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_RINGS 1 59 #define GFX9_MEC_HPD_SIZE 4096 60 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L 61 #define RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET 0x00000000L 62 63 #define mmGCEA_PROBE_MAP 0x070c 64 #define mmGCEA_PROBE_MAP_BASE_IDX 0 65 66 MODULE_FIRMWARE("amdgpu/vega10_ce.bin"); 67 MODULE_FIRMWARE("amdgpu/vega10_pfp.bin"); 68 MODULE_FIRMWARE("amdgpu/vega10_me.bin"); 69 MODULE_FIRMWARE("amdgpu/vega10_mec.bin"); 70 MODULE_FIRMWARE("amdgpu/vega10_mec2.bin"); 71 MODULE_FIRMWARE("amdgpu/vega10_rlc.bin"); 72 73 MODULE_FIRMWARE("amdgpu/vega12_ce.bin"); 74 MODULE_FIRMWARE("amdgpu/vega12_pfp.bin"); 75 MODULE_FIRMWARE("amdgpu/vega12_me.bin"); 76 MODULE_FIRMWARE("amdgpu/vega12_mec.bin"); 77 MODULE_FIRMWARE("amdgpu/vega12_mec2.bin"); 78 MODULE_FIRMWARE("amdgpu/vega12_rlc.bin"); 79 80 MODULE_FIRMWARE("amdgpu/vega20_ce.bin"); 81 MODULE_FIRMWARE("amdgpu/vega20_pfp.bin"); 82 MODULE_FIRMWARE("amdgpu/vega20_me.bin"); 83 MODULE_FIRMWARE("amdgpu/vega20_mec.bin"); 84 MODULE_FIRMWARE("amdgpu/vega20_mec2.bin"); 85 MODULE_FIRMWARE("amdgpu/vega20_rlc.bin"); 86 87 MODULE_FIRMWARE("amdgpu/raven_ce.bin"); 88 MODULE_FIRMWARE("amdgpu/raven_pfp.bin"); 89 MODULE_FIRMWARE("amdgpu/raven_me.bin"); 90 MODULE_FIRMWARE("amdgpu/raven_mec.bin"); 91 MODULE_FIRMWARE("amdgpu/raven_mec2.bin"); 92 MODULE_FIRMWARE("amdgpu/raven_rlc.bin"); 93 94 MODULE_FIRMWARE("amdgpu/picasso_ce.bin"); 95 MODULE_FIRMWARE("amdgpu/picasso_pfp.bin"); 96 MODULE_FIRMWARE("amdgpu/picasso_me.bin"); 97 MODULE_FIRMWARE("amdgpu/picasso_mec.bin"); 98 MODULE_FIRMWARE("amdgpu/picasso_mec2.bin"); 99 MODULE_FIRMWARE("amdgpu/picasso_rlc.bin"); 100 MODULE_FIRMWARE("amdgpu/picasso_rlc_am4.bin"); 101 102 MODULE_FIRMWARE("amdgpu/raven2_ce.bin"); 103 MODULE_FIRMWARE("amdgpu/raven2_pfp.bin"); 104 MODULE_FIRMWARE("amdgpu/raven2_me.bin"); 105 MODULE_FIRMWARE("amdgpu/raven2_mec.bin"); 106 MODULE_FIRMWARE("amdgpu/raven2_mec2.bin"); 107 MODULE_FIRMWARE("amdgpu/raven2_rlc.bin"); 108 MODULE_FIRMWARE("amdgpu/raven_kicker_rlc.bin"); 109 110 MODULE_FIRMWARE("amdgpu/arcturus_mec.bin"); 111 MODULE_FIRMWARE("amdgpu/arcturus_rlc.bin"); 112 113 MODULE_FIRMWARE("amdgpu/renoir_ce.bin"); 114 MODULE_FIRMWARE("amdgpu/renoir_pfp.bin"); 115 MODULE_FIRMWARE("amdgpu/renoir_me.bin"); 116 MODULE_FIRMWARE("amdgpu/renoir_mec.bin"); 117 MODULE_FIRMWARE("amdgpu/renoir_rlc.bin"); 118 119 MODULE_FIRMWARE("amdgpu/green_sardine_ce.bin"); 120 MODULE_FIRMWARE("amdgpu/green_sardine_pfp.bin"); 121 MODULE_FIRMWARE("amdgpu/green_sardine_me.bin"); 122 MODULE_FIRMWARE("amdgpu/green_sardine_mec.bin"); 123 MODULE_FIRMWARE("amdgpu/green_sardine_mec2.bin"); 124 MODULE_FIRMWARE("amdgpu/green_sardine_rlc.bin"); 125 126 MODULE_FIRMWARE("amdgpu/aldebaran_mec.bin"); 127 MODULE_FIRMWARE("amdgpu/aldebaran_mec2.bin"); 128 MODULE_FIRMWARE("amdgpu/aldebaran_rlc.bin"); 129 MODULE_FIRMWARE("amdgpu/aldebaran_sjt_mec.bin"); 130 MODULE_FIRMWARE("amdgpu/aldebaran_sjt_mec2.bin"); 131 132 #define mmTCP_CHAN_STEER_0_ARCT 0x0b03 133 #define mmTCP_CHAN_STEER_0_ARCT_BASE_IDX 0 134 #define mmTCP_CHAN_STEER_1_ARCT 0x0b04 135 #define mmTCP_CHAN_STEER_1_ARCT_BASE_IDX 0 136 #define mmTCP_CHAN_STEER_2_ARCT 0x0b09 137 #define mmTCP_CHAN_STEER_2_ARCT_BASE_IDX 0 138 #define mmTCP_CHAN_STEER_3_ARCT 0x0b0a 139 #define mmTCP_CHAN_STEER_3_ARCT_BASE_IDX 0 140 #define mmTCP_CHAN_STEER_4_ARCT 0x0b0b 141 #define mmTCP_CHAN_STEER_4_ARCT_BASE_IDX 0 142 #define mmTCP_CHAN_STEER_5_ARCT 0x0b0c 143 #define mmTCP_CHAN_STEER_5_ARCT_BASE_IDX 0 144 145 #define mmGOLDEN_TSC_COUNT_UPPER_Renoir 0x0025 146 #define mmGOLDEN_TSC_COUNT_UPPER_Renoir_BASE_IDX 1 147 #define mmGOLDEN_TSC_COUNT_LOWER_Renoir 0x0026 148 #define mmGOLDEN_TSC_COUNT_LOWER_Renoir_BASE_IDX 1 149 150 enum 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 }; 351 352 struct 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) \ 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 } 366 367 static const struct ras_gfx_subblock ras_gfx_subblocks[] = { 368 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_SCRATCH, 0, 1, 1, 1, 1, 0, 0, 1), 369 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_UCODE, 0, 1, 1, 1, 1, 0, 0, 1), 370 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME1, 1, 0, 0, 1, 0, 0, 1, 0), 371 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME1, 1, 0, 0, 1, 0, 0, 0, 0), 372 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME1, 1, 0, 0, 1, 0, 0, 0, 0), 373 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME2, 1, 0, 0, 1, 0, 0, 0, 0), 374 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME2, 1, 0, 0, 1, 0, 0, 0, 0), 375 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME2, 1, 0, 0, 1, 0, 0, 0, 0), 376 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME2, 1, 0, 0, 1, 0, 0, 0, 0), 377 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME1, 1, 0, 0, 1, 0, 0, 1, 0), 378 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_TAG, 0, 1, 1, 1, 1, 0, 0, 1), 379 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_ROQ, 1, 0, 0, 1, 0, 0, 1, 0), 380 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_TAG, 0, 1, 1, 1, 0, 1, 0, 1), 381 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_TAG, 0, 1, 1, 1, 1, 1, 0, 1), 382 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_MEM, 0, 1, 1, 1, 0, 0, 0, 0), 383 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_INPUT_QUEUE, 1, 0, 0, 1, 0, 0, 0, 0), 384 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_CMD_RAM_MEM, 0, 1, 1, 1, 0, 0, 0, 385 0), 386 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_DATA_RAM_MEM, 1, 0, 0, 1, 0, 0, 0, 387 0), 388 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PIPE_MEM, 0, 1, 1, 1, 0, 0, 0, 0), 389 AMDGPU_RAS_SUB_BLOCK(GFX_SPI_SR_MEM, 1, 0, 0, 1, 0, 0, 0, 0), 390 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_SGPR, 0, 1, 1, 1, 0, 0, 0, 0), 391 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_D, 0, 1, 1, 1, 1, 0, 0, 1), 392 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_I, 0, 1, 1, 1, 0, 0, 0, 0), 393 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_VGPR, 0, 1, 1, 1, 0, 0, 0, 0), 394 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 1), 395 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0, 396 0, 0), 397 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 398 0), 399 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0, 400 0, 0), 401 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_UTCL1_LFIFO, 0, 1, 1, 1, 1, 0, 0, 402 0), 403 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0, 404 0, 0), 405 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 406 0), 407 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_TAG_RAM, 0, 1, 1, 1, 1, 0, 0, 408 1), 409 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0, 410 0, 0, 0), 411 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0, 412 0), 413 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0, 414 0), 415 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_TAG_RAM, 0, 1, 1, 1, 0, 0, 0, 416 0), 417 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0, 418 0), 419 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0, 420 0), 421 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0, 422 0, 0), 423 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0, 424 0), 425 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_TAG_RAM, 0, 1, 1, 1, 1, 0, 0, 426 0), 427 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0, 428 0, 0, 0), 429 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0, 430 0), 431 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0, 432 0), 433 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_TAG_RAM, 0, 1, 1, 1, 0, 0, 0, 434 0), 435 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0, 436 0), 437 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0, 438 0), 439 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0, 440 0, 0), 441 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0, 442 0), 443 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_DFIFO, 0, 1, 1, 1, 1, 0, 0, 1), 444 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_AFIFO, 1, 0, 0, 1, 0, 0, 0, 0), 445 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FL_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0), 446 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FX_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0), 447 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_CFIFO, 1, 0, 0, 1, 0, 0, 0, 0), 448 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_HOLE_FIFO, 1, 0, 0, 1, 0, 1, 1, 0), 449 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_REQ_FIFO, 1, 0, 0, 1, 0, 0, 0, 0), 450 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA, 0, 1, 1, 1, 1, 0, 0, 1), 451 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_0_1, 0, 1, 1, 1, 1, 0, 0, 452 1), 453 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_0, 0, 1, 1, 1, 1, 0, 0, 454 1), 455 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_1, 0, 1, 1, 1, 1, 0, 0, 456 1), 457 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_0, 0, 1, 1, 1, 0, 0, 0, 458 0), 459 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_1, 0, 1, 1, 1, 0, 0, 0, 460 0), 461 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_HIGH_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0), 462 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LOW_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0), 463 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_DEC, 1, 0, 0, 1, 0, 0, 0, 0), 464 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_TRANSFER, 1, 0, 0, 1, 0, 0, 0, 0), 465 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_DATA, 1, 0, 0, 1, 0, 0, 0, 0), 466 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_CONTROL, 1, 0, 0, 1, 0, 0, 0, 0), 467 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_UC_ATOMIC_FIFO, 1, 0, 0, 1, 0, 0, 0, 0), 468 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_RETURN, 1, 0, 0, 1, 0, 1, 1, 0), 469 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_CACHE_READ, 1, 0, 0, 1, 0, 0, 0, 0), 470 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO, 0, 1, 1, 1, 0, 0, 0, 0), 471 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 1, 0), 472 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_TAG_PROBE_FIFO, 1, 0, 0, 1, 0, 0, 0, 473 0), 474 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO, 1, 0, 0, 1, 0, 0, 0, 0), 475 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 0, 476 0), 477 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRRET_TAG_WRITE_RETURN, 1, 0, 0, 1, 0, 0, 478 0, 0), 479 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_ATOMIC_RETURN_BUFFER, 1, 0, 0, 1, 0, 0, 0, 480 0), 481 AMDGPU_RAS_SUB_BLOCK(GFX_TCI_WRITE_RAM, 1, 0, 0, 1, 0, 0, 0, 0), 482 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CACHE_RAM, 0, 1, 1, 1, 1, 0, 0, 1), 483 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_LFIFO_RAM, 0, 1, 1, 1, 0, 0, 0, 0), 484 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CMD_FIFO, 1, 0, 0, 1, 0, 0, 0, 0), 485 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_VM_FIFO, 0, 1, 1, 1, 0, 0, 0, 0), 486 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_DB_RAM, 1, 0, 0, 1, 0, 0, 0, 0), 487 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO0, 0, 1, 1, 1, 0, 0, 0, 0), 488 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO1, 0, 1, 1, 1, 0, 0, 0, 0), 489 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_LO, 0, 1, 1, 1, 1, 0, 0, 1), 490 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_HI, 0, 1, 1, 1, 0, 0, 0, 0), 491 AMDGPU_RAS_SUB_BLOCK(GFX_TD_CS_FIFO, 1, 0, 0, 1, 0, 0, 0, 0), 492 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_CMDMEM, 0, 1, 1, 1, 1, 0, 0, 1), 493 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0), 494 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0), 495 AMDGPU_RAS_SUB_BLOCK(GFX_EA_RRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0), 496 AMDGPU_RAS_SUB_BLOCK(GFX_EA_WRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0), 497 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0), 498 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0), 499 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0), 500 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0), 501 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0), 502 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IORD_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0), 503 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0), 504 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_DATAMEM, 1, 0, 0, 1, 0, 0, 0, 0), 505 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0), 506 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0), 507 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D0MEM, 1, 0, 0, 1, 0, 0, 0, 0), 508 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D1MEM, 1, 0, 0, 1, 0, 0, 0, 0), 509 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D2MEM, 1, 0, 0, 1, 0, 0, 0, 0), 510 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D3MEM, 1, 0, 0, 1, 0, 0, 0, 0), 511 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_BANK_CACHE, 0, 1, 1, 1, 0, 0, 0, 0), 512 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_WALKER, 0, 1, 1, 1, 0, 0, 0, 0), 513 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_2M_BANK, 1, 0, 0, 1, 0, 0, 0, 0), 514 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_4K_BANK, 0, 1, 1, 1, 0, 0, 0, 0), 515 }; 516 517 static const struct soc15_reg_golden golden_settings_gc_9_0[] = 518 { 519 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400), 520 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG3, 0x80000000, 0x80000000), 521 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000), 522 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024), 523 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001), 524 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000), 525 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000), 526 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800), 527 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800), 528 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x00ffff87), 529 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x00ffff8f), 530 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000), 531 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000), 532 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68), 533 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197), 534 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000), 535 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff), 536 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800), 537 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800), 538 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000) 539 }; 540 541 static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] = 542 { 543 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0x0000f000, 0x00012107), 544 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000), 545 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080), 546 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080), 547 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080), 548 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042), 549 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x2a114042), 550 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080), 551 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0x00008000, 0x00048000), 552 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080), 553 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080), 554 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080), 555 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080), 556 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080), 557 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000), 558 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x01000107), 559 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x00001800, 0x00000800), 560 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080) 561 }; 562 563 static const struct soc15_reg_golden golden_settings_gc_9_0_vg20[] = 564 { 565 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x0f000080, 0x04000080), 566 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000), 567 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000), 568 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x22014042), 569 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x22014042), 570 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0x00003e00, 0x00000400), 571 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xff840000, 0x04040000), 572 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00030000), 573 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff010f, 0x01000107), 574 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0x000b0000, 0x000b0000), 575 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01000000, 0x01000000) 576 }; 577 578 static const struct soc15_reg_golden golden_settings_gc_9_1[] = 579 { 580 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104), 581 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080), 582 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080), 583 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080), 584 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420), 585 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000), 586 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080), 587 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024), 588 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001), 589 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000), 590 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080), 591 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080), 592 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080), 593 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080), 594 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080), 595 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000), 596 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000), 597 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120), 598 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000), 599 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000000ff), 600 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080), 601 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800), 602 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800), 603 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000) 604 }; 605 606 static 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), 609 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24000042), 610 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24000042), 611 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04048000), 612 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_MODE_CNTL_1, 0x06000000, 0x06000000), 613 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000), 614 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x00000800) 615 }; 616 617 static const struct soc15_reg_golden golden_settings_gc_9_1_rv2[] = 618 { 619 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0xff7fffff, 0x04000000), 620 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104), 621 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000), 622 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x7f0fffff, 0x08000080), 623 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0xff8fffff, 0x08000080), 624 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x7f8fffff, 0x08000080), 625 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x26013041), 626 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x26013041), 627 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x3f8fffff, 0x08000080), 628 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000), 629 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0xff0fffff, 0x08000080), 630 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0xff0fffff, 0x08000080), 631 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0xff0fffff, 0x08000080), 632 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0xff0fffff, 0x08000080), 633 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0xff0fffff, 0x08000080), 634 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000), 635 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010), 636 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000), 637 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x3f8fffff, 0x08000080), 638 }; 639 640 static const struct soc15_reg_golden golden_settings_gc_9_1_rn[] = 641 { 642 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104), 643 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000), 644 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400), 645 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x24000042), 646 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x24000042), 647 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001), 648 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000), 649 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000), 650 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000), 651 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000), 652 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120), 653 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGCEA_PROBE_MAP, 0xffffffff, 0x0000cccc), 654 }; 655 656 static const struct soc15_reg_golden golden_settings_gc_9_x_common[] = 657 { 658 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0xffffffff, 0x000001ff), 659 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_INDEX, 0xffffffff, 0x00000000), 660 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_DATA, 0xffffffff, 0x2544c382) 661 }; 662 663 static const struct soc15_reg_golden golden_settings_gc_9_2_1[] = 664 { 665 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420), 666 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000), 667 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024), 668 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001), 669 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000), 670 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000), 671 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800), 672 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800), 673 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87), 674 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f), 675 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000), 676 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000), 677 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68), 678 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197), 679 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000), 680 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff) 681 }; 682 683 static 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), 686 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104), 687 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000), 688 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24104041), 689 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24104041), 690 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000), 691 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff03ff, 0x01000107), 692 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000), 693 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x76325410), 694 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000), 695 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800), 696 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800), 697 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000) 698 }; 699 700 static 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), 703 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x10b0000), 704 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_0_ARCT, 0x3fffffff, 0x346f0a4e), 705 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_1_ARCT, 0x3fffffff, 0x1c642ca), 706 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_2_ARCT, 0x3fffffff, 0x26f45098), 707 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_3_ARCT, 0x3fffffff, 0x2ebd9fe3), 708 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_4_ARCT, 0x3fffffff, 0xb90f5b1), 709 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_5_ARCT, 0x3ff, 0x135), 710 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_CONFIG, 0xffffffff, 0x011A0000), 711 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_FIFO_SIZES, 0xffffffff, 0x00000f00), 712 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_UTCL1_CNTL1, 0x30000000, 0x30000000) 713 }; 714 715 static const struct soc15_reg_rlcg rlcg_access_gc_9_0[] = { 716 {SOC15_REG_ENTRY(GC, 0, mmGRBM_GFX_INDEX)}, 717 {SOC15_REG_ENTRY(GC, 0, mmSQ_IND_INDEX)}, 718 }; 719 720 static const u32 GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[] = 721 { 722 mmRLC_SRM_INDEX_CNTL_ADDR_0 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 723 mmRLC_SRM_INDEX_CNTL_ADDR_1 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 724 mmRLC_SRM_INDEX_CNTL_ADDR_2 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 725 mmRLC_SRM_INDEX_CNTL_ADDR_3 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 726 mmRLC_SRM_INDEX_CNTL_ADDR_4 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 727 mmRLC_SRM_INDEX_CNTL_ADDR_5 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 728 mmRLC_SRM_INDEX_CNTL_ADDR_6 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 729 mmRLC_SRM_INDEX_CNTL_ADDR_7 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 730 }; 731 732 static const u32 GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[] = 733 { 734 mmRLC_SRM_INDEX_CNTL_DATA_0 - mmRLC_SRM_INDEX_CNTL_DATA_0, 735 mmRLC_SRM_INDEX_CNTL_DATA_1 - mmRLC_SRM_INDEX_CNTL_DATA_0, 736 mmRLC_SRM_INDEX_CNTL_DATA_2 - mmRLC_SRM_INDEX_CNTL_DATA_0, 737 mmRLC_SRM_INDEX_CNTL_DATA_3 - mmRLC_SRM_INDEX_CNTL_DATA_0, 738 mmRLC_SRM_INDEX_CNTL_DATA_4 - mmRLC_SRM_INDEX_CNTL_DATA_0, 739 mmRLC_SRM_INDEX_CNTL_DATA_5 - mmRLC_SRM_INDEX_CNTL_DATA_0, 740 mmRLC_SRM_INDEX_CNTL_DATA_6 - mmRLC_SRM_INDEX_CNTL_DATA_0, 741 mmRLC_SRM_INDEX_CNTL_DATA_7 - mmRLC_SRM_INDEX_CNTL_DATA_0, 742 }; 743 744 #define VEGA10_GB_ADDR_CONFIG_GOLDEN 0x2a114042 745 #define VEGA12_GB_ADDR_CONFIG_GOLDEN 0x24104041 746 #define RAVEN_GB_ADDR_CONFIG_GOLDEN 0x24000042 747 #define RAVEN2_GB_ADDR_CONFIG_GOLDEN 0x26013041 748 749 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev); 750 