,
845 PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
846 amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
847 amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
848 amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
849 amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
850 }
851
852 static void gfx_v9_0_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
853 struct amdgpu_ring *ring,
854 enum amdgpu_unmap_queues_action action,
855 u64 gpu_addr, u64 seq)
856 {
857 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
858
859 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
860 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
861 PACKET3_UNMAP_QUEUES_ACTION(action) |
862 PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
863 PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
864 PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
865 amdgpu_ring_write(kiq_ring,
866 PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
867
868 if (action == PREEMPT_QUEUES_NO_UNMAP) {
869 amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr));
870 amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr));
871 amdgpu_ring_write(kiq_ring, seq);
872 } else {
873 amdgpu_ring_write(kiq_ring, 0);
874 amdgpu_ring_write(kiq_ring, 0);
875 amdgpu_ring_write(kiq_ring, 0);
876 }
877 }
878
879 static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
880 struct amdgpu_ring *ring,
881 u64 addr,
882 u64 seq)
883 {
884 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
885
886 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
887 amdgpu_ring_write(kiq_ring,
888 PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
889 PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
890 PACKET3_QUERY_STATUS_COMMAND(2));
891 /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
892 amdgpu_ring_write(kiq_ring,
893 PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
894 PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
895 amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
896 amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
897 amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
898 amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
899 }
900
901 static void gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
902 uint16_t pasid, uint32_t flush_type,
903 bool all_hub)
904 {
905 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
906 amdgpu_ring_write(kiq_ring,
907 PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
908 PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
909 PACKET3_INVALIDATE_TLBS_PASID(pasid) |
910 PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
911 }
912
913 static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
914 .kiq_set_resources = gfx_v9_0_kiq_set_resources,
915 .kiq_map_queues = gfx_v9_0_kiq_map_queues,
916 .kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
917 .kiq_query_status = gfx_v9_0_kiq_query_status,
918 .kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs,
919 .set_resources_size = 8,
920 .map_queues_size = 7,
921 .unmap_queues_size = 6,
922 .query_status_size = 7,
923 .invalidate_tlbs_size = 2,
924 };
925
926 static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
927 {
928 adev->gfx.kiq.pmf = &gfx_v9_0_kiq_pm4_funcs;
929 }
930
931 static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev)
932 {
933 switch (adev->asic_type) {
934 case CHIP_VEGA10:
935 soc15_program_register_sequence(adev,
936 golden_settings_gc_9_0,
937 ARRAY_SIZE(golden_settings_gc_9_0));
938 soc15_program_register_sequence(adev,
939 golden_settings_gc_9_0_vg10,
940 ARRAY_SIZE(golden_settings_gc_9_0_vg10));
941 break;
942 case CHIP_VEGA12:
943 soc15_program_register_sequence(adev,
944 golden_settings_gc_9_2_1,
945 ARRAY_SIZE(golden_settings_gc_9_2_1));
946 soc15_program_register_sequence(adev,
947 golden_settings_gc_9_2_1_vg12,
948 ARRAY_SIZE(golden_settings_gc_9_2_1_vg12));
949 break;
950 case CHIP_VEGA20:
951 soc15_program_register_sequence(adev,
952 golden_settings_gc_9_0,
953 ARRAY_SIZE(golden_settings_gc_9_0));
954 soc15_program_register_sequence(adev,
955 golden_settings_gc_9_0_vg20,
956 ARRAY_SIZE(golden_settings_gc_9_0_vg20));
957 break;
958 case CHIP_ARCTURUS:
959 soc15_program_register_sequence(adev,
960 golden_settings_gc_9_4_1_arct,
961 ARRAY_SIZE(golden_settings_gc_9_4_1_arct));
962 break;
963 case CHIP_RAVEN:
964 soc15_program_register_sequence(adev, golden_settings_gc_9_1,
965 ARRAY_SIZE(golden_settings_gc_9_1));
966 if (adev->rev_id >= 8)
967 soc15_program_register_sequence(adev,
968 golden_settings_gc_9_1_rv2,
969 ARRAY_SIZE(golden_settings_gc_9_1_rv2));
970 else
971 soc15_program_register_sequence(adev,
972 golden_settings_gc_9_1_rv1,
973 ARRAY_SIZE(golden_settings_gc_9_1_rv1));
974 break;
975 case CHIP_RENOIR:
976 soc15_program_register_sequence(adev,
977 golden_settings_gc_9_1_rn,
978 ARRAY_SIZE(golden_settings_gc_9_1_rn));
979 return; /* for renoir, don't need common goldensetting */
980 default:
981 break;
982 }
983
984 if (adev->asic_type != CHIP_ARCTURUS)
985 soc15_program_register_sequence(adev, golden_settings_gc_9_x_common,
986 (const u32)ARRAY_SIZE(golden_settings_gc_9_x_common));
987 }
988
989 static void gfx_v9_0_scratch_init(struct amdgpu_device *adev)
990 {
991 adev->gfx.scratch.num_reg = 8;
992 adev->gfx.scratch.reg_base = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0);
993 adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
994 }
995
996 static void gfx_v9_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
997 bool wc, uint32_t reg, uint32_t val)
998 {
999 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
1000 amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) |
1001 WRITE_DATA_DST_SEL(0) |
1002 (wc ? WR_CONFIRM : 0));
1003 amdgpu_ring_write(ring, reg);
1004 amdgpu_ring_write(ring, 0);
1005 amdgpu_ring_write(ring, val);
1006 }
1007
1008 static void gfx_v9_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
1009 int mem_space, int opt, uint32_t addr0,
1010 uint32_t addr1, uint32_t ref, uint32_t mask,
1011 uint32_t inv)
1012 {
1013 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
1014 amdgpu_ring_write(ring,
1015 /* memory (1) or register (0) */
1016 (WAIT_REG_MEM_MEM_SPACE(mem_space) |
1017 WAIT_REG_MEM_OPERATION(opt) | /* wait */
1018 WAIT_REG_MEM_FUNCTION(3) | /* equal */
1019 WAIT_REG_MEM_ENGINE(eng_sel)));
1020
1021 if (mem_space)
1022 BUG_ON(addr0 & 0x3); /* Dword align */
1023 amdgpu_ring_write(ring, addr0);
1024 amdgpu_ring_write(ring, addr1);
1025 amdgpu_ring_write(ring, ref);
1026 amdgpu_ring_write(ring, mask);
1027 amdgpu_ring_write(ring, inv); /* poll interval */
1028 }
1029
1030 static int gfx_v9_0_ring_test_ring(struct amdgpu_ring *ring)
1031 {
1032 struct amdgpu_device *adev = ring->adev;
1033 uint32_t scratch;
1034 uint32_t tmp = 0;
1035 unsigned i;
1036 int r;
1037
1038 r = amdgpu_gfx_scratch_get(adev, &scratch);
1039 if (r)
1040 return r;
1041
1042 WREG32(scratch, 0xCAFEDEAD);
1043 r = amdgpu_ring_alloc(ring, 3);
1044 if (r)
1045 goto error_free_scratch;
1046
1047 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
1048 amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
1049 amdgpu_ring_write(ring, 0xDEADBEEF);
1050 amdgpu_ring_commit(ring);
1051
1052 for (i = 0; i < adev->usec_timeout; i++) {
1053 tmp = RREG32(scratch);
1054 if (tmp == 0xDEADBEEF)
1055 break;
1056 udelay(1);
1057 }
1058
1059 if (i >= adev->usec_timeout)
1060 r = -ETIMEDOUT;
1061
1062 error_free_scratch:
1063 amdgpu_gfx_scratch_free(adev, scratch);
1064 return r;
1065 }
1066
1067 static int gfx_v9_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1068 {
1069 struct amdgpu_device *adev = ring->adev;
1070 struct amdgpu_ib ib;
1071 struct dma_fence *f = NULL;
1072
1073 unsigned index;
1074 uint64_t gpu_addr;
1075 uint32_t tmp;
1076 long r;
1077
1078 r = amdgpu_device_wb_get(adev, &index);
1079 if (r)
1080 return r;
1081
1082 gpu_addr = adev->wb.gpu_addr + (index * 4);
1083 adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
1084 memset(&ib, 0, sizeof(ib));
1085 r = amdgpu_ib_get(adev, NULL, 16, &ib);
1086 if (r)
1087 goto err1;
1088
1089 ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
1090 ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
1091 ib.ptr[2] = lower_32_bits(gpu_addr);
1092 ib.ptr[3] = upper_32_bits(gpu_addr);
1093 ib.ptr[4] = 0xDEADBEEF;
1094 ib.length_dw = 5;
1095
1096 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1097 if (r)
1098 goto err2;
1099
1100 r = dma_fence_wait_timeout(f, false, timeout);
1101 if (r == 0) {
1102 r = -ETIMEDOUT;
1103 goto err2;
1104 } else if (r < 0) {
1105 goto err2;
1106 }
1107
1108 tmp = adev->wb.wb[index];
1109 if (tmp == 0xDEADBEEF)
1110 r = 0;
1111 else
1112 r = -EINVAL;
1113
1114 err2:
1115 amdgpu_ib_free(adev, &ib, NULL);
1116 dma_fence_put(f);
1117 err1:
1118 amdgpu_device_wb_free(adev, index);
1119 return r;
1120 }
1121
1122
1123 static void gfx_v9_0_free_microcode(struct amdgpu_device *adev)
1124 {
1125 release_firmware(adev->gfx.pfp_fw);
1126 adev->gfx.pfp_fw = NULL;
1127 release_firmware(adev->gfx.me_fw);
1128 adev->gfx.me_fw = NULL;
1129 release_firmware(adev->gfx.ce_fw);
1130 adev->gfx.ce_fw = NULL;
1131 release_firmware(adev->gfx.rlc_fw);
1132 adev->gfx.rlc_fw = NULL;
1133 release_firmware(adev->gfx.mec_fw);
1134 adev->gfx.mec_fw = NULL;
1135 release_firmware(adev->gfx.mec2_fw);
1136 adev->gfx.mec2_fw = NULL;
1137
1138 kfree(adev->gfx.rlc.register_list_format);
1139 }
1140
1141 static void gfx_v9_0_init_rlc_ext_microcode(struct amdgpu_device *adev)
1142 {
1143 const struct rlc_firmware_header_v2_1 *rlc_hdr;
1144
1145 rlc_hdr = (const struct rlc_firmware_header_v2_1 *)adev->gfx.rlc_fw->data;
1146 adev->gfx.rlc_srlc_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_ucode_ver);
1147 adev->gfx.rlc_srlc_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_feature_ver);
1148 adev->gfx.rlc.save_restore_list_cntl_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_cntl_size_bytes);
1149 adev->gfx.rlc.save_restore_list_cntl = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_cntl_offset_bytes);
1150 adev->gfx.rlc_srlg_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_ucode_ver);
1151 adev->gfx.rlc_srlg_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_feature_ver);
1152 adev->gfx.rlc.save_restore_list_gpm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_gpm_size_bytes);
1153 adev->gfx.rlc.save_restore_list_gpm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_gpm_offset_bytes);
1154 adev->gfx.rlc_srls_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_ucode_ver);
1155 adev->gfx.rlc_srls_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_feature_ver);
1156 adev->gfx.rlc.save_restore_list_srm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_srm_size_bytes);
1157 adev->gfx.rlc.save_restore_list_srm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_srm_offset_bytes);
1158 adev->gfx.rlc.reg_list_format_direct_reg_list_length =
1159 le32_to_cpu(rlc_hdr->reg_list_format_direct_reg_list_length);
1160 }
1161
1162 static void gfx_v9_0_check_fw_write_wait(struct amdgpu_device *adev)
1163 {
1164 adev->gfx.me_fw_write_wait = false;
1165 adev->gfx.mec_fw_write_wait = false;
1166
1167 if ((adev->asic_type != CHIP_ARCTURUS) &&
1168 ((adev->gfx.mec_fw_version < 0x000001a5) ||
1169 (adev->gfx.mec_feature_version < 46) ||
1170 (adev->gfx.pfp_fw_version < 0x000000b7) ||
1171 (adev->gfx.pfp_feature_version < 46)))
1172 DRM_WARN_ONCE("CP firmware version too old, please update!");
1173
1174 switch (adev->asic_type) {
1175 case CHIP_VEGA10:
1176 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1177 (adev->gfx.me_feature_version >= 42) &&
1178 (adev->gfx.pfp_fw_version >= 0x000000b1) &&
1179 (adev->gfx.pfp_feature_version >= 42))
1180 adev->gfx.me_fw_write_wait = true;
1181
1182 if ((adev->gfx.mec_fw_version >= 0x00000193) &&
1183 (adev->gfx.mec_feature_version >= 42))
1184 adev->gfx.mec_fw_write_wait = true;
1185 break;
1186 case CHIP_VEGA12:
1187 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1188 (adev->gfx.me_feature_version >= 44) &&
1189 (adev->gfx.pfp_fw_version >= 0x000000b2) &&
1190 (adev->gfx.pfp_feature_version >= 44))
1191 adev->gfx.me_fw_write_wait = true;
1192
1193 if ((adev->gfx.mec_fw_version >= 0x00000196) &&
1194 (adev->gfx.mec_feature_version >= 44))
1195 adev->gfx.mec_fw_write_wait = true;
1196 break;
1197 case CHIP_VEGA20:
1198 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1199 (adev->gfx.me_feature_version >= 44) &&
1200 (adev->gfx.pfp_fw_version >= 0x000000b2) &&
1201 (adev->gfx.pfp_feature_version >= 44))
1202 adev->gfx.me_fw_write_wait = true;
1203
1204 if ((adev->gfx.mec_fw_version >= 0x00000197) &&
1205 (adev->gfx.mec_feature_version >= 44))
1206 adev->gfx.mec_fw_write_wait = true;
1207 break;
1208 case CHIP_RAVEN:
1209 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1210 (adev->gfx.me_feature_version >= 42) &&
1211 (adev->gfx.pfp_fw_version >= 0x000000b1) &&
1212 (adev->gfx.pfp_feature_version >= 42))
1213 adev->gfx.me_fw_write_wait = true;
1214
1215 if ((adev->gfx.mec_fw_version >= 0x00000192) &&
1216 (adev->gfx.mec_feature_version >= 42))
1217 adev->gfx.mec_fw_write_wait = true;
1218 break;
1219 default:
1220 adev->gfx.me_fw_write_wait = true;
1221 adev->gfx.mec_fw_write_wait = true;
1222 break;
1223 }
1224 }
1225
1226 struct amdgpu_gfxoff_quirk {
1227 u16 chip_vendor;
1228 u16 chip_device;
1229 u16 subsys_vendor;
1230 u16 subsys_device;
1231 u8 revision;
1232 };
1233
1234 static const struct amdgpu_gfxoff_quirk amdgpu_gfxoff_quirk_list[] = {
1235 /* https://bugzilla.kernel.org/show_bug.cgi?id=204689 */
1236 { 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc8 },
1237 /* https://bugzilla.kernel.org/show_bug.cgi?id=207171 */
1238 { 0x1002, 0x15dd, 0x103c, 0x83e7, 0xd3 },
1239 /* GFXOFF is unstable on C6 parts with a VBIOS 113-RAVEN-114 */
1240 { 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc6 },
1241 { 0, 0, 0, 0, 0 },
1242 };
1243
1244 static bool gfx_v9_0_should_disable_gfxoff(struct pci_dev *pdev)
1245 {
1246 const struct amdgpu_gfxoff_quirk *p = amdgpu_gfxoff_quirk_list;
1247
1248 while (p && p->chip_device != 0) {
1249 if (pdev->vendor == p->chip_vendor &&
1250 pdev->device == p->chip_device &&
1251 pdev->subsystem_vendor == p->subsys_vendor &&
1252 pdev->subsystem_device == p->subsys_device &&
1253 pdev->revision == p->revision) {
1254 return true;
1255 }
1256 ++p;
1257 }
1258 return false;
1259 }
1260
1261 static bool is_raven_kicker(struct amdgpu_device *adev)
1262 {
1263 if (adev->pm.fw_version >= 0x41e2b)
1264 return true;
1265 else
1266 return false;
1267 }
1268
1269 static void gfx_v9_0_check_if_need_gfxoff(struct amdgpu_device *adev)
1270 {
1271 if (gfx_v9_0_should_disable_gfxoff(adev->pdev))
1272 adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1273
1274 switch (adev->asic_type) {
1275 case CHIP_VEGA10:
1276 case CHIP_VEGA12:
1277 case CHIP_VEGA20:
1278 break;
1279 case CHIP_RAVEN:
1280 if (!(adev->rev_id >= 0x8 || adev->pdev->device == 0x15d8) &&
1281 ((!is_raven_kicker(adev) &&
1282 adev->gfx.rlc_fw_version < 531) ||
1283 (adev->gfx.rlc_feature_version < 1) ||
1284 !adev->gfx.rlc.is_rlc_v2_1))
1285 adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1286
1287 if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1288 adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG |
1289 AMD_PG_SUPPORT_CP |
1290 AMD_PG_SUPPORT_RLC_SMU_HS;
1291 break;
1292 case CHIP_RENOIR:
1293 if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1294 adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG |
1295 AMD_PG_SUPPORT_CP |
1296 AMD_PG_SUPPORT_RLC_SMU_HS;
1297 break;
1298 default:
1299 break;
1300 }
1301 }
1302
1303 static int gfx_v9_0_init_cp_gfx_microcode(struct amdgpu_device *adev,
1304 const char *chip_name)
1305 {
1306 char fw_name[30];
1307 int err;
1308 struct amdgpu_firmware_info *info = NULL;
1309 const struct common_firmware_header *header = NULL;
1310 const struct gfx_firmware_header_v1_0 *cp_hdr;
1311
1312 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
1313 err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1314 if (err)
1315 goto out;
1316 err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
1317 if (err)
1318 goto out;
1319 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
1320 adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1321 adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1322
1323 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1324 err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1325 if (err)
1326 goto out;
1327 err = amdgpu_ucode_validate(adev->gfx.me_fw);
1328 if (err)
1329 goto out;
1330 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
1331 adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1332 adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1333
1334 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1335 err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1336 if (err)
1337 goto out;
1338 err = amdgpu_ucode_validate(adev->gfx.ce_fw);
1339 if (err)
1340 goto out;
1341 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
1342 adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1343 adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1344
1345 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1346 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
1347 info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
1348 info->fw = adev->gfx.pfp_fw;
1349 header = (const struct common_firmware_header *)info->fw->data;
1350 adev->firmware.fw_size +=
1351 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1352
1353 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
1354 info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
1355 info->fw = adev->gfx.me_fw;
1356 header = (const struct common_firmware_header *)info->fw->data;
1357 adev->firmware.fw_size +=
1358 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1359
1360 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE];
1361 info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
1362 info->fw = adev->gfx.ce_fw;
1363 header = (const struct common_firmware_header *)info->fw->data;
1364 adev->firmware.fw_size +=
1365 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1366 }
1367
1368 out:
1369 if (err) {
1370 dev_err(adev->dev,
1371 "gfx9: Failed to load firmware \"%s\"\n",
1372 fw_name);
1373 release_firmware(adev->gfx.pfp_fw);
1374 adev->gfx.pfp_fw = NULL;
1375 release_firmware(adev->gfx.me_fw);
1376 adev->gfx.me_fw = NULL;
1377 release_firmware(adev->gfx.ce_fw);
1378 adev->gfx.ce_fw = NULL;
1379 }
1380 return err;
1381 }
1382
1383 static int gfx_v9_0_init_rlc_microcode(struct amdgpu_device *adev,
1384 const char *chip_name)
1385 {
1386 char fw_name[30];
1387 int err;
1388 struct amdgpu_firmware_info *info = NULL;
1389 const struct common_firmware_header *header = NULL;
1390 const struct rlc_firmware_header_v2_0 *rlc_hdr;
1391 unsigned int *tmp = NULL;
1392 unsigned int i = 0;
1393 uint16_t version_major;
1394 uint16_t version_minor;
1395 uint32_t smu_version;
1396
1397 /*
1398 * For Picasso && AM4 SOCKET board, we use picasso_rlc_am4.bin
1399 * instead of picasso_rlc.bin.
1400 * Judgment method:
1401 * PCO AM4: revision >= 0xC8 && revision <= 0xCF
1402 * or revision >= 0xD8 && revision <= 0xDF
1403 * otherwise is PCO FP5
1404 */
1405 if (!strcmp(chip_name, "picasso") &&
1406 (((adev->pdev->revision >= 0xC8) && (adev->pdev->revision <= 0xCF)) ||
1407 ((adev->pdev->revision >= 0xD8) && (adev->pdev->revision <= 0xDF))))
1408 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc_am4.bin", chip_name);
1409 else if (!strcmp(chip_name, "raven") && (amdgpu_pm_load_smu_firmware(adev, &smu_version) == 0) &&
1410 (smu_version >= 0x41e2b))
1411 /**
1412 *SMC is loaded by SBIOS on APU and it's able to get the SMU version directly.
