Bug Summary

File:dev/pci/drm/amd/amdgpu/gfx_v9_0.c
Warning:line 2076, column 22
Value stored to 'ring' during its initialization is never read

Annotated Source Code

Press '?' to see keyboard shortcuts

clang -cc1 -cc1 -triple amd64-unknown-openbsd7.4 -analyze -disable-free -clear-ast-before-backend -disable-llvm-verifier -discard-value-names -main-file-name gfx_v9_0.c -analyzer-checker=core -analyzer-checker=apiModeling -analyzer-checker=unix -analyzer-checker=deadcode -analyzer-checker=security.insecureAPI.UncheckedReturn -analyzer-checker=security.insecureAPI.getpw -analyzer-checker=security.insecureAPI.gets -analyzer-checker=security.insecureAPI.mktemp -analyzer-checker=security.insecureAPI.mkstemp -analyzer-checker=security.insecureAPI.vfork -analyzer-checker=nullability.NullPassedToNonnull -analyzer-checker=nullability.NullReturnedFromNonnull -analyzer-output plist -w -setup-static-analyzer -mrelocation-model static -mframe-pointer=all -relaxed-aliasing -ffp-contract=on -fno-rounding-math -mconstructor-aliases -ffreestanding -mcmodel=kernel -target-cpu x86-64 -target-feature +retpoline-indirect-calls -target-feature +retpoline-indirect-branches -target-feature -sse2 -target-feature -sse -target-feature -3dnow -target-feature -mmx -target-feature +save-args -target-feature +retpoline-external-thunk -disable-red-zone -no-implicit-float -tune-cpu generic -debugger-tuning=gdb -fcoverage-compilation-dir=/usr/src/sys/arch/amd64/compile/GENERIC.MP/obj -nostdsysteminc -nobuiltininc -resource-dir /usr/local/llvm16/lib/clang/16 -I /usr/src/sys -I /usr/src/sys/arch/amd64/compile/GENERIC.MP/obj -I /usr/src/sys/arch -I /usr/src/sys/dev/pci/drm/include -I /usr/src/sys/dev/pci/drm/include/uapi -I /usr/src/sys/dev/pci/drm/amd/include/asic_reg -I /usr/src/sys/dev/pci/drm/amd/include -I /usr/src/sys/dev/pci/drm/amd/amdgpu -I /usr/src/sys/dev/pci/drm/amd/display -I /usr/src/sys/dev/pci/drm/amd/display/include -I /usr/src/sys/dev/pci/drm/amd/display/dc -I /usr/src/sys/dev/pci/drm/amd/display/amdgpu_dm -I /usr/src/sys/dev/pci/drm/amd/pm/inc -I /usr/src/sys/dev/pci/drm/amd/pm/legacy-dpm -I /usr/src/sys/dev/pci/drm/amd/pm/swsmu -I /usr/src/sys/dev/pci/drm/amd/pm/swsmu/inc -I /usr/src/sys/dev/pci/drm/amd/pm/swsmu/smu11 -I /usr/src/sys/dev/pci/drm/amd/pm/swsmu/smu12 -I /usr/src/sys/dev/pci/drm/amd/pm/swsmu/smu13 -I /usr/src/sys/dev/pci/drm/amd/pm/powerplay/inc -I /usr/src/sys/dev/pci/drm/amd/pm/powerplay/hwmgr -I /usr/src/sys/dev/pci/drm/amd/pm/powerplay/smumgr -I /usr/src/sys/dev/pci/drm/amd/pm/swsmu/inc -I /usr/src/sys/dev/pci/drm/amd/pm/swsmu/inc/pmfw_if -I /usr/src/sys/dev/pci/drm/amd/display/dc/inc -I /usr/src/sys/dev/pci/drm/amd/display/dc/inc/hw -I /usr/src/sys/dev/pci/drm/amd/display/dc/clk_mgr -I /usr/src/sys/dev/pci/drm/amd/display/modules/inc -I /usr/src/sys/dev/pci/drm/amd/display/modules/hdcp -I /usr/src/sys/dev/pci/drm/amd/display/dmub/inc -I /usr/src/sys/dev/pci/drm/i915 -D DDB -D DIAGNOSTIC -D KTRACE -D ACCOUNTING -D KMEMSTATS -D PTRACE -D POOL_DEBUG -D CRYPTO -D SYSVMSG -D SYSVSEM -D SYSVSHM -D UVM_SWAP_ENCRYPT -D FFS -D FFS2 -D FFS_SOFTUPDATES -D UFS_DIRHASH -D QUOTA -D EXT2FS -D MFS -D NFSCLIENT -D NFSSERVER -D CD9660 -D UDF -D MSDOSFS -D FIFO -D FUSE -D SOCKET_SPLICE -D TCP_ECN -D TCP_SIGNATURE -D INET6 -D IPSEC -D PPP_BSDCOMP -D PPP_DEFLATE -D PIPEX -D MROUTING -D MPLS -D BOOT_CONFIG -D USER_PCICONF -D APERTURE -D MTRR -D NTFS -D SUSPEND -D HIBERNATE -D PCIVERBOSE -D USBVERBOSE -D WSDISPLAY_COMPAT_USL -D WSDISPLAY_COMPAT_RAWKBD -D WSDISPLAY_DEFAULTSCREENS=6 -D X86EMU -D ONEWIREVERBOSE -D MULTIPROCESSOR -D MAXUSERS=80 -D _KERNEL -O2 -Wno-pointer-sign -Wno-address-of-packed-member -Wno-constant-conversion -Wno-unused-but-set-variable -Wno-gnu-folding-constant -fdebug-compilation-dir=/usr/src/sys/arch/amd64/compile/GENERIC.MP/obj -ferror-limit 19 -fwrapv -D_RET_PROTECTOR -ret-protector -fcf-protection=branch -fgnuc-version=4.2.1 -vectorize-loops -vectorize-slp -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -fno-builtin-valloc -fno-builtin-free -fno-builtin-strdup -fno-builtin-strndup -analyzer-output=html -faddrsig -o /home/ben/Projects/scan/2024-01-11-110808-61670-1 -x c /usr/src/sys/dev/pci/drm/amd/amdgpu/gfx_v9_0.c
1/*
2 * Copyright 2016 Advanced Micro Devices, Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22 */
23
24#include <linux/delay.h>
25#include <linux/kernel.h>
26#include <linux/firmware.h>
27#include <linux/module.h>
28#include <linux/pci.h>
29
30#include "amdgpu.h"
31#include "amdgpu_gfx.h"
32#include "soc15.h"
33#include "soc15d.h"
34#include "amdgpu_atomfirmware.h"
35#include "amdgpu_pm.h"
36
37#include "gc/gc_9_0_offset.h"
38#include "gc/gc_9_0_sh_mask.h"
39
40#include "vega10_enum.h"
41
42#include "soc15_common.h"
43#include "clearstate_gfx9.h"
44#include "v9_structs.h"
45
46#include "ivsrcid/gfx/irqsrcs_gfx_9_0.h"
47
48#include "amdgpu_ras.h"
49
50#include "gfx_v9_4.h"
51#include "gfx_v9_0.h"
52#include "gfx_v9_4_2.h"
53
54#include "asic_reg/pwr/pwr_10_0_offset.h"
55#include "asic_reg/pwr/pwr_10_0_sh_mask.h"
56#include "asic_reg/gc/gc_9_0_default.h"
57
58#define GFX9_NUM_GFX_RINGS1 1
59#define GFX9_MEC_HPD_SIZE4096 4096
60#define RLCG_UCODE_LOADING_START_ADDRESS0x00002000L 0x00002000L
61#define RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET0x00000000L 0x00000000L
62
63#define mmGCEA_PROBE_MAP0x070c 0x070c
64#define mmGCEA_PROBE_MAP_BASE_IDX0 0
65
66MODULE_FIRMWARE("amdgpu/vega10_ce.bin");
67MODULE_FIRMWARE("amdgpu/vega10_pfp.bin");
68MODULE_FIRMWARE("amdgpu/vega10_me.bin");
69MODULE_FIRMWARE("amdgpu/vega10_mec.bin");
70MODULE_FIRMWARE("amdgpu/vega10_mec2.bin");
71MODULE_FIRMWARE("amdgpu/vega10_rlc.bin");
72
73MODULE_FIRMWARE("amdgpu/vega12_ce.bin");
74MODULE_FIRMWARE("amdgpu/vega12_pfp.bin");
75MODULE_FIRMWARE("amdgpu/vega12_me.bin");
76MODULE_FIRMWARE("amdgpu/vega12_mec.bin");
77MODULE_FIRMWARE("amdgpu/vega12_mec2.bin");
78MODULE_FIRMWARE("amdgpu/vega12_rlc.bin");
79
80MODULE_FIRMWARE("amdgpu/vega20_ce.bin");
81MODULE_FIRMWARE("amdgpu/vega20_pfp.bin");
82MODULE_FIRMWARE("amdgpu/vega20_me.bin");
83MODULE_FIRMWARE("amdgpu/vega20_mec.bin");
84MODULE_FIRMWARE("amdgpu/vega20_mec2.bin");
85MODULE_FIRMWARE("amdgpu/vega20_rlc.bin");
86
87MODULE_FIRMWARE("amdgpu/raven_ce.bin");
88MODULE_FIRMWARE("amdgpu/raven_pfp.bin");
89MODULE_FIRMWARE("amdgpu/raven_me.bin");
90MODULE_FIRMWARE("amdgpu/raven_mec.bin");
91MODULE_FIRMWARE("amdgpu/raven_mec2.bin");
92MODULE_FIRMWARE("amdgpu/raven_rlc.bin");
93
94MODULE_FIRMWARE("amdgpu/picasso_ce.bin");
95MODULE_FIRMWARE("amdgpu/picasso_pfp.bin");
96MODULE_FIRMWARE("amdgpu/picasso_me.bin");
97MODULE_FIRMWARE("amdgpu/picasso_mec.bin");
98MODULE_FIRMWARE("amdgpu/picasso_mec2.bin");
99MODULE_FIRMWARE("amdgpu/picasso_rlc.bin");
100MODULE_FIRMWARE("amdgpu/picasso_rlc_am4.bin");
101
102MODULE_FIRMWARE("amdgpu/raven2_ce.bin");
103MODULE_FIRMWARE("amdgpu/raven2_pfp.bin");
104MODULE_FIRMWARE("amdgpu/raven2_me.bin");
105MODULE_FIRMWARE("amdgpu/raven2_mec.bin");
106MODULE_FIRMWARE("amdgpu/raven2_mec2.bin");
107MODULE_FIRMWARE("amdgpu/raven2_rlc.bin");
108MODULE_FIRMWARE("amdgpu/raven_kicker_rlc.bin");
109
110MODULE_FIRMWARE("amdgpu/arcturus_mec.bin");
111MODULE_FIRMWARE("amdgpu/arcturus_rlc.bin");
112
113MODULE_FIRMWARE("amdgpu/renoir_ce.bin");
114MODULE_FIRMWARE("amdgpu/renoir_pfp.bin");
115MODULE_FIRMWARE("amdgpu/renoir_me.bin");
116MODULE_FIRMWARE("amdgpu/renoir_mec.bin");
117MODULE_FIRMWARE("amdgpu/renoir_rlc.bin");
118
119MODULE_FIRMWARE("amdgpu/green_sardine_ce.bin");
120MODULE_FIRMWARE("amdgpu/green_sardine_pfp.bin");
121MODULE_FIRMWARE("amdgpu/green_sardine_me.bin");
122MODULE_FIRMWARE("amdgpu/green_sardine_mec.bin");
123MODULE_FIRMWARE("amdgpu/green_sardine_mec2.bin");
124MODULE_FIRMWARE("amdgpu/green_sardine_rlc.bin");
125
126MODULE_FIRMWARE("amdgpu/aldebaran_mec.bin");
127MODULE_FIRMWARE("amdgpu/aldebaran_mec2.bin");
128MODULE_FIRMWARE("amdgpu/aldebaran_rlc.bin");
129MODULE_FIRMWARE("amdgpu/aldebaran_sjt_mec.bin");
130MODULE_FIRMWARE("amdgpu/aldebaran_sjt_mec2.bin");
131
132#define mmTCP_CHAN_STEER_0_ARCT0x0b03 0x0b03
133#define mmTCP_CHAN_STEER_0_ARCT_BASE_IDX0 0
134#define mmTCP_CHAN_STEER_1_ARCT0x0b04 0x0b04
135#define mmTCP_CHAN_STEER_1_ARCT_BASE_IDX0 0
136#define mmTCP_CHAN_STEER_2_ARCT0x0b09 0x0b09
137#define mmTCP_CHAN_STEER_2_ARCT_BASE_IDX0 0
138#define mmTCP_CHAN_STEER_3_ARCT0x0b0a 0x0b0a
139#define mmTCP_CHAN_STEER_3_ARCT_BASE_IDX0 0
140#define mmTCP_CHAN_STEER_4_ARCT0x0b0b 0x0b0b
141#define mmTCP_CHAN_STEER_4_ARCT_BASE_IDX0 0
142#define mmTCP_CHAN_STEER_5_ARCT0x0b0c 0x0b0c
143#define mmTCP_CHAN_STEER_5_ARCT_BASE_IDX0 0
144
145#define mmGOLDEN_TSC_COUNT_UPPER_Renoir0x0025 0x0025
146#define mmGOLDEN_TSC_COUNT_UPPER_Renoir_BASE_IDX1 1
147#define mmGOLDEN_TSC_COUNT_LOWER_Renoir0x0026 0x0026
148#define mmGOLDEN_TSC_COUNT_LOWER_Renoir_BASE_IDX1 1
149
150enum ta_ras_gfx_subblock {
151 /*CPC*/
152 TA_RAS_BLOCK__GFX_CPC_INDEX_START = 0,
153 TA_RAS_BLOCK__GFX_CPC_SCRATCH = TA_RAS_BLOCK__GFX_CPC_INDEX_START,
154 TA_RAS_BLOCK__GFX_CPC_UCODE,
155 TA_RAS_BLOCK__GFX_DC_STATE_ME1,
156 TA_RAS_BLOCK__GFX_DC_CSINVOC_ME1,
157 TA_RAS_BLOCK__GFX_DC_RESTORE_ME1,
158 TA_RAS_BLOCK__GFX_DC_STATE_ME2,
159 TA_RAS_BLOCK__GFX_DC_CSINVOC_ME2,
160 TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
161 TA_RAS_BLOCK__GFX_CPC_INDEX_END = TA_RAS_BLOCK__GFX_DC_RESTORE_ME2,
162 /* CPF*/
163 TA_RAS_BLOCK__GFX_CPF_INDEX_START,
164 TA_RAS_BLOCK__GFX_CPF_ROQ_ME2 = TA_RAS_BLOCK__GFX_CPF_INDEX_START,
165 TA_RAS_BLOCK__GFX_CPF_ROQ_ME1,
166 TA_RAS_BLOCK__GFX_CPF_TAG,
167 TA_RAS_BLOCK__GFX_CPF_INDEX_END = TA_RAS_BLOCK__GFX_CPF_TAG,
168 /* CPG*/
169 TA_RAS_BLOCK__GFX_CPG_INDEX_START,
170 TA_RAS_BLOCK__GFX_CPG_DMA_ROQ = TA_RAS_BLOCK__GFX_CPG_INDEX_START,
171 TA_RAS_BLOCK__GFX_CPG_DMA_TAG,
172 TA_RAS_BLOCK__GFX_CPG_TAG,
173 TA_RAS_BLOCK__GFX_CPG_INDEX_END = TA_RAS_BLOCK__GFX_CPG_TAG,
174 /* GDS*/
175 TA_RAS_BLOCK__GFX_GDS_INDEX_START,
176 TA_RAS_BLOCK__GFX_GDS_MEM = TA_RAS_BLOCK__GFX_GDS_INDEX_START,
177 TA_RAS_BLOCK__GFX_GDS_INPUT_QUEUE,
178 TA_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM,
179 TA_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM,
180 TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
181 TA_RAS_BLOCK__GFX_GDS_INDEX_END = TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM,
182 /* SPI*/
183 TA_RAS_BLOCK__GFX_SPI_SR_MEM,
184 /* SQ*/
185 TA_RAS_BLOCK__GFX_SQ_INDEX_START,
186 TA_RAS_BLOCK__GFX_SQ_SGPR = TA_RAS_BLOCK__GFX_SQ_INDEX_START,
187 TA_RAS_BLOCK__GFX_SQ_LDS_D,
188 TA_RAS_BLOCK__GFX_SQ_LDS_I,
189 TA_RAS_BLOCK__GFX_SQ_VGPR, /* VGPR = SP*/
190 TA_RAS_BLOCK__GFX_SQ_INDEX_END = TA_RAS_BLOCK__GFX_SQ_VGPR,
191 /* SQC (3 ranges)*/
192 TA_RAS_BLOCK__GFX_SQC_INDEX_START,
193 /* SQC range 0*/
194 TA_RAS_BLOCK__GFX_SQC_INDEX0_START = TA_RAS_BLOCK__GFX_SQC_INDEX_START,
195 TA_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO =
196 TA_RAS_BLOCK__GFX_SQC_INDEX0_START,
197 TA_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF,
198 TA_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO,
199 TA_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF,
200 TA_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO,
201 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF,
202 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
203 TA_RAS_BLOCK__GFX_SQC_INDEX0_END =
204 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO,
205 /* SQC range 1*/
206 TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
207 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM =
208 TA_RAS_BLOCK__GFX_SQC_INDEX1_START,
209 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO,
210 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO,
211 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM,
212 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM,
213 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO,
214 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO,
215 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM,
216 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
217 TA_RAS_BLOCK__GFX_SQC_INDEX1_END =
218 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM,
219 /* SQC range 2*/
220 TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
221 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM =
222 TA_RAS_BLOCK__GFX_SQC_INDEX2_START,
223 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO,
224 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO,
225 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM,
226 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM,
227 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO,
228 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO,
229 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM,
230 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
231 TA_RAS_BLOCK__GFX_SQC_INDEX2_END =
232 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM,
233 TA_RAS_BLOCK__GFX_SQC_INDEX_END = TA_RAS_BLOCK__GFX_SQC_INDEX2_END,
234 /* TA*/
235 TA_RAS_BLOCK__GFX_TA_INDEX_START,
236 TA_RAS_BLOCK__GFX_TA_FS_DFIFO = TA_RAS_BLOCK__GFX_TA_INDEX_START,
237 TA_RAS_BLOCK__GFX_TA_FS_AFIFO,
238 TA_RAS_BLOCK__GFX_TA_FL_LFIFO,
239 TA_RAS_BLOCK__GFX_TA_FX_LFIFO,
240 TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
241 TA_RAS_BLOCK__GFX_TA_INDEX_END = TA_RAS_BLOCK__GFX_TA_FS_CFIFO,
242 /* TCA*/
243 TA_RAS_BLOCK__GFX_TCA_INDEX_START,
244 TA_RAS_BLOCK__GFX_TCA_HOLE_FIFO = TA_RAS_BLOCK__GFX_TCA_INDEX_START,
245 TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
246 TA_RAS_BLOCK__GFX_TCA_INDEX_END = TA_RAS_BLOCK__GFX_TCA_REQ_FIFO,
247 /* TCC (5 sub-ranges)*/
248 TA_RAS_BLOCK__GFX_TCC_INDEX_START,
249 /* TCC range 0*/
250 TA_RAS_BLOCK__GFX_TCC_INDEX0_START = TA_RAS_BLOCK__GFX_TCC_INDEX_START,
251 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX0_START,
252 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1,
253 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0,
254 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1,
255 TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0,
256 TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1,
257 TA_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG,
258 TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
259 TA_RAS_BLOCK__GFX_TCC_INDEX0_END = TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG,
260 /* TCC range 1*/
261 TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
262 TA_RAS_BLOCK__GFX_TCC_IN_USE_DEC = TA_RAS_BLOCK__GFX_TCC_INDEX1_START,
263 TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
264 TA_RAS_BLOCK__GFX_TCC_INDEX1_END =
265 TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER,
266 /* TCC range 2*/
267 TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
268 TA_RAS_BLOCK__GFX_TCC_RETURN_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX2_START,
269 TA_RAS_BLOCK__GFX_TCC_RETURN_CONTROL,
270 TA_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO,
271 TA_RAS_BLOCK__GFX_TCC_WRITE_RETURN,
272 TA_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ,
273 TA_RAS_BLOCK__GFX_TCC_SRC_FIFO,
274 TA_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM,
275 TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
276 TA_RAS_BLOCK__GFX_TCC_INDEX2_END =
277 TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO,
278 /* TCC range 3*/
279 TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
280 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO = TA_RAS_BLOCK__GFX_TCC_INDEX3_START,
281 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
282 TA_RAS_BLOCK__GFX_TCC_INDEX3_END =
283 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM,
284 /* TCC range 4*/
285 TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
286 TA_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN =
287 TA_RAS_BLOCK__GFX_TCC_INDEX4_START,
288 TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
289 TA_RAS_BLOCK__GFX_TCC_INDEX4_END =
290 TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER,
291 TA_RAS_BLOCK__GFX_TCC_INDEX_END = TA_RAS_BLOCK__GFX_TCC_INDEX4_END,
292 /* TCI*/
293 TA_RAS_BLOCK__GFX_TCI_WRITE_RAM,
294 /* TCP*/
295 TA_RAS_BLOCK__GFX_TCP_INDEX_START,
296 TA_RAS_BLOCK__GFX_TCP_CACHE_RAM = TA_RAS_BLOCK__GFX_TCP_INDEX_START,
297 TA_RAS_BLOCK__GFX_TCP_LFIFO_RAM,
298 TA_RAS_BLOCK__GFX_TCP_CMD_FIFO,
299 TA_RAS_BLOCK__GFX_TCP_VM_FIFO,
300 TA_RAS_BLOCK__GFX_TCP_DB_RAM,
301 TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0,
302 TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
303 TA_RAS_BLOCK__GFX_TCP_INDEX_END = TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1,
304 /* TD*/
305 TA_RAS_BLOCK__GFX_TD_INDEX_START,
306 TA_RAS_BLOCK__GFX_TD_SS_FIFO_LO = TA_RAS_BLOCK__GFX_TD_INDEX_START,
307 TA_RAS_BLOCK__GFX_TD_SS_FIFO_HI,
308 TA_RAS_BLOCK__GFX_TD_CS_FIFO,
309 TA_RAS_BLOCK__GFX_TD_INDEX_END = TA_RAS_BLOCK__GFX_TD_CS_FIFO,
310 /* EA (3 sub-ranges)*/
311 TA_RAS_BLOCK__GFX_EA_INDEX_START,
312 /* EA range 0*/
313 TA_RAS_BLOCK__GFX_EA_INDEX0_START = TA_RAS_BLOCK__GFX_EA_INDEX_START,
314 TA_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM = TA_RAS_BLOCK__GFX_EA_INDEX0_START,
315 TA_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM,
316 TA_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM,
317 TA_RAS_BLOCK__GFX_EA_RRET_TAGMEM,
318 TA_RAS_BLOCK__GFX_EA_WRET_TAGMEM,
319 TA_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM,
320 TA_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM,
321 TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
322 TA_RAS_BLOCK__GFX_EA_INDEX0_END = TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM,
323 /* EA range 1*/
324 TA_RAS_BLOCK__GFX_EA_INDEX1_START,
325 TA_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM = TA_RAS_BLOCK__GFX_EA_INDEX1_START,
326 TA_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM,
327 TA_RAS_BLOCK__GFX_EA_IORD_CMDMEM,
328 TA_RAS_BLOCK__GFX_EA_IOWR_CMDMEM,
329 TA_RAS_BLOCK__GFX_EA_IOWR_DATAMEM,
330 TA_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM,
331 TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
332 TA_RAS_BLOCK__GFX_EA_INDEX1_END = TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM,
333 /* EA range 2*/
334 TA_RAS_BLOCK__GFX_EA_INDEX2_START,
335 TA_RAS_BLOCK__GFX_EA_MAM_D0MEM = TA_RAS_BLOCK__GFX_EA_INDEX2_START,
336 TA_RAS_BLOCK__GFX_EA_MAM_D1MEM,
337 TA_RAS_BLOCK__GFX_EA_MAM_D2MEM,
338 TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
339 TA_RAS_BLOCK__GFX_EA_INDEX2_END = TA_RAS_BLOCK__GFX_EA_MAM_D3MEM,
340 TA_RAS_BLOCK__GFX_EA_INDEX_END = TA_RAS_BLOCK__GFX_EA_INDEX2_END,
341 /* UTC VM L2 bank*/
342 TA_RAS_BLOCK__UTC_VML2_BANK_CACHE,
343 /* UTC VM walker*/
344 TA_RAS_BLOCK__UTC_VML2_WALKER,
345 /* UTC ATC L2 2MB cache*/
346 TA_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK,
347 /* UTC ATC L2 4KB cache*/
348 TA_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK,
349 TA_RAS_BLOCK__GFX_MAX
350};
351
352struct ras_gfx_subblock {
353 unsigned char *name;
354 int ta_subblock;
355 int hw_supported_error_type;
356 int sw_supported_error_type;
357};
358
359#define AMDGPU_RAS_SUB_BLOCK(subblock, a, b, c, d, e, f, g, h)[AMDGPU_RAS_BLOCK__subblock] = { "subblock", TA_RAS_BLOCK__subblock
, ((a) | ((b) << 1) | ((c) << 2) | ((d) << 3
)), (((e) << 1) | ((f) << 3) | (g) | ((h) <<
2)), } \
360 [AMDGPU_RAS_BLOCK__##subblock] = { \
361 #subblock, \
362 TA_RAS_BLOCK__##subblock, \
363 ((a) | ((b) << 1) | ((c) << 2) | ((d) << 3)), \
364 (((e) << 1) | ((f) << 3) | (g) | ((h) << 2)), \
365 }
366
367static const struct ras_gfx_subblock ras_gfx_subblocks[] = {
368 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_SCRATCH, 0, 1, 1, 1, 1, 0, 0, 1)[AMDGPU_RAS_BLOCK__GFX_CPC_SCRATCH] = { "GFX_CPC_SCRATCH", TA_RAS_BLOCK__GFX_CPC_SCRATCH
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((1) << 1) | ((0) << 3) | (0) | ((1) <<
2)), },
369 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_UCODE, 0, 1, 1, 1, 1, 0, 0, 1)[AMDGPU_RAS_BLOCK__GFX_CPC_UCODE] = { "GFX_CPC_UCODE", TA_RAS_BLOCK__GFX_CPC_UCODE
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((1) << 1) | ((0) << 3) | (0) | ((1) <<
2)), },
370 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME1, 1, 0, 0, 1, 0, 0, 1, 0)[AMDGPU_RAS_BLOCK__GFX_DC_STATE_ME1] = { "GFX_DC_STATE_ME1", TA_RAS_BLOCK__GFX_DC_STATE_ME1
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (1) | ((0) <<
2)), },
371 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME1, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_DC_CSINVOC_ME1] = { "GFX_DC_CSINVOC_ME1"
, TA_RAS_BLOCK__GFX_DC_CSINVOC_ME1, ((1) | ((0) << 1) |
((0) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
372 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME1, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_DC_RESTORE_ME1] = { "GFX_DC_RESTORE_ME1"
, TA_RAS_BLOCK__GFX_DC_RESTORE_ME1, ((1) | ((0) << 1) |
((0) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
373 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME2, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_DC_STATE_ME2] = { "GFX_DC_STATE_ME2", TA_RAS_BLOCK__GFX_DC_STATE_ME2
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
374 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME2, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_DC_CSINVOC_ME2] = { "GFX_DC_CSINVOC_ME2"
, TA_RAS_BLOCK__GFX_DC_CSINVOC_ME2, ((1) | ((0) << 1) |
((0) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
375 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME2, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_DC_RESTORE_ME2] = { "GFX_DC_RESTORE_ME2"
, TA_RAS_BLOCK__GFX_DC_RESTORE_ME2, ((1) | ((0) << 1) |
((0) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
376 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME2, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_CPF_ROQ_ME2] = { "GFX_CPF_ROQ_ME2", TA_RAS_BLOCK__GFX_CPF_ROQ_ME2
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
377 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME1, 1, 0, 0, 1, 0, 0, 1, 0)[AMDGPU_RAS_BLOCK__GFX_CPF_ROQ_ME1] = { "GFX_CPF_ROQ_ME1", TA_RAS_BLOCK__GFX_CPF_ROQ_ME1
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (1) | ((0) <<
2)), },
378 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_TAG, 0, 1, 1, 1, 1, 0, 0, 1)[AMDGPU_RAS_BLOCK__GFX_CPF_TAG] = { "GFX_CPF_TAG", TA_RAS_BLOCK__GFX_CPF_TAG
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((1) << 1) | ((0) << 3) | (0) | ((1) <<
2)), },
379 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_ROQ, 1, 0, 0, 1, 0, 0, 1, 0)[AMDGPU_RAS_BLOCK__GFX_CPG_DMA_ROQ] = { "GFX_CPG_DMA_ROQ", TA_RAS_BLOCK__GFX_CPG_DMA_ROQ
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (1) | ((0) <<
2)), },
380 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_TAG, 0, 1, 1, 1, 0, 1, 0, 1)[AMDGPU_RAS_BLOCK__GFX_CPG_DMA_TAG] = { "GFX_CPG_DMA_TAG", TA_RAS_BLOCK__GFX_CPG_DMA_TAG
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((0) << 1) | ((1) << 3) | (0) | ((1) <<
2)), },
381 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_TAG, 0, 1, 1, 1, 1, 1, 0, 1)[AMDGPU_RAS_BLOCK__GFX_CPG_TAG] = { "GFX_CPG_TAG", TA_RAS_BLOCK__GFX_CPG_TAG
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((1) << 1) | ((1) << 3) | (0) | ((1) <<
2)), },
382 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_MEM, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_GDS_MEM] = { "GFX_GDS_MEM", TA_RAS_BLOCK__GFX_GDS_MEM
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
383 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_INPUT_QUEUE, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_GDS_INPUT_QUEUE] = { "GFX_GDS_INPUT_QUEUE"
, TA_RAS_BLOCK__GFX_GDS_INPUT_QUEUE, ((1) | ((0) << 1) |
((0) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
384 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_CMD_RAM_MEM, 0, 1, 1, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM] = { "GFX_GDS_OA_PHY_CMD_RAM_MEM"
, TA_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
385 0)[AMDGPU_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM] = { "GFX_GDS_OA_PHY_CMD_RAM_MEM"
, TA_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
386 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_DATA_RAM_MEM, 1, 0, 0, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM] = { "GFX_GDS_OA_PHY_DATA_RAM_MEM"
, TA_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
387 0)[AMDGPU_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM] = { "GFX_GDS_OA_PHY_DATA_RAM_MEM"
, TA_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
388 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PIPE_MEM, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM] = { "GFX_GDS_OA_PIPE_MEM"
, TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM, ((0) | ((1) << 1) |
((1) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
389 AMDGPU_RAS_SUB_BLOCK(GFX_SPI_SR_MEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_SPI_SR_MEM] = { "GFX_SPI_SR_MEM", TA_RAS_BLOCK__GFX_SPI_SR_MEM
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
390 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_SGPR, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_SQ_SGPR] = { "GFX_SQ_SGPR", TA_RAS_BLOCK__GFX_SQ_SGPR
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
391 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_D, 0, 1, 1, 1, 1, 0, 0, 1)[AMDGPU_RAS_BLOCK__GFX_SQ_LDS_D] = { "GFX_SQ_LDS_D", TA_RAS_BLOCK__GFX_SQ_LDS_D
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((1) << 1) | ((0) << 3) | (0) | ((1) <<
2)), },
392 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_I, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_SQ_LDS_I] = { "GFX_SQ_LDS_I", TA_RAS_BLOCK__GFX_SQ_LDS_I
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
393 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_VGPR, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_SQ_VGPR] = { "GFX_SQ_VGPR", TA_RAS_BLOCK__GFX_SQ_VGPR
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
394 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 1)[AMDGPU_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO] = { "GFX_SQC_INST_UTCL1_LFIFO"
, TA_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((1) << 2)), },
395 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF] = { "GFX_SQC_DATA_CU0_WRITE_DATA_BUF"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
396 0, 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF] = { "GFX_SQC_DATA_CU0_WRITE_DATA_BUF"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
397 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO] = { "GFX_SQC_DATA_CU0_UTCL1_LFIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
398 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO] = { "GFX_SQC_DATA_CU0_UTCL1_LFIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
399 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF] = { "GFX_SQC_DATA_CU1_WRITE_DATA_BUF"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
400 0, 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF] = { "GFX_SQC_DATA_CU1_WRITE_DATA_BUF"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
401 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_UTCL1_LFIFO, 0, 1, 1, 1, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO] = { "GFX_SQC_DATA_CU1_UTCL1_LFIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
402 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO] = { "GFX_SQC_DATA_CU1_UTCL1_LFIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
403 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF] = { "GFX_SQC_DATA_CU2_WRITE_DATA_BUF"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
404 0, 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF] = { "GFX_SQC_DATA_CU2_WRITE_DATA_BUF"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
405 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO] = { "GFX_SQC_DATA_CU2_UTCL1_LFIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
406 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO] = { "GFX_SQC_DATA_CU2_UTCL1_LFIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
407 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM] = { "GFX_SQC_INST_BANKA_TAG_RAM"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((1) << 2)), }
408 1)[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM] = { "GFX_SQC_INST_BANKA_TAG_RAM"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((1) << 2)), },
409 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO] = { "GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO, ((1) | ((
0) << 1) | ((0) << 2) | ((1) << 3)), (((0) <<
1) | ((0) << 3) | (0) | ((0) << 2)), }
410 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO] = { "GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO, ((1) | ((
0) << 1) | ((0) << 2) | ((1) << 3)), (((0) <<
1) | ((0) << 3) | (0) | ((0) << 2)), },
411 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO] = { "GFX_SQC_INST_BANKA_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
412 0)[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO] = { "GFX_SQC_INST_BANKA_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
413 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM] = { "GFX_SQC_INST_BANKA_BANK_RAM"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
414 0)[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM] = { "GFX_SQC_INST_BANKA_BANK_RAM"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
415 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM] = { "GFX_SQC_DATA_BANKA_TAG_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
416 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM] = { "GFX_SQC_DATA_BANKA_TAG_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
417 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO] = { "GFX_SQC_DATA_BANKA_HIT_FIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
418 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO] = { "GFX_SQC_DATA_BANKA_HIT_FIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
419 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO] = { "GFX_SQC_DATA_BANKA_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
420 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO] = { "GFX_SQC_DATA_BANKA_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
421 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM] = { "GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM, ((1) | ((0)
<< 1) | ((0) << 2) | ((1) << 3)), (((0) <<
1) | ((0) << 3) | (0) | ((0) << 2)), }
422 0, 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM] = { "GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM, ((1) | ((0)
<< 1) | ((0) << 2) | ((1) << 3)), (((0) <<
1) | ((0) << 3) | (0) | ((0) << 2)), },
423 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM] = { "GFX_SQC_DATA_BANKA_BANK_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
424 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM] = { "GFX_SQC_DATA_BANKA_BANK_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
425 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_TAG_RAM, 0, 1, 1, 1, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM] = { "GFX_SQC_INST_BANKB_TAG_RAM"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
426 0)[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM] = { "GFX_SQC_INST_BANKB_TAG_RAM"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
427 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO] = { "GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO, ((1) | ((
0) << 1) | ((0) << 2) | ((1) << 3)), (((0) <<
1) | ((0) << 3) | (0) | ((0) << 2)), }
428 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO] = { "GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO, ((1) | ((
0) << 1) | ((0) << 2) | ((1) << 3)), (((0) <<
1) | ((0) << 3) | (0) | ((0) << 2)), },
429 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO] = { "GFX_SQC_INST_BANKB_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
430 0)[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO] = { "GFX_SQC_INST_BANKB_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
431 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM] = { "GFX_SQC_INST_BANKB_BANK_RAM"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
