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
1043 r = amdgpu_ib_get(adev, NULL, 20, 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 /* Now only limit the quirk on the APU gfx9 series and already
3037 * confirmed that the APU gfx10/gfx11 needn't such update.
3038 */
3039 if (adev->flags & AMD_IS_APU &&
3040 adev->in_s3 && !adev->suspend_complete) {
3041 DRM_INFO(" Will skip the CSB packet resubmit\n");
3042 return 0;
3043 }
3044 r = amdgpu_ring_alloc(ring, gfx_v9_0_get_csb_size(adev) + 4 + 3);
3045 if (r) {
3046 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
3047 return r;
3048 }
3049
3050 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3051 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3052
3053 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3054 amdgpu_ring_write(ring, 0x80000000);
3055 amdgpu_ring_write(ring, 0x80000000);
3056
3057 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) {
3058 for (ext = sect->section; ext->extent != NULL; ++ext) {
3059 if (sect->id == SECT_CONTEXT) {
3060 amdgpu_ring_write(ring,
3061 PACKET3(PACKET3_SET_CONTEXT_REG,
3062 ext->reg_count));
3063 amdgpu_ring_write(ring,
3064 ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
3065 for (i = 0; i < ext->reg_count; i++)
3066 amdgpu_ring_write(ring, ext->extent[i]);
3067 }
3068 }
3069 }
3070
3071 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3072 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
3073
3074 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3075 amdgpu_ring_write(ring, 0);
3076
3077 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
3078 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
3079 amdgpu_ring_write(ring, 0x8000);
3080 amdgpu_ring_write(ring, 0x8000);
3081
3082 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG,1));
3083 tmp = (PACKET3_SET_UCONFIG_REG_INDEX_TYPE |
3084 (SOC15_REG_OFFSET(GC, 0, mmVGT_INDEX_TYPE) - PACKET3_SET_UCONFIG_REG_START));
3085 amdgpu_ring_write(ring, tmp);
3086 amdgpu_ring_write(ring, 0);
3087
3088 amdgpu_ring_commit(ring);
3089
3090 return 0;
3091 }
3092
gfx_v9_0_cp_gfx_resume(struct amdgpu_device * adev)3093 static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev)
3094 {
3095 struct amdgpu_ring *ring;
3096 u32 tmp;
3097 u32 rb_bufsz;
3098 u64 rb_addr, rptr_addr, wptr_gpu_addr;
3099
3100 /* Set the write pointer delay */
3101 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_DELAY, 0);
3102
3103 /* set the RB to use vmid 0 */
3104 WREG32_SOC15(GC, 0, mmCP_RB_VMID, 0);
3105
3106 /* Set ring buffer size */
3107 ring = &adev->gfx.gfx_ring[0];
3108 rb_bufsz = order_base_2(ring->ring_size / 8);
3109 tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
3110 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
3111 #ifdef __BIG_ENDIAN
3112 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
3113 #endif
3114 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3115
3116 /* Initialize the ring buffer's write pointers */
3117 ring->wptr = 0;
3118 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
3119 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
3120
3121 /* set the wb address wether it's enabled or not */
3122 rptr_addr = ring->rptr_gpu_addr;
3123 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
3124 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3125
3126 wptr_gpu_addr = ring->wptr_gpu_addr;
3127 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
3128 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
3129
3130 mdelay(1);
3131 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
3132
3133 rb_addr = ring->gpu_addr >> 8;
3134 WREG32_SOC15(GC, 0, mmCP_RB0_BASE, rb_addr);
3135 WREG32_SOC15(GC, 0, mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
3136
3137 tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL);
3138 if (ring->use_doorbell) {
3139 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3140 DOORBELL_OFFSET, ring->doorbell_index);
3141 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3142 DOORBELL_EN, 1);
3143 } else {
3144 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
3145 }
3146 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, tmp);
3147
3148 tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
3149 DOORBELL_RANGE_LOWER, ring->doorbell_index);
3150 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
3151
3152 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER,
3153 CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
3154
3155
3156 /* start the ring */
3157 gfx_v9_0_cp_gfx_start(adev);
3158
3159 return 0;
3160 }
3161
gfx_v9_0_cp_compute_enable(struct amdgpu_device * adev,bool enable)3162 static void gfx_v9_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
3163 {
3164 if (enable) {
3165 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL, 0);
3166 } else {
3167 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL,
3168 (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
3169 adev->gfx.kiq[0].ring.sched.ready = false;
3170 }
3171 udelay(50);
3172 }
3173
gfx_v9_0_cp_compute_load_microcode(struct amdgpu_device * adev)3174 static int gfx_v9_0_cp_compute_load_microcode(struct amdgpu_device *adev)
3175 {
3176 const struct gfx_firmware_header_v1_0 *mec_hdr;
3177 const __le32 *fw_data;
3178 unsigned i;
3179 u32 tmp;
3180
3181 if (!adev->gfx.mec_fw)
3182 return -EINVAL;
3183
3184 gfx_v9_0_cp_compute_enable(adev, false);
3185
3186 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3187 amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3188
3189 fw_data = (const __le32 *)
3190 (adev->gfx.mec_fw->data +
3191 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
3192 tmp = 0;
3193 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
3194 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
3195 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL, tmp);
3196
3197 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO,
3198 adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000);
3199 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI,
3200 upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
3201
3202 /* MEC1 */
3203 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3204 mec_hdr->jt_offset);
3205 for (i = 0; i < mec_hdr->jt_size; i++)
3206 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_DATA,
3207 le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
3208
3209 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,
3210 adev->gfx.mec_fw_version);
3211 /* Todo : Loading MEC2 firmware is only necessary if MEC2 should run different microcode than MEC1. */
3212
3213 return 0;
3214 }
3215
3216 /* KIQ functions */
gfx_v9_0_kiq_setting(struct amdgpu_ring * ring)3217 static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)
3218 {
3219 uint32_t tmp;
3220 struct amdgpu_device *adev = ring->adev;
3221
3222 /* tell RLC which is KIQ queue */
3223 tmp = RREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS);
3224 tmp &= 0xffffff00;
3225 tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
3226 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3227 tmp |= 0x80;
3228 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
3229 }
3230
gfx_v9_0_mqd_set_priority(struct amdgpu_ring * ring,struct v9_mqd * mqd)3231 static void gfx_v9_0_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd)
3232 {
3233 struct amdgpu_device *adev = ring->adev;
3234
3235 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
3236 if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) {
3237 mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
3238 mqd->cp_hqd_queue_priority =
3239 AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
3240 }
3241 }
3242 }
3243
gfx_v9_0_mqd_init(struct amdgpu_ring * ring)3244 static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring)
3245 {
3246 struct amdgpu_device *adev = ring->adev;
3247 struct v9_mqd *mqd = ring->mqd_ptr;
3248 uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
3249 uint32_t tmp;
3250
3251 mqd->header = 0xC0310800;
3252 mqd->compute_pipelinestat_enable = 0x00000001;
3253 mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
3254 mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
3255 mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
3256 mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
3257 mqd->compute_static_thread_mgmt_se4 = 0xffffffff;
3258 mqd->compute_static_thread_mgmt_se5 = 0xffffffff;
3259 mqd->compute_static_thread_mgmt_se6 = 0xffffffff;
3260 mqd->compute_static_thread_mgmt_se7 = 0xffffffff;
3261 mqd->compute_misc_reserved = 0x00000003;
3262
3263 mqd->dynamic_cu_mask_addr_lo =
3264 lower_32_bits(ring->mqd_gpu_addr
3265 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3266 mqd->dynamic_cu_mask_addr_hi =
3267 upper_32_bits(ring->mqd_gpu_addr
3268 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
3269
3270 eop_base_addr = ring->eop_gpu_addr >> 8;
3271 mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
3272 mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
3273
3274 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3275 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL);
3276 tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
3277 (order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1));
3278
3279 mqd->cp_hqd_eop_control = tmp;
3280
3281 /* enable doorbell? */
3282 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
3283
3284 if (ring->use_doorbell) {
3285 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3286 DOORBELL_OFFSET, ring->doorbell_index);
3287 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3288 DOORBELL_EN, 1);
3289 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3290 DOORBELL_SOURCE, 0);
3291 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3292 DOORBELL_HIT, 0);
3293 } else {
3294 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3295 DOORBELL_EN, 0);
3296 }
3297
3298 mqd->cp_hqd_pq_doorbell_control = tmp;
3299
3300 /* disable the queue if it's active */
3301 ring->wptr = 0;
3302 mqd->cp_hqd_dequeue_request = 0;
3303 mqd->cp_hqd_pq_rptr = 0;
3304 mqd->cp_hqd_pq_wptr_lo = 0;
3305 mqd->cp_hqd_pq_wptr_hi = 0;
3306
3307 /* set the pointer to the MQD */
3308 mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
3309 mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
3310
3311 /* set MQD vmid to 0 */
3312 tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL);
3313 tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
3314 mqd->cp_mqd_control = tmp;
3315
3316 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3317 hqd_gpu_addr = ring->gpu_addr >> 8;
3318 mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
3319 mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3320
3321 /* set up the HQD, this is similar to CP_RB0_CNTL */
3322 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL);
3323 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
3324 (order_base_2(ring->ring_size / 4) - 1));
3325 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
3326 (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1));
3327 #ifdef __BIG_ENDIAN
3328 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
3329 #endif
3330 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
3331 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
3332 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
3333 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
3334 mqd->cp_hqd_pq_control = tmp;
3335
3336 /* set the wb address whether it's enabled or not */
3337 wb_gpu_addr = ring->rptr_gpu_addr;
3338 mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
3339 mqd->cp_hqd_pq_rptr_report_addr_hi =
3340 upper_32_bits(wb_gpu_addr) & 0xffff;
3341
3342 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3343 wb_gpu_addr = ring->wptr_gpu_addr;
3344 mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3345 mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3346
3347 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3348 ring->wptr = 0;
3349 mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR);
3350
3351 /* set the vmid for the queue */
3352 mqd->cp_hqd_vmid = 0;
3353
3354 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE);
3355 tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
3356 mqd->cp_hqd_persistent_state = tmp;
3357
3358 /* set MIN_IB_AVAIL_SIZE */
3359 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL);
3360 tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
3361 mqd->cp_hqd_ib_control = tmp;
3362
3363 /* set static priority for a queue/ring */
3364 gfx_v9_0_mqd_set_priority(ring, mqd);
3365 mqd->cp_hqd_quantum = RREG32_SOC15(GC, 0, mmCP_HQD_QUANTUM);
3366
3367 /* map_queues packet doesn't need activate the queue,
3368 * so only kiq need set this field.
3369 */
3370 if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
3371 mqd->cp_hqd_active = 1;
3372
3373 return 0;
3374 }
3375
gfx_v9_0_kiq_init_register(struct amdgpu_ring * ring)3376 static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring)
3377 {
3378 struct amdgpu_device *adev = ring->adev;
3379 struct v9_mqd *mqd = ring->mqd_ptr;
3380 int j;
3381
3382 /* disable wptr polling */
3383 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3384
3385 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR,
3386 mqd->cp_hqd_eop_base_addr_lo);
3387 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI,
3388 mqd->cp_hqd_eop_base_addr_hi);
3389
3390 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3391 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_CONTROL,
3392 mqd->cp_hqd_eop_control);
3393
3394 /* enable doorbell? */
3395 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3396 mqd->cp_hqd_pq_doorbell_control);
3397
3398 /* disable the queue if it's active */
3399 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3400 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3401 for (j = 0; j < adev->usec_timeout; j++) {
3402 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3403 break;
3404 udelay(1);
3405 }
3406 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3407 mqd->cp_hqd_dequeue_request);
3408 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR,
3409 mqd->cp_hqd_pq_rptr);
3410 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3411 mqd->cp_hqd_pq_wptr_lo);
3412 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3413 mqd->cp_hqd_pq_wptr_hi);
3414 }
3415
3416 /* set the pointer to the MQD */
3417 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR,
3418 mqd->cp_mqd_base_addr_lo);
3419 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR_HI,
3420 mqd->cp_mqd_base_addr_hi);
3421
3422 /* set MQD vmid to 0 */
3423 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_CONTROL,
3424 mqd->cp_mqd_control);
3425
3426 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3427 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE,
3428 mqd->cp_hqd_pq_base_lo);
3429 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE_HI,
3430 mqd->cp_hqd_pq_base_hi);
3431
3432 /* set up the HQD, this is similar to CP_RB0_CNTL */
3433 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_CONTROL,
3434 mqd->cp_hqd_pq_control);
3435
3436 /* set the wb address whether it's enabled or not */
3437 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR,
3438 mqd->cp_hqd_pq_rptr_report_addr_lo);
3439 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3440 mqd->cp_hqd_pq_rptr_report_addr_hi);
3441
3442 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3443 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR,
3444 mqd->cp_hqd_pq_wptr_poll_addr_lo);
3445 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
3446 mqd->cp_hqd_pq_wptr_poll_addr_hi);
3447
3448 /* enable the doorbell if requested */
3449 if (ring->use_doorbell) {
3450 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER,
3451 (adev->doorbell_index.kiq * 2) << 2);
3452 /* If GC has entered CGPG, ringing doorbell > first page
3453 * doesn't wakeup GC. Enlarge CP_MEC_DOORBELL_RANGE_UPPER to
3454 * workaround this issue. And this change has to align with firmware
3455 * update.
3456 */
3457 if (check_if_enlarge_doorbell_range(adev))
3458 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
3459 (adev->doorbell.size - 4));
3460 else
3461 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
3462 (adev->doorbell_index.userqueue_end * 2) << 2);
3463 }
3464
3465 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3466 mqd->cp_hqd_pq_doorbell_control);
3467
3468 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3469 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3470 mqd->cp_hqd_pq_wptr_lo);
3471 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3472 mqd->cp_hqd_pq_wptr_hi);
3473
3474 /* set the vmid for the queue */
3475 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_VMID, mqd->cp_hqd_vmid);
3476
3477 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE,
3478 mqd->cp_hqd_persistent_state);
3479
3480 /* activate the queue */
3481 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE,
3482 mqd->cp_hqd_active);
3483
3484 if (ring->use_doorbell)
3485 WREG32_FIELD15(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
3486
3487 return 0;
3488 }
3489
gfx_v9_0_kiq_fini_register(struct amdgpu_ring * ring)3490 static int gfx_v9_0_kiq_fini_register(struct amdgpu_ring *ring)
3491 {
3492 struct amdgpu_device *adev = ring->adev;
3493 int j;
3494
3495 /* disable the queue if it's active */
3496 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3497
3498 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3499
3500 for (j = 0; j < adev->usec_timeout; j++) {
3501 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3502 break;
3503 udelay(1);
3504 }
3505
3506 if (j == AMDGPU_MAX_USEC_TIMEOUT) {
3507 DRM_DEBUG("KIQ dequeue request failed.\n");
3508
3509 /* Manual disable if dequeue request times out */
3510 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE, 0);
3511 }
3512
3513 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3514 0);
3515 }
3516
3517 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IQ_TIMER, 0);
3518 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IB_CONTROL, 0);
3519 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE, 0);
3520 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000);
3521 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0);
3522 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR, 0);
3523 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 0);
3524 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 0);
3525
3526 return 0;
3527 }
3528
gfx_v9_0_kiq_init_queue(struct amdgpu_ring * ring)3529 static int gfx_v9_0_kiq_init_queue(struct amdgpu_ring *ring)
3530 {
3531 struct amdgpu_device *adev = ring->adev;
3532 struct v9_mqd *mqd = ring->mqd_ptr;
3533 struct v9_mqd *tmp_mqd;
3534
3535 gfx_v9_0_kiq_setting(ring);
3536
3537 /* GPU could be in bad state during probe, driver trigger the reset
3538 * after load the SMU, in this case , the mqd is not be initialized.
3539 * driver need to re-init the mqd.