1413 */
1414 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_kicker_rlc.bin", chip_name);
1415 else
1416 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
1417 err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
1418 if (err)
1419 goto out;
1420 err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
1421 rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1422
1423 version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
1424 version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
1425 if (version_major == 2 && version_minor == 1)
1426 adev->gfx.rlc.is_rlc_v2_1 = true;
1427
1428 adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
1429 adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
1430 adev->gfx.rlc.save_and_restore_offset =
1431 le32_to_cpu(rlc_hdr->save_and_restore_offset);
1432 adev->gfx.rlc.clear_state_descriptor_offset =
1433 le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
1434 adev->gfx.rlc.avail_scratch_ram_locations =
1435 le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
1436 adev->gfx.rlc.reg_restore_list_size =
1437 le32_to_cpu(rlc_hdr->reg_restore_list_size);
1438 adev->gfx.rlc.reg_list_format_start =
1439 le32_to_cpu(rlc_hdr->reg_list_format_start);
1440 adev->gfx.rlc.reg_list_format_separate_start =
1441 le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
1442 adev->gfx.rlc.starting_offsets_start =
1443 le32_to_cpu(rlc_hdr->starting_offsets_start);
1444 adev->gfx.rlc.reg_list_format_size_bytes =
1445 le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
1446 adev->gfx.rlc.reg_list_size_bytes =
1447 le32_to_cpu(rlc_hdr->reg_list_size_bytes);
1448 adev->gfx.rlc.register_list_format =
1449 kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
1450 adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
1451 if (!adev->gfx.rlc.register_list_format) {
1452 err = -ENOMEM;
1453 goto out;
1454 }
1455
1456 tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1457 le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
1458 for (i = 0 ; i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2); i++)
1459 adev->gfx.rlc.register_list_format[i] = le32_to_cpu(tmp[i]);
1460
1461 adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
1462
1463 tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1464 le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
1465 for (i = 0 ; i < (adev->gfx.rlc.reg_list_size_bytes >> 2); i++)
1466 adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
1467
1468 if (adev->gfx.rlc.is_rlc_v2_1)
1469 gfx_v9_0_init_rlc_ext_microcode(adev);
1470
1471 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1472 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
1473 info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
1474 info->fw = adev->gfx.rlc_fw;
1475 header = (const struct common_firmware_header *)info->fw->data;
1476 adev->firmware.fw_size +=
1477 ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1478
1479 if (adev->gfx.rlc.is_rlc_v2_1 &&
1480 adev->gfx.rlc.save_restore_list_cntl_size_bytes &&
1481 adev->gfx.rlc.save_restore_list_gpm_size_bytes &&
1482 adev->gfx.rlc.save_restore_list_srm_size_bytes) {
1483 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL];
1484 info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL;
1485 info->fw = adev->gfx.rlc_fw;
1486 adev->firmware.fw_size +=
1487 ALIGN(adev->gfx.rlc.save_restore_list_cntl_size_bytes, PAGE_SIZE);
1488
1489 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM];
1490 info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM;
1491 info->fw = adev->gfx.rlc_fw;
1492 adev->firmware.fw_size +=
1493 ALIGN(adev->gfx.rlc.save_restore_list_gpm_size_bytes, PAGE_SIZE);
1494
1495 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM];
1496 info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM;
1497 info->fw = adev->gfx.rlc_fw;
1498 adev->firmware.fw_size +=
1499 ALIGN(adev->gfx.rlc.save_restore_list_srm_size_bytes, PAGE_SIZE);
1500 }
1501 }
1502
1503 out:
1504 if (err) {
1505 dev_err(adev->dev,
1506 "gfx9: Failed to load firmware \"%s\"\n",
1507 fw_name);
1508 release_firmware(adev->gfx.rlc_fw);
1509 adev->gfx.rlc_fw = NULL;
1510 }
1511 return err;
1512 }
1513
1514 static int gfx_v9_0_init_cp_compute_microcode(struct amdgpu_device *adev,
1515 const char *chip_name)
1516 {
1517 char fw_name[30];
1518 int err;
1519 struct amdgpu_firmware_info *info = NULL;
1520 const struct common_firmware_header *header = NULL;
1521 const struct gfx_firmware_header_v1_0 *cp_hdr;
1522
1523 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1524 err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1525 if (err)
1526 goto out;
1527 err = amdgpu_ucode_validate(adev->gfx.mec_fw);
1528 if (err)
1529 goto out;
1530 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1531 adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1532 adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1533
1534
1535 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1536 err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1537 if (!err) {
1538 err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
1539 if (err)
1540 goto out;
1541 cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1542 adev->gfx.mec2_fw->data;
1543 adev->gfx.mec2_fw_version =
1544 le32_to_cpu(cp_hdr->header.ucode_version);
1545 adev->gfx.mec2_feature_version =
1546 le32_to_cpu(cp_hdr->ucode_feature_version);
1547 } else {
1548 err = 0;
1549 adev->gfx.mec2_fw = NULL;
1550 }
1551
1552 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1553 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
1554 info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
1555 info->fw = adev->gfx.mec_fw;
1556 header = (const struct common_firmware_header *)info->fw->data;
1557 cp_hdr = (const struct gfx_firmware_header_v1_0 *)info->fw->data;
1558 adev->firmware.fw_size +=
1559 ALIGN(le32_to_cpu(header->ucode_size_bytes) - le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
1560
1561 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1_JT];
1562 info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1_JT;
1563 info->fw = adev->gfx.mec_fw;
1564 adev->firmware.fw_size +=
1565 ALIGN(le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
1566
1567 if (adev->gfx.mec2_fw) {
1568 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
1569 info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
1570 info->fw = adev->gfx.mec2_fw;
1571 header = (const struct common_firmware_header *)info->fw->data;
1572 cp_hdr = (const struct gfx_firmware_header_v1_0 *)info->fw->data;
1573 adev->firmware.fw_size +=
1574 ALIGN(le32_to_cpu(header->ucode_size_bytes) - le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
1575
1576 /* TODO: Determine if MEC2 JT FW loading can be removed
1577 for all GFX V9 asic and above */
1578 if (adev->asic_type != CHIP_ARCTURUS &&
1579 adev->asic_type != CHIP_RENOIR) {
1580 info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2_JT];
1581 info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2_JT;
1582 info->fw = adev->gfx.mec2_fw;
1583 adev->firmware.fw_size +=
1584 ALIGN(le32_to_cpu(cp_hdr->jt_size) * 4,
1585 PAGE_SIZE);
1586 }
1587 }
1588 }
1589
1590 out:
1591 gfx_v9_0_check_if_need_gfxoff(adev);
1592 gfx_v9_0_check_fw_write_wait(adev);
1593 if (err) {
1594 dev_err(adev->dev,
1595 "gfx9: Failed to load firmware \"%s\"\n",
1596 fw_name);
1597 release_firmware(adev->gfx.mec_fw);
1598 adev->gfx.mec_fw = NULL;
1599 release_firmware(adev->gfx.mec2_fw);
1600 adev->gfx.mec2_fw = NULL;
1601 }
1602 return err;
1603 }
1604
1605 static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
1606 {
1607 const char *chip_name;
1608 int r;
1609
1610 DRM_DEBUG("\n");
1611
1612 switch (adev->asic_type) {
1613 case CHIP_VEGA10:
1614 chip_name = "vega10";
1615 break;
1616 case CHIP_VEGA12:
1617 chip_name = "vega12";
1618 break;
1619 case CHIP_VEGA20:
1620 chip_name = "vega20";
1621 break;
1622 case CHIP_RAVEN:
1623 if (adev->rev_id >= 8)
1624 chip_name = "raven2";
1625 else if (adev->pdev->device == 0x15d8)
1626 chip_name = "picasso";
1627 else
1628 chip_name = "raven";
1629 break;
1630 case CHIP_ARCTURUS:
1631 chip_name = "arcturus";
1632 break;
1633 case CHIP_RENOIR:
1634 chip_name = "renoir";
1635 break;
1636 default:
1637 BUG();
1638 }
1639
1640 /* No CPG in Arcturus */
1641 if (adev->asic_type != CHIP_ARCTURUS) {
1642 r = gfx_v9_0_init_cp_gfx_microcode(adev, chip_name);
1643 if (r)
1644 return r;
1645 }
1646
1647 r = gfx_v9_0_init_rlc_microcode(adev, chip_name);
1648 if (r)
1649 return r;
1650
1651 r = gfx_v9_0_init_cp_compute_microcode(adev, chip_name);
1652 if (r)
1653 return r;
1654
1655 return r;
1656 }
1657
1658 static u32 gfx_v9_0_get_csb_size(struct amdgpu_device *adev)
1659 {
1660 u32 count = 0;
1661 const struct cs_section_def *sect = NULL;
1662 const struct cs_extent_def *ext = NULL;
1663
1664 /* begin clear state */
1665 count += 2;
1666 /* context control state */
1667 count += 3;
1668
1669 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
1670 for (ext = sect->section; ext->extent != NULL; ++ext) {
1671 if (sect->id == SECT_CONTEXT)
1672 count += 2 + ext->reg_count;
1673 else
1674 return 0;
1675 }
1676 }
1677
1678 /* end clear state */
1679 count += 2;
1680 /* clear state */
1681 count += 2;
1682
1683 return count;
1684 }
1685
1686 static void gfx_v9_0_get_csb_buffer(struct amdgpu_device *adev,
1687 volatile u32 *buffer)
1688 {
1689 u32 count = 0, i;
1690 const struct cs_section_def *sect = NULL;
1691 const struct cs_extent_def *ext = NULL;
1692
1693 if (adev->gfx.rlc.cs_data == NULL)
1694 return;
1695 if (buffer == NULL)
1696 return;
1697
1698 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1699 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1700
1701 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1702 buffer[count++] = cpu_to_le32(0x80000000);
1703 buffer[count++] = cpu_to_le32(0x80000000);
1704
1705 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1706 for (ext = sect->section; ext->extent != NULL; ++ext) {
1707 if (sect->id == SECT_CONTEXT) {
1708 buffer[count++] =
1709 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1710 buffer[count++] = cpu_to_le32(ext->reg_index -
1711 PACKET3_SET_CONTEXT_REG_START);
1712 for (i = 0; i < ext->reg_count; i++)
1713 buffer[count++] = cpu_to_le32(ext->extent[i]);
1714 } else {
1715 return;
1716 }
1717 }
1718 }
1719
1720 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1721 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1722
1723 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1724 buffer[count++] = cpu_to_le32(0);
1725 }
1726
1727 static void gfx_v9_0_init_always_on_cu_mask(struct amdgpu_device *adev)
1728 {
1729 struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
1730 uint32_t pg_always_on_cu_num = 2;
1731 uint32_t always_on_cu_num;
1732 uint32_t i, j, k;
1733 uint32_t mask, cu_bitmap, counter;
1734
1735 if (adev->flags & AMD_IS_APU)
1736 always_on_cu_num = 4;
1737 else if (adev->asic_type == CHIP_VEGA12)
1738 always_on_cu_num = 8;
1739 else
1740 always_on_cu_num = 12;
1741
1742 mutex_lock(&adev->grbm_idx_mutex);
1743 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1744 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1745 mask = 1;
1746 cu_bitmap = 0;
1747 counter = 0;
1748 gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
1749
1750 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
1751 if (cu_info->bitmap[i][j] & mask) {
1752 if (counter == pg_always_on_cu_num)
1753 WREG32_SOC15(GC, 0, mmRLC_PG_ALWAYS_ON_CU_MASK, cu_bitmap);
1754 if (counter < always_on_cu_num)
1755 cu_bitmap |= mask;
1756 else
1757 break;
1758 counter++;
1759 }
1760 mask <<= 1;
1761 }
1762
1763 WREG32_SOC15(GC, 0, mmRLC_LB_ALWAYS_ACTIVE_CU_MASK, cu_bitmap);
1764 cu_info->ao_cu_bitmap[i][j] = cu_bitmap;
1765 }
1766 }
1767 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1768 mutex_unlock(&adev->grbm_idx_mutex);
1769 }
1770
1771 static void gfx_v9_0_init_lbpw(struct amdgpu_device *adev)
1772 {
1773 uint32_t data;
1774
1775 /* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1776 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F);
1777 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x0333A5A7);
1778 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077);
1779 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x30 | 0x40 << 8 | 0x02FA << 16));
1780
1781 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1782 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000);
1783
1784 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1785 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000500);
1786
1787 mutex_lock(&adev->grbm_idx_mutex);
1788 /* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1789 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1790 WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff);
1791
1792 /* set mmRLC_LB_PARAMS = 0x003F_1006 */
1793 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003);
1794 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010);
1795 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F);
1796 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data);
1797
1798 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1799 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7);
1800 data &= 0x0000FFFF;
1801 data |= 0x00C00000;
1802 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data);
1803
1804 /*
1805 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xF (4 CUs AON for Raven),
1806 * programmed in gfx_v9_0_init_always_on_cu_mask()
1807 */
1808
1809 /* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1810 * but used for RLC_LB_CNTL configuration */
1811 data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK;
1812 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09);
1813 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000);
1814 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data);
1815 mutex_unlock(&adev->grbm_idx_mutex);
1816
1817 gfx_v9_0_init_always_on_cu_mask(adev);
1818 }
1819
1820 static void gfx_v9_4_init_lbpw(struct amdgpu_device *adev)
1821 {
1822 uint32_t data;
1823
1824 /* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1825 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F);
1826 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x033388F8);
1827 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077);
1828 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x10 | 0x27 << 8 | 0x02FA << 16));
1829
1830 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1831 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000);
1832
1833 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1834 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000800);
1835
1836 mutex_lock(&adev->grbm_idx_mutex);
1837 /* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1838 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1839 WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff);
1840
1841 /* set mmRLC_LB_PARAMS = 0x003F_1006 */
1842 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003);
1843 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010);
1844 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F);
1845 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data);
1846
1847 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1848 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7);
1849 data &= 0x0000FFFF;
1850 data |= 0x00C00000;
1851 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data);
1852
1853 /*
1854 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xFFF (12 CUs AON),
1855 * programmed in gfx_v9_0_init_always_on_cu_mask()
1856 */
1857
1858 /* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1859 * but used for RLC_LB_CNTL configuration */
1860 data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK;
1861 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09);
1862 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000);
1863 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data);
1864 mutex_unlock(&adev->grbm_idx_mutex);
1865
1866 gfx_v9_0_init_always_on_cu_mask(adev);
1867 }
1868
1869 static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
1870 {
1871 WREG32_FIELD15(GC, 0, RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0);
1872 }
1873
1874 static int gfx_v9_0_cp_jump_table_num(struct amdgpu_device *adev)
1875 {
1876 return 5;
1877 }
1878
1879 static int gfx_v9_0_rlc_init(struct amdgpu_device *adev)
1880 {
1881 const struct cs_section_def *cs_data;
1882 int r;
1883
1884 adev->gfx.rlc.cs_data = gfx9_cs_data;
1885
1886 cs_data = adev->gfx.rlc.cs_data;
1887
1888 if (cs_data) {
1889 /* init clear state block */
1890 r = amdgpu_gfx_rlc_init_csb(adev);
1891 if (r)
1892 return r;
1893 }
1894
1895 if (adev->asic_type == CHIP_RAVEN || adev->asic_type == CHIP_RENOIR) {
1896 /* TODO: double check the cp_table_size for RV */
1897 adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1898 r = amdgpu_gfx_rlc_init_cpt(adev);
1899 if (r)
1900 return r;
1901 }
1902
1903 switch (adev->asic_type) {
1904 case CHIP_RAVEN:
1905 gfx_v9_0_init_lbpw(adev);
1906 break;
1907 case CHIP_VEGA20:
1908 gfx_v9_4_init_lbpw(adev);
1909 break;
1910 default:
1911 break;
1912 }
1913
1914 /* init spm vmid with 0xf */
1915 if (adev->gfx.rlc.funcs->update_spm_vmid)
1916 adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
1917
1918 return 0;
1919 }
1920
1921 static void gfx_v9_0_mec_fini(struct amdgpu_device *adev)
1922 {
1923 amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1924 amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL);
1925 }
1926
1927 static int gfx_v9_0_mec_init(struct amdgpu_device *adev)
1928 {
1929 int r;
1930 u32 *hpd;
1931 const __le32 *fw_data;
1932 unsigned fw_size;
1933 u32 *fw;
1934 size_t mec_hpd_size;
1935
1936 const struct gfx_firmware_header_v1_0 *mec_hdr;
1937
1938 bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1939
1940 /* take ownership of the relevant compute queues */
1941 amdgpu_gfx_compute_queue_acquire(adev);
1942 mec_hpd_size = adev->gfx.num_compute_rings * GFX9_MEC_HPD_SIZE;
1943
1944 r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1945 AMDGPU_GEM_DOMAIN_VRAM,
1946 &adev->gfx.mec.hpd_eop_obj,
1947 &adev->gfx.mec.hpd_eop_gpu_addr,
1948 (void **)&hpd);
1949 if (r) {
1950 dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1951 gfx_v9_0_mec_fini(adev);
1952 return r;
1953 }
1954
1955 memset(hpd, 0, mec_hpd_size);
1956
1957 amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1958 amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1959
1960 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1961
1962 fw_data = (const __le32 *)
1963 (adev->gfx.mec_fw->data +
1964 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
1965 fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4;
1966
1967 r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
1968 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
1969 &adev->gfx.mec.mec_fw_obj,
1970 &adev->gfx.mec.mec_fw_gpu_addr,
1971 (void **)&fw);
1972 if (r) {
1973 dev_warn(adev->dev, "(%d) create mec firmware bo failed\n", r);
1974 gfx_v9_0_mec_fini(adev);
1975 return r;
1976 }
1977
1978 memcpy(fw, fw_data, fw_size);
1979
1980 amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
1981 amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
1982
1983 return 0;
1984 }
1985
1986 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
1987 {
1988 WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX,
1989 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1990 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
1991 (address << SQ_IND_INDEX__INDEX__SHIFT) |
1992 (SQ_IND_INDEX__FORCE_READ_MASK));
1993 return RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
1994 }
1995
1996 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
1997 uint32_t wave, uint32_t thread,
1998 uint32_t regno, uint32_t num, uint32_t *out)
1999 {
2000 WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX,
2001 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
2002 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
2003 (regno << SQ_IND_INDEX__INDEX__SHIFT) |
2004 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
2005 (SQ_IND_INDEX__FORCE_READ_MASK) |
2006 (SQ_IND_INDEX__AUTO_INCR_MASK));
2007 while (num--)
2008 *(out++) = RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
2009 }
2010
2011 static void gfx_v9_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
2012 {
2013 /* type 1 wave data */
2014 dst[(*no_fields)++] = 1;
2015 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
2016 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
2017 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
2018 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
2019 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
2020 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
2021 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
2022 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
2023 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
2024 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
2025 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
2026 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
2027 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
2028 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
2029 }
2030
2031 static void gfx_v9_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
2032 uint32_t wave, uint32_t start,
2033 uint32_t size, uint32_t *dst)
2034 {
2035 wave_read_regs(
2036 adev, simd, wave, 0,
2037 start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
2038 }
2039
2040 static void gfx_v9_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t simd,
2041 uint32_t wave, uint32_t thread,
2042 uint32_t start, uint32_t size,
2043 uint32_t *dst)
2044 {
2045 wave_read_regs(
2046 adev, simd, wave, thread,
2047 start + SQIND_WAVE_VGPRS_OFFSET, size, dst);
2048 }
2049
2050 static void gfx_v9_0_select_me_pipe_q(struct amdgpu_device *adev,
2051 u32 me, u32 pipe, u32 q, u32 vm)
2052 {
2053 soc15_grbm_select(adev, me, pipe, q, vm);
2054 }
2055
2056 static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = {
2057 .get_gpu_clock_counter = &gfx_v9_0_get_gpu_clock_counter,
2058 .select_se_sh = &gfx_v9_0_select_se_sh,
2059 .read_wave_data = &gfx_v9_0_read_wave_data,
2060 .read_wave_sgprs = &gfx_v9_0_read_wave_sgprs,
2061 .read_wave_vgprs = &gfx_v9_0_read_wave_vgprs,
2062 .select_me_pipe_q = &gfx_v9_0_select_me_pipe_q,
2063 .ras_error_inject = &gfx_v9_0_ras_error_inject,
2064 .query_ras_error_count = &gfx_v9_0_query_ras_error_count,
2065 .reset_ras_error_count = &gfx_v9_0_reset_ras_error_count,
2066 };
2067
2068 static const struct amdgpu_gfx_funcs gfx_v9_4_gfx_funcs = {
2069 .get_gpu_clock_counter = &gfx_v9_0_get_gpu_clock_counter,
2070 .select_se_sh = &gfx_v9_0_select_se_sh,
2071 .read_wave_data = &gfx_v9_0_read_wave_data,
2072 .read_wave_sgprs = &gfx_v9_0_read_wave_sgprs,
2073 .read_wave_vgprs = &gfx_v9_0_read_wave_vgprs,
2074 .select_me_pipe_q = &gfx_v9_0_select_me_pipe_q,
2075 .ras_error_inject = &gfx_v9_4_ras_error_inject,
2076 .query_ras_error_count = &gfx_v9_4_query_ras_error_count,
2077 .reset_ras_error_count = &gfx_v9_4_reset_ras_error_count,
2078 };
2079
2080 static int gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
2081 {
2082 u32 gb_addr_config;
2083 int err;
2084
2085 adev->gfx.funcs = &gfx_v9_0_gfx_funcs;
2086
2087 switch (adev->asic_type) {
2088 case CHIP_VEGA10:
2089 adev->gfx.config.max_hw_contexts = 8;
2090 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2091 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2092 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2093 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2094 gb_addr_config = VEGA10_GB_ADDR_CONFIG_GOLDEN;
2095 break;
2096 case CHIP_VEGA12:
2097 adev->gfx.config.max_hw_contexts = 8;
2098 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2099 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2100 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2101 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2102 gb_addr_config = VEGA12_GB_ADDR_CONFIG_GOLDEN;
2103 DRM_INFO("fix gfx.config for vega12\n");
2104 break;
2105 case CHIP_VEGA20:
2106 adev->gfx.config.max_hw_contexts = 8;
2107 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2108 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2109 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2110 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2111 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2112 gb_addr_config &= ~0xf3e777ff;
2113 gb_addr_config |= 0x22014042;
2114 /* check vbios table if gpu info is not available */
2115 err = amdgpu_atomfirmware_get_gfx_info(adev);
2116 if (err)
2117 return err;
2118 break;
2119 case CHIP_RAVEN:
2120 adev->gfx.config.max_hw_contexts = 8;
2121 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2122 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2123 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2124 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2125 if (adev->rev_id >= 8)
2126 gb_addr_config = RAVEN2_GB_ADDR_CONFIG_GOLDEN;
2127 else
2128 gb_addr_config = RAVEN_GB_ADDR_CONFIG_GOLDEN;
2129 break;
2130 case CHIP_ARCTURUS:
2131 adev->gfx.funcs = &gfx_v9_4_gfx_funcs;
2132 adev->gfx.config.max_hw_contexts = 8;
2133 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2134 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2135 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
2136 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2137 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2138 gb_addr_config &= ~0xf3e777ff;
2139 gb_addr_config |= 0x22014042;
2140 break;
2141 case CHIP_RENOIR:
2142 adev->gfx.config.max_hw_contexts = 8;
2143 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
2144 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
2145 adev->gfx.config.sc_hiz_tile_fifo_size = 0x80;
2146 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
2147 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
2148 gb_addr_config &= ~0xf3e777ff;
2149 gb_addr_config |= 0x22010042;
2150 break;
2151 default:
2152 BUG();
2153 break;
2154 }
2155
2156 adev->gfx.config.gb_addr_config = gb_addr_config;
2157
2158 adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
2159 REG_GET_FIELD(
2160 adev->gfx.config.gb_addr_config,
2161 GB_ADDR_CONFIG,
2162 NUM_PIPES);
2163
2164 adev->gfx.config.max_tile_pipes =
2165 adev->gfx.config.gb_addr_config_fields.num_pipes;
2166
2167 adev->gfx.config.gb_addr_config_fields.num_banks = 1 <<
2168 REG_GET_FIELD(
2169 adev->gfx.config.gb_addr_config,
2170 GB_ADDR_CONFIG,
2171 NUM_BANKS);
2172 adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
2173 REG_GET_FIELD(
2174 adev->gfx.config.gb_addr_config,
2175 GB_ADDR_CONFIG,
2176 MAX_COMPRESSED_FRAGS);
2177 adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
2178 REG_GET_FIELD(
2179 adev->gfx.config.gb_addr_config,
2180 GB_ADDR_CONFIG,
2181 NUM_RB_PER_SE);
2182 adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
2183 REG_GET_FIELD(
2184 adev->gfx.config.gb_addr_config,
2185 GB_ADDR_CONFIG,
2186 NUM_SHADER_ENGINES);
2187 adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
2188 REG_GET_FIELD(
2189 adev->gfx.config.gb_addr_config,
2190 GB_ADDR_CONFIG,
2191 PIPE_INTERLEAVE_SIZE));
2192
2193 return 0;
2194 }
2195
2196 static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
2197 int mec, int pipe, int queue)
2198 {
2199 int r;
2200 unsigned irq_type;
2201 struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
2202
2203 ring = &adev->gfx.compute_ring[ring_id];
2204
2205 /* mec0 is me1 */
2206 ring->me = mec + 1;
2207 ring->pipe = pipe;
2208 ring->queue = queue;
2209
2210 ring->ring_obj = NULL;
2211 ring->use_doorbell = true;
2212 ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1;
2213 ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
2214 + (ring_id * GFX9_MEC_HPD_SIZE);
2215 sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
2216
2217 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
2218 + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
2219 + ring->pipe;
2220
2221 /* type-2 packets are deprecated on MEC, use type-3 instead */
2222 r = amdgpu_ring_init(adev, ring, 1024,
2223 &adev->gfx.eop_irq, irq_type);
2224 if (r)
2225 return r;
2226
2227
2228 return 0;
2229 }
2230
2231 static int gfx_v9_0_sw_init(void *handle)
2232 {
2233 int i, j, k, r, ring_id;
2234 struct amdgpu_ring *ring;
2235 struct amdgpu_kiq *kiq;
2236 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2237
2238 switch (adev->asic_type) {
2239 case CHIP_VEGA10:
2240 case CHIP_VEGA12:
2241 case CHIP_VEGA20:
2242 case CHIP_RAVEN:
2243 case CHIP_ARCTURUS:
2244 case CHIP_RENOIR:
2245 adev->gfx.mec.num_mec = 2;
2246 break;
2247 default:
2248 adev->gfx.mec.num_mec = 1;
2249 break;
2250 }
2251
2252 adev->gfx.mec.num_pipe_per_mec = 4;
2253 adev->gfx.mec.num_queue_per_pipe = 8;
2254
2255 /* EOP Event */
2256 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_EOP_INTERRUPT, &adev->gfx.eop_irq);
2257 if (r)
2258 return r;
2259
2260 /* Privileged reg */
2261 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_REG_FAULT,
2262 &adev->gfx.priv_reg_irq);
2263 if (r)
2264 return r;
2265
2266 /* Privileged inst */
2267 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_INSTR_FAULT,
2268 &adev->gfx.priv_inst_irq);
2269 if (r)
2270 return r;
2271
2272 /* ECC error */
2273 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_ECC_ERROR,
2274 &adev->gfx.cp_ecc_error_irq);
2275 if (r)
2276 return r;
2277
2278 /* FUE error */
2279 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_FUE_ERROR,
2280 &adev->gfx.cp_ecc_error_irq);
2281 if (r)
2282 return r;
2283
2284 adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
2285
2286 gfx_v9_0_scratch_init(adev);
2287
2288 r = gfx_v9_0_init_microcode(adev);
2289 if (r) {
2290 DRM_ERROR("Failed to load gfx firmware!\n");
2291 return r;
2292 }
2293
2294 r = adev->gfx.rlc.funcs->init(adev);
2295 if (r) {
2296 DRM_ERROR("Failed to init rlc BOs!\n");
2297 return r;
2298 }
2299
2300 r = gfx_v9_0_mec_init(adev);
2301 if (r) {
2302 DRM_ERROR("Failed to init MEC BOs!\n");
2303 return r;
2304 }
2305
2306 /* set up the gfx ring */
2307 for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
2308 ring = &adev->gfx.gfx_ring[i];
2309 ring->ring_obj = NULL;
2310 if (!i)
2311 sprintf(ring->name, "gfx");
2312 else
2313 sprintf(ring->name, "gfx_%d", i);
2314 ring->use_doorbell = true;
2315 ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
2316 r = amdgpu_ring_init(adev, ring, 1024,
2317 &adev->gfx.eop_irq, AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP);
2318 if (r)
2319 return r;
2320 }
2321
2322 /* set up the compute queues - allocate horizontally across pipes */
2323 ring_id = 0;
2324 for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
2325 for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2326 for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2327 if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
2328 continue;
2329
2330 r = gfx_v9_0_compute_ring_init(adev,
2331 ring_id,
2332 i, k, j);
2333 if (r)
2334 return r;
2335
2336 ring_id++;
2337 }
2338 }
2339 }
2340
2341 r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE);
2342 if (r) {
2343 DRM_ERROR("Failed to init KIQ BOs!\n");
2344 return r;
2345 }
2346
2347 kiq = &adev->gfx.kiq;
2348 r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
2349 if (r)
2350 return r;
2351
2352 /* create MQD for all compute queues as wel as KIQ for SRIOV case */
2353 r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v9_mqd_allocation));
2354 if (r)
2355 return r;
2356
2357 adev->gfx.ce_ram_size = 0x8000;
2358
2359 r = gfx_v9_0_gpu_early_init(adev);
2360 if (r)
2361 return r;
2362
2363 return 0;
2364 }
2365
2366
2367 static int gfx_v9_0_sw_fini(void *handle)
2368 {
2369 int i;
2370 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2371
2372 amdgpu_gfx_ras_fini(adev);
2373
2374 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2375 amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2376 for (i = 0; i < adev->gfx.num_compute_rings; i++)
2377 amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2378
2379 amdgpu_gfx_mqd_sw_fini(adev);
2380 amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring);
2381 amdgpu_gfx_kiq_fini(adev);
2382
2383 gfx_v9_0_mec_fini(adev);
2384 amdgpu_bo_unref(&adev->gfx.rlc.clear_state_obj);
2385 if (adev->asic_type == CHIP_RAVEN || adev->asic_type == CHIP_RENOIR) {
2386 amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2387 &adev->gfx.rlc.cp_table_gpu_addr,
2388 (void **)&adev->gfx.rlc.cp_table_ptr);
2389 }
2390 gfx_v9_0_free_microcode(adev);
2391
2392 return 0;
2393 }
2394
2395
2396 static void gfx_v9_0_tiling_mode_table_init(struct amdgpu_device *adev)
2397 {
2398 /* TODO */
2399 }
2400
2401 static void gfx_v9_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num, u32 instance)
2402 {
2403 u32 data;
2404
2405 if (instance == 0xffffffff)
2406 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
2407 else
2408 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
2409
2410 if (se_num == 0xffffffff)
2411 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
2412 else
2413 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
2414
2415 if (sh_num == 0xffffffff)
2416 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
2417 else
2418 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
2419
2420 WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_INDEX, data);
2421 }
2422
2423 static u32 gfx_v9_0_get_rb_active_bitmap(struct amdgpu_device *adev)
2424 {
2425 u32 data, mask;
2426
2427 data = RREG32_SOC15(GC, 0, mmCC_RB_BACKEND_DISABLE);
2428 data |= RREG32_SOC15(GC, 0, mmGC_USER_RB_BACKEND_DISABLE);
2429
2430 data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
2431 data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
2432
2433 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
2434 adev->gfx.config.max_sh_per_se);
2435
2436 return (~data) & mask;
2437 }
2438
2439 static void gfx_v9_0_setup_rb(struct amdgpu_device *adev)
2440 {
2441 int i, j;
2442 u32 data;
2443 u32 active_rbs = 0;
2444 u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
2445 adev->gfx.config.max_sh_per_se;
2446
2447 mutex_lock(&adev->grbm_idx_mutex);
2448 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2449 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2450 gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
2451 data = gfx_v9_0_get_rb_active_bitmap(adev);
2452 active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
2453 rb_bitmap_width_per_sh);
2454 }
2455 }
2456 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
2457 mutex_unlock(&adev->grbm_idx_mutex);
2458
2459 adev->gfx.config.backend_enable_mask = active_rbs;
2460 adev->gfx.config.num_rbs = hweight32(active_rbs);
2461 }
2462
2463 #define DEFAULT_SH_MEM_BASES (0x6000)
2464 #define FIRST_COMPUTE_VMID (8)
2465 #define LAST_COMPUTE_VMID (16)
2466 static void gfx_v9_0_init_compute_vmid(struct amdgpu_device *adev)
2467 {
2468 int i;
2469 uint32_t sh_mem_config;
2470 uint32_t sh_mem_bases;
2471
2472 /*
2473 * Configure apertures:
2474 * LDS: 0x60000000'00000000 - 0x60000001'00000000 (4GB)
2475 * Scratch: 0x60000001'00000000 - 0x60000002'00000000 (4GB)
2476 * GPUVM: 0x60010000'00000000 - 0x60020000'00000000 (1TB)
2477 */
2478 sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
2479
2480 sh_mem_config = SH_MEM_ADDRESS_MODE_64 |
2481 SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
2482 SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
2483
2484 mutex_lock(&adev->srbm_mutex);
2485 for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
2486 soc15_grbm_select(adev, 0, 0, 0, i);
2487 /* CP and shaders */
2488 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, sh_mem_config);
2489 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, sh_mem_bases);
2490 }
2491 soc15_grbm_select(adev, 0, 0, 0, 0);
2492 mutex_unlock(&adev->srbm_mutex);
2493
2494 /* Initialize all compute VMIDs to have no GDS, GWS, or OA
2495 acccess. These should be enabled by FW for target VMIDs. */
2496 for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
2497 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * i, 0);
2498 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * i, 0);
2499 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, i, 0);
2500 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, i, 0);
2501 }
2502 }
2503
2504 static void gfx_v9_0_init_gds_vmid(struct amdgpu_device *adev)
2505 {
2506 int vmid;
2507
2508 /*
2509 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
2510 * access. Compute VMIDs should be enabled by FW for target VMIDs,
2511 * the driver can enable them for graphics. VMID0 should maintain
2512 * access so that HWS firmware can save/restore entries.