432 0)[AMDGPU_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM] = { "GFX_SQC_INST_BANKB_BANK_RAM"
, TA_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
433 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_TAG_RAM, 0, 1, 1, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM] = { "GFX_SQC_DATA_BANKB_TAG_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
434 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM] = { "GFX_SQC_DATA_BANKB_TAG_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
435 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO] = { "GFX_SQC_DATA_BANKB_HIT_FIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
436 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO] = { "GFX_SQC_DATA_BANKB_HIT_FIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
437 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO] = { "GFX_SQC_DATA_BANKB_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
438 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO] = { "GFX_SQC_DATA_BANKB_MISS_FIFO"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
439 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM] = { "GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM, ((1) | ((0)
<< 1) | ((0) << 2) | ((1) << 3)), (((0) <<
1) | ((0) << 3) | (0) | ((0) << 2)), }
440 0, 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM] = { "GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM, ((1) | ((0)
<< 1) | ((0) << 2) | ((1) << 3)), (((0) <<
1) | ((0) << 3) | (0) | ((0) << 2)), },
441 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM] = { "GFX_SQC_DATA_BANKB_BANK_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
442 0)[AMDGPU_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM] = { "GFX_SQC_DATA_BANKB_BANK_RAM"
, TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
443 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_DFIFO, 0, 1, 1, 1, 1, 0, 0, 1)[AMDGPU_RAS_BLOCK__GFX_TA_FS_DFIFO] = { "GFX_TA_FS_DFIFO", TA_RAS_BLOCK__GFX_TA_FS_DFIFO
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((1) << 1) | ((0) << 3) | (0) | ((1) <<
2)), },
444 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_AFIFO, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TA_FS_AFIFO] = { "GFX_TA_FS_AFIFO", TA_RAS_BLOCK__GFX_TA_FS_AFIFO
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
445 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FL_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TA_FL_LFIFO] = { "GFX_TA_FL_LFIFO", TA_RAS_BLOCK__GFX_TA_FL_LFIFO
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
446 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FX_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TA_FX_LFIFO] = { "GFX_TA_FX_LFIFO", TA_RAS_BLOCK__GFX_TA_FX_LFIFO
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
447 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_CFIFO, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TA_FS_CFIFO] = { "GFX_TA_FS_CFIFO", TA_RAS_BLOCK__GFX_TA_FS_CFIFO
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
448 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_HOLE_FIFO, 1, 0, 0, 1, 0, 1, 1, 0)[AMDGPU_RAS_BLOCK__GFX_TCA_HOLE_FIFO] = { "GFX_TCA_HOLE_FIFO"
, TA_RAS_BLOCK__GFX_TCA_HOLE_FIFO, ((1) | ((0) << 1) | (
(0) << 2) | ((1) << 3)), (((0) << 1) | ((1)
<< 3) | (1) | ((0) << 2)), },
449 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_REQ_FIFO, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCA_REQ_FIFO] = { "GFX_TCA_REQ_FIFO", TA_RAS_BLOCK__GFX_TCA_REQ_FIFO
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
450 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA, 0, 1, 1, 1, 1, 0, 0, 1)[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA] = { "GFX_TCC_CACHE_DATA"
, TA_RAS_BLOCK__GFX_TCC_CACHE_DATA, ((0) | ((1) << 1) |
((1) << 2) | ((1) << 3)), (((1) << 1) | ((
0) << 3) | (0) | ((1) << 2)), },
451 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_0_1, 0, 1, 1, 1, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1] = { "GFX_TCC_CACHE_DATA_BANK_0_1"
, TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((1) << 2)), }
452 1)[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1] = { "GFX_TCC_CACHE_DATA_BANK_0_1"
, TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((1) << 2)), },
453 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_0, 0, 1, 1, 1, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0] = { "GFX_TCC_CACHE_DATA_BANK_1_0"
, TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((1) << 2)), }
454 1)[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0] = { "GFX_TCC_CACHE_DATA_BANK_1_0"
, TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((1) << 2)), },
455 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_1, 0, 1, 1, 1, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1] = { "GFX_TCC_CACHE_DATA_BANK_1_1"
, TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((1) << 2)), }
456 1)[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1] = { "GFX_TCC_CACHE_DATA_BANK_1_1"
, TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((1) << 1)
| ((0) << 3) | (0) | ((1) << 2)), },
457 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_0, 0, 1, 1, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0] = { "GFX_TCC_CACHE_DIRTY_BANK_0"
, TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
458 0)[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0] = { "GFX_TCC_CACHE_DIRTY_BANK_0"
, TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
459 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_1, 0, 1, 1, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1] = { "GFX_TCC_CACHE_DIRTY_BANK_1"
, TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
460 0)[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1] = { "GFX_TCC_CACHE_DIRTY_BANK_1"
, TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
461 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_HIGH_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG] = { "GFX_TCC_HIGH_RATE_TAG"
, TA_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG, ((0) | ((1) << 1
) | ((1) << 2) | ((1) << 3)), (((0) << 1) |
((0) << 3) | (0) | ((0) << 2)), },
462 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LOW_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG] = { "GFX_TCC_LOW_RATE_TAG"
, TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG, ((0) | ((1) << 1)
| ((1) << 2) | ((1) << 3)), (((0) << 1) | (
(0) << 3) | (0) | ((0) << 2)), },
463 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_DEC, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_IN_USE_DEC] = { "GFX_TCC_IN_USE_DEC"
, TA_RAS_BLOCK__GFX_TCC_IN_USE_DEC, ((1) | ((0) << 1) |
((0) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
464 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_TRANSFER, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER] = { "GFX_TCC_IN_USE_TRANSFER"
, TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
465 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_DATA, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_RETURN_DATA] = { "GFX_TCC_RETURN_DATA"
, TA_RAS_BLOCK__GFX_TCC_RETURN_DATA, ((1) | ((0) << 1) |
((0) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
466 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_CONTROL, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_RETURN_CONTROL] = { "GFX_TCC_RETURN_CONTROL"
, TA_RAS_BLOCK__GFX_TCC_RETURN_CONTROL, ((1) | ((0) << 1
) | ((0) << 2) | ((1) << 3)), (((0) << 1) |
((0) << 3) | (0) | ((0) << 2)), },
467 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_UC_ATOMIC_FIFO, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO] = { "GFX_TCC_UC_ATOMIC_FIFO"
, TA_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO, ((1) | ((0) << 1
) | ((0) << 2) | ((1) << 3)), (((0) << 1) |
((0) << 3) | (0) | ((0) << 2)), },
468 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_RETURN, 1, 0, 0, 1, 0, 1, 1, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_WRITE_RETURN] = { "GFX_TCC_WRITE_RETURN"
, TA_RAS_BLOCK__GFX_TCC_WRITE_RETURN, ((1) | ((0) << 1)
| ((0) << 2) | ((1) << 3)), (((0) << 1) | (
(1) << 3) | (1) | ((0) << 2)), },
469 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_CACHE_READ, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ] = { "GFX_TCC_WRITE_CACHE_READ"
, TA_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
470 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_SRC_FIFO] = { "GFX_TCC_SRC_FIFO", TA_RAS_BLOCK__GFX_TCC_SRC_FIFO
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
471 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 1, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM] = { "GFX_TCC_SRC_FIFO_NEXT_RAM"
, TA_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (1) | ((0) << 2)), },
472 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_TAG_PROBE_FIFO, 1, 0, 0, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO] = { "GFX_TCC_CACHE_TAG_PROBE_FIFO"
, TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
473 0)[AMDGPU_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO] = { "GFX_TCC_CACHE_TAG_PROBE_FIFO"
, TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
474 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO] = { "GFX_TCC_LATENCY_FIFO"
, TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO, ((1) | ((0) << 1)
| ((0) << 2) | ((1) << 3)), (((0) << 1) | (
(0) << 3) | (0) | ((0) << 2)), },
475 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM] = { "GFX_TCC_LATENCY_FIFO_NEXT_RAM"
, TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
476 0)[AMDGPU_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM] = { "GFX_TCC_LATENCY_FIFO_NEXT_RAM"
, TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
477 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRRET_TAG_WRITE_RETURN, 1, 0, 0, 1, 0, 0,[AMDGPU_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN] = { "GFX_TCC_WRRET_TAG_WRITE_RETURN"
, TA_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
478 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN] = { "GFX_TCC_WRRET_TAG_WRITE_RETURN"
, TA_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
479 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_ATOMIC_RETURN_BUFFER, 1, 0, 0, 1, 0, 0, 0,[AMDGPU_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER] = { "GFX_TCC_ATOMIC_RETURN_BUFFER"
, TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), }
480 0)[AMDGPU_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER] = { "GFX_TCC_ATOMIC_RETURN_BUFFER"
, TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
481 AMDGPU_RAS_SUB_BLOCK(GFX_TCI_WRITE_RAM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCI_WRITE_RAM] = { "GFX_TCI_WRITE_RAM"
, TA_RAS_BLOCK__GFX_TCI_WRITE_RAM, ((1) | ((0) << 1) | (
(0) << 2) | ((1) << 3)), (((0) << 1) | ((0)
<< 3) | (0) | ((0) << 2)), },
482 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CACHE_RAM, 0, 1, 1, 1, 1, 0, 0, 1)[AMDGPU_RAS_BLOCK__GFX_TCP_CACHE_RAM] = { "GFX_TCP_CACHE_RAM"
, TA_RAS_BLOCK__GFX_TCP_CACHE_RAM, ((0) | ((1) << 1) | (
(1) << 2) | ((1) << 3)), (((1) << 1) | ((0)
<< 3) | (0) | ((1) << 2)), },
483 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_LFIFO_RAM, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCP_LFIFO_RAM] = { "GFX_TCP_LFIFO_RAM"
, TA_RAS_BLOCK__GFX_TCP_LFIFO_RAM, ((0) | ((1) << 1) | (
(1) << 2) | ((1) << 3)), (((0) << 1) | ((0)
<< 3) | (0) | ((0) << 2)), },
484 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CMD_FIFO, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCP_CMD_FIFO] = { "GFX_TCP_CMD_FIFO", TA_RAS_BLOCK__GFX_TCP_CMD_FIFO
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
485 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_VM_FIFO, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCP_VM_FIFO] = { "GFX_TCP_VM_FIFO", TA_RAS_BLOCK__GFX_TCP_VM_FIFO
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
486 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_DB_RAM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCP_DB_RAM] = { "GFX_TCP_DB_RAM", TA_RAS_BLOCK__GFX_TCP_DB_RAM
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
487 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO0, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0] = { "GFX_TCP_UTCL1_LFIFO0"
, TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0, ((0) | ((1) << 1)
| ((1) << 2) | ((1) << 3)), (((0) << 1) | (
(0) << 3) | (0) | ((0) << 2)), },
488 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO1, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1] = { "GFX_TCP_UTCL1_LFIFO1"
, TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1, ((0) | ((1) << 1)
| ((1) << 2) | ((1) << 3)), (((0) << 1) | (
(0) << 3) | (0) | ((0) << 2)), },
489 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_LO, 0, 1, 1, 1, 1, 0, 0, 1)[AMDGPU_RAS_BLOCK__GFX_TD_SS_FIFO_LO] = { "GFX_TD_SS_FIFO_LO"
, TA_RAS_BLOCK__GFX_TD_SS_FIFO_LO, ((0) | ((1) << 1) | (
(1) << 2) | ((1) << 3)), (((1) << 1) | ((0)
<< 3) | (0) | ((1) << 2)), },
490 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_HI, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TD_SS_FIFO_HI] = { "GFX_TD_SS_FIFO_HI"
, TA_RAS_BLOCK__GFX_TD_SS_FIFO_HI, ((0) | ((1) << 1) | (
(1) << 2) | ((1) << 3)), (((0) << 1) | ((0)
<< 3) | (0) | ((0) << 2)), },
491 AMDGPU_RAS_SUB_BLOCK(GFX_TD_CS_FIFO, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_TD_CS_FIFO] = { "GFX_TD_CS_FIFO", TA_RAS_BLOCK__GFX_TD_CS_FIFO
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
492 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_CMDMEM, 0, 1, 1, 1, 1, 0, 0, 1)[AMDGPU_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM] = { "GFX_EA_DRAMRD_CMDMEM"
, TA_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM, ((0) | ((1) << 1)
| ((1) << 2) | ((1) << 3)), (((1) << 1) | (
(0) << 3) | (0) | ((1) << 2)), },
493 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM] = { "GFX_EA_DRAMWR_CMDMEM"
, TA_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM, ((0) | ((1) << 1)
| ((1) << 2) | ((1) << 3)), (((0) << 1) | (
(0) << 3) | (0) | ((0) << 2)), },
494 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM] = { "GFX_EA_DRAMWR_DATAMEM"
, TA_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM, ((0) | ((1) << 1
) | ((1) << 2) | ((1) << 3)), (((0) << 1) |
((0) << 3) | (0) | ((0) << 2)), },
495 AMDGPU_RAS_SUB_BLOCK(GFX_EA_RRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_RRET_TAGMEM] = { "GFX_EA_RRET_TAGMEM"
, TA_RAS_BLOCK__GFX_EA_RRET_TAGMEM, ((0) | ((1) << 1) |
((1) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
496 AMDGPU_RAS_SUB_BLOCK(GFX_EA_WRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_WRET_TAGMEM] = { "GFX_EA_WRET_TAGMEM"
, TA_RAS_BLOCK__GFX_EA_WRET_TAGMEM, ((0) | ((1) << 1) |
((1) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
497 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM] = { "GFX_EA_GMIRD_CMDMEM"
, TA_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM, ((0) | ((1) << 1) |
((1) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
498 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM] = { "GFX_EA_GMIWR_CMDMEM"
, TA_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM, ((0) | ((1) << 1) |
((1) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
499 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM] = { "GFX_EA_GMIWR_DATAMEM"
, TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM, ((0) | ((1) << 1)
| ((1) << 2) | ((1) << 3)), (((0) << 1) | (
(0) << 3) | (0) | ((0) << 2)), },
500 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM] = { "GFX_EA_DRAMRD_PAGEMEM"
, TA_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM, ((1) | ((0) << 1
) | ((0) << 2) | ((1) << 3)), (((0) << 1) |
((0) << 3) | (0) | ((0) << 2)), },
501 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM] = { "GFX_EA_DRAMWR_PAGEMEM"
, TA_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM, ((1) | ((0) << 1
) | ((0) << 2) | ((1) << 3)), (((0) << 1) |
((0) << 3) | (0) | ((0) << 2)), },
502 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IORD_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_IORD_CMDMEM] = { "GFX_EA_IORD_CMDMEM"
, TA_RAS_BLOCK__GFX_EA_IORD_CMDMEM, ((1) | ((0) << 1) |
((0) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
503 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_IOWR_CMDMEM] = { "GFX_EA_IOWR_CMDMEM"
, TA_RAS_BLOCK__GFX_EA_IOWR_CMDMEM, ((1) | ((0) << 1) |
((0) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
504 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_DATAMEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_IOWR_DATAMEM] = { "GFX_EA_IOWR_DATAMEM"
, TA_RAS_BLOCK__GFX_EA_IOWR_DATAMEM, ((1) | ((0) << 1) |
((0) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
505 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM] = { "GFX_EA_GMIRD_PAGEMEM"
, TA_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM, ((1) | ((0) << 1)
| ((0) << 2) | ((1) << 3)), (((0) << 1) | (
(0) << 3) | (0) | ((0) << 2)), },
506 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM] = { "GFX_EA_GMIWR_PAGEMEM"
, TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM, ((1) | ((0) << 1)
| ((0) << 2) | ((1) << 3)), (((0) << 1) | (
(0) << 3) | (0) | ((0) << 2)), },
507 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D0MEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_MAM_D0MEM] = { "GFX_EA_MAM_D0MEM", TA_RAS_BLOCK__GFX_EA_MAM_D0MEM
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
508 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D1MEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_MAM_D1MEM] = { "GFX_EA_MAM_D1MEM", TA_RAS_BLOCK__GFX_EA_MAM_D1MEM
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
509 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D2MEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_MAM_D2MEM] = { "GFX_EA_MAM_D2MEM", TA_RAS_BLOCK__GFX_EA_MAM_D2MEM
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
510 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D3MEM, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__GFX_EA_MAM_D3MEM] = { "GFX_EA_MAM_D3MEM", TA_RAS_BLOCK__GFX_EA_MAM_D3MEM
, ((1) | ((0) << 1) | ((0) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
511 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_BANK_CACHE, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__UTC_VML2_BANK_CACHE] = { "UTC_VML2_BANK_CACHE"
, TA_RAS_BLOCK__UTC_VML2_BANK_CACHE, ((0) | ((1) << 1) |
((1) << 2) | ((1) << 3)), (((0) << 1) | ((
0) << 3) | (0) | ((0) << 2)), },
512 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_WALKER, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__UTC_VML2_WALKER] = { "UTC_VML2_WALKER", TA_RAS_BLOCK__UTC_VML2_WALKER
, ((0) | ((1) << 1) | ((1) << 2) | ((1) << 3
)), (((0) << 1) | ((0) << 3) | (0) | ((0) <<
2)), },
513 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_2M_BANK, 1, 0, 0, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK] = { "UTC_ATCL2_CACHE_2M_BANK"
, TA_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK, ((1) | ((0) <<
1) | ((0) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
514 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_4K_BANK, 0, 1, 1, 1, 0, 0, 0, 0)[AMDGPU_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK] = { "UTC_ATCL2_CACHE_4K_BANK"
, TA_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK, ((0) | ((1) <<
1) | ((1) << 2) | ((1) << 3)), (((0) << 1)
| ((0) << 3) | (0) | ((0) << 2)), },
515};
516
517static const struct soc15_reg_golden golden_settings_gc_9_0[] =
518{
519 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400){ GC_HWIP, 0, 0, 0x060d, 0xf00fffff, 0x00000400 },
520 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG3, 0x80000000, 0x80000000){ GC_HWIP, 0, 0, 0x060e, 0x80000000, 0x80000000 },
521 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000){ GC_HWIP, 0, 0, 0x0640, 0x0000000f, 0x00000000 },
522 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024){ GC_HWIP, 0, 0, 0x02cf, 0x00000003, 0x82400024 },
523 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001){ GC_HWIP, 0, 0, 0x02fc, 0x3fffffff, 0x00000001 },
524 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000){ GC_HWIP, 0, 1, 0x2281, 0x0000ff0f, 0x00000000 },
525 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000){ GC_HWIP, 0, 0, 0x030d, 0x00001000, 0x00001000 },
526 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800){ GC_HWIP, 0, 0, 0x11dc, 0x0007ffff, 0x00000800 },
527 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800){ GC_HWIP, 0, 0, 0x11dd, 0x0007ffff, 0x00000800 },
528 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x00ffff87){ GC_HWIP, 0, 0, 0x11e6, 0x01ffffff, 0x00ffff87 },
529 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x00ffff8f){ GC_HWIP, 0, 0, 0x11e7, 0x01ffffff, 0x00ffff8f },
530 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000){ GC_HWIP, 0, 0, 0x0301, 0x03000000, 0x020a2000 },
531 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000){ GC_HWIP, 0, 0, 0x0542, 0xfffffeef, 0x010b0000 },
532 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68){ GC_HWIP, 0, 0, 0x0b04, 0xffffffff, 0x4a2c0e68 },
533 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197){ GC_HWIP, 0, 0, 0x0b03, 0xffffffff, 0xb5d3f197 },
534 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000){ GC_HWIP, 0, 0, 0x0231, 0x3fff3af3, 0x19200000 },
535 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff){ GC_HWIP, 0, 0, 0x0269, 0x00000fff, 0x000003ff },
536 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800){ GC_HWIP, 0, 0, 0x10bd, 0x00000800, 0x00000800 },
537 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800){ GC_HWIP, 0, 0, 0x10be, 0x00000800, 0x00000800 },
538 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000){ GC_HWIP, 0, 0, 0x107f, 0x00008000, 0x00008000 }
539};
540
541static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] =
542{
543 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0x0000f000, 0x00012107){ GC_HWIP, 0, 0, 0x0680, 0x0000f000, 0x00012107 },
544 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000){ GC_HWIP, 0, 0, 0x0683, 0x30000000, 0x10000000 },
545 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 0, 0x103d, 0x08000000, 0x08000080 },
546 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 0, 0x103e, 0x08000000, 0x08000080 },
547 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 0, 0x103c, 0x08000000, 0x08000080 },
548 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042){ GC_HWIP, 0, 0, 0x063e, 0xffff77ff, 0x2a114042 },
549 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x2a114042){ GC_HWIP, 0, 0, 0x0642, 0xffff77ff, 0x2a114042 },
550 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 0, 0x0246, 0x08000000, 0x08000080 },
551 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0x00008000, 0x00048000){ GC_HWIP, 0, 0, 0x02fd, 0x00008000, 0x00048000 },
552 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080){ GC_HWIP, 0, 1, 0x4cb2, 0x08000000, 0x08000080 },
553 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080){ GC_HWIP, 0, 1, 0x4cb3, 0x08000000, 0x08000080 },
554 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080){ GC_HWIP, 0, 1, 0x4cb4, 0x08000000, 0x08000080 },
555 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 1, 0x4ccd, 0x08000000, 0x08000080 },
556 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 1, 0x4cb5, 0x08000000, 0x08000080 },
557 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000){ GC_HWIP, 0, 0, 0x078c, 0x00030000, 0x00020000 },
558 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x01000107){ GC_HWIP, 0, 1, 0x2441, 0x0000000f, 0x01000107 },
559 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x00001800, 0x00000800){ GC_HWIP, 0, 0, 0x0525, 0x00001800, 0x00000800 },
560 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 0, 0x0243, 0x08000000, 0x08000080 }
561};
562
563static const struct soc15_reg_golden golden_settings_gc_9_0_vg20[] =
564{
565 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x0f000080, 0x04000080){ GC_HWIP, 0, 0, 0x0688, 0x0f000080, 0x04000080 },
566 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000){ GC_HWIP, 0, 0, 0x0682, 0x0f000000, 0x0a000000 },
567 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000){ GC_HWIP, 0, 0, 0x0683, 0x30000000, 0x10000000 },
568 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x22014042){ GC_HWIP, 0, 0, 0x063e, 0xf3e777ff, 0x22014042 },
569 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x22014042){ GC_HWIP, 0, 0, 0x0642, 0xf3e777ff, 0x22014042 },
570 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0x00003e00, 0x00000400){ GC_HWIP, 0, 0, 0x060d, 0x00003e00, 0x00000400 },
571 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xff840000, 0x04040000){ GC_HWIP, 0, 0, 0x02fd, 0xff840000, 0x04040000 },
572 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00030000){ GC_HWIP, 0, 0, 0x078c, 0x00030000, 0x00030000 },
573 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff010f, 0x01000107){ GC_HWIP, 0, 1, 0x2441, 0xffff010f, 0x01000107 },
574 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0x000b0000, 0x000b0000){ GC_HWIP, 0, 0, 0x0542, 0x000b0000, 0x000b0000 },
575 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01000000, 0x01000000){ GC_HWIP, 0, 0, 0x0525, 0x01000000, 0x01000000 }
576};
577
578static const struct soc15_reg_golden golden_settings_gc_9_1[] =
579{
580 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104){ GC_HWIP, 0, 0, 0x0680, 0xfffdf3cf, 0x00014104 },
581 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 0, 0x103d, 0x08000000, 0x08000080 },
582 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 0, 0x103e, 0x08000000, 0x08000080 },
583 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 0, 0x103c, 0x08000000, 0x08000080 },
584 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420){ GC_HWIP, 0, 0, 0x060d, 0xf00fffff, 0x00000420 },
585 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000){ GC_HWIP, 0, 0, 0x0640, 0x0000000f, 0x00000000 },
586 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 0, 0x0246, 0x08000000, 0x08000080 },
587 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024){ GC_HWIP, 0, 0, 0x02cf, 0x00000003, 0x82400024 },
588 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001){ GC_HWIP, 0, 0, 0x02fc, 0x3fffffff, 0x00000001 },
589 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000){ GC_HWIP, 0, 1, 0x2281, 0x0000ff0f, 0x00000000 },
590 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080){ GC_HWIP, 0, 1, 0x4cb2, 0x08000000, 0x08000080 },
591 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080){ GC_HWIP, 0, 1, 0x4cb3, 0x08000000, 0x08000080 },
592 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080){ GC_HWIP, 0, 1, 0x4cb4, 0x08000000, 0x08000080 },
593 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 1, 0x4ccd, 0x08000000, 0x08000080 },
594 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 1, 0x4cb5, 0x08000000, 0x08000080 },
595 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000){ GC_HWIP, 0, 0, 0x0542, 0xfffffeef, 0x010b0000 },
596 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000){ GC_HWIP, 0, 0, 0x0b04, 0xffffffff, 0x00000000 },
597 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120){ GC_HWIP, 0, 0, 0x0b03, 0xffffffff, 0x00003120 },
598 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000){ GC_HWIP, 0, 0, 0x0231, 0x3fff3af3, 0x19200000 },
599 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000000ff){ GC_HWIP, 0, 0, 0x0269, 0x00000fff, 0x000000ff },
600 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080){ GC_HWIP, 0, 0, 0x0243, 0x08000000, 0x08000080 },
601 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800){ GC_HWIP, 0, 0, 0x10bd, 0x00000800, 0x00000800 },
602 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800){ GC_HWIP, 0, 0, 0x10be, 0x00000800, 0x00000800 },
603 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000){ GC_HWIP, 0, 0, 0x107f, 0x00008000, 0x00008000 }
604};
605
606static const struct soc15_reg_golden golden_settings_gc_9_1_rv1[] =
607{
608 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000){ GC_HWIP, 0, 0, 0x0683, 0x30000000, 0x10000000 },
609 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24000042){ GC_HWIP, 0, 0, 0x063e, 0xffff77ff, 0x24000042 },
610 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24000042){ GC_HWIP, 0, 0, 0x0642, 0xffff77ff, 0x24000042 },
611 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04048000){ GC_HWIP, 0, 0, 0x02fd, 0xffffffff, 0x04048000 },
612 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_MODE_CNTL_1, 0x06000000, 0x06000000){ GC_HWIP, 0, 1, 0x0293, 0x06000000, 0x06000000 },
613 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000){ GC_HWIP, 0, 0, 0x078c, 0x00030000, 0x00020000 },
614 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x00000800){ GC_HWIP, 0, 0, 0x0525, 0x01bd9f33, 0x00000800 }
615};
616
617static const struct soc15_reg_golden golden_settings_gc_9_1_rv2[] =
618{
619 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0xff7fffff, 0x04000000){ GC_HWIP, 0, 0, 0x0688, 0xff7fffff, 0x04000000 },
620 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104){ GC_HWIP, 0, 0, 0x0680, 0xfffdf3cf, 0x00014104 },
621 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000){ GC_HWIP, 0, 0, 0x0682, 0xff7fffff, 0x0a000000 },
622 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x7f0fffff, 0x08000080){ GC_HWIP, 0, 0, 0x103d, 0x7f0fffff, 0x08000080 },
623 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0xff8fffff, 0x08000080){ GC_HWIP, 0, 0, 0x103e, 0xff8fffff, 0x08000080 },
624 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x7f8fffff, 0x08000080){ GC_HWIP, 0, 0, 0x103c, 0x7f8fffff, 0x08000080 },
625 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x26013041){ GC_HWIP, 0, 0, 0x063e, 0xffff77ff, 0x26013041 },
626 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x26013041){ GC_HWIP, 0, 0, 0x0642, 0xffff77ff, 0x26013041 },
627 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x3f8fffff, 0x08000080){ GC_HWIP, 0, 0, 0x0246, 0x3f8fffff, 0x08000080 },
628 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000){ GC_HWIP, 0, 0, 0x02fd, 0xffffffff, 0x04040000 },
629 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0xff0fffff, 0x08000080){ GC_HWIP, 0, 1, 0x4cb2, 0xff0fffff, 0x08000080 },
630 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0xff0fffff, 0x08000080){ GC_HWIP, 0, 1, 0x4cb3, 0xff0fffff, 0x08000080 },
631 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0xff0fffff, 0x08000080){ GC_HWIP, 0, 1, 0x4cb4, 0xff0fffff, 0x08000080 },
632 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0xff0fffff, 0x08000080){ GC_HWIP, 0, 1, 0x4ccd, 0xff0fffff, 0x08000080 },
633 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0xff0fffff, 0x08000080){ GC_HWIP, 0, 1, 0x4cb5, 0xff0fffff, 0x08000080 },
634 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000){ GC_HWIP, 0, 0, 0x0b04, 0xffffffff, 0x00000000 },
635 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010){ GC_HWIP, 0, 0, 0x0b03, 0xffffffff, 0x00000010 },
636 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000){ GC_HWIP, 0, 0, 0x0525, 0x01bd9f33, 0x01000000 },
637 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x3f8fffff, 0x08000080){ GC_HWIP, 0, 0, 0x0243, 0x3f8fffff, 0x08000080 },
638};
639
640static const struct soc15_reg_golden golden_settings_gc_9_1_rn[] =
641{
642 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104){ GC_HWIP, 0, 0, 0x0680, 0xfffdf3cf, 0x00014104 },
643 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000){ GC_HWIP, 0, 0, 0x0682, 0xff7fffff, 0x0a000000 },
644 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400){ GC_HWIP, 0, 0, 0x060d, 0xf00fffff, 0x00000400 },
645 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x24000042){ GC_HWIP, 0, 0, 0x063e, 0xf3e777ff, 0x24000042 },
646 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x24000042){ GC_HWIP, 0, 0, 0x0642, 0xf3e777ff, 0x24000042 },
647 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001){ GC_HWIP, 0, 0, 0x02fc, 0x3fffffff, 0x00000001 },