3540 * check mqd->cp_hqd_pq_control since this value should not be 0
3541 */
3542 tmp_mqd = (struct v9_mqd *)adev->gfx.kiq[0].mqd_backup;
3543 if (amdgpu_in_reset(adev) && tmp_mqd->cp_hqd_pq_control){
3544 /* for GPU_RESET case , reset MQD to a clean status */
3545 if (adev->gfx.kiq[0].mqd_backup)
3546 memcpy(mqd, adev->gfx.kiq[0].mqd_backup, sizeof(struct v9_mqd_allocation));
3547
3548 /* reset ring buffer */
3549 ring->wptr = 0;
3550 amdgpu_ring_clear_ring(ring);
3551
3552 mutex_lock(&adev->srbm_mutex);
3553 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
3554 gfx_v9_0_kiq_init_register(ring);
3555 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
3556 mutex_unlock(&adev->srbm_mutex);
3557 } else {
3558 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3559 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3560 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3561 if (amdgpu_sriov_vf(adev) && adev->in_suspend)
3562 amdgpu_ring_clear_ring(ring);
3563 mutex_lock(&adev->srbm_mutex);
3564 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
3565 gfx_v9_0_mqd_init(ring);
3566 gfx_v9_0_kiq_init_register(ring);
3567 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
3568 mutex_unlock(&adev->srbm_mutex);
3569
3570 if (adev->gfx.kiq[0].mqd_backup)
3571 memcpy(adev->gfx.kiq[0].mqd_backup, mqd, sizeof(struct v9_mqd_allocation));
3572 }
3573
3574 return 0;
3575 }
3576
gfx_v9_0_kcq_init_queue(struct amdgpu_ring * ring)3577 static int gfx_v9_0_kcq_init_queue(struct amdgpu_ring *ring)
3578 {
3579 struct amdgpu_device *adev = ring->adev;
3580 struct v9_mqd *mqd = ring->mqd_ptr;
3581 int mqd_idx = ring - &adev->gfx.compute_ring[0];
3582 struct v9_mqd *tmp_mqd;
3583
3584 /* Same as above kiq init, driver need to re-init the mqd if mqd->cp_hqd_pq_control
3585 * is not be initialized before
3586 */
3587 tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx];
3588
3589 if (!tmp_mqd->cp_hqd_pq_control ||
3590 (!amdgpu_in_reset(adev) && !adev->in_suspend)) {
3591 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
3592 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3593 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3594 mutex_lock(&adev->srbm_mutex);
3595 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, 0);
3596 gfx_v9_0_mqd_init(ring);
3597 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
3598 mutex_unlock(&adev->srbm_mutex);
3599
3600 if (adev->gfx.mec.mqd_backup[mqd_idx])
3601 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
3602 } else {
3603 /* restore MQD to a clean status */
3604 if (adev->gfx.mec.mqd_backup[mqd_idx])
3605 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
3606 /* reset ring buffer */
3607 ring->wptr = 0;
3608 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, 0);
3609 amdgpu_ring_clear_ring(ring);
3610 }
3611
3612 return 0;
3613 }
3614
gfx_v9_0_kiq_resume(struct amdgpu_device * adev)3615 static int gfx_v9_0_kiq_resume(struct amdgpu_device *adev)
3616 {
3617 struct amdgpu_ring *ring;
3618 int r;
3619
3620 ring = &adev->gfx.kiq[0].ring;
3621
3622 r = amdgpu_bo_reserve(ring->mqd_obj, false);
3623 if (unlikely(r != 0))
3624 return r;
3625
3626 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3627 if (unlikely(r != 0)) {
3628 amdgpu_bo_unreserve(ring->mqd_obj);
3629 return r;
3630 }
3631
3632 gfx_v9_0_kiq_init_queue(ring);
3633 amdgpu_bo_kunmap(ring->mqd_obj);
3634 ring->mqd_ptr = NULL;
3635 amdgpu_bo_unreserve(ring->mqd_obj);
3636 return 0;
3637 }
3638
gfx_v9_0_kcq_resume(struct amdgpu_device * adev)3639 static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev)
3640 {
3641 struct amdgpu_ring *ring = NULL;
3642 int r = 0, i;
3643
3644 gfx_v9_0_cp_compute_enable(adev, true);
3645
3646 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3647 ring = &adev->gfx.compute_ring[i];
3648
3649 r = amdgpu_bo_reserve(ring->mqd_obj, false);
3650 if (unlikely(r != 0))
3651 goto done;
3652 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3653 if (!r) {
3654 r = gfx_v9_0_kcq_init_queue(ring);
3655 amdgpu_bo_kunmap(ring->mqd_obj);
3656 ring->mqd_ptr = NULL;
3657 }
3658 amdgpu_bo_unreserve(ring->mqd_obj);
3659 if (r)
3660 goto done;
3661 }
3662
3663 r = amdgpu_gfx_enable_kcq(adev, 0);
3664 done:
3665 return r;
3666 }
3667
gfx_v9_0_cp_resume(struct amdgpu_device * adev)3668 static int gfx_v9_0_cp_resume(struct amdgpu_device *adev)
3669 {
3670 int r, i;
3671 struct amdgpu_ring *ring;
3672
3673 if (!(adev->flags & AMD_IS_APU))
3674 gfx_v9_0_enable_gui_idle_interrupt(adev, false);
3675
3676 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
3677 if (adev->gfx.num_gfx_rings) {
3678 /* legacy firmware loading */
3679 r = gfx_v9_0_cp_gfx_load_microcode(adev);
3680 if (r)
3681 return r;
3682 }
3683
3684 r = gfx_v9_0_cp_compute_load_microcode(adev);
3685 if (r)
3686 return r;
3687 }
3688
3689 r = gfx_v9_0_kiq_resume(adev);
3690 if (r)
3691 return r;
3692
3693 if (adev->gfx.num_gfx_rings) {
3694 r = gfx_v9_0_cp_gfx_resume(adev);
3695 if (r)
3696 return r;
3697 }
3698
3699 r = gfx_v9_0_kcq_resume(adev);
3700 if (r)
3701 return r;
3702
3703 if (adev->gfx.num_gfx_rings) {
3704 ring = &adev->gfx.gfx_ring[0];
3705 r = amdgpu_ring_test_helper(ring);
3706 if (r)
3707 return r;
3708 }
3709
3710 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3711 ring = &adev->gfx.compute_ring[i];
3712 amdgpu_ring_test_helper(ring);
3713 }
3714
3715 gfx_v9_0_enable_gui_idle_interrupt(adev, true);
3716
3717 return 0;
3718 }
3719
gfx_v9_0_init_tcp_config(struct amdgpu_device * adev)3720 static void gfx_v9_0_init_tcp_config(struct amdgpu_device *adev)
3721 {
3722 u32 tmp;
3723
3724 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1) &&
3725 adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 2))
3726 return;
3727
3728 tmp = RREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG);
3729 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE64KHASH,
3730 adev->df.hash_status.hash_64k);
3731 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE2MHASH,
3732 adev->df.hash_status.hash_2m);
3733 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE1GHASH,
3734 adev->df.hash_status.hash_1g);
3735 WREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG, tmp);
3736 }
3737
gfx_v9_0_cp_enable(struct amdgpu_device * adev,bool enable)3738 static void gfx_v9_0_cp_enable(struct amdgpu_device *adev, bool enable)
3739 {
3740 if (adev->gfx.num_gfx_rings)
3741 gfx_v9_0_cp_gfx_enable(adev, enable);
3742 gfx_v9_0_cp_compute_enable(adev, enable);
3743 }
3744
gfx_v9_0_hw_init(void * handle)3745 static int gfx_v9_0_hw_init(void *handle)
3746 {
3747 int r;
3748 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3749
3750 if (!amdgpu_sriov_vf(adev))
3751 gfx_v9_0_init_golden_registers(adev);
3752
3753 gfx_v9_0_constants_init(adev);
3754
3755 gfx_v9_0_init_tcp_config(adev);
3756
3757 r = adev->gfx.rlc.funcs->resume(adev);
3758 if (r)
3759 return r;
3760
3761 r = gfx_v9_0_cp_resume(adev);
3762 if (r)
3763 return r;
3764
3765 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
3766 gfx_v9_4_2_set_power_brake_sequence(adev);
3767
3768 return r;
3769 }
3770
gfx_v9_0_hw_fini(void * handle)3771 static int gfx_v9_0_hw_fini(void *handle)
3772 {
3773 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3774
3775 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
3776 amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
3777 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
3778 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
3779
3780 /* DF freeze and kcq disable will fail */
3781 if (!amdgpu_ras_intr_triggered())
3782 /* disable KCQ to avoid CPC touch memory not valid anymore */
3783 amdgpu_gfx_disable_kcq(adev, 0);
3784
3785 if (amdgpu_sriov_vf(adev)) {
3786 gfx_v9_0_cp_gfx_enable(adev, false);
3787 /* must disable polling for SRIOV when hw finished, otherwise
3788 * CPC engine may still keep fetching WB address which is already
3789 * invalid after sw finished and trigger DMAR reading error in
3790 * hypervisor side.
3791 */
3792 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3793 return 0;
3794 }
3795
3796 /* Use deinitialize sequence from CAIL when unbinding device from driver,
3797 * otherwise KIQ is hanging when binding back
3798 */
3799 if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
3800 mutex_lock(&adev->srbm_mutex);
3801 soc15_grbm_select(adev, adev->gfx.kiq[0].ring.me,
3802 adev->gfx.kiq[0].ring.pipe,
3803 adev->gfx.kiq[0].ring.queue, 0, 0);
3804 gfx_v9_0_kiq_fini_register(&adev->gfx.kiq[0].ring);
3805 soc15_grbm_select(adev, 0, 0, 0, 0, 0);
3806 mutex_unlock(&adev->srbm_mutex);
3807 }
3808
3809 gfx_v9_0_cp_enable(adev, false);
3810
3811 /* Skip stopping RLC with A+A reset or when RLC controls GFX clock */
3812 if ((adev->gmc.xgmi.connected_to_cpu && amdgpu_in_reset(adev)) ||
3813 (adev->ip_versions[GC_HWIP][0] >= IP_VERSION(9, 4, 2))) {
3814 dev_dbg(adev->dev, "Skipping RLC halt\n");
3815 return 0;
3816 }
3817
3818 adev->gfx.rlc.funcs->stop(adev);
3819 return 0;
3820 }
3821
gfx_v9_0_suspend(void * handle)3822 static int gfx_v9_0_suspend(void *handle)
3823 {
3824 return gfx_v9_0_hw_fini(handle);
3825 }
3826
gfx_v9_0_resume(void * handle)3827 static int gfx_v9_0_resume(void *handle)
3828 {
3829 return gfx_v9_0_hw_init(handle);
3830 }
3831
gfx_v9_0_is_idle(void * handle)3832 static bool gfx_v9_0_is_idle(void *handle)
3833 {
3834 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3835
3836 if (REG_GET_FIELD(RREG32_SOC15(GC, 0, mmGRBM_STATUS),
3837 GRBM_STATUS, GUI_ACTIVE))
3838 return false;
3839 else
3840 return true;
3841 }
3842
gfx_v9_0_wait_for_idle(void * handle)3843 static int gfx_v9_0_wait_for_idle(void *handle)
3844 {
3845 unsigned i;
3846 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3847
3848 for (i = 0; i < adev->usec_timeout; i++) {
3849 if (gfx_v9_0_is_idle(handle))
3850 return 0;
3851 udelay(1);
3852 }
3853 return -ETIMEDOUT;
3854 }
3855
gfx_v9_0_soft_reset(void * handle)3856 static int gfx_v9_0_soft_reset(void *handle)
3857 {
3858 u32 grbm_soft_reset = 0;
3859 u32 tmp;
3860 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3861
3862 /* GRBM_STATUS */
3863 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS);
3864 if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
3865 GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
3866 GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
3867 GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
3868 GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
3869 GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) {
3870 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3871 GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
3872 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3873 GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
3874 }
3875
3876 if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
3877 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3878 GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
3879 }
3880
3881 /* GRBM_STATUS2 */
3882 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2);
3883 if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
3884 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3885 GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
3886
3887
3888 if (grbm_soft_reset) {
3889 /* stop the rlc */
3890 adev->gfx.rlc.funcs->stop(adev);
3891
3892 if (adev->gfx.num_gfx_rings)
3893 /* Disable GFX parsing/prefetching */
3894 gfx_v9_0_cp_gfx_enable(adev, false);
3895
3896 /* Disable MEC parsing/prefetching */
3897 gfx_v9_0_cp_compute_enable(adev, false);
3898
3899 if (grbm_soft_reset) {
3900 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3901 tmp |= grbm_soft_reset;
3902 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
3903 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
3904 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3905
3906 udelay(50);
3907
3908 tmp &= ~grbm_soft_reset;
3909 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
3910 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3911 }
3912
3913 /* Wait a little for things to settle down */
3914 udelay(50);
3915 }
3916 return 0;
3917 }
3918
gfx_v9_0_kiq_read_clock(struct amdgpu_device * adev)3919 static uint64_t gfx_v9_0_kiq_read_clock(struct amdgpu_device *adev)
3920 {
3921 signed long r, cnt = 0;
3922 unsigned long flags;
3923 uint32_t seq, reg_val_offs = 0;
3924 uint64_t value = 0;
3925 struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
3926 struct amdgpu_ring *ring = &kiq->ring;
3927
3928 BUG_ON(!ring->funcs->emit_rreg);
3929
3930 spin_lock_irqsave(&kiq->ring_lock, flags);
3931 if (amdgpu_device_wb_get(adev, ®_val_offs)) {
3932 pr_err("critical bug! too many kiq readers\n");
3933 goto failed_unlock;
3934 }
3935 amdgpu_ring_alloc(ring, 32);
3936 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
3937 amdgpu_ring_write(ring, 9 | /* src: register*/
3938 (5 << 8) | /* dst: memory */
3939 (1 << 16) | /* count sel */
3940 (1 << 20)); /* write confirm */
3941 amdgpu_ring_write(ring, 0);
3942 amdgpu_ring_write(ring, 0);
3943 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
3944 reg_val_offs * 4));
3945 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
3946 reg_val_offs * 4));
3947 r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT);
3948 if (r)
3949 goto failed_undo;
3950
3951 amdgpu_ring_commit(ring);
3952 spin_unlock_irqrestore(&kiq->ring_lock, flags);
3953
3954 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
3955
3956 /* don't wait anymore for gpu reset case because this way may
3957 * block gpu_recover() routine forever, e.g. this virt_kiq_rreg
3958 * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will
3959 * never return if we keep waiting in virt_kiq_rreg, which cause
3960 * gpu_recover() hang there.
3961 *
3962 * also don't wait anymore for IRQ context
3963 * */
3964 if (r < 1 && (amdgpu_in_reset(adev)))
3965 goto failed_kiq_read;
3966
3967 might_sleep();
3968 while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) {
3969 msleep(MAX_KIQ_REG_BAILOUT_INTERVAL);
3970 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT);
3971 }
3972
3973 if (cnt > MAX_KIQ_REG_TRY)
3974 goto failed_kiq_read;
3975
3976 mb();
3977 value = (uint64_t)adev->wb.wb[reg_val_offs] |
3978 (uint64_t)adev->wb.wb[reg_val_offs + 1 ] << 32ULL;
3979 amdgpu_device_wb_free(adev, reg_val_offs);
3980 return value;
3981
3982 failed_undo:
3983 amdgpu_ring_undo(ring);
3984 failed_unlock:
3985 spin_unlock_irqrestore(&kiq->ring_lock, flags);
3986 failed_kiq_read:
3987 if (reg_val_offs)
3988 amdgpu_device_wb_free(adev, reg_val_offs);
3989 pr_err("failed to read gpu clock\n");
3990 return ~0;
3991 }
3992
gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device * adev)3993 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev)
3994 {
3995 uint64_t clock, clock_lo, clock_hi, hi_check;
3996
3997 switch (adev->ip_versions[GC_HWIP][0]) {
3998 case IP_VERSION(9, 3, 0):
3999 preempt_disable();
4000 clock_hi = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir);
4001 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir);
4002 hi_check = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir);
4003 /* The SMUIO TSC clock frequency is 100MHz, which sets 32-bit carry over
4004 * roughly every 42 seconds.
4005 */
4006 if (hi_check != clock_hi) {
4007 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir);
4008 clock_hi = hi_check;
4009 }
4010 preempt_enable();
4011 clock = clock_lo | (clock_hi << 32ULL);
4012 break;
4013 default:
4014 amdgpu_gfx_off_ctrl(adev, false);
4015 mutex_lock(&adev->gfx.gpu_clock_mutex);
4016 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 0, 1) && amdgpu_sriov_runtime(adev)) {
4017 clock = gfx_v9_0_kiq_read_clock(adev);
4018 } else {
4019 WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4020 clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB) |
4021 ((uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4022 }
4023 mutex_unlock(&adev->gfx.gpu_clock_mutex);
4024 amdgpu_gfx_off_ctrl(adev, true);
4025 break;
4026 }
4027 return clock;
4028 }
4029
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)4030 static void gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4031 uint32_t vmid,
4032 uint32_t gds_base, uint32_t gds_size,
4033 uint32_t gws_base, uint32_t gws_size,
4034 uint32_t oa_base, uint32_t oa_size)
4035 {
4036 struct amdgpu_device *adev = ring->adev;
4037
4038 /* GDS Base */
4039 gfx_v9_0_write_data_to_reg(ring, 0, false,
4040 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_BASE) + 2 * vmid,
4041 gds_base);
4042
4043 /* GDS Size */
4044 gfx_v9_0_write_data_to_reg(ring, 0, false,
4045 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE) + 2 * vmid,
4046 gds_size);
4047
4048 /* GWS */
4049 gfx_v9_0_write_data_to_reg(ring, 0, false,
4050 SOC15_REG_OFFSET(GC, 0, mmGDS_GWS_VMID0) + vmid,
4051 gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4052
4053 /* OA */
4054 gfx_v9_0_write_data_to_reg(ring, 0, false,
4055 SOC15_REG_OFFSET(GC, 0, mmGDS_OA_VMID0) + vmid,
4056 (1 << (oa_size + oa_base)) - (1 << oa_base));
4057 }
4058
4059 static const u32 vgpr_init_compute_shader[] =
4060 {
4061 0xb07c0000, 0xbe8000ff,
4062 0x000000f8, 0xbf110800,
4063 0x7e000280, 0x7e020280,
4064 0x7e040280, 0x7e060280,
4065 0x7e080280, 0x7e0a0280,
4066 0x7e0c0280, 0x7e0e0280,
4067 0x80808800, 0xbe803200,
4068 0xbf84fff5, 0xbf9c0000,
4069 0xd28c0001, 0x0001007f,
4070 0xd28d0001, 0x0002027e,
4071 0x10020288, 0xb8810904,
4072 0xb7814000, 0xd1196a01,
4073 0x00000301, 0xbe800087,
4074 0xbefc00c1, 0xd89c4000,
4075 0x00020201, 0xd89cc080,
4076 0x00040401, 0x320202ff,
4077 0x00000800, 0x80808100,
4078 0xbf84fff8, 0x7e020280,
4079 0xbf810000, 0x00000000,
4080 };
4081
4082 static const u32 sgpr_init_compute_shader[] =
4083 {
4084 0xb07c0000, 0xbe8000ff,
4085 0x0000005f, 0xbee50080,
4086 0xbe812c65, 0xbe822c65,
4087 0xbe832c65, 0xbe842c65,
4088 0xbe852c65, 0xb77c0005,
4089 0x80808500, 0xbf84fff8,
4090 0xbe800080, 0xbf810000,
4091 };
4092
4093 static const u32 vgpr_init_compute_shader_arcturus[] = {
4094 0xd3d94000, 0x18000080, 0xd3d94001, 0x18000080, 0xd3d94002, 0x18000080,
4095 0xd3d94003, 0x18000080, 0xd3d94004, 0x18000080, 0xd3d94005, 0x18000080,
4096 0xd3d94006, 0x18000080, 0xd3d94007, 0x18000080, 0xd3d94008, 0x18000080,
4097 0xd3d94009, 0x18000080, 0xd3d9400a, 0x18000080, 0xd3d9400b, 0x18000080,
4098 0xd3d9400c, 0x18000080, 0xd3d9400d, 0x18000080, 0xd3d9400e, 0x18000080,
4099 0xd3d9400f, 0x18000080, 0xd3d94010, 0x18000080, 0xd3d94011, 0x18000080,
4100 0xd3d94012, 0x18000080, 0xd3d94013, 0x18000080, 0xd3d94014, 0x18000080,
4101 0xd3d94015, 0x18000080, 0xd3d94016, 0x18000080, 0xd3d94017, 0x18000080,
4102 0xd3d94018, 0x18000080, 0xd3d94019, 0x18000080, 0xd3d9401a, 0x18000080,
4103 0xd3d9401b, 0x18000080, 0xd3d9401c, 0x18000080, 0xd3d9401d, 0x18000080,
4104 0xd3d9401e, 0x18000080, 0xd3d9401f, 0x18000080, 0xd3d94020, 0x18000080,
4105 0xd3d94021, 0x18000080, 0xd3d94022, 0x18000080, 0xd3d94023, 0x18000080,
4106 0xd3d94024, 0x18000080, 0xd3d94025, 0x18000080, 0xd3d94026, 0x18000080,
4107 0xd3d94027, 0x18000080, 0xd3d94028, 0x18000080, 0xd3d94029, 0x18000080,
4108 0xd3d9402a, 0x18000080, 0xd3d9402b, 0x18000080, 0xd3d9402c, 0x18000080,
4109 0xd3d9402d, 0x18000080, 0xd3d9402e, 0x18000080, 0xd3d9402f, 0x18000080,
4110 0xd3d94030, 0x18000080, 0xd3d94031, 0x18000080, 0xd3d94032, 0x18000080,
4111 0xd3d94033, 0x18000080, 0xd3d94034, 0x18000080, 0xd3d94035, 0x18000080,
4112 0xd3d94036, 0x18000080, 0xd3d94037, 0x18000080, 0xd3d94038, 0x18000080,
4113 0xd3d94039, 0x18000080, 0xd3d9403a, 0x18000080, 0xd3d9403b, 0x18000080,
4114 0xd3d9403c, 0x18000080, 0xd3d9403d, 0x18000080, 0xd3d9403e, 0x18000080,
4115 0xd3d9403f, 0x18000080, 0xd3d94040, 0x18000080, 0xd3d94041, 0x18000080,
4116 0xd3d94042, 0x18000080, 0xd3d94043, 0x18000080, 0xd3d94044, 0x18000080,
4117 0xd3d94045, 0x18000080, 0xd3d94046, 0x18000080, 0xd3d94047, 0x18000080,
4118 0xd3d94048, 0x18000080, 0xd3d94049, 0x18000080, 0xd3d9404a, 0x18000080,
4119 0xd3d9404b, 0x18000080, 0xd3d9404c, 0x18000080, 0xd3d9404d, 0x18000080,
4120 0xd3d9404e, 0x18000080, 0xd3d9404f, 0x18000080, 0xd3d94050, 0x18000080,
4121 0xd3d94051, 0x18000080, 0xd3d94052, 0x18000080, 0xd3d94053, 0x18000080,
4122 0xd3d94054, 0x18000080, 0xd3d94055, 0x18000080, 0xd3d94056, 0x18000080,
4123 0xd3d94057, 0x18000080, 0xd3d94058, 0x18000080, 0xd3d94059, 0x18000080,
4124 0xd3d9405a, 0x18000080, 0xd3d9405b, 0x18000080, 0xd3d9405c, 0x18000080,
4125 0xd3d9405d, 0x18000080, 0xd3d9405e, 0x18000080, 0xd3d9405f, 0x18000080,
4126 0xd3d94060, 0x18000080, 0xd3d94061, 0x18000080, 0xd3d94062, 0x18000080,