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev); 751 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev); 752 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev); 753 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev, 754 struct amdgpu_cu_info *cu_info); 755 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev); 756 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring); 757 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring); 758 static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev, 759 void *ras_error_status); 760 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev, 761 void *inject_if); 762 static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev); 763 764 static 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)); 768 amdgpu_ring_write(kiq_ring, 769 PACKET3_SET_RESOURCES_VMID_MASK(0) | 770 /* vmid_mask:0* queue_type:0 (KIQ) */ 771 PACKET3_SET_RESOURCES_QUEUE_TYPE(0)); 772 amdgpu_ring_write(kiq_ring, 773 lower_32_bits(queue_mask)); /* queue mask lo */ 774 amdgpu_ring_write(kiq_ring, 775 upper_32_bits(queue_mask)); /* 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 } 781 782 static 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)); 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) | /* Queue_Sel */ 793 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */ 794 PACKET3_MAP_QUEUES_QUEUE(ring->queue) | 795 PACKET3_MAP_QUEUES_PIPE(ring->pipe) | 796 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) | 797 /*queue_type: normal compute queue */ 798 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | 799 /* alloc format: all_on_one_pipe */ 800 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | 801 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) | 802 /* num_queues: must be 1 */ 803 PACKET3_MAP_QUEUES_NUM_QUEUES(1)); 804 amdgpu_ring_write(kiq_ring, 805 PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index)); 806 amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr)); 807 amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr)); 808 amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr)); 809 amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr)); 810 } 811 812 static 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)); 820 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */ 821 PACKET3_UNMAP_QUEUES_ACTION(action) | 822 PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) | 823 PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) | 824 PACKET3_UNMAP_QUEUES_NUM_QUEUES(1)); 825 amdgpu_ring_write(kiq_ring, 826 PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index)); 827 828 if (action == PREEMPT_QUEUES_NO_UNMAP) { 829 amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr)); 830 amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr)); 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 } 838 839 static 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)); 847 amdgpu_ring_write(kiq_ring, 848 PACKET3_QUERY_STATUS_CONTEXT_ID(0) | 849 PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) | 850 PACKET3_QUERY_STATUS_COMMAND(2)); 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) | 854 PACKET3_QUERY_STATUS_ENG_SEL(eng_sel)); 855 amdgpu_ring_write(kiq_ring, lower_32_bits(addr)); 856 amdgpu_ring_write(kiq_ring, upper_32_bits(addr)); 857 amdgpu_ring_write(kiq_ring, lower_32_bits(seq)); 858 amdgpu_ring_write(kiq_ring, upper_32_bits(seq)); 859 } 860 861 static void gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring, 862 uint16_t pasid, uint32_t flush_type, 863 bool all_hub) 864 { 865 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0)); 866 amdgpu_ring_write(kiq_ring, 867 PACKET3_INVALIDATE_TLBS_DST_SEL(1) | 868 PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) | 869 PACKET3_INVALIDATE_TLBS_PASID(pasid) | 870 PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type)); 871 } 872 873 static 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 }; 885 886 static 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 } 890 891 static 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): 895 soc15_program_register_sequence(adev, 896 golden_settings_gc_9_0, 897 ARRAY_SIZE(golden_settings_gc_9_0)); 898 soc15_program_register_sequence(adev, 899 golden_settings_gc_9_0_vg10, 900 ARRAY_SIZE(golden_settings_gc_9_0_vg10)); 901 break; 902 case IP_VERSION(9, 2, 1): 903 soc15_program_register_sequence(adev, 904 golden_settings_gc_9_2_1, 905 ARRAY_SIZE(golden_settings_gc_9_2_1)); 906 soc15_program_register_sequence(adev, 907 golden_settings_gc_9_2_1_vg12, 908 ARRAY_SIZE(golden_settings_gc_9_2_1_vg12)); 909 break; 910 case IP_VERSION(9, 4, 0): 911 soc15_program_register_sequence(adev, 912 golden_settings_gc_9_0, 913 ARRAY_SIZE(golden_settings_gc_9_0)); 914 soc15_program_register_sequence(adev, 915 golden_settings_gc_9_0_vg20, 916 ARRAY_SIZE(golden_settings_gc_9_0_vg20)); 917 break; 918 case IP_VERSION(9, 4, 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)); 922 break; 923 case IP_VERSION(9, 2, 2): 924 case IP_VERSION(9, 1, 0): 925 soc15_program_register_sequence(adev, golden_settings_gc_9_1, 926 ARRAY_SIZE(golden_settings_gc_9_1)); 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)); 931 else 932 soc15_program_register_sequence(adev, 933 golden_settings_gc_9_1_rv1, 934 ARRAY_SIZE(golden_settings_gc_9_1_rv1)); 935 break; 936 case IP_VERSION(9, 3, 0): 937 soc15_program_register_sequence(adev, 938 golden_settings_gc_9_1_rn, 939 ARRAY_SIZE(golden_settings_gc_9_1_rn)); 940 return; /* for renoir, don't need common goldensetting */ 941 case IP_VERSION(9, 4, 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)) && 950 (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 2))) 951 soc15_program_register_sequence(adev, golden_settings_gc_9_x_common, 952 (const u32)ARRAY_SIZE(golden_settings_gc_9_x_common)); 953 } 954 955 static void gfx_v9_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel, 956 bool wc, uint32_t reg, uint32_t val) 957 { 958 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 959 amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) | 960 WRITE_DATA_DST_SEL(0) | 961 (wc ? WR_CONFIRM : 0)); 962 amdgpu_ring_write(ring, reg); 963 amdgpu_ring_write(ring, 0); 964 amdgpu_ring_write(ring, val); 965 } 966 967 static 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)); 973 amdgpu_ring_write(ring, 974 /* memory (1) or register (0) */ 975 (WAIT_REG_MEM_MEM_SPACE(mem_space) | 976 WAIT_REG_MEM_OPERATION(opt) | /* wait */ 977 WAIT_REG_MEM_FUNCTION(3) | /* equal */ 978 WAIT_REG_MEM_ENGINE(eng_sel))); 979 980 if (mem_space) 981 BUG_ON(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 } 988 989 static 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); 993 uint32_t tmp = 0; 994 unsigned i; 995 int r; 996 997 WREG32(scratch, 0xCAFEDEAD); 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)); 1003 amdgpu_ring_write(ring, scratch - PACKET3_SET_UCONFIG_REG_START); 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); 1009 if (tmp == 0xDEADBEEF) 1010 break; 1011 udelay(1); 1012 } 1013 1014 if (i >= adev->usec_timeout) 1015 r = -ETIMEDOUT; 1016 return r; 1017 } 1018 1019 static 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; 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); 1036 memset(&ib, 0, sizeof(ib)); 1037 r = amdgpu_ib_get(adev, NULL, 16, 1038 AMDGPU_IB_POOL_DIRECT, &ib); 1039 if (r) 1040 goto err1; 1041 1042 ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3); 1043 ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM; 1044 ib.ptr[2] = lower_32_bits(gpu_addr); 1045 ib.ptr[3] = upper_32_bits(gpu_addr); 1046 ib.ptr[4] = 0xDEADBEEF; 1047 ib.length_dw = 5; 1048 1049 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); 1050 if (r) 1051 goto err2; 1052 1053 r = dma_fence_wait_timeout(f, false, timeout); 1054 if (r == 0) { 1055 r = -ETIMEDOUT; 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 = -EINVAL; 1066 1067 err2: 1068 amdgpu_ib_free(adev, &ib, NULL); 1069 dma_fence_put(f); 1070 err1: 1071 amdgpu_device_wb_free(adev, index); 1072 return r; 1073 } 1074 1075 1076 static void gfx_v9_0_free_microcode(struct amdgpu_device *adev) 1077 { 1078 release_firmware(adev->gfx.pfp_fw); 1079 adev->gfx.pfp_fw = NULL; 1080 release_firmware(adev->gfx.me_fw); 1081 adev->gfx.me_fw = NULL; 1082 release_firmware(adev->gfx.ce_fw); 1083 adev->gfx.ce_fw = NULL; 1084 release_firmware(adev->gfx.rlc_fw); 1085 adev->gfx.rlc_fw = NULL; 1086 release_firmware(adev->gfx.mec_fw); 1087 adev->gfx.mec_fw = NULL; 1088 release_firmware(adev->gfx.mec2_fw); 1089 adev->gfx.mec2_fw = NULL; 1090 1091 kfree(adev->gfx.rlc.register_list_format); 1092 } 1093 1094 static void gfx_v9_0_check_fw_write_wait(struct amdgpu_device *adev) 1095 { 1096 adev->gfx.me_fw_write_wait = false; 1097 adev->gfx.mec_fw_write_wait = false; 1098 1099 if ((adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 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!"); 1105 1106 switch (adev->ip_versions[GC_HWIP][0]) { 1107 case IP_VERSION(9, 0, 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 = true; 1113 1114 if ((adev->gfx.mec_fw_version >= 0x00000193) && 1115 (adev->gfx.mec_feature_version >= 42)) 1116 adev->gfx.mec_fw_write_wait = true; 1117 break; 1118 case IP_VERSION(9, 2, 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 = true; 1124 1125 if ((adev->gfx.mec_fw_version >= 0x00000196) && 1126 (adev->gfx.mec_feature_version >= 44)) 1127 adev->gfx.mec_fw_write_wait = true; 1128 break; 1129 case IP_VERSION(9, 4, 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 = true; 1135 1136 if ((adev->gfx.mec_fw_version >= 0x00000197) && 1137 (adev->gfx.mec_feature_version >= 44)) 1138 adev->gfx.mec_fw_write_wait = true; 1139 break; 1140 case IP_VERSION(9, 1, 0): 1141 case IP_VERSION(9, 2, 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 = true; 1147 1148 if ((adev->gfx.mec_fw_version >= 0x00000192) && 1149 (adev->gfx.mec_feature_version >= 42)) 1150 adev->gfx.mec_fw_write_wait = true; 1151 break; 1152 default: 1153 adev->gfx.me_fw_write_wait = true; 1154 adev->gfx.mec_fw_write_wait = true; 1155 break; 1156 } 1157 } 1158 1159 struct amdgpu_gfxoff_quirk { 1160 u16 chip_vendor; 1161 u16 chip_device; 1162 u16 subsys_vendor; 1163 u16 subsys_device; 1164 u8 revision; 1165 }; 1166 1167 static 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 1179 static 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 true; 1190 } 1191 ++p; 1192 } 1193 return false; 1194 } 1195 1196 static bool is_raven_kicker(struct amdgpu_device *adev) 1197 { 1198 if (adev->pm.fw_version >= 0x41e2b) 1199 return true; 1200 else 1201 return false; 1202 } 1203 1204 static bool check_if_enlarge_doorbell_range(struct amdgpu_device *adev) 1205 { 1206 if ((adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 3, 0)) && 1207 (adev->gfx.me_fw_version >= 0x000000a5) && 1208 (adev->gfx.me_feature_version >= 52)) 1209 return true; 1210 else 1211 return false; 1212 } 1213 1214 static 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): 1221 case IP_VERSION(9, 2, 1): 1222 case IP_VERSION(9, 4, 0): 1223 break; 1224 case IP_VERSION(9, 2, 2): 1225 case IP_VERSION(9, 1, 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 | 1236 AMD_PG_SUPPORT_CP | 1237 AMD_PG_SUPPORT_RLC_SMU_HS; 1238 break; 1239 case IP_VERSION(9, 3, 0): 1240 if (adev->pm.pp_feature & PP_GFXOFF_MASK) 1241 adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG | 1242 AMD_PG_SUPPORT_CP | 1243 AMD_PG_SUPPORT_RLC_SMU_HS; 1244 break; 1245 default: 1246 break; 1247 } 1248 } 1249 1250 static 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 1283 out: 1284 if (err) { 1285 dev_err(adev->dev, 1286 "gfx9: Failed to init firmware \"%s\"\n", 1287 fw_name); 1288 release_firmware(adev->gfx.pfp_fw); 1289 adev->gfx.pfp_fw = NULL; 1290 release_firmware(adev->gfx.me_fw); 1291 adev->gfx.me_fw = NULL; 1292 release_firmware(adev->gfx.ce_fw); 1293 adev->gfx.ce_fw = NULL; 1294 } 1295 return err; 1296 } 1297 1298 static 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); 1337 version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor); 1338 err = amdgpu_gfx_rlc_init_microcode(adev, version_major, version_minor); 1339 out: 1340 if (err) { 1341 dev_err(adev->dev, 1342 "gfx9: Failed to init firmware \"%s\"\n", 1343 fw_name); 1344 release_firmware(adev->gfx.rlc_fw); 1345 adev->gfx.rlc_fw = NULL; 1346 } 1347 return err; 1348 } 1349 1350 static 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) || 1353 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1) || 1354 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 3, 0)) 1355 return false; 1356 1357 return true; 1358 } 1359 1360 static 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->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->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; 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 1402 out: 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, 1407 "gfx9: Failed to init firmware \"%s\"\n", 1408 fw_name); 1409 release_firmware(adev->gfx.mec_fw); 1410 adev->gfx.mec_fw = NULL; 1411 release_firmware(adev->gfx.mec2_fw); 1412 adev->gfx.mec2_fw = NULL; 1413 } 1414 return err; 1415 } 1416 1417 static int gfx_v9_0_init_microcode(struct amdgpu_device *adev) 1418 { 1419 const char *chip_name; 1420 int r; 1421 1422 DRM_DEBUG("\n"); 1423 1424 switch (adev->ip_versions[GC_HWIP][0]) { 1425 case IP_VERSION(9, 0, 1): 1426 chip_name = "vega10"; 1427 break; 1428 case IP_VERSION(9, 2, 1): 1429 chip_name = "vega12"; 1430 break; 1431 case IP_VERSION(9, 4, 0): 1432 chip_name = "vega20"; 1433 break; 1434 case IP_VERSION(9, 2, 2): 1435 case IP_VERSION(9, 1, 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): 1444 chip_name = "arcturus"; 1445 break; 1446 case IP_VERSION(9, 3, 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): 1453 chip_name = "aldebaran"; 1454 break; 1455 default: 1456 BUG(); 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 1477 static u32 gfx_v9_0_get_csb_size(struct amdgpu_device *adev) 1478 { 1479 u32 count = 0; 1480 const struct cs_section_def *sect = NULL; 1481 const struct cs_extent_def *ext = NULL; 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; ++sect) { 1489 for (ext = sect->section; ext->extent != NULL; ++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 1505 static 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; 1510 const struct cs_extent_def *ext = NULL; 1511 1512 if (adev->gfx.rlc.cs_data == NULL) 1513 return; 1514 if (buffer == NULL) 1515 return; 1516 1517 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 1518 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 1519 1520 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 1521 buffer[count++] = cpu_to_le32(0x80000000); 1522 buffer[count++] = cpu_to_le32(0x80000000); 1523 1524 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 1525 for (ext = sect->section; ext->extent != NULL; ++ext) { 1526 if (sect->id == SECT_CONTEXT) { 1527 buffer[count++] = 1528 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count)); 1529 buffer[count++] = cpu_to_le32(ext->reg_index - 1530 PACKET3_SET_CONTEXT_REG_START); 1531 for (i = 0; i < ext->reg_count; i++) 1532 buffer[count++] = cpu_to_le32(ext->extent[i]); 1533 } else { 1534 return; 1535 } 1536 } 1537 } 1538 1539 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 1540 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE); 1541 1542 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0)); 1543 buffer[count++] = cpu_to_le32(0); 1544 } 1545 1546 static 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)) 1557 always_on_cu_num = 8; 1558 else 1559 always_on_cu_num = 12; 1560 1561 mutex_lock(&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); 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); 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); 1588 } 1589 1590 static 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); 1596 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x0333A5A7); 1597 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077); 1598 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x30 | 0x40 << 8 | 0x02FA << 16)); 1599 1600 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */ 1601 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000); 1602 1603 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */ 1604 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000500); 1605 1606 mutex_lock(&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); 1610 1611 /* set mmRLC_LB_PARAMS = 0x003F_1006 */ 1612 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003); 1613 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010); 1614 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F); 1615 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data); 1616 1617 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */ 1618 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7); 1619 data &= 0x0000FFFF; 1620 data |= 0x00C00000; 1621 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data); 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_MASK; 1631 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09); 1632 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000); 1633 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data); 1634 mutex_unlock(&adev->grbm_idx_mutex); 1635 1636 gfx_v9_0_init_always_on_cu_mask(adev); 1637 } 1638 1639 static 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); 1645 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x033388F8); 1646 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077); 1647 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x10 | 0x27 << 8 | 0x02FA << 16)); 1648 1649 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */ 1650 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000); 1651 1652 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */ 1653 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000800); 1654 1655 mutex_lock(&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); 1659 1660 /* set mmRLC_LB_PARAMS = 0x003F_1006 */ 1661 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003); 1662 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010); 1663 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F); 1664 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data); 1665 1666 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */ 1667 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7); 1668 data &= 0x0000FFFF; 1669 data |= 0x00C00000; 1670 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data); 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_MASK; 1680 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09); 1681 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000); 1682 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data); 1683 mutex_unlock(&adev->grbm_idx_mutex); 1684 1685 gfx_v9_0_init_always_on_cu_mask(adev); 1686 } 1687 1688 static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool enable) 1689 { 1690 WREG32_FIELD15(GC, 0, RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0); 1691 } 1692 1693 static 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 1701 static 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); 1707 reg_access_ctrl->scratch_reg1 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG1); 1708 reg_access_ctrl->scratch_reg2 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG2); 1709 reg_access_ctrl->scratch_reg3 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG3); 1710 reg_access_ctrl->grbm_cntl = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_CNTL); 1711 reg_access_ctrl->grbm_idx = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_INDEX); 1712 reg_access_ctrl->spare_int = SOC15_REG_OFFSET(GC, 0, mmRLC_SPARE_INT); 1713 adev->gfx.rlc.rlcg_reg_access_supported = true; 1714 } 1715 1716 static 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) + (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): 1742 case IP_VERSION(9, 1, 0): 1743 gfx_v9_0_init_lbpw(adev); 1744 break; 1745 case IP_VERSION(9, 4, 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 1759 static void gfx_v9_0_mec_fini(struct amdgpu_device *adev) 1760 { 1761 amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL); 1762 amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL); 1763 } 1764 1765 static 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_QUEUES); 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_SIZE; 1781 if (mec_hpd_size) { 1782 r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE, 1783 AMDGPU_GEM_DOMAIN_VRAM, 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); 1789 gfx_v9_0_mec_fini(adev); 1790 return r; 1791 } 1792 1793 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)); 1804 fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes); 1805 1806 r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes, 1807 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT, 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); 1813 gfx_v9_0_mec_fini(adev); 1814 return r; 1815 } 1816 1817 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 1825 static 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, 1828 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 1829 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 1830 (address << SQ_IND_INDEX__INDEX__SHIFT) | 1831 (SQ_IND_INDEX__FORCE_READ_MASK)); 1832 return RREG32_SOC15(GC, 0, mmSQ_IND_DATA); 1833 } 1834 1835 static 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, 1840 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 1841 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 1842 (regno << SQ_IND_INDEX__INDEX__SHIFT) | 1843 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) | 1844 (SQ_IND_INDEX__FORCE_READ_MASK) | 1845 (SQ_IND_INDEX__AUTO_INCR_MASK)); 1846 while (num--) 1847 *(out++) = RREG32_SOC15(GC, 0, mmSQ_IND_DATA); 1848 } 1849 1850 static 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_STATUS); 1855 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO); 1856 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI); 1857 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO); 1858 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI); 1859 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID); 1860 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0); 1861 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1); 1862 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC); 1863 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC); 1864 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS); 1865 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS); 1866 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0); 1867 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0); 1868 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE); 1869 } 1870 1871 static 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_OFFSET, size, dst); 1878 } 1879 1880 static 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_OFFSET, size, dst); 1888 } 1889 1890 static 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 1896 static 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 1905 const 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 1911 static struct amdgpu_gfx_ras gfx_v9_0_ras = { 1912 .ras_block = { 1913 .hw_ops = &gfx_v9_0_ras_ops, 1914 }, 1915 }; 1916 1917 static 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): 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_GOLDEN; 1932 break; 1933 case IP_VERSION(9, 2, 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_GOLDEN; 1940 DRM_INFO("fix gfx.config for vega12\n"); 1941 break; 1942 case IP_VERSION(9, 4, 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); 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): 1958 case IP_VERSION(9, 1, 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_GOLDEN; 1966 else 1967 gb_addr_config = RAVEN_GB_ADDR_CONFIG_GOLDEN; 1968 break; 1969 case IP_VERSION(9, 4, 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); 1977 gb_addr_config &= ~0xf3e777ff; 1978 gb_addr_config |= 0x22014042; 1979 break; 1980 case IP_VERSION(9, 3, 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); 1987 gb_addr_config &= ~0xf3e777ff; 1988 gb_addr_config |= 0x22010042; 1989 break; 1990 case IP_VERSION(9, 4, 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); 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(); 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"); 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( 2036 adev->gfx.config.gb_addr_config, 2037 GB_ADDR_CONFIG, 2038 NUM_PIPES); 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( 2045 adev->gfx.config.gb_addr_config, 2046 GB_ADDR_CONFIG, 2047 NUM_BANKS); 2048 adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 << 2049 REG_GET_FIELD( 2050 adev->gfx.config.gb_addr_config, 2051 GB_ADDR_CONFIG, 2052 MAX_COMPRESSED_FRAGS); 2053 adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 << 2054 REG_GET_FIELD( 2055 adev->gfx.config.gb_addr_config, 2056 GB_ADDR_CONFIG, 2057 NUM_RB_PER_SE); 2058 adev->gfx.config.gb_addr_config_fields.num_se = 1 << 2059 REG_GET_FIELD( 2060 adev->gfx.config.gb_addr_config, 2061 GB_ADDR_CONFIG, 2062 NUM_SHADER_ENGINES); 2063 adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 + 2064 REG_GET_FIELD( 2065 adev->gfx.config.gb_addr_config, 2066 GB_ADDR_CONFIG, 2067 PIPE_INTERLEAVE_SIZE)); 2068 2069 return 0; 2070 } 2071 2072 static 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]; 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; 2087 ring->use_doorbell = true; 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_SIZE); 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); 2101 } 2102 2103 static 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): 2112 case IP_VERSION(9, 2, 1): 2113 case IP_VERSION(9, 4, 0): 2114 case IP_VERSION(9, 2, 2): 2115 case IP_VERSION(9, 1, 0): 2116 case IP_VERSION(9, 4, 1): 2117 case IP_VERSION(9, 3, 0): 2118 case IP_VERSION(9, 4, 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_INTERRUPT, &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_FAULT, 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_FAULT, 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_ERROR, 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_ERROR, 2154 &adev->gfx.cp_ecc_error_irq); 2155 if (r) 2156 return r; 2157 2158 adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE; 2159 2160 r = gfx_v9_0_init_microcode(adev); 2161 if (r) { 2162 DRM_ERROR("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"); 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"); 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; 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 = true; 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); 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_SIZE); 2219 if (r) { 2220 DRM_ERROR("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 2244 static 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 2273 static void gfx_v9_0_tiling_mode_table_init(struct amdgpu_device *adev) 2274 { 2275 /* TODO */ 2276 } 2277 2278 void 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); 2285 else 2286 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance); 2287 2288 if (se_num == 0xffffffff) 2289 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1); 2290 else 2291 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num); 2292 2293 if (sh_num == 0xffffffff) 2294 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1); 2295 else 2296 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num); 2297 2298 WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_INDEX, data); 2299 } 2300 2301 static 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); 2306 data |= RREG32_SOC15(GC, 0, mmGC_USER_RB_BACKEND_DISABLE); 2307 2308 data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK; 2309 data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT; 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 2317 static 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); 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); 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) 2342 static 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 | (DEFAULT_SH_MEM_BASES << 16); 2355 2356 sh_mem_config = SH_MEM_ADDRESS_MODE_64 | 2357 SH_MEM_ALIGNMENT_MODE_UNALIGNED << 2358 SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT; 2359 2360 mutex_lock(&adev->srbm_mutex); 2361 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; 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); 2365 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, sh_mem_bases); 2366 } 2367 soc15_grbm_select(adev, 0, 0, 0, 0); 2368 mutex_unlock(&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_VMID; i++) { 2373 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * i, 0); 2374 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * i, 0); 2375 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, i, 0); 2376 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, i, 0); 2377 } 2378 } 2379 2380 static 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_VMID; vmid++) { 2391 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * vmid, 0); 2392 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * vmid, 0); 2393 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, vmid, 0); 2394 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, vmid, 0); 2395 } 2396 } 2397 2398 static 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): 2404 tmp = RREG32_SOC15(GC, 0, mmSQ_CONFIG); 2405 tmp = REG_SET_FIELD(tmp, SQ_CONFIG, 2406 DISABLE_BARRIER_WAITCNT, 1); 2407 WREG32_SOC15(GC, 0, mmSQ_CONFIG, tmp); 2408 break; 2409 default: 2410 break; 2411 } 2412 } 2413 2414 static 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); 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); 2427 2428 /* XXX SH_MEM regs */ 2429 /* where to put LDS, scratch, GPUVM in FSA64 space */ 2430 mutex_lock(&adev->srbm_mutex); 2431 for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_0].