2513 */
2514 for (vmid = 1; vmid < 16; vmid++) {
2515 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * vmid, 0);
2516 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * vmid, 0);
2517 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, vmid, 0);
2518 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, vmid, 0);
2519 }
2520 }
2521
2522 static void gfx_v9_0_init_sq_config(struct amdgpu_device *adev)
2523 {
2524 uint32_t tmp;
2525
2526 switch (adev->asic_type) {
2527 case CHIP_ARCTURUS:
2528 tmp = RREG32_SOC15(GC, 0, mmSQ_CONFIG);
2529 tmp = REG_SET_FIELD(tmp, SQ_CONFIG,
2530 DISABLE_BARRIER_WAITCNT, 1);
2531 WREG32_SOC15(GC, 0, mmSQ_CONFIG, tmp);
2532 break;
2533 default:
2534 break;
2535 };
2536 }
2537
2538 static void gfx_v9_0_constants_init(struct amdgpu_device *adev)
2539 {
2540 u32 tmp;
2541 int i;
2542
2543 WREG32_FIELD15_RLC(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff);
2544
2545 gfx_v9_0_tiling_mode_table_init(adev);
2546
2547 gfx_v9_0_setup_rb(adev);
2548 gfx_v9_0_get_cu_info(adev, &adev->gfx.cu_info);
2549 adev->gfx.config.db_debug2 = RREG32_SOC15(GC, 0, mmDB_DEBUG2);
2550
2551 /* XXX SH_MEM regs */
2552 /* where to put LDS, scratch, GPUVM in FSA64 space */
2553 mutex_lock(&adev->srbm_mutex);
2554 for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_0].num_ids; i++) {
2555 soc15_grbm_select(adev, 0, 0, 0, i);
2556 /* CP and shaders */
2557 if (i == 0) {
2558 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
2559 SH_MEM_ALIGNMENT_MODE_UNALIGNED);
2560 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
2561 !!amdgpu_noretry);
2562 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp);
2563 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, 0);
2564 } else {
2565 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
2566 SH_MEM_ALIGNMENT_MODE_UNALIGNED);
2567 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
2568 !!amdgpu_noretry);
2569 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp);
2570 tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
2571 (adev->gmc.private_aperture_start >> 48));
2572 tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
2573 (adev->gmc.shared_aperture_start >> 48));
2574 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, tmp);
2575 }
2576 }
2577 soc15_grbm_select(adev, 0, 0, 0, 0);
2578
2579 mutex_unlock(&adev->srbm_mutex);
2580
2581 gfx_v9_0_init_compute_vmid(adev);
2582 gfx_v9_0_init_gds_vmid(adev);
2583 gfx_v9_0_init_sq_config(adev);
2584 }
2585
2586 static void gfx_v9_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
2587 {
2588 u32 i, j, k;
2589 u32 mask;
2590
2591 mutex_lock(&adev->grbm_idx_mutex);
2592 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2593 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2594 gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
2595 for (k = 0; k < adev->usec_timeout; k++) {
2596 if (RREG32_SOC15(GC, 0, mmRLC_SERDES_CU_MASTER_BUSY) == 0)
2597 break;
2598 udelay(1);
2599 }
2600 if (k == adev->usec_timeout) {
2601 gfx_v9_0_select_se_sh(adev, 0xffffffff,
2602 0xffffffff, 0xffffffff);
2603 mutex_unlock(&adev->grbm_idx_mutex);
2604 DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
2605 i, j);
2606 return;
2607 }
2608 }
2609 }
2610 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
2611 mutex_unlock(&adev->grbm_idx_mutex);
2612
2613 mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
2614 RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
2615 RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
2616 RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
2617 for (k = 0; k < adev->usec_timeout; k++) {
2618 if ((RREG32_SOC15(GC, 0, mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
2619 break;
2620 udelay(1);
2621 }
2622 }
2623
2624 static void gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
2625 bool enable)
2626 {
2627 u32 tmp = RREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0);
2628
2629 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
2630 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
2631 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
2632 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
2633
2634 WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp);
2635 }
2636
2637 static void gfx_v9_0_init_csb(struct amdgpu_device *adev)
2638 {
2639 adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
2640 /* csib */
2641 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_HI),
2642 adev->gfx.rlc.clear_state_gpu_addr >> 32);
2643 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_LO),
2644 adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
2645 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_LENGTH),
2646 adev->gfx.rlc.clear_state_size);
2647 }
2648
2649 static void gfx_v9_1_parse_ind_reg_list(int *register_list_format,
2650 int indirect_offset,
2651 int list_size,
2652 int *unique_indirect_regs,
2653 int unique_indirect_reg_count,
2654 int *indirect_start_offsets,
2655 int *indirect_start_offsets_count,
2656 int max_start_offsets_count)
2657 {
2658 int idx;
2659
2660 for (; indirect_offset < list_size; indirect_offset++) {
2661 WARN_ON(*indirect_start_offsets_count >= max_start_offsets_count);
2662 indirect_start_offsets[*indirect_start_offsets_count] = indirect_offset;
2663 *indirect_start_offsets_count = *indirect_start_offsets_count + 1;
2664
2665 while (register_list_format[indirect_offset] != 0xFFFFFFFF) {
2666 indirect_offset += 2;
2667
2668 /* look for the matching indice */
2669 for (idx = 0; idx < unique_indirect_reg_count; idx++) {
2670 if (unique_indirect_regs[idx] ==
2671 register_list_format[indirect_offset] ||
2672 !unique_indirect_regs[idx])
2673 break;
2674 }
2675
2676 BUG_ON(idx >= unique_indirect_reg_count);
2677
2678 if (!unique_indirect_regs[idx])
2679 unique_indirect_regs[idx] = register_list_format[indirect_offset];
2680
2681 indirect_offset++;
2682 }
2683 }
2684 }
2685
2686 static int gfx_v9_1_init_rlc_save_restore_list(struct amdgpu_device *adev)
2687 {
2688 int unique_indirect_regs[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2689 int unique_indirect_reg_count = 0;
2690
2691 int indirect_start_offsets[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2692 int indirect_start_offsets_count = 0;
2693
2694 int list_size = 0;
2695 int i = 0, j = 0;
2696 u32 tmp = 0;
2697
2698 u32 *register_list_format =
2699 kmemdup(adev->gfx.rlc.register_list_format,
2700 adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
2701 if (!register_list_format)
2702 return -ENOMEM;
2703
2704 /* setup unique_indirect_regs array and indirect_start_offsets array */
2705 unique_indirect_reg_count = ARRAY_SIZE(unique_indirect_regs);
2706 gfx_v9_1_parse_ind_reg_list(register_list_format,
2707 adev->gfx.rlc.reg_list_format_direct_reg_list_length,
2708 adev->gfx.rlc.reg_list_format_size_bytes >> 2,
2709 unique_indirect_regs,
2710 unique_indirect_reg_count,
2711 indirect_start_offsets,
2712 &indirect_start_offsets_count,
2713 ARRAY_SIZE(indirect_start_offsets));
2714
2715 /* enable auto inc in case it is disabled */
2716 tmp = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL));
2717 tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK;
2718 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL), tmp);
2719
2720 /* write register_restore table to offset 0x0 using RLC_SRM_ARAM_ADDR/DATA */
2721 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_ADDR),
2722 RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET);
2723 for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
2724 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_DATA),
2725 adev->gfx.rlc.register_restore[i]);
2726
2727 /* load indirect register */
2728 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2729 adev->gfx.rlc.reg_list_format_start);
2730
2731 /* direct register portion */
2732 for (i = 0; i < adev->gfx.rlc.reg_list_format_direct_reg_list_length; i++)
2733 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
2734 register_list_format[i]);
2735
2736 /* indirect register portion */
2737 while (i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2)) {
2738 if (register_list_format[i] == 0xFFFFFFFF) {
2739 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2740 continue;
2741 }
2742
2743 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2744 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2745
2746 for (j = 0; j < unique_indirect_reg_count; j++) {
2747 if (register_list_format[i] == unique_indirect_regs[j]) {
2748 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, j);
2749 break;
2750 }
2751 }
2752
2753 BUG_ON(j >= unique_indirect_reg_count);
2754
2755 i++;
2756 }
2757
2758 /* set save/restore list size */
2759 list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
2760 list_size = list_size >> 1;
2761 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2762 adev->gfx.rlc.reg_restore_list_size);
2763 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), list_size);
2764
2765 /* write the starting offsets to RLC scratch ram */
2766 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2767 adev->gfx.rlc.starting_offsets_start);
2768 for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
2769 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
2770 indirect_start_offsets[i]);
2771
2772 /* load unique indirect regs*/
2773 for (i = 0; i < ARRAY_SIZE(unique_indirect_regs); i++) {
2774 if (unique_indirect_regs[i] != 0) {
2775 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_ADDR_0)
2776 + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i],
2777 unique_indirect_regs[i] & 0x3FFFF);
2778
2779 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_DATA_0)
2780 + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i],
2781 unique_indirect_regs[i] >> 20);
2782 }
2783 }
2784
2785 kfree(register_list_format);
2786 return 0;
2787 }
2788
2789 static void gfx_v9_0_enable_save_restore_machine(struct amdgpu_device *adev)
2790 {
2791 WREG32_FIELD15(GC, 0, RLC_SRM_CNTL, SRM_ENABLE, 1);
2792 }
2793
2794 static void pwr_10_0_gfxip_control_over_cgpg(struct amdgpu_device *adev,
2795 bool enable)
2796 {
2797 uint32_t data = 0;
2798 uint32_t default_data = 0;
2799
2800 default_data = data = RREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS));
2801 if (enable == true) {
2802 /* enable GFXIP control over CGPG */
2803 data |= PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
2804 if(default_data != data)
2805 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2806
2807 /* update status */
2808 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK;
2809 data |= (2 << PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT);
2810 if(default_data != data)
2811 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2812 } else {
2813 /* restore GFXIP control over GCPG */
2814 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
2815 if(default_data != data)
2816 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2817 }
2818 }
2819
2820 static void gfx_v9_0_init_gfx_power_gating(struct amdgpu_device *adev)
2821 {
2822 uint32_t data = 0;
2823
2824 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
2825 AMD_PG_SUPPORT_GFX_SMG |
2826 AMD_PG_SUPPORT_GFX_DMG)) {
2827 /* init IDLE_POLL_COUNT = 60 */
2828 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL));
2829 data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
2830 data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
2831 WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL), data);
2832
2833 /* init RLC PG Delay */
2834 data = 0;
2835 data |= (0x10 << RLC_PG_DELAY__POWER_UP_DELAY__SHIFT);
2836 data |= (0x10 << RLC_PG_DELAY__POWER_DOWN_DELAY__SHIFT);
2837 data |= (0x10 << RLC_PG_DELAY__CMD_PROPAGATE_DELAY__SHIFT);
2838 data |= (0x40 << RLC_PG_DELAY__MEM_SLEEP_DELAY__SHIFT);
2839 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY), data);
2840
2841 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2));
2842 data &= ~RLC_PG_DELAY_2__SERDES_CMD_DELAY_MASK;
2843 data |= (0x4 << RLC_PG_DELAY_2__SERDES_CMD_DELAY__SHIFT);
2844 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2), data);
2845
2846 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3));
2847 data &= ~RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG_MASK;
2848 data |= (0xff << RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG__SHIFT);
2849 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3), data);
2850
2851 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL));
2852 data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
2853
2854 /* program GRBM_REG_SAVE_GFX_IDLE_THRESHOLD to 0x55f0 */
2855 data |= (0x55f0 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
2856 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL), data);
2857
2858 pwr_10_0_gfxip_control_over_cgpg(adev, true);
2859 }
2860 }
2861
2862 static void gfx_v9_0_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
2863 bool enable)
2864 {
2865 uint32_t data = 0;
2866 uint32_t default_data = 0;
2867
2868 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2869 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2870 SMU_CLK_SLOWDOWN_ON_PU_ENABLE,
2871 enable ? 1 : 0);
2872 if (default_data != data)
2873 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2874 }
2875
2876 static void gfx_v9_0_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
2877 bool enable)
2878 {
2879 uint32_t data = 0;
2880 uint32_t default_data = 0;
2881
2882 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2883 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2884 SMU_CLK_SLOWDOWN_ON_PD_ENABLE,
2885 enable ? 1 : 0);
2886 if(default_data != data)
2887 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2888 }
2889
2890 static void gfx_v9_0_enable_cp_power_gating(struct amdgpu_device *adev,
2891 bool enable)
2892 {
2893 uint32_t data = 0;
2894 uint32_t default_data = 0;
2895
2896 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2897 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2898 CP_PG_DISABLE,
2899 enable ? 0 : 1);
2900 if(default_data != data)
2901 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2902 }
2903
2904 static void gfx_v9_0_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
2905 bool enable)
2906 {
2907 uint32_t data, default_data;
2908
2909 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2910 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2911 GFX_POWER_GATING_ENABLE,
2912 enable ? 1 : 0);
2913 if(default_data != data)
2914 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2915 }
2916
2917 static void gfx_v9_0_enable_gfx_pipeline_powergating(struct amdgpu_device *adev,
2918 bool enable)
2919 {
2920 uint32_t data, default_data;
2921
2922 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2923 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2924 GFX_PIPELINE_PG_ENABLE,
2925 enable ? 1 : 0);
2926 if(default_data != data)
2927 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2928
2929 if (!enable)
2930 /* read any GFX register to wake up GFX */
2931 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmDB_RENDER_CONTROL));
2932 }
2933
2934 static void gfx_v9_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
2935 bool enable)
2936 {
2937 uint32_t data, default_data;
2938
2939 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2940 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2941 STATIC_PER_CU_PG_ENABLE,
2942 enable ? 1 : 0);
2943 if(default_data != data)
2944 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2945 }
2946
2947 static void gfx_v9_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
2948 bool enable)
2949 {
2950 uint32_t data, default_data;
2951
2952 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2953 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2954 DYN_PER_CU_PG_ENABLE,
2955 enable ? 1 : 0);
2956 if(default_data != data)
2957 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2958 }
2959
2960 static void gfx_v9_0_init_pg(struct amdgpu_device *adev)
2961 {
2962 gfx_v9_0_init_csb(adev);
2963
2964 /*
2965 * Rlc save restore list is workable since v2_1.
2966 * And it's needed by gfxoff feature.
