1 /*
2 * Copyright 2016 Advanced Micro Devices, Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22 */
23
24 #include <linux/delay.h>
25 #include <linux/kernel.h>
26 #include <linux/firmware.h>
27 #include <linux/module.h>
28 #include <linux/pci.h>
29
30 #include "amdgpu.h"
31 #include "amdgpu_gfx.h"
32 #include "soc15.h"
33 #include "soc15d.h"
34 #include "amdgpu_atomfirmware.h"
35 #include "amdgpu_pm.h"
36
37 #include "gc/gc_9_0_offset.h"
38 #include "gc/gc_9_0_sh_mask.h"
39
40 #include "vega10_enum.h"
41
42 #include "soc15_common.h"
43 #include "clearstate_gfx9.h"
44 #include "v9_structs.h"
45
46 #include "ivsrcid/gfx/irqsrcs_gfx_9_0.h"
47
48 #include "amdgpu_ras.h"
49
50 #include "amdgpu_ring_mux.h"
51 #include "gfx_v9_4.h"
52 #include "gfx_v9_0.h"
53 #include "gfx_v9_4_2.h"
54
55 #include "asic_reg/pwr/pwr_10_0_offset.h"
56 #include "asic_reg/pwr/pwr_10_0_sh_mask.h"
57 #include "asic_reg/gc/gc_9_0_default.h"
58
59 #define GFX9_NUM_GFX_RINGS 1
60 #define GFX9_NUM_SW_GFX_RINGS 2
61 #define GFX9_MEC_HPD_SIZE 4096
62 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L
63 #define RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET 0x00000000L
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_rlc.bin");
114
115 MODULE_FIRMWARE("amdgpu/renoir_ce.bin");
116 MODULE_FIRMWARE("amdgpu/renoir_pfp.bin");
117 MODULE_FIRMWARE("amdgpu/renoir_me.bin");
118 MODULE_FIRMWARE("amdgpu/renoir_mec.bin");
119 MODULE_FIRMWARE("amdgpu/renoir_rlc.bin");
120
121 MODULE_FIRMWARE("amdgpu/green_sardine_ce.bin");
122 MODULE_FIRMWARE("amdgpu/green_sardine_pfp.bin");
123 MODULE_FIRMWARE("amdgpu/green_sardine_me.bin");
124 MODULE_FIRMWARE("amdgpu/green_sardine_mec.bin");
125 MODULE_FIRMWARE("amdgpu/green_sardine_mec2.bin");
126 MODULE_FIRMWARE("amdgpu/green_sardine_rlc.bin");
127
128 MODULE_FIRMWARE("amdgpu/aldebaran_mec.bin");
129 MODULE_FIRMWARE("amdgpu/aldebaran_mec2.bin");
130 MODULE_FIRMWARE("amdgpu/aldebaran_rlc.bin");
131 MODULE_FIRMWARE("amdgpu/aldebaran_sjt_mec.bin");
132 MODULE_FIRMWARE("amdgpu/aldebaran_sjt_mec2.bin");
133
134 #define mmTCP_CHAN_STEER_0_ARCT 0x0b03
135 #define mmTCP_CHAN_STEER_0_ARCT_BASE_IDX 0
136 #define mmTCP_CHAN_STEER_1_ARCT 0x0b04
137 #define mmTCP_CHAN_STEER_1_ARCT_BASE_IDX 0
138 #define mmTCP_CHAN_STEER_2_ARCT 0x0b09
139 #define mmTCP_CHAN_STEER_2_ARCT_BASE_IDX 0
140 #define mmTCP_CHAN_STEER_3_ARCT 0x0b0a
141 #define mmTCP_CHAN_STEER_3_ARCT_BASE_IDX 0
142 #define mmTCP_CHAN_STEER_4_ARCT 0x0b0b
143 #define mmTCP_CHAN_STEER_4_ARCT_BASE_IDX 0
144 #define mmTCP_CHAN_STEER_5_ARCT 0x0b0c
145 #define mmTCP_CHAN_STEER_5_ARCT_BASE_IDX 0
146
147 #define mmGOLDEN_TSC_COUNT_UPPER_Renoir 0x0025
148 #define mmGOLDEN_TSC_COUNT_UPPER_Renoir_BASE_IDX 1
149 #define mmGOLDEN_TSC_COUNT_LOWER_Renoir 0x0026
150 #define mmGOLDEN_TSC_COUNT_LOWER_Renoir_BASE_IDX 1
151
152 enum ta_ras_gfx_subblock {
153 /*CPC*/
154 TA_RAS_BLOCK__GFX_CPC_INDEX_START = 0,
155 TA_RAS_BLOCK__GFX_CPC_SCRATCH = TA_RAS_BLOCK__GFX_CPC_INDEX_START,
156 TA_RAS_BLOCK__GFX_CPC_UCODE,
157 TA_RAS_BLOCK__GFX_DC_STATE_ME1,
158 TA_RAS_BLOCK__GFX_DC_CSINVOC_ME1,
159 TA_RAS_BLOCK__GFX_DC_RESTORE_ME1,
160 TA_RAS_BLOCK__GFX_DC_STATE_ME2,
161 TA_RAS_BLOCK__GFX_DC_CSINVOC_ME2,
162 TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
163 TA_RAS_BLOCK__GFX_CPC_INDEX_END = TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
164 /* CPF*/
165 TA_RAS_BLOCK__GFX_CPF_INDEX_START,
166 TA_RAS_BLOCK__GFX_CPF_ROQ_ME2 = TA_RAS_BLOCK__GFX_CPF_INDEX_START,
167 TA_RAS_BLOCK__GFX_CPF_ROQ_ME1,
168 TA_RAS_BLOCK__GFX_CPF_TAG,
169 TA_RAS_BLOCK__GFX_CPF_INDEX_END = TA_RAS_BLOCK__GFX_CPF_TAG,
170 /* CPG*/
171 TA_RAS_BLOCK__GFX_CPG_INDEX_START,
172 TA_RAS_BLOCK__GFX_CPG_DMA_ROQ = TA_RAS_BLOCK__GFX_CPG_INDEX_START,
173 TA_RAS_BLOCK__GFX_CPG_DMA_TAG,
174 TA_RAS_BLOCK__GFX_CPG_TAG,
175 TA_RAS_BLOCK__GFX_CPG_INDEX_END = TA_RAS_BLOCK__GFX_CPG_TAG,
176 /* GDS*/
177 TA_RAS_BLOCK__GFX_GDS_INDEX_START,
178 TA_RAS_BLOCK__GFX_GDS_MEM = TA_RAS_BLOCK__GFX_GDS_INDEX_START,
179 TA_RAS_BLOCK__GFX_GDS_INPUT_QUEUE,
180 TA_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM,
181 TA_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM,
182 TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
183 TA_RAS_BLOCK__GFX_GDS_INDEX_END = TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
184 /* SPI*/
185 TA_RAS_BLOCK__GFX_SPI_SR_MEM,
186 /* SQ*/
187 TA_RAS_BLOCK__GFX_SQ_INDEX_START,
188 TA_RAS_BLOCK__GFX_SQ_SGPR = TA_RAS_BLOCK__GFX_SQ_INDEX_START,
189 TA_RAS_BLOCK__GFX_SQ_LDS_D,
190 TA_RAS_BLOCK__GFX_SQ_LDS_I,
191 TA_RAS_BLOCK__GFX_SQ_VGPR, /* VGPR = SP*/
192 TA_RAS_BLOCK__GFX_SQ_INDEX_END = TA_RAS_BLOCK__GFX_SQ_VGPR,
193 /* SQC (3 ranges)*/
194 TA_RAS_BLOCK__GFX_SQC_INDEX_START,
195 /* SQC range 0*/
196 TA_RAS_BLOCK__GFX_SQC_INDEX0_START = TA_RAS_BLOCK__GFX_SQC_INDEX_START,
197 TA_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO =
198 TA_RAS_BLOCK__GFX_SQC_INDEX0_START,
199 TA_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF,
200 TA_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO,
201 TA_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF,
202 TA_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO,
203 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF,
204 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
205 TA_RAS_BLOCK__GFX_SQC_INDEX0_END =
206 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
207 /* SQC range 1*/
208 TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
209 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM =
210 TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
211 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO,
212 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO,
213 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM,
214 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM,
215 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO,
216 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO,
217 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM,
218 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
219 TA_RAS_BLOCK__GFX_SQC_INDEX1_END =
220 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
221 /* SQC range 2*/
222 TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
223 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM =
224 TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
225 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO,
226 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO,
227 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM,
228 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM,
229 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO,
230 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO,
231 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM,
232 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
233 TA_RAS_BLOCK__GFX_SQC_INDEX2_END =
234 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
235 TA_RAS_BLOCK__GFX_SQC_INDEX_END = TA_RAS_BLOCK__GFX_SQC_INDEX2_END,
236 /* TA*/
237 TA_RAS_BLOCK__GFX_TA_INDEX_START,
238 TA_RAS_BLOCK__GFX_TA_FS_DFIFO = TA_RAS_BLOCK__GFX_TA_INDEX_START,
239 TA_RAS_BLOCK__GFX_TA_FS_AFIFO,
240 TA_RAS_BLOCK__GFX_TA_FL_LFIFO,
241 TA_RAS_BLOCK__GFX_TA_FX_LFIFO,
242 TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
243 TA_RAS_BLOCK__GFX_TA_INDEX_END = TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
244 /* TCA*/
245 TA_RAS_BLOCK__GFX_TCA_INDEX_START,
246 TA_RAS_BLOCK__GFX_TCA_HOLE_FIFO = TA_RAS_BLOCK__GFX_TCA_INDEX_START,
247 TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
248 TA_RAS_BLOCK__GFX_TCA_INDEX_END = TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
249 /* TCC (5 sub-ranges)*/
250 TA_RAS_BLOCK__GFX_TCC_INDEX_START,
251 /* TCC range 0*/
252 TA_RAS_BLOCK__GFX_TCC_INDEX0_START = TA_RAS_BLOCK__GFX_TCC_INDEX_START,
253 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX0_START,
254 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1,
255 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0,
256 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1,
257 TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0,
258 TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1,
259 TA_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG,
260 TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
261 TA_RAS_BLOCK__GFX_TCC_INDEX0_END = TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
262 /* TCC range 1*/
263 TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
264 TA_RAS_BLOCK__GFX_TCC_IN_USE_DEC = TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
265 TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
266 TA_RAS_BLOCK__GFX_TCC_INDEX1_END =
267 TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
268 /* TCC range 2*/
269 TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
270 TA_RAS_BLOCK__GFX_TCC_RETURN_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
271 TA_RAS_BLOCK__GFX_TCC_RETURN_CONTROL,
272 TA_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO,
273 TA_RAS_BLOCK__GFX_TCC_WRITE_RETURN,
274 TA_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ,
275 TA_RAS_BLOCK__GFX_TCC_SRC_FIFO,
276 TA_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM,
277 TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
278 TA_RAS_BLOCK__GFX_TCC_INDEX2_END =
279 TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
280 /* TCC range 3*/
281 TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
282 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO = TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
283 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
284 TA_RAS_BLOCK__GFX_TCC_INDEX3_END =
285 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
286 /* TCC range 4*/
287 TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
288 TA_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN =
289 TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
290 TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
291 TA_RAS_BLOCK__GFX_TCC_INDEX4_END =
292 TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
293 TA_RAS_BLOCK__GFX_TCC_INDEX_END = TA_RAS_BLOCK__GFX_TCC_INDEX4_END,
294 /* TCI*/
295 TA_RAS_BLOCK__GFX_TCI_WRITE_RAM,
296 /* TCP*/
297 TA_RAS_BLOCK__GFX_TCP_INDEX_START,
298 TA_RAS_BLOCK__GFX_TCP_CACHE_RAM = TA_RAS_BLOCK__GFX_TCP_INDEX_START,
299 TA_RAS_BLOCK__GFX_TCP_LFIFO_RAM,
300 TA_RAS_BLOCK__GFX_TCP_CMD_FIFO,
301 TA_RAS_BLOCK__GFX_TCP_VM_FIFO,
302 TA_RAS_BLOCK__GFX_TCP_DB_RAM,
303 TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0,
304 TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
305 TA_RAS_BLOCK__GFX_TCP_INDEX_END = TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
306 /* TD*/
307 TA_RAS_BLOCK__GFX_TD_INDEX_START,
308 TA_RAS_BLOCK__GFX_TD_SS_FIFO_LO = TA_RAS_BLOCK__GFX_TD_INDEX_START,
309 TA_RAS_BLOCK__GFX_TD_SS_FIFO_HI,
310 TA_RAS_BLOCK__GFX_TD_CS_FIFO,
311 TA_RAS_BLOCK__GFX_TD_INDEX_END = TA_RAS_BLOCK__GFX_TD_CS_FIFO,
312 /* EA (3 sub-ranges)*/
313 TA_RAS_BLOCK__GFX_EA_INDEX_START,
314 /* EA range 0*/
315 TA_RAS_BLOCK__GFX_EA_INDEX0_START = TA_RAS_BLOCK__GFX_EA_INDEX_START,
316 TA_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM = TA_RAS_BLOCK__GFX_EA_INDEX0_START,
317 TA_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM,
318 TA_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM,
319 TA_RAS_BLOCK__GFX_EA_RRET_TAGMEM,
320 TA_RAS_BLOCK__GFX_EA_WRET_TAGMEM,
321 TA_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM,
322 TA_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM,
323 TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
324 TA_RAS_BLOCK__GFX_EA_INDEX0_END = TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
325 /* EA range 1*/
326 TA_RAS_BLOCK__GFX_EA_INDEX1_START,
327 TA_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM = TA_RAS_BLOCK__GFX_EA_INDEX1_START,
328 TA_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM,
329 TA_RAS_BLOCK__GFX_EA_IORD_CMDMEM,
330 TA_RAS_BLOCK__GFX_EA_IOWR_CMDMEM,
331 TA_RAS_BLOCK__GFX_EA_IOWR_DATAMEM,
332 TA_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM,
333 TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
334 TA_RAS_BLOCK__GFX_EA_INDEX1_END = TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
335 /* EA range 2*/
336 TA_RAS_BLOCK__GFX_EA_INDEX2_START,
337 TA_RAS_BLOCK__GFX_EA_MAM_D0MEM = TA_RAS_BLOCK__GFX_EA_INDEX2_START,
338 TA_RAS_BLOCK__GFX_EA_MAM_D1MEM,
339 TA_RAS_BLOCK__GFX_EA_MAM_D2MEM,
340 TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
341 TA_RAS_BLOCK__GFX_EA_INDEX2_END = TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
342 TA_RAS_BLOCK__GFX_EA_INDEX_END = TA_RAS_BLOCK__GFX_EA_INDEX2_END,
343 /* UTC VM L2 bank*/
344 TA_RAS_BLOCK__UTC_VML2_BANK_CACHE,
345 /* UTC VM walker*/
346 TA_RAS_BLOCK__UTC_VML2_WALKER,
347 /* UTC ATC L2 2MB cache*/
348 TA_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK,
349 /* UTC ATC L2 4KB cache*/
350 TA_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK,
351 TA_RAS_BLOCK__GFX_MAX
352 };
353
354 struct ras_gfx_subblock {
355 unsigned char *name;
356 int ta_subblock;
357 int hw_supported_error_type;
358 int sw_supported_error_type;
359 };
360
361 #define AMDGPU_RAS_SUB_BLOCK(subblock, a, b, c, d, e, f, g, h) \
362 [AMDGPU_RAS_BLOCK__##subblock] = { \
363 #subblock, \
364 TA_RAS_BLOCK__##subblock, \
365 ((a) | ((b) << 1) | ((c) << 2) | ((d) << 3)), \
366 (((e) << 1) | ((f) << 3) | (g) | ((h) << 2)), \
367 }
368
369 static const struct ras_gfx_subblock ras_gfx_subblocks[] = {
370 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_SCRATCH, 0, 1, 1, 1, 1, 0, 0, 1),
371 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_UCODE, 0, 1, 1, 1, 1, 0, 0, 1),
372 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME1, 1, 0, 0, 1, 0, 0, 1, 0),
373 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME1, 1, 0, 0, 1, 0, 0, 0, 0),
374 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME1, 1, 0, 0, 1, 0, 0, 0, 0),
375 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
376 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
377 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
378 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME2, 1, 0, 0, 1, 0, 0, 0, 0),
379 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME1, 1, 0, 0, 1, 0, 0, 1, 0),
380 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_TAG, 0, 1, 1, 1, 1, 0, 0, 1),
381 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_ROQ, 1, 0, 0, 1, 0, 0, 1, 0),
382 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_TAG, 0, 1, 1, 1, 0, 1, 0, 1),
383 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_TAG, 0, 1, 1, 1, 1, 1, 0, 1),
384 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_MEM, 0, 1, 1, 1, 0, 0, 0, 0),
385 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_INPUT_QUEUE, 1, 0, 0, 1, 0, 0, 0, 0),
386 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_CMD_RAM_MEM, 0, 1, 1, 1, 0, 0, 0,
387 0),
388 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_DATA_RAM_MEM, 1, 0, 0, 1, 0, 0, 0,
389 0),
390 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PIPE_MEM, 0, 1, 1, 1, 0, 0, 0, 0),
391 AMDGPU_RAS_SUB_BLOCK(GFX_SPI_SR_MEM, 1, 0, 0, 1, 0, 0, 0, 0),
392 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_SGPR, 0, 1, 1, 1, 0, 0, 0, 0),
393 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_D, 0, 1, 1, 1, 1, 0, 0, 1),
394 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_I, 0, 1, 1, 1, 0, 0, 0, 0),
395 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_VGPR, 0, 1, 1, 1, 0, 0, 0, 0),
396 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 1),
397 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
398 0, 0),
399 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,
400 0),
401 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
402 0, 0),
403 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_UTCL1_LFIFO, 0, 1, 1, 1, 1, 0, 0,
404 0),
405 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,
406 0, 0),
407 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,
408 0),
409 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,
410 1),
411 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,
412 0, 0, 0),
413 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
414 0),
415 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
416 0),
417 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,
418 0),
419 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,
420 0),
421 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
422 0),
423 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,
424 0, 0),
425 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
426 0),
427 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,
428 0),
429 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,
430 0, 0, 0),
431 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
432 0),
433 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
434 0),
435 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,
436 0),
437 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,
438 0),
439 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,
440 0),
441 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,
442 0, 0),
443 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,
444 0),
445 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_DFIFO, 0, 1, 1, 1, 1, 0, 0, 1),
446 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_AFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
447 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FL_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
448 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FX_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
449 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_CFIFO, 1, 0, 0, 1, 0, 0, 0, 0),
450 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_HOLE_FIFO, 1, 0, 0, 1, 0, 1, 1, 0),
451 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_REQ_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
452 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA, 0, 1, 1, 1, 1, 0, 0, 1),
453 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_0_1, 0, 1, 1, 1, 1, 0, 0,
454 1),
455 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_0, 0, 1, 1, 1, 1, 0, 0,
456 1),
457 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_1, 0, 1, 1, 1, 1, 0, 0,
458 1),
459 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_0, 0, 1, 1, 1, 0, 0, 0,
460 0),
461 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_1, 0, 1, 1, 1, 0, 0, 0,
462 0),
463 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_HIGH_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0),
464 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LOW_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0),
465 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_DEC, 1, 0, 0, 1, 0, 0, 0, 0),
466 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_TRANSFER, 1, 0, 0, 1, 0, 0, 0, 0),
467 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_DATA, 1, 0, 0, 1, 0, 0, 0, 0),
468 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_CONTROL, 1, 0, 0, 1, 0, 0, 0, 0),
469 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_UC_ATOMIC_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
470 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_RETURN, 1, 0, 0, 1, 0, 1, 1, 0),
471 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_CACHE_READ, 1, 0, 0, 1, 0, 0, 0, 0),
472 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO, 0, 1, 1, 1, 0, 0, 0, 0),
473 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 1, 0),
474 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_TAG_PROBE_FIFO, 1, 0, 0, 1, 0, 0, 0,
475 0),
476 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
477 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 0,
478 0),
479 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRRET_TAG_WRITE_RETURN, 1, 0, 0, 1, 0, 0,
480 0, 0),
481 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_ATOMIC_RETURN_BUFFER, 1, 0, 0, 1, 0, 0, 0,
482 0),
483 AMDGPU_RAS_SUB_BLOCK(GFX_TCI_WRITE_RAM, 1, 0, 0, 1, 0, 0, 0, 0),
484 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CACHE_RAM, 0, 1, 1, 1, 1, 0, 0, 1),
485 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_LFIFO_RAM, 0, 1, 1, 1, 0, 0, 0, 0),
486 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CMD_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
487 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_VM_FIFO, 0, 1, 1, 1, 0, 0, 0, 0),
488 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_DB_RAM, 1, 0, 0, 1, 0, 0, 0, 0),
489 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO0, 0, 1, 1, 1, 0, 0, 0, 0),
490 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO1, 0, 1, 1, 1, 0, 0, 0, 0),
491 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_LO, 0, 1, 1, 1, 1, 0, 0, 1),
492 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_HI, 0, 1, 1, 1, 0, 0, 0, 0),
493 AMDGPU_RAS_SUB_BLOCK(GFX_TD_CS_FIFO, 1, 0, 0, 1, 0, 0, 0, 0),
494 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_CMDMEM, 0, 1, 1, 1, 1, 0, 0, 1),
495 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
496 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0),
497 AMDGPU_RAS_SUB_BLOCK(GFX_EA_RRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0),
498 AMDGPU_RAS_SUB_BLOCK(GFX_EA_WRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0),
499 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
500 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0),
501 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0),
502 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
503 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
504 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IORD_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0),
505 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0),
506 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_DATAMEM, 1, 0, 0, 1, 0, 0, 0, 0),
507 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
508 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0),
509 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D0MEM, 1, 0, 0, 1, 0, 0, 0, 0),
510 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D1MEM, 1, 0, 0, 1, 0, 0, 0, 0),
511 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D2MEM, 1, 0, 0, 1, 0, 0, 0, 0),
512 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D3MEM, 1, 0, 0, 1, 0, 0, 0, 0),
513 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_BANK_CACHE, 0, 1, 1, 1, 0, 0, 0, 0),
514 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_WALKER, 0, 1, 1, 1, 0, 0, 0, 0),
515 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_2M_BANK, 1, 0, 0, 1, 0, 0, 0, 0),
516 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_4K_BANK, 0, 1, 1, 1, 0, 0, 0, 0),
517 };
518
519 static const struct soc15_reg_golden golden_settings_gc_9_0[] =
520 {
521 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400),
522 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG3, 0x80000000, 0x80000000),
523 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
524 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
525 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
526 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
527 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
528 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
529 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
530 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x00ffff87),
531 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x00ffff8f),
532 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
533 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
534 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
535 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
536 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
537 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff),
538 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
539 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
540 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
541 };
542
543 static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] =
544 {
545 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0x0000f000, 0x00012107),
546 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
547 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
548 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
549 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
550 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
551 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x2a114042),
552 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
553 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0x00008000, 0x00048000),
554 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
555 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
556 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
557 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
558 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
559 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
560 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x01000107),
561 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x00001800, 0x00000800),
562 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080)
563 };
564
565 static const struct soc15_reg_golden golden_settings_gc_9_0_vg20[] =
566 {
567 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x0f000080, 0x04000080),
568 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
569 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
570 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x22014042),
571 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x22014042),
572 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0x00003e00, 0x00000400),
573 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xff840000, 0x04040000),
574 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00030000),
575 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff010f, 0x01000107),
576 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0x000b0000, 0x000b0000),
577 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01000000, 0x01000000)
578 };
579
580 static const struct soc15_reg_golden golden_settings_gc_9_1[] =
581 {
582 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
583 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080),
584 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080),
585 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080),
586 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
587 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
588 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080),
589 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
590 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
591 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
592 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080),
593 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080),
594 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080),
595 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080),
596 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080),
597 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
598 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
599 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120),
600 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
601 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000000ff),
602 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080),
603 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
604 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
605 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
606 };
607
608 static const struct soc15_reg_golden golden_settings_gc_9_1_rv1[] =
609 {
610 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000),
611 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24000042),
612 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24000042),
613 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04048000),
614 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_MODE_CNTL_1, 0x06000000, 0x06000000),
615 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000),
616 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x00000800)
617 };
618
619 static const struct soc15_reg_golden golden_settings_gc_9_1_rv2[] =
620 {
621 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0xff7fffff, 0x04000000),
622 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
623 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000),
624 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x7f0fffff, 0x08000080),
625 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0xff8fffff, 0x08000080),
626 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x7f8fffff, 0x08000080),
627 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x26013041),
628 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x26013041),
629 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x3f8fffff, 0x08000080),
630 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
631 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0xff0fffff, 0x08000080),
632 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0xff0fffff, 0x08000080),
633 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0xff0fffff, 0x08000080),
634 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0xff0fffff, 0x08000080),
635 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0xff0fffff, 0x08000080),
636 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
637 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010),
638 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000),
639 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x3f8fffff, 0x08000080),
640 };
641
642 static const struct soc15_reg_golden golden_settings_gc_9_1_rn[] =
643 {
644 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
645 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000),
646 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400),
647 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x24000042),
648 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x24000042),
649 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
650 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
651 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
652 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
653 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
654 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120),
655 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGCEA_PROBE_MAP, 0xffffffff, 0x0000cccc),
656 };
657
658 static const struct soc15_reg_golden golden_settings_gc_9_x_common[] =
659 {
660 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0xffffffff, 0x000001ff),
661 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_INDEX, 0xffffffff, 0x00000000),
662 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_DATA, 0xffffffff, 0x2544c382)
663 };
664
665 static const struct soc15_reg_golden golden_settings_gc_9_2_1[] =
666 {
667 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420),
668 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000),
669 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024),
670 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001),
671 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
672 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000),
673 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800),
674 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800),
675 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87),
676 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f),
677 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000),
678 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000),
679 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68),
680 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197),
681 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000),
682 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff)
683 };
684
685 static const struct soc15_reg_golden golden_settings_gc_9_2_1_vg12[] =
686 {
687 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x00000080, 0x04000080),
688 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104),
689 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000),
690 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24104041),
691 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24104041),
692 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000),
693 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff03ff, 0x01000107),
694 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000),
695 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x76325410),
696 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000),
697 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800),
698 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800),
699 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000)
700 };
701
702 static const struct soc15_reg_golden golden_settings_gc_9_4_1_arct[] =
703 {
704 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042),
705 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x10b0000),
706 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_0_ARCT, 0x3fffffff, 0x346f0a4e),
707 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_1_ARCT, 0x3fffffff, 0x1c642ca),
708 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_2_ARCT, 0x3fffffff, 0x26f45098),
709 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_3_ARCT, 0x3fffffff, 0x2ebd9fe3),
710 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_4_ARCT, 0x3fffffff, 0xb90f5b1),
711 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_5_ARCT, 0x3ff, 0x135),
712 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_CONFIG, 0xffffffff, 0x011A0000),
713 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_FIFO_SIZES, 0xffffffff, 0x00000f00),
714 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_UTCL1_CNTL1, 0x30000000, 0x30000000)
715 };
716
717 static const struct soc15_reg_rlcg rlcg_access_gc_9_0[] = {
718 {SOC15_REG_ENTRY(GC, 0, mmGRBM_GFX_INDEX)},
719 {SOC15_REG_ENTRY(GC, 0, mmSQ_IND_INDEX)},
720 };
721
722 static const u32 GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[] =
723 {
724 mmRLC_SRM_INDEX_CNTL_ADDR_0 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
725 mmRLC_SRM_INDEX_CNTL_ADDR_1 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
726 mmRLC_SRM_INDEX_CNTL_ADDR_2 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
727 mmRLC_SRM_INDEX_CNTL_ADDR_3 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
728 mmRLC_SRM_INDEX_CNTL_ADDR_4 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
729 mmRLC_SRM_INDEX_CNTL_ADDR_5 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
730 mmRLC_SRM_INDEX_CNTL_ADDR_6 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
731 mmRLC_SRM_INDEX_CNTL_ADDR_7 - mmRLC_SRM_INDEX_CNTL_ADDR_0,
732 };
733
734 static const u32 GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[] =
735 {
736 mmRLC_SRM_INDEX_CNTL_DATA_0 - mmRLC_SRM_INDEX_CNTL_DATA_0,
737 mmRLC_SRM_INDEX_CNTL_DATA_1 - mmRLC_SRM_INDEX_CNTL_DATA_0,
738 mmRLC_SRM_INDEX_CNTL_DATA_2 - mmRLC_SRM_INDEX_CNTL_DATA_0,
739 mmRLC_SRM_INDEX_CNTL_DATA_3 - mmRLC_SRM_INDEX_CNTL_DATA_0,
740 mmRLC_SRM_INDEX_CNTL_DATA_4 - mmRLC_SRM_INDEX_CNTL_DATA_0,
741 mmRLC_SRM_INDEX_CNTL_DATA_5 - mmRLC_SRM_INDEX_CNTL_DATA_0,
742 mmRLC_SRM_INDEX_CNTL_DATA_6 - mmRLC_SRM_INDEX_CNTL_DATA_0,
743 mmRLC_SRM_INDEX_CNTL_DATA_7 - mmRLC_SRM_INDEX_CNTL_DATA_0,
744 };
745
746 #define VEGA10_GB_ADDR_CONFIG_GOLDEN 0x2a114042
747 #define VEGA12_GB_ADDR_CONFIG_GOLDEN 0x24104041
748 #define RAVEN_GB_ADDR_CONFIG_GOLDEN 0x24000042
749 #define RAVEN2_GB_ADDR_CONFIG_GOLDEN 0x26013041
750
751 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev);
752 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev);
753 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev);
754 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev);
755 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
756 struct amdgpu_cu_info *cu_info);
757 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev);
758 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume, bool usegds);
759 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring);
760 static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
761 void *ras_error_status);
762 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
763 void *inject_if, uint32_t instance_mask);
764 static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev);
765 static void gfx_v9_0_update_spm_vmid_internal(struct amdgpu_device *adev,
766 unsigned int vmid);
767
gfx_v9_0_kiq_set_resources(struct amdgpu_ring * kiq_ring,uint64_t queue_mask)768 static void gfx_v9_0_kiq_set_resources(struct amdgpu_ring *kiq_ring,
769 uint64_t queue_mask)
770 {
771 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
772 amdgpu_ring_write(kiq_ring,
773 PACKET3_SET_RESOURCES_VMID_MASK(0) |
774 /* vmid_mask:0* queue_type:0 (KIQ) */
775 PACKET3_SET_RESOURCES_QUEUE_TYPE(0));
776 amdgpu_ring_write(kiq_ring,
777 lower_32_bits(queue_mask)); /* queue mask lo */
778 amdgpu_ring_write(kiq_ring,
779 upper_32_bits(queue_mask)); /* queue mask hi */
