1 /* 2 * Copyright 2016 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 */ 23 24 #include <linux/delay.h> 25 #include <linux/kernel.h> 26 #include <linux/firmware.h> 27 #include <linux/module.h> 28 #include <linux/pci.h> 29 30 #include "amdgpu.h" 31 #include "amdgpu_gfx.h" 32 #include "soc15.h" 33 #include "soc15d.h" 34 #include "amdgpu_atomfirmware.h" 35 #include "amdgpu_pm.h" 36 37 #include "gc/gc_9_0_offset.h" 38 #include "gc/gc_9_0_sh_mask.h" 39 40 #include "vega10_enum.h" 41 42 #include "soc15_common.h" 43 #include "clearstate_gfx9.h" 44 #include "v9_structs.h" 45 46 #include "ivsrcid/gfx/irqsrcs_gfx_9_0.h" 47 48 #include "amdgpu_ras.h" 49 50 #include "amdgpu_ring_mux.h" 51 #include "gfx_v9_4.h" 52 #include "gfx_v9_0.h" 53 #include "gfx_v9_4_2.h" 54 55 #include "asic_reg/pwr/pwr_10_0_offset.h" 56 #include "asic_reg/pwr/pwr_10_0_sh_mask.h" 57 #include "asic_reg/gc/gc_9_0_default.h" 58 59 #define GFX9_NUM_GFX_RINGS 1 60 #define GFX9_NUM_SW_GFX_RINGS 2 61 #define GFX9_MEC_HPD_SIZE 4096 62 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L 63 #define RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET 0x00000000L 64 65 #define mmGCEA_PROBE_MAP 0x070c 66 #define mmGCEA_PROBE_MAP_BASE_IDX 0 67 68 MODULE_FIRMWARE("amdgpu/vega10_ce.bin"); 69 MODULE_FIRMWARE("amdgpu/vega10_pfp.bin"); 70 MODULE_FIRMWARE("amdgpu/vega10_me.bin"); 71 MODULE_FIRMWARE("amdgpu/vega10_mec.bin"); 72 MODULE_FIRMWARE("amdgpu/vega10_mec2.bin"); 73 MODULE_FIRMWARE("amdgpu/vega10_rlc.bin"); 74 75 MODULE_FIRMWARE("amdgpu/vega12_ce.bin"); 76 MODULE_FIRMWARE("amdgpu/vega12_pfp.bin"); 77 MODULE_FIRMWARE("amdgpu/vega12_me.bin"); 78 MODULE_FIRMWARE("amdgpu/vega12_mec.bin"); 79 MODULE_FIRMWARE("amdgpu/vega12_mec2.bin"); 80 MODULE_FIRMWARE("amdgpu/vega12_rlc.bin"); 81 82 MODULE_FIRMWARE("amdgpu/vega20_ce.bin"); 83 MODULE_FIRMWARE("amdgpu/vega20_pfp.bin"); 84 MODULE_FIRMWARE("amdgpu/vega20_me.bin"); 85 MODULE_FIRMWARE("amdgpu/vega20_mec.bin"); 86 MODULE_FIRMWARE("amdgpu/vega20_mec2.bin"); 87 MODULE_FIRMWARE("amdgpu/vega20_rlc.bin"); 88 89 MODULE_FIRMWARE("amdgpu/raven_ce.bin"); 90 MODULE_FIRMWARE("amdgpu/raven_pfp.bin"); 91 MODULE_FIRMWARE("amdgpu/raven_me.bin"); 92 MODULE_FIRMWARE("amdgpu/raven_mec.bin"); 93 MODULE_FIRMWARE("amdgpu/raven_mec2.bin"); 94 MODULE_FIRMWARE("amdgpu/raven_rlc.bin"); 95 96 MODULE_FIRMWARE("amdgpu/picasso_ce.bin"); 97 MODULE_FIRMWARE("amdgpu/picasso_pfp.bin"); 98 MODULE_FIRMWARE("amdgpu/picasso_me.bin"); 99 MODULE_FIRMWARE("amdgpu/picasso_mec.bin"); 100 MODULE_FIRMWARE("amdgpu/picasso_mec2.bin"); 101 MODULE_FIRMWARE("amdgpu/picasso_rlc.bin"); 102 MODULE_FIRMWARE("amdgpu/picasso_rlc_am4.bin"); 103 104 MODULE_FIRMWARE("amdgpu/raven2_ce.bin"); 105 MODULE_FIRMWARE("amdgpu/raven2_pfp.bin"); 106 MODULE_FIRMWARE("amdgpu/raven2_me.bin"); 107 MODULE_FIRMWARE("amdgpu/raven2_mec.bin"); 108 MODULE_FIRMWARE("amdgpu/raven2_mec2.bin"); 109 MODULE_FIRMWARE("amdgpu/raven2_rlc.bin"); 110 MODULE_FIRMWARE("amdgpu/raven_kicker_rlc.bin"); 111 112 MODULE_FIRMWARE("amdgpu/arcturus_mec.bin"); 113 MODULE_FIRMWARE("amdgpu/arcturus_rlc.bin"); 114 115 MODULE_FIRMWARE("amdgpu/renoir_ce.bin"); 116 MODULE_FIRMWARE("amdgpu/renoir_pfp.bin"); 117 MODULE_FIRMWARE("amdgpu/renoir_me.bin"); 118 MODULE_FIRMWARE("amdgpu/renoir_mec.bin"); 119 MODULE_FIRMWARE("amdgpu/renoir_rlc.bin"); 120 121 MODULE_FIRMWARE("amdgpu/green_sardine_ce.bin"); 122 MODULE_FIRMWARE("amdgpu/green_sardine_pfp.bin"); 123 MODULE_FIRMWARE("amdgpu/green_sardine_me.bin"); 124 MODULE_FIRMWARE("amdgpu/green_sardine_mec.bin"); 125 MODULE_FIRMWARE("amdgpu/green_sardine_mec2.bin"); 126 MODULE_FIRMWARE("amdgpu/green_sardine_rlc.bin"); 127 128 MODULE_FIRMWARE("amdgpu/aldebaran_mec.bin"); 129 MODULE_FIRMWARE("amdgpu/aldebaran_mec2.bin"); 130 MODULE_FIRMWARE("amdgpu/aldebaran_rlc.bin"); 131 MODULE_FIRMWARE("amdgpu/aldebaran_sjt_mec.bin"); 132 MODULE_FIRMWARE("amdgpu/aldebaran_sjt_mec2.bin"); 133 134 #define mmTCP_CHAN_STEER_0_ARCT 0x0b03 135 #define mmTCP_CHAN_STEER_0_ARCT_BASE_IDX 0 136 #define mmTCP_CHAN_STEER_1_ARCT 0x0b04 137 #define mmTCP_CHAN_STEER_1_ARCT_BASE_IDX 0 138 #define mmTCP_CHAN_STEER_2_ARCT 0x0b09 139 #define mmTCP_CHAN_STEER_2_ARCT_BASE_IDX 0 140 #define mmTCP_CHAN_STEER_3_ARCT 0x0b0a 141 #define mmTCP_CHAN_STEER_3_ARCT_BASE_IDX 0 142 #define mmTCP_CHAN_STEER_4_ARCT 0x0b0b 143 #define mmTCP_CHAN_STEER_4_ARCT_BASE_IDX 0 144 #define mmTCP_CHAN_STEER_5_ARCT 0x0b0c 145 #define mmTCP_CHAN_STEER_5_ARCT_BASE_IDX 0 146 147 #define mmGOLDEN_TSC_COUNT_UPPER_Renoir 0x0025 148 #define mmGOLDEN_TSC_COUNT_UPPER_Renoir_BASE_IDX 1 149 #define mmGOLDEN_TSC_COUNT_LOWER_Renoir 0x0026 150 #define mmGOLDEN_TSC_COUNT_LOWER_Renoir_BASE_IDX 1 151 152 enum ta_ras_gfx_subblock { 153 /*CPC*/ 154 TA_RAS_BLOCK__GFX_CPC_INDEX_START = 0, 155 TA_RAS_BLOCK__GFX_CPC_SCRATCH = TA_RAS_BLOCK__GFX_CPC_INDEX_START, 156 TA_RAS_BLOCK__GFX_CPC_UCODE, 157 TA_RAS_BLOCK__GFX_DC_STATE_ME1, 158 TA_RAS_BLOCK__GFX_DC_CSINVOC_ME1, 159 TA_RAS_BLOCK__GFX_DC_RESTORE_ME1, 160 TA_RAS_BLOCK__GFX_DC_STATE_ME2, 161 TA_RAS_BLOCK__GFX_DC_CSINVOC_ME2, 162 TA_RAS_BLOCK__GFX_DC_RESTORE_ME2, 163 TA_RAS_BLOCK__GFX_CPC_INDEX_END = TA_RAS_BLOCK__GFX_DC_RESTORE_ME2, 164 /* CPF*/ 165 TA_RAS_BLOCK__GFX_CPF_INDEX_START, 166 TA_RAS_BLOCK__GFX_CPF_ROQ_ME2 = TA_RAS_BLOCK__GFX_CPF_INDEX_START, 167 TA_RAS_BLOCK__GFX_CPF_ROQ_ME1, 168 TA_RAS_BLOCK__GFX_CPF_TAG, 169 TA_RAS_BLOCK__GFX_CPF_INDEX_END = TA_RAS_BLOCK__GFX_CPF_TAG, 170 /* CPG*/ 171 TA_RAS_BLOCK__GFX_CPG_INDEX_START, 172 TA_RAS_BLOCK__GFX_CPG_DMA_ROQ = TA_RAS_BLOCK__GFX_CPG_INDEX_START, 173 TA_RAS_BLOCK__GFX_CPG_DMA_TAG, 174 TA_RAS_BLOCK__GFX_CPG_TAG, 175 TA_RAS_BLOCK__GFX_CPG_INDEX_END = TA_RAS_BLOCK__GFX_CPG_TAG, 176 /* GDS*/ 177 TA_RAS_BLOCK__GFX_GDS_INDEX_START, 178 TA_RAS_BLOCK__GFX_GDS_MEM = TA_RAS_BLOCK__GFX_GDS_INDEX_START, 179 TA_RAS_BLOCK__GFX_GDS_INPUT_QUEUE, 180 TA_RAS_BLOCK__GFX_GDS_OA_PHY_CMD_RAM_MEM, 181 TA_RAS_BLOCK__GFX_GDS_OA_PHY_DATA_RAM_MEM, 182 TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM, 183 TA_RAS_BLOCK__GFX_GDS_INDEX_END = TA_RAS_BLOCK__GFX_GDS_OA_PIPE_MEM, 184 /* SPI*/ 185 TA_RAS_BLOCK__GFX_SPI_SR_MEM, 186 /* SQ*/ 187 TA_RAS_BLOCK__GFX_SQ_INDEX_START, 188 TA_RAS_BLOCK__GFX_SQ_SGPR = TA_RAS_BLOCK__GFX_SQ_INDEX_START, 189 TA_RAS_BLOCK__GFX_SQ_LDS_D, 190 TA_RAS_BLOCK__GFX_SQ_LDS_I, 191 TA_RAS_BLOCK__GFX_SQ_VGPR, /* VGPR = SP*/ 192 TA_RAS_BLOCK__GFX_SQ_INDEX_END = TA_RAS_BLOCK__GFX_SQ_VGPR, 193 /* SQC (3 ranges)*/ 194 TA_RAS_BLOCK__GFX_SQC_INDEX_START, 195 /* SQC range 0*/ 196 TA_RAS_BLOCK__GFX_SQC_INDEX0_START = TA_RAS_BLOCK__GFX_SQC_INDEX_START, 197 TA_RAS_BLOCK__GFX_SQC_INST_UTCL1_LFIFO = 198 TA_RAS_BLOCK__GFX_SQC_INDEX0_START, 199 TA_RAS_BLOCK__GFX_SQC_DATA_CU0_WRITE_DATA_BUF, 200 TA_RAS_BLOCK__GFX_SQC_DATA_CU0_UTCL1_LFIFO, 201 TA_RAS_BLOCK__GFX_SQC_DATA_CU1_WRITE_DATA_BUF, 202 TA_RAS_BLOCK__GFX_SQC_DATA_CU1_UTCL1_LFIFO, 203 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_WRITE_DATA_BUF, 204 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO, 205 TA_RAS_BLOCK__GFX_SQC_INDEX0_END = 206 TA_RAS_BLOCK__GFX_SQC_DATA_CU2_UTCL1_LFIFO, 207 /* SQC range 1*/ 208 TA_RAS_BLOCK__GFX_SQC_INDEX1_START, 209 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_TAG_RAM = 210 TA_RAS_BLOCK__GFX_SQC_INDEX1_START, 211 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO, 212 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_MISS_FIFO, 213 TA_RAS_BLOCK__GFX_SQC_INST_BANKA_BANK_RAM, 214 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_TAG_RAM, 215 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_HIT_FIFO, 216 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_MISS_FIFO, 217 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM, 218 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM, 219 TA_RAS_BLOCK__GFX_SQC_INDEX1_END = 220 TA_RAS_BLOCK__GFX_SQC_DATA_BANKA_BANK_RAM, 221 /* SQC range 2*/ 222 TA_RAS_BLOCK__GFX_SQC_INDEX2_START, 223 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_TAG_RAM = 224 TA_RAS_BLOCK__GFX_SQC_INDEX2_START, 225 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO, 226 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_MISS_FIFO, 227 TA_RAS_BLOCK__GFX_SQC_INST_BANKB_BANK_RAM, 228 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_TAG_RAM, 229 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_HIT_FIFO, 230 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_MISS_FIFO, 231 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM, 232 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM, 233 TA_RAS_BLOCK__GFX_SQC_INDEX2_END = 234 TA_RAS_BLOCK__GFX_SQC_DATA_BANKB_BANK_RAM, 235 TA_RAS_BLOCK__GFX_SQC_INDEX_END = TA_RAS_BLOCK__GFX_SQC_INDEX2_END, 236 /* TA*/ 237 TA_RAS_BLOCK__GFX_TA_INDEX_START, 238 TA_RAS_BLOCK__GFX_TA_FS_DFIFO = TA_RAS_BLOCK__GFX_TA_INDEX_START, 239 TA_RAS_BLOCK__GFX_TA_FS_AFIFO, 240 TA_RAS_BLOCK__GFX_TA_FL_LFIFO, 241 TA_RAS_BLOCK__GFX_TA_FX_LFIFO, 242 TA_RAS_BLOCK__GFX_TA_FS_CFIFO, 243 TA_RAS_BLOCK__GFX_TA_INDEX_END = TA_RAS_BLOCK__GFX_TA_FS_CFIFO, 244 /* TCA*/ 245 TA_RAS_BLOCK__GFX_TCA_INDEX_START, 246 TA_RAS_BLOCK__GFX_TCA_HOLE_FIFO = TA_RAS_BLOCK__GFX_TCA_INDEX_START, 247 TA_RAS_BLOCK__GFX_TCA_REQ_FIFO, 248 TA_RAS_BLOCK__GFX_TCA_INDEX_END = TA_RAS_BLOCK__GFX_TCA_REQ_FIFO, 249 /* TCC (5 sub-ranges)*/ 250 TA_RAS_BLOCK__GFX_TCC_INDEX_START, 251 /* TCC range 0*/ 252 TA_RAS_BLOCK__GFX_TCC_INDEX0_START = TA_RAS_BLOCK__GFX_TCC_INDEX_START, 253 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX0_START, 254 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_0_1, 255 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_0, 256 TA_RAS_BLOCK__GFX_TCC_CACHE_DATA_BANK_1_1, 257 TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_0, 258 TA_RAS_BLOCK__GFX_TCC_CACHE_DIRTY_BANK_1, 259 TA_RAS_BLOCK__GFX_TCC_HIGH_RATE_TAG, 260 TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG, 261 TA_RAS_BLOCK__GFX_TCC_INDEX0_END = TA_RAS_BLOCK__GFX_TCC_LOW_RATE_TAG, 262 /* TCC range 1*/ 263 TA_RAS_BLOCK__GFX_TCC_INDEX1_START, 264 TA_RAS_BLOCK__GFX_TCC_IN_USE_DEC = TA_RAS_BLOCK__GFX_TCC_INDEX1_START, 265 TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER, 266 TA_RAS_BLOCK__GFX_TCC_INDEX1_END = 267 TA_RAS_BLOCK__GFX_TCC_IN_USE_TRANSFER, 268 /* TCC range 2*/ 269 TA_RAS_BLOCK__GFX_TCC_INDEX2_START, 270 TA_RAS_BLOCK__GFX_TCC_RETURN_DATA = TA_RAS_BLOCK__GFX_TCC_INDEX2_START, 271 TA_RAS_BLOCK__GFX_TCC_RETURN_CONTROL, 272 TA_RAS_BLOCK__GFX_TCC_UC_ATOMIC_FIFO, 273 TA_RAS_BLOCK__GFX_TCC_WRITE_RETURN, 274 TA_RAS_BLOCK__GFX_TCC_WRITE_CACHE_READ, 275 TA_RAS_BLOCK__GFX_TCC_SRC_FIFO, 276 TA_RAS_BLOCK__GFX_TCC_SRC_FIFO_NEXT_RAM, 277 TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO, 278 TA_RAS_BLOCK__GFX_TCC_INDEX2_END = 279 TA_RAS_BLOCK__GFX_TCC_CACHE_TAG_PROBE_FIFO, 280 /* TCC range 3*/ 281 TA_RAS_BLOCK__GFX_TCC_INDEX3_START, 282 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO = TA_RAS_BLOCK__GFX_TCC_INDEX3_START, 283 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM, 284 TA_RAS_BLOCK__GFX_TCC_INDEX3_END = 285 TA_RAS_BLOCK__GFX_TCC_LATENCY_FIFO_NEXT_RAM, 286 /* TCC range 4*/ 287 TA_RAS_BLOCK__GFX_TCC_INDEX4_START, 288 TA_RAS_BLOCK__GFX_TCC_WRRET_TAG_WRITE_RETURN = 289 TA_RAS_BLOCK__GFX_TCC_INDEX4_START, 290 TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER, 291 TA_RAS_BLOCK__GFX_TCC_INDEX4_END = 292 TA_RAS_BLOCK__GFX_TCC_ATOMIC_RETURN_BUFFER, 293 TA_RAS_BLOCK__GFX_TCC_INDEX_END = TA_RAS_BLOCK__GFX_TCC_INDEX4_END, 294 /* TCI*/ 295 TA_RAS_BLOCK__GFX_TCI_WRITE_RAM, 296 /* TCP*/ 297 TA_RAS_BLOCK__GFX_TCP_INDEX_START, 298 TA_RAS_BLOCK__GFX_TCP_CACHE_RAM = TA_RAS_BLOCK__GFX_TCP_INDEX_START, 299 TA_RAS_BLOCK__GFX_TCP_LFIFO_RAM, 300 TA_RAS_BLOCK__GFX_TCP_CMD_FIFO, 301 TA_RAS_BLOCK__GFX_TCP_VM_FIFO, 302 TA_RAS_BLOCK__GFX_TCP_DB_RAM, 303 TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO0, 304 TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1, 305 TA_RAS_BLOCK__GFX_TCP_INDEX_END = TA_RAS_BLOCK__GFX_TCP_UTCL1_LFIFO1, 306 /* TD*/ 307 TA_RAS_BLOCK__GFX_TD_INDEX_START, 308 TA_RAS_BLOCK__GFX_TD_SS_FIFO_LO = TA_RAS_BLOCK__GFX_TD_INDEX_START, 309 TA_RAS_BLOCK__GFX_TD_SS_FIFO_HI, 310 TA_RAS_BLOCK__GFX_TD_CS_FIFO, 311 TA_RAS_BLOCK__GFX_TD_INDEX_END = TA_RAS_BLOCK__GFX_TD_CS_FIFO, 312 /* EA (3 sub-ranges)*/ 313 TA_RAS_BLOCK__GFX_EA_INDEX_START, 314 /* EA range 0*/ 315 TA_RAS_BLOCK__GFX_EA_INDEX0_START = TA_RAS_BLOCK__GFX_EA_INDEX_START, 316 TA_RAS_BLOCK__GFX_EA_DRAMRD_CMDMEM = TA_RAS_BLOCK__GFX_EA_INDEX0_START, 317 TA_RAS_BLOCK__GFX_EA_DRAMWR_CMDMEM, 318 TA_RAS_BLOCK__GFX_EA_DRAMWR_DATAMEM, 319 TA_RAS_BLOCK__GFX_EA_RRET_TAGMEM, 320 TA_RAS_BLOCK__GFX_EA_WRET_TAGMEM, 321 TA_RAS_BLOCK__GFX_EA_GMIRD_CMDMEM, 322 TA_RAS_BLOCK__GFX_EA_GMIWR_CMDMEM, 323 TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM, 324 TA_RAS_BLOCK__GFX_EA_INDEX0_END = TA_RAS_BLOCK__GFX_EA_GMIWR_DATAMEM, 325 /* EA range 1*/ 326 TA_RAS_BLOCK__GFX_EA_INDEX1_START, 327 TA_RAS_BLOCK__GFX_EA_DRAMRD_PAGEMEM = TA_RAS_BLOCK__GFX_EA_INDEX1_START, 328 TA_RAS_BLOCK__GFX_EA_DRAMWR_PAGEMEM, 329 TA_RAS_BLOCK__GFX_EA_IORD_CMDMEM, 330 TA_RAS_BLOCK__GFX_EA_IOWR_CMDMEM, 331 TA_RAS_BLOCK__GFX_EA_IOWR_DATAMEM, 332 TA_RAS_BLOCK__GFX_EA_GMIRD_PAGEMEM, 333 TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM, 334 TA_RAS_BLOCK__GFX_EA_INDEX1_END = TA_RAS_BLOCK__GFX_EA_GMIWR_PAGEMEM, 335 /* EA range 2*/ 336 TA_RAS_BLOCK__GFX_EA_INDEX2_START, 337 TA_RAS_BLOCK__GFX_EA_MAM_D0MEM = TA_RAS_BLOCK__GFX_EA_INDEX2_START, 338 TA_RAS_BLOCK__GFX_EA_MAM_D1MEM, 339 TA_RAS_BLOCK__GFX_EA_MAM_D2MEM, 340 TA_RAS_BLOCK__GFX_EA_MAM_D3MEM, 341 TA_RAS_BLOCK__GFX_EA_INDEX2_END = TA_RAS_BLOCK__GFX_EA_MAM_D3MEM, 342 TA_RAS_BLOCK__GFX_EA_INDEX_END = TA_RAS_BLOCK__GFX_EA_INDEX2_END, 343 /* UTC VM L2 bank*/ 344 TA_RAS_BLOCK__UTC_VML2_BANK_CACHE, 345 /* UTC VM walker*/ 346 TA_RAS_BLOCK__UTC_VML2_WALKER, 347 /* UTC ATC L2 2MB cache*/ 348 TA_RAS_BLOCK__UTC_ATCL2_CACHE_2M_BANK, 349 /* UTC ATC L2 4KB cache*/ 350 TA_RAS_BLOCK__UTC_ATCL2_CACHE_4K_BANK, 351 TA_RAS_BLOCK__GFX_MAX 352 }; 353 354 struct ras_gfx_subblock { 355 unsigned char *name; 356 int ta_subblock; 357 int hw_supported_error_type; 358 int sw_supported_error_type; 359 }; 360 361 #define AMDGPU_RAS_SUB_BLOCK(subblock, a, b, c, d, e, f, g, h) \ 362 [AMDGPU_RAS_BLOCK__##subblock] = { \ 363 #subblock, \ 364 TA_RAS_BLOCK__##subblock, \ 365 ((a) | ((b) << 1) | ((c) << 2) | ((d) << 3)), \ 366 (((e) << 1) | ((f) << 3) | (g) | ((h) << 2)), \ 367 } 368 369 static const struct ras_gfx_subblock ras_gfx_subblocks[] = { 370 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_SCRATCH, 0, 1, 1, 1, 1, 0, 0, 1), 371 AMDGPU_RAS_SUB_BLOCK(GFX_CPC_UCODE, 0, 1, 1, 1, 1, 0, 0, 1), 372 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME1, 1, 0, 0, 1, 0, 0, 1, 0), 373 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME1, 1, 0, 0, 1, 0, 0, 0, 0), 374 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME1, 1, 0, 0, 1, 0, 0, 0, 0), 375 AMDGPU_RAS_SUB_BLOCK(GFX_DC_STATE_ME2, 1, 0, 0, 1, 0, 0, 0, 0), 376 AMDGPU_RAS_SUB_BLOCK(GFX_DC_CSINVOC_ME2, 1, 0, 0, 1, 0, 0, 0, 0), 377 AMDGPU_RAS_SUB_BLOCK(GFX_DC_RESTORE_ME2, 1, 0, 0, 1, 0, 0, 0, 0), 378 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME2, 1, 0, 0, 1, 0, 0, 0, 0), 379 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_ROQ_ME1, 1, 0, 0, 1, 0, 0, 1, 0), 380 AMDGPU_RAS_SUB_BLOCK(GFX_CPF_TAG, 0, 1, 1, 1, 1, 0, 0, 1), 381 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_ROQ, 1, 0, 0, 1, 0, 0, 1, 0), 382 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_DMA_TAG, 0, 1, 1, 1, 0, 1, 0, 1), 383 AMDGPU_RAS_SUB_BLOCK(GFX_CPG_TAG, 0, 1, 1, 1, 1, 1, 0, 1), 384 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_MEM, 0, 1, 1, 1, 0, 0, 0, 0), 385 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_INPUT_QUEUE, 1, 0, 0, 1, 0, 0, 0, 0), 386 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_CMD_RAM_MEM, 0, 1, 1, 1, 0, 0, 0, 387 0), 388 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PHY_DATA_RAM_MEM, 1, 0, 0, 1, 0, 0, 0, 389 0), 390 AMDGPU_RAS_SUB_BLOCK(GFX_GDS_OA_PIPE_MEM, 0, 1, 1, 1, 0, 0, 0, 0), 391 AMDGPU_RAS_SUB_BLOCK(GFX_SPI_SR_MEM, 1, 0, 0, 1, 0, 0, 0, 0), 392 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_SGPR, 0, 1, 1, 1, 0, 0, 0, 0), 393 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_D, 0, 1, 1, 1, 1, 0, 0, 1), 394 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_LDS_I, 0, 1, 1, 1, 0, 0, 0, 0), 395 AMDGPU_RAS_SUB_BLOCK(GFX_SQ_VGPR, 0, 1, 1, 1, 0, 0, 0, 0), 396 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 1), 397 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0, 398 0, 0), 399 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU0_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 400 0), 401 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0, 402 0, 0), 403 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU1_UTCL1_LFIFO, 0, 1, 1, 1, 1, 0, 0, 404 0), 405 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_WRITE_DATA_BUF, 0, 1, 1, 1, 0, 0, 406 0, 0), 407 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_CU2_UTCL1_LFIFO, 0, 1, 1, 1, 0, 0, 0, 408 0), 409 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_TAG_RAM, 0, 1, 1, 1, 1, 0, 0, 410 1), 411 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0, 412 0, 0, 0), 413 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0, 414 0), 415 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0, 416 0), 417 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_TAG_RAM, 0, 1, 1, 1, 0, 0, 0, 418 0), 419 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0, 420 0), 421 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0, 422 0), 423 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0, 424 0, 0), 425 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKA_BANK_RAM, 0, 1, 1, 1, 0, 0, 0, 426 0), 427 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_TAG_RAM, 0, 1, 1, 1, 1, 0, 0, 428 0), 429 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_UTCL1_MISS_FIFO, 1, 0, 0, 1, 0, 430 0, 0, 0), 431 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0, 432 0), 433 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_INST_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0, 434 0), 435 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_TAG_RAM, 0, 1, 1, 1, 0, 0, 0, 436 0), 437 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_HIT_FIFO, 1, 0, 0, 1, 0, 0, 0, 438 0), 439 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_MISS_FIFO, 1, 0, 0, 1, 0, 0, 0, 440 0), 441 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_DIRTY_BIT_RAM, 1, 0, 0, 1, 0, 0, 442 0, 0), 443 