4127 0xd3d94063, 0x18000080, 0xd3d94064, 0x18000080, 0xd3d94065, 0x18000080,
4128 0xd3d94066, 0x18000080, 0xd3d94067, 0x18000080, 0xd3d94068, 0x18000080,
4129 0xd3d94069, 0x18000080, 0xd3d9406a, 0x18000080, 0xd3d9406b, 0x18000080,
4130 0xd3d9406c, 0x18000080, 0xd3d9406d, 0x18000080, 0xd3d9406e, 0x18000080,
4131 0xd3d9406f, 0x18000080, 0xd3d94070, 0x18000080, 0xd3d94071, 0x18000080,
4132 0xd3d94072, 0x18000080, 0xd3d94073, 0x18000080, 0xd3d94074, 0x18000080,
4133 0xd3d94075, 0x18000080, 0xd3d94076, 0x18000080, 0xd3d94077, 0x18000080,
4134 0xd3d94078, 0x18000080, 0xd3d94079, 0x18000080, 0xd3d9407a, 0x18000080,
4135 0xd3d9407b, 0x18000080, 0xd3d9407c, 0x18000080, 0xd3d9407d, 0x18000080,
4136 0xd3d9407e, 0x18000080, 0xd3d9407f, 0x18000080, 0xd3d94080, 0x18000080,
4137 0xd3d94081, 0x18000080, 0xd3d94082, 0x18000080, 0xd3d94083, 0x18000080,
4138 0xd3d94084, 0x18000080, 0xd3d94085, 0x18000080, 0xd3d94086, 0x18000080,
4139 0xd3d94087, 0x18000080, 0xd3d94088, 0x18000080, 0xd3d94089, 0x18000080,
4140 0xd3d9408a, 0x18000080, 0xd3d9408b, 0x18000080, 0xd3d9408c, 0x18000080,
4141 0xd3d9408d, 0x18000080, 0xd3d9408e, 0x18000080, 0xd3d9408f, 0x18000080,
4142 0xd3d94090, 0x18000080, 0xd3d94091, 0x18000080, 0xd3d94092, 0x18000080,
4143 0xd3d94093, 0x18000080, 0xd3d94094, 0x18000080, 0xd3d94095, 0x18000080,
4144 0xd3d94096, 0x18000080, 0xd3d94097, 0x18000080, 0xd3d94098, 0x18000080,
4145 0xd3d94099, 0x18000080, 0xd3d9409a, 0x18000080, 0xd3d9409b, 0x18000080,
4146 0xd3d9409c, 0x18000080, 0xd3d9409d, 0x18000080, 0xd3d9409e, 0x18000080,
4147 0xd3d9409f, 0x18000080, 0xd3d940a0, 0x18000080, 0xd3d940a1, 0x18000080,
4148 0xd3d940a2, 0x18000080, 0xd3d940a3, 0x18000080, 0xd3d940a4, 0x18000080,
4149 0xd3d940a5, 0x18000080, 0xd3d940a6, 0x18000080, 0xd3d940a7, 0x18000080,
4150 0xd3d940a8, 0x18000080, 0xd3d940a9, 0x18000080, 0xd3d940aa, 0x18000080,
4151 0xd3d940ab, 0x18000080, 0xd3d940ac, 0x18000080, 0xd3d940ad, 0x18000080,
4152 0xd3d940ae, 0x18000080, 0xd3d940af, 0x18000080, 0xd3d940b0, 0x18000080,
4153 0xd3d940b1, 0x18000080, 0xd3d940b2, 0x18000080, 0xd3d940b3, 0x18000080,
4154 0xd3d940b4, 0x18000080, 0xd3d940b5, 0x18000080, 0xd3d940b6, 0x18000080,
4155 0xd3d940b7, 0x18000080, 0xd3d940b8, 0x18000080, 0xd3d940b9, 0x18000080,
4156 0xd3d940ba, 0x18000080, 0xd3d940bb, 0x18000080, 0xd3d940bc, 0x18000080,
4157 0xd3d940bd, 0x18000080, 0xd3d940be, 0x18000080, 0xd3d940bf, 0x18000080,
4158 0xd3d940c0, 0x18000080, 0xd3d940c1, 0x18000080, 0xd3d940c2, 0x18000080,
4159 0xd3d940c3, 0x18000080, 0xd3d940c4, 0x18000080, 0xd3d940c5, 0x18000080,
4160 0xd3d940c6, 0x18000080, 0xd3d940c7, 0x18000080, 0xd3d940c8, 0x18000080,
4161 0xd3d940c9, 0x18000080, 0xd3d940ca, 0x18000080, 0xd3d940cb, 0x18000080,
4162 0xd3d940cc, 0x18000080, 0xd3d940cd, 0x18000080, 0xd3d940ce, 0x18000080,
4163 0xd3d940cf, 0x18000080, 0xd3d940d0, 0x18000080, 0xd3d940d1, 0x18000080,
4164 0xd3d940d2, 0x18000080, 0xd3d940d3, 0x18000080, 0xd3d940d4, 0x18000080,
4165 0xd3d940d5, 0x18000080, 0xd3d940d6, 0x18000080, 0xd3d940d7, 0x18000080,
4166 0xd3d940d8, 0x18000080, 0xd3d940d9, 0x18000080, 0xd3d940da, 0x18000080,
4167 0xd3d940db, 0x18000080, 0xd3d940dc, 0x18000080, 0xd3d940dd, 0x18000080,
4168 0xd3d940de, 0x18000080, 0xd3d940df, 0x18000080, 0xd3d940e0, 0x18000080,
4169 0xd3d940e1, 0x18000080, 0xd3d940e2, 0x18000080, 0xd3d940e3, 0x18000080,
4170 0xd3d940e4, 0x18000080, 0xd3d940e5, 0x18000080, 0xd3d940e6, 0x18000080,
4171 0xd3d940e7, 0x18000080, 0xd3d940e8, 0x18000080, 0xd3d940e9, 0x18000080,
4172 0xd3d940ea, 0x18000080, 0xd3d940eb, 0x18000080, 0xd3d940ec, 0x18000080,
4173 0xd3d940ed, 0x18000080, 0xd3d940ee, 0x18000080, 0xd3d940ef, 0x18000080,
4174 0xd3d940f0, 0x18000080, 0xd3d940f1, 0x18000080, 0xd3d940f2, 0x18000080,
4175 0xd3d940f3, 0x18000080, 0xd3d940f4, 0x18000080, 0xd3d940f5, 0x18000080,
4176 0xd3d940f6, 0x18000080, 0xd3d940f7, 0x18000080, 0xd3d940f8, 0x18000080,
4177 0xd3d940f9, 0x18000080, 0xd3d940fa, 0x18000080, 0xd3d940fb, 0x18000080,
4178 0xd3d940fc, 0x18000080, 0xd3d940fd, 0x18000080, 0xd3d940fe, 0x18000080,
4179 0xd3d940ff, 0x18000080, 0xb07c0000, 0xbe8a00ff, 0x000000f8, 0xbf11080a,
4180 0x7e000280, 0x7e020280, 0x7e040280, 0x7e060280, 0x7e080280, 0x7e0a0280,
4181 0x7e0c0280, 0x7e0e0280, 0x808a880a, 0xbe80320a, 0xbf84fff5, 0xbf9c0000,
4182 0xd28c0001, 0x0001007f, 0xd28d0001, 0x0002027e, 0x10020288, 0xb88b0904,
4183 0xb78b4000, 0xd1196a01, 0x00001701, 0xbe8a0087, 0xbefc00c1, 0xd89c4000,
4184 0x00020201, 0xd89cc080, 0x00040401, 0x320202ff, 0x00000800, 0x808a810a,
4185 0xbf84fff8, 0xbf810000,
4186 };
4187
4188 /* When below register arrays changed, please update gpr_reg_size,
4189 and sec_ded_counter_reg_size in function gfx_v9_0_do_edc_gpr_workarounds,
4190 to cover all gfx9 ASICs */
4191 static const struct soc15_reg_entry vgpr_init_regs[] = {
4192 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4193 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4194 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 },
4195 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4196 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x3f },
4197 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */
4198 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
4199 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
4200 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
4201 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
4202 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff },
4203 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff },
4204 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff },
4205 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff },
4206 };
4207
4208 static const struct soc15_reg_entry vgpr_init_regs_arcturus[] = {
4209 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4210 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4211 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 },
4212 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4213 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0xbf },
4214 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */
4215 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff },
4216 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff },
4217 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff },
4218 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff },
4219 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff },
4220 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff },
4221 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff },
4222 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff },
4223 };
4224
4225 static const struct soc15_reg_entry sgpr1_init_regs[] = {
4226 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4227 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4228 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 },
4229 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4230 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */
4231 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
4232 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x000000ff },
4233 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x000000ff },
4234 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x000000ff },
4235 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x000000ff },
4236 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x000000ff },
4237 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x000000ff },
4238 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x000000ff },
4239 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x000000ff },
4240 };
4241
4242 static const struct soc15_reg_entry sgpr2_init_regs[] = {
4243 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 },
4244 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 },
4245 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 },
4246 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 },
4247 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */
4248 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 },
4249 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x0000ff00 },
4250 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x0000ff00 },
4251 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x0000ff00 },
4252 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x0000ff00 },
4253 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x0000ff00 },
4254 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x0000ff00 },
4255 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x0000ff00 },
4256 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x0000ff00 },
4257 };
4258
4259 static const struct soc15_reg_entry gfx_v9_0_edc_counter_regs[] = {
4260 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT), 0, 1, 1},
4261 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT), 0, 1, 1},
4262 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 0, 1, 1},
4263 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT), 0, 1, 1},
4264 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 0, 1, 1},
4265 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT), 0, 1, 1},
4266 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT), 0, 1, 1},
4267 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT), 0, 1, 1},
4268 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT), 0, 1, 1},
4269 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 0, 1, 1},
4270 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_GRBM_CNT), 0, 1, 1},
4271 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_DED), 0, 1, 1},
4272 { SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT), 0, 4, 1},
4273 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 0, 4, 6},
4274 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_DED_CNT), 0, 4, 16},
4275 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_INFO), 0, 4, 16},
4276 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_SEC_CNT), 0, 4, 16},
4277 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 0, 1, 16},
4278 { SOC15_REG_ENTRY(GC, 0, mmTCP_ATC_EDC_GATCL1_CNT), 0, 4, 16},
4279 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT), 0, 4, 16},
4280 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 0, 4, 16},
4281 { SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 0, 4, 16},
4282 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 0, 4, 6},
4283 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 0, 4, 16},
4284 { SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 0, 4, 16},
4285 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 0, 1, 1},
4286 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 0, 1, 1},
4287 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 0, 1, 32},
4288 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 0, 1, 32},
4289 { SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT), 0, 1, 72},
4290 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 0, 1, 16},
4291 { SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 0, 1, 2},
4292 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 0, 4, 6},
4293 };
4294
gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device * adev)4295 static int gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device *adev)
4296 {
4297 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4298 int i, r;
4299
4300 /* only support when RAS is enabled */
4301 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4302 return 0;
4303
4304 r = amdgpu_ring_alloc(ring, 7);
4305 if (r) {
4306 DRM_ERROR("amdgpu: GDS workarounds failed to lock ring %s (%d).\n",
4307 ring->name, r);
4308 return r;
4309 }
4310
4311 WREG32_SOC15(GC, 0, mmGDS_VMID0_BASE, 0x00000000);
4312 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, adev->gds.gds_size);
4313
4314 amdgpu_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5));
4315 amdgpu_ring_write(ring, (PACKET3_DMA_DATA_CP_SYNC |
4316 PACKET3_DMA_DATA_DST_SEL(1) |
4317 PACKET3_DMA_DATA_SRC_SEL(2) |
4318 PACKET3_DMA_DATA_ENGINE(0)));
4319 amdgpu_ring_write(ring, 0);
4320 amdgpu_ring_write(ring, 0);
4321 amdgpu_ring_write(ring, 0);
4322 amdgpu_ring_write(ring, 0);
4323 amdgpu_ring_write(ring, PACKET3_DMA_DATA_CMD_RAW_WAIT |
4324 adev->gds.gds_size);
4325
4326 amdgpu_ring_commit(ring);
4327
4328 for (i = 0; i < adev->usec_timeout; i++) {
4329 if (ring->wptr == gfx_v9_0_ring_get_rptr_compute(ring))
4330 break;
4331 udelay(1);
4332 }
4333
4334 if (i >= adev->usec_timeout)
4335 r = -ETIMEDOUT;
4336
4337 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, 0x00000000);
4338
4339 return r;
4340 }
4341
gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device * adev)4342 static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
4343 {
4344 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4345 struct amdgpu_ib ib;
4346 struct dma_fence *f = NULL;
4347 int r, i;
4348 unsigned total_size, vgpr_offset, sgpr_offset;
4349 u64 gpu_addr;
4350
4351 int compute_dim_x = adev->gfx.config.max_shader_engines *
4352 adev->gfx.config.max_cu_per_sh *
4353 adev->gfx.config.max_sh_per_se;
4354 int sgpr_work_group_size = 5;
4355 int gpr_reg_size = adev->gfx.config.max_shader_engines + 6;
4356 int vgpr_init_shader_size;
4357 const u32 *vgpr_init_shader_ptr;
4358 const struct soc15_reg_entry *vgpr_init_regs_ptr;
4359
4360 /* only support when RAS is enabled */
4361 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4362 return 0;
4363
4364 /* bail if the compute ring is not ready */
4365 if (!ring->sched.ready)
4366 return 0;
4367
4368 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1)) {
4369 vgpr_init_shader_ptr = vgpr_init_compute_shader_arcturus;
4370 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader_arcturus);
4371 vgpr_init_regs_ptr = vgpr_init_regs_arcturus;
4372 } else {
4373 vgpr_init_shader_ptr = vgpr_init_compute_shader;
4374 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader);
4375 vgpr_init_regs_ptr = vgpr_init_regs;
4376 }
4377
4378 total_size =
4379 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* VGPRS */
4380 total_size +=
4381 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS1 */
4382 total_size +=
4383 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS2 */
4384 total_size = ALIGN(total_size, 256);
4385 vgpr_offset = total_size;
4386 total_size += ALIGN(vgpr_init_shader_size, 256);
4387 sgpr_offset = total_size;
4388 total_size += sizeof(sgpr_init_compute_shader);
4389
4390 /* allocate an indirect buffer to put the commands in */
4391 memset(&ib, 0, sizeof(ib));
4392 r = amdgpu_ib_get(adev, NULL, total_size,
4393 AMDGPU_IB_POOL_DIRECT, &ib);
4394 if (r) {
4395 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
4396 return r;
4397 }
4398
4399 /* load the compute shaders */
4400 for (i = 0; i < vgpr_init_shader_size/sizeof(u32); i++)
4401 ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_shader_ptr[i];
4402
4403 for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
4404 ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
4405
4406 /* init the ib length to 0 */
4407 ib.length_dw = 0;
4408
4409 /* VGPR */
4410 /* write the register state for the compute dispatch */
4411 for (i = 0; i < gpr_reg_size; i++) {
4412 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4413 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(vgpr_init_regs_ptr[i])
4414 - PACKET3_SET_SH_REG_START;
4415 ib.ptr[ib.length_dw++] = vgpr_init_regs_ptr[i].reg_value;
4416 }
4417 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4418 gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
4419 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4420 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4421 - PACKET3_SET_SH_REG_START;
4422 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4423 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4424
4425 /* write dispatch packet */
4426 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4427 ib.ptr[ib.length_dw++] = compute_dim_x * 2; /* x */
4428 ib.ptr[ib.length_dw++] = 1; /* y */
4429 ib.ptr[ib.length_dw++] = 1; /* z */
4430 ib.ptr[ib.length_dw++] =
4431 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4432
4433 /* write CS partial flush packet */
4434 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4435 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4436
4437 /* SGPR1 */
4438 /* write the register state for the compute dispatch */
4439 for (i = 0; i < gpr_reg_size; i++) {
4440 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4441 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr1_init_regs[i])
4442 - PACKET3_SET_SH_REG_START;
4443 ib.ptr[ib.length_dw++] = sgpr1_init_regs[i].reg_value;
4444 }
4445 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4446 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4447 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4448 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4449 - PACKET3_SET_SH_REG_START;
4450 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4451 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4452
4453 /* write dispatch packet */
4454 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4455 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */
4456 ib.ptr[ib.length_dw++] = 1; /* y */
4457 ib.ptr[ib.length_dw++] = 1; /* z */
4458 ib.ptr[ib.length_dw++] =
4459 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4460
4461 /* write CS partial flush packet */
4462 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4463 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4464
4465 /* SGPR2 */
4466 /* write the register state for the compute dispatch */
4467 for (i = 0; i < gpr_reg_size; i++) {
4468 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
4469 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr2_init_regs[i])
4470 - PACKET3_SET_SH_REG_START;
4471 ib.ptr[ib.length_dw++] = sgpr2_init_regs[i].reg_value;
4472 }
4473 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4474 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4475 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
4476 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)
4477 - PACKET3_SET_SH_REG_START;
4478 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
4479 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
4480
4481 /* write dispatch packet */
4482 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
4483 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */
4484 ib.ptr[ib.length_dw++] = 1; /* y */
4485 ib.ptr[ib.length_dw++] = 1; /* z */
4486 ib.ptr[ib.length_dw++] =
4487 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
4488
4489 /* write CS partial flush packet */
4490 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
4491 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
4492
4493 /* shedule the ib on the ring */
4494 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
4495 if (r) {
4496 DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
4497 goto fail;
4498 }
4499
4500 /* wait for the GPU to finish processing the IB */
4501 r = dma_fence_wait(f, false);
4502 if (r) {
4503 DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
4504 goto fail;
4505 }
4506
4507 fail:
4508 amdgpu_ib_free(adev, &ib, NULL);
4509 dma_fence_put(f);
4510
4511 return r;
4512 }
4513
gfx_v9_0_early_init(void * handle)4514 static int gfx_v9_0_early_init(void *handle)
4515 {
4516 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4517
4518 adev->gfx.funcs = &gfx_v9_0_gfx_funcs;
4519
4520 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1) ||
4521 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
4522 adev->gfx.num_gfx_rings = 0;
4523 else
4524 adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS;
4525 adev->gfx.xcc_mask = 1;
4526 adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
4527 AMDGPU_MAX_COMPUTE_RINGS);
4528 gfx_v9_0_set_kiq_pm4_funcs(adev);
4529 gfx_v9_0_set_ring_funcs(adev);
4530 gfx_v9_0_set_irq_funcs(adev);
4531 gfx_v9_0_set_gds_init(adev);
4532 gfx_v9_0_set_rlc_funcs(adev);
4533
4534 /* init rlcg reg access ctrl */
4535 gfx_v9_0_init_rlcg_reg_access_ctrl(adev);
4536
4537 return gfx_v9_0_init_microcode(adev);
4538 }
4539
gfx_v9_0_ecc_late_init(void * handle)4540 static int gfx_v9_0_ecc_late_init(void *handle)
4541 {
4542 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4543 int r;
4544
4545 /*
4546 * Temp workaround to fix the issue that CP firmware fails to
4547 * update read pointer when CPDMA is writing clearing operation
4548 * to GDS in suspend/resume sequence on several cards. So just
4549 * limit this operation in cold boot sequence.