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, 2436 SH_MEM_ALIGNMENT_MODE_UNALIGNED); 2437 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE, 2438 !!adev->gmc.noretry); 2439 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp); 2440 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, 0); 2441 } else { 2442 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE, 2443 SH_MEM_ALIGNMENT_MODE_UNALIGNED); 2444 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE, 2445 !!adev->gmc.noretry); 2446 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp); 2447 tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE, 2448 (adev->gmc.private_aperture_start >> 48)); 2449 tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE, 2450 (adev->gmc.shared_aperture_start >> 48)); 2451 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, tmp); 2452 } 2453 } 2454 soc15_grbm_select(adev, 0, 0, 0, 0); 2455 2456 mutex_unlock(&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 2463 static 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); 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) == 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); 2481 DRM_INFO("Timeout wait for RLC serdes %u,%u\n", 2482 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); 2489 2490 mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK | 2491 RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK | 2492 RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK | 2493 RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK; 2494 for (k = 0; k < adev->usec_timeout; k++) { 2495 if ((RREG32_SOC15(GC, 0, mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0) 2496 break; 2497 udelay(1); 2498 } 2499 } 2500 2501 static void gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device *adev, 2502 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); 2509 2510 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0); 2511 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0); 2512 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0); 2513 if(adev->gfx.num_gfx_rings) 2514 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0); 2515 2516 WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp); 2517 } 2518 2519 static 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), 2524 adev->gfx.rlc.clear_state_gpu_addr >> 32); 2525 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_LO), 2526 adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc); 2527 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_LENGTH), 2528 adev->gfx.rlc.clear_state_size); 2529 } 2530 2531 static 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); 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); 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 2568 static 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); 2583 if (!register_list_format) 2584 return -ENOMEM; 2585 2586 /* setup unique_indirect_regs array and indirect_start_offsets array */ 2587 unique_indirect_reg_count = ARRAY_SIZE(unique_indirect_regs); 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)); 2596 2597 /* enable auto inc in case it is disabled */ 2598 tmp = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL)); 2599 tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK; 2600 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL), tmp); 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), 2604 RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET); 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), 2607 adev->gfx.rlc.register_restore[i]); 2608 2609 /* load indirect register */ 2610 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR), 2611 adev->gfx.rlc.reg_list_format_start); 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), 2616 register_list_format[i]); 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++]); 2622 continue; 2623 } 2624 2625 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]); 2626 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]); 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); 2631 break; 2632 } 2633 } 2634 2635 BUG_ON(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), 2644 adev->gfx.rlc.reg_restore_list_size); 2645 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), list_size); 2646 2647 /* write the starting offsets to RLC scratch ram */ 2648 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR), 2649 adev->gfx.rlc.starting_offsets_start); 2650 for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++) 2651 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), 2652 indirect_start_offsets[i]); 2653 2654 /* load unique indirect regs*/ 2655 for (i = 0; i < ARRAY_SIZE(unique_indirect_regs); i++) { 2656 if (unique_indirect_regs[i] != 0) { 2657 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_ADDR_0) 2658 + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i], 2659 unique_indirect_regs[i] & 0x3FFFF); 2660 2661 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_DATA_0) 2662 + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i], 2663 unique_indirect_regs[i] >> 20); 2664 } 2665 } 2666 2667 kfree(register_list_format); 2668 return 0; 2669 } 2670 2671 static void gfx_v9_0_enable_save_restore_machine(struct amdgpu_device *adev) 2672 { 2673 WREG32_FIELD15(GC, 0, RLC_SRM_CNTL, SRM_ENABLE, 1); 2674 } 2675 2676 static void pwr_10_0_gfxip_control_over_cgpg(struct amdgpu_device *adev, 2677 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)); 2683 if (enable) { 2684 /* enable GFXIP control over CGPG */ 2685 data |= PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK; 2686 if(default_data != data) 2687 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data); 2688 2689 /* update status */ 2690 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK; 2691 data |= (2 << PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT); 2692 if(default_data != data) 2693 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data); 2694 } else { 2695 /* restore GFXIP control over GCPG */ 2696 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK; 2697 if(default_data != data) 2698 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data); 2699 } 2700 } 2701 2702 static 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 | 2707 AMD_PG_SUPPORT_GFX_SMG | 2708 AMD_PG_SUPPORT_GFX_DMG)) { 2709 /* init IDLE_POLL_COUNT = 60 */ 2710 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL)); 2711 data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK; 2712 data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT); 2713 WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL), data); 2714 2715 /* init RLC PG Delay */ 2716 data = 0; 2717 data |= (0x10 << RLC_PG_DELAY__POWER_UP_DELAY__SHIFT); 2718 data |= (0x10 << RLC_PG_DELAY__POWER_DOWN_DELAY__SHIFT); 2719 data |= (0x10 << RLC_PG_DELAY__CMD_PROPAGATE_DELAY__SHIFT); 2720 data |= (0x40 << RLC_PG_DELAY__MEM_SLEEP_DELAY__SHIFT); 2721 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY), data); 2722 2723 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2)); 2724 data &= ~RLC_PG_DELAY_2__SERDES_CMD_DELAY_MASK; 2725 data |= (0x4 << RLC_PG_DELAY_2__SERDES_CMD_DELAY__SHIFT); 2726 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2), data); 2727 2728 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3)); 2729 data &= ~RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG_MASK; 2730 data |= (0xff << RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG__SHIFT); 2731 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3), data); 2732 2733 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL)); 2734 data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK; 2735 2736 /* program GRBM_REG_SAVE_GFX_IDLE_THRESHOLD to 0x55f0 */ 2737 data |= (0x55f0 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT); 2738 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL), data); 2739 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 3, 0)) 2740 pwr_10_0_gfxip_control_over_cgpg(adev, true); 2741 } 2742 } 2743 2744 static void gfx_v9_0_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev, 2745 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)); 2751 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2752 SMU_CLK_SLOWDOWN_ON_PU_ENABLE, 2753 enable ? 1 : 0); 2754 if (default_data != data) 2755 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2756 } 2757 2758 static void gfx_v9_0_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev, 2759 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)); 2765 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2766 SMU_CLK_SLOWDOWN_ON_PD_ENABLE, 2767 enable ? 1 : 0); 2768 if(default_data != data) 2769 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2770 } 2771 2772 static void gfx_v9_0_enable_cp_power_gating(struct amdgpu_device *adev, 2773 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)); 2779 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2780 CP_PG_DISABLE, 2781 enable ? 0 : 1); 2782 if(default_data != data) 2783 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2784 } 2785 2786 static void gfx_v9_0_enable_gfx_cg_power_gating(struct amdgpu_device *adev, 2787 bool enable) 2788 { 2789 uint32_t data, default_data; 2790 2791 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL)); 2792 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2793 GFX_POWER_GATING_ENABLE, 2794 enable ? 1 : 0); 2795 if(default_data != data) 2796 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2797 } 2798 2799 static void gfx_v9_0_enable_gfx_pipeline_powergating(struct amdgpu_device *adev, 2800 bool enable) 2801 { 2802 uint32_t data, default_data; 2803 2804 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL)); 2805 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2806 GFX_PIPELINE_PG_ENABLE, 2807 enable ? 1 : 0); 2808 if(default_data != data) 2809 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2810 2811 if (!enable) 2812 /* read any GFX register to wake up GFX */ 2813 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmDB_RENDER_CONTROL)); 2814 } 2815 2816 static void gfx_v9_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev, 2817 bool enable) 2818 { 2819 uint32_t data, default_data; 2820 2821 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL)); 2822 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2823 STATIC_PER_CU_PG_ENABLE, 2824 enable ? 1 : 0); 2825 if(default_data != data) 2826 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2827 } 2828 2829 static void gfx_v9_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev, 2830 bool enable) 2831 { 2832 uint32_t data, default_data; 2833 2834 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL)); 2835 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2836 DYN_PER_CU_PG_ENABLE, 2837 enable ? 1 : 0); 2838 if(default_data != data) 2839 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2840 } 2841 2842 static 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) || 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 | 2858 AMD_PG_SUPPORT_GFX_SMG | 2859 AMD_PG_SUPPORT_GFX_DMG | 2860 AMD_PG_SUPPORT_CP | 2861 AMD_PG_SUPPORT_GDS | 2862 AMD_PG_SUPPORT_RLC_SMU_HS)) { 2863 WREG32_SOC15(GC, 0, mmRLC_JUMP_TABLE_RESTORE, 2864 adev->gfx.rlc.cp_table_gpu_addr >> 8); 2865 gfx_v9_0_init_gfx_power_gating(adev); 2866 } 2867 } 2868 2869 static void gfx_v9_0_rlc_stop(struct amdgpu_device *adev) 2870 { 2871 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 0); 2872 gfx_v9_0_enable_gui_idle_interrupt(adev, false); 2873 gfx_v9_0_wait_for_rlc_serdes(adev); 2874 } 2875 2876 static void gfx_v9_0_rlc_reset(struct amdgpu_device *adev) 2877 { 2878 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1); 2879 udelay(50); 2880 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0); 2881 udelay(50); 2882 } 2883 2884 static 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); 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, true); 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); 2902 if(rlc_ucode_ver == 0x108) { 2903 DRM_INFO("Using rlc debug ucode. mmRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n", 2904 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); 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); 2912 } 2913 #endif 2914 } 2915 2916 static 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 -EINVAL; 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)); 2930 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; 2931 2932 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, 2933 RLCG_UCODE_LOADING_START_ADDRESS); 2934 for (i = 0; i < fw_size; i++) 2935 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++)); 2936 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version); 2937 2938 return 0; 2939 } 2940 2941 static int gfx_v9_0_rlc_resume(struct amdgpu_device *adev) 2942 { 2943 int r; 2944 2945 if (amdgpu_sriov_vf(adev)) { 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); 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): 2966 case IP_VERSION(9, 1, 0): 2967 if (amdgpu_lbpw == 0) 2968 gfx_v9_0_enable_lbpw(adev, false); 2969 else 2970 gfx_v9_0_enable_lbpw(adev, true); 2971 break; 2972 case IP_VERSION(9, 4, 0): 2973 if (amdgpu_lbpw > 0) 2974 gfx_v9_0_enable_lbpw(adev, true); 2975 else 2976 gfx_v9_0_enable_lbpw(adev, false); 2977 break; 2978 default: 2979 break; 2980 } 2981 2982 adev->gfx.rlc.funcs->start(adev); 2983 2984 return 0; 2985 } 2986 2987 static void gfx_v9_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable) 2988 { 2989 u32 tmp = RREG32_SOC15(GC, 0, mmCP_ME_CNTL); 2990 2991 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1); 2992 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1); 2993 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, enable ? 0 : 1); 2994 WREG32_SOC15_RLC(GC, 0, mmCP_ME_CNTL, tmp); 2995 udelay(50); 2996 } 2997 2998 static 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 -EINVAL; 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, false); 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)); 3026 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4; 3027 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, 0); 3028 for (i = 0; i < fw_size; i++) 3029 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++)); 3030 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version); 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)); 3036 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4; 3037 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, 0); 3038 for (i = 0; i < fw_size; i++) 3039 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++)); 3040 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version); 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)); 3046 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4; 3047 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, 0); 3048 for (i = 0; i < fw_size; i++) 3049 WREG32_SOC15(GC, 0, mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++)); 3050 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version); 3051 3052 return 0; 3053 } 3054 3055 static 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; 3059 const struct cs_extent_def *ext = NULL; 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); 3064 WREG32_SOC15(GC, 0, mmCP_DEVICE_ID, 1); 3065 3066 gfx_v9_0_cp_gfx_enable(adev, true); 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); 3071 return r; 3072 } 3073 3074 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 3075 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 3076 3077 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 3078 amdgpu_ring_write(ring, 0x80000000); 3079 amdgpu_ring_write(ring, 0x80000000); 3080 3081 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) { 3082 for (ext = sect->section; ext->extent != NULL; ++ext) { 3083 if (sect->id == SECT_CONTEXT) { 3084 amdgpu_ring_write(ring, 3085 PACKET3(PACKET3_SET_CONTEXT_REG, 3086 ext->reg_count)); 3087 amdgpu_ring_write(ring, 3088 ext->reg_index - PACKET3_SET_CONTEXT_REG_START); 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)); 3096 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE); 3097 3098 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0)); 3099 amdgpu_ring_write(ring, 0); 3100 3101 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2)); 3102 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE)); 3103 amdgpu_ring_write(ring, 0x8000); 3104 amdgpu_ring_write(ring, 0x8000); 3105 3106 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG,1)); 3107 tmp = (PACKET3_SET_UCONFIG_REG_INDEX_TYPE | 3108 (SOC15_REG_OFFSET(GC, 0, mmVGT_INDEX_TYPE) - PACKET3_SET_UCONFIG_REG_START)); 3109 amdgpu_ring_write(ring, tmp); 3110 amdgpu_ring_write(ring, 0); 3111 3112 amdgpu_ring_commit(ring); 3113 3114 return 0; 3115 } 3116 3117 static 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); 3126 3127 /* set the RB to use vmid 0 */ 3128 WREG32_SOC15(GC, 0, mmCP_RB_VMID, 0); 3129 3130 /* Set ring buffer size */ 3131 ring = &adev->gfx.gfx_ring[0]; 3132 rb_bufsz = order_base_2(ring->ring_size / 8); 3133 tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz); 3134 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2); 3135 #ifdef __BIG_ENDIAN 3136 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1); 3137 #endif 3138 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp); 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)); 3143 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr)); 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)); 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); 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)); 3152 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr)); 3153 3154 mdelay(1); 3155 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp); 3156 3157 rb_addr = ring->gpu_addr >> 8; 3158 WREG32_SOC15(GC, 0, mmCP_RB0_BASE, rb_addr); 3159 WREG32_SOC15(GC, 0, mmCP_RB0_BASE_HI, upper_32_bits(rb_addr)); 3160 3161 tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL); 3162 if (ring->use_doorbell) { 3163 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, 3164 DOORBELL_OFFSET, ring->doorbell_index); 3165 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, 3166 DOORBELL_EN, 1); 3167 } else { 3168 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0); 3169 } 3170 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, tmp); 3171 3172 tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER, 3173 DOORBELL_RANGE_LOWER, ring->doorbell_index); 3174 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp); 3175 3176 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER, 3177 CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK); 3178 3179 3180 /* start the ring */ 3181 gfx_v9_0_cp_gfx_start(adev); 3182 ring->sched.ready = true; 3183 3184 return 0; 3185 } 3186 3187 static void gfx_v9_0_cp_compute_enable(struct amdgpu_device *adev, bool enable) 3188 { 3189 if (enable) { 3190 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL, 0); 3191 } else { 3192 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL, 3193 (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK)); 3194 adev->gfx.kiq.ring.sched.ready = false; 3195 } 3196 udelay(50); 3197 } 3198 3199 static 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 -EINVAL; 3208 3209 gfx_v9_0_cp_compute_enable(adev, false); 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)); 3217 tmp = 0; 3218 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0); 3219 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0); 3220 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL, tmp); 3221 3222 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO, 3223 adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000); 3224 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI, 3225 upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr)); 3226 3227 /* MEC1 */ 3228 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR, 3229 mec_hdr->jt_offset); 3230 for (i = 0; i < mec_hdr->jt_size; i++) 3231 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_DATA, 3232 le32_to_cpup(fw_data + mec_hdr->jt_offset + i)); 3233 3234 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR, 3235 adev->gfx.mec_fw_version); 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 */ 3242 static 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); 3249 tmp &= 0xffffff00; 3250 tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue); 3251 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp); 3252 tmp |= 0x80; 3253 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp); 3254 } 3255 3256 static 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_MAXIMUM; 3265 } 3266 } 3267 } 3268 3269 static 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 3290 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask)); 3291 mqd->dynamic_cu_mask_addr_hi = 3292 upper_32_bits(ring->mqd_gpu_addr 3293 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask)); 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); 3298 3299 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */ 3300 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL); 3301 tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE, 3302 (order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1)); 3303 3304 mqd->cp_hqd_eop_control = tmp; 3305 3306 /* enable doorbell? */ 3307 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL); 3308 3309 if (ring->use_doorbell) { 3310 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 3311 DOORBELL_OFFSET, ring->doorbell_index); 3312 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 3313 DOORBELL_EN, 1); 3314 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 3315 DOORBELL_SOURCE, 0); 3316 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 3317 DOORBELL_HIT, 0); 3318 } else { 3319 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 3320 DOORBELL_EN, 0); 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); 3335 3336 /* set MQD vmid to 0 */ 3337 tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL); 3338 tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0); 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); 3345 3346 /* set up the HQD, this is similar to CP_RB0_CNTL */ 3347 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL); 3348 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE, 3349 (order_base_2(ring->ring_size / 4) - 1)); 3350 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE, 3351 (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1)); 3352 #ifdef __BIG_ENDIAN 3353 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1); 3354 #endif 3355 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0); 3356 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0); 3357 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1); 3358 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1); 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) & 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) & 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); 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); 3380 tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53); 3381 mqd->cp_hqd_persistent_state = tmp; 3382 3383 /* set MIN_IB_AVAIL_SIZE */ 3384 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL); 3385 tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3); 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); 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 3401 static 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); 3409 3410 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR, 3411 mqd->cp_hqd_eop_base_addr_lo); 3412 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI, 3413 mqd->cp_hqd_eop_base_addr_hi); 3414 3415 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */ 3416 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_CONTROL, 3417 mqd->cp_hqd_eop_control); 3418 3419 /* enable doorbell? */ 3420 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 3421 mqd->cp_hqd_pq_doorbell_control); 3422 3423 /* disable the queue if it's active */ 3424 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) { 3425 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1); 3426 for (j = 0; j < adev->usec_timeout; j++) { 3427 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1)) 3428 break; 3429 udelay(1); 3430 } 3431 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 3432 mqd->cp_hqd_dequeue_request); 3433 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR, 3434 mqd->cp_hqd_pq_rptr); 3435 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 3436 mqd->cp_hqd_pq_wptr_lo); 3437 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 3438 mqd->cp_hqd_pq_wptr_hi); 3439 } 3440 3441 /* set the pointer to the MQD */ 3442 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR, 3443 mqd->cp_mqd_base_addr_lo); 3444 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR_HI, 3445 mqd->cp_mqd_base_addr_hi); 3446 3447 /* set MQD vmid to 0 */ 3448 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_CONTROL, 3449 mqd->cp_mqd_control); 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, 3453 mqd->cp_hqd_pq_base_lo); 3454 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE_HI, 3455 mqd->cp_hqd_pq_base_hi); 3456 3457 /* set up the HQD, this is similar to CP_RB0_CNTL */ 3458 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_CONTROL, 3459 mqd->cp_hqd_pq_control); 3460 3461 /* set the wb address whether it's enabled or not */ 3462 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR, 3463 mqd->cp_hqd_pq_rptr_report_addr_lo); 3464 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI, 3465 mqd->cp_hqd_pq_rptr_report_addr_hi); 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, 3469 mqd->cp_hqd_pq_wptr_poll_addr_lo); 3470 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI, 3471 mqd->cp_hqd_pq_wptr_poll_addr_hi); 3472 3473 /* enable the doorbell if requested */ 3474 if (ring->use_doorbell) { 3475 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER, 3476 (adev->doorbell_index.kiq * 2) << 2); 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, 3484 (adev->doorbell.size - 4)); 3485 else 3486 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER, 3487 (adev->doorbell_index.userqueue_end * 2) << 2); 3488 } 3489 3490 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 3491 mqd->cp_hqd_pq_doorbell_control); 3492 3493 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */ 3494 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 3495 mqd->cp_hqd_pq_wptr_lo); 3496 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 3497 mqd->cp_hqd_pq_wptr_hi); 3498 3499 /* set the vmid for the queue */ 3500 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_VMID, mqd->cp_hqd_vmid); 3501 3502 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE, 3503 mqd->cp_hqd_persistent_state); 3504 3505 /* activate the queue */ 3506 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE, 3507 mqd->cp_hqd_active); 3508 3509 if (ring->use_doorbell) 3510 WREG32_FIELD15(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1); 3511 3512 return 0; 3513 } 3514 3515 static 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) & 1) { 3522 3523 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1); 3524 3525 for (j = 0; j < adev->usec_timeout; j++) { 3526 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1)) 3527 break; 3528 udelay(1); 3529 } 3530 3531 if (j == AMDGPU_MAX_USEC_TIMEOUT) { 3532 DRM_DEBUG("KIQ dequeue request failed.\n"); 3533 3534 /* Manual disable if dequeue request times out */ 3535 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE, 0); 3536 } 3537 3538 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 3539 0); 3540 } 3541 3542 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IQ_TIMER, 0); 3543 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IB_CONTROL, 0); 3544 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE, 0); 3545 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000); 3546 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0); 3547 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR, 0); 3548 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 0); 3549 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 0); 3550 3551 return 0; 3552 } 3553 3554 static 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_RINGS; 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)); 3573 3574 /* reset ring buffer */ 3575 ring->wptr = 0; 3576 amdgpu_ring_clear_ring(ring); 3577 3578 mutex_lock(&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); 3583 } else { 3584 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); 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); 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)); 3596 } 3597 3598 return 0; 3599 } 3600 3601 static 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)); 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); 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); 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)); 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)); 3630 3631 /* reset ring buffer */ 3632 ring->wptr = 0; 3633 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, 0); 3634 amdgpu_ring_clear_ring(ring); 3635 } else { 3636 amdgpu_ring_clear_ring(ring); 3637 } 3638 3639 return 0; 3640 } 3641 3642 static 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, false); 3650 if (unlikely(r != 0)) 3651 return r; 3652 3653 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); 3654 if (unlikely(r != 0)) 3655 return r; 3656 3657 gfx_v9_0_kiq_init_queue(ring); 3658 amdgpu_bo_kunmap(ring->mqd_obj); 3659 ring->mqd_ptr = NULL; 3660 amdgpu_bo_unreserve(ring->mqd_obj); 3661 ring->sched.ready = true; 3662 return 0; 3663 } 3664 3665 static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev) 3666 { 3667 struct amdgpu_ring *ring = NULL; 3668 int r = 0, i; 3669 3670 gfx_v9_0_cp_compute_enable(adev, true); 3671 3672 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 3673 ring = &adev->gfx.compute_ring[i]; 3674 3675 r = amdgpu_bo_reserve(ring->mqd_obj, false); 3676 if (unlikely(r != 0)) 3677 goto done; 3678 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); 3679 if (!r) { 3680 r = gfx_v9_0_kcq_init_queue(ring); 3681 amdgpu_bo_kunmap(ring->mqd_obj); 3682 ring->mqd_ptr = NULL; 3683 } 3684 amdgpu_bo_unreserve(ring->mqd_obj); 3685 if (r) 3686 goto done; 3687 } 3688 3689 r = amdgpu_gfx_enable_kcq(adev); 3690 done: 3691 return r; 3692 } 3693 3694 static int gfx_v9_0_cp_resume(struct amdgpu_device *adev) 3695 { 3696 int r, i; 3697 struct amdgpu_ring *ring; 3698 3699 if (!(adev->flags & AMD_IS_APU)) 3700 gfx_v9_0_enable_gui_idle_interrupt(adev, false); 3701 3702 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { 3703 if (adev->gfx.num_gfx_rings) { 3704 /* legacy firmware loading */ 3705 r = gfx_v9_0_cp_gfx_load_microcode(adev); 3706 if (r) 3707 return r; 3708 } 3709 3710 r = gfx_v9_0_cp_compute_load_microcode(adev); 3711 if (r) 3712 return r; 3713 } 3714 3715 r = gfx_v9_0_kiq_resume(adev); 3716 if (r) 3717 return r; 3718 3719 if (adev->gfx.num_gfx_rings) { 3720 r = gfx_v9_0_cp_gfx_resume(adev); 3721 if (r) 3722 return r; 3723 } 3724 3725 r = gfx_v9_0_kcq_resume(adev); 3726 if (r) 3727 return r; 3728 3729 if (adev->gfx.num_gfx_rings) { 3730 ring = &adev->gfx.gfx_ring[0]; 3731 r = amdgpu_ring_test_helper(ring); 3732 if (r) 3733 return r; 3734 } 3735 3736 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 3737 ring = &adev->gfx.compute_ring[i]; 3738 amdgpu_ring_test_helper(ring); 3739 } 3740 3741 gfx_v9_0_enable_gui_idle_interrupt(adev, true); 3742 3743 return 0; 3744 } 3745 3746 static void gfx_v9_0_init_tcp_config(struct amdgpu_device *adev) 3747 { 3748 u32 tmp; 3749 3750 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1) && 3751 adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 2)) 3752 return; 3753 3754 tmp = RREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG); 3755 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE64KHASH, 3756 adev->df.hash_status.hash_64k); 3757 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE2MHASH, 3758 adev->df.hash_status.hash_2m); 3759 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE1GHASH, 3760 adev->df.hash_status.hash_1g); 3761 WREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG, tmp); 3762 } 3763 3764 static void gfx_v9_0_cp_enable(struct amdgpu_device *adev, bool enable) 3765 { 3766 if (adev->gfx.num_gfx_rings) 3767 gfx_v9_0_cp_gfx_enable(adev, enable); 3768 gfx_v9_0_cp_compute_enable(adev, enable); 3769 } 3770 3771 static int gfx_v9_0_hw_init(void *handle) 3772 { 3773 int r; 3774 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3775 3776 if (!amdgpu_sriov_vf(adev)) 3777 gfx_v9_0_init_golden_registers(adev); 3778 3779 gfx_v9_0_constants_init(adev); 3780 3781 gfx_v9_0_init_tcp_config(adev); 3782 3783 r = adev->gfx.rlc.funcs->resume(adev); 3784 if (r) 3785 return r; 3786 3787 r = gfx_v9_0_cp_resume(adev); 3788 if (r) 3789 return r; 3790 3791 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2)) 3792 gfx_v9_4_2_set_power_brake_sequence(adev); 3793 3794 return r; 3795 } 3796 3797 static int gfx_v9_0_hw_fini(void *handle) 3798 { 3799 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3800 3801 amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0); 3802 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); 3803 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); 3804 3805 /* DF freeze and kcq disable will fail */ 3806 if (!amdgpu_ras_intr_triggered()) 3807 /* disable KCQ to avoid CPC touch memory not valid anymore */ 3808 amdgpu_gfx_disable_kcq(adev); 3809 3810 if (amdgpu_sriov_vf(adev)) { 3811 gfx_v9_0_cp_gfx_enable(adev, false); 3812 /* must disable polling for SRIOV when hw finished, otherwise 3813 * CPC engine may still keep fetching WB address which is already 3814 * invalid after sw finished and trigger DMAR reading error in 3815 * hypervisor side. 3816 */ 3817 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0); 3818 return 0; 3819 } 3820 3821 /* Use deinitialize sequence from CAIL when unbinding device from driver, 3822 * otherwise KIQ is hanging when binding back 3823 */ 3824 if (!amdgpu_in_reset(adev) && !adev->in_suspend) { 3825 mutex_lock(&adev->srbm_mutex); 3826 soc15_grbm_select(adev, adev->gfx.kiq.ring.me, 3827 adev->gfx.kiq.ring.pipe, 3828 adev->gfx.kiq.ring.queue, 0); 3829 gfx_v9_0_kiq_fini_register(&adev->gfx.kiq.ring); 3830 soc15_grbm_select(adev, 0, 0, 0, 0); 3831 mutex_unlock(&adev->srbm_mutex); 3832 } 3833 3834 gfx_v9_0_cp_enable(adev, false); 3835 3836 /* Skip stopping RLC with A+A reset or when RLC controls GFX clock */ 3837 if ((adev->gmc.xgmi.connected_to_cpu && amdgpu_in_reset(adev)) || 3838 (adev->ip_versions[GC_HWIP][0] >= IP_VERSION(9, 4, 2))) { 3839 dev_dbg(adev->dev, "Skipping RLC halt\n"); 3840 return 0; 3841 } 3842 3843 adev->gfx.rlc.funcs->stop(adev); 3844 return 0; 3845 } 3846 3847 static int gfx_v9_0_suspend(void *handle) 3848 { 3849 return gfx_v9_0_hw_fini(handle); 3850 } 3851 3852 static int gfx_v9_0_resume(void *handle) 3853 { 3854 return gfx_v9_0_hw_init(handle); 3855 } 3856 3857 static bool gfx_v9_0_is_idle(void *handle) 3858 { 3859 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3860 3861 if (REG_GET_FIELD(RREG32_SOC15(GC, 0, mmGRBM_STATUS), 3862 GRBM_STATUS, GUI_ACTIVE)) 3863 return false; 3864 else 3865 return true; 3866 } 3867 3868 static int gfx_v9_0_wait_for_idle(void *handle) 3869 { 3870 unsigned i; 3871 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3872 3873 for (i = 0; i < adev->usec_timeout; i++) { 3874 if (gfx_v9_0_is_idle(handle)) 3875 return 0; 3876 udelay(1); 3877 } 3878 return -ETIMEDOUT; 3879 } 3880 3881 static int gfx_v9_0_soft_reset(void *handle) 3882 { 3883 u32 grbm_soft_reset = 0; 3884 u32 tmp; 3885 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3886 3887 /* GRBM_STATUS */ 3888 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS); 3889 if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK | 3890 GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK | 3891 GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK | 3892 GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK | 3893 GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK | 3894 GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) { 3895 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 3896 GRBM_SOFT_RESET, SOFT_RESET_CP, 1); 3897 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 3898 GRBM_SOFT_RESET, SOFT_RESET_GFX, 1); 3899 } 3900 3901 if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) { 3902 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 3903 GRBM_SOFT_RESET, SOFT_RESET_CP, 1); 3904 } 3905 3906 /* GRBM_STATUS2 */ 3907 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2); 3908 if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY)) 3909 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 3910 GRBM_SOFT_RESET, SOFT_RESET_RLC, 1); 3911 3912 3913 if (grbm_soft_reset) { 3914 /* stop the rlc */ 3915 adev->gfx.rlc.funcs->stop(adev); 3916 3917 if (adev->gfx.num_gfx_rings) 3918 /* Disable GFX parsing/prefetching */ 3919 gfx_v9_0_cp_gfx_enable(adev, false); 3920 3921 /* Disable MEC parsing/prefetching */ 3922 gfx_v9_0_cp_compute_enable(adev, false); 3923 3924 if (grbm_soft_reset) { 3925 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET); 3926 tmp |= grbm_soft_reset; 3927 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp); 3928 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp); 3929 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET); 3930 3931 udelay(50); 3932 3933 tmp &= ~grbm_soft_reset; 3934 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp); 3935 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET); 3936 } 3937 3938 /* Wait a little for things to settle down */ 3939 udelay(50); 3940 } 3941 return 0; 3942 } 3943 3944 static uint64_t gfx_v9_0_kiq_read_clock(struct amdgpu_device *adev) 3945 { 3946 signed long r, cnt = 0; 3947 unsigned long flags; 3948 uint32_t seq, reg_val_offs = 0; 3949 uint64_t value = 0; 3950 struct amdgpu_kiq *kiq = &adev->gfx.kiq; 3951 struct amdgpu_ring *ring = &kiq->ring; 3952 3953 BUG_ON(!ring->funcs->emit_rreg); 3954 3955 spin_lock_irqsave(&kiq->ring_lock, flags); 3956 if (amdgpu_device_wb_get(adev, ®_val_offs)) { 3957 pr_err("critical bug! too many kiq readers\n"); 3958 goto failed_unlock; 3959 } 3960 amdgpu_ring_alloc(ring, 32); 3961 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4)); 3962 amdgpu_ring_write(ring, 9 | /* src: register*/ 3963 (5 << 8) | /* dst: memory */ 3964 (1 << 16) | /* count sel */ 3965 (1 << 20)); /* write confirm */ 3966 amdgpu_ring_write(ring, 0); 3967 amdgpu_ring_write(ring, 0); 3968 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr + 3969 reg_val_offs * 4)); 3970 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr + 3971 reg_val_offs * 4)); 3972 r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT); 3973 if (r) 3974 goto failed_undo; 3975 3976 amdgpu_ring_commit(ring); 3977 spin_unlock_irqrestore(&kiq->ring_lock, flags); 3978 3979 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); 3980 3981 /* don't wait anymore for gpu reset case because this way may 3982 * block gpu_recover() routine forever, e.g. this virt_kiq_rreg 3983 * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will 3984 * never return if we keep waiting in virt_kiq_rreg, which cause 3985 * gpu_recover() hang there. 3986 * 3987 * also don't wait anymore for IRQ context 3988 * */ 3989 if (r < 1 && (amdgpu_in_reset(adev))) 3990 goto failed_kiq_read; 3991 3992 might_sleep(); 3993 while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) { 3994 drm_msleep(MAX_KIQ_REG_BAILOUT_INTERVAL); 3995 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); 3996 } 3997 3998 if (cnt > MAX_KIQ_REG_TRY) 3999 goto failed_kiq_read; 4000 4001 mb(); 4002 value = (uint64_t)adev->wb.wb[reg_val_offs] | 4003 (uint64_t)adev->wb.wb[reg_val_offs + 1 ] << 32ULL; 4004 amdgpu_device_wb_free(adev, reg_val_offs); 4005 return value; 4006 4007 failed_undo: 4008 amdgpu_ring_undo(ring); 4009 failed_unlock: 4010 spin_unlock_irqrestore(&kiq->ring_lock, flags); 4011 failed_kiq_read: 4012 if (reg_val_offs) 4013 amdgpu_device_wb_free(adev, reg_val_offs); 4014 pr_err("failed to read gpu clock\n"); 4015 return ~0; 4016 } 4017 4018 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev) 4019 { 4020 uint64_t clock, clock_lo, clock_hi, hi_check; 4021 4022 switch (adev->ip_versions[GC_HWIP][0]) { 4023 case IP_VERSION(9, 3, 0): 4024 preempt_disable(); 4025 clock_hi = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir); 4026 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir); 4027 hi_check = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir); 4028 /* The SMUIO TSC clock frequency is 100MHz, which sets 32-bit carry over 4029 * roughly every 42 seconds. 4030 */ 4031 if (hi_check != clock_hi) { 4032 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir); 4033 clock_hi = hi_check; 4034 } 4035 preempt_enable(); 4036 clock = clock_lo | (clock_hi << 32ULL); 4037 break; 4038 default: 4039 amdgpu_gfx_off_ctrl(adev, false); 4040 mutex_lock(&adev->gfx.gpu_clock_mutex); 4041 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 0, 1) && amdgpu_sriov_runtime(adev)) { 4042 clock = gfx_v9_0_kiq_read_clock(adev); 4043 } else { 4044 WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1); 4045 clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB) | 4046 ((uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL); 4047 } 4048 mutex_unlock(&adev->gfx.gpu_clock_mutex); 4049 amdgpu_gfx_off_ctrl(adev, true); 4050 break; 4051 } 4052 return clock; 4053 } 4054 4055 static void gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring *ring, 4056 uint32_t vmid, 4057 uint32_t gds_base, uint32_t gds_size, 4058 uint32_t gws_base, uint32_t gws_size, 4059 uint32_t oa_base, uint32_t oa_size) 4060 { 4061 struct amdgpu_device *adev = ring->adev; 4062 4063 /* GDS Base */ 4064 gfx_v9_0_write_data_to_reg(ring, 0, false, 4065 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_BASE) + 2 * vmid, 4066 gds_base); 4067 4068 /* GDS Size */ 4069 gfx_v9_0_write_data_to_reg(ring, 0, false, 4070 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE) + 2 * vmid, 4071 gds_size); 4072 4073 /* GWS */ 4074 gfx_v9_0_write_data_to_reg(ring, 0, false, 4075 SOC15_REG_OFFSET(GC, 0, mmGDS_GWS_VMID0) + vmid, 4076 gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base); 4077 4078 /* OA */ 4079 gfx_v9_0_write_data_to_reg(ring, 0, false, 4080 SOC15_REG_OFFSET(GC, 0, mmGDS_OA_VMID0) + vmid, 4081 (1 << (oa_size + oa_base)) - (1 << oa_base)); 4082 } 4083 4084 static const u32 vgpr_init_compute_shader[] = 4085 { 4086 0xb07c0000, 0xbe8000ff, 4087 0x000000f8, 0xbf110800, 4088 0x7e000280, 0x7e020280, 4089 0x7e040280, 0x7e060280, 4090 0x7e080280, 0x7e0a0280, 4091 0x7e0c0280, 0x7e0e0280, 4092 0x80808800, 0xbe803200, 4093 0xbf84fff5, 0xbf9c0000, 4094 0xd28c0001, 0x0001007f, 4095 0xd28d0001, 0x0002027e, 4096 0x10020288, 0xb8810904, 4097 0xb7814000, 0xd1196a01, 4098 0x00000301, 0xbe800087, 4099 0xbefc00c1, 0xd89c4000, 4100 0x00020201, 0xd89cc080, 4101 0x00040401, 0x320202ff, 4102 0x00000800, 0x80808100, 4103 0xbf84fff8, 0x7e020280, 4104 0xbf810000, 0x00000000, 4105 }; 4106 4107 static const u32 sgpr_init_compute_shader[] = 4108 { 4109 0xb07c0000, 0xbe8000ff, 4110 0x0000005f, 0xbee50080, 4111 0xbe812c65, 0xbe822c65, 4112 0xbe832c65, 0xbe842c65, 4113 0xbe852c65, 0xb77c0005, 4114 0x80808500, 0xbf84fff8, 4115 0xbe800080, 0xbf810000, 4116 }; 4117 4118 static const u32 vgpr_init_compute_shader_arcturus[] = { 4119 0xd3d94000, 0x18000080, 0xd3d94001, 0x18000080, 0xd3d94002, 0x18000080, 4120 0xd3d94003, 0x18000080, 0xd3d94004, 0x18000080, 0xd3d94005, 0x18000080, 4121 0xd3d94006, 0x18000080, 0xd3d94007, 0x18000080, 0xd3d94008, 0x18000080, 4122 0xd3d94009, 0x18000080, 0xd3d9400a, 0x18000080, 0xd3d9400b, 0x18000080, 4123 0xd3d9400c, 0x18000080, 0xd3d9400d, 0x18000080, 0xd3d9400e, 0x18000080, 4124 0xd3d9400f, 0x18000080, 0xd3d94010, 0x18000080, 0xd3d94011, 0x18000080, 4125 0xd3d94012, 0x18000080, 0xd3d94013, 0x18000080, 0xd3d94014, 0x18000080, 4126 0xd3d94015, 0x18000080, 0xd3d94016, 0x18000080, 0xd3d94017, 0x18000080, 4127 0xd3d94018, 0x18000080, 0xd3d94019, 0x18000080, 0xd3d9401a, 0x18000080, 4128 0xd3d9401b, 0x18000080, 0xd3d9401c, 0x18000080, 0xd3d9401d, 0x18000080, 4129 0xd3d9401e, 0x18000080, 0xd3d9401f, 0x18000080, 0xd3d94020, 0x18000080, 4130 0xd3d94021, 0x18000080, 0xd3d94022, 0x18000080, 0xd3d94023, 0x18000080, 4131 0xd3d94024, 0x18000080, 0xd3d94025, 0x18000080, 0xd3d94026, 0x18000080, 4132 0xd3d94027, 0x18000080, 0xd3d94028, 0x18000080, 0xd3d94029, 0x18000080, 4133 0xd3d9402a, 0x18000080, 0xd3d9402b, 0x18000080, 0xd3d9402c, 0x18000080, 4134 0xd3d9402d, 0x18000080, 0xd3d9402e, 0x18000080, 0xd3d9402f, 0x18000080, 4135 0xd3d94030, 0x18000080, 0xd3d94031, 0x18000080, 0xd3d94032, 0x18000080, 4136 0xd3d94033, 0x18000080, 0xd3d94034, 0x18000080, 0xd3d94035, 0x18000080, 4137 0xd3d94036, 0x18000080, 0xd3d94037, 0x18000080, 0xd3d94038, 0x18000080, 4138 0xd3d94039, 0x18000080, 0xd3d9403a, 0x18000080, 0xd3d9403b, 0x18000080, 4139 0xd3d9403c, 0x18000080, 0xd3d9403d, 0x18000080, 0xd3d9403e, 0x18000080, 4140 0xd3d9403f, 0x18000080, 0xd3d94040, 0x18000080, 0xd3d94041, 0x18000080, 4141 0xd3d94042, 0x18000080, 0xd3d94043, 0x18000080, 0xd3d94044, 0x18000080, 4142 0xd3d94045, 0x18000080, 0xd3d94046, 0x18000080, 0xd3d94047, 0x18000080, 4143 0xd3d94048, 0x18000080, 0xd3d94049, 0x18000080, 0xd3d9404a, 0x18000080, 4144 0xd3d9404b, 0x18000080, 0xd3d9404c, 0x18000080, 0xd3d9404d, 0x18000080, 4145 0xd3d9404e, 0x18000080, 0xd3d9404f, 0x18000080, 0xd3d94050, 0x18000080, 4146 0xd3d94051, 0x18000080, 0xd3d94052, 0x18000080, 0xd3d94053, 0x18000080, 4147 0xd3d94054, 0x18000080, 0xd3d94055, 0x18000080, 0xd3d94056, 0x18000080, 4148 0xd3d94057, 0x18000080, 0xd3d94058, 0x18000080, 0xd3d94059, 0x18000080, 4149 0xd3d9405a, 0x18000080, 0xd3d9405b, 0x18000080, 0xd3d9405c, 0x18000080, 4150 0xd3d9405d, 0x18000080, 0xd3d9405e, 0x18000080, 0xd3d9405f, 0x18000080, 4151 0xd3d94060, 0x18000080, 0xd3d94061, 0x18000080, 0xd3d94062, 0x18000080, 4152 0xd3d94063, 0x18000080, 0xd3d94064, 0x18000080, 0xd3d94065, 0x18000080, 4153 0xd3d94066, 0x18000080, 0xd3d94067, 0x18000080, 0xd3d94068, 0x18000080, 4154 0xd3d94069, 0x18000080, 0xd3d9406a, 0x18000080, 0xd3d9406b, 0x18000080, 4155 0xd3d9406c, 0x18000080, 0xd3d9406d, 0x18000080, 0xd3d9406e, 0x18000080, 4156 0xd3d9406f, 0x18000080, 0xd3d94070, 0x18000080, 0xd3d94071, 0x18000080, 4157 0xd3d94072, 0x18000080, 0xd3d94073, 0x18000080, 0xd3d94074, 0x18000080, 4158 0xd3d94075, 0x18000080, 0xd3d94076, 0x18000080, 0xd3d94077, 0x18000080, 4159 0xd3d94078, 0x18000080, 0xd3d94079, 0x18000080, 0xd3d9407a, 0x18000080, 4160 0xd3d9407b, 0x18000080, 0xd3d9407c, 0x18000080, 0xd3d9407d, 0x18000080, 4161 0xd3d9407e, 0x18000080, 0xd3d9407f, 0x18000080, 0xd3d94080, 0x18000080, 4162 0xd3d94081, 0x18000080, 0xd3d94082, 0x18000080, 0xd3d94083, 0x18000080, 4163 0xd3d94084, 0x18000080, 0xd3d94085, 0x18000080, 0xd3d94086, 0x18000080, 4164 0xd3d94087, 0x18000080, 0xd3d94088, 0x18000080, 0xd3d94089, 0x18000080, 4165 0xd3d9408a, 0x18000080, 0xd3d9408b, 0x18000080, 0xd3d9408c, 0x18000080, 4166 0xd3d9408d, 0x18000080, 0xd3d9408e, 0x18000080, 0xd3d9408f, 0x18000080, 4167 0xd3d94090, 0x18000080, 0xd3d94091, 0x18000080, 0xd3d94092, 0x18000080, 4168 0xd3d94093, 0x18000080, 0xd3d94094, 0x18000080, 0xd3d94095, 0x18000080, 4169 0xd3d94096, 0x18000080, 0xd3d94097, 0x18000080, 0xd3d94098, 0x18000080, 4170 0xd3d94099, 0x18000080, 0xd3d9409a, 0x18000080, 0xd3d9409b, 0x18000080, 4171 0xd3d9409c, 0x18000080, 0xd3d9409d, 0x18000080, 0xd3d9409e, 0x18000080, 4172 0xd3d9409f, 0x18000080, 0xd3d940a0, 0x18000080, 0xd3d940a1, 0x18000080, 4173 0xd3d940a2, 0x18000080, 0xd3d940a3, 0x18000080, 0xd3d940a4, 0x18000080, 4174 0xd3d940a5, 0x18000080, 0xd3d940a6, 0x18000080, 0xd3d940a7, 0x18000080, 4175 0xd3d940a8, 0x18000080, 0xd3d940a9, 0x18000080, 0xd3d940aa, 0x18000080, 4176 0xd3d940ab, 0x18000080, 0xd3d940ac, 0x18000080, 0xd3d940ad, 0x18000080, 4177 0xd3d940ae, 0x18000080, 0xd3d940af, 0x18000080, 0xd3d940b0, 0x18000080, 4178 0xd3d940b1, 0x18000080, 0xd3d940b2, 0x18000080, 0xd3d940b3, 0x18000080, 4179 0xd3d940b4, 0x18000080, 0xd3d940b5, 0x18000080, 0xd3d940b6, 0x18000080, 4180 0xd3d940b7, 0x18000080, 0xd3d940b8, 0x18000080, 0xd3d940b9, 0x18000080, 4181 0xd3d940ba, 0x18000080, 0xd3d940bb, 0x18000080, 0xd3d940bc, 0x18000080, 4182 0xd3d940bd, 0x18000080, 0xd3d940be, 0x18000080, 0xd3d940bf, 0x18000080, 4183 0xd3d940c0, 0x18000080, 0xd3d940c1, 0x18000080, 0xd3d940c2, 0x18000080, 4184 0xd3d940c3, 0x18000080, 0xd3d940c4, 0x18000080, 0xd3d940c5, 0x18000080, 4185 0xd3d940c6, 0x18000080, 0xd3d940c7, 0x18000080, 0xd3d940c8, 0x18000080, 4186 0xd3d940c9, 0x18000080, 0xd3d940ca, 0x18000080, 0xd3d940cb, 0x18000080, 4187 0xd3d940cc, 0x18000080, 0xd3d940cd, 0x18000080, 0xd3d940ce, 0x18000080, 4188 0xd3d940cf, 0x18000080, 0xd3d940d0, 0x18000080, 0xd3d940d1, 0x18000080, 4189 0xd3d940d2, 0x18000080, 0xd3d940d3, 0x18000080, 0xd3d940d4, 0x18000080, 4190 0xd3d940d5, 0x18000080, 0xd3d940d6, 0x18000080, 0xd3d940d7, 0x18000080, 4191 0xd3d940d8, 0x18000080, 0xd3d940d9, 0x18000080, 0xd3d940da, 0x18000080, 4192 0xd3d940db, 0x18000080, 0xd3d940dc, 0x18000080, 0xd3d940dd, 0x18000080, 4193 0xd3d940de, 0x18000080, 0xd3d940df, 0x18000080, 0xd3d940e0, 0x18000080, 4194 0xd3d940e1, 0x18000080, 0xd3d940e2, 0x18000080, 0xd3d940e3, 0x18000080, 4195 0xd3d940e4, 0x18000080, 0xd3d940e5, 0x18000080, 0xd3d940e6, 0x18000080, 4196 0xd3d940e7, 0x18000080, 0xd3d940e8, 0x18000080, 0xd3d940e9, 0x18000080, 4197 0xd3d940ea, 0x18000080, 0xd3d940eb, 0x18000080, 0xd3d940ec, 0x18000080, 4198 0xd3d940ed, 0x18000080, 0xd3d940ee, 0x18000080, 0xd3d940ef, 0x18000080, 4199 0xd3d940f0, 0x18000080, 0xd3d940f1, 