AMDGPU_RAS_SUB_BLOCK(GFX_SQC_DATA_BANKB_BANK_RAM, 0, 1, 1, 1, 0, 0, 0, 444 0), 445 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_DFIFO, 0, 1, 1, 1, 1, 0, 0, 1), 446 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_AFIFO, 1, 0, 0, 1, 0, 0, 0, 0), 447 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FL_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0), 448 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FX_LFIFO, 1, 0, 0, 1, 0, 0, 0, 0), 449 AMDGPU_RAS_SUB_BLOCK(GFX_TA_FS_CFIFO, 1, 0, 0, 1, 0, 0, 0, 0), 450 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_HOLE_FIFO, 1, 0, 0, 1, 0, 1, 1, 0), 451 AMDGPU_RAS_SUB_BLOCK(GFX_TCA_REQ_FIFO, 1, 0, 0, 1, 0, 0, 0, 0), 452 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA, 0, 1, 1, 1, 1, 0, 0, 1), 453 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_0_1, 0, 1, 1, 1, 1, 0, 0, 454 1), 455 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_0, 0, 1, 1, 1, 1, 0, 0, 456 1), 457 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DATA_BANK_1_1, 0, 1, 1, 1, 1, 0, 0, 458 1), 459 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_0, 0, 1, 1, 1, 0, 0, 0, 460 0), 461 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_DIRTY_BANK_1, 0, 1, 1, 1, 0, 0, 0, 462 0), 463 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_HIGH_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0), 464 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LOW_RATE_TAG, 0, 1, 1, 1, 0, 0, 0, 0), 465 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_DEC, 1, 0, 0, 1, 0, 0, 0, 0), 466 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_IN_USE_TRANSFER, 1, 0, 0, 1, 0, 0, 0, 0), 467 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_DATA, 1, 0, 0, 1, 0, 0, 0, 0), 468 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_RETURN_CONTROL, 1, 0, 0, 1, 0, 0, 0, 0), 469 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_UC_ATOMIC_FIFO, 1, 0, 0, 1, 0, 0, 0, 0), 470 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_RETURN, 1, 0, 0, 1, 0, 1, 1, 0), 471 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRITE_CACHE_READ, 1, 0, 0, 1, 0, 0, 0, 0), 472 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO, 0, 1, 1, 1, 0, 0, 0, 0), 473 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_SRC_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 1, 0), 474 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_CACHE_TAG_PROBE_FIFO, 1, 0, 0, 1, 0, 0, 0, 475 0), 476 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO, 1, 0, 0, 1, 0, 0, 0, 0), 477 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_LATENCY_FIFO_NEXT_RAM, 1, 0, 0, 1, 0, 0, 0, 478 0), 479 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_WRRET_TAG_WRITE_RETURN, 1, 0, 0, 1, 0, 0, 480 0, 0), 481 AMDGPU_RAS_SUB_BLOCK(GFX_TCC_ATOMIC_RETURN_BUFFER, 1, 0, 0, 1, 0, 0, 0, 482 0), 483 AMDGPU_RAS_SUB_BLOCK(GFX_TCI_WRITE_RAM, 1, 0, 0, 1, 0, 0, 0, 0), 484 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CACHE_RAM, 0, 1, 1, 1, 1, 0, 0, 1), 485 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_LFIFO_RAM, 0, 1, 1, 1, 0, 0, 0, 0), 486 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_CMD_FIFO, 1, 0, 0, 1, 0, 0, 0, 0), 487 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_VM_FIFO, 0, 1, 1, 1, 0, 0, 0, 0), 488 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_DB_RAM, 1, 0, 0, 1, 0, 0, 0, 0), 489 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO0, 0, 1, 1, 1, 0, 0, 0, 0), 490 AMDGPU_RAS_SUB_BLOCK(GFX_TCP_UTCL1_LFIFO1, 0, 1, 1, 1, 0, 0, 0, 0), 491 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_LO, 0, 1, 1, 1, 1, 0, 0, 1), 492 AMDGPU_RAS_SUB_BLOCK(GFX_TD_SS_FIFO_HI, 0, 1, 1, 1, 0, 0, 0, 0), 493 AMDGPU_RAS_SUB_BLOCK(GFX_TD_CS_FIFO, 1, 0, 0, 1, 0, 0, 0, 0), 494 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_CMDMEM, 0, 1, 1, 1, 1, 0, 0, 1), 495 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0), 496 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0), 497 AMDGPU_RAS_SUB_BLOCK(GFX_EA_RRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0), 498 AMDGPU_RAS_SUB_BLOCK(GFX_EA_WRET_TAGMEM, 0, 1, 1, 1, 0, 0, 0, 0), 499 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0), 500 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_CMDMEM, 0, 1, 1, 1, 0, 0, 0, 0), 501 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_DATAMEM, 0, 1, 1, 1, 0, 0, 0, 0), 502 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0), 503 AMDGPU_RAS_SUB_BLOCK(GFX_EA_DRAMWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0), 504 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IORD_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0), 505 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_CMDMEM, 1, 0, 0, 1, 0, 0, 0, 0), 506 AMDGPU_RAS_SUB_BLOCK(GFX_EA_IOWR_DATAMEM, 1, 0, 0, 1, 0, 0, 0, 0), 507 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIRD_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0), 508 AMDGPU_RAS_SUB_BLOCK(GFX_EA_GMIWR_PAGEMEM, 1, 0, 0, 1, 0, 0, 0, 0), 509 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D0MEM, 1, 0, 0, 1, 0, 0, 0, 0), 510 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D1MEM, 1, 0, 0, 1, 0, 0, 0, 0), 511 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D2MEM, 1, 0, 0, 1, 0, 0, 0, 0), 512 AMDGPU_RAS_SUB_BLOCK(GFX_EA_MAM_D3MEM, 1, 0, 0, 1, 0, 0, 0, 0), 513 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_BANK_CACHE, 0, 1, 1, 1, 0, 0, 0, 0), 514 AMDGPU_RAS_SUB_BLOCK(UTC_VML2_WALKER, 0, 1, 1, 1, 0, 0, 0, 0), 515 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_2M_BANK, 1, 0, 0, 1, 0, 0, 0, 0), 516 AMDGPU_RAS_SUB_BLOCK(UTC_ATCL2_CACHE_4K_BANK, 0, 1, 1, 1, 0, 0, 0, 0), 517 }; 518 519 static const struct soc15_reg_golden golden_settings_gc_9_0[] = 520 { 521 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400), 522 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG3, 0x80000000, 0x80000000), 523 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000), 524 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024), 525 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001), 526 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000), 527 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000), 528 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800), 529 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800), 530 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x00ffff87), 531 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x00ffff8f), 532 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000), 533 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000), 534 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68), 535 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197), 536 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000), 537 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff), 538 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800), 539 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800), 540 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000) 541 }; 542 543 static const struct soc15_reg_golden golden_settings_gc_9_0_vg10[] = 544 { 545 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0x0000f000, 0x00012107), 546 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000), 547 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080), 548 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080), 549 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080), 550 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042), 551 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x2a114042), 552 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080), 553 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0x00008000, 0x00048000), 554 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080), 555 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080), 556 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080), 557 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080), 558 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080), 559 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000), 560 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x01000107), 561 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x00001800, 0x00000800), 562 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080) 563 }; 564 565 static const struct soc15_reg_golden golden_settings_gc_9_0_vg20[] = 566 { 567 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x0f000080, 0x04000080), 568 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000), 569 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000), 570 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x22014042), 571 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x22014042), 572 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0x00003e00, 0x00000400), 573 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xff840000, 0x04040000), 574 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00030000), 575 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff010f, 0x01000107), 576 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0x000b0000, 0x000b0000), 577 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01000000, 0x01000000) 578 }; 579 580 static const struct soc15_reg_golden golden_settings_gc_9_1[] = 581 { 582 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104), 583 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x08000000, 0x08000080), 584 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0x08000000, 0x08000080), 585 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x08000000, 0x08000080), 586 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420), 587 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000), 588 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x08000000, 0x08000080), 589 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024), 590 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001), 591 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000), 592 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0x08000000, 0x08000080), 593 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0x08000000, 0x08000080), 594 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0x08000000, 0x08000080), 595 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0x08000000, 0x08000080), 596 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0x08000000, 0x08000080), 597 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000), 598 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000), 599 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120), 600 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000), 601 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000000ff), 602 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x08000000, 0x08000080), 603 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800), 604 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800), 605 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000) 606 }; 607 608 static const struct soc15_reg_golden golden_settings_gc_9_1_rv1[] = 609 { 610 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_3, 0x30000000, 0x10000000), 611 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24000042), 612 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24000042), 613 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04048000), 614 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_MODE_CNTL_1, 0x06000000, 0x06000000), 615 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_UTCL1_CNTL2, 0x00030000, 0x00020000), 616 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x00000800) 617 }; 618 619 static const struct soc15_reg_golden golden_settings_gc_9_1_rv2[] = 620 { 621 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0xff7fffff, 0x04000000), 622 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104), 623 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000), 624 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPC_UTCL1_CNTL, 0x7f0fffff, 0x08000080), 625 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPF_UTCL1_CNTL, 0xff8fffff, 0x08000080), 626 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCPG_UTCL1_CNTL, 0x7f8fffff, 0x08000080), 627 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x26013041), 628 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x26013041), 629 SOC15_REG_GOLDEN_VALUE(GC, 0, mmIA_UTCL1_CNTL, 0x3f8fffff, 0x08000080), 630 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000), 631 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_0, 0xff0fffff, 0x08000080), 632 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_1, 0xff0fffff, 0x08000080), 633 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_GPM_UTCL1_CNTL_2, 0xff0fffff, 0x08000080), 634 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_PREWALKER_UTCL1_CNTL, 0xff0fffff, 0x08000080), 635 SOC15_REG_GOLDEN_VALUE(GC, 0, mmRLC_SPM_UTCL1_CNTL, 0xff0fffff, 0x08000080), 636 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000), 637 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010), 638 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000), 639 SOC15_REG_GOLDEN_VALUE(GC, 0, mmWD_UTCL1_CNTL, 0x3f8fffff, 0x08000080), 640 }; 641 642 static const struct soc15_reg_golden golden_settings_gc_9_1_rn[] = 643 { 644 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104), 645 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0xff7fffff, 0x0a000000), 646 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000400), 647 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xf3e777ff, 0x24000042), 648 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xf3e777ff, 0x24000042), 649 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001), 650 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000), 651 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000), 652 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000), 653 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000), 654 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003120), 655 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGCEA_PROBE_MAP, 0xffffffff, 0x0000cccc), 656 }; 657 658 static const struct soc15_reg_golden golden_settings_gc_9_x_common[] = 659 { 660 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0xffffffff, 0x000001ff), 661 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_INDEX, 0xffffffff, 0x00000000), 662 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGRBM_CAM_DATA, 0xffffffff, 0x2544c382) 663 }; 664 665 static const struct soc15_reg_golden golden_settings_gc_9_2_1[] = 666 { 667 SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xf00fffff, 0x00000420), 668 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_GPU_ID, 0x0000000f, 0x00000000), 669 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3, 0x00000003, 0x82400024), 670 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x00000001), 671 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000), 672 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSH_MEM_CONFIG, 0x00001000, 0x00001000), 673 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_0, 0x0007ffff, 0x00000800), 674 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_CU_1, 0x0007ffff, 0x00000800), 675 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_0, 0x01ffffff, 0x0000ff87), 676 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_RESOURCE_RESERVE_EN_CU_1, 0x01ffffff, 0x0000ff8f), 677 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQC_CONFIG, 0x03000000, 0x020a2000), 678 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x010b0000), 679 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x4a2c0e68), 680 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0xb5d3f197), 681 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_CACHE_INVALIDATION, 0x3fff3af3, 0x19200000), 682 SOC15_REG_GOLDEN_VALUE(GC, 0, mmVGT_GS_MAX_WAVE_ID, 0x00000fff, 0x000003ff) 683 }; 684 685 static const struct soc15_reg_golden golden_settings_gc_9_2_1_vg12[] = 686 { 687 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_DCC_CONFIG, 0x00000080, 0x04000080), 688 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL, 0xfffdf3cf, 0x00014104), 689 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_2, 0x0f000000, 0x0a000000), 690 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x24104041), 691 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG_READ, 0xffff77ff, 0x24104041), 692 SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04040000), 693 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSPI_CONFIG_CNTL_1, 0xffff03ff, 0x01000107), 694 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000), 695 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_LO, 0xffffffff, 0x76325410), 696 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTD_CNTL, 0x01bd9f33, 0x01000000), 697 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC1_F32_INT_DIS, 0x00000800, 0x00000800), 698 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_MEC2_F32_INT_DIS, 0x00000800, 0x00000800), 699 SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_DEBUG, 0x00008000, 0x00008000) 700 }; 701 702 static const struct soc15_reg_golden golden_settings_gc_9_4_1_arct[] = 703 { 704 SOC15_REG_GOLDEN_VALUE(GC, 0, mmGB_ADDR_CONFIG, 0xffff77ff, 0x2a114042), 705 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfffffeef, 0x10b0000), 706 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_0_ARCT, 0x3fffffff, 0x346f0a4e), 707 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_1_ARCT, 0x3fffffff, 0x1c642ca), 708 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_2_ARCT, 0x3fffffff, 0x26f45098), 709 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_3_ARCT, 0x3fffffff, 0x2ebd9fe3), 710 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_4_ARCT, 0x3fffffff, 0xb90f5b1), 711 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CHAN_STEER_5_ARCT, 0x3ff, 0x135), 712 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_CONFIG, 0xffffffff, 0x011A0000), 713 SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_FIFO_SIZES, 0xffffffff, 0x00000f00), 714 SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_UTCL1_CNTL1, 0x30000000, 0x30000000) 715 }; 716 717 static const struct soc15_reg_rlcg rlcg_access_gc_9_0[] = { 718 {SOC15_REG_ENTRY(GC, 0, mmGRBM_GFX_INDEX)}, 719 {SOC15_REG_ENTRY(GC, 0, mmSQ_IND_INDEX)}, 720 }; 721 722 static const u32 GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[] = 