4550 */
4551 if ((!adev->in_suspend) &&
4552 (adev->gds.gds_size)) {
4553 r = gfx_v9_0_do_edc_gds_workarounds(adev);
4554 if (r)
4555 return r;
4556 }
4557
4558 /* requires IBs so do in late init after IB pool is initialized */
4559 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
4560 r = gfx_v9_4_2_do_edc_gpr_workarounds(adev);
4561 else
4562 r = gfx_v9_0_do_edc_gpr_workarounds(adev);
4563
4564 if (r)
4565 return r;
4566
4567 if (adev->gfx.ras &&
4568 adev->gfx.ras->enable_watchdog_timer)
4569 adev->gfx.ras->enable_watchdog_timer(adev);
4570
4571 return 0;
4572 }
4573
gfx_v9_0_late_init(void * handle)4574 static int gfx_v9_0_late_init(void *handle)
4575 {
4576 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4577 int r;
4578
4579 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4580 if (r)
4581 return r;
4582
4583 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4584 if (r)
4585 return r;
4586
4587 r = gfx_v9_0_ecc_late_init(handle);
4588 if (r)
4589 return r;
4590
4591 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2))
4592 gfx_v9_4_2_debug_trap_config_init(adev,
4593 adev->vm_manager.first_kfd_vmid, AMDGPU_NUM_VMID);
4594 else
4595 gfx_v9_0_debug_trap_config_init(adev,
4596 adev->vm_manager.first_kfd_vmid, AMDGPU_NUM_VMID);
4597
4598 return 0;
4599 }
4600
gfx_v9_0_is_rlc_enabled(struct amdgpu_device * adev)4601 static bool gfx_v9_0_is_rlc_enabled(struct amdgpu_device *adev)
4602 {
4603 uint32_t rlc_setting;
4604
4605 /* if RLC is not enabled, do nothing */
4606 rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL);
4607 if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
4608 return false;
4609
4610 return true;
4611 }
4612
gfx_v9_0_set_safe_mode(struct amdgpu_device * adev,int xcc_id)4613 static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id)
4614 {
4615 uint32_t data;
4616 unsigned i;
4617
4618 data = RLC_SAFE_MODE__CMD_MASK;
4619 data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
4620 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4621
4622 /* wait for RLC_SAFE_MODE */
4623 for (i = 0; i < adev->usec_timeout; i++) {
4624 if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
4625 break;
4626 udelay(1);
4627 }
4628 }
4629
gfx_v9_0_unset_safe_mode(struct amdgpu_device * adev,int xcc_id)4630 static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id)
4631 {
4632 uint32_t data;
4633
4634 data = RLC_SAFE_MODE__CMD_MASK;
4635 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4636 }
4637
gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device * adev,bool enable)4638 static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev,
4639 bool enable)
4640 {
4641 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4642
4643 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
4644 gfx_v9_0_enable_gfx_cg_power_gating(adev, true);
4645 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4646 gfx_v9_0_enable_gfx_pipeline_powergating(adev, true);
4647 } else {
4648 gfx_v9_0_enable_gfx_cg_power_gating(adev, false);
4649 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
4650 gfx_v9_0_enable_gfx_pipeline_powergating(adev, false);
4651 }
4652
4653 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4654 }
4655
gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device * adev,bool enable)4656 static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev,
4657 bool enable)
4658 {
4659 /* TODO: double check if we need to perform under safe mode */
4660 /* gfx_v9_0_enter_rlc_safe_mode(adev); */
4661
4662 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
4663 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, true);
4664 else
4665 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, false);
4666
4667 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
4668 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, true);
4669 else
4670 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, false);
4671
4672 /* gfx_v9_0_exit_rlc_safe_mode(adev); */
4673 }
4674
gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device * adev,bool enable)4675 static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
4676 bool enable)
4677 {
4678 uint32_t data, def;
4679
4680 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4681
4682 /* It is disabled by HW by default */
4683 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
4684 /* 1 - RLC_CGTT_MGCG_OVERRIDE */
4685 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4686
4687 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1))
4688 data &= ~RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
4689
4690 data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4691 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4692 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4693
4694 /* only for Vega10 & Raven1 */
4695 data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK;
4696
4697 if (def != data)
4698 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4699
4700 /* MGLS is a global flag to control all MGLS in GFX */
4701 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
4702 /* 2 - RLC memory Light sleep */
4703 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
4704 def = data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4705 data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4706 if (def != data)
4707 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4708 }
4709 /* 3 - CP memory Light sleep */
4710 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
4711 def = data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4712 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4713 if (def != data)
4714 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4715 }
4716 }
4717 } else {
4718 /* 1 - MGCG_OVERRIDE */
4719 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4720
4721 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1))
4722 data |= RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK;
4723
4724 data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4725 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4726 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4727 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4728
4729 if (def != data)
4730 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4731
4732 /* 2 - disable MGLS in RLC */
4733 data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4734 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
4735 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4736 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4737 }
4738
4739 /* 3 - disable MGLS in CP */
4740 data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4741 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
4742 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4743 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4744 }
4745 }
4746
4747 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4748 }
4749
gfx_v9_0_update_3d_clock_gating(struct amdgpu_device * adev,bool enable)4750 static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev,
4751 bool enable)
4752 {
4753 uint32_t data, def;
4754
4755 if (!adev->gfx.num_gfx_rings)
4756 return;
4757
4758 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4759
4760 /* Enable 3D CGCG/CGLS */
4761 if (enable) {
4762 /* write cmd to clear cgcg/cgls ov */
4763 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4764 /* unset CGCG override */
4765 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK;
4766 /* update CGCG and CGLS override bits */
4767 if (def != data)
4768 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4769
4770 /* enable 3Dcgcg FSM(0x0000363f) */
4771 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4772
4773 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)
4774 data = (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4775 RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
4776 else
4777 data = 0x0 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT;
4778
4779 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
4780 data |= (0x000F << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4781 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
4782 if (def != data)
4783 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
4784
4785 /* set IDLE_POLL_COUNT(0x00900100) */
4786 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
4787 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4788 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4789 if (def != data)
4790 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
4791 } else {
4792 /* Disable CGCG/CGLS */
4793 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4794 /* disable cgcg, cgls should be disabled */
4795 data &= ~(RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK |
4796 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK);
4797 /* disable cgcg and cgls in FSM */
4798 if (def != data)
4799 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
4800 }
4801
4802 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4803 }
4804
gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device * adev,bool enable)4805 static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
4806 bool enable)
4807 {
4808 uint32_t def, data;
4809
4810 amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4811
4812 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
4813 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4814 /* unset CGCG override */
4815 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
4816 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4817 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4818 else
4819 data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4820 /* update CGCG and CGLS override bits */
4821 if (def != data)
4822 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4823
4824 /* enable cgcg FSM(0x0000363F) */
4825 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4826
4827 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1))
4828 data = (0x2000 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4829 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4830 else
4831 data = (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4832 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4833 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4834 data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4835 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
4836 if (def != data)
4837 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
4838
4839 /* set IDLE_POLL_COUNT(0x00900100) */
4840 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
4841 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4842 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4843 if (def != data)
4844 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
4845 } else {
4846 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4847 /* reset CGCG/CGLS bits */
4848 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
4849 /* disable cgcg and cgls in FSM */
4850 if (def != data)
4851 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
4852 }
4853
4854 amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4855 }
4856
gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device * adev,bool enable)4857 static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev,
4858 bool enable)
4859 {
4860 if (enable) {
4861 /* CGCG/CGLS should be enabled after MGCG/MGLS
4862 * === MGCG + MGLS ===
4863 */
4864 gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
4865 /* === CGCG /CGLS for GFX 3D Only === */
4866 gfx_v9_0_update_3d_clock_gating(adev, enable);
4867 /* === CGCG + CGLS === */
4868 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
4869 } else {
4870 /* CGCG/CGLS should be disabled before MGCG/MGLS
4871 * === CGCG + CGLS ===
4872 */
4873 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable);
4874 /* === CGCG /CGLS for GFX 3D Only === */
4875 gfx_v9_0_update_3d_clock_gating(adev, enable);
4876 /* === MGCG + MGLS === */
4877 gfx_v9_0_update_medium_grain_clock_gating(adev, enable);
4878 }
4879 return 0;
4880 }
4881
gfx_v9_0_update_spm_vmid_internal(struct amdgpu_device * adev,unsigned int vmid)4882 static void gfx_v9_0_update_spm_vmid_internal(struct amdgpu_device *adev,
4883 unsigned int vmid)
4884 {
4885 u32 reg, data;
4886
4887 reg = SOC15_REG_OFFSET(GC, 0, mmRLC_SPM_MC_CNTL);
4888 if (amdgpu_sriov_is_pp_one_vf(adev))
4889 data = RREG32_NO_KIQ(reg);
4890 else
4891 data = RREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL);
4892
4893 data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK;
4894 data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT;
4895
4896 if (amdgpu_sriov_is_pp_one_vf(adev))
4897 WREG32_SOC15_NO_KIQ(GC, 0, mmRLC_SPM_MC_CNTL, data);
4898 else
4899 WREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL, data);
4900 }
4901
gfx_v9_0_update_spm_vmid(struct amdgpu_device * adev,unsigned int vmid)4902 static void gfx_v9_0_update_spm_vmid(struct amdgpu_device *adev, unsigned int vmid)
4903 {
4904 amdgpu_gfx_off_ctrl(adev, false);
4905
4906 gfx_v9_0_update_spm_vmid_internal(adev, vmid);
4907
4908 amdgpu_gfx_off_ctrl(adev, true);
4909 }
4910
gfx_v9_0_check_rlcg_range(struct amdgpu_device * adev,uint32_t offset,struct soc15_reg_rlcg * entries,int arr_size)4911 static bool gfx_v9_0_check_rlcg_range(struct amdgpu_device *adev,
4912 uint32_t offset,
4913 struct soc15_reg_rlcg *entries, int arr_size)
4914 {
4915 int i;
4916 uint32_t reg;
4917
4918 if (!entries)
4919 return false;
4920
4921 for (i = 0; i < arr_size; i++) {
4922 const struct soc15_reg_rlcg *entry;
4923
4924 entry = &entries[i];
4925 reg = adev->reg_offset[entry->hwip][entry->instance][entry->segment] + entry->reg;
4926 if (offset == reg)
4927 return true;
4928 }
4929
4930 return false;
4931 }
4932
gfx_v9_0_is_rlcg_access_range(struct amdgpu_device * adev,u32 offset)4933 static bool gfx_v9_0_is_rlcg_access_range(struct amdgpu_device *adev, u32 offset)
4934 {
4935 return gfx_v9_0_check_rlcg_range(adev, offset,
4936 (void *)rlcg_access_gc_9_0,
4937 ARRAY_SIZE(rlcg_access_gc_9_0));
4938 }
4939
4940 static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = {
4941 .is_rlc_enabled = gfx_v9_0_is_rlc_enabled,
4942 .set_safe_mode = gfx_v9_0_set_safe_mode,
4943 .unset_safe_mode = gfx_v9_0_unset_safe_mode,
4944 .init = gfx_v9_0_rlc_init,
4945 .get_csb_size = gfx_v9_0_get_csb_size,
4946 .get_csb_buffer = gfx_v9_0_get_csb_buffer,
4947 .get_cp_table_num = gfx_v9_0_cp_jump_table_num,
4948 .resume = gfx_v9_0_rlc_resume,
4949 .stop = gfx_v9_0_rlc_stop,
4950 .reset = gfx_v9_0_rlc_reset,
4951 .start = gfx_v9_0_rlc_start,
4952 .update_spm_vmid = gfx_v9_0_update_spm_vmid,
4953 .is_rlcg_access_range = gfx_v9_0_is_rlcg_access_range,
4954 };
4955
gfx_v9_0_set_powergating_state(void * handle,enum amd_powergating_state state)4956 static int gfx_v9_0_set_powergating_state(void *handle,
4957 enum amd_powergating_state state)
4958 {
4959 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4960 bool enable = (state == AMD_PG_STATE_GATE);
4961
4962 switch (adev->ip_versions[GC_HWIP][0]) {
4963 case IP_VERSION(9, 2, 2):
4964 case IP_VERSION(9, 1, 0):
4965 case IP_VERSION(9, 3, 0):
4966 if (!enable)
4967 amdgpu_gfx_off_ctrl(adev, false);
4968
4969 if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
4970 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, true);
4971 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, true);
4972 } else {
4973 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, false);
4974 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, false);
4975 }
4976
4977 if (adev->pg_flags & AMD_PG_SUPPORT_CP)
4978 gfx_v9_0_enable_cp_power_gating(adev, true);
4979 else
4980 gfx_v9_0_enable_cp_power_gating(adev, false);
4981
4982 /* update gfx cgpg state */
4983 gfx_v9_0_update_gfx_cg_power_gating(adev, enable);
4984
4985 /* update mgcg state */
4986 gfx_v9_0_update_gfx_mg_power_gating(adev, enable);
4987
4988 if (enable)
4989 amdgpu_gfx_off_ctrl(adev, true);
4990 break;
4991 case IP_VERSION(9, 2, 1):
4992 amdgpu_gfx_off_ctrl(adev, enable);
4993 break;
4994 default:
4995 break;
4996 }
4997
4998 return 0;
4999 }
5000
gfx_v9_0_set_clockgating_state(void * handle,enum amd_clockgating_state state)5001 static int gfx_v9_0_set_clockgating_state(void *handle,
5002 enum amd_clockgating_state state)
5003 {
5004 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5005
5006 if (amdgpu_sriov_vf(adev))
5007 return 0;
5008
5009 switch (adev->ip_versions[GC_HWIP][0]) {
5010 case IP_VERSION(9, 0, 1):
5011 case IP_VERSION(9, 2, 1):
5012 case IP_VERSION(9, 4, 0):
5013 case IP_VERSION(9, 2, 2):
5014 case IP_VERSION(9, 1, 0):
5015 case IP_VERSION(9, 4, 1):
5016 case IP_VERSION(9, 3, 0):
5017 case IP_VERSION(9, 4, 2):
5018 gfx_v9_0_update_gfx_clock_gating(adev,
5019 state == AMD_CG_STATE_GATE);
5020 break;
5021 default:
5022 break;
5023 }
5024 return 0;
5025 }
5026
gfx_v9_0_get_clockgating_state(void * handle,u64 * flags)5027 static void gfx_v9_0_get_clockgating_state(void *handle, u64 *flags)
5028 {
5029 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5030 int data;
5031
5032 if (amdgpu_sriov_vf(adev))
5033 *flags = 0;
5034
5035 /* AMD_CG_SUPPORT_GFX_MGCG */
5036 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE));
5037 if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
5038 *flags |= AMD_CG_SUPPORT_GFX_MGCG;
5039
5040 /* AMD_CG_SUPPORT_GFX_CGCG */
5041 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL));
5042 if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5043 *flags |= AMD_CG_SUPPORT_GFX_CGCG;
5044
5045 /* AMD_CG_SUPPORT_GFX_CGLS */
5046 if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5047 *flags |= AMD_CG_SUPPORT_GFX_CGLS;
5048
5049 /* AMD_CG_SUPPORT_GFX_RLC_LS */
5050 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_MEM_SLP_CNTL));
5051 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5052 *flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5053
5054 /* AMD_CG_SUPPORT_GFX_CP_LS */
5055 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmCP_MEM_SLP_CNTL));
5056 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5057 *flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5058
5059 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)) {
5060 /* AMD_CG_SUPPORT_GFX_3D_CGCG */
5061 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D));
5062 if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK)
5063 *flags |= AMD_CG_SUPPORT_GFX_3D_CGCG;
5064
5065 /* AMD_CG_SUPPORT_GFX_3D_CGLS */
5066 if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK)
5067 *flags |= AMD_CG_SUPPORT_GFX_3D_CGLS;
5068 }
5069 }
5070
gfx_v9_0_ring_get_rptr_gfx(struct amdgpu_ring * ring)5071 static u64 gfx_v9_0_ring_get_rptr_gfx(struct amdgpu_ring *ring)
5072 {
5073 return *ring->rptr_cpu_addr; /* gfx9 is 32bit rptr*/
5074 }
5075
gfx_v9_0_ring_get_wptr_gfx(struct amdgpu_ring * ring)5076 static u64 gfx_v9_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
5077 {
5078 struct amdgpu_device *adev = ring->adev;
5079 u64 wptr;
5080
5081 /* XXX check if swapping is necessary on BE */
5082 if (ring->use_doorbell) {
5083 wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5084 } else {
5085 wptr = RREG32_SOC15(GC, 0, mmCP_RB0_WPTR);
5086 wptr += (u64)RREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI) << 32;
5087 }
5088
5089 return wptr;
5090 }
5091
gfx_v9_0_ring_set_wptr_gfx(struct amdgpu_ring * ring)5092 static void gfx_v9_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
5093 {
5094 struct amdgpu_device *adev = ring->adev;
5095
5096 if (ring->use_doorbell) {
5097 /* XXX check if swapping is necessary on BE */
5098 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr);
5099 WDOORBELL64(ring->doorbell_index, ring->wptr);
5100 } else {
5101 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
5102 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
5103 }
5104 }
5105
gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring * ring)5106 static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
5107 {
5108 struct amdgpu_device *adev = ring->adev;
5109 u32 ref_and_mask, reg_mem_engine;
5110 const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
5111
5112 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
5113 switch (ring->me) {
5114 case 1:
5115 ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
5116 break;
5117 case 2:
5118 ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
5119 break;
5120 default:
5121 return;
5122 }
5123 reg_mem_engine = 0;
5124 } else {
5125 ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
5126 reg_mem_engine = 1; /* pfp */
5127 }
5128
5129 gfx_v9_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
5130 adev->nbio.funcs->get_hdp_flush_req_offset(adev),
5131 adev->nbio.funcs->get_hdp_flush_done_offset(adev),
5132 ref_and_mask, ref_and_mask, 0x20);
5133 }
5134
gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)5135 static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
5136 struct amdgpu_job *job,
5137 struct amdgpu_ib *ib,
5138 uint32_t flags)
5139 {
5140 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5141 u32 header, control = 0;
5142
5143 if (ib->flags & AMDGPU_IB_FLAG_CE)
5144 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
5145 else
5146 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
5147
5148 control |= ib->length_dw | (vmid << 24);
5149
5150 if (ib->flags & AMDGPU_IB_FLAG_PREEMPT) {
5151 control |= INDIRECT_BUFFER_PRE_ENB(1);
5152
5153 if (flags & AMDGPU_IB_PREEMPTED)
5154 control |= INDIRECT_BUFFER_PRE_RESUME(1);
5155
5156 if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid)
5157 gfx_v9_0_ring_emit_de_meta(ring,
5158 (!amdgpu_sriov_vf(ring->adev) &&
5159 flags & AMDGPU_IB_PREEMPTED) ?