780 amdgpu_ring_write(kiq_ring, 0); /* gws mask lo */
781 amdgpu_ring_write(kiq_ring, 0); /* gws mask hi */
782 amdgpu_ring_write(kiq_ring, 0); /* oac mask */
783 amdgpu_ring_write(kiq_ring, 0); /* gds heap base:0, gds heap size:0 */
784 }
785
gfx_v9_0_kiq_map_queues(struct amdgpu_ring * kiq_ring,struct amdgpu_ring * ring)786 static void gfx_v9_0_kiq_map_queues(struct amdgpu_ring *kiq_ring,
787 struct amdgpu_ring *ring)
788 {
789 uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
790 uint64_t wptr_addr = ring->wptr_gpu_addr;
791 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
792
793 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
794 /* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
795 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
796 PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
797 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
798 PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
799 PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
800 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
801 /*queue_type: normal compute queue */
802 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) |
803 /* alloc format: all_on_one_pipe */
804 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) |
805 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
806 /* num_queues: must be 1 */
807 PACKET3_MAP_QUEUES_NUM_QUEUES(1));
808 amdgpu_ring_write(kiq_ring,
809 PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
810 amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
811 amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
812 amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
813 amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
814 }
815
gfx_v9_0_kiq_unmap_queues(struct amdgpu_ring * kiq_ring,struct amdgpu_ring * ring,enum amdgpu_unmap_queues_action action,u64 gpu_addr,u64 seq)816 static void gfx_v9_0_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
817 struct amdgpu_ring *ring,
818 enum amdgpu_unmap_queues_action action,
819 u64 gpu_addr, u64 seq)
820 {
821 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
822
823 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
824 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
825 PACKET3_UNMAP_QUEUES_ACTION(action) |
826 PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
827 PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
828 PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
829 amdgpu_ring_write(kiq_ring,
830 PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
831
832 if (action == PREEMPT_QUEUES_NO_UNMAP) {
833 amdgpu_ring_write(kiq_ring, lower_32_bits(ring->wptr & ring->buf_mask));
834 amdgpu_ring_write(kiq_ring, 0);
835 amdgpu_ring_write(kiq_ring, 0);
836
837 } else {
838 amdgpu_ring_write(kiq_ring, 0);
839 amdgpu_ring_write(kiq_ring, 0);
840 amdgpu_ring_write(kiq_ring, 0);
841 }
842 }
843
gfx_v9_0_kiq_query_status(struct amdgpu_ring * kiq_ring,struct amdgpu_ring * ring,u64 addr,u64 seq)844 static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
845 struct amdgpu_ring *ring,
846 u64 addr,
847 u64 seq)
848 {
849 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
850
851 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
852 amdgpu_ring_write(kiq_ring,
853 PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
854 PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
855 PACKET3_QUERY_STATUS_COMMAND(2));
856 /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
857 amdgpu_ring_write(kiq_ring,
858 PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
859 PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
860 amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
861 amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
862 amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
863 amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
864 }
865
gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring * kiq_ring,uint16_t pasid,uint32_t flush_type,bool all_hub)866 static void gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
867 uint16_t pasid, uint32_t flush_type,
868 bool all_hub)
869 {
870 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
871 amdgpu_ring_write(kiq_ring,
872 PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
873 PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
874 PACKET3_INVALIDATE_TLBS_PASID(pasid) |
875 PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
876 }
877
878 static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
879 .kiq_set_resources = gfx_v9_0_kiq_set_resources,
880 .kiq_map_queues = gfx_v9_0_kiq_map_queues,
881 .kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
882 .kiq_query_status = gfx_v9_0_kiq_query_status,
883 .kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs,
884 .set_resources_size = 8,
885 .map_queues_size = 7,
886 .unmap_queues_size = 6,
887 .query_status_size = 7,
888 .invalidate_tlbs_size = 2,
889 };
890
gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device * adev)891 static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
892 {
893 adev->gfx.kiq[0].pmf = &gfx_v9_0_kiq_pm4_funcs;
894 }
895
gfx_v9_0_init_golden_registers(struct amdgpu_device * adev)896 static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev)
897 {
898 switch (adev->ip_versions[GC_HWIP][0]) {
899 case IP_VERSION(9, 0, 1):
900 soc15_program_register_sequence(adev,
901 golden_settings_gc_9_0,
902 ARRAY_SIZE(golden_settings_gc_9_0));
903 soc15_program_register_sequence(adev,
904 golden_settings_gc_9_0_vg10,
905 ARRAY_SIZE(golden_settings_gc_9_0_vg10));
906 break;
907 case IP_VERSION(9, 2, 1):
908 soc15_program_register_sequence(adev,
909 golden_settings_gc_9_2_1,
910 ARRAY_SIZE(golden_settings_gc_9_2_1));
911 soc15_program_register_sequence(adev,
912 golden_settings_gc_9_2_1_vg12,
913 ARRAY_SIZE(golden_settings_gc_9_2_1_vg12));
914 break;
915 case IP_VERSION(9, 4, 0):
916 soc15_program_register_sequence(adev,
917 golden_settings_gc_9_0,
918 ARRAY_SIZE(golden_settings_gc_9_0));
919 soc15_program_register_sequence(adev,
920 golden_settings_gc_9_0_vg20,
921 ARRAY_SIZE(golden_settings_gc_9_0_vg20));
922 break;
923 case IP_VERSION(9, 4, 1):
924 soc15_program_register_sequence(adev,
925 golden_settings_gc_9_4_1_arct,
926 ARRAY_SIZE(golden_settings_gc_9_4_1_arct));
927 break;
928 case IP_VERSION(9, 2, 2):
929 case IP_VERSION(9, 1, 0):
930 soc15_program_register_sequence(adev, golden_settings_gc_9_1,
931 ARRAY_SIZE(golden_settings_gc_9_1));
932 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
933 soc15_program_register_sequence(adev,
934 golden_settings_gc_9_1_rv2,
935 ARRAY_SIZE(golden_settings_gc_9_1_rv2));
936 else
937 soc15_program_register_sequence(adev,
938 golden_settings_gc_9_1_rv1,
939 ARRAY_SIZE(golden_settings_gc_9_1_rv1));
940 break;
941 case IP_VERSION(9, 3, 0):
942 soc15_program_register_sequence(adev,
943 golden_settings_gc_9_1_rn,
944 ARRAY_SIZE(golden_settings_gc_9_1_rn));
945 return; /* for renoir, don't need common goldensetting */
946 case IP_VERSION(9, 4, 2):
947 gfx_v9_4_2_init_golden_registers(adev,
948 adev->smuio.funcs->get_die_id(adev));
949 break;
950 default:
951 break;
952 }
953
954 if ((adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)) &&
955 (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 2)))
956 soc15_program_register_sequence(adev, golden_settings_gc_9_x_common,
957 (const u32)ARRAY_SIZE(golden_settings_gc_9_x_common));
958 }
959
gfx_v9_0_write_data_to_reg(struct amdgpu_ring * ring,int eng_sel,bool wc,uint32_t reg,uint32_t val)960 static void gfx_v9_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
961 bool wc, uint32_t reg, uint32_t val)
962 {
963 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
964 amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) |
965 WRITE_DATA_DST_SEL(0) |
966 (wc ? WR_CONFIRM : 0));
967 amdgpu_ring_write(ring, reg);
968 amdgpu_ring_write(ring, 0);
969 amdgpu_ring_write(ring, val);
970 }
971
gfx_v9_0_wait_reg_mem(struct amdgpu_ring * ring,int eng_sel,int mem_space,int opt,uint32_t addr0,uint32_t addr1,uint32_t ref,uint32_t mask,uint32_t inv)972 static void gfx_v9_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
973 int mem_space, int opt, uint32_t addr0,
974 uint32_t addr1, uint32_t ref, uint32_t mask,
975 uint32_t inv)
976 {
977 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
978 amdgpu_ring_write(ring,
979 /* memory (1) or register (0) */
980 (WAIT_REG_MEM_MEM_SPACE(mem_space) |
981 WAIT_REG_MEM_OPERATION(opt) | /* wait */
982 WAIT_REG_MEM_FUNCTION(3) | /* equal */
983 WAIT_REG_MEM_ENGINE(eng_sel)));
984
985 if (mem_space)
986 BUG_ON(addr0 & 0x3); /* Dword align */
987 amdgpu_ring_write(ring, addr0);
988 amdgpu_ring_write(ring, addr1);
989 amdgpu_ring_write(ring, ref);
990 amdgpu_ring_write(ring, mask);
991 amdgpu_ring_write(ring, inv); /* poll interval */
992 }
993
gfx_v9_0_ring_test_ring(struct amdgpu_ring * ring)994 static int gfx_v9_0_ring_test_ring(struct amdgpu_ring *ring)
995 {
996 struct amdgpu_device *adev = ring->adev;
997 uint32_t scratch = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0);
998 uint32_t tmp = 0;
999 unsigned i;
1000 int r;
1001
1002 WREG32(scratch, 0xCAFEDEAD);
1003 r = amdgpu_ring_alloc(ring, 3);
1004 if (r)
1005 return r;
1006
1007 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
1008 amdgpu_ring_write(ring, scratch - PACKET3_SET_UCONFIG_REG_START);
1009 amdgpu_ring_write(ring, 0xDEADBEEF);
1010 amdgpu_ring_commit(ring);
1011
1012 for (i = 0; i < adev->usec_timeout; i++) {
1013 tmp = RREG32(scratch);
1014 if (tmp == 0xDEADBEEF)
1015 break;
1016 udelay(1);
1017 }
1018
1019 if (i >= adev->usec_timeout)
1020 r = -ETIMEDOUT;
1021 return r;
1022 }
1023
gfx_v9_0_ring_test_ib(struct amdgpu_ring * ring,long timeout)1024 static int gfx_v9_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1025 {
1026 struct amdgpu_device *adev = ring->adev;
1027 struct amdgpu_ib ib;
1028 struct dma_fence *f = NULL;
1029
1030 unsigned index;
1031 uint64_t gpu_addr;
1032 uint32_t tmp;
1033 long r;
1034
1035 r = amdgpu_device_wb_get(adev, &index);
1036 if (r)
1037 return r;
1038
1039 gpu_addr = adev->wb.gpu_addr + (index * 4);
1040 adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
1041 memset(&ib, 0, sizeof(ib));
1042 r = amdgpu_ib_get(adev, NULL, 16,
1043 AMDGPU_IB_POOL_DIRECT, &ib);
1044 if (r)
1045 goto err1;
1046
1047 ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
1048 ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
1049 ib.ptr[2] = lower_32_bits(gpu_addr);
1050 ib.ptr[3] = upper_32_bits(gpu_addr);
1051 ib.ptr[4] = 0xDEADBEEF;
1052 ib.length_dw = 5;
1053
1054 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1055 if (r)
1056 goto err2;
1057
1058 r = dma_fence_wait_timeout(f, false, timeout);
1059 if (r == 0) {
1060 r = -ETIMEDOUT;
1061 goto err2;
1062 } else if (r < 0) {
1063 goto err2;
1064 }
1065
1066 tmp = adev->wb.wb[index];
1067 if (tmp == 0xDEADBEEF)
1068 r = 0;
1069 else
1070 r = -EINVAL;
1071
1072 err2:
1073 amdgpu_ib_free(adev, &ib, NULL);
1074 dma_fence_put(f);
1075 err1:
1076 amdgpu_device_wb_free(adev, index);
1077 return r;
1078 }
1079
1080
gfx_v9_0_free_microcode(struct amdgpu_device * adev)1081 static void gfx_v9_0_free_microcode(struct amdgpu_device *adev)
1082 {
1083 amdgpu_ucode_release(&adev->gfx.pfp_fw);
1084 amdgpu_ucode_release(&adev->gfx.me_fw);
1085 amdgpu_ucode_release(&adev->gfx.ce_fw);
1086 amdgpu_ucode_release(&adev->gfx.rlc_fw);
1087 amdgpu_ucode_release(&adev->gfx.mec_fw);
1088 amdgpu_ucode_release(&adev->gfx.mec2_fw);
1089
1090 kfree(adev->gfx.rlc.register_list_format);
1091 }
1092
gfx_v9_0_check_fw_write_wait(struct amdgpu_device * adev)1093 static void gfx_v9_0_check_fw_write_wait(struct amdgpu_device *adev)
1094 {
1095 adev->gfx.me_fw_write_wait = false;
1096 adev->gfx.mec_fw_write_wait = false;
1097
1098 if ((adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)) &&
1099 ((adev->gfx.mec_fw_version < 0x000001a5) ||
1100 (adev->gfx.mec_feature_version < 46) ||
1101 (adev->gfx.pfp_fw_version < 0x000000b7) ||
1102 (adev->gfx.pfp_feature_version < 46)))
1103 DRM_WARN_ONCE("CP firmware version too old, please update!");
1104
1105 switch (adev->ip_versions[GC_HWIP][0]) {
1106 case IP_VERSION(9, 0, 1):
1107 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1108 (adev->gfx.me_feature_version >= 42) &&
1109 (adev->gfx.pfp_fw_version >= 0x000000b1) &&
1110 (adev->gfx.pfp_feature_version >= 42))
1111 adev->gfx.me_fw_write_wait = true;
1112
1113 if ((adev->gfx.mec_fw_version >= 0x00000193) &&
1114 (adev->gfx.mec_feature_version >= 42))
1115 adev->gfx.mec_fw_write_wait = true;
1116 break;
1117 case IP_VERSION(9, 2, 1):
1118 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1119 (adev->gfx.me_feature_version >= 44) &&
1120 (adev->gfx.pfp_fw_version >= 0x000000b2) &&
1121 (adev->gfx.pfp_feature_version >= 44))
1122 adev->gfx.me_fw_write_wait = true;
1123
1124 if ((adev->gfx.mec_fw_version >= 0x00000196) &&
1125 (adev->gfx.mec_feature_version >= 44))
1126 adev->gfx.mec_fw_write_wait = true;
1127 break;
1128 case IP_VERSION(9, 4, 0):
1129 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1130 (adev->gfx.me_feature_version >= 44) &&
1131 (adev->gfx.pfp_fw_version >= 0x000000b2) &&
1132 (adev->gfx.pfp_feature_version >= 44))
1133 adev->gfx.me_fw_write_wait = true;
1134
1135 if ((adev->gfx.mec_fw_version >= 0x00000197) &&
1136 (adev->gfx.mec_feature_version >= 44))
1137 adev->gfx.mec_fw_write_wait = true;
1138 break;
1139 case IP_VERSION(9, 1, 0):
1140 case IP_VERSION(9, 2, 2):
1141 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1142 (adev->gfx.me_feature_version >= 42) &&
1143 (adev->gfx.pfp_fw_version >= 0x000000b1) &&
1144 (adev->gfx.pfp_feature_version >= 42))
1145 adev->gfx.me_fw_write_wait = true;
1146
1147 if ((adev->gfx.mec_fw_version >= 0x00000192) &&
1148 (adev->gfx.mec_feature_version >= 42))
1149 adev->gfx.mec_fw_write_wait = true;
1150 break;
1151 default:
1152 adev->gfx.me_fw_write_wait = true;
1153 adev->gfx.mec_fw_write_wait = true;
1154 break;
1155 }
1156 }
1157
1158 struct amdgpu_gfxoff_quirk {
1159 u16 chip_vendor;
1160 u16 chip_device;
1161 u16 subsys_vendor;
1162 u16 subsys_device;
1163 u8 revision;
1164 };
1165
1166 static const struct amdgpu_gfxoff_quirk amdgpu_gfxoff_quirk_list[] = {
1167 /* https://bugzilla.kernel.org/show_bug.cgi?id=204689 */
1168 { 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc8 },
1169 /* https://bugzilla.kernel.org/show_bug.cgi?id=207171 */
1170 { 0x1002, 0x15dd, 0x103c, 0x83e7, 0xd3 },
1171 /* GFXOFF is unstable on C6 parts with a VBIOS 113-RAVEN-114 */
1172 { 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc6 },
1173 /* Apple MacBook Pro (15-inch, 2019) Radeon Pro Vega 20 4 GB */
1174 { 0x1002, 0x69af, 0x106b, 0x019a, 0xc0 },
1175 { 0, 0, 0, 0, 0 },
1176 };
1177
gfx_v9_0_should_disable_gfxoff(struct pci_dev * pdev)1178 static bool gfx_v9_0_should_disable_gfxoff(struct pci_dev *pdev)
1179 {
1180 const struct amdgpu_gfxoff_quirk *p = amdgpu_gfxoff_quirk_list;
1181
1182 while (p && p->chip_device != 0) {
1183 if (pdev->vendor == p->chip_vendor &&
1184 pdev->device == p->chip_device &&
1185 pdev->subsystem_vendor == p->subsys_vendor &&
1186 pdev->subsystem_device == p->subsys_device &&
1187 pdev->revision == p->revision) {
1188 return true;
1189 }
1190 ++p;
1191 }
1192 return false;
1193 }
1194
is_raven_kicker(struct amdgpu_device * adev)1195 static bool is_raven_kicker(struct amdgpu_device *adev)
1196 {
1197 if (adev->pm.fw_version >= 0x41e2b)
1198 return true;
1199 else
1200 return false;
1201 }
1202
check_if_enlarge_doorbell_range(struct amdgpu_device * adev)1203 static bool check_if_enlarge_doorbell_range(struct amdgpu_device *adev)
1204 {
1205 if ((adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 3, 0)) &&
1206 (adev->gfx.me_fw_version >= 0x000000a5) &&
1207 (adev->gfx.me_feature_version >= 52))
1208 return true;
1209 else
1210 return false;
1211 }
1212
gfx_v9_0_check_if_need_gfxoff(struct amdgpu_device * adev)1213 static void gfx_v9_0_check_if_need_gfxoff(struct amdgpu_device *adev)
1214 {
1215 if (gfx_v9_0_should_disable_gfxoff(adev->pdev))
1216 adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1217
1218 switch (adev->ip_versions[GC_HWIP][0]) {
1219 case IP_VERSION(9, 0, 1):
1220 case IP_VERSION(9, 2, 1):
1221 case IP_VERSION(9, 4, 0):
1222 break;
1223 case IP_VERSION(9, 2, 2):
1224 case IP_VERSION(9, 1, 0):
1225 if (!((adev->apu_flags & AMD_APU_IS_RAVEN2) ||
1226 (adev->apu_flags & AMD_APU_IS_PICASSO)) &&
1227 ((!is_raven_kicker(adev) &&
1228 adev->gfx.rlc_fw_version < 531) ||
1229 (adev->gfx.rlc_feature_version < 1) ||
1230 !adev->gfx.rlc.is_rlc_v2_1))
1231 adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1232
1233 if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1234 adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG |
1235 AMD_PG_SUPPORT_CP |
1236 AMD_PG_SUPPORT_RLC_SMU_HS;
1237 break;
1238 case IP_VERSION(9, 3, 0):
1239 if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1240 adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG |
1241 AMD_PG_SUPPORT_CP |
1242 AMD_PG_SUPPORT_RLC_SMU_HS;
1243 break;
1244 default:
1245 break;
1246 }
1247 }
1248
gfx_v9_0_init_cp_gfx_microcode(struct amdgpu_device * adev,char * chip_name)1249 static int gfx_v9_0_init_cp_gfx_microcode(struct amdgpu_device *adev,
1250 char *chip_name)
1251 {
1252 char fw_name[30];
1253 int err;
1254
1255 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
1256 err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw, fw_name);
1257 if (err)
1258 goto out;
1259 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_PFP);
1260
1261 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1262 err = amdgpu_ucode_request(adev, &adev->gfx.me_fw, fw_name);
1263 if (err)
1264 goto out;
1265 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_ME);
1266
1267 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1268 err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw, fw_name);
1269 if (err)
1270 goto out;
1271 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_CE);
1272
1273 out:
1274 if (err) {
1275 amdgpu_ucode_release(&adev->gfx.pfp_fw);
1276 amdgpu_ucode_release(&adev->gfx.me_fw);
1277 amdgpu_ucode_release(&adev->gfx.ce_fw);
1278 }
1279 return err;
1280 }
1281
gfx_v9_0_init_rlc_microcode(struct amdgpu_device * adev,char * chip_name)1282 static int gfx_v9_0_init_rlc_microcode(struct amdgpu_device *adev,
1283 char *chip_name)
1284 {
1285 char fw_name[30];
1286 int err;
1287 const struct rlc_firmware_header_v2_0 *rlc_hdr;
1288 uint16_t version_major;
1289 uint16_t version_minor;
1290 uint32_t smu_version;
1291
1292 /*
1293 * For Picasso && AM4 SOCKET board, we use picasso_rlc_am4.bin
1294 * instead of picasso_rlc.bin.
1295 * Judgment method:
1296 * PCO AM4: revision >= 0xC8 && revision <= 0xCF
1297 * or revision >= 0xD8 && revision <= 0xDF
1298 * otherwise is PCO FP5
1299 */
1300 if (!strcmp(chip_name, "picasso") &&
1301 (((adev->pdev->revision >= 0xC8) && (adev->pdev->revision <= 0xCF)) ||
1302 ((adev->pdev->revision >= 0xD8) && (adev->pdev->revision <= 0xDF))))
1303 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc_am4.bin", chip_name);
1304 else if (!strcmp(chip_name, "raven") && (amdgpu_pm_load_smu_firmware(adev, &smu_version) == 0) &&
1305 (smu_version >= 0x41e2b))
1306 /**
1307 *SMC is loaded by SBIOS on APU and it's able to get the SMU version directly.
1308 */
1309 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_kicker_rlc.bin", chip_name);
1310 else
1311 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
1312 err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, fw_name);
1313 if (err)
1314 goto out;
1315 rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1316
1317 version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
1318 version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
1319 err = amdgpu_gfx_rlc_init_microcode(adev, version_major, version_minor);
1320 out:
1321 if (err)
1322 amdgpu_ucode_release(&adev->gfx.rlc_fw);
1323
1324 return err;
1325 }
1326
gfx_v9_0_load_mec2_fw_bin_support(struct amdgpu_device * adev)1327 static bool gfx_v9_0_load_mec2_fw_bin_support(struct amdgpu_device *adev)
1328 {
1329 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2) ||
1330 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1) ||
1331 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 3, 0))
1332 return false;
1333
1334 return true;
1335 }
1336
gfx_v9_0_init_cp_compute_microcode(struct amdgpu_device * adev,char * chip_name)1337 static int gfx_v9_0_init_cp_compute_microcode(struct amdgpu_device *adev,
1338 char *chip_name)
1339 {
1340 char fw_name[30];
1341 int err;
1342
1343 if (amdgpu_sriov_vf(adev) && (adev->asic_type == CHIP_ALDEBARAN))
1344 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sjt_mec.bin", chip_name);
1345 else
1346 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1347
1348 err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, fw_name);
1349 if (err)
1350 goto out;
1351 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1);
1352 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1_JT);
1353
1354 if (gfx_v9_0_load_mec2_fw_bin_support(adev)) {
1355 if (amdgpu_sriov_vf(adev) && (adev->asic_type == CHIP_ALDEBARAN))
1356 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sjt_mec2.bin", chip_name);
1357 else
1358 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1359
1360 /* ignore failures to load */
1361 err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw, fw_name);
1362 if (!err) {
1363 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC2);
1364 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC2_JT);
1365 } else {
1366 err = 0;
1367 amdgpu_ucode_release(&adev->gfx.mec2_fw);
1368 }
1369 } else {
1370 adev->gfx.mec2_fw_version = adev->gfx.mec_fw_version;
1371 adev->gfx.mec2_feature_version = adev->gfx.mec_feature_version;
1372 }
1373
1374 gfx_v9_0_check_if_need_gfxoff(adev);
1375 gfx_v9_0_check_fw_write_wait(adev);
1376
1377 out:
1378 if (err)
1379 amdgpu_ucode_release(&adev->gfx.mec_fw);
1380 return err;
1381 }
1382
gfx_v9_0_init_microcode(struct amdgpu_device * adev)1383 static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
1384 {
1385 char ucode_prefix[30];
1386 int r;
1387
1388 DRM_DEBUG("\n");
1389 amdgpu_ucode_ip_version_decode(adev, GC_HWIP, ucode_prefix, sizeof(ucode_prefix));
1390
1391 /* No CPG in Arcturus */
1392 if (adev->gfx.num_gfx_rings) {
1393 r = gfx_v9_0_init_cp_gfx_microcode(adev, ucode_prefix);
1394 if (r)
1395 return r;
1396 }
1397
1398 r = gfx_v9_0_init_rlc_microcode(adev, ucode_prefix);
1399 if (r)
1400 return r;
1401
1402 r = gfx_v9_0_init_cp_compute_microcode(adev, ucode_prefix);
1403 if (r)
1404 return r;
1405
1406 return r;
1407 }
1408
gfx_v9_0_get_csb_size(struct amdgpu_device * adev)1409 static u32 gfx_v9_0_get_csb_size(struct amdgpu_device *adev)
1410 {
1411 u32 count = 0;
1412 const struct cs_section_def *sect = NULL;
1413 const struct cs_extent_def *ext = NULL;
1414
1415 /* begin clear state */
1416 count += 2;
1417 /* context control state */
1418 count += 3;
1419
1420 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
1421 for (ext = sect->section; ext->extent != NULL; ++ext) {
1422 if (sect->id == SECT_CONTEXT)
1423 count += 2 + ext->reg_count;
1424 else
1425 return 0;
1426 }
1427 }
1428
1429 /* end clear state */
1430 count += 2;
1431 /* clear state */
1432 count += 2;
1433
1434 return count;
1435 }
1436
gfx_v9_0_get_csb_buffer(struct amdgpu_device * adev,volatile u32 * buffer)1437 static void gfx_v9_0_get_csb_buffer(struct amdgpu_device *adev,
1438 volatile u32 *buffer)
1439 {
1440 u32 count = 0, i;
1441 const struct cs_section_def *sect = NULL;
1442 const struct cs_extent_def *ext = NULL;
1443
1444 if (adev->gfx.rlc.cs_data == NULL)
1445 return;
1446 if (buffer == NULL)
1447 return;
1448
1449 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1450 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1451
1452 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1453 buffer[count++] = cpu_to_le32(0x80000000);
1454 buffer[count++] = cpu_to_le32(0x80000000);
1455
1456 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1457 for (ext = sect->section; ext->extent != NULL; ++ext) {
1458 if (sect->id == SECT_CONTEXT) {
1459 buffer[count++] =
1460 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1461 buffer[count++] = cpu_to_le32(ext->reg_index -
1462 PACKET3_SET_CONTEXT_REG_START);
1463 for (i = 0; i < ext->reg_count; i++)
1464 buffer[count++] = cpu_to_le32(ext->extent[i]);
1465 } else {
1466 return;
1467 }
1468 }
1469 }
1470
1471 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1472 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1473
1474 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1475 buffer[count++] = cpu_to_le32(0);
1476 }
1477
gfx_v9_0_init_always_on_cu_mask(struct amdgpu_device * adev)1478 static void gfx_v9_0_init_always_on_cu_mask(struct amdgpu_device *adev)
1479 {
1480 struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
1481 uint32_t pg_always_on_cu_num = 2;
1482 uint32_t always_on_cu_num;
1483 uint32_t i, j, k;
1484 uint32_t mask, cu_bitmap, counter;
1485
1486 if (adev->flags & AMD_IS_APU)
1487 always_on_cu_num = 4;
1488 else if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 2, 1))
1489 always_on_cu_num = 8;
1490 else
1491 always_on_cu_num = 12;
1492
1493 mutex_lock(&adev->grbm_idx_mutex);
1494 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1495 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1496 mask = 1;
1497 cu_bitmap = 0;
1498 counter = 0;
1499 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff, 0);
1500
1501 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
1502 if (cu_info->bitmap[0][i][j] & mask) {
1503 if (counter == pg_always_on_cu_num)
1504 WREG32_SOC15(GC, 0, mmRLC_PG_ALWAYS_ON_CU_MASK, cu_bitmap);
1505 if (counter < always_on_cu_num)
1506 cu_bitmap |= mask;
1507 else
1508 break;
1509 counter++;
1510 }
1511 mask <<= 1;
1512 }
1513
1514 WREG32_SOC15(GC, 0, mmRLC_LB_ALWAYS_ACTIVE_CU_MASK, cu_bitmap);
1515 cu_info->ao_cu_bitmap[i][j] = cu_bitmap;
1516 }
1517 }
1518 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1519 mutex_unlock(&adev->grbm_idx_mutex);
1520 }
1521
gfx_v9_0_init_lbpw(struct amdgpu_device * adev)1522 static void gfx_v9_0_init_lbpw(struct amdgpu_device *adev)
1523 {
1524 uint32_t data;
1525
1526 /* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1527 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F);
1528 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x0333A5A7);
1529 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077);
1530 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x30 | 0x40 << 8 | 0x02FA << 16));
1531
1532 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1533 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000);
1534
1535 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1536 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000500);
1537
1538 mutex_lock(&adev->grbm_idx_mutex);
1539 /* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1540 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1541 WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff);
1542
1543 /* set mmRLC_LB_PARAMS = 0x003F_1006 */
1544 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003);
1545 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010);
1546 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F);
1547 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data);
1548
1549 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1550 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7);
1551 data &= 0x0000FFFF;
1552 data |= 0x00C00000;
1553 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data);
1554
1555 /*
1556 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xF (4 CUs AON for Raven),
1557 * programmed in gfx_v9_0_init_always_on_cu_mask()
1558 */
1559
1560 /* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1561 * but used for RLC_LB_CNTL configuration */
1562 data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK;
1563 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09);
1564 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000);
1565 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data);
1566 mutex_unlock(&adev->grbm_idx_mutex);
1567
1568 gfx_v9_0_init_always_on_cu_mask(adev);
1569 }
1570
gfx_v9_4_init_lbpw(struct amdgpu_device * adev)1571 static void gfx_v9_4_init_lbpw(struct amdgpu_device *adev)
1572 {
1573 uint32_t data;
1574
1575 /* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1576 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F);
1577 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x033388F8);
1578 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077);
1579 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x10 | 0x27 << 8 | 0x02FA << 16));
1580
1581 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1582 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000);
1583
1584 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1585 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000800);
1586
1587 mutex_lock(&adev->grbm_idx_mutex);
1588 /* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1589 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
1590 WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff);
1591
1592 /* set mmRLC_LB_PARAMS = 0x003F_1006 */
1593 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003);
1594 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010);
1595 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F);
1596 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data);
1597
1598 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1599 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7);
1600 data &= 0x0000FFFF;
1601 data |= 0x00C00000;
1602 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data);
1603
1604 /*
1605 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xFFF (12 CUs AON),
1606 * programmed in gfx_v9_0_init_always_on_cu_mask()
1607 */
1608
1609 /* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1610 * but used for RLC_LB_CNTL configuration */
1611 data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK;
1612 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09);
1613 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000);
1614 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data);
1615 mutex_unlock(&adev->grbm_idx_mutex);
1616
1617 gfx_v9_0_init_always_on_cu_mask(adev);
1618 }
1619
gfx_v9_0_enable_lbpw(struct amdgpu_device * adev,bool enable)1620 static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool enable)
1621 {
1622 WREG32_FIELD15(GC, 0, RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0);
1623 }
1624
gfx_v9_0_cp_jump_table_num(struct amdgpu_device * adev)1625 static int gfx_v9_0_cp_jump_table_num(struct amdgpu_device *adev)
1626 {
1627 if (gfx_v9_0_load_mec2_fw_bin_support(adev))
1628 return 5;
1629 else
1630 return 4;
1631 }
1632
gfx_v9_0_init_rlcg_reg_access_ctrl(struct amdgpu_device * adev)1633 static void gfx_v9_0_init_rlcg_reg_access_ctrl(struct amdgpu_device *adev)
1634 {
1635 struct amdgpu_rlcg_reg_access_ctrl *reg_access_ctrl;
1636
1637 reg_access_ctrl = &adev->gfx.rlc.reg_access_ctrl[0];
1638 reg_access_ctrl->scratch_reg0 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0);
1639 reg_access_ctrl->scratch_reg1 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG1);
1640 reg_access_ctrl->scratch_reg2 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG2);
1641 reg_access_ctrl->scratch_reg3 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG3);
1642 reg_access_ctrl->grbm_cntl = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_CNTL);
1643 reg_access_ctrl->grbm_idx = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_INDEX);
1644 reg_access_ctrl->spare_int = SOC15_REG_OFFSET(GC, 0, mmRLC_SPARE_INT);
1645 adev->gfx.rlc.rlcg_reg_access_supported = true;
1646 }
1647
gfx_v9_0_rlc_init(struct amdgpu_device * adev)1648 static int gfx_v9_0_rlc_init(struct amdgpu_device *adev)
1649 {
1650 const struct cs_section_def *cs_data;
1651 int r;
1652
1653 adev->gfx.rlc.cs_data = gfx9_cs_data;
1654
1655 cs_data = adev->gfx.rlc.cs_data;
1656
1657 if (cs_data) {
1658 /* init clear state block */
1659 r = amdgpu_gfx_rlc_init_csb(adev);
1660 if (r)
1661 return r;
1662 }
1663
1664 if (adev->flags & AMD_IS_APU) {
1665 /* TODO: double check the cp_table_size for RV */
1666 adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1667 r = amdgpu_gfx_rlc_init_cpt(adev);
1668 if (r)
1669 return r;
1670 }
1671
1672 return 0;
1673 }
1674
gfx_v9_0_mec_fini(struct amdgpu_device * adev)1675 static void gfx_v9_0_mec_fini(struct amdgpu_device *adev)
1676 {
1677 amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1678 amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL);
1679 }
1680
gfx_v9_0_mec_init(struct amdgpu_device * adev)1681 static int gfx_v9_0_mec_init(struct amdgpu_device *adev)
1682 {
1683 int r;
1684 u32 *hpd;
1685 const __le32 *fw_data;
1686 unsigned fw_size;
1687 u32 *fw;
1688 size_t mec_hpd_size;
1689
1690 const struct gfx_firmware_header_v1_0 *mec_hdr;
1691
1692 bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1693
1694 /* take ownership of the relevant compute queues */
1695 amdgpu_gfx_compute_queue_acquire(adev);
1696 mec_hpd_size = adev->gfx.num_compute_rings * GFX9_MEC_HPD_SIZE;
1697 if (mec_hpd_size) {
1698 r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1699 AMDGPU_GEM_DOMAIN_VRAM |
1700 AMDGPU_GEM_DOMAIN_GTT,
1701 &adev->gfx.mec.hpd_eop_obj,
1702 &adev->gfx.mec.hpd_eop_gpu_addr,
1703 (void **)&hpd);
1704 if (r) {
1705 dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1706 gfx_v9_0_mec_fini(adev);
1707 return r;
1708 }
1709
1710 memset(hpd, 0, mec_hpd_size);
1711
1712 amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1713 amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1714 }
1715
1716 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1717
1718 fw_data = (const __le32 *)
1719 (adev->gfx.mec_fw->data +
1720 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
1721 fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes);
1722
1723 r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
1724 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
1725 &adev->gfx.mec.mec_fw_obj,
1726 &adev->gfx.mec.mec_fw_gpu_addr,
1727 (void **)&fw);
1728 if (r) {
1729 dev_warn(adev->dev, "(%d) create mec firmware bo failed\n", r);
1730 gfx_v9_0_mec_fini(adev);
1731 return r;
1732 }
1733
1734 memcpy(fw, fw_data, fw_size);
1735
1736 amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
1737 amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
1738
1739 return 0;
1740 }
1741
wave_read_ind(struct amdgpu_device * adev,uint32_t simd,uint32_t wave,uint32_t address)1742 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
1743 {
1744 WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX,
1745 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1746 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
1747 (address << SQ_IND_INDEX__INDEX__SHIFT) |
1748 (SQ_IND_INDEX__FORCE_READ_MASK));
1749 return RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
1750 }
1751
wave_read_regs(struct amdgpu_device * adev,uint32_t simd,uint32_t wave,uint32_t thread,uint32_t regno,uint32_t num,uint32_t * out)1752 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
1753 uint32_t wave, uint32_t thread,
1754 uint32_t regno, uint32_t num, uint32_t *out)
1755 {
1756 WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX,
1757 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1758 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
1759 (regno << SQ_IND_INDEX__INDEX__SHIFT) |
1760 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
1761 (SQ_IND_INDEX__FORCE_READ_MASK) |
1762 (SQ_IND_INDEX__AUTO_INCR_MASK));
1763 while (num--)
1764 *(out++) = RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
1765 }
1766
gfx_v9_0_read_wave_data(struct amdgpu_device * adev,uint32_t xcc_id,uint32_t simd,uint32_t wave,uint32_t * dst,int * no_fields)1767 static void gfx_v9_0_read_wave_data(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
1768 {
1769 /* type 1 wave data */
1770 dst[(*no_fields)++] = 1;
1771 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
1772 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
1773 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
1774 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
1775 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
1776 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
1777 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
1778 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
1779 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
1780 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
1781 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
1782 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
1783 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
1784 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
1785 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE);
1786 }
1787
gfx_v9_0_read_wave_sgprs(struct amdgpu_device * adev,uint32_t xcc_id,uint32_t simd,uint32_t wave,uint32_t start,uint32_t size,uint32_t * dst)1788 static void gfx_v9_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