723 { 724 mmRLC_SRM_INDEX_CNTL_ADDR_0 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 725 mmRLC_SRM_INDEX_CNTL_ADDR_1 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 726 mmRLC_SRM_INDEX_CNTL_ADDR_2 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 727 mmRLC_SRM_INDEX_CNTL_ADDR_3 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 728 mmRLC_SRM_INDEX_CNTL_ADDR_4 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 729 mmRLC_SRM_INDEX_CNTL_ADDR_5 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 730 mmRLC_SRM_INDEX_CNTL_ADDR_6 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 731 mmRLC_SRM_INDEX_CNTL_ADDR_7 - mmRLC_SRM_INDEX_CNTL_ADDR_0, 732 }; 733 734 static const u32 GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[] = 735 { 736 mmRLC_SRM_INDEX_CNTL_DATA_0 - mmRLC_SRM_INDEX_CNTL_DATA_0, 737 mmRLC_SRM_INDEX_CNTL_DATA_1 - mmRLC_SRM_INDEX_CNTL_DATA_0, 738 mmRLC_SRM_INDEX_CNTL_DATA_2 - mmRLC_SRM_INDEX_CNTL_DATA_0, 739 mmRLC_SRM_INDEX_CNTL_DATA_3 - mmRLC_SRM_INDEX_CNTL_DATA_0, 740 mmRLC_SRM_INDEX_CNTL_DATA_4 - mmRLC_SRM_INDEX_CNTL_DATA_0, 741 mmRLC_SRM_INDEX_CNTL_DATA_5 - mmRLC_SRM_INDEX_CNTL_DATA_0, 742 mmRLC_SRM_INDEX_CNTL_DATA_6 - mmRLC_SRM_INDEX_CNTL_DATA_0, 743 mmRLC_SRM_INDEX_CNTL_DATA_7 - mmRLC_SRM_INDEX_CNTL_DATA_0, 744 }; 745 746 #define VEGA10_GB_ADDR_CONFIG_GOLDEN 0x2a114042 747 #define VEGA12_GB_ADDR_CONFIG_GOLDEN 0x24104041 748 #define RAVEN_GB_ADDR_CONFIG_GOLDEN 0x24000042 749 #define RAVEN2_GB_ADDR_CONFIG_GOLDEN 0x26013041 750 751 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev); 752 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev); 753 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev); 754 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev); 755 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev, 756 struct amdgpu_cu_info *cu_info); 757 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev); 758 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume); 759 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring); 760 static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev, 761 void *ras_error_status); 762 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev, 763 void *inject_if); 764 static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev); 765 766 static void gfx_v9_0_kiq_set_resources(struct amdgpu_ring *kiq_ring, 767 uint64_t queue_mask) 768 { 769 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6)); 770 amdgpu_ring_write(kiq_ring, 771 PACKET3_SET_RESOURCES_VMID_MASK(0) | 772 /* vmid_mask:0* queue_type:0 (KIQ) */ 773 PACKET3_SET_RESOURCES_QUEUE_TYPE(0)); 774 amdgpu_ring_write(kiq_ring, 775 lower_32_bits(queue_mask)); /* queue mask lo */ 776 amdgpu_ring_write(kiq_ring, 777 upper_32_bits(queue_mask)); /* queue mask hi */ 778 amdgpu_ring_write(kiq_ring, 0); /* gws mask lo */ 779 amdgpu_ring_write(kiq_ring, 0); /* gws mask hi */ 780 amdgpu_ring_write(kiq_ring, 0); /* oac mask */ 781 amdgpu_ring_write(kiq_ring, 0); /* gds heap base:0, gds heap size:0 */ 782 } 783 784 static void gfx_v9_0_kiq_map_queues(struct amdgpu_ring *kiq_ring, 785 struct amdgpu_ring *ring) 786 { 787 uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj); 788 uint64_t wptr_addr = ring->wptr_gpu_addr; 789 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0; 790 791 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5)); 792 /* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/ 793 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */ 794 PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */ 795 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */ 796 PACKET3_MAP_QUEUES_QUEUE(ring->queue) | 797 PACKET3_MAP_QUEUES_PIPE(ring->pipe) | 798 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) | 799 /*queue_type: normal compute queue */ 800 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | 801 /* alloc format: all_on_one_pipe */ 802 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | 803 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) | 804 /* num_queues: must be 1 */ 805 PACKET3_MAP_QUEUES_NUM_QUEUES(1)); 806 amdgpu_ring_write(kiq_ring, 807 PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index)); 808 amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr)); 809 amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr)); 810 amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr)); 811 amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr)); 812 } 813 814 static void gfx_v9_0_kiq_unmap_queues(struct amdgpu_ring *kiq_ring, 815 struct amdgpu_ring *ring, 816 enum amdgpu_unmap_queues_action action, 817 u64 gpu_addr, u64 seq) 818 { 819 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0; 820 821 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4)); 822 amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */ 823 PACKET3_UNMAP_QUEUES_ACTION(action) | 824 PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) | 825 PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) | 826 PACKET3_UNMAP_QUEUES_NUM_QUEUES(1)); 827 amdgpu_ring_write(kiq_ring, 828 PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index)); 829 830 if (action == PREEMPT_QUEUES_NO_UNMAP) { 831 amdgpu_ring_write(kiq_ring, lower_32_bits(ring->wptr & ring->buf_mask)); 832 amdgpu_ring_write(kiq_ring, 0); 833 amdgpu_ring_write(kiq_ring, 0); 834 835 } else { 836 amdgpu_ring_write(kiq_ring, 0); 837 amdgpu_ring_write(kiq_ring, 0); 838 amdgpu_ring_write(kiq_ring, 0); 839 } 840 } 841 842 static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring, 843 struct amdgpu_ring *ring, 844 u64 addr, 845 u64 seq) 846 { 847 uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0; 848 849 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5)); 850 amdgpu_ring_write(kiq_ring, 851 PACKET3_QUERY_STATUS_CONTEXT_ID(0) | 852 PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) | 853 PACKET3_QUERY_STATUS_COMMAND(2)); 854 /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */ 855 amdgpu_ring_write(kiq_ring, 856 PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) | 857 PACKET3_QUERY_STATUS_ENG_SEL(eng_sel)); 858 amdgpu_ring_write(kiq_ring, lower_32_bits(addr)); 859 amdgpu_ring_write(kiq_ring, upper_32_bits(addr)); 860 amdgpu_ring_write(kiq_ring, lower_32_bits(seq)); 861 amdgpu_ring_write(kiq_ring, upper_32_bits(seq)); 862 } 863 864 static void gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring, 865 uint16_t pasid, uint32_t flush_type, 866 bool all_hub) 867 { 868 amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0)); 869 amdgpu_ring_write(kiq_ring, 870 PACKET3_INVALIDATE_TLBS_DST_SEL(1) | 871 PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) | 872 PACKET3_INVALIDATE_TLBS_PASID(pasid) | 873 PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type)); 874 } 875 876 static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = { 877 .kiq_set_resources = gfx_v9_0_kiq_set_resources, 878 .kiq_map_queues = gfx_v9_0_kiq_map_queues, 879 .kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues, 880 .kiq_query_status = gfx_v9_0_kiq_query_status, 881 .kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs, 882 .set_resources_size = 8, 883 .map_queues_size = 7, 884 .unmap_queues_size = 6, 885 .query_status_size = 7, 886 .invalidate_tlbs_size = 2, 887 }; 888 889 static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev) 890 { 891 adev->gfx.kiq.pmf = &gfx_v9_0_kiq_pm4_funcs; 892 } 893 894 static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev) 895 { 896 switch (adev->ip_versions[GC_HWIP][0]) { 897 case IP_VERSION(9, 0, 1): 898 soc15_program_register_sequence(adev, 899 golden_settings_gc_9_0, 900 ARRAY_SIZE(golden_settings_gc_9_0)); 901 soc15_program_register_sequence(adev, 902 golden_settings_gc_9_0_vg10, 903 ARRAY_SIZE(golden_settings_gc_9_0_vg10)); 904 break; 905 case IP_VERSION(9, 2, 1): 906 soc15_program_register_sequence(adev, 907 golden_settings_gc_9_2_1, 908 ARRAY_SIZE(golden_settings_gc_9_2_1)); 909 soc15_program_register_sequence(adev, 910 golden_settings_gc_9_2_1_vg12, 911 ARRAY_SIZE(golden_settings_gc_9_2_1_vg12)); 912 break; 913 case IP_VERSION(9, 4, 0): 914 soc15_program_register_sequence(adev, 915 golden_settings_gc_9_0, 916 ARRAY_SIZE(golden_settings_gc_9_0)); 917 soc15_program_register_sequence(adev, 918 golden_settings_gc_9_0_vg20, 919 ARRAY_SIZE(golden_settings_gc_9_0_vg20)); 920 break; 921 case IP_VERSION(9, 4, 1): 922 soc15_program_register_sequence(adev, 923 golden_settings_gc_9_4_1_arct, 924 ARRAY_SIZE(golden_settings_gc_9_4_1_arct)); 925 break; 926 case IP_VERSION(9, 2, 2): 927 case IP_VERSION(9, 1, 0): 928 soc15_program_register_sequence(adev, golden_settings_gc_9_1, 929 ARRAY_SIZE(golden_settings_gc_9_1)); 930 if (adev->apu_flags & AMD_APU_IS_RAVEN2) 931 soc15_program_register_sequence(adev, 932 golden_settings_gc_9_1_rv2, 933 ARRAY_SIZE(golden_settings_gc_9_1_rv2)); 934 else 935 soc15_program_register_sequence(adev, 936 golden_settings_gc_9_1_rv1, 937 ARRAY_SIZE(golden_settings_gc_9_1_rv1)); 938 break; 939 case IP_VERSION(9, 3, 0): 940 soc15_program_register_sequence(adev, 941 golden_settings_gc_9_1_rn, 942 ARRAY_SIZE(golden_settings_gc_9_1_rn)); 943 return; /* for renoir, don't need common goldensetting */ 944 case IP_VERSION(9, 4, 2): 945 gfx_v9_4_2_init_golden_registers(adev, 946 adev->smuio.funcs->get_die_id(adev)); 947 break; 948 default: 949 break; 950 } 951 952 if ((adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)) && 953 (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 2))) 954 soc15_program_register_sequence(adev, golden_settings_gc_9_x_common, 955 (const u32)ARRAY_SIZE(golden_settings_gc_9_x_common)); 956 } 957 958 static void gfx_v9_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel, 959 bool wc, uint32_t reg, uint32_t val) 960 { 961 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 962 amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) | 963 WRITE_DATA_DST_SEL(0) | 964 (wc ? WR_CONFIRM : 0)); 965 amdgpu_ring_write(ring, reg); 966 amdgpu_ring_write(ring, 0); 967 amdgpu_ring_write(ring, val); 968 } 969 970 static void gfx_v9_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel, 971 int mem_space, int opt, uint32_t addr0, 972 uint32_t addr1, uint32_t ref, uint32_t mask, 973 uint32_t inv) 974 { 975 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 976 amdgpu_ring_write(ring, 977 /* memory (1) or register (0) */ 978 (WAIT_REG_MEM_MEM_SPACE(mem_space) | 979 WAIT_REG_MEM_OPERATION(opt) | /* wait */ 980 WAIT_REG_MEM_FUNCTION(3) | /* equal */ 981 WAIT_REG_MEM_ENGINE(eng_sel))); 982 983 if (mem_space) 984 BUG_ON(addr0 & 0x3); /* Dword align */ 985 amdgpu_ring_write(ring, addr0); 986 amdgpu_ring_write(ring, addr1); 987 amdgpu_ring_write(ring, ref); 988 amdgpu_ring_write(ring, mask); 989 amdgpu_ring_write(ring, inv); /* poll interval */ 990 } 991 992 static int gfx_v9_0_ring_test_ring(struct amdgpu_ring *ring) 993 { 994 struct amdgpu_device *adev = ring->adev; 995 uint32_t scratch = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0); 996 uint32_t tmp = 0; 997 unsigned i; 998 int r; 999 1000 WREG32(scratch, 0xCAFEDEAD); 1001 r = amdgpu_ring_alloc(ring, 3); 1002 if (r) 1003 return r; 1004 1005 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)); 1006 amdgpu_ring_write(ring, scratch - PACKET3_SET_UCONFIG_REG_START); 1007 amdgpu_ring_write(ring, 0xDEADBEEF); 1008 amdgpu_ring_commit(ring); 1009 1010 for (i = 0; i < adev->usec_timeout; i++) { 1011 tmp = RREG32(scratch); 1012 if (tmp == 0xDEADBEEF) 1013 break; 1014 udelay(1); 1015 } 1016 1017 if (i >= adev->usec_timeout) 1018 r = -ETIMEDOUT; 1019 return r; 1020 } 1021 1022 static int gfx_v9_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) 1023 { 1024 struct amdgpu_device *adev = ring->adev; 1025 struct amdgpu_ib ib; 1026 struct dma_fence *f = NULL; 1027 1028 unsigned index; 1029 uint64_t gpu_addr; 1030 uint32_t tmp; 1031 long r; 1032 1033 r = amdgpu_device_wb_get(adev, &index); 1034 if (r) 1035 return r; 1036 1037 gpu_addr = adev->wb.gpu_addr + (index * 4); 1038 adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD); 1039 memset(&ib, 0, sizeof(ib)); 1040 r = amdgpu_ib_get(adev, NULL, 16, 1041 AMDGPU_IB_POOL_DIRECT, &ib); 1042 if (r) 1043 goto err1; 1044 1045 ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3); 1046 ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM; 1047 ib.ptr[2] = lower_32_bits(gpu_addr); 1048 ib.ptr[3] = upper_32_bits(gpu_addr); 1049 ib.ptr[4] = 0xDEADBEEF; 1050 ib.length_dw = 5; 1051 1052 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); 1053 if (r) 1054 goto err2; 1055 1056 r = dma_fence_wait_timeout(f, false, timeout); 1057 if (r == 0) { 1058 r = -ETIMEDOUT; 1059 goto err2; 1060 } else if (r < 0) { 1061 goto err2; 1062 } 1063 1064 tmp = adev->wb.wb[index]; 1065 if (tmp == 0xDEADBEEF) 1066 r = 0; 1067 else 1068 r = -EINVAL; 1069 1070 err2: 1071 amdgpu_ib_free(adev, &ib, NULL); 1072 dma_fence_put(f); 1073 err1: 1074 amdgpu_device_wb_free(adev, index); 1075 return r; 1076 } 1077 1078 1079 static void gfx_v9_0_free_microcode(struct amdgpu_device *adev) 1080 { 1081 amdgpu_ucode_release(&adev->gfx.pfp_fw); 1082 amdgpu_ucode_release(&adev->gfx.me_fw); 1083 amdgpu_ucode_release(&adev->gfx.ce_fw); 1084 amdgpu_ucode_release(&adev->gfx.rlc_fw); 1085 amdgpu_ucode_release(&adev->gfx.mec_fw); 1086 amdgpu_ucode_release(&adev->gfx.mec2_fw); 1087 1088 kfree(adev->gfx.rlc.register_list_format); 1089 } 1090 1091 static void gfx_v9_0_check_fw_write_wait(struct amdgpu_device *adev) 1092 { 1093 adev->gfx.me_fw_write_wait = false; 1094 adev->gfx.mec_fw_write_wait = false; 1095 1096 if ((adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)) && 1097 ((adev->gfx.mec_fw_version < 0x000001a5) || 1098 (adev->gfx.mec_feature_version < 46) || 1099 (adev->gfx.pfp_fw_version < 0x000000b7) || 1100 (adev->gfx.pfp_feature_version < 46))) 1101 DRM_WARN_ONCE("CP firmware version too old, please update!"); 1102 1103 switch (adev->ip_versions[GC_HWIP][0]) { 1104 case IP_VERSION(9, 0, 1): 1105 if ((adev->gfx.me_fw_version >= 0x0000009c) && 1106 (adev->gfx.me_feature_version >= 42) && 1107 (adev->gfx.pfp_fw_version >= 0x000000b1) && 1108 (adev->gfx.pfp_feature_version >= 42)) 1109 adev->gfx.me_fw_write_wait = true; 1110 1111 if ((adev->gfx.mec_fw_version >= 0x00000193) && 1112 (adev->gfx.mec_feature_version >= 42)) 1113 adev->gfx.mec_fw_write_wait = true; 1114 break; 1115 case IP_VERSION(9, 2, 1): 1116 if ((adev->gfx.me_fw_version >= 0x0000009c) && 1117 (adev->gfx.me_feature_version >= 44) && 1118 (adev->gfx.pfp_fw_version >= 0x000000b2) && 1119 (adev->gfx.pfp_feature_version >= 44)) 1120 adev->gfx.me_fw_write_wait = true; 1121 1122 if ((adev->gfx.mec_fw_version >= 0x00000196) && 1123 (adev->gfx.mec_feature_version >= 44)) 1124 adev->gfx.mec_fw_write_wait = true; 1125 break; 1126 case IP_VERSION(9, 4, 0): 1127 if ((adev->gfx.me_fw_version >= 0x0000009c) && 1128 (adev->gfx.me_feature_version >= 44) && 1129 (adev->gfx.pfp_fw_version >= 0x000000b2) && 1130 (adev->gfx.pfp_feature_version >= 44)) 1131 adev->gfx.me_fw_write_wait = true; 1132 1133 if ((adev->gfx.mec_fw_version >= 0x00000197) && 1134 (adev->gfx.mec_feature_version >= 44)) 1135 adev->gfx.mec_fw_write_wait = true; 1136 break; 1137 case IP_VERSION(9, 1, 0): 1138 case IP_VERSION(9, 2, 2): 1139 if ((adev->gfx.me_fw_version >= 0x0000009c) && 1140 (adev->gfx.me_feature_version >= 42) && 1141 (adev->gfx.pfp_fw_version >= 0x000000b1) && 1142 (adev->gfx.pfp_feature_version >= 42)) 1143 adev->gfx.me_fw_write_wait = true; 1144 1145 if ((adev->gfx.mec_fw_version >= 0x00000192) && 1146 (adev->gfx.mec_feature_version >= 42)) 1147 adev->gfx.mec_fw_write_wait = true; 1148 break; 1149 default: 1150 adev->gfx.me_fw_write_wait = true; 1151 adev->gfx.mec_fw_write_wait = true; 1152 break; 1153 } 1154 } 1155 1156 struct amdgpu_gfxoff_quirk { 1157 u16 chip_vendor; 1158 u16 chip_device; 1159 u16 subsys_vendor; 1160 u16 subsys_device; 1161 u8 revision; 1162 }; 1163 1164 static const struct amdgpu_gfxoff_quirk amdgpu_gfxoff_quirk_list[] = { 1165 /* https://bugzilla.kernel.org/show_bug.cgi?id=204689 */ 1166 { 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc8 }, 1167 /* https://bugzilla.kernel.org/show_bug.cgi?id=207171 */ 1168 { 0x1002, 0x15dd, 0x103c, 0x83e7, 0xd3 }, 1169 /* GFXOFF is unstable on C6 parts with a VBIOS 113-RAVEN-114 */ 1170 { 0x1002, 0x15dd, 0x1002, 0x15dd, 0xc6 }, 1171 /* Apple MacBook Pro (15-inch, 2019) Radeon Pro Vega 20 4 GB */ 1172 { 0x1002, 0x69af, 0x106b, 0x019a, 0xc0 }, 1173 { 0, 0, 0, 0, 0 }, 1174 }; 1175 1176 static bool gfx_v9_0_should_disable_gfxoff(struct pci_dev *pdev) 1177 { 1178 const struct amdgpu_gfxoff_quirk *p = amdgpu_gfxoff_quirk_list; 1179 1180 while (p && p->chip_device != 0) { 1181 if (pdev->vendor == p->chip_vendor && 1182 pdev->device == p->chip_device && 1183 pdev->subsystem_vendor == p->subsys_vendor && 1184 pdev->subsystem_device == p->subsys_device && 1185 pdev->revision == p->revision) { 1186 return true; 1187 } 1188 ++p; 1189 } 1190 return false; 1191 } 1192 1193 static bool is_raven_kicker(struct amdgpu_device *adev) 1194 { 1195 if (adev->pm.fw_version >= 0x41e2b) 1196 return true; 1197 else 1198 return false; 1199 } 1200 1201 static bool check_if_enlarge_doorbell_range(struct amdgpu_device *adev) 1202 { 1203 if ((adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 3, 0)) && 1204 (adev->gfx.me_fw_version >= 0x000000a5) && 1205 (adev->gfx.me_feature_version >= 52)) 1206 return true; 1207 else 1208 return false; 1209 } 1210 1211 static void gfx_v9_0_check_if_need_gfxoff(struct amdgpu_device *adev) 1212 { 1213 if (gfx_v9_0_should_disable_gfxoff(adev->pdev)) 1214 adev->pm.pp_feature &= ~PP_GFXOFF_MASK; 1215 1216 switch (adev->ip_versions[GC_HWIP][0]) { 1217 case IP_VERSION(9, 0, 1): 1218 case IP_VERSION(9, 2, 1): 1219 case IP_VERSION(9, 4, 0): 1220 break; 1221 case IP_VERSION(9, 2, 2): 1222 case IP_VERSION(9, 1, 0): 1223 if (!