2967 */
2968 if (adev->gfx.rlc.is_rlc_v2_1) {
2969 if (adev->asic_type == CHIP_VEGA12 ||
2970 (adev->asic_type == CHIP_RAVEN &&
2971 adev->rev_id >= 8))
2972 gfx_v9_1_init_rlc_save_restore_list(adev);
2973 gfx_v9_0_enable_save_restore_machine(adev);
2974 }
2975
2976 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
2977 AMD_PG_SUPPORT_GFX_SMG |
2978 AMD_PG_SUPPORT_GFX_DMG |
2979 AMD_PG_SUPPORT_CP |
2980 AMD_PG_SUPPORT_GDS |
2981 AMD_PG_SUPPORT_RLC_SMU_HS)) {
2982 WREG32(mmRLC_JUMP_TABLE_RESTORE,
2983 adev->gfx.rlc.cp_table_gpu_addr >> 8);
2984 gfx_v9_0_init_gfx_power_gating(adev);
2985 }
2986 }
2987
2988 void gfx_v9_0_rlc_stop(struct amdgpu_device *adev)
2989 {
2990 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 0);
2991 gfx_v9_0_enable_gui_idle_interrupt(adev, false);
2992 gfx_v9_0_wait_for_rlc_serdes(adev);
2993 }
2994
2995 static void gfx_v9_0_rlc_reset(struct amdgpu_device *adev)
2996 {
2997 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
2998 udelay(50);
2999 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
3000 udelay(50);
3001 }
3002
3003 static void gfx_v9_0_rlc_start(struct amdgpu_device *adev)
3004 {
3005 #ifdef AMDGPU_RLC_DEBUG_RETRY
3006 u32 rlc_ucode_ver;
3007 #endif
3008
3009 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1);
3010 udelay(50);
3011
3012 /* carrizo do enable cp interrupt after cp inited */
3013 if (!(adev->flags & AMD_IS_APU)) {
3014 gfx_v9_0_enable_gui_idle_interrupt(adev, true);
3015 udelay(50);
3016 }
3017
3018 #ifdef AMDGPU_RLC_DEBUG_RETRY
3019 /* RLC_GPM_GENERAL_6 : RLC Ucode version */
3020 rlc_ucode_ver = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_6);
3021 if(rlc_ucode_ver == 0x108) {
3022 DRM_INFO("Using rlc debug ucode. mmRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n",
3023 rlc_ucode_ver, adev->gfx.rlc_fw_version);
3024 /* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles,
3025 * default is 0x9C4 to create a 100us interval */
3026 WREG32_SOC15(GC, 0, mmRLC_GPM_TIMER_INT_3, 0x9C4);
3027 /* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr
3028 * to disable the page fault retry interrupts, default is
3029 * 0x100 (256) */
3030 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_12, 0x100);
3031 }
3032 #endif
3033 }
3034
3035 static int gfx_v9_0_rlc_load_microcode(struct amdgpu_device *adev)
3036 {
3037 const struct rlc_firmware_header_v2_0 *hdr;
3038 const __le32 *fw_data;
3039 unsigned i, fw_size;
3040
3041 if (!adev->gfx.rlc_fw)
3042 return -EINVAL;
3043
3044 hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
3045 amdgpu_ucode_print_rlc_hdr(&hdr->header);
3046
3047 fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
3048 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
3049 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
3050
3051 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR,
3052 RLCG_UCODE_LOADING_START_ADDRESS);
3053 for (i = 0; i < fw_size; i++)
3054 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
3055 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
3056
3057 return 0;
3058 }
3059
3060 static int gfx_v9_0_rlc_resume(struct amdgpu_device *adev)
3061 {
3062 int r;
3063
3064 if (amdgpu_sriov_vf(adev)) {
3065 gfx_v9_0_init_csb(adev);
3066 return 0;
3067 }
3068
3069 adev->gfx.rlc.funcs->stop(adev);
3070
3071 /* disable CG */
3072 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, 0);
3073
3074 gfx_v9_0_init_pg(adev);
3075
3076 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
3077 /* legacy rlc firmware loading */
3078 r = gfx_v9_0_rlc_load_microcode(adev);
3079 if (r)
3080 return r;
3081 }
3082
3083 switch (adev->asic_type) {
3084 case CHIP_RAVEN:
3085 if (amdgpu_lbpw == 0)
3086 gfx_v9_0_enable_lbpw(adev, false);
3087 else
3088 gfx_v9_0_enable_lbpw(adev, true);
3089 break;
3090 case CHIP_VEGA20:
3091 if (amdgpu_lbpw > 0)
3092 gfx_v9_0_enable_lbpw(adev, true);
3093 else
3094 gfx_v9_0_enable_lbpw(adev, false);
3095 break;
3096 default:
3097 break;
3098 }
3099
3100 adev->gfx.rlc.funcs->start(adev);
3101
3102 return 0;
3103 }
3104
3105 static void gfx_v9_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
3106 {
3107 int i;
3108 u32 tmp = RREG32_SOC15(GC, 0, mmCP_ME_CNTL);
3109
3110 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1);
3111 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1);
3112 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, enable ? 0 : 1);
3113 if (!enable) {
3114 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
3115 adev->gfx.gfx_ring[i].sched.ready = false;
3116 }
3117 WREG32_SOC15_RLC(GC, 0, mmCP_ME_CNTL, tmp);
3118 udelay(50);
3119 }
3120
3121 static int gfx_v9_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
3122 {
3123 const struct gfx_firmware_header_v1_0 *pfp_hdr;
3124 const struct gfx_firmware_header_v1_0 *ce_hdr;
3125 const struct gfx_firmware_header_v1_0 *me_hdr;
3126 const __le32 *fw_data;
3127 unsigned i, fw_size;
3128
3129 if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
3130 return -EINVAL;
3131
3132 pfp_hdr = (const struct gfx_firmware_header_v1_0 *)
3133 adev->gfx.pfp_fw->data;
3134 ce_hdr = (const struct gfx_firmware_header_v1_0 *)
3135 adev->gfx.ce_fw->data;
3136 me_hdr = (const struct gfx_firmware_header_v1_0 *)
3137 adev->gfx.me_fw->data;
3138
3139 amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
3140 amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
3141 amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
3142
3143 gfx_v9_0_cp_gfx_enable(adev, false);
3144
3145 /* PFP */
3146 fw_data = (const __le32 *)
3147 (adev->gfx.pfp_fw->data +
3148 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
3149 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
3150 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, 0);
3151 for (i = 0; i < fw_size; i++)
3152 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
3153 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
3154
3155 /* CE */
3156 fw_data = (const __le32 *)
3157 (adev->gfx.ce_fw->data +
3158 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
3159 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
3160 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, 0);
3161 for (i = 0; i < fw_size; i++)
3162 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
3163 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
3164
3165 /* ME */
3166 fw_data = (const __le32 *)
3167 (adev->gfx.me_fw->data +
3168 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
3169 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
3170 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, 0);
3171 for (i = 0; i < fw_size; i++)
3172 WREG32_SOC15(GC, 0, mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
3173 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
3174
3175 return 0;
3176 }
3177
3178 static int gfx_v9_0_cp_gfx_start(struct amdgpu_device *adev)
3179 {
3180 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
3181 const struct cs_section_def *sect = NULL;
3182 const struct cs_extent_def *ext = NULL;
3183 int r, i, tmp;
3184
3185 /* init the CP */
3186 WREG32_SOC15(GC, 0, mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
3187 WREG32_SOC15(GC, 0, mmCP_DEVICE_ID, 1);
3188
3189 gfx_v9_0_cp_gfx_enable(adev, true);
3190
3191 r = amdgpu_ring_alloc(ring, gfx_v9_0_get_csb_size(adev) + 4 + 3);
3192 if (r) {
3193 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
3194 return r;
3195 }
3196
3197 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3198 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3199
3200 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3201 amdgpu_ring_write(ring, 0x80000000);
3202 amdgpu_ring_write(ring, 0x80000000);
3203
3204 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
3205 for (ext = sect->section; ext->extent != NULL; ++ext) {
3206 if (sect->id == SECT_CONTEXT) {
3207 amdgpu_ring_write(ring,
3208 PACKET3(PACKET3_SET_CONTEXT_REG,
3209 ext->reg_count));
3210 amdgpu_ring_write(ring,
3211 ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
3212 for (i = 0; i < ext->reg_count; i++)
3213 amdgpu_ring_write(ring, ext->extent[i]);
3214 }
3215 }
3216 }
3217
3218 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3219 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
3220
3221 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3222 amdgpu_ring_write(ring, 0);
3223
3224 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
3225 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
3226 amdgpu_ring_write(ring, 0x8000);
3227 amdgpu_ring_write(ring, 0x8000);
3228
3229 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG,1));
3230 tmp = (PACKET3_SET_UCONFIG_REG_INDEX_TYPE |
3231 (SOC15_REG_OFFSET(GC, 0, mmVGT_INDEX_TYPE) - PACKET3_SET_UCONFIG_REG_START));
3232 amdgpu_ring_write(ring, tmp);
3233 amdgpu_ring_write(ring, 0);
3234
3235 amdgpu_ring_commit(ring);
3236
3237 return 0;
3238 }
3239
3240 static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev)
3241 {
3242 struct amdgpu_ring *ring;
3243 u32 tmp;
3244 u32 rb_bufsz;
3245 u64 rb_addr, rptr_addr, wptr_gpu_addr;
3246
3247 /* Set the write pointer delay */
3248 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_DELAY, 0);
3249
3250 /* set the RB to use vmid 0 */
3251 WREG32_SOC15(GC, 0, mmCP_RB_VMID, 0);
3252
3253 /* Set ring buffer size */
3254 ring = &adev->gfx.gfx_ring[0];
3255 rb_bufsz = order_base_2(ring->ring_size / 8);
3256 tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
3257 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
3258 #ifdef __BIG_ENDIAN
3259 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
3260 #endif
3261 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3262
3263 /* Initialize the ring buffer's write pointers */
3264 ring->wptr = 0;
3265 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
3266 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
3267
3268 /* set the wb address wether it's enabled or not */
3269 rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
3270 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
3271 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3272
3273 wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
3274 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
3275 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
3276
3277 mdelay(1);
3278 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3279
3280 rb_addr = ring->gpu_addr >> 8;
3281 WREG32_SOC15(GC, 0, mmCP_RB0_BASE, rb_addr);
3282 WREG32_SOC15(GC, 0, mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
3283
3284 tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL);
3285 if (ring->use_doorbell) {
3286 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3287 DOORBELL_OFFSET, ring->doorbell_index);
3288 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3289 DOORBELL_EN, 1);
3290 } else {
3291 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
3292 }
3293 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, tmp);
3294
3295 tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
3296 DOORBELL_RANGE_LOWER, ring->doorbell_index);
3297 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
3298
3299 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER,
3300 CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
3301
3302
3303 /* start the ring */
3304 gfx_v9_0_cp_gfx_start(adev);
3305 ring->sched.ready = true;
3306
3307 return 0;
3308 }
3309
3310 static void gfx_v9_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
3311 {
3312 int i;
3313
3314 if (enable) {
3315 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL, 0);
3316 } else {
3317 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL,
3318 (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
3319 for (i = 0; i < adev->gfx.num_compute_rings; i++)
3320 adev->gfx.compute_ring[i].sched.ready = false;
3321 adev->gfx.kiq.ring.sched.ready = false;
3322 }
3323 udelay(50);
3324 }
3325
3326 static int gfx_v9_0_cp_compute_load_microcode(struct amdgpu_device *adev)
3327 {
3328 const struct gfx_firmware_header_v1_0 *mec_hdr;
3329 const __le32 *fw_data;
3330 unsigned i;
3331 u32 tmp;
3332
3333 if (!adev->gfx.mec_fw)
3334 return -EINVAL;
3335
3336 gfx_v9_0_cp_compute_enable(adev, false);
3337
3338 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3339 amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3340
3341 fw_data = (const __le32 *)
3342 (adev->gfx.mec_fw->data +
3343 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
3344 tmp = 0;
3345 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
3346 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
3347 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL, tmp);
3348
3349 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO,
3350 adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000);
3351 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI,
3352 upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
3353
3354 /* MEC1 */
3355 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3356 mec_hdr->jt_offset);
3357 for (i = 0; i < mec_hdr->jt_size; i++)
3358 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_DATA,
3359 le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
3360
3361 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3362 adev->gfx.mec_fw_version);
3363 /* Todo : Loading MEC2 firmware is only necessary if MEC2 should run different microcode than MEC1. */
3364
3365 return 0;
3366 }
3367
3368 /* KIQ functions */
3369 static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)
3370 {
3371 uint32_t tmp;
3372 struct amdgpu_device *adev = ring->adev;
3373
3374 /* tell RLC which is KIQ queue */
3375 tmp = RREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS);
3376 tmp &= 0xffffff00;
3377 tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
3378 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3379 tmp |= 0x80;
3380 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3381 }
3382
3383 static void gfx_v9_0_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd)
3384 {
3385 struct amdgpu_device *adev = ring->adev;
3386
3387 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
3388 if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring->queue)) {
3389 mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
3390 ring->has_high_prio = true;
3391 mqd->cp_hqd_queue_priority =
3392 AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
3393 } else {
3394 ring->has_high_prio = false;
3395 }
3396 }
3397 }
3398
3399 static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring)
3400 {
3401 struct amdgpu_device *adev = ring->adev;
3402 struct v9_mqd *mqd = ring->mqd_ptr;
3403 uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
3404 uint32_t tmp;
3405
3406 mqd->header = 0xC0310800;
3407 mqd->compute_pipelinestat_enable = 0x00000001;
3408 mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
3409 mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
3410 mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
3411 mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
3412 mqd->compute_static_thread_mgmt_se4 = 0xffffffff;
3413 mqd->compute_static_thread_mgmt_se5 = 0xffffffff;
3414 mqd->compute_static_thread_mgmt_se6 = 0xffffffff;
3415 mqd->compute_static_thread_mgmt_se7 = 0xffffffff;
3416 mqd->compute_misc_reserved = 0x00000003;
3417
3418 mqd->dynamic_cu_mask_addr_lo =
3419 lower_32_bits(ring->mqd_gpu_addr
3420 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3421 mqd->dynamic_cu_mask_addr_hi =
3422 upper_32_bits(ring->mqd_gpu_addr
3423 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3424
3425 eop_base_addr = ring->eop_gpu_addr >> 8;
3426 mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
3427 mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
3428
3429 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3430 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL);
3431 tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
3432 (order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1));
3433
3434 mqd->cp_hqd_eop_control = tmp;
3435
3436 /* enable doorbell? */
3437 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
3438
3439 if (ring->use_doorbell) {
3440 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3441 DOORBELL_OFFSET, ring->doorbell_index);
3442 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3443 DOORBELL_EN, 1);
3444 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3445 DOORBELL_SOURCE, 0);
3446 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3447 DOORBELL_HIT, 0);
3448 } else {
3449 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3450 DOORBELL_EN, 0);
3451 }
3452
3453 mqd->cp_hqd_pq_doorbell_control = tmp;
3454
3455 /* disable the queue if it's active */
3456 ring->wptr = 0;
3457 mqd->cp_hqd_dequeue_request = 0;
3458 mqd->cp_hqd_pq_rptr = 0;
3459 mqd->cp_hqd_pq_wptr_lo = 0;
3460 mqd->cp_hqd_pq_wptr_hi = 0;
3461
3462 /* set the pointer to the MQD */
3463 mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
3464 mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
3465
3466 /* set MQD vmid to 0 */
3467 tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL);
3468 tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
3469 mqd->cp_mqd_control = tmp;
3470
3471 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3472 hqd_gpu_addr = ring->gpu_addr >> 8;
3473 mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
3474 mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3475
3476 /* set up the HQD, this is similar to CP_RB0_CNTL */
3477 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL);
3478 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
3479 (order_base_2(ring->ring_size / 4) - 1));
3480 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
3481 ((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8));
3482 #ifdef __BIG_ENDIAN
3483 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
3484 #endif
3485 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
3486 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
3487 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
3488 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
3489 mqd->cp_hqd_pq_control = tmp;
3490
3491 /* set the wb address whether it's enabled or not */
3492 wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
3493 mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
3494 mqd->cp_hqd_pq_rptr_report_addr_hi =
3495 upper_32_bits(wb_gpu_addr) & 0xffff;
3496
3497 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3498 wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
3499 mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3500 mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3501
3502 tmp = 0;
3503 /* enable the doorbell if requested */
3504 if (ring->use_doorbell) {
3505 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
3506 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3507 DOORBELL_OFFSET, ring->doorbell_index);
3508
3509 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3510 DOORBELL_EN, 1);
3511 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3512 DOORBELL_SOURCE, 0);
3513 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3514 DOORBELL_HIT, 0);
3515 }
3516
3517 mqd->cp_hqd_pq_doorbell_control = tmp;
3518
3519 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3520 ring->wptr = 0;
3521 mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR);
3522
3523 /* set the vmid for the queue */
3524 mqd->cp_hqd_vmid = 0;
3525
3526 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE);
3527 tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
3528 mqd->cp_hqd_persistent_state = tmp;
3529
3530 /* set MIN_IB_AVAIL_SIZE */
3531 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL);
3532 tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
3533 mqd->cp_hqd_ib_control = tmp;
3534
3535 /* set static priority for a queue/ring */
3536 gfx_v9_0_mqd_set_priority(ring, mqd);
3537 mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
3538
3539 /* map_queues packet doesn't need activate the queue,
3540 * so only kiq need set this field.
3541 */
3542 if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
3543 mqd->cp_hqd_active = 1;
3544
3545 return 0;
3546 }
3547
3548 static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring)
3549 {
3550 struct amdgpu_device *adev = ring->adev;
3551 struct v9_mqd *mqd = ring->mqd_ptr;
3552 int j;
3553
3554 /* disable wptr polling */
3555 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3556
3557 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR,
3558 mqd->cp_hqd_eop_base_addr_lo);
3559 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI,
3560 mqd->cp_hqd_eop_base_addr_hi);
3561
3562 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3563 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_CONTROL,
3564 mqd->cp_hqd_eop_control);
3565
3566 /* enable doorbell? */
3567 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3568 mqd->cp_hqd_pq_doorbell_control);
3569
3570 /* disable the queue if it's active */
3571 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3572 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3573 for (j = 0; j < adev->usec_timeout; j++) {
3574 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3575 break;
3576 udelay(1);
3577 }
3578 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3579 mqd->cp_hqd_dequeue_request);
3580 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR,
3581 mqd->cp_hqd_pq_rptr);
3582 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3583 mqd->cp_hqd_pq_wptr_lo);
3584 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3585 mqd->cp_hqd_pq_wptr_hi);
3586 }
3587
3588 /* set the pointer to the MQD */
3589 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR,
3590 mqd->cp_mqd_base_addr_lo);
3591 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR_HI,
3592 mqd->cp_mqd_base_addr_hi);
3593
3594 /* set MQD vmid to 0 */
3595 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_CONTROL,
3596 mqd->cp_mqd_control);
3597
3598 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3599 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE,
3600 mqd->cp_hqd_pq_base_lo);
3601 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE_HI,
3602 mqd->cp_hqd_pq_base_hi);
3603
3604 /* set up the HQD, this is similar to CP_RB0_CNTL */
3605 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_CONTROL,
3606 mqd->cp_hqd_pq_control);
3607
3608 /* set the wb address whether it's enabled or not */
3609 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR,
3610 mqd->cp_hqd_pq_rptr_report_addr_lo);
3611 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3612 mqd->cp_hqd_pq_rptr_report_addr_hi);
3613
3614 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3615 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR,
3616 mqd->cp_hqd_pq_wptr_poll_addr_lo);
3617 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
3618 mqd->cp_hqd_pq_wptr_poll_addr_hi);
3619
3620 /* enable the doorbell if requested */
3621 if (ring->use_doorbell) {
3622 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER,
3623 (adev->doorbell_index.kiq * 2) << 2);
3624 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
3625 (adev->doorbell_index.userqueue_end * 2) << 2);
3626 }
3627
3628 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3629 mqd->cp_hqd_pq_doorbell_control);
3630
3631 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3632 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3633 mqd->cp_hqd_pq_wptr_lo);
3634 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3635 mqd->cp_hqd_pq_wptr_hi);
3636
3637 /* set the vmid for the queue */
3638 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_VMID, mqd->cp_hqd_vmid);
3639
3640 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE,
3641 mqd->cp_hqd_persistent_state);
3642
3643 /* activate the queue */
3644 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE,
3645 mqd->cp_hqd_active);
3646
3647 if (ring->use_doorbell)
3648 WREG32_FIELD15(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
3649
3650 return 0;
3651 }
3652
3653 static int gfx_v9_0_kiq_fini_register(struct amdgpu_ring *ring)
3654 {
3655 struct amdgpu_device *adev = ring->adev;
3656 int j;
3657
3658 /* disable the queue if it's active */
3659 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3660
3661 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3662
3663 for (j = 0; j < adev->usec_timeout; j++) {
3664 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3665 break;
3666 udelay(1);
3667 }
3668
3669 if (j == AMDGPU_MAX_USEC_TIMEOUT) {
3670 DRM_DEBUG("KIQ dequeue request failed.\n");
3671
3672 /* Manual disable if dequeue request times out */
3673 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE, 0);
3674 }
3675
3676 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3677 0);
3678 }
3679
3680 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IQ_TIMER, 0);
3681 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IB_CONTROL, 0);
3682 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE, 0);
3683 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000);
3684 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0);
3685 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR, 0);
3686 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 0);
3687 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 0);
3688
3689 return 0;
3690 }
3691
3692 static int gfx_v9_0_kiq_init_queue(struct amdgpu_ring *ring)
3693 {
3694 struct amdgpu_device *adev = ring->adev;
3695 struct v9_mqd *mqd = ring->mqd_ptr;
3696 int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
3697
3698 gfx_v9_0_kiq_setting(ring);
3699
3700 if (adev->in_gpu_reset) { /* for GPU_RESET case */
3701 /* reset MQD to a clean status */
3702 if (adev->gfx.mec.mqd_backup[mqd_idx])
3703 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
3704
3705 /* reset ring buffer */
3706 ring->wptr = 0;
3707 amdgpu_ring_clear_ring(ring);
3708
3709 mutex_lock(&adev->srbm_mutex);
3710 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3711 gfx_v9_0_kiq_init_register(ring);
3712 soc15_grbm_select(adev, 0, 0, 0, 0);
3713 mutex_unlock(&adev->srbm_mutex);
3714 } else {
3715 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3716 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3717 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3718 mutex_lock(&adev->srbm_mutex);
3719 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3720 gfx_v9_0_mqd_init(ring);
3721 gfx_v9_0_kiq_init_register(ring);
3722 soc15_grbm_select(adev, 0, 0, 0, 0);
3723 mutex_unlock(&adev->srbm_mutex);
3724
3725 if (adev->gfx.mec.mqd_backup[mqd_idx])
3726 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
3727 }
3728
3729 return 0;
3730 }
3731
3732 static int gfx_v9_0_kcq_init_queue(struct amdgpu_ring *ring)
3733 {
3734 struct amdgpu_device *adev = ring->adev;
3735 struct v9_mqd *mqd = ring->mqd_ptr;
3736 int mqd_idx = ring - &adev->gfx.compute_ring[0];
3737
3738 if (!adev->in_gpu_reset && !adev->in_suspend) {
3739 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3740 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3741 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3742 mutex_lock(&adev->srbm_mutex);
3743 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3744 gfx_v9_0_mqd_init(ring);
3745 soc15_grbm_select(adev, 0, 0, 0, 0);
3746 mutex_unlock(&adev->srbm_mutex);
3747
3748 if (adev->gfx.mec.mqd_backup[mqd_idx])
3749 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
3750 } else if (adev->in_gpu_reset) { /* for GPU_RESET case */
3751 /* reset MQD to a clean status */
3752 if (adev->gfx.mec.mqd_backup[mqd_idx])
3753 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
3754
3755 /* reset ring buffer */
3756 ring->wptr = 0;
3757 atomic64_set((atomic64_t *)&adev->wb.wb[ring->wptr_offs], 0);
3758 amdgpu_ring_clear_ring(ring);
3759 } else {
3760 amdgpu_ring_clear_ring(ring);
3761 }
3762
3763 return 0;
3764 }
3765
3766 static int gfx_v9_0_kiq_resume(struct amdgpu_device *adev)
3767 {
3768 struct amdgpu_ring *ring;
3769 int r;
3770
3771 ring = &adev->gfx.kiq.ring;
3772
3773 r = amdgpu_bo_reserve(ring->mqd_obj, false);
3774 if (unlikely(r != 0))
3775 return r;
3776
3777 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3778 if (unlikely(r != 0))
3779 return r;
3780
3781 gfx_v9_0_kiq_init_queue(ring);
3782 amdgpu_bo_kunmap(ring->mqd_obj);
3783 ring->mqd_ptr = NULL;
3784 amdgpu_bo_unreserve(ring->mqd_obj);
3785 ring->sched.ready = true;
3786 return 0;
3787 }
3788
3789 static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev)
3790 {
3791 struct amdgpu_ring *ring = NULL;
3792 int r = 0, i;
3793
3794 gfx_v9_0_cp_compute_enable(adev, true);
3795
3796 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3797 ring = &adev->gfx.compute_ring[i];
3798
3799 r = amdgpu_bo_reserve(ring->mqd_obj, false);
3800 if (unlikely(r != 0))
3801 goto done;
3802 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3803 if (!r) {
3804 r = gfx_v9_0_kcq_init_queue(ring);
3805 amdgpu_bo_kunmap(ring->mqd_obj);
3806 ring->mqd_ptr = NULL;
3807 }
3808 amdgpu_bo_unreserve(ring->mqd_obj);
3809 if (r)
3810 goto done;
3811 }
3812
3813 r = amdgpu_gfx_enable_kcq(adev);
3814 done:
3815 return r;
3816 }
3817
3818 static int gfx_v9_0_cp_resume(struct amdgpu_device *adev)
3819 {
3820 int r, i;
3821 struct amdgpu_ring *ring;
3822
3823 if (!(adev->flags & AMD_IS_APU))
3824 gfx_v9_0_enable_gui_idle_interrupt(adev, false);
3825
3826 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
3827 if (adev->asic_type != CHIP_ARCTURUS) {
3828 /* legacy firmware loading */
3829 r = gfx_v9_0_cp_gfx_load_microcode(adev);
3830 if (r)
3831 return r;
3832 }
3833
3834 r = gfx_v9_0_cp_compute_load_microcode(adev);
3835 if (r)
3836 return r;
3837 }
3838
3839 r = gfx_v9_0_kiq_resume(adev);
3840 if (r)
3841 return r;
3842
3843 if (adev->asic_type != CHIP_ARCTURUS) {
3844 r = gfx_v9_0_cp_gfx_resume(adev);
3845 if (r)
3846 return r;
3847 }
3848
3849 r = gfx_v9_0_kcq_resume(adev);
3850 if (r)
3851 return r;
3852
3853 if (adev->asic_type != CHIP_ARCTURUS) {
3854 ring = &adev->gfx.gfx_ring[0];
3855 r = amdgpu_ring_test_helper(ring);
3856 if (r)
3857 return r;
3858 }
3859
3860 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3861 ring = &adev->gfx.compute_ring[i];
3862 amdgpu_ring_test_helper(ring);
3863 }
3864
3865 gfx_v9_0_enable_gui_idle_interrupt(adev, true);
3866
3867 return 0;
3868 }
3869
3870 static void gfx_v9_0_init_tcp_config(struct amdgpu_device *adev)
3871 {
3872 u32 tmp;
3873
3874 if (adev->asic_type != CHIP_ARCTURUS)
3875 return;
3876
3877 tmp = RREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG);
3878 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE64KHASH,
3879 adev->df.hash_status.hash_64k);
3880 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE2MHASH,
3881 adev->df.hash_status.hash_2m);
3882 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE1GHASH,
3883 adev->df.hash_status.hash_1g);
3884 WREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG, tmp);
3885 }
3886
3887 static void gfx_v9_0_cp_enable(struct amdgpu_device *adev, bool enable)
3888 {
3889 if (adev->asic_type != CHIP_ARCTURUS)
3890 gfx_v9_0_cp_gfx_enable(adev, enable);
3891 gfx_v9_0_cp_compute_enable(adev, enable);
3892 }
3893
3894 static int gfx_v9_0_hw_init(void *handle)
3895 {
3896 int r;
3897 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3898
3899 if (!amdgpu_sriov_vf(adev))
3900 gfx_v9_0_init_golden_registers(adev);
3901
3902 gfx_v9_0_constants_init(adev);
3903
3904 gfx_v9_0_init_tcp_config(adev);
3905
3906 r = adev->gfx.rlc.funcs->resume(adev);
3907 if (r)
3908 return r;
3909
3910 r = gfx_v9_0_cp_resume(adev);
3911 if (r)
3912 return r;
3913
3914 return r;
3915 }
3916
3917 static int gfx_v9_0_hw_fini(void *handle)
3918 {
3919 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3920
3921 amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
3922 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
3923 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
3924
3925 /* DF freeze and kcq disable will fail */
3926 if (!amdgpu_ras_intr_triggered())
3927 /* disable KCQ to avoid CPC touch memory not valid anymore */
3928 amdgpu_gfx_disable_kcq(adev);
3929
3930 if (amdgpu_sriov_vf(adev)) {
3931 gfx_v9_0_cp_gfx_enable(adev, false);
3932 /* must disable polling for SRIOV when hw finished, otherwise
3933 * CPC engine may still keep fetching WB address which is already
3934 * invalid after sw finished and trigger DMAR reading error in
3935 * hypervisor side.
3936 */
3937 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3938 return 0;
3939 }
3940
3941 /* Use deinitialize sequence from CAIL when unbinding device from driver,
3942 * otherwise KIQ is hanging when binding back
3943 */
3944 if (!adev->in_gpu_reset && !adev->in_suspend) {
3945 mutex_lock(&adev->srbm_mutex);
3946 soc15_grbm_select(adev, adev->gfx.kiq.ring.me,
3947 adev->gfx.kiq.ring.pipe,
3948 adev->gfx.kiq.ring.queue, 0);
3949 gfx_v9_0_kiq_fini_register(&adev->gfx.kiq.ring);
3950 soc15_grbm_select(adev, 0, 0, 0, 0);
3951 mutex_unlock(&adev->srbm_mutex);
3952 }
3953
3954 gfx_v9_0_cp_enable(adev, false);
3955 adev->gfx.rlc.funcs->stop(adev);
3956
3957 return 0;
3958 }
3959
3960 static int gfx_v9_0_suspend(void *handle)
3961 {
3962 return gfx_v9_0_hw_fini(handle);
3963 }
3964
3965 static int gfx_v9_0_resume(void *handle)
3966 {
3967 return gfx_v9_0_hw_init(handle);
3968 }
3969
3970 static bool gfx_v9_0_is_idle(void *handle)
3971 {
3972 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3973
3974 if (REG_GET_FIELD(RREG32_SOC15(GC, 0, mmGRBM_STATUS),
3975 GRBM_STATUS, GUI_ACTIVE))
3976 return false;
3977 else
3978 return true;
3979 }
3980
3981 static int gfx_v9_0_wait_for_idle(void *handle)
3982 {
3983 unsigned i;
3984 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3985
3986 for (i = 0; i < adev->usec_timeout; i++) {
3987 if (gfx_v9_0_is_idle(handle))
3988 return 0;
3989 udelay(1);
3990 }
3991 return -ETIMEDOUT;
3992 }
3993
3994 static int gfx_v9_0_soft_reset(void *handle)
3995 {
3996 u32 grbm_soft_reset = 0;
3997 u32 tmp;
3998 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3999
4000 /* GRBM_STATUS */
4001 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS);
4002 if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4003 GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4004 GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4005 GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4006 GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4007 GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) {
4008 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4009 GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4010 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4011 GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
4012 }
4013
4014 if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4015 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4016 GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4017 }
4018
4019 /* GRBM_STATUS2 */
4020 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2);
4021 if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
4022 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4023 GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4024
4025
4026 if (grbm_soft_reset) {
4027 /* stop the rlc */
4028 adev->gfx.rlc.funcs->stop(adev);
4029
4030 if (adev->asic_type != CHIP_ARCTURUS)
4031 /* Disable GFX parsing/prefetching */
4032 gfx_v9_0_cp_gfx_enable(adev, false);
4033
4034 /* Disable MEC parsing/prefetching */
4035 gfx_v9_0_cp_compute_enable(adev, false);
4036
4037 if (grbm_soft_reset) {
4038 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
4039 tmp |= grbm_soft_reset;
4040 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4041 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
4042 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
4043
4044 udelay(50);
4045
4046 tmp &= ~grbm_soft_reset;
4047 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
4048 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
4049 }
4050
4051 /* Wait a little for things to settle down */
4052 udelay(50);
4053 }
4054 return 0;
4055 }
4056
4057 static uint64_t gfx_v9_0_kiq_read_clock(struct amdgpu_device *adev)
4058 {
4059 signed long r, cnt = 0;
4060 unsigned long flags;
4061 uint32_t seq;
4062 struct amdgpu_kiq *kiq = &adev->gfx.kiq;
4063 struct amdgpu_ring *ring = &kiq->ring;
4064
4065 BUG_ON(!ring->funcs->emit_rreg);
4066
4067 spin_lock_irqsave(&kiq->ring_lock, flags);
4068 amdgpu_ring_alloc(ring, 32);
4069 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
4070 amdgpu_ring_write(ring, 9 | /* src: register*/
4071 (5 << 8) | /* dst: memory */
4072 (1 << 16) | /* count sel */
4073 (1 << 20)); /* write confirm */
4074 amdgpu_ring_write(ring, 0);
4075 amdgpu_ring_write(ring, 0);
4076 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
4077 kiq->reg_val_offs * 4));
4078 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
4079 kiq->reg_val_offs * 4));
4080 amdgpu_fence_emit_polling(ring, &seq);
4081 amdgpu_ring_commit(ring);
4082 spin_unlock_irqrestore(&kiq->ring_lock, flags);
4083
4084 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
4085
4086 /* don't wait anymore for gpu reset case because this way may
4087 * block gpu_recover() routine forever, e.g. this virt_kiq_rreg
4088 * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will
4089 * never return if we keep waiting in virt_kiq_rreg, which cause
4090 * gpu_recover() hang there.