1789 uint32_t wave, uint32_t start,
1790 uint32_t size, uint32_t *dst)
1791 {
1792 wave_read_regs(
1793 adev, simd, wave, 0,
1794 start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
1795 }
1796
gfx_v9_0_read_wave_vgprs(struct amdgpu_device * adev,uint32_t xcc_id,uint32_t simd,uint32_t wave,uint32_t thread,uint32_t start,uint32_t size,uint32_t * dst)1797 static void gfx_v9_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
1798 uint32_t wave, uint32_t thread,
1799 uint32_t start, uint32_t size,
1800 uint32_t *dst)
1801 {
1802 wave_read_regs(
1803 adev, simd, wave, thread,
1804 start + SQIND_WAVE_VGPRS_OFFSET, size, dst);
1805 }
1806
gfx_v9_0_select_me_pipe_q(struct amdgpu_device * adev,u32 me,u32 pipe,u32 q,u32 vm,u32 xcc_id)1807 static void gfx_v9_0_select_me_pipe_q(struct amdgpu_device *adev,
1808 u32 me, u32 pipe, u32 q, u32 vm, u32 xcc_id)
1809 {
1810 soc15_grbm_select(adev, me, pipe, q, vm, 0);
1811 }
1812
1813 static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = {
1814 .get_gpu_clock_counter = &gfx_v9_0_get_gpu_clock_counter,
1815 .select_se_sh = &gfx_v9_0_select_se_sh,
1816 .read_wave_data = &gfx_v9_0_read_wave_data,
1817 .read_wave_sgprs = &gfx_v9_0_read_wave_sgprs,
1818 .read_wave_vgprs = &gfx_v9_0_read_wave_vgprs,
1819 .select_me_pipe_q = &gfx_v9_0_select_me_pipe_q,
1820 };
1821
1822 const struct amdgpu_ras_block_hw_ops gfx_v9_0_ras_ops = {
1823 .ras_error_inject = &gfx_v9_0_ras_error_inject,
1824 .query_ras_error_count = &gfx_v9_0_query_ras_error_count,
1825 .reset_ras_error_count = &gfx_v9_0_reset_ras_error_count,
1826 };
1827
1828 static struct amdgpu_gfx_ras gfx_v9_0_ras = {
1829 .ras_block = {
1830 .hw_ops = &gfx_v9_0_ras_ops,
1831 },
1832 };
1833
gfx_v9_0_gpu_early_init(struct amdgpu_device * adev)1834 static int gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
1835 {
1836 u32 gb_addr_config;
1837 int err;
1838
1839 switch (adev->ip_versions[GC_HWIP][0]) {
1840 case IP_VERSION(9, 0, 1):
1841 adev->gfx.config.max_hw_contexts = 8;
1842 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1843 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1844 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1845 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1846 gb_addr_config = VEGA10_GB_ADDR_CONFIG_GOLDEN;
1847 break;
1848 case IP_VERSION(9, 2, 1):
1849 adev->gfx.config.max_hw_contexts = 8;
1850 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1851 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1852 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1853 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1854 gb_addr_config = VEGA12_GB_ADDR_CONFIG_GOLDEN;
1855 DRM_INFO("fix gfx.config for vega12\n");
1856 break;
1857 case IP_VERSION(9, 4, 0):
1858 adev->gfx.ras = &gfx_v9_0_ras;
1859 adev->gfx.config.max_hw_contexts = 8;
1860 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1861 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1862 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1863 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1864 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
1865 gb_addr_config &= ~0xf3e777ff;
1866 gb_addr_config |= 0x22014042;
1867 /* check vbios table if gpu info is not available */
1868 err = amdgpu_atomfirmware_get_gfx_info(adev);
1869 if (err)
1870 return err;
1871 break;
1872 case IP_VERSION(9, 2, 2):
1873 case IP_VERSION(9, 1, 0):
1874 adev->gfx.config.max_hw_contexts = 8;
1875 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1876 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1877 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1878 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1879 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
1880 gb_addr_config = RAVEN2_GB_ADDR_CONFIG_GOLDEN;
1881 else
1882 gb_addr_config = RAVEN_GB_ADDR_CONFIG_GOLDEN;
1883 break;
1884 case IP_VERSION(9, 4, 1):
1885 adev->gfx.ras = &gfx_v9_4_ras;
1886 adev->gfx.config.max_hw_contexts = 8;
1887 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1888 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1889 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1890 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1891 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
1892 gb_addr_config &= ~0xf3e777ff;
1893 gb_addr_config |= 0x22014042;
1894 break;
1895 case IP_VERSION(9, 3, 0):
1896 adev->gfx.config.max_hw_contexts = 8;
1897 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1898 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1899 adev->gfx.config.sc_hiz_tile_fifo_size = 0x80;
1900 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1901 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
1902 gb_addr_config &= ~0xf3e777ff;
1903 gb_addr_config |= 0x22010042;
1904 break;
1905 case IP_VERSION(9, 4, 2):
1906 adev->gfx.ras = &gfx_v9_4_2_ras;
1907 adev->gfx.config.max_hw_contexts = 8;
1908 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1909 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1910 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1911 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1912 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
1913 gb_addr_config &= ~0xf3e777ff;
1914 gb_addr_config |= 0x22014042;
1915 /* check vbios table if gpu info is not available */
1916 err = amdgpu_atomfirmware_get_gfx_info(adev);
1917 if (err)
1918 return err;
1919 break;
1920 default:
1921 BUG();
1922 break;
1923 }
1924
1925 adev->gfx.config.gb_addr_config = gb_addr_config;
1926
1927 adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
1928 REG_GET_FIELD(
1929 adev->gfx.config.gb_addr_config,
1930 GB_ADDR_CONFIG,
1931 NUM_PIPES);
1932
1933 adev->gfx.config.max_tile_pipes =
1934 adev->gfx.config.gb_addr_config_fields.num_pipes;
1935
1936 adev->gfx.config.gb_addr_config_fields.num_banks = 1 <<
1937 REG_GET_FIELD(
1938 adev->gfx.config.gb_addr_config,
1939 GB_ADDR_CONFIG,
1940 NUM_BANKS);
1941 adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
1942 REG_GET_FIELD(
1943 adev->gfx.config.gb_addr_config,
1944 GB_ADDR_CONFIG,
1945 MAX_COMPRESSED_FRAGS);
1946 adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
1947 REG_GET_FIELD(
1948 adev->gfx.config.gb_addr_config,
1949 GB_ADDR_CONFIG,
1950 NUM_RB_PER_SE);
1951 adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
1952 REG_GET_FIELD(
1953 adev->gfx.config.gb_addr_config,
1954 GB_ADDR_CONFIG,
1955 NUM_SHADER_ENGINES);
1956 adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
1957 REG_GET_FIELD(
1958 adev->gfx.config.gb_addr_config,
1959 GB_ADDR_CONFIG,
1960 PIPE_INTERLEAVE_SIZE));
1961
1962 return 0;
1963 }
1964
gfx_v9_0_compute_ring_init(struct amdgpu_device * adev,int ring_id,int mec,int pipe,int queue)1965 static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1966 int mec, int pipe, int queue)
1967 {
1968 unsigned irq_type;
1969 struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
1970 unsigned int hw_prio;
1971
1972 ring = &adev->gfx.compute_ring[ring_id];
1973
1974 /* mec0 is me1 */
1975 ring->me = mec + 1;
1976 ring->pipe = pipe;
1977 ring->queue = queue;
1978
1979 ring->ring_obj = NULL;
1980 ring->use_doorbell = true;
1981 ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1;
1982 ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1983 + (ring_id * GFX9_MEC_HPD_SIZE);
1984 ring->vm_hub = AMDGPU_GFXHUB(0);
1985 sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1986
1987 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1988 + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1989 + ring->pipe;
1990 hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
1991 AMDGPU_RING_PRIO_2 : AMDGPU_RING_PRIO_DEFAULT;
1992 /* type-2 packets are deprecated on MEC, use type-3 instead */
1993 return amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
1994 hw_prio, NULL);
1995 }
1996
gfx_v9_0_sw_init(void * handle)1997 static int gfx_v9_0_sw_init(void *handle)
1998 {
1999 int i, j, k, r, ring_id;
2000 struct amdgpu_ring *ring;
2001 struct amdgpu_kiq *kiq;
2002 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2003 unsigned int hw_prio;
2004
2005 switch (adev->ip_versions[GC_HWIP][0]) {
2006 case IP_VERSION(9, 0, 1):
2007 case IP_VERSION(9, 2, 1):
2008 case IP_VERSION(9, 4, 0):
2009 case IP_VERSION(9, 2, 2):
2010 case IP_VERSION(9, 1, 0):
2011 case IP_VERSION(9, 4, 1):
2012 case IP_VERSION(9, 3, 0):
2013 case IP_VERSION(9, 4, 2):
2014 adev->gfx.mec.num_mec = 2;
2015 break;
2016 default:
2017 adev->gfx.mec.num_mec = 1;
2018 break;
2019 }
2020
2021 adev->gfx.mec.num_pipe_per_mec = 4;
2022 adev->gfx.mec.num_queue_per_pipe = 8;
2023
2024 /* EOP Event */
2025 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_EOP_INTERRUPT, &adev->gfx.eop_irq);
2026 if (r)
2027 return r;
2028
2029 /* Privileged reg */
2030 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_REG_FAULT,
2031 &adev->gfx.priv_reg_irq);
2032 if (r)
2033 return r;
2034
2035 /* Privileged inst */
2036 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_INSTR_FAULT,
2037 &adev->gfx.priv_inst_irq);
2038 if (r)
2039 return r;
2040
2041 /* ECC error */
2042 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_ECC_ERROR,
2043 &adev->gfx.cp_ecc_error_irq);
2044 if (r)
2045 return r;
2046
2047 /* FUE error */
2048 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_FUE_ERROR,
2049 &adev->gfx.cp_ecc_error_irq);
2050 if (r)
2051 return r;
2052
2053 adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
2054
2055 if (adev->gfx.rlc.funcs) {
2056 if (adev->gfx.rlc.funcs->init) {
2057 r = adev->gfx.rlc.funcs->init(adev);
2058 if (r) {
2059 dev_err(adev->dev, "Failed to init rlc BOs!\n");
2060 return r;
2061 }
2062 }
2063 }
2064
2065 r = gfx_v9_0_mec_init(adev);
2066 if (r) {
2067 DRM_ERROR("Failed to init MEC BOs!\n");
2068 return r;
2069 }
2070
2071 /* set up the gfx ring */
2072 for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
2073 ring = &adev->gfx.gfx_ring[i];
2074 ring->ring_obj = NULL;
2075 if (!i)
2076 sprintf(ring->name, "gfx");
2077 else
2078 sprintf(ring->name, "gfx_%d", i);
2079 ring->use_doorbell = true;
2080 ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
2081
2082 /* disable scheduler on the real ring */
2083 ring->no_scheduler = true;
2084 ring->vm_hub = AMDGPU_GFXHUB(0);
2085 r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
2086 AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
2087 AMDGPU_RING_PRIO_DEFAULT, NULL);
2088 if (r)
2089 return r;
2090 }
2091
2092 /* set up the software rings */
2093 if (adev->gfx.num_gfx_rings) {
2094 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++) {
2095 ring = &adev->gfx.sw_gfx_ring[i];
2096 ring->ring_obj = NULL;
2097 sprintf(ring->name, amdgpu_sw_ring_name(i));
2098 ring->use_doorbell = true;
2099 ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
2100 ring->is_sw_ring = true;
2101 hw_prio = amdgpu_sw_ring_priority(i);
2102 ring->vm_hub = AMDGPU_GFXHUB(0);
2103 r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
2104 AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP, hw_prio,
2105 NULL);
2106 if (r)
2107 return r;
2108 ring->wptr = 0;
2109 }
2110
2111 /* init the muxer and add software rings */
2112 r = amdgpu_ring_mux_init(&adev->gfx.muxer, &adev->gfx.gfx_ring[0],
2113 GFX9_NUM_SW_GFX_RINGS);
2114 if (r) {
2115 DRM_ERROR("amdgpu_ring_mux_init failed(%d)\n", r);
2116 return r;
2117 }
2118 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++) {
2119 r = amdgpu_ring_mux_add_sw_ring(&adev->gfx.muxer,
2120 &adev->gfx.sw_gfx_ring[i]);
2121 if (r) {
2122 DRM_ERROR("amdgpu_ring_mux_add_sw_ring failed(%d)\n", r);
2123 return r;
2124 }
2125 }
2126 }
2127
2128 /* set up the compute queues - allocate horizontally across pipes */
2129 ring_id = 0;
2130 for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
2131 for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2132 for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2133 if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
2134 k, j))
2135 continue;
2136
2137 r = gfx_v9_0_compute_ring_init(adev,
2138 ring_id,
2139 i, k, j);
2140 if (r)
2141 return r;
2142
2143 ring_id++;
2144 }
2145 }
2146 }
2147
2148 r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE, 0);
2149 if (r) {
2150 DRM_ERROR("Failed to init KIQ BOs!\n");
2151 return r;
2152 }
2153
2154 kiq = &adev->gfx.kiq[0];
2155 r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq, 0);
2156 if (r)
2157 return r;
2158
2159 /* create MQD for all compute queues as wel as KIQ for SRIOV case */
2160 r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v9_mqd_allocation), 0);
2161 if (r)
2162 return r;
2163
2164 adev->gfx.ce_ram_size = 0x8000;
2165
2166 r = gfx_v9_0_gpu_early_init(adev);
2167 if (r)
2168 return r;
2169
2170 if (amdgpu_gfx_ras_sw_init(adev)) {
2171 dev_err(adev->dev, "Failed to initialize gfx ras block!\n");
2172 return -EINVAL;
2173 }
2174
2175 return 0;
2176 }
2177
2178
gfx_v9_0_sw_fini(void * handle)2179 static int gfx_v9_0_sw_fini(void *handle)
2180 {
2181 int i;
2182 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2183
2184 if (adev->gfx.num_gfx_rings) {
2185 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++)
2186 amdgpu_ring_fini(&adev->gfx.sw_gfx_ring[i]);
2187 amdgpu_ring_mux_fini(&adev->gfx.muxer);
2188 }
2189
2190 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2191 amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2192 for (i = 0; i < adev->gfx.num_compute_rings; i++)
2193 amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2194
2195 amdgpu_gfx_mqd_sw_fini(adev, 0);
2196 amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq[0].ring);
2197 amdgpu_gfx_kiq_fini(adev, 0);
2198
2199 gfx_v9_0_mec_fini(adev);
2200 amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
2201 &adev->gfx.rlc.clear_state_gpu_addr,
2202 (void **)&adev->gfx.rlc.cs_ptr);
2203 if (adev->flags & AMD_IS_APU) {
2204 amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2205 &adev->gfx.rlc.cp_table_gpu_addr,
2206 (void **)&adev->gfx.rlc.cp_table_ptr);
2207 }
2208 gfx_v9_0_free_microcode(adev);
2209
2210 return 0;
2211 }
2212
2213
gfx_v9_0_tiling_mode_table_init(struct amdgpu_device * adev)2214 static void gfx_v9_0_tiling_mode_table_init(struct amdgpu_device *adev)
2215 {
2216 /* TODO */
2217 }
2218
gfx_v9_0_select_se_sh(struct amdgpu_device * adev,u32 se_num,u32 sh_num,u32 instance,int xcc_id)2219 void gfx_v9_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num,
2220 u32 instance, int xcc_id)
2221 {
2222 u32 data;
2223
2224 if (instance == 0xffffffff)
2225 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
2226 else
2227 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
2228
2229 if (se_num == 0xffffffff)
2230 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
2231 else
2232 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
2233
2234 if (sh_num == 0xffffffff)
2235 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
2236 else
2237 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
2238
2239 WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_INDEX, data);
2240 }
2241
gfx_v9_0_get_rb_active_bitmap(struct amdgpu_device * adev)2242 static u32 gfx_v9_0_get_rb_active_bitmap(struct amdgpu_device *adev)
2243 {
2244 u32 data, mask;
2245
2246 data = RREG32_SOC15(GC, 0, mmCC_RB_BACKEND_DISABLE);
2247 data |= RREG32_SOC15(GC, 0, mmGC_USER_RB_BACKEND_DISABLE);
2248
2249 data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
2250 data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
2251
2252 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
2253 adev->gfx.config.max_sh_per_se);
2254
2255 return (~data) & mask;
2256 }
2257
gfx_v9_0_setup_rb(struct amdgpu_device * adev)2258 static void gfx_v9_0_setup_rb(struct amdgpu_device *adev)
2259 {
2260 int i, j;
2261 u32 data;
2262 u32 active_rbs = 0;
2263 u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
2264 adev->gfx.config.max_sh_per_se;
2265
2266 mutex_lock(&adev->grbm_idx_mutex);
2267 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2268 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2269 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff, 0);
2270 data = gfx_v9_0_get_rb_active_bitmap(adev);
2271 active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
2272 rb_bitmap_width_per_sh);
2273 }
2274 }
2275 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
2276 mutex_unlock(&adev->grbm_idx_mutex);
2277
2278 adev->gfx.config.backend_enable_mask = active_rbs;
2279 adev->gfx.config.num_rbs = hweight32(active_rbs);
2280 }
2281
gfx_v9_0_debug_trap_config_init(struct amdgpu_device * adev,uint32_t first_vmid,uint32_t last_vmid)2282 static void gfx_v9_0_debug_trap_config_init(struct amdgpu_device *adev,
2283 uint32_t first_vmid,
2284 uint32_t last_vmid)
2285 {
2286 uint32_t data;
2287 uint32_t trap_config_vmid_mask = 0;
2288 int i;
2289
2290 /* Calculate trap config vmid mask */
2291 for (i = first_vmid; i < last_vmid; i++)
2292 trap_config_vmid_mask |= (1 << i);
2293
2294 data = REG_SET_FIELD(0, SPI_GDBG_TRAP_CONFIG,
2295 VMID_SEL, trap_config_vmid_mask);
2296 data = REG_SET_FIELD(data, SPI_GDBG_TRAP_CONFIG,
2297 TRAP_EN, 1);
2298 WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_CONFIG), data);
2299 WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_MASK), 0);
2300
2301 WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_DATA0), 0);
2302 WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_TRAP_DATA1), 0);
2303 }
2304
2305 #define DEFAULT_SH_MEM_BASES (0x6000)
gfx_v9_0_init_compute_vmid(struct amdgpu_device * adev)2306 static void gfx_v9_0_init_compute_vmid(struct amdgpu_device *adev)
2307 {
2308 int i;
2309 uint32_t sh_mem_config;
2310 uint32_t sh_mem_bases;
2311
2312 /*
2313 * Configure apertures:
2314 * LDS: 0x60000000'00000000 - 0x60000001'00000000 (4GB)
2315 * Scratch: 0x60000001'00000000 - 0x60000002'00000000 (4GB)
2316 * GPUVM: 0x60010000'00000000 - 0x60020000'00000000 (1TB)
2317 */
2318 sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
2319
2320 sh_mem_config = SH_MEM_ADDRESS_MODE_64 |
2321 SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
2322 SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
2323
2324 mutex_lock(&adev->srbm_mutex);
2325 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
2326 soc15_grbm_select(adev, 0, 0, 0, i, 0);
2327 /* CP and shaders */
2328 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, sh_mem_config);
2329 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, sh_mem_bases);
2330 }
2331 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
2332 mutex_unlock(&adev->srbm_mutex);
2333
2334 /* Initialize all compute VMIDs to have no GDS, GWS, or OA
2335 access. These should be enabled by FW for target VMIDs. */
2336 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
2337 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * i, 0);
2338 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * i, 0);
2339 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, i, 0);
2340 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, i, 0);
2341 }
2342 }
2343
gfx_v9_0_init_gds_vmid(struct amdgpu_device * adev)2344 static void gfx_v9_0_init_gds_vmid(struct amdgpu_device *adev)
2345 {
2346 int vmid;
2347
2348 /*
2349 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
2350 * access. Compute VMIDs should be enabled by FW for target VMIDs,
2351 * the driver can enable them for graphics. VMID0 should maintain
2352 * access so that HWS firmware can save/restore entries.
2353 */
2354 for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
2355 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * vmid, 0);
2356 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * vmid, 0);
2357 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, vmid, 0);
2358 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, vmid, 0);
2359 }
2360 }
2361
gfx_v9_0_init_sq_config(struct amdgpu_device * adev)2362 static void gfx_v9_0_init_sq_config(struct amdgpu_device *adev)
2363 {
2364 uint32_t tmp;
2365
2366 switch (adev->ip_versions[GC_HWIP][0]) {
2367 case IP_VERSION(9, 4, 1):
2368 tmp = RREG32_SOC15(GC, 0, mmSQ_CONFIG);
2369 tmp = REG_SET_FIELD(tmp, SQ_CONFIG, DISABLE_BARRIER_WAITCNT,
2370 !READ_ONCE(adev->barrier_has_auto_waitcnt));
2371 WREG32_SOC15(GC, 0, mmSQ_CONFIG, tmp);
2372 break;
2373 default:
2374 break;
2375 }
2376 }
2377
gfx_v9_0_constants_init(struct amdgpu_device * adev)2378 static void gfx_v9_0_constants_init(struct amdgpu_device *adev)
2379 {
2380 u32 tmp;
2381 int i;
2382
2383 WREG32_FIELD15_RLC(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff);
2384
2385 gfx_v9_0_tiling_mode_table_init(adev);
2386
2387 if (adev->gfx.num_gfx_rings)
2388 gfx_v9_0_setup_rb(adev);
2389 gfx_v9_0_get_cu_info(adev, &adev->gfx.cu_info);
2390 adev->gfx.config.db_debug2 = RREG32_SOC15(GC, 0, mmDB_DEBUG2);
2391
2392 /* XXX SH_MEM regs */
2393 /* where to put LDS, scratch, GPUVM in FSA64 space */
2394 mutex_lock(&adev->srbm_mutex);
2395 for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB(0)].num_ids; i++) {
2396 soc15_grbm_select(adev, 0, 0, 0, i, 0);
2397 /* CP and shaders */
2398 if (i == 0) {
2399 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
2400 SH_MEM_ALIGNMENT_MODE_UNALIGNED);
2401 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
2402 !!adev->gmc.noretry);
2403 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp);
2404 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, 0);
2405 } else {
2406 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
2407 SH_MEM_ALIGNMENT_MODE_UNALIGNED);
2408 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
2409 !!adev->gmc.noretry);
2410 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp);
2411 tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
2412 (adev->gmc.private_aperture_start >> 48));
2413 tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
2414 (adev->gmc.shared_aperture_start >> 48));
2415 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, tmp);
2416 }
2417 }
2418 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
2419
2420 mutex_unlock(&adev->srbm_mutex);
2421
2422 gfx_v9_0_init_compute_vmid(adev);
2423 gfx_v9_0_init_gds_vmid(adev);
2424 gfx_v9_0_init_sq_config(adev);
2425 }
2426
gfx_v9_0_wait_for_rlc_serdes(struct amdgpu_device * adev)2427 static void gfx_v9_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
2428 {
2429 u32 i, j, k;
2430 u32 mask;
2431
2432 mutex_lock(&adev->grbm_idx_mutex);
2433 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2434 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2435 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff, 0);
2436 for (k = 0; k < adev->usec_timeout; k++) {
2437 if (RREG32_SOC15(GC, 0, mmRLC_SERDES_CU_MASTER_BUSY) == 0)
2438 break;
2439 udelay(1);
2440 }
2441 if (k == adev->usec_timeout) {
2442 amdgpu_gfx_select_se_sh(adev, 0xffffffff,
2443 0xffffffff, 0xffffffff, 0);
2444 mutex_unlock(&adev->grbm_idx_mutex);
2445 DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
2446 i, j);
2447 return;
2448 }
2449 }
2450 }
2451 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
2452 mutex_unlock(&adev->grbm_idx_mutex);
2453
2454 mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
2455 RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
2456 RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
2457 RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
2458 for (k = 0; k < adev->usec_timeout; k++) {
2459 if ((RREG32_SOC15(GC, 0, mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
2460 break;
2461 udelay(1);
2462 }
2463 }
2464
gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device * adev,bool enable)2465 static void gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
2466 bool enable)
2467 {
2468 u32 tmp;
2469
2470 /* These interrupts should be enabled to drive DS clock */
2471
2472 tmp= RREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0);
2473
2474 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
2475 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
2476 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
2477 if(adev->gfx.num_gfx_rings)
2478 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
2479
2480 WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp);
2481 }
2482
gfx_v9_0_init_csb(struct amdgpu_device * adev)2483 static void gfx_v9_0_init_csb(struct amdgpu_device *adev)
2484 {
2485 adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
2486 /* csib */
2487 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_HI),
2488 adev->gfx.rlc.clear_state_gpu_addr >> 32);
2489 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_LO),
2490 adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
2491 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_LENGTH),
2492 adev->gfx.rlc.clear_state_size);
2493 }
2494
gfx_v9_1_parse_ind_reg_list(int * register_list_format,int indirect_offset,int list_size,int * unique_indirect_regs,int unique_indirect_reg_count,int * indirect_start_offsets,int * indirect_start_offsets_count,int max_start_offsets_count)2495 static void gfx_v9_1_parse_ind_reg_list(int *register_list_format,
2496 int indirect_offset,
2497 int list_size,
2498 int *unique_indirect_regs,
2499 int unique_indirect_reg_count,
2500 int *indirect_start_offsets,
2501 int *indirect_start_offsets_count,
2502 int max_start_offsets_count)
2503 {
2504 int idx;
2505
2506 for (; indirect_offset < list_size; indirect_offset++) {
2507 WARN_ON(*indirect_start_offsets_count >= max_start_offsets_count);
2508 indirect_start_offsets[*indirect_start_offsets_count] = indirect_offset;
2509 *indirect_start_offsets_count = *indirect_start_offsets_count + 1;
2510
2511 while (register_list_format[indirect_offset] != 0xFFFFFFFF) {
2512 indirect_offset += 2;
2513
2514 /* look for the matching indice */
2515 for (idx = 0; idx < unique_indirect_reg_count; idx++) {
2516 if (unique_indirect_regs[idx] ==
2517 register_list_format[indirect_offset] ||
2518 !unique_indirect_regs[idx])
2519 break;
2520 }
2521
2522 BUG_ON(idx >= unique_indirect_reg_count);
2523
2524 if (!unique_indirect_regs[idx])
2525 unique_indirect_regs[idx] = register_list_format[indirect_offset];
2526
2527 indirect_offset++;
2528 }
2529 }
2530 }
2531
gfx_v9_1_init_rlc_save_restore_list(struct amdgpu_device * adev)2532 static int gfx_v9_1_init_rlc_save_restore_list(struct amdgpu_device *adev)
2533 {
2534 int unique_indirect_regs[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2535 int unique_indirect_reg_count = 0;
2536
2537 int indirect_start_offsets[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2538 int indirect_start_offsets_count = 0;
2539
2540 int list_size = 0;
2541 int i = 0, j = 0;
2542 u32 tmp = 0;
2543
2544 u32 *register_list_format =
2545 kmemdup(adev->gfx.rlc.register_list_format,
2546 adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
2547 if (!register_list_format)
2548 return -ENOMEM;
2549
2550 /* setup unique_indirect_regs array and indirect_start_offsets array */
2551 unique_indirect_reg_count = ARRAY_SIZE(unique_indirect_regs);
2552 gfx_v9_1_parse_ind_reg_list(register_list_format,
2553 adev->gfx.rlc.reg_list_format_direct_reg_list_length,
2554 adev->gfx.rlc.reg_list_format_size_bytes >> 2,
2555 unique_indirect_regs,
2556 unique_indirect_reg_count,
2557 indirect_start_offsets,
2558 &indirect_start_offsets_count,
2559 ARRAY_SIZE(indirect_start_offsets));
2560
2561 /* enable auto inc in case it is disabled */
2562 tmp = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL));
2563 tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK;
2564 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL), tmp);
2565
2566 /* write register_restore table to offset 0x0 using RLC_SRM_ARAM_ADDR/DATA */
2567 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_ADDR),
2568 RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET);
2569 for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
2570 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_DATA),
2571 adev->gfx.rlc.register_restore[i]);
2572
2573 /* load indirect register */
2574 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2575 adev->gfx.rlc.reg_list_format_start);
2576
2577 /* direct register portion */
2578 for (i = 0; i < adev->gfx.rlc.reg_list_format_direct_reg_list_length; i++)
2579 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
2580 register_list_format[i]);
2581
2582 /* indirect register portion */
2583 while (i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2)) {
2584 if (register_list_format[i] == 0xFFFFFFFF) {
2585 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2586 continue;
2587 }
2588
2589 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2590 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]);
2591
2592 for (j = 0; j < unique_indirect_reg_count; j++) {
2593 if (register_list_format[i] == unique_indirect_regs[j]) {
2594 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, j);
2595 break;
2596 }
2597 }
2598
2599 BUG_ON(j >= unique_indirect_reg_count);
2600
2601 i++;
2602 }
2603
2604 /* set save/restore list size */
2605 list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
2606 list_size = list_size >> 1;
2607 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2608 adev->gfx.rlc.reg_restore_list_size);
2609 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), list_size);
2610
2611 /* write the starting offsets to RLC scratch ram */
2612 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),
2613 adev->gfx.rlc.starting_offsets_start);
2614 for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
2615 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),
2616 indirect_start_offsets[i]);
2617
2618 /* load unique indirect regs*/
2619 for (i = 0; i < ARRAY_SIZE(unique_indirect_regs); i++) {
2620 if (unique_indirect_regs[i] != 0) {
2621 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_ADDR_0)
2622 + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i],
2623 unique_indirect_regs[i] & 0x3FFFF);
2624
2625 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_DATA_0)
2626 + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i],
2627 unique_indirect_regs[i] >> 20);
2628 }
2629 }
2630
2631 kfree(register_list_format);
2632 return 0;
2633 }
2634
gfx_v9_0_enable_save_restore_machine(struct amdgpu_device * adev)2635 static void gfx_v9_0_enable_save_restore_machine(struct amdgpu_device *adev)
2636 {
2637 WREG32_FIELD15(GC, 0, RLC_SRM_CNTL, SRM_ENABLE, 1);
2638 }
2639
pwr_10_0_gfxip_control_over_cgpg(struct amdgpu_device * adev,bool enable)2640 static void pwr_10_0_gfxip_control_over_cgpg(struct amdgpu_device *adev,
2641 bool enable)
2642 {
2643 uint32_t data = 0;
2644 uint32_t default_data = 0;
2645
2646 default_data = data = RREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS));
2647 if (enable) {
2648 /* enable GFXIP control over CGPG */
2649 data |= PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
2650 if(default_data != data)
2651 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2652
2653 /* update status */
2654 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK;
2655 data |= (2 << PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT);
2656 if(default_data != data)
2657 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2658 } else {
2659 /* restore GFXIP control over GCPG */
2660 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK;
2661 if(default_data != data)
2662 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data);
2663 }
2664 }
2665
gfx_v9_0_init_gfx_power_gating(struct amdgpu_device * adev)2666 static void gfx_v9_0_init_gfx_power_gating(struct amdgpu_device *adev)
2667 {
2668 uint32_t data = 0;
2669
2670 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
2671 AMD_PG_SUPPORT_GFX_SMG |
2672 AMD_PG_SUPPORT_GFX_DMG)) {
2673 /* init IDLE_POLL_COUNT = 60 */
2674 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL));
2675 data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK;
2676 data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
2677 WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL), data);
2678
2679 /* init RLC PG Delay */
2680 data = 0;
2681 data |= (0x10 << RLC_PG_DELAY__POWER_UP_DELAY__SHIFT);
2682 data |= (0x10 << RLC_PG_DELAY__POWER_DOWN_DELAY__SHIFT);
2683 data |= (0x10 << RLC_PG_DELAY__CMD_PROPAGATE_DELAY__SHIFT);
2684 data |= (0x40 << RLC_PG_DELAY__MEM_SLEEP_DELAY__SHIFT);
2685 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY), data);
2686
2687 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2));
2688 data &= ~RLC_PG_DELAY_2__SERDES_CMD_DELAY_MASK;
2689 data |= (0x4 << RLC_PG_DELAY_2__SERDES_CMD_DELAY__SHIFT);
2690 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2), data);
2691
2692 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3));
2693 data &= ~RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG_MASK;
2694 data |= (0xff << RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG__SHIFT);
2695 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3), data);
2696
2697 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL));
2698 data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK;
2699
2700 /* program GRBM_REG_SAVE_GFX_IDLE_THRESHOLD to 0x55f0 */
2701 data |= (0x55f0 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT);
2702 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL), data);
2703 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 3, 0))
2704 pwr_10_0_gfxip_control_over_cgpg(adev, true);
2705 }
2706 }
2707
gfx_v9_0_enable_sck_slow_down_on_power_up(struct amdgpu_device * adev,bool enable)2708 static void gfx_v9_0_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
2709 bool enable)
2710 {
2711 uint32_t data = 0;
2712 uint32_t default_data = 0;
2713
2714 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2715 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2716 SMU_CLK_SLOWDOWN_ON_PU_ENABLE,
2717 enable ? 1 : 0);
2718 if (default_data != data)
2719 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2720 }
2721
gfx_v9_0_enable_sck_slow_down_on_power_down(struct amdgpu_device * adev,bool enable)2722 static void gfx_v9_0_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
2723 bool enable)
2724 {
2725 uint32_t data = 0;
2726 uint32_t default_data = 0;
2727
2728 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2729 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2730 SMU_CLK_SLOWDOWN_ON_PD_ENABLE,
2731 enable ? 1 : 0);
2732 if(default_data != data)
2733 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2734 }
2735
gfx_v9_0_enable_cp_power_gating(struct amdgpu_device * adev,bool enable)2736 static void gfx_v9_0_enable_cp_power_gating(struct amdgpu_device *adev,
2737 bool enable)
2738 {
2739 uint32_t data = 0;
2740 uint32_t default_data = 0;
2741
2742 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2743 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2744 CP_PG_DISABLE,
2745 enable ? 0 : 1);
2746 if(default_data != data)
2747 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2748 }
2749
gfx_v9_0_enable_gfx_cg_power_gating(struct amdgpu_device * adev,bool enable)2750 static void gfx_v9_0_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
2751 bool enable)
2752 {
2753 uint32_t data, default_data;
2754
2755 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2756 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2757 GFX_POWER_GATING_ENABLE,
2758 enable ? 1 : 0);
2759 if(default_data != data)
2760 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2761 }
2762
gfx_v9_0_enable_gfx_pipeline_powergating(struct amdgpu_device * adev,bool enable)2763 static void gfx_v9_0_enable_gfx_pipeline_powergating(struct amdgpu_device *adev,
2764 bool enable)
2765 {
2766 uint32_t data, default_data;
2767
2768 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2769 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2770 GFX_PIPELINE_PG_ENABLE,
2771 enable ? 1 : 0);
2772 if(default_data != data)
2773 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2774
2775 if (!enable)
2776 /* read any GFX register to wake up GFX */
2777 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmDB_RENDER_CONTROL));
2778 }
2779
gfx_v9_0_enable_gfx_static_mg_power_gating(struct amdgpu_device * adev,bool enable)2780 static void gfx_v9_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
2781 bool enable)
2782 {
2783 uint32_t data, default_data;
2784
2785 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2786 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2787 STATIC_PER_CU_PG_ENABLE,
2788 enable ? 1 : 0);
2789 if(default_data != data)
2790 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2791 }
2792
gfx_v9_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device * adev,bool enable)2793 static void gfx_v9_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
2794 bool enable)
2795 {
2796 uint32_t data, default_data;
2797
2798 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL));
2799 data = REG_SET_FIELD(data, RLC_PG_CNTL,
2800 DYN_PER_CU_PG_ENABLE,
2801 enable ? 1 : 0);
2802 if(default_data != data)
2803 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data);
2804 }
2805
gfx_v9_0_init_pg(struct amdgpu_device * adev)2806 static void gfx_v9_0_init_pg(struct amdgpu_device *adev)
2807 {
2808 gfx_v9_0_init_csb(adev);
2809
2810 /*
2811 * Rlc save restore list is workable since v2_1.