((adev->apu_flags & AMD_APU_IS_RAVEN2) || 1224 (adev->apu_flags & AMD_APU_IS_PICASSO)) && 1225 ((!is_raven_kicker(adev) && 1226 adev->gfx.rlc_fw_version < 531) || 1227 (adev->gfx.rlc_feature_version < 1) || 1228 !adev->gfx.rlc.is_rlc_v2_1)) 1229 adev->pm.pp_feature &= ~PP_GFXOFF_MASK; 1230 1231 if (adev->pm.pp_feature & PP_GFXOFF_MASK) 1232 adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG | 1233 AMD_PG_SUPPORT_CP | 1234 AMD_PG_SUPPORT_RLC_SMU_HS; 1235 break; 1236 case IP_VERSION(9, 3, 0): 1237 if (adev->pm.pp_feature & PP_GFXOFF_MASK) 1238 adev->pg_flags |= AMD_PG_SUPPORT_GFX_PG | 1239 AMD_PG_SUPPORT_CP | 1240 AMD_PG_SUPPORT_RLC_SMU_HS; 1241 break; 1242 default: 1243 break; 1244 } 1245 } 1246 1247 static int gfx_v9_0_init_cp_gfx_microcode(struct amdgpu_device *adev, 1248 char *chip_name) 1249 { 1250 char fw_name[30]; 1251 int err; 1252 1253 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name); 1254 err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw, fw_name); 1255 if (err) 1256 goto out; 1257 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_PFP); 1258 1259 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name); 1260 err = amdgpu_ucode_request(adev, &adev->gfx.me_fw, fw_name); 1261 if (err) 1262 goto out; 1263 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_ME); 1264 1265 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name); 1266 err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw, fw_name); 1267 if (err) 1268 goto out; 1269 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_CE); 1270 1271 out: 1272 if (err) { 1273 amdgpu_ucode_release(&adev->gfx.pfp_fw); 1274 amdgpu_ucode_release(&adev->gfx.me_fw); 1275 amdgpu_ucode_release(&adev->gfx.ce_fw); 1276 } 1277 return err; 1278 } 1279 1280 static int gfx_v9_0_init_rlc_microcode(struct amdgpu_device *adev, 1281 char *chip_name) 1282 { 1283 char fw_name[30]; 1284 int err; 1285 const struct rlc_firmware_header_v2_0 *rlc_hdr; 1286 uint16_t version_major; 1287 uint16_t version_minor; 1288 uint32_t smu_version; 1289 1290 /* 1291 * For Picasso && AM4 SOCKET board, we use picasso_rlc_am4.bin 1292 * instead of picasso_rlc.bin. 1293 * Judgment method: 1294 * PCO AM4: revision >= 0xC8 && revision <= 0xCF 1295 * or revision >= 0xD8 && revision <= 0xDF 1296 * otherwise is PCO FP5 1297 */ 1298 if (!strcmp(chip_name, "picasso") && 1299 (((adev->pdev->revision >= 0xC8) && (adev->pdev->revision <= 0xCF)) || 1300 ((adev->pdev->revision >= 0xD8) && (adev->pdev->revision <= 0xDF)))) 1301 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc_am4.bin", chip_name); 1302 else if (!strcmp(chip_name, "raven") && (amdgpu_pm_load_smu_firmware(adev, &smu_version) == 0) && 1303 (smu_version >= 0x41e2b)) 1304 /** 1305 *SMC is loaded by SBIOS on APU and it's able to get the SMU version directly. 1306 */ 1307 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_kicker_rlc.bin", chip_name); 1308 else 1309 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name); 1310 err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, fw_name); 1311 if (err) 1312 goto out; 1313 rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data; 1314 1315 version_major = le16_to_cpu(rlc_hdr->header.header_version_major); 1316 version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor); 1317 err = amdgpu_gfx_rlc_init_microcode(adev, version_major, version_minor); 1318 out: 1319 if (err) 1320 amdgpu_ucode_release(&adev->gfx.rlc_fw); 1321 1322 return err; 1323 } 1324 1325 static bool gfx_v9_0_load_mec2_fw_bin_support(struct amdgpu_device *adev) 1326 { 1327 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2) || 1328 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1) || 1329 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 3, 0)) 1330 return false; 1331 1332 return true; 1333 } 1334 1335 static int gfx_v9_0_init_cp_compute_microcode(struct amdgpu_device *adev, 1336 char *chip_name) 1337 { 1338 char fw_name[30]; 1339 int err; 1340 1341 if (amdgpu_sriov_vf(adev) && (adev->asic_type == CHIP_ALDEBARAN)) 1342 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sjt_mec.bin", chip_name); 1343 else 1344 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name); 1345 1346 err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, fw_name); 1347 if (err) 1348 goto out; 1349 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1); 1350 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1_JT); 1351 1352 if (gfx_v9_0_load_mec2_fw_bin_support(adev)) { 1353 if (amdgpu_sriov_vf(adev) && (adev->asic_type == CHIP_ALDEBARAN)) 1354 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sjt_mec2.bin", chip_name); 1355 else 1356 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name); 1357 1358 /* ignore failures to load */ 1359 err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw, fw_name); 1360 if (!err) { 1361 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC2); 1362 amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC2_JT); 1363 } else { 1364 err = 0; 1365 amdgpu_ucode_release(&adev->gfx.mec2_fw); 1366 } 1367 } else { 1368 adev->gfx.mec2_fw_version = adev->gfx.mec_fw_version; 1369 adev->gfx.mec2_feature_version = adev->gfx.mec_feature_version; 1370 } 1371 1372 gfx_v9_0_check_if_need_gfxoff(adev); 1373 gfx_v9_0_check_fw_write_wait(adev); 1374 1375 out: 1376 if (err) 1377 amdgpu_ucode_release(&adev->gfx.mec_fw); 1378 return err; 1379 } 1380 1381 static int gfx_v9_0_init_microcode(struct amdgpu_device *adev) 1382 { 1383 char ucode_prefix[30]; 1384 int r; 1385 1386 DRM_DEBUG("\n"); 1387 amdgpu_ucode_ip_version_decode(adev, GC_HWIP, ucode_prefix, sizeof(ucode_prefix)); 1388 1389 /* No CPG in Arcturus */ 1390 if (adev->gfx.num_gfx_rings) { 1391 r = gfx_v9_0_init_cp_gfx_microcode(adev, ucode_prefix); 1392 if (r) 1393 return r; 1394 } 1395 1396 r = gfx_v9_0_init_rlc_microcode(adev, ucode_prefix); 1397 if (r) 1398 return r; 1399 1400 r = gfx_v9_0_init_cp_compute_microcode(adev, ucode_prefix); 1401 if (r) 1402 return r; 1403 1404 return r; 1405 } 1406 1407 static u32 gfx_v9_0_get_csb_size(struct amdgpu_device *adev) 1408 { 1409 u32 count = 0; 1410 const struct cs_section_def *sect = NULL; 1411 const struct cs_extent_def *ext = NULL; 1412 1413 /* begin clear state */ 1414 count += 2; 1415 /* context control state */ 1416 count += 3; 1417 1418 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) { 1419 for (ext = sect->section; ext->extent != NULL; ++ext) { 1420 if (sect->id == SECT_CONTEXT) 1421 count += 2 + ext->reg_count; 1422 else 1423 return 0; 1424 } 1425 } 1426 1427 /* end clear state */ 1428 count += 2; 1429 /* clear state */ 1430 count += 2; 1431 1432 return count; 1433 } 1434 1435 static void gfx_v9_0_get_csb_buffer(struct amdgpu_device *adev, 1436 volatile u32 *buffer) 1437 { 1438 u32 count = 0, i; 1439 const struct cs_section_def *sect = NULL; 1440 const struct cs_extent_def *ext = NULL; 1441 1442 if (adev->gfx.rlc.cs_data == NULL) 1443 return; 1444 if (buffer == NULL) 1445 return; 1446 1447 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 1448 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 1449 1450 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 1451 buffer[count++] = cpu_to_le32(0x80000000); 1452 buffer[count++] = cpu_to_le32(0x80000000); 1453 1454 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 1455 for (ext = sect->section; ext->extent != NULL; ++ext) { 1456 if (sect->id == SECT_CONTEXT) { 1457 buffer[count++] = 1458 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count)); 1459 buffer[count++] = cpu_to_le32(ext->reg_index - 1460 PACKET3_SET_CONTEXT_REG_START); 1461 for (i = 0; i < ext->reg_count; i++) 1462 buffer[count++] = cpu_to_le32(ext->extent[i]); 1463 } else { 1464 return; 1465 } 1466 } 1467 } 1468 1469 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 1470 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE); 1471 1472 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0)); 1473 buffer[count++] = cpu_to_le32(0); 1474 } 1475 1476 static void gfx_v9_0_init_always_on_cu_mask(struct amdgpu_device *adev) 1477 { 1478 struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info; 1479 uint32_t pg_always_on_cu_num = 2; 1480 uint32_t always_on_cu_num; 1481 uint32_t i, j, k; 1482 uint32_t mask, cu_bitmap, counter; 1483 1484 if (adev->flags & AMD_IS_APU) 1485 always_on_cu_num = 4; 1486 else if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 2, 1)) 1487 always_on_cu_num = 8; 1488 else 1489 always_on_cu_num = 12; 1490 1491 mutex_lock(&adev->grbm_idx_mutex); 1492 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 1493 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 1494 mask = 1; 1495 cu_bitmap = 0; 1496 counter = 0; 1497 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff); 1498 1499 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) { 1500 if (cu_info->bitmap[i][j] & mask) { 1501 if (counter == pg_always_on_cu_num) 1502 WREG32_SOC15(GC, 0, mmRLC_PG_ALWAYS_ON_CU_MASK, cu_bitmap); 1503 if (counter < always_on_cu_num) 1504 cu_bitmap |= mask; 1505 else 1506 break; 1507 counter++; 1508 } 1509 mask <<= 1; 1510 } 1511 1512 WREG32_SOC15(GC, 0, mmRLC_LB_ALWAYS_ACTIVE_CU_MASK, cu_bitmap); 1513 cu_info->ao_cu_bitmap[i][j] = cu_bitmap; 1514 } 1515 } 1516 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 1517 mutex_unlock(&adev->grbm_idx_mutex); 1518 } 1519 1520 static void gfx_v9_0_init_lbpw(struct amdgpu_device *adev) 1521 { 1522 uint32_t data; 1523 1524 /* set mmRLC_LB_THR_CONFIG_1/2/3/4 */ 1525 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F); 1526 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x0333A5A7); 1527 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077); 1528 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x30 | 0x40 << 8 | 0x02FA << 16)); 1529 1530 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */ 1531 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000); 1532 1533 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */ 1534 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000500); 1535 1536 mutex_lock(&adev->grbm_idx_mutex); 1537 /* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/ 1538 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 1539 WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff); 1540 1541 /* set mmRLC_LB_PARAMS = 0x003F_1006 */ 1542 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003); 1543 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010); 1544 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F); 1545 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data); 1546 1547 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */ 1548 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7); 1549 data &= 0x0000FFFF; 1550 data |= 0x00C00000; 1551 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data); 1552 1553 /* 1554 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xF (4 CUs AON for Raven), 1555 * programmed in gfx_v9_0_init_always_on_cu_mask() 1556 */ 1557 1558 /* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved, 1559 * but used for RLC_LB_CNTL configuration */ 1560 data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK; 1561 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09); 1562 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000); 1563 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data); 1564 mutex_unlock(&adev->grbm_idx_mutex); 1565 1566 gfx_v9_0_init_always_on_cu_mask(adev); 1567 } 1568 1569 static void gfx_v9_4_init_lbpw(struct amdgpu_device *adev) 1570 { 1571 uint32_t data; 1572 1573 /* set mmRLC_LB_THR_CONFIG_1/2/3/4 */ 1574 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_1, 0x0000007F); 1575 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_2, 0x033388F8); 1576 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_3, 0x00000077); 1577 WREG32_SOC15(GC, 0, mmRLC_LB_THR_CONFIG_4, (0x10 | 0x27 << 8 | 0x02FA << 16)); 1578 1579 /* set mmRLC_LB_CNTR_INIT = 0x0000_0000 */ 1580 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_INIT, 0x00000000); 1581 1582 /* set mmRLC_LB_CNTR_MAX = 0x0000_0500 */ 1583 WREG32_SOC15(GC, 0, mmRLC_LB_CNTR_MAX, 0x00000800); 1584 1585 mutex_lock(&adev->grbm_idx_mutex); 1586 /* set mmRLC_LB_INIT_CU_MASK thru broadcast mode to enable all SE/SH*/ 1587 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 1588 WREG32_SOC15(GC, 0, mmRLC_LB_INIT_CU_MASK, 0xffffffff); 1589 1590 /* set mmRLC_LB_PARAMS = 0x003F_1006 */ 1591 data = REG_SET_FIELD(0, RLC_LB_PARAMS, FIFO_SAMPLES, 0x0003); 1592 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLES, 0x0010); 1593 data |= REG_SET_FIELD(data, RLC_LB_PARAMS, PG_IDLE_SAMPLE_INTERVAL, 0x033F); 1594 WREG32_SOC15(GC, 0, mmRLC_LB_PARAMS, data); 1595 1596 /* set mmRLC_GPM_GENERAL_7[31-16] = 0x00C0 */ 1597 data = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7); 1598 data &= 0x0000FFFF; 1599 data |= 0x00C00000; 1600 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_7, data); 1601 1602 /* 1603 * RLC_LB_ALWAYS_ACTIVE_CU_MASK = 0xFFF (12 CUs AON), 1604 * programmed in gfx_v9_0_init_always_on_cu_mask() 1605 */ 1606 1607 /* set RLC_LB_CNTL = 0x8000_0095, 31 bit is reserved, 1608 * but used for RLC_LB_CNTL configuration */ 1609 data = RLC_LB_CNTL__LB_CNT_SPIM_ACTIVE_MASK; 1610 data |= REG_SET_FIELD(data, RLC_LB_CNTL, CU_MASK_USED_OFF_HYST, 0x09); 1611 data |= REG_SET_FIELD(data, RLC_LB_CNTL, RESERVED, 0x80000); 1612 WREG32_SOC15(GC, 0, mmRLC_LB_CNTL, data); 1613 mutex_unlock(&adev->grbm_idx_mutex); 1614 1615 gfx_v9_0_init_always_on_cu_mask(adev); 1616 } 1617 1618 static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool enable) 1619 { 1620 WREG32_FIELD15(GC, 0, RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0); 1621 } 1622 1623 static int gfx_v9_0_cp_jump_table_num(struct amdgpu_device *adev) 1624 { 1625 if (gfx_v9_0_load_mec2_fw_bin_support(adev)) 1626 return 5; 1627 else 1628 return 4; 1629 } 1630 1631 static void gfx_v9_0_init_rlcg_reg_access_ctrl(struct amdgpu_device *adev) 1632 { 1633 struct amdgpu_rlcg_reg_access_ctrl *reg_access_ctrl; 1634 1635 reg_access_ctrl = &adev->gfx.rlc.reg_access_ctrl; 1636 reg_access_ctrl->scratch_reg0 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0); 1637 reg_access_ctrl->scratch_reg1 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG1); 1638 reg_access_ctrl->scratch_reg2 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG2); 1639 reg_access_ctrl->scratch_reg3 = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG3); 1640 reg_access_ctrl->grbm_cntl = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_CNTL); 1641 reg_access_ctrl->grbm_idx = SOC15_REG_OFFSET(GC, 0, mmGRBM_GFX_INDEX); 1642 reg_access_ctrl->spare_int = SOC15_REG_OFFSET(GC, 0, mmRLC_SPARE_INT); 1643 adev->gfx.rlc.rlcg_reg_access_supported = true; 1644 } 1645 1646 static int gfx_v9_0_rlc_init(struct amdgpu_device *adev) 1647 { 1648 const struct cs_section_def *cs_data; 1649 int r; 1650 1651 adev->gfx.rlc.cs_data = gfx9_cs_data; 1652 1653 cs_data = adev->gfx.rlc.cs_data; 1654 1655 if (cs_data) { 1656 /* init clear state block */ 1657 r = amdgpu_gfx_rlc_init_csb(adev); 1658 if (r) 1659 return r; 1660 } 1661 1662 if (adev->flags & AMD_IS_APU) { 1663 /* TODO: double check the cp_table_size for RV */ 1664 adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */ 1665 r = amdgpu_gfx_rlc_init_cpt(adev); 1666 if (r) 1667 return r; 1668 } 1669 1670 switch (adev->ip_versions[GC_HWIP][0]) { 1671 case IP_VERSION(9, 2, 2): 1672 case IP_VERSION(9, 1, 0): 1673 gfx_v9_0_init_lbpw(adev); 1674 break; 1675 case IP_VERSION(9, 4, 0): 1676 gfx_v9_4_init_lbpw(adev); 1677 break; 1678 default: 1679 break; 1680 } 1681 1682 /* init spm vmid with 0xf */ 1683 if (adev->gfx.rlc.funcs->update_spm_vmid) 1684 adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf); 1685 1686 return 0; 1687 } 1688 1689 static void gfx_v9_0_mec_fini(struct amdgpu_device *adev) 1690 { 1691 amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL); 1692 amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL); 1693 } 1694 1695 static int gfx_v9_0_mec_init(struct amdgpu_device *adev) 1696 { 1697 int r; 1698 u32 *hpd; 1699 const __le32 *fw_data; 1700 unsigned fw_size; 1701 u32 *fw; 1702 size_t mec_hpd_size; 1703 1704 const struct gfx_firmware_header_v1_0 *mec_hdr; 1705 1706 bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); 1707 1708 /* take ownership of the relevant compute queues */ 1709 amdgpu_gfx_compute_queue_acquire(adev); 1710 mec_hpd_size = adev->gfx.num_compute_rings * GFX9_MEC_HPD_SIZE; 1711 if (mec_hpd_size) { 1712 r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE, 1713 AMDGPU_GEM_DOMAIN_VRAM | 1714 AMDGPU_GEM_DOMAIN_GTT, 1715 &adev->gfx.mec.hpd_eop_obj, 1716 &adev->gfx.mec.hpd_eop_gpu_addr, 1717 (void **)&hpd); 1718 if (r) { 1719 dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r); 1720 gfx_v9_0_mec_fini(adev); 1721 return r; 1722 } 1723 1724 memset(hpd, 0, mec_hpd_size); 1725 1726 amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj); 1727 amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj); 1728 } 1729 1730 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; 1731 1732 fw_data = (const __le32 *) 1733 (adev->gfx.mec_fw->data + 1734 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes)); 1735 fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes); 1736 1737 r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes, 1738 PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT, 1739 &adev->gfx.mec.mec_fw_obj, 1740 &adev->gfx.mec.mec_fw_gpu_addr, 1741 (void **)&fw); 1742 if (r) { 1743 dev_warn(adev->dev, "(%d) create mec firmware bo failed\n", r); 1744 gfx_v9_0_mec_fini(adev); 1745 return r; 1746 } 1747 1748 memcpy(fw, fw_data, fw_size); 1749 1750 amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj); 1751 amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj); 1752 1753 return 0; 1754 } 1755 1756 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address) 1757 { 1758 WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX, 1759 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 1760 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 1761 (address << SQ_IND_INDEX__INDEX__SHIFT) | 1762 (SQ_IND_INDEX__FORCE_READ_MASK)); 1763 return RREG32_SOC15(GC, 0, mmSQ_IND_DATA); 1764 } 1765 1766 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd, 1767 uint32_t wave, uint32_t thread, 1768 uint32_t regno, uint32_t num, uint32_t *out) 1769 { 1770 WREG32_SOC15_RLC(GC, 0, mmSQ_IND_INDEX, 1771 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 1772 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 1773 (regno << SQ_IND_INDEX__INDEX__SHIFT) | 1774 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) | 1775 (SQ_IND_INDEX__FORCE_READ_MASK) | 1776 (SQ_IND_INDEX__AUTO_INCR_MASK)); 1777 while (num--) 1778 *(out++) = RREG32_SOC15(GC, 0, mmSQ_IND_DATA); 1779 } 1780 1781 static void gfx_v9_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields) 1782 { 1783 /* type 1 wave data */ 1784 dst[(*no_fields)++] = 1; 1785 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS); 1786 