648 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000){ GC_HWIP, 0, 0, 0x02fd, 0xffffffff, 0x04040000 },
649 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000){ GC_HWIP, 0, 1, 0x2281, 0x0000ff0f, 0x00000000 },
650 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000){ GC_HWIP, 0, 0, 0x0542, 0xfffffeef, 0x010b0000 },
651 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000){ GC_HWIP, 0, 0, 0x0b04, 0xffffffff, 0x00000000 },
652 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120){ GC_HWIP, 0, 0, 0x0b03, 0xffffffff, 0x00003120 },
653 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGCEA_PROBE_MAP, 0xffffffff, 0x0000cccc){ GC_HWIP, 0, 0, 0x070c, 0xffffffff, 0x0000cccc },
654};
655
656static const struct soc15_reg_golden golden_settings_gc_9_x_common[] =
657{
658 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0xffffffff, 0x000001ff){ GC_HWIP, 0, 0, 0x11b7, 0xffffffff, 0x000001ff },
659 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_INDEX, 0xffffffff, 0x00000000){ GC_HWIP, 0, 1, 0x5a04, 0xffffffff, 0x00000000 },
660 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_DATA, 0xffffffff, 0x2544c382){ GC_HWIP, 0, 1, 0x5a05, 0xffffffff, 0x2544c382 }
661};
662
663static const struct soc15_reg_golden golden_settings_gc_9_2_1[] =
664{
665 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420){ GC_HWIP, 0, 0, 0x060d, 0xf00fffff, 0x00000420 },
666 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000){ GC_HWIP, 0, 0, 0x0640, 0x0000000f, 0x00000000 },
667 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024){ GC_HWIP, 0, 0, 0x02cf, 0x00000003, 0x82400024 },
668 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001){ GC_HWIP, 0, 0, 0x02fc, 0x3fffffff, 0x00000001 },
669 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000){ GC_HWIP, 0, 1, 0x2281, 0x0000ff0f, 0x00000000 },
670 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000){ GC_HWIP, 0, 0, 0x030d, 0x00001000, 0x00001000 },
671 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800){ GC_HWIP, 0, 0, 0x11dc, 0x0007ffff, 0x00000800 },
672 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800){ GC_HWIP, 0, 0, 0x11dd, 0x0007ffff, 0x00000800 },
673 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87){ GC_HWIP, 0, 0, 0x11e6, 0x01ffffff, 0x0000ff87 },
674 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f){ GC_HWIP, 0, 0, 0x11e7, 0x01ffffff, 0x0000ff8f },
675 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000){ GC_HWIP, 0, 0, 0x0301, 0x03000000, 0x020a2000 },
676 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000){ GC_HWIP, 0, 0, 0x0542, 0xfffffeef, 0x010b0000 },
677 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68){ GC_HWIP, 0, 0, 0x0b04, 0xffffffff, 0x4a2c0e68 },
678 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197){ GC_HWIP, 0, 0, 0x0b03, 0xffffffff, 0xb5d3f197 },
679 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000){ GC_HWIP, 0, 0, 0x0231, 0x3fff3af3, 0x19200000 },
680 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff){ GC_HWIP, 0, 0, 0x0269, 0x00000fff, 0x000003ff }
681};
682
683static const struct soc15_reg_golden golden_settings_gc_9_2_1_vg12[] =
684{
685 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x00000080, 0x04000080){ GC_HWIP, 0, 0, 0x0688, 0x00000080, 0x04000080 },
686 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104){ GC_HWIP, 0, 0, 0x0680, 0xfffdf3cf, 0x00014104 },
687 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000){ GC_HWIP, 0, 0, 0x0682, 0x0f000000, 0x0a000000 },
688 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24104041){ GC_HWIP, 0, 0, 0x063e, 0xffff77ff, 0x24104041 },
689 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24104041){ GC_HWIP, 0, 0, 0x0642, 0xffff77ff, 0x24104041 },
690 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000){ GC_HWIP, 0, 0, 0x02fd, 0xffffffff, 0x04040000 },
691 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff03ff, 0x01000107){ GC_HWIP, 0, 1, 0x2441, 0xffff03ff, 0x01000107 },
692 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000){ GC_HWIP, 0, 0, 0x0b04, 0xffffffff, 0x00000000 },
693 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x76325410){ GC_HWIP, 0, 0, 0x0b03, 0xffffffff, 0x76325410 },
694 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000){ GC_HWIP, 0, 0, 0x0525, 0x01bd9f33, 0x01000000 },
695 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800){ GC_HWIP, 0, 0, 0x10bd, 0x00000800, 0x00000800 },
696 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800){ GC_HWIP, 0, 0, 0x10be, 0x00000800, 0x00000800 },
697 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000){ GC_HWIP, 0, 0, 0x107f, 0x00008000, 0x00008000 }
698};
699
700static const struct soc15_reg_golden golden_settings_gc_9_4_1_arct[] =
701{
702 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042){ GC_HWIP, 0, 0, 0x063e, 0xffff77ff, 0x2a114042 },
703 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x10b0000){ GC_HWIP, 0, 0, 0x0542, 0xfffffeef, 0x10b0000 },
704 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_0_ARCT, 0x3fffffff, 0x346f0a4e){ GC_HWIP, 0, 0, 0x0b03, 0x3fffffff, 0x346f0a4e },
705 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_1_ARCT, 0x3fffffff, 0x1c642ca){ GC_HWIP, 0, 0, 0x0b04, 0x3fffffff, 0x1c642ca },
706 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_2_ARCT, 0x3fffffff, 0x26f45098){ GC_HWIP, 0, 0, 0x0b09, 0x3fffffff, 0x26f45098 },
707 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_3_ARCT, 0x3fffffff, 0x2ebd9fe3){ GC_HWIP, 0, 0, 0x0b0a, 0x3fffffff, 0x2ebd9fe3 },
708 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_4_ARCT, 0x3fffffff, 0xb90f5b1){ GC_HWIP, 0, 0, 0x0b0b, 0x3fffffff, 0xb90f5b1 },
709 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_5_ARCT, 0x3ff, 0x135){ GC_HWIP, 0, 0, 0x0b0c, 0x3ff, 0x135 },
710 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_CONFIG, 0xffffffff, 0x011A0000){ GC_HWIP, 0, 0, 0x0300, 0xffffffff, 0x011A0000 },
711 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_FIFO_SIZES, 0xffffffff, 0x00000f00){ GC_HWIP, 0, 0, 0x0305, 0xffffffff, 0x00000f00 },
712 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_UTCL1_CNTL1, 0x30000000, 0x30000000){ GC_HWIP, 0, 0, 0x12b5, 0x30000000, 0x30000000 }
713};
714
715static const struct soc15_reg_rlcg rlcg_access_gc_9_0[] = {
716 {SOC15_REG_ENTRY(GC, 0, mmGRBM_GFX_INDEX)GC_HWIP, 0, 1, 0x2200},
717 {SOC15_REG_ENTRY(GC, 0, mmSQ_IND_INDEX)GC_HWIP, 0, 0, 0x0378},
718};
719
720static const u32 GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[] =
721{
722 mmRLC_SRM_INDEX_CNTL_ADDR_00x4c8b - mmRLC_SRM_INDEX_CNTL_ADDR_00x4c8b,
723 mmRLC_SRM_INDEX_CNTL_ADDR_10x4c8c - mmRLC_SRM_INDEX_CNTL_ADDR_00x4c8b,
724 mmRLC_SRM_INDEX_CNTL_ADDR_20x4c8d - mmRLC_SRM_INDEX_CNTL_ADDR_00x4c8b,
725 mmRLC_SRM_INDEX_CNTL_ADDR_30x4c8e - mmRLC_SRM_INDEX_CNTL_ADDR_00x4c8b,
726 mmRLC_SRM_INDEX_CNTL_ADDR_40x4c8f - mmRLC_SRM_INDEX_CNTL_ADDR_00x4c8b,
727 mmRLC_SRM_INDEX_CNTL_ADDR_50x4c90 - mmRLC_SRM_INDEX_CNTL_ADDR_00x4c8b,
728 mmRLC_SRM_INDEX_CNTL_ADDR_60x4c91 - mmRLC_SRM_INDEX_CNTL_ADDR_00x4c8b,
729 mmRLC_SRM_INDEX_CNTL_ADDR_70x4c92 - mmRLC_SRM_INDEX_CNTL_ADDR_00x4c8b,
730};
731
732static const u32 GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[] =
733{
734 mmRLC_SRM_INDEX_CNTL_DATA_00x4c93 - mmRLC_SRM_INDEX_CNTL_DATA_00x4c93,
735 mmRLC_SRM_INDEX_CNTL_DATA_10x4c94 - mmRLC_SRM_INDEX_CNTL_DATA_00x4c93,
736 mmRLC_SRM_INDEX_CNTL_DATA_20x4c95 - mmRLC_SRM_INDEX_CNTL_DATA_00x4c93,
737 mmRLC_SRM_INDEX_CNTL_DATA_30x4c96 - mmRLC_SRM_INDEX_CNTL_DATA_00x4c93,
738 mmRLC_SRM_INDEX_CNTL_DATA_40x4c97 - mmRLC_SRM_INDEX_CNTL_DATA_00x4c93,
739 mmRLC_SRM_INDEX_CNTL_DATA_50x4c98 - mmRLC_SRM_INDEX_CNTL_DATA_00x4c93,
740 mmRLC_SRM_INDEX_CNTL_DATA_60x4c99 - mmRLC_SRM_INDEX_CNTL_DATA_00x4c93,
741 mmRLC_SRM_INDEX_CNTL_DATA_70x4c9a - mmRLC_SRM_INDEX_CNTL_DATA_00x4c93,
742};
743
744#define VEGA10_GB_ADDR_CONFIG_GOLDEN0x2a114042 0x2a114042
745#define VEGA12_GB_ADDR_CONFIG_GOLDEN0x24104041 0x24104041
746#define RAVEN_GB_ADDR_CONFIG_GOLDEN0x24000042 0x24000042
747#define RAVEN2_GB_ADDR_CONFIG_GOLDEN0x26013041 0x26013041
748
749static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev);
750static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev);
751static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev);
752static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev);
753static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev,
754 struct amdgpu_cu_info *cu_info);
755static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev);
756static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring);
757static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring);
758static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
759 void *ras_error_status);
760static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
761 void *inject_if);
762static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev);
763
764static void gfx_v9_0_kiq_set_resources(struct amdgpu_ring *kiq_ring,
765 uint64_t queue_mask)
766{
767 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6)((3 << 30) | (((0xA0) & 0xFF) << 8) | ((6) &
0x3FFF) << 16));
768 amdgpu_ring_write(kiq_ring,
769 PACKET3_SET_RESOURCES_VMID_MASK(0)((0) << 0) |
770 /* vmid_mask:0* queue_type:0 (KIQ) */
771 PACKET3_SET_RESOURCES_QUEUE_TYPE(0)((0) << 29));
772 amdgpu_ring_write(kiq_ring,
773 lower_32_bits(queue_mask)((u32)(queue_mask))); /* queue mask lo */
774 amdgpu_ring_write(kiq_ring,
775 upper_32_bits(queue_mask)((u32)(((queue_mask) >> 16) >> 16))); /* queue mask hi */
776 amdgpu_ring_write(kiq_ring, 0); /* gws mask lo */
777 amdgpu_ring_write(kiq_ring, 0); /* gws mask hi */
778 amdgpu_ring_write(kiq_ring, 0); /* oac mask */
779 amdgpu_ring_write(kiq_ring, 0); /* gds heap base:0, gds heap size:0 */
780}
781
782static void gfx_v9_0_kiq_map_queues(struct amdgpu_ring *kiq_ring,
783 struct amdgpu_ring *ring)
784{
785 uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
786 uint64_t wptr_addr = ring->wptr_gpu_addr;
787 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
788
789 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5)((3 << 30) | (((0xA2) & 0xFF) << 8) | ((5) &
0x3FFF) << 16));
790 /* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
791 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
792 PACKET3_MAP_QUEUES_QUEUE_SEL(0)((0) << 4) | /* Queue_Sel */
793 PACKET3_MAP_QUEUES_VMID(0)((0) << 8) | /* VMID */
794 PACKET3_MAP_QUEUES_QUEUE(ring->queue)((ring->queue) << 13) |
795 PACKET3_MAP_QUEUES_PIPE(ring->pipe)((ring->pipe) << 16) |
796 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1))(((ring->me == 1 ? 0 : 1)) << 18) |
797 /*queue_type: normal compute queue */
798 PACKET3_MAP_QUEUES_QUEUE_TYPE(0)((0) << 21) |
799 /* alloc format: all_on_one_pipe */
800 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0)((0) << 24) |
801 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel)((eng_sel) << 26) |
802 /* num_queues: must be 1 */
803 PACKET3_MAP_QUEUES_NUM_QUEUES(1)((1) << 29));
804 amdgpu_ring_write(kiq_ring,
805 PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index)((ring->doorbell_index) << 2));
806 amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr)((u32)(mqd_addr)));
807 amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr)((u32)(((mqd_addr) >> 16) >> 16)));
808 amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr)((u32)(wptr_addr)));
809 amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr)((u32)(((wptr_addr) >> 16) >> 16)));
810}
811
812static void gfx_v9_0_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
813 struct amdgpu_ring *ring,
814 enum amdgpu_unmap_queues_action action,
815 u64 gpu_addr, u64 seq)
816{
817 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
818
819 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4)((3 << 30) | (((0xA3) & 0xFF) << 8) | ((4) &
0x3FFF) << 16));
820 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
821 PACKET3_UNMAP_QUEUES_ACTION(action)((action) << 0) |
822 PACKET3_UNMAP_QUEUES_QUEUE_SEL(0)((0) << 4) |
823 PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel)((eng_sel) << 26) |
824 PACKET3_UNMAP_QUEUES_NUM_QUEUES(1)((1) << 29));
825 amdgpu_ring_write(kiq_ring,
826 PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index)((ring->doorbell_index) << 2));
827
828 if (action == PREEMPT_QUEUES_NO_UNMAP) {
829 amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr)((u32)(gpu_addr)));
830 amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr)((u32)(((gpu_addr) >> 16) >> 16)));
831 amdgpu_ring_write(kiq_ring, seq);
832 } else {
833 amdgpu_ring_write(kiq_ring, 0);
834 amdgpu_ring_write(kiq_ring, 0);
835 amdgpu_ring_write(kiq_ring, 0);
836 }
837}
838
839static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
840 struct amdgpu_ring *ring,
841 u64 addr,
842 u64 seq)
843{
844 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
845
846 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5)((3 << 30) | (((0xA4) & 0xFF) << 8) | ((5) &
0x3FFF) << 16));
847 amdgpu_ring_write(kiq_ring,
848 PACKET3_QUERY_STATUS_CONTEXT_ID(0)((0) << 0) |
849 PACKET3_QUERY_STATUS_INTERRUPT_SEL(0)((0) << 28) |
850 PACKET3_QUERY_STATUS_COMMAND(2)((2) << 30));
851 /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
852 amdgpu_ring_write(kiq_ring,
853 PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index)((ring->doorbell_index) << 2) |
854 PACKET3_QUERY_STATUS_ENG_SEL(eng_sel)((eng_sel) << 25));
855 amdgpu_ring_write(kiq_ring, lower_32_bits(addr)((u32)(addr)));
856 amdgpu_ring_write(kiq_ring, upper_32_bits(addr)((u32)(((addr) >> 16) >> 16)));
857 amdgpu_ring_write(kiq_ring, lower_32_bits(seq)((u32)(seq)));
858 amdgpu_ring_write(kiq_ring, upper_32_bits(seq)((u32)(((seq) >> 16) >> 16)));
859}
860
861static void gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
862 uint16_t pasid, uint32_t flush_type,
863 bool_Bool all_hub)
864{
865 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0)((3 << 30) | (((0x98) & 0xFF) << 8) | ((0) &
0x3FFF) << 16));
866 amdgpu_ring_write(kiq_ring,
867 PACKET3_INVALIDATE_TLBS_DST_SEL(1)((1) << 0) |
868 PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub)((all_hub) << 4) |
869 PACKET3_INVALIDATE_TLBS_PASID(pasid)((pasid) << 5) |
870 PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type)((flush_type) << 29));
871}
872
873static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
874 .kiq_set_resources = gfx_v9_0_kiq_set_resources,
875 .kiq_map_queues = gfx_v9_0_kiq_map_queues,
876 .kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
877 .kiq_query_status = gfx_v9_0_kiq_query_status,
878 .kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs,
879 .set_resources_size = 8,
880 .map_queues_size = 7,
881 .unmap_queues_size = 6,
882 .query_status_size = 7,
883 .invalidate_tlbs_size = 2,
884};
885
886static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
887{
888 adev->gfx.kiq.pmf = &gfx_v9_0_kiq_pm4_funcs;
889}
890
891static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev)
892{
893 switch (adev->ip_versions[GC_HWIP][0]) {
894 case IP_VERSION(9, 0, 1)(((9) << 16) | ((0) << 8) | (1)):
895 soc15_program_register_sequence(adev,
896 golden_settings_gc_9_0,
897 ARRAY_SIZE(golden_settings_gc_9_0)(sizeof((golden_settings_gc_9_0)) / sizeof((golden_settings_gc_9_0
)[0])));
898 soc15_program_register_sequence(adev,
899 golden_settings_gc_9_0_vg10,
900 ARRAY_SIZE(golden_settings_gc_9_0_vg10)(sizeof((golden_settings_gc_9_0_vg10)) / sizeof((golden_settings_gc_9_0_vg10
)[0])));
901 break;
902 case IP_VERSION(9, 2, 1)(((9) << 16) | ((2) << 8) | (1)):
903 soc15_program_register_sequence(adev,
904 golden_settings_gc_9_2_1,
905 ARRAY_SIZE(golden_settings_gc_9_2_1)(sizeof((golden_settings_gc_9_2_1)) / sizeof((golden_settings_gc_9_2_1
)[0])));
906 soc15_program_register_sequence(adev,
907 golden_settings_gc_9_2_1_vg12,
908 ARRAY_SIZE(golden_settings_gc_9_2_1_vg12)(sizeof((golden_settings_gc_9_2_1_vg12)) / sizeof((golden_settings_gc_9_2_1_vg12
)[0])));
909 break;
910 case IP_VERSION(9, 4, 0)(((9) << 16) | ((4) << 8) | (0)):
911 soc15_program_register_sequence(adev,
912 golden_settings_gc_9_0,
913 ARRAY_SIZE(golden_settings_gc_9_0)(sizeof((golden_settings_gc_9_0)) / sizeof((golden_settings_gc_9_0
)[0])));
914 soc15_program_register_sequence(adev,
915 golden_settings_gc_9_0_vg20,
916 ARRAY_SIZE(golden_settings_gc_9_0_vg20)(sizeof((golden_settings_gc_9_0_vg20)) / sizeof((golden_settings_gc_9_0_vg20
)[0])));
917 break;
918 case IP_VERSION(9, 4, 1)(((9) << 16) | ((4) << 8) | (1)):
919 soc15_program_register_sequence(adev,
920 golden_settings_gc_9_4_1_arct,
921 ARRAY_SIZE(golden_settings_gc_9_4_1_arct)(sizeof((golden_settings_gc_9_4_1_arct)) / sizeof((golden_settings_gc_9_4_1_arct
)[0])));
922 break;
923 case IP_VERSION(9, 2, 2)(((9) << 16) | ((2) << 8) | (2)):
924 case IP_VERSION(9, 1, 0)(((9) << 16) | ((1) << 8) | (0)):
925 soc15_program_register_sequence(adev, golden_settings_gc_9_1,
926 ARRAY_SIZE(golden_settings_gc_9_1)(sizeof((golden_settings_gc_9_1)) / sizeof((golden_settings_gc_9_1
)[0])));
927 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
928 soc15_program_register_sequence(adev,
929 golden_settings_gc_9_1_rv2,
930 ARRAY_SIZE(golden_settings_gc_9_1_rv2)(sizeof((golden_settings_gc_9_1_rv2)) / sizeof((golden_settings_gc_9_1_rv2
)[0])));
931 else
932 soc15_program_register_sequence(adev,
933 golden_settings_gc_9_1_rv1,
934 ARRAY_SIZE(golden_settings_gc_9_1_rv1)(sizeof((golden_settings_gc_9_1_rv1)) / sizeof((golden_settings_gc_9_1_rv1
)[0])));
935 break;
936 case IP_VERSION(9, 3, 0)(((9) << 16) | ((3) << 8) | (0)):
937 soc15_program_register_sequence(adev,
938 golden_settings_gc_9_1_rn,
939 ARRAY_SIZE(golden_settings_gc_9_1_rn)(sizeof((golden_settings_gc_9_1_rn)) / sizeof((golden_settings_gc_9_1_rn
)[0])));
940 return; /* for renoir, don't need common goldensetting */
941 case IP_VERSION(9, 4, 2)(((9) << 16) | ((4) << 8) | (2)):
942 gfx_v9_4_2_init_golden_registers(adev,
943 adev->smuio.funcs->get_die_id(adev));
944 break;
945 default:
946 break;
947 }
948
949 if ((adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)(((9) << 16) | ((4) << 8) | (1))) &&
950 (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 2)(((9) << 16) | ((4) << 8) | (2))))
951 soc15_program_register_sequence(adev, golden_settings_gc_9_x_common,
952 (const u32)ARRAY_SIZE(golden_settings_gc_9_x_common)(sizeof((golden_settings_gc_9_x_common)) / sizeof((golden_settings_gc_9_x_common
)[0])));
953}
954
955static void gfx_v9_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
956 bool_Bool wc, uint32_t reg, uint32_t val)
957{
958 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)((3 << 30) | (((0x37) & 0xFF) << 8) | ((3) &
0x3FFF) << 16));
959 amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel)((eng_sel) << 30) |
960 WRITE_DATA_DST_SEL(0)((0) << 8) |
961 (wc ? WR_CONFIRM(1 << 20) : 0));
962 amdgpu_ring_write(ring, reg);
963 amdgpu_ring_write(ring, 0);
964 amdgpu_ring_write(ring, val);
965}
966
967static void gfx_v9_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
968 int mem_space, int opt, uint32_t addr0,
969 uint32_t addr1, uint32_t ref, uint32_t mask,
970 uint32_t inv)
971{
972 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)((3 << 30) | (((0x3C) & 0xFF) << 8) | ((5) &
0x3FFF) << 16));
973 amdgpu_ring_write(ring,
974 /* memory (1) or register (0) */
975 (WAIT_REG_MEM_MEM_SPACE(mem_space)((mem_space) << 4) |
976 WAIT_REG_MEM_OPERATION(opt)((opt) << 6) | /* wait */
977 WAIT_REG_MEM_FUNCTION(3)((3) << 0) | /* equal */
978 WAIT_REG_MEM_ENGINE(eng_sel)((eng_sel) << 8)));
979
980 if (mem_space)
981 BUG_ON(addr0 & 0x3)((!(addr0 & 0x3)) ? (void)0 : __assert("diagnostic ", "/usr/src/sys/dev/pci/drm/amd/amdgpu/gfx_v9_0.c"
, 981, "!(addr0 & 0x3)")); /* Dword align */
982 amdgpu_ring_write(ring, addr0);
983 amdgpu_ring_write(ring, addr1);
984 amdgpu_ring_write(ring, ref);
985 amdgpu_ring_write(ring, mask);
986 amdgpu_ring_write(ring, inv); /* poll interval */
987}
988
989static int gfx_v9_0_ring_test_ring(struct amdgpu_ring *ring)
990{
991 struct amdgpu_device *adev = ring->adev;
992 uint32_t scratch = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0)(adev->reg_offset[GC_HWIP][0][1] + 0x2040);
993 uint32_t tmp = 0;
994 unsigned i;
995 int r;
996
997 WREG32(scratch, 0xCAFEDEAD)amdgpu_device_wreg(adev, (scratch), (0xCAFEDEAD), 0);
998 r = amdgpu_ring_alloc(ring, 3);
999 if (r)
1000 return r;
1001
1002 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)((3 << 30) | (((0x79) & 0xFF) << 8) | ((1) &
0x3FFF) << 16));
1003 amdgpu_ring_write(ring, scratch - PACKET3_SET_UCONFIG_REG_START0x0000c000);
1004 amdgpu_ring_write(ring, 0xDEADBEEF);
1005 amdgpu_ring_commit(ring);
1006
1007 for (i = 0; i < adev->usec_timeout; i++) {
1008 tmp = RREG32(scratch)amdgpu_device_rreg(adev, (scratch), 0);
1009 if (tmp == 0xDEADBEEF)
1010 break;
1011 udelay(1);
1012 }
1013
1014 if (i >= adev->usec_timeout)
1015 r = -ETIMEDOUT60;
1016 return r;
1017}
1018
1019static int gfx_v9_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1020{
1021 struct amdgpu_device *adev = ring->adev;
1022 struct amdgpu_ib ib;
1023 struct dma_fence *f = NULL((void *)0);
1024
1025 unsigned index;
1026 uint64_t gpu_addr;
1027 uint32_t tmp;
1028 long r;
1029
1030 r = amdgpu_device_wb_get(adev, &index);
1031 if (r)
1032 return r;
1033
1034 gpu_addr = adev->wb.gpu_addr + (index * 4);
1035 adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD)((__uint32_t)(0xCAFEDEAD));
1036 memset(&ib, 0, sizeof(ib))__builtin_memset((&ib), (0), (sizeof(ib)));
1037
1038 r = amdgpu_ib_get(adev, NULL((void *)0), 20, AMDGPU_IB_POOL_DIRECT, &ib);
1039 if (r)
1040 goto err1;
1041
1042 ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3)((3 << 30) | (((0x37) & 0xFF) << 8) | ((3) &
0x3FFF) << 16);
1043 ib.ptr[1] = WRITE_DATA_DST_SEL(5)((5) << 8) | WR_CONFIRM(1 << 20);
1044 ib.ptr[2] = lower_32_bits(gpu_addr)((u32)(gpu_addr));
1045 ib.ptr[3] = upper_32_bits(gpu_addr)((u32)(((gpu_addr) >> 16) >> 16));
1046 ib.ptr[4] = 0xDEADBEEF;
1047 ib.length_dw = 5;
1048
1049 r = amdgpu_ib_schedule(ring, 1, &ib, NULL((void *)0), &f);
1050 if (r)
1051 goto err2;
1052
1053 r = dma_fence_wait_timeout(f, false0, timeout);
1054 if (r == 0) {
1055 r = -ETIMEDOUT60;
1056 goto err2;
1057 } else if (r < 0) {
1058 goto err2;
1059 }
1060
1061 tmp = adev->wb.wb[index];
1062 if (tmp == 0xDEADBEEF)
1063 r = 0;
1064 else
1065 r = -EINVAL22;
1066
1067err2:
1068 amdgpu_ib_free(adev, &ib, NULL((void *)0));
1069 dma_fence_put(f);
1070err1:
1071 amdgpu_device_wb_free(adev, index);
1072 return r;
1073}
1074
1075
1076static void gfx_v9_0_free_microcode(struct amdgpu_device *adev)
1077{
1078 release_firmware(adev->gfx.pfp_fw);
1079 adev->gfx.pfp_fw = NULL((void *)0);
1080 release_firmware(adev->gfx.me_fw);
1081 adev->gfx.me_fw = NULL((void *)0);
1082 release_firmware(adev->gfx.ce_fw);
1083 adev->gfx.ce_fw = NULL((void *)0);
1084 release_firmware(adev->gfx.rlc_fw);
1085 adev->gfx.rlc_fw = NULL((void *)0);
1086 release_firmware(adev->gfx.mec_fw);
1087 adev->gfx.mec_fw = NULL((void *)0);
1088 release_firmware(adev->gfx.mec2_fw);
1089 adev->gfx.mec2_fw = NULL((void *)0);
1090
1091 kfree(adev->gfx.rlc.register_list_format);
1092}
1093
1094static void gfx_v9_0_check_fw_write_wait(struct amdgpu_device *adev)
1095{
1096 adev->gfx.me_fw_write_wait = false0;
1097 adev->gfx.mec_fw_write_wait = false0;
1098
1099 if ((adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)(((9) << 16) | ((4) << 8) | (1))) &&
1100 ((adev->gfx.mec_fw_version < 0x000001a5) ||
1101 (adev->gfx.mec_feature_version < 46) ||
1102 (adev->gfx.pfp_fw_version < 0x000000b7) ||
1103 (adev->gfx.pfp_feature_version < 46)))
1104 DRM_WARN_ONCE("CP firmware version too old, please update!")({ static int __warned; if (!__warned) { printk("\0014" "[" "drm"
"] " "CP firmware version too old, please update!"); __warned
= 1; } });
1105
1106 switch (adev->ip_versions[GC_HWIP][0]) {
1107 case IP_VERSION(9, 0, 1)(((9) << 16) | ((0) << 8) | (1)):
1108 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1109 (adev->gfx.me_feature_version >= 42) &&
1110 (adev->gfx.pfp_fw_version >= 0x000000b1) &&
1111 (adev->gfx.pfp_feature_version >= 42))
1112 adev->gfx.me_fw_write_wait = true1;
1113
1114 if ((adev->gfx.mec_fw_version >= 0x00000193) &&
1115 (adev->gfx.mec_feature_version >= 42))
1116 adev->gfx.mec_fw_write_wait = true1;
1117 break;
1118 case IP_VERSION(9, 2, 1)(((9) << 16) | ((2) << 8) | (1)):
1119 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1120 (adev->gfx.me_feature_version >= 44) &&
1121 (adev->gfx.pfp_fw_version >= 0x000000b2) &&
1122 (adev->gfx.pfp_feature_version >= 44))
1123 adev->gfx.me_fw_write_wait = true1;
1124
1125 if ((adev->gfx.mec_fw_version >= 0x00000196) &&
1126 (adev->gfx.mec_feature_version >= 44))
1127 adev->gfx.mec_fw_write_wait = true1;
1128 break;
1129 case IP_VERSION(9, 4, 0)(((9) << 16) | ((4) << 8) | (0)):
1130 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1131 (adev->gfx.me_feature_version >= 44) &&
1132 (adev->gfx.pfp_fw_version >= 0x000000b2) &&
1133 (adev->gfx.pfp_feature_version >= 44))
1134 adev->gfx.me_fw_write_wait = true1;
1135
1136 if ((adev->gfx.mec_fw_version >= 0x00000197) &&
1137 (adev->gfx.mec_feature_version >= 44))
1138 adev->gfx.mec_fw_write_wait = true1;
1139 break;
1140 case IP_VERSION(9, 1, 0)(((9) << 16) | ((1) << 8) | (0)):
1141 case IP_VERSION(9, 2, 2)(((9) << 16) | ((2) << 8) | (2)):
1142 if ((adev->gfx.me_fw_version >= 0x0000009c) &&
1143 (adev->gfx.me_feature_version >= 42) &&
1144 (adev->gfx.pfp_fw_version >= 0x000000b1) &&
1145 (adev->gfx.pfp_feature_version >= 42))
1146 adev->gfx.me_fw_write_wait = true1;
1147
1148 if ((adev->gfx.mec_fw_version >= 0x00000192) &&
1149 (adev->gfx.mec_feature_version >= 42))
1150 adev->gfx.mec_fw_write_wait = true1;
1151 break;
1152 default:
1153 adev->gfx.me_fw_write_wait = true1;
1154 adev->gfx.mec_fw_write_wait = true1;
1155 break;
1156 }
1157}
1158
1159struct amdgpu_gfxoff_quirk {
1160 u16 chip_vendor;
1161 u16 chip_device;
1162 u16 subsys_vendor;
1163 u16 subsys_device;
1164 u8 revision;
1165};
1166
1167static const struct amdgpu_gfxoff_quirk amdgpu_gfxoff_quirk_list[] = {
1168 /* https://bugzilla.kernel.org/show_bug.cgi?id=204689 */
1169 { 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc8 },
1170 /* https://bugzilla.kernel.org/show_bug.cgi?id=207171 */
1171 { 0x1002, 0x15dd, 0x103c, 0x83e7, 0xd3 },
1172 /* GFXOFF is unstable on C6 parts with a VBIOS 113-RAVEN-114 */
1173 { 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc6 },
1174 /* Apple MacBook Pro (15-inch, 2019) Radeon Pro Vega 20 4 GB */
1175 { 0x1002, 0x69af, 0x106b, 0x019a, 0xc0 },
1176 { 0, 0, 0, 0, 0 },
1177};
1178
1179static bool_Bool gfx_v9_0_should_disable_gfxoff(struct pci_dev *pdev)
1180{
1181 const struct amdgpu_gfxoff_quirk *p = amdgpu_gfxoff_quirk_list;
1182
1183 while (p && p->chip_device != 0) {
1184 if (pdev->vendor == p->chip_vendor &&
1185 pdev->device == p->chip_device &&
1186 pdev->subsystem_vendor == p->subsys_vendor &&
1187 pdev->subsystem_device == p->subsys_device &&
1188 pdev->revision == p->revision) {
1189 return true1;
1190 }
1191 ++p;
1192 }
1193 return false0;
1194}
1195
1196static bool_Bool is_raven_kicker(struct amdgpu_device *adev)
1197{
1198 if (adev->pm.fw_version >= 0x41e2b)
1199 return true1;
1200 else
1201 return false0;
1202}
1203
1204static bool_Bool check_if_enlarge_doorbell_range(struct amdgpu_device *adev)
1205{
1206 if ((adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 3, 0)(((9) << 16) | ((3) << 8) | (0))) &&
1207 (adev->gfx.me_fw_version >= 0x000000a5) &&
1208 (adev->gfx.me_feature_version >= 52))
1209 return true1;
1210 else
1211 return false0;
1212}
1213
1214static void gfx_v9_0_check_if_need_gfxoff(struct amdgpu_device *adev)
1215{
1216 if (gfx_v9_0_should_disable_gfxoff(adev->pdev))
1217 adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1218
1219 switch (adev->ip_versions[GC_HWIP][0]) {
1220 case IP_VERSION(9, 0, 1)(((9) << 16) | ((0) << 8) | (1)):
1221 case IP_VERSION(9, 2, 1)(((9) << 16) | ((2) << 8) | (1)):
1222 case IP_VERSION(9, 4, 0)(((9) << 16) | ((4) << 8) | (0)):
1223 break;
1224 case IP_VERSION(9, 2, 2)(((9) << 16) | ((2) << 8) | (2)):
1225 case IP_VERSION(9, 1, 0)(((9) << 16) | ((1) << 8) | (0)):
1226 if (!((adev->apu_flags & AMD_APU_IS_RAVEN2) ||
1227 (adev->apu_flags & AMD_APU_IS_PICASSO)) &&
1228 ((!is_raven_kicker(adev) &&
1229 adev->gfx.rlc_fw_version < 531) ||
1230 (adev->gfx.rlc_feature_version < 1) ||
1231 !adev->gfx.rlc.is_rlc_v2_1))
1232 adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
1233
1234 if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1235 adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG(1 << 0) |
1236 AMD_PG_SUPPORT_CP(1 << 5) |
1237 AMD_PG_SUPPORT_RLC_SMU_HS(1 << 7);
1238 break;
1239 case IP_VERSION(9, 3, 0)(((9) << 16) | ((3) << 8) | (0)):
1240 if (adev->pm.pp_feature & PP_GFXOFF_MASK)
1241 adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG(1 << 0) |
1242 AMD_PG_SUPPORT_CP(1 << 5) |
1243 AMD_PG_SUPPORT_RLC_SMU_HS(1 << 7);
1244 break;
1245 default:
1246 break;
1247 }
1248}
1249
1250static int gfx_v9_0_init_cp_gfx_microcode(struct amdgpu_device *adev,
1251 const char *chip_name)
1252{
1253 char fw_name[30];
1254 int err;
1255
1256 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
1257 err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1258 if (err)
1259 goto out;
1260 err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
1261 if (err)
1262 goto out;
1263 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_PFP);
1264
1265 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1266 err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1267 if (err)
1268 goto out;
1269 err = amdgpu_ucode_validate(adev->gfx.me_fw);
1270 if (err)
1271 goto out;
1272 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_ME);
1273
1274 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1275 err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1276 if (err)
1277 goto out;
1278 err = amdgpu_ucode_validate(adev->gfx.ce_fw);
1279 if (err)
1280 goto out;
1281 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_CE);
1282
1283out:
1284 if (err) {
1285 dev_err(adev->dev,printf("drm:pid%d:%s *ERROR* " "gfx9: Failed to init firmware \"%s\"\n"
, ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__ , fw_name
)
1286 "gfx9: Failed to init firmware \"%s\"\n",printf("drm:pid%d:%s *ERROR* " "gfx9: Failed to init firmware \"%s\"\n"
, ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__ , fw_name
)
1287 fw_name)printf("drm:pid%d:%s *ERROR* " "gfx9: Failed to init firmware \"%s\"\n"
, ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__ , fw_name
);
1288 release_firmware(adev->gfx.pfp_fw);
1289 adev->gfx.pfp_fw = NULL((void *)0);
1290 release_firmware(adev->gfx.me_fw);
1291 adev->gfx.me_fw = NULL((void *)0);
1292 release_firmware(adev->gfx.ce_fw);
1293 adev->gfx.ce_fw = NULL((void *)0);
1294 }
1295 return err;
1296}
1297
1298static int gfx_v9_0_init_rlc_microcode(struct amdgpu_device *adev,
1299 const char *chip_name)
1300{
1301 char fw_name[30];
1302 int err;
1303 const struct rlc_firmware_header_v2_0 *rlc_hdr;
1304 uint16_t version_major;
1305 uint16_t version_minor;
1306 uint32_t smu_version;
1307
1308 /*
1309 * For Picasso && AM4 SOCKET board, we use picasso_rlc_am4.bin
1310 * instead of picasso_rlc.bin.
1311 * Judgment method:
1312 * PCO AM4: revision >= 0xC8 && revision <= 0xCF
1313 * or revision >= 0xD8 && revision <= 0xDF
1314 * otherwise is PCO FP5
1315 */
1316 if (!strcmp(chip_name, "picasso") &&
1317 (((adev->pdev->revision >= 0xC8) && (adev->pdev->revision <= 0xCF)) ||
1318 ((adev->pdev->revision >= 0xD8) && (adev->pdev->revision <= 0xDF))))
1319 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc_am4.bin", chip_name);
1320 else if (!strcmp(chip_name, "raven") && (amdgpu_pm_load_smu_firmware(adev, &smu_version) == 0) &&
1321 (smu_version >= 0x41e2b))
1322 /**
1323 *SMC is loaded by SBIOS on APU and it's able to get the SMU version directly.