0x18000080, 0xd3d940f2, 0x18000080, 4200 0xd3d940f3, 0x18000080, 0xd3d940f4, 0x18000080, 0xd3d940f5, 0x18000080, 4201 0xd3d940f6, 0x18000080, 0xd3d940f7, 0x18000080, 0xd3d940f8, 0x18000080, 4202 0xd3d940f9, 0x18000080, 0xd3d940fa, 0x18000080, 0xd3d940fb, 0x18000080, 4203 0xd3d940fc, 0x18000080, 0xd3d940fd, 0x18000080, 0xd3d940fe, 0x18000080, 4204 0xd3d940ff, 0x18000080, 0xb07c0000, 0xbe8a00ff, 0x000000f8, 0xbf11080a, 4205 0x7e000280, 0x7e020280, 0x7e040280, 0x7e060280, 0x7e080280, 0x7e0a0280, 4206 0x7e0c0280, 0x7e0e0280, 0x808a880a, 0xbe80320a, 0xbf84fff5, 0xbf9c0000, 4207 0xd28c0001, 0x0001007f, 0xd28d0001, 0x0002027e, 0x10020288, 0xb88b0904, 4208 0xb78b4000, 0xd1196a01, 0x00001701, 0xbe8a0087, 0xbefc00c1, 0xd89c4000, 4209 0x00020201, 0xd89cc080, 0x00040401, 0x320202ff, 0x00000800, 0x808a810a, 4210 0xbf84fff8, 0xbf810000, 4211 }; 4212 4213 /* When below register arrays changed, please update gpr_reg_size, 4214 and sec_ded_counter_reg_size in function gfx_v9_0_do_edc_gpr_workarounds, 4215 to cover all gfx9 ASICs */ 4216 static const struct soc15_reg_entry vgpr_init_regs[] = { 4217 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 }, 4218 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 }, 4219 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 }, 4220 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 }, 4221 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x3f }, 4222 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */ 4223 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff }, 4224 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff }, 4225 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff }, 4226 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff }, 4227 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff }, 4228 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff }, 4229 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff }, 4230 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff }, 4231 }; 4232 4233 static const struct soc15_reg_entry vgpr_init_regs_arcturus[] = { 4234 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 }, 4235 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 }, 4236 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 }, 4237 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 }, 4238 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0xbf }, 4239 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */ 4240 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff }, 4241 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff }, 4242 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff }, 4243 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff }, 4244 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff }, 4245 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff }, 4246 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff }, 4247 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff }, 4248 }; 4249 4250 static const struct soc15_reg_entry sgpr1_init_regs[] = { 4251 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 }, 4252 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 }, 4253 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 }, 4254 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 }, 4255 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */ 4256 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 }, 4257 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x000000ff }, 4258 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x000000ff }, 4259 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x000000ff }, 4260 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x000000ff }, 4261 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x000000ff }, 4262 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x000000ff }, 4263 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x000000ff }, 4264 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x000000ff }, 4265 }; 4266 4267 static const struct soc15_reg_entry sgpr2_init_regs[] = { 4268 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 }, 4269 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 }, 4270 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 }, 4271 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 }, 4272 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */ 4273 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 }, 4274 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x0000ff00 }, 4275 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x0000ff00 }, 4276 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x0000ff00 }, 4277 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x0000ff00 }, 4278 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x0000ff00 }, 4279 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x0000ff00 }, 4280 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x0000ff00 }, 4281 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x0000ff00 }, 4282 }; 4283 4284 static const struct soc15_reg_entry gfx_v9_0_edc_counter_regs[] = { 4285 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT), 0, 1, 1}, 4286 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT), 0, 1, 1}, 4287 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 0, 1, 1}, 4288 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT), 0, 1, 1}, 4289 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 0, 1, 1}, 4290 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT), 0, 1, 1}, 4291 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT), 0, 1, 1}, 4292 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT), 0, 1, 1}, 4293 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT), 0, 1, 1}, 4294 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 0, 1, 1}, 4295 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_GRBM_CNT), 0, 1, 1}, 4296 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_DED), 0, 1, 1}, 4297 { SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT), 0, 4, 1}, 4298 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 0, 4, 6}, 4299 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_DED_CNT), 0, 4, 16}, 4300 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_INFO), 0, 4, 16}, 4301 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_SEC_CNT), 0, 4, 16}, 4302 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 0, 1, 16}, 4303 { SOC15_REG_ENTRY(GC, 0, mmTCP_ATC_EDC_GATCL1_CNT), 0, 4, 16}, 4304 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT), 0, 4, 16}, 4305 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 0, 4, 16}, 4306 { SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 0, 4, 16}, 4307 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 0, 4, 6}, 4308 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 0, 4, 16}, 4309 { SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 0, 4, 16}, 4310 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 0, 1, 1}, 4311 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 0, 1, 1}, 4312 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 0, 1, 32}, 4313 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 0, 1, 32}, 4314 { SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT), 0, 1, 72}, 4315 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 0, 1, 16}, 4316 { SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 0, 1, 2}, 4317 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 0, 4, 6}, 4318 }; 4319 4320 static int gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device *adev) 4321 { 4322 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0]; 4323 int i, r; 4324 4325 /* only support when RAS is enabled */ 4326 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) 4327 return 0; 4328 4329 r = amdgpu_ring_alloc(ring, 7); 4330 if (r) { 4331 DRM_ERROR("amdgpu: GDS workarounds failed to lock ring %s (%d).\n", 4332 ring->name, r); 4333 return r; 4334 } 4335 4336 WREG32_SOC15(GC, 0, mmGDS_VMID0_BASE, 0x00000000); 4337 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, adev->gds.gds_size); 4338 4339 amdgpu_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5)); 4340 amdgpu_ring_write(ring, (PACKET3_DMA_DATA_CP_SYNC | 4341 PACKET3_DMA_DATA_DST_SEL(1) | 4342 PACKET3_DMA_DATA_SRC_SEL(2) | 4343 PACKET3_DMA_DATA_ENGINE(0))); 4344 amdgpu_ring_write(ring, 0); 4345 amdgpu_ring_write(ring, 0); 4346 amdgpu_ring_write(ring, 0); 4347 amdgpu_ring_write(ring, 0); 4348 amdgpu_ring_write(ring, PACKET3_DMA_DATA_CMD_RAW_WAIT | 4349 adev->gds.gds_size); 4350 4351 amdgpu_ring_commit(ring); 4352 4353 for (i = 0; i < adev->usec_timeout; i++) { 4354 if (ring->wptr == gfx_v9_0_ring_get_rptr_compute(ring)) 4355 break; 4356 udelay(1); 4357 } 4358 4359 if (i >= adev->usec_timeout) 4360 r = -ETIMEDOUT; 4361 4362 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, 0x00000000); 4363 4364 return r; 4365 } 4366 4367 static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev) 4368 { 4369 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0]; 4370 struct amdgpu_ib ib; 4371 struct dma_fence *f = NULL; 4372 int r, i; 4373 unsigned total_size, vgpr_offset, sgpr_offset; 4374 u64 gpu_addr; 4375 4376 int compute_dim_x = adev->gfx.config.max_shader_engines * 4377 adev->gfx.config.max_cu_per_sh * 4378 adev->gfx.config.max_sh_per_se; 4379 int sgpr_work_group_size = 5; 4380 int gpr_reg_size = adev->gfx.config.max_shader_engines + 6; 4381 int vgpr_init_shader_size; 4382 const u32 *vgpr_init_shader_ptr; 4383 const struct soc15_reg_entry *vgpr_init_regs_ptr; 4384 4385 /* only support when RAS is enabled */ 4386 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) 4387 return 0; 4388 4389 /* bail if the compute ring is not ready */ 4390 if (!ring->sched.ready) 4391 return 0; 4392 4393 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1)) { 4394 vgpr_init_shader_ptr = vgpr_init_compute_shader_arcturus; 4395 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader_arcturus); 4396 vgpr_init_regs_ptr = vgpr_init_regs_arcturus; 4397 } else { 4398 vgpr_init_shader_ptr = vgpr_init_compute_shader; 4399 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader); 4400 vgpr_init_regs_ptr = vgpr_init_regs; 4401 } 4402 4403 total_size = 4404 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* VGPRS */ 4405 total_size += 4406 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS1 */ 4407 total_size += 4408 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS2 */ 4409 total_size = roundup2(total_size, 256); 4410 vgpr_offset = total_size; 4411 total_size += roundup2(vgpr_init_shader_size, 256); 4412 sgpr_offset = total_size; 4413 total_size += sizeof(sgpr_init_compute_shader); 4414 4415 /* allocate an indirect buffer to put the commands in */ 4416 memset(&ib, 0, sizeof(ib)); 4417 r = amdgpu_ib_get(adev, NULL, total_size, 4418 AMDGPU_IB_POOL_DIRECT, &ib); 4419 if (r) { 4420 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); 4421 return r; 4422 } 4423 4424 /* load the compute shaders */ 4425 for (i = 0; i < vgpr_init_shader_size/sizeof(u32); i++) 4426 ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_shader_ptr[i]; 4427 4428 for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++) 4429 ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i]; 4430 4431 /* init the ib length to 0 */ 4432 ib.length_dw = 0; 4433 4434 /* VGPR */ 4435 /* write the register state for the compute dispatch */ 4436 for (i = 0; i < gpr_reg_size; i++) { 4437 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); 4438 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(vgpr_init_regs_ptr[i]) 4439 - PACKET3_SET_SH_REG_START; 4440 ib.ptr[ib.length_dw++] = vgpr_init_regs_ptr[i].reg_value; 4441 } 4442 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ 4443 gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8; 4444 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); 4445 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO) 4446 - PACKET3_SET_SH_REG_START; 4447 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); 4448 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); 4449 4450 /* write dispatch packet */ 4451 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); 4452 ib.ptr[ib.length_dw++] = compute_dim_x * 2; /* x */ 4453 ib.ptr[ib.length_dw++] = 1; /* y */ 4454 ib.ptr[ib.length_dw++] = 1; /* z */ 4455 ib.ptr[ib.length_dw++] = 4456 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); 4457 4458 /* write CS partial flush packet */ 4459 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); 4460 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); 4461 4462 /* SGPR1 */ 4463 /* write the register state for the compute dispatch */ 4464 for (i = 0; i < gpr_reg_size; i++) { 4465 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); 4466 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr1_init_regs[i]) 4467 - PACKET3_SET_SH_REG_START; 4468 ib.ptr[ib.length_dw++] = sgpr1_init_regs[i].reg_value; 4469 } 4470 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ 4471 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8; 4472 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); 4473 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO) 4474 - PACKET3_SET_SH_REG_START; 4475 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); 4476 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); 4477 4478 /* write dispatch packet */ 4479 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); 4480 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */ 4481 ib.ptr[ib.length_dw++] = 1; /* y */ 4482 ib.ptr[ib.length_dw++] = 1; /* z */ 4483 ib.ptr[ib.length_dw++] = 4484 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); 4485 4486 /* write CS partial flush packet */ 4487 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); 4488 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); 4489 4490 /* SGPR2 */ 4491 /* write the register state for the compute dispatch */ 4492 for (i = 0; i < gpr_reg_size; i++) { 4493 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); 4494 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr2_init_regs[i]) 4495 - PACKET3_SET_SH_REG_START; 4496 ib.ptr[ib.length_dw++] = sgpr2_init_regs[i].reg_value; 4497 } 4498 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ 4499 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8; 4500 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); 4501 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO) 4502 - PACKET3_SET_SH_REG_START; 4503 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); 4504 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); 4505 4506 /* write dispatch packet */ 4507 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); 4508 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */ 4509 ib.ptr[ib.length_dw++] = 1; /* y */ 4510 ib.ptr[ib.length_dw++] = 1; /* z */ 4511 ib.ptr[ib.length_dw++] = 4512 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); 4513 4514 /* write CS partial flush packet */ 4515 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); 4516 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); 4517 4518 /* shedule the ib on the ring */ 4519 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); 4520 if (r) { 4521 DRM_ERROR("amdgpu: ib submit failed (%d).\n", r); 4522 goto fail; 4523 } 4524 4525 /* wait for the GPU to finish processing the IB */ 4526 r = dma_fence_wait(f, false); 4527 if (r) { 4528 DRM_ERROR("amdgpu: fence wait failed (%d).\n", r); 4529 goto fail; 4530 } 4531 4532 fail: 4533 amdgpu_ib_free(adev, &ib, NULL); 4534 dma_fence_put(f); 4535 4536 return r; 4537 } 4538 4539 static int gfx_v9_0_early_init(void *handle) 4540 { 4541 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4542 4543 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1) || 4544 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2)) 4545 adev->gfx.num_gfx_rings = 0; 4546 else 4547 adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS; 4548 adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev), 4549 AMDGPU_MAX_COMPUTE_RINGS); 4550 gfx_v9_0_set_kiq_pm4_funcs(adev); 4551 gfx_v9_0_set_ring_funcs(adev); 4552 gfx_v9_0_set_irq_funcs(adev); 4553 gfx_v9_0_set_gds_init(adev); 4554 gfx_v9_0_set_rlc_funcs(adev); 4555 4556 /* init rlcg reg access ctrl */ 4557 gfx_v9_0_init_rlcg_reg_access_ctrl(adev); 4558 4559 return 0; 4560 } 4561 4562 static int gfx_v9_0_ecc_late_init(void *handle) 4563 { 4564 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4565 int r; 4566 4567 /* 4568 * Temp workaround to fix the issue that CP firmware fails to 4569 * update read pointer when CPDMA is writing clearing operation 4570 * to GDS in suspend/resume sequence on several cards. So just 4571 * limit this operation in cold boot sequence. 4572 */ 4573 if ((!adev->in_suspend) && 4574 (adev->gds.gds_size)) { 4575 r = gfx_v9_0_do_edc_gds_workarounds(adev); 4576 if (r) 4577 return r; 4578 } 4579 4580 /* requires IBs so do in late init after IB pool is initialized */ 4581 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2)) 4582 r = gfx_v9_4_2_do_edc_gpr_workarounds(adev); 4583 else 4584 r = gfx_v9_0_do_edc_gpr_workarounds(adev); 4585 4586 if (r) 4587 return r; 4588 4589 if (adev->gfx.ras && 4590 adev->gfx.ras->enable_watchdog_timer) 4591 adev->gfx.ras->enable_watchdog_timer(adev); 4592 4593 return 0; 4594 } 4595 4596 static int gfx_v9_0_late_init(void *handle) 4597 { 4598 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4599 int r; 4600 4601 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0); 4602 if (r) 4603 return r; 4604 4605 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0); 4606 if (r) 4607 return r; 4608 4609 r = gfx_v9_0_ecc_late_init(handle); 4610 if (r) 4611 return r; 4612 4613 return 0; 4614 } 4615 4616 static bool gfx_v9_0_is_rlc_enabled(struct amdgpu_device *adev) 4617 { 4618 uint32_t rlc_setting; 4619 4620 /* if RLC is not enabled, do nothing */ 4621 rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL); 4622 if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK)) 4623 return false; 4624 4625 return true; 4626 } 4627 4628 static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev) 4629 { 4630 uint32_t data; 4631 unsigned i; 4632 4633 data = RLC_SAFE_MODE__CMD_MASK; 4634 data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); 4635 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); 4636 4637 /* wait for RLC_SAFE_MODE */ 4638 for (i = 0; i < adev->usec_timeout; i++) { 4639 if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) 4640 break; 4641 udelay(1); 4642 } 4643 } 4644 4645 static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev) 4646 { 4647 uint32_t data; 4648 4649 data = RLC_SAFE_MODE__CMD_MASK; 4650 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); 4651 } 4652 4653 static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev, 4654 bool enable) 4655 { 4656 amdgpu_gfx_rlc_enter_safe_mode(adev); 4657 4658 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) { 4659 gfx_v9_0_enable_gfx_cg_power_gating(adev, true); 4660 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE) 4661 gfx_v9_0_enable_gfx_pipeline_powergating(adev, true); 4662 } else { 4663 gfx_v9_0_enable_gfx_cg_power_gating(adev, false); 4664 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE) 4665 gfx_v9_0_enable_gfx_pipeline_powergating(adev, false); 4666 } 4667 4668 amdgpu_gfx_rlc_exit_safe_mode(adev); 4669 } 4670 4671 static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev, 4672 bool enable) 4673 { 4674 /* TODO: double check if we need to perform under safe mode */ 4675 /* gfx_v9_0_enter_rlc_safe_mode(adev); */ 4676 4677 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable) 4678 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, true); 4679 else 4680 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, false); 4681 4682 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable) 4683 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, true); 4684 else 4685 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, false); 4686 4687 /* gfx_v9_0_exit_rlc_safe_mode(adev); */ 4688 } 4689 4690 static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev, 4691 bool enable) 4692 { 4693 uint32_t data, def; 4694 4695 amdgpu_gfx_rlc_enter_safe_mode(adev); 4696 4697 /* It is disabled by HW by default */ 4698 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) { 4699 /* 1 - RLC_CGTT_MGCG_OVERRIDE */ 4700 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE); 4701 4702 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1)) 4703 data &= ~RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK; 4704 4705 data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK | 4706 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK | 4707 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK); 4708 4709 /* only for Vega10 & Raven1 */ 4710 data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK; 4711 4712 if (def != data) 4713 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data); 4714 4715 /* MGLS is a global flag to control all MGLS in GFX */ 4716 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) { 4717 /* 2 - RLC memory Light sleep */ 4718 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) { 4719 def = data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL); 4720 data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK; 4721 if (def != data) 4722 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data); 4723 } 4724 /* 3 - CP memory Light sleep */ 4725 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) { 4726 def = data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL); 4727 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 4728 if (def != data) 4729 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data); 4730 } 4731 } 4732 } else { 4733 /* 1 - MGCG_OVERRIDE */ 4734 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE); 4735 4736 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1)) 4737 data |= RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK; 4738 4739 data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK | 4740 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK | 4741 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK | 4742 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK); 4743 4744 if (def != data) 4745 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data); 4746 4747 /* 2 - disable MGLS in RLC */ 4748 data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL); 4749 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) { 4750 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK; 4751 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data); 4752 } 4753 4754 /* 3 - disable MGLS in CP */ 4755 data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL); 4756 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) { 4757 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 4758 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data); 4759 } 4760 } 4761 4762 amdgpu_gfx_rlc_exit_safe_mode(adev); 4763 } 4764 4765 static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev, 4766 bool enable) 4767 { 4768 uint32_t data, def; 4769 4770 if (!adev->gfx.num_gfx_rings) 4771 return; 4772 4773 amdgpu_gfx_rlc_enter_safe_mode(adev); 4774 4775 /* Enable 3D CGCG/CGLS */ 4776 if (enable) { 4777 /* write cmd to clear cgcg/cgls ov */ 4778 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE); 4779 /* unset CGCG override */ 4780 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK; 4781 /* update CGCG and CGLS override bits */ 4782 if (def != data) 4783 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data); 4784 4785 /* enable 3Dcgcg FSM(0x0000363f) */ 4786 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D); 4787 4788 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) 4789 data = (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) | 4790 RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK; 4791 else 4792 data = 0x0 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT; 4793 4794 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) 4795 data |= (0x000F << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) | 4796 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK; 4797 if (def != data) 4798 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data); 4799 4800 /* set IDLE_POLL_COUNT(0x00900100) */ 4801 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL); 4802 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) | 4803 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT); 4804 if (def != data) 4805 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data); 4806 } else { 4807 /* Disable CGCG/CGLS */ 4808 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D); 4809 /* disable cgcg, cgls should be disabled */ 4810 data &= ~(RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK | 4811 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK); 4812 /* disable cgcg and cgls in FSM */ 4813 if (def != data) 4814 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data); 4815 } 4816 4817 amdgpu_gfx_rlc_exit_safe_mode(adev); 4818 } 4819 4820 static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev, 4821 bool enable) 4822 { 4823 uint32_t def, data; 4824 4825 amdgpu_gfx_rlc_enter_safe_mode(adev); 4826 4827 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { 4828 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE); 4829 /* unset CGCG override */ 4830 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK; 4831 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) 4832 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK; 4833 else 4834 data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK; 4835 /* update CGCG and CGLS override bits */ 4836 if (def != data) 4837 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data); 4838 4839 /* enable cgcg FSM(0x0000363F) */ 4840 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL); 4841 4842 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1)) 4843 data = (0x2000 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) | 4844 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK; 4845 else 4846 data = (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) | 4847 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK; 4848 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) 4849 data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) | 4850 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK; 4851 if (def != data) 4852 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data); 4853 4854 /* set IDLE_POLL_COUNT(0x00900100) */ 4855 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL); 4856 