4091 *
4092 * also don't wait anymore for IRQ context
4093 * */
4094 if (r < 1 && (adev->in_gpu_reset || in_interrupt()))
4095 goto failed_kiq_read;
4096
4097 might_sleep();
4098 while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) {
4099 msleep(MAX_KIQ_REG_BAILOUT_INTERVAL);
4100 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
4101 }
4102
4103 if (cnt > MAX_KIQ_REG_TRY)
4104 goto failed_kiq_read;
4105
4106 return (uint64_t)adev->wb.wb[kiq->reg_val_offs] |
4107 (uint64_t)adev->wb.wb[kiq->reg_val_offs + 1 ] << 32ULL;
4108
4109 failed_kiq_read:
4110 pr_err("failed to read gpu clock\n");
4111 return ~0;
4112 }
4113
4114 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4115 {
4116 uint64_t clock;
4117
4118 amdgpu_gfx_off_ctrl(adev, false);
4119 mutex_lock(&adev->gfx.gpu_clock_mutex);
4120 if (adev->asic_type == CHIP_VEGA10 && amdgpu_sriov_runtime(adev)) {
4121 clock = gfx_v9_0_kiq_read_clock(adev);
4122 } else {
4123 WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4124 clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB) |
4125 ((uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4126 }
4127 mutex_unlock(&adev->gfx.gpu_clock_mutex);
4128 amdgpu_gfx_off_ctrl(adev, true);
4129 return clock;
4130 }
4131
4132 static void gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4133 uint32_t vmid,
4134 uint32_t gds_base, uint32_t gds_size,
4135 uint32_t gws_base, uint32_t gws_size,
4136 uint32_t oa_base, uint32_t oa_size)
4137 {
4138 struct amdgpu_device *adev = ring->adev;
4139
4140 /* GDS Base */
4141 gfx_v9_0_write_data_to_reg(ring, 0, false,
4142 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_BASE) + 2 * vmid,
4143 gds_base);
4144
4145 /* GDS Size */
4146 gfx_v9_0_write_data_to_reg(ring, 0, false,
4147 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE) + 2 * vmid,
4148 gds_size);
4149
4150 /* GWS */
4151 gfx_v9_0_write_data_to_reg(ring, 0, false,
4152 SOC15_REG_OFFSET(GC, 0, mmGDS_GWS_VMID0) + vmid,
4153 gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4154
4155 /* OA */
4156 gfx_v9_0_write_data_to_reg(ring, 0, false,
4157 SOC15_REG_OFFSET(GC, 0, mmGDS_OA_VMID0) + vmid,
4158 (1 << (oa_size + oa_base)) - (1 << oa_base));
4159 }
4160
4161 static const u32 vgpr_init_compute_shader[] =
4162 {
4163 0xb07c0000, 0xbe8000ff,
4164 0x000000f8, 0xbf110800,
4165 0x7e000280, 0x7e020280,
4166 0x7e040280, 0x7e060280,
4167 0x7e080280, 0x7e0a0280,
4168 0x7e0c0280, 0x7e0e0280,
4169 0x80808800, 0xbe803200,
4170 0xbf84fff5, 0xbf9c0000,
4171 0xd28c0001, 0x0001007f,
4172 0xd28d0001, 0x0002027e,
4173 0x10020288, 0xb8810904,
4174 0xb7814000, 0xd1196a01,
4175 0x00000301, 0xbe800087,
4176 0xbefc00c1, 0xd89c4000,
4177 0x00020201, 0xd89cc080,
4178 0x00040401, 0x320202ff,
4179 0x00000800, 0x80808100,
4180 0xbf84fff8, 0x7e020280,
4181 0xbf810000, 0x00000000,
4182 };
4183
4184 static const u32 sgpr_init_compute_shader[] =
4185 {
4186 0xb07c0000, 0xbe8000ff,
4187 0x0000005f, 0xbee50080,
4188 0xbe812c65, 0xbe822c65,
4189 0xbe832c65, 0xbe842c65,
4190 0xbe852c65, 0xb77c0005,
4191 0x80808500, 0xbf84fff8,
4192 0xbe800080, 0xbf810000,
4193 };
4194
4195 static const u32 vgpr_init_compute_shader_arcturus[] = {
4196 0xd3d94000, 0x18000080, 0xd3d94001, 0x18000080, 0xd3d94002, 0x18000080,
4197 0xd3d94003, 0x18000080, 0xd3d94004, 0x18000080, 0xd3d94005, 0x18000080,
4198 0xd3d94006, 0x18000080, 0xd3d94007, 0x18000080, 0xd3d94008, 0x18000080,
4199 0xd3d94009, 0x18000080, 0xd3d9400a, 0x18000080, 0xd3d9400b, 0x18000080,
4200 0xd3d9400c, 0x18000080, 0xd3d9400d, 0x18000080, 0xd3d9400e, 0x18000080,
4201 0xd3d9400f, 0x18000080, 0xd3d94010, 0x18000080, 0xd3d94011, 0x18000080,
4202 0xd3d94012, 0x18000080, 0xd3d94013, 0x18000080, 0xd3d94014, 0x18000080,
4203 0xd3d94015, 0x18000080, 0xd3d94016, 0x18000080, 0xd3d94017, 0x18000080,
4204 0xd3d94018, 0x18000080, 0xd3d94019, 0x18000080, 0xd3d9401a, 0x18000080,
4205 0xd3d9401b, 0x18000080, 0xd3d9401c, 0x18000080, 0xd3d9401d, 0x18000080,
4206 0xd3d9401e, 0x18000080, 0xd3d9401f, 0x18000080, 0xd3d94020, 0x18000080,
4207 0xd3d94021, 0x18000080, 0xd3d94022, 0x18000080, 0xd3d94023, 0x18000080,
4208 0xd3d94024, 0x18000080, 0xd3d94025, 0x18000080, 0xd3d94026, 0x18000080,
4209 0xd3d94027, 0x18000080, 0xd3d94028, 0x18000080, 0xd3d94029, 0x18000080,
4210 0xd3d9402a, 0x18000080, 0xd3d9402b, 0x18000080, 0xd3d9402c, 0x18000080,
4211 0xd3d9402d, 0x18000080, 0xd3d9402e, 0x18000080, 0xd3d9402f, 0x18000080,
4212 0xd3d94030, 0x18000080, 0xd3d94031, 0x18000080, 0xd3d94032, 0x18000080,
4213 0xd3d94033, 0x18000080, 0xd3d94034, 0x18000080, 0xd3d94035, 0x18000080,
4214 0xd3d94036, 0x18000080, 0xd3d94037, 0x18000080, 0xd3d94038, 0x18000080,
4215 0xd3d94039, 0x18000080, 0xd3d9403a, 0x18000080, 0xd3d9403b, 0x18000080,
4216 0xd3d9403c, 0x18000080, 0xd3d9403d, 0x18000080, 0xd3d9403e, 0x18000080,
4217 0xd3d9403f, 0x18000080, 0xd3d94040, 0x18000080, 0xd3d94041, 0x18000080,
4218 0xd3d94042, 0x18000080, 0xd3d94043, 0x18000080, 0xd3d94044, 0x18000080,
4219 0xd3d94045, 0x18000080, 0xd3d94046, 0x18000080, 0xd3d94047, 0x18000080,
4220 0xd3d94048, 0x18000080, 0xd3d94049, 0x18000080, 0xd3d9404a, 0x18000080,
4221 0xd3d9404b, 0x18000080, 0xd3d9404c, 0x18000080, 0xd3d9404d, 0x18000080,
4222 0xd3d9404e, 0x18000080, 0xd3d9404f, 0x18000080, 0xd3d94050, 0x18000080,
4223 0xd3d94051, 0x18000080, 0xd3d94052, 0x18000080, 0xd3d94053, 0x18000080,
4224 0xd3d94054, 0x18000080, 0xd3d94055, 0x18000080, 0xd3d94056, 0x18000080,
4225 0xd3d94057, 0x18000080, 0xd3d94058, 0x18000080, 0xd3d94059, 0x18000080,
4226 0xd3d9405a, 0x18000080, 0xd3d9405b, 0x18000080, 0xd3d9405c, 0x18000080,
4227 0xd3d9405d, 0x18000080, 0xd3d9405e, 0x18000080, 0xd3d9405f, 0x18000080,
4228 0xd3d94060, 0x18000080, 0xd3d94061, 0x18000080, 0xd3d94062, 0x18000080,
4229 0xd3d94063, 0x18000080, 0xd3d94064, 0x18000080, 0xd3d94065, 0x18000080,
4230 0xd3d94066, 0x18000080, 0xd3d94067, 0x18000080, 0xd3d94068, 0x18000080,
4231 0xd3d94069, 0x18000080, 0xd3d9406a, 0x18000080, 0xd3d9406b, 0x18000080,
4232 0xd3d9406c, 0x18000080, 0xd3d9406d, 0x18000080, 0xd3d9406e, 0x18000080,
4233 0xd3d9406f, 0x18000080, 0xd3d94070, 0x18000080, 0xd3d94071, 0x18000080,
4234 0xd3d94072, 0x18000080, 0xd3d94073, 0x18000080, 0xd3d94074, 0x18000080,
4235 0xd3d94075, 0x18000080, 0xd3d94076, 0x18000080, 0xd3d94077, 0x18000080,
4236 0xd3d94078, 0x18000080, 0xd3d94079, 0x18000080, 0xd3d9407a, 0x18000080,
4237 0xd3d9407b, 0x18000080, 0xd3d9407c, 0x18000080, 0xd3d9407d, 0x18000080,
4238 0xd3d9407e, 0x18000080, 0xd3d9407f, 0x18000080, 0xd3d94080, 0x18000080,
4239 0xd3d94081, 0x18000080, 0xd3d94082, 0x18000080, 0xd3d94083, 0x18000080,
4240 0xd3d94084, 0x18000080, 0xd3d94085, 0x18000080, 0xd3d94086, 0x18000080,
4241 0xd3d94087, 0x18000080, 0xd3d94088, 0x18000080, 0xd3d94089, 0x18000080,
4242 0xd3d9408a, 0x18000080, 0xd3d9408b, 0x18000080, 0xd3d9408c, 0x18000080,
4243 0xd3d9408d, 0x18000080, 0xd3d9408e, 0x18000080, 0xd3d9408f, 0x18000080,
4244 0xd3d94090, 0x18000080, 0xd3d94091, 0x18000080, 0xd3d94092, 0x18000080,
4245 0xd3d94093, 0x18000080, 0xd3d94094, 0x18000080, 0xd3d94095, 0x18000080,
4246 0xd3d94096, 0x18000080, 0xd3d94097, 0x18000080, 0xd3d94098, 0x18000080,
4247 0xd3d94099, 0x18000080, 0xd3d9409a, 0x18000080, 0xd3d9409b, 0x18000080,
4248 0xd3d9409c, 0x18000080, 0xd3d9409d, 0x18000080, 0xd3d9409e, 0x18000080,
4249 0xd3d9409f, 0x18000080, 0xd3d940a0, 0x18000080, 0xd3d940a1, 0x18000080,
4250 0xd3d940a2, 0x18000080, 0xd3d940a3, 0x18000080, 0xd3d940a4, 0x18000080,
4251 0xd3d940a5, 0x18000080, 0xd3d940a6, 0x18000080, 0xd3d940a7, 0x18000080,
4252 0xd3d940a8, 0x18000080, 0xd3d940a9, 0x18000080, 0xd3d940aa, 0x18000080,
4253 0xd3d940ab, 0x18000080, 0xd3d940ac, 0x18000080, 0xd3d940ad, 0x18000080,
4254 0xd3d940ae, 0x18000080, 0xd3d940af, 0x18000080, 0xd3d940b0, 0x18000080,
4255 0xd3d940b1, 0x18000080, 0xd3d940b2, 0x18000080, 0xd3d940b3, 0x18000080,
4256 0xd3d940b4, 0x18000080, 0xd3d940b5, 0x18000080, 0xd3d940b6, 0x18000080,
4257 0xd3d940b7, 0x18000080, 0xd3d940b8, 0x18000080, 0xd3d940b9, 0x18000080,
4258 0xd3d940ba, 0x18000080, 0xd3d940bb, 0x18000080, 0xd3d940bc, 0x18000080,
4259 0xd3d940bd, 0x18000080, 0xd3d940be, 0x18000080, 0xd3d940bf, 0x18000080,
4260 0xd3d940c0, 0x18000080, 0xd3d940c1, 0x18000080, 0xd3d940c2, 0x18000080,
4261 0xd3d940c3, 0x18000080, 0xd3d940c4, 0x18000080, 0xd3d940c5, 0x18000080,
4262 0xd3d940c6, 0x18000080, 0xd3d940c7, 0x18000080, 0xd3d940c8, 0x18000080,
4263 0xd3d940c9, 0x18000080, 0xd3d940ca, 0x18000080, 0xd3d940cb, 0x18000080,
4264 0xd3d940cc, 0x18000080, 0xd3d940cd, 0x18000080, 0xd3d940ce, 0x18000080,
4265 0xd3d940cf, 0x18000080, 0xd3d940d0, 0x18000080, 0xd3d940d1, 0x18000080,
4266 0xd3d940d2, 0x18000080, 0xd3d940d3, 0x18000080, 0xd3d940d4, 0x18000080,
4267 0xd3d940d5, 0x18000080, 0xd3d940d6, 0x18000080, 0xd3d940d7, 0x18000080,
4268 0xd3d940d8, 0x18000080, 0xd3d940d9, 0x18000080, 0xd3d940da, 0x18000080,
4269 0xd3d940db, 0x18000080, 0xd3d940dc, 0x18000080, 0xd3d940dd, 0x18000080,
4270 0xd3d940de, 0x18000080, 0xd3d940df, 0x18000080, 0xd3d940e0, 0x18000080,
4271 0xd3d940e1, 0x18000080, 0xd3d940e2, 0x18000080, 0xd3d940e3, 0x18000080,
4272 0xd3d940e4, 0x18000080, 0xd3d940e5, 0x18000080, 0xd3d940e6, 0x18000080,
4273 0xd3d940e7, 0x18000080, 0xd3d940e8, 0x18000080, 0xd3d940e9, 0x18000080,
4274 0xd3d940ea, 0x18000080, 0xd3d940eb, 0x18000080, 0xd3d940ec, 0x18000080,
4275 0xd3d940ed, 0x18000080, 0xd3d940ee, 0x18000080, 0xd3d940ef, 0x18000080,
4276 0xd3d940f0, 0x18000080, 0xd3d940f1, 0x18000080, 0xd3d940f2, 0x18000080,
4277 0xd3d940f3, 0x18000080, 0xd3d940f4, 0x18000080, 0xd3d940f5, 0x18000080,
4278 0xd3d940f6, 0x18000080, 0xd3d940f7, 0x18000080, 0xd3d940f8, 0x18000080,
4279 0xd3d940f9, 0x18000080, 0xd3d940fa, 0x18000080, 0xd3d940fb, 0x18000080,
4280 0xd3d940fc, 0x18000080, 0xd3d940fd, 0x18000080, 0xd3d940fe, 0x18000080,
4281 0xd3d940ff, 0x18000080, 0xb07c0000, 0xbe8a00ff, 0x000000f8, 0xbf11080a,
4282 0x7e000280, 0x7e020280, 0x7e040280, 0x7e060280, 0x7e080280, 0x7e0a0280,
4283 0x7e0c0280, 0x7e0e0280, 0x808a880a, 0xbe80320a, 0xbf84fff5, 0xbf9c0000,
4284 0xd28c0001, 0x0001007f, 0xd28d0001, 0x0002027e, 0x10020288, 0xb88b0904,
4285 0xb78b4000, 0xd1196a01, 0x00001701, 0xbe8a0087, 0xbefc00c1, 0xd89c4000,
4286 0x00020201, 0xd89cc080, 0x00040401, 0x320202ff, 0x00000800, 0x808a810a,
4287 0xbf84fff8, 0xbf810000,
4288 };
4289
4290 /* When below register arrays changed, please update gpr_reg_size,
4291 and sec_ded_counter_reg_size in function gfx_v9_0_do_edc_gpr_workarounds,
4292 to cover all gfx9 ASICs */
4293 static const struct soc15_reg_entry vgpr_init_regs[] = {
4294 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4295 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4296 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 },
4297 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4298 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x3f },
4299 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */
4300 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
4301 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
4302 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
4303 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
4304 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff },
4305 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff },
4306 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff },
4307 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff },
4308 };
4309
4310 static const struct soc15_reg_entry vgpr_init_regs_arcturus[] = {
4311 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4312 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4313 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 },
4314 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4315 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0xbf },
4316 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */
4317 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
4318 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
4319 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
4320 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
4321 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff },
4322 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff },
4323 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff },
4324 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff },
4325 };
4326
4327 static const struct soc15_reg_entry sgpr1_init_regs[] = {
4328 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4329 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4330 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 },
4331 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4332 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */
4333 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
4334 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x000000ff },
4335 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x000000ff },
4336 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x000000ff },
4337 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x000000ff },
4338 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x000000ff },
4339 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x000000ff },
4340 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x000000ff },
4341 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x000000ff },
4342 };
4343
4344 static const struct soc15_reg_entry sgpr2_init_regs[] = {
4345 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4346 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4347 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 },
4348 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4349 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */
4350 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
4351 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x0000ff00 },
4352 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x0000ff00 },
4353 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x0000ff00 },
4354 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x0000ff00 },
4355 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x0000ff00 },
4356 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x0000ff00 },
4357 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x0000ff00 },
4358 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x0000ff00 },
4359 };
4360
4361 static const struct soc15_reg_entry gfx_v9_0_edc_counter_regs[] = {
4362 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT), 0, 1, 1},
4363 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT), 0, 1, 1},
4364 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 0, 1, 1},
4365 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT), 0, 1, 1},
4366 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 0, 1, 1},
4367 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT), 0, 1, 1},
4368 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT), 0, 1, 1},
4369 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT), 0, 1, 1},
4370 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT), 0, 1, 1},
4371 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 0, 1, 1},
4372 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_GRBM_CNT), 0, 1, 1},
4373 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_DED), 0, 1, 1},
4374 { SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT), 0, 4, 1},
4375 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 0, 4, 6},
4376 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_DED_CNT), 0, 4, 16},
4377 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_INFO), 0, 4, 16},
4378 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_SEC_CNT), 0, 4, 16},
4379 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 0, 1, 16},
4380 { SOC15_REG_ENTRY(GC, 0, mmTCP_ATC_EDC_GATCL1_CNT), 0, 4, 16},
4381 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT), 0, 4, 16},
4382 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 0, 4, 16},
4383 { SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 0, 4, 16},
4384 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 0, 4, 6},
4385 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 0, 4, 16},
4386 { SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 0, 4, 16},
4387 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 0, 1, 1},
4388 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 0, 1, 1},
4389 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 0, 1, 32},
4390 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 0, 1, 32},
4391 { SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT), 0, 1, 72},
4392 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 0, 1, 16},
4393 { SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 0, 1, 2},
4394 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 0, 4, 6},
4395 };
4396
4397 static int gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device *adev)
4398 {
4399 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4400 int i, r;
4401
4402 /* only support when RAS is enabled */
4403 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4404 return 0;
4405
4406 r = amdgpu_ring_alloc(ring, 7);
4407 if (r) {
4408 DRM_ERROR("amdgpu: GDS workarounds failed to lock ring %s (%d).\n",
4409 ring->name, r);
4410 return r;
4411 }
4412
4413 WREG32_SOC15(GC, 0, mmGDS_VMID0_BASE, 0x00000000);
4414 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, adev->gds.gds_size);
4415
4416 amdgpu_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5));
4417 amdgpu_ring_write(ring, (PACKET3_DMA_DATA_CP_SYNC |
4418 PACKET3_DMA_DATA_DST_SEL(1) |
4419 PACKET3_DMA_DATA_SRC_SEL(2) |
4420 PACKET3_DMA_DATA_ENGINE(0)));
4421 amdgpu_ring_write(ring, 0);
4422 amdgpu_ring_write(ring, 0);
4423 amdgpu_ring_write(ring, 0);
4424 amdgpu_ring_write(ring, 0);
4425 amdgpu_ring_write(ring, PACKET3_DMA_DATA_CMD_RAW_WAIT |
4426 adev->gds.gds_size);
4427
4428 amdgpu_ring_commit(ring);
4429
4430 for (i = 0; i < adev->usec_timeout; i++) {
4431 if (ring->wptr == gfx_v9_0_ring_get_rptr_compute(ring))
4432 break;
4433 udelay(1);
4434 }
4435
4436 if (i >= adev->usec_timeout)
4437 r = -ETIMEDOUT;
4438
4439 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, 0x00000000);
4440
4441 return r;
4442 }
4443
4444 static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
4445 {
4446 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4447 struct amdgpu_ib ib;
4448 struct dma_fence *f = NULL;
4449 int r, i;
4450 unsigned total_size, vgpr_offset, sgpr_offset;
4451 u64 gpu_addr;
4452
4453 int compute_dim_x = adev->gfx.config.max_shader_engines *
4454 adev->gfx.config.max_cu_per_sh *
4455 adev->gfx.config.max_sh_per_se;
4456 int sgpr_work_group_size = 5;
4457 int gpr_reg_size = adev->gfx.config.max_shader_engines + 6;
4458 int vgpr_init_shader_size;
4459 const u32 *vgpr_init_shader_ptr;
4460 const struct soc15_reg_entry *vgpr_init_regs_ptr;
4461
4462 /* only support when RAS is enabled */
4463 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4464 return 0;
4465
4466 /* bail if the compute ring is not ready */
4467 if (!ring->sched.ready)
4468 return 0;
4469
4470 if (adev->asic_type == CHIP_ARCTURUS) {
4471 vgpr_init_shader_ptr = vgpr_init_compute_shader_arcturus;
4472 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader_arcturus);
4473 vgpr_init_regs_ptr = vgpr_init_regs_arcturus;
4474 } else {
4475 vgpr_init_shader_ptr = vgpr_init_compute_shader;
4476 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader);
4477 vgpr_init_regs_ptr = vgpr_init_regs;
4478 }
4479
4480 total_size =
4481 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* VGPRS */
4482 total_size +=
4483 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS1 */
4484 total_size +=
4485 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS2 */
4486 total_size = ALIGN(total_size, 256);
4487 vgpr_offset = total_size;
4488 total_size += ALIGN(vgpr_init_shader_size, 256);
4489 sgpr_offset = total_size;
4490 total_size += sizeof(sgpr_init_compute_shader);
4491
4492 /* allocate an indirect buffer to put the commands in */
4493 memset(&ib, 0, sizeof(ib));
4494 r = amdgpu_ib_get(adev, NULL, total_size, &ib);
4495 if (r) {
4496 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
4497 return r;
4498 }
4499
4500 /* load the compute shaders */
4501 for (i = 0; i < vgpr_init_shader_size/sizeof(u32); i++)
4502 ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_shader_ptr[i];
4503
4504 for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
4505 ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
4506
4507 /* init the ib length to 0 */
4508 ib.length_dw = 0;
4509
4510 /* VGPR */
4511 /* write the register state for the compute dispatch */
4512 for (i = 0; i < gpr_reg_size; i++) {
4513 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4514 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(vgpr_init_regs_ptr[i])
4515 - PACKET3_SET_SH_REG_START;
4516 ib.ptr[ib.length_dw++] = vgpr_init_regs_ptr[i].reg_value;
4517 }
4518 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4519 gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
4520 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4521 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4522 - PACKET3_SET_SH_REG_START;
4523 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4524 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4525
4526 /* write dispatch packet */
4527 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4528 ib.ptr[ib.length_dw++] = compute_dim_x * 2; /* x */
4529 ib.ptr[ib.length_dw++] = 1; /* y */
4530 ib.ptr[ib.length_dw++] = 1; /* z */
4531 ib.ptr[ib.length_dw++] =
4532 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4533
4534 /* write CS partial flush packet */
4535 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4536 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4537
4538 /* SGPR1 */
4539 /* write the register state for the compute dispatch */
4540 for (i = 0; i < gpr_reg_size; i++) {
4541 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4542 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr1_init_regs[i])
4543 - PACKET3_SET_SH_REG_START;
4544 ib.ptr[ib.length_dw++] = sgpr1_init_regs[i].reg_value;
4545 }
4546 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4547 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4548 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4549 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4550 - PACKET3_SET_SH_REG_START;
4551 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4552 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4553
4554 /* write dispatch packet */
4555 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4556 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */
4557 ib.ptr[ib.length_dw++] = 1; /* y */
4558 ib.ptr[ib.length_dw++] = 1; /* z */
4559 ib.ptr[ib.length_dw++] =
4560 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4561
4562 /* write CS partial flush packet */
4563 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4564 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4565
4566 /* SGPR2 */
4567 /* write the register state for the compute dispatch */
4568 for (i = 0; i < gpr_reg_size; i++) {
4569 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4570 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr2_init_regs[i])
4571 - PACKET3_SET_SH_REG_START;
4572 ib.ptr[ib.length_dw++] = sgpr2_init_regs[i].reg_value;
4573 }
4574 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4575 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4576 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4577 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4578 - PACKET3_SET_SH_REG_START;
4579 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4580 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4581
4582 /* write dispatch packet */
4583 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4584 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */
4585 ib.ptr[ib.length_dw++] = 1; /* y */
4586 ib.ptr[ib.length_dw++] = 1; /* z */
4587 ib.ptr[ib.length_dw++] =
4588 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4589
4590 /* write CS partial flush packet */
4591 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4592 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4593
4594 /* shedule the ib on the ring */
4595 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
4596 if (r) {
4597 DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
4598 goto fail;
4599 }
4600
4601 /* wait for the GPU to finish processing the IB */
4602 r = dma_fence_wait(f, false);
4603 if (r) {
4604 DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
4605 goto fail;
4606 }
4607
4608 fail:
4609 amdgpu_ib_free(adev, &ib, NULL);
4610 dma_fence_put(f);
4611
4612 return r;
4613 }
4614
4615 static int gfx_v9_0_early_init(void *handle)
4616 {
4617 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4618
4619 if (adev->asic_type == CHIP_ARCTURUS)
4620 adev->gfx.num_gfx_rings = 0;
4621 else
4622 adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS;
4623 adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
4624 gfx_v9_0_set_kiq_pm4_funcs(adev);
4625 gfx_v9_0_set_ring_funcs(adev);
4626 gfx_v9_0_set_irq_funcs(adev);
4627 gfx_v9_0_set_gds_init(adev);
4628 gfx_v9_0_set_rlc_funcs(adev);
4629
4630 return 0;
4631 }
4632
4633 static int gfx_v9_0_ecc_late_init(void *handle)
4634 {
4635 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4636 int r;
4637
4638 /*
4639 * Temp workaround to fix the issue that CP firmware fails to
4640 * update read pointer when CPDMA is writing clearing operation
4641 * to GDS in suspend/resume sequence on several cards. So just
4642 * limit this operation in cold boot sequence.