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO); 1787 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI); 1788 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO); 1789 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI); 1790 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID); 1791 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0); 1792 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1); 1793 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC); 1794 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC); 1795 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS); 1796 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS); 1797 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0); 1798 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0); 1799 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE); 1800 } 1801 1802 static void gfx_v9_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd, 1803 uint32_t wave, uint32_t start, 1804 uint32_t size, uint32_t *dst) 1805 { 1806 wave_read_regs( 1807 adev, simd, wave, 0, 1808 start + SQIND_WAVE_SGPRS_OFFSET, size, dst); 1809 } 1810 1811 static void gfx_v9_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t simd, 1812 uint32_t wave, uint32_t thread, 1813 uint32_t start, uint32_t size, 1814 uint32_t *dst) 1815 { 1816 wave_read_regs( 1817 adev, simd, wave, thread, 1818 start + SQIND_WAVE_VGPRS_OFFSET, size, dst); 1819 } 1820 1821 static void gfx_v9_0_select_me_pipe_q(struct amdgpu_device *adev, 1822 u32 me, u32 pipe, u32 q, u32 vm) 1823 { 1824 soc15_grbm_select(adev, me, pipe, q, vm); 1825 } 1826 1827 static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = { 1828 .get_gpu_clock_counter = &gfx_v9_0_get_gpu_clock_counter, 1829 .select_se_sh = &gfx_v9_0_select_se_sh, 1830 .read_wave_data = &gfx_v9_0_read_wave_data, 1831 .read_wave_sgprs = &gfx_v9_0_read_wave_sgprs, 1832 .read_wave_vgprs = &gfx_v9_0_read_wave_vgprs, 1833 .select_me_pipe_q = &gfx_v9_0_select_me_pipe_q, 1834 }; 1835 1836 const struct amdgpu_ras_block_hw_ops gfx_v9_0_ras_ops = { 1837 .ras_error_inject = &gfx_v9_0_ras_error_inject, 1838 .query_ras_error_count = &gfx_v9_0_query_ras_error_count, 1839 .reset_ras_error_count = &gfx_v9_0_reset_ras_error_count, 1840 }; 1841 1842 static struct amdgpu_gfx_ras gfx_v9_0_ras = { 1843 .ras_block = { 1844 .hw_ops = &gfx_v9_0_ras_ops, 1845 }, 1846 }; 1847 1848 static int gfx_v9_0_gpu_early_init(struct amdgpu_device *adev) 1849 { 1850 u32 gb_addr_config; 1851 int err; 1852 1853 switch (adev->ip_versions[GC_HWIP][0]) { 1854 case IP_VERSION(9, 0, 1): 1855 adev->gfx.config.max_hw_contexts = 8; 1856 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1857 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1858 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1859 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0; 1860 gb_addr_config = VEGA10_GB_ADDR_CONFIG_GOLDEN; 1861 break; 1862 case IP_VERSION(9, 2, 1): 1863 adev->gfx.config.max_hw_contexts = 8; 1864 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1865 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1866 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1867 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0; 1868 gb_addr_config = VEGA12_GB_ADDR_CONFIG_GOLDEN; 1869 DRM_INFO("fix gfx.config for vega12\n"); 1870 break; 1871 case IP_VERSION(9, 4, 0): 1872 adev->gfx.ras = &gfx_v9_0_ras; 1873 adev->gfx.config.max_hw_contexts = 8; 1874 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1875 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1876 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1877 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0; 1878 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG); 1879 gb_addr_config &= ~0xf3e777ff; 1880 gb_addr_config |= 0x22014042; 1881 /* check vbios table if gpu info is not available */ 1882 err = amdgpu_atomfirmware_get_gfx_info(adev); 1883 if (err) 1884 return err; 1885 break; 1886 case IP_VERSION(9, 2, 2): 1887 case IP_VERSION(9, 1, 0): 1888 adev->gfx.config.max_hw_contexts = 8; 1889 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1890 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1891 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1892 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0; 1893 if (adev->apu_flags & AMD_APU_IS_RAVEN2) 1894 gb_addr_config = RAVEN2_GB_ADDR_CONFIG_GOLDEN; 1895 else 1896 gb_addr_config = RAVEN_GB_ADDR_CONFIG_GOLDEN; 1897 break; 1898 case IP_VERSION(9, 4, 1): 1899 adev->gfx.ras = &gfx_v9_4_ras; 1900 adev->gfx.config.max_hw_contexts = 8; 1901 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1902 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1903 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1904 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0; 1905 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG); 1906 gb_addr_config &= ~0xf3e777ff; 1907 gb_addr_config |= 0x22014042; 1908 break; 1909 case IP_VERSION(9, 3, 0): 1910 adev->gfx.config.max_hw_contexts = 8; 1911 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1912 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1913 adev->gfx.config.sc_hiz_tile_fifo_size = 0x80; 1914 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0; 1915 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG); 1916 gb_addr_config &= ~0xf3e777ff; 1917 gb_addr_config |= 0x22010042; 1918 break; 1919 case IP_VERSION(9, 4, 2): 1920 adev->gfx.ras = &gfx_v9_4_2_ras; 1921 adev->gfx.config.max_hw_contexts = 8; 1922 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1923 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1924 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1925 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0; 1926 gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG); 1927 gb_addr_config &= ~0xf3e777ff; 1928 gb_addr_config |= 0x22014042; 1929 /* check vbios table if gpu info is not available */ 1930 err = amdgpu_atomfirmware_get_gfx_info(adev); 1931 if (err) 1932 return err; 1933 break; 1934 default: 1935 BUG(); 1936 break; 1937 } 1938 1939 adev->gfx.config.gb_addr_config = gb_addr_config; 1940 1941 adev->gfx.config.gb_addr_config_fields.num_pipes = 1 << 1942 REG_GET_FIELD( 1943 adev->gfx.config.gb_addr_config, 1944 GB_ADDR_CONFIG, 1945 NUM_PIPES); 1946 1947 adev->gfx.config.max_tile_pipes = 1948 adev->gfx.config.gb_addr_config_fields.num_pipes; 1949 1950 adev->gfx.config.gb_addr_config_fields.num_banks = 1 << 1951 REG_GET_FIELD( 1952 adev->gfx.config.gb_addr_config, 1953 GB_ADDR_CONFIG, 1954 NUM_BANKS); 1955 adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 << 1956 REG_GET_FIELD( 1957 adev->gfx.config.gb_addr_config, 1958 GB_ADDR_CONFIG, 1959 MAX_COMPRESSED_FRAGS); 1960 adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 << 1961 REG_GET_FIELD( 1962 adev->gfx.config.gb_addr_config, 1963 GB_ADDR_CONFIG, 1964 NUM_RB_PER_SE); 1965 adev->gfx.config.gb_addr_config_fields.num_se = 1 << 1966 REG_GET_FIELD( 1967 adev->gfx.config.gb_addr_config, 1968 GB_ADDR_CONFIG, 1969 NUM_SHADER_ENGINES); 1970 adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 + 1971 REG_GET_FIELD( 1972 adev->gfx.config.gb_addr_config, 1973 GB_ADDR_CONFIG, 1974 PIPE_INTERLEAVE_SIZE)); 1975 1976 return 0; 1977 } 1978 1979 static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id, 1980 int mec, int pipe, int queue) 1981 { 1982 unsigned irq_type; 1983 struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id]; 1984 unsigned int hw_prio; 1985 1986 ring = &adev->gfx.compute_ring[ring_id]; 1987 1988 /* mec0 is me1 */ 1989 ring->me = mec + 1; 1990 ring->pipe = pipe; 1991 ring->queue = queue; 1992 1993 ring->ring_obj = NULL; 1994 ring->use_doorbell = true; 1995 ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1; 1996 ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr 1997 + (ring_id * GFX9_MEC_HPD_SIZE); 1998 ring->vm_hub = AMDGPU_GFXHUB_0; 1999 sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue); 2000 2001 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP 2002 + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec) 2003 + ring->pipe; 2004 hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ? 2005 AMDGPU_RING_PRIO_2 : AMDGPU_RING_PRIO_DEFAULT; 2006 /* type-2 packets are deprecated on MEC, use type-3 instead */ 2007 return amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type, 2008 hw_prio, NULL); 2009 } 2010 2011 static int gfx_v9_0_sw_init(void *handle) 2012 { 2013 int i, j, k, r, ring_id; 2014 struct amdgpu_ring *ring; 2015 struct amdgpu_kiq *kiq; 2016 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 2017 unsigned int hw_prio; 2018 2019 switch (adev->ip_versions[GC_HWIP][0]) { 2020 case IP_VERSION(9, 0, 1): 2021 case IP_VERSION(9, 2, 1): 2022 case IP_VERSION(9, 4, 0): 2023 case IP_VERSION(9, 2, 2): 2024 case IP_VERSION(9, 1, 0): 2025 case IP_VERSION(9, 4, 1): 2026 case IP_VERSION(9, 3, 0): 2027 case IP_VERSION(9, 4, 2): 2028 adev->gfx.mec.num_mec = 2; 2029 break; 2030 default: 2031 adev->gfx.mec.num_mec = 1; 2032 break; 2033 } 2034 2035 adev->gfx.mec.num_pipe_per_mec = 4; 2036 adev->gfx.mec.num_queue_per_pipe = 8; 2037 2038 /* EOP Event */ 2039 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_EOP_INTERRUPT, &adev->gfx.eop_irq); 2040 if (r) 2041 return r; 2042 2043 /* Privileged reg */ 2044 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_REG_FAULT, 2045 &adev->gfx.priv_reg_irq); 2046 if (r) 2047 return r; 2048 2049 /* Privileged inst */ 2050 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_INSTR_FAULT, 2051 &adev->gfx.priv_inst_irq); 2052 if (r) 2053 return r; 2054 2055 /* ECC error */ 2056 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_ECC_ERROR, 2057 &adev->gfx.cp_ecc_error_irq); 2058 if (r) 2059 return r; 2060 2061 /* FUE error */ 2062 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_FUE_ERROR, 2063 &adev->gfx.cp_ecc_error_irq); 2064 if (r) 2065 return r; 2066 2067 adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE; 2068 2069 if (adev->gfx.rlc.funcs) { 2070 if (adev->gfx.rlc.funcs->init) { 2071 r = adev->gfx.rlc.funcs->init(adev); 2072 if (r) { 2073 dev_err(adev->dev, "Failed to init rlc BOs!\n"); 2074 return r; 2075 } 2076 } 2077 } 2078 2079 r = gfx_v9_0_mec_init(adev); 2080 if (r) { 2081 DRM_ERROR("Failed to init MEC BOs!\n"); 2082 return r; 2083 } 2084 2085 /* set up the gfx ring */ 2086 for (i = 0; i < adev->gfx.num_gfx_rings; i++) { 2087 ring = &adev->gfx.gfx_ring[i]; 2088 ring->ring_obj = NULL; 2089 if (!i) 2090 sprintf(ring->name, "gfx"); 2091 else 2092 sprintf(ring->name, "gfx_%d", i); 2093 ring->use_doorbell = true; 2094 ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1; 2095 2096 /* disable scheduler on the real ring */ 2097 ring->no_scheduler = true; 2098 ring->vm_hub = AMDGPU_GFXHUB_0; 2099 r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, 2100 AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP, 2101 AMDGPU_RING_PRIO_DEFAULT, NULL); 2102 if (r) 2103 return r; 2104 } 2105 2106 /* set up the software rings */ 2107 if (adev->gfx.num_gfx_rings) { 2108 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++) { 2109 ring = &adev->gfx.sw_gfx_ring[i]; 2110 ring->ring_obj = NULL; 2111 sprintf(ring->name, amdgpu_sw_ring_name(i)); 2112 ring->use_doorbell = true; 2113 ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1; 2114 ring->is_sw_ring = true; 2115 hw_prio = amdgpu_sw_ring_priority(i); 2116 ring->vm_hub = AMDGPU_GFXHUB_0; 2117 r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, 2118 AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP, hw_prio, 2119 NULL); 2120 if (r) 2121 return r; 2122 ring->wptr = 0; 2123 } 2124 2125 /* init the muxer and add software rings */ 2126 r = amdgpu_ring_mux_init(&adev->gfx.muxer, &adev->gfx.gfx_ring[0], 2127 GFX9_NUM_SW_GFX_RINGS); 2128 if (r) { 2129 DRM_ERROR("amdgpu_ring_mux_init failed(%d)\n", r); 2130 return r; 2131 } 2132 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++) { 2133 r = amdgpu_ring_mux_add_sw_ring(&adev->gfx.muxer, 2134 &adev->gfx.sw_gfx_ring[i]); 2135 if (r) { 2136 DRM_ERROR("amdgpu_ring_mux_add_sw_ring failed(%d)\n", r); 2137 return r; 2138 } 2139 } 2140 } 2141 2142 /* set up the compute queues - allocate horizontally across pipes */ 2143 ring_id = 0; 2144 for (i = 0; i < adev->gfx.mec.num_mec; ++i) { 2145 for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) { 2146 for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) { 2147 if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j)) 2148 continue; 2149 2150 r = gfx_v9_0_compute_ring_init(adev, 2151 ring_id, 2152 i, k, j); 2153 if (r) 2154 return r; 2155 2156 ring_id++; 2157 } 2158 } 2159 } 2160 2161 r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE); 2162 if (r) { 2163 DRM_ERROR("Failed to init KIQ BOs!\n"); 2164 return r; 2165 } 2166 2167 kiq = &adev->gfx.kiq; 2168 r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq); 2169 if (r) 2170 return r; 2171 2172 /* create MQD for all compute queues as wel as KIQ for SRIOV case */ 2173 r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v9_mqd_allocation)); 2174 if (r) 2175 return r; 2176 2177 adev->gfx.ce_ram_size = 0x8000; 2178 2179 r = gfx_v9_0_gpu_early_init(adev); 2180 if (r) 2181 return r; 2182 2183 if (amdgpu_gfx_ras_sw_init(adev)) { 2184 dev_err(adev->dev, "Failed to initialize gfx ras block!\n"); 2185 return -EINVAL; 2186 } 2187 2188 return 0; 2189 } 2190 2191 2192 static int gfx_v9_0_sw_fini(void *handle) 2193 { 2194 int i; 2195 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 2196 2197 if (adev->gfx.num_gfx_rings) { 2198 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++) 2199 amdgpu_ring_fini(&adev->gfx.sw_gfx_ring[i]); 2200 amdgpu_ring_mux_fini(&adev->gfx.muxer); 2201 } 2202 2203 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 2204 amdgpu_ring_fini(&adev->gfx.gfx_ring[i]); 2205 for (i = 0; i < adev->gfx.num_compute_rings; i++) 2206 amdgpu_ring_fini(&adev->gfx.compute_ring[i]); 2207 2208 amdgpu_gfx_mqd_sw_fini(adev); 2209 amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring); 2210 amdgpu_gfx_kiq_fini(adev); 2211 2212 gfx_v9_0_mec_fini(adev); 2213 amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj, 2214 &adev->gfx.rlc.clear_state_gpu_addr, 2215 (void **)&adev->gfx.rlc.cs_ptr); 2216 if (adev->flags & AMD_IS_APU) { 2217 amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj, 2218 &adev->gfx.rlc.cp_table_gpu_addr, 2219 (void **)&adev->gfx.rlc.cp_table_ptr); 2220 } 2221 gfx_v9_0_free_microcode(adev); 2222 2223 return 0; 2224 } 2225 2226 2227 static void gfx_v9_0_tiling_mode_table_init(struct amdgpu_device *adev) 2228 { 2229 /* TODO */ 2230 } 2231 2232 void gfx_v9_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, u32 sh_num, 2233 u32 instance) 2234 { 2235 u32 data; 2236 2237 if (instance == 0xffffffff) 2238 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1); 2239 else 2240 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance); 2241 2242 if (se_num == 0xffffffff) 2243 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1); 2244 else 2245 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num); 2246 2247 if (sh_num == 0xffffffff) 2248 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1); 2249 else 2250 data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num); 2251 2252 WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_INDEX, data); 2253 } 2254 2255 static u32 gfx_v9_0_get_rb_active_bitmap(struct amdgpu_device *adev) 2256 { 2257 u32 data, mask; 2258 2259 data = RREG32_SOC15(GC, 0, mmCC_RB_BACKEND_DISABLE); 2260 data |= RREG32_SOC15(GC, 0, mmGC_USER_RB_BACKEND_DISABLE); 2261 2262 data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK; 2263 data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT; 2264 2265 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se / 2266 adev->gfx.config.max_sh_per_se); 2267 2268 return (~data) & mask; 2269 } 2270 2271 static void gfx_v9_0_setup_rb(struct amdgpu_device *adev) 2272 { 2273 int i, j; 2274 u32 data; 2275 u32 active_rbs = 0; 2276 u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se / 2277 adev->gfx.config.max_sh_per_se; 2278 2279 mutex_lock(&adev->grbm_idx_mutex); 2280 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 2281 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 2282 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff); 2283 data = gfx_v9_0_get_rb_active_bitmap(adev); 2284 active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) * 2285 rb_bitmap_width_per_sh); 2286 } 2287 } 2288 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 2289 mutex_unlock(&adev->grbm_idx_mutex); 2290 2291 adev->gfx.config.backend_enable_mask = active_rbs; 2292 adev->gfx.config.num_rbs = hweight32(active_rbs); 2293 } 2294 2295 #define DEFAULT_SH_MEM_BASES (0x6000) 2296 static void gfx_v9_0_init_compute_vmid(struct amdgpu_device *adev) 2297 { 2298 int i; 2299 uint32_t sh_mem_config; 2300 uint32_t sh_mem_bases; 2301 2302 /* 2303 * Configure apertures: 2304 * LDS: 0x60000000'00000000 - 0x60000001'00000000 (4GB) 2305 * Scratch: 0x60000001'00000000 - 0x60000002'00000000 (4GB) 2306 * GPUVM: 0x60010000'00000000 - 0x60020000'00000000 (1TB) 2307 */ 2308 sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16); 2309 2310 sh_mem_config = SH_MEM_ADDRESS_MODE_64 | 2311 SH_MEM_ALIGNMENT_MODE_UNALIGNED << 2312 SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT; 2313 2314 mutex_lock(&adev->srbm_mutex); 2315 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) { 2316 soc15_grbm_select(adev, 0, 0, 0, i); 2317 /* CP and shaders */ 2318 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, sh_mem_config); 2319 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, sh_mem_bases); 2320 } 2321 soc15_grbm_select(adev, 0, 0, 0, 0); 2322 mutex_unlock(&adev->srbm_mutex); 2323 2324 /* Initialize all compute VMIDs to have no GDS, GWS, or OA 2325 access. These should be enabled by FW for target VMIDs. */ 2326 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) { 2327 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * i, 0); 2328 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * i, 0); 2329 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, i, 0); 2330 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, i, 0); 2331 } 2332 } 2333 2334 static void gfx_v9_0_init_gds_vmid(struct amdgpu_device *adev) 2335 { 2336 int vmid; 2337 2338 /* 2339 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA 2340 * access. Compute VMIDs should be enabled by FW for target VMIDs, 2341 * the driver can enable them for graphics. VMID0 should maintain 2342 * access so that HWS firmware can save/restore entries. 