5160 true : false,
5161 job->gds_size > 0 && job->gds_base != 0);
5162 }
5163
5164 amdgpu_ring_write(ring, header);
5165 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5166 amdgpu_ring_write(ring,
5167 #ifdef __BIG_ENDIAN
5168 (2 << 0) |
5169 #endif
5170 lower_32_bits(ib->gpu_addr));
5171 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5172 amdgpu_ring_ib_on_emit_cntl(ring);
5173 amdgpu_ring_write(ring, control);
5174 }
5175
gfx_v9_0_ring_patch_cntl(struct amdgpu_ring * ring,unsigned offset)5176 static void gfx_v9_0_ring_patch_cntl(struct amdgpu_ring *ring,
5177 unsigned offset)
5178 {
5179 u32 control = ring->ring[offset];
5180
5181 control |= INDIRECT_BUFFER_PRE_RESUME(1);
5182 ring->ring[offset] = control;
5183 }
5184
gfx_v9_0_ring_patch_ce_meta(struct amdgpu_ring * ring,unsigned offset)5185 static void gfx_v9_0_ring_patch_ce_meta(struct amdgpu_ring *ring,
5186 unsigned offset)
5187 {
5188 struct amdgpu_device *adev = ring->adev;
5189 void *ce_payload_cpu_addr;
5190 uint64_t payload_offset, payload_size;
5191
5192 payload_size = sizeof(struct v9_ce_ib_state);
5193
5194 if (ring->is_mes_queue) {
5195 payload_offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5196 gfx[0].gfx_meta_data) +
5197 offsetof(struct v9_gfx_meta_data, ce_payload);
5198 ce_payload_cpu_addr =
5199 amdgpu_mes_ctx_get_offs_cpu_addr(ring, payload_offset);
5200 } else {
5201 payload_offset = offsetof(struct v9_gfx_meta_data, ce_payload);
5202 ce_payload_cpu_addr = adev->virt.csa_cpu_addr + payload_offset;
5203 }
5204
5205 if (offset + (payload_size >> 2) <= ring->buf_mask + 1) {
5206 memcpy((void *)&ring->ring[offset], ce_payload_cpu_addr, payload_size);
5207 } else {
5208 memcpy((void *)&ring->ring[offset], ce_payload_cpu_addr,
5209 (ring->buf_mask + 1 - offset) << 2);
5210 payload_size -= (ring->buf_mask + 1 - offset) << 2;
5211 memcpy((void *)&ring->ring[0],
5212 ce_payload_cpu_addr + ((ring->buf_mask + 1 - offset) << 2),
5213 payload_size);
5214 }
5215 }
5216
gfx_v9_0_ring_patch_de_meta(struct amdgpu_ring * ring,unsigned offset)5217 static void gfx_v9_0_ring_patch_de_meta(struct amdgpu_ring *ring,
5218 unsigned offset)
5219 {
5220 struct amdgpu_device *adev = ring->adev;
5221 void *de_payload_cpu_addr;
5222 uint64_t payload_offset, payload_size;
5223
5224 payload_size = sizeof(struct v9_de_ib_state);
5225
5226 if (ring->is_mes_queue) {
5227 payload_offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5228 gfx[0].gfx_meta_data) +
5229 offsetof(struct v9_gfx_meta_data, de_payload);
5230 de_payload_cpu_addr =
5231 amdgpu_mes_ctx_get_offs_cpu_addr(ring, payload_offset);
5232 } else {
5233 payload_offset = offsetof(struct v9_gfx_meta_data, de_payload);
5234 de_payload_cpu_addr = adev->virt.csa_cpu_addr + payload_offset;
5235 }
5236
5237 ((struct v9_de_ib_state *)de_payload_cpu_addr)->ib_completion_status =
5238 IB_COMPLETION_STATUS_PREEMPTED;
5239
5240 if (offset + (payload_size >> 2) <= ring->buf_mask + 1) {
5241 memcpy((void *)&ring->ring[offset], de_payload_cpu_addr, payload_size);
5242 } else {
5243 memcpy((void *)&ring->ring[offset], de_payload_cpu_addr,
5244 (ring->buf_mask + 1 - offset) << 2);
5245 payload_size -= (ring->buf_mask + 1 - offset) << 2;
5246 memcpy((void *)&ring->ring[0],
5247 de_payload_cpu_addr + ((ring->buf_mask + 1 - offset) << 2),
5248 payload_size);
5249 }
5250 }
5251
gfx_v9_0_ring_emit_ib_compute(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)5252 static void gfx_v9_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
5253 struct amdgpu_job *job,
5254 struct amdgpu_ib *ib,
5255 uint32_t flags)
5256 {
5257 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5258 u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
5259
5260 /* Currently, there is a high possibility to get wave ID mismatch
5261 * between ME and GDS, leading to a hw deadlock, because ME generates
5262 * different wave IDs than the GDS expects. This situation happens
5263 * randomly when at least 5 compute pipes use GDS ordered append.
5264 * The wave IDs generated by ME are also wrong after suspend/resume.
5265 * Those are probably bugs somewhere else in the kernel driver.
5266 *
5267 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
5268 * GDS to 0 for this ring (me/pipe).
5269 */
5270 if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
5271 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
5272 amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID);
5273 amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
5274 }
5275
5276 amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
5277 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5278 amdgpu_ring_write(ring,
5279 #ifdef __BIG_ENDIAN
5280 (2 << 0) |
5281 #endif
5282 lower_32_bits(ib->gpu_addr));
5283 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5284 amdgpu_ring_write(ring, control);
5285 }
5286
gfx_v9_0_ring_emit_fence(struct amdgpu_ring * ring,u64 addr,u64 seq,unsigned flags)5287 static void gfx_v9_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
5288 u64 seq, unsigned flags)
5289 {
5290 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
5291 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
5292 bool writeback = flags & AMDGPU_FENCE_FLAG_TC_WB_ONLY;
5293 bool exec = flags & AMDGPU_FENCE_FLAG_EXEC;
5294 uint32_t dw2 = 0;
5295
5296 /* RELEASE_MEM - flush caches, send int */
5297 amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
5298
5299 if (writeback) {
5300 dw2 = EOP_TC_NC_ACTION_EN;
5301 } else {
5302 dw2 = EOP_TCL1_ACTION_EN | EOP_TC_ACTION_EN |
5303 EOP_TC_MD_ACTION_EN;
5304 }
5305 dw2 |= EOP_TC_WB_ACTION_EN | EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
5306 EVENT_INDEX(5);
5307 if (exec)
5308 dw2 |= EOP_EXEC;
5309
5310 amdgpu_ring_write(ring, dw2);
5311 amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
5312
5313 /*
5314 * the address should be Qword aligned if 64bit write, Dword
5315 * aligned if only send 32bit data low (discard data high)
5316 */
5317 if (write64bit)
5318 BUG_ON(addr & 0x7);
5319 else
5320 BUG_ON(addr & 0x3);
5321 amdgpu_ring_write(ring, lower_32_bits(addr));
5322 amdgpu_ring_write(ring, upper_32_bits(addr));
5323 amdgpu_ring_write(ring, lower_32_bits(seq));
5324 amdgpu_ring_write(ring, upper_32_bits(seq));
5325 amdgpu_ring_write(ring, 0);
5326 }
5327
gfx_v9_0_ring_emit_pipeline_sync(struct amdgpu_ring * ring)5328 static void gfx_v9_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
5329 {
5330 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5331 uint32_t seq = ring->fence_drv.sync_seq;
5332 uint64_t addr = ring->fence_drv.gpu_addr;
5333
5334 gfx_v9_0_wait_reg_mem(ring, usepfp, 1, 0,
5335 lower_32_bits(addr), upper_32_bits(addr),
5336 seq, 0xffffffff, 4);
5337 }
5338
gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring * ring,unsigned vmid,uint64_t pd_addr)5339 static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
5340 unsigned vmid, uint64_t pd_addr)
5341 {
5342 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
5343
5344 /* compute doesn't have PFP */
5345 if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
5346 /* sync PFP to ME, otherwise we might get invalid PFP reads */
5347 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
5348 amdgpu_ring_write(ring, 0x0);
5349 }
5350 }
5351
gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring * ring)5352 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring)
5353 {
5354 return *ring->rptr_cpu_addr; /* gfx9 hardware is 32bit rptr */
5355 }
5356
gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring * ring)5357 static u64 gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
5358 {
5359 u64 wptr;
5360
5361 /* XXX check if swapping is necessary on BE */
5362 if (ring->use_doorbell)
5363 wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5364 else
5365 BUG();
5366 return wptr;
5367 }
5368
gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring * ring)5369 static void gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
5370 {
5371 struct amdgpu_device *adev = ring->adev;
5372
5373 /* XXX check if swapping is necessary on BE */
5374 if (ring->use_doorbell) {
5375 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr);
5376 WDOORBELL64(ring->doorbell_index, ring->wptr);
5377 } else{
5378 BUG(); /* only DOORBELL method supported on gfx9 now */
5379 }
5380 }
5381
gfx_v9_0_ring_emit_fence_kiq(struct amdgpu_ring * ring,u64 addr,u64 seq,unsigned int flags)5382 static void gfx_v9_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
5383 u64 seq, unsigned int flags)
5384 {
5385 struct amdgpu_device *adev = ring->adev;
5386
5387 /* we only allocate 32bit for each seq wb address */
5388 BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
5389
5390 /* write fence seq to the "addr" */
5391 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5392 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5393 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
5394 amdgpu_ring_write(ring, lower_32_bits(addr));
5395 amdgpu_ring_write(ring, upper_32_bits(addr));
5396 amdgpu_ring_write(ring, lower_32_bits(seq));
5397
5398 if (flags & AMDGPU_FENCE_FLAG_INT) {
5399 /* set register to trigger INT */
5400 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5401 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5402 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
5403 amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, mmCPC_INT_STATUS));
5404 amdgpu_ring_write(ring, 0);
5405 amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
5406 }
5407 }
5408
gfx_v9_ring_emit_sb(struct amdgpu_ring * ring)5409 static void gfx_v9_ring_emit_sb(struct amdgpu_ring *ring)
5410 {
5411 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
5412 amdgpu_ring_write(ring, 0);
5413 }
5414
gfx_v9_0_ring_emit_ce_meta(struct amdgpu_ring * ring,bool resume)5415 static void gfx_v9_0_ring_emit_ce_meta(struct amdgpu_ring *ring, bool resume)
5416 {
5417 struct amdgpu_device *adev = ring->adev;
5418 struct v9_ce_ib_state ce_payload = {0};
5419 uint64_t offset, ce_payload_gpu_addr;
5420 void *ce_payload_cpu_addr;
5421 int cnt;
5422
5423 cnt = (sizeof(ce_payload) >> 2) + 4 - 2;
5424
5425 if (ring->is_mes_queue) {
5426 offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5427 gfx[0].gfx_meta_data) +
5428 offsetof(struct v9_gfx_meta_data, ce_payload);
5429 ce_payload_gpu_addr =
5430 amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5431 ce_payload_cpu_addr =
5432 amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
5433 } else {
5434 offset = offsetof(struct v9_gfx_meta_data, ce_payload);
5435 ce_payload_gpu_addr = amdgpu_csa_vaddr(ring->adev) + offset;
5436 ce_payload_cpu_addr = adev->virt.csa_cpu_addr + offset;
5437 }
5438
5439 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5440 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
5441 WRITE_DATA_DST_SEL(8) |
5442 WR_CONFIRM) |
5443 WRITE_DATA_CACHE_POLICY(0));
5444 amdgpu_ring_write(ring, lower_32_bits(ce_payload_gpu_addr));
5445 amdgpu_ring_write(ring, upper_32_bits(ce_payload_gpu_addr));
5446
5447 amdgpu_ring_ib_on_emit_ce(ring);
5448
5449 if (resume)
5450 amdgpu_ring_write_multiple(ring, ce_payload_cpu_addr,
5451 sizeof(ce_payload) >> 2);
5452 else
5453 amdgpu_ring_write_multiple(ring, (void *)&ce_payload,
5454 sizeof(ce_payload) >> 2);
5455 }
5456
gfx_v9_0_ring_preempt_ib(struct amdgpu_ring * ring)5457 static int gfx_v9_0_ring_preempt_ib(struct amdgpu_ring *ring)
5458 {
5459 int i, r = 0;
5460 struct amdgpu_device *adev = ring->adev;
5461 struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
5462 struct amdgpu_ring *kiq_ring = &kiq->ring;
5463 unsigned long flags;
5464
5465 if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
5466 return -EINVAL;
5467
5468 spin_lock_irqsave(&kiq->ring_lock, flags);
5469
5470 if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) {
5471 spin_unlock_irqrestore(&kiq->ring_lock, flags);
5472 return -ENOMEM;
5473 }
5474
5475 /* assert preemption condition */
5476 amdgpu_ring_set_preempt_cond_exec(ring, false);
5477
5478 ring->trail_seq += 1;
5479 amdgpu_ring_alloc(ring, 13);
5480 gfx_v9_0_ring_emit_fence(ring, ring->trail_fence_gpu_addr,
5481 ring->trail_seq, AMDGPU_FENCE_FLAG_EXEC | AMDGPU_FENCE_FLAG_INT);
5482
5483 /* assert IB preemption, emit the trailing fence */
5484 kiq->pmf->kiq_unmap_queues(kiq_ring, ring, PREEMPT_QUEUES_NO_UNMAP,
5485 ring->trail_fence_gpu_addr,
5486 ring->trail_seq);
5487
5488 amdgpu_ring_commit(kiq_ring);
5489 spin_unlock_irqrestore(&kiq->ring_lock, flags);
5490
5491 /* poll the trailing fence */
5492 for (i = 0; i < adev->usec_timeout; i++) {
5493 if (ring->trail_seq ==
5494 le32_to_cpu(*ring->trail_fence_cpu_addr))
5495 break;
5496 udelay(1);
5497 }
5498
5499 if (i >= adev->usec_timeout) {
5500 r = -EINVAL;
5501 DRM_WARN("ring %d timeout to preempt ib\n", ring->idx);
5502 }
5503
5504 /*reset the CP_VMID_PREEMPT after trailing fence*/
5505 amdgpu_ring_emit_wreg(ring,
5506 SOC15_REG_OFFSET(GC, 0, mmCP_VMID_PREEMPT),
5507 0x0);
5508 amdgpu_ring_commit(ring);
5509
5510 /* deassert preemption condition */
5511 amdgpu_ring_set_preempt_cond_exec(ring, true);
5512 return r;
5513 }
5514
gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring * ring,bool resume,bool usegds)5515 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume, bool usegds)
5516 {
5517 struct amdgpu_device *adev = ring->adev;
5518 struct v9_de_ib_state de_payload = {0};
5519 uint64_t offset, gds_addr, de_payload_gpu_addr;
5520 void *de_payload_cpu_addr;
5521 int cnt;
5522
5523 if (ring->is_mes_queue) {
5524 offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5525 gfx[0].gfx_meta_data) +
5526 offsetof(struct v9_gfx_meta_data, de_payload);
5527 de_payload_gpu_addr =
5528 amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5529 de_payload_cpu_addr =
5530 amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
5531
5532 offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5533 gfx[0].gds_backup) +
5534 offsetof(struct v9_gfx_meta_data, de_payload);
5535 gds_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5536 } else {
5537 offset = offsetof(struct v9_gfx_meta_data, de_payload);
5538 de_payload_gpu_addr = amdgpu_csa_vaddr(ring->adev) + offset;
5539 de_payload_cpu_addr = adev->virt.csa_cpu_addr + offset;
5540
5541 gds_addr = ALIGN(amdgpu_csa_vaddr(ring->adev) +
5542 AMDGPU_CSA_SIZE - adev->gds.gds_size,
5543 PAGE_SIZE);
5544 }
5545
5546 if (usegds) {
5547 de_payload.gds_backup_addrlo = lower_32_bits(gds_addr);
5548 de_payload.gds_backup_addrhi = upper_32_bits(gds_addr);
5549 }
5550
5551 cnt = (sizeof(de_payload) >> 2) + 4 - 2;
5552 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5553 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
5554 WRITE_DATA_DST_SEL(8) |
5555 WR_CONFIRM) |
5556 WRITE_DATA_CACHE_POLICY(0));
5557 amdgpu_ring_write(ring, lower_32_bits(de_payload_gpu_addr));
5558 amdgpu_ring_write(ring, upper_32_bits(de_payload_gpu_addr));
5559
5560 amdgpu_ring_ib_on_emit_de(ring);
5561 if (resume)
5562 amdgpu_ring_write_multiple(ring, de_payload_cpu_addr,
5563 sizeof(de_payload) >> 2);
5564 else
5565 amdgpu_ring_write_multiple(ring, (void *)&de_payload,
5566 sizeof(de_payload) >> 2);
5567 }
5568
gfx_v9_0_ring_emit_frame_cntl(struct amdgpu_ring * ring,bool start,bool secure)5569 static void gfx_v9_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start,
5570 bool secure)
5571 {
5572 uint32_t v = secure ? FRAME_TMZ : 0;
5573
5574 amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0));
5575 amdgpu_ring_write(ring, v | FRAME_CMD(start ? 0 : 1));
5576 }
5577
gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring * ring,uint32_t flags)5578 static void gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
5579 {
5580 uint32_t dw2 = 0;
5581
5582 gfx_v9_0_ring_emit_ce_meta(ring,
5583 (!amdgpu_sriov_vf(ring->adev) &&
5584 flags & AMDGPU_IB_PREEMPTED) ? true : false);
5585
5586 dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
5587 if (flags & AMDGPU_HAVE_CTX_SWITCH) {
5588 /* set load_global_config & load_global_uconfig */
5589 dw2 |= 0x8001;
5590 /* set load_cs_sh_regs */
5591 dw2 |= 0x01000000;
5592 /* set load_per_context_state & load_gfx_sh_regs for GFX */
5593 dw2 |= 0x10002;
5594
5595 /* set load_ce_ram if preamble presented */
5596 if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
5597 dw2 |= 0x10000000;
5598 } else {
5599 /* still load_ce_ram if this is the first time preamble presented
5600 * although there is no context switch happens.