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) | 4857 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT); 4858 if (def != data) 4859 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data); 4860 } else { 4861 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL); 4862 /* reset CGCG/CGLS bits */ 4863 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK); 4864 /* disable cgcg and cgls in FSM */ 4865 if (def != data) 4866 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data); 4867 } 4868 4869 amdgpu_gfx_rlc_exit_safe_mode(adev); 4870 } 4871 4872 static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev, 4873 bool enable) 4874 { 4875 if (enable) { 4876 /* CGCG/CGLS should be enabled after MGCG/MGLS 4877 * === MGCG + MGLS === 4878 */ 4879 gfx_v9_0_update_medium_grain_clock_gating(adev, enable); 4880 /* === CGCG /CGLS for GFX 3D Only === */ 4881 gfx_v9_0_update_3d_clock_gating(adev, enable); 4882 /* === CGCG + CGLS === */ 4883 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable); 4884 } else { 4885 /* CGCG/CGLS should be disabled before MGCG/MGLS 4886 * === CGCG + CGLS === 4887 */ 4888 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable); 4889 /* === CGCG /CGLS for GFX 3D Only === */ 4890 gfx_v9_0_update_3d_clock_gating(adev, enable); 4891 /* === MGCG + MGLS === */ 4892 gfx_v9_0_update_medium_grain_clock_gating(adev, enable); 4893 } 4894 return 0; 4895 } 4896 4897 static void gfx_v9_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid) 4898 { 4899 u32 reg, data; 4900 4901 amdgpu_gfx_off_ctrl(adev, false); 4902 4903 reg = SOC15_REG_OFFSET(GC, 0, mmRLC_SPM_MC_CNTL); 4904 if (amdgpu_sriov_is_pp_one_vf(adev)) 4905 data = RREG32_NO_KIQ(reg); 4906 else 4907 data = RREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL); 4908 4909 data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK; 4910 data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT; 4911 4912 if (amdgpu_sriov_is_pp_one_vf(adev)) 4913 WREG32_SOC15_NO_KIQ(GC, 0, mmRLC_SPM_MC_CNTL, data); 4914 else 4915 WREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL, data); 4916 4917 amdgpu_gfx_off_ctrl(adev, true); 4918 } 4919 4920 static bool gfx_v9_0_check_rlcg_range(struct amdgpu_device *adev, 4921 uint32_t offset, 4922 struct soc15_reg_rlcg *entries, int arr_size) 4923 { 4924 int i; 4925 uint32_t reg; 4926 4927 if (!entries) 4928 return false; 4929 4930 for (i = 0; i < arr_size; i++) { 4931 const struct soc15_reg_rlcg *entry; 4932 4933 entry = &entries[i]; 4934 reg = adev->reg_offset[entry->hwip][entry->instance][entry->segment] + entry->reg; 4935 if (offset == reg) 4936 return true; 4937 } 4938 4939 return false; 4940 } 4941 4942 static bool gfx_v9_0_is_rlcg_access_range(struct amdgpu_device *adev, u32 offset) 4943 { 4944 return gfx_v9_0_check_rlcg_range(adev, offset, 4945 (void *)rlcg_access_gc_9_0, 4946 ARRAY_SIZE(rlcg_access_gc_9_0)); 4947 } 4948 4949 static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = { 4950 .is_rlc_enabled = gfx_v9_0_is_rlc_enabled, 4951 .set_safe_mode = gfx_v9_0_set_safe_mode, 4952 .unset_safe_mode = gfx_v9_0_unset_safe_mode, 4953 .init = gfx_v9_0_rlc_init, 4954 .get_csb_size = gfx_v9_0_get_csb_size, 4955 .get_csb_buffer = gfx_v9_0_get_csb_buffer, 4956 .get_cp_table_num = gfx_v9_0_cp_jump_table_num, 4957 .resume = gfx_v9_0_rlc_resume, 4958 .stop = gfx_v9_0_rlc_stop, 4959 .reset = gfx_v9_0_rlc_reset, 4960 .start = gfx_v9_0_rlc_start, 4961 .update_spm_vmid = gfx_v9_0_update_spm_vmid, 4962 .is_rlcg_access_range = gfx_v9_0_is_rlcg_access_range, 4963 }; 4964 4965 static int gfx_v9_0_set_powergating_state(void *handle, 4966 enum amd_powergating_state state) 4967 { 4968 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4969 bool enable = (state == AMD_PG_STATE_GATE); 4970 4971 switch (adev->ip_versions[GC_HWIP][0]) { 4972 case IP_VERSION(9, 2, 2): 4973 case IP_VERSION(9, 1, 0): 4974 case IP_VERSION(9, 3, 0): 4975 if (!enable) 4976 amdgpu_gfx_off_ctrl(adev, false); 4977 4978 if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) { 4979 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, true); 4980 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, true); 4981 } else { 4982 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, false); 4983 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, false); 4984 } 4985 4986 if (adev->pg_flags & AMD_PG_SUPPORT_CP) 4987 gfx_v9_0_enable_cp_power_gating(adev, true); 4988 else 4989 gfx_v9_0_enable_cp_power_gating(adev, false); 4990 4991 /* update gfx cgpg state */ 4992 gfx_v9_0_update_gfx_cg_power_gating(adev, enable); 4993 4994 /* update mgcg state */ 4995 gfx_v9_0_update_gfx_mg_power_gating(adev, enable); 4996 4997 if (enable) 4998 amdgpu_gfx_off_ctrl(adev, true); 4999 break; 5000 case IP_VERSION(9, 2, 1): 5001 amdgpu_gfx_off_ctrl(adev, enable); 5002 break; 5003 default: 5004 break; 5005 } 5006 5007 return 0; 5008 } 5009 5010 static int gfx_v9_0_set_clockgating_state(void *handle, 5011 enum amd_clockgating_state state) 5012 { 5013 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 5014 5015 if (amdgpu_sriov_vf(adev)) 5016 return 0; 5017 5018 switch (adev->ip_versions[GC_HWIP][0]) { 5019 case IP_VERSION(9, 0, 1): 5020 case IP_VERSION(9, 2, 1): 5021 case IP_VERSION(9, 4, 0): 5022 case IP_VERSION(9, 2, 2): 5023 case IP_VERSION(9, 1, 0): 5024 case IP_VERSION(9, 4, 1): 5025 case IP_VERSION(9, 3, 0): 5026 case IP_VERSION(9, 4, 2): 5027 gfx_v9_0_update_gfx_clock_gating(adev, 5028 state == AMD_CG_STATE_GATE); 5029 break; 5030 default: 5031 break; 5032 } 5033 return 0; 5034 } 5035 5036 static void gfx_v9_0_get_clockgating_state(void *handle, u64 *flags) 5037 { 5038 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 5039 int data; 5040 5041 if (amdgpu_sriov_vf(adev)) 5042 *flags = 0; 5043 5044 /* AMD_CG_SUPPORT_GFX_MGCG */ 5045 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE)); 5046 if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK)) 5047 *flags |= AMD_CG_SUPPORT_GFX_MGCG; 5048 5049 /* AMD_CG_SUPPORT_GFX_CGCG */ 5050 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL)); 5051 if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK) 5052 *flags |= AMD_CG_SUPPORT_GFX_CGCG; 5053 5054 /* AMD_CG_SUPPORT_GFX_CGLS */ 5055 if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK) 5056 *flags |= AMD_CG_SUPPORT_GFX_CGLS; 5057 5058 /* AMD_CG_SUPPORT_GFX_RLC_LS */ 5059 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_MEM_SLP_CNTL)); 5060 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) 5061 *flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS; 5062 5063 /* AMD_CG_SUPPORT_GFX_CP_LS */ 5064 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmCP_MEM_SLP_CNTL)); 5065 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) 5066 *flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS; 5067 5068 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)) { 5069 /* AMD_CG_SUPPORT_GFX_3D_CGCG */ 5070 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D)); 5071 if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK) 5072 *flags |= AMD_CG_SUPPORT_GFX_3D_CGCG; 5073 5074 /* AMD_CG_SUPPORT_GFX_3D_CGLS */ 5075 if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK) 5076 *flags |= AMD_CG_SUPPORT_GFX_3D_CGLS; 5077 } 5078 } 5079 5080 static u64 gfx_v9_0_ring_get_rptr_gfx(struct amdgpu_ring *ring) 5081 { 5082 return *ring->rptr_cpu_addr; /* gfx9 is 32bit rptr*/ 5083 } 5084 5085 static u64 gfx_v9_0_ring_get_wptr_gfx(struct amdgpu_ring *ring) 5086 { 5087 struct amdgpu_device *adev = ring->adev; 5088 u64 wptr; 5089 5090 /* XXX check if swapping is necessary on BE */ 5091 if (ring->use_doorbell) { 5092 wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr); 5093 } else { 5094 wptr = RREG32_SOC15(GC, 0, mmCP_RB0_WPTR); 5095 wptr += (u64)RREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI) << 32; 5096 } 5097 5098 return wptr; 5099 } 5100 5101 static void gfx_v9_0_ring_set_wptr_gfx(struct amdgpu_ring *ring) 5102 { 5103 struct amdgpu_device *adev = ring->adev; 5104 5105 if (ring->use_doorbell) { 5106 /* XXX check if swapping is necessary on BE */ 5107 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr); 5108 WDOORBELL64(ring->doorbell_index, ring->wptr); 5109 } else { 5110 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr)); 5111 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr)); 5112 } 5113 } 5114 5115 static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) 5116 { 5117 struct amdgpu_device *adev = ring->adev; 5118 u32 ref_and_mask, reg_mem_engine; 5119 const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg; 5120 5121 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) { 5122 switch (ring->me) { 5123 case 1: 5124 ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe; 5125 break; 5126 case 2: 5127 ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe; 5128 break; 5129 default: 5130 return; 5131 } 5132 reg_mem_engine = 0; 5133 } else { 5134 ref_and_mask = nbio_hf_reg->ref_and_mask_cp0; 5135 reg_mem_engine = 1; /* pfp */ 5136 } 5137 5138 gfx_v9_0_wait_reg_mem(ring, reg_mem_engine, 0, 1, 5139 adev->nbio.funcs->get_hdp_flush_req_offset(adev), 5140 adev->nbio.funcs->get_hdp_flush_done_offset(adev), 5141 ref_and_mask, ref_and_mask, 0x20); 5142 } 5143 5144 static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, 5145 struct amdgpu_job *job, 5146 struct amdgpu_ib *ib, 5147 uint32_t flags) 5148 { 5149 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 5150 u32 header, control = 0; 5151 5152 if (ib->flags & AMDGPU_IB_FLAG_CE) 5153 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2); 5154 else 5155 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); 5156 5157 control |= ib->length_dw | (vmid << 24); 5158 5159 if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) { 5160 control |= INDIRECT_BUFFER_PRE_ENB(1); 5161 5162 if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid) 5163 gfx_v9_0_ring_emit_de_meta(ring); 5164 } 5165 5166 amdgpu_ring_write(ring, header); 5167 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */ 5168 amdgpu_ring_write(ring, 5169 #ifdef __BIG_ENDIAN 5170 (2 << 0) | 5171 #endif 5172 lower_32_bits(ib->gpu_addr)); 5173 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 5174 amdgpu_ring_write(ring, control); 5175 } 5176 5177 static void gfx_v9_0_ring_emit_ib_compute(struct amdgpu_ring *ring, 5178 struct amdgpu_job *job, 5179 struct amdgpu_ib *ib, 5180 uint32_t flags) 5181 { 5182 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 5183 u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); 5184 5185 /* Currently, there is a high possibility to get wave ID mismatch 5186 * between ME and GDS, leading to a hw deadlock, because ME generates 5187 * different wave IDs than the GDS expects. This situation happens 5188 * randomly when at least 5 compute pipes use GDS ordered append. 5189 * The wave IDs generated by ME are also wrong after suspend/resume. 5190 * Those are probably bugs somewhere else in the kernel driver. 5191 * 5192 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and 5193 * GDS to 0 for this ring (me/pipe). 5194 */ 5195 if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) { 5196 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 5197 amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID); 5198 amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id); 5199 } 5200 5201 amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); 5202 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */ 5203 amdgpu_ring_write(ring, 5204 #ifdef __BIG_ENDIAN 5205 (2 << 0) | 5206 #endif 5207 lower_32_bits(ib->gpu_addr)); 5208 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 5209 amdgpu_ring_write(ring, control); 5210 } 5211 5212 static void gfx_v9_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, 5213 u64 seq, unsigned flags) 5214 { 5215 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT; 5216 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT; 5217 bool writeback = flags & AMDGPU_FENCE_FLAG_TC_WB_ONLY; 5218 5219 /* RELEASE_MEM - flush caches, send int */ 5220 amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6)); 5221 amdgpu_ring_write(ring, ((writeback ? (EOP_TC_WB_ACTION_EN | 5222 EOP_TC_NC_ACTION_EN) : 5223 (EOP_TCL1_ACTION_EN | 5224 EOP_TC_ACTION_EN | 5225 EOP_TC_WB_ACTION_EN | 5226 EOP_TC_MD_ACTION_EN)) | 5227 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | 5228 EVENT_INDEX(5))); 5229 amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0)); 5230 5231 /* 5232 * the address should be Qword aligned if 64bit write, Dword 5233 * aligned if only send 32bit data low (discard data high) 5234 */ 5235 if (write64bit) 5236 BUG_ON(addr & 0x7); 5237 else 5238 BUG_ON(addr & 0x3); 5239 amdgpu_ring_write(ring, lower_32_bits(addr)); 5240 amdgpu_ring_write(ring, upper_32_bits(addr)); 5241 amdgpu_ring_write(ring, lower_32_bits(seq)); 5242 amdgpu_ring_write(ring, upper_32_bits(seq)); 5243 amdgpu_ring_write(ring, 0); 5244 } 5245 5246 static void gfx_v9_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring) 5247 { 5248 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 5249 uint32_t seq = ring->fence_drv.sync_seq; 5250 uint64_t addr = ring->fence_drv.gpu_addr; 5251 5252 gfx_v9_0_wait_reg_mem(ring, usepfp, 1, 0, 5253 lower_32_bits(addr), upper_32_bits(addr), 5254 seq, 0xffffffff, 4); 5255 } 5256 5257 static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring, 5258 unsigned vmid, uint64_t pd_addr) 5259 { 5260 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 5261 5262 /* compute doesn't have PFP */ 5263 if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) { 5264 /* sync PFP to ME, otherwise we might get invalid PFP reads */ 5265 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0)); 5266 amdgpu_ring_write(ring, 0x0); 5267 } 5268 } 5269 5270 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring) 5271 { 5272 return *ring->rptr_cpu_addr; /* gfx9 hardware is 32bit rptr */ 5273 } 5274 5275 static u64 gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring *ring) 5276 { 5277 u64 wptr; 5278 5279 /* XXX check if swapping is necessary on BE */ 5280 if (ring->use_doorbell) 5281 wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr); 5282 else 5283 BUG(); 5284 return wptr; 5285 } 5286 5287 static void gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring *ring) 5288 { 5289 struct amdgpu_device *adev = ring->adev; 5290 5291 /* XXX check if swapping is necessary on BE */ 5292 if (ring->use_doorbell) { 5293 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr); 5294 WDOORBELL64(ring->doorbell_index, ring->wptr); 5295 } else{ 5296 BUG(); /* only DOORBELL method supported on gfx9 now */ 5297 } 5298 } 5299 5300 static void gfx_v9_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr, 5301 u64 seq, unsigned int flags) 5302 { 5303 struct amdgpu_device *adev = ring->adev; 5304 5305 /* we only allocate 32bit for each seq wb address */ 5306 BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT); 5307 5308 /* write fence seq to the "addr" */ 5309 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5310 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 5311 WRITE_DATA_DST_SEL(5) | WR_CONFIRM)); 5312 amdgpu_ring_write(ring, lower_32_bits(addr)); 5313 amdgpu_ring_write(ring, upper_32_bits(addr)); 5314 amdgpu_ring_write(ring, lower_32_bits(seq)); 5315 5316 if (flags & AMDGPU_FENCE_FLAG_INT) { 5317 /* set register to trigger INT */ 5318 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5319 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 5320 WRITE_DATA_DST_SEL(0) | WR_CONFIRM)); 5321 amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, mmCPC_INT_STATUS)); 5322 amdgpu_ring_write(ring, 0); 5323 amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */ 5324 } 5325 } 5326 5327 static void gfx_v9_ring_emit_sb(struct amdgpu_ring *ring) 5328 { 5329 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 5330 amdgpu_ring_write(ring, 0); 5331 } 5332 5333 static void gfx_v9_0_ring_emit_ce_meta(struct amdgpu_ring *ring) 5334 { 5335 struct v9_ce_ib_state ce_payload = {0}; 5336 uint64_t csa_addr; 5337 int cnt; 5338 5339 cnt = (sizeof(ce_payload) >> 2) + 4 - 2; 5340 csa_addr = amdgpu_csa_vaddr(ring->adev); 5341 5342 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt)); 5343 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) | 5344 WRITE_DATA_DST_SEL(8) | 5345 WR_CONFIRM) | 5346 WRITE_DATA_CACHE_POLICY(0)); 5347 amdgpu_ring_write(ring, lower_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, ce_payload))); 5348 amdgpu_ring_write(ring, upper_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, ce_payload))); 5349 amdgpu_ring_write_multiple(ring, (void *)&ce_payload, sizeof(ce_payload) >> 2); 5350 } 5351 5352 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring) 5353 { 5354 struct v9_de_ib_state de_payload = {0}; 5355 uint64_t csa_addr, gds_addr; 5356 int cnt; 5357 5358 csa_addr = amdgpu_csa_vaddr(ring->adev); 5359 gds_addr = csa_addr + 4096; 5360 de_payload.gds_backup_addrlo = lower_32_bits(gds_addr); 5361 de_payload.gds_backup_addrhi = upper_32_bits(gds_addr); 5362 5363 cnt = (sizeof(de_payload) >> 2) + 4 - 2; 5364 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt)); 5365 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) | 5366 WRITE_DATA_DST_SEL(8) | 5367 WR_CONFIRM) | 5368 WRITE_DATA_CACHE_POLICY(0)); 5369 amdgpu_ring_write(ring, lower_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, de_payload))); 5370 amdgpu_ring_write(ring, upper_32_bits(csa_addr + offsetof(struct v9_gfx_meta_data, de_payload))); 5371 amdgpu_ring_write_multiple(ring, (void *)&de_payload, sizeof(de_payload) >> 2); 5372 } 5373 5374 static void gfx_v9_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start, 5375 bool secure) 5376 { 5377 uint32_t v = secure ? FRAME_TMZ : 0; 5378 5379 amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0)); 5380 amdgpu_ring_write(ring, v | FRAME_CMD(start ? 0 : 1)); 5381 } 5382 5383 static void gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags) 5384 { 5385 uint32_t dw2 = 0; 5386 5387 if (amdgpu_sriov_vf(ring->adev)) 5388 gfx_v9_0_ring_emit_ce_meta(ring); 5389 5390 dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */ 5391 if (flags & AMDGPU_HAVE_CTX_SWITCH) { 5392 /* set load_global_config & load_global_uconfig */ 5393 dw2 |= 0x8001; 5394 /* set load_cs_sh_regs */ 5395 dw2 |= 0x01000000; 5396 /* set load_per_context_state & load_gfx_sh_regs for GFX */ 5397 dw2 |= 0x10002; 5398 5399 /* set load_ce_ram if preamble presented */ 5400 if (AMDGPU_PREAMBLE_IB_PRESENT & flags) 5401 dw2 |= 0x10000000; 5402 } else { 5403 /* still load_ce_ram if this is the first time preamble presented 5404 * although there is no context switch happens. 5405 */ 5406 if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags) 5407 dw2 |= 0x10000000; 5408 } 5409 5410 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 5411 amdgpu_ring_write(ring, dw2); 5412 amdgpu_ring_write(ring, 0); 5413 } 5414 5415 static unsigned gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring) 5416 { 5417 unsigned ret; 5418 amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3)); 5419 amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr)); 5420 amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr)); 5421 amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */ 5422 ret = ring->wptr & ring->buf_mask; 5423 amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */ 5424 return ret; 5425 } 5426 5427 static void gfx_v9_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset) 5428 { 5429 unsigned cur; 5430 BUG_ON(offset > ring->buf_mask); 5431 BUG_ON(ring->ring[offset] != 0x55aa55aa); 5432 5433 cur = (ring->wptr - 1) & ring->buf_mask; 5434 if (likely(cur > offset)) 5435 ring->ring[offset] = cur - offset; 5436 else 5437 ring->ring[offset] = (ring->ring_size>>2) - offset + cur; 5438 } 5439 5440 static void gfx_v9_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg, 5441 uint32_t reg_val_offs) 5442 { 5443 struct amdgpu_device *adev = ring->adev; 5444 5445 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4)); 5446 amdgpu_ring_write(ring, 0 | /* src: register*/ 5447 (5 << 8) | /* dst: memory */ 5448 (1 << 20)); /* write confirm */ 5449 amdgpu_ring_write(ring, reg); 5450 amdgpu_ring_write(ring, 0); 5451 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr + 5452 reg_val_offs * 4)); 5453 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr + 5454 reg_val_offs * 4)); 5455 } 5456 5457 static void gfx_v9_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, 5458 uint32_t val) 5459 { 5460 uint32_t cmd = 0; 5461 5462 switch (ring->funcs->type) { 5463 case AMDGPU_RING_TYPE_GFX: 5464 cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM; 5465 break; 5466 case AMDGPU_RING_TYPE_KIQ: 5467 cmd = (1 << 16); /* no inc addr */ 5468 break; 5469 default: 5470 cmd = WR_CONFIRM; 5471 break; 5472 } 5473 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5474 amdgpu_ring_write(ring, cmd); 5475 amdgpu_ring_write(ring, reg); 5476 amdgpu_ring_write(ring, 0); 5477 amdgpu_ring_write(ring, val); 5478 } 5479 5480 static void gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg, 5481 uint32_t val, uint32_t mask) 5482 { 5483 gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20); 5484 } 5485 5486 static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring, 5487 uint32_t reg0, uint32_t reg1, 5488 uint32_t ref, uint32_t mask) 5489 { 5490 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 5491 struct amdgpu_device *adev = ring->adev; 5492 bool fw_version_ok = (ring->funcs->type == AMDGPU_RING_TYPE_GFX) ? 5493 adev->gfx.me_fw_write_wait : adev->gfx.mec_fw_write_wait; 5494 5495 if (fw_version_ok) 5496 gfx_v9_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1, 5497 ref, mask, 0x20); 5498 else 5499 amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1, 5500 ref, mask); 5501 } 5502 5503 static void gfx_v9_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid) 5504 { 5505 struct amdgpu_device *adev = ring->adev; 5506 uint32_t value = 0; 5507 5508 value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03); 5509 value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01); 5510 value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1); 5511 value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid); 5512 WREG32_SOC15(GC, 0, mmSQ_CMD, value); 5513 } 5514 5515 static void gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev, 5516 enum amdgpu_interrupt_state state) 5517 { 5518 switch (state) { 5519 case AMDGPU_IRQ_STATE_DISABLE: 5520 case AMDGPU_IRQ_STATE_ENABLE: 5521 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0, 5522 TIME_STAMP_INT_ENABLE, 5523 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); 5524 break; 5525 default: 5526 break; 5527 } 5528 } 5529 5530 static void gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev, 5531 int me, int pipe, 5532 enum amdgpu_interrupt_state state) 5533 { 5534 u32 mec_int_cntl, mec_int_cntl_reg; 5535 5536 /* 5537 * amdgpu controls only the first MEC. That's why this function only 5538 * handles the setting of interrupts for this specific MEC. All other 5539 * pipes' interrupts are set by amdkfd. 5540 */ 5541 5542 if (me == 1) { 5543 switch (pipe) { 5544 case 0: 5545 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL); 5546 break; 5547 case 1: 5548 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL); 5549 break; 5550 case 2: 5551 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL); 5552 break; 5553 case 3: 5554 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL); 5555 break; 5556 default: 5557 DRM_DEBUG("invalid pipe %d\n", pipe); 5558 return; 5559 } 5560 } else { 5561 DRM_DEBUG("invalid me %d\n", me); 5562 return; 5563 } 5564 5565 switch (state) { 5566 case AMDGPU_IRQ_STATE_DISABLE: 5567 mec_int_cntl = RREG32_SOC15_IP(GC,mec_int_cntl_reg); 5568 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL, 5569 TIME_STAMP_INT_ENABLE, 0); 5570 WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl); 5571 break; 5572 case AMDGPU_IRQ_STATE_ENABLE: 5573 mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg); 5574 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL, 5575 TIME_STAMP_INT_ENABLE, 1); 5576 WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl); 5577 break; 5578 default: 5579 break; 5580 } 5581 } 5582 5583 static int gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device *adev, 5584 struct amdgpu_irq_src *source, 5585 unsigned type, 5586 enum amdgpu_interrupt_state state) 5587 { 5588 switch (state) { 5589 case AMDGPU_IRQ_STATE_DISABLE: 5590 case AMDGPU_IRQ_STATE_ENABLE: 5591 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0, 5592 PRIV_REG_INT_ENABLE, 5593 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); 5594 break; 5595 default: 5596 break; 5597 } 5598 5599 return 0; 5600 } 5601 5602 static int gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device *adev, 5603 struct amdgpu_irq_src *source, 5604 unsigned type, 5605 enum amdgpu_interrupt_state state) 5606 { 5607 switch (state) { 5608 case AMDGPU_IRQ_STATE_DISABLE: 5609 case AMDGPU_IRQ_STATE_ENABLE: 5610 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0, 5611 PRIV_INSTR_INT_ENABLE, 5612 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); 5613 break; 5614 default: 5615 break; 5616 } 5617 5618 return 0; 5619 } 5620 5621 #define ENABLE_ECC_ON_ME_PIPE(me, pipe) \ 5622 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\ 5623 CP_ECC_ERROR_INT_ENABLE, 1) 5624 5625 #define DISABLE_ECC_ON_ME_PIPE(me, pipe) \ 5626 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\ 5627 CP_ECC_ERROR_INT_ENABLE, 0) 5628 5629 static int gfx_v9_0_set_cp_ecc_error_state(struct amdgpu_device *adev, 5630 struct amdgpu_irq_src *source, 5631 unsigned type, 5632 enum amdgpu_interrupt_state state) 5633 { 5634 switch (state) { 5635 case AMDGPU_IRQ_STATE_DISABLE: 5636 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0, 5637 CP_ECC_ERROR_INT_ENABLE, 0); 5638 DISABLE_ECC_ON_ME_PIPE(1, 0); 5639 DISABLE_ECC_ON_ME_PIPE(1, 1); 5640 DISABLE_ECC_ON_ME_PIPE(1, 2); 5641 DISABLE_ECC_ON_ME_PIPE(1, 3); 5642 break; 5643 5644 case AMDGPU_IRQ_STATE_ENABLE: 5645 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0, 5646 CP_ECC_ERROR_INT_ENABLE, 1); 5647 ENABLE_ECC_ON_ME_PIPE(1, 0); 5648 ENABLE_ECC_ON_ME_PIPE(1, 1); 5649 ENABLE_ECC_ON_ME_PIPE(1, 2); 5650 ENABLE_ECC_ON_ME_PIPE(1, 3); 5651 break; 5652 default: 5653 break; 5654 } 5655 5656 return 0; 5657 } 5658 5659 5660 static int gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device *adev, 5661 struct amdgpu_irq_src *src, 5662 unsigned type, 5663 enum amdgpu_interrupt_state state) 5664 { 5665 switch (type) { 5666 case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP: 5667 gfx_v9_0_set_gfx_eop_interrupt_state(adev, state); 5668 break; 5669 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP: 5670 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 0, state); 5671 break; 5672 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP: 5673 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 1, state); 5674 break; 5675 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP: 5676 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 2, state); 5677 break; 5678 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP: 5679 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 3, state); 5680 break; 5681 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP: 5682 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 0, state); 5683 break; 5684 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP: 5685 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 1, state); 5686 break; 5687 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP: 5688 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 2, state); 5689 break; 5690 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP: 5691 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 3, state); 5692 break; 5693 default: 5694 break; 5695 } 5696 return 0; 5697 } 5698 5699 static int gfx_v9_0_eop_irq(struct amdgpu_device *adev, 5700 struct amdgpu_irq_src *source, 5701 struct amdgpu_iv_entry *entry) 5702 { 5703 int i; 5704 u8 me_id, pipe_id, queue_id; 5705 struct amdgpu_ring *ring; 5706 5707 DRM_DEBUG("IH: CP EOP\n"); 5708 me_id = (entry->ring_id & 0x0c) >> 2; 5709 pipe_id = (entry->ring_id & 0x03) >> 0; 5710 queue_id = (entry->ring_id & 0x70) >> 4; 5711 5712 switch (me_id) { 5713 case 0: 5714 amdgpu_fence_process(&adev->gfx.gfx_ring[0]); 5715 break; 5716 case 1: 5717 case 2: 5718 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 5719 ring = &adev->gfx.compute_ring[i]; 5720 /* Per-queue interrupt is supported for MEC starting from VI. 5721 * The interrupt can only be enabled/disabled per pipe instead of per queue. 