2812 * And it's needed by gfxoff feature.
2813 */
2814 if (adev->gfx.rlc.is_rlc_v2_1) {
2815 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 2, 1) ||
2816 (adev->apu_flags & AMD_APU_IS_RAVEN2))
2817 gfx_v9_1_init_rlc_save_restore_list(adev);
2818 gfx_v9_0_enable_save_restore_machine(adev);
2819 }
2820
2821 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG |
2822 AMD_PG_SUPPORT_GFX_SMG |
2823 AMD_PG_SUPPORT_GFX_DMG |
2824 AMD_PG_SUPPORT_CP |
2825 AMD_PG_SUPPORT_GDS |
2826 AMD_PG_SUPPORT_RLC_SMU_HS)) {
2827 WREG32_SOC15(GC, 0, mmRLC_JUMP_TABLE_RESTORE,
2828 adev->gfx.rlc.cp_table_gpu_addr >> 8);
2829 gfx_v9_0_init_gfx_power_gating(adev);
2830 }
2831 }
2832
gfx_v9_0_rlc_stop(struct amdgpu_device * adev)2833 static void gfx_v9_0_rlc_stop(struct amdgpu_device *adev)
2834 {
2835 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 0);
2836 gfx_v9_0_enable_gui_idle_interrupt(adev, false);
2837 gfx_v9_0_wait_for_rlc_serdes(adev);
2838 }
2839
gfx_v9_0_rlc_reset(struct amdgpu_device * adev)2840 static void gfx_v9_0_rlc_reset(struct amdgpu_device *adev)
2841 {
2842 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
2843 udelay(50);
2844 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
2845 udelay(50);
2846 }
2847
gfx_v9_0_rlc_start(struct amdgpu_device * adev)2848 static void gfx_v9_0_rlc_start(struct amdgpu_device *adev)
2849 {
2850 #ifdef AMDGPU_RLC_DEBUG_RETRY
2851 u32 rlc_ucode_ver;
2852 #endif
2853
2854 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1);
2855 udelay(50);
2856
2857 /* carrizo do enable cp interrupt after cp inited */
2858 if (!(adev->flags & AMD_IS_APU)) {
2859 gfx_v9_0_enable_gui_idle_interrupt(adev, true);
2860 udelay(50);
2861 }
2862
2863 #ifdef AMDGPU_RLC_DEBUG_RETRY
2864 /* RLC_GPM_GENERAL_6 : RLC Ucode version */
2865 rlc_ucode_ver = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_6);
2866 if(rlc_ucode_ver == 0x108) {
2867 DRM_INFO("Using rlc debug ucode. mmRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n",
2868 rlc_ucode_ver, adev->gfx.rlc_fw_version);
2869 /* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles,
2870 * default is 0x9C4 to create a 100us interval */
2871 WREG32_SOC15(GC, 0, mmRLC_GPM_TIMER_INT_3, 0x9C4);
2872 /* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr
2873 * to disable the page fault retry interrupts, default is
2874 * 0x100 (256) */
2875 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_12, 0x100);
2876 }
2877 #endif
2878 }
2879
gfx_v9_0_rlc_load_microcode(struct amdgpu_device * adev)2880 static int gfx_v9_0_rlc_load_microcode(struct amdgpu_device *adev)
2881 {
2882 const struct rlc_firmware_header_v2_0 *hdr;
2883 const __le32 *fw_data;
2884 unsigned i, fw_size;
2885
2886 if (!adev->gfx.rlc_fw)
2887 return -EINVAL;
2888
2889 hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
2890 amdgpu_ucode_print_rlc_hdr(&hdr->header);
2891
2892 fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
2893 le32_to_cpu(hdr->header.ucode_array_offset_bytes));
2894 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
2895
2896 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR,
2897 RLCG_UCODE_LOADING_START_ADDRESS);
2898 for (i = 0; i < fw_size; i++)
2899 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
2900 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
2901
2902 return 0;
2903 }
2904
gfx_v9_0_rlc_resume(struct amdgpu_device * adev)2905 static int gfx_v9_0_rlc_resume(struct amdgpu_device *adev)
2906 {
2907 int r;
2908
2909 if (amdgpu_sriov_vf(adev)) {
2910 gfx_v9_0_init_csb(adev);
2911 return 0;
2912 }
2913
2914 adev->gfx.rlc.funcs->stop(adev);
2915
2916 /* disable CG */
2917 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, 0);
2918
2919 gfx_v9_0_init_pg(adev);
2920
2921 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
2922 /* legacy rlc firmware loading */
2923 r = gfx_v9_0_rlc_load_microcode(adev);
2924 if (r)
2925 return r;
2926 }
2927
2928 switch (adev->ip_versions[GC_HWIP][0]) {
2929 case IP_VERSION(9, 2, 2):
2930 case IP_VERSION(9, 1, 0):
2931 gfx_v9_0_init_lbpw(adev);
2932 if (amdgpu_lbpw == 0)
2933 gfx_v9_0_enable_lbpw(adev, false);
2934 else
2935 gfx_v9_0_enable_lbpw(adev, true);
2936 break;
2937 case IP_VERSION(9, 4, 0):
2938 gfx_v9_4_init_lbpw(adev);
2939 if (amdgpu_lbpw > 0)
2940 gfx_v9_0_enable_lbpw(adev, true);
2941 else
2942 gfx_v9_0_enable_lbpw(adev, false);
2943 break;
2944 default:
2945 break;
2946 }
2947
2948 gfx_v9_0_update_spm_vmid_internal(adev, 0xf);
2949
2950 adev->gfx.rlc.funcs->start(adev);
2951
2952 return 0;
2953 }
2954
gfx_v9_0_cp_gfx_enable(struct amdgpu_device * adev,bool enable)2955 static void gfx_v9_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2956 {
2957 u32 tmp = RREG32_SOC15(GC, 0, mmCP_ME_CNTL);
2958
2959 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1);
2960 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1);
2961 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, enable ? 0 : 1);
2962 WREG32_SOC15_RLC(GC, 0, mmCP_ME_CNTL, tmp);
2963 udelay(50);
2964 }
2965
gfx_v9_0_cp_gfx_load_microcode(struct amdgpu_device * adev)2966 static int gfx_v9_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2967 {
2968 const struct gfx_firmware_header_v1_0 *pfp_hdr;
2969 const struct gfx_firmware_header_v1_0 *ce_hdr;
2970 const struct gfx_firmware_header_v1_0 *me_hdr;
2971 const __le32 *fw_data;
2972 unsigned i, fw_size;
2973
2974 if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2975 return -EINVAL;
2976
2977 pfp_hdr = (const struct gfx_firmware_header_v1_0 *)
2978 adev->gfx.pfp_fw->data;
2979 ce_hdr = (const struct gfx_firmware_header_v1_0 *)
2980 adev->gfx.ce_fw->data;
2981 me_hdr = (const struct gfx_firmware_header_v1_0 *)
2982 adev->gfx.me_fw->data;
2983
2984 amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2985 amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2986 amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2987
2988 gfx_v9_0_cp_gfx_enable(adev, false);
2989
2990 /* PFP */
2991 fw_data = (const __le32 *)
2992 (adev->gfx.pfp_fw->data +
2993 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2994 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4;
2995 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, 0);
2996 for (i = 0; i < fw_size; i++)
2997 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++));
2998 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2999
3000 /* CE */
3001 fw_data = (const __le32 *)
3002 (adev->gfx.ce_fw->data +
3003 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
3004 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4;
3005 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, 0);
3006 for (i = 0; i < fw_size; i++)
3007 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++));
3008 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version);
3009
3010 /* ME */
3011 fw_data = (const __le32 *)
3012 (adev->gfx.me_fw->data +
3013 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
3014 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4;
3015 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, 0);
3016 for (i = 0; i < fw_size; i++)
3017 WREG32_SOC15(GC, 0, mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++));
3018 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version);
3019
3020 return 0;
3021 }
3022
gfx_v9_0_cp_gfx_start(struct amdgpu_device * adev)3023 static int gfx_v9_0_cp_gfx_start(struct amdgpu_device *adev)
3024 {
3025 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
3026 const struct cs_section_def *sect = NULL;
3027 const struct cs_extent_def *ext = NULL;
3028 int r, i, tmp;
3029
3030 /* init the CP */
3031 WREG32_SOC15(GC, 0, mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
3032 WREG32_SOC15(GC, 0, mmCP_DEVICE_ID, 1);
3033
3034 gfx_v9_0_cp_gfx_enable(adev, true);
3035
3036 r = amdgpu_ring_alloc(ring, gfx_v9_0_get_csb_size(adev) + 4 + 3);
3037 if (r) {
3038 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
3039 return r;
3040 }
3041
3042 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3043 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3044
3045 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3046 amdgpu_ring_write(ring, 0x80000000);
3047 amdgpu_ring_write(ring, 0x80000000);
3048
3049 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
3050 for (ext = sect->section; ext->extent != NULL; ++ext) {
3051 if (sect->id == SECT_CONTEXT) {
3052 amdgpu_ring_write(ring,
3053 PACKET3(PACKET3_SET_CONTEXT_REG,
3054 ext->reg_count));
3055 amdgpu_ring_write(ring,
3056 ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
3057 for (i = 0; i < ext->reg_count; i++)
3058 amdgpu_ring_write(ring, ext->extent[i]);
3059 }
3060 }
3061 }
3062
3063 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3064 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
3065
3066 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3067 amdgpu_ring_write(ring, 0);
3068
3069 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
3070 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
3071 amdgpu_ring_write(ring, 0x8000);
3072 amdgpu_ring_write(ring, 0x8000);
3073
3074 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG,1));
3075 tmp = (PACKET3_SET_UCONFIG_REG_INDEX_TYPE |
3076 (SOC15_REG_OFFSET(GC, 0, mmVGT_INDEX_TYPE) - PACKET3_SET_UCONFIG_REG_START));
3077 amdgpu_ring_write(ring, tmp);
3078 amdgpu_ring_write(ring, 0);
3079
3080 amdgpu_ring_commit(ring);
3081
3082 return 0;
3083 }
3084
gfx_v9_0_cp_gfx_resume(struct amdgpu_device * adev)3085 static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev)
3086 {
3087 struct amdgpu_ring *ring;
3088 u32 tmp;
3089 u32 rb_bufsz;
3090 u64 rb_addr, rptr_addr, wptr_gpu_addr;
3091
3092 /* Set the write pointer delay */
3093 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_DELAY, 0);
3094
3095 /* set the RB to use vmid 0 */
3096 WREG32_SOC15(GC, 0, mmCP_RB_VMID, 0);
3097
3098 /* Set ring buffer size */
3099 ring = &adev->gfx.gfx_ring[0];
3100 rb_bufsz = order_base_2(ring->ring_size / 8);
3101 tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
3102 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
3103 #ifdef __BIG_ENDIAN
3104 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
3105 #endif
3106 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3107
3108 /* Initialize the ring buffer's write pointers */
3109 ring->wptr = 0;
3110 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
3111 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
3112
3113 /* set the wb address wether it's enabled or not */
3114 rptr_addr = ring->rptr_gpu_addr;
3115 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
3116 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3117
3118 wptr_gpu_addr = ring->wptr_gpu_addr;
3119 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
3120 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
3121
3122 mdelay(1);
3123 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3124
3125 rb_addr = ring->gpu_addr >> 8;
3126 WREG32_SOC15(GC, 0, mmCP_RB0_BASE, rb_addr);
3127 WREG32_SOC15(GC, 0, mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
3128
3129 tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL);
3130 if (ring->use_doorbell) {
3131 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3132 DOORBELL_OFFSET, ring->doorbell_index);
3133 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3134 DOORBELL_EN, 1);
3135 } else {
3136 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
3137 }
3138 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, tmp);
3139
3140 tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
3141 DOORBELL_RANGE_LOWER, ring->doorbell_index);
3142 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
3143
3144 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER,
3145 CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
3146
3147
3148 /* start the ring */
3149 gfx_v9_0_cp_gfx_start(adev);
3150
3151 return 0;
3152 }
3153
gfx_v9_0_cp_compute_enable(struct amdgpu_device * adev,bool enable)3154 static void gfx_v9_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
3155 {
3156 if (enable) {
3157 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL, 0);
3158 } else {
3159 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL,
3160 (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
3161 adev->gfx.kiq[0].ring.sched.ready = false;
3162 }
3163 udelay(50);
3164 }
3165
gfx_v9_0_cp_compute_load_microcode(struct amdgpu_device * adev)3166 static int gfx_v9_0_cp_compute_load_microcode(struct amdgpu_device *adev)
3167 {
3168 const struct gfx_firmware_header_v1_0 *mec_hdr;
3169 const __le32 *fw_data;
3170 unsigned i;
3171 u32 tmp;
3172
3173 if (!adev->gfx.mec_fw)
3174 return -EINVAL;
3175
3176 gfx_v9_0_cp_compute_enable(adev, false);
3177
3178 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3179 amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3180
3181 fw_data = (const __le32 *)
3182 (adev->gfx.mec_fw->data +
3183 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
3184 tmp = 0;
3185 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
3186 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
3187 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL, tmp);
3188
3189 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO,
3190 adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000);
3191 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI,
3192 upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
3193
3194 /* MEC1 */
3195 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3196 mec_hdr->jt_offset);
3197 for (i = 0; i < mec_hdr->jt_size; i++)
3198 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_DATA,
3199 le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
3200
3201 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3202 adev->gfx.mec_fw_version);
3203 /* Todo : Loading MEC2 firmware is only necessary if MEC2 should run different microcode than MEC1. */
3204
3205 return 0;
3206 }
3207
3208 /* KIQ functions */
gfx_v9_0_kiq_setting(struct amdgpu_ring * ring)3209 static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)
3210 {
3211 uint32_t tmp;
3212 struct amdgpu_device *adev = ring->adev;
3213
3214 /* tell RLC which is KIQ queue */
3215 tmp = RREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS);
3216 tmp &= 0xffffff00;
3217 tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
3218 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3219 tmp |= 0x80;
3220 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3221 }
3222
gfx_v9_0_mqd_set_priority(struct amdgpu_ring * ring,struct v9_mqd * mqd)3223 static void gfx_v9_0_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd)
3224 {
3225 struct amdgpu_device *adev = ring->adev;
3226
3227 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
3228 if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) {
3229 mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
3230 mqd->cp_hqd_queue_priority =
3231 AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
3232 }
3233 }
3234 }
3235
gfx_v9_0_mqd_init(struct amdgpu_ring * ring)3236 static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring)
3237 {
3238 struct amdgpu_device *adev = ring->adev;
3239 struct v9_mqd *mqd = ring->mqd_ptr;
3240 uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
3241 uint32_t tmp;
3242
3243 mqd->header = 0xC0310800;
3244 mqd->compute_pipelinestat_enable = 0x00000001;
3245 mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
3246 mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
3247 mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
3248 mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
3249 mqd->compute_static_thread_mgmt_se4 = 0xffffffff;
3250 mqd->compute_static_thread_mgmt_se5 = 0xffffffff;
3251 mqd->compute_static_thread_mgmt_se6 = 0xffffffff;
3252 mqd->compute_static_thread_mgmt_se7 = 0xffffffff;
3253 mqd->compute_misc_reserved = 0x00000003;
3254
3255 mqd->dynamic_cu_mask_addr_lo =
3256 lower_32_bits(ring->mqd_gpu_addr
3257 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3258 mqd->dynamic_cu_mask_addr_hi =
3259 upper_32_bits(ring->mqd_gpu_addr
3260 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3261
3262 eop_base_addr = ring->eop_gpu_addr >> 8;
3263 mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
3264 mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
3265
3266 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3267 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL);
3268 tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
3269 (order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1));
3270
3271 mqd->cp_hqd_eop_control = tmp;
3272
3273 /* enable doorbell? */
3274 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
3275
3276 if (ring->use_doorbell) {
3277 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3278 DOORBELL_OFFSET, ring->doorbell_index);
3279 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3280 DOORBELL_EN, 1);
3281 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3282 DOORBELL_SOURCE, 0);
3283 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3284 DOORBELL_HIT, 0);
3285 } else {
3286 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3287 DOORBELL_EN, 0);
3288 }
3289
3290 mqd->cp_hqd_pq_doorbell_control = tmp;
3291
3292 /* disable the queue if it's active */
3293 ring->wptr = 0;
3294 mqd->cp_hqd_dequeue_request = 0;
3295 mqd->cp_hqd_pq_rptr = 0;
3296 mqd->cp_hqd_pq_wptr_lo = 0;
3297 mqd->cp_hqd_pq_wptr_hi = 0;
3298
3299 /* set the pointer to the MQD */
3300 mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
3301 mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
3302
3303 /* set MQD vmid to 0 */
3304 tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL);
3305 tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
3306 mqd->cp_mqd_control = tmp;
3307
3308 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3309 hqd_gpu_addr = ring->gpu_addr >> 8;
3310 mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
3311 mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3312
3313 /* set up the HQD, this is similar to CP_RB0_CNTL */
3314 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL);
3315 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
3316 (order_base_2(ring->ring_size / 4) - 1));
3317 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
3318 (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1));
3319 #ifdef __BIG_ENDIAN
3320 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
3321 #endif
3322 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
3323 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
3324 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
3325 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
3326 mqd->cp_hqd_pq_control = tmp;
3327
3328 /* set the wb address whether it's enabled or not */
3329 wb_gpu_addr = ring->rptr_gpu_addr;
3330 mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
3331 mqd->cp_hqd_pq_rptr_report_addr_hi =
3332 upper_32_bits(wb_gpu_addr) & 0xffff;
3333
3334 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3335 wb_gpu_addr = ring->wptr_gpu_addr;
3336 mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3337 mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3338
3339 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3340 ring->wptr = 0;
3341 mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR);
3342
3343 /* set the vmid for the queue */
3344 mqd->cp_hqd_vmid = 0;
3345
3346 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE);
3347 tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
3348 mqd->cp_hqd_persistent_state = tmp;
3349
3350 /* set MIN_IB_AVAIL_SIZE */
3351 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL);
3352 tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
3353 mqd->cp_hqd_ib_control = tmp;
3354
3355 /* set static priority for a queue/ring */
3356 gfx_v9_0_mqd_set_priority(ring, mqd);
3357 mqd->cp_hqd_quantum = RREG32_SOC15(GC, 0, mmCP_HQD_QUANTUM);
3358
3359 /* map_queues packet doesn't need activate the queue,
3360 * so only kiq need set this field.
3361 */
3362 if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
3363 mqd->cp_hqd_active = 1;
3364
3365 return 0;
3366 }
3367
gfx_v9_0_kiq_init_register(struct amdgpu_ring * ring)3368 static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring)
3369 {
3370 struct amdgpu_device *adev = ring->adev;
3371 struct v9_mqd *mqd = ring->mqd_ptr;
3372 int j;
3373
3374 /* disable wptr polling */
3375 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3376
3377 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR,
3378 mqd->cp_hqd_eop_base_addr_lo);
3379 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI,
3380 mqd->cp_hqd_eop_base_addr_hi);
3381
3382 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3383 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_CONTROL,
3384 mqd->cp_hqd_eop_control);
3385
3386 /* enable doorbell? */
3387 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3388 mqd->cp_hqd_pq_doorbell_control);
3389
3390 /* disable the queue if it's active */
3391 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3392 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3393 for (j = 0; j < adev->usec_timeout; j++) {
3394 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3395 break;
3396 udelay(1);
3397 }
3398 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3399 mqd->cp_hqd_dequeue_request);
3400 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR,
3401 mqd->cp_hqd_pq_rptr);
3402 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3403 mqd->cp_hqd_pq_wptr_lo);
3404 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3405 mqd->cp_hqd_pq_wptr_hi);
3406 }
3407
3408 /* set the pointer to the MQD */
3409 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR,
3410 mqd->cp_mqd_base_addr_lo);
3411 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR_HI,
3412 mqd->cp_mqd_base_addr_hi);
3413
3414 /* set MQD vmid to 0 */
3415 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_CONTROL,
3416 mqd->cp_mqd_control);
3417
3418 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3419 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE,
3420 mqd->cp_hqd_pq_base_lo);
3421 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE_HI,
3422 mqd->cp_hqd_pq_base_hi);
3423
3424 /* set up the HQD, this is similar to CP_RB0_CNTL */
3425 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_CONTROL,
3426 mqd->cp_hqd_pq_control);
3427
3428 /* set the wb address whether it's enabled or not */
3429 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR,
3430 mqd->cp_hqd_pq_rptr_report_addr_lo);
3431 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3432 mqd->cp_hqd_pq_rptr_report_addr_hi);
3433
3434 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3435 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR,
3436 mqd->cp_hqd_pq_wptr_poll_addr_lo);
3437 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
3438 mqd->cp_hqd_pq_wptr_poll_addr_hi);
3439
3440 /* enable the doorbell if requested */
3441 if (ring->use_doorbell) {
3442 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER,
3443 (adev->doorbell_index.kiq * 2) << 2);
3444 /* If GC has entered CGPG, ringing doorbell > first page
3445 * doesn't wakeup GC. Enlarge CP_MEC_DOORBELL_RANGE_UPPER to
3446 * workaround this issue. And this change has to align with firmware
3447 * update.
3448 */
3449 if (check_if_enlarge_doorbell_range(adev))
3450 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
3451 (adev->doorbell.size - 4));
3452 else
3453 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
3454 (adev->doorbell_index.userqueue_end * 2) << 2);
3455 }
3456
3457 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3458 mqd->cp_hqd_pq_doorbell_control);
3459
3460 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3461 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3462 mqd->cp_hqd_pq_wptr_lo);
3463 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3464 mqd->cp_hqd_pq_wptr_hi);
3465
3466 /* set the vmid for the queue */
3467 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_VMID, mqd->cp_hqd_vmid);
3468
3469 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE,
3470 mqd->cp_hqd_persistent_state);
3471
3472 /* activate the queue */
3473 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE,
3474 mqd->cp_hqd_active);
3475
3476 if (ring->use_doorbell)
3477 WREG32_FIELD15(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
3478
3479 return 0;
3480 }
3481
gfx_v9_0_kiq_fini_register(struct amdgpu_ring * ring)3482 static int gfx_v9_0_kiq_fini_register(struct amdgpu_ring *ring)
3483 {
3484 struct amdgpu_device *adev = ring->adev;
3485 int j;
3486
3487 /* disable the queue if it's active */
3488 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3489
3490 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3491
3492 for (j = 0; j < adev->usec_timeout; j++) {
3493 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3494 break;
3495 udelay(1);
3496 }
3497
3498 if (j == AMDGPU_MAX_USEC_TIMEOUT) {
3499 DRM_DEBUG("KIQ dequeue request failed.\n");
3500
3501 /* Manual disable if dequeue request times out */
3502 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE, 0);
3503 }
3504
3505 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3506 0);
3507 }
3508
3509 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IQ_TIMER, 0);
3510 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IB_CONTROL, 0);
3511 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE, 0);
3512 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000);
3513 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0);
3514 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR, 0);
3515 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 0);
3516 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 0);
3517
3518 return 0;
3519 }
3520
gfx_v9_0_kiq_init_queue(struct amdgpu_ring * ring)3521 static int gfx_v9_0_kiq_init_queue(struct amdgpu_ring *ring)
3522 {
3523 struct amdgpu_device *adev = ring->adev;
3524 struct v9_mqd *mqd = ring->mqd_ptr;
3525 struct v9_mqd *tmp_mqd;
3526
3527 gfx_v9_0_kiq_setting(ring);
3528
3529 /* GPU could be in bad state during probe, driver trigger the reset
3530 * after load the SMU, in this case , the mqd is not be initialized.
3531 * driver need to re-init the mqd.
3532 * check mqd->cp_hqd_pq_control since this value should not be 0
3533 */
3534 tmp_mqd = (struct v9_mqd *)adev->gfx.kiq[0].mqd_backup;
3535 if (amdgpu_in_reset(adev) && tmp_mqd->cp_hqd_pq_control){
3536 /* for GPU_RESET case , reset MQD to a clean status */
3537 if (adev->gfx.kiq[0].mqd_backup)
3538 memcpy(mqd, adev->gfx.kiq[0].mqd_backup, sizeof(struct v9_mqd_allocation));
3539
3540 /* reset ring buffer */
3541 ring->wptr = 0;
3542 amdgpu_ring_clear_ring(ring);
3543
3544 mutex_lock(&adev->srbm_mutex);
3545 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
3546 gfx_v9_0_kiq_init_register(ring);
3547 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
3548 mutex_unlock(&adev->srbm_mutex);
3549 } else {
3550 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3551 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3552 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3553 if (amdgpu_sriov_vf(adev) && adev->in_suspend)
3554 amdgpu_ring_clear_ring(ring);
3555 mutex_lock(&adev->srbm_mutex);
3556 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
3557 gfx_v9_0_mqd_init(ring);
3558 gfx_v9_0_kiq_init_register(ring);
3559 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
3560 mutex_unlock(&adev->srbm_mutex);
3561
3562 if (adev->gfx.kiq[0].mqd_backup)
3563 memcpy(adev->gfx.kiq[0].mqd_backup, mqd, sizeof(struct v9_mqd_allocation));
3564 }
3565
3566 return 0;
3567 }
3568
gfx_v9_0_kcq_init_queue(struct amdgpu_ring * ring)3569 static int gfx_v9_0_kcq_init_queue(struct amdgpu_ring *ring)
3570 {
3571 struct amdgpu_device *adev = ring->adev;
3572 struct v9_mqd *mqd = ring->mqd_ptr;
3573 int mqd_idx = ring - &adev->gfx.compute_ring[0];
3574 struct v9_mqd *tmp_mqd;
3575
3576 /* Same as above kiq init, driver need to re-init the mqd if mqd->cp_hqd_pq_control
3577 * is not be initialized before
3578 */
3579 tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx];
3580
3581 if (!tmp_mqd->cp_hqd_pq_control ||
3582 (!amdgpu_in_reset(adev) && !adev->in_suspend)) {
3583 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3584 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3585 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3586 mutex_lock(&adev->srbm_mutex);
3587 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
3588 gfx_v9_0_mqd_init(ring);
3589 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
3590 mutex_unlock(&adev->srbm_mutex);
3591
3592 if (adev->gfx.mec.mqd_backup[mqd_idx])
3593 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
3594 } else {
3595 /* restore MQD to a clean status */
3596 if (adev->gfx.mec.mqd_backup[mqd_idx])
3597 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
3598 /* reset ring buffer */
3599 ring->wptr = 0;
3600 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, 0);
3601 amdgpu_ring_clear_ring(ring);
3602 }
3603
3604 return 0;
3605 }
3606
gfx_v9_0_kiq_resume(struct amdgpu_device * adev)3607 static int gfx_v9_0_kiq_resume(struct amdgpu_device *adev)
3608 {
3609 struct amdgpu_ring *ring;
3610 int r;
3611
3612 ring = &adev->gfx.kiq[0].ring;
3613
3614 r = amdgpu_bo_reserve(ring->mqd_obj, false);
3615 if (unlikely(r != 0))
3616 return r;
3617
3618 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3619 if (unlikely(r != 0)) {
3620 amdgpu_bo_unreserve(ring->mqd_obj);
3621 return r;
3622 }
3623
3624 gfx_v9_0_kiq_init_queue(ring);
3625 amdgpu_bo_kunmap(ring->mqd_obj);
3626 ring->mqd_ptr = NULL;
3627 amdgpu_bo_unreserve(ring->mqd_obj);
3628 return 0;
3629 }
3630
gfx_v9_0_kcq_resume(struct amdgpu_device * adev)3631 static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev)
3632 {
3633 struct amdgpu_ring *ring = NULL;
3634 int r = 0, i;
3635
3636 gfx_v9_0_cp_compute_enable(adev, true);
3637
3638 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3639 ring = &adev->gfx.compute_ring[i];
3640
3641 r = amdgpu_bo_reserve(ring->mqd_obj, false);
3642 if (unlikely(r != 0))
3643 goto done;
3644 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3645 if (!r) {
3646 r = gfx_v9_0_kcq_init_queue(ring);
3647 amdgpu_bo_kunmap(ring->mqd_obj);
3648 ring->mqd_ptr = NULL;
3649 }
3650 amdgpu_bo_unreserve(ring->mqd_obj);
3651 if (r)
3652 goto done;
3653 }
3654
3655 r = amdgpu_gfx_enable_kcq(adev, 0);
3656 done:
3657 return r;
3658 }
3659
gfx_v9_0_cp_resume(struct amdgpu_device * adev)3660 static int gfx_v9_0_cp_resume(struct amdgpu_device *adev)
3661 {
3662 int r, i;
3663 struct amdgpu_ring *ring;
3664
3665 if (!(adev->flags & AMD_IS_APU))
3666 gfx_v9_0_enable_gui_idle_interrupt(adev, false);
3667
3668 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
3669 if (adev->gfx.num_gfx_rings) {
3670 /* legacy firmware loading */
3671 r = gfx_v9_0_cp_gfx_load_microcode(adev);
3672 if (r)
3673 return r;
3674 }
3675
3676 r = gfx_v9_0_cp_compute_load_microcode(adev);
3677 if (r)
3678 return r;
3679 }
3680
3681 r = gfx_v9_0_kiq_resume(adev);
3682 if (r)
3683 return r;
3684
3685 if (adev->gfx.num_gfx_rings) {
3686 r = gfx_v9_0_cp_gfx_resume(adev);
3687 if (r)
3688 return r;
3689 }
3690
3691 r = gfx_v9_0_kcq_resume(adev);
3692 if (r)
3693 return r;
3694
3695 if (adev->gfx.num_gfx_rings) {
3696 ring = &adev->gfx.gfx_ring[0];
3697 r = amdgpu_ring_test_helper(ring);
3698 if (r)
3699 return r;
3700 }
3701
3702 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3703 ring = &adev->gfx.compute_ring[i];
3704 amdgpu_ring_test_helper(ring);
3705 }
3706
3707 gfx_v9_0_enable_gui_idle_interrupt(adev, true);
3708
3709 return 0;
3710 }
3711
gfx_v9_0_init_tcp_config(struct amdgpu_device * adev)3712 static void gfx_v9_0_init_tcp_config(struct amdgpu_device *adev)
3713 {
3714 u32 tmp;
3715
3716 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1) &&
3717 adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 2))
3718 return;
3719
3720 tmp = RREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG);
3721 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE64KHASH,
3722 adev->df.hash_status.hash_64k);
3723 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE2MHASH,
3724 adev->df.hash_status.hash_2m);
3725 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE1GHASH,
3726 adev->df.hash_status.hash_1g);
3727 WREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG, tmp);
3728 }
3729
gfx_v9_0_cp_enable(struct amdgpu_device * adev,bool enable)3730 static void gfx_v9_0_cp_enable(struct amdgpu_device *adev, bool enable)
3731 {
3732 if (adev->gfx.num_gfx_rings)
3733 gfx_v9_0_cp_gfx_enable(adev, enable);
3734 gfx_v9_0_cp_compute_enable(adev, enable);
3735 }
3736
gfx_v9_0_hw_init(void * handle)3737 static int gfx_v9_0_hw_init(void *handle)
3738 {
3739 int r;
3740 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3741
3742 if (!amdgpu_sriov_vf(adev))
3743 gfx_v9_0_init_golden_registers(adev);
3744
3745 gfx_v9_0_constants_init(adev);
3746
3747 gfx_v9_0_init_tcp_config(adev);
3748
3749 r = adev->gfx.rlc.funcs->resume(adev);
3750 if (r)
3751 return r;
3752
3753 r = gfx_v9_0_cp_resume(adev);
3754 if (r)
3755 return r;
3756
3757 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
3758 gfx_v9_4_2_set_power_brake_sequence(adev);
3759
3760 return r;
3761 }
3762
gfx_v9_0_hw_fini(void * handle)3763 static int gfx_v9_0_hw_fini(void *handle)
3764 {
3765 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3766
3767 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
3768 amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
3769 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
3770 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
3771
3772 /* DF freeze and kcq disable will fail */
3773 if (!amdgpu_ras_intr_triggered())
3774 /* disable KCQ to avoid CPC touch memory not valid anymore */
3775 amdgpu_gfx_disable_kcq(adev, 0);
3776
3777 if (amdgpu_sriov_vf(adev)) {
3778 gfx_v9_0_cp_gfx_enable(adev, false);
3779 /* must disable polling for SRIOV when hw finished, otherwise
3780 * CPC engine may still keep fetching WB address which is already
3781 * invalid after sw finished and trigger DMAR reading error in
3782 * hypervisor side.