5601 */
5602 if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
5603 dw2 |= 0x10000000;
5604 }
5605
5606 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
5607 amdgpu_ring_write(ring, dw2);
5608 amdgpu_ring_write(ring, 0);
5609 }
5610
gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring * ring)5611 static unsigned gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
5612 {
5613 unsigned ret;
5614 amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
5615 amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
5616 amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
5617 amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
5618 ret = ring->wptr & ring->buf_mask;
5619 amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
5620 return ret;
5621 }
5622
gfx_v9_0_ring_emit_patch_cond_exec(struct amdgpu_ring * ring,unsigned offset)5623 static void gfx_v9_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
5624 {
5625 unsigned cur;
5626 BUG_ON(offset > ring->buf_mask);
5627 BUG_ON(ring->ring[offset] != 0x55aa55aa);
5628
5629 cur = (ring->wptr - 1) & ring->buf_mask;
5630 if (likely(cur > offset))
5631 ring->ring[offset] = cur - offset;
5632 else
5633 ring->ring[offset] = (ring->ring_size>>2) - offset + cur;
5634 }
5635
gfx_v9_0_ring_emit_rreg(struct amdgpu_ring * ring,uint32_t reg,uint32_t reg_val_offs)5636 static void gfx_v9_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
5637 uint32_t reg_val_offs)
5638 {
5639 struct amdgpu_device *adev = ring->adev;
5640
5641 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
5642 amdgpu_ring_write(ring, 0 | /* src: register*/
5643 (5 << 8) | /* dst: memory */
5644 (1 << 20)); /* write confirm */
5645 amdgpu_ring_write(ring, reg);
5646 amdgpu_ring_write(ring, 0);
5647 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
5648 reg_val_offs * 4));
5649 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
5650 reg_val_offs * 4));
5651 }
5652
gfx_v9_0_ring_emit_wreg(struct amdgpu_ring * ring,uint32_t reg,uint32_t val)5653 static void gfx_v9_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
5654 uint32_t val)
5655 {
5656 uint32_t cmd = 0;
5657
5658 switch (ring->funcs->type) {
5659 case AMDGPU_RING_TYPE_GFX:
5660 cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
5661 break;
5662 case AMDGPU_RING_TYPE_KIQ:
5663 cmd = (1 << 16); /* no inc addr */
5664 break;
5665 default:
5666 cmd = WR_CONFIRM;
5667 break;
5668 }
5669 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5670 amdgpu_ring_write(ring, cmd);
5671 amdgpu_ring_write(ring, reg);
5672 amdgpu_ring_write(ring, 0);
5673 amdgpu_ring_write(ring, val);
5674 }
5675
gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring * ring,uint32_t reg,uint32_t val,uint32_t mask)5676 static void gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
5677 uint32_t val, uint32_t mask)
5678 {
5679 gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
5680 }
5681
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)5682 static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
5683 uint32_t reg0, uint32_t reg1,
5684 uint32_t ref, uint32_t mask)
5685 {
5686 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5687 struct amdgpu_device *adev = ring->adev;
5688 bool fw_version_ok = (ring->funcs->type == AMDGPU_RING_TYPE_GFX) ?
5689 adev->gfx.me_fw_write_wait : adev->gfx.mec_fw_write_wait;
5690
5691 if (fw_version_ok)
5692 gfx_v9_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1,
5693 ref, mask, 0x20);
5694 else
5695 amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1,
5696 ref, mask);
5697 }
5698
gfx_v9_0_ring_soft_recovery(struct amdgpu_ring * ring,unsigned vmid)5699 static void gfx_v9_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
5700 {
5701 struct amdgpu_device *adev = ring->adev;
5702 uint32_t value = 0;
5703
5704 value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
5705 value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
5706 value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
5707 value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
5708 WREG32_SOC15(GC, 0, mmSQ_CMD, value);
5709 }
5710
gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device * adev,enum amdgpu_interrupt_state state)5711 static void gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
5712 enum amdgpu_interrupt_state state)
5713 {
5714 switch (state) {
5715 case AMDGPU_IRQ_STATE_DISABLE:
5716 case AMDGPU_IRQ_STATE_ENABLE:
5717 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5718 TIME_STAMP_INT_ENABLE,
5719 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5720 break;
5721 default:
5722 break;
5723 }
5724 }
5725
gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device * adev,int me,int pipe,enum amdgpu_interrupt_state state)5726 static void gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
5727 int me, int pipe,
5728 enum amdgpu_interrupt_state state)
5729 {
5730 u32 mec_int_cntl, mec_int_cntl_reg;
5731
5732 /*
5733 * amdgpu controls only the first MEC. That's why this function only
5734 * handles the setting of interrupts for this specific MEC. All other
5735 * pipes' interrupts are set by amdkfd.
5736 */
5737
5738 if (me == 1) {
5739 switch (pipe) {
5740 case 0:
5741 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
5742 break;
5743 case 1:
5744 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL);
5745 break;
5746 case 2:
5747 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL);
5748 break;
5749 case 3:
5750 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL);
5751 break;
5752 default:
5753 DRM_DEBUG("invalid pipe %d\n", pipe);
5754 return;
5755 }
5756 } else {
5757 DRM_DEBUG("invalid me %d\n", me);
5758 return;
5759 }
5760
5761 switch (state) {
5762 case AMDGPU_IRQ_STATE_DISABLE:
5763 mec_int_cntl = RREG32_SOC15_IP(GC,mec_int_cntl_reg);
5764 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5765 TIME_STAMP_INT_ENABLE, 0);
5766 WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5767 break;
5768 case AMDGPU_IRQ_STATE_ENABLE:
5769 mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg);
5770 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5771 TIME_STAMP_INT_ENABLE, 1);
5772 WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5773 break;
5774 default:
5775 break;
5776 }
5777 }
5778
gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)5779 static int gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
5780 struct amdgpu_irq_src *source,
5781 unsigned type,
5782 enum amdgpu_interrupt_state state)
5783 {
5784 switch (state) {
5785 case AMDGPU_IRQ_STATE_DISABLE:
5786 case AMDGPU_IRQ_STATE_ENABLE:
5787 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5788 PRIV_REG_INT_ENABLE,
5789 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5790 break;
5791 default:
5792 break;
5793 }
5794
5795 return 0;
5796 }
5797
gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)5798 static int gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
5799 struct amdgpu_irq_src *source,
5800 unsigned type,
5801 enum amdgpu_interrupt_state state)
5802 {
5803 switch (state) {
5804 case AMDGPU_IRQ_STATE_DISABLE:
5805 case AMDGPU_IRQ_STATE_ENABLE:
5806 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5807 PRIV_INSTR_INT_ENABLE,
5808 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5809 break;
5810 default:
5811 break;
5812 }
5813
5814 return 0;
5815 }
5816
5817 #define ENABLE_ECC_ON_ME_PIPE(me, pipe) \
5818 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
5819 CP_ECC_ERROR_INT_ENABLE, 1)
5820
5821 #define DISABLE_ECC_ON_ME_PIPE(me, pipe) \
5822 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\
5823 CP_ECC_ERROR_INT_ENABLE, 0)
5824
gfx_v9_0_set_cp_ecc_error_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)5825 static int gfx_v9_0_set_cp_ecc_error_state(struct amdgpu_device *adev,
5826 struct amdgpu_irq_src *source,
5827 unsigned type,
5828 enum amdgpu_interrupt_state state)
5829 {
5830 switch (state) {
5831 case AMDGPU_IRQ_STATE_DISABLE:
5832 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5833 CP_ECC_ERROR_INT_ENABLE, 0);
5834 DISABLE_ECC_ON_ME_PIPE(1, 0);
5835 DISABLE_ECC_ON_ME_PIPE(1, 1);
5836 DISABLE_ECC_ON_ME_PIPE(1, 2);
5837 DISABLE_ECC_ON_ME_PIPE(1, 3);
5838 break;
5839
5840 case AMDGPU_IRQ_STATE_ENABLE:
5841 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
5842 CP_ECC_ERROR_INT_ENABLE, 1);
5843 ENABLE_ECC_ON_ME_PIPE(1, 0);
5844 ENABLE_ECC_ON_ME_PIPE(1, 1);
5845 ENABLE_ECC_ON_ME_PIPE(1, 2);
5846 ENABLE_ECC_ON_ME_PIPE(1, 3);
5847 break;
5848 default:
5849 break;
5850 }
5851
5852 return 0;
5853 }
5854
5855
gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device * adev,struct amdgpu_irq_src * src,unsigned type,enum amdgpu_interrupt_state state)5856 static int gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device *adev,
5857 struct amdgpu_irq_src *src,
5858 unsigned type,
5859 enum amdgpu_interrupt_state state)
5860 {
5861 switch (type) {
5862 case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
5863 gfx_v9_0_set_gfx_eop_interrupt_state(adev, state);
5864 break;
5865 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
5866 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
5867 break;
5868 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
5869 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
5870 break;
5871 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
5872 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
5873 break;
5874 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
5875 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
5876 break;
5877 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
5878 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
5879 break;
5880 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
5881 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
5882 break;
5883 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
5884 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
5885 break;
5886 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
5887 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
5888 break;
5889 default:
5890 break;
5891 }
5892 return 0;
5893 }
5894
gfx_v9_0_eop_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)5895 static int gfx_v9_0_eop_irq(struct amdgpu_device *adev,
5896 struct amdgpu_irq_src *source,
5897 struct amdgpu_iv_entry *entry)
5898 {
5899 int i;
5900 u8 me_id, pipe_id, queue_id;
5901 struct amdgpu_ring *ring;
5902
5903 DRM_DEBUG("IH: CP EOP\n");
5904 me_id = (entry->ring_id & 0x0c) >> 2;
5905 pipe_id = (entry->ring_id & 0x03) >> 0;
5906 queue_id = (entry->ring_id & 0x70) >> 4;
5907
5908 switch (me_id) {
5909 case 0:
5910 if (adev->gfx.num_gfx_rings &&
5911 !amdgpu_mcbp_handle_trailing_fence_irq(&adev->gfx.muxer)) {
5912 /* Fence signals are handled on the software rings*/
5913 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++)
5914 amdgpu_fence_process(&adev->gfx.sw_gfx_ring[i]);
5915 }
5916 break;
5917 case 1:
5918 case 2:
5919 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5920 ring = &adev->gfx.compute_ring[i];
5921 /* Per-queue interrupt is supported for MEC starting from VI.
5922 * The interrupt can only be enabled/disabled per pipe instead of per queue.
5923 */
5924 if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
5925 amdgpu_fence_process(ring);
5926 }
5927 break;
5928 }
5929 return 0;
5930 }
5931
gfx_v9_0_fault(struct amdgpu_device * adev,struct amdgpu_iv_entry * entry)5932 static void gfx_v9_0_fault(struct amdgpu_device *adev,
5933 struct amdgpu_iv_entry *entry)
5934 {
5935 u8 me_id, pipe_id, queue_id;
5936 struct amdgpu_ring *ring;
5937 int i;
5938
5939 me_id = (entry->ring_id & 0x0c) >> 2;
5940 pipe_id = (entry->ring_id & 0x03) >> 0;
5941 queue_id = (entry->ring_id & 0x70) >> 4;
5942
5943 switch (me_id) {
5944 case 0:
5945 drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
5946 break;
5947 case 1:
5948 case 2:
5949 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5950 ring = &adev->gfx.compute_ring[i];
5951 if (ring->me == me_id && ring->pipe == pipe_id &&
5952 ring->queue == queue_id)
5953 drm_sched_fault(&ring->sched);
5954 }
5955 break;
5956 }
5957 }
5958
gfx_v9_0_priv_reg_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)5959 static int gfx_v9_0_priv_reg_irq(struct amdgpu_device *adev,
5960 struct amdgpu_irq_src *source,
5961 struct amdgpu_iv_entry *entry)
5962 {
5963 DRM_ERROR("Illegal register access in command stream\n");
5964 gfx_v9_0_fault(adev, entry);
5965 return 0;
5966 }
5967
gfx_v9_0_priv_inst_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)5968 static int gfx_v9_0_priv_inst_irq(struct amdgpu_device *adev,
5969 struct amdgpu_irq_src *source,
5970 struct amdgpu_iv_entry *entry)
5971 {
5972 DRM_ERROR("Illegal instruction in command stream\n");
5973 gfx_v9_0_fault(adev, entry);
5974 return 0;
5975 }
5976
5977
5978 static const struct soc15_ras_field_entry gfx_v9_0_ras_fields[] = {
5979 { "CPC_SCRATCH", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT),
5980 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, SEC_COUNT),
5981 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, DED_COUNT)
5982 },
5983 { "CPC_UCODE", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT),
5984 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, SEC_COUNT),
5985 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, DED_COUNT)
5986 },
5987 { "CPF_ROQ_ME1", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
5988 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME1),
5989 0, 0
5990 },
5991 { "CPF_ROQ_ME2", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT),
5992 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME2),
5993 0, 0
5994 },
5995 { "CPF_TAG", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT),
5996 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, SEC_COUNT),
5997 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, DED_COUNT)
5998 },
5999 { "CPG_DMA_ROQ", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
6000 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, ROQ_COUNT),
6001 0, 0
6002 },
6003 { "CPG_DMA_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT),
6004 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_SEC_COUNT),
6005 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_DED_COUNT)
6006 },
6007 { "CPG_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT),
6008 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, SEC_COUNT),
6009 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, DED_COUNT)
6010 },
6011 { "DC_CSINVOC", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT),
6012 SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, COUNT_ME1),
6013 0, 0
6014 },
6015 { "DC_RESTORE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT),
6016 SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, COUNT_ME1),
6017 0, 0
6018 },
6019 { "DC_STATE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT),
6020 SOC15_REG_FIELD(DC_EDC_STATE_CNT, COUNT_ME1),
6021 0, 0
6022 },
6023 { "GDS_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
6024 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_SEC),
6025 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_DED)
6026 },
6027 { "GDS_INPUT_QUEUE", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT),
6028 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_INPUT_QUEUE_SED),
6029 0, 0
6030 },
6031 { "GDS_ME0_CS_PIPE_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6032 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_SEC),
6033 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_DED)
6034 },
6035 { "GDS_OA_PHY_PHY_CMD_RAM_MEM",
6036 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6037 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_SEC),
6038 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_DED)
6039 },
6040 { "GDS_OA_PHY_PHY_DATA_RAM_MEM",
6041 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT),
6042 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_DATA_RAM_MEM_SED),
6043 0, 0
6044 },
6045 { "GDS_OA_PIPE_ME1_PIPE0_PIPE_MEM",
6046 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6047 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_SEC),
6048 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_DED)
6049 },
6050 { "GDS_OA_PIPE_ME1_PIPE1_PIPE_MEM",
6051 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6052 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_SEC),
6053 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_DED)
6054 },
6055 { "GDS_OA_PIPE_ME1_PIPE2_PIPE_MEM",
6056 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6057 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_SEC),
6058 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_DED)
6059 },
6060 { "GDS_OA_PIPE_ME1_PIPE3_PIPE_MEM",
6061 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT),
6062 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_SEC),
6063 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_DED)
6064 },
6065 { "SPI_SR_MEM", SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT),
6066 SOC15_REG_FIELD(SPI_EDC_CNT, SPI_SR_MEM_SED_COUNT),
6067 0, 0
6068 },
6069 { "TA_FS_DFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6070 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_SEC_COUNT),
6071 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_DED_COUNT)
6072 },
6073 { "TA_FS_AFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6074 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_SED_COUNT),
6075 0, 0
6076 },
6077 { "TA_FL_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6078 SOC15_REG_FIELD(TA_EDC_CNT, TA_FL_LFIFO_SED_COUNT),
6079 0, 0
6080 },
6081 { "TA_FX_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6082 SOC15_REG_FIELD(TA_EDC_CNT, TA_FX_LFIFO_SED_COUNT),
6083 0, 0
6084 },
6085 { "TA_FS_CFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT),
6086 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_CFIFO_SED_COUNT),
6087 0, 0
6088 },
6089 { "TCA_HOLE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
6090 SOC15_REG_FIELD(TCA_EDC_CNT, HOLE_FIFO_SED_COUNT),
6091 0, 0
6092 },
6093 { "TCA_REQ_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT),
6094 SOC15_REG_FIELD(TCA_EDC_CNT, REQ_FIFO_SED_COUNT),
6095 0, 0
6096 },
6097 { "TCC_CACHE_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6098 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_SEC_COUNT),
6099 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_DED_COUNT)
6100 },
6101 { "TCC_CACHE_DIRTY", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6102 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_SEC_COUNT),
6103 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_DED_COUNT)
6104 },
6105 { "TCC_HIGH_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6106 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_SEC_COUNT),
6107 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_DED_COUNT)
6108 },
6109 { "TCC_LOW_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6110 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_SEC_COUNT),
6111 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_DED_COUNT)
6112 },
6113 { "TCC_SRC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6114 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_SEC_COUNT),
6115 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_DED_COUNT)
6116 },
6117 { "TCC_IN_USE_DEC", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6118 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_DEC_SED_COUNT),
6119 0, 0
6120 },
6121 { "TCC_IN_USE_TRANSFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6122 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_TRANSFER_SED_COUNT),
6123 0, 0
6124 },
6125 { "TCC_LATENCY_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6126 SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_SED_COUNT),
6127 0, 0
6128 },
6129 { "TCC_RETURN_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6130 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_DATA_SED_COUNT),
6131 0, 0
6132 },
6133 { "TCC_RETURN_CONTROL", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6134 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_CONTROL_SED_COUNT),
6135 0, 0
6136 },
6137 { "TCC_UC_ATOMIC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT),
6138 SOC15_REG_FIELD(TCC_EDC_CNT, UC_ATOMIC_FIFO_SED_COUNT),
6139 0, 0
6140 },
6141 { "TCC_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6142 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_RETURN_SED_COUNT),
6143 0, 0
6144 },
6145 { "TCC_WRITE_CACHE_READ", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6146 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_CACHE_READ_SED_COUNT),
6147 0, 0
6148 },
6149 { "TCC_SRC_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6150 SOC15_REG_FIELD(TCC_EDC_CNT2, SRC_FIFO_NEXT_RAM_SED_COUNT),
6151 0, 0
6152 },
6153 { "TCC_LATENCY_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6154 SOC15_REG_FIELD(TCC_EDC_CNT2, LATENCY_FIFO_NEXT_RAM_SED_COUNT),
6155 0, 0
6156 },
6157 { "TCC_CACHE_TAG_PROBE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6158 SOC15_REG_FIELD(TCC_EDC_CNT2, CACHE_TAG_PROBE_FIFO_SED_COUNT),
6159 0, 0
6160 },
6161 { "TCC_WRRET_TAG_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6162 SOC15_REG_FIELD(TCC_EDC_CNT2, WRRET_TAG_WRITE_RETURN_SED_COUNT),
6163 0, 0
6164 },
6165 { "TCC_ATOMIC_RETURN_BUFFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2),
6166 SOC15_REG_FIELD(TCC_EDC_CNT2, ATOMIC_RETURN_BUFFER_SED_COUNT),
6167 0, 0
6168 },
6169 { "TCI_WRITE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT),
6170 SOC15_REG_FIELD(TCI_EDC_CNT, WRITE_RAM_SED_COUNT),
6171 0, 0
6172 },
6173 { "TCP_CACHE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6174 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_SEC_COUNT),
6175 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_DED_COUNT)
6176 },
6177 { "TCP_LFIFO_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6178 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_SEC_COUNT),
6179 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_DED_COUNT)
6180 },
6181 { "TCP_CMD_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6182 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CMD_FIFO_SED_COUNT),
6183 0, 0
6184 },
6185 { "TCP_VM_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6186 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, VM_FIFO_SEC_COUNT),
6187 0, 0
6188 },
6189 { "TCP_DB_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6190 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, DB_RAM_SED_COUNT),
6191 0, 0
6192 },
6193 { "TCP_UTCL1_LFIFO0", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6194 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_SEC_COUNT),
6195 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_DED_COUNT)
6196 },
6197 { "TCP_UTCL1_LFIFO1", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW),
6198 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_SEC_COUNT),
6199 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_DED_COUNT)
6200 },
6201 { "TD_SS_FIFO_LO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6202 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_SEC_COUNT),
6203 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_DED_COUNT)
6204 },
6205 { "TD_SS_FIFO_HI", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6206 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_SEC_COUNT),
6207 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_DED_COUNT)
6208 },
6209 { "TD_CS_FIFO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT),
6210 SOC15_REG_FIELD(TD_EDC_CNT, CS_FIFO_SED_COUNT),
6211 0, 0
6212 },
6213 { "SQ_LDS_D", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6214 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_SEC_COUNT),
6215 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_DED_COUNT)
6216 },
6217 { "SQ_LDS_I", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6218 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_SEC_COUNT),
6219 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_DED_COUNT)
6220 },
6221 { "SQ_SGPR", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6222 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_SEC_COUNT),
6223 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_DED_COUNT)
6224 },
6225 { "SQ_VGPR0", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6226 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_SEC_COUNT),
6227 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_DED_COUNT)
6228 },
6229 { "SQ_VGPR1", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6230 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_SEC_COUNT),