2343 */ 2344 for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) { 2345 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * vmid, 0); 2346 WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * vmid, 0); 2347 WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, vmid, 0); 2348 WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, vmid, 0); 2349 } 2350 } 2351 2352 static void gfx_v9_0_init_sq_config(struct amdgpu_device *adev) 2353 { 2354 uint32_t tmp; 2355 2356 switch (adev->ip_versions[GC_HWIP][0]) { 2357 case IP_VERSION(9, 4, 1): 2358 tmp = RREG32_SOC15(GC, 0, mmSQ_CONFIG); 2359 tmp = REG_SET_FIELD(tmp, SQ_CONFIG, 2360 DISABLE_BARRIER_WAITCNT, 1); 2361 WREG32_SOC15(GC, 0, mmSQ_CONFIG, tmp); 2362 break; 2363 default: 2364 break; 2365 } 2366 } 2367 2368 static void gfx_v9_0_constants_init(struct amdgpu_device *adev) 2369 { 2370 u32 tmp; 2371 int i; 2372 2373 WREG32_FIELD15_RLC(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff); 2374 2375 gfx_v9_0_tiling_mode_table_init(adev); 2376 2377 if (adev->gfx.num_gfx_rings) 2378 gfx_v9_0_setup_rb(adev); 2379 gfx_v9_0_get_cu_info(adev, &adev->gfx.cu_info); 2380 adev->gfx.config.db_debug2 = RREG32_SOC15(GC, 0, mmDB_DEBUG2); 2381 2382 /* XXX SH_MEM regs */ 2383 /* where to put LDS, scratch, GPUVM in FSA64 space */ 2384 mutex_lock(&adev->srbm_mutex); 2385 for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_0].num_ids; i++) { 2386 soc15_grbm_select(adev, 0, 0, 0, i); 2387 /* CP and shaders */ 2388 if (i == 0) { 2389 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE, 2390 SH_MEM_ALIGNMENT_MODE_UNALIGNED); 2391 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE, 2392 !!adev->gmc.noretry); 2393 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp); 2394 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, 0); 2395 } else { 2396 tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE, 2397 SH_MEM_ALIGNMENT_MODE_UNALIGNED); 2398 tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE, 2399 !!adev->gmc.noretry); 2400 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_CONFIG, tmp); 2401 tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE, 2402 (adev->gmc.private_aperture_start >> 48)); 2403 tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE, 2404 (adev->gmc.shared_aperture_start >> 48)); 2405 WREG32_SOC15_RLC(GC, 0, mmSH_MEM_BASES, tmp); 2406 } 2407 } 2408 soc15_grbm_select(adev, 0, 0, 0, 0); 2409 2410 mutex_unlock(&adev->srbm_mutex); 2411 2412 gfx_v9_0_init_compute_vmid(adev); 2413 gfx_v9_0_init_gds_vmid(adev); 2414 gfx_v9_0_init_sq_config(adev); 2415 } 2416 2417 static void gfx_v9_0_wait_for_rlc_serdes(struct amdgpu_device *adev) 2418 { 2419 u32 i, j, k; 2420 u32 mask; 2421 2422 mutex_lock(&adev->grbm_idx_mutex); 2423 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 2424 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 2425 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff); 2426 for (k = 0; k < adev->usec_timeout; k++) { 2427 if (RREG32_SOC15(GC, 0, mmRLC_SERDES_CU_MASTER_BUSY) == 0) 2428 break; 2429 udelay(1); 2430 } 2431 if (k == adev->usec_timeout) { 2432 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 2433 0xffffffff, 0xffffffff); 2434 mutex_unlock(&adev->grbm_idx_mutex); 2435 DRM_INFO("Timeout wait for RLC serdes %u,%u\n", 2436 i, j); 2437 return; 2438 } 2439 } 2440 } 2441 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 2442 mutex_unlock(&adev->grbm_idx_mutex); 2443 2444 mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK | 2445 RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK | 2446 RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK | 2447 RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK; 2448 for (k = 0; k < adev->usec_timeout; k++) { 2449 if ((RREG32_SOC15(GC, 0, mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0) 2450 break; 2451 udelay(1); 2452 } 2453 } 2454 2455 static void gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device *adev, 2456 bool enable) 2457 { 2458 u32 tmp; 2459 2460 /* These interrupts should be enabled to drive DS clock */ 2461 2462 tmp= RREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0); 2463 2464 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0); 2465 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0); 2466 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0); 2467 if(adev->gfx.num_gfx_rings) 2468 tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0); 2469 2470 WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp); 2471 } 2472 2473 static void gfx_v9_0_init_csb(struct amdgpu_device *adev) 2474 { 2475 adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr); 2476 /* csib */ 2477 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_HI), 2478 adev->gfx.rlc.clear_state_gpu_addr >> 32); 2479 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_ADDR_LO), 2480 adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc); 2481 WREG32_RLC(SOC15_REG_OFFSET(GC, 0, mmRLC_CSIB_LENGTH), 2482 adev->gfx.rlc.clear_state_size); 2483 } 2484 2485 static void gfx_v9_1_parse_ind_reg_list(int *register_list_format, 2486 int indirect_offset, 2487 int list_size, 2488 int *unique_indirect_regs, 2489 int unique_indirect_reg_count, 2490 int *indirect_start_offsets, 2491 int *indirect_start_offsets_count, 2492 int max_start_offsets_count) 2493 { 2494 int idx; 2495 2496 for (; indirect_offset < list_size; indirect_offset++) { 2497 WARN_ON(*indirect_start_offsets_count >= max_start_offsets_count); 2498 indirect_start_offsets[*indirect_start_offsets_count] = indirect_offset; 2499 *indirect_start_offsets_count = *indirect_start_offsets_count + 1; 2500 2501 while (register_list_format[indirect_offset] != 0xFFFFFFFF) { 2502 indirect_offset += 2; 2503 2504 /* look for the matching indice */ 2505 for (idx = 0; idx < unique_indirect_reg_count; idx++) { 2506 if (unique_indirect_regs[idx] == 2507 register_list_format[indirect_offset] || 2508 !unique_indirect_regs[idx]) 2509 break; 2510 } 2511 2512 BUG_ON(idx >= unique_indirect_reg_count); 2513 2514 if (!unique_indirect_regs[idx]) 2515 unique_indirect_regs[idx] = register_list_format[indirect_offset]; 2516 2517 indirect_offset++; 2518 } 2519 } 2520 } 2521 2522 static int gfx_v9_1_init_rlc_save_restore_list(struct amdgpu_device *adev) 2523 { 2524 int unique_indirect_regs[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}; 2525 int unique_indirect_reg_count = 0; 2526 2527 int indirect_start_offsets[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0}; 2528 int indirect_start_offsets_count = 0; 2529 2530 int list_size = 0; 2531 int i = 0, j = 0; 2532 u32 tmp = 0; 2533 2534 u32 *register_list_format = 2535 kmemdup(adev->gfx.rlc.register_list_format, 2536 adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL); 2537 if (!register_list_format) 2538 return -ENOMEM; 2539 2540 /* setup unique_indirect_regs array and indirect_start_offsets array */ 2541 unique_indirect_reg_count = ARRAY_SIZE(unique_indirect_regs); 2542 gfx_v9_1_parse_ind_reg_list(register_list_format, 2543 adev->gfx.rlc.reg_list_format_direct_reg_list_length, 2544 adev->gfx.rlc.reg_list_format_size_bytes >> 2, 2545 unique_indirect_regs, 2546 unique_indirect_reg_count, 2547 indirect_start_offsets, 2548 &indirect_start_offsets_count, 2549 ARRAY_SIZE(indirect_start_offsets)); 2550 2551 /* enable auto inc in case it is disabled */ 2552 tmp = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL)); 2553 tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK; 2554 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL), tmp); 2555 2556 /* write register_restore table to offset 0x0 using RLC_SRM_ARAM_ADDR/DATA */ 2557 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_ADDR), 2558 RLC_SAVE_RESTORE_ADDR_STARTING_OFFSET); 2559 for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++) 2560 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_ARAM_DATA), 2561 adev->gfx.rlc.register_restore[i]); 2562 2563 /* load indirect register */ 2564 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR), 2565 adev->gfx.rlc.reg_list_format_start); 2566 2567 /* direct register portion */ 2568 for (i = 0; i < adev->gfx.rlc.reg_list_format_direct_reg_list_length; i++) 2569 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), 2570 register_list_format[i]); 2571 2572 /* indirect register portion */ 2573 while (i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2)) { 2574 if (register_list_format[i] == 0xFFFFFFFF) { 2575 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]); 2576 continue; 2577 } 2578 2579 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]); 2580 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, register_list_format[i++]); 2581 2582 for (j = 0; j < unique_indirect_reg_count; j++) { 2583 if (register_list_format[i] == unique_indirect_regs[j]) { 2584 WREG32_SOC15(GC, 0, mmRLC_GPM_SCRATCH_DATA, j); 2585 break; 2586 } 2587 } 2588 2589 BUG_ON(j >= unique_indirect_reg_count); 2590 2591 i++; 2592 } 2593 2594 /* set save/restore list size */ 2595 list_size = adev->gfx.rlc.reg_list_size_bytes >> 2; 2596 list_size = list_size >> 1; 2597 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR), 2598 adev->gfx.rlc.reg_restore_list_size); 2599 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), list_size); 2600 2601 /* write the starting offsets to RLC scratch ram */ 2602 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_ADDR), 2603 adev->gfx.rlc.starting_offsets_start); 2604 for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++) 2605 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_GPM_SCRATCH_DATA), 2606 indirect_start_offsets[i]); 2607 2608 /* load unique indirect regs*/ 2609 for (i = 0; i < ARRAY_SIZE(unique_indirect_regs); i++) { 2610 if (unique_indirect_regs[i] != 0) { 2611 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_ADDR_0) 2612 + GFX_RLC_SRM_INDEX_CNTL_ADDR_OFFSETS[i], 2613 unique_indirect_regs[i] & 0x3FFFF); 2614 2615 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_INDEX_CNTL_DATA_0) 2616 + GFX_RLC_SRM_INDEX_CNTL_DATA_OFFSETS[i], 2617 unique_indirect_regs[i] >> 20); 2618 } 2619 } 2620 2621 kfree(register_list_format); 2622 return 0; 2623 } 2624 2625 static void gfx_v9_0_enable_save_restore_machine(struct amdgpu_device *adev) 2626 { 2627 WREG32_FIELD15(GC, 0, RLC_SRM_CNTL, SRM_ENABLE, 1); 2628 } 2629 2630 static void pwr_10_0_gfxip_control_over_cgpg(struct amdgpu_device *adev, 2631 bool enable) 2632 { 2633 uint32_t data = 0; 2634 uint32_t default_data = 0; 2635 2636 default_data = data = RREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS)); 2637 if (enable) { 2638 /* enable GFXIP control over CGPG */ 2639 data |= PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK; 2640 if(default_data != data) 2641 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data); 2642 2643 /* update status */ 2644 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS_MASK; 2645 data |= (2 << PWR_MISC_CNTL_STATUS__PWR_GFXOFF_STATUS__SHIFT); 2646 if(default_data != data) 2647 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data); 2648 } else { 2649 /* restore GFXIP control over GCPG */ 2650 data &= ~PWR_MISC_CNTL_STATUS__PWR_GFX_RLC_CGPG_EN_MASK; 2651 if(default_data != data) 2652 WREG32(SOC15_REG_OFFSET(PWR, 0, mmPWR_MISC_CNTL_STATUS), data); 2653 } 2654 } 2655 2656 static void gfx_v9_0_init_gfx_power_gating(struct amdgpu_device *adev) 2657 { 2658 uint32_t data = 0; 2659 2660 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | 2661 AMD_PG_SUPPORT_GFX_SMG | 2662 AMD_PG_SUPPORT_GFX_DMG)) { 2663 /* init IDLE_POLL_COUNT = 60 */ 2664 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL)); 2665 data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK; 2666 data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT); 2667 WREG32(SOC15_REG_OFFSET(GC, 0, mmCP_RB_WPTR_POLL_CNTL), data); 2668 2669 /* init RLC PG Delay */ 2670 data = 0; 2671 data |= (0x10 << RLC_PG_DELAY__POWER_UP_DELAY__SHIFT); 2672 data |= (0x10 << RLC_PG_DELAY__POWER_DOWN_DELAY__SHIFT); 2673 data |= (0x10 << RLC_PG_DELAY__CMD_PROPAGATE_DELAY__SHIFT); 2674 data |= (0x40 << RLC_PG_DELAY__MEM_SLEEP_DELAY__SHIFT); 2675 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY), data); 2676 2677 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2)); 2678 data &= ~RLC_PG_DELAY_2__SERDES_CMD_DELAY_MASK; 2679 data |= (0x4 << RLC_PG_DELAY_2__SERDES_CMD_DELAY__SHIFT); 2680 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_2), data); 2681 2682 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3)); 2683 data &= ~RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG_MASK; 2684 data |= (0xff << RLC_PG_DELAY_3__CGCG_ACTIVE_BEFORE_CGPG__SHIFT); 2685 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_DELAY_3), data); 2686 2687 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL)); 2688 data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK; 2689 2690 /* program GRBM_REG_SAVE_GFX_IDLE_THRESHOLD to 0x55f0 */ 2691 data |= (0x55f0 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT); 2692 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_AUTO_PG_CTRL), data); 2693 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 3, 0)) 2694 pwr_10_0_gfxip_control_over_cgpg(adev, true); 2695 } 2696 } 2697 2698 static void gfx_v9_0_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev, 2699 bool enable) 2700 { 2701 uint32_t data = 0; 2702 uint32_t default_data = 0; 2703 2704 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL)); 2705 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2706 SMU_CLK_SLOWDOWN_ON_PU_ENABLE, 2707 enable ? 1 : 0); 2708 if (default_data != data) 2709 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2710 } 2711 2712 static void gfx_v9_0_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev, 2713 bool enable) 2714 { 2715 uint32_t data = 0; 2716 uint32_t default_data = 0; 2717 2718 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL)); 2719 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2720 SMU_CLK_SLOWDOWN_ON_PD_ENABLE, 2721 enable ? 1 : 0); 2722 if(default_data != data) 2723 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2724 } 2725 2726 static void gfx_v9_0_enable_cp_power_gating(struct amdgpu_device *adev, 2727 bool enable) 2728 { 2729 uint32_t data = 0; 2730 uint32_t default_data = 0; 2731 2732 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL)); 2733 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2734 CP_PG_DISABLE, 2735 enable ? 0 : 1); 2736 if(default_data != data) 2737 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2738 } 2739 2740 static void gfx_v9_0_enable_gfx_cg_power_gating(struct amdgpu_device *adev, 2741 bool enable) 2742 { 2743 uint32_t data, default_data; 2744 2745 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL)); 2746 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2747 GFX_POWER_GATING_ENABLE, 2748 enable ? 1 : 0); 2749 if(default_data != data) 2750 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2751 } 2752 2753 static void gfx_v9_0_enable_gfx_pipeline_powergating(struct amdgpu_device *adev, 2754 bool enable) 2755 { 2756 uint32_t data, default_data; 2757 2758 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL)); 2759 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2760 GFX_PIPELINE_PG_ENABLE, 2761 enable ? 1 : 0); 2762 if(default_data != data) 2763 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2764 2765 if (!enable) 2766 /* read any GFX register to wake up GFX */ 2767 data = RREG32(SOC15_REG_OFFSET(GC, 0, mmDB_RENDER_CONTROL)); 2768 } 2769 2770 static void gfx_v9_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev, 2771 bool enable) 2772 { 2773 uint32_t data, default_data; 2774 2775 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL)); 2776 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2777 STATIC_PER_CU_PG_ENABLE, 2778 enable ? 1 : 0); 2779 if(default_data != data) 2780 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2781 } 2782 2783 static void gfx_v9_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev, 2784 bool enable) 2785 { 2786 uint32_t data, default_data; 2787 2788 default_data = data = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL)); 2789 data = REG_SET_FIELD(data, RLC_PG_CNTL, 2790 DYN_PER_CU_PG_ENABLE, 2791 enable ? 1 : 0); 2792 if(default_data != data) 2793 WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_PG_CNTL), data); 2794 } 2795 2796 static void gfx_v9_0_init_pg(struct amdgpu_device *adev) 2797 { 2798 gfx_v9_0_init_csb(adev); 2799 2800 /* 2801 * Rlc save restore list is workable since v2_1. 2802 * And it's needed by gfxoff feature. 2803 */ 2804 if (adev->gfx.rlc.is_rlc_v2_1) { 2805 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 2, 1) || 2806 (adev->apu_flags & AMD_APU_IS_RAVEN2)) 2807 gfx_v9_1_init_rlc_save_restore_list(adev); 2808 gfx_v9_0_enable_save_restore_machine(adev); 2809 } 2810 2811 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | 2812 AMD_PG_SUPPORT_GFX_SMG | 2813 AMD_PG_SUPPORT_GFX_DMG | 2814 AMD_PG_SUPPORT_CP | 2815 AMD_PG_SUPPORT_GDS | 2816 AMD_PG_SUPPORT_RLC_SMU_HS)) { 2817 WREG32_SOC15(GC, 0, mmRLC_JUMP_TABLE_RESTORE, 2818 adev->gfx.rlc.cp_table_gpu_addr >> 8); 2819 gfx_v9_0_init_gfx_power_gating(adev); 2820 } 2821 } 2822 2823 static void gfx_v9_0_rlc_stop(struct amdgpu_device *adev) 2824 { 2825 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 0); 2826 gfx_v9_0_enable_gui_idle_interrupt(adev, false); 2827 gfx_v9_0_wait_for_rlc_serdes(adev); 2828 } 2829 2830 static void gfx_v9_0_rlc_reset(struct amdgpu_device *adev) 2831 { 2832 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1); 2833 udelay(50); 2834 WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0); 2835 udelay(50); 2836 } 2837 2838 static void gfx_v9_0_rlc_start(struct amdgpu_device *adev) 2839 { 2840 #ifdef AMDGPU_RLC_DEBUG_RETRY 2841 u32 rlc_ucode_ver; 2842 #endif 2843 2844 WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1); 2845 udelay(50); 2846 2847 /* carrizo do enable cp interrupt after cp inited */ 2848 if (!