3783 */
3784 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3785 return 0;
3786 }
3787
3788 /* Use deinitialize sequence from CAIL when unbinding device from driver,
3789 * otherwise KIQ is hanging when binding back
3790 */
3791 if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
3792 mutex_lock(&adev->srbm_mutex);
3793 soc15_grbm_select(adev, adev->gfx.kiq[0].ring.me,
3794 adev->gfx.kiq[0].ring.pipe,
3795 adev->gfx.kiq[0].ring.queue, 0, 0);
3796 gfx_v9_0_kiq_fini_register(&adev->gfx.kiq[0].ring);
3797 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
3798 mutex_unlock(&adev->srbm_mutex);
3799 }
3800
3801 gfx_v9_0_cp_enable(adev, false);
3802
3803 /* Skip stopping RLC with A+A reset or when RLC controls GFX clock */
3804 if ((adev->gmc.xgmi.connected_to_cpu && amdgpu_in_reset(adev)) ||
3805 (adev->ip_versions[GC_HWIP][0] >= IP_VERSION(9, 4, 2))) {
3806 dev_dbg(adev->dev, "Skipping RLC halt\n");
3807 return 0;
3808 }
3809
3810 adev->gfx.rlc.funcs->stop(adev);
3811 return 0;
3812 }
3813
gfx_v9_0_suspend(void * handle)3814 static int gfx_v9_0_suspend(void *handle)
3815 {
3816 return gfx_v9_0_hw_fini(handle);
3817 }
3818
gfx_v9_0_resume(void * handle)3819 static int gfx_v9_0_resume(void *handle)
3820 {
3821 return gfx_v9_0_hw_init(handle);
3822 }
3823
gfx_v9_0_is_idle(void * handle)3824 static bool gfx_v9_0_is_idle(void *handle)
3825 {
3826 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3827
3828 if (REG_GET_FIELD(RREG32_SOC15(GC, 0, mmGRBM_STATUS),
3829 GRBM_STATUS, GUI_ACTIVE))
3830 return false;
3831 else
3832 return true;
3833 }
3834
gfx_v9_0_wait_for_idle(void * handle)3835 static int gfx_v9_0_wait_for_idle(void *handle)
3836 {
3837 unsigned i;
3838 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3839
3840 for (i = 0; i < adev->usec_timeout; i++) {
3841 if (gfx_v9_0_is_idle(handle))
3842 return 0;
3843 udelay(1);
3844 }
3845 return -ETIMEDOUT;
3846 }
3847
gfx_v9_0_soft_reset(void * handle)3848 static int gfx_v9_0_soft_reset(void *handle)
3849 {
3850 u32 grbm_soft_reset = 0;
3851 u32 tmp;
3852 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3853
3854 /* GRBM_STATUS */
3855 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS);
3856 if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
3857 GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
3858 GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
3859 GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
3860 GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
3861 GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) {
3862 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3863 GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
3864 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3865 GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
3866 }
3867
3868 if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
3869 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3870 GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
3871 }
3872
3873 /* GRBM_STATUS2 */
3874 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2);
3875 if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
3876 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3877 GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
3878
3879
3880 if (grbm_soft_reset) {
3881 /* stop the rlc */
3882 adev->gfx.rlc.funcs->stop(adev);
3883
3884 if (adev->gfx.num_gfx_rings)
3885 /* Disable GFX parsing/prefetching */
3886 gfx_v9_0_cp_gfx_enable(adev, false);
3887
3888 /* Disable MEC parsing/prefetching */
3889 gfx_v9_0_cp_compute_enable(adev, false);
3890
3891 if (grbm_soft_reset) {
3892 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3893 tmp |= grbm_soft_reset;
3894 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
3895 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
3896 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3897
3898 udelay(50);
3899
3900 tmp &= ~grbm_soft_reset;
3901 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
3902 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3903 }
3904
3905 /* Wait a little for things to settle down */
3906 udelay(50);
3907 }
3908 return 0;
3909 }
3910
gfx_v9_0_kiq_read_clock(struct amdgpu_device * adev)3911 static uint64_t gfx_v9_0_kiq_read_clock(struct amdgpu_device *adev)
3912 {
3913 signed long r, cnt = 0;
3914 unsigned long flags;
3915 uint32_t seq, reg_val_offs = 0;
3916 uint64_t value = 0;
3917 struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
3918 struct amdgpu_ring *ring = &kiq->ring;
3919
3920 BUG_ON(!ring->funcs->emit_rreg);
3921
3922 spin_lock_irqsave(&kiq->ring_lock, flags);
3923 if (amdgpu_device_wb_get(adev, ®_val_offs)) {
3924 pr_err("critical bug! too many kiq readers\n");
3925 goto failed_unlock;
3926 }
3927 amdgpu_ring_alloc(ring, 32);
3928 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
3929 amdgpu_ring_write(ring, 9 | /* src: register*/
3930 (5 << 8) | /* dst: memory */
3931 (1 << 16) | /* count sel */
3932 (1 << 20)); /* write confirm */
3933 amdgpu_ring_write(ring, 0);
3934 amdgpu_ring_write(ring, 0);
3935 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
3936 reg_val_offs * 4));
3937 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
3938 reg_val_offs * 4));
3939 r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT);
3940 if (r)
3941 goto failed_undo;
3942
3943 amdgpu_ring_commit(ring);
3944 spin_unlock_irqrestore(&kiq->ring_lock, flags);
3945
3946 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
3947
3948 /* don't wait anymore for gpu reset case because this way may
3949 * block gpu_recover() routine forever, e.g. this virt_kiq_rreg
3950 * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will
3951 * never return if we keep waiting in virt_kiq_rreg, which cause
3952 * gpu_recover() hang there.
3953 *
3954 * also don't wait anymore for IRQ context
3955 * */
3956 if (r < 1 && (amdgpu_in_reset(adev)))
3957 goto failed_kiq_read;
3958
3959 might_sleep();
3960 while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) {
3961 msleep(MAX_KIQ_REG_BAILOUT_INTERVAL);
3962 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
3963 }
3964
3965 if (cnt > MAX_KIQ_REG_TRY)
3966 goto failed_kiq_read;
3967
3968 mb();
3969 value = (uint64_t)adev->wb.wb[reg_val_offs] |
3970 (uint64_t)adev->wb.wb[reg_val_offs + 1 ] << 32ULL;
3971 amdgpu_device_wb_free(adev, reg_val_offs);
3972 return value;
3973
3974 failed_undo:
3975 amdgpu_ring_undo(ring);
3976 failed_unlock:
3977 spin_unlock_irqrestore(&kiq->ring_lock, flags);
3978 failed_kiq_read:
3979 if (reg_val_offs)
3980 amdgpu_device_wb_free(adev, reg_val_offs);
3981 pr_err("failed to read gpu clock\n");
3982 return ~0;
3983 }
3984
gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device * adev)3985 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev)
3986 {
3987 uint64_t clock, clock_lo, clock_hi, hi_check;
3988
3989 switch (adev->ip_versions[GC_HWIP][0]) {
3990 case IP_VERSION(9, 3, 0):
3991 preempt_disable();
3992 clock_hi = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir);
3993 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir);
3994 hi_check = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir);
3995 /* The SMUIO TSC clock frequency is 100MHz, which sets 32-bit carry over
3996 * roughly every 42 seconds.
3997 */
3998 if (hi_check != clock_hi) {
3999 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir);
4000 clock_hi = hi_check;
4001 }
4002 preempt_enable();
4003 clock = clock_lo | (clock_hi << 32ULL);
4004 break;
4005 default:
4006 amdgpu_gfx_off_ctrl(adev, false);
4007 mutex_lock(&adev->gfx.gpu_clock_mutex);
4008 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 0, 1) && amdgpu_sriov_runtime(adev)) {
4009 clock = gfx_v9_0_kiq_read_clock(adev);
4010 } else {
4011 WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4012 clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB) |
4013 ((uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4014 }
4015 mutex_unlock(&adev->gfx.gpu_clock_mutex);
4016 amdgpu_gfx_off_ctrl(adev, true);
4017 break;
4018 }
4019 return clock;
4020 }
4021
gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring * ring,uint32_t vmid,uint32_t gds_base,uint32_t gds_size,uint32_t gws_base,uint32_t gws_size,uint32_t oa_base,uint32_t oa_size)4022 static void gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4023 uint32_t vmid,
4024 uint32_t gds_base, uint32_t gds_size,
4025 uint32_t gws_base, uint32_t gws_size,
4026 uint32_t oa_base, uint32_t oa_size)
4027 {
4028 struct amdgpu_device *adev = ring->adev;
4029
4030 /* GDS Base */
4031 gfx_v9_0_write_data_to_reg(ring, 0, false,
4032 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_BASE) + 2 * vmid,
4033 gds_base);
4034
4035 /* GDS Size */
4036 gfx_v9_0_write_data_to_reg(ring, 0, false,
4037 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE) + 2 * vmid,
4038 gds_size);
4039
4040 /* GWS */
4041 gfx_v9_0_write_data_to_reg(ring, 0, false,
4042 SOC15_REG_OFFSET(GC, 0, mmGDS_GWS_VMID0) + vmid,
4043 gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4044
4045 /* OA */
4046 gfx_v9_0_write_data_to_reg(ring, 0, false,
4047 SOC15_REG_OFFSET(GC, 0, mmGDS_OA_VMID0) + vmid,
4048 (1 << (oa_size + oa_base)) - (1 << oa_base));
4049 }
4050
4051 static const u32 vgpr_init_compute_shader[] =
4052 {
4053 0xb07c0000, 0xbe8000ff,
4054 0x000000f8, 0xbf110800,
4055 0x7e000280, 0x7e020280,
4056 0x7e040280, 0x7e060280,
4057 0x7e080280, 0x7e0a0280,
4058 0x7e0c0280, 0x7e0e0280,
4059 0x80808800, 0xbe803200,
4060 0xbf84fff5, 0xbf9c0000,
4061 0xd28c0001, 0x0001007f,
4062 0xd28d0001, 0x0002027e,
4063 0x10020288, 0xb8810904,
4064 0xb7814000, 0xd1196a01,
4065 0x00000301, 0xbe800087,
4066 0xbefc00c1, 0xd89c4000,
4067 0x00020201, 0xd89cc080,
4068 0x00040401, 0x320202ff,
4069 0x00000800, 0x80808100,
4070 0xbf84fff8, 0x7e020280,
4071 0xbf810000, 0x00000000,
4072 };
4073
4074 static const u32 sgpr_init_compute_shader[] =
4075 {
4076 0xb07c0000, 0xbe8000ff,
4077 0x0000005f, 0xbee50080,
4078 0xbe812c65, 0xbe822c65,
4079 0xbe832c65, 0xbe842c65,
4080 0xbe852c65, 0xb77c0005,
4081 0x80808500, 0xbf84fff8,
4082 0xbe800080, 0xbf810000,
4083 };
4084
4085 static const u32 vgpr_init_compute_shader_arcturus[] = {
4086 0xd3d94000, 0x18000080, 0xd3d94001, 0x18000080, 0xd3d94002, 0x18000080,
4087 0xd3d94003, 0x18000080, 0xd3d94004, 0x18000080, 0xd3d94005, 0x18000080,
4088 0xd3d94006, 0x18000080, 0xd3d94007, 0x18000080, 0xd3d94008, 0x18000080,
4089 0xd3d94009, 0x18000080, 0xd3d9400a, 0x18000080, 0xd3d9400b, 0x18000080,
4090 0xd3d9400c, 0x18000080, 0xd3d9400d, 0x18000080, 0xd3d9400e, 0x18000080,
4091 0xd3d9400f, 0x18000080, 0xd3d94010, 0x18000080, 0xd3d94011, 0x18000080,
4092 0xd3d94012, 0x18000080, 0xd3d94013, 0x18000080, 0xd3d94014, 0x18000080,
4093 0xd3d94015, 0x18000080, 0xd3d94016, 0x18000080, 0xd3d94017, 0x18000080,
4094 0xd3d94018, 0x18000080, 0xd3d94019, 0x18000080, 0xd3d9401a, 0x18000080,
4095 0xd3d9401b, 0x18000080, 0xd3d9401c, 0x18000080, 0xd3d9401d, 0x18000080,
4096 0xd3d9401e, 0x18000080, 0xd3d9401f, 0x18000080, 0xd3d94020, 0x18000080,
4097 0xd3d94021, 0x18000080, 0xd3d94022, 0x18000080, 0xd3d94023, 0x18000080,
4098 0xd3d94024, 0x18000080, 0xd3d94025, 0x18000080, 0xd3d94026, 0x18000080,
4099 0xd3d94027, 0x18000080, 0xd3d94028, 0x18000080, 0xd3d94029, 0x18000080,
4100 0xd3d9402a, 0x18000080, 0xd3d9402b, 0x18000080, 0xd3d9402c, 0x18000080,
4101 0xd3d9402d, 0x18000080, 0xd3d9402e, 0x18000080, 0xd3d9402f, 0x18000080,
4102 0xd3d94030, 0x18000080, 0xd3d94031, 0x18000080, 0xd3d94032, 0x18000080,
4103 0xd3d94033, 0x18000080, 0xd3d94034, 0x18000080, 0xd3d94035, 0x18000080,
4104 0xd3d94036, 0x18000080, 0xd3d94037, 0x18000080, 0xd3d94038, 0x18000080,
4105 0xd3d94039, 0x18000080, 0xd3d9403a, 0x18000080, 0xd3d9403b, 0x18000080,
4106 0xd3d9403c, 0x18000080, 0xd3d9403d, 0x18000080, 0xd3d9403e, 0x18000080,
4107 0xd3d9403f, 0x18000080, 0xd3d94040, 0x18000080, 0xd3d94041, 0x18000080,
4108 0xd3d94042, 0x18000080, 0xd3d94043, 0x18000080, 0xd3d94044, 0x18000080,
4109 0xd3d94045, 0x18000080, 0xd3d94046, 0x18000080, 0xd3d94047, 0x18000080,
4110 0xd3d94048, 0x18000080, 0xd3d94049, 0x18000080, 0xd3d9404a, 0x18000080,
4111 0xd3d9404b, 0x18000080, 0xd3d9404c, 0x18000080, 0xd3d9404d, 0x18000080,
4112 0xd3d9404e, 0x18000080, 0xd3d9404f, 0x18000080, 0xd3d94050, 0x18000080,
4113 0xd3d94051, 0x18000080, 0xd3d94052, 0x18000080, 0xd3d94053, 0x18000080,
4114 0xd3d94054, 0x18000080, 0xd3d94055, 0x18000080, 0xd3d94056, 0x18000080,
4115 0xd3d94057, 0x18000080, 0xd3d94058, 0x18000080, 0xd3d94059, 0x18000080,
4116 0xd3d9405a, 0x18000080, 0xd3d9405b, 0x18000080, 0xd3d9405c, 0x18000080,
4117 0xd3d9405d, 0x18000080, 0xd3d9405e, 0x18000080, 0xd3d9405f, 0x18000080,
4118 0xd3d94060, 0x18000080, 0xd3d94061, 0x18000080, 0xd3d94062, 0x18000080,
4119 0xd3d94063, 0x18000080, 0xd3d94064, 0x18000080, 0xd3d94065, 0x18000080,
4120 0xd3d94066, 0x18000080, 0xd3d94067, 0x18000080, 0xd3d94068, 0x18000080,
4121 0xd3d94069, 0x18000080, 0xd3d9406a, 0x18000080, 0xd3d9406b, 0x18000080,
4122 0xd3d9406c, 0x18000080, 0xd3d9406d, 0x18000080, 0xd3d9406e, 0x18000080,
4123 0xd3d9406f, 0x18000080, 0xd3d94070, 0x18000080, 0xd3d94071, 0x18000080,
4124 0xd3d94072, 0x18000080, 0xd3d94073, 0x18000080, 0xd3d94074, 0x18000080,
4125 0xd3d94075, 0x18000080, 0xd3d94076, 0x18000080, 0xd3d94077, 0x18000080,
4126 0xd3d94078, 0x18000080, 0xd3d94079, 0x18000080, 0xd3d9407a, 0x18000080,
4127 0xd3d9407b, 0x18000080, 0xd3d9407c, 0x18000080, 0xd3d9407d, 0x18000080,
4128 0xd3d9407e, 0x18000080, 0xd3d9407f, 0x18000080, 0xd3d94080, 0x18000080,
4129 0xd3d94081, 0x18000080, 0xd3d94082, 0x18000080, 0xd3d94083, 0x18000080,
4130 0xd3d94084, 0x18000080, 0xd3d94085, 0x18000080, 0xd3d94086, 0x18000080,
4131 0xd3d94087, 0x18000080, 0xd3d94088, 0x18000080, 0xd3d94089, 0x18000080,
4132 0xd3d9408a, 0x18000080, 0xd3d9408b, 0x18000080, 0xd3d9408c, 0x18000080,
4133 0xd3d9408d, 0x18000080, 0xd3d9408e, 0x18000080, 0xd3d9408f, 0x18000080,
4134 0xd3d94090, 0x18000080, 0xd3d94091, 0x18000080, 0xd3d94092, 0x18000080,
4135 0xd3d94093, 0x18000080, 0xd3d94094, 0x18000080, 0xd3d94095, 0x18000080,
4136 0xd3d94096, 0x18000080, 0xd3d94097, 0x18000080, 0xd3d94098, 0x18000080,
4137 0xd3d94099, 0x18000080, 0xd3d9409a, 0x18000080, 0xd3d9409b, 0x18000080,
4138 0xd3d9409c, 0x18000080, 0xd3d9409d, 0x18000080, 0xd3d9409e, 0x18000080,
4139 0xd3d9409f, 0x18000080, 0xd3d940a0, 0x18000080, 0xd3d940a1, 0x18000080,
4140 0xd3d940a2, 0x18000080, 0xd3d940a3, 0x18000080, 0xd3d940a4, 0x18000080,
4141 0xd3d940a5, 0x18000080, 0xd3d940a6, 0x18000080, 0xd3d940a7, 0x18000080,
4142 0xd3d940a8, 0x18000080, 0xd3d940a9, 0x18000080, 0xd3d940aa, 0x18000080,
4143 0xd3d940ab, 0x18000080, 0xd3d940ac, 0x18000080, 0xd3d940ad, 0x18000080,
4144 0xd3d940ae, 0x18000080, 0xd3d940af, 0x18000080, 0xd3d940b0, 0x18000080,
4145 0xd3d940b1, 0x18000080, 0xd3d940b2, 0x18000080, 0xd3d940b3, 0x18000080,
4146 0xd3d940b4, 0x18000080, 0xd3d940b5, 0x18000080, 0xd3d940b6, 0x18000080,
4147 0xd3d940b7, 0x18000080, 0xd3d940b8, 0x18000080, 0xd3d940b9, 0x18000080,
4148 0xd3d940ba, 0x18000080, 0xd3d940bb, 0x18000080, 0xd3d940bc, 0x18000080,
4149 0xd3d940bd, 0x18000080, 0xd3d940be, 0x18000080, 0xd3d940bf, 0x18000080,
4150 0xd3d940c0, 0x18000080, 0xd3d940c1, 0x18000080, 0xd3d940c2, 0x18000080,
4151 0xd3d940c3, 0x18000080, 0xd3d940c4, 0x18000080, 0xd3d940c5, 0x18000080,
4152 0xd3d940c6, 0x18000080, 0xd3d940c7, 0x18000080, 0xd3d940c8, 0x18000080,
4153 0xd3d940c9, 0x18000080, 0xd3d940ca, 0x18000080, 0xd3d940cb, 0x18000080,
4154 0xd3d940cc, 0x18000080, 0xd3d940cd, 0x18000080, 0xd3d940ce, 0x18000080,
4155 0xd3d940cf, 0x18000080, 0xd3d940d0, 0x18000080, 0xd3d940d1, 0x18000080,
4156 0xd3d940d2, 0x18000080, 0xd3d940d3, 0x18000080, 0xd3d940d4, 0x18000080,
4157 0xd3d940d5, 0x18000080, 0xd3d940d6, 0x18000080, 0xd3d940d7, 0x18000080,
4158 0xd3d940d8, 0x18000080, 0xd3d940d9, 0x18000080, 0xd3d940da, 0x18000080,
4159 0xd3d940db, 0x18000080, 0xd3d940dc, 0x18000080, 0xd3d940dd, 0x18000080,
4160 0xd3d940de, 0x18000080, 0xd3d940df, 0x18000080, 0xd3d940e0, 0x18000080,
4161 0xd3d940e1, 0x18000080, 0xd3d940e2, 0x18000080, 0xd3d940e3, 0x18000080,
4162 0xd3d940e4, 0x18000080, 0xd3d940e5, 0x18000080, 0xd3d940e6, 0x18000080,
4163 0xd3d940e7, 0x18000080, 0xd3d940e8, 0x18000080, 0xd3d940e9, 0x18000080,
4164 0xd3d940ea, 0x18000080, 0xd3d940eb, 0x18000080, 0xd3d940ec, 0x18000080,
4165 0xd3d940ed, 0x18000080, 0xd3d940ee, 0x18000080, 0xd3d940ef, 0x18000080,
4166 0xd3d940f0, 0x18000080, 0xd3d940f1, 0x18000080, 0xd3d940f2, 0x18000080,
4167 0xd3d940f3, 0x18000080, 0xd3d940f4, 0x18000080, 0xd3d940f5, 0x18000080,
4168 0xd3d940f6, 0x18000080, 0xd3d940f7, 0x18000080, 0xd3d940f8, 0x18000080,
4169 0xd3d940f9, 0x18000080, 0xd3d940fa, 0x18000080, 0xd3d940fb, 0x18000080,
4170 0xd3d940fc, 0x18000080, 0xd3d940fd, 0x18000080, 0xd3d940fe, 0x18000080,
4171 0xd3d940ff, 0x18000080, 0xb07c0000, 0xbe8a00ff, 0x000000f8, 0xbf11080a,
4172 0x7e000280, 0x7e020280, 0x7e040280, 0x7e060280, 0x7e080280, 0x7e0a0280,
4173 0x7e0c0280, 0x7e0e0280, 0x808a880a, 0xbe80320a, 0xbf84fff5, 0xbf9c0000,
4174 0xd28c0001, 0x0001007f, 0xd28d0001, 0x0002027e, 0x10020288, 0xb88b0904,
4175 0xb78b4000, 0xd1196a01, 0x00001701, 0xbe8a0087, 0xbefc00c1, 0xd89c4000,
4176 0x00020201, 0xd89cc080, 0x00040401, 0x320202ff, 0x00000800, 0x808a810a,
4177 0xbf84fff8, 0xbf810000,
4178 };
4179
4180 /* When below register arrays changed, please update gpr_reg_size,
4181 and sec_ded_counter_reg_size in function gfx_v9_0_do_edc_gpr_workarounds,
4182 to cover all gfx9 ASICs */
4183 static const struct soc15_reg_entry vgpr_init_regs[] = {
4184 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4185 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4186 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 },
4187 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4188 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x3f },
4189 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */
4190 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
4191 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
4192 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
4193 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
4194 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff },
4195 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff },
4196 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff },
4197 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff },
4198 };
4199
4200 static const struct soc15_reg_entry vgpr_init_regs_arcturus[] = {
4201 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4202 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4203 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 },
4204 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4205 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0xbf },
4206 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */
4207 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
4208 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
4209 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
4210 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
4211 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff },
4212 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff },
4213 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff },
4214 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff },
4215 };
4216
4217 static const struct soc15_reg_entry sgpr1_init_regs[] = {
4218 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4219 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4220 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 },
4221 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4222 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */
4223 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
4224 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x000000ff },
4225 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x000000ff },
4226 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x000000ff },
4227 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x000000ff },
4228 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x000000ff },
4229 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x000000ff },
4230 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x000000ff },
4231 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x000000ff },
4232 };
4233
4234 static const struct soc15_reg_entry sgpr2_init_regs[] = {
4235 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4236 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4237 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 },
4238 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4239 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */
4240 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
4241 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x0000ff00 },
4242 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x0000ff00 },
4243 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x0000ff00 },
4244 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x0000ff00 },
4245 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x0000ff00 },
4246 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x0000ff00 },
4247 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x0000ff00 },
4248 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x0000ff00 },
4249 };
4250
4251 static const struct soc15_reg_entry gfx_v9_0_edc_counter_regs[] = {
4252 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT), 0, 1, 1},
4253 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT), 0, 1, 1},
4254 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 0, 1, 1},
4255 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT), 0, 1, 1},
4256 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 0, 1, 1},
4257 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT), 0, 1, 1},
4258 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT), 0, 1, 1},
4259 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT), 0, 1, 1},
4260 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT), 0, 1, 1},
4261 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 0, 1, 1},
4262 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_GRBM_CNT), 0, 1, 1},
4263 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_DED), 0, 1, 1},
4264 { SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT), 0, 4, 1},
4265 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 0, 4, 6},
4266 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_DED_CNT), 0, 4, 16},
4267 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_INFO), 0, 4, 16},
4268 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_SEC_CNT), 0, 4, 16},
4269 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 0, 1, 16},
4270 { SOC15_REG_ENTRY(GC, 0, mmTCP_ATC_EDC_GATCL1_CNT), 0, 4, 16},
4271 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT), 0, 4, 16},
4272 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 0, 4, 16},
4273 { SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 0, 4, 16},
4274 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 0, 4, 6},
4275 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 0, 4, 16},
4276 { SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 0, 4, 16},
4277 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 0, 1, 1},
4278 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 0, 1, 1},
4279 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 0, 1, 32},
4280 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 0, 1, 32},
4281 { SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT), 0, 1, 72},
4282 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 0, 1, 16},
4283 { SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 0, 1, 2},
4284 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 0, 4, 6},
4285 };
4286
gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device * adev)4287 static int gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device *adev)
4288 {
4289 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4290 int i, r;
4291
4292 /* only support when RAS is enabled */
4293 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4294 return 0;
4295
4296 r = amdgpu_ring_alloc(ring, 7);
4297 if (r) {
4298 DRM_ERROR("amdgpu: GDS workarounds failed to lock ring %s (%d).\n",
4299 ring->name, r);
4300 return r;
4301 }
4302
4303 WREG32_SOC15(GC, 0, mmGDS_VMID0_BASE, 0x00000000);
4304 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, adev->gds.gds_size);
4305
4306 amdgpu_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5));
4307 amdgpu_ring_write(ring, (PACKET3_DMA_DATA_CP_SYNC |
4308 PACKET3_DMA_DATA_DST_SEL(1) |
4309 PACKET3_DMA_DATA_SRC_SEL(2) |
4310 PACKET3_DMA_DATA_ENGINE(0)));
4311 amdgpu_ring_write(ring, 0);
4312 amdgpu_ring_write(ring, 0);
4313 amdgpu_ring_write(ring, 0);
4314 amdgpu_ring_write(ring, 0);
4315 amdgpu_ring_write(ring, PACKET3_DMA_DATA_CMD_RAW_WAIT |
4316 adev->gds.gds_size);
4317
4318 amdgpu_ring_commit(ring);
4319
4320 for (i = 0; i < adev->usec_timeout; i++) {
4321 if (ring->wptr == gfx_v9_0_ring_get_rptr_compute(ring))
4322 break;
4323 udelay(1);
4324 }
4325
4326 if (i >= adev->usec_timeout)
4327 r = -ETIMEDOUT;
4328
4329 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, 0x00000000);
4330
4331 return r;
4332 }
4333
gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device * adev)4334 static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
4335 {
4336 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4337 struct amdgpu_ib ib;
4338 struct dma_fence *f = NULL;
4339 int r, i;
4340 unsigned total_size, vgpr_offset, sgpr_offset;
4341 u64 gpu_addr;
4342
4343 int compute_dim_x = adev->gfx.config.max_shader_engines *
4344 adev->gfx.config.max_cu_per_sh *
4345 adev->gfx.config.max_sh_per_se;
4346 int sgpr_work_group_size = 5;
4347 int gpr_reg_size = adev->gfx.config.max_shader_engines + 6;
4348 int vgpr_init_shader_size;
4349 const u32 *vgpr_init_shader_ptr;
4350 const struct soc15_reg_entry *vgpr_init_regs_ptr;
4351
4352 /* only support when RAS is enabled */
4353 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4354 return 0;
4355
4356 /* bail if the compute ring is not ready */
4357 if (!ring->sched.ready)
4358 return 0;
4359
4360 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1)) {
4361 vgpr_init_shader_ptr = vgpr_init_compute_shader_arcturus;
4362 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader_arcturus);
4363 vgpr_init_regs_ptr = vgpr_init_regs_arcturus;
4364 } else {
4365 vgpr_init_shader_ptr = vgpr_init_compute_shader;
4366 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader);
4367 vgpr_init_regs_ptr = vgpr_init_regs;
4368 }
4369
4370 total_size =
4371 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* VGPRS */
4372 total_size +=
4373 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS1 */
4374 total_size +=
4375 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS2 */
4376 total_size = ALIGN(total_size, 256);
4377 vgpr_offset = total_size;
4378 total_size += ALIGN(vgpr_init_shader_size, 256);
4379 sgpr_offset = total_size;
4380 total_size += sizeof(sgpr_init_compute_shader);
4381
4382 /* allocate an indirect buffer to put the commands in */
4383 memset(&ib, 0, sizeof(ib));
4384 r = amdgpu_ib_get(adev, NULL, total_size,
4385 AMDGPU_IB_POOL_DIRECT, &ib);
4386 if (r) {
4387 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
4388 return r;
4389 }
4390
4391 /* load the compute shaders */
4392 for (i = 0; i < vgpr_init_shader_size/sizeof(u32); i++)
4393 ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_shader_ptr[i];
4394
4395 for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
4396 ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
4397
4398 /* init the ib length to 0 */
4399 ib.length_dw = 0;
4400
4401 /* VGPR */
4402 /* write the register state for the compute dispatch */
4403 for (i = 0; i < gpr_reg_size; i++) {
4404 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4405 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(vgpr_init_regs_ptr[i])
4406 - PACKET3_SET_SH_REG_START;
4407 ib.ptr[ib.length_dw++] = vgpr_init_regs_ptr[i].reg_value;
4408 }
4409 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4410 gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
4411 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4412 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4413 - PACKET3_SET_SH_REG_START;
4414 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4415 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4416
4417 /* write dispatch packet */
4418 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4419 ib.ptr[ib.length_dw++] = compute_dim_x * 2; /* x */
4420 ib.ptr[ib.length_dw++] = 1; /* y */
4421 ib.ptr[ib.length_dw++] = 1; /* z */
4422 ib.ptr[ib.length_dw++] =
4423 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4424
4425 /* write CS partial flush packet */
4426 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4427 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4428
4429 /* SGPR1 */
4430 /* write the register state for the compute dispatch */
4431 for (i = 0; i < gpr_reg_size; i++) {
4432 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4433 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr1_init_regs[i])
4434 - PACKET3_SET_SH_REG_START;
4435 ib.ptr[ib.length_dw++] = sgpr1_init_regs[i].reg_value;
4436 }
4437 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4438 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4439 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4440 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4441 - PACKET3_SET_SH_REG_START;
4442 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4443 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4444
4445 /* write dispatch packet */
4446 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4447 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */
4448 ib.ptr[ib.length_dw++] = 1; /* y */
4449 ib.ptr[ib.length_dw++] = 1; /* z */
4450 ib.ptr[ib.length_dw++] =
4451 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4452
4453 /* write CS partial flush packet */
4454 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4455 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4456
4457 /* SGPR2 */
4458 /* write the register state for the compute dispatch */
4459 for (i = 0; i < gpr_reg_size; i++) {
4460 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4461 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr2_init_regs[i])
4462 - PACKET3_SET_SH_REG_START;
4463 ib.ptr[ib.length_dw++] = sgpr2_init_regs[i].reg_value;
4464 }
4465 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4466 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4467 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4468 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4469 - PACKET3_SET_SH_REG_START;
4470 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4471 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4472
4473 /* write dispatch packet */
4474 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4475 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */
4476 ib.ptr[ib.length_dw++] = 1; /* y */
4477 ib.ptr[ib.length_dw++] = 1; /* z */
4478 ib.ptr[ib.length_dw++] =
4479 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4480
4481 /* write CS partial flush packet */
4482 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4483 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4484
4485 /* shedule the ib on the ring */
4486 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
4487 if (r) {
4488 DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
4489 goto fail;
4490 }
4491
4492 /* wait for the GPU to finish processing the IB */
4493 r = dma_fence_wait(f, false);
4494 if (r) {
4495 DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
4496 goto fail;
4497 }
4498
4499 fail:
4500 amdgpu_ib_free(adev, &ib, NULL);
4501 dma_fence_put(f);
4502
4503 return r;
4504 }
4505
gfx_v9_0_early_init(void * handle)4506 static int gfx_v9_0_early_init(void *handle)
4507 {
4508 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4509
4510 adev->gfx.funcs = &gfx_v9_0_gfx_funcs;
4511
4512 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1) ||
4513 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
4514 adev->gfx.num_gfx_rings = 0;
4515 else
4516 adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS;
4517 adev->gfx.xcc_mask = 1;
4518 adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
4519 AMDGPU_MAX_COMPUTE_RINGS);
4520 gfx_v9_0_set_kiq_pm4_funcs(adev);
4521 gfx_v9_0_set_ring_funcs(adev);
4522 gfx_v9_0_set_irq_funcs(adev);
4523 gfx_v9_0_set_gds_init(adev);
4524 gfx_v9_0_set_rlc_funcs(adev);
4525
4526 /* init rlcg reg access ctrl */
4527 gfx_v9_0_init_rlcg_reg_access_ctrl(adev);
4528
4529 return gfx_v9_0_init_microcode(adev);
4530 }
4531
gfx_v9_0_ecc_late_init(void * handle)4532 static int gfx_v9_0_ecc_late_init(void *handle)
4533 {
4534 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4535 int r;
4536
4537 /*
4538 * Temp workaround to fix the issue that CP firmware fails to
4539 * update read pointer when CPDMA is writing clearing operation
4540 * to GDS in suspend/resume sequence on several cards. So just
4541 * limit this operation in cold boot sequence.