1324 */
1325 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_kicker_rlc.bin", chip_name);
1326 else
1327 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
1328 err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
1329 if (err)
1330 goto out;
1331 err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
1332 if (err)
1333 goto out;
1334 rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1335
1336 version_major = le16_to_cpu(rlc_hdr->header.header_version_major)((__uint16_t)(rlc_hdr->header.header_version_major));
1337 version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor)((__uint16_t)(rlc_hdr->header.header_version_minor));
1338 err = amdgpu_gfx_rlc_init_microcode(adev, version_major, version_minor);
1339out:
1340 if (err) {
1341 dev_err(adev->dev,printf("drm:pid%d:%s *ERROR* " "gfx9: Failed to init firmware \"%s\"\n"
, ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__ , fw_name
)
1342 "gfx9: Failed to init firmware \"%s\"\n",printf("drm:pid%d:%s *ERROR* " "gfx9: Failed to init firmware \"%s\"\n"
, ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__ , fw_name
)
1343 fw_name)printf("drm:pid%d:%s *ERROR* " "gfx9: Failed to init firmware \"%s\"\n"
, ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__ , fw_name
);
1344 release_firmware(adev->gfx.rlc_fw);
1345 adev->gfx.rlc_fw = NULL((void *)0);
1346 }
1347 return err;
1348}
1349
1350static bool_Bool gfx_v9_0_load_mec2_fw_bin_support(struct amdgpu_device *adev)
1351{
1352 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2)(((9) << 16) | ((4) << 8) | (2)) ||
1353 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1)(((9) << 16) | ((4) << 8) | (1)) ||
1354 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 3, 0)(((9) << 16) | ((3) << 8) | (0)))
1355 return false0;
1356
1357 return true1;
1358}
1359
1360static int gfx_v9_0_init_cp_compute_microcode(struct amdgpu_device *adev,
1361 const char *chip_name)
1362{
1363 char fw_name[30];
1364 int err;
1365
1366 if (amdgpu_sriov_vf(adev)((adev)->virt.caps & (1 << 2)) && (adev->asic_type == CHIP_ALDEBARAN))
1367 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sjt_mec.bin", chip_name);
1368 else
1369 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1370
1371 err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1372 if (err)
1373 goto out;
1374 err = amdgpu_ucode_validate(adev->gfx.mec_fw);
1375 if (err)
1376 goto out;
1377 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1);
1378 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1_JT);
1379
1380 if (gfx_v9_0_load_mec2_fw_bin_support(adev)) {
1381 if (amdgpu_sriov_vf(adev)((adev)->virt.caps & (1 << 2)) && (adev->asic_type == CHIP_ALDEBARAN))
1382 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sjt_mec2.bin", chip_name);
1383 else
1384 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1385
1386 err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1387 if (!err) {
1388 err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
1389 if (err)
1390 goto out;
1391 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC2);
1392 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC2_JT);
1393 } else {
1394 err = 0;
1395 adev->gfx.mec2_fw = NULL((void *)0);
1396 }
1397 } else {
1398 adev->gfx.mec2_fw_version = adev->gfx.mec_fw_version;
1399 adev->gfx.mec2_feature_version = adev->gfx.mec_feature_version;
1400 }
1401
1402out:
1403 gfx_v9_0_check_if_need_gfxoff(adev);
1404 gfx_v9_0_check_fw_write_wait(adev);
1405 if (err) {
1406 dev_err(adev->dev,printf("drm:pid%d:%s *ERROR* " "gfx9: Failed to init firmware \"%s\"\n"
, ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__ , fw_name
)
1407 "gfx9: Failed to init firmware \"%s\"\n",printf("drm:pid%d:%s *ERROR* " "gfx9: Failed to init firmware \"%s\"\n"
, ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__ , fw_name
)
1408 fw_name)printf("drm:pid%d:%s *ERROR* " "gfx9: Failed to init firmware \"%s\"\n"
, ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__ , fw_name
);
1409 release_firmware(adev->gfx.mec_fw);
1410 adev->gfx.mec_fw = NULL((void *)0);
1411 release_firmware(adev->gfx.mec2_fw);
1412 adev->gfx.mec2_fw = NULL((void *)0);
1413 }
1414 return err;
1415}
1416
1417static int gfx_v9_0_init_microcode(struct amdgpu_device *adev)
1418{
1419 const char *chip_name;
1420 int r;
1421
1422 DRM_DEBUG("\n")___drm_dbg(((void *)0), DRM_UT_CORE, "\n");
1423
1424 switch (adev->ip_versions[GC_HWIP][0]) {
1425 case IP_VERSION(9, 0, 1)(((9) << 16) | ((0) << 8) | (1)):
1426 chip_name = "vega10";
1427 break;
1428 case IP_VERSION(9, 2, 1)(((9) << 16) | ((2) << 8) | (1)):
1429 chip_name = "vega12";
1430 break;
1431 case IP_VERSION(9, 4, 0)(((9) << 16) | ((4) << 8) | (0)):
1432 chip_name = "vega20";
1433 break;
1434 case IP_VERSION(9, 2, 2)(((9) << 16) | ((2) << 8) | (2)):
1435 case IP_VERSION(9, 1, 0)(((9) << 16) | ((1) << 8) | (0)):
1436 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
1437 chip_name = "raven2";
1438 else if (adev->apu_flags & AMD_APU_IS_PICASSO)
1439 chip_name = "picasso";
1440 else
1441 chip_name = "raven";
1442 break;
1443 case IP_VERSION(9, 4, 1)(((9) << 16) | ((4) << 8) | (1)):
1444 chip_name = "arcturus";
1445 break;
1446 case IP_VERSION(9, 3, 0)(((9) << 16) | ((3) << 8) | (0)):
1447 if (adev->apu_flags & AMD_APU_IS_RENOIR)
1448 chip_name = "renoir";
1449 else
1450 chip_name = "green_sardine";
1451 break;
1452 case IP_VERSION(9, 4, 2)(((9) << 16) | ((4) << 8) | (2)):
1453 chip_name = "aldebaran";
1454 break;
1455 default:
1456 BUG()do { panic("BUG at %s:%d", "/usr/src/sys/dev/pci/drm/amd/amdgpu/gfx_v9_0.c"
, 1456); } while (0);
1457 }
1458
1459 /* No CPG in Arcturus */
1460 if (adev->gfx.num_gfx_rings) {
1461 r = gfx_v9_0_init_cp_gfx_microcode(adev, chip_name);
1462 if (r)
1463 return r;
1464 }
1465
1466 r = gfx_v9_0_init_rlc_microcode(adev, chip_name);
1467 if (r)
1468 return r;
1469
1470 r = gfx_v9_0_init_cp_compute_microcode(adev, chip_name);
1471 if (r)
1472 return r;
1473
1474 return r;
1475}
1476
1477static u32 gfx_v9_0_get_csb_size(struct amdgpu_device *adev)
1478{
1479 u32 count = 0;
1480 const struct cs_section_def *sect = NULL((void *)0);
1481 const struct cs_extent_def *ext = NULL((void *)0);
1482
1483 /* begin clear state */
1484 count += 2;
1485 /* context control state */
1486 count += 3;
1487
1488 for (sect = gfx9_cs_data; sect->section != NULL((void *)0); ++sect) {
1489 for (ext = sect->section; ext->extent != NULL((void *)0); ++ext) {
1490 if (sect->id == SECT_CONTEXT)
1491 count += 2 + ext->reg_count;
1492 else
1493 return 0;
1494 }
1495 }
1496
1497 /* end clear state */
1498 count += 2;
1499 /* clear state */
1500 count += 2;
1501
1502 return count;
1503}
1504
1505static void gfx_v9_0_get_csb_buffer(struct amdgpu_device *adev,
1506 volatile u32 *buffer)
1507{
1508 u32 count = 0, i;
1509 const struct cs_section_def *sect = NULL((void *)0);
1510 const struct cs_extent_def *ext = NULL((void *)0);
1511
1512 if (adev->gfx.rlc.cs_data == NULL((void *)0))
1513 return;
1514 if (buffer == NULL((void *)0))
1515 return;
1516
1517 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0))((__uint32_t)(((3 << 30) | (((0x4A) & 0xFF) <<
8) | ((0) & 0x3FFF) << 16)));
1518 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE)((__uint32_t)((2 << 28)));
1519
1520 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1))((__uint32_t)(((3 << 30) | (((0x28) & 0xFF) <<
8) | ((1) & 0x3FFF) << 16)));
1521 buffer[count++] = cpu_to_le32(0x80000000)((__uint32_t)(0x80000000));
1522 buffer[count++] = cpu_to_le32(0x80000000)((__uint32_t)(0x80000000));
1523
1524 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL((void *)0); ++sect) {
1525 for (ext = sect->section; ext->extent != NULL((void *)0); ++ext) {
1526 if (sect->id == SECT_CONTEXT) {
1527 buffer[count++] =
1528 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count))((__uint32_t)(((3 << 30) | (((0x69) & 0xFF) <<
8) | ((ext->reg_count) & 0x3FFF) << 16)));
1529 buffer[count++] = cpu_to_le32(ext->reg_index -((__uint32_t)(ext->reg_index - 0x0000a000))
1530 PACKET3_SET_CONTEXT_REG_START)((__uint32_t)(ext->reg_index - 0x0000a000));
1531 for (i = 0; i < ext->reg_count; i++)
1532 buffer[count++] = cpu_to_le32(ext->extent[i])((__uint32_t)(ext->extent[i]));
1533 } else {
1534 return;
1535 }
1536 }
1537 }
1538
1539 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0))((__uint32_t)(((3 << 30) | (((0x4A) & 0xFF) <<
8) | ((0) & 0x3FFF) << 16)));
1540 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE)((__uint32_t)((3 << 28)));
1541
1542 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0))((__uint32_t)(((3 << 30) | (((0x12) & 0xFF) <<
8) | ((0) & 0x3FFF) << 16)));
1543 buffer[count++] = cpu_to_le32(0)((__uint32_t)(0));
1544}
1545
1546static void gfx_v9_0_init_always_on_cu_mask(struct amdgpu_device *adev)
1547{
1548 struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
1549 uint32_t pg_always_on_cu_num = 2;
1550 uint32_t always_on_cu_num;
1551 uint32_t i, j, k;
1552 uint32_t mask, cu_bitmap, counter;
1553
1554 if (adev->flags & AMD_IS_APU)
1555 always_on_cu_num = 4;
1556 else if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 2, 1)(((9) << 16) | ((2) << 8) | (1)))
1557 always_on_cu_num = 8;
1558 else
1559 always_on_cu_num = 12;
1560
1561 mutex_lock(&adev->grbm_idx_mutex)rw_enter_write(&adev->grbm_idx_mutex);
1562 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1563 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1564 mask = 1;
1565 cu_bitmap = 0;
1566 counter = 0;
1567 gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
1568
1569 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
1570 if (cu_info->bitmap[i][j] & mask) {
1571 if (counter == pg_always_on_cu_num)
1572 WREG32_SOC15(GC, 0, mmRLC_PG_ALWAYS_ON_CU_MASK, cu_bitmap)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c53), cu_bitmap, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4c53)), (cu_bitmap
), 0));
1573 if (counter < always_on_cu_num)
1574 cu_bitmap |= mask;
1575 else
1576 break;
1577 counter++;
1578 }
1579 mask <<= 1;
1580 }
1581
1582 WREG32_SOC15(GC, 0, mmRLC_LB_ALWAYS_ACTIVE_CU_MASK, cu_bitmap)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c50), cu_bitmap, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4c50)), (cu_bitmap
), 0));
1583 cu_info->ao_cu_bitmap[i][j] = cu_bitmap;
1584 }
1585 }
1586 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1587 mutex_unlock(&adev->grbm_idx_mutex)rw_exit_write(&adev->grbm_idx_mutex);
1588}
1589
1590static void gfx_v9_0_init_lbpw(struct amdgpu_device *adev)
1591{
1592 uint32_t data;
1593
1594 /* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1595 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4cbf), 0x0000007F, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4cbf)), (0x0000007F
), 0));
1596 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x0333A5A7)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4cb8), 0x0333A5A7, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4cb8)), (0x0333A5A7
), 0));
1597 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4cb9), 0x00000077, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4cb9)), (0x00000077
), 0));
1598 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x30 | 0x40 << 8 | 0x02FA << 16))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4cba), (0x30 | 0x40 << 8 | 0x02FA << 16), 0
, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP
][0][1] + 0x4cba)), ((0x30 | 0x40 << 8 | 0x02FA <<
16)), 0));
1599
1600 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1601 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c1b), 0x00000000, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4c1b)), (0x00000000
), 0));
1602
1603 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1604 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000500)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c12), 0x00000500, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4c12)), (0x00000500
), 0));
1605
1606 mutex_lock(&adev->grbm_idx_mutex)rw_enter_write(&adev->grbm_idx_mutex);
1607 /* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1608 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1609 WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c4f), 0xffffffff, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4c4f)), (0xffffffff
), 0));
1610
1611 /* set mmRLC_LB_PARAMS = 0x003F_1006 */
1612 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003)(((0) & ~0x000000FEL) | (0x000000FEL & ((0x0003) <<
0x1)));
1613 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010)(((data) & ~0x0000FF00L) | (0x0000FF00L & ((0x0010) <<
0x8)));
1614 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F)(((data) & ~0xFFFF0000L) | (0xFFFF0000L & ((0x033F) <<
0x10)));
1615 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c51), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c51)), (data), 0));
1616
1617 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1618 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c6a, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c6a), 0));
1619 data &= 0x0000FFFF;
1620 data |= 0x00C00000;
1621 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c6a), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c6a)), (data), 0));
1622
1623 /*
1624 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xF (4 CUs AON for Raven),
1625 * programmed in gfx_v9_0_init_always_on_cu_mask()
1626 */
1627
1628 /* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1629 * but used for RLC_LB_CNTL configuration */
1630 data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK0x00000004L;
1631 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09)(((data) & ~0x00000FF0L) | (0x00000FF0L & ((0x09) <<
0x4)));
1632 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000)(((data) & ~0xFFFFF000L) | (0xFFFFF000L & ((0x80000) <<
0xc)));
1633 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c19), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c19)), (data), 0));
1634 mutex_unlock(&adev->grbm_idx_mutex)rw_exit_write(&adev->grbm_idx_mutex);
1635
1636 gfx_v9_0_init_always_on_cu_mask(adev);
1637}
1638
1639static void gfx_v9_4_init_lbpw(struct amdgpu_device *adev)
1640{
1641 uint32_t data;
1642
1643 /* set mmRLC_LB_THR_CONFIG_1/2/3/4 */
1644 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4cbf), 0x0000007F, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4cbf)), (0x0000007F
), 0));
1645 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x033388F8)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4cb8), 0x033388F8, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4cb8)), (0x033388F8
), 0));
1646 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4cb9), 0x00000077, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4cb9)), (0x00000077
), 0));
1647 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x10 | 0x27 << 8 | 0x02FA << 16))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4cba), (0x10 | 0x27 << 8 | 0x02FA << 16), 0
, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP
][0][1] + 0x4cba)), ((0x10 | 0x27 << 8 | 0x02FA <<
16)), 0));
1648
1649 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */
1650 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c1b), 0x00000000, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4c1b)), (0x00000000
), 0));
1651
1652 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */
1653 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000800)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c12), 0x00000800, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4c12)), (0x00000800
), 0));
1654
1655 mutex_lock(&adev->grbm_idx_mutex)rw_enter_write(&adev->grbm_idx_mutex);
1656 /* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/
1657 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1658 WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c4f), 0xffffffff, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x4c4f)), (0xffffffff
), 0));
1659
1660 /* set mmRLC_LB_PARAMS = 0x003F_1006 */
1661 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003)(((0) & ~0x000000FEL) | (0x000000FEL & ((0x0003) <<
0x1)));
1662 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010)(((data) & ~0x0000FF00L) | (0x0000FF00L & ((0x0010) <<
0x8)));
1663 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F)(((data) & ~0xFFFF0000L) | (0xFFFF0000L & ((0x033F) <<
0x10)));
1664 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c51), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c51)), (data), 0));
1665
1666 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */
1667 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c6a, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c6a), 0));
1668 data &= 0x0000FFFF;
1669 data |= 0x00C00000;
1670 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c6a), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c6a)), (data), 0));
1671
1672 /*
1673 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xFFF (12 CUs AON),
1674 * programmed in gfx_v9_0_init_always_on_cu_mask()
1675 */
1676
1677 /* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved,
1678 * but used for RLC_LB_CNTL configuration */
1679 data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK0x00000004L;
1680 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09)(((data) & ~0x00000FF0L) | (0x00000FF0L & ((0x09) <<
0x4)));
1681 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000)(((data) & ~0xFFFFF000L) | (0xFFFFF000L & ((0x80000) <<
0xc)));
1682 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c19), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c19)), (data), 0));
1683 mutex_unlock(&adev->grbm_idx_mutex)rw_exit_write(&adev->grbm_idx_mutex);
1684
1685 gfx_v9_0_init_always_on_cu_mask(adev);
1686}
1687
1688static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool_Bool enable)
1689{
1690 WREG32_FIELD15(GC, 0, RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c19, (((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c19, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c19), 0)) & ~0x00000001L) |
(enable ? 1 : 0) << 0x0, 0, GC_HWIP) : amdgpu_device_wreg
(adev, (adev->reg_offset[GC_HWIP][0][1] + 0x4c19), ((((((adev
)->virt.caps & (1 << 2)) && adev->gfx
.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c19, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c19), 0)) & ~0x00000001L) |
(enable ? 1 : 0) << 0x0), 0));
1691}
1692
1693static int gfx_v9_0_cp_jump_table_num(struct amdgpu_device *adev)
1694{
1695 if (gfx_v9_0_load_mec2_fw_bin_support(adev))
1696 return 5;
1697 else
1698 return 4;
1699}
1700
1701static void gfx_v9_0_init_rlcg_reg_access_ctrl(struct amdgpu_device *adev)
1702{
1703 struct amdgpu_rlcg_reg_access_ctrl *reg_access_ctrl;
1704
1705 reg_access_ctrl = &adev->gfx.rlc.reg_access_ctrl;
1706 reg_access_ctrl->scratch_reg0 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0)(adev->reg_offset[GC_HWIP][0][1] + 0x2040);
1707 reg_access_ctrl->scratch_reg1 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG1)(adev->reg_offset[GC_HWIP][0][1] + 0x2041);
1708 reg_access_ctrl->scratch_reg2 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG2)(adev->reg_offset[GC_HWIP][0][1] + 0x2042);
1709 reg_access_ctrl->scratch_reg3 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG3)(adev->reg_offset[GC_HWIP][0][1] + 0x2043);
1710 reg_access_ctrl->grbm_cntl = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_CNTL)(adev->reg_offset[GC_HWIP][0][0] + 0x0022);
1711 reg_access_ctrl->grbm_idx = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_INDEX)(adev->reg_offset[GC_HWIP][0][1] + 0x2200);
1712 reg_access_ctrl->spare_int = SOC15_REG_OFFSET(GC, 0, mmRLC_SPARE_INT)(adev->reg_offset[GC_HWIP][0][1] + 0x4ccc);
1713 adev->gfx.rlc.rlcg_reg_access_supported = true1;
1714}
1715
1716static int gfx_v9_0_rlc_init(struct amdgpu_device *adev)
1717{
1718 const struct cs_section_def *cs_data;
1719 int r;
1720
1721 adev->gfx.rlc.cs_data = gfx9_cs_data;
1722
1723 cs_data = adev->gfx.rlc.cs_data;
1724
1725 if (cs_data) {
1726 /* init clear state block */
1727 r = amdgpu_gfx_rlc_init_csb(adev);
1728 if (r)
1729 return r;
1730 }
1731
1732 if (adev->flags & AMD_IS_APU) {
1733 /* TODO: double check the cp_table_size for RV */
1734 adev->gfx.rlc.cp_table_size = roundup2(96 * 5 * 4, 2048)(((96 * 5 * 4) + ((2048) - 1)) & (~((__typeof(96 * 5 * 4)
)(2048) - 1))) + (64 * 1024); /* JT + GDS */
1735 r = amdgpu_gfx_rlc_init_cpt(adev);
1736 if (r)
1737 return r;
1738 }
1739
1740 switch (adev->ip_versions[GC_HWIP][0]) {
1741 case IP_VERSION(9, 2, 2)(((9) << 16) | ((2) << 8) | (2)):
1742 case IP_VERSION(9, 1, 0)(((9) << 16) | ((1) << 8) | (0)):
1743 gfx_v9_0_init_lbpw(adev);
1744 break;
1745 case IP_VERSION(9, 4, 0)(((9) << 16) | ((4) << 8) | (0)):
1746 gfx_v9_4_init_lbpw(adev);
1747 break;
1748 default:
1749 break;
1750 }
1751
1752 /* init spm vmid with 0xf */
1753 if (adev->gfx.rlc.funcs->update_spm_vmid)
1754 adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
1755
1756 return 0;
1757}
1758
1759static void gfx_v9_0_mec_fini(struct amdgpu_device *adev)
1760{
1761 amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL((void *)0), NULL((void *)0));
1762 amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL((void *)0), NULL((void *)0));
1763}
1764
1765static int gfx_v9_0_mec_init(struct amdgpu_device *adev)
1766{
1767 int r;
1768 u32 *hpd;
1769 const __le32 *fw_data;
1770 unsigned fw_size;
1771 u32 *fw;
1772 size_t mec_hpd_size;
1773
1774 const struct gfx_firmware_header_v1_0 *mec_hdr;
1775
1776 bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES128);
1777
1778 /* take ownership of the relevant compute queues */
1779 amdgpu_gfx_compute_queue_acquire(adev);
1780 mec_hpd_size = adev->gfx.num_compute_rings * GFX9_MEC_HPD_SIZE4096;
1781 if (mec_hpd_size) {
1782 r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE(1 << 12),
1783 AMDGPU_GEM_DOMAIN_VRAM0x4,
1784 &adev->gfx.mec.hpd_eop_obj,
1785 &adev->gfx.mec.hpd_eop_gpu_addr,
1786 (void **)&hpd);
1787 if (r) {
1788 dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r)printf("drm:pid%d:%s *WARNING* " "(%d) create HDP EOP bo failed\n"
, ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__ , r);
1789 gfx_v9_0_mec_fini(adev);
1790 return r;
1791 }
1792
1793 memset(hpd, 0, mec_hpd_size)__builtin_memset((hpd), (0), (mec_hpd_size));
1794
1795 amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1796 amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1797 }
1798
1799 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1800
1801 fw_data = (const __le32 *)
1802 (adev->gfx.mec_fw->data +
1803 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes)((__uint32_t)(mec_hdr->header.ucode_array_offset_bytes)));
1804 fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes)((__uint32_t)(mec_hdr->header.ucode_size_bytes));
1805
1806 r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
1807 PAGE_SIZE(1 << 12), AMDGPU_GEM_DOMAIN_GTT0x2,
1808 &adev->gfx.mec.mec_fw_obj,
1809 &adev->gfx.mec.mec_fw_gpu_addr,
1810 (void **)&fw);
1811 if (r) {
1812 dev_warn(adev->dev, "(%d) create mec firmware bo failed\n", r)printf("drm:pid%d:%s *WARNING* " "(%d) create mec firmware bo failed\n"
, ({struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__ , r);
1813 gfx_v9_0_mec_fini(adev);
1814 return r;
1815 }
1816
1817 memcpy(fw, fw_data, fw_size)__builtin_memcpy((fw), (fw_data), (fw_size));
1818
1819 amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
1820 amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
1821
1822 return 0;
1823}
1824
1825static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
1826{
1827 WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (address << 0x10) | (0x00002000L
), (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), ((wave << 0x0) | (simd << 0x4) | (address <<
0x10) | (0x00002000L)), 0)); } while (0)
1828 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (address << 0x10) | (0x00002000L
), (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), ((wave << 0x0) | (simd << 0x4) | (address <<
0x10) | (0x00002000L)), 0)); } while (0)
1829 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (address << 0x10) | (0x00002000L
), (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), ((wave << 0x0) | (simd << 0x4) | (address <<
0x10) | (0x00002000L)), 0)); } while (0)
1830 (address << SQ_IND_INDEX__INDEX__SHIFT) |do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (address << 0x10) | (0x00002000L
), (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), ((wave << 0x0) | (simd << 0x4) | (address <<
0x10) | (0x00002000L)), 0)); } while (0)
1831 (SQ_IND_INDEX__FORCE_READ_MASK))do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (address << 0x10) | (0x00002000L
), (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), ((wave << 0x0) | (simd << 0x4) | (address <<
0x10) | (0x00002000L)), 0)); } while (0);
1832 return RREG32_SOC15(GC, 0, mmSQ_IND_DATA)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0379, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0379), 0));
1833}
1834
1835static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
1836 uint32_t wave, uint32_t thread,
1837 uint32_t regno, uint32_t num, uint32_t *out)
1838{
1839 WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L), (1<<2), GC_HWIP)
: amdgpu_device_wreg(adev, (target_reg), ((wave << 0x0
) | (simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L)), 0)); } while (0)
1840 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L), (1<<2), GC_HWIP)
: amdgpu_device_wreg(adev, (target_reg), ((wave << 0x0
) | (simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L)), 0)); } while (0)
1841 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L), (1<<2), GC_HWIP)
: amdgpu_device_wreg(adev, (target_reg), ((wave << 0x0
) | (simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L)), 0)); } while (0)
1842 (regno << SQ_IND_INDEX__INDEX__SHIFT) |do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L), (1<<2), GC_HWIP)
: amdgpu_device_wreg(adev, (target_reg), ((wave << 0x0
) | (simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L)), 0)); } while (0)
1843 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L), (1<<2), GC_HWIP)
: amdgpu_device_wreg(adev, (target_reg), ((wave << 0x0
) | (simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L)), 0)); } while (0)
1844 (SQ_IND_INDEX__FORCE_READ_MASK) |do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L), (1<<2), GC_HWIP)
: amdgpu_device_wreg(adev, (target_reg), ((wave << 0x0
) | (simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L)), 0)); } while (0)
1845 (SQ_IND_INDEX__AUTO_INCR_MASK))do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x0378; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (wave << 0x0) |
(simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L), (1<<2), GC_HWIP)
: amdgpu_device_wreg(adev, (target_reg), ((wave << 0x0
) | (simd << 0x4) | (regno << 0x10) | (thread <<
0x6) | (0x00002000L) | (0x00001000L)), 0)); } while (0);
1846 while (num--)
1847 *(out++) = RREG32_SOC15(GC, 0, mmSQ_IND_DATA)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0379, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0379), 0));
1848}
1849
1850static void gfx_v9_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
1851{
1852 /* type 1 wave data */
1853 dst[(*no_fields)++] = 1;
1854 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS0x0012);
1855 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO0x0018);
1856 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI0x0019);
1857 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO0x027e);
1858 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI0x027f);
1859 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID0x0014);
1860 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW00x001a);
1861 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW10x001b);
1862 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC0x0015);
1863 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC0x0016);
1864 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS0x0013);
1865 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS0x0017);
1866 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG00x001c);
1867 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M00x027c);
1868 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE0x0011);
1869}
1870
1871static void gfx_v9_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
1872 uint32_t wave, uint32_t start,
1873 uint32_t size, uint32_t *dst)
1874{
1875 wave_read_regs(
1876 adev, simd, wave, 0,
1877 start + SQIND_WAVE_SGPRS_OFFSET0x00000200, size, dst);
1878}
1879
1880static void gfx_v9_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t simd,
1881 uint32_t wave, uint32_t thread,
1882 uint32_t start, uint32_t size,
1883 uint32_t *dst)
1884{
1885 wave_read_regs(
1886 adev, simd, wave, thread,
1887 start + SQIND_WAVE_VGPRS_OFFSET0x00000400, size, dst);
1888}
1889
1890static void gfx_v9_0_select_me_pipe_q(struct amdgpu_device *adev,
1891 u32 me, u32 pipe, u32 q, u32 vm)
1892{
1893 soc15_grbm_select(adev, me, pipe, q, vm);
1894}
1895
1896static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = {
1897 .get_gpu_clock_counter = &gfx_v9_0_get_gpu_clock_counter,
1898 .select_se_sh = &gfx_v9_0_select_se_sh,
1899 .read_wave_data = &gfx_v9_0_read_wave_data,
1900 .read_wave_sgprs = &gfx_v9_0_read_wave_sgprs,
1901 .read_wave_vgprs = &gfx_v9_0_read_wave_vgprs,
1902 .select_me_pipe_q = &gfx_v9_0_select_me_pipe_q,
1903};
1904
1905const struct amdgpu_ras_block_hw_ops gfx_v9_0_ras_ops = {
1906 .ras_error_inject = &gfx_v9_0_ras_error_inject,
1907 .query_ras_error_count = &gfx_v9_0_query_ras_error_count,
1908 .reset_ras_error_count = &gfx_v9_0_reset_ras_error_count,
1909};
1910
1911static struct amdgpu_gfx_ras gfx_v9_0_ras = {
1912 .ras_block = {
1913 .hw_ops = &gfx_v9_0_ras_ops,
1914 },
1915};
1916
1917static int gfx_v9_0_gpu_early_init(struct amdgpu_device *adev)
1918{
1919 u32 gb_addr_config;
1920 int err;
1921
1922 adev->gfx.funcs = &gfx_v9_0_gfx_funcs;
1923
1924 switch (adev->ip_versions[GC_HWIP][0]) {
1925 case IP_VERSION(9, 0, 1)(((9) << 16) | ((0) << 8) | (1)):
1926 adev->gfx.config.max_hw_contexts = 8;
1927 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1928 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1929 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1930 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1931 gb_addr_config = VEGA10_GB_ADDR_CONFIG_GOLDEN0x2a114042;
1932 break;
1933 case IP_VERSION(9, 2, 1)(((9) << 16) | ((2) << 8) | (1)):
1934 adev->gfx.config.max_hw_contexts = 8;
1935 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1936 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1937 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1938 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1939 gb_addr_config = VEGA12_GB_ADDR_CONFIG_GOLDEN0x24104041;
1940 DRM_INFO("fix gfx.config for vega12\n")printk("\0016" "[" "drm" "] " "fix gfx.config for vega12\n");
1941 break;
1942 case IP_VERSION(9, 4, 0)(((9) << 16) | ((4) << 8) | (0)):
1943 adev->gfx.ras = &gfx_v9_0_ras;
1944 adev->gfx.config.max_hw_contexts = 8;
1945 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1946 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1947 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1948 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1949 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x063e, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x063e), 0));
1950 gb_addr_config &= ~0xf3e777ff;
1951 gb_addr_config |= 0x22014042;
1952 /* check vbios table if gpu info is not available */
1953 err = amdgpu_atomfirmware_get_gfx_info(adev);
1954 if (err)
1955 return err;
1956 break;
1957 case IP_VERSION(9, 2, 2)(((9) << 16) | ((2) << 8) | (2)):
1958 case IP_VERSION(9, 1, 0)(((9) << 16) | ((1) << 8) | (0)):
1959 adev->gfx.config.max_hw_contexts = 8;
1960 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1961 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1962 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1963 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1964 if (adev->apu_flags & AMD_APU_IS_RAVEN2)
1965 gb_addr_config = RAVEN2_GB_ADDR_CONFIG_GOLDEN0x26013041;
1966 else
1967 gb_addr_config = RAVEN_GB_ADDR_CONFIG_GOLDEN0x24000042;
1968 break;
1969 case IP_VERSION(9, 4, 1)(((9) << 16) | ((4) << 8) | (1)):
1970 adev->gfx.ras = &gfx_v9_4_ras;
1971 adev->gfx.config.max_hw_contexts = 8;
1972 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1973 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1974 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1975 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1976 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x063e, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x063e), 0));
1977 gb_addr_config &= ~0xf3e777ff;
1978 gb_addr_config |= 0x22014042;
1979 break;
1980 case IP_VERSION(9, 3, 0)(((9) << 16) | ((3) << 8) | (0)):
1981 adev->gfx.config.max_hw_contexts = 8;
1982 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1983 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1984 adev->gfx.config.sc_hiz_tile_fifo_size = 0x80;
1985 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1986 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x063e, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x063e), 0));
1987 gb_addr_config &= ~0xf3e777ff;
1988 gb_addr_config |= 0x22010042;
1989 break;
1990 case IP_VERSION(9, 4, 2)(((9) << 16) | ((4) << 8) | (2)):
1991 adev->gfx.ras = &gfx_v9_4_2_ras;
1992 adev->gfx.config.max_hw_contexts = 8;
1993 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1994 adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1995 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1996 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1997 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x063e, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x063e), 0));
1998 gb_addr_config &= ~0xf3e777ff;
1999 gb_addr_config |= 0x22014042;
2000 /* check vbios table if gpu info is not available */
2001 err = amdgpu_atomfirmware_get_gfx_info(adev);
2002 if (err)
2003 return err;
2004 break;
2005 default:
2006 BUG()do { panic("BUG at %s:%d", "/usr/src/sys/dev/pci/drm/amd/amdgpu/gfx_v9_0.c"
, 2006); } while (0);
2007 break;
2008 }
2009
2010 if (adev->gfx.ras) {
2011 err = amdgpu_ras_register_ras_block(adev, &adev->gfx.ras->ras_block);
2012 if (err) {
2013 DRM_ERROR("Failed to register gfx ras block!\n")__drm_err("Failed to register gfx ras block!\n");
2014 return err;
2015 }
2016
2017 strlcpy(adev->gfx.ras->ras_block.ras_comm.name, "gfx",
2018 sizeof(adev->gfx.ras->ras_block.ras_comm.name));
2019 adev->gfx.ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__GFX;
2020 adev->gfx.ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
2021 adev->gfx.ras_if = &adev->gfx.ras->ras_block.ras_comm;
2022
2023 /* If not define special ras_late_init function, use gfx default ras_late_init */
2024 if (!adev->gfx.ras->ras_block.ras_late_init)
2025 adev->gfx.ras->ras_block.ras_late_init = amdgpu_gfx_ras_late_init;
2026
2027 /* If not defined special ras_cb function, use default ras_cb */
2028 if (!adev->gfx.ras->ras_block.ras_cb)
2029 adev->gfx.ras->ras_block.ras_cb = amdgpu_gfx_process_ras_data_cb;
2030 }
2031
2032 adev->gfx.config.gb_addr_config = gb_addr_config;
2033
2034 adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
2035 REG_GET_FIELD((((adev->gfx.config.gb_addr_config) & 0x00000007L) >>
0x0)
2036 adev->gfx.config.gb_addr_config,(((adev->gfx.config.gb_addr_config) & 0x00000007L) >>
0x0)
2037 GB_ADDR_CONFIG,(((adev->gfx.config.gb_addr_config) & 0x00000007L) >>
0x0)
2038 NUM_PIPES)(((adev->gfx.config.gb_addr_config) & 0x00000007L) >>
0x0);
2039
2040 adev->gfx.config.max_tile_pipes =
2041 adev->gfx.config.gb_addr_config_fields.num_pipes;
2042
2043 adev->gfx.config.gb_addr_config_fields.num_banks = 1 <<
2044 REG_GET_FIELD((((adev->gfx.config.gb_addr_config) & 0x00007000L) >>
0xc)
2045 adev->gfx.config.gb_addr_config,(((adev->gfx.config.gb_addr_config) & 0x00007000L) >>
0xc)
2046 GB_ADDR_CONFIG,(((adev->gfx.config.gb_addr_config) & 0x00007000L) >>
0xc)
2047 NUM_BANKS)(((adev->gfx.config.gb_addr_config) & 0x00007000L) >>
0xc);
2048 adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
2049 REG_GET_FIELD((((adev->gfx.config.gb_addr_config) & 0x000000C0L) >>
0x6)
2050 adev->gfx.config.gb_addr_config,(((adev->gfx.config.gb_addr_config) & 0x000000C0L) >>
0x6)
2051 GB_ADDR_CONFIG,(((adev->gfx.config.gb_addr_config) & 0x000000C0L) >>
0x6)
2052 MAX_COMPRESSED_FRAGS)(((adev->gfx.config.gb_addr_config) & 0x000000C0L) >>
0x6);
2053 adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
2054 REG_GET_FIELD((((adev->gfx.config.gb_addr_config) & 0x0C000000L) >>
0x1a)
2055 adev->gfx.config.gb_addr_config,(((adev->gfx.config.gb_addr_config) & 0x0C000000L) >>
0x1a)
2056 GB_ADDR_CONFIG,(((adev->gfx.config.gb_addr_config) & 0x0C000000L) >>
0x1a)
2057 NUM_RB_PER_SE)(((adev->gfx.config.gb_addr_config) & 0x0C000000L) >>
0x1a);
2058 adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
2059 REG_GET_FIELD((((adev->gfx.config.gb_addr_config) & 0x00180000L) >>
0x13)
2060 adev->gfx.config.gb_addr_config,(((adev->gfx.config.gb_addr_config) & 0x00180000L) >>
0x13)
2061 GB_ADDR_CONFIG,(((adev->gfx.config.gb_addr_config) & 0x00180000L) >>
0x13)
2062 NUM_SHADER_ENGINES)(((adev->gfx.config.gb_addr_config) & 0x00180000L) >>
0x13);
2063 adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
2064 REG_GET_FIELD((((adev->gfx.config.gb_addr_config) & 0x00000038L) >>
0x3)
2065 adev->gfx.config.gb_addr_config,(((adev->gfx.config.gb_addr_config) & 0x00000038L) >>
0x3)
2066 GB_ADDR_CONFIG,(((adev->gfx.config.gb_addr_config) & 0x00000038L) >>
0x3)
2067 PIPE_INTERLEAVE_SIZE)(((adev->gfx.config.gb_addr_config) & 0x00000038L) >>
0x3));
2068
2069 return 0;
2070}
2071
2072static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
2073 int mec, int pipe, int queue)
2074{
2075 unsigned irq_type;
2076 struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
Value stored to 'ring' during its initialization is never read
2077 unsigned int hw_prio;
2078
2079 ring = &adev->gfx.compute_ring[ring_id];
2080
2081 /* mec0 is me1 */
2082 ring->me = mec + 1;
2083 ring->pipe = pipe;
2084 ring->queue = queue;
2085
2086 ring->ring_obj = NULL((void *)0);
2087 ring->use_doorbell = true1;
2088 ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1;
2089 ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
2090 + (ring_id * GFX9_MEC_HPD_SIZE4096);
2091 snprintf(ring->name, sizeof(ring->name), "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
2092
2093 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
2094 + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
2095 + ring->pipe;
2096 hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
2097 AMDGPU_RING_PRIO_2 : AMDGPU_RING_PRIO_DEFAULT;
2098 /* type-2 packets are deprecated on MEC, use type-3 instead */
2099 return amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
2100 hw_prio, NULL((void *)0));
2101}
2102
2103static int gfx_v9_0_sw_init(void *handle)
2104{
2105 int i, j, k, r, ring_id;
2106 struct amdgpu_ring *ring;
2107 struct amdgpu_kiq *kiq;
2108 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2109
2110 switch (adev->ip_versions[GC_HWIP][0]) {
2111 case IP_VERSION(9, 0, 1)(((9) << 16) | ((0) << 8) | (1)):
2112 case IP_VERSION(9, 2, 1)(((9) << 16) | ((2) << 8) | (1)):
2113 case IP_VERSION(9, 4, 0)(((9) << 16) | ((4) << 8) | (0)):
2114 case IP_VERSION(9, 2, 2)(((9) << 16) | ((2) << 8) | (2)):
2115 case IP_VERSION(9, 1, 0)(((9) << 16) | ((1) << 8) | (0)):
2116 case IP_VERSION(9, 4, 1)(((9) << 16) | ((4) << 8) | (1)):
2117 case IP_VERSION(9, 3, 0)(((9) << 16) | ((3) << 8) | (0)):
2118 case IP_VERSION(9, 4, 2)(((9) << 16) | ((4) << 8) | (2)):
2119 adev->gfx.mec.num_mec = 2;
2120 break;
2121 default:
2122 adev->gfx.mec.num_mec = 1;
2123 break;
2124 }
2125
2126 adev->gfx.mec.num_pipe_per_mec = 4;
2127 adev->gfx.mec.num_queue_per_pipe = 8;
2128
2129 /* EOP Event */
2130 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_EOP_INTERRUPT181, &adev->gfx.eop_irq);
2131 if (r)
2132 return r;
2133
2134 /* Privileged reg */
2135 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_REG_FAULT184,
2136 &adev->gfx.priv_reg_irq);
2137 if (r)
2138 return r;
2139
2140 /* Privileged inst */
2141 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_INSTR_FAULT185,
2142 &adev->gfx.priv_inst_irq);
2143 if (r)
2144 return r;
2145
2146 /* ECC error */
2147 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_ECC_ERROR197,
2148 &adev->gfx.cp_ecc_error_irq);
2149 if (r)
2150 return r;
2151
2152 /* FUE error */
2153 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_FUE_ERROR201,
2154 &adev->gfx.cp_ecc_error_irq);
2155 if (r)
2156 return r;
2157
2158 adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE0x00000000L;
2159
2160 r = gfx_v9_0_init_microcode(adev);
2161 if (r) {
2162 DRM_ERROR("Failed to load gfx firmware!\n")__drm_err("Failed to load gfx firmware!\n");
2163 return r;
2164 }
2165
2166 if (adev->gfx.rlc.funcs) {
2167 if (adev->gfx.rlc.funcs->init) {
2168 r = adev->gfx.rlc.funcs->init(adev);
2169 if (r) {
2170 dev_err(adev->dev, "Failed to init rlc BOs!\n")printf("drm:pid%d:%s *ERROR* " "Failed to init rlc BOs!\n", (
{struct cpu_info *__ci; asm volatile("movq %%gs:%P1,%0" : "=r"
(__ci) :"n" (__builtin_offsetof(struct cpu_info, ci_self)));
__ci;})->ci_curproc->p_p->ps_pid, __func__);
2171 return r;
2172 }
2173 }
2174 }
2175
2176 r = gfx_v9_0_mec_init(adev);
2177 if (r) {
2178 DRM_ERROR("Failed to init MEC BOs!\n")__drm_err("Failed to init MEC BOs!\n");
2179 return r;
2180 }
2181
2182 /* set up the gfx ring */
2183 for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
2184 ring = &adev->gfx.gfx_ring[i];
2185 ring->ring_obj = NULL((void *)0);
2186 if (!i)
2187 snprintf(ring->name, sizeof(ring->name), "gfx");
2188 else
2189 snprintf(ring->name, sizeof(ring->name), "gfx_%d", i);
2190 ring->use_doorbell = true1;
2191 ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
2192 r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
2193 AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
2194 AMDGPU_RING_PRIO_DEFAULT, NULL((void *)0));
2195 if (r)
2196 return r;
2197 }
2198
2199 /* set up the compute queues - allocate horizontally across pipes */
2200 ring_id = 0;
2201 for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
2202 for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2203 for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2204 if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
2205 continue;
2206
2207 r = gfx_v9_0_compute_ring_init(adev,
2208 ring_id,
2209 i, k, j);
2210 if (r)
2211 return r;
2212
2213 ring_id++;
2214 }
2215 }
2216 }
2217
2218 r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE4096);
2219 if (r) {
2220 DRM_ERROR("Failed to init KIQ BOs!\n")__drm_err("Failed to init KIQ BOs!\n");
2221 return r;
2222 }
2223
2224 kiq = &adev->gfx.kiq;
2225 r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
2226 if (r)
2227 return r;
2228
2229 /* create MQD for all compute queues as wel as KIQ for SRIOV case */
2230 r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v9_mqd_allocation));
2231 if (r)
2232 return r;
2233
2234 adev->gfx.ce_ram_size = 0x8000;
2235
2236 r = gfx_v9_0_gpu_early_init(adev);
2237 if (r)
2238 return r;
2239
2240 return 0;
2241}
2242
2243
2244static int gfx_v9_0_sw_fini(void *handle)
2245{
2246 int i;
2247 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2248
2249 for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2250 amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2251 for (i = 0; i < adev->gfx.num_compute_rings; i++)
2252 amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2253
2254 amdgpu_gfx_mqd_sw_fini(adev);
2255 amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring);
2256 amdgpu_gfx_kiq_fini(adev);
2257
2258 gfx_v9_0_mec_fini(adev);
2259 amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
2260 &adev->gfx.rlc.clear_state_gpu_addr,
2261 (void **)&adev->gfx.rlc.cs_ptr);
2262 if (adev->flags & AMD_IS_APU) {
2263 amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2264 &adev->gfx.rlc.cp_table_gpu_addr,
2265 (void **)&adev->gfx.rlc.cp_table_ptr);
2266 }
2267 gfx_v9_0_free_microcode(adev);
2268
2269 return 0;
2270}
2271
2272
2273static void gfx_v9_0_tiling_mode_table_init(struct amdgpu_device *adev)
2274{
2275 /* TODO */
2276}
2277
2278void gfx_v9_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num,
2279 u32 instance)
2280{
2281 u32 data;
2282
2283 if (instance == 0xffffffff)
2284 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1)(((0) & ~0x40000000L) | (0x40000000L & ((1) << 0x1e
)));
2285 else
2286 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance)(((0) & ~0x000000FFL) | (0x000000FFL & ((instance) <<
0x0)));
2287
2288 if (se_num == 0xffffffff)
2289 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1)(((data) & ~0x80000000L) | (0x80000000L & ((1) <<
0x1f)));
2290 else
2291 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num)(((data) & ~0x00FF0000L) | (0x00FF0000L & ((se_num) <<
0x10)));
2292
2293 if (sh_num == 0xffffffff)
2294 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1)(((data) & ~0x20000000L) | (0x20000000L & ((1) <<
0x1d)));
2295 else
2296 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num)(((data) & ~0x0000FF00L) | (0x0000FF00L & ((sh_num) <<
0x8)));
2297
2298 WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_INDEX, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x2200), data, (1<<2), GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x2200)), (data
), 0));
2299}
2300
2301static u32 gfx_v9_0_get_rb_active_bitmap(struct amdgpu_device *adev)
2302{
2303 u32 data, mask;
2304
2305 data = RREG32_SOC15(GC, 0, mmCC_RB_BACKEND_DISABLE)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x063d, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x063d), 0));
2306 data |= RREG32_SOC15(GC, 0, mmGC_USER_RB_BACKEND_DISABLE)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x06df, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x06df), 0));
2307
2308 data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK0x00FF0000L;
2309 data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT0x10;
2310
2311 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
2312 adev->gfx.config.max_sh_per_se);
2313
2314 return (~data) & mask;
2315}
2316
2317static void gfx_v9_0_setup_rb(struct amdgpu_device *adev)
2318{
2319 int i, j;
2320 u32 data;
2321 u32 active_rbs = 0;
2322 u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
2323 adev->gfx.config.max_sh_per_se;
2324
2325 mutex_lock(&adev->grbm_idx_mutex)rw_enter_write(&adev->grbm_idx_mutex);
2326 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2327 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2328 gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
2329 data = gfx_v9_0_get_rb_active_bitmap(adev);
2330 active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
2331 rb_bitmap_width_per_sh);
2332 }
2333 }
2334 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
2335 mutex_unlock(&adev->grbm_idx_mutex)rw_exit_write(&adev->grbm_idx_mutex);
2336
2337 adev->gfx.config.backend_enable_mask = active_rbs;
2338 adev->gfx.config.num_rbs = hweight32(active_rbs);
2339}
2340
2341#define DEFAULT_SH_MEM_BASES(0x6000) (0x6000)
2342static void gfx_v9_0_init_compute_vmid(struct amdgpu_device *adev)
2343{
2344 int i;
2345 uint32_t sh_mem_config;
2346 uint32_t sh_mem_bases;
2347
2348 /*
2349 * Configure apertures:
2350 * LDS: 0x60000000'00000000 - 0x60000001'00000000 (4GB)
2351 * Scratch: 0x60000001'00000000 - 0x60000002'00000000 (4GB)
2352 * GPUVM: 0x60010000'00000000 - 0x60020000'00000000 (1TB)
2353 */
2354 sh_mem_bases = DEFAULT_SH_MEM_BASES(0x6000) | (DEFAULT_SH_MEM_BASES(0x6000) << 16);
2355
2356 sh_mem_config = SH_MEM_ADDRESS_MODE_64 |
2357 SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
2358 SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT0x3;
2359
2360 mutex_lock(&adev->srbm_mutex)rw_enter_write(&adev->srbm_mutex);
2361 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID16; i++) {
2362 soc15_grbm_select(adev, 0, 0, 0, i);
2363 /* CP and shaders */
2364 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, sh_mem_config)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x030d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, sh_mem_config, (1<<
2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg), (sh_mem_config
), 0)); } while (0);
2365 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, sh_mem_bases)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x030a; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, sh_mem_bases, (1<<
2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg), (sh_mem_bases
), 0)); } while (0);
2366 }
2367 soc15_grbm_select(adev, 0, 0, 0, 0);
2368 mutex_unlock(&adev->srbm_mutex)rw_exit_write(&adev->srbm_mutex);
2369
2370 /* Initialize all compute VMIDs to have no GDS, GWS, or OA
2371 access. These should be enabled by FW for target VMIDs. */
2372 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID16; i++) {
2373 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * i, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1300) + 2 * i, 0, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][0] + 0x1300) + 2 * i), (0
), 0));
2374 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * i, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1301) + 2 * i, 0, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][0] + 0x1301) + 2 * i), (0
), 0));
2375 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, i, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1320) + i, 0, 0, GC_HWIP) : amdgpu_device_wreg(adev, (
(adev->reg_offset[GC_HWIP][0][0] + 0x1320) + i), (0), 0));
2376 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, i, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1330) + i, 0, 0, GC_HWIP) : amdgpu_device_wreg(adev, (
(adev->reg_offset[GC_HWIP][0][0] + 0x1330) + i), (0), 0));
2377 }
2378}
2379
2380static void gfx_v9_0_init_gds_vmid(struct amdgpu_device *adev)
2381{
2382 int vmid;
2383
2384 /*
2385 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
2386 * access. Compute VMIDs should be enabled by FW for target VMIDs,
2387 * the driver can enable them for graphics. VMID0 should maintain
2388 * access so that HWS firmware can save/restore entries.