4643 */
4644 if (!adev->in_suspend) {
4645 r = gfx_v9_0_do_edc_gds_workarounds(adev);
4646 if (r)
4647 return r;
4648 }
4649
4650 /* requires IBs so do in late init after IB pool is initialized */
4651 r = gfx_v9_0_do_edc_gpr_workarounds(adev);
4652 if (r)
4653 return r;
4654
4655 if (adev->gfx.funcs &&
4656 adev->gfx.funcs->reset_ras_error_count)
4657 adev->gfx.funcs->reset_ras_error_count(adev);
4658
4659 r = amdgpu_gfx_ras_late_init(adev);
4660 if (r)
4661 return r;
4662
4663 return 0;
4664 }
4665
4666 static int gfx_v9_0_late_init(void *handle)
4667 {
4668 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4669 int r;
4670
4671 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4672 if (r)
4673 return r;
4674
4675 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4676 if (r)
4677 return r;
4678
4679 r = gfx_v9_0_ecc_late_init(handle);
4680 if (r)
4681 return r;
4682
4683 return 0;
4684 }
4685
4686 static bool gfx_v9_0_is_rlc_enabled(struct amdgpu_device *adev)
4687 {
4688 uint32_t rlc_setting;
4689
4690 /* if RLC is not enabled, do nothing */
4691 rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL);
4692 if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
4693 return false;
4694
4695 return true;
4696 }
4697
4698 static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev)
4699 {
4700 uint32_t data;
4701 unsigned i;
4702
4703 data = RLC_SAFE_MODE__CMD_MASK;
4704 data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
4705 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4706
4707 /* wait for RLC_SAFE_MODE */
4708 for (i = 0; i < adev->usec_timeout; i++) {
4709 if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
4710 break;
4711 udelay(1);
4712 }
4713 }
4714
4715 static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev)
4716 {
4717 uint32_t data;
4718
4719 data = RLC_SAFE_MODE__CMD_MASK;
4720 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4721 }
4722
4723 static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev,
4724 bool enable)
4725 {
4726 amdgpu_gfx_rlc_enter_safe_mode(adev);
4727
4728 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
4729 gfx_v9_0_enable_gfx_cg_power_gating(adev, true);
4730 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4731 gfx_v9_0_enable_gfx_pipeline_powergating(adev, true);
4732 } else {
4733 gfx_v9_0_enable_gfx_cg_power_gating(adev, false);
4734 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4735 gfx_v9_0_enable_gfx_pipeline_powergating(adev, false);
4736 }
4737
4738 amdgpu_gfx_rlc_exit_safe_mode(adev);
4739 }
4740
4741 static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev,
4742 bool enable)
4743 {
4744 /* TODO: double check if we need to perform under safe mode */
4745 /* gfx_v9_0_enter_rlc_safe_mode(adev); */
4746
4747 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
4748 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, true);
4749 else
4750 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, false);
4751
4752 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
4753 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, true);
4754 else
4755 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, false);
4756
4757 /* gfx_v9_0_exit_rlc_safe_mode(adev); */
4758 }
4759
4760 static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
4761 bool enable)
4762 {
4763 uint32_t data, def;
4764
4765 amdgpu_gfx_rlc_enter_safe_mode(adev);
4766
4767 /* It is disabled by HW by default */
4768 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
4769 /* 1 - RLC_CGTT_MGCG_OVERRIDE */
4770 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4771
4772 if (adev->asic_type != CHIP_VEGA12)
4773 data &= ~RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
4774
4775 data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4776 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4777 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4778
4779 /* only for Vega10 & Raven1 */
4780 data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK;
4781
4782 if (def != data)
4783 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4784
4785 /* MGLS is a global flag to control all MGLS in GFX */
4786 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
4787 /* 2 - RLC memory Light sleep */
4788 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
4789 def = data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4790 data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4791 if (def != data)
4792 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4793 }
4794 /* 3 - CP memory Light sleep */
4795 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
4796 def = data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4797 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4798 if (def != data)
4799 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4800 }
4801 }
4802 } else {
4803 /* 1 - MGCG_OVERRIDE */
4804 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4805
4806 if (adev->asic_type != CHIP_VEGA12)
4807 data |= RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
4808
4809 data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4810 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4811 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4812 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4813
4814 if (def != data)
4815 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4816
4817 /* 2 - disable MGLS in RLC */
4818 data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4819 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
4820 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4821 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4822 }
4823
4824 /* 3 - disable MGLS in CP */
4825 data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4826 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
4827 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4828 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4829 }
4830 }
4831
4832 amdgpu_gfx_rlc_exit_safe_mode(adev);
4833 }
4834
4835 static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev,
4836 bool enable)
4837 {
4838 uint32_t data, def;
4839
4840 if (adev->asic_type == CHIP_ARCTURUS)
4841 return;
4842
4843 amdgpu_gfx_rlc_enter_safe_mode(adev);
4844
4845 /* Enable 3D CGCG/CGLS */
4846 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)) {
4847 /* write cmd to clear cgcg/cgls ov */
4848 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4849 /* unset CGCG override */
4850 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK;
4851 /* update CGCG and CGLS override bits */
4852 if (def != data)
4853 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4854
4855 /* enable 3Dcgcg FSM(0x0000363f) */
4856 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4857
4858 data = (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4859 RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
4860 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
4861 data |= (0x000F << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4862 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
4863 if (def != data)
4864 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
4865
4866 /* set IDLE_POLL_COUNT(0x00900100) */
4867 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
4868 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4869 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4870 if (def != data)
4871 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
4872 } else {
4873 /* Disable CGCG/CGLS */
4874 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4875 /* disable cgcg, cgls should be disabled */
4876 data &= ~(RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK |
4877 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK);
4878 /* disable cgcg and cgls in FSM */
4879 if (def != data)
4880 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
4881 }
4882
4883 amdgpu_gfx_rlc_exit_safe_mode(adev);
4884 }
4885
4886 static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
4887 bool enable)
4888 {
4889 uint32_t def, data;
4890
4891 amdgpu_gfx_rlc_enter_safe_mode(adev);
4892
4893 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
4894 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4895 /* unset CGCG override */
4896 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
4897 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4898 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4899 else
4900 data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4901 /* update CGCG and CGLS override bits */
4902 if (def != data)
4903 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4904
4905 /* enable cgcg FSM(0x0000363F) */
4906 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4907
4908 if (adev->asic_type == CHIP_ARCTURUS)
4909 data = (0x2000 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4910 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4911 else
4912 data = (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4913 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4914 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4915 data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4916 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
4917 if (def != data)
4918 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
4919
4920 /* set IDLE_POLL_COUNT(0x00900100) */
4921 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
4922 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4923 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4924 if (def != data)
4925 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
4926 } else {
4927 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4928 /* reset CGCG/CGLS bits */
4929 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
4930 /* disable cgcg and cgls in FSM */
4931 if (def != data)
4932 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
4933 }
4934
4935 amdgpu_gfx_rlc_exit_safe_mode(adev);
4936 }
4937
4938 static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev,
4939 bool enable)
4940 {
4941 if (enable) {
4942 /* CGCG/CGLS should be enabled after MGCG/MGLS
4943 * === MGCG + MGLS ===
4944 */
4945 gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
4946 /* === CGCG /CGLS for GFX 3D Only === */
4947 gfx_v9_0_update_3d_clock_gating(adev, enable);
4948 /* === CGCG + CGLS === */
4949 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
4950 } else {
4951 /* CGCG/CGLS should be disabled before MGCG/MGLS
4952 * === CGCG + CGLS ===
4953 */
4954 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
4955 /* === CGCG /CGLS for GFX 3D Only === */
4956 gfx_v9_0_update_3d_clock_gating(adev, enable);
4957 /* === MGCG + MGLS === */
4958 gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
4959 }
4960 return 0;
4961 }
4962
4963 static void gfx_v9_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
4964 {
4965 u32 data;
4966
4967 data = RREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL);
4968
4969 data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK;
4970 data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT;
4971
4972 WREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL, data);
4973 }
4974
4975 static bool gfx_v9_0_check_rlcg_range(struct amdgpu_device *adev,
4976 uint32_t offset,
4977 struct soc15_reg_rlcg *entries, int arr_size)
4978 {
4979 int i;
4980 uint32_t reg;
4981
4982 if (!entries)
4983 return false;
4984
4985 for (i = 0; i < arr_size; i++) {
4986 const struct soc15_reg_rlcg *entry;
4987
4988 entry = &entries[i];
4989 reg = adev->reg_offset[entry->hwip][entry->instance][entry->segment] + entry->reg;
4990 if (offset == reg)
4991 return true;
4992 }
4993
4994 return false;
4995 }
4996
4997 static bool gfx_v9_0_is_rlcg_access_range(struct amdgpu_device *adev, u32 offset)
4998 {
4999 return gfx_v9_0_check_rlcg_range(adev, offset,
5000 (void *)rlcg_access_gc_9_0,
5001 ARRAY_SIZE(rlcg_access_gc_9_0));
5002 }
5003
5004 static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = {
5005 .is_rlc_enabled = gfx_v9_0_is_rlc_enabled,
5006 .set_safe_mode = gfx_v9_0_set_safe_mode,
5007 .unset_safe_mode = gfx_v9_0_unset_safe_mode,
5008 .init = gfx_v9_0_rlc_init,
5009 .get_csb_size = gfx_v9_0_get_csb_size,
5010 .get_csb_buffer = gfx_v9_0_get_csb_buffer,
5011 .get_cp_table_num = gfx_v9_0_cp_jump_table_num,
5012 .resume = gfx_v9_0_rlc_resume,
5013 .stop = gfx_v9_0_rlc_stop,
5014 .reset = gfx_v9_0_rlc_reset,
5015 .start = gfx_v9_0_rlc_start,
5016 .update_spm_vmid = gfx_v9_0_update_spm_vmid,
5017 .rlcg_wreg = gfx_v9_0_rlcg_wreg,
5018 .is_rlcg_access_range = gfx_v9_0_is_rlcg_access_range,
5019 };
5020
5021 static int gfx_v9_0_set_powergating_state(void *handle,
5022 enum amd_powergating_state state)
5023 {
5024 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5025 bool enable = (state == AMD_PG_STATE_GATE);
5026
5027 switch (adev->asic_type) {
5028 case CHIP_RAVEN:
5029 case CHIP_RENOIR:
5030 if (!enable)
5031 amdgpu_gfx_off_ctrl(adev, false);
5032
5033 if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
5034 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, true);
5035 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, true);
5036 } else {
5037 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, false);
5038 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, false);
5039 }
5040
5041 if (adev->pg_flags & AMD_PG_SUPPORT_CP)
5042 gfx_v9_0_enable_cp_power_gating(adev, true);
5043 else
5044 gfx_v9_0_enable_cp_power_gating(adev, false);
5045
5046 /* update gfx cgpg state */
5047 gfx_v9_0_update_gfx_cg_power_gating(adev, enable);
5048
5049 /* update mgcg state */
5050 gfx_v9_0_update_gfx_mg_power_gating(adev, enable);
5051
5052 if (enable)
5053 amdgpu_gfx_off_ctrl(adev, true);
5054 break;
5055 case CHIP_VEGA12:
5056 amdgpu_gfx_off_ctrl(adev, enable);
5057 break;
5058 default:
5059 break;
5060 }
5061
5062 return 0;
5063 }
5064
5065 static int gfx_v9_0_set_clockgating_state(void *handle,
5066 enum amd_clockgating_state state)
5067 {
5068 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5069
5070 if (amdgpu_sriov_vf(adev))
5071 return 0;
5072
5073 switch (adev->asic_type) {
5074 case CHIP_VEGA10:
5075 case CHIP_VEGA12:
5076 case CHIP_VEGA20:
5077 case CHIP_RAVEN:
5078 case CHIP_ARCTURUS:
5079 case CHIP_RENOIR:
5080 gfx_v9_0_update_gfx_clock_gating(adev,
5081 state == AMD_CG_STATE_GATE);
5082 break;
5083 default:
5084 break;
5085 }
5086 return 0;
5087 }
5088
5089 static void gfx_v9_0_get_clockgating_state(void *handle, u32 *flags)
5090 {
5091 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5092 int data;
5093
5094 if (amdgpu_sriov_vf(adev))
5095 *flags = 0;
5096
5097 /* AMD_CG_SUPPORT_GFX_MGCG */
5098 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE));
5099 if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
5100 *flags |= AMD_CG_SUPPORT_GFX_MGCG;
5101
5102 /* AMD_CG_SUPPORT_GFX_CGCG */
5103 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL));
5104 if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5105 *flags |= AMD_CG_SUPPORT_GFX_CGCG;
5106
5107 /* AMD_CG_SUPPORT_GFX_CGLS */
5108 if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5109 *flags |= AMD_CG_SUPPORT_GFX_CGLS;
5110
5111 /* AMD_CG_SUPPORT_GFX_RLC_LS */
5112 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_MEM_SLP_CNTL));
5113 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5114 *flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5115
5116 /* AMD_CG_SUPPORT_GFX_CP_LS */
5117 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmCP_MEM_SLP_CNTL));
5118 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5119 *flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5120
5121 if (adev->asic_type != CHIP_ARCTURUS) {
5122 /* AMD_CG_SUPPORT_GFX_3D_CGCG */
5123 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D));
5124 if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK)
5125 *flags |= AMD_CG_SUPPORT_GFX_3D_CGCG;
5126
5127 /* AMD_CG_SUPPORT_GFX_3D_CGLS */
5128 if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK)
5129 *flags |= AMD_CG_SUPPORT_GFX_3D_CGLS;
5130 }
5131 }
5132
5133 static u64 gfx_v9_0_ring_get_rptr_gfx(struct amdgpu_ring *ring)
5134 {
5135 return ring->adev->wb.wb[ring->rptr_offs]; /* gfx9 is 32bit rptr*/
5136 }
5137
5138 static u64 gfx_v9_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
5139 {
5140 struct amdgpu_device *adev = ring->adev;
5141 u64 wptr;
5142
5143 /* XXX check if swapping is necessary on BE */
5144 if (ring->use_doorbell) {
5145 wptr = atomic64_read((atomic64_t *)&adev->wb.wb[ring->wptr_offs]);
5146 } else {
5147 wptr = RREG32_SOC15(GC, 0, mmCP_RB0_WPTR);
5148 wptr += (u64)RREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI) << 32;
5149 }
5150
5151 return wptr;
5152 }
5153
5154 static void gfx_v9_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
5155 {
5156 struct amdgpu_device *adev = ring->adev;
5157
5158 if (ring->use_doorbell) {
5159 /* XXX check if swapping is necessary on BE */
5160 atomic64_set((atomic64_t*)&adev->wb.wb[ring->wptr_offs], ring->wptr);
5161 WDOORBELL64(ring->doorbell_index, ring->wptr);
5162 } else {
5163 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
5164 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
5165 }
5166 }
5167
5168 static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
5169 {
5170 struct amdgpu_device *adev = ring->adev;
5171 u32 ref_and_mask, reg_mem_engine;
5172 const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
5173
5174 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
5175 switch (ring->me) {
5176 case 1:
5177 ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
5178 break;
5179 case 2:
5180 ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
5181 break;
5182 default:
5183 return;
5184 }
5185 reg_mem_engine = 0;
5186 } else {
5187 ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
5188 reg_mem_engine = 1; /* pfp */
5189 }
5190
5191 gfx_v9_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
5192 adev->nbio.funcs->get_hdp_flush_req_offset(adev),
5193 adev->nbio.funcs->get_hdp_flush_done_offset(adev),
5194 ref_and_mask, ref_and_mask, 0x20);
5195 }
5196
5197 static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
5198 struct amdgpu_job *job,
5199 struct amdgpu_ib *ib,
5200 uint32_t flags)
5201 {
5202 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5203 u32 header, control = 0;
5204
5205 if (ib->flags & AMDGPU_IB_FLAG_CE)
5206 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
5207 else
5208 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
5209
5210 control |= ib->length_dw | (vmid << 24);
5211
5212 if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
5213 control |= INDIRECT_BUFFER_PRE_ENB(1);
5214
5215 if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid)
5216 gfx_v9_0_ring_emit_de_meta(ring);
5217 }
5218
5219 amdgpu_ring_write(ring, header);
5220 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5221 amdgpu_ring_write(ring,
5222 #ifdef __BIG_ENDIAN
5223 (2 << 0) |
5224 #endif
5225 lower_32_bits(ib->gpu_addr));
5226 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5227 amdgpu_ring_write(ring, control);
5228 }
5229
5230 static void gfx_v9_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
5231 struct amdgpu_job *job,
5232 struct amdgpu_ib *ib,
5233 uint32_t flags)
5234 {
5235 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5236 u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
5237
5238 /* Currently, there is a high possibility to get wave ID mismatch
5239 * between ME and GDS, leading to a hw deadlock, because ME generates
5240 * different wave IDs than the GDS expects. This situation happens
5241 * randomly when at least 5 compute pipes use GDS ordered append.
5242 * The wave IDs generated by ME are also wrong after suspend/resume.
5243 * Those are probably bugs somewhere else in the kernel driver.
5244 *
5245 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
5246 * GDS to 0 for this ring (me/pipe).
5247 */
5248 if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
5249 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
5250 amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID);
5251 amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
5252 }
5253
5254 amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
5255 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5256 amdgpu_ring_write(ring,
5257 #ifdef __BIG_ENDIAN
5258 (2 << 0) |
5259 #endif
5260 lower_32_bits(ib->gpu_addr));
5261 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5262 amdgpu_ring_write(ring, control);
5263 }
5264
5265 static void gfx_v9_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
5266 u64 seq, unsigned flags)
5267 {
5268 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
5269 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
5270 bool writeback = flags & AMDGPU_FENCE_FLAG_TC_WB_ONLY;
5271
5272 /* RELEASE_MEM - flush caches, send int */
5273 amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
5274 amdgpu_ring_write(ring, ((writeback ? (EOP_TC_WB_ACTION_EN |
5275 EOP_TC_NC_ACTION_EN) :
5276 (EOP_TCL1_ACTION_EN |
5277 EOP_TC_ACTION_EN |
5278 EOP_TC_WB_ACTION_EN |
5279 EOP_TC_MD_ACTION_EN)) |
5280 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
5281 EVENT_INDEX(5)));
5282 amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
5283
5284 /*
5285 * the address should be Qword aligned if 64bit write, Dword
5286 * aligned if only send 32bit data low (discard data high)
5287 */
5288 if (write64bit)
5289 BUG_ON(addr & 0x7);
5290 else
5291 BUG_ON(addr & 0x3);
5292 amdgpu_ring_write(ring, lower_32_bits(addr));
5293 amdgpu_ring_write(ring, upper_32_bits(addr));
5294 amdgpu_ring_write(ring, lower_32_bits(seq));
5295 amdgpu_ring_write(ring, upper_32_bits(seq));
5296 amdgpu_ring_write(ring, 0);
5297 }
5298
5299 static void gfx_v9_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
5300 {
5301 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5302 uint32_t seq = ring->fence_drv.sync_seq;
5303 uint64_t addr = ring->fence_drv.gpu_addr;
5304
5305 gfx_v9_0_wait_reg_mem(ring, usepfp, 1, 0,
5306 lower_32_bits(addr), upper_32_bits(addr),
5307 seq, 0xffffffff, 4);
5308 }
5309
5310 static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
5311 unsigned vmid, uint64_t pd_addr)
5312 {
5313 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
5314
5315 /* compute doesn't have PFP */
5316 if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
5317 /* sync PFP to ME, otherwise we might get invalid PFP reads */
5318 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
5319 amdgpu_ring_write(ring, 0x0);
5320 }
5321 }
5322
5323 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring)
5324 {
5325 return ring->adev->wb.wb[ring->rptr_offs]; /* gfx9 hardware is 32bit rptr */
5326 }
5327
5328 static u64 gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
5329 {
5330 u64 wptr;
5331
5332 /* XXX check if swapping is necessary on BE */
5333 if (ring->use_doorbell)
5334 wptr = atomic64_read((atomic64_t *)&ring->adev->wb.wb[ring->wptr_offs]);
5335 else
5336 BUG();
5337 return wptr;
5338 }
5339
5340 static void gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
5341 {
5342 struct amdgpu_device *adev = ring->adev;
5343
5344 /* XXX check if swapping is necessary on BE */
5345 if (ring->use_doorbell) {
5346 atomic64_set((atomic64_t*)&adev->wb.wb[ring->wptr_offs], ring->wptr);
5347 WDOORBELL64(ring->doorbell_index, ring->wptr);
5348 } else{
5349 BUG(); /* only DOORBELL method supported on gfx9 now */
5350 }
5351 }
5352
5353 static void gfx_v9_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
5354 u64 seq, unsigned int flags)
5355 {
5356 struct amdgpu_device *adev = ring->adev;
5357
5358 /* we only allocate 32bit for each seq wb address */
5359 BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
5360
5361 /* write fence seq to the "addr" */
5362 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5363 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5364 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
5365 amdgpu_ring_write(ring, lower_32_bits(addr));
5366 amdgpu_ring_write(ring, upper_32_bits(addr));
5367 amdgpu_ring_write(ring, lower_32_bits(seq));
5368
5369 if (flags & AMDGPU_FENCE_FLAG_INT) {
5370 /* set register to trigger INT */
5371 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5372 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5373 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
5374 amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, mmCPC_INT_STATUS));
5375 amdgpu_ring_write(ring, 0);
5376 amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
5377 }
5378 }
5379
5380 static void gfx_v9_ring_emit_sb(struct amdgpu_ring *ring)
5381 {
5382 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
5383 amdgpu_ring_write(ring, 0);
5384 }
5385
5386 static void gfx_v9_0_ring_emit_ce_meta(struct amdgpu_ring *ring)
5387 {
5388 struct v9_ce_ib_state ce_payload = {0};
5389 uint64_t csa_addr;
5390 int cnt;
5391
5392 cnt = (sizeof(ce_payload) >> 2) + 4 - 2;
5393 csa_addr = amdgpu_csa_vaddr(ring->adev);
5394
5395 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5396 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
5397 WRITE_DATA_DST_SEL(8) |
5398 WR_CONFIRM) |
5399 WRITE_DATA_CACHE_POLICY(0));
5400 amdgpu_ring_write(ring, lower_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, ce_payload)));
5401 amdgpu_ring_write(ring, upper_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, ce_payload)));
5402 amdgpu_ring_write_multiple(ring, (void *)&ce_payload, sizeof(ce_payload) >> 2);
5403 }
5404
5405 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring)
5406 {
5407 struct v9_de_ib_state de_payload = {0};
5408 uint64_t csa_addr, gds_addr;
5409 int cnt;
5410
5411 csa_addr = amdgpu_csa_vaddr(ring->adev);
5412 gds_addr = csa_addr + 4096;
5413 de_payload.gds_backup_addrlo = lower_32_bits(gds_addr);
5414 de_payload.gds_backup_addrhi = upper_32_bits(gds_addr);
5415
5416 cnt = (sizeof(de_payload) >> 2) + 4 - 2;
5417 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5418 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
5419 WRITE_DATA_DST_SEL(8) |
5420 WR_CONFIRM) |
5421 WRITE_DATA_CACHE_POLICY(0));
5422 amdgpu_ring_write(ring, lower_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, de_payload)));
5423 amdgpu_ring_write(ring, upper_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, de_payload)));
5424 amdgpu_ring_write_multiple(ring, (void *)&de_payload, sizeof(de_payload) >> 2);
5425 }
5426
5427 static void gfx_v9_0_ring_emit_tmz(struct amdgpu_ring *ring, bool start)
5428 {
5429 amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0));
5430 amdgpu_ring_write(ring, FRAME_CMD(start ? 0 : 1)); /* frame_end */
5431 }
5432
5433 static void gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
5434 {
5435 uint32_t dw2 = 0;
5436
5437 if (amdgpu_sriov_vf(ring->adev))
5438 gfx_v9_0_ring_emit_ce_meta(ring);
5439
5440 gfx_v9_0_ring_emit_tmz(ring, true);
5441
5442 dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
5443 if (flags & AMDGPU_HAVE_CTX_SWITCH) {
5444 /* set load_global_config & load_global_uconfig */
5445 dw2 |= 0x8001;
5446 /* set load_cs_sh_regs */
5447 dw2 |= 0x01000000;
5448 /* set load_per_context_state & load_gfx_sh_regs for GFX */
5449 dw2 |= 0x10002;
5450
5451 /* set load_ce_ram if preamble presented */
5452 if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
5453 dw2 |= 0x10000000;
5454 } else {
5455 /* still load_ce_ram if this is the first time preamble presented
5456 * although there is no context switch happens.