4542 */
4543 if ((!adev->in_suspend) &&
4544 (adev->gds.gds_size)) {
4545 r = gfx_v9_0_do_edc_gds_workarounds(adev);
4546 if (r)
4547 return r;
4548 }
4549
4550 /* requires IBs so do in late init after IB pool is initialized */
4551 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
4552 r = gfx_v9_4_2_do_edc_gpr_workarounds(adev);
4553 else
4554 r = gfx_v9_0_do_edc_gpr_workarounds(adev);
4555
4556 if (r)
4557 return r;
4558
4559 if (adev->gfx.ras &&
4560 adev->gfx.ras->enable_watchdog_timer)
4561 adev->gfx.ras->enable_watchdog_timer(adev);
4562
4563 return 0;
4564 }
4565
gfx_v9_0_late_init(void * handle)4566 static int gfx_v9_0_late_init(void *handle)
4567 {
4568 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4569 int r;
4570
4571 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4572 if (r)
4573 return r;
4574
4575 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4576 if (r)
4577 return r;
4578
4579 r = gfx_v9_0_ecc_late_init(handle);
4580 if (r)
4581 return r;
4582
4583 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
4584 gfx_v9_4_2_debug_trap_config_init(adev,
4585 adev->vm_manager.first_kfd_vmid, AMDGPU_NUM_VMID);
4586 else
4587 gfx_v9_0_debug_trap_config_init(adev,
4588 adev->vm_manager.first_kfd_vmid, AMDGPU_NUM_VMID);
4589
4590 return 0;
4591 }
4592
gfx_v9_0_is_rlc_enabled(struct amdgpu_device * adev)4593 static bool gfx_v9_0_is_rlc_enabled(struct amdgpu_device *adev)
4594 {
4595 uint32_t rlc_setting;
4596
4597 /* if RLC is not enabled, do nothing */
4598 rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL);
4599 if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
4600 return false;
4601
4602 return true;
4603 }
4604
gfx_v9_0_set_safe_mode(struct amdgpu_device * adev,int xcc_id)4605 static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id)
4606 {
4607 uint32_t data;
4608 unsigned i;
4609
4610 data = RLC_SAFE_MODE__CMD_MASK;
4611 data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
4612 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4613
4614 /* wait for RLC_SAFE_MODE */
4615 for (i = 0; i < adev->usec_timeout; i++) {
4616 if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
4617 break;
4618 udelay(1);
4619 }
4620 }
4621
gfx_v9_0_unset_safe_mode(struct amdgpu_device * adev,int xcc_id)4622 static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id)
4623 {
4624 uint32_t data;
4625
4626 data = RLC_SAFE_MODE__CMD_MASK;
4627 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4628 }
4629
gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device * adev,bool enable)4630 static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev,
4631 bool enable)
4632 {
4633 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4634
4635 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
4636 gfx_v9_0_enable_gfx_cg_power_gating(adev, true);
4637 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4638 gfx_v9_0_enable_gfx_pipeline_powergating(adev, true);
4639 } else {
4640 gfx_v9_0_enable_gfx_cg_power_gating(adev, false);
4641 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4642 gfx_v9_0_enable_gfx_pipeline_powergating(adev, false);
4643 }
4644
4645 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4646 }
4647
gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device * adev,bool enable)4648 static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev,
4649 bool enable)
4650 {
4651 /* TODO: double check if we need to perform under safe mode */
4652 /* gfx_v9_0_enter_rlc_safe_mode(adev); */
4653
4654 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
4655 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, true);
4656 else
4657 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, false);
4658
4659 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
4660 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, true);
4661 else
4662 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, false);
4663
4664 /* gfx_v9_0_exit_rlc_safe_mode(adev); */
4665 }
4666
gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device * adev,bool enable)4667 static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
4668 bool enable)
4669 {
4670 uint32_t data, def;
4671
4672 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4673
4674 /* It is disabled by HW by default */
4675 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
4676 /* 1 - RLC_CGTT_MGCG_OVERRIDE */
4677 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4678
4679 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1))
4680 data &= ~RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
4681
4682 data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4683 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4684 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4685
4686 /* only for Vega10 & Raven1 */
4687 data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK;
4688
4689 if (def != data)
4690 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4691
4692 /* MGLS is a global flag to control all MGLS in GFX */
4693 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
4694 /* 2 - RLC memory Light sleep */
4695 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
4696 def = data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4697 data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4698 if (def != data)
4699 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4700 }
4701 /* 3 - CP memory Light sleep */
4702 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
4703 def = data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4704 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4705 if (def != data)
4706 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4707 }
4708 }
4709 } else {
4710 /* 1 - MGCG_OVERRIDE */
4711 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4712
4713 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1))
4714 data |= RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
4715
4716 data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4717 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4718 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4719 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4720
4721 if (def != data)
4722 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4723
4724 /* 2 - disable MGLS in RLC */
4725 data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4726 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
4727 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4728 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4729 }
4730
4731 /* 3 - disable MGLS in CP */
4732 data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4733 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
4734 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4735 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4736 }
4737 }
4738
4739 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4740 }
4741
gfx_v9_0_update_3d_clock_gating(struct amdgpu_device * adev,bool enable)4742 static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev,
4743 bool enable)
4744 {
4745 uint32_t data, def;
4746
4747 if (!adev->gfx.num_gfx_rings)
4748 return;
4749
4750 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4751
4752 /* Enable 3D CGCG/CGLS */
4753 if (enable) {
4754 /* write cmd to clear cgcg/cgls ov */
4755 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4756 /* unset CGCG override */
4757 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK;
4758 /* update CGCG and CGLS override bits */
4759 if (def != data)
4760 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4761
4762 /* enable 3Dcgcg FSM(0x0000363f) */
4763 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4764
4765 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)
4766 data = (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4767 RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
4768 else
4769 data = 0x0 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT;
4770
4771 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
4772 data |= (0x000F << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4773 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
4774 if (def != data)
4775 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
4776
4777 /* set IDLE_POLL_COUNT(0x00900100) */
4778 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
4779 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4780 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4781 if (def != data)
4782 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
4783 } else {
4784 /* Disable CGCG/CGLS */
4785 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4786 /* disable cgcg, cgls should be disabled */
4787 data &= ~(RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK |
4788 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK);
4789 /* disable cgcg and cgls in FSM */
4790 if (def != data)
4791 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
4792 }
4793
4794 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4795 }
4796
gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device * adev,bool enable)4797 static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
4798 bool enable)
4799 {
4800 uint32_t def, data;
4801
4802 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4803
4804 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
4805 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4806 /* unset CGCG override */
4807 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
4808 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4809 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4810 else
4811 data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4812 /* update CGCG and CGLS override bits */
4813 if (def != data)
4814 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4815
4816 /* enable cgcg FSM(0x0000363F) */
4817 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4818
4819 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1))
4820 data = (0x2000 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4821 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4822 else
4823 data = (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4824 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4825 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4826 data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4827 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
4828 if (def != data)
4829 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
4830
4831 /* set IDLE_POLL_COUNT(0x00900100) */
4832 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
4833 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4834 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4835 if (def != data)
4836 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
4837 } else {
4838 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4839 /* reset CGCG/CGLS bits */
4840 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
4841 /* disable cgcg and cgls in FSM */
4842 if (def != data)
4843 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
4844 }
4845
4846 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4847 }
4848
gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device * adev,bool enable)4849 static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev,
4850 bool enable)
4851 {
4852 if (enable) {
4853 /* CGCG/CGLS should be enabled after MGCG/MGLS
4854 * === MGCG + MGLS ===
4855 */
4856 gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
4857 /* === CGCG /CGLS for GFX 3D Only === */
4858 gfx_v9_0_update_3d_clock_gating(adev, enable);
4859 /* === CGCG + CGLS === */
4860 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
4861 } else {
4862 /* CGCG/CGLS should be disabled before MGCG/MGLS
4863 * === CGCG + CGLS ===
4864 */
4865 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
4866 /* === CGCG /CGLS for GFX 3D Only === */
4867 gfx_v9_0_update_3d_clock_gating(adev, enable);
4868 /* === MGCG + MGLS === */
4869 gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
4870 }
4871 return 0;
4872 }
4873
gfx_v9_0_update_spm_vmid_internal(struct amdgpu_device * adev,unsigned int vmid)4874 static void gfx_v9_0_update_spm_vmid_internal(struct amdgpu_device *adev,
4875 unsigned int vmid)
4876 {
4877 u32 reg, data;
4878
4879 reg = SOC15_REG_OFFSET(GC, 0, mmRLC_SPM_MC_CNTL);
4880 if (amdgpu_sriov_is_pp_one_vf(adev))
4881 data = RREG32_NO_KIQ(reg);
4882 else
4883 data = RREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL);
4884
4885 data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK;
4886 data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT;
4887
4888 if (amdgpu_sriov_is_pp_one_vf(adev))
4889 WREG32_SOC15_NO_KIQ(GC, 0, mmRLC_SPM_MC_CNTL, data);
4890 else
4891 WREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL, data);
4892 }
4893
gfx_v9_0_update_spm_vmid(struct amdgpu_device * adev,unsigned int vmid)4894 static void gfx_v9_0_update_spm_vmid(struct amdgpu_device *adev, unsigned int vmid)
4895 {
4896 amdgpu_gfx_off_ctrl(adev, false);
4897
4898 gfx_v9_0_update_spm_vmid_internal(adev, vmid);
4899
4900 amdgpu_gfx_off_ctrl(adev, true);
4901 }
4902
gfx_v9_0_check_rlcg_range(struct amdgpu_device * adev,uint32_t offset,struct soc15_reg_rlcg * entries,int arr_size)4903 static bool gfx_v9_0_check_rlcg_range(struct amdgpu_device *adev,
4904 uint32_t offset,
4905 struct soc15_reg_rlcg *entries, int arr_size)
4906 {
4907 int i;
4908 uint32_t reg;
4909
4910 if (!entries)
4911 return false;
4912
4913 for (i = 0; i < arr_size; i++) {
4914 const struct soc15_reg_rlcg *entry;
4915
4916 entry = &entries[i];
4917 reg = adev->reg_offset[entry->hwip][entry->instance][entry->segment] + entry->reg;
4918 if (offset == reg)
4919 return true;
4920 }
4921
4922 return false;
4923 }
4924
gfx_v9_0_is_rlcg_access_range(struct amdgpu_device * adev,u32 offset)4925 static bool gfx_v9_0_is_rlcg_access_range(struct amdgpu_device *adev, u32 offset)
4926 {
4927 return gfx_v9_0_check_rlcg_range(adev, offset,
4928 (void *)rlcg_access_gc_9_0,
4929 ARRAY_SIZE(rlcg_access_gc_9_0));
4930 }
4931
4932 static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = {
4933 .is_rlc_enabled = gfx_v9_0_is_rlc_enabled,
4934 .set_safe_mode = gfx_v9_0_set_safe_mode,
4935 .unset_safe_mode = gfx_v9_0_unset_safe_mode,
4936 .init = gfx_v9_0_rlc_init,
4937 .get_csb_size = gfx_v9_0_get_csb_size,
4938 .get_csb_buffer = gfx_v9_0_get_csb_buffer,
4939 .get_cp_table_num = gfx_v9_0_cp_jump_table_num,
4940 .resume = gfx_v9_0_rlc_resume,
4941 .stop = gfx_v9_0_rlc_stop,
4942 .reset = gfx_v9_0_rlc_reset,
4943 .start = gfx_v9_0_rlc_start,
4944 .update_spm_vmid = gfx_v9_0_update_spm_vmid,
4945 .is_rlcg_access_range = gfx_v9_0_is_rlcg_access_range,
4946 };
4947
gfx_v9_0_set_powergating_state(void * handle,enum amd_powergating_state state)4948 static int gfx_v9_0_set_powergating_state(void *handle,
4949 enum amd_powergating_state state)
4950 {
4951 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4952 bool enable = (state == AMD_PG_STATE_GATE);
4953
4954 switch (adev->ip_versions[GC_HWIP][0]) {
4955 case IP_VERSION(9, 2, 2):
4956 case IP_VERSION(9, 1, 0):
4957 case IP_VERSION(9, 3, 0):
4958 if (!enable)
4959 amdgpu_gfx_off_ctrl(adev, false);
4960
4961 if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
4962 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, true);
4963 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, true);
4964 } else {
4965 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, false);
4966 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, false);
4967 }
4968
4969 if (adev->pg_flags & AMD_PG_SUPPORT_CP)
4970 gfx_v9_0_enable_cp_power_gating(adev, true);
4971 else
4972 gfx_v9_0_enable_cp_power_gating(adev, false);
4973
4974 /* update gfx cgpg state */
4975 gfx_v9_0_update_gfx_cg_power_gating(adev, enable);
4976
4977 /* update mgcg state */
4978 gfx_v9_0_update_gfx_mg_power_gating(adev, enable);
4979
4980 if (enable)
4981 amdgpu_gfx_off_ctrl(adev, true);
4982 break;
4983 case IP_VERSION(9, 2, 1):
4984 amdgpu_gfx_off_ctrl(adev, enable);
4985 break;
4986 default:
4987 break;
4988 }
4989
4990 return 0;
4991 }
4992
gfx_v9_0_set_clockgating_state(void * handle,enum amd_clockgating_state state)4993 static int gfx_v9_0_set_clockgating_state(void *handle,
4994 enum amd_clockgating_state state)
4995 {
4996 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4997
4998 if (amdgpu_sriov_vf(adev))
4999 return 0;
5000
5001 switch (adev->ip_versions[GC_HWIP][0]) {
5002 case IP_VERSION(9, 0, 1):
5003 case IP_VERSION(9, 2, 1):
5004 case IP_VERSION(9, 4, 0):
5005 case IP_VERSION(9, 2, 2):
5006 case IP_VERSION(9, 1, 0):
5007 case IP_VERSION(9, 4, 1):
5008 case IP_VERSION(9, 3, 0):
5009 case IP_VERSION(9, 4, 2):
5010 gfx_v9_0_update_gfx_clock_gating(adev,
5011 state == AMD_CG_STATE_GATE);
5012 break;
5013 default:
5014 break;
5015 }
5016 return 0;
5017 }
5018
gfx_v9_0_get_clockgating_state(void * handle,u64 * flags)5019 static void gfx_v9_0_get_clockgating_state(void *handle, u64 *flags)
5020 {
5021 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5022 int data;
5023
5024 if (amdgpu_sriov_vf(adev))
5025 *flags = 0;
5026
5027 /* AMD_CG_SUPPORT_GFX_MGCG */
5028 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE));
5029 if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
5030 *flags |= AMD_CG_SUPPORT_GFX_MGCG;
5031
5032 /* AMD_CG_SUPPORT_GFX_CGCG */
5033 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL));
5034 if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5035 *flags |= AMD_CG_SUPPORT_GFX_CGCG;
5036
5037 /* AMD_CG_SUPPORT_GFX_CGLS */
5038 if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5039 *flags |= AMD_CG_SUPPORT_GFX_CGLS;
5040
5041 /* AMD_CG_SUPPORT_GFX_RLC_LS */
5042 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_MEM_SLP_CNTL));
5043 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5044 *flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5045
5046 /* AMD_CG_SUPPORT_GFX_CP_LS */
5047 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmCP_MEM_SLP_CNTL));
5048 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5049 *flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5050
5051 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)) {
5052 /* AMD_CG_SUPPORT_GFX_3D_CGCG */
5053 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D));
5054 if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK)
5055 *flags |= AMD_CG_SUPPORT_GFX_3D_CGCG;
5056
5057 /* AMD_CG_SUPPORT_GFX_3D_CGLS */
5058 if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK)
5059 *flags |= AMD_CG_SUPPORT_GFX_3D_CGLS;
5060 }
5061 }
5062
gfx_v9_0_ring_get_rptr_gfx(struct amdgpu_ring * ring)5063 static u64 gfx_v9_0_ring_get_rptr_gfx(struct amdgpu_ring *ring)
5064 {
5065 return *ring->rptr_cpu_addr; /* gfx9 is 32bit rptr*/
5066 }
5067
gfx_v9_0_ring_get_wptr_gfx(struct amdgpu_ring * ring)5068 static u64 gfx_v9_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
5069 {
5070 struct amdgpu_device *adev = ring->adev;
5071 u64 wptr;
5072
5073 /* XXX check if swapping is necessary on BE */
5074 if (ring->use_doorbell) {
5075 wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5076 } else {
5077 wptr = RREG32_SOC15(GC, 0, mmCP_RB0_WPTR);
5078 wptr += (u64)RREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI) << 32;
5079 }
5080
5081 return wptr;
5082 }
5083
gfx_v9_0_ring_set_wptr_gfx(struct amdgpu_ring * ring)5084 static void gfx_v9_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
5085 {
5086 struct amdgpu_device *adev = ring->adev;
5087
5088 if (ring->use_doorbell) {
5089 /* XXX check if swapping is necessary on BE */
5090 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr);
5091 WDOORBELL64(ring->doorbell_index, ring->wptr);
5092 } else {
5093 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
5094 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
5095 }
5096 }
5097
gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring * ring)5098 static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
5099 {
5100 struct amdgpu_device *adev = ring->adev;
5101 u32 ref_and_mask, reg_mem_engine;
5102 const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
5103
5104 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
5105 switch (ring->me) {
5106 case 1:
5107 ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
5108 break;
5109 case 2:
5110 ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
5111 break;
5112 default:
5113 return;
5114 }
5115 reg_mem_engine = 0;
5116 } else {
5117 ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
5118 reg_mem_engine = 1; /* pfp */
5119 }
5120
5121 gfx_v9_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
5122 adev->nbio.funcs->get_hdp_flush_req_offset(adev),
5123 adev->nbio.funcs->get_hdp_flush_done_offset(adev),
5124 ref_and_mask, ref_and_mask, 0x20);
5125 }
5126
gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)5127 static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
5128 struct amdgpu_job *job,
5129 struct amdgpu_ib *ib,
5130 uint32_t flags)
5131 {
5132 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5133 u32 header, control = 0;
5134
5135 if (ib->flags & AMDGPU_IB_FLAG_CE)
5136 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
5137 else
5138 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
5139
5140 control |= ib->length_dw | (vmid << 24);
5141
5142 if (ib->flags & AMDGPU_IB_FLAG_PREEMPT) {
5143 control |= INDIRECT_BUFFER_PRE_ENB(1);
5144
5145 if (flags & AMDGPU_IB_PREEMPTED)
5146 control |= INDIRECT_BUFFER_PRE_RESUME(1);
5147
5148 if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid)
5149 gfx_v9_0_ring_emit_de_meta(ring,
5150 (!amdgpu_sriov_vf(ring->adev) &&
5151 flags & AMDGPU_IB_PREEMPTED) ?
5152 true : false,
5153 job->gds_size > 0 && job->gds_base != 0);
5154 }
5155
5156 amdgpu_ring_write(ring, header);
5157 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5158 amdgpu_ring_write(ring,
5159 #ifdef __BIG_ENDIAN
5160 (2 << 0) |
5161 #endif
5162 lower_32_bits(ib->gpu_addr));
5163 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5164 amdgpu_ring_ib_on_emit_cntl(ring);
5165 amdgpu_ring_write(ring, control);
5166 }
5167
gfx_v9_0_ring_patch_cntl(struct amdgpu_ring * ring,unsigned offset)5168 static void gfx_v9_0_ring_patch_cntl(struct amdgpu_ring *ring,
5169 unsigned offset)
5170 {
5171 u32 control = ring->ring[offset];
5172
5173 control |= INDIRECT_BUFFER_PRE_RESUME(1);
5174 ring->ring[offset] = control;
5175 }
5176
gfx_v9_0_ring_patch_ce_meta(struct amdgpu_ring * ring,unsigned offset)5177 static void gfx_v9_0_ring_patch_ce_meta(struct amdgpu_ring *ring,
5178 unsigned offset)
5179 {
5180 struct amdgpu_device *adev = ring->adev;
5181 void *ce_payload_cpu_addr;
5182 uint64_t payload_offset, payload_size;
5183
5184 payload_size = sizeof(struct v9_ce_ib_state);
5185
5186 if (ring->is_mes_queue) {
5187 payload_offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5188 gfx[0].gfx_meta_data) +
5189 offsetof(struct v9_gfx_meta_data, ce_payload);
5190 ce_payload_cpu_addr =
5191 amdgpu_mes_ctx_get_offs_cpu_addr(ring, payload_offset);
5192 } else {
5193 payload_offset = offsetof(struct v9_gfx_meta_data, ce_payload);
5194 ce_payload_cpu_addr = adev->virt.csa_cpu_addr + payload_offset;
5195 }
5196
5197 if (offset + (payload_size >> 2) <= ring->buf_mask + 1) {
5198 memcpy((void *)&ring->ring[offset], ce_payload_cpu_addr, payload_size);
5199 } else {
5200 memcpy((void *)&ring->ring[offset], ce_payload_cpu_addr,
5201 (ring->buf_mask + 1 - offset) << 2);
5202 payload_size -= (ring->buf_mask + 1 - offset) << 2;
5203 memcpy((void *)&ring->ring[0],
5204 ce_payload_cpu_addr + ((ring->buf_mask + 1 - offset) << 2),
5205 payload_size);
5206 }
5207 }
5208
gfx_v9_0_ring_patch_de_meta(struct amdgpu_ring * ring,unsigned offset)5209 static void gfx_v9_0_ring_patch_de_meta(struct amdgpu_ring *ring,
5210 unsigned offset)
5211 {
5212 struct amdgpu_device *adev = ring->adev;
5213 void *de_payload_cpu_addr;
5214 uint64_t payload_offset, payload_size;
5215
5216 payload_size = sizeof(struct v9_de_ib_state);
5217
5218 if (ring->is_mes_queue) {
5219 payload_offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5220 gfx[0].gfx_meta_data) +
5221 offsetof(struct v9_gfx_meta_data, de_payload);
5222 de_payload_cpu_addr =
5223 amdgpu_mes_ctx_get_offs_cpu_addr(ring, payload_offset);
5224 } else {
5225 payload_offset = offsetof(struct v9_gfx_meta_data, de_payload);
5226 de_payload_cpu_addr = adev->virt.csa_cpu_addr + payload_offset;
5227 }
5228
5229 ((struct v9_de_ib_state *)de_payload_cpu_addr)->ib_completion_status =
5230 IB_COMPLETION_STATUS_PREEMPTED;
5231
5232 if (offset + (payload_size >> 2) <= ring->buf_mask + 1) {
5233 memcpy((void *)&ring->ring[offset], de_payload_cpu_addr, payload_size);
5234 } else {
5235 memcpy((void *)&ring->ring[offset], de_payload_cpu_addr,
5236 (ring->buf_mask + 1 - offset) << 2);
5237 payload_size -= (ring->buf_mask + 1 - offset) << 2;
5238 memcpy((void *)&ring->ring[0],
5239 de_payload_cpu_addr + ((ring->buf_mask + 1 - offset) << 2),
5240 payload_size);
5241 }
5242 }
5243
gfx_v9_0_ring_emit_ib_compute(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)5244 static void gfx_v9_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
5245 struct amdgpu_job *job,
5246 struct amdgpu_ib *ib,
5247 uint32_t flags)
5248 {
5249 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5250 u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
5251
5252 /* Currently, there is a high possibility to get wave ID mismatch
5253 * between ME and GDS, leading to a hw deadlock, because ME generates
5254 * different wave IDs than the GDS expects. This situation happens
5255 * randomly when at least 5 compute pipes use GDS ordered append.
5256 * The wave IDs generated by ME are also wrong after suspend/resume.
5257 * Those are probably bugs somewhere else in the kernel driver.
5258 *
5259 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
5260 * GDS to 0 for this ring (me/pipe).