6231 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_DED_COUNT)
6232 },
6233 { "SQ_VGPR2", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6234 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_SEC_COUNT),
6235 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_DED_COUNT)
6236 },
6237 { "SQ_VGPR3", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT),
6238 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_SEC_COUNT),
6239 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_DED_COUNT)
6240 },
6241 { "SQC_DATA_CU0_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6242 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_SEC_COUNT),
6243 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_DED_COUNT)
6244 },
6245 { "SQC_DATA_CU0_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6246 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_SEC_COUNT),
6247 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_DED_COUNT)
6248 },
6249 { "SQC_DATA_CU1_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6250 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_SEC_COUNT),
6251 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_DED_COUNT)
6252 },
6253 { "SQC_DATA_CU1_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6254 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_SEC_COUNT),
6255 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_DED_COUNT)
6256 },
6257 { "SQC_DATA_CU2_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6258 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_SEC_COUNT),
6259 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_DED_COUNT)
6260 },
6261 { "SQC_DATA_CU2_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT),
6262 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_SEC_COUNT),
6263 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_DED_COUNT)
6264 },
6265 { "SQC_INST_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6266 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_SEC_COUNT),
6267 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_DED_COUNT)
6268 },
6269 { "SQC_INST_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6270 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_SEC_COUNT),
6271 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_DED_COUNT)
6272 },
6273 { "SQC_DATA_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6274 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_SEC_COUNT),
6275 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_DED_COUNT)
6276 },
6277 { "SQC_DATA_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6278 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_SEC_COUNT),
6279 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_DED_COUNT)
6280 },
6281 { "SQC_INST_BANKA_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6282 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_UTCL1_MISS_FIFO_SED_COUNT),
6283 0, 0
6284 },
6285 { "SQC_INST_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6286 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_MISS_FIFO_SED_COUNT),
6287 0, 0
6288 },
6289 { "SQC_DATA_BANKA_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6290 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_HIT_FIFO_SED_COUNT),
6291 0, 0
6292 },
6293 { "SQC_DATA_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6294 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_MISS_FIFO_SED_COUNT),
6295 0, 0
6296 },
6297 { "SQC_DATA_BANKA_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6298 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_DIRTY_BIT_RAM_SED_COUNT),
6299 0, 0
6300 },
6301 { "SQC_INST_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2),
6302 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_SEC_COUNT),
6303 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_DED_COUNT)
6304 },
6305 { "SQC_INST_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6306 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_SEC_COUNT),
6307 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_DED_COUNT)
6308 },
6309 { "SQC_INST_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6310 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_SEC_COUNT),
6311 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_DED_COUNT)
6312 },
6313 { "SQC_DATA_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6314 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_SEC_COUNT),
6315 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_DED_COUNT)
6316 },
6317 { "SQC_DATA_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6318 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_SEC_COUNT),
6319 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_DED_COUNT)
6320 },
6321 { "SQC_INST_BANKB_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6322 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_UTCL1_MISS_FIFO_SED_COUNT),
6323 0, 0
6324 },
6325 { "SQC_INST_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6326 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_MISS_FIFO_SED_COUNT),
6327 0, 0
6328 },
6329 { "SQC_DATA_BANKB_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6330 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_HIT_FIFO_SED_COUNT),
6331 0, 0
6332 },
6333 { "SQC_DATA_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6334 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_MISS_FIFO_SED_COUNT),
6335 0, 0
6336 },
6337 { "SQC_DATA_BANKB_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3),
6338 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_DIRTY_BIT_RAM_SED_COUNT),
6339 0, 0
6340 },
6341 { "EA_DRAMRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6342 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT),
6343 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT)
6344 },
6345 { "EA_DRAMWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6346 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT),
6347 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT)
6348 },
6349 { "EA_DRAMWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6350 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT),
6351 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT)
6352 },
6353 { "EA_RRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6354 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_SEC_COUNT),
6355 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_DED_COUNT)
6356 },
6357 { "EA_WRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6358 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_SEC_COUNT),
6359 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_DED_COUNT)
6360 },
6361 { "EA_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6362 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT),
6363 0, 0
6364 },
6365 { "EA_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6366 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT),
6367 0, 0
6368 },
6369 { "EA_IORD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6370 SOC15_REG_FIELD(GCEA_EDC_CNT, IORD_CMDMEM_SED_COUNT),
6371 0, 0
6372 },
6373 { "EA_IOWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6374 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_CMDMEM_SED_COUNT),
6375 0, 0
6376 },
6377 { "EA_IOWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT),
6378 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_DATAMEM_SED_COUNT),
6379 0, 0
6380 },
6381 { "GMIRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6382 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT),
6383 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT)
6384 },
6385 { "GMIWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6386 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT),
6387 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT)
6388 },
6389 { "GMIWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6390 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT),
6391 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT)
6392 },
6393 { "GMIRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6394 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT),
6395 0, 0
6396 },
6397 { "GMIWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6398 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT),
6399 0, 0
6400 },
6401 { "MAM_D0MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6402 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D0MEM_SED_COUNT),
6403 0, 0
6404 },
6405 { "MAM_D1MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6406 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D1MEM_SED_COUNT),
6407 0, 0
6408 },
6409 { "MAM_D2MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6410 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D2MEM_SED_COUNT),
6411 0, 0
6412 },
6413 { "MAM_D3MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2),
6414 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D3MEM_SED_COUNT),
6415 0, 0
6416 }
6417 };
6418
gfx_v9_0_ras_error_inject(struct amdgpu_device * adev,void * inject_if,uint32_t instance_mask)6419 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
6420 void *inject_if, uint32_t instance_mask)
6421 {
6422 struct ras_inject_if *info = (struct ras_inject_if *)inject_if;
6423 int ret;
6424 struct ta_ras_trigger_error_input block_info = { 0 };
6425
6426 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6427 return -EINVAL;
6428
6429 if (info->head.sub_block_index >= ARRAY_SIZE(ras_gfx_subblocks))
6430 return -EINVAL;
6431
6432 if (!ras_gfx_subblocks[info->head.sub_block_index].name)
6433 return -EPERM;
6434
6435 if (!(ras_gfx_subblocks[info->head.sub_block_index].hw_supported_error_type &
6436 info->head.type)) {
6437 DRM_ERROR("GFX Subblock %s, hardware 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 if (!(ras_gfx_subblocks[info->head.sub_block_index].sw_supported_error_type &
6444 info->head.type)) {
6445 DRM_ERROR("GFX Subblock %s, driver do not support type 0x%x\n",
6446 ras_gfx_subblocks[info->head.sub_block_index].name,
6447 info->head.type);
6448 return -EPERM;
6449 }
6450
6451 block_info.block_id = amdgpu_ras_block_to_ta(info->head.block);
6452 block_info.sub_block_index =
6453 ras_gfx_subblocks[info->head.sub_block_index].ta_subblock;
6454 block_info.inject_error_type = amdgpu_ras_error_to_ta(info->head.type);
6455 block_info.address = info->address;
6456 block_info.value = info->value;
6457
6458 mutex_lock(&adev->grbm_idx_mutex);
6459 ret = psp_ras_trigger_error(&adev->psp, &block_info, instance_mask);
6460 mutex_unlock(&adev->grbm_idx_mutex);
6461
6462 return ret;
6463 }
6464
6465 static const char *vml2_mems[] = {
6466 "UTC_VML2_BANK_CACHE_0_BIGK_MEM0",
6467 "UTC_VML2_BANK_CACHE_0_BIGK_MEM1",
6468 "UTC_VML2_BANK_CACHE_0_4K_MEM0",
6469 "UTC_VML2_BANK_CACHE_0_4K_MEM1",
6470 "UTC_VML2_BANK_CACHE_1_BIGK_MEM0",
6471 "UTC_VML2_BANK_CACHE_1_BIGK_MEM1",
6472 "UTC_VML2_BANK_CACHE_1_4K_MEM0",
6473 "UTC_VML2_BANK_CACHE_1_4K_MEM1",
6474 "UTC_VML2_BANK_CACHE_2_BIGK_MEM0",
6475 "UTC_VML2_BANK_CACHE_2_BIGK_MEM1",
6476 "UTC_VML2_BANK_CACHE_2_4K_MEM0",
6477 "UTC_VML2_BANK_CACHE_2_4K_MEM1",
6478 "UTC_VML2_BANK_CACHE_3_BIGK_MEM0",
6479 "UTC_VML2_BANK_CACHE_3_BIGK_MEM1",
6480 "UTC_VML2_BANK_CACHE_3_4K_MEM0",
6481 "UTC_VML2_BANK_CACHE_3_4K_MEM1",
6482 };
6483
6484 static const char *vml2_walker_mems[] = {
6485 "UTC_VML2_CACHE_PDE0_MEM0",
6486 "UTC_VML2_CACHE_PDE0_MEM1",
6487 "UTC_VML2_CACHE_PDE1_MEM0",
6488 "UTC_VML2_CACHE_PDE1_MEM1",
6489 "UTC_VML2_CACHE_PDE2_MEM0",
6490 "UTC_VML2_CACHE_PDE2_MEM1",
6491 "UTC_VML2_RDIF_LOG_FIFO",
6492 };
6493
6494 static const char *atc_l2_cache_2m_mems[] = {
6495 "UTC_ATCL2_CACHE_2M_BANK0_WAY0_MEM",
6496 "UTC_ATCL2_CACHE_2M_BANK0_WAY1_MEM",
6497 "UTC_ATCL2_CACHE_2M_BANK1_WAY0_MEM",
6498 "UTC_ATCL2_CACHE_2M_BANK1_WAY1_MEM",
6499 };
6500
6501 static const char *atc_l2_cache_4k_mems[] = {
6502 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM0",
6503 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM1",
6504 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM2",
6505 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM3",
6506 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM4",
6507 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM5",
6508 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM6",
6509 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM7",
6510 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM0",
6511 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM1",
6512 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM2",
6513 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM3",
6514 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM4",
6515 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM5",
6516 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM6",
6517 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM7",
6518 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM0",
6519 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM1",
6520 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM2",
6521 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM3",
6522 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM4",
6523 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM5",
6524 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM6",
6525 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM7",
6526 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM0",
6527 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM1",
6528 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM2",
6529 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM3",
6530 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM4",
6531 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM5",
6532 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM6",
6533 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM7",
6534 };
6535
gfx_v9_0_query_utc_edc_status(struct amdgpu_device * adev,struct ras_err_data * err_data)6536 static int gfx_v9_0_query_utc_edc_status(struct amdgpu_device *adev,
6537 struct ras_err_data *err_data)
6538 {
6539 uint32_t i, data;
6540 uint32_t sec_count, ded_count;
6541
6542 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6543 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0);
6544 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6545 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0);
6546 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6547 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0);
6548 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6549 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0);
6550
6551 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) {
6552 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i);
6553 data = RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT);
6554
6555 sec_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, SEC_COUNT);
6556 if (sec_count) {
6557 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6558 "SEC %d\n", i, vml2_mems[i], sec_count);
6559 err_data->ce_count += sec_count;
6560 }
6561
6562 ded_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, DED_COUNT);
6563 if (ded_count) {
6564 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6565 "DED %d\n", i, vml2_mems[i], ded_count);
6566 err_data->ue_count += ded_count;
6567 }
6568 }
6569
6570 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) {
6571 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i);
6572 data = RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT);
6573
6574 sec_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6575 SEC_COUNT);
6576 if (sec_count) {
6577 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6578 "SEC %d\n", i, vml2_walker_mems[i], sec_count);
6579 err_data->ce_count += sec_count;
6580 }
6581
6582 ded_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT,
6583 DED_COUNT);
6584 if (ded_count) {
6585 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6586 "DED %d\n", i, vml2_walker_mems[i], ded_count);
6587 err_data->ue_count += ded_count;
6588 }
6589 }
6590
6591 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) {
6592 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i);
6593 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT);
6594
6595 sec_count = (data & 0x00006000L) >> 0xd;
6596 if (sec_count) {
6597 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6598 "SEC %d\n", i, atc_l2_cache_2m_mems[i],
6599 sec_count);
6600 err_data->ce_count += sec_count;
6601 }
6602 }
6603
6604 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) {
6605 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i);
6606 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT);
6607
6608 sec_count = (data & 0x00006000L) >> 0xd;
6609 if (sec_count) {
6610 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6611 "SEC %d\n", i, atc_l2_cache_4k_mems[i],
6612 sec_count);
6613 err_data->ce_count += sec_count;
6614 }
6615
6616 ded_count = (data & 0x00018000L) >> 0xf;
6617 if (ded_count) {
6618 dev_info(adev->dev, "Instance[%d]: SubBlock %s, "
6619 "DED %d\n", i, atc_l2_cache_4k_mems[i],
6620 ded_count);
6621 err_data->ue_count += ded_count;
6622 }
6623 }
6624
6625 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6626 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6627 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6628 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6629
6630 return 0;
6631 }
6632
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)6633 static int gfx_v9_0_ras_error_count(struct amdgpu_device *adev,
6634 const struct soc15_reg_entry *reg,
6635 uint32_t se_id, uint32_t inst_id, uint32_t value,
6636 uint32_t *sec_count, uint32_t *ded_count)
6637 {
6638 uint32_t i;
6639 uint32_t sec_cnt, ded_cnt;
6640
6641 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_ras_fields); i++) {
6642 if(gfx_v9_0_ras_fields[i].reg_offset != reg->reg_offset ||
6643 gfx_v9_0_ras_fields[i].seg != reg->seg ||
6644 gfx_v9_0_ras_fields[i].inst != reg->inst)
6645 continue;
6646
6647 sec_cnt = (value &
6648 gfx_v9_0_ras_fields[i].sec_count_mask) >>
6649 gfx_v9_0_ras_fields[i].sec_count_shift;
6650 if (sec_cnt) {
6651 dev_info(adev->dev, "GFX SubBlock %s, "
6652 "Instance[%d][%d], SEC %d\n",
6653 gfx_v9_0_ras_fields[i].name,
6654 se_id, inst_id,
6655 sec_cnt);
6656 *sec_count += sec_cnt;
6657 }
6658
6659 ded_cnt = (value &
6660 gfx_v9_0_ras_fields[i].ded_count_mask) >>
6661 gfx_v9_0_ras_fields[i].ded_count_shift;
6662 if (ded_cnt) {
6663 dev_info(adev->dev, "GFX SubBlock %s, "
6664 "Instance[%d][%d], DED %d\n",
6665 gfx_v9_0_ras_fields[i].name,
6666 se_id, inst_id,
6667 ded_cnt);
6668 *ded_count += ded_cnt;
6669 }
6670 }
6671
6672 return 0;
6673 }
6674
gfx_v9_0_reset_ras_error_count(struct amdgpu_device * adev)6675 static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev)
6676 {
6677 int i, j, k;
6678
6679 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6680 return;
6681
6682 /* read back registers to clear the counters */
6683 mutex_lock(&adev->grbm_idx_mutex);
6684 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) {
6685 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) {
6686 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) {
6687 amdgpu_gfx_select_se_sh(adev, j, 0x0, k, 0);
6688 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i]));
6689 }
6690 }
6691 }
6692 WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, 0xe0000000);
6693 mutex_unlock(&adev->grbm_idx_mutex);
6694
6695 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6696 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0);
6697 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6698 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0);
6699 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6700 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0);
6701 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6702 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0);
6703
6704 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) {
6705 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i);
6706 RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT);
6707 }
6708
6709 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) {
6710 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i);
6711 RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT);
6712 }
6713
6714 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) {
6715 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i);
6716 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT);
6717 }
6718
6719 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) {
6720 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i);
6721 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT);
6722 }
6723
6724 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255);
6725 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255);
6726 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255);
6727 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255);
6728 }
6729
gfx_v9_0_query_ras_error_count(struct amdgpu_device * adev,void * ras_error_status)6730 static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
6731 void *ras_error_status)
6732 {
6733 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
6734 uint32_t sec_count = 0, ded_count = 0;
6735 uint32_t i, j, k;
6736 uint32_t reg_value;
6737
6738 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
6739 return;
6740
6741 err_data->ue_count = 0;
6742 err_data->ce_count = 0;
6743
6744 mutex_lock(&adev->grbm_idx_mutex);
6745
6746 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) {
6747 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) {
6748 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) {
6749 amdgpu_gfx_select_se_sh(adev, j, 0, k, 0);
6750 reg_value =
6751 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i]));
6752 if (reg_value)
6753 gfx_v9_0_ras_error_count(adev,
6754 &gfx_v9_0_edc_counter_regs[i],
6755 j, k, reg_value,
6756 &sec_count, &ded_count);
6757 }
6758 }
6759 }
6760
6761 err_data->ce_count += sec_count;
6762 err_data->ue_count += ded_count;
6763
6764 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
6765 mutex_unlock(&adev->grbm_idx_mutex);
6766
6767 gfx_v9_0_query_utc_edc_status(adev, err_data);
6768 }
6769
gfx_v9_0_emit_mem_sync(struct amdgpu_ring * ring)6770 static void gfx_v9_0_emit_mem_sync(struct amdgpu_ring *ring)
6771 {
6772 const unsigned int cp_coher_cntl =
6773 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_ICACHE_ACTION_ENA(1) |
6774 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_KCACHE_ACTION_ENA(1) |
6775 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_ACTION_ENA(1) |
6776 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TCL1_ACTION_ENA(1) |
6777 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_WB_ACTION_ENA(1);
6778
6779 /* ACQUIRE_MEM -make one or more surfaces valid for use by the subsequent operations */
6780 amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
6781 amdgpu_ring_write(ring, cp_coher_cntl); /* CP_COHER_CNTL */
6782 amdgpu_ring_write(ring, 0xffffffff); /* CP_COHER_SIZE */
6783 amdgpu_ring_write(ring, 0xffffff); /* CP_COHER_SIZE_HI */
6784 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */
6785 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE_HI */
6786 amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */
6787 }
6788
gfx_v9_0_emit_wave_limit_cs(struct amdgpu_ring * ring,uint32_t pipe,bool enable)6789 static void gfx_v9_0_emit_wave_limit_cs(struct amdgpu_ring *ring,
6790 uint32_t pipe, bool enable)
6791 {
6792 struct amdgpu_device *adev = ring->adev;
6793 uint32_t val;
6794 uint32_t wcl_cs_reg;
6795
6796 /* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */
6797 val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS0_DEFAULT;
6798
6799 switch (pipe) {
6800 case 0:
6801 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS0);
6802 break;
6803 case 1:
6804 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS1);
6805 break;
6806 case 2:
6807 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS2);
6808 break;
6809 case 3:
6810 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS3);
6811 break;
6812 default:
6813 DRM_DEBUG("invalid pipe %d\n", pipe);
6814 return;
6815 }
6816
6817 amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val);
6818
6819 }
gfx_v9_0_emit_wave_limit(struct amdgpu_ring * ring,bool enable)6820 static void gfx_v9_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable)
6821 {
6822 struct amdgpu_device *adev = ring->adev;
6823 uint32_t val;
6824 int i;
6825
6826
6827 /* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit
6828 * number of gfx waves. Setting 5 bit will make sure gfx only gets
6829 * around 25% of gpu resources.