5457 */
5458 if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
5459 dw2 |= 0x10000000;
5460 }
5461
5462 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
5463 amdgpu_ring_write(ring, dw2);
5464 amdgpu_ring_write(ring, 0);
5465 }
5466
5467 static unsigned gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
5468 {
5469 unsigned ret;
5470 amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
5471 amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
5472 amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
5473 amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
5474 ret = ring->wptr & ring->buf_mask;
5475 amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
5476 return ret;
5477 }
5478
5479 static void gfx_v9_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
5480 {
5481 unsigned cur;
5482 BUG_ON(offset > ring->buf_mask);
5483 BUG_ON(ring->ring[offset] != 0x55aa55aa);
5484
5485 cur = (ring->wptr & ring->buf_mask) - 1;
5486 if (likely(cur > offset))
5487 ring->ring[offset] = cur - offset;
5488 else
5489 ring->ring[offset] = (ring->ring_size>>2) - offset + cur;
5490 }
5491
5492 static void gfx_v9_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg)
5493 {
5494 struct amdgpu_device *adev = ring->adev;
5495 struct amdgpu_kiq *kiq = &adev->gfx.kiq;
5496
5497 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
5498 amdgpu_ring_write(ring, 0 | /* src: register*/
5499 (5 << 8) | /* dst: memory */
5500 (1 << 20)); /* write confirm */
5501 amdgpu_ring_write(ring, reg);
5502 amdgpu_ring_write(ring, 0);
5503 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
5504 kiq->reg_val_offs * 4));
5505 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
5506 kiq->reg_val_offs * 4));
5507 }
5508
5509 static void gfx_v9_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
5510 uint32_t val)
5511 {
5512 uint32_t cmd = 0;
5513
5514 switch (ring->funcs->type) {
5515 case AMDGPU_RING_TYPE_GFX:
5516 cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
5517 break;
5518 case AMDGPU_RING_TYPE_KIQ:
5519 cmd = (1 << 16); /* no inc addr */
5520 break;
5521 default:
5522 cmd = WR_CONFIRM;
5523 break;
5524 }
5525 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5526 amdgpu_ring_write(ring, cmd);
5527 amdgpu_ring_write(ring, reg);
5528 amdgpu_ring_write(ring, 0);
5529 amdgpu_ring_write(ring, val);
5530 }
5531
5532 static void gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
5533 uint32_t val, uint32_t mask)
5534 {
5535 gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
5536 }
5537
5538 static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
5539 uint32_t reg0, uint32_t reg1,
5540 uint32_t ref, uint32_t mask)
5541 {
5542 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5543 struct amdgpu_device *adev = ring->adev;
5544 bool fw_version_ok = (ring->funcs->type == AMDGPU_RING_TYPE_GFX) ?
5545 adev->gfx.me_fw_write_wait : adev->gfx.mec_fw_write_wait;
5546
5547 if (fw_version_ok)
5548 gfx_v9_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1,
5549 ref, mask, 0x20);
5550 else
5551 amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1,
5552 ref, mask);
5553 }
5554
5555 static void gfx_v9_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
5556 {
5557 struct amdgpu_device *adev = ring->adev;
5558 uint32_t value = 0;
5559
5560 value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
5561 value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
5562 value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
5563 value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
5564 WREG32_SOC15(GC, 0, mmSQ_CMD, value);
5565 }
5566
5567 static void gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
5568 enum amdgpu_interrupt_state state)
5569 {
5570 switch (state) {
5571 case AMDGPU_IRQ_STATE_DISABLE:
5572 case AMDGPU_IRQ_STATE_ENABLE:
5573 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5574 TIME_STAMP_INT_ENABLE,
5575 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5576 break;
5577 default:
5578 break;
5579 }
5580 }
5581
5582 static void gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
5583 int me, int pipe,
5584 enum amdgpu_interrupt_state state)
5585 {
5586 u32 mec_int_cntl, mec_int_cntl_reg;
5587
5588 /*
5589 * amdgpu controls only the first MEC. That's why this function only
5590 * handles the setting of interrupts for this specific MEC. All other
5591 * pipes' interrupts are set by amdkfd.
5592 */
5593
5594 if (me == 1) {
5595 switch (pipe) {
5596 case 0:
5597 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
5598 break;
5599 case 1:
5600 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL);
5601 break;
5602 case 2:
5603 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL);
5604 break;
5605 case 3:
5606 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL);
5607 break;
5608 default:
5609 DRM_DEBUG("invalid pipe %d\n", pipe);
5610 return;
5611 }
5612 } else {
5613 DRM_DEBUG("invalid me %d\n", me);
5614 return;
5615 }
5616
5617 switch (state) {
5618 case AMDGPU_IRQ_STATE_DISABLE:
5619 mec_int_cntl = RREG32(mec_int_cntl_reg);
5620 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5621 TIME_STAMP_INT_ENABLE, 0);
5622 WREG32(mec_int_cntl_reg, mec_int_cntl);
5623 break;
5624 case AMDGPU_IRQ_STATE_ENABLE:
5625 mec_int_cntl = RREG32(mec_int_cntl_reg);
5626 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5627 TIME_STAMP_INT_ENABLE, 1);
5628 WREG32(mec_int_cntl_reg, mec_int_cntl);
5629 break;
5630 default:
5631 break;
5632 }
5633 }
5634
5635 static int gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
5636 struct amdgpu_irq_src *source,
5637 unsigned type,
5638 enum amdgpu_interrupt_state state)
5639 {
5640 switch (state) {
5641 case AMDGPU_IRQ_STATE_DISABLE:
5642 case AMDGPU_IRQ_STATE_ENABLE:
5643 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5644 PRIV_REG_INT_ENABLE,
5645 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5646 break;
5647 default:
5648 break;
5649 }
5650
5651 return 0;
5652 }
5653
5654 static int gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
5655 struct amdgpu_irq_src *source,
5656 unsigned type,
5657 enum amdgpu_interrupt_state state)
5658 {
5659 switch (state) {
5660 case AMDGPU_IRQ_STATE_DISABLE:
5661 case AMDGPU_IRQ_STATE_ENABLE:
5662 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5663 PRIV_INSTR_INT_ENABLE,
5664 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5665 default:
5666 break;
5667 }
5668
5669 return 0;
5670 }
5671
5672 #define ENABLE_ECC_ON_ME_PIPE(me, pipe) \
5673 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
5674 CP_ECC_ERROR_INT_ENABLE, 1)
5675
5676 #define DISABLE_ECC_ON_ME_PIPE(me, pipe) \
5677 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
5678 CP_ECC_ERROR_INT_ENABLE, 0)
5679
5680 static int gfx_v9_0_set_cp_ecc_error_state(struct amdgpu_device *adev,
5681 struct amdgpu_irq_src *source,
5682 unsigned type,
5683 enum amdgpu_interrupt_state state)
5684 {
5685 switch (state) {
5686 case AMDGPU_IRQ_STATE_DISABLE:
5687 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5688 CP_ECC_ERROR_INT_ENABLE, 0);
5689 DISABLE_ECC_ON_ME_PIPE(1, 0);
5690 DISABLE_ECC_ON_ME_PIPE(1, 1);
5691 DISABLE_ECC_ON_ME_PIPE(1, 2);
5692 DISABLE_ECC_ON_ME_PIPE(1, 3);
5693 break;
5694
5695 case AMDGPU_IRQ_STATE_ENABLE:
5696 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5697 CP_ECC_ERROR_INT_ENABLE, 1);
5698 ENABLE_ECC_ON_ME_PIPE(1, 0);
5699 ENABLE_ECC_ON_ME_PIPE(1, 1);
5700 ENABLE_ECC_ON_ME_PIPE(1, 2);
5701 ENABLE_ECC_ON_ME_PIPE(1, 3);
5702 break;
5703 default:
5704 break;
5705 }
5706
5707 return 0;
5708 }
5709
5710
5711 static int gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device *adev,
5712 struct amdgpu_irq_src *src,
5713 unsigned type,
5714 enum amdgpu_interrupt_state state)
5715 {
5716 switch (type) {
5717 case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
5718 gfx_v9_0_set_gfx_eop_interrupt_state(adev, state);
5719 break;
5720 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
5721 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
5722 break;
5723 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
5724 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
5725 break;
5726 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
5727 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
5728 break;
5729 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
5730 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
5731 break;
5732 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
5733 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
5734 break;
5735 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
5736 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
5737 break;
5738 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
5739 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
5740 break;
5741 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
5742 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
5743 break;
5744 default:
5745 break;
5746 }
5747 return 0;
5748 }
5749
5750 static int gfx_v9_0_eop_irq(struct amdgpu_device *adev,
5751 struct amdgpu_irq_src *source,
5752 struct amdgpu_iv_entry *entry)
5753 {
5754 int i;
5755 u8 me_id, pipe_id, queue_id;
5756 struct amdgpu_ring *ring;
5757
5758 DRM_DEBUG("IH: CP EOP\n");
5759 me_id = (entry->ring_id & 0x0c) >> 2;
5760 pipe_id = (entry->ring_id & 0x03) >> 0;
5761 queue_id = (entry->ring_id & 0x70) >> 4;
5762
5763 switch (me_id) {
5764 case 0:
5765 amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
5766 break;
5767 case 1:
5768 case 2:
5769 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5770 ring = &adev->gfx.compute_ring[i];
5771 /* Per-queue interrupt is supported for MEC starting from VI.
5772 * The interrupt can only be enabled/disabled per pipe instead of per queue.
5773 */
5774 if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
5775 amdgpu_fence_process(ring);
5776 }
5777 break;
5778 }
5779 return 0;
5780 }
5781
5782 static void gfx_v9_0_fault(struct amdgpu_device *adev,
5783 struct amdgpu_iv_entry *entry)
5784 {
5785 u8 me_id, pipe_id, queue_id;
5786 struct amdgpu_ring *ring;
5787 int i;
5788
5789 me_id = (entry->ring_id & 0x0c) >> 2;
5790 pipe_id = (entry->ring_id & 0x03) >> 0;
5791 queue_id = (entry->ring_id & 0x70) >> 4;
5792
5793 switch (me_id) {
5794 case 0:
5795 drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
5796 break;
5797 case 1:
5798 case 2:
5799 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5800 ring = &adev->gfx.compute_ring[i];
5801 if (ring->me == me_id && ring->pipe == pipe_id &&
5802 ring->queue == queue_id)
5803 drm_sched_fault(&ring->sched);
5804 }
5805 break;
5806 }
5807 }
5808
5809 static int gfx_v9_0_priv_reg_irq(struct amdgpu_device *adev,
5810 struct amdgpu_irq_src *source,
5811 struct amdgpu_iv_entry *entry)
5812 {
5813 DRM_ERROR("Illegal register access in command stream\n");
5814 gfx_v9_0_fault(adev, entry);
5815 return 0;
5816 }
5817
5818 static int gfx_v9_0_priv_inst_irq(struct amdgpu_device *adev,
5819 struct amdgpu_irq_src *source,
5820 struct amdgpu_iv_entry *entry)
5821 {
5822 DRM_ERROR("Illegal instruction in command stream\n");
5823 gfx_v9_0_fault(adev, entry);
5824 return 0;
5825 }
5826
5827
5828 static const struct soc15_ras_field_entry gfx_v9_0_ras_fields[] = {
5829 { "CPC_SCRATCH", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT),
5830 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, SEC_COUNT),
5831 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, DED_COUNT)
5832 },
5833 { "CPC_UCODE", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT),
5834 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, SEC_COUNT),
5835 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, DED_COUNT)
5836 },
5837 { "CPF_ROQ_ME1", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
5838 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME1),
5839 0, 0
5840 },
5841 { "CPF_ROQ_ME2", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
5842 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME2),
5843 0, 0
5844 },
5845 { "CPF_TAG", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT),
5846 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, SEC_COUNT),
5847 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, DED_COUNT)
5848 },
5849 { "CPG_DMA_ROQ", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
5850 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, ROQ_COUNT),
5851 0, 0
5852 },
5853 { "CPG_DMA_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
5854 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_SEC_COUNT),
5855 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_DED_COUNT)
5856 },
5857 { "CPG_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT),
5858 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, SEC_COUNT),
5859 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, DED_COUNT)
5860 },
5861 { "DC_CSINVOC", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT),
5862 SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, COUNT_ME1),
5863 0, 0
5864 },
5865 { "DC_RESTORE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT),
5866 SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, COUNT_ME1),
5867 0, 0
5868 },
5869 { "DC_STATE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT),
5870 SOC15_REG_FIELD(DC_EDC_STATE_CNT, COUNT_ME1),
5871 0, 0
5872 },
5873 { "GDS_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
5874 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_SEC),
5875 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_DED)
5876 },
5877 { "GDS_INPUT_QUEUE", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
5878 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_INPUT_QUEUE_SED),
5879 0, 0
5880 },
5881 { "GDS_ME0_CS_PIPE_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
5882 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_SEC),
5883 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_DED)
5884 },
5885 { "GDS_OA_PHY_PHY_CMD_RAM_MEM",
5886 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
5887 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_SEC),
5888 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_DED)
5889 },
5890 { "GDS_OA_PHY_PHY_DATA_RAM_MEM",
5891 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
5892 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_DATA_RAM_MEM_SED),
5893 0, 0
5894 },
5895 { "GDS_OA_PIPE_ME1_PIPE0_PIPE_MEM",
5896 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
5897 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_SEC),
5898 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_DED)
5899 },
5900 { "GDS_OA_PIPE_ME1_PIPE1_PIPE_MEM",
5901 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
5902 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_SEC),
5903 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_DED)
5904 },
5905 { "GDS_OA_PIPE_ME1_PIPE2_PIPE_MEM",
5906 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
5907 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_SEC),
5908 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_DED)
5909 },
5910 { "GDS_OA_PIPE_ME1_PIPE3_PIPE_MEM",
5911 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
5912 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_SEC),
5913 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_DED)
5914 },
5915 { "SPI_SR_MEM", SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT),
5916 SOC15_REG_FIELD(SPI_EDC_CNT, SPI_SR_MEM_SED_COUNT),
5917 0, 0
5918 },
5919 { "TA_FS_DFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
5920 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_SEC_COUNT),
5921 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_DED_COUNT)
5922 },
5923 { "TA_FS_AFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
5924 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_SED_COUNT),
5925 0, 0
5926 },
5927 { "TA_FL_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
5928 SOC15_REG_FIELD(TA_EDC_CNT, TA_FL_LFIFO_SED_COUNT),
5929 0, 0
5930 },
5931 { "TA_FX_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
5932 SOC15_REG_FIELD(TA_EDC_CNT, TA_FX_LFIFO_SED_COUNT),
5933 0, 0
5934 },
5935 { "TA_FS_CFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
5936 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_CFIFO_SED_COUNT),
5937 0, 0
5938 },
5939 { "TCA_HOLE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
5940 SOC15_REG_FIELD(TCA_EDC_CNT, HOLE_FIFO_SED_COUNT),
5941 0, 0
5942 },
5943 { "TCA_REQ_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
5944 SOC15_REG_FIELD(TCA_EDC_CNT, REQ_FIFO_SED_COUNT),
5945 0, 0
5946 },
5947 { "TCC_CACHE_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5948 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_SEC_COUNT),
5949 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_DED_COUNT)
5950 },
5951 { "TCC_CACHE_DIRTY", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5952 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_SEC_COUNT),
5953 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_DED_COUNT)
5954 },
5955 { "TCC_HIGH_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5956 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_SEC_COUNT),
5957 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_DED_COUNT)
5958 },
5959 { "TCC_LOW_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5960 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_SEC_COUNT),
5961 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_DED_COUNT)
5962 },
5963 { "TCC_SRC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5964 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_SEC_COUNT),
5965 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_DED_COUNT)
5966 },
5967 { "TCC_IN_USE_DEC", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5968 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_DEC_SED_COUNT),
5969 0, 0
5970 },
5971 { "TCC_IN_USE_TRANSFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5972 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_TRANSFER_SED_COUNT),
5973 0, 0
5974 },
5975 { "TCC_LATENCY_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5976 SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_SED_COUNT),
5977 0, 0
5978 },
5979 { "TCC_RETURN_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5980 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_DATA_SED_COUNT),
5981 0, 0
5982 },
5983 { "TCC_RETURN_CONTROL", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5984 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_CONTROL_SED_COUNT),
5985 0, 0
5986 },
5987 { "TCC_UC_ATOMIC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
5988 SOC15_REG_FIELD(TCC_EDC_CNT, UC_ATOMIC_FIFO_SED_COUNT),
5989 0, 0
5990 },
5991 { "TCC_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
5992 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_RETURN_SED_COUNT),
5993 0, 0
5994 },
5995 { "TCC_WRITE_CACHE_READ", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
5996 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_CACHE_READ_SED_COUNT),
5997 0, 0
5998 },
5999 { "TCC_SRC_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6000 SOC15_REG_FIELD(TCC_EDC_CNT2, SRC_FIFO_NEXT_RAM_SED_COUNT),
6001 0, 0
6002 },
6003 { "TCC_LATENCY_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6004 SOC15_REG_FIELD(TCC_EDC_CNT2, LATENCY_FIFO_NEXT_RAM_SED_COUNT),
6005 0, 0
6006 },
6007 { "TCC_CACHE_TAG_PROBE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6008 SOC15_REG_FIELD(TCC_EDC_CNT2, CACHE_TAG_PROBE_FIFO_SED_COUNT),
6009 0, 0
6010 },
6011 { "TCC_WRRET_TAG_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6012 SOC15_REG_FIELD(TCC_EDC_CNT2, WRRET_TAG_WRITE_RETURN_SED_COUNT),
6013 0, 0
6014 },
6015 { "TCC_ATOMIC_RETURN_BUFFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6016 SOC15_REG_FIELD(TCC_EDC_CNT2, ATOMIC_RETURN_BUFFER_SED_COUNT),
6017 0, 0
6018 },
6019 { "TCI_WRITE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT),
6020 SOC15_REG_FIELD(TCI_EDC_CNT, WRITE_RAM_SED_COUNT),
6021 0, 0
6022 },
6023 { "TCP_CACHE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6024 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_SEC_COUNT),
6025 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_DED_COUNT)
6026 },
6027 { "TCP_LFIFO_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6028 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_SEC_COUNT),
6029 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_DED_COUNT)
6030 },
6031 { "TCP_CMD_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6032 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CMD_FIFO_SED_COUNT),
6033 0, 0
6034 },
6035 { "TCP_VM_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6036 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, VM_FIFO_SEC_COUNT),
6037 0, 0
6038 },
6039 { "TCP_DB_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6040 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, DB_RAM_SED_COUNT),
6041 0, 0
6042 },
6043 { "TCP_UTCL1_LFIFO0", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6044 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_SEC_COUNT),
6045 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_DED_COUNT)
6046 },
6047 { "TCP_UTCL1_LFIFO1", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6048 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_SEC_COUNT),
6049 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_DED_COUNT)
6050 },
6051 { "TD_SS_FIFO_LO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6052 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_SEC_COUNT),
6053 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_DED_COUNT)
6054 },
6055 { "TD_SS_FIFO_HI", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6056 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_SEC_COUNT),
6057 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_DED_COUNT)
6058 },
6059 { "TD_CS_FIFO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6060 SOC15_REG_FIELD(TD_EDC_CNT, CS_FIFO_SED_COUNT),
6061 0, 0
6062 },
6063 { "SQ_LDS_D", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6064 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_SEC_COUNT),
6065 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_DED_COUNT)
6066 },
6067 { "SQ_LDS_I", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6068 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_SEC_COUNT),
6069 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_DED_COUNT)
6070 },
6071 { "SQ_SGPR", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6072 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_SEC_COUNT),
6073 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_DED_COUNT)
6074 },
6075 { "SQ_VGPR0", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6076 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_SEC_COUNT),
6077 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_DED_COUNT)
6078 },
6079 { "SQ_VGPR1", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6080 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_SEC_COUNT),
6081 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_DED_COUNT)
6082 },
6083 { "SQ_VGPR2", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6084 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_SEC_COUNT),
6085 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_DED_COUNT)
6086 },
6087 { "SQ_VGPR3", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6088 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_SEC_COUNT),
6089 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_DED_COUNT)
6090 },
6091 { "SQC_DATA_CU0_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6092 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_SEC_COUNT),
6093 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_DED_COUNT)
6094 },
6095 { "SQC_DATA_CU0_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6096 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_SEC_COUNT),
6097 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_DED_COUNT)
6098 },
6099 { "SQC_DATA_CU1_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6100 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_SEC_COUNT),
6101 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_DED_COUNT)
6102 },
6103 { "SQC_DATA_CU1_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6104 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_SEC_COUNT),
6105 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_DED_COUNT)
6106 },
6107 { "SQC_DATA_CU2_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6108 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_SEC_COUNT),
6109 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_DED_COUNT)
6110 },
6111 { "SQC_DATA_CU2_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6112 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_SEC_COUNT),
6113 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_DED_COUNT)
6114 },
6115 { "SQC_INST_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6116 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_SEC_COUNT),
6117 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_DED_COUNT)
6118 },
6119 { "SQC_INST_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6120 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_SEC_COUNT),
6121 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_DED_COUNT)
6122 },
6123 { "SQC_DATA_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6124 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_SEC_COUNT),
6125 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_DED_COUNT)
6126 },
6127 { "SQC_DATA_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6128 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_SEC_COUNT),
6129 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_DED_COUNT)
6130 },
6131 { "SQC_INST_BANKA_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6132 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_UTCL1_MISS_FIFO_SED_COUNT),
6133 0, 0
6134 },
6135 { "SQC_INST_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6136 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_MISS_FIFO_SED_COUNT),
6137 0, 0
6138 },
6139 { "SQC_DATA_BANKA_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6140 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_HIT_FIFO_SED_COUNT),
6141 0, 0
6142 },
6143 { "SQC_DATA_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6144 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_MISS_FIFO_SED_COUNT),
6145 0, 0
6146 },
6147 { "SQC_DATA_BANKA_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6148 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_DIRTY_BIT_RAM_SED_COUNT),
6149 0, 0
6150 },
6151 { "SQC_INST_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6152 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_SEC_COUNT),
6153 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_DED_COUNT)
6154 },
6155 { "SQC_INST_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6156 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_SEC_COUNT),
6157 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_DED_COUNT)
6158 },
6159 { "SQC_INST_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6160 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_SEC_COUNT),
6161 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_DED_COUNT)
6162 },
6163 { "SQC_DATA_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6164 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_SEC_COUNT),
6165 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_DED_COUNT)
6166 },
6167 { "SQC_DATA_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6168 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_SEC_COUNT),
6169 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_DED_COUNT)
6170 },
6171 { "SQC_INST_BANKB_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6172 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_UTCL1_MISS_FIFO_SED_COUNT),
6173 0, 0
6174 },
6175 { "SQC_INST_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6176 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_MISS_FIFO_SED_COUNT),
6177 0, 0
6178 },
6179 { "SQC_DATA_BANKB_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6180 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_HIT_FIFO_SED_COUNT),
6181 0, 0
6182 },
6183 { "SQC_DATA_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6184 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_MISS_FIFO_SED_COUNT),