5261 */
5262 if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
5263 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
5264 amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID);
5265 amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
5266 }
5267
5268 amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
5269 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5270 amdgpu_ring_write(ring,
5271 #ifdef __BIG_ENDIAN
5272 (2 << 0) |
5273 #endif
5274 lower_32_bits(ib->gpu_addr));
5275 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5276 amdgpu_ring_write(ring, control);
5277 }
5278
gfx_v9_0_ring_emit_fence(struct amdgpu_ring * ring,u64 addr,u64 seq,unsigned flags)5279 static void gfx_v9_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
5280 u64 seq, unsigned flags)
5281 {
5282 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
5283 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
5284 bool writeback = flags & AMDGPU_FENCE_FLAG_TC_WB_ONLY;
5285 bool exec = flags & AMDGPU_FENCE_FLAG_EXEC;
5286 uint32_t dw2 = 0;
5287
5288 /* RELEASE_MEM - flush caches, send int */
5289 amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
5290
5291 if (writeback) {
5292 dw2 = EOP_TC_NC_ACTION_EN;
5293 } else {
5294 dw2 = EOP_TCL1_ACTION_EN | EOP_TC_ACTION_EN |
5295 EOP_TC_MD_ACTION_EN;
5296 }
5297 dw2 |= EOP_TC_WB_ACTION_EN | EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
5298 EVENT_INDEX(5);
5299 if (exec)
5300 dw2 |= EOP_EXEC;
5301
5302 amdgpu_ring_write(ring, dw2);
5303 amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
5304
5305 /*
5306 * the address should be Qword aligned if 64bit write, Dword
5307 * aligned if only send 32bit data low (discard data high)
5308 */
5309 if (write64bit)
5310 BUG_ON(addr & 0x7);
5311 else
5312 BUG_ON(addr & 0x3);
5313 amdgpu_ring_write(ring, lower_32_bits(addr));
5314 amdgpu_ring_write(ring, upper_32_bits(addr));
5315 amdgpu_ring_write(ring, lower_32_bits(seq));
5316 amdgpu_ring_write(ring, upper_32_bits(seq));
5317 amdgpu_ring_write(ring, 0);
5318 }
5319
gfx_v9_0_ring_emit_pipeline_sync(struct amdgpu_ring * ring)5320 static void gfx_v9_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
5321 {
5322 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5323 uint32_t seq = ring->fence_drv.sync_seq;
5324 uint64_t addr = ring->fence_drv.gpu_addr;
5325
5326 gfx_v9_0_wait_reg_mem(ring, usepfp, 1, 0,
5327 lower_32_bits(addr), upper_32_bits(addr),
5328 seq, 0xffffffff, 4);
5329 }
5330
gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring * ring,unsigned vmid,uint64_t pd_addr)5331 static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
5332 unsigned vmid, uint64_t pd_addr)
5333 {
5334 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
5335
5336 /* compute doesn't have PFP */
5337 if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
5338 /* sync PFP to ME, otherwise we might get invalid PFP reads */
5339 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
5340 amdgpu_ring_write(ring, 0x0);
5341 }
5342 }
5343
gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring * ring)5344 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring)
5345 {
5346 return *ring->rptr_cpu_addr; /* gfx9 hardware is 32bit rptr */
5347 }
5348
gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring * ring)5349 static u64 gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
5350 {
5351 u64 wptr;
5352
5353 /* XXX check if swapping is necessary on BE */
5354 if (ring->use_doorbell)
5355 wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5356 else
5357 BUG();
5358 return wptr;
5359 }
5360
gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring * ring)5361 static void gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
5362 {
5363 struct amdgpu_device *adev = ring->adev;
5364
5365 /* XXX check if swapping is necessary on BE */
5366 if (ring->use_doorbell) {
5367 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr);
5368 WDOORBELL64(ring->doorbell_index, ring->wptr);
5369 } else{
5370 BUG(); /* only DOORBELL method supported on gfx9 now */
5371 }
5372 }
5373
gfx_v9_0_ring_emit_fence_kiq(struct amdgpu_ring * ring,u64 addr,u64 seq,unsigned int flags)5374 static void gfx_v9_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
5375 u64 seq, unsigned int flags)
5376 {
5377 struct amdgpu_device *adev = ring->adev;
5378
5379 /* we only allocate 32bit for each seq wb address */
5380 BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
5381
5382 /* write fence seq to the "addr" */
5383 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5384 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5385 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
5386 amdgpu_ring_write(ring, lower_32_bits(addr));
5387 amdgpu_ring_write(ring, upper_32_bits(addr));
5388 amdgpu_ring_write(ring, lower_32_bits(seq));
5389
5390 if (flags & AMDGPU_FENCE_FLAG_INT) {
5391 /* set register to trigger INT */
5392 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5393 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5394 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
5395 amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, mmCPC_INT_STATUS));
5396 amdgpu_ring_write(ring, 0);
5397 amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
5398 }
5399 }
5400
gfx_v9_ring_emit_sb(struct amdgpu_ring * ring)5401 static void gfx_v9_ring_emit_sb(struct amdgpu_ring *ring)
5402 {
5403 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
5404 amdgpu_ring_write(ring, 0);
5405 }
5406
gfx_v9_0_ring_emit_ce_meta(struct amdgpu_ring * ring,bool resume)5407 static void gfx_v9_0_ring_emit_ce_meta(struct amdgpu_ring *ring, bool resume)
5408 {
5409 struct amdgpu_device *adev = ring->adev;
5410 struct v9_ce_ib_state ce_payload = {0};
5411 uint64_t offset, ce_payload_gpu_addr;
5412 void *ce_payload_cpu_addr;
5413 int cnt;
5414
5415 cnt = (sizeof(ce_payload) >> 2) + 4 - 2;
5416
5417 if (ring->is_mes_queue) {
5418 offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5419 gfx[0].gfx_meta_data) +
5420 offsetof(struct v9_gfx_meta_data, ce_payload);
5421 ce_payload_gpu_addr =
5422 amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5423 ce_payload_cpu_addr =
5424 amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
5425 } else {
5426 offset = offsetof(struct v9_gfx_meta_data, ce_payload);
5427 ce_payload_gpu_addr = amdgpu_csa_vaddr(ring->adev) + offset;
5428 ce_payload_cpu_addr = adev->virt.csa_cpu_addr + offset;
5429 }
5430
5431 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5432 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
5433 WRITE_DATA_DST_SEL(8) |
5434 WR_CONFIRM) |
5435 WRITE_DATA_CACHE_POLICY(0));
5436 amdgpu_ring_write(ring, lower_32_bits(ce_payload_gpu_addr));
5437 amdgpu_ring_write(ring, upper_32_bits(ce_payload_gpu_addr));
5438
5439 amdgpu_ring_ib_on_emit_ce(ring);
5440
5441 if (resume)
5442 amdgpu_ring_write_multiple(ring, ce_payload_cpu_addr,
5443 sizeof(ce_payload) >> 2);
5444 else
5445 amdgpu_ring_write_multiple(ring, (void *)&ce_payload,
5446 sizeof(ce_payload) >> 2);
5447 }
5448
gfx_v9_0_ring_preempt_ib(struct amdgpu_ring * ring)5449 static int gfx_v9_0_ring_preempt_ib(struct amdgpu_ring *ring)
5450 {
5451 int i, r = 0;
5452 struct amdgpu_device *adev = ring->adev;
5453 struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
5454 struct amdgpu_ring *kiq_ring = &kiq->ring;
5455 unsigned long flags;
5456
5457 if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
5458 return -EINVAL;
5459
5460 spin_lock_irqsave(&kiq->ring_lock, flags);
5461
5462 if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) {
5463 spin_unlock_irqrestore(&kiq->ring_lock, flags);
5464 return -ENOMEM;
5465 }
5466
5467 /* assert preemption condition */
5468 amdgpu_ring_set_preempt_cond_exec(ring, false);
5469
5470 ring->trail_seq += 1;
5471 amdgpu_ring_alloc(ring, 13);
5472 gfx_v9_0_ring_emit_fence(ring, ring->trail_fence_gpu_addr,
5473 ring->trail_seq, AMDGPU_FENCE_FLAG_EXEC | AMDGPU_FENCE_FLAG_INT);
5474
5475 /* assert IB preemption, emit the trailing fence */
5476 kiq->pmf->kiq_unmap_queues(kiq_ring, ring, PREEMPT_QUEUES_NO_UNMAP,
5477 ring->trail_fence_gpu_addr,
5478 ring->trail_seq);
5479
5480 amdgpu_ring_commit(kiq_ring);
5481 spin_unlock_irqrestore(&kiq->ring_lock, flags);
5482
5483 /* poll the trailing fence */
5484 for (i = 0; i < adev->usec_timeout; i++) {
5485 if (ring->trail_seq ==
5486 le32_to_cpu(*ring->trail_fence_cpu_addr))
5487 break;
5488 udelay(1);
5489 }
5490
5491 if (i >= adev->usec_timeout) {
5492 r = -EINVAL;
5493 DRM_WARN("ring %d timeout to preempt ib\n", ring->idx);
5494 }
5495
5496 /*reset the CP_VMID_PREEMPT after trailing fence*/
5497 amdgpu_ring_emit_wreg(ring,
5498 SOC15_REG_OFFSET(GC, 0, mmCP_VMID_PREEMPT),
5499 0x0);
5500 amdgpu_ring_commit(ring);
5501
5502 /* deassert preemption condition */
5503 amdgpu_ring_set_preempt_cond_exec(ring, true);
5504 return r;
5505 }
5506
gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring * ring,bool resume,bool usegds)5507 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume, bool usegds)
5508 {
5509 struct amdgpu_device *adev = ring->adev;
5510 struct v9_de_ib_state de_payload = {0};
5511 uint64_t offset, gds_addr, de_payload_gpu_addr;
5512 void *de_payload_cpu_addr;
5513 int cnt;
5514
5515 if (ring->is_mes_queue) {
5516 offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5517 gfx[0].gfx_meta_data) +
5518 offsetof(struct v9_gfx_meta_data, de_payload);
5519 de_payload_gpu_addr =
5520 amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5521 de_payload_cpu_addr =
5522 amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
5523
5524 offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5525 gfx[0].gds_backup) +
5526 offsetof(struct v9_gfx_meta_data, de_payload);
5527 gds_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5528 } else {
5529 offset = offsetof(struct v9_gfx_meta_data, de_payload);
5530 de_payload_gpu_addr = amdgpu_csa_vaddr(ring->adev) + offset;
5531 de_payload_cpu_addr = adev->virt.csa_cpu_addr + offset;
5532
5533 gds_addr = ALIGN(amdgpu_csa_vaddr(ring->adev) +
5534 AMDGPU_CSA_SIZE - adev->gds.gds_size,
5535 PAGE_SIZE);
5536 }
5537
5538 if (usegds) {
5539 de_payload.gds_backup_addrlo = lower_32_bits(gds_addr);
5540 de_payload.gds_backup_addrhi = upper_32_bits(gds_addr);
5541 }
5542
5543 cnt = (sizeof(de_payload) >> 2) + 4 - 2;
5544 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5545 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
5546 WRITE_DATA_DST_SEL(8) |
5547 WR_CONFIRM) |
5548 WRITE_DATA_CACHE_POLICY(0));
5549 amdgpu_ring_write(ring, lower_32_bits(de_payload_gpu_addr));
5550 amdgpu_ring_write(ring, upper_32_bits(de_payload_gpu_addr));
5551
5552 amdgpu_ring_ib_on_emit_de(ring);
5553 if (resume)
5554 amdgpu_ring_write_multiple(ring, de_payload_cpu_addr,
5555 sizeof(de_payload) >> 2);
5556 else
5557 amdgpu_ring_write_multiple(ring, (void *)&de_payload,
5558 sizeof(de_payload) >> 2);
5559 }
5560
gfx_v9_0_ring_emit_frame_cntl(struct amdgpu_ring * ring,bool start,bool secure)5561 static void gfx_v9_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start,
5562 bool secure)
5563 {
5564 uint32_t v = secure ? FRAME_TMZ : 0;
5565
5566 amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0));
5567 amdgpu_ring_write(ring, v | FRAME_CMD(start ? 0 : 1));
5568 }
5569
gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring * ring,uint32_t flags)5570 static void gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
5571 {
5572 uint32_t dw2 = 0;
5573
5574 gfx_v9_0_ring_emit_ce_meta(ring,
5575 (!amdgpu_sriov_vf(ring->adev) &&
5576 flags & AMDGPU_IB_PREEMPTED) ? true : false);
5577
5578 dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
5579 if (flags & AMDGPU_HAVE_CTX_SWITCH) {
5580 /* set load_global_config & load_global_uconfig */
5581 dw2 |= 0x8001;
5582 /* set load_cs_sh_regs */
5583 dw2 |= 0x01000000;
5584 /* set load_per_context_state & load_gfx_sh_regs for GFX */
5585 dw2 |= 0x10002;
5586
5587 /* set load_ce_ram if preamble presented */
5588 if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
5589 dw2 |= 0x10000000;
5590 } else {
5591 /* still load_ce_ram if this is the first time preamble presented
5592 * although there is no context switch happens.
5593 */
5594 if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
5595 dw2 |= 0x10000000;
5596 }
5597
5598 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
5599 amdgpu_ring_write(ring, dw2);
5600 amdgpu_ring_write(ring, 0);
5601 }
5602
gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring * ring)5603 static unsigned gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
5604 {
5605 unsigned ret;
5606 amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
5607 amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
5608 amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
5609 amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
5610 ret = ring->wptr & ring->buf_mask;
5611 amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
5612 return ret;
5613 }
5614
gfx_v9_0_ring_emit_patch_cond_exec(struct amdgpu_ring * ring,unsigned offset)5615 static void gfx_v9_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
5616 {
5617 unsigned cur;
5618 BUG_ON(offset > ring->buf_mask);
5619 BUG_ON(ring->ring[offset] != 0x55aa55aa);
5620
5621 cur = (ring->wptr - 1) & ring->buf_mask;
5622 if (likely(cur > offset))
5623 ring->ring[offset] = cur - offset;
5624 else
5625 ring->ring[offset] = (ring->ring_size>>2) - offset + cur;
5626 }
5627
gfx_v9_0_ring_emit_rreg(struct amdgpu_ring * ring,uint32_t reg,uint32_t reg_val_offs)5628 static void gfx_v9_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
5629 uint32_t reg_val_offs)
5630 {
5631 struct amdgpu_device *adev = ring->adev;
5632
5633 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
5634 amdgpu_ring_write(ring, 0 | /* src: register*/
5635 (5 << 8) | /* dst: memory */
5636 (1 << 20)); /* write confirm */
5637 amdgpu_ring_write(ring, reg);
5638 amdgpu_ring_write(ring, 0);
5639 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
5640 reg_val_offs * 4));
5641 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
5642 reg_val_offs * 4));
5643 }
5644
gfx_v9_0_ring_emit_wreg(struct amdgpu_ring * ring,uint32_t reg,uint32_t val)5645 static void gfx_v9_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
5646 uint32_t val)
5647 {
5648 uint32_t cmd = 0;
5649
5650 switch (ring->funcs->type) {
5651 case AMDGPU_RING_TYPE_GFX:
5652 cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
5653 break;
5654 case AMDGPU_RING_TYPE_KIQ:
5655 cmd = (1 << 16); /* no inc addr */
5656 break;
5657 default:
5658 cmd = WR_CONFIRM;
5659 break;
5660 }
5661 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5662 amdgpu_ring_write(ring, cmd);
5663 amdgpu_ring_write(ring, reg);
5664 amdgpu_ring_write(ring, 0);
5665 amdgpu_ring_write(ring, val);
5666 }
5667
gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring * ring,uint32_t reg,uint32_t val,uint32_t mask)5668 static void gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
5669 uint32_t val, uint32_t mask)
5670 {
5671 gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
5672 }
5673
gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring * ring,uint32_t reg0,uint32_t reg1,uint32_t ref,uint32_t mask)5674 static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
5675 uint32_t reg0, uint32_t reg1,
5676 uint32_t ref, uint32_t mask)
5677 {
5678 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5679 struct amdgpu_device *adev = ring->adev;
5680 bool fw_version_ok = (ring->funcs->type == AMDGPU_RING_TYPE_GFX) ?
5681 adev->gfx.me_fw_write_wait : adev->gfx.mec_fw_write_wait;
5682
5683 if (fw_version_ok)
5684 gfx_v9_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1,
5685 ref, mask, 0x20);
5686 else
5687 amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1,
5688 ref, mask);
5689 }
5690
gfx_v9_0_ring_soft_recovery(struct amdgpu_ring * ring,unsigned vmid)5691 static void gfx_v9_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
5692 {
5693 struct amdgpu_device *adev = ring->adev;
5694 uint32_t value = 0;
5695
5696 value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
5697 value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
5698 value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
5699 value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
5700 WREG32_SOC15(GC, 0, mmSQ_CMD, value);
5701 }
5702
gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device * adev,enum amdgpu_interrupt_state state)5703 static void gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
5704 enum amdgpu_interrupt_state state)
5705 {
5706 switch (state) {
5707 case AMDGPU_IRQ_STATE_DISABLE:
5708 case AMDGPU_IRQ_STATE_ENABLE:
5709 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5710 TIME_STAMP_INT_ENABLE,
5711 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5712 break;
5713 default:
5714 break;
5715 }
5716 }
5717
gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device * adev,int me,int pipe,enum amdgpu_interrupt_state state)5718 static void gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
5719 int me, int pipe,
5720 enum amdgpu_interrupt_state state)
5721 {
5722 u32 mec_int_cntl, mec_int_cntl_reg;
5723
5724 /*
5725 * amdgpu controls only the first MEC. That's why this function only
5726 * handles the setting of interrupts for this specific MEC. All other
5727 * pipes' interrupts are set by amdkfd.
5728 */
5729
5730 if (me == 1) {
5731 switch (pipe) {
5732 case 0:
5733 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
5734 break;
5735 case 1:
5736 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL);
5737 break;
5738 case 2:
5739 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL);
5740 break;
5741 case 3:
5742 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL);
5743 break;
5744 default:
5745 DRM_DEBUG("invalid pipe %d\n", pipe);
5746 return;
5747 }
5748 } else {
5749 DRM_DEBUG("invalid me %d\n", me);
5750 return;
5751 }
5752
5753 switch (state) {
5754 case AMDGPU_IRQ_STATE_DISABLE:
5755 mec_int_cntl = RREG32_SOC15_IP(GC,mec_int_cntl_reg);
5756 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5757 TIME_STAMP_INT_ENABLE, 0);
5758 WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5759 break;
5760 case AMDGPU_IRQ_STATE_ENABLE:
5761 mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg);
5762 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5763 TIME_STAMP_INT_ENABLE, 1);
5764 WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5765 break;
5766 default:
5767 break;
5768 }
5769 }
5770
gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)5771 static int gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
5772 struct amdgpu_irq_src *source,
5773 unsigned type,
5774 enum amdgpu_interrupt_state state)
5775 {
5776 switch (state) {
5777 case AMDGPU_IRQ_STATE_DISABLE:
5778 case AMDGPU_IRQ_STATE_ENABLE:
5779 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5780 PRIV_REG_INT_ENABLE,
5781 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5782 break;
5783 default:
5784 break;
5785 }
5786
5787 return 0;
5788 }
5789
gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)5790 static int gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
5791 struct amdgpu_irq_src *source,
5792 unsigned type,
5793 enum amdgpu_interrupt_state state)
5794 {
5795 switch (state) {
5796 case AMDGPU_IRQ_STATE_DISABLE:
5797 case AMDGPU_IRQ_STATE_ENABLE:
5798 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5799 PRIV_INSTR_INT_ENABLE,
5800 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5801 break;
5802 default:
5803 break;
5804 }
5805
5806 return 0;
5807 }
5808
5809 #define ENABLE_ECC_ON_ME_PIPE(me, pipe) \
5810 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
5811 CP_ECC_ERROR_INT_ENABLE, 1)
5812
5813 #define DISABLE_ECC_ON_ME_PIPE(me, pipe) \
5814 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
5815 CP_ECC_ERROR_INT_ENABLE, 0)
5816
gfx_v9_0_set_cp_ecc_error_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)5817 static int gfx_v9_0_set_cp_ecc_error_state(struct amdgpu_device *adev,
5818 struct amdgpu_irq_src *source,
5819 unsigned type,
5820 enum amdgpu_interrupt_state state)
5821 {
5822 switch (state) {
5823 case AMDGPU_IRQ_STATE_DISABLE:
5824 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5825 CP_ECC_ERROR_INT_ENABLE, 0);
5826 DISABLE_ECC_ON_ME_PIPE(1, 0);
5827 DISABLE_ECC_ON_ME_PIPE(1, 1);
5828 DISABLE_ECC_ON_ME_PIPE(1, 2);
5829 DISABLE_ECC_ON_ME_PIPE(1, 3);
5830 break;
5831
5832 case AMDGPU_IRQ_STATE_ENABLE:
5833 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5834 CP_ECC_ERROR_INT_ENABLE, 1);
5835 ENABLE_ECC_ON_ME_PIPE(1, 0);
5836 ENABLE_ECC_ON_ME_PIPE(1, 1);
5837 ENABLE_ECC_ON_ME_PIPE(1, 2);
5838 ENABLE_ECC_ON_ME_PIPE(1, 3);
5839 break;
5840 default:
5841 break;
5842 }
5843
5844 return 0;
5845 }
5846
5847
gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device * adev,struct amdgpu_irq_src * src,unsigned type,enum amdgpu_interrupt_state state)5848 static int gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device *adev,
5849 struct amdgpu_irq_src *src,
5850 unsigned type,
5851 enum amdgpu_interrupt_state state)
5852 {
5853 switch (type) {
5854 case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
5855 gfx_v9_0_set_gfx_eop_interrupt_state(adev, state);
5856 break;
5857 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
5858 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
5859 break;
5860 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
5861 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
5862 break;
5863 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
5864 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
5865 break;
5866 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
5867 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
5868 break;
5869 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
5870 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
5871 break;
5872 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
5873 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
5874 break;
5875 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
5876 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
5877 break;
5878 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
5879 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
5880 break;
5881 default:
5882 break;
5883 }
5884 return 0;
5885 }
5886
gfx_v9_0_eop_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)5887 static int gfx_v9_0_eop_irq(struct amdgpu_device *adev,
5888 struct amdgpu_irq_src *source,
5889 struct amdgpu_iv_entry *entry)
5890 {
5891 int i;
5892 u8 me_id, pipe_id, queue_id;
5893 struct amdgpu_ring *ring;
5894
5895 DRM_DEBUG("IH: CP EOP\n");
5896 me_id = (entry->ring_id & 0x0c) >> 2;
5897 pipe_id = (entry->ring_id & 0x03) >> 0;
5898 queue_id = (entry->ring_id & 0x70) >> 4;
5899
5900 switch (me_id) {
5901 case 0:
5902 if (adev->gfx.num_gfx_rings &&
5903 !amdgpu_mcbp_handle_trailing_fence_irq(&adev->gfx.muxer)) {
5904 /* Fence signals are handled on the software rings*/
5905 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++)
5906 amdgpu_fence_process(&adev->gfx.sw_gfx_ring[i]);
5907 }
5908 break;
5909 case 1:
5910 case 2:
5911 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5912 ring = &adev->gfx.compute_ring[i];
5913 /* Per-queue interrupt is supported for MEC starting from VI.
5914 * The interrupt can only be enabled/disabled per pipe instead of per queue.
5915 */
5916 if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
5917 amdgpu_fence_process(ring);
5918 }
5919 break;
5920 }
5921 return 0;
5922 }
5923
gfx_v9_0_fault(struct amdgpu_device * adev,struct amdgpu_iv_entry * entry)5924 static void gfx_v9_0_fault(struct amdgpu_device *adev,
5925 struct amdgpu_iv_entry *entry)
5926 {
5927 u8 me_id, pipe_id, queue_id;
5928 struct amdgpu_ring *ring;
5929 int i;
5930
5931 me_id = (entry->ring_id & 0x0c) >> 2;
5932 pipe_id = (entry->ring_id & 0x03) >> 0;
5933 queue_id = (entry->ring_id & 0x70) >> 4;
5934
5935 switch (me_id) {
5936 case 0:
5937 drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
5938 break;
5939 case 1:
5940 case 2:
5941 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5942 ring = &adev->gfx.compute_ring[i];
5943 if (ring->me == me_id && ring->pipe == pipe_id &&
5944 ring->queue == queue_id)
5945 drm_sched_fault(&ring->sched);
5946 }
5947 break;
5948 }
5949 }
5950
gfx_v9_0_priv_reg_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)5951 static int gfx_v9_0_priv_reg_irq(struct amdgpu_device *adev,
5952 struct amdgpu_irq_src *source,
5953 struct amdgpu_iv_entry *entry)
5954 {
5955 DRM_ERROR("Illegal register access in command stream\n");
5956 gfx_v9_0_fault(adev, entry);
5957 return 0;
5958 }
5959
gfx_v9_0_priv_inst_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)5960 static int gfx_v9_0_priv_inst_irq(struct amdgpu_device *adev,
5961 struct amdgpu_irq_src *source,
5962 struct amdgpu_iv_entry *entry)
5963 {
5964 DRM_ERROR("Illegal instruction in command stream\n");
5965 gfx_v9_0_fault(adev, entry);
5966 return 0;
5967 }
5968
5969
5970 static const struct soc15_ras_field_entry gfx_v9_0_ras_fields[] = {
5971 { "CPC_SCRATCH", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT),
5972 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, SEC_COUNT),
5973 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, DED_COUNT)
5974 },
5975 { "CPC_UCODE", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT),
5976 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, SEC_COUNT),
5977 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, DED_COUNT)
5978 },
5979 { "CPF_ROQ_ME1", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
5980 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME1),
5981 0, 0
5982 },
5983 { "CPF_ROQ_ME2", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
5984 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME2),
5985 0, 0
5986 },
5987 { "CPF_TAG", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT),
5988 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, SEC_COUNT),
5989 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, DED_COUNT)
5990 },
5991 { "CPG_DMA_ROQ", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
5992 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, ROQ_COUNT),
5993 0, 0
5994 },
5995 { "CPG_DMA_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
5996 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_SEC_COUNT),
5997 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_DED_COUNT)
5998 },
5999 { "CPG_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT),
6000 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, SEC_COUNT),
6001 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, DED_COUNT)
6002 },
6003 { "DC_CSINVOC", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT),
6004 SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, COUNT_ME1),
6005 0, 0
6006 },
6007 { "DC_RESTORE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT),
6008 SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, COUNT_ME1),
6009 0, 0
6010 },
6011 { "DC_STATE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT),
6012 SOC15_REG_FIELD(DC_EDC_STATE_CNT, COUNT_ME1),
6013 0, 0
6014 },
6015 { "GDS_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
6016 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_SEC),
6017 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_DED)
6018 },
6019 { "GDS_INPUT_QUEUE", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
6020 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_INPUT_QUEUE_SED),
6021 0, 0
6022 },
6023 { "GDS_ME0_CS_PIPE_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6024 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_SEC),
6025 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_DED)
6026 },
6027 { "GDS_OA_PHY_PHY_CMD_RAM_MEM",
6028 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6029 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_SEC),
6030 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_DED)
6031 },
6032 { "GDS_OA_PHY_PHY_DATA_RAM_MEM",
6033 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6034 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_DATA_RAM_MEM_SED),
6035 0, 0
6036 },
6037 { "GDS_OA_PIPE_ME1_PIPE0_PIPE_MEM",
6038 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6039 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_SEC),
6040 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_DED)
6041 },
6042 { "GDS_OA_PIPE_ME1_PIPE1_PIPE_MEM",
6043 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6044 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_SEC),
6045 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_DED)
6046 },
6047 { "GDS_OA_PIPE_ME1_PIPE2_PIPE_MEM",
6048 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6049 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_SEC),
6050 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_DED)
6051 },
6052 { "GDS_OA_PIPE_ME1_PIPE3_PIPE_MEM",
6053 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6054 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_SEC),
6055 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_DED)
6056 },
6057 { "SPI_SR_MEM", SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT),
6058 SOC15_REG_FIELD(SPI_EDC_CNT, SPI_SR_MEM_SED_COUNT),
6059 0, 0
6060 },
6061 { "TA_FS_DFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6062 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_SEC_COUNT),
6063 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_DED_COUNT)
6064 },
6065 { "TA_FS_AFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6066 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_SED_COUNT),
6067 0, 0
6068 },
6069 { "TA_FL_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6070 SOC15_REG_FIELD(TA_EDC_CNT, TA_FL_LFIFO_SED_COUNT),
6071 0, 0
6072 },
6073 { "TA_FX_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6074 SOC15_REG_FIELD(TA_EDC_CNT, TA_FX_LFIFO_SED_COUNT),
6075 0, 0
6076 },
6077 { "TA_FS_CFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6078 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_CFIFO_SED_COUNT),
6079 0, 0
6080 },
6081 { "TCA_HOLE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
6082 SOC15_REG_FIELD(TCA_EDC_CNT, HOLE_FIFO_SED_COUNT),
6083 0, 0
6084 },
6085 { "TCA_REQ_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
6086 SOC15_REG_FIELD(TCA_EDC_CNT, REQ_FIFO_SED_COUNT),
6087 0, 0
6088 },
6089 { "TCC_CACHE_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6090 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_SEC_COUNT),
6091 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_DED_COUNT)
6092 },
6093 { "TCC_CACHE_DIRTY", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6094 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_SEC_COUNT),
6095 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_DED_COUNT)
6096 },
6097 { "TCC_HIGH_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6098 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_SEC_COUNT),
6099 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_DED_COUNT)
6100 },
6101 { "TCC_LOW_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6102 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_SEC_COUNT),
6103 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_DED_COUNT)
6104 },
6105 { "TCC_SRC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6106 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_SEC_COUNT),
6107 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_DED_COUNT)
6108 },
6109 { "TCC_IN_USE_DEC", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6110 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_DEC_SED_COUNT),
6111 0, 0
6112 },
6113 { "TCC_IN_USE_TRANSFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6114 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_TRANSFER_SED_COUNT),
6115 0, 0
6116 },
6117 { "TCC_LATENCY_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6118 SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_SED_COUNT),
6119 0, 0
6120 },
6121 { "TCC_RETURN_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6122 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_DATA_SED_COUNT),
6123 0, 0
6124 },
6125 { "TCC_RETURN_CONTROL", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6126 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_CONTROL_SED_COUNT),
6127 0, 0
6128 },
6129 { "TCC_UC_ATOMIC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6130 SOC15_REG_FIELD(TCC_EDC_CNT, UC_ATOMIC_FIFO_SED_COUNT),
6131 0, 0
6132 },
6133 { "TCC_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6134 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_RETURN_SED_COUNT),
6135 0, 0
6136 },
6137 { "TCC_WRITE_CACHE_READ", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6138 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_CACHE_READ_SED_COUNT),
6139 0, 0
6140 },
6141 { "TCC_SRC_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6142 SOC15_REG_FIELD(TCC_EDC_CNT2, SRC_FIFO_NEXT_RAM_SED_COUNT),
6143 0, 0
6144 },
6145 { "TCC_LATENCY_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6146 SOC15_REG_FIELD(TCC_EDC_CNT2, LATENCY_FIFO_NEXT_RAM_SED_COUNT),
6147 0, 0
6148 },
6149 { "TCC_CACHE_TAG_PROBE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6150 SOC15_REG_FIELD(TCC_EDC_CNT2, CACHE_TAG_PROBE_FIFO_SED_COUNT),
6151 0, 0
6152 },
6153 { "TCC_WRRET_TAG_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6154 SOC15_REG_FIELD(TCC_EDC_CNT2, WRRET_TAG_WRITE_RETURN_SED_COUNT),
6155 0, 0
6156 },
6157 { "TCC_ATOMIC_RETURN_BUFFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6158 SOC15_REG_FIELD(TCC_EDC_CNT2, ATOMIC_RETURN_BUFFER_SED_COUNT),
6159 0, 0
6160 },
6161 { "TCI_WRITE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT),
6162 SOC15_REG_FIELD(TCI_EDC_CNT, WRITE_RAM_SED_COUNT),
6163 0, 0
6164 },
6165 { "TCP_CACHE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6166 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_SEC_COUNT),
6167 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_DED_COUNT)
6168 },
6169 { "TCP_LFIFO_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6170 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_SEC_COUNT),
6171 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_DED_COUNT)
6172 },
6173 { "TCP_CMD_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6174 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CMD_FIFO_SED_COUNT),
6175 0, 0
6176 },
6177 { "TCP_VM_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6178 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, VM_FIFO_SEC_COUNT),
6179 0, 0
6180 },
6181 { "TCP_DB_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6182 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, DB_RAM_SED_COUNT),
6183 0, 0
6184 },
6185 { "TCP_UTCL1_LFIFO0", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6186 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_SEC_COUNT),
6187 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_DED_COUNT)
6188 },
6189 { "TCP_UTCL1_LFIFO1", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6190 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_SEC_COUNT),
6191 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_DED_COUNT)
6192 },
6193 { "TD_SS_FIFO_LO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6194 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_SEC_COUNT),
6195 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_DED_COUNT)
6196 },
6197 { "TD_SS_FIFO_HI", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6198 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_SEC_COUNT),
6199 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_DED_COUNT)
6200 },
6201 { "TD_CS_FIFO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6202 SOC15_REG_FIELD(TD_EDC_CNT, CS_FIFO_SED_COUNT),
6203 0, 0
6204 },
6205 { "SQ_LDS_D", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6206 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_SEC_COUNT),
6207 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_DED_COUNT)
6208 },
6209 { "SQ_LDS_I", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6210 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_SEC_COUNT),
6211 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_DED_COUNT)
6212 },
6213 { "SQ_SGPR", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6214 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_SEC_COUNT),
6215 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_DED_COUNT)
6216 },
6217 { "SQ_VGPR0", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6218 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_SEC_COUNT),
6219 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_DED_COUNT)
6220 },
6221 { "SQ_VGPR1", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6222 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_SEC_COUNT),
6223 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_DED_COUNT)
6224 },
6225 { "SQ_VGPR2", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6226 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_SEC_COUNT),
6227 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_DED_COUNT)
6228 },
6229 { "SQ_VGPR3", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6230 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_SEC_COUNT),
6231 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_DED_COUNT)
6232 },
6233 { "SQC_DATA_CU0_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6234 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_SEC_COUNT),
6235 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_DED_COUNT)
6236 },
6237 { "SQC_DATA_CU0_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6238 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_SEC_COUNT),
6239 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_DED_COUNT)
6240 },
6241 { "SQC_DATA_CU1_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6242 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_SEC_COUNT),
6243 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_DED_COUNT)
6244 },
6245 { "SQC_DATA_CU1_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6246 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_SEC_COUNT),
6247 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_DED_COUNT)
6248 },
6249 { "SQC_DATA_CU2_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6250 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_SEC_COUNT),
6251 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_DED_COUNT)
6252 },
6253 { "SQC_DATA_CU2_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6254 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_SEC_COUNT),
6255 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_DED_COUNT)
6256 },
6257 { "SQC_INST_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6258 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_SEC_COUNT),
6259 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_DED_COUNT)
6260 },
6261 { "SQC_INST_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6262 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_SEC_COUNT),