(adev->flags & AMD_IS_APU)) { 2849 gfx_v9_0_enable_gui_idle_interrupt(adev, true); 2850 udelay(50); 2851 } 2852 2853 #ifdef AMDGPU_RLC_DEBUG_RETRY 2854 /* RLC_GPM_GENERAL_6 : RLC Ucode version */ 2855 rlc_ucode_ver = RREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_6); 2856 if(rlc_ucode_ver == 0x108) { 2857 DRM_INFO("Using rlc debug ucode. mmRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n", 2858 rlc_ucode_ver, adev->gfx.rlc_fw_version); 2859 /* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles, 2860 * default is 0x9C4 to create a 100us interval */ 2861 WREG32_SOC15(GC, 0, mmRLC_GPM_TIMER_INT_3, 0x9C4); 2862 /* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr 2863 * to disable the page fault retry interrupts, default is 2864 * 0x100 (256) */ 2865 WREG32_SOC15(GC, 0, mmRLC_GPM_GENERAL_12, 0x100); 2866 } 2867 #endif 2868 } 2869 2870 static int gfx_v9_0_rlc_load_microcode(struct amdgpu_device *adev) 2871 { 2872 const struct rlc_firmware_header_v2_0 *hdr; 2873 const __le32 *fw_data; 2874 unsigned i, fw_size; 2875 2876 if (!adev->gfx.rlc_fw) 2877 return -EINVAL; 2878 2879 hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data; 2880 amdgpu_ucode_print_rlc_hdr(&hdr->header); 2881 2882 fw_data = (const __le32 *)(adev->gfx.rlc_fw->data + 2883 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2884 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; 2885 2886 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, 2887 RLCG_UCODE_LOADING_START_ADDRESS); 2888 for (i = 0; i < fw_size; i++) 2889 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++)); 2890 WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version); 2891 2892 return 0; 2893 } 2894 2895 static int gfx_v9_0_rlc_resume(struct amdgpu_device *adev) 2896 { 2897 int r; 2898 2899 if (amdgpu_sriov_vf(adev)) { 2900 gfx_v9_0_init_csb(adev); 2901 return 0; 2902 } 2903 2904 adev->gfx.rlc.funcs->stop(adev); 2905 2906 /* disable CG */ 2907 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, 0); 2908 2909 gfx_v9_0_init_pg(adev); 2910 2911 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { 2912 /* legacy rlc firmware loading */ 2913 r = gfx_v9_0_rlc_load_microcode(adev); 2914 if (r) 2915 return r; 2916 } 2917 2918 switch (adev->ip_versions[GC_HWIP][0]) { 2919 case IP_VERSION(9, 2, 2): 2920 case IP_VERSION(9, 1, 0): 2921 if (amdgpu_lbpw == 0) 2922 gfx_v9_0_enable_lbpw(adev, false); 2923 else 2924 gfx_v9_0_enable_lbpw(adev, true); 2925 break; 2926 case IP_VERSION(9, 4, 0): 2927 if (amdgpu_lbpw > 0) 2928 gfx_v9_0_enable_lbpw(adev, true); 2929 else 2930 gfx_v9_0_enable_lbpw(adev, false); 2931 break; 2932 default: 2933 break; 2934 } 2935 2936 adev->gfx.rlc.funcs->start(adev); 2937 2938 return 0; 2939 } 2940 2941 static void gfx_v9_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable) 2942 { 2943 u32 tmp = RREG32_SOC15(GC, 0, mmCP_ME_CNTL); 2944 2945 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1); 2946 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1); 2947 tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, enable ? 0 : 1); 2948 WREG32_SOC15_RLC(GC, 0, mmCP_ME_CNTL, tmp); 2949 udelay(50); 2950 } 2951 2952 static int gfx_v9_0_cp_gfx_load_microcode(struct amdgpu_device *adev) 2953 { 2954 const struct gfx_firmware_header_v1_0 *pfp_hdr; 2955 const struct gfx_firmware_header_v1_0 *ce_hdr; 2956 const struct gfx_firmware_header_v1_0 *me_hdr; 2957 const __le32 *fw_data; 2958 unsigned i, fw_size; 2959 2960 if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw) 2961 return -EINVAL; 2962 2963 pfp_hdr = (const struct gfx_firmware_header_v1_0 *) 2964 adev->gfx.pfp_fw->data; 2965 ce_hdr = (const struct gfx_firmware_header_v1_0 *) 2966 adev->gfx.ce_fw->data; 2967 me_hdr = (const struct gfx_firmware_header_v1_0 *) 2968 adev->gfx.me_fw->data; 2969 2970 amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header); 2971 amdgpu_ucode_print_gfx_hdr(&ce_hdr->header); 2972 amdgpu_ucode_print_gfx_hdr(&me_hdr->header); 2973 2974 gfx_v9_0_cp_gfx_enable(adev, false); 2975 2976 /* PFP */ 2977 fw_data = (const __le32 *) 2978 (adev->gfx.pfp_fw->data + 2979 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes)); 2980 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4; 2981 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, 0); 2982 for (i = 0; i < fw_size; i++) 2983 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++)); 2984 WREG32_SOC15(GC, 0, mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version); 2985 2986 /* CE */ 2987 fw_data = (const __le32 *) 2988 (adev->gfx.ce_fw->data + 2989 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes)); 2990 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4; 2991 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, 0); 2992 for (i = 0; i < fw_size; i++) 2993 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++)); 2994 WREG32_SOC15(GC, 0, mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version); 2995 2996 /* ME */ 2997 fw_data = (const __le32 *) 2998 (adev->gfx.me_fw->data + 2999 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes)); 3000 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4; 3001 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, 0); 3002 for (i = 0; i < fw_size; i++) 3003 WREG32_SOC15(GC, 0, mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++)); 3004 WREG32_SOC15(GC, 0, mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version); 3005 3006 return 0; 3007 } 3008 3009 static int gfx_v9_0_cp_gfx_start(struct amdgpu_device *adev) 3010 { 3011 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0]; 3012 const struct cs_section_def *sect = NULL; 3013 const struct cs_extent_def *ext = NULL; 3014 int r, i, tmp; 3015 3016 /* init the CP */ 3017 WREG32_SOC15(GC, 0, mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1); 3018 WREG32_SOC15(GC, 0, mmCP_DEVICE_ID, 1); 3019 3020 gfx_v9_0_cp_gfx_enable(adev, true); 3021 3022 r = amdgpu_ring_alloc(ring, gfx_v9_0_get_csb_size(adev) + 4 + 3); 3023 if (r) { 3024 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r); 3025 return r; 3026 } 3027 3028 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 3029 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 3030 3031 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 3032 amdgpu_ring_write(ring, 0x80000000); 3033 amdgpu_ring_write(ring, 0x80000000); 3034 3035 for (sect = gfx9_cs_data; sect->section != NULL; ++sect) { 3036 for (ext = sect->section; ext->extent != NULL; ++ext) { 3037 if (sect->id == SECT_CONTEXT) { 3038 amdgpu_ring_write(ring, 3039 PACKET3(PACKET3_SET_CONTEXT_REG, 3040 ext->reg_count)); 3041 amdgpu_ring_write(ring, 3042 ext->reg_index - PACKET3_SET_CONTEXT_REG_START); 3043 for (i = 0; i < ext->reg_count; i++) 3044 amdgpu_ring_write(ring, ext->extent[i]); 3045 } 3046 } 3047 } 3048 3049 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 3050 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE); 3051 3052 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0)); 3053 amdgpu_ring_write(ring, 0); 3054 3055 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2)); 3056 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE)); 3057 amdgpu_ring_write(ring, 0x8000); 3058 amdgpu_ring_write(ring, 0x8000); 3059 3060 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG,1)); 3061 tmp = (PACKET3_SET_UCONFIG_REG_INDEX_TYPE | 3062 (SOC15_REG_OFFSET(GC, 0, mmVGT_INDEX_TYPE) - PACKET3_SET_UCONFIG_REG_START)); 3063 amdgpu_ring_write(ring, tmp); 3064 amdgpu_ring_write(ring, 0); 3065 3066 amdgpu_ring_commit(ring); 3067 3068 return 0; 3069 } 3070 3071 static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev) 3072 { 3073 struct amdgpu_ring *ring; 3074 u32 tmp; 3075 u32 rb_bufsz; 3076 u64 rb_addr, rptr_addr, wptr_gpu_addr; 3077 3078 /* Set the write pointer delay */ 3079 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_DELAY, 0); 3080 3081 /* set the RB to use vmid 0 */ 3082 WREG32_SOC15(GC, 0, mmCP_RB_VMID, 0); 3083 3084 /* Set ring buffer size */ 3085 ring = &adev->gfx.gfx_ring[0]; 3086 rb_bufsz = order_base_2(ring->ring_size / 8); 3087 tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz); 3088 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2); 3089 #ifdef __BIG_ENDIAN 3090 tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1); 3091 #endif 3092 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp); 3093 3094 /* Initialize the ring buffer's write pointers */ 3095 ring->wptr = 0; 3096 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr)); 3097 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr)); 3098 3099 /* set the wb address wether it's enabled or not */ 3100 rptr_addr = ring->rptr_gpu_addr; 3101 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr)); 3102 WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK); 3103 3104 wptr_gpu_addr = ring->wptr_gpu_addr; 3105 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr)); 3106 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr)); 3107 3108 mdelay(1); 3109 WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp); 3110 3111 rb_addr = ring->gpu_addr >> 8; 3112 WREG32_SOC15(GC, 0, mmCP_RB0_BASE, rb_addr); 3113 WREG32_SOC15(GC, 0, mmCP_RB0_BASE_HI, upper_32_bits(rb_addr)); 3114 3115 tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL); 3116 if (ring->use_doorbell) { 3117 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, 3118 DOORBELL_OFFSET, ring->doorbell_index); 3119 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, 3120 DOORBELL_EN, 1); 3121 } else { 3122 tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0); 3123 } 3124 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, tmp); 3125 3126 tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER, 3127 DOORBELL_RANGE_LOWER, ring->doorbell_index); 3128 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp); 3129 3130 WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER, 3131 CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK); 3132 3133 3134 /* start the ring */ 3135 gfx_v9_0_cp_gfx_start(adev); 3136 ring->sched.ready = true; 3137 3138 return 0; 3139 } 3140 3141 static void gfx_v9_0_cp_compute_enable(struct amdgpu_device *adev, bool enable) 3142 { 3143 if (enable) { 3144 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL, 0); 3145 } else { 3146 WREG32_SOC15_RLC(GC, 0, mmCP_MEC_CNTL, 3147 (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK)); 3148 adev->gfx.kiq.ring.sched.ready = false; 3149 } 3150 udelay(50); 3151 } 3152 3153 static int gfx_v9_0_cp_compute_load_microcode(struct amdgpu_device *adev) 3154 { 3155 const struct gfx_firmware_header_v1_0 *mec_hdr; 3156 const __le32 *fw_data; 3157 unsigned i; 3158 u32 tmp; 3159 3160 if (!adev->gfx.mec_fw) 3161 return -EINVAL; 3162 3163 gfx_v9_0_cp_compute_enable(adev, false); 3164 3165 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; 3166 amdgpu_ucode_print_gfx_hdr(&mec_hdr->header); 3167 3168 fw_data = (const __le32 *) 3169 (adev->gfx.mec_fw->data + 3170 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes)); 3171 tmp = 0; 3172 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0); 3173 tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0); 3174 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL, tmp); 3175 3176 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO, 3177 adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000); 3178 WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI, 3179 upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr)); 3180 3181 /* MEC1 */ 3182 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR, 3183 mec_hdr->jt_offset); 3184 for (i = 0; i < mec_hdr->jt_size; i++) 3185 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_DATA, 3186 le32_to_cpup(fw_data + mec_hdr->jt_offset + i)); 3187 3188 WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR, 3189 adev->gfx.mec_fw_version); 3190 /* Todo : Loading MEC2 firmware is only necessary if MEC2 should run different microcode than MEC1. */ 3191 3192 return 0; 3193 } 3194 3195 /* KIQ functions */ 3196 static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring) 3197 { 3198 uint32_t tmp; 3199 struct amdgpu_device *adev = ring->adev; 3200 3201 /* tell RLC which is KIQ queue */ 3202 tmp = RREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS); 3203 tmp &= 0xffffff00; 3204 tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue); 3205 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp); 3206 tmp |= 0x80; 3207 WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp); 3208 } 3209 3210 static void gfx_v9_0_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd) 3211 { 3212 struct amdgpu_device *adev = ring->adev; 3213 3214 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) { 3215 if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) { 3216 mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH; 3217 mqd->cp_hqd_queue_priority = 3218 AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM; 3219 } 3220 } 3221 } 3222 3223 static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring) 3224 { 3225 struct amdgpu_device *adev = ring->adev; 3226 struct v9_mqd *mqd = ring->mqd_ptr; 3227 uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr; 3228 uint32_t tmp; 3229 3230 mqd->header = 0xC0310800; 3231 mqd->compute_pipelinestat_enable = 0x00000001; 3232 mqd->compute_static_thread_mgmt_se0 = 0xffffffff; 3233 mqd->compute_static_thread_mgmt_se1 = 0xffffffff; 3234 mqd->compute_static_thread_mgmt_se2 = 0xffffffff; 3235 mqd->compute_static_thread_mgmt_se3 = 0xffffffff; 3236 mqd->compute_static_thread_mgmt_se4 = 0xffffffff; 3237 mqd->compute_static_thread_mgmt_se5 = 0xffffffff; 3238 mqd->compute_static_thread_mgmt_se6 = 0xffffffff; 3239 mqd->compute_static_thread_mgmt_se7 = 0xffffffff; 3240 mqd->compute_misc_reserved = 0x00000003; 3241 3242 mqd->dynamic_cu_mask_addr_lo = 3243 lower_32_bits(ring->mqd_gpu_addr 3244 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask)); 3245 mqd->dynamic_cu_mask_addr_hi = 3246 upper_32_bits(ring->mqd_gpu_addr 3247 + offsetof(struct v9_mqd_allocation, dynamic_cu_mask)); 3248 3249 eop_base_addr = ring->eop_gpu_addr >> 8; 3250 mqd->cp_hqd_eop_base_addr_lo = eop_base_addr; 3251 mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr); 3252 3253 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */ 3254 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL); 3255 tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE, 3256 (order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1)); 3257 3258 mqd->cp_hqd_eop_control = tmp; 3259 3260 /* enable doorbell? */ 3261 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL); 3262 3263 if (ring->use_doorbell) { 3264 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 3265 DOORBELL_OFFSET, ring->doorbell_index); 3266 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 3267 DOORBELL_EN, 1); 3268 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 3269 DOORBELL_SOURCE, 0); 3270 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 3271 DOORBELL_HIT, 0); 3272 } else { 3273 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL, 3274 DOORBELL_EN, 0); 3275 } 3276 3277 mqd->cp_hqd_pq_doorbell_control = tmp; 3278 3279 /* disable the queue if it's active */ 3280 ring->wptr = 0; 3281 mqd->cp_hqd_dequeue_request = 0; 3282 mqd->cp_hqd_pq_rptr = 0; 3283 mqd->cp_hqd_pq_wptr_lo = 0; 3284 mqd->cp_hqd_pq_wptr_hi = 0; 3285 3286 /* set the pointer to the MQD */ 3287 mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc; 3288 mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr); 3289 3290 /* set MQD vmid to 0 */ 3291 tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL); 3292 tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0); 3293 mqd->cp_mqd_control = tmp; 3294 3295 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */ 3296 hqd_gpu_addr = ring->gpu_addr >> 8; 3297 mqd->cp_hqd_pq_base_lo = hqd_gpu_addr; 3298 mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr); 3299 3300 /* set up the HQD, this is similar to CP_RB0_CNTL */ 3301 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL); 3302 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE, 3303 (order_base_2(ring->ring_size / 4) - 1)); 3304 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE, 3305 (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1)); 3306 #ifdef __BIG_ENDIAN 3307 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1); 3308 #endif 3309 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0); 3310 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0); 3311 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1); 3312 tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1); 3313 mqd->cp_hqd_pq_control = tmp; 3314 3315 /* set the wb address whether it's enabled or not */ 3316 wb_gpu_addr = ring->rptr_gpu_addr; 3317 mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc; 3318 mqd->cp_hqd_pq_rptr_report_addr_hi = 3319 upper_32_bits(wb_gpu_addr) & 0xffff; 3320 3321 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */ 3322 wb_gpu_addr = ring->wptr_gpu_addr; 3323 mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc; 3324 mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff; 3325 3326 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */ 3327 ring->wptr = 0; 3328 mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR); 3329 3330 /* set the vmid for the queue */ 3331 mqd->cp_hqd_vmid = 0; 3332 3333 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE); 3334 tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53); 3335 mqd->cp_hqd_persistent_state = tmp; 3336 3337 /* set MIN_IB_AVAIL_SIZE */ 3338 tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL); 3339 tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3); 3340 mqd->cp_hqd_ib_control = tmp; 3341 3342 /* set static priority for a queue/ring */ 3343 gfx_v9_0_mqd_set_priority(ring, mqd); 3344 mqd->cp_hqd_quantum = RREG32_SOC15(GC, 0, mmCP_HQD_QUANTUM); 3345 3346 /* map_queues packet doesn't need activate the queue, 3347 * so only kiq need set this field. 