6185 0, 0
6186 },
6187 { "SQC_DATA_BANKB_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6188 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_DIRTY_BIT_RAM_SED_COUNT),
6189 0, 0
6190 },
6191 { "EA_DRAMRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6192 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
6193 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT)
6194 },
6195 { "EA_DRAMWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6196 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
6197 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT)
6198 },
6199 { "EA_DRAMWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6200 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
6201 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT)
6202 },
6203 { "EA_RRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6204 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
6205 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_DED_COUNT)
6206 },
6207 { "EA_WRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6208 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
6209 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_DED_COUNT)
6210 },
6211 { "EA_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6212 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
6213 0, 0
6214 },
6215 { "EA_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6216 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
6217 0, 0
6218 },
6219 { "EA_IORD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6220 SOC15_REG_FIELD(GCEA_EDC_CNT, IORD_CMDMEM_SED_COUNT),
6221 0, 0
6222 },
6223 { "EA_IOWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6224 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
6225 0, 0
6226 },
6227 { "EA_IOWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6228 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_DATAMEM_SED_COUNT),
6229 0, 0
6230 },
6231 { "GMIRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6232 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
6233 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT)
6234 },
6235 { "GMIWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6236 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
6237 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT)
6238 },
6239 { "GMIWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6240 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
6241 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT)
6242 },
6243 { "GMIRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6244 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
6245 0, 0
6246 },
6247 { "GMIWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6248 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
6249 0, 0
6250 },
6251 { "MAM_D0MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6252 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D0MEM_SED_COUNT),
6253 0, 0
6254 },
6255 { "MAM_D1MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6256 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D1MEM_SED_COUNT),
6257 0, 0
6258 },
6259 { "MAM_D2MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6260 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D2MEM_SED_COUNT),
6261 0, 0
6262 },
6263 { "MAM_D3MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6264 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D3MEM_SED_COUNT),
6265 0, 0
6266 }
6267 };
6268
6269 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
6270 void *inject_if)
6271 {
6272 struct ras_inject_if *info = (struct ras_inject_if *)inject_if;
6273 int ret;
6274 struct ta_ras_trigger_error_input block_info = { 0 };
6275
6276 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6277 return -EINVAL;
6278
6279 if (info->head.sub_block_index >= ARRAY_SIZE(ras_gfx_subblocks))
6280 return -EINVAL;
6281
6282 if (!ras_gfx_subblocks[info->head.sub_block_index].name)
6283 return -EPERM;
6284
6285 if (!(ras_gfx_subblocks[info->head.sub_block_index].hw_supported_error_type &
6286 info->head.type)) {
6287 DRM_ERROR("GFX Subblock %s, hardware do not support type 0x%x\n",
6288 ras_gfx_subblocks[info->head.sub_block_index].name,
6289 info->head.type);
6290 return -EPERM;
6291 }
6292
6293 if (!(ras_gfx_subblocks[info->head.sub_block_index].sw_supported_error_type &
6294 info->head.type)) {
6295 DRM_ERROR("GFX Subblock %s, driver do not support type 0x%x\n",
6296 ras_gfx_subblocks[info->head.sub_block_index].name,
6297 info->head.type);
6298 return -EPERM;
6299 }
6300
6301 block_info.block_id = amdgpu_ras_block_to_ta(info->head.block);
6302 block_info.sub_block_index =
6303 ras_gfx_subblocks[info->head.sub_block_index].ta_subblock;
6304 block_info.inject_error_type = amdgpu_ras_error_to_ta(info->head.type);
6305 block_info.address = info->address;
6306 block_info.value = info->value;
6307
6308 mutex_lock(&adev->grbm_idx_mutex);
6309 ret = psp_ras_trigger_error(&adev->psp, &block_info);
6310 mutex_unlock(&adev->grbm_idx_mutex);
6311
6312 return ret;
6313 }
6314
6315 static const char *vml2_mems[] = {
6316 "UTC_VML2_BANK_CACHE_0_BIGK_MEM0",
6317 "UTC_VML2_BANK_CACHE_0_BIGK_MEM1",
6318 "UTC_VML2_BANK_CACHE_0_4K_MEM0",
6319 "UTC_VML2_BANK_CACHE_0_4K_MEM1",
6320 "UTC_VML2_BANK_CACHE_1_BIGK_MEM0",
6321 "UTC_VML2_BANK_CACHE_1_BIGK_MEM1",
6322 "UTC_VML2_BANK_CACHE_1_4K_MEM0",
6323 "UTC_VML2_BANK_CACHE_1_4K_MEM1",
6324 "UTC_VML2_BANK_CACHE_2_BIGK_MEM0",
6325 "UTC_VML2_BANK_CACHE_2_BIGK_MEM1",
6326 "UTC_VML2_BANK_CACHE_2_4K_MEM0",
6327 "UTC_VML2_BANK_CACHE_2_4K_MEM1",
6328 "UTC_VML2_BANK_CACHE_3_BIGK_MEM0",
6329 "UTC_VML2_BANK_CACHE_3_BIGK_MEM1",
6330 "UTC_VML2_BANK_CACHE_3_4K_MEM0",
6331 "UTC_VML2_BANK_CACHE_3_4K_MEM1",
6332 };
6333
6334 static const char *vml2_walker_mems[] = {
6335 "UTC_VML2_CACHE_PDE0_MEM0",
6336 "UTC_VML2_CACHE_PDE0_MEM1",
6337 "UTC_VML2_CACHE_PDE1_MEM0",
6338 "UTC_VML2_CACHE_PDE1_MEM1",
6339 "UTC_VML2_CACHE_PDE2_MEM0",
6340 "UTC_VML2_CACHE_PDE2_MEM1",
6341 "UTC_VML2_RDIF_LOG_FIFO",
6342 };
6343
6344 static const char *atc_l2_cache_2m_mems[] = {
6345 "UTC_ATCL2_CACHE_2M_BANK0_WAY0_MEM",
6346 "UTC_ATCL2_CACHE_2M_BANK0_WAY1_MEM",
6347 "UTC_ATCL2_CACHE_2M_BANK1_WAY0_MEM",
6348 "UTC_ATCL2_CACHE_2M_BANK1_WAY1_MEM",
6349 };
6350
6351 static const char *atc_l2_cache_4k_mems[] = {
6352 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM0",
6353 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM1",
6354 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM2",
6355 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM3",
6356 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM4",
6357 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM5",
6358 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM6",
6359 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM7",
6360 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM0",
6361 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM1",
6362 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM2",
6363 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM3",
6364 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM4",
6365 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM5",
6366 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM6",
6367 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM7",
6368 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM0",
6369 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM1",
6370 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM2",
6371 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM3",
6372 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM4",
6373 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM5",
6374 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM6",
6375 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM7",
6376 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM0",
6377 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM1",
6378 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM2",
6379 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM3",
6380 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM4",
6381 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM5",
6382 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM6",
6383 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM7",
6384 };
6385
6386 static int gfx_v9_0_query_utc_edc_status(struct amdgpu_device *adev,
6387 struct ras_err_data *err_data)
6388 {
6389 uint32_t i, data;
6390 uint32_t sec_count, ded_count;
6391
6392 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6393 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0);
6394 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6395 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0);
6396 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6397 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0);
6398 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6399 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0);
6400
6401 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) {
6402 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i);
6403 data = RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT);
6404
6405 sec_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, SEC_COUNT);
6406 if (sec_count) {
6407 DRM_INFO("Instance[%d]: SubBlock %s, SEC %d\n", i,
6408 vml2_mems[i], sec_count);
6409 err_data->ce_count += sec_count;
6410 }
6411
6412 ded_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, DED_COUNT);
6413 if (ded_count) {
6414 DRM_INFO("Instance[%d]: SubBlock %s, DED %d\n", i,
6415 vml2_mems[i], ded_count);
6416 err_data->ue_count += ded_count;
6417 }
6418 }
6419
6420 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) {
6421 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i);
6422 data = RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT);
6423
6424 sec_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6425 SEC_COUNT);
6426 if (sec_count) {
6427 DRM_INFO("Instance[%d]: SubBlock %s, SEC %d\n", i,
6428 vml2_walker_mems[i], sec_count);
6429 err_data->ce_count += sec_count;
6430 }
6431
6432 ded_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6433 DED_COUNT);
6434 if (ded_count) {
6435 DRM_INFO("Instance[%d]: SubBlock %s, DED %d\n", i,
6436 vml2_walker_mems[i], ded_count);
6437 err_data->ue_count += ded_count;
6438 }
6439 }
6440
6441 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) {
6442 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i);
6443 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT);
6444
6445 sec_count = (data & 0x00006000L) >> 0xd;
6446 if (sec_count) {
6447 DRM_INFO("Instance[%d]: SubBlock %s, SEC %d\n", i,
6448 atc_l2_cache_2m_mems[i], sec_count);
6449 err_data->ce_count += sec_count;
6450 }
6451 }
6452
6453 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) {
6454 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i);
6455 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT);
6456
6457 sec_count = (data & 0x00006000L) >> 0xd;
6458 if (sec_count) {
6459 DRM_INFO("Instance[%d]: SubBlock %s, SEC %d\n", i,
6460 atc_l2_cache_4k_mems[i], sec_count);
6461 err_data->ce_count += sec_count;
6462 }
6463
6464 ded_count = (data & 0x00018000L) >> 0xf;
6465 if (ded_count) {
6466 DRM_INFO("Instance[%d]: SubBlock %s, DED %d\n", i,
6467 atc_l2_cache_4k_mems[i], ded_count);
6468 err_data->ue_count += ded_count;
6469 }
6470 }
6471
6472 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6473 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6474 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6475 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6476
6477 return 0;
6478 }
6479
6480 static int gfx_v9_0_ras_error_count(const struct soc15_reg_entry *reg,
6481 uint32_t se_id, uint32_t inst_id, uint32_t value,
6482 uint32_t *sec_count, uint32_t *ded_count)
6483 {
6484 uint32_t i;
6485 uint32_t sec_cnt, ded_cnt;
6486
6487 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_ras_fields); i++) {
6488 if(gfx_v9_0_ras_fields[i].reg_offset != reg->reg_offset ||
6489 gfx_v9_0_ras_fields[i].seg != reg->seg ||
6490 gfx_v9_0_ras_fields[i].inst != reg->inst)
6491 continue;
6492
6493 sec_cnt = (value &
6494 gfx_v9_0_ras_fields[i].sec_count_mask) >>
6495 gfx_v9_0_ras_fields[i].sec_count_shift;
6496 if (sec_cnt) {
6497 DRM_INFO("GFX SubBlock %s, Instance[%d][%d], SEC %d\n",
6498 gfx_v9_0_ras_fields[i].name,
6499 se_id, inst_id,
6500 sec_cnt);
6501 *sec_count += sec_cnt;
6502 }
6503
6504 ded_cnt = (value &
6505 gfx_v9_0_ras_fields[i].ded_count_mask) >>
6506 gfx_v9_0_ras_fields[i].ded_count_shift;
6507 if (ded_cnt) {
6508 DRM_INFO("GFX SubBlock %s, Instance[%d][%d], DED %d\n",
6509 gfx_v9_0_ras_fields[i].name,
6510 se_id, inst_id,
6511 ded_cnt);
6512 *ded_count += ded_cnt;
6513 }
6514 }
6515
6516 return 0;
6517 }
6518
6519 static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev)
6520 {
6521 int i, j, k;
6522
6523 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6524 return;
6525
6526 /* read back registers to clear the counters */
6527 mutex_lock(&adev->grbm_idx_mutex);
6528 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) {
6529 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) {
6530 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) {
6531 gfx_v9_0_select_se_sh(adev, j, 0x0, k);
6532 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i]));
6533 }
6534 }
6535 }
6536 WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, 0xe0000000);
6537 mutex_unlock(&adev->grbm_idx_mutex);
6538
6539 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6540 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0);
6541 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6542 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0);
6543 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6544 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0);
6545 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6546 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0);
6547
6548 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) {
6549 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i);
6550 RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT);
6551 }
6552
6553 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) {
6554 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i);
6555 RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT);
6556 }
6557
6558 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) {
6559 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i);
6560 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT);
6561 }
6562
6563 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) {
6564 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i);
6565 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT);
6566 }
6567
6568 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6569 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6570 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6571 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6572 }
6573
6574 static int gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
6575 void *ras_error_status)
6576 {
6577 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
6578 uint32_t sec_count = 0, ded_count = 0;
6579 uint32_t i, j, k;
6580 uint32_t reg_value;
6581
6582 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6583 return -EINVAL;
6584
6585 err_data->ue_count = 0;
6586 err_data->ce_count = 0;
6587
6588 mutex_lock(&adev->grbm_idx_mutex);
6589
6590 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) {
6591 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) {
6592 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) {
6593 gfx_v9_0_select_se_sh(adev, j, 0, k);
6594 reg_value =
6595 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i]));
6596 if (reg_value)
6597 gfx_v9_0_ras_error_count(&gfx_v9_0_edc_counter_regs[i],
6598 j, k, reg_value,
6599 &sec_count, &ded_count);
6600 }
6601 }
6602 }
6603
6604 err_data->ce_count += sec_count;
6605 err_data->ue_count += ded_count;
6606
6607 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
6608 mutex_unlock(&adev->grbm_idx_mutex);
6609
6610 gfx_v9_0_query_utc_edc_status(adev, err_data);
6611
6612 return 0;
6613 }
6614
6615 static const struct amd_ip_funcs gfx_v9_0_ip_funcs = {
6616 .name = "gfx_v9_0",
6617 .early_init = gfx_v9_0_early_init,
6618 .late_init = gfx_v9_0_late_init,
6619 .sw_init = gfx_v9_0_sw_init,
6620 .sw_fini = gfx_v9_0_sw_fini,
6621 .hw_init = gfx_v9_0_hw_init,
6622 .hw_fini = gfx_v9_0_hw_fini,
6623 .suspend = gfx_v9_0_suspend,
6624 .resume = gfx_v9_0_resume,
6625 .is_idle = gfx_v9_0_is_idle,
6626 .wait_for_idle = gfx_v9_0_wait_for_idle,
6627 .soft_reset = gfx_v9_0_soft_reset,
6628 .set_clockgating_state = gfx_v9_0_set_clockgating_state,
6629 .set_powergating_state = gfx_v9_0_set_powergating_state,
6630 .get_clockgating_state = gfx_v9_0_get_clockgating_state,
6631 };
6632
6633 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
6634 .type = AMDGPU_RING_TYPE_GFX,
6635 .align_mask = 0xff,
6636 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
6637 .support_64bit_ptrs = true,
6638 .vmhub = AMDGPU_GFXHUB_0,
6639 .get_rptr = gfx_v9_0_ring_get_rptr_gfx,
6640 .get_wptr = gfx_v9_0_ring_get_wptr_gfx,
6641 .set_wptr = gfx_v9_0_ring_set_wptr_gfx,
6642 .emit_frame_size = /* totally 242 maximum if 16 IBs */
6643 5 + /* COND_EXEC */
6644 7 + /* PIPELINE_SYNC */
6645 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6646 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6647 2 + /* VM_FLUSH */
6648 8 + /* FENCE for VM_FLUSH */
6649 20 + /* GDS switch */
6650 4 + /* double SWITCH_BUFFER,
6651 the first COND_EXEC jump to the place just
6652 prior to this double SWITCH_BUFFER */
6653 5 + /* COND_EXEC */
6654 7 + /* HDP_flush */
6655 4 + /* VGT_flush */
6656 14 + /* CE_META */
6657 31 + /* DE_META */
6658 3 + /* CNTX_CTRL */
6659 5 + /* HDP_INVL */
6660 8 + 8 + /* FENCE x2 */
6661 2, /* SWITCH_BUFFER */
6662 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_gfx */
6663 .emit_ib = gfx_v9_0_ring_emit_ib_gfx,
6664 .emit_fence = gfx_v9_0_ring_emit_fence,
6665 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
6666 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
6667 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
6668 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
6669 .test_ring = gfx_v9_0_ring_test_ring,
6670 .test_ib = gfx_v9_0_ring_test_ib,
6671 .insert_nop = amdgpu_ring_insert_nop,
6672 .pad_ib = amdgpu_ring_generic_pad_ib,
6673 .emit_switch_buffer = gfx_v9_ring_emit_sb,
6674 .emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
6675 .init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
6676 .patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec,
6677 .emit_tmz = gfx_v9_0_ring_emit_tmz,
6678 .emit_wreg = gfx_v9_0_ring_emit_wreg,
6679 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6680 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6681 .soft_recovery = gfx_v9_0_ring_soft_recovery,
6682 };
6683
6684 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
6685 .type = AMDGPU_RING_TYPE_COMPUTE,
6686 .align_mask = 0xff,
6687 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
6688 .support_64bit_ptrs = true,
6689 .vmhub = AMDGPU_GFXHUB_0,
6690 .get_rptr = gfx_v9_0_ring_get_rptr_compute,
6691 .get_wptr = gfx_v9_0_ring_get_wptr_compute,
6692 .set_wptr = gfx_v9_0_ring_set_wptr_compute,
6693 .emit_frame_size =
6694 20 + /* gfx_v9_0_ring_emit_gds_switch */
6695 7 + /* gfx_v9_0_ring_emit_hdp_flush */
6696 5 + /* hdp invalidate */
6697 7 + /* gfx_v9_0_ring_emit_pipeline_sync */
6698 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6699 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6700 2 + /* gfx_v9_0_ring_emit_vm_flush */
6701 8 + 8 + 8, /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
6702 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */
6703 .emit_ib = gfx_v9_0_ring_emit_ib_compute,
6704 .emit_fence = gfx_v9_0_ring_emit_fence,
6705 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
6706 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
6707 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
6708 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
6709 .test_ring = gfx_v9_0_ring_test_ring,
6710 .test_ib = gfx_v9_0_ring_test_ib,
6711 .insert_nop = amdgpu_ring_insert_nop,
6712 .pad_ib = amdgpu_ring_generic_pad_ib,
6713 .emit_wreg = gfx_v9_0_ring_emit_wreg,
6714 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6715 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6716 };
6717
6718 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
6719 .type = AMDGPU_RING_TYPE_KIQ,
6720 .align_mask = 0xff,
6721 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
6722 .support_64bit_ptrs = true,
6723 .vmhub = AMDGPU_GFXHUB_0,
6724 .get_rptr = gfx_v9_0_ring_get_rptr_compute,
6725 .get_wptr = gfx_v9_0_ring_get_wptr_compute,
6726 .set_wptr = gfx_v9_0_ring_set_wptr_compute,
6727 .emit_frame_size =
6728 20 + /* gfx_v9_0_ring_emit_gds_switch */
6729 7 + /* gfx_v9_0_ring_emit_hdp_flush */
6730 5 + /* hdp invalidate */
6731 7 + /* gfx_v9_0_ring_emit_pipeline_sync */
6732 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6733 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6734 2 + /* gfx_v9_0_ring_emit_vm_flush */
6735 8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */
6736 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */
6737 .emit_fence = gfx_v9_0_ring_emit_fence_kiq,
6738 .test_ring = gfx_v9_0_ring_test_ring,
6739 .insert_nop = amdgpu_ring_insert_nop,
6740 .pad_ib = amdgpu_ring_generic_pad_ib,
6741 .emit_rreg = gfx_v9_0_ring_emit_rreg,
6742 .emit_wreg = gfx_v9_0_ring_emit_wreg,
6743 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6744 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6745 };
6746
6747 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev)
6748 {
6749 int i;
6750
6751 adev->gfx.kiq.ring.funcs = &gfx_v9_0_ring_funcs_kiq;
6752
6753 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
6754 adev->gfx.gfx_ring[i].funcs = &gfx_v9_0_ring_funcs_gfx;
6755
6756 for (i = 0; i < adev->gfx.num_compute_rings; i++)
6757 adev->gfx.compute_ring[i].funcs = &gfx_v9_0_ring_funcs_compute;
6758 }
6759
6760 static const struct amdgpu_irq_src_funcs gfx_v9_0_eop_irq_funcs = {
6761 .set = gfx_v9_0_set_eop_interrupt_state,
6762 .process = gfx_v9_0_eop_irq,
6763 };
6764
6765 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_reg_irq_funcs = {
6766 .set = gfx_v9_0_set_priv_reg_fault_state,
6767 .process = gfx_v9_0_priv_reg_irq,
6768 };
6769
6770 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_inst_irq_funcs = {
6771 .set = gfx_v9_0_set_priv_inst_fault_state,
6772 .process = gfx_v9_0_priv_inst_irq,
6773 };
6774
6775 static const struct amdgpu_irq_src_funcs gfx_v9_0_cp_ecc_error_irq_funcs = {
6776 .set = gfx_v9_0_set_cp_ecc_error_state,
6777 .process = amdgpu_gfx_cp_ecc_error_irq,
6778 };
6779
6780
6781 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev)
6782 {
6783 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
6784 adev->gfx.eop_irq.funcs = &gfx_v9_0_eop_irq_funcs;
6785
6786 adev->gfx.priv_reg_irq.num_types = 1;
6787 adev->gfx.priv_reg_irq.funcs = &gfx_v9_0_priv_reg_irq_funcs;
6788
6789 adev->gfx.priv_inst_irq.num_types = 1;
6790 adev->gfx.priv_inst_irq.funcs = &gfx_v9_0_priv_inst_irq_funcs;
6791
6792 adev->gfx.cp_ecc_error_irq.num_types = 2; /*C5 ECC error and C9 FUE error*/
6793 adev->gfx.cp_ecc_error_irq.funcs = &gfx_v9_0_cp_ecc_error_irq_funcs;
6794 }
6795
6796 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev)
6797 {
6798 switch (adev->asic_type) {
6799 case CHIP_VEGA10:
6800 case CHIP_VEGA12:
6801 case CHIP_VEGA20:
6802 case CHIP_RAVEN:
6803 case CHIP_ARCTURUS:
6804 case CHIP_RENOIR:
6805 adev->gfx.rlc.funcs = &gfx_v9_0_rlc_funcs;
6806 break;
6807 default:
6808 break;
6809 }
6810 }
6811
6812 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev)
6813 {
6814 /* init asci gds info */
6815 switch (adev->asic_type) {
6816 case CHIP_VEGA10:
6817 case CHIP_VEGA12:
6818 case CHIP_VEGA20:
6819 adev->gds.gds_size = 0x10000;
6820 break;
6821 case CHIP_RAVEN:
6822 case CHIP_ARCTURUS:
6823 adev->gds.gds_size = 0x1000;
6824 break;
6825 default:
6826 adev->gds.gds_size = 0x10000;
6827 break;
6828 }
6829
6830 switch (adev->asic_type) {
6831 case CHIP_VEGA10:
6832 case CHIP_VEGA20:
6833 adev->gds.gds_compute_max_wave_id = 0x7ff;
6834 break;
6835 case CHIP_VEGA12:
6836 adev->gds.gds_compute_max_wave_id = 0x27f;
6837 break;
6838 case CHIP_RAVEN:
6839 if (adev->rev_id >= 0x8)
6840 adev->gds.gds_compute_max_wave_id = 0x77; /* raven2 */
6841 else
6842 adev->gds.gds_compute_max_wave_id = 0x15f; /* raven1 */
6843 break;
6844 case CHIP_ARCTURUS:
6845 adev->gds.gds_compute_max_wave_id = 0xfff;
6846 break;
6847 default:
6848 /* this really depends on the chip */
6849 adev->gds.gds_compute_max_wave_id = 0x7ff;
6850 break;
6851 }
6852
6853 adev->gds.gws_size = 64;
6854 adev->gds.oa_size = 16;
6855 }
6856
6857 static void gfx_v9_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
6858 u32 bitmap)
6859 {
6860 u32 data;
6861
6862 if (!bitmap)
6863 return;
6864
6865 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
6866 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
6867
6868 WREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG, data);
6869 }
6870
6871 static u32 gfx_v9_0_get_cu_active_bitmap(struct amdgpu_device *adev)
6872 {
6873 u32 data, mask;
6874
6875 data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
6876 data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
6877
6878 data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
6879 data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
6880
6881 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
6882
6883 return (~data) & mask;
6884 }
6885
6886 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
6887 struct amdgpu_cu_info *cu_info)
6888 {
6889 int i, j, k, counter, active_cu_number = 0;
6890 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
6891 unsigned disable_masks[4 * 4];
6892
6893 if (!adev || !cu_info)
6894 return -EINVAL;
6895
6896 /*
6897 * 16 comes from bitmap array size 4*4, and it can cover all gfx9 ASICs
6898 */
6899 if (adev->gfx.config.max_shader_engines *
6900 adev->gfx.config.max_sh_per_se > 16)
6901 return -EINVAL;
6902
6903 amdgpu_gfx_parse_disable_cu(disable_masks,
6904 adev->gfx.config.max_shader_engines,
6905 adev->gfx.config.max_sh_per_se);
6906
6907 mutex_lock(&adev->grbm_idx_mutex);
6908 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
6909 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
6910 mask = 1;
6911 ao_bitmap = 0;
6912 counter = 0;
6913 gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
6914 gfx_v9_0_set_user_cu_inactive_bitmap(
6915 adev, disable_masks[i * adev->gfx.config.max_sh_per_se + j]);
6916 bitmap = gfx_v9_0_get_cu_active_bitmap(adev);
6917
6918 /*
6919 * The bitmap(and ao_cu_bitmap) in cu_info structure is
6920 * 4x4 size array, and it's usually suitable for Vega
6921 * ASICs which has 4*2 SE/SH layout.
6922 * But for Arcturus, SE/SH layout is changed to 8*1.
6923 * To mostly reduce the impact, we make it compatible
6924 * with current bitmap array as below:
6925 * SE4,SH0 --> bitmap[0][1]
6926 * SE5,SH0 --> bitmap[1][1]
6927 * SE6,SH0 --> bitmap[2][1]
6928 * SE7,SH0 --> bitmap[3][1]
6929 */
6930 cu_info->bitmap[i % 4][j + i / 4] = bitmap;
6931
6932 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
6933 if (bitmap & mask) {
6934 if (counter < adev->gfx.config.max_cu_per_sh)
6935 ao_bitmap |= mask;
6936 counter ++;
6937 }
6938 mask <<= 1;
6939 }
6940 active_cu_number += counter;
6941 if (i < 2 && j < 2)
6942 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
6943 cu_info->ao_cu_bitmap[i % 4][j + i / 4] = ao_bitmap;
6944 }
6945 }
6946 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
6947 mutex_unlock(&adev->grbm_idx_mutex);
6948
6949 cu_info->number = active_cu_number;
6950 cu_info->ao_cu_mask = ao_cu_mask;
6951 cu_info->simd_per_cu = NUM_SIMD_PER_CU;
6952
6953 return 0;
6954 }
6955
6956 const struct amdgpu_ip_block_version gfx_v9_0_ip_block =
6957 {
6958 .type = AMD_IP_BLOCK_TYPE_GFX,
6959 .major = 9,
6960 .minor = 0,
6961 .rev = 0,
6962 .funcs = &gfx_v9_0_ip_funcs,
6963 };
A mirror of Linus' kernel repository