2389 */
2390 for (vmid = 1; vmid < AMDGPU_NUM_VMID16; vmid++) {
2391 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * vmid, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1300) + 2 * vmid, 0, 0, GC_HWIP) : amdgpu_device_wreg(
adev, ((adev->reg_offset[GC_HWIP][0][0] + 0x1300) + 2 * vmid
), (0), 0));
2392 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * vmid, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1301) + 2 * vmid, 0, 0, GC_HWIP) : amdgpu_device_wreg(
adev, ((adev->reg_offset[GC_HWIP][0][0] + 0x1301) + 2 * vmid
), (0), 0));
2393 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, vmid, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1320) + vmid, 0, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][0] + 0x1320) + vmid), (0)
, 0));
2394 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, vmid, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1330) + vmid, 0, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][0] + 0x1330) + vmid), (0)
, 0));
2395 }
2396}
2397
2398static void gfx_v9_0_init_sq_config(struct amdgpu_device *adev)
2399{
2400 uint32_t tmp;
2401
2402 switch (adev->ip_versions[GC_HWIP][0]) {
2403 case IP_VERSION(9, 4, 1)(((9) << 16) | ((4) << 8) | (1)):
2404 tmp = RREG32_SOC15(GC, 0, mmSQ_CONFIG)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0300, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0300), 0));
2405 tmp = REG_SET_FIELD(tmp, SQ_CONFIG,(((tmp) & ~0x00000001L) | (0x00000001L & ((1) <<
0x0)))
2406 DISABLE_BARRIER_WAITCNT, 1)(((tmp) & ~0x00000001L) | (0x00000001L & ((1) <<
0x0)));
2407 WREG32_SOC15(GC, 0, mmSQ_CONFIG, tmp)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x0300), tmp, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x0300)), (tmp), 0));
2408 break;
2409 default:
2410 break;
2411 }
2412}
2413
2414static void gfx_v9_0_constants_init(struct amdgpu_device *adev)
2415{
2416 u32 tmp;
2417 int i;
2418
2419 WREG32_FIELD15_RLC(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x0000), (((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0000, (1<<2), GC_HWIP) : amdgpu_device_rreg(adev,
(adev->reg_offset[GC_HWIP][0][0] + 0x0000), 0)) & ~0x000000FFL
) | (0xff) << 0x0, (1<<2), GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][0] + 0x0000)), (((((
(adev)->virt.caps & (1 << 2)) && adev->
gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0000, (1<<2), GC_HWIP) : amdgpu_device_rreg(adev,
(adev->reg_offset[GC_HWIP][0][0] + 0x0000), 0)) & ~0x000000FFL
) | (0xff) << 0x0), 0));
2420
2421 gfx_v9_0_tiling_mode_table_init(adev);
2422
2423 if (adev->gfx.num_gfx_rings)
2424 gfx_v9_0_setup_rb(adev);
2425 gfx_v9_0_get_cu_info(adev, &adev->gfx.cu_info);
2426 adev->gfx.config.db_debug2 = RREG32_SOC15(GC, 0, mmDB_DEBUG2)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x060d, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x060d), 0));
2427
2428 /* XXX SH_MEM regs */
2429 /* where to put LDS, scratch, GPUVM in FSA64 space */
2430 mutex_lock(&adev->srbm_mutex)rw_enter_write(&adev->srbm_mutex);
2431 for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_00].num_ids; i++) {
2432 soc15_grbm_select(adev, 0, 0, 0, i);
2433 /* CP and shaders */
2434 if (i == 0) {
2435 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,(((0) & ~0x00000018L) | (0x00000018L & ((SH_MEM_ALIGNMENT_MODE_UNALIGNED
) << 0x3)))
2436 SH_MEM_ALIGNMENT_MODE_UNALIGNED)(((0) & ~0x00000018L) | (0x00000018L & ((SH_MEM_ALIGNMENT_MODE_UNALIGNED
) << 0x3)));
2437 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,(((tmp) & ~0x00001000L) | (0x00001000L & ((!!adev->
gmc.noretry) << 0xc)))
2438 !!adev->gmc.noretry)(((tmp) & ~0x00001000L) | (0x00001000L & ((!!adev->
gmc.noretry) << 0xc)));
2439 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x030d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, tmp, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (tmp), 0)); } while
(0);
2440 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, 0)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x030a; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0);
2441 } else {
2442 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,(((0) & ~0x00000018L) | (0x00000018L & ((SH_MEM_ALIGNMENT_MODE_UNALIGNED
) << 0x3)))
2443 SH_MEM_ALIGNMENT_MODE_UNALIGNED)(((0) & ~0x00000018L) | (0x00000018L & ((SH_MEM_ALIGNMENT_MODE_UNALIGNED
) << 0x3)));
2444 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,(((tmp) & ~0x00001000L) | (0x00001000L & ((!!adev->
gmc.noretry) << 0xc)))
2445 !!adev->gmc.noretry)(((tmp) & ~0x00001000L) | (0x00001000L & ((!!adev->
gmc.noretry) << 0xc)));
2446 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x030d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, tmp, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (tmp), 0)); } while
(0);
2447 tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,(((0) & ~0x0000FFFFL) | (0x0000FFFFL & (((adev->gmc
.private_aperture_start >> 48)) << 0x0)))
2448 (adev->gmc.private_aperture_start >> 48))(((0) & ~0x0000FFFFL) | (0x0000FFFFL & (((adev->gmc
.private_aperture_start >> 48)) << 0x0)));
2449 tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,(((tmp) & ~0xFFFF0000L) | (0xFFFF0000L & (((adev->
gmc.shared_aperture_start >> 48)) << 0x10)))
2450 (adev->gmc.shared_aperture_start >> 48))(((tmp) & ~0xFFFF0000L) | (0xFFFF0000L & (((adev->
gmc.shared_aperture_start >> 48)) << 0x10)));
2451 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, tmp)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x030a; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, tmp, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (tmp), 0)); } while
(0);
2452 }
2453 }
2454 soc15_grbm_select(adev, 0, 0, 0, 0);
2455
2456 mutex_unlock(&adev->srbm_mutex)rw_exit_write(&adev->srbm_mutex);
2457
2458 gfx_v9_0_init_compute_vmid(adev);
2459 gfx_v9_0_init_gds_vmid(adev);
2460 gfx_v9_0_init_sq_config(adev);
2461}
2462
2463static void gfx_v9_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
2464{
2465 u32 i, j, k;
2466 u32 mask;
2467
2468 mutex_lock(&adev->grbm_idx_mutex)rw_enter_write(&adev->grbm_idx_mutex);
2469 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
2470 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
2471 gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff);
2472 for (k = 0; k < adev->usec_timeout; k++) {
2473 if (RREG32_SOC15(GC, 0, mmRLC_SERDES_CU_MASTER_BUSY)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c61, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c61), 0)) == 0)
2474 break;
2475 udelay(1);
2476 }
2477 if (k == adev->usec_timeout) {
2478 gfx_v9_0_select_se_sh(adev, 0xffffffff,
2479 0xffffffff, 0xffffffff);
2480 mutex_unlock(&adev->grbm_idx_mutex)rw_exit_write(&adev->grbm_idx_mutex);
2481 DRM_INFO("Timeout wait for RLC serdes %u,%u\n",printk("\0016" "[" "drm" "] " "Timeout wait for RLC serdes %u,%u\n"
, i, j)
2482 i, j)printk("\0016" "[" "drm" "] " "Timeout wait for RLC serdes %u,%u\n"
, i, j);
2483 return;
2484 }
2485 }
2486 }
2487 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
2488 mutex_unlock(&adev->grbm_idx_mutex)rw_exit_write(&adev->grbm_idx_mutex);
2489
2490 mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK0x0000FFFFL |
2491 RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK0x00010000L |
2492 RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK0x00040000L |
2493 RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK0x00080000L;
2494 for (k = 0; k < adev->usec_timeout; k++) {
2495 if ((RREG32_SOC15(GC, 0, mmRLC_SERDES_NONCU_MASTER_BUSY)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c62, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c62), 0)) & mask) == 0)
2496 break;
2497 udelay(1);
2498 }
2499}
2500
2501static void gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
2502 bool_Bool enable)
2503{
2504 u32 tmp;
2505
2506 /* These interrupts should be enabled to drive DS clock */
2507
2508 tmp= RREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x106a, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x106a), 0));
2509
2510 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0)(((tmp) & ~0x00080000L) | (0x00080000L & ((enable ? 1
: 0) << 0x13)));
2511 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0)(((tmp) & ~0x00100000L) | (0x00100000L & ((enable ? 1
: 0) << 0x14)));
2512 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0)(((tmp) & ~0x00040000L) | (0x00040000L & ((enable ? 1
: 0) << 0x12)));
2513 if(adev->gfx.num_gfx_rings)
2514 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0)(((tmp) & ~0x00200000L) | (0x00200000L & ((enable ? 1
: 0) << 0x15)));
2515
2516 WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x106a), tmp, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x106a)), (tmp), 0));
2517}
2518
2519static void gfx_v9_0_init_csb(struct amdgpu_device *adev)
2520{
2521 adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
2522 /* csib */
2523 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_HI),((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4ca3), adev->gfx.rlc.clear_state_gpu_addr >> 32
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, ((adev->
reg_offset[GC_HWIP][0][1] + 0x4ca3)), (adev->gfx.rlc.clear_state_gpu_addr
>> 32), 0))
2524 adev->gfx.rlc.clear_state_gpu_addr >> 32)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4ca3), adev->gfx.rlc.clear_state_gpu_addr >> 32
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, ((adev->
reg_offset[GC_HWIP][0][1] + 0x4ca3)), (adev->gfx.rlc.clear_state_gpu_addr
>> 32), 0));
2525 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_LO),((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4ca2), adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, ((adev->
reg_offset[GC_HWIP][0][1] + 0x4ca2)), (adev->gfx.rlc.clear_state_gpu_addr
& 0xfffffffc), 0))
2526 adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4ca2), adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, ((adev->
reg_offset[GC_HWIP][0][1] + 0x4ca2)), (adev->gfx.rlc.clear_state_gpu_addr
& 0xfffffffc), 0));
2527 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_LENGTH),((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4ca4), adev->gfx.rlc.clear_state_size, (1<<2)
, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP
][0][1] + 0x4ca4)), (adev->gfx.rlc.clear_state_size), 0))
2528 adev->gfx.rlc.clear_state_size)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4ca4), adev->gfx.rlc.clear_state_size, (1<<2)
, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP
][0][1] + 0x4ca4)), (adev->gfx.rlc.clear_state_size), 0));
2529}
2530
2531static void gfx_v9_1_parse_ind_reg_list(int *register_list_format,
2532 int indirect_offset,
2533 int list_size,
2534 int *unique_indirect_regs,
2535 int unique_indirect_reg_count,
2536 int *indirect_start_offsets,
2537 int *indirect_start_offsets_count,
2538 int max_start_offsets_count)
2539{
2540 int idx;
2541
2542 for (; indirect_offset < list_size; indirect_offset++) {
2543 WARN_ON(*indirect_start_offsets_count >= max_start_offsets_count)({ int __ret = !!(*indirect_start_offsets_count >= max_start_offsets_count
); if (__ret) printf("WARNING %s failed at %s:%d\n", "*indirect_start_offsets_count >= max_start_offsets_count"
, "/usr/src/sys/dev/pci/drm/amd/amdgpu/gfx_v9_0.c", 2543); __builtin_expect
(!!(__ret), 0); });
2544 indirect_start_offsets[*indirect_start_offsets_count] = indirect_offset;
2545 *indirect_start_offsets_count = *indirect_start_offsets_count + 1;
2546
2547 while (register_list_format[indirect_offset] != 0xFFFFFFFF) {
2548 indirect_offset += 2;
2549
2550 /* look for the matching indice */
2551 for (idx = 0; idx < unique_indirect_reg_count; idx++) {
2552 if (unique_indirect_regs[idx] ==
2553 register_list_format[indirect_offset] ||
2554 !unique_indirect_regs[idx])
2555 break;
2556 }
2557
2558 BUG_ON(idx >= unique_indirect_reg_count)((!(idx >= unique_indirect_reg_count)) ? (void)0 : __assert
("diagnostic ", "/usr/src/sys/dev/pci/drm/amd/amdgpu/gfx_v9_0.c"
, 2558, "!(idx >= unique_indirect_reg_count)"));
2559
2560 if (!unique_indirect_regs[idx])
2561 unique_indirect_regs[idx] = register_list_format[indirect_offset];
2562
2563 indirect_offset++;
2564 }
2565 }
2566}
2567
2568static int gfx_v9_1_init_rlc_save_restore_list(struct amdgpu_device *adev)
2569{
2570 int unique_indirect_regs[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2571 int unique_indirect_reg_count = 0;
2572
2573 int indirect_start_offsets[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0};
2574 int indirect_start_offsets_count = 0;
2575
2576 int list_size = 0;
2577 int i = 0, j = 0;
2578 u32 tmp = 0;
2579
2580 u32 *register_list_format =
2581 kmemdup(adev->gfx.rlc.register_list_format,
2582 adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL(0x0001 | 0x0004));
2583 if (!register_list_format)
2584 return -ENOMEM12;
2585
2586 /* setup unique_indirect_regs array and indirect_start_offsets array */
2587 unique_indirect_reg_count = ARRAY_SIZE(unique_indirect_regs)(sizeof((unique_indirect_regs)) / sizeof((unique_indirect_regs
)[0]));
2588 gfx_v9_1_parse_ind_reg_list(register_list_format,
2589 adev->gfx.rlc.reg_list_format_direct_reg_list_length,
2590 adev->gfx.rlc.reg_list_format_size_bytes >> 2,
2591 unique_indirect_regs,
2592 unique_indirect_reg_count,
2593 indirect_start_offsets,
2594 &indirect_start_offsets_count,
2595 ARRAY_SIZE(indirect_start_offsets)(sizeof((indirect_start_offsets)) / sizeof((indirect_start_offsets
)[0])));
2596
2597 /* enable auto inc in case it is disabled */
2598 tmp = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c80)), 0);
2599 tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK0x00000002L;
2600 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL), tmp)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c80)), (tmp), 0);
2601
2602 /* write register_restore table to offset 0x0 using RLC_SRM_ARAM_ADDR/DATA */
2603 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_ADDR),amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c83)), (0x00000000L), 0)
2604 RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c83)), (0x00000000L), 0);
2605 for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
2606 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_DATA),amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c84)), (adev->gfx.rlc.register_restore[i]), 0)
2607 adev->gfx.rlc.register_restore[i])amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c84)), (adev->gfx.rlc.register_restore[i]), 0);
2608
2609 /* load indirect register */
2610 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c6c)), (adev->gfx.rlc.reg_list_format_start), 0)
2611 adev->gfx.rlc.reg_list_format_start)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c6c)), (adev->gfx.rlc.reg_list_format_start), 0);
2612
2613 /* direct register portion */
2614 for (i = 0; i < adev->gfx.rlc.reg_list_format_direct_reg_list_length; i++)
2615 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c6d)), (register_list_format[i]), 0)
2616 register_list_format[i])amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c6d)), (register_list_format[i]), 0);
2617
2618 /* indirect register portion */
2619 while (i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2)) {
2620 if (register_list_format[i] == 0xFFFFFFFF) {
2621 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++])((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c6d), register_list_format[i++], 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x4c6d)), (register_list_format
[i++]), 0));
2622 continue;
2623 }
2624
2625 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++])((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c6d), register_list_format[i++], 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x4c6d)), (register_list_format
[i++]), 0));
2626 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++])((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c6d), register_list_format[i++], 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x4c6d)), (register_list_format
[i++]), 0));
2627
2628 for (j = 0; j < unique_indirect_reg_count; j++) {
2629 if (register_list_format[i] == unique_indirect_regs[j]) {
2630 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, j)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c6d), j, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][1] + 0x4c6d)), (j), 0));
2631 break;
2632 }
2633 }
2634
2635 BUG_ON(j >= unique_indirect_reg_count)((!(j >= unique_indirect_reg_count)) ? (void)0 : __assert(
"diagnostic ", "/usr/src/sys/dev/pci/drm/amd/amdgpu/gfx_v9_0.c"
, 2635, "!(j >= unique_indirect_reg_count)"));
2636
2637 i++;
2638 }
2639
2640 /* set save/restore list size */
2641 list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
2642 list_size = list_size >> 1;
2643 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c6c)), (adev->gfx.rlc.reg_restore_list_size), 0)
2644 adev->gfx.rlc.reg_restore_list_size)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c6c)), (adev->gfx.rlc.reg_restore_list_size), 0);
2645 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), list_size)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c6d)), (list_size), 0);
2646
2647 /* write the starting offsets to RLC scratch ram */
2648 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR),amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c6c)), (adev->gfx.rlc.starting_offsets_start), 0)
2649 adev->gfx.rlc.starting_offsets_start)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c6c)), (adev->gfx.rlc.starting_offsets_start), 0);
2650 for (i = 0; i < ARRAY_SIZE(indirect_start_offsets)(sizeof((indirect_start_offsets)) / sizeof((indirect_start_offsets
)[0])); i++)
2651 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA),amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c6d)), (indirect_start_offsets[i]), 0)
2652 indirect_start_offsets[i])amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c6d)), (indirect_start_offsets[i]), 0);
2653
2654 /* load unique indirect regs*/
2655 for (i = 0; i < ARRAY_SIZE(unique_indirect_regs)(sizeof((unique_indirect_regs)) / sizeof((unique_indirect_regs
)[0])); i++) {
2656 if (unique_indirect_regs[i] != 0) {
2657 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_ADDR_0)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c8b) + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i]), (unique_indirect_regs
[i] & 0x3FFFF), 0)
2658 + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i],amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c8b) + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i]), (unique_indirect_regs
[i] & 0x3FFFF), 0)
2659 unique_indirect_regs[i] & 0x3FFFF)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c8b) + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i]), (unique_indirect_regs
[i] & 0x3FFFF), 0);
2660
2661 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_DATA_0)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c93) + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i]), (unique_indirect_regs
[i] >> 20), 0)
2662 + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i],amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c93) + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i]), (unique_indirect_regs
[i] >> 20), 0)
2663 unique_indirect_regs[i] >> 20)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c93) + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i]), (unique_indirect_regs
[i] >> 20), 0);
2664 }
2665 }
2666
2667 kfree(register_list_format);
2668 return 0;
2669}
2670
2671static void gfx_v9_0_enable_save_restore_machine(struct amdgpu_device *adev)
2672{
2673 WREG32_FIELD15(GC, 0, RLC_SRM_CNTL, SRM_ENABLE, 1)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c80, (((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c80, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c80), 0)) & ~0x00000001L) |
(1) << 0x0, 0, GC_HWIP) : amdgpu_device_wreg(adev, (adev
->reg_offset[GC_HWIP][0][1] + 0x4c80), ((((((adev)->virt
.caps & (1 << 2)) && adev->gfx.rlc.funcs
&& adev->gfx.rlc.rlcg_reg_access_supported) ? amdgpu_sriov_rreg
(adev, adev->reg_offset[GC_HWIP][0][1] + 0x4c80, 0, GC_HWIP
) : amdgpu_device_rreg(adev, (adev->reg_offset[GC_HWIP][0]
[1] + 0x4c80), 0)) & ~0x00000001L) | (1) << 0x0), 0
));
2674}
2675
2676static void pwr_10_0_gfxip_control_over_cgpg(struct amdgpu_device *adev,
2677 bool_Bool enable)
2678{
2679 uint32_t data = 0;
2680 uint32_t default_data = 0;
2681
2682 default_data = data = RREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS))amdgpu_device_rreg(adev, ((adev->reg_offset[PWR_HWIP][0][0
] + 0x0183)), 0);
2683 if (enable) {
2684 /* enable GFXIP control over CGPG */
2685 data |= PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK0x00000001L;
2686 if(default_data != data)
2687 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data)amdgpu_device_wreg(adev, ((adev->reg_offset[PWR_HWIP][0][0
] + 0x0183)), (data), 0);
2688
2689 /* update status */
2690 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK0x00000006L;
2691 data |= (2 << PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT0x1);
2692 if(default_data != data)
2693 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data)amdgpu_device_wreg(adev, ((adev->reg_offset[PWR_HWIP][0][0
] + 0x0183)), (data), 0);
2694 } else {
2695 /* restore GFXIP control over GCPG */
2696 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK0x00000001L;
2697 if(default_data != data)
2698 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data)amdgpu_device_wreg(adev, ((adev->reg_offset[PWR_HWIP][0][0
] + 0x0183)), (data), 0);
2699 }
2700}
2701
2702static void gfx_v9_0_init_gfx_power_gating(struct amdgpu_device *adev)
2703{
2704 uint32_t data = 0;
2705
2706 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG(1 << 0) |
2707 AMD_PG_SUPPORT_GFX_SMG(1 << 1) |
2708 AMD_PG_SUPPORT_GFX_DMG(1 << 2))) {
2709 /* init IDLE_POLL_COUNT = 60 */
2710 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][0]
+ 0x01c2)), 0);
2711 data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK0xFFFF0000L;
2712 data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT0x10);
2713 WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][0]
+ 0x01c2)), (data), 0);
2714
2715 /* init RLC PG Delay */
2716 data = 0;
2717 data |= (0x10 << RLC_PG_DELAY__POWER_UP_DELAY__SHIFT0x0);
2718 data |= (0x10 << RLC_PG_DELAY__POWER_DOWN_DELAY__SHIFT0x8);
2719 data |= (0x10 << RLC_PG_DELAY__CMD_PROPAGATE_DELAY__SHIFT0x10);
2720 data |= (0x40 << RLC_PG_DELAY__MEM_SLEEP_DELAY__SHIFT0x18);
2721 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c4d)), (data), 0);
2722
2723 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c1f)), 0);
2724 data &= ~RLC_PG_DELAY_2__SERDES_CMD_DELAY_MASK0x0000FF00L;
2725 data |= (0x4 << RLC_PG_DELAY_2__SERDES_CMD_DELAY__SHIFT0x8);
2726 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c1f)), (data), 0);
2727
2728 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c78)), 0);
2729 data &= ~RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG_MASK0x000000FFL;
2730 data |= (0xff << RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG__SHIFT0x0);
2731 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c78)), (data), 0);
2732
2733 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c55)), 0);
2734 data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK0x0007FFF8L;
2735
2736 /* program GRBM_REG_SAVE_GFX_IDLE_THRESHOLD to 0x55f0 */
2737 data |= (0x55f0 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT0x3);
2738 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c55)), (data), 0);
2739 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 3, 0)(((9) << 16) | ((3) << 8) | (0)))
2740 pwr_10_0_gfxip_control_over_cgpg(adev, true1);
2741 }
2742}
2743
2744static void gfx_v9_0_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
2745 bool_Bool enable)
2746{
2747 uint32_t data = 0;
2748 uint32_t default_data = 0;
2749
2750 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), 0);
2751 data = REG_SET_FIELD(data, RLC_PG_CNTL,(((data) & ~0x00020000L) | (0x00020000L & ((enable ? 1
: 0) << 0x11)))
2752 SMU_CLK_SLOWDOWN_ON_PU_ENABLE,(((data) & ~0x00020000L) | (0x00020000L & ((enable ? 1
: 0) << 0x11)))
2753 enable ? 1 : 0)(((data) & ~0x00020000L) | (0x00020000L & ((enable ? 1
: 0) << 0x11)));
2754 if (default_data != data)
2755 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), (data), 0);
2756}
2757
2758static void gfx_v9_0_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
2759 bool_Bool enable)
2760{
2761 uint32_t data = 0;
2762 uint32_t default_data = 0;
2763
2764 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), 0);
2765 data = REG_SET_FIELD(data, RLC_PG_CNTL,(((data) & ~0x00040000L) | (0x00040000L & ((enable ? 1
: 0) << 0x12)))
2766 SMU_CLK_SLOWDOWN_ON_PD_ENABLE,(((data) & ~0x00040000L) | (0x00040000L & ((enable ? 1
: 0) << 0x12)))
2767 enable ? 1 : 0)(((data) & ~0x00040000L) | (0x00040000L & ((enable ? 1
: 0) << 0x12)));
2768 if(default_data != data)
2769 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), (data), 0);
2770}
2771
2772static void gfx_v9_0_enable_cp_power_gating(struct amdgpu_device *adev,
2773 bool_Bool enable)
2774{
2775 uint32_t data = 0;
2776 uint32_t default_data = 0;
2777
2778 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), 0);
2779 data = REG_SET_FIELD(data, RLC_PG_CNTL,(((data) & ~0x00008000L) | (0x00008000L & ((enable ? 0
: 1) << 0xf)))
2780 CP_PG_DISABLE,(((data) & ~0x00008000L) | (0x00008000L & ((enable ? 0
: 1) << 0xf)))
2781 enable ? 0 : 1)(((data) & ~0x00008000L) | (0x00008000L & ((enable ? 0
: 1) << 0xf)));
2782 if(default_data != data)
2783 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), (data), 0);
2784}
2785
2786static void gfx_v9_0_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
2787 bool_Bool enable)
2788{
2789 uint32_t data, default_data;
2790
2791 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), 0);
2792 data = REG_SET_FIELD(data, RLC_PG_CNTL,(((data) & ~0x00000001L) | (0x00000001L & ((enable ? 1
: 0) << 0x0)))
2793 GFX_POWER_GATING_ENABLE,(((data) & ~0x00000001L) | (0x00000001L & ((enable ? 1
: 0) << 0x0)))
2794 enable ? 1 : 0)(((data) & ~0x00000001L) | (0x00000001L & ((enable ? 1
: 0) << 0x0)));
2795 if(default_data != data)
2796 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), (data), 0);
2797}
2798
2799static void gfx_v9_0_enable_gfx_pipeline_powergating(struct amdgpu_device *adev,
2800 bool_Bool enable)
2801{
2802 uint32_t data, default_data;
2803
2804 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), 0);
2805 data = REG_SET_FIELD(data, RLC_PG_CNTL,(((data) & ~0x00000010L) | (0x00000010L & ((enable ? 1
: 0) << 0x4)))
2806 GFX_PIPELINE_PG_ENABLE,(((data) & ~0x00000010L) | (0x00000010L & ((enable ? 1
: 0) << 0x4)))
2807 enable ? 1 : 0)(((data) & ~0x00000010L) | (0x00000010L & ((enable ? 1
: 0) << 0x4)));
2808 if(default_data != data)
2809 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), (data), 0);
2810
2811 if (!enable)
2812 /* read any GFX register to wake up GFX */
2813 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmDB_RENDER_CONTROL))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x0000)), 0);
2814}
2815
2816static void gfx_v9_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
2817 bool_Bool enable)
2818{
2819 uint32_t data, default_data;
2820
2821 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), 0);
2822 data = REG_SET_FIELD(data, RLC_PG_CNTL,(((data) & ~0x00000008L) | (0x00000008L & ((enable ? 1
: 0) << 0x3)))
2823 STATIC_PER_CU_PG_ENABLE,(((data) & ~0x00000008L) | (0x00000008L & ((enable ? 1
: 0) << 0x3)))
2824 enable ? 1 : 0)(((data) & ~0x00000008L) | (0x00000008L & ((enable ? 1
: 0) << 0x3)));
2825 if(default_data != data)
2826 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), (data), 0);
2827}
2828
2829static void gfx_v9_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
2830 bool_Bool enable)
2831{
2832 uint32_t data, default_data;
2833
2834 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL))amdgpu_device_rreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), 0);
2835 data = REG_SET_FIELD(data, RLC_PG_CNTL,(((data) & ~0x00000004L) | (0x00000004L & ((enable ? 1
: 0) << 0x2)))
2836 DYN_PER_CU_PG_ENABLE,(((data) & ~0x00000004L) | (0x00000004L & ((enable ? 1
: 0) << 0x2)))
2837 enable ? 1 : 0)(((data) & ~0x00000004L) | (0x00000004L & ((enable ? 1
: 0) << 0x2)));
2838 if(default_data != data)
2839 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data)amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0][1]
+ 0x4c43)), (data), 0);
2840}
2841
2842static void gfx_v9_0_init_pg(struct amdgpu_device *adev)
2843{
2844 gfx_v9_0_init_csb(adev);
2845
2846 /*
2847 * Rlc save restore list is workable since v2_1.
2848 * And it's needed by gfxoff feature.