6263 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_DED_COUNT)
6264 },
6265 { "SQC_DATA_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6266 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_SEC_COUNT),
6267 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_DED_COUNT)
6268 },
6269 { "SQC_DATA_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6270 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_SEC_COUNT),
6271 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_DED_COUNT)
6272 },
6273 { "SQC_INST_BANKA_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6274 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_UTCL1_MISS_FIFO_SED_COUNT),
6275 0, 0
6276 },
6277 { "SQC_INST_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6278 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_MISS_FIFO_SED_COUNT),
6279 0, 0
6280 },
6281 { "SQC_DATA_BANKA_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6282 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_HIT_FIFO_SED_COUNT),
6283 0, 0
6284 },
6285 { "SQC_DATA_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6286 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_MISS_FIFO_SED_COUNT),
6287 0, 0
6288 },
6289 { "SQC_DATA_BANKA_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6290 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_DIRTY_BIT_RAM_SED_COUNT),
6291 0, 0
6292 },
6293 { "SQC_INST_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6294 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_SEC_COUNT),
6295 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_DED_COUNT)
6296 },
6297 { "SQC_INST_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6298 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_SEC_COUNT),
6299 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_DED_COUNT)
6300 },
6301 { "SQC_INST_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6302 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_SEC_COUNT),
6303 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_DED_COUNT)
6304 },
6305 { "SQC_DATA_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6306 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_SEC_COUNT),
6307 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_DED_COUNT)
6308 },
6309 { "SQC_DATA_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6310 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_SEC_COUNT),
6311 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_DED_COUNT)
6312 },
6313 { "SQC_INST_BANKB_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6314 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_UTCL1_MISS_FIFO_SED_COUNT),
6315 0, 0
6316 },
6317 { "SQC_INST_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6318 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_MISS_FIFO_SED_COUNT),
6319 0, 0
6320 },
6321 { "SQC_DATA_BANKB_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6322 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_HIT_FIFO_SED_COUNT),
6323 0, 0
6324 },
6325 { "SQC_DATA_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6326 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_MISS_FIFO_SED_COUNT),
6327 0, 0
6328 },
6329 { "SQC_DATA_BANKB_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6330 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_DIRTY_BIT_RAM_SED_COUNT),
6331 0, 0
6332 },
6333 { "EA_DRAMRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6334 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
6335 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT)
6336 },
6337 { "EA_DRAMWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6338 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
6339 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT)
6340 },
6341 { "EA_DRAMWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6342 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
6343 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT)
6344 },
6345 { "EA_RRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6346 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
6347 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_DED_COUNT)
6348 },
6349 { "EA_WRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6350 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
6351 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_DED_COUNT)
6352 },
6353 { "EA_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6354 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
6355 0, 0
6356 },
6357 { "EA_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6358 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
6359 0, 0
6360 },
6361 { "EA_IORD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6362 SOC15_REG_FIELD(GCEA_EDC_CNT, IORD_CMDMEM_SED_COUNT),
6363 0, 0
6364 },
6365 { "EA_IOWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6366 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
6367 0, 0
6368 },
6369 { "EA_IOWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6370 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_DATAMEM_SED_COUNT),
6371 0, 0
6372 },
6373 { "GMIRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6374 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
6375 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT)
6376 },
6377 { "GMIWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6378 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
6379 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT)
6380 },
6381 { "GMIWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6382 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
6383 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT)
6384 },
6385 { "GMIRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6386 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
6387 0, 0
6388 },
6389 { "GMIWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6390 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
6391 0, 0
6392 },
6393 { "MAM_D0MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6394 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D0MEM_SED_COUNT),
6395 0, 0
6396 },
6397 { "MAM_D1MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6398 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D1MEM_SED_COUNT),
6399 0, 0
6400 },
6401 { "MAM_D2MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6402 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D2MEM_SED_COUNT),
6403 0, 0
6404 },
6405 { "MAM_D3MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6406 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D3MEM_SED_COUNT),
6407 0, 0
6408 }
6409 };
6410
gfx_v9_0_ras_error_inject(struct amdgpu_device * adev,void * inject_if,uint32_t instance_mask)6411 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
6412 void *inject_if, uint32_t instance_mask)
6413 {
6414 struct ras_inject_if *info = (struct ras_inject_if *)inject_if;
6415 int ret;
6416 struct ta_ras_trigger_error_input block_info = { 0 };
6417
6418 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6419 return -EINVAL;
6420
6421 if (info->head.sub_block_index >= ARRAY_SIZE(ras_gfx_subblocks))
6422 return -EINVAL;
6423
6424 if (!ras_gfx_subblocks[info->head.sub_block_index].name)
6425 return -EPERM;
6426
6427 if (!(ras_gfx_subblocks[info->head.sub_block_index].hw_supported_error_type &
6428 info->head.type)) {
6429 DRM_ERROR("GFX Subblock %s, hardware do not support type 0x%x\n",
6430 ras_gfx_subblocks[info->head.sub_block_index].name,
6431 info->head.type);
6432 return -EPERM;
6433 }
6434
6435 if (!(ras_gfx_subblocks[info->head.sub_block_index].sw_supported_error_type &
6436 info->head.type)) {
6437 DRM_ERROR("GFX Subblock %s, driver do not support type 0x%x\n",
6438 ras_gfx_subblocks[info->head.sub_block_index].name,
6439 info->head.type);
6440 return -EPERM;
6441 }
6442
6443 block_info.block_id = amdgpu_ras_block_to_ta(info->head.block);
6444 block_info.sub_block_index =
6445 ras_gfx_subblocks[info->head.sub_block_index].ta_subblock;
6446 block_info.inject_error_type = amdgpu_ras_error_to_ta(info->head.type);
6447 block_info.address = info->address;
6448 block_info.value = info->value;
6449
6450 mutex_lock(&adev->grbm_idx_mutex);
6451 ret = psp_ras_trigger_error(&adev->psp, &block_info, instance_mask);
6452 mutex_unlock(&adev->grbm_idx_mutex);
6453
6454 return ret;
6455 }
6456
6457 static const char *vml2_mems[] = {
6458 "UTC_VML2_BANK_CACHE_0_BIGK_MEM0",
6459 "UTC_VML2_BANK_CACHE_0_BIGK_MEM1",
6460 "UTC_VML2_BANK_CACHE_0_4K_MEM0",
6461 "UTC_VML2_BANK_CACHE_0_4K_MEM1",
6462 "UTC_VML2_BANK_CACHE_1_BIGK_MEM0",
6463 "UTC_VML2_BANK_CACHE_1_BIGK_MEM1",
6464 "UTC_VML2_BANK_CACHE_1_4K_MEM0",
6465 "UTC_VML2_BANK_CACHE_1_4K_MEM1",
6466 "UTC_VML2_BANK_CACHE_2_BIGK_MEM0",
6467 "UTC_VML2_BANK_CACHE_2_BIGK_MEM1",
6468 "UTC_VML2_BANK_CACHE_2_4K_MEM0",
6469 "UTC_VML2_BANK_CACHE_2_4K_MEM1",
6470 "UTC_VML2_BANK_CACHE_3_BIGK_MEM0",
6471 "UTC_VML2_BANK_CACHE_3_BIGK_MEM1",
6472 "UTC_VML2_BANK_CACHE_3_4K_MEM0",
6473 "UTC_VML2_BANK_CACHE_3_4K_MEM1",
6474 };
6475
6476 static const char *vml2_walker_mems[] = {
6477 "UTC_VML2_CACHE_PDE0_MEM0",
6478 "UTC_VML2_CACHE_PDE0_MEM1",
6479 "UTC_VML2_CACHE_PDE1_MEM0",
6480 "UTC_VML2_CACHE_PDE1_MEM1",
6481 "UTC_VML2_CACHE_PDE2_MEM0",
6482 "UTC_VML2_CACHE_PDE2_MEM1",
6483 "UTC_VML2_RDIF_LOG_FIFO",
6484 };
6485
6486 static const char *atc_l2_cache_2m_mems[] = {
6487 "UTC_ATCL2_CACHE_2M_BANK0_WAY0_MEM",
6488 "UTC_ATCL2_CACHE_2M_BANK0_WAY1_MEM",
6489 "UTC_ATCL2_CACHE_2M_BANK1_WAY0_MEM",
6490 "UTC_ATCL2_CACHE_2M_BANK1_WAY1_MEM",
6491 };
6492
6493 static const char *atc_l2_cache_4k_mems[] = {
6494 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM0",
6495 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM1",
6496 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM2",
6497 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM3",
6498 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM4",
6499 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM5",
6500 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM6",
6501 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM7",
6502 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM0",
6503 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM1",
6504 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM2",
6505 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM3",
6506 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM4",
6507 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM5",
6508 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM6",
6509 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM7",
6510 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM0",
6511 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM1",
6512 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM2",
6513 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM3",
6514 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM4",
6515 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM5",
6516 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM6",
6517 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM7",
6518 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM0",
6519 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM1",
6520 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM2",
6521 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM3",
6522 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM4",
6523 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM5",
6524 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM6",
6525 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM7",
6526 };
6527
gfx_v9_0_query_utc_edc_status(struct amdgpu_device * adev,struct ras_err_data * err_data)6528 static int gfx_v9_0_query_utc_edc_status(struct amdgpu_device *adev,
6529 struct ras_err_data *err_data)
6530 {
6531 uint32_t i, data;
6532 uint32_t sec_count, ded_count;
6533
6534 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6535 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0);
6536 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6537 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0);
6538 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6539 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0);
6540 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6541 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0);
6542
6543 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) {
6544 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i);
6545 data = RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT);
6546
6547 sec_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, SEC_COUNT);
6548 if (sec_count) {
6549 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6550 "SEC %d\n", i, vml2_mems[i], sec_count);
6551 err_data->ce_count += sec_count;
6552 }
6553
6554 ded_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, DED_COUNT);
6555 if (ded_count) {
6556 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6557 "DED %d\n", i, vml2_mems[i], ded_count);
6558 err_data->ue_count += ded_count;
6559 }
6560 }
6561
6562 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) {
6563 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i);
6564 data = RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT);
6565
6566 sec_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6567 SEC_COUNT);
6568 if (sec_count) {
6569 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6570 "SEC %d\n", i, vml2_walker_mems[i], sec_count);
6571 err_data->ce_count += sec_count;
6572 }
6573
6574 ded_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6575 DED_COUNT);
6576 if (ded_count) {
6577 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6578 "DED %d\n", i, vml2_walker_mems[i], ded_count);
6579 err_data->ue_count += ded_count;
6580 }
6581 }
6582
6583 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) {
6584 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i);
6585 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT);
6586
6587 sec_count = (data & 0x00006000L) >> 0xd;
6588 if (sec_count) {
6589 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6590 "SEC %d\n", i, atc_l2_cache_2m_mems[i],
6591 sec_count);
6592 err_data->ce_count += sec_count;
6593 }
6594 }
6595
6596 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) {
6597 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i);
6598 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT);
6599
6600 sec_count = (data & 0x00006000L) >> 0xd;
6601 if (sec_count) {
6602 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6603 "SEC %d\n", i, atc_l2_cache_4k_mems[i],
6604 sec_count);
6605 err_data->ce_count += sec_count;
6606 }
6607
6608 ded_count = (data & 0x00018000L) >> 0xf;
6609 if (ded_count) {
6610 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6611 "DED %d\n", i, atc_l2_cache_4k_mems[i],
6612 ded_count);
6613 err_data->ue_count += ded_count;
6614 }
6615 }
6616
6617 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6618 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6619 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6620 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6621
6622 return 0;
6623 }
6624
gfx_v9_0_ras_error_count(struct amdgpu_device * adev,const struct soc15_reg_entry * reg,uint32_t se_id,uint32_t inst_id,uint32_t value,uint32_t * sec_count,uint32_t * ded_count)6625 static int gfx_v9_0_ras_error_count(struct amdgpu_device *adev,
6626 const struct soc15_reg_entry *reg,
6627 uint32_t se_id, uint32_t inst_id, uint32_t value,
6628 uint32_t *sec_count, uint32_t *ded_count)
6629 {
6630 uint32_t i;
6631 uint32_t sec_cnt, ded_cnt;
6632
6633 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_ras_fields); i++) {
6634 if(gfx_v9_0_ras_fields[i].reg_offset != reg->reg_offset ||
6635 gfx_v9_0_ras_fields[i].seg != reg->seg ||
6636 gfx_v9_0_ras_fields[i].inst != reg->inst)
6637 continue;
6638
6639 sec_cnt = (value &
6640 gfx_v9_0_ras_fields[i].sec_count_mask) >>
6641 gfx_v9_0_ras_fields[i].sec_count_shift;
6642 if (sec_cnt) {
6643 dev_info(adev->dev, "GFX SubBlock %s, "
6644 "Instance[%d][%d], SEC %d\n",
6645 gfx_v9_0_ras_fields[i].name,
6646 se_id, inst_id,
6647 sec_cnt);
6648 *sec_count += sec_cnt;
6649 }
6650
6651 ded_cnt = (value &
6652 gfx_v9_0_ras_fields[i].ded_count_mask) >>
6653 gfx_v9_0_ras_fields[i].ded_count_shift;
6654 if (ded_cnt) {
6655 dev_info(adev->dev, "GFX SubBlock %s, "
6656 "Instance[%d][%d], DED %d\n",
6657 gfx_v9_0_ras_fields[i].name,
6658 se_id, inst_id,
6659 ded_cnt);
6660 *ded_count += ded_cnt;
6661 }
6662 }
6663
6664 return 0;
6665 }
6666
gfx_v9_0_reset_ras_error_count(struct amdgpu_device * adev)6667 static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev)
6668 {
6669 int i, j, k;
6670
6671 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6672 return;
6673
6674 /* read back registers to clear the counters */
6675 mutex_lock(&adev->grbm_idx_mutex);
6676 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) {
6677 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) {
6678 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) {
6679 amdgpu_gfx_select_se_sh(adev, j, 0x0, k, 0);
6680 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i]));
6681 }
6682 }
6683 }
6684 WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, 0xe0000000);
6685 mutex_unlock(&adev->grbm_idx_mutex);
6686
6687 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6688 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0);
6689 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6690 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0);
6691 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6692 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0);
6693 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6694 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0);
6695
6696 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) {
6697 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i);
6698 RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT);
6699 }
6700
6701 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) {
6702 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i);
6703 RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT);
6704 }
6705
6706 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) {
6707 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i);
6708 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT);
6709 }
6710
6711 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) {
6712 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i);
6713 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT);
6714 }
6715
6716 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6717 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6718 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6719 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6720 }
6721
gfx_v9_0_query_ras_error_count(struct amdgpu_device * adev,void * ras_error_status)6722 static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
6723 void *ras_error_status)
6724 {
6725 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
6726 uint32_t sec_count = 0, ded_count = 0;
6727 uint32_t i, j, k;
6728 uint32_t reg_value;
6729
6730 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6731 return;
6732
6733 err_data->ue_count = 0;
6734 err_data->ce_count = 0;
6735
6736 mutex_lock(&adev->grbm_idx_mutex);
6737
6738 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) {
6739 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) {
6740 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) {
6741 amdgpu_gfx_select_se_sh(adev, j, 0, k, 0);
6742 reg_value =
6743 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i]));
6744 if (reg_value)
6745 gfx_v9_0_ras_error_count(adev,
6746 &gfx_v9_0_edc_counter_regs[i],
6747 j, k, reg_value,
6748 &sec_count, &ded_count);
6749 }
6750 }
6751 }
6752
6753 err_data->ce_count += sec_count;
6754 err_data->ue_count += ded_count;
6755
6756 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
6757 mutex_unlock(&adev->grbm_idx_mutex);
6758
6759 gfx_v9_0_query_utc_edc_status(adev, err_data);
6760 }
6761
gfx_v9_0_emit_mem_sync(struct amdgpu_ring * ring)6762 static void gfx_v9_0_emit_mem_sync(struct amdgpu_ring *ring)
6763 {
6764 const unsigned int cp_coher_cntl =
6765 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_ICACHE_ACTION_ENA(1) |
6766 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_KCACHE_ACTION_ENA(1) |
6767 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_ACTION_ENA(1) |
6768 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TCL1_ACTION_ENA(1) |
6769 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_WB_ACTION_ENA(1);
6770
6771 /* ACQUIRE_MEM -make one or more surfaces valid for use by the subsequent operations */
6772 amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
6773 amdgpu_ring_write(ring, cp_coher_cntl); /* CP_COHER_CNTL */
6774 amdgpu_ring_write(ring, 0xffffffff); /* CP_COHER_SIZE */
6775 amdgpu_ring_write(ring, 0xffffff); /* CP_COHER_SIZE_HI */
6776 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */
6777 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE_HI */
6778 amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */
6779 }
6780
gfx_v9_0_emit_wave_limit_cs(struct amdgpu_ring * ring,uint32_t pipe,bool enable)6781 static void gfx_v9_0_emit_wave_limit_cs(struct amdgpu_ring *ring,
6782 uint32_t pipe, bool enable)
6783 {
6784 struct amdgpu_device *adev = ring->adev;
6785 uint32_t val;
6786 uint32_t wcl_cs_reg;
6787
6788 /* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */
6789 val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS0_DEFAULT;
6790
6791 switch (pipe) {
6792 case 0:
6793 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS0);
6794 break;
6795 case 1:
6796 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS1);
6797 break;
6798 case 2:
6799 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS2);
6800 break;
6801 case 3:
6802 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS3);
6803 break;
6804 default:
6805 DRM_DEBUG("invalid pipe %d\n", pipe);
6806 return;
6807 }
6808
6809 amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val);
6810
6811 }
gfx_v9_0_emit_wave_limit(struct amdgpu_ring * ring,bool enable)6812 static void gfx_v9_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable)
6813 {
6814 struct amdgpu_device *adev = ring->adev;
6815 uint32_t val;
6816 int i;
6817
6818
6819 /* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit
6820 * number of gfx waves. Setting 5 bit will make sure gfx only gets
6821 * around 25% of gpu resources.
6822 */
6823 val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT;
6824 amdgpu_ring_emit_wreg(ring,
6825 SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_GFX),
6826 val);
6827
6828 /* Restrict waves for normal/low priority compute queues as well
6829 * to get best QoS for high priority compute jobs.
6830 *
6831 * amdgpu controls only 1st ME(0-3 CS pipes).
6832 */
6833 for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
6834 if (i != ring->pipe)
6835 gfx_v9_0_emit_wave_limit_cs(ring, i, enable);
6836
6837 }
6838 }
6839
6840 static const struct amd_ip_funcs gfx_v9_0_ip_funcs = {
6841 .name = "gfx_v9_0",
6842 .early_init = gfx_v9_0_early_init,
6843 .late_init = gfx_v9_0_late_init,
6844 .sw_init = gfx_v9_0_sw_init,
6845 .sw_fini = gfx_v9_0_sw_fini,
6846 .hw_init = gfx_v9_0_hw_init,
6847 .hw_fini = gfx_v9_0_hw_fini,
6848 .suspend = gfx_v9_0_suspend,
6849 .resume = gfx_v9_0_resume,
6850 .is_idle = gfx_v9_0_is_idle,
6851 .wait_for_idle = gfx_v9_0_wait_for_idle,
6852 .soft_reset = gfx_v9_0_soft_reset,
6853 .set_clockgating_state = gfx_v9_0_set_clockgating_state,
6854 .set_powergating_state = gfx_v9_0_set_powergating_state,
6855 .get_clockgating_state = gfx_v9_0_get_clockgating_state,
6856 };
6857
6858 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
6859 .type = AMDGPU_RING_TYPE_GFX,
6860 .align_mask = 0xff,
6861 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
6862 .support_64bit_ptrs = true,
6863 .secure_submission_supported = true,
6864 .get_rptr = gfx_v9_0_ring_get_rptr_gfx,
6865 .get_wptr = gfx_v9_0_ring_get_wptr_gfx,
6866 .set_wptr = gfx_v9_0_ring_set_wptr_gfx,
6867 .emit_frame_size = /* totally 242 maximum if 16 IBs */
6868 5 + /* COND_EXEC */
6869 7 + /* PIPELINE_SYNC */
6870 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6871 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6872 2 + /* VM_FLUSH */
6873 8 + /* FENCE for VM_FLUSH */
6874 20 + /* GDS switch */
6875 4 + /* double SWITCH_BUFFER,
6876 the first COND_EXEC jump to the place just
6877 prior to this double SWITCH_BUFFER */
6878 5 + /* COND_EXEC */
6879 7 + /* HDP_flush */
6880 4 + /* VGT_flush */
6881 14 + /* CE_META */
6882 31 + /* DE_META */
6883 3 + /* CNTX_CTRL */
6884 5 + /* HDP_INVL */
6885 8 + 8 + /* FENCE x2 */
6886 2 + /* SWITCH_BUFFER */
6887 7, /* gfx_v9_0_emit_mem_sync */
6888 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_gfx */
6889 .emit_ib = gfx_v9_0_ring_emit_ib_gfx,
6890 .emit_fence = gfx_v9_0_ring_emit_fence,
6891 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
6892 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
6893 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
6894 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
6895 .test_ring = gfx_v9_0_ring_test_ring,
6896 .insert_nop = amdgpu_ring_insert_nop,
6897 .pad_ib = amdgpu_ring_generic_pad_ib,
6898 .emit_switch_buffer = gfx_v9_ring_emit_sb,
6899 .emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
6900 .init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
6901 .patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec,
6902 .preempt_ib = gfx_v9_0_ring_preempt_ib,
6903 .emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl,
6904 .emit_wreg = gfx_v9_0_ring_emit_wreg,
6905 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6906 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6907 .soft_recovery = gfx_v9_0_ring_soft_recovery,
6908 .emit_mem_sync = gfx_v9_0_emit_mem_sync,
6909 };
6910
6911 static const struct amdgpu_ring_funcs gfx_v9_0_sw_ring_funcs_gfx = {
6912 .type = AMDGPU_RING_TYPE_GFX,
6913 .align_mask = 0xff,
6914 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
6915 .support_64bit_ptrs = true,
6916 .secure_submission_supported = true,
6917 .get_rptr = amdgpu_sw_ring_get_rptr_gfx,
6918 .get_wptr = amdgpu_sw_ring_get_wptr_gfx,
6919 .set_wptr = amdgpu_sw_ring_set_wptr_gfx,
6920 .emit_frame_size = /* totally 242 maximum if 16 IBs */
6921 5 + /* COND_EXEC */
6922 7 + /* PIPELINE_SYNC */
6923 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6924 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6925 2 + /* VM_FLUSH */
6926 8 + /* FENCE for VM_FLUSH */
6927 20 + /* GDS switch */
6928 4 + /* double SWITCH_BUFFER,
6929 * the first COND_EXEC jump to the place just
6930 * prior to this double SWITCH_BUFFER
6931 */
6932 5 + /* COND_EXEC */
6933 7 + /* HDP_flush */
6934 4 + /* VGT_flush */
6935 14 + /* CE_META */
6936 31 + /* DE_META */
6937 3 + /* CNTX_CTRL */
6938 5 + /* HDP_INVL */
6939 8 + 8 + /* FENCE x2 */
6940 2 + /* SWITCH_BUFFER */
6941 7, /* gfx_v9_0_emit_mem_sync */
6942 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_gfx */
6943 .emit_ib = gfx_v9_0_ring_emit_ib_gfx,
6944 .emit_fence = gfx_v9_0_ring_emit_fence,
6945 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
6946 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
6947 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
6948 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
6949 .test_ring = gfx_v9_0_ring_test_ring,
6950 .test_ib = gfx_v9_0_ring_test_ib,
6951 .insert_nop = amdgpu_sw_ring_insert_nop,
6952 .pad_ib = amdgpu_ring_generic_pad_ib,
6953 .emit_switch_buffer = gfx_v9_ring_emit_sb,
6954 .emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
6955 .init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
6956 .patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec,
6957 .emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl,
6958 .emit_wreg = gfx_v9_0_ring_emit_wreg,
6959 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6960 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6961 .soft_recovery = gfx_v9_0_ring_soft_recovery,
6962 .emit_mem_sync = gfx_v9_0_emit_mem_sync,
6963 .patch_cntl = gfx_v9_0_ring_patch_cntl,
6964 .patch_de = gfx_v9_0_ring_patch_de_meta,
6965 .patch_ce = gfx_v9_0_ring_patch_ce_meta,
6966 };
6967
6968 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
6969 .type = AMDGPU_RING_TYPE_COMPUTE,
6970 .align_mask = 0xff,
6971 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
6972 .support_64bit_ptrs = true,
6973 .get_rptr = gfx_v9_0_ring_get_rptr_compute,
6974 .get_wptr = gfx_v9_0_ring_get_wptr_compute,
6975 .set_wptr = gfx_v9_0_ring_set_wptr_compute,
6976 .emit_frame_size =
6977 20 + /* gfx_v9_0_ring_emit_gds_switch */
6978 7 + /* gfx_v9_0_ring_emit_hdp_flush */
6979 5 + /* hdp invalidate */
6980 7 + /* gfx_v9_0_ring_emit_pipeline_sync */
6981 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6982 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6983 2 + /* gfx_v9_0_ring_emit_vm_flush */
6984 8 + 8 + 8 + /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
6985 7 + /* gfx_v9_0_emit_mem_sync */
6986 5 + /* gfx_v9_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */
6987 15, /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */
6988 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */
6989 .emit_ib = gfx_v9_0_ring_emit_ib_compute,
6990 .emit_fence = gfx_v9_0_ring_emit_fence,
6991 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
6992 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
6993 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
6994 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
6995 .test_ring = gfx_v9_0_ring_test_ring,
6996 .test_ib = gfx_v9_0_ring_test_ib,
6997 .insert_nop = amdgpu_ring_insert_nop,
6998 .pad_ib = amdgpu_ring_generic_pad_ib,
6999 .emit_wreg = gfx_v9_0_ring_emit_wreg,
7000 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
7001 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
7002 .emit_mem_sync = gfx_v9_0_emit_mem_sync,
7003 .emit_wave_limit = gfx_v9_0_emit_wave_limit,
7004 };
7005
7006 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
7007 .type = AMDGPU_RING_TYPE_KIQ,
7008 .align_mask = 0xff,
7009 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
7010 .support_64bit_ptrs = true,
7011 .get_rptr = gfx_v9_0_ring_get_rptr_compute,
7012 .get_wptr = gfx_v9_0_ring_get_wptr_compute,
7013 .set_wptr = gfx_v9_0_ring_set_wptr_compute,
7014 .emit_frame_size =
7015 20 + /* gfx_v9_0_ring_emit_gds_switch */
7016 7 + /* gfx_v9_0_ring_emit_hdp_flush */
7017 5 + /* hdp invalidate */
7018 7 + /* gfx_v9_0_ring_emit_pipeline_sync */
7019 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
7020 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
7021 2 + /* gfx_v9_0_ring_emit_vm_flush */
7022 8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */
7023 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */
7024 .emit_fence = gfx_v9_0_ring_emit_fence_kiq,
7025 .test_ring = gfx_v9_0_ring_test_ring,
7026 .insert_nop = amdgpu_ring_insert_nop,
7027 .pad_ib = amdgpu_ring_generic_pad_ib,
7028 .emit_rreg = gfx_v9_0_ring_emit_rreg,
7029 .emit_wreg = gfx_v9_0_ring_emit_wreg,
7030 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
7031 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
7032 };
7033
gfx_v9_0_set_ring_funcs(struct amdgpu_device * adev)7034 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev)
7035 {
7036 int i;
7037
7038 adev->gfx.kiq[0].ring.funcs = &gfx_v9_0_ring_funcs_kiq;
7039
7040 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
7041 adev->gfx.gfx_ring[i].funcs = &gfx_v9_0_ring_funcs_gfx;
7042
7043 if (adev->gfx.num_gfx_rings) {
7044 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++)
7045 adev->gfx.sw_gfx_ring[i].funcs = &gfx_v9_0_sw_ring_funcs_gfx;
7046 }
7047
7048 for (i = 0; i < adev->gfx.num_compute_rings; i++)
7049 adev->gfx.compute_ring[i].funcs = &gfx_v9_0_ring_funcs_compute;
7050 }
7051
7052 static const struct amdgpu_irq_src_funcs gfx_v9_0_eop_irq_funcs = {
7053 .set = gfx_v9_0_set_eop_interrupt_state,
7054 .process = gfx_v9_0_eop_irq,
7055 };
7056
7057 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_reg_irq_funcs = {
7058 .set = gfx_v9_0_set_priv_reg_fault_state,
7059 .process = gfx_v9_0_priv_reg_irq,
7060 };
7061
7062 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_inst_irq_funcs = {
7063 .set = gfx_v9_0_set_priv_inst_fault_state,
7064 .process = gfx_v9_0_priv_inst_irq,
7065 };
7066
7067 static const struct amdgpu_irq_src_funcs gfx_v9_0_cp_ecc_error_irq_funcs = {
7068 .set = gfx_v9_0_set_cp_ecc_error_state,
7069 .process = amdgpu_gfx_cp_ecc_error_irq,
7070 };
7071
7072
gfx_v9_0_set_irq_funcs(struct amdgpu_device * adev)7073 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev)
7074 {
7075 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
7076 adev->gfx.eop_irq.funcs = &gfx_v9_0_eop_irq_funcs;
7077
7078 adev->gfx.priv_reg_irq.num_types = 1;
7079 adev->gfx.priv_reg_irq.funcs = &gfx_v9_0_priv_reg_irq_funcs;
7080
7081 adev->gfx.priv_inst_irq.num_types = 1;
7082 adev->gfx.priv_inst_irq.funcs = &gfx_v9_0_priv_inst_irq_funcs;
7083
7084 adev->gfx.cp_ecc_error_irq.num_types = 2; /*C5 ECC error and C9 FUE error*/
7085 adev->gfx.cp_ecc_error_irq.funcs = &gfx_v9_0_cp_ecc_error_irq_funcs;
7086 }
7087
gfx_v9_0_set_rlc_funcs(struct amdgpu_device * adev)7088 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev)
7089 {
7090 switch (adev->ip_versions[GC_HWIP][0]) {
7091 case IP_VERSION(9, 0, 1):
7092 case IP_VERSION(9, 2, 1):
7093 case IP_VERSION(9, 4, 0):
7094 case IP_VERSION(9, 2, 2):
7095 case IP_VERSION(9, 1, 0):
7096 case IP_VERSION(9, 4, 1):
7097 case IP_VERSION(9, 3, 0):
7098 case IP_VERSION(9, 4, 2):
7099 adev->gfx.rlc.funcs = &gfx_v9_0_rlc_funcs;
7100 break;
7101 default:
7102 break;
7103 }
7104 }
7105
gfx_v9_0_set_gds_init(struct amdgpu_device * adev)7106 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev)
7107 {
7108 /* init asci gds info */
7109 switch (adev->ip_versions[GC_HWIP][0]) {
7110 case IP_VERSION(9, 0, 1):
7111 case IP_VERSION(9, 2, 1):
7112 case IP_VERSION(9, 4, 0):
7113 adev->gds.gds_size = 0x10000;
7114 break;
7115 case IP_VERSION(9, 2, 2):
7116 case IP_VERSION(9, 1, 0):
7117 case IP_VERSION(9, 4, 1):
7118 adev->gds.gds_size = 0x1000;
7119 break;
7120 case IP_VERSION(9, 4, 2):
7121 /* aldebaran removed all the GDS internal memory,
7122 * only support GWS opcode in kernel, like barrier
7123 * semaphore.etc */
7124 adev->gds.gds_size = 0;
7125 break;
7126 default:
7127 adev->gds.gds_size = 0x10000;
7128 break;
7129 }
7130
7131 switch (adev->ip_versions[GC_HWIP][0]) {
7132 case IP_VERSION(9, 0, 1):
7133 case IP_VERSION(9, 4, 0):
7134 adev->gds.gds_compute_max_wave_id = 0x7ff;
7135 break;
7136 case IP_VERSION(9, 2, 1):
7137 adev->gds.gds_compute_max_wave_id = 0x27f;
7138 break;
7139 case IP_VERSION(9, 2, 2):
7140 case IP_VERSION(9, 1, 0):
7141 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
7142 adev->gds.gds_compute_max_wave_id = 0x77; /* raven2 */
7143 else
7144 adev->gds.gds_compute_max_wave_id = 0x15f; /* raven1 */
7145 break;
7146 case IP_VERSION(9, 4, 1):
7147 adev->gds.gds_compute_max_wave_id = 0xfff;
7148 break;
7149 case IP_VERSION(9, 4, 2):
7150 /* deprecated for Aldebaran, no usage at all */
7151 adev->gds.gds_compute_max_wave_id = 0;
7152 break;
7153 default:
7154 /* this really depends on the chip */
7155 adev->gds.gds_compute_max_wave_id = 0x7ff;
7156 break;
7157 }
7158
7159 adev->gds.gws_size = 64;
7160 adev->gds.oa_size = 16;
7161 }
7162
gfx_v9_0_set_user_cu_inactive_bitmap(struct amdgpu_device * adev,u32 bitmap)7163 static void gfx_v9_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
7164 u32 bitmap)
7165 {
7166 u32 data;
7167
7168 if (!bitmap)
7169 return;
7170
7171 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7172 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7173
7174 WREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG, data);
7175 }
7176
gfx_v9_0_get_cu_active_bitmap(struct amdgpu_device * adev)7177 static u32 gfx_v9_0_get_cu_active_bitmap(struct amdgpu_device *adev)
7178 {
7179 u32 data, mask;
7180
7181 data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
7182 data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
7183
7184 data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7185 data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7186
7187 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
7188
7189 return (~data) & mask;
7190 }
7191
gfx_v9_0_get_cu_info(struct amdgpu_device * adev,struct amdgpu_cu_info * cu_info)7192 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
7193 struct amdgpu_cu_info *cu_info)
7194 {
7195 int i, j, k, counter, active_cu_number = 0;
7196 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
7197 unsigned disable_masks[4 * 4];
7198
7199 if (!adev || !cu_info)
7200 return -EINVAL;
7201
7202 /*
7203 * 16 comes from bitmap array size 4*4, and it can cover all gfx9 ASICs
7204 */
7205 if (adev->gfx.config.max_shader_engines *
7206 adev->gfx.config.max_sh_per_se > 16)
7207 return -EINVAL;
7208
7209 amdgpu_gfx_parse_disable_cu(disable_masks,
7210 adev->gfx.config.max_shader_engines,
7211 adev->gfx.config.max_sh_per_se);
7212
7213 mutex_lock(&adev->grbm_idx_mutex);
7214 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
7215 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
7216 mask = 1;
7217 ao_bitmap = 0;
7218 counter = 0;
7219 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff, 0);
7220 gfx_v9_0_set_user_cu_inactive_bitmap(
7221 adev, disable_masks[i * adev->gfx.config.max_sh_per_se + j]);
7222 bitmap = gfx_v9_0_get_cu_active_bitmap(adev);
7223
7224 /*
7225 * The bitmap(and ao_cu_bitmap) in cu_info structure is
7226 * 4x4 size array, and it's usually suitable for Vega
7227 * ASICs which has 4*2 SE/SH layout.
7228 * But for Arcturus, SE/SH layout is changed to 8*1.
7229 * To mostly reduce the impact, we make it compatible
7230 * with current bitmap array as below:
7231 * SE4,SH0 --> bitmap[0][1]
7232 * SE5,SH0 --> bitmap[1][1]
7233 * SE6,SH0 --> bitmap[2][1]
7234 * SE7,SH0 --> bitmap[3][1]
7235 */
7236 cu_info->bitmap[0][i % 4][j + i / 4] = bitmap;
7237
7238 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
7239 if (bitmap & mask) {
7240 if (counter < adev->gfx.config.max_cu_per_sh)
7241 ao_bitmap |= mask;
7242 counter ++;
7243 }
7244 mask <<= 1;
7245 }
7246 active_cu_number += counter;
7247 if (i < 2 && j < 2)
7248 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
7249 cu_info->ao_cu_bitmap[i % 4][j + i / 4] = ao_bitmap;
7250 }
7251 }
7252 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
7253 mutex_unlock(&adev->grbm_idx_mutex);
7254
7255 cu_info->number = active_cu_number;
7256 cu_info->ao_cu_mask = ao_cu_mask;
7257 cu_info->simd_per_cu = NUM_SIMD_PER_CU;
7258
7259 return 0;
7260 }
7261
7262 const struct amdgpu_ip_block_version gfx_v9_0_ip_block =
7263 {
7264 .type = AMD_IP_BLOCK_TYPE_GFX,
7265 .major = 9,
7266 .minor = 0,
7267 .rev = 0,
7268 .funcs = &gfx_v9_0_ip_funcs,
7269 };
7270