5722 */ 5723 if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id)) 5724 amdgpu_fence_process(ring); 5725 } 5726 break; 5727 } 5728 return 0; 5729 } 5730 5731 static void gfx_v9_0_fault(struct amdgpu_device *adev, 5732 struct amdgpu_iv_entry *entry) 5733 { 5734 u8 me_id, pipe_id, queue_id; 5735 struct amdgpu_ring *ring; 5736 int i; 5737 5738 me_id = (entry->ring_id & 0x0c) >> 2; 5739 pipe_id = (entry->ring_id & 0x03) >> 0; 5740 queue_id = (entry->ring_id & 0x70) >> 4; 5741 5742 switch (me_id) { 5743 case 0: 5744 drm_sched_fault(&adev->gfx.gfx_ring[0].sched); 5745 break; 5746 case 1: 5747 case 2: 5748 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 5749 ring = &adev->gfx.compute_ring[i]; 5750 if (ring->me == me_id && ring->pipe == pipe_id && 5751 ring->queue == queue_id) 5752 drm_sched_fault(&ring->sched); 5753 } 5754 break; 5755 } 5756 } 5757 5758 static int gfx_v9_0_priv_reg_irq(struct amdgpu_device *adev, 5759 struct amdgpu_irq_src *source, 5760 struct amdgpu_iv_entry *entry) 5761 { 5762 DRM_ERROR("Illegal register access in command stream\n"); 5763 gfx_v9_0_fault(adev, entry); 5764 return 0; 5765 } 5766 5767 static int gfx_v9_0_priv_inst_irq(struct amdgpu_device *adev, 5768 struct amdgpu_irq_src *source, 5769 struct amdgpu_iv_entry *entry) 5770 { 5771 DRM_ERROR("Illegal instruction in command stream\n"); 5772 gfx_v9_0_fault(adev, entry); 5773 return 0; 5774 } 5775 5776 5777 static const struct soc15_ras_field_entry gfx_v9_0_ras_fields[] = { 5778 { "CPC_SCRATCH", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT), 5779 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, SEC_COUNT), 5780 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, DED_COUNT) 5781 }, 5782 { "CPC_UCODE", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT), 5783 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, SEC_COUNT), 5784 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, DED_COUNT) 5785 }, 5786 { "CPF_ROQ_ME1", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 5787 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME1), 5788 0, 0 5789 }, 5790 { "CPF_ROQ_ME2", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 5791 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME2), 5792 0, 0 5793 }, 5794 { "CPF_TAG", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT), 5795 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, SEC_COUNT), 5796 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, DED_COUNT) 5797 }, 5798 { "CPG_DMA_ROQ", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 5799 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, ROQ_COUNT), 5800 0, 0 5801 }, 5802 { "CPG_DMA_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 5803 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_SEC_COUNT), 5804 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_DED_COUNT) 5805 }, 5806 { "CPG_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT), 5807 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, SEC_COUNT), 5808 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, DED_COUNT) 5809 }, 5810 { "DC_CSINVOC", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT), 5811 SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, COUNT_ME1), 5812 0, 0 5813 }, 5814 { "DC_RESTORE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT), 5815 SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, COUNT_ME1), 5816 0, 0 5817 }, 5818 { "DC_STATE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT), 5819 SOC15_REG_FIELD(DC_EDC_STATE_CNT, COUNT_ME1), 5820 0, 0 5821 }, 5822 { "GDS_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 5823 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_SEC), 5824 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_DED) 5825 }, 5826 { "GDS_INPUT_QUEUE", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 5827 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_INPUT_QUEUE_SED), 5828 0, 0 5829 }, 5830 { "GDS_ME0_CS_PIPE_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 5831 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_SEC), 5832 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_DED) 5833 }, 5834 { "GDS_OA_PHY_PHY_CMD_RAM_MEM", 5835 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 5836 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_SEC), 5837 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_DED) 5838 }, 5839 { "GDS_OA_PHY_PHY_DATA_RAM_MEM", 5840 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 5841 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_DATA_RAM_MEM_SED), 5842 0, 0 5843 }, 5844 { "GDS_OA_PIPE_ME1_PIPE0_PIPE_MEM", 5845 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 5846 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_SEC), 5847 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_DED) 5848 }, 5849 { "GDS_OA_PIPE_ME1_PIPE1_PIPE_MEM", 5850 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 5851 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_SEC), 5852 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_DED) 5853 }, 5854 { "GDS_OA_PIPE_ME1_PIPE2_PIPE_MEM", 5855 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 5856 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_SEC), 5857 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_DED) 5858 }, 5859 { "GDS_OA_PIPE_ME1_PIPE3_PIPE_MEM", 5860 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 5861 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_SEC), 5862 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_DED) 5863 }, 5864 { "SPI_SR_MEM", SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT), 5865 SOC15_REG_FIELD(SPI_EDC_CNT, SPI_SR_MEM_SED_COUNT), 5866 0, 0 5867 }, 5868 { "TA_FS_DFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 5869 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_SEC_COUNT), 5870 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_DED_COUNT) 5871 }, 5872 { "TA_FS_AFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 5873 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_SED_COUNT), 5874 0, 0 5875 }, 5876 { "TA_FL_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 5877 SOC15_REG_FIELD(TA_EDC_CNT, TA_FL_LFIFO_SED_COUNT), 5878 0, 0 5879 }, 5880 { "TA_FX_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 5881 SOC15_REG_FIELD(TA_EDC_CNT, TA_FX_LFIFO_SED_COUNT), 5882 0, 0 5883 }, 5884 { "TA_FS_CFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 5885 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_CFIFO_SED_COUNT), 5886 0, 0 5887 }, 5888 { "TCA_HOLE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 5889 SOC15_REG_FIELD(TCA_EDC_CNT, HOLE_FIFO_SED_COUNT), 5890 0, 0 5891 }, 5892 { "TCA_REQ_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 5893 SOC15_REG_FIELD(TCA_EDC_CNT, REQ_FIFO_SED_COUNT), 5894 0, 0 5895 }, 5896 { "TCC_CACHE_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5897 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_SEC_COUNT), 5898 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_DED_COUNT) 5899 }, 5900 { "TCC_CACHE_DIRTY", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5901 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_SEC_COUNT), 5902 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_DED_COUNT) 5903 }, 5904 { "TCC_HIGH_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5905 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_SEC_COUNT), 5906 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_DED_COUNT) 5907 }, 5908 { "TCC_LOW_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5909 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_SEC_COUNT), 5910 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_DED_COUNT) 5911 }, 5912 { "TCC_SRC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5913 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_SEC_COUNT), 5914 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_DED_COUNT) 5915 }, 5916 { "TCC_IN_USE_DEC", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5917 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_DEC_SED_COUNT), 5918 0, 0 5919 }, 5920 { "TCC_IN_USE_TRANSFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5921 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_TRANSFER_SED_COUNT), 5922 0, 0 5923 }, 5924 { "TCC_LATENCY_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5925 SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_SED_COUNT), 5926 0, 0 5927 }, 5928 { "TCC_RETURN_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5929 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_DATA_SED_COUNT), 5930 0, 0 5931 }, 5932 { "TCC_RETURN_CONTROL", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5933 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_CONTROL_SED_COUNT), 5934 0, 0 5935 }, 5936 { "TCC_UC_ATOMIC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5937 SOC15_REG_FIELD(TCC_EDC_CNT, UC_ATOMIC_FIFO_SED_COUNT), 5938 0, 0 5939 }, 5940 { "TCC_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 5941 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_RETURN_SED_COUNT), 5942 0, 0 5943 }, 5944 { "TCC_WRITE_CACHE_READ", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 5945 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_CACHE_READ_SED_COUNT), 5946 0, 0 5947 }, 5948 { "TCC_SRC_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 5949 SOC15_REG_FIELD(TCC_EDC_CNT2, SRC_FIFO_NEXT_RAM_SED_COUNT), 5950 0, 0 5951 }, 5952 { "TCC_LATENCY_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 5953 SOC15_REG_FIELD(TCC_EDC_CNT2, LATENCY_FIFO_NEXT_RAM_SED_COUNT), 5954 0, 0 5955 }, 5956 { "TCC_CACHE_TAG_PROBE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 5957 SOC15_REG_FIELD(TCC_EDC_CNT2, CACHE_TAG_PROBE_FIFO_SED_COUNT), 5958 0, 0 5959 }, 5960 { "TCC_WRRET_TAG_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 5961 SOC15_REG_FIELD(TCC_EDC_CNT2, WRRET_TAG_WRITE_RETURN_SED_COUNT), 5962 0, 0 5963 }, 5964 { "TCC_ATOMIC_RETURN_BUFFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 5965 SOC15_REG_FIELD(TCC_EDC_CNT2, ATOMIC_RETURN_BUFFER_SED_COUNT), 5966 0, 0 5967 }, 5968 { "TCI_WRITE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT), 5969 SOC15_REG_FIELD(TCI_EDC_CNT, WRITE_RAM_SED_COUNT), 5970 0, 0 5971 }, 5972 { "TCP_CACHE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 5973 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_SEC_COUNT), 5974 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_DED_COUNT) 5975 }, 5976 { "TCP_LFIFO_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 5977 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_SEC_COUNT), 5978 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_DED_COUNT) 5979 }, 5980 { "TCP_CMD_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 5981 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CMD_FIFO_SED_COUNT), 5982 0, 0 5983 }, 5984 { "TCP_VM_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 5985 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, VM_FIFO_SEC_COUNT), 5986 0, 0 5987 }, 5988 { "TCP_DB_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 5989 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, DB_RAM_SED_COUNT), 5990 0, 0 5991 }, 5992 { "TCP_UTCL1_LFIFO0", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 5993 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_SEC_COUNT), 5994 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_DED_COUNT) 5995 }, 5996 { "TCP_UTCL1_LFIFO1", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 5997 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_SEC_COUNT), 5998 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_DED_COUNT) 5999 }, 6000 { "TD_SS_FIFO_LO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 6001 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_SEC_COUNT), 6002 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_DED_COUNT) 6003 }, 6004 { "TD_SS_FIFO_HI", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 6005 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_SEC_COUNT), 6006 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_DED_COUNT) 6007 }, 6008 { "TD_CS_FIFO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 6009 SOC15_REG_FIELD(TD_EDC_CNT, CS_FIFO_SED_COUNT), 6010 0, 0 6011 }, 6012 { "SQ_LDS_D", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6013 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_SEC_COUNT), 6014 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_DED_COUNT) 6015 }, 6016 { "SQ_LDS_I", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6017 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_SEC_COUNT), 6018 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_DED_COUNT) 6019 }, 6020 { "SQ_SGPR", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6021 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_SEC_COUNT), 6022 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_DED_COUNT) 6023 }, 6024 { "SQ_VGPR0", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6025 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_SEC_COUNT), 6026 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_DED_COUNT) 6027 }, 6028 { "SQ_VGPR1", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6029 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_SEC_COUNT), 6030 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_DED_COUNT) 6031 }, 6032 { "SQ_VGPR2", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6033 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_SEC_COUNT), 6034 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_DED_COUNT) 6035 }, 6036 { "SQ_VGPR3", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6037 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_SEC_COUNT), 6038 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_DED_COUNT) 6039 }, 6040 { "SQC_DATA_CU0_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6041 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_SEC_COUNT), 6042 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_DED_COUNT) 6043 }, 6044 { "SQC_DATA_CU0_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6045 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_SEC_COUNT), 6046 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_DED_COUNT) 6047 }, 6048 { "SQC_DATA_CU1_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6049 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_SEC_COUNT), 6050 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_DED_COUNT) 6051 }, 6052 { "SQC_DATA_CU1_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6053 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_SEC_COUNT), 6054 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_DED_COUNT) 6055 }, 6056 { "SQC_DATA_CU2_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6057 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_SEC_COUNT), 6058 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_DED_COUNT) 6059 }, 6060 { "SQC_DATA_CU2_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6061 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_SEC_COUNT), 6062 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_DED_COUNT) 6063 }, 6064 { "SQC_INST_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6065 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_SEC_COUNT), 6066 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_DED_COUNT) 6067 }, 6068 { "SQC_INST_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6069 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_SEC_COUNT), 6070 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_DED_COUNT) 6071 }, 6072 { "SQC_DATA_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6073 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_SEC_COUNT), 6074 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_DED_COUNT) 6075 }, 6076 { "SQC_DATA_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6077 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_SEC_COUNT), 6078 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_DED_COUNT) 6079 }, 6080 { "SQC_INST_BANKA_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6081 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_UTCL1_MISS_FIFO_SED_COUNT), 6082 0, 0 6083 }, 6084 { "SQC_INST_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6085 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_MISS_FIFO_SED_COUNT), 6086 0, 0 6087 }, 6088 { "SQC_DATA_BANKA_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6089 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_HIT_FIFO_SED_COUNT), 6090 0, 0 6091 }, 6092 { "SQC_DATA_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6093 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_MISS_FIFO_SED_COUNT), 6094 0, 0 6095 }, 6096 { "SQC_DATA_BANKA_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6097 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_DIRTY_BIT_RAM_SED_COUNT), 6098 0, 0 6099 }, 6100 { "SQC_INST_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6101 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_SEC_COUNT), 6102 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_DED_COUNT) 6103 }, 6104 { "SQC_INST_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6105 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_SEC_COUNT), 6106 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_DED_COUNT) 6107 }, 6108 { "SQC_INST_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6109 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_SEC_COUNT), 6110 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_DED_COUNT) 6111 }, 6112 { "SQC_DATA_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6113 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_SEC_COUNT), 6114 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_DED_COUNT) 6115 }, 6116 { "SQC_DATA_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6117 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_SEC_COUNT), 6118 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_DED_COUNT) 6119 }, 6120 { "SQC_INST_BANKB_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6121 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_UTCL1_MISS_FIFO_SED_COUNT), 6122 0, 0 6123 }, 6124 { "SQC_INST_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6125 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_MISS_FIFO_SED_COUNT), 6126 0, 0 6127 }, 6128 { "SQC_DATA_BANKB_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6129 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_HIT_FIFO_SED_COUNT), 6130 0, 0 6131 }, 6132 { "SQC_DATA_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6133 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_MISS_FIFO_SED_COUNT), 6134 0, 0 6135 }, 6136 { "SQC_DATA_BANKB_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6137 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_DIRTY_BIT_RAM_SED_COUNT), 6138 0, 0 6139 }, 6140 { "EA_DRAMRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6141 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT), 6142 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT) 6143 }, 6144 { "EA_DRAMWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6145 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT), 6146 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT) 6147 }, 6148 { "EA_DRAMWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6149 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT), 6150 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT) 6151 }, 6152 { "EA_RRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6153 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_SEC_COUNT), 6154 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_DED_COUNT) 6155 }, 6156 { "EA_WRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6157 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_SEC_COUNT), 6158 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_DED_COUNT) 6159 }, 6160 { "EA_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6161 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT), 6162 0, 0 6163 }, 6164 { "EA_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6165 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT), 6166 0, 0 6167 }, 6168 { "EA_IORD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6169 SOC15_REG_FIELD(GCEA_EDC_CNT, IORD_CMDMEM_SED_COUNT), 6170 0, 0 6171 }, 6172 { "EA_IOWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6173 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_CMDMEM_SED_COUNT), 6174 0, 0 6175 }, 6176 { "EA_IOWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6177 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_DATAMEM_SED_COUNT), 6178 0, 0 6179 }, 6180 { "GMIRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6181 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT), 6182 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT) 6183 }, 6184 { "GMIWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6185 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT), 6186 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT) 6187 }, 6188 { "GMIWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6189 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT), 6190 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT) 6191 }, 6192 { "GMIRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6193 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT), 6194 0, 0 6195 }, 6196 { "GMIWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6197 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT), 6198 0, 0 6199 }, 6200 { "MAM_D0MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6201 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D0MEM_SED_COUNT), 6202 0, 0 6203 }, 6204 { "MAM_D1MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6205 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D1MEM_SED_COUNT), 6206 0, 0 6207 }, 6208 { "MAM_D2MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6209 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D2MEM_SED_COUNT), 6210 0, 0 6211 }, 6212 { "MAM_D3MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6213 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D3MEM_SED_COUNT), 6214 0, 0 6215 } 6216 }; 6217 6218 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev, 6219 void *inject_if) 6220 { 6221 struct ras_inject_if *info = (struct ras_inject_if *)inject_if; 6222 int ret; 6223 struct ta_ras_trigger_error_input block_info = { 0 }; 6224 6225 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) 6226 return -EINVAL; 6227 6228 if (info->head.sub_block_index >= ARRAY_SIZE(ras_gfx_subblocks)) 6229 return -EINVAL; 6230 6231 if (!ras_gfx_subblocks[info->head.sub_block_index].name) 6232 return -EPERM; 6233 6234 if (!(ras_gfx_subblocks[info->head.sub_block_index].hw_supported_error_type & 6235 info->head.type)) { 6236 DRM_ERROR("GFX Subblock %s, hardware do not support type 0x%x\n", 6237 ras_gfx_subblocks[info->head.sub_block_index].name, 6238 info->head.type); 6239 return -EPERM; 6240 } 6241 6242 if (!