6830 */
6831 val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT;
6832 amdgpu_ring_emit_wreg(ring,
6833 SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_GFX),
6834 val);
6835
6836 /* Restrict waves for normal/low priority compute queues as well
6837 * to get best QoS for high priority compute jobs.
6838 *
6839 * amdgpu controls only 1st ME(0-3 CS pipes).
6840 */
6841 for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
6842 if (i != ring->pipe)
6843 gfx_v9_0_emit_wave_limit_cs(ring, i, enable);
6844
6845 }
6846 }
6847
6848 static const struct amd_ip_funcs gfx_v9_0_ip_funcs = {
6849 .name = "gfx_v9_0",
6850 .early_init = gfx_v9_0_early_init,
6851 .late_init = gfx_v9_0_late_init,
6852 .sw_init = gfx_v9_0_sw_init,
6853 .sw_fini = gfx_v9_0_sw_fini,
6854 .hw_init = gfx_v9_0_hw_init,
6855 .hw_fini = gfx_v9_0_hw_fini,
6856 .suspend = gfx_v9_0_suspend,
6857 .resume = gfx_v9_0_resume,
6858 .is_idle = gfx_v9_0_is_idle,
6859 .wait_for_idle = gfx_v9_0_wait_for_idle,
6860 .soft_reset = gfx_v9_0_soft_reset,
6861 .set_clockgating_state = gfx_v9_0_set_clockgating_state,
6862 .set_powergating_state = gfx_v9_0_set_powergating_state,
6863 .get_clockgating_state = gfx_v9_0_get_clockgating_state,
6864 };
6865
6866 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = {
6867 .type = AMDGPU_RING_TYPE_GFX,
6868 .align_mask = 0xff,
6869 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
6870 .support_64bit_ptrs = true,
6871 .secure_submission_supported = true,
6872 .get_rptr = gfx_v9_0_ring_get_rptr_gfx,
6873 .get_wptr = gfx_v9_0_ring_get_wptr_gfx,
6874 .set_wptr = gfx_v9_0_ring_set_wptr_gfx,
6875 .emit_frame_size = /* totally 242 maximum if 16 IBs */
6876 5 + /* COND_EXEC */
6877 7 + /* PIPELINE_SYNC */
6878 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6879 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6880 2 + /* VM_FLUSH */
6881 8 + /* FENCE for VM_FLUSH */
6882 20 + /* GDS switch */
6883 4 + /* double SWITCH_BUFFER,
6884 the first COND_EXEC jump to the place just
6885 prior to this double SWITCH_BUFFER */
6886 5 + /* COND_EXEC */
6887 7 + /* HDP_flush */
6888 4 + /* VGT_flush */
6889 14 + /* CE_META */
6890 31 + /* DE_META */
6891 3 + /* CNTX_CTRL */
6892 5 + /* HDP_INVL */
6893 8 + 8 + /* FENCE x2 */
6894 2 + /* SWITCH_BUFFER */
6895 7, /* gfx_v9_0_emit_mem_sync */
6896 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_gfx */
6897 .emit_ib = gfx_v9_0_ring_emit_ib_gfx,
6898 .emit_fence = gfx_v9_0_ring_emit_fence,
6899 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
6900 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
6901 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
6902 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
6903 .test_ring = gfx_v9_0_ring_test_ring,
6904 .insert_nop = amdgpu_ring_insert_nop,
6905 .pad_ib = amdgpu_ring_generic_pad_ib,
6906 .emit_switch_buffer = gfx_v9_ring_emit_sb,
6907 .emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
6908 .init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
6909 .patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec,
6910 .preempt_ib = gfx_v9_0_ring_preempt_ib,
6911 .emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl,
6912 .emit_wreg = gfx_v9_0_ring_emit_wreg,
6913 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6914 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6915 .soft_recovery = gfx_v9_0_ring_soft_recovery,
6916 .emit_mem_sync = gfx_v9_0_emit_mem_sync,
6917 };
6918
6919 static const struct amdgpu_ring_funcs gfx_v9_0_sw_ring_funcs_gfx = {
6920 .type = AMDGPU_RING_TYPE_GFX,
6921 .align_mask = 0xff,
6922 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
6923 .support_64bit_ptrs = true,
6924 .secure_submission_supported = true,
6925 .get_rptr = amdgpu_sw_ring_get_rptr_gfx,
6926 .get_wptr = amdgpu_sw_ring_get_wptr_gfx,
6927 .set_wptr = amdgpu_sw_ring_set_wptr_gfx,
6928 .emit_frame_size = /* totally 242 maximum if 16 IBs */
6929 5 + /* COND_EXEC */
6930 7 + /* PIPELINE_SYNC */
6931 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6932 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6933 2 + /* VM_FLUSH */
6934 8 + /* FENCE for VM_FLUSH */
6935 20 + /* GDS switch */
6936 4 + /* double SWITCH_BUFFER,
6937 * the first COND_EXEC jump to the place just
6938 * prior to this double SWITCH_BUFFER
6939 */
6940 5 + /* COND_EXEC */
6941 7 + /* HDP_flush */
6942 4 + /* VGT_flush */
6943 14 + /* CE_META */
6944 31 + /* DE_META */
6945 3 + /* CNTX_CTRL */
6946 5 + /* HDP_INVL */
6947 8 + 8 + /* FENCE x2 */
6948 2 + /* SWITCH_BUFFER */
6949 7, /* gfx_v9_0_emit_mem_sync */
6950 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_gfx */
6951 .emit_ib = gfx_v9_0_ring_emit_ib_gfx,
6952 .emit_fence = gfx_v9_0_ring_emit_fence,
6953 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
6954 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
6955 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
6956 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
6957 .test_ring = gfx_v9_0_ring_test_ring,
6958 .test_ib = gfx_v9_0_ring_test_ib,
6959 .insert_nop = amdgpu_sw_ring_insert_nop,
6960 .pad_ib = amdgpu_ring_generic_pad_ib,
6961 .emit_switch_buffer = gfx_v9_ring_emit_sb,
6962 .emit_cntxcntl = gfx_v9_ring_emit_cntxcntl,
6963 .init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec,
6964 .patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec,
6965 .emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl,
6966 .emit_wreg = gfx_v9_0_ring_emit_wreg,
6967 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
6968 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
6969 .soft_recovery = gfx_v9_0_ring_soft_recovery,
6970 .emit_mem_sync = gfx_v9_0_emit_mem_sync,
6971 .patch_cntl = gfx_v9_0_ring_patch_cntl,
6972 .patch_de = gfx_v9_0_ring_patch_de_meta,
6973 .patch_ce = gfx_v9_0_ring_patch_ce_meta,
6974 };
6975
6976 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = {
6977 .type = AMDGPU_RING_TYPE_COMPUTE,
6978 .align_mask = 0xff,
6979 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
6980 .support_64bit_ptrs = true,
6981 .get_rptr = gfx_v9_0_ring_get_rptr_compute,
6982 .get_wptr = gfx_v9_0_ring_get_wptr_compute,
6983 .set_wptr = gfx_v9_0_ring_set_wptr_compute,
6984 .emit_frame_size =
6985 20 + /* gfx_v9_0_ring_emit_gds_switch */
6986 7 + /* gfx_v9_0_ring_emit_hdp_flush */
6987 5 + /* hdp invalidate */
6988 7 + /* gfx_v9_0_ring_emit_pipeline_sync */
6989 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6990 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6991 2 + /* gfx_v9_0_ring_emit_vm_flush */
6992 8 + 8 + 8 + /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */
6993 7 + /* gfx_v9_0_emit_mem_sync */
6994 5 + /* gfx_v9_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */
6995 15, /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */
6996 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */
6997 .emit_ib = gfx_v9_0_ring_emit_ib_compute,
6998 .emit_fence = gfx_v9_0_ring_emit_fence,
6999 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync,
7000 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush,
7001 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch,
7002 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush,
7003 .test_ring = gfx_v9_0_ring_test_ring,
7004 .test_ib = gfx_v9_0_ring_test_ib,
7005 .insert_nop = amdgpu_ring_insert_nop,
7006 .pad_ib = amdgpu_ring_generic_pad_ib,
7007 .emit_wreg = gfx_v9_0_ring_emit_wreg,
7008 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
7009 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
7010 .emit_mem_sync = gfx_v9_0_emit_mem_sync,
7011 .emit_wave_limit = gfx_v9_0_emit_wave_limit,
7012 };
7013
7014 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = {
7015 .type = AMDGPU_RING_TYPE_KIQ,
7016 .align_mask = 0xff,
7017 .nop = PACKET3(PACKET3_NOP, 0x3FFF),
7018 .support_64bit_ptrs = true,
7019 .get_rptr = gfx_v9_0_ring_get_rptr_compute,
7020 .get_wptr = gfx_v9_0_ring_get_wptr_compute,
7021 .set_wptr = gfx_v9_0_ring_set_wptr_compute,
7022 .emit_frame_size =
7023 20 + /* gfx_v9_0_ring_emit_gds_switch */
7024 7 + /* gfx_v9_0_ring_emit_hdp_flush */
7025 5 + /* hdp invalidate */
7026 7 + /* gfx_v9_0_ring_emit_pipeline_sync */
7027 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
7028 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
7029 2 + /* gfx_v9_0_ring_emit_vm_flush */
7030 8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */
7031 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */
7032 .emit_fence = gfx_v9_0_ring_emit_fence_kiq,
7033 .test_ring = gfx_v9_0_ring_test_ring,
7034 .insert_nop = amdgpu_ring_insert_nop,
7035 .pad_ib = amdgpu_ring_generic_pad_ib,
7036 .emit_rreg = gfx_v9_0_ring_emit_rreg,
7037 .emit_wreg = gfx_v9_0_ring_emit_wreg,
7038 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait,
7039 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait,
7040 };
7041
gfx_v9_0_set_ring_funcs(struct amdgpu_device * adev)7042 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev)
7043 {
7044 int i;
7045
7046 adev->gfx.kiq[0].ring.funcs = &gfx_v9_0_ring_funcs_kiq;
7047
7048 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
7049 adev->gfx.gfx_ring[i].funcs = &gfx_v9_0_ring_funcs_gfx;
7050
7051 if (adev->gfx.num_gfx_rings) {
7052 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++)
7053 adev->gfx.sw_gfx_ring[i].funcs = &gfx_v9_0_sw_ring_funcs_gfx;
7054 }
7055
7056 for (i = 0; i < adev->gfx.num_compute_rings; i++)
7057 adev->gfx.compute_ring[i].funcs = &gfx_v9_0_ring_funcs_compute;
7058 }
7059
7060 static const struct amdgpu_irq_src_funcs gfx_v9_0_eop_irq_funcs = {
7061 .set = gfx_v9_0_set_eop_interrupt_state,
7062 .process = gfx_v9_0_eop_irq,
7063 };
7064
7065 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_reg_irq_funcs = {
7066 .set = gfx_v9_0_set_priv_reg_fault_state,
7067 .process = gfx_v9_0_priv_reg_irq,
7068 };
7069
7070 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_inst_irq_funcs = {
7071 .set = gfx_v9_0_set_priv_inst_fault_state,
7072 .process = gfx_v9_0_priv_inst_irq,
7073 };
7074
7075 static const struct amdgpu_irq_src_funcs gfx_v9_0_cp_ecc_error_irq_funcs = {
7076 .set = gfx_v9_0_set_cp_ecc_error_state,
7077 .process = amdgpu_gfx_cp_ecc_error_irq,
7078 };
7079
7080
gfx_v9_0_set_irq_funcs(struct amdgpu_device * adev)7081 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev)
7082 {
7083 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
7084 adev->gfx.eop_irq.funcs = &gfx_v9_0_eop_irq_funcs;
7085
7086 adev->gfx.priv_reg_irq.num_types = 1;
7087 adev->gfx.priv_reg_irq.funcs = &gfx_v9_0_priv_reg_irq_funcs;
7088
7089 adev->gfx.priv_inst_irq.num_types = 1;
7090 adev->gfx.priv_inst_irq.funcs = &gfx_v9_0_priv_inst_irq_funcs;
7091
7092 adev->gfx.cp_ecc_error_irq.num_types = 2; /*C5 ECC error and C9 FUE error*/
7093 adev->gfx.cp_ecc_error_irq.funcs = &gfx_v9_0_cp_ecc_error_irq_funcs;
7094 }
7095
gfx_v9_0_set_rlc_funcs(struct amdgpu_device * adev)7096 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev)
7097 {
7098 switch (adev->ip_versions[GC_HWIP][0]) {
7099 case IP_VERSION(9, 0, 1):
7100 case IP_VERSION(9, 2, 1):
7101 case IP_VERSION(9, 4, 0):
7102 case IP_VERSION(9, 2, 2):
7103 case IP_VERSION(9, 1, 0):
7104 case IP_VERSION(9, 4, 1):
7105 case IP_VERSION(9, 3, 0):
7106 case IP_VERSION(9, 4, 2):
7107 adev->gfx.rlc.funcs = &gfx_v9_0_rlc_funcs;
7108 break;
7109 default:
7110 break;
7111 }
7112 }
7113
gfx_v9_0_set_gds_init(struct amdgpu_device * adev)7114 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev)
7115 {
7116 /* init asci gds info */
7117 switch (adev->ip_versions[GC_HWIP][0]) {
7118 case IP_VERSION(9, 0, 1):
7119 case IP_VERSION(9, 2, 1):
7120 case IP_VERSION(9, 4, 0):
7121 adev->gds.gds_size = 0x10000;
7122 break;
7123 case IP_VERSION(9, 2, 2):
7124 case IP_VERSION(9, 1, 0):
7125 case IP_VERSION(9, 4, 1):
7126 adev->gds.gds_size = 0x1000;
7127 break;
7128 case IP_VERSION(9, 4, 2):
7129 /* aldebaran removed all the GDS internal memory,
7130 * only support GWS opcode in kernel, like barrier
7131 * semaphore.etc */
7132 adev->gds.gds_size = 0;
7133 break;
7134 default:
7135 adev->gds.gds_size = 0x10000;
7136 break;
7137 }
7138
7139 switch (adev->ip_versions[GC_HWIP][0]) {
7140 case IP_VERSION(9, 0, 1):
7141 case IP_VERSION(9, 4, 0):
7142 adev->gds.gds_compute_max_wave_id = 0x7ff;
7143 break;
7144 case IP_VERSION(9, 2, 1):
7145 adev->gds.gds_compute_max_wave_id = 0x27f;
7146 break;
7147 case IP_VERSION(9, 2, 2):
7148 case IP_VERSION(9, 1, 0):
7149 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
7150 adev->gds.gds_compute_max_wave_id = 0x77; /* raven2 */
7151 else
7152 adev->gds.gds_compute_max_wave_id = 0x15f; /* raven1 */
7153 break;
7154 case IP_VERSION(9, 4, 1):
7155 adev->gds.gds_compute_max_wave_id = 0xfff;
7156 break;
7157 case IP_VERSION(9, 4, 2):
7158 /* deprecated for Aldebaran, no usage at all */
7159 adev->gds.gds_compute_max_wave_id = 0;
7160 break;
7161 default:
7162 /* this really depends on the chip */
7163 adev->gds.gds_compute_max_wave_id = 0x7ff;
7164 break;
7165 }
7166
7167 adev->gds.gws_size = 64;
7168 adev->gds.oa_size = 16;
7169 }
7170
gfx_v9_0_set_user_cu_inactive_bitmap(struct amdgpu_device * adev,u32 bitmap)7171 static void gfx_v9_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
7172 u32 bitmap)
7173 {
7174 u32 data;
7175
7176 if (!bitmap)
7177 return;
7178
7179 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7180 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7181
7182 WREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG, data);
7183 }
7184
gfx_v9_0_get_cu_active_bitmap(struct amdgpu_device * adev)7185 static u32 gfx_v9_0_get_cu_active_bitmap(struct amdgpu_device *adev)
7186 {
7187 u32 data, mask;
7188
7189 data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
7190 data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
7191
7192 data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7193 data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7194
7195 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
7196
7197 return (~data) & mask;
7198 }
7199
gfx_v9_0_get_cu_info(struct amdgpu_device * adev,struct amdgpu_cu_info * cu_info)7200 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
7201 struct amdgpu_cu_info *cu_info)
7202 {
7203 int i, j, k, counter, active_cu_number = 0;
7204 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
7205 unsigned disable_masks[4 * 4];
7206
7207 if (!adev || !cu_info)
7208 return -EINVAL;
7209
7210 /*
7211 * 16 comes from bitmap array size 4*4, and it can cover all gfx9 ASICs
7212 */
7213 if (adev->gfx.config.max_shader_engines *
7214 adev->gfx.config.max_sh_per_se > 16)
7215 return -EINVAL;
7216
7217 amdgpu_gfx_parse_disable_cu(disable_masks,
7218 adev->gfx.config.max_shader_engines,
7219 adev->gfx.config.max_sh_per_se);
7220
7221 mutex_lock(&adev->grbm_idx_mutex);
7222 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
7223 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
7224 mask = 1;
7225 ao_bitmap = 0;
7226 counter = 0;
7227 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff, 0);
7228 gfx_v9_0_set_user_cu_inactive_bitmap(
7229 adev, disable_masks[i * adev->gfx.config.max_sh_per_se + j]);
7230 bitmap = gfx_v9_0_get_cu_active_bitmap(adev);
7231
7232 /*
7233 * The bitmap(and ao_cu_bitmap) in cu_info structure is
7234 * 4x4 size array, and it's usually suitable for Vega
7235 * ASICs which has 4*2 SE/SH layout.
7236 * But for Arcturus, SE/SH layout is changed to 8*1.
7237 * To mostly reduce the impact, we make it compatible
7238 * with current bitmap array as below:
7239 * SE4,SH0 --> bitmap[0][1]
7240 * SE5,SH0 --> bitmap[1][1]
7241 * SE6,SH0 --> bitmap[2][1]
7242 * SE7,SH0 --> bitmap[3][1]
7243 */
7244 cu_info->bitmap[0][i % 4][j + i / 4] = bitmap;
7245
7246 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
7247 if (bitmap & mask) {
7248 if (counter < adev->gfx.config.max_cu_per_sh)
7249 ao_bitmap |= mask;
7250 counter ++;
7251 }
7252 mask <<= 1;
7253 }
7254 active_cu_number += counter;
7255 if (i < 2 && j < 2)
7256 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
7257 cu_info->ao_cu_bitmap[i % 4][j + i / 4] = ao_bitmap;
7258 }
7259 }
7260 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
7261 mutex_unlock(&adev->grbm_idx_mutex);
7262
7263 cu_info->number = active_cu_number;
7264 cu_info->ao_cu_mask = ao_cu_mask;
7265 cu_info->simd_per_cu = NUM_SIMD_PER_CU;
7266
7267 return 0;
7268 }
7269
7270 const struct amdgpu_ip_block_version gfx_v9_0_ip_block =
7271 {
7272 .type = AMD_IP_BLOCK_TYPE_GFX,
7273 .major = 9,
7274 .minor = 0,
7275 .rev = 0,
7276 .funcs = &gfx_v9_0_ip_funcs,
7277 };
7278