2849 */
2850 if (adev->gfx.rlc.is_rlc_v2_1) {
2851 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 2, 1)(((9) << 16) | ((2) << 8) | (1)) ||
2852 (adev->apu_flags & AMD_APU_IS_RAVEN2))
2853 gfx_v9_1_init_rlc_save_restore_list(adev);
2854 gfx_v9_0_enable_save_restore_machine(adev);
2855 }
2856
2857 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG(1 << 0) |
2858 AMD_PG_SUPPORT_GFX_SMG(1 << 1) |
2859 AMD_PG_SUPPORT_GFX_DMG(1 << 2) |
2860 AMD_PG_SUPPORT_CP(1 << 5) |
2861 AMD_PG_SUPPORT_GDS(1 << 6) |
2862 AMD_PG_SUPPORT_RLC_SMU_HS(1 << 7))) {
2863 WREG32_SOC15(GC, 0, mmRLC_JUMP_TABLE_RESTORE,((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c1e), adev->gfx.rlc.cp_table_gpu_addr >> 8, 0
, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP
][0][1] + 0x4c1e)), (adev->gfx.rlc.cp_table_gpu_addr >>
8), 0))
2864 adev->gfx.rlc.cp_table_gpu_addr >> 8)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c1e), adev->gfx.rlc.cp_table_gpu_addr >> 8, 0
, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP
][0][1] + 0x4c1e)), (adev->gfx.rlc.cp_table_gpu_addr >>
8), 0));
2865 gfx_v9_0_init_gfx_power_gating(adev);
2866 }
2867}
2868
2869static void gfx_v9_0_rlc_stop(struct amdgpu_device *adev)
2870{
2871 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c00, (((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c00, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c00), 0)) & ~0x00000001L) |
(0) << 0x0, 0, GC_HWIP) : amdgpu_device_wreg(adev, (adev
->reg_offset[GC_HWIP][0][1] + 0x4c00), ((((((adev)->virt
.caps & (1 << 2)) && adev->gfx.rlc.funcs
&& adev->gfx.rlc.rlcg_reg_access_supported) ? amdgpu_sriov_rreg
(adev, adev->reg_offset[GC_HWIP][0][1] + 0x4c00, 0, GC_HWIP
) : amdgpu_device_rreg(adev, (adev->reg_offset[GC_HWIP][0]
[1] + 0x4c00), 0)) & ~0x00000001L) | (0) << 0x0), 0
));
2872 gfx_v9_0_enable_gui_idle_interrupt(adev, false0);
2873 gfx_v9_0_wait_for_rlc_serdes(adev);
2874}
2875
2876static void gfx_v9_0_rlc_reset(struct amdgpu_device *adev)
2877{
2878 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0008, (((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0008, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0008), 0)) & ~0x00000004L) |
(1) << 0x2, 0, GC_HWIP) : amdgpu_device_wreg(adev, (adev
->reg_offset[GC_HWIP][0][0] + 0x0008), ((((((adev)->virt
.caps & (1 << 2)) && adev->gfx.rlc.funcs
&& adev->gfx.rlc.rlcg_reg_access_supported) ? amdgpu_sriov_rreg
(adev, adev->reg_offset[GC_HWIP][0][0] + 0x0008, 0, GC_HWIP
) : amdgpu_device_rreg(adev, (adev->reg_offset[GC_HWIP][0]
[0] + 0x0008), 0)) & ~0x00000004L) | (1) << 0x2), 0
));
2879 udelay(50);
2880 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0008, (((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0008, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0008), 0)) & ~0x00000004L) |
(0) << 0x2, 0, GC_HWIP) : amdgpu_device_wreg(adev, (adev
->reg_offset[GC_HWIP][0][0] + 0x0008), ((((((adev)->virt
.caps & (1 << 2)) && adev->gfx.rlc.funcs
&& adev->gfx.rlc.rlcg_reg_access_supported) ? amdgpu_sriov_rreg
(adev, adev->reg_offset[GC_HWIP][0][0] + 0x0008, 0, GC_HWIP
) : amdgpu_device_rreg(adev, (adev->reg_offset[GC_HWIP][0]
[0] + 0x0008), 0)) & ~0x00000004L) | (0) << 0x2), 0
));
2881 udelay(50);
2882}
2883
2884static void gfx_v9_0_rlc_start(struct amdgpu_device *adev)
2885{
2886#ifdef AMDGPU_RLC_DEBUG_RETRY
2887 u32 rlc_ucode_ver;
2888#endif
2889
2890 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c00, (((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c00, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c00), 0)) & ~0x00000001L) |
(1) << 0x0, 0, GC_HWIP) : amdgpu_device_wreg(adev, (adev
->reg_offset[GC_HWIP][0][1] + 0x4c00), ((((((adev)->virt
.caps & (1 << 2)) && adev->gfx.rlc.funcs
&& adev->gfx.rlc.rlcg_reg_access_supported) ? amdgpu_sriov_rreg
(adev, adev->reg_offset[GC_HWIP][0][1] + 0x4c00, 0, GC_HWIP
) : amdgpu_device_rreg(adev, (adev->reg_offset[GC_HWIP][0]
[1] + 0x4c00), 0)) & ~0x00000001L) | (1) << 0x0), 0
));
2891 udelay(50);
2892
2893 /* carrizo do enable cp interrupt after cp inited */
2894 if (!(adev->flags & AMD_IS_APU)) {
2895 gfx_v9_0_enable_gui_idle_interrupt(adev, true1);
2896 udelay(50);
2897 }
2898
2899#ifdef AMDGPU_RLC_DEBUG_RETRY
2900 /* RLC_GPM_GENERAL_6 : RLC Ucode version */
2901 rlc_ucode_ver = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_6)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c69, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c69), 0));
2902 if(rlc_ucode_ver == 0x108) {
2903 DRM_INFO("Using rlc debug ucode. mmRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n",printk("\0016" "[" "drm" "] " "Using rlc debug ucode. mmRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n"
, rlc_ucode_ver, adev->gfx.rlc_fw_version)
2904 rlc_ucode_ver, adev->gfx.rlc_fw_version)printk("\0016" "[" "drm" "] " "Using rlc debug ucode. mmRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n"
, rlc_ucode_ver, adev->gfx.rlc_fw_version);
2905 /* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles,
2906 * default is 0x9C4 to create a 100us interval */
2907 WREG32_SOC15(GC, 0, mmRLC_GPM_TIMER_INT_3, 0x9C4)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c15), 0x9C4, 0, GC_HWIP) : amdgpu_device_wreg(adev, (
(adev->reg_offset[GC_HWIP][0][1] + 0x4c15)), (0x9C4), 0));
2908 /* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr
2909 * to disable the page fault retry interrupts, default is
2910 * 0x100 (256) */
2911 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_12, 0x100)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4cb1), 0x100, 0, GC_HWIP) : amdgpu_device_wreg(adev, (
(adev->reg_offset[GC_HWIP][0][1] + 0x4cb1)), (0x100), 0));
2912 }
2913#endif
2914}
2915
2916static int gfx_v9_0_rlc_load_microcode(struct amdgpu_device *adev)
2917{
2918 const struct rlc_firmware_header_v2_0 *hdr;
2919 const __le32 *fw_data;
2920 unsigned i, fw_size;
2921
2922 if (!adev->gfx.rlc_fw)
2923 return -EINVAL22;
2924
2925 hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
2926 amdgpu_ucode_print_rlc_hdr(&hdr->header);
2927
2928 fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
2929 le32_to_cpu(hdr->header.ucode_array_offset_bytes)((__uint32_t)(hdr->header.ucode_array_offset_bytes)));
2930 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes)((__uint32_t)(hdr->header.ucode_size_bytes)) / 4;
2931
2932 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR,((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x583c), 0x00002000L, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x583c)), (0x00002000L
), 0))
2933 RLCG_UCODE_LOADING_START_ADDRESS)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x583c), 0x00002000L, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][1] + 0x583c)), (0x00002000L
), 0));
2934 for (i = 0; i < fw_size; i++)
2935 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x583d), ((__uint32_t)(*(__uint32_t *)(fw_data++))), 0, GC_HWIP
) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0
][1] + 0x583d)), (((__uint32_t)(*(__uint32_t *)(fw_data++))))
, 0));
2936 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x583c), adev->gfx.rlc_fw_version, 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x583c)), (adev
->gfx.rlc_fw_version), 0));
2937
2938 return 0;
2939}
2940
2941static int gfx_v9_0_rlc_resume(struct amdgpu_device *adev)
2942{
2943 int r;
2944
2945 if (amdgpu_sriov_vf(adev)((adev)->virt.caps & (1 << 2))) {
2946 gfx_v9_0_init_csb(adev);
2947 return 0;
2948 }
2949
2950 adev->gfx.rlc.funcs->stop(adev);
2951
2952 /* disable CG */
2953 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c49), 0, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][1] + 0x4c49)), (0), 0));
2954
2955 gfx_v9_0_init_pg(adev);
2956
2957 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
2958 /* legacy rlc firmware loading */
2959 r = gfx_v9_0_rlc_load_microcode(adev);
2960 if (r)
2961 return r;
2962 }
2963
2964 switch (adev->ip_versions[GC_HWIP][0]) {
2965 case IP_VERSION(9, 2, 2)(((9) << 16) | ((2) << 8) | (2)):
2966 case IP_VERSION(9, 1, 0)(((9) << 16) | ((1) << 8) | (0)):
2967 if (amdgpu_lbpw == 0)
2968 gfx_v9_0_enable_lbpw(adev, false0);
2969 else
2970 gfx_v9_0_enable_lbpw(adev, true1);
2971 break;
2972 case IP_VERSION(9, 4, 0)(((9) << 16) | ((4) << 8) | (0)):
2973 if (amdgpu_lbpw > 0)
2974 gfx_v9_0_enable_lbpw(adev, true1);
2975 else
2976 gfx_v9_0_enable_lbpw(adev, false0);
2977 break;
2978 default:
2979 break;
2980 }
2981
2982 adev->gfx.rlc.funcs->start(adev);
2983
2984 return 0;
2985}
2986
2987static void gfx_v9_0_cp_gfx_enable(struct amdgpu_device *adev, bool_Bool enable)
2988{
2989 u32 tmp = RREG32_SOC15(GC, 0, mmCP_ME_CNTL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x01b6, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x01b6), 0));
2990
2991 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1)(((tmp) & ~0x10000000L) | (0x10000000L & ((enable ? 0
: 1) << 0x1c)));
2992 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1)(((tmp) & ~0x04000000L) | (0x04000000L & ((enable ? 0
: 1) << 0x1a)));
2993 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, enable ? 0 : 1)(((tmp) & ~0x01000000L) | (0x01000000L & ((enable ? 0
: 1) << 0x18)));
2994 WREG32_SOC15_RLC(GC, 0, mmCP_ME_CNTL, tmp)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x01b6; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, tmp, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (tmp), 0)); } while
(0);
2995 udelay(50);
2996}
2997
2998static int gfx_v9_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2999{
3000 const struct gfx_firmware_header_v1_0 *pfp_hdr;
3001 const struct gfx_firmware_header_v1_0 *ce_hdr;
3002 const struct gfx_firmware_header_v1_0 *me_hdr;
3003 const __le32 *fw_data;
3004 unsigned i, fw_size;
3005
3006 if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
3007 return -EINVAL22;
3008
3009 pfp_hdr = (const struct gfx_firmware_header_v1_0 *)
3010 adev->gfx.pfp_fw->data;
3011 ce_hdr = (const struct gfx_firmware_header_v1_0 *)
3012 adev->gfx.ce_fw->data;
3013 me_hdr = (const struct gfx_firmware_header_v1_0 *)
3014 adev->gfx.me_fw->data;
3015
3016 amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
3017 amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
3018 amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
3019
3020 gfx_v9_0_cp_gfx_enable(adev, false0);
3021
3022 /* PFP */
3023 fw_data = (const __le32 *)
3024 (adev->gfx.pfp_fw->data +
3025 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes)((__uint32_t)(pfp_hdr->header.ucode_array_offset_bytes)));
3026 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes)((__uint32_t)(pfp_hdr->header.ucode_size_bytes)) / 4;
3027 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x5814), 0, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][1] + 0x5814)), (0), 0));
3028 for (i = 0; i < fw_size; i++)
3029 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x5815), ((__uint32_t)(*(__uint32_t *)(fw_data++))), 0, GC_HWIP
) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0
][1] + 0x5815)), (((__uint32_t)(*(__uint32_t *)(fw_data++))))
, 0));
3030 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x5814), adev->gfx.pfp_fw_version, 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x5814)), (adev
->gfx.pfp_fw_version), 0));
3031
3032 /* CE */
3033 fw_data = (const __le32 *)
3034 (adev->gfx.ce_fw->data +
3035 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes)((__uint32_t)(ce_hdr->header.ucode_array_offset_bytes)));
3036 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes)((__uint32_t)(ce_hdr->header.ucode_size_bytes)) / 4;
3037 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x5818), 0, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][1] + 0x5818)), (0), 0));
3038 for (i = 0; i < fw_size; i++)
3039 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x5819), ((__uint32_t)(*(__uint32_t *)(fw_data++))), 0, GC_HWIP
) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0
][1] + 0x5819)), (((__uint32_t)(*(__uint32_t *)(fw_data++))))
, 0));
3040 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x5818), adev->gfx.ce_fw_version, 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x5818)), (adev
->gfx.ce_fw_version), 0));
3041
3042 /* ME */
3043 fw_data = (const __le32 *)
3044 (adev->gfx.me_fw->data +
3045 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes)((__uint32_t)(me_hdr->header.ucode_array_offset_bytes)));
3046 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes)((__uint32_t)(me_hdr->header.ucode_size_bytes)) / 4;
3047 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x5816), 0, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][1] + 0x5816)), (0), 0));
3048 for (i = 0; i < fw_size; i++)
3049 WREG32_SOC15(GC, 0, mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x5817), ((__uint32_t)(*(__uint32_t *)(fw_data++))), 0, GC_HWIP
) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0
][1] + 0x5817)), (((__uint32_t)(*(__uint32_t *)(fw_data++))))
, 0));
3050 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x5816), adev->gfx.me_fw_version, 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x5816)), (adev
->gfx.me_fw_version), 0));
3051
3052 return 0;
3053}
3054
3055static int gfx_v9_0_cp_gfx_start(struct amdgpu_device *adev)
3056{
3057 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
3058 const struct cs_section_def *sect = NULL((void *)0);
3059 const struct cs_extent_def *ext = NULL((void *)0);
3060 int r, i, tmp;
3061
3062 /* init the CP */
3063 WREG32_SOC15(GC, 0, mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x10ae), adev->gfx.config.max_hw_contexts - 1, 0, GC_HWIP
) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP][0
][0] + 0x10ae)), (adev->gfx.config.max_hw_contexts - 1), 0
));
3064 WREG32_SOC15(GC, 0, mmCP_DEVICE_ID, 1)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x104b), 1, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x104b)), (1), 0));
3065
3066 gfx_v9_0_cp_gfx_enable(adev, true1);
3067
3068 r = amdgpu_ring_alloc(ring, gfx_v9_0_get_csb_size(adev) + 4 + 3);
3069 if (r) {
3070 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r)__drm_err("amdgpu: cp failed to lock ring (%d).\n", r);
3071 return r;
3072 }
3073
3074 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)((3 << 30) | (((0x4A) & 0xFF) << 8) | ((0) &
0x3FFF) << 16));
3075 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE(2 << 28));
3076
3077 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)((3 << 30) | (((0x28) & 0xFF) << 8) | ((1) &
0x3FFF) << 16));
3078 amdgpu_ring_write(ring, 0x80000000);
3079 amdgpu_ring_write(ring, 0x80000000);
3080
3081 for (sect = gfx9_cs_data; sect->section != NULL((void *)0); ++sect) {
3082 for (ext = sect->section; ext->extent != NULL((void *)0); ++ext) {
3083 if (sect->id == SECT_CONTEXT) {
3084 amdgpu_ring_write(ring,
3085 PACKET3(PACKET3_SET_CONTEXT_REG,((3 << 30) | (((0x69) & 0xFF) << 8) | ((ext->
reg_count) & 0x3FFF) << 16)
3086 ext->reg_count)((3 << 30) | (((0x69) & 0xFF) << 8) | ((ext->
reg_count) & 0x3FFF) << 16));
3087 amdgpu_ring_write(ring,
3088 ext->reg_index - PACKET3_SET_CONTEXT_REG_START0x0000a000);
3089 for (i = 0; i < ext->reg_count; i++)
3090 amdgpu_ring_write(ring, ext->extent[i]);
3091 }
3092 }
3093 }
3094
3095 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)((3 << 30) | (((0x4A) & 0xFF) << 8) | ((0) &
0x3FFF) << 16));
3096 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE(3 << 28));
3097
3098 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0)((3 << 30) | (((0x12) & 0xFF) << 8) | ((0) &
0x3FFF) << 16));
3099 amdgpu_ring_write(ring, 0);
3100
3101 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2)((3 << 30) | (((0x11) & 0xFF) << 8) | ((2) &
0x3FFF) << 16));
3102 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE)((3) << 0));
3103 amdgpu_ring_write(ring, 0x8000);
3104 amdgpu_ring_write(ring, 0x8000);
3105
3106 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG,1)((3 << 30) | (((0x79) & 0xFF) << 8) | ((1) &
0x3FFF) << 16));
3107 tmp = (PACKET3_SET_UCONFIG_REG_INDEX_TYPE(2 << 28) |
3108 (SOC15_REG_OFFSET(GC, 0, mmVGT_INDEX_TYPE)(adev->reg_offset[GC_HWIP][0][1] + 0x2243) - PACKET3_SET_UCONFIG_REG_START0x0000c000));
3109 amdgpu_ring_write(ring, tmp);
3110 amdgpu_ring_write(ring, 0);
3111
3112 amdgpu_ring_commit(ring);
3113
3114 return 0;
3115}
3116
3117static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev)
3118{
3119 struct amdgpu_ring *ring;
3120 u32 tmp;
3121 u32 rb_bufsz;
3122 u64 rb_addr, rptr_addr, wptr_gpu_addr;
3123
3124 /* Set the write pointer delay */
3125 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_DELAY, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x01c1), 0, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x01c1)), (0), 0));
3126
3127 /* set the RB to use vmid 0 */
3128 WREG32_SOC15(GC, 0, mmCP_RB_VMID, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1051), 0, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x1051)), (0), 0));
3129
3130 /* Set ring buffer size */
3131 ring = &adev->gfx.gfx_ring[0];
3132 rb_bufsz = order_base_2(ring->ring_size / 8)drm_order(ring->ring_size / 8);
3133 tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz)(((0) & ~0x0000003FL) | (0x0000003FL & ((rb_bufsz) <<
0x0)));
3134 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2)(((tmp) & ~0x00003F00L) | (0x00003F00L & ((rb_bufsz -
2) << 0x8)));
3135#ifdef __BIG_ENDIAN
3136 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1)(((tmp) & ~0x00060000L) | (0x00060000L & ((1) <<
0x11)));
3137#endif
3138 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1041), tmp, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x1041)), (tmp), 0));
3139
3140 /* Initialize the ring buffer's write pointers */
3141 ring->wptr = 0;
3142 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1054), ((u32)(ring->wptr)), 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][0] + 0x1054)), (((u32
)(ring->wptr))), 0));
3143 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1055), ((u32)(((ring->wptr) >> 16) >> 16
)), 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset
[GC_HWIP][0][0] + 0x1055)), (((u32)(((ring->wptr) >>
16) >> 16))), 0));
3144
3145 /* set the wb address wether it's enabled or not */
3146 rptr_addr = ring->rptr_gpu_addr;
3147 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1043), ((u32)(rptr_addr)), 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][0] + 0x1043)), (((u32
)(rptr_addr))), 0));
3148 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1044), ((u32)(((rptr_addr) >> 16) >> 16)) &
0x0000FFFFL, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->
reg_offset[GC_HWIP][0][0] + 0x1044)), (((u32)(((rptr_addr) >>
16) >> 16)) & 0x0000FFFFL), 0));
3149
3150 wptr_gpu_addr = ring->wptr_gpu_addr;
3151 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1046), ((u32)(wptr_gpu_addr)), 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][0] + 0x1046)), (((u32
)(wptr_gpu_addr))), 0));
3152 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1047), ((u32)(((wptr_gpu_addr) >> 16) >> 16
)), 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset
[GC_HWIP][0][0] + 0x1047)), (((u32)(((wptr_gpu_addr) >>
16) >> 16))), 0));
3153
3154 mdelay(1);
3155 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1041), tmp, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x1041)), (tmp), 0));
3156
3157 rb_addr = ring->gpu_addr >> 8;
3158 WREG32_SOC15(GC, 0, mmCP_RB0_BASE, rb_addr)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1040), rb_addr, 0, GC_HWIP) : amdgpu_device_wreg(adev,
((adev->reg_offset[GC_HWIP][0][0] + 0x1040)), (rb_addr), 0
));
3159 WREG32_SOC15(GC, 0, mmCP_RB0_BASE_HI, upper_32_bits(rb_addr))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x10b1), ((u32)(((rb_addr) >> 16) >> 16)), 0
, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP
][0][0] + 0x10b1)), (((u32)(((rb_addr) >> 16) >> 16
))), 0));
3160
3161 tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1059, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1059), 0));
3162 if (ring->use_doorbell) {
3163 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,(((tmp) & ~0x0FFFFFFCL) | (0x0FFFFFFCL & ((ring->doorbell_index
) << 0x2)))
3164 DOORBELL_OFFSET, ring->doorbell_index)(((tmp) & ~0x0FFFFFFCL) | (0x0FFFFFFCL & ((ring->doorbell_index
) << 0x2)));
3165 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,(((tmp) & ~0x40000000L) | (0x40000000L & ((1) <<
0x1e)))
3166 DOORBELL_EN, 1)(((tmp) & ~0x40000000L) | (0x40000000L & ((1) <<
0x1e)));
3167 } else {
3168 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0)(((tmp) & ~0x40000000L) | (0x40000000L & ((0) <<
0x1e)));
3169 }
3170 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, tmp)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1059), tmp, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x1059)), (tmp), 0));
3171
3172 tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,(((0) & ~0x0FFFFFFCL) | (0x0FFFFFFCL & ((ring->doorbell_index
) << 0x2)))
3173 DOORBELL_RANGE_LOWER, ring->doorbell_index)(((0) & ~0x0FFFFFFCL) | (0x0FFFFFFCL & ((ring->doorbell_index
) << 0x2)));
3174 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x105a), tmp, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x105a)), (tmp), 0));
3175
3176 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER,((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x105b), 0x0FFFFFFCL, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][0] + 0x105b)), (0x0FFFFFFCL
), 0))
3177 CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x105b), 0x0FFFFFFCL, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][0] + 0x105b)), (0x0FFFFFFCL
), 0));
3178
3179
3180 /* start the ring */
3181 gfx_v9_0_cp_gfx_start(adev);
3182 ring->sched.ready = true1;
3183
3184 return 0;
3185}
3186
3187static void gfx_v9_0_cp_compute_enable(struct amdgpu_device *adev, bool_Bool enable)
3188{
3189 if (enable) {
3190 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL, 0)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x008d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0);
3191 } else {
3192 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x008d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (0x40000000L | 0x10000000L
), (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), ((0x40000000L | 0x10000000L)), 0)); } while (0)
3193 (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK))do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x008d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, (0x40000000L | 0x10000000L
), (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), ((0x40000000L | 0x10000000L)), 0)); } while (0);
3194 adev->gfx.kiq.ring.sched.ready = false0;
3195 }
3196 udelay(50);
3197}
3198
3199static int gfx_v9_0_cp_compute_load_microcode(struct amdgpu_device *adev)
3200{
3201 const struct gfx_firmware_header_v1_0 *mec_hdr;
3202 const __le32 *fw_data;
3203 unsigned i;
3204 u32 tmp;
3205
3206 if (!adev->gfx.mec_fw)
3207 return -EINVAL22;
3208
3209 gfx_v9_0_cp_compute_enable(adev, false0);
3210
3211 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3212 amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3213
3214 fw_data = (const __le32 *)
3215 (adev->gfx.mec_fw->data +
3216 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes)((__uint32_t)(mec_hdr->header.ucode_array_offset_bytes)));
3217 tmp = 0;
3218 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0)(((tmp) & ~0x0000000FL) | (0x0000000FL & ((0) <<
0x0)));
3219 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0)(((tmp) & ~0x01000000L) | (0x01000000L & ((0) <<
0x18)));
3220 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL, tmp)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x10bb), tmp, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x10bb)), (tmp), 0));
3221
3222 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO,((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x10b9), adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000
, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset
[GC_HWIP][0][0] + 0x10b9)), (adev->gfx.mec.mec_fw_gpu_addr
& 0xFFFFF000), 0))
3223 adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x10b9), adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000
, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset
[GC_HWIP][0][0] + 0x10b9)), (adev->gfx.mec.mec_fw_gpu_addr
& 0xFFFFF000), 0));
3224 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI,((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x10ba), ((u32)(((adev->gfx.mec.mec_fw_gpu_addr) >>
16) >> 16)), 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][0] + 0x10ba)), (((u32)(((adev
->gfx.mec.mec_fw_gpu_addr) >> 16) >> 16))), 0)
)
3225 upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x10ba), ((u32)(((adev->gfx.mec.mec_fw_gpu_addr) >>
16) >> 16)), 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][0] + 0x10ba)), (((u32)(((adev
->gfx.mec.mec_fw_gpu_addr) >> 16) >> 16))), 0)
);
3226
3227 /* MEC1 */
3228 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x581a), mec_hdr->jt_offset, 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x581a)), (mec_hdr
->jt_offset), 0))
3229 mec_hdr->jt_offset)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x581a), mec_hdr->jt_offset, 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x581a)), (mec_hdr
->jt_offset), 0));
3230 for (i = 0; i < mec_hdr->jt_size; i++)
3231 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_DATA,((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x581b), ((__uint32_t)(*(__uint32_t *)(fw_data + mec_hdr
->jt_offset + i))), 0, GC_HWIP) : amdgpu_device_wreg(adev,
((adev->reg_offset[GC_HWIP][0][1] + 0x581b)), (((__uint32_t
)(*(__uint32_t *)(fw_data + mec_hdr->jt_offset + i)))), 0)
)
3232 le32_to_cpup(fw_data + mec_hdr->jt_offset + i))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x581b), ((__uint32_t)(*(__uint32_t *)(fw_data + mec_hdr
->jt_offset + i))), 0, GC_HWIP) : amdgpu_device_wreg(adev,
((adev->reg_offset[GC_HWIP][0][1] + 0x581b)), (((__uint32_t
)(*(__uint32_t *)(fw_data + mec_hdr->jt_offset + i)))), 0)
);
3233
3234 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR,((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x581a), adev->gfx.mec_fw_version, 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x581a)), (adev
->gfx.mec_fw_version), 0))
3235 adev->gfx.mec_fw_version)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x581a), adev->gfx.mec_fw_version, 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][1] + 0x581a)), (adev
->gfx.mec_fw_version), 0));
3236 /* Todo : Loading MEC2 firmware is only necessary if MEC2 should run different microcode than MEC1. */
3237
3238 return 0;
3239}
3240
3241/* KIQ functions */
3242static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)
3243{
3244 uint32_t tmp;
3245 struct amdgpu_device *adev = ring->adev;
3246
3247 /* tell RLC which is KIQ queue */
3248 tmp = RREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4caa, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4caa), 0));
3249 tmp &= 0xffffff00;
3250 tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
3251 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][1]
+ 0x4caa; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, tmp, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (tmp), 0)); } while
(0);
3252 tmp |= 0x80;
3253 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][1]
+ 0x4caa; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, tmp, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (tmp), 0)); } while
(0);
3254}
3255
3256static void gfx_v9_0_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd)
3257{
3258 struct amdgpu_device *adev = ring->adev;
3259
3260 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
3261 if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) {
3262 mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
3263 mqd->cp_hqd_queue_priority =
3264 AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM15;
3265 }
3266 }
3267}
3268
3269static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring)
3270{
3271 struct amdgpu_device *adev = ring->adev;
3272 struct v9_mqd *mqd = ring->mqd_ptr;
3273 uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
3274 uint32_t tmp;
3275
3276 mqd->header = 0xC0310800;
3277 mqd->compute_pipelinestat_enable = 0x00000001;
3278 mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
3279 mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
3280 mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
3281 mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
3282 mqd->compute_static_thread_mgmt_se4 = 0xffffffff;
3283 mqd->compute_static_thread_mgmt_se5 = 0xffffffff;
3284 mqd->compute_static_thread_mgmt_se6 = 0xffffffff;
3285 mqd->compute_static_thread_mgmt_se7 = 0xffffffff;
3286 mqd->compute_misc_reserved = 0x00000003;
3287
3288 mqd->dynamic_cu_mask_addr_lo =
3289 lower_32_bits(ring->mqd_gpu_addr((u32)(ring->mqd_gpu_addr + __builtin_offsetof(struct v9_mqd_allocation
, dynamic_cu_mask)))
3290 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask))((u32)(ring->mqd_gpu_addr + __builtin_offsetof(struct v9_mqd_allocation
, dynamic_cu_mask)));
3291 mqd->dynamic_cu_mask_addr_hi =
3292 upper_32_bits(ring->mqd_gpu_addr((u32)(((ring->mqd_gpu_addr + __builtin_offsetof(struct v9_mqd_allocation
, dynamic_cu_mask)) >> 16) >> 16))
3293 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask))((u32)(((ring->mqd_gpu_addr + __builtin_offsetof(struct v9_mqd_allocation
, dynamic_cu_mask)) >> 16) >> 16));
3294
3295 eop_base_addr = ring->eop_gpu_addr >> 8;
3296 mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
3297 mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr)((u32)(((eop_base_addr) >> 16) >> 16));
3298
3299 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3300 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x126c, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x126c), 0));
3301 tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,(((tmp) & ~0x0000003FL) | (0x0000003FL & (((drm_order
(4096 / 4) - 1)) << 0x0)))
3302 (order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1))(((tmp) & ~0x0000003FL) | (0x0000003FL & (((drm_order
(4096 / 4) - 1)) << 0x0)));
3303
3304 mqd->cp_hqd_eop_control = tmp;
3305
3306 /* enable doorbell? */
3307 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1254, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1254), 0));
3308
3309 if (ring->use_doorbell) {
3310 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,(((tmp) & ~0x0FFFFFFCL) | (0x0FFFFFFCL & ((ring->doorbell_index
) << 0x2)))
3311 DOORBELL_OFFSET, ring->doorbell_index)(((tmp) & ~0x0FFFFFFCL) | (0x0FFFFFFCL & ((ring->doorbell_index
) << 0x2)));
3312 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,(((tmp) & ~0x40000000L) | (0x40000000L & ((1) <<
0x1e)))
3313 DOORBELL_EN, 1)(((tmp) & ~0x40000000L) | (0x40000000L & ((1) <<
0x1e)));
3314 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,(((tmp) & ~0x10000000L) | (0x10000000L & ((0) <<
0x1c)))
3315 DOORBELL_SOURCE, 0)(((tmp) & ~0x10000000L) | (0x10000000L & ((0) <<
0x1c)));
3316 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,(((tmp) & ~0x80000000L) | (0x80000000L & ((0) <<
0x1f)))
3317 DOORBELL_HIT, 0)(((tmp) & ~0x80000000L) | (0x80000000L & ((0) <<
0x1f)));
3318 } else {
3319 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,(((tmp) & ~0x40000000L) | (0x40000000L & ((0) <<
0x1e)))
3320 DOORBELL_EN, 0)(((tmp) & ~0x40000000L) | (0x40000000L & ((0) <<
0x1e)));
3321 }
3322
3323 mqd->cp_hqd_pq_doorbell_control = tmp;
3324
3325 /* disable the queue if it's active */
3326 ring->wptr = 0;
3327 mqd->cp_hqd_dequeue_request = 0;
3328 mqd->cp_hqd_pq_rptr = 0;
3329 mqd->cp_hqd_pq_wptr_lo = 0;
3330 mqd->cp_hqd_pq_wptr_hi = 0;
3331
3332 /* set the pointer to the MQD */
3333 mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
3334 mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr)((u32)(((ring->mqd_gpu_addr) >> 16) >> 16));
3335
3336 /* set MQD vmid to 0 */
3337 tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1267, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1267), 0));
3338 tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0)(((tmp) & ~0x0000000FL) | (0x0000000FL & ((0) <<
0x0)));
3339 mqd->cp_mqd_control = tmp;
3340
3341 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3342 hqd_gpu_addr = ring->gpu_addr >> 8;
3343 mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
3344 mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr)((u32)(((hqd_gpu_addr) >> 16) >> 16));
3345
3346 /* set up the HQD, this is similar to CP_RB0_CNTL */
3347 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1256, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1256), 0));
3348 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,(((tmp) & ~0x0000003FL) | (0x0000003FL & (((drm_order
(ring->ring_size / 4) - 1)) << 0x0)))
3349 (order_base_2(ring->ring_size / 4) - 1))(((tmp) & ~0x0000003FL) | (0x0000003FL & (((drm_order
(ring->ring_size / 4) - 1)) << 0x0)));
3350 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,(((tmp) & ~0x00003F00L) | (0x00003F00L & (((drm_order
(4096 / 4) - 1)) << 0x8)))
3351 (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1))(((tmp) & ~0x00003F00L) | (0x00003F00L & (((drm_order
(4096 / 4) - 1)) << 0x8)));
3352#ifdef __BIG_ENDIAN
3353 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1)(((tmp) & ~0x00060000L) | (0x00060000L & ((1) <<
0x11)));
3354#endif
3355 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0)(((tmp) & ~0x10000000L) | (0x10000000L & ((0) <<
0x1c)));
3356 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0)(((tmp) & ~0x20000000L) | (0x20000000L & ((0) <<
0x1d)));
3357 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1)(((tmp) & ~0x40000000L) | (0x40000000L & ((1) <<
0x1e)));
3358 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1)(((tmp) & ~0x80000000L) | (0x80000000L & ((1) <<
0x1f)));
3359 mqd->cp_hqd_pq_control = tmp;
3360
3361 /* set the wb address whether it's enabled or not */
3362 wb_gpu_addr = ring->rptr_gpu_addr;
3363 mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
3364 mqd->cp_hqd_pq_rptr_report_addr_hi =
3365 upper_32_bits(wb_gpu_addr)((u32)(((wb_gpu_addr) >> 16) >> 16)) & 0xffff;
3366
3367 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3368 wb_gpu_addr = ring->wptr_gpu_addr;
3369 mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3370 mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr)((u32)(((wb_gpu_addr) >> 16) >> 16)) & 0xffff;
3371
3372 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3373 ring->wptr = 0;
3374 mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x124f, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x124f), 0));
3375
3376 /* set the vmid for the queue */
3377 mqd->cp_hqd_vmid = 0;
3378
3379 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1249, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1249), 0));
3380 tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53)(((tmp) & ~0x0003FF00L) | (0x0003FF00L & ((0x53) <<
0x8)));
3381 mqd->cp_hqd_persistent_state = tmp;
3382
3383 /* set MIN_IB_AVAIL_SIZE */
3384 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x125a, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x125a), 0));
3385 tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3)(((tmp) & ~0x00300000L) | (0x00300000L & ((3) <<
0x14)));
3386 mqd->cp_hqd_ib_control = tmp;
3387
3388 /* set static priority for a queue/ring */
3389 gfx_v9_0_mqd_set_priority(ring, mqd);
3390 mqd->cp_hqd_quantum = RREG32_SOC15(GC, 0, mmCP_HQD_QUANTUM)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x124c, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x124c), 0));
3391
3392 /* map_queues packet doesn't need activate the queue,
3393 * so only kiq need set this field.
3394 */
3395 if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
3396 mqd->cp_hqd_active = 1;
3397
3398 return 0;
3399}
3400
3401static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring)
3402{
3403 struct amdgpu_device *adev = ring->adev;
3404 struct v9_mqd *mqd = ring->mqd_ptr;
3405 int j;
3406
3407 /* disable wptr polling */
3408 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1083, (((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1083, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1083), 0)) & ~0x80000000L) |
(0) << 0x1f, 0, GC_HWIP) : amdgpu_device_wreg(adev, (adev
->reg_offset[GC_HWIP][0][0] + 0x1083), ((((((adev)->virt
.caps & (1 << 2)) && adev->gfx.rlc.funcs
&& adev->gfx.rlc.rlcg_reg_access_supported) ? amdgpu_sriov_rreg
(adev, adev->reg_offset[GC_HWIP][0][0] + 0x1083, 0, GC_HWIP
) : amdgpu_device_rreg(adev, (adev->reg_offset[GC_HWIP][0]
[0] + 0x1083), 0)) & ~0x80000000L) | (0) << 0x1f), 0
));
3409
3410 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x126a; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_eop_base_addr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_eop_base_addr_lo), 0)); } while (0)
3411 mqd->cp_hqd_eop_base_addr_lo)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x126a; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_eop_base_addr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_eop_base_addr_lo), 0)); } while (0);
3412 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x126b; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_eop_base_addr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_eop_base_addr_hi), 0)); } while (0)
3413 mqd->cp_hqd_eop_base_addr_hi)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x126b; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_eop_base_addr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_eop_base_addr_hi), 0)); } while (0);
3414
3415 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3416 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_CONTROL,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x126c; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_eop_control
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_eop_control), 0)); } while (0)
3417 mqd->cp_hqd_eop_control)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x126c; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_eop_control
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_eop_control), 0)); } while (0);
3418
3419 /* enable doorbell? */
3420 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1254; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_doorbell_control
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_doorbell_control), 0)); } while (0)
3421 mqd->cp_hqd_pq_doorbell_control)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1254; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_doorbell_control
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_doorbell_control), 0)); } while (0);
3422
3423 /* disable the queue if it's active */
3424 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1247, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1247), 0)) & 1) {
3425 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x125d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 1, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (1), 0)); } while (
0);
3426 for (j = 0; j < adev->usec_timeout; j++) {
3427 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1247, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1247), 0)) & 1))
3428 break;
3429 udelay(1);
3430 }
3431 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x125d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_dequeue_request
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_dequeue_request), 0)); } while (0)
3432 mqd->cp_hqd_dequeue_request)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x125d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_dequeue_request
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_dequeue_request), 0)); } while (0);
3433 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x124f; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_rptr
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_rptr), 0)); } while (0)
3434 mqd->cp_hqd_pq_rptr)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x124f; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_rptr
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_rptr), 0)); } while (0);
3435 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x127b; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_lo), 0)); } while (0)
3436 mqd->cp_hqd_pq_wptr_lo)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x127b; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_lo), 0)); } while (0);
3437 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x127c; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_hi), 0)); } while (0)
3438 mqd->cp_hqd_pq_wptr_hi)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x127c; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_hi), 0)); } while (0);
3439 }
3440
3441 /* set the pointer to the MQD */
3442 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1245; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_mqd_base_addr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_mqd_base_addr_lo), 0)); } while (0)
3443 mqd->cp_mqd_base_addr_lo)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1245; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_mqd_base_addr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_mqd_base_addr_lo), 0)); } while (0);
3444 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR_HI,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1246; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_mqd_base_addr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_mqd_base_addr_hi), 0)); } while (0)
3445 mqd->cp_mqd_base_addr_hi)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1246; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_mqd_base_addr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_mqd_base_addr_hi), 0)); } while (0);
3446
3447 /* set MQD vmid to 0 */
3448 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_CONTROL,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1267; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_mqd_control
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_mqd_control), 0)); } while (0)
3449 mqd->cp_mqd_control)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1267; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_mqd_control
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_mqd_control), 0)); } while (0);
3450
3451 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3452 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x124d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_base_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_base_lo), 0)); } while (0)
3453 mqd->cp_hqd_pq_base_lo)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x124d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_base_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_base_lo), 0)); } while (0);
3454 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE_HI,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x124e; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_base_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_base_hi), 0)); } while (0)
3455 mqd->cp_hqd_pq_base_hi)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x124e; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_base_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_base_hi), 0)); } while (0);
3456
3457 /* set up the HQD, this is similar to CP_RB0_CNTL */
3458 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_CONTROL,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1256; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_control
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_control), 0)); } while (0)
3459 mqd->cp_hqd_pq_control)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1256; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_control
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_control), 0)); } while (0);
3460
3461 /* set the wb address whether it's enabled or not */
3462 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1250; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_rptr_report_addr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_rptr_report_addr_lo), 0)); } while (0)
3463 mqd->cp_hqd_pq_rptr_report_addr_lo)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1250; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_rptr_report_addr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_rptr_report_addr_lo), 0)); } while (0);
3464 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1251; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_rptr_report_addr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_rptr_report_addr_hi), 0)); } while (0)
3465 mqd->cp_hqd_pq_rptr_report_addr_hi)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1251; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_rptr_report_addr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_rptr_report_addr_hi), 0)); } while (0);
3466
3467 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3468 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1252; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_poll_addr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_poll_addr_lo), 0)); } while (0)
3469 mqd->cp_hqd_pq_wptr_poll_addr_lo)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1252; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_poll_addr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_poll_addr_lo), 0)); } while (0);
3470 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1253; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_poll_addr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_poll_addr_hi), 0)); } while (0)
3471 mqd->cp_hqd_pq_wptr_poll_addr_hi)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1253; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_poll_addr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_poll_addr_hi), 0)); } while (0);
3472
3473 /* enable the doorbell if requested */
3474 if (ring->use_doorbell) {
3475 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER,((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x105c), (adev->doorbell_index.kiq * 2) << 2, 0
, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP
][0][0] + 0x105c)), ((adev->doorbell_index.kiq * 2) <<
2), 0))
3476 (adev->doorbell_index.kiq * 2) << 2)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x105c), (adev->doorbell_index.kiq * 2) << 2, 0
, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset[GC_HWIP
][0][0] + 0x105c)), ((adev->doorbell_index.kiq * 2) <<
2), 0));
3477 /* If GC has entered CGPG, ringing doorbell > first page
3478 * doesn't wakeup GC. Enlarge CP_MEC_DOORBELL_RANGE_UPPER to
3479 * workaround this issue. And this change has to align with firmware
3480 * update.