(ras_gfx_subblocks[info->head.sub_block_index].sw_supported_error_type & 6243 info->head.type)) { 6244 DRM_ERROR("GFX Subblock %s, driver do not support type 0x%x\n", 6245 ras_gfx_subblocks[info->head.sub_block_index].name, 6246 info->head.type); 6247 return -EPERM; 6248 } 6249 6250 block_info.block_id = amdgpu_ras_block_to_ta(info->head.block); 6251 block_info.sub_block_index = 6252 ras_gfx_subblocks[info->head.sub_block_index].ta_subblock; 6253 block_info.inject_error_type = amdgpu_ras_error_to_ta(info->head.type); 6254 block_info.address = info->address; 6255 block_info.value = info->value; 6256 6257 mutex_lock(&adev->grbm_idx_mutex); 6258 ret = psp_ras_trigger_error(&adev->psp, &block_info); 6259 mutex_unlock(&adev->grbm_idx_mutex); 6260 6261 return ret; 6262 } 6263 6264 static const char *vml2_mems[] = { 6265 "UTC_VML2_BANK_CACHE_0_BIGK_MEM0", 6266 "UTC_VML2_BANK_CACHE_0_BIGK_MEM1", 6267 "UTC_VML2_BANK_CACHE_0_4K_MEM0", 6268 "UTC_VML2_BANK_CACHE_0_4K_MEM1", 6269 "UTC_VML2_BANK_CACHE_1_BIGK_MEM0", 6270 "UTC_VML2_BANK_CACHE_1_BIGK_MEM1", 6271 "UTC_VML2_BANK_CACHE_1_4K_MEM0", 6272 "UTC_VML2_BANK_CACHE_1_4K_MEM1", 6273 "UTC_VML2_BANK_CACHE_2_BIGK_MEM0", 6274 "UTC_VML2_BANK_CACHE_2_BIGK_MEM1", 6275 "UTC_VML2_BANK_CACHE_2_4K_MEM0", 6276 "UTC_VML2_BANK_CACHE_2_4K_MEM1", 6277 "UTC_VML2_BANK_CACHE_3_BIGK_MEM0", 6278 "UTC_VML2_BANK_CACHE_3_BIGK_MEM1", 6279 "UTC_VML2_BANK_CACHE_3_4K_MEM0", 6280 "UTC_VML2_BANK_CACHE_3_4K_MEM1", 6281 }; 6282 6283 static const char *vml2_walker_mems[] = { 6284 "UTC_VML2_CACHE_PDE0_MEM0", 6285 "UTC_VML2_CACHE_PDE0_MEM1", 6286 "UTC_VML2_CACHE_PDE1_MEM0", 6287 "UTC_VML2_CACHE_PDE1_MEM1", 6288 "UTC_VML2_CACHE_PDE2_MEM0", 6289 "UTC_VML2_CACHE_PDE2_MEM1", 6290 "UTC_VML2_RDIF_LOG_FIFO", 6291 }; 6292 6293 static const char *atc_l2_cache_2m_mems[] = { 6294 "UTC_ATCL2_CACHE_2M_BANK0_WAY0_MEM", 6295 "UTC_ATCL2_CACHE_2M_BANK0_WAY1_MEM", 6296 "UTC_ATCL2_CACHE_2M_BANK1_WAY0_MEM", 6297 "UTC_ATCL2_CACHE_2M_BANK1_WAY1_MEM", 6298 }; 6299 6300 static const char *atc_l2_cache_4k_mems[] = { 6301 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM0", 6302 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM1", 6303 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM2", 6304 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM3", 6305 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM4", 6306 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM5", 6307 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM6", 6308 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM7", 6309 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM0", 6310 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM1", 6311 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM2", 6312 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM3", 6313 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM4", 6314 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM5", 6315 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM6", 6316 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM7", 6317 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM0", 6318 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM1", 6319 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM2", 6320 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM3", 6321 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM4", 6322 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM5", 6323 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM6", 6324 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM7", 6325 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM0", 6326 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM1", 6327 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM2", 6328 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM3", 6329 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM4", 6330 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM5", 6331 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM6", 6332 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM7", 6333 }; 6334 6335 static int gfx_v9_0_query_utc_edc_status(struct amdgpu_device *adev, 6336 struct ras_err_data *err_data) 6337 { 6338 uint32_t i, data; 6339 uint32_t sec_count, ded_count; 6340 6341 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255); 6342 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0); 6343 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255); 6344 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0); 6345 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255); 6346 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0); 6347 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255); 6348 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0); 6349 6350 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) { 6351 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i); 6352 data = RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT); 6353 6354 sec_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, SEC_COUNT); 6355 if (sec_count) { 6356 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6357 "SEC %d\n", i, vml2_mems[i], sec_count); 6358 err_data->ce_count += sec_count; 6359 } 6360 6361 ded_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, DED_COUNT); 6362 if (ded_count) { 6363 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6364 "DED %d\n", i, vml2_mems[i], ded_count); 6365 err_data->ue_count += ded_count; 6366 } 6367 } 6368 6369 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) { 6370 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i); 6371 data = RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT); 6372 6373 sec_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT, 6374 SEC_COUNT); 6375 if (sec_count) { 6376 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6377 "SEC %d\n", i, vml2_walker_mems[i], sec_count); 6378 err_data->ce_count += sec_count; 6379 } 6380 6381 ded_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT, 6382 DED_COUNT); 6383 if (ded_count) { 6384 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6385 "DED %d\n", i, vml2_walker_mems[i], ded_count); 6386 err_data->ue_count += ded_count; 6387 } 6388 } 6389 6390 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) { 6391 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i); 6392 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT); 6393 6394 sec_count = (data & 0x00006000L) >> 0xd; 6395 if (sec_count) { 6396 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6397 "SEC %d\n", i, atc_l2_cache_2m_mems[i], 6398 sec_count); 6399 err_data->ce_count += sec_count; 6400 } 6401 } 6402 6403 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) { 6404 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i); 6405 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT); 6406 6407 sec_count = (data & 0x00006000L) >> 0xd; 6408 if (sec_count) { 6409 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6410 "SEC %d\n", i, atc_l2_cache_4k_mems[i], 6411 sec_count); 6412 err_data->ce_count += sec_count; 6413 } 6414 6415 ded_count = (data & 0x00018000L) >> 0xf; 6416 if (ded_count) { 6417 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6418 "DED %d\n", i, atc_l2_cache_4k_mems[i], 6419 ded_count); 6420 err_data->ue_count += ded_count; 6421 } 6422 } 6423 6424 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255); 6425 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255); 6426 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255); 6427 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255); 6428 6429 return 0; 6430 } 6431 6432 static int gfx_v9_0_ras_error_count(struct amdgpu_device *adev, 6433 const struct soc15_reg_entry *reg, 6434 uint32_t se_id, uint32_t inst_id, uint32_t value, 6435 uint32_t *sec_count, uint32_t *ded_count) 6436 { 6437 uint32_t i; 6438 uint32_t sec_cnt, ded_cnt; 6439 6440 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_ras_fields); i++) { 6441 if(gfx_v9_0_ras_fields[i].reg_offset != reg->reg_offset || 6442 gfx_v9_0_ras_fields[i].seg != reg->seg || 6443 gfx_v9_0_ras_fields[i].inst != reg->inst) 6444 continue; 6445 6446 sec_cnt = (value & 6447 gfx_v9_0_ras_fields[i].sec_count_mask) >> 6448 gfx_v9_0_ras_fields[i].sec_count_shift; 6449 if (sec_cnt) { 6450 dev_info(adev->dev, "GFX SubBlock %s, " 6451 "Instance[%d][%d], SEC %d\n", 6452 gfx_v9_0_ras_fields[i].name, 6453 se_id, inst_id, 6454 sec_cnt); 6455 *sec_count += sec_cnt; 6456 } 6457 6458 ded_cnt = (value & 6459 gfx_v9_0_ras_fields[i].ded_count_mask) >> 6460 gfx_v9_0_ras_fields[i].ded_count_shift; 6461 if (ded_cnt) { 6462 dev_info(adev->dev, "GFX SubBlock %s, " 6463 "Instance[%d][%d], DED %d\n", 6464 gfx_v9_0_ras_fields[i].name, 6465 se_id, inst_id, 6466 ded_cnt); 6467 *ded_count += ded_cnt; 6468 } 6469 } 6470 6471 return 0; 6472 } 6473 6474 static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev) 6475 { 6476 int i, j, k; 6477 6478 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) 6479 return; 6480 6481 /* read back registers to clear the counters */ 6482 mutex_lock(&adev->grbm_idx_mutex); 6483 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) { 6484 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) { 6485 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) { 6486 gfx_v9_0_select_se_sh(adev, j, 0x0, k); 6487 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i])); 6488 } 6489 } 6490 } 6491 WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, 0xe0000000); 6492 mutex_unlock(&adev->grbm_idx_mutex); 6493 6494 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255); 6495 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0); 6496 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255); 6497 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0); 6498 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255); 6499 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0); 6500 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255); 6501 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0); 6502 6503 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) { 6504 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i); 6505 RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT); 6506 } 6507 6508 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) { 6509 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i); 6510 RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT); 6511 } 6512 6513 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) { 6514 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i); 6515 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT); 6516 } 6517 6518 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) { 6519 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i); 6520 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT); 6521 } 6522 6523 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255); 6524 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255); 6525 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255); 6526 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255); 6527 } 6528 6529 static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev, 6530 void *ras_error_status) 6531 { 6532 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; 6533 uint32_t sec_count = 0, ded_count = 0; 6534 uint32_t i, j, k; 6535 uint32_t reg_value; 6536 6537 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) 6538 return; 6539 6540 err_data->ue_count = 0; 6541 err_data->ce_count = 0; 6542 6543 mutex_lock(&adev->grbm_idx_mutex); 6544 6545 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) { 6546 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) { 6547 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) { 6548 gfx_v9_0_select_se_sh(adev, j, 0, k); 6549 reg_value = 6550 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i])); 6551 if (reg_value) 6552 gfx_v9_0_ras_error_count(adev, 6553 &gfx_v9_0_edc_counter_regs[i], 6554 j, k, reg_value, 6555 &sec_count, &ded_count); 6556 } 6557 } 6558 } 6559 6560 err_data->ce_count += sec_count; 6561 err_data->ue_count += ded_count; 6562 6563 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 6564 mutex_unlock(&adev->grbm_idx_mutex); 6565 6566 gfx_v9_0_query_utc_edc_status(adev, err_data); 6567 } 6568 6569 static void gfx_v9_0_emit_mem_sync(struct amdgpu_ring *ring) 6570 { 6571 const unsigned int cp_coher_cntl = 6572 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_ICACHE_ACTION_ENA(1) | 6573 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_KCACHE_ACTION_ENA(1) | 6574 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_ACTION_ENA(1) | 6575 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TCL1_ACTION_ENA(1) | 6576 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_WB_ACTION_ENA(1); 6577 6578 /* ACQUIRE_MEM -make one or more surfaces valid for use by the subsequent operations */ 6579 amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5)); 6580 amdgpu_ring_write(ring, cp_coher_cntl); /* CP_COHER_CNTL */ 6581 amdgpu_ring_write(ring, 0xffffffff); /* CP_COHER_SIZE */ 6582 amdgpu_ring_write(ring, 0xffffff); /* CP_COHER_SIZE_HI */ 6583 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */ 6584 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE_HI */ 6585 amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */ 6586 } 6587 6588 static void gfx_v9_0_emit_wave_limit_cs(struct amdgpu_ring *ring, 6589 uint32_t pipe, bool enable) 6590 { 6591 struct amdgpu_device *adev = ring->adev; 6592 uint32_t val; 6593 uint32_t wcl_cs_reg; 6594 6595 /* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */ 6596 val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS0_DEFAULT; 6597 6598 switch (pipe) { 6599 case 0: 6600 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS0); 6601 break; 6602 case 1: 6603 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS1); 6604 break; 6605 case 2: 6606 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS2); 6607 break; 6608 case 3: 6609 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS3); 6610 break; 6611 default: 6612 DRM_DEBUG("invalid pipe %d\n", pipe); 6613 return; 6614 } 6615 6616 amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val); 6617 6618 } 6619 static void gfx_v9_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable) 6620 { 6621 struct amdgpu_device *adev = ring->adev; 6622 uint32_t val; 6623 int i; 6624 6625 6626 /* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit 6627 * number of gfx waves. Setting 5 bit will make sure gfx only gets 6628 * around 25% of gpu resources. 6629 */ 6630 val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT; 6631 amdgpu_ring_emit_wreg(ring, 6632 SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_GFX), 6633 val); 6634 6635 /* Restrict waves for normal/low priority compute queues as well 6636 * to get best QoS for high priority compute jobs. 6637 * 6638 * amdgpu controls only 1st ME(0-3 CS pipes). 6639 */ 6640 for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) { 6641 if (i != ring->pipe) 6642 gfx_v9_0_emit_wave_limit_cs(ring, i, enable); 6643 6644 } 6645 } 6646 6647 static const struct amd_ip_funcs gfx_v9_0_ip_funcs = { 6648 .name = "gfx_v9_0", 6649 .early_init = gfx_v9_0_early_init, 6650 .late_init = gfx_v9_0_late_init, 6651 .sw_init = gfx_v9_0_sw_init, 6652 .sw_fini = gfx_v9_0_sw_fini, 6653 .hw_init = gfx_v9_0_hw_init, 6654 .hw_fini = gfx_v9_0_hw_fini, 6655 .suspend = gfx_v9_0_suspend, 6656 .resume = gfx_v9_0_resume, 6657 .is_idle = gfx_v9_0_is_idle, 6658 .wait_for_idle = gfx_v9_0_wait_for_idle, 6659 .soft_reset = gfx_v9_0_soft_reset, 6660 .set_clockgating_state = gfx_v9_0_set_clockgating_state, 6661 .set_powergating_state = gfx_v9_0_set_powergating_state, 6662 .get_clockgating_state = gfx_v9_0_get_clockgating_state, 6663 }; 6664 6665 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = { 6666 .type = AMDGPU_RING_TYPE_GFX, 6667 .align_mask = 0xff, 6668 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 6669 .support_64bit_ptrs = true, 6670 .secure_submission_supported = true, 6671 .vmhub = AMDGPU_GFXHUB_0, 6672 .get_rptr = gfx_v9_0_ring_get_rptr_gfx, 6673 .get_wptr = gfx_v9_0_ring_get_wptr_gfx, 6674 .set_wptr = gfx_v9_0_ring_set_wptr_gfx, 6675 .emit_frame_size = /* totally 242 maximum if 16 IBs */ 6676 5 + /* COND_EXEC */ 6677 7 + /* PIPELINE_SYNC */ 6678 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 6679 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 6680 2 + /* VM_FLUSH */ 6681 8 + /* FENCE for VM_FLUSH */ 6682 20 + /* GDS switch */ 6683 4 + /* double SWITCH_BUFFER, 6684 the first COND_EXEC jump to the place just 6685 prior to this double SWITCH_BUFFER */ 6686 5 + /* COND_EXEC */ 6687 7 + /* HDP_flush */ 6688 4 + /* VGT_flush */ 6689 14 + /* CE_META */ 6690 31 + /* DE_META */ 6691 3 + /* CNTX_CTRL */ 6692 5 + /* HDP_INVL */ 6693 8 + 8 + /* FENCE x2 */ 6694 2 + /* SWITCH_BUFFER */ 6695 7, /* gfx_v9_0_emit_mem_sync */ 6696 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_gfx */ 6697 .emit_ib = gfx_v9_0_ring_emit_ib_gfx, 6698 .emit_fence = gfx_v9_0_ring_emit_fence, 6699 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync, 6700 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush, 6701 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch, 6702 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush, 6703 .test_ring = gfx_v9_0_ring_test_ring, 6704 .test_ib = gfx_v9_0_ring_test_ib, 6705 .insert_nop = amdgpu_ring_insert_nop, 6706 .pad_ib = amdgpu_ring_generic_pad_ib, 6707 .emit_switch_buffer = gfx_v9_ring_emit_sb, 6708 .emit_cntxcntl = gfx_v9_ring_emit_cntxcntl, 6709 .init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec, 6710 .patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec, 6711 .emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl, 6712 .emit_wreg = gfx_v9_0_ring_emit_wreg, 6713 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait, 6714 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait, 6715 .soft_recovery = gfx_v9_0_ring_soft_recovery, 6716 .emit_mem_sync = gfx_v9_0_emit_mem_sync, 6717 }; 6718 6719 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = { 6720 .type = AMDGPU_RING_TYPE_COMPUTE, 6721 .align_mask = 0xff, 6722 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 6723 .support_64bit_ptrs = true, 6724 .vmhub = AMDGPU_GFXHUB_0, 6725 .get_rptr = gfx_v9_0_ring_get_rptr_compute, 6726 .get_wptr = gfx_v9_0_ring_get_wptr_compute, 6727 .set_wptr = gfx_v9_0_ring_set_wptr_compute, 6728 .emit_frame_size = 6729 20 + /* gfx_v9_0_ring_emit_gds_switch */ 6730 7 + /* gfx_v9_0_ring_emit_hdp_flush */ 6731 5 + /* hdp invalidate */ 6732 7 + /* gfx_v9_0_ring_emit_pipeline_sync */ 6733 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 6734 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 6735 2 + /* gfx_v9_0_ring_emit_vm_flush */ 6736 8 + 8 + 8 + /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */ 6737 7 + /* gfx_v9_0_emit_mem_sync */ 6738 5 + /* gfx_v9_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */ 6739 15, /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */ 6740 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */ 6741 .emit_ib = gfx_v9_0_ring_emit_ib_compute, 6742 .emit_fence = gfx_v9_0_ring_emit_fence, 6743 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync, 6744 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush, 6745 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch, 6746 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush, 6747 .test_ring = gfx_v9_0_ring_test_ring, 6748 .test_ib = gfx_v9_0_ring_test_ib, 6749 .insert_nop = amdgpu_ring_insert_nop, 6750 .pad_ib = amdgpu_ring_generic_pad_ib, 6751 .emit_wreg = gfx_v9_0_ring_emit_wreg, 6752 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait, 6753 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait, 6754 .emit_mem_sync = gfx_v9_0_emit_mem_sync, 6755 .emit_wave_limit = gfx_v9_0_emit_wave_limit, 6756 }; 6757 6758 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = { 6759 .type = AMDGPU_RING_TYPE_KIQ, 6760 .align_mask = 0xff, 6761 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 6762 .support_64bit_ptrs = true, 6763 .vmhub = AMDGPU_GFXHUB_0, 6764 .get_rptr = gfx_v9_0_ring_get_rptr_compute, 6765 .get_wptr = gfx_v9_0_ring_get_wptr_compute, 6766 .set_wptr = gfx_v9_0_ring_set_wptr_compute, 6767 .emit_frame_size = 6768 20 + /* gfx_v9_0_ring_emit_gds_switch */ 6769 7 + /* gfx_v9_0_ring_emit_hdp_flush */ 6770 5 + /* hdp invalidate */ 6771 7 + /* gfx_v9_0_ring_emit_pipeline_sync */ 6772 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 6773 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 6774 2 + /* gfx_v9_0_ring_emit_vm_flush */ 6775 8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */ 6776 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */ 6777 .emit_fence = gfx_v9_0_ring_emit_fence_kiq, 6778 .test_ring = gfx_v9_0_ring_test_ring, 6779 .insert_nop = amdgpu_ring_insert_nop, 6780 .pad_ib = amdgpu_ring_generic_pad_ib, 6781 .emit_rreg = gfx_v9_0_ring_emit_rreg, 6782 .emit_wreg = gfx_v9_0_ring_emit_wreg, 6783 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait, 6784 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait, 6785 }; 6786 6787 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev) 6788 { 6789 int i; 6790 6791 adev->gfx.kiq.ring.funcs = &gfx_v9_0_ring_funcs_kiq; 6792 6793 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 6794 adev->gfx.gfx_ring[i].funcs = &gfx_v9_0_ring_funcs_gfx; 6795 6796 for (i = 0; i < adev->gfx.num_compute_rings; i++) 6797 adev->gfx.compute_ring[i].funcs = &gfx_v9_0_ring_funcs_compute; 6798 } 6799 6800 static const struct amdgpu_irq_src_funcs gfx_v9_0_eop_irq_funcs = { 6801 .set = gfx_v9_0_set_eop_interrupt_state, 6802 .process = gfx_v9_0_eop_irq, 6803 }; 6804 6805 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_reg_irq_funcs = { 6806 .set = gfx_v9_0_set_priv_reg_fault_state, 6807 .process = gfx_v9_0_priv_reg_irq, 6808 }; 6809 6810 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_inst_irq_funcs = { 6811 .set = gfx_v9_0_set_priv_inst_fault_state, 6812 .process = gfx_v9_0_priv_inst_irq, 6813 }; 6814 6815 static const struct amdgpu_irq_src_funcs gfx_v9_0_cp_ecc_error_irq_funcs = { 6816 .set = gfx_v9_0_set_cp_ecc_error_state, 6817 .process = amdgpu_gfx_cp_ecc_error_irq, 6818 }; 6819 6820 6821 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev) 6822 { 6823 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST; 6824 adev->gfx.eop_irq.funcs = &gfx_v9_0_eop_irq_funcs; 6825 6826 adev->gfx.priv_reg_irq.num_types = 1; 6827 adev->gfx.priv_reg_irq.funcs = &gfx_v9_0_priv_reg_irq_funcs; 6828 6829 adev->gfx.priv_inst_irq.num_types = 1; 6830 adev->gfx.priv_inst_irq.funcs = &gfx_v9_0_priv_inst_irq_funcs; 6831 6832 adev->gfx.cp_ecc_error_irq.num_types = 2; /*C5 ECC error and C9 FUE error*/ 6833 adev->gfx.cp_ecc_error_irq.funcs = &gfx_v9_0_cp_ecc_error_irq_funcs; 6834 } 6835 6836 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev) 6837 { 6838 switch (adev->ip_versions[GC_HWIP][0]) { 6839 case IP_VERSION(9, 0, 1): 6840 case IP_VERSION(9, 2, 1): 6841 case IP_VERSION(9, 4, 0): 6842 case IP_VERSION(9, 2, 2): 6843 case IP_VERSION(9, 1, 0): 6844 case IP_VERSION(9, 4, 1): 6845 case IP_VERSION(9, 3, 0): 6846 case IP_VERSION(9, 4, 2): 6847 adev->gfx.rlc.funcs = &gfx_v9_0_rlc_funcs; 6848 break; 6849 default: 6850 break; 6851 } 6852 } 6853 6854 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev) 6855 { 6856 /* init asci gds info */ 6857 switch (adev->ip_versions[GC_HWIP][0]) { 6858 case IP_VERSION(9, 0, 1): 6859 case IP_VERSION(9, 2, 1): 6860 case IP_VERSION(9, 4, 0): 6861 adev->gds.gds_size = 0x10000; 6862 break; 6863 case IP_VERSION(9, 2, 2): 6864 case IP_VERSION(9, 1, 0): 6865 case IP_VERSION(9, 4, 1): 6866 adev->gds.gds_size = 0x1000; 6867 break; 6868 case IP_VERSION(9, 4, 2): 6869 /* aldebaran removed all the GDS internal memory, 6870 * only support GWS opcode in kernel, like barrier 6871 * semaphore.etc */ 6872 adev->gds.gds_size = 0; 6873 break; 6874 default: 6875 adev->gds.gds_size = 0x10000; 6876 break; 6877 } 6878 6879 switch (adev->ip_versions[GC_HWIP][0]) { 6880 case IP_VERSION(9, 0, 1): 6881 case IP_VERSION(9, 4, 0): 6882 adev->gds.gds_compute_max_wave_id = 0x7ff; 6883 break; 6884 case IP_VERSION(9, 2, 1): 6885 adev->gds.gds_compute_max_wave_id = 0x27f; 6886 break; 6887 case IP_VERSION(9, 2, 2): 6888 case IP_VERSION(9, 1, 0): 6889 if (adev->apu_flags & AMD_APU_IS_RAVEN2) 6890 adev->gds.gds_compute_max_wave_id = 0x77; /* raven2 */ 6891 else 6892 adev->gds.gds_compute_max_wave_id = 0x15f; /* raven1 */ 6893 break; 6894 case IP_VERSION(9, 4, 1): 6895 adev->gds.gds_compute_max_wave_id = 0xfff; 6896 break; 6897 case IP_VERSION(9, 4, 2): 6898 /* deprecated for Aldebaran, no usage at all */ 6899 adev->gds.gds_compute_max_wave_id = 0; 6900 break; 6901 default: 6902 /* this really depends on the chip */ 6903 adev->gds.gds_compute_max_wave_id = 0x7ff; 6904 break; 6905 } 6906 6907 adev->gds.gws_size = 64; 6908 adev->gds.oa_size = 16; 6909 } 6910 6911 static void gfx_v9_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev, 6912 u32 bitmap) 6913 { 6914 u32 data; 6915 6916 if (!bitmap) 6917 return; 6918 6919 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; 6920 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; 6921 6922 WREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG, data); 6923 } 6924 6925 static u32 gfx_v9_0_get_cu_active_bitmap(struct amdgpu_device *adev) 6926 { 6927 u32 data, mask; 6928 6929 data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG); 6930 data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG); 6931 6932 data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; 6933 data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; 6934 6935 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh); 6936 6937 return (~data) & mask; 6938 } 6939 6940 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev, 6941 struct amdgpu_cu_info *cu_info) 6942 { 6943 int i, j, k, counter, active_cu_number = 0; 6944 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0; 6945 unsigned disable_masks[4 * 4]; 6946 6947 if (!adev || !cu_info) 6948 return -EINVAL; 6949 6950 /* 6951 * 16 comes from bitmap array size 4*4, and it can cover all gfx9 ASICs 6952 */ 6953 if (adev->gfx.config.max_shader_engines * 6954 adev->gfx.config.max_sh_per_se > 16) 6955 return -EINVAL; 6956 6957 amdgpu_gfx_parse_disable_cu(disable_masks, 6958 adev->gfx.config.max_shader_engines, 6959 adev->gfx.config.max_sh_per_se); 6960 6961 mutex_lock(&adev->grbm_idx_mutex); 6962 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 6963 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 6964 mask = 1; 6965 ao_bitmap = 0; 6966 counter = 0; 6967 gfx_v9_0_select_se_sh(adev, i, j, 0xffffffff); 6968 gfx_v9_0_set_user_cu_inactive_bitmap( 6969 adev, disable_masks[i * adev->gfx.config.max_sh_per_se + j]); 6970 bitmap = gfx_v9_0_get_cu_active_bitmap(adev); 6971 6972 /* 6973 * The bitmap(and ao_cu_bitmap) in cu_info structure is 6974 * 4x4 size array, and it's usually suitable for Vega 6975 * ASICs which has 4*2 SE/SH layout. 6976 * But for Arcturus, SE/SH layout is changed to 8*1. 6977 * To mostly reduce the impact, we make it compatible 6978 * with current bitmap array as below: 6979 * SE4,SH0 --> bitmap[0][1] 6980 * SE5,SH0 --> bitmap[1][1] 6981 * SE6,SH0 --> bitmap[2][1] 6982 * SE7,SH0 --> bitmap[3][1] 6983 */ 6984 cu_info->bitmap[i % 4][j + i / 4] = bitmap; 6985 6986 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) { 6987 if (bitmap & mask) { 6988 if (counter < adev->gfx.config.max_cu_per_sh) 6989 ao_bitmap |= mask; 6990 counter ++; 6991 } 6992 mask <<= 1; 6993 } 6994 active_cu_number += counter; 6995 if (i < 2 && j < 2) 6996 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8)); 6997 cu_info->ao_cu_bitmap[i % 4][j + i / 4] = ao_bitmap; 6998 } 6999 } 7000 gfx_v9_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 7001 mutex_unlock(&adev->grbm_idx_mutex); 7002 7003 cu_info->number = active_cu_number; 7004 cu_info->ao_cu_mask = ao_cu_mask; 7005 cu_info->simd_per_cu = NUM_SIMD_PER_CU; 7006 7007 return 0; 7008 } 7009 7010 const struct amdgpu_ip_block_version gfx_v9_0_ip_block = 7011 { 7012 .type = AMD_IP_BLOCK_TYPE_GFX, 7013 .major = 9, 7014 .minor = 0, 7015 .rev = 0, 7016 .funcs = &gfx_v9_0_ip_funcs, 7017 }; 7018