3481 */
3482 if (check_if_enlarge_doorbell_range(adev))
3483 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x105d), (adev->doorbell.size - 4), 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][0] + 0x105d)), ((adev
->doorbell.size - 4)), 0))
3484 (adev->doorbell.size - 4))((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x105d), (adev->doorbell.size - 4), 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][0] + 0x105d)), ((adev
->doorbell.size - 4)), 0));
3485 else
3486 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x105d), (adev->doorbell_index.userqueue_end * 2) <<
2, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset
[GC_HWIP][0][0] + 0x105d)), ((adev->doorbell_index.userqueue_end
* 2) << 2), 0))
3487 (adev->doorbell_index.userqueue_end * 2) << 2)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x105d), (adev->doorbell_index.userqueue_end * 2) <<
2, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev->reg_offset
[GC_HWIP][0][0] + 0x105d)), ((adev->doorbell_index.userqueue_end
* 2) << 2), 0));
3488 }
3489
3490 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1254; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_doorbell_control
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_doorbell_control), 0)); } while (0)
3491 mqd->cp_hqd_pq_doorbell_control)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1254; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_doorbell_control
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_doorbell_control), 0)); } while (0);
3492
3493 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3494 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x127b; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_lo), 0)); } while (0)
3495 mqd->cp_hqd_pq_wptr_lo)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x127b; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_lo
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_lo), 0)); } while (0);
3496 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x127c; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_hi), 0)); } while (0)
3497 mqd->cp_hqd_pq_wptr_hi)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x127c; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_pq_wptr_hi
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_pq_wptr_hi), 0)); } while (0);
3498
3499 /* set the vmid for the queue */
3500 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_VMID, mqd->cp_hqd_vmid)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1248; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_vmid, (
1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg)
, (mqd->cp_hqd_vmid), 0)); } while (0);
3501
3502 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1249; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_persistent_state
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_persistent_state), 0)); } while (0)
3503 mqd->cp_hqd_persistent_state)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1249; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_persistent_state
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_persistent_state), 0)); } while (0);
3504
3505 /* activate the queue */
3506 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1247; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_active
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_active), 0)); } while (0)
3507 mqd->cp_hqd_active)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1247; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, mqd->cp_hqd_active
, (1<<2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg
), (mqd->cp_hqd_active), 0)); } while (0);
3508
3509 if (ring->use_doorbell)
3510 WREG32_FIELD15(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x10b8, (((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x10b8, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x10b8), 0)) & ~0x00000002L) |
(1) << 0x1, 0, GC_HWIP) : amdgpu_device_wreg(adev, (adev
->reg_offset[GC_HWIP][0][0] + 0x10b8), ((((((adev)->virt
.caps & (1 << 2)) && adev->gfx.rlc.funcs
&& adev->gfx.rlc.rlcg_reg_access_supported) ? amdgpu_sriov_rreg
(adev, adev->reg_offset[GC_HWIP][0][0] + 0x10b8, 0, GC_HWIP
) : amdgpu_device_rreg(adev, (adev->reg_offset[GC_HWIP][0]
[0] + 0x10b8), 0)) & ~0x00000002L) | (1) << 0x1), 0
));
3511
3512 return 0;
3513}
3514
3515static int gfx_v9_0_kiq_fini_register(struct amdgpu_ring *ring)
3516{
3517 struct amdgpu_device *adev = ring->adev;
3518 int j;
3519
3520 /* disable the queue if it's active */
3521 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1247, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1247), 0)) & 1) {
3522
3523 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x125d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 1, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (1), 0)); } while (
0);
3524
3525 for (j = 0; j < adev->usec_timeout; j++) {
3526 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1247, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1247), 0)) & 1))
3527 break;
3528 udelay(1);
3529 }
3530
3531 if (j == AMDGPU_MAX_USEC_TIMEOUT100000) {
3532 DRM_DEBUG("KIQ dequeue request failed.\n")___drm_dbg(((void *)0), DRM_UT_CORE, "KIQ dequeue request failed.\n"
);
3533
3534 /* Manual disable if dequeue request times out */
3535 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE, 0)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1247; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0);
3536 }
3537
3538 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x125d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0)
3539 0)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x125d; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0);
3540 }
3541
3542 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IQ_TIMER, 0)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x125b; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0);
3543 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IB_CONTROL, 0)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x125a; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0);
3544 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE, 0)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1249; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0);
3545 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1254; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0x40000000, (1<<
2), GC_HWIP) : amdgpu_device_wreg(adev, (target_reg), (0x40000000
), 0)); } while (0);
3546 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x1254; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0);
3547 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR, 0)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x124f; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0);
3548 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 0)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x127c; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0);
3549 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 0)do { uint32_t target_reg = adev->reg_offset[GC_HWIP][0][0]
+ 0x127b; ((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, target_reg, 0, (1<<2), GC_HWIP
) : amdgpu_device_wreg(adev, (target_reg), (0), 0)); } while (
0);
3550
3551 return 0;
3552}
3553
3554static int gfx_v9_0_kiq_init_queue(struct amdgpu_ring *ring)
3555{
3556 struct amdgpu_device *adev = ring->adev;
3557 struct v9_mqd *mqd = ring->mqd_ptr;
3558 int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS8;
3559 struct v9_mqd *tmp_mqd;
3560
3561 gfx_v9_0_kiq_setting(ring);
3562
3563 /* GPU could be in bad state during probe, driver trigger the reset
3564 * after load the SMU, in this case , the mqd is not be initialized.
3565 * driver need to re-init the mqd.
3566 * check mqd->cp_hqd_pq_control since this value should not be 0
3567 */
3568 tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx];
3569 if (amdgpu_in_reset(adev) && tmp_mqd->cp_hqd_pq_control){
3570 /* for GPU_RESET case , reset MQD to a clean status */
3571 if (adev->gfx.mec.mqd_backup[mqd_idx])
3572 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation))__builtin_memcpy((mqd), (adev->gfx.mec.mqd_backup[mqd_idx]
), (sizeof(struct v9_mqd_allocation)));
3573
3574 /* reset ring buffer */
3575 ring->wptr = 0;
3576 amdgpu_ring_clear_ring(ring);
3577
3578 mutex_lock(&adev->srbm_mutex)rw_enter_write(&adev->srbm_mutex);
3579 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3580 gfx_v9_0_kiq_init_register(ring);
3581 soc15_grbm_select(adev, 0, 0, 0, 0);
3582 mutex_unlock(&adev->srbm_mutex)rw_exit_write(&adev->srbm_mutex);
3583 } else {
3584 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation))__builtin_memset(((void *)mqd), (0), (sizeof(struct v9_mqd_allocation
)));
3585 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3586 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3587 mutex_lock(&adev->srbm_mutex)rw_enter_write(&adev->srbm_mutex);
3588 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3589 gfx_v9_0_mqd_init(ring);
3590 gfx_v9_0_kiq_init_register(ring);
3591 soc15_grbm_select(adev, 0, 0, 0, 0);
3592 mutex_unlock(&adev->srbm_mutex)rw_exit_write(&adev->srbm_mutex);
3593
3594 if (adev->gfx.mec.mqd_backup[mqd_idx])
3595 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation))__builtin_memcpy((adev->gfx.mec.mqd_backup[mqd_idx]), (mqd
), (sizeof(struct v9_mqd_allocation)));
3596 }
3597
3598 return 0;
3599}
3600
3601static int gfx_v9_0_kcq_init_queue(struct amdgpu_ring *ring)
3602{
3603 struct amdgpu_device *adev = ring->adev;
3604 struct v9_mqd *mqd = ring->mqd_ptr;
3605 int mqd_idx = ring - &adev->gfx.compute_ring[0];
3606 struct v9_mqd *tmp_mqd;
3607
3608 /* Same as above kiq init, driver need to re-init the mqd if mqd->cp_hqd_pq_control
3609 * is not be initialized before
3610 */
3611 tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx];
3612
3613 if (!tmp_mqd->cp_hqd_pq_control ||
3614 (!amdgpu_in_reset(adev) && !adev->in_suspend)) {
3615 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation))__builtin_memset(((void *)mqd), (0), (sizeof(struct v9_mqd_allocation
)));
3616 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
3617 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
3618 mutex_lock(&adev->srbm_mutex)rw_enter_write(&adev->srbm_mutex);
3619 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3620 gfx_v9_0_mqd_init(ring);
3621 soc15_grbm_select(adev, 0, 0, 0, 0);
3622 mutex_unlock(&adev->srbm_mutex)rw_exit_write(&adev->srbm_mutex);
3623
3624 if (adev->gfx.mec.mqd_backup[mqd_idx])
3625 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation))__builtin_memcpy((adev->gfx.mec.mqd_backup[mqd_idx]), (mqd
), (sizeof(struct v9_mqd_allocation)));
3626 } else if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
3627 /* reset MQD to a clean status */
3628 if (adev->gfx.mec.mqd_backup[mqd_idx])
3629 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation))__builtin_memcpy((mqd), (adev->gfx.mec.mqd_backup[mqd_idx]
), (sizeof(struct v9_mqd_allocation)));
3630
3631 /* reset ring buffer */
3632 ring->wptr = 0;
3633 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, 0)({ typeof(*((atomic64_t *)ring->wptr_cpu_addr)) __tmp = ((
0)); *(volatile typeof(*((atomic64_t *)ring->wptr_cpu_addr
)) *)&(*((atomic64_t *)ring->wptr_cpu_addr)) = __tmp; __tmp
; });
3634 amdgpu_ring_clear_ring(ring);
3635 } else {
3636 amdgpu_ring_clear_ring(ring);
3637 }
3638
3639 return 0;
3640}
3641
3642static int gfx_v9_0_kiq_resume(struct amdgpu_device *adev)
3643{
3644 struct amdgpu_ring *ring;
3645 int r;
3646
3647 ring = &adev->gfx.kiq.ring;
3648
3649 r = amdgpu_bo_reserve(ring->mqd_obj, false0);
3650 if (unlikely(r != 0)__builtin_expect(!!(r != 0), 0))
3651 return r;
3652
3653 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3654 if (unlikely(r != 0)__builtin_expect(!!(r != 0), 0)) {
3655 amdgpu_bo_unreserve(ring->mqd_obj);
3656 return r;
3657 }
3658
3659 gfx_v9_0_kiq_init_queue(ring);
3660 amdgpu_bo_kunmap(ring->mqd_obj);
3661 ring->mqd_ptr = NULL((void *)0);
3662 amdgpu_bo_unreserve(ring->mqd_obj);
3663 ring->sched.ready = true1;
3664 return 0;
3665}
3666
3667static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev)
3668{
3669 struct amdgpu_ring *ring = NULL((void *)0);
3670 int r = 0, i;
3671
3672 gfx_v9_0_cp_compute_enable(adev, true1);
3673
3674 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3675 ring = &adev->gfx.compute_ring[i];
3676
3677 r = amdgpu_bo_reserve(ring->mqd_obj, false0);
3678 if (unlikely(r != 0)__builtin_expect(!!(r != 0), 0))
3679 goto done;
3680 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3681 if (!r) {
3682 r = gfx_v9_0_kcq_init_queue(ring);
3683 amdgpu_bo_kunmap(ring->mqd_obj);
3684 ring->mqd_ptr = NULL((void *)0);
3685 }
3686 amdgpu_bo_unreserve(ring->mqd_obj);
3687 if (r)
3688 goto done;
3689 }
3690
3691 r = amdgpu_gfx_enable_kcq(adev);
3692done:
3693 return r;
3694}
3695
3696static int gfx_v9_0_cp_resume(struct amdgpu_device *adev)
3697{
3698 int r, i;
3699 struct amdgpu_ring *ring;
3700
3701 if (!(adev->flags & AMD_IS_APU))
3702 gfx_v9_0_enable_gui_idle_interrupt(adev, false0);
3703
3704 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
3705 if (adev->gfx.num_gfx_rings) {
3706 /* legacy firmware loading */
3707 r = gfx_v9_0_cp_gfx_load_microcode(adev);
3708 if (r)
3709 return r;
3710 }
3711
3712 r = gfx_v9_0_cp_compute_load_microcode(adev);
3713 if (r)
3714 return r;
3715 }
3716
3717 r = gfx_v9_0_kiq_resume(adev);
3718 if (r)
3719 return r;
3720
3721 if (adev->gfx.num_gfx_rings) {
3722 r = gfx_v9_0_cp_gfx_resume(adev);
3723 if (r)
3724 return r;
3725 }
3726
3727 r = gfx_v9_0_kcq_resume(adev);
3728 if (r)
3729 return r;
3730
3731 if (adev->gfx.num_gfx_rings) {
3732 ring = &adev->gfx.gfx_ring[0];
3733 r = amdgpu_ring_test_helper(ring);
3734 if (r)
3735 return r;
3736 }
3737
3738 for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3739 ring = &adev->gfx.compute_ring[i];
3740 amdgpu_ring_test_helper(ring);
3741 }
3742
3743 gfx_v9_0_enable_gui_idle_interrupt(adev, true1);
3744
3745 return 0;
3746}
3747
3748static void gfx_v9_0_init_tcp_config(struct amdgpu_device *adev)
3749{
3750 u32 tmp;
3751
3752 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)(((9) << 16) | ((4) << 8) | (1)) &&
3753 adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 2)(((9) << 16) | ((4) << 8) | (2)))
3754 return;
3755
3756 tmp = RREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0b05, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0b05), 0));
3757 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE64KHASH,(((tmp) & ~0x00000800L) | (0x00000800L & ((adev->df
.hash_status.hash_64k) << 0xb)))
3758 adev->df.hash_status.hash_64k)(((tmp) & ~0x00000800L) | (0x00000800L & ((adev->df
.hash_status.hash_64k) << 0xb)));
3759 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE2MHASH,(((tmp) & ~0x00001000L) | (0x00001000L & ((adev->df
.hash_status.hash_2m) << 0xc)))
3760 adev->df.hash_status.hash_2m)(((tmp) & ~0x00001000L) | (0x00001000L & ((adev->df
.hash_status.hash_2m) << 0xc)));
3761 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE1GHASH,(((tmp) & ~0x00002000L) | (0x00002000L & ((adev->df
.hash_status.hash_1g) << 0xd)))
3762 adev->df.hash_status.hash_1g)(((tmp) & ~0x00002000L) | (0x00002000L & ((adev->df
.hash_status.hash_1g) << 0xd)));
3763 WREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG, tmp)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x0b05), tmp, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x0b05)), (tmp), 0));
3764}
3765
3766static void gfx_v9_0_cp_enable(struct amdgpu_device *adev, bool_Bool enable)
3767{
3768 if (adev->gfx.num_gfx_rings)
3769 gfx_v9_0_cp_gfx_enable(adev, enable);
3770 gfx_v9_0_cp_compute_enable(adev, enable);
3771}
3772
3773static int gfx_v9_0_hw_init(void *handle)
3774{
3775 int r;
3776 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3777
3778 if (!amdgpu_sriov_vf(adev)((adev)->virt.caps & (1 << 2)))
3779 gfx_v9_0_init_golden_registers(adev);
3780
3781 gfx_v9_0_constants_init(adev);
3782
3783 gfx_v9_0_init_tcp_config(adev);
3784
3785 r = adev->gfx.rlc.funcs->resume(adev);
3786 if (r)
3787 return r;
3788
3789 r = gfx_v9_0_cp_resume(adev);
3790 if (r)
3791 return r;
3792
3793 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2)(((9) << 16) | ((4) << 8) | (2)))
3794 gfx_v9_4_2_set_power_brake_sequence(adev);
3795
3796 return r;
3797}
3798
3799static int gfx_v9_0_hw_fini(void *handle)
3800{
3801 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3802
3803 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
3804 amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
3805 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
3806 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
3807
3808 /* DF freeze and kcq disable will fail */
3809 if (!amdgpu_ras_intr_triggered())
3810 /* disable KCQ to avoid CPC touch memory not valid anymore */
3811 amdgpu_gfx_disable_kcq(adev);
3812
3813 if (amdgpu_sriov_vf(adev)((adev)->virt.caps & (1 << 2))) {
3814 gfx_v9_0_cp_gfx_enable(adev, false0);
3815 /* must disable polling for SRIOV when hw finished, otherwise
3816 * CPC engine may still keep fetching WB address which is already
3817 * invalid after sw finished and trigger DMAR reading error in
3818 * hypervisor side.
3819 */
3820 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1083, (((((adev)->virt.caps & (1 << 2)) &&
adev->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1083, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1083), 0)) & ~0x80000000L) |
(0) << 0x1f, 0, GC_HWIP) : amdgpu_device_wreg(adev, (adev
->reg_offset[GC_HWIP][0][0] + 0x1083), ((((((adev)->virt
.caps & (1 << 2)) && adev->gfx.rlc.funcs
&& adev->gfx.rlc.rlcg_reg_access_supported) ? amdgpu_sriov_rreg
(adev, adev->reg_offset[GC_HWIP][0][0] + 0x1083, 0, GC_HWIP
) : amdgpu_device_rreg(adev, (adev->reg_offset[GC_HWIP][0]
[0] + 0x1083), 0)) & ~0x80000000L) | (0) << 0x1f), 0
));
3821 return 0;
3822 }
3823
3824 /* Use deinitialize sequence from CAIL when unbinding device from driver,
3825 * otherwise KIQ is hanging when binding back
3826 */
3827 if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
3828 mutex_lock(&adev->srbm_mutex)rw_enter_write(&adev->srbm_mutex);
3829 soc15_grbm_select(adev, adev->gfx.kiq.ring.me,
3830 adev->gfx.kiq.ring.pipe,
3831 adev->gfx.kiq.ring.queue, 0);
3832 gfx_v9_0_kiq_fini_register(&adev->gfx.kiq.ring);
3833 soc15_grbm_select(adev, 0, 0, 0, 0);
3834 mutex_unlock(&adev->srbm_mutex)rw_exit_write(&adev->srbm_mutex);
3835 }
3836
3837 gfx_v9_0_cp_enable(adev, false0);
3838
3839 /* Skip stopping RLC with A+A reset or when RLC controls GFX clock */
3840 if ((adev->gmc.xgmi.connected_to_cpu && amdgpu_in_reset(adev)) ||
3841 (adev->ip_versions[GC_HWIP][0] >= IP_VERSION(9, 4, 2)(((9) << 16) | ((4) << 8) | (2)))) {
3842 dev_dbg(adev->dev, "Skipping RLC halt\n")do { } while(0);
3843 return 0;
3844 }
3845
3846 adev->gfx.rlc.funcs->stop(adev);
3847 return 0;
3848}
3849
3850static int gfx_v9_0_suspend(void *handle)
3851{
3852 return gfx_v9_0_hw_fini(handle);
3853}
3854
3855static int gfx_v9_0_resume(void *handle)
3856{
3857 return gfx_v9_0_hw_init(handle);
3858}
3859
3860static bool_Bool gfx_v9_0_is_idle(void *handle)
3861{
3862 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3863
3864 if (REG_GET_FIELD(RREG32_SOC15(GC, 0, mmGRBM_STATUS),(((((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0004, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0004), 0))) & 0x80000000L) >>
0x1f)
3865 GRBM_STATUS, GUI_ACTIVE)(((((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0004, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0004), 0))) & 0x80000000L) >>
0x1f))
3866 return false0;
3867 else
3868 return true1;
3869}
3870
3871static int gfx_v9_0_wait_for_idle(void *handle)
3872{
3873 unsigned i;
3874 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3875
3876 for (i = 0; i < adev->usec_timeout; i++) {
3877 if (gfx_v9_0_is_idle(handle))
3878 return 0;
3879 udelay(1);
3880 }
3881 return -ETIMEDOUT60;
3882}
3883
3884static int gfx_v9_0_soft_reset(void *handle)
3885{
3886 u32 grbm_soft_reset = 0;
3887 u32 tmp;
3888 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3889
3890 /* GRBM_STATUS */
3891 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0004, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0004), 0));
3892 if (tmp & (GRBM_STATUS__PA_BUSY_MASK0x02000000L | GRBM_STATUS__SC_BUSY_MASK0x01000000L |
3893 GRBM_STATUS__BCI_BUSY_MASK0x00800000L | GRBM_STATUS__SX_BUSY_MASK0x00100000L |
3894 GRBM_STATUS__TA_BUSY_MASK0x00004000L | GRBM_STATUS__VGT_BUSY_MASK0x00020000L |
3895 GRBM_STATUS__DB_BUSY_MASK0x04000000L | GRBM_STATUS__CB_BUSY_MASK0x40000000L |
3896 GRBM_STATUS__GDS_BUSY_MASK0x00008000L | GRBM_STATUS__SPI_BUSY_MASK0x00400000L |
3897 GRBM_STATUS__IA_BUSY_MASK0x00080000L | GRBM_STATUS__IA_BUSY_NO_DMA_MASK0x00040000L)) {
3898 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,(((grbm_soft_reset) & ~0x00000001L) | (0x00000001L & (
(1) << 0x0)))
3899 GRBM_SOFT_RESET, SOFT_RESET_CP, 1)(((grbm_soft_reset) & ~0x00000001L) | (0x00000001L & (
(1) << 0x0)));
3900 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,(((grbm_soft_reset) & ~0x00010000L) | (0x00010000L & (
(1) << 0x10)))
3901 GRBM_SOFT_RESET, SOFT_RESET_GFX, 1)(((grbm_soft_reset) & ~0x00010000L) | (0x00010000L & (
(1) << 0x10)));
3902 }
3903
3904 if (tmp & (GRBM_STATUS__CP_BUSY_MASK0x20000000L | GRBM_STATUS__CP_COHERENCY_BUSY_MASK0x10000000L)) {
3905 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,(((grbm_soft_reset) & ~0x00000001L) | (0x00000001L & (
(1) << 0x0)))
3906 GRBM_SOFT_RESET, SOFT_RESET_CP, 1)(((grbm_soft_reset) & ~0x00000001L) | (0x00000001L & (
(1) << 0x0)));
3907 }
3908
3909 /* GRBM_STATUS2 */
3910 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0002, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0002), 0));
3911 if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY)(((tmp) & 0x01000000L) >> 0x18))
3912 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,(((grbm_soft_reset) & ~0x00000004L) | (0x00000004L & (
(1) << 0x2)))
3913 GRBM_SOFT_RESET, SOFT_RESET_RLC, 1)(((grbm_soft_reset) & ~0x00000004L) | (0x00000004L & (
(1) << 0x2)));
3914
3915
3916 if (grbm_soft_reset) {
3917 /* stop the rlc */
3918 adev->gfx.rlc.funcs->stop(adev);
3919
3920 if (adev->gfx.num_gfx_rings)
3921 /* Disable GFX parsing/prefetching */
3922 gfx_v9_0_cp_gfx_enable(adev, false0);
3923
3924 /* Disable MEC parsing/prefetching */
3925 gfx_v9_0_cp_compute_enable(adev, false0);
3926
3927 if (grbm_soft_reset) {
3928 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0008, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0008), 0));
3929 tmp |= grbm_soft_reset;
3930 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp)do { } while(0);
3931 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x0008), tmp, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x0008)), (tmp), 0));
3932 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0008, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0008), 0));
3933
3934 udelay(50);
3935
3936 tmp &= ~grbm_soft_reset;
3937 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x0008), tmp, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][0] + 0x0008)), (tmp), 0));
3938 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x0008, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x0008), 0));
3939 }
3940
3941 /* Wait a little for things to settle down */
3942 udelay(50);
3943 }
3944 return 0;
3945}
3946
3947static uint64_t gfx_v9_0_kiq_read_clock(struct amdgpu_device *adev)
3948{
3949 signed long r, cnt = 0;
3950 unsigned long flags;
3951 uint32_t seq, reg_val_offs = 0;
3952 uint64_t value = 0;
3953 struct amdgpu_kiq *kiq = &adev->gfx.kiq;
3954 struct amdgpu_ring *ring = &kiq->ring;
3955
3956 BUG_ON(!ring->funcs->emit_rreg)((!(!ring->funcs->emit_rreg)) ? (void)0 : __assert("diagnostic "
, "/usr/src/sys/dev/pci/drm/amd/amdgpu/gfx_v9_0.c", 3956, "!(!ring->funcs->emit_rreg)"
));
3957
3958 spin_lock_irqsave(&kiq->ring_lock, flags)do { flags = 0; mtx_enter(&kiq->ring_lock); } while (0
);
3959 if (amdgpu_device_wb_get(adev, &reg_val_offs)) {
3960 pr_err("critical bug! too many kiq readers\n")printk("\0013" "amdgpu: " "critical bug! too many kiq readers\n"
);
3961 goto failed_unlock;
3962 }
3963 amdgpu_ring_alloc(ring, 32);
3964 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4)((3 << 30) | (((0x40) & 0xFF) << 8) | ((4) &
0x3FFF) << 16));
3965 amdgpu_ring_write(ring, 9 | /* src: register*/
3966 (5 << 8) | /* dst: memory */
3967 (1 << 16) | /* count sel */
3968 (1 << 20)); /* write confirm */
3969 amdgpu_ring_write(ring, 0);
3970 amdgpu_ring_write(ring, 0);
3971 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +((u32)(adev->wb.gpu_addr + reg_val_offs * 4))
3972 reg_val_offs * 4)((u32)(adev->wb.gpu_addr + reg_val_offs * 4)));
3973 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +((u32)(((adev->wb.gpu_addr + reg_val_offs * 4) >> 16
) >> 16))
3974 reg_val_offs * 4)((u32)(((adev->wb.gpu_addr + reg_val_offs * 4) >> 16
) >> 16)));
3975 r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT5000);
3976 if (r)
3977 goto failed_undo;
3978
3979 amdgpu_ring_commit(ring);
3980 spin_unlock_irqrestore(&kiq->ring_lock, flags)do { (void)(flags); mtx_leave(&kiq->ring_lock); } while
(0);
3981
3982 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT5000);
3983
3984 /* don't wait anymore for gpu reset case because this way may
3985 * block gpu_recover() routine forever, e.g. this virt_kiq_rreg
3986 * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will
3987 * never return if we keep waiting in virt_kiq_rreg, which cause
3988 * gpu_recover() hang there.
3989 *
3990 * also don't wait anymore for IRQ context
3991 * */
3992 if (r < 1 && (amdgpu_in_reset(adev)))
3993 goto failed_kiq_read;
3994
3995 might_sleep()assertwaitok();
3996 while (r < 1 && cnt++ < MAX_KIQ_REG_TRY1000) {
3997 drm_msleep(MAX_KIQ_REG_BAILOUT_INTERVAL)mdelay(5);
3998 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT5000);
3999 }
4000
4001 if (cnt > MAX_KIQ_REG_TRY1000)
4002 goto failed_kiq_read;
4003
4004 mb()do { __asm volatile("mfence" ::: "memory"); } while (0);
4005 value = (uint64_t)adev->wb.wb[reg_val_offs] |
4006 (uint64_t)adev->wb.wb[reg_val_offs + 1 ] << 32ULL;
4007 amdgpu_device_wb_free(adev, reg_val_offs);
4008 return value;
4009
4010failed_undo:
4011 amdgpu_ring_undo(ring);
4012failed_unlock:
4013 spin_unlock_irqrestore(&kiq->ring_lock, flags)do { (void)(flags); mtx_leave(&kiq->ring_lock); } while
(0);
4014failed_kiq_read:
4015 if (reg_val_offs)
4016 amdgpu_device_wb_free(adev, reg_val_offs);
4017 pr_err("failed to read gpu clock\n")printk("\0013" "amdgpu: " "failed to read gpu clock\n");
4018 return ~0;
4019}
4020
4021static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4022{
4023 uint64_t clock, clock_lo, clock_hi, hi_check;
4024
4025 switch (adev->ip_versions[GC_HWIP][0]) {
4026 case IP_VERSION(9, 3, 0)(((9) << 16) | ((3) << 8) | (0)):
4027 preempt_disable();
4028 clock_hi = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[SMUIO_HWIP][0
][1] + 0x0025, (1<<1), SMUIO_HWIP) : amdgpu_device_rreg
(adev, (adev->reg_offset[SMUIO_HWIP][0][1] + 0x0025), 0));
4029 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[SMUIO_HWIP][0
][1] + 0x0026, (1<<1), SMUIO_HWIP) : amdgpu_device_rreg
(adev, (adev->reg_offset[SMUIO_HWIP][0][1] + 0x0026), 0));
4030 hi_check = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[SMUIO_HWIP][0
][1] + 0x0025, (1<<1), SMUIO_HWIP) : amdgpu_device_rreg
(adev, (adev->reg_offset[SMUIO_HWIP][0][1] + 0x0025), 0));
4031 /* The SMUIO TSC clock frequency is 100MHz, which sets 32-bit carry over
4032 * roughly every 42 seconds.
4033 */
4034 if (hi_check != clock_hi) {
4035 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[SMUIO_HWIP][0
][1] + 0x0026, (1<<1), SMUIO_HWIP) : amdgpu_device_rreg
(adev, (adev->reg_offset[SMUIO_HWIP][0][1] + 0x0026), 0));
4036 clock_hi = hi_check;
4037 }
4038 preempt_enable();
4039 clock = clock_lo | (clock_hi << 32ULL);
4040 break;
4041 default:
4042 amdgpu_gfx_off_ctrl(adev, false0);
4043 mutex_lock(&adev->gfx.gpu_clock_mutex)rw_enter_write(&adev->gfx.gpu_clock_mutex);
4044 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 0, 1)(((9) << 16) | ((0) << 8) | (1)) && amdgpu_sriov_runtime(adev)((adev)->virt.caps & (1 << 4))) {
4045 clock = gfx_v9_0_kiq_read_clock(adev);
4046 } else {
4047 WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c26), 1, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((adev
->reg_offset[GC_HWIP][0][1] + 0x4c26)), (1), 0));
4048 clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c24, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c24), 0)) |
4049 ((uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_MSB)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c25, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c25), 0)) << 32ULL);
4050 }
4051 mutex_unlock(&adev->gfx.gpu_clock_mutex)rw_exit_write(&adev->gfx.gpu_clock_mutex);
4052 amdgpu_gfx_off_ctrl(adev, true1);
4053 break;
4054 }
4055 return clock;
4056}
4057
4058static void gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4059 uint32_t vmid,
4060 uint32_t gds_base, uint32_t gds_size,
4061 uint32_t gws_base, uint32_t gws_size,
4062 uint32_t oa_base, uint32_t oa_size)
4063{
4064 struct amdgpu_device *adev = ring->adev;
4065
4066 /* GDS Base */
4067 gfx_v9_0_write_data_to_reg(ring, 0, false0,
4068 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_BASE)(adev->reg_offset[GC_HWIP][0][0] + 0x1300) + 2 * vmid,
4069 gds_base);
4070
4071 /* GDS Size */
4072 gfx_v9_0_write_data_to_reg(ring, 0, false0,
4073 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE)(adev->reg_offset[GC_HWIP][0][0] + 0x1301) + 2 * vmid,
4074 gds_size);
4075
4076 /* GWS */
4077 gfx_v9_0_write_data_to_reg(ring, 0, false0,
4078 SOC15_REG_OFFSET(GC, 0, mmGDS_GWS_VMID0)(adev->reg_offset[GC_HWIP][0][0] + 0x1320) + vmid,
4079 gws_size << GDS_GWS_VMID0__SIZE__SHIFT0x10 | gws_base);
4080
4081 /* OA */
4082 gfx_v9_0_write_data_to_reg(ring, 0, false0,
4083 SOC15_REG_OFFSET(GC, 0, mmGDS_OA_VMID0)(adev->reg_offset[GC_HWIP][0][0] + 0x1330) + vmid,
4084 (1 << (oa_size + oa_base)) - (1 << oa_base));
4085}
4086
4087static const u32 vgpr_init_compute_shader[] =
4088{
4089 0xb07c0000, 0xbe8000ff,
4090 0x000000f8, 0xbf110800,
4091 0x7e000280, 0x7e020280,
4092 0x7e040280, 0x7e060280,
4093 0x7e080280, 0x7e0a0280,
4094 0x7e0c0280, 0x7e0e0280,
4095 0x80808800, 0xbe803200,
4096 0xbf84fff5, 0xbf9c0000,
4097 0xd28c0001, 0x0001007f,
4098 0xd28d0001, 0x0002027e,
4099 0x10020288, 0xb8810904,
4100 0xb7814000, 0xd1196a01,
4101 0x00000301, 0xbe800087,
4102 0xbefc00c1, 0xd89c4000,
4103 0x00020201, 0xd89cc080,
4104 0x00040401, 0x320202ff,
4105 0x00000800, 0x80808100,
4106 0xbf84fff8, 0x7e020280,
4107 0xbf810000, 0x00000000,
4108};
4109
4110static const u32 sgpr_init_compute_shader[] =
4111{
4112 0xb07c0000, 0xbe8000ff,
4113 0x0000005f, 0xbee50080,
4114 0xbe812c65, 0xbe822c65,
4115 0xbe832c65, 0xbe842c65,
4116 0xbe852c65, 0xb77c0005,
4117 0x80808500, 0xbf84fff8,
4118 0xbe800080, 0xbf810000,
4119};
4120
4121static const u32 vgpr_init_compute_shader_arcturus[] = {
4122 0xd3d94000, 0x18000080, 0xd3d94001, 0x18000080, 0xd3d94002, 0x18000080,
4123 0xd3d94003, 0x18000080, 0xd3d94004, 0x18000080, 0xd3d94005, 0x18000080,
4124 0xd3d94006, 0x18000080, 0xd3d94007, 0x18000080, 0xd3d94008, 0x18000080,
4125 0xd3d94009, 0x18000080, 0xd3d9400a, 0x18000080, 0xd3d9400b, 0x18000080,
4126 0xd3d9400c, 0x18000080, 0xd3d9400d, 0x18000080, 0xd3d9400e, 0x18000080,
4127 0xd3d9400f, 0x18000080, 0xd3d94010, 0x18000080, 0xd3d94011, 0x18000080,
4128 0xd3d94012, 0x18000080, 0xd3d94013, 0x18000080, 0xd3d94014, 0x18000080,
4129 0xd3d94015, 0x18000080, 0xd3d94016, 0x18000080, 0xd3d94017, 0x18000080,
4130 0xd3d94018, 0x18000080, 0xd3d94019, 0x18000080, 0xd3d9401a, 0x18000080,
4131 0xd3d9401b, 0x18000080, 0xd3d9401c, 0x18000080, 0xd3d9401d, 0x18000080,
4132 0xd3d9401e, 0x18000080, 0xd3d9401f, 0x18000080, 0xd3d94020, 0x18000080,
4133 0xd3d94021, 0x18000080, 0xd3d94022, 0x18000080, 0xd3d94023, 0x18000080,
4134 0xd3d94024, 0x18000080, 0xd3d94025, 0x18000080, 0xd3d94026, 0x18000080,
4135 0xd3d94027, 0x18000080, 0xd3d94028, 0x18000080, 0xd3d94029, 0x18000080,
4136 0xd3d9402a, 0x18000080, 0xd3d9402b, 0x18000080, 0xd3d9402c, 0x18000080,
4137 0xd3d9402d, 0x18000080, 0xd3d9402e, 0x18000080, 0xd3d9402f, 0x18000080,
4138 0xd3d94030, 0x18000080, 0xd3d94031, 0x18000080, 0xd3d94032, 0x18000080,
4139 0xd3d94033, 0x18000080, 0xd3d94034, 0x18000080, 0xd3d94035, 0x18000080,
4140 0xd3d94036, 0x18000080, 0xd3d94037, 0x18000080, 0xd3d94038, 0x18000080,
4141 0xd3d94039, 0x18000080, 0xd3d9403a, 0x18000080, 0xd3d9403b, 0x18000080,
4142 0xd3d9403c, 0x18000080, 0xd3d9403d, 0x18000080, 0xd3d9403e, 0x18000080,
4143 0xd3d9403f, 0x18000080, 0xd3d94040, 0x18000080, 0xd3d94041, 0x18000080,
4144 0xd3d94042, 0x18000080, 0xd3d94043, 0x18000080, 0xd3d94044, 0x18000080,
4145 0xd3d94045, 0x18000080, 0xd3d94046, 0x18000080, 0xd3d94047, 0x18000080,
4146 0xd3d94048, 0x18000080, 0xd3d94049, 0x18000080, 0xd3d9404a, 0x18000080,
4147 0xd3d9404b, 0x18000080, 0xd3d9404c, 0x18000080, 0xd3d9404d, 0x18000080,
4148 0xd3d9404e, 0x18000080, 0xd3d9404f, 0x18000080, 0xd3d94050, 0x18000080,
4149 0xd3d94051, 0x18000080, 0xd3d94052, 0x18000080, 0xd3d94053, 0x18000080,
4150 0xd3d94054, 0x18000080, 0xd3d94055, 0x18000080, 0xd3d94056, 0x18000080,
4151 0xd3d94057, 0x18000080, 0xd3d94058, 0x18000080, 0xd3d94059, 0x18000080,
4152 0xd3d9405a, 0x18000080, 0xd3d9405b, 0x18000080, 0xd3d9405c, 0x18000080,
4153 0xd3d9405d, 0x18000080, 0xd3d9405e, 0x18000080, 0xd3d9405f, 0x18000080,
4154 0xd3d94060, 0x18000080, 0xd3d94061, 0x18000080, 0xd3d94062, 0x18000080,
4155 0xd3d94063, 0x18000080, 0xd3d94064, 0x18000080, 0xd3d94065, 0x18000080,
4156 0xd3d94066, 0x18000080, 0xd3d94067, 0x18000080, 0xd3d94068, 0x18000080,
4157 0xd3d94069, 0x18000080, 0xd3d9406a, 0x18000080, 0xd3d9406b, 0x18000080,
4158 0xd3d9406c, 0x18000080, 0xd3d9406d, 0x18000080, 0xd3d9406e, 0x18000080,
4159 0xd3d9406f, 0x18000080, 0xd3d94070, 0x18000080, 0xd3d94071, 0x18000080,
4160 0xd3d94072, 0x18000080, 0xd3d94073, 0x18000080, 0xd3d94074, 0x18000080,
4161 0xd3d94075, 0x18000080, 0xd3d94076, 0x18000080, 0xd3d94077, 0x18000080,
4162 0xd3d94078, 0x18000080, 0xd3d94079, 0x18000080, 0xd3d9407a, 0x18000080,
4163 0xd3d9407b, 0x18000080, 0xd3d9407c, 0x18000080, 0xd3d9407d, 0x18000080,
4164 0xd3d9407e, 0x18000080, 0xd3d9407f, 0x18000080, 0xd3d94080, 0x18000080,
4165 0xd3d94081, 0x18000080, 0xd3d94082, 0x18000080, 0xd3d94083, 0x18000080,
4166 0xd3d94084, 0x18000080, 0xd3d94085, 0x18000080, 0xd3d94086, 0x18000080,
4167 0xd3d94087, 0x18000080, 0xd3d94088, 0x18000080, 0xd3d94089, 0x18000080,
4168 0xd3d9408a, 0x18000080, 0xd3d9408b, 0x18000080, 0xd3d9408c, 0x18000080,
4169 0xd3d9408d, 0x18000080, 0xd3d9408e, 0x18000080, 0xd3d9408f, 0x18000080,
4170 0xd3d94090, 0x18000080, 0xd3d94091, 0x18000080, 0xd3d94092, 0x18000080,
4171 0xd3d94093, 0x18000080, 0xd3d94094, 0x18000080, 0xd3d94095, 0x18000080,
4172 0xd3d94096, 0x18000080, 0xd3d94097, 0x18000080, 0xd3d94098, 0x18000080,
4173 0xd3d94099, 0x18000080, 0xd3d9409a, 0x18000080, 0xd3d9409b, 0x18000080,
4174 0xd3d9409c, 0x18000080, 0xd3d9409d, 0x18000080, 0xd3d9409e, 0x18000080,
4175 0xd3d9409f, 0x18000080, 0xd3d940a0, 0x18000080, 0xd3d940a1, 0x18000080,
4176 0xd3d940a2, 0x18000080, 0xd3d940a3, 0x18000080, 0xd3d940a4, 0x18000080,
4177 0xd3d940a5, 0x18000080, 0xd3d940a6, 0x18000080, 0xd3d940a7, 0x18000080,
4178 0xd3d940a8, 0x18000080, 0xd3d940a9, 0x18000080, 0xd3d940aa, 0x18000080,
4179 0xd3d940ab, 0x18000080, 0xd3d940ac, 0x18000080, 0xd3d940ad, 0x18000080,
4180 0xd3d940ae, 0x18000080, 0xd3d940af, 0x18000080, 0xd3d940b0, 0x18000080,
4181 0xd3d940b1, 0x18000080, 0xd3d940b2, 0x18000080, 0xd3d940b3, 0x18000080,
4182 0xd3d940b4, 0x18000080, 0xd3d940b5, 0x18000080, 0xd3d940b6, 0x18000080,
4183 0xd3d940b7, 0x18000080, 0xd3d940b8, 0x18000080, 0xd3d940b9, 0x18000080,
4184 0xd3d940ba, 0x18000080, 0xd3d940bb, 0x18000080, 0xd3d940bc, 0x18000080,
4185 0xd3d940bd, 0x18000080, 0xd3d940be, 0x18000080, 0xd3d940bf, 0x18000080,
4186 0xd3d940c0, 0x18000080, 0xd3d940c1, 0x18000080, 0xd3d940c2, 0x18000080,
4187 0xd3d940c3, 0x18000080, 0xd3d940c4, 0x18000080, 0xd3d940c5, 0x18000080,
4188 0xd3d940c6, 0x18000080, 0xd3d940c7, 0x18000080, 0xd3d940c8, 0x18000080,
4189 0xd3d940c9, 0x18000080, 0xd3d940ca, 0x18000080, 0xd3d940cb, 0x18000080,
4190 0xd3d940cc, 0x18000080, 0xd3d940cd, 0x18000080, 0xd3d940ce, 0x18000080,
4191 0xd3d940cf, 0x18000080, 0xd3d940d0, 0x18000080, 0xd3d940d1, 0x18000080,
4192 0xd3d940d2, 0x18000080, 0xd3d940d3, 0x18000080, 0xd3d940d4, 0x18000080,
4193 0xd3d940d5, 0x18000080, 0xd3d940d6, 0x18000080, 0xd3d940d7, 0x18000080,
4194 0xd3d940d8, 0x18000080, 0xd3d940d9, 0x18000080, 0xd3d940da, 0x18000080,
4195 0xd3d940db, 0x18000080, 0xd3d940dc, 0x18000080, 0xd3d940dd, 0x18000080,
4196 0xd3d940de, 0x18000080, 0xd3d940df, 0x18000080, 0xd3d940e0, 0x18000080,
4197 0xd3d940e1, 0x18000080, 0xd3d940e2, 0x18000080, 0xd3d940e3, 0x18000080,
4198 0xd3d940e4, 0x18000080, 0xd3d940e5, 0x18000080, 0xd3d940e6, 0x18000080,
4199 0xd3d940e7, 0x18000080, 0xd3d940e8, 0x18000080, 0xd3d940e9, 0x18000080,
4200 0xd3d940ea, 0x18000080, 0xd3d940eb, 0x18000080, 0xd3d940ec, 0x18000080,
4201 0xd3d940ed, 0x18000080, 0xd3d940ee, 0x18000080, 0xd3d940ef, 0x18000080,
4202 0xd3d940f0, 0x18000080, 0xd3d940f1, 0x18000080, 0xd3d940f2, 0x18000080,
4203 0xd3d940f3, 0x18000080, 0xd3d940f4, 0x18000080, 0xd3d940f5, 0x18000080,
4204 0xd3d940f6, 0x18000080, 0xd3d940f7, 0x18000080, 0xd3d940f8, 0x18000080,
4205 0xd3d940f9, 0x18000080, 0xd3d940fa, 0x18000080, 0xd3d940fb, 0x18000080,
4206 0xd3d940fc, 0x18000080, 0xd3d940fd, 0x18000080, 0xd3d940fe, 0x18000080,
4207 0xd3d940ff, 0x18000080, 0xb07c0000, 0xbe8a00ff, 0x000000f8, 0xbf11080a,
4208 0x7e000280, 0x7e020280, 0x7e040280, 0x7e060280, 0x7e080280, 0x7e0a0280,
4209 0x7e0c0280, 0x7e0e0280, 0x808a880a, 0xbe80320a, 0xbf84fff5, 0xbf9c0000,
4210 0xd28c0001, 0x0001007f, 0xd28d0001, 0x0002027e, 0x10020288, 0xb88b0904,
4211 0xb78b4000, 0xd1196a01, 0x00001701, 0xbe8a0087, 0xbefc00c1, 0xd89c4000,
4212 0x00020201, 0xd89cc080, 0x00040401, 0x320202ff, 0x00000800, 0x808a810a,
4213 0xbf84fff8, 0xbf810000,
4214};
4215
4216/* When below register arrays changed, please update gpr_reg_size,
4217 and sec_ded_counter_reg_size in function gfx_v9_0_do_edc_gpr_workarounds,
4218 to cover all gfx9 ASICs */
4219static const struct soc15_reg_entry vgpr_init_regs[] = {
4220 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS)GC_HWIP, 0, 0, 0x0e15, 0x0000000 },
4221 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X)GC_HWIP, 0, 0, 0x0e07, 0x40 },
4222 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y)GC_HWIP, 0, 0, 0x0e08, 4 },
4223 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z)GC_HWIP, 0, 0, 0x0e09, 1 },
4224 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1)GC_HWIP, 0, 0, 0x0e12, 0x3f },
4225 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2)GC_HWIP, 0, 0, 0x0e13, 0x400000 }, /* 64KB LDS */
4226 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0)GC_HWIP, 0, 0, 0x0e16, 0xffffffff },
4227 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1)GC_HWIP, 0, 0, 0x0e17, 0xffffffff },
4228 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2)GC_HWIP, 0, 0, 0x0e19, 0xffffffff },
4229 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3)GC_HWIP, 0, 0, 0x0e1a, 0xffffffff },
4230 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4)GC_HWIP, 0, 0, 0x0e25, 0xffffffff },
4231 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5)GC_HWIP, 0, 0, 0x0e26, 0xffffffff },
4232 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6)GC_HWIP, 0, 0, 0x0e27, 0xffffffff },
4233 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7)GC_HWIP, 0, 0, 0x0e28, 0xffffffff },
4234};
4235
4236static const struct soc15_reg_entry vgpr_init_regs_arcturus[] = {
4237 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS)GC_HWIP, 0, 0, 0x0e15, 0x0000000 },
4238 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X)GC_HWIP, 0, 0, 0x0e07, 0x40 },
4239 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y)GC_HWIP, 0, 0, 0x0e08, 4 },
4240 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z)GC_HWIP, 0, 0, 0x0e09, 1 },
4241 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1)GC_HWIP, 0, 0, 0x0e12, 0xbf },
4242 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2)GC_HWIP, 0, 0, 0x0e13, 0x400000 }, /* 64KB LDS */
4243 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0)GC_HWIP, 0, 0, 0x0e16, 0xffffffff },
4244 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1)GC_HWIP, 0, 0, 0x0e17, 0xffffffff },
4245 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2)GC_HWIP, 0, 0, 0x0e19, 0xffffffff },
4246 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3)GC_HWIP, 0, 0, 0x0e1a, 0xffffffff },
4247 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4)GC_HWIP, 0, 0, 0x0e25, 0xffffffff },
4248 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5)GC_HWIP, 0, 0, 0x0e26, 0xffffffff },
4249 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6)GC_HWIP, 0, 0, 0x0e27, 0xffffffff },
4250 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7)GC_HWIP, 0, 0, 0x0e28, 0xffffffff },
4251};
4252
4253static const struct soc15_reg_entry sgpr1_init_regs[] = {
4254 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS)GC_HWIP, 0, 0, 0x0e15, 0x0000000 },
4255 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X)GC_HWIP, 0, 0, 0x0e07, 0x40 },
4256 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y)GC_HWIP, 0, 0, 0x0e08, 8 },
4257 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z)GC_HWIP, 0, 0, 0x0e09, 1 },
4258 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1)GC_HWIP, 0, 0, 0x0e12, 0x240 }, /* (80 GPRS) */
4259 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2)GC_HWIP, 0, 0, 0x0e13, 0x0 },
4260 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0)GC_HWIP, 0, 0, 0x0e16, 0x000000ff },
4261 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1)GC_HWIP, 0, 0, 0x0e17, 0x000000ff },
4262 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2)GC_HWIP, 0, 0, 0x0e19, 0x000000ff },
4263 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3)GC_HWIP, 0, 0, 0x0e1a, 0x000000ff },
4264 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4)GC_HWIP, 0, 0, 0x0e25, 0x000000ff },
4265 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5)GC_HWIP, 0, 0, 0x0e26, 0x000000ff },
4266 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6)GC_HWIP, 0, 0, 0x0e27, 0x000000ff },
4267 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7)GC_HWIP, 0, 0, 0x0e28, 0x000000ff },
4268};
4269
4270static const struct soc15_reg_entry sgpr2_init_regs[] = {
4271 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS)GC_HWIP, 0, 0, 0x0e15, 0x0000000 },
4272 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X)GC_HWIP, 0, 0, 0x0e07, 0x40 },
4273 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y)GC_HWIP, 0, 0, 0x0e08, 8 },
4274 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z)GC_HWIP, 0, 0, 0x0e09, 1 },
4275 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1)GC_HWIP, 0, 0, 0x0e12, 0x240 }, /* (80 GPRS) */
4276 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2)GC_HWIP, 0, 0, 0x0e13, 0x0 },
4277 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0)GC_HWIP, 0, 0, 0x0e16, 0x0000ff00 },
4278 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1)GC_HWIP, 0, 0, 0x0e17, 0x0000ff00 },
4279 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2)GC_HWIP, 0, 0, 0x0e19, 0x0000ff00 },
4280 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3)GC_HWIP, 0, 0, 0x0e1a, 0x0000ff00 },
4281 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4)GC_HWIP, 0, 0, 0x0e25, 0x0000ff00 },
4282 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5)GC_HWIP, 0, 0, 0x0e26, 0x0000ff00 },
4283 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6)GC_HWIP, 0, 0, 0x0e27, 0x0000ff00 },
4284 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7)GC_HWIP, 0, 0, 0x0e28, 0x0000ff00 },
4285};
4286
4287static const struct soc15_reg_entry gfx_v9_0_edc_counter_regs[] = {
4288 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT)GC_HWIP, 0, 0, 0x118e, 0, 1, 1},
4289 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT)GC_HWIP, 0, 0, 0x118f, 0, 1, 1},
4290 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT)GC_HWIP, 0, 0, 0x118a, 0, 1, 1},
4291 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT)GC_HWIP, 0, 0, 0x1189, 0, 1, 1},
4292 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT)GC_HWIP, 0, 0, 0x118d, 0, 1, 1},
4293 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT)GC_HWIP, 0, 0, 0x118b, 0, 1, 1},
4294 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT)GC_HWIP, 0, 0, 0x1192, 0, 1, 1},
4295 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT)GC_HWIP, 0, 0, 0x1193, 0, 1, 1},
4296 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT)GC_HWIP, 0, 0, 0x1191, 0, 1, 1},
4297 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT)GC_HWIP, 0, 0, 0x05c5, 0, 1, 1},
4298 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_GRBM_CNT)GC_HWIP, 0, 0, 0x05c6, 0, 1, 1},
4299 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_DED)GC_HWIP, 0, 0, 0x05c7, 0, 1, 1},
4300 { SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT)GC_HWIP, 0, 0, 0x0445, 0, 4, 1},
4301 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT)GC_HWIP, 0, 0, 0x03a2, 0, 4, 6},
4302 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_DED_CNT)GC_HWIP, 0, 0, 0x03a4, 0, 4, 16},
4303 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_INFO)GC_HWIP, 0, 0, 0x03a5, 0, 4, 16},
4304 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_SEC_CNT)GC_HWIP, 0, 0, 0x03a3, 0, 4, 16},
4305 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT)GC_HWIP, 0, 0, 0x0b82, 0, 1, 16},
4306 { SOC15_REG_ENTRY(GC, 0, mmTCP_ATC_EDC_GATCL1_CNT)GC_HWIP, 0, 0, 0x12b1, 0, 4, 16},
4307 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT)GC_HWIP, 0, 0, 0x0b17, 0, 4, 16},
4308 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW)GC_HWIP, 0, 0, 0x0b18, 0, 4, 16},
4309 { SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT)GC_HWIP, 0, 0, 0x052e, 0, 4, 16},
4310 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2)GC_HWIP, 0, 0, 0x032c, 0, 4, 6},
4311 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT)GC_HWIP, 0, 0, 0x03a6, 0, 4, 16},
4312 { SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT)GC_HWIP, 0, 0, 0x0586, 0, 4, 16},
4313 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT)GC_HWIP, 0, 0, 0x05cb, 0, 1, 1},
4314 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT)GC_HWIP, 0, 0, 0x05cc, 0, 1, 1},
4315 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT)GC_HWIP, 0, 0, 0x0706, 0, 1, 32},
4316 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2)GC_HWIP, 0, 0, 0x0707, 0, 1, 32},
4317 { SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT)GC_HWIP, 0, 0, 0x0b60, 0, 1, 72},
4318 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2)GC_HWIP, 0, 0, 0x0b83, 0, 1, 16},
4319 { SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT)GC_HWIP, 0, 0, 0x0bc5, 0, 1, 2},
4320 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3)GC_HWIP, 0, 0, 0x032d, 0, 4, 6},
4321};
4322
4323static int gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device *adev)
4324{
4325 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4326 int i, r;
4327
4328 /* only support when RAS is enabled */
4329 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4330 return 0;
4331
4332 r = amdgpu_ring_alloc(ring, 7);
4333 if (r) {
4334 DRM_ERROR("amdgpu: GDS workarounds failed to lock ring %s (%d).\n",__drm_err("amdgpu: GDS workarounds failed to lock ring %s (%d).\n"
, ring->name, r)
4335 ring->name, r)__drm_err("amdgpu: GDS workarounds failed to lock ring %s (%d).\n"
, ring->name, r);
4336 return r;
4337 }
4338
4339 WREG32_SOC15(GC, 0, mmGDS_VMID0_BASE, 0x00000000)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1300), 0x00000000, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][0] + 0x1300)), (0x00000000
), 0));
4340 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, adev->gds.gds_size)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1301), adev->gds.gds_size, 0, GC_HWIP) : amdgpu_device_wreg
(adev, ((adev->reg_offset[GC_HWIP][0][0] + 0x1301)), (adev
->gds.gds_size), 0));
4341
4342 amdgpu_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5)((3 << 30) | (((0x50) & 0xFF) << 8) | ((5) &
0x3FFF) << 16));
4343 amdgpu_ring_write(ring, (PACKET3_DMA_DATA_CP_SYNC(1 << 31) |
4344 PACKET3_DMA_DATA_DST_SEL(1)((1) << 20) |
4345 PACKET3_DMA_DATA_SRC_SEL(2)((2) << 29) |
4346 PACKET3_DMA_DATA_ENGINE(0)((0) << 0)));
4347 amdgpu_ring_write(ring, 0);
4348 amdgpu_ring_write(ring, 0);
4349 amdgpu_ring_write(ring, 0);
4350 amdgpu_ring_write(ring, 0);
4351 amdgpu_ring_write(ring, PACKET3_DMA_DATA_CMD_RAW_WAIT(1 << 30) |
4352 adev->gds.gds_size);
4353
4354 amdgpu_ring_commit(ring);
4355
4356 for (i = 0; i < adev->usec_timeout; i++) {
4357 if (ring->wptr == gfx_v9_0_ring_get_rptr_compute(ring))
4358 break;
4359 udelay(1);
4360 }
4361
4362 if (i >= adev->usec_timeout)
4363 r = -ETIMEDOUT60;
4364
4365 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, 0x00000000)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1301), 0x00000000, 0, GC_HWIP) : amdgpu_device_wreg(adev
, ((adev->reg_offset[GC_HWIP][0][0] + 0x1301)), (0x00000000
), 0));
4366
4367 return r;
4368}
4369
4370static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
4371{
4372 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
4373 struct amdgpu_ib ib;
4374 struct dma_fence *f = NULL((void *)0);
4375 int r, i;
4376 unsigned total_size, vgpr_offset, sgpr_offset;
4377 u64 gpu_addr;
4378
4379 int compute_dim_x = adev->gfx.config.max_shader_engines *
4380 adev->gfx.config.max_cu_per_sh *
4381 adev->gfx.config.max_sh_per_se;
4382 int sgpr_work_group_size = 5;
4383 int gpr_reg_size = adev->gfx.config.max_shader_engines + 6;
4384 int vgpr_init_shader_size;
4385 const u32 *vgpr_init_shader_ptr;
4386 const struct soc15_reg_entry *vgpr_init_regs_ptr;
4387
4388 /* only support when RAS is enabled */
4389 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
4390 return 0;
4391
4392 /* bail if the compute ring is not ready */
4393 if (!ring->sched.ready)
4394 return 0;
4395
4396 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1)(((9) << 16) | ((4) << 8) | (1))) {
4397 vgpr_init_shader_ptr = vgpr_init_compute_shader_arcturus;
4398 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader_arcturus);
4399 vgpr_init_regs_ptr = vgpr_init_regs_arcturus;
4400 } else {
4401 vgpr_init_shader_ptr = vgpr_init_compute_shader;
4402 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader);
4403 vgpr_init_regs_ptr = vgpr_init_regs;
4404 }
4405
4406 total_size =
4407 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* VGPRS */
4408 total_size +=
4409 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS1 */
4410 total_size +=
4411 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS2 */
4412 total_size = roundup2(total_size, 256)(((total_size) + ((256) - 1)) & (~((__typeof(total_size))
(256) - 1)));
4413 vgpr_offset = total_size;
4414 total_size += roundup2(vgpr_init_shader_size, 256)(((vgpr_init_shader_size) + ((256) - 1)) & (~((__typeof(vgpr_init_shader_size
))(256) - 1)));
4415 sgpr_offset = total_size;
4416 total_size += sizeof(sgpr_init_compute_shader);
4417
4418 /* allocate an indirect buffer to put the commands in */
4419 memset(&ib, 0, sizeof(ib))__builtin_memset((&ib), (0), (sizeof(ib)));
4420 r = amdgpu_ib_get(adev, NULL((void *)0), total_size,
4421 AMDGPU_IB_POOL_DIRECT, &ib);
4422 if (r) {
4423 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r)__drm_err("amdgpu: failed to get ib (%d).\n", r);
4424 return r;
4425 }
4426
4427 /* load the compute shaders */
4428 for (i = 0; i < vgpr_init_shader_size/sizeof(u32); i++)
4429 ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_shader_ptr[i];
4430
4431 for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader)(sizeof((sgpr_init_compute_shader)) / sizeof((sgpr_init_compute_shader
)[0])); i++)
4432 ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
4433
4434 /* init the ib length to 0 */
4435 ib.length_dw = 0;
4436
4437 /* VGPR */
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)((3 << 30) | (((0x76) & 0xFF) << 8) | ((1) &
0x3FFF) << 16);
4441 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(vgpr_init_regs_ptr[i])(adev->reg_offset[vgpr_init_regs_ptr[i].hwip][vgpr_init_regs_ptr
[i].inst][vgpr_init_regs_ptr[i].seg] + vgpr_init_regs_ptr[i].
reg_offset)
4442 - PACKET3_SET_SH_REG_START0x00002c00;
4443 ib.ptr[ib.length_dw++] = vgpr_init_regs_ptr[i].reg_value;
4444 }
4445 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4446 gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
4447 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2)((3 << 30) | (((0x76) & 0xFF) << 8) | ((2) &
0x3FFF) << 16);
4448 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)(adev->reg_offset[GC_HWIP][0][0] + 0x0e0c)
4449 - PACKET3_SET_SH_REG_START0x00002c00;
4450 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr)((u32)(gpu_addr));
4451 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr)((u32)(((gpu_addr) >> 16) >> 16));
4452
4453 /* write dispatch packet */
4454 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3)((3 << 30) | (((0x15) & 0xFF) << 8) | ((3) &
0x3FFF) << 16);
4455 ib.ptr[ib.length_dw++] = compute_dim_x * 2; /* 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)(((0) & ~0x00000001L) | (0x00000001L & ((1) << 0x0
)));
4460
4461 /* write CS partial flush packet */
4462 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0)((3 << 30) | (((0x46) & 0xFF) << 8) | ((0) &
0x3FFF) << 16);
4463 ib.ptr[ib.length_dw++] = EVENT_TYPE(7)((7) << 0) | EVENT_INDEX(4)((4) << 8);
4464
4465 /* SGPR1 */
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)((3 << 30) | (((0x76) & 0xFF) << 8) | ((1) &
0x3FFF) << 16);
4469 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr1_init_regs[i])(adev->reg_offset[sgpr1_init_regs[i].hwip][sgpr1_init_regs
[i].inst][sgpr1_init_regs[i].seg] + sgpr1_init_regs[i].reg_offset
)
4470 - PACKET3_SET_SH_REG_START0x00002c00;
4471 ib.ptr[ib.length_dw++] = sgpr1_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)((3 << 30) | (((0x76) & 0xFF) << 8) | ((2) &
0x3FFF) << 16);
4476 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)(adev->reg_offset[GC_HWIP][0][0] + 0x0e0c)
4477 - PACKET3_SET_SH_REG_START0x00002c00;
4478 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr)((u32)(gpu_addr));
4479 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr)((u32)(((gpu_addr) >> 16) >> 16));
4480
4481 /* write dispatch packet */
4482 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3)((3 << 30) | (((0x15) & 0xFF) << 8) | ((3) &
0x3FFF) << 16);
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)(((0) & ~0x00000001L) | (0x00000001L & ((1) << 0x0
)));
4488
4489 /* write CS partial flush packet */
4490 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0)((3 << 30) | (((0x46) & 0xFF) << 8) | ((0) &
0x3FFF) << 16);
4491 ib.ptr[ib.length_dw++] = EVENT_TYPE(7)((7) << 0) | EVENT_INDEX(4)((4) << 8);
4492
4493 /* SGPR2 */
4494 /* write the register state for the compute dispatch */
4495 for (i = 0; i < gpr_reg_size; i++) {
4496 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1)((3 << 30) | (((0x76) & 0xFF) << 8) | ((1) &
0x3FFF) << 16);
4497 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr2_init_regs[i])(adev->reg_offset[sgpr2_init_regs[i].hwip][sgpr2_init_regs
[i].inst][sgpr2_init_regs[i].seg] + sgpr2_init_regs[i].reg_offset
)
4498 - PACKET3_SET_SH_REG_START0x00002c00;
4499 ib.ptr[ib.length_dw++] = sgpr2_init_regs[i].reg_value;
4500 }
4501 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
4502 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
4503 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2)((3 << 30) | (((0x76) & 0xFF) << 8) | ((2) &
0x3FFF) << 16);
4504 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO)(adev->reg_offset[GC_HWIP][0][0] + 0x0e0c)
4505 - PACKET3_SET_SH_REG_START0x00002c00;
4506 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr)((u32)(gpu_addr));
4507 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr)((u32)(((gpu_addr) >> 16) >> 16));
4508
4509 /* write dispatch packet */
4510 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3)((3 << 30) | (((0x15) & 0xFF) << 8) | ((3) &
0x3FFF) << 16);
4511 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */
4512 ib.ptr[ib.length_dw++] = 1; /* y */
4513 ib.ptr[ib.length_dw++] = 1; /* z */
4514 ib.ptr[ib.length_dw++] =
4515 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1)(((0) & ~0x00000001L) | (0x00000001L & ((1) << 0x0
)));
4516
4517 /* write CS partial flush packet */
4518 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0)((3 << 30) | (((0x46) & 0xFF) << 8) | ((0) &
0x3FFF) << 16);
4519 ib.ptr[ib.length_dw++] = EVENT_TYPE(7)((7) << 0) | EVENT_INDEX(4)((4) << 8);
4520
4521 /* shedule the ib on the ring */
4522 r = amdgpu_ib_schedule(ring, 1, &ib, NULL((void *)0), &f);
4523 if (r) {
4524 DRM_ERROR("amdgpu: ib submit failed (%d).\n", r)__drm_err("amdgpu: ib submit failed (%d).\n", r);
4525 goto fail;
4526 }
4527
4528 /* wait for the GPU to finish processing the IB */
4529 r = dma_fence_wait(f, false0);
4530 if (r) {
4531 DRM_ERROR("amdgpu: fence wait failed (%d).\n", r)__drm_err("amdgpu: fence wait failed (%d).\n", r);
4532 goto fail;
4533 }
4534
4535fail:
4536 amdgpu_ib_free(adev, &ib, NULL((void *)0));
4537 dma_fence_put(f);
4538
4539 return r;
4540}
4541
4542static int gfx_v9_0_early_init(void *handle)
4543{
4544 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4545
4546 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1)(((9) << 16) | ((4) << 8) | (1)) ||
4547 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2)(((9) << 16) | ((4) << 8) | (2)))
4548 adev->gfx.num_gfx_rings = 0;
4549 else
4550 adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS1;
4551 adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),(((amdgpu_gfx_get_num_kcq(adev))<(8))?(amdgpu_gfx_get_num_kcq
(adev)):(8))
4552 AMDGPU_MAX_COMPUTE_RINGS)(((amdgpu_gfx_get_num_kcq(adev))<(8))?(amdgpu_gfx_get_num_kcq
(adev)):(8));
4553 gfx_v9_0_set_kiq_pm4_funcs(adev);
4554 gfx_v9_0_set_ring_funcs(adev);
4555 gfx_v9_0_set_irq_funcs(adev);
4556 gfx_v9_0_set_gds_init(adev);
4557 gfx_v9_0_set_rlc_funcs(adev);
4558
4559 /* init rlcg reg access ctrl */
4560 gfx_v9_0_init_rlcg_reg_access_ctrl(adev);
4561
4562 return 0;
4563}
4564
4565static int gfx_v9_0_ecc_late_init(void *handle)
4566{
4567 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4568 int r;
4569
4570 /*
4571 * Temp workaround to fix the issue that CP firmware fails to
4572 * update read pointer when CPDMA is writing clearing operation
4573 * to GDS in suspend/resume sequence on several cards. So just
4574 * limit this operation in cold boot sequence.
4575 */
4576 if ((!adev->in_suspend) &&
4577 (adev->gds.gds_size)) {
4578 r = gfx_v9_0_do_edc_gds_workarounds(adev);
4579 if (r)
4580 return r;
4581 }
4582
4583 /* requires IBs so do in late init after IB pool is initialized */
4584 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2)(((9) << 16) | ((4) << 8) | (2)))
4585 r = gfx_v9_4_2_do_edc_gpr_workarounds(adev);
4586 else
4587 r = gfx_v9_0_do_edc_gpr_workarounds(adev);
4588
4589 if (r)
4590 return r;
4591
4592 if (adev->gfx.ras &&
4593 adev->gfx.ras->enable_watchdog_timer)
4594 adev->gfx.ras->enable_watchdog_timer(adev);
4595
4596 return 0;
4597}
4598
4599static int gfx_v9_0_late_init(void *handle)
4600{
4601 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4602 int r;
4603
4604 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4605 if (r)
4606 return r;
4607
4608 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4609 if (r)
4610 return r;
4611
4612 r = gfx_v9_0_ecc_late_init(handle);
4613 if (r)
4614 return r;
4615
4616 return 0;
4617}
4618
4619static bool_Bool gfx_v9_0_is_rlc_enabled(struct amdgpu_device *adev)
4620{
4621 uint32_t rlc_setting;
4622
4623 /* if RLC is not enabled, do nothing */
4624 rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c00, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c00), 0));
4625 if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK0x00000001L))
4626 return false0;
4627
4628 return true1;
4629}
4630
4631static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev)
4632{
4633 uint32_t data;
4634 unsigned i;
4635
4636 data = RLC_SAFE_MODE__CMD_MASK0x00000001L;
4637 data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT0x1);
4638 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c05), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c05)), (data), 0));
4639
4640 /* wait for RLC_SAFE_MODE */
4641 for (i = 0; i < adev->usec_timeout; i++) {
4642 if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)(((((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c05, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c05), 0))) & 0x00000001L) >>
0x0))
4643 break;
4644 udelay(1);
4645 }
4646}
4647
4648static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev)
4649{
4650 uint32_t data;
4651
4652 data = RLC_SAFE_MODE__CMD_MASK0x00000001L;
4653 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c05), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c05)), (data), 0));
4654}
4655
4656static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev,
4657 bool_Bool enable)
4658{
4659 amdgpu_gfx_rlc_enter_safe_mode(adev);
4660
4661 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG(1 << 0)) && enable) {
4662 gfx_v9_0_enable_gfx_cg_power_gating(adev, true1);
4663 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE(1 << 12))
4664 gfx_v9_0_enable_gfx_pipeline_powergating(adev, true1);
4665 } else {
4666 gfx_v9_0_enable_gfx_cg_power_gating(adev, false0);
4667 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE(1 << 12))
4668 gfx_v9_0_enable_gfx_pipeline_powergating(adev, false0);
4669 }
4670
4671 amdgpu_gfx_rlc_exit_safe_mode(adev);
4672}
4673
4674static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev,
4675 bool_Bool enable)
4676{
4677 /* TODO: double check if we need to perform under safe mode */
4678 /* gfx_v9_0_enter_rlc_safe_mode(adev); */
4679
4680 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG(1 << 1)) && enable)
4681 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, true1);
4682 else
4683 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, false0);
4684
4685 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG(1 << 2)) && enable)
4686 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, true1);
4687 else
4688 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, false0);
4689
4690 /* gfx_v9_0_exit_rlc_safe_mode(adev); */
4691}
4692
4693static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
4694 bool_Bool enable)
4695{
4696 uint32_t data, def;
4697
4698 amdgpu_gfx_rlc_enter_safe_mode(adev);
4699
4700 /* It is disabled by HW by default */
4701 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG(1ULL << 0))) {
4702 /* 1 - RLC_CGTT_MGCG_OVERRIDE */
4703 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c48, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c48), 0));
4704
4705 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1)(((9) << 16) | ((2) << 8) | (1)))
4706 data &= ~RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK0x00000001L;
4707
4708 data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK0x00000020L |
4709 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK0x00000004L |
4710 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK0x00000040L);
4711
4712 /* only for Vega10 & Raven1 */
4713 data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK0x00000002L;
4714
4715 if (def != data)
4716 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c48), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c48)), (data), 0));
4717
4718 /* MGLS is a global flag to control all MGLS in GFX */
4719 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS(1ULL << 1)) {
4720 /* 2 - RLC memory Light sleep */
4721 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS(1ULL << 7)) {
4722 def = data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c06, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c06), 0));
4723 data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK0x00000001L;
4724 if (def != data)
4725 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c06), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c06)), (data), 0));
4726 }
4727 /* 3 - CP memory Light sleep */
4728 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS(1ULL << 6)) {
4729 def = data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1079, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1079), 0));
4730 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK0x00000001L;
4731 if (def != data)
4732 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1079), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][0] + 0x1079)), (data), 0));
4733 }
4734 }
4735 } else {
4736 /* 1 - MGCG_OVERRIDE */
4737 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c48, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c48), 0));
4738
4739 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1)(((9) << 16) | ((2) << 8) | (1)))
4740 data |= RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK0x00000001L;
4741
4742 data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK0x00000002L |
4743 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK0x00000020L |
4744 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK0x00000004L |
4745 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK0x00000040L);
4746
4747 if (def != data)
4748 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c48), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c48)), (data), 0));
4749
4750 /* 2 - disable MGLS in RLC */
4751 data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c06, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c06), 0));
4752 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK0x00000001L) {
4753 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK0x00000001L;
4754 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c06), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c06)), (data), 0));
4755 }
4756
4757 /* 3 - disable MGLS in CP */
4758 data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][0
] + 0x1079, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][0] + 0x1079), 0));
4759 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK0x00000001L) {
4760 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK0x00000001L;
4761 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
0] + 0x1079), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][0] + 0x1079)), (data), 0));
4762 }
4763 }
4764
4765 amdgpu_gfx_rlc_exit_safe_mode(adev);
4766}
4767
4768static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev,
4769 bool_Bool enable)
4770{
4771 uint32_t data, def;
4772
4773 if (!adev->gfx.num_gfx_rings)
4774 return;
4775
4776 amdgpu_gfx_rlc_enter_safe_mode(adev);
4777
4778 /* Enable 3D CGCG/CGLS */
4779 if (enable) {
4780 /* write cmd to clear cgcg/cgls ov */
4781 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4c48, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4c48), 0));
4782 /* unset CGCG override */
4783 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK0x00000080L;
4784 /* update CGCG and CGLS override bits */
4785 if (def != data)
4786 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_wreg(adev, (adev->reg_offset[GC_HWIP][0][
1] + 0x4c48), data, 0, GC_HWIP) : amdgpu_device_wreg(adev, ((
adev->reg_offset[GC_HWIP][0][1] + 0x4c48)), (data), 0));
4787
4788 /* enable 3Dcgcg FSM(0x0000363f) */
4789 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D)((((adev)->virt.caps & (1 << 2)) && adev
->gfx.rlc.funcs && adev->gfx.rlc.rlcg_reg_access_supported
) ? amdgpu_sriov_rreg(adev, adev->reg_offset[GC_HWIP][0][1
] + 0x4cc5, 0, GC_HWIP) : amdgpu_device_rreg(adev, (adev->
reg_offset[GC_HWIP][0][1] + 0x4cc5), 0));
4790
4791 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG(1ULL << 20))
479