3348 */ 3349 if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ) 3350 mqd->cp_hqd_active = 1; 3351 3352 return 0; 3353 } 3354 3355 static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring) 3356 { 3357 struct amdgpu_device *adev = ring->adev; 3358 struct v9_mqd *mqd = ring->mqd_ptr; 3359 int j; 3360 3361 /* disable wptr polling */ 3362 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0); 3363 3364 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR, 3365 mqd->cp_hqd_eop_base_addr_lo); 3366 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI, 3367 mqd->cp_hqd_eop_base_addr_hi); 3368 3369 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */ 3370 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_EOP_CONTROL, 3371 mqd->cp_hqd_eop_control); 3372 3373 /* enable doorbell? */ 3374 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 3375 mqd->cp_hqd_pq_doorbell_control); 3376 3377 /* disable the queue if it's active */ 3378 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) { 3379 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1); 3380 for (j = 0; j < adev->usec_timeout; j++) { 3381 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1)) 3382 break; 3383 udelay(1); 3384 } 3385 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 3386 mqd->cp_hqd_dequeue_request); 3387 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR, 3388 mqd->cp_hqd_pq_rptr); 3389 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 3390 mqd->cp_hqd_pq_wptr_lo); 3391 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 3392 mqd->cp_hqd_pq_wptr_hi); 3393 } 3394 3395 /* set the pointer to the MQD */ 3396 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR, 3397 mqd->cp_mqd_base_addr_lo); 3398 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_BASE_ADDR_HI, 3399 mqd->cp_mqd_base_addr_hi); 3400 3401 /* set MQD vmid to 0 */ 3402 WREG32_SOC15_RLC(GC, 0, mmCP_MQD_CONTROL, 3403 mqd->cp_mqd_control); 3404 3405 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */ 3406 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE, 3407 mqd->cp_hqd_pq_base_lo); 3408 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_BASE_HI, 3409 mqd->cp_hqd_pq_base_hi); 3410 3411 /* set up the HQD, this is similar to CP_RB0_CNTL */ 3412 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_CONTROL, 3413 mqd->cp_hqd_pq_control); 3414 3415 /* set the wb address whether it's enabled or not */ 3416 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR, 3417 mqd->cp_hqd_pq_rptr_report_addr_lo); 3418 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI, 3419 mqd->cp_hqd_pq_rptr_report_addr_hi); 3420 3421 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */ 3422 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR, 3423 mqd->cp_hqd_pq_wptr_poll_addr_lo); 3424 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI, 3425 mqd->cp_hqd_pq_wptr_poll_addr_hi); 3426 3427 /* enable the doorbell if requested */ 3428 if (ring->use_doorbell) { 3429 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER, 3430 (adev->doorbell_index.kiq * 2) << 2); 3431 /* If GC has entered CGPG, ringing doorbell > first page 3432 * doesn't wakeup GC. Enlarge CP_MEC_DOORBELL_RANGE_UPPER to 3433 * workaround this issue. And this change has to align with firmware 3434 * update. 3435 */ 3436 if (check_if_enlarge_doorbell_range(adev)) 3437 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER, 3438 (adev->doorbell.size - 4)); 3439 else 3440 WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER, 3441 (adev->doorbell_index.userqueue_end * 2) << 2); 3442 } 3443 3444 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 3445 mqd->cp_hqd_pq_doorbell_control); 3446 3447 /* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */ 3448 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 3449 mqd->cp_hqd_pq_wptr_lo); 3450 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 3451 mqd->cp_hqd_pq_wptr_hi); 3452 3453 /* set the vmid for the queue */ 3454 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_VMID, mqd->cp_hqd_vmid); 3455 3456 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE, 3457 mqd->cp_hqd_persistent_state); 3458 3459 /* activate the queue */ 3460 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE, 3461 mqd->cp_hqd_active); 3462 3463 if (ring->use_doorbell) 3464 WREG32_FIELD15(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1); 3465 3466 return 0; 3467 } 3468 3469 static int gfx_v9_0_kiq_fini_register(struct amdgpu_ring *ring) 3470 { 3471 struct amdgpu_device *adev = ring->adev; 3472 int j; 3473 3474 /* disable the queue if it's active */ 3475 if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) { 3476 3477 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1); 3478 3479 for (j = 0; j < adev->usec_timeout; j++) { 3480 if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1)) 3481 break; 3482 udelay(1); 3483 } 3484 3485 if (j == AMDGPU_MAX_USEC_TIMEOUT) { 3486 DRM_DEBUG("KIQ dequeue request failed.\n"); 3487 3488 /* Manual disable if dequeue request times out */ 3489 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_ACTIVE, 0); 3490 } 3491 3492 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 3493 0); 3494 } 3495 3496 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IQ_TIMER, 0); 3497 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_IB_CONTROL, 0); 3498 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PERSISTENT_STATE, 0); 3499 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000); 3500 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, 0); 3501 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_RPTR, 0); 3502 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_HI, 0); 3503 WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PQ_WPTR_LO, 0); 3504 3505 return 0; 3506 } 3507 3508 static int gfx_v9_0_kiq_init_queue(struct amdgpu_ring *ring) 3509 { 3510 struct amdgpu_device *adev = ring->adev; 3511 struct v9_mqd *mqd = ring->mqd_ptr; 3512 int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS; 3513 struct v9_mqd *tmp_mqd; 3514 3515 gfx_v9_0_kiq_setting(ring); 3516 3517 /* GPU could be in bad state during probe, driver trigger the reset 3518 * after load the SMU, in this case , the mqd is not be initialized. 3519 * driver need to re-init the mqd. 3520 * check mqd->cp_hqd_pq_control since this value should not be 0 3521 */ 3522 tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx]; 3523 if (amdgpu_in_reset(adev) && tmp_mqd->cp_hqd_pq_control){ 3524 /* for GPU_RESET case , reset MQD to a clean status */ 3525 if (adev->gfx.mec.mqd_backup[mqd_idx]) 3526 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation)); 3527 3528 /* reset ring buffer */ 3529 ring->wptr = 0; 3530 amdgpu_ring_clear_ring(ring); 3531 3532 mutex_lock(&adev->srbm_mutex); 3533 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); 3534 gfx_v9_0_kiq_init_register(ring); 3535 soc15_grbm_select(adev, 0, 0, 0, 0); 3536 mutex_unlock(&adev->srbm_mutex); 3537 } else { 3538 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation)); 3539 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF; 3540 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF; 3541 if (amdgpu_sriov_vf(adev) && adev->in_suspend) 3542 amdgpu_ring_clear_ring(ring); 3543 mutex_lock(&adev->srbm_mutex); 3544 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); 3545 gfx_v9_0_mqd_init(ring); 3546 gfx_v9_0_kiq_init_register(ring); 3547 soc15_grbm_select(adev, 0, 0, 0, 0); 3548 mutex_unlock(&adev->srbm_mutex); 3549 3550 if (adev->gfx.mec.mqd_backup[mqd_idx]) 3551 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation)); 3552 } 3553 3554 return 0; 3555 } 3556 3557 static int gfx_v9_0_kcq_init_queue(struct amdgpu_ring *ring) 3558 { 3559 struct amdgpu_device *adev = ring->adev; 3560 struct v9_mqd *mqd = ring->mqd_ptr; 3561 int mqd_idx = ring - &adev->gfx.compute_ring[0]; 3562 struct v9_mqd *tmp_mqd; 3563 3564 /* Same as above kiq init, driver need to re-init the mqd if mqd->cp_hqd_pq_control 3565 * is not be initialized before 3566 */ 3567 tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx]; 3568 3569 if (!tmp_mqd->cp_hqd_pq_control || 3570 (!amdgpu_in_reset(adev) && !adev->in_suspend)) { 3571 memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation)); 3572 ((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF; 3573 ((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF; 3574 mutex_lock(&adev->srbm_mutex); 3575 soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); 3576 gfx_v9_0_mqd_init(ring); 3577 soc15_grbm_select(adev, 0, 0, 0, 0); 3578 mutex_unlock(&adev->srbm_mutex); 3579 3580 if (adev->gfx.mec.mqd_backup[mqd_idx]) 3581 memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation)); 3582 } else if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */ 3583 /* reset MQD to a clean status */ 3584 if (adev->gfx.mec.mqd_backup[mqd_idx]) 3585 memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation)); 3586 3587 /* reset ring buffer */ 3588 ring->wptr = 0; 3589 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, 0); 3590 amdgpu_ring_clear_ring(ring); 3591 } else { 3592 amdgpu_ring_clear_ring(ring); 3593 } 3594 3595 return 0; 3596 } 3597 3598 static int gfx_v9_0_kiq_resume(struct amdgpu_device *adev) 3599 { 3600 struct amdgpu_ring *ring; 3601 int r; 3602 3603 ring = &adev->gfx.kiq.ring; 3604 3605 r = amdgpu_bo_reserve(ring->mqd_obj, false); 3606 if (unlikely(r != 0)) 3607 return r; 3608 3609 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); 3610 if (unlikely(r != 0)) { 3611 amdgpu_bo_unreserve(ring->mqd_obj); 3612 return r; 3613 } 3614 3615 gfx_v9_0_kiq_init_queue(ring); 3616 amdgpu_bo_kunmap(ring->mqd_obj); 3617 ring->mqd_ptr = NULL; 3618 amdgpu_bo_unreserve(ring->mqd_obj); 3619 ring->sched.ready = true; 3620 return 0; 3621 } 3622 3623 static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev) 3624 { 3625 struct amdgpu_ring *ring = NULL; 3626 int r = 0, i; 3627 3628 gfx_v9_0_cp_compute_enable(adev, true); 3629 3630 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 3631 ring = &adev->gfx.compute_ring[i]; 3632 3633 r = amdgpu_bo_reserve(ring->mqd_obj, false); 3634 if (unlikely(r != 0)) 3635 goto done; 3636 r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr); 3637 if (!r) { 3638 r = gfx_v9_0_kcq_init_queue(ring); 3639 amdgpu_bo_kunmap(ring->mqd_obj); 3640 ring->mqd_ptr = NULL; 3641 } 3642 amdgpu_bo_unreserve(ring->mqd_obj); 3643 if (r) 3644 goto done; 3645 } 3646 3647 r = amdgpu_gfx_enable_kcq(adev); 3648 done: 3649 return r; 3650 } 3651 3652 static int gfx_v9_0_cp_resume(struct amdgpu_device *adev) 3653 { 3654 int r, i; 3655 struct amdgpu_ring *ring; 3656 3657 if (!(adev->flags & AMD_IS_APU)) 3658 gfx_v9_0_enable_gui_idle_interrupt(adev, false); 3659 3660 if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) { 3661 if (adev->gfx.num_gfx_rings) { 3662 /* legacy firmware loading */ 3663 r = gfx_v9_0_cp_gfx_load_microcode(adev); 3664 if (r) 3665 return r; 3666 } 3667 3668 r = gfx_v9_0_cp_compute_load_microcode(adev); 3669 if (r) 3670 return r; 3671 } 3672 3673 r = gfx_v9_0_kiq_resume(adev); 3674 if (r) 3675 return r; 3676 3677 if (adev->gfx.num_gfx_rings) { 3678 r = gfx_v9_0_cp_gfx_resume(adev); 3679 if (r) 3680 return r; 3681 } 3682 3683 r = gfx_v9_0_kcq_resume(adev); 3684 if (r) 3685 return r; 3686 3687 if (adev->gfx.num_gfx_rings) { 3688 ring = &adev->gfx.gfx_ring[0]; 3689 r = amdgpu_ring_test_helper(ring); 3690 if (r) 3691 return r; 3692 } 3693 3694 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 3695 ring = &adev->gfx.compute_ring[i]; 3696 amdgpu_ring_test_helper(ring); 3697 } 3698 3699 gfx_v9_0_enable_gui_idle_interrupt(adev, true); 3700 3701 return 0; 3702 } 3703 3704 static void gfx_v9_0_init_tcp_config(struct amdgpu_device *adev) 3705 { 3706 u32 tmp; 3707 3708 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1) && 3709 adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 2)) 3710 return; 3711 3712 tmp = RREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG); 3713 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE64KHASH, 3714 adev->df.hash_status.hash_64k); 3715 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE2MHASH, 3716 adev->df.hash_status.hash_2m); 3717 tmp = REG_SET_FIELD(tmp, TCP_ADDR_CONFIG, ENABLE1GHASH, 3718 adev->df.hash_status.hash_1g); 3719 WREG32_SOC15(GC, 0, mmTCP_ADDR_CONFIG, tmp); 3720 } 3721 3722 static void gfx_v9_0_cp_enable(struct amdgpu_device *adev, bool enable) 3723 { 3724 if (adev->gfx.num_gfx_rings) 3725 gfx_v9_0_cp_gfx_enable(adev, enable); 3726 gfx_v9_0_cp_compute_enable(adev, enable); 3727 } 3728 3729 static int gfx_v9_0_hw_init(void *handle) 3730 { 3731 int r; 3732 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3733 3734 if (!amdgpu_sriov_vf(adev)) 3735 gfx_v9_0_init_golden_registers(adev); 3736 3737 gfx_v9_0_constants_init(adev); 3738 3739 gfx_v9_0_init_tcp_config(adev); 3740 3741 r = adev->gfx.rlc.funcs->resume(adev); 3742 if (r) 3743 return r; 3744 3745 r = gfx_v9_0_cp_resume(adev); 3746 if (r) 3747 return r; 3748 3749 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2)) 3750 gfx_v9_4_2_set_power_brake_sequence(adev); 3751 3752 return r; 3753 } 3754 3755 static int gfx_v9_0_hw_fini(void *handle) 3756 { 3757 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3758 3759 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) 3760 amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0); 3761 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); 3762 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); 3763 3764 /* DF freeze and kcq disable will fail */ 3765 if (!amdgpu_ras_intr_triggered()) 3766 /* disable KCQ to avoid CPC touch memory not valid anymore */ 3767 amdgpu_gfx_disable_kcq(adev); 3768 3769 if (amdgpu_sriov_vf(adev)) { 3770 gfx_v9_0_cp_gfx_enable(adev, false); 3771 /* must disable polling for SRIOV when hw finished, otherwise 3772 * CPC engine may still keep fetching WB address which is already 3773 * invalid after sw finished and trigger DMAR reading error in 3774 * hypervisor side. 3775 */ 3776 WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0); 3777 return 0; 3778 } 3779 3780 /* Use deinitialize sequence from CAIL when unbinding device from driver, 3781 * otherwise KIQ is hanging when binding back 3782 */ 3783 if (!amdgpu_in_reset(adev) && !adev->in_suspend) { 3784 mutex_lock(&adev->srbm_mutex); 3785 soc15_grbm_select(adev, adev->gfx.kiq.ring.me, 3786 adev->gfx.kiq.ring.pipe, 3787 adev->gfx.kiq.ring.queue, 0); 3788 gfx_v9_0_kiq_fini_register(&adev->gfx.kiq.ring); 3789 soc15_grbm_select(adev, 0, 0, 0, 0); 3790 mutex_unlock(&adev->srbm_mutex); 3791 } 3792 3793 gfx_v9_0_cp_enable(adev, false); 3794 3795 /* Skip stopping RLC with A+A reset or when RLC controls GFX clock */ 3796 if ((adev->gmc.xgmi.connected_to_cpu && amdgpu_in_reset(adev)) || 3797 (adev->ip_versions[GC_HWIP][0] >= IP_VERSION(9, 4, 2))) { 3798 dev_dbg(adev->dev, "Skipping RLC halt\n"); 3799 return 0; 3800 } 3801 3802 adev->gfx.rlc.funcs->stop(adev); 3803 return 0; 3804 } 3805 3806 static int gfx_v9_0_suspend(void *handle) 3807 { 3808 return gfx_v9_0_hw_fini(handle); 3809 } 3810 3811 static int gfx_v9_0_resume(void *handle) 3812 { 3813 return gfx_v9_0_hw_init(handle); 3814 } 3815 3816 static bool gfx_v9_0_is_idle(void *handle) 3817 { 3818 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3819 3820 if (REG_GET_FIELD(RREG32_SOC15(GC, 0, mmGRBM_STATUS), 3821 GRBM_STATUS, GUI_ACTIVE)) 3822 return false; 3823 else 3824 return true; 3825 } 3826 3827 static int gfx_v9_0_wait_for_idle(void *handle) 3828 { 3829 unsigned i; 3830 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3831 3832 for (i = 0; i < adev->usec_timeout; i++) { 3833 if (gfx_v9_0_is_idle(handle)) 3834 return 0; 3835 udelay(1); 3836 } 3837 return -ETIMEDOUT; 3838 } 3839 3840 static int gfx_v9_0_soft_reset(void *handle) 3841 { 3842 u32 grbm_soft_reset = 0; 3843 u32 tmp; 3844 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3845 3846 /* GRBM_STATUS */ 3847 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS); 3848 if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK | 3849 GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK | 3850 GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK | 3851 GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK | 3852 GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK | 3853 GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) { 3854 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 3855 GRBM_SOFT_RESET, SOFT_RESET_CP, 1); 3856 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 3857 GRBM_SOFT_RESET, SOFT_RESET_GFX, 1); 3858 } 3859 3860 if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) { 3861 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 3862 GRBM_SOFT_RESET, SOFT_RESET_CP, 1); 3863 } 3864 3865 /* GRBM_STATUS2 */ 3866 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2); 3867 if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY)) 3868 grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, 3869 GRBM_SOFT_RESET, SOFT_RESET_RLC, 1); 3870 3871 3872 if (grbm_soft_reset) { 3873 /* stop the rlc */ 3874 adev->gfx.rlc.funcs->stop(adev); 3875 3876 if (adev->gfx.num_gfx_rings) 3877 /* Disable GFX parsing/prefetching */ 3878 gfx_v9_0_cp_gfx_enable(adev, false); 3879 3880 /* Disable MEC parsing/prefetching */ 3881 gfx_v9_0_cp_compute_enable(adev, false); 3882 3883 if (grbm_soft_reset) { 3884 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET); 3885 tmp |= grbm_soft_reset; 3886 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp); 3887 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp); 3888 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET); 3889 3890 udelay(50); 3891 3892 tmp &= ~grbm_soft_reset; 3893 WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp); 3894 tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET); 3895 } 3896 3897 /* Wait a little for things to settle down */ 3898 udelay(50); 3899 } 3900 return 0; 3901 } 3902 3903 static uint64_t gfx_v9_0_kiq_read_clock(struct amdgpu_device *adev) 3904 { 3905 signed long r, cnt = 0; 3906 unsigned long flags; 3907 uint32_t seq, reg_val_offs = 0; 3908 uint64_t value = 0; 3909 struct amdgpu_kiq *kiq = &adev->gfx.kiq; 3910 struct amdgpu_ring *ring = &kiq->ring; 3911 3912 BUG_ON(!ring->funcs->emit_rreg); 3913 3914 spin_lock_irqsave(&kiq->ring_lock, flags); 3915 if (amdgpu_device_wb_get(adev, ®_val_offs)) { 3916 pr_err("critical bug! too many kiq readers\n"); 3917 goto failed_unlock; 3918 } 3919 amdgpu_ring_alloc(ring, 32); 3920 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4)); 3921 amdgpu_ring_write(ring, 9 | /* src: register*/ 3922 (5 << 8) | /* dst: memory */ 3923 (1 << 16) | /* count sel */ 3924 (1 << 20)); /* write confirm */ 3925 amdgpu_ring_write(ring, 0); 3926 amdgpu_ring_write(ring, 0); 3927 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr + 3928 reg_val_offs * 4)); 3929 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr + 3930 reg_val_offs * 4)); 3931 r = amdgpu_fence_emit_polling(ring, &seq, MAX_KIQ_REG_WAIT); 3932 if (r) 3933 goto failed_undo; 3934 3935 amdgpu_ring_commit(ring); 3936 spin_unlock_irqrestore(&kiq->ring_lock, flags); 3937 3938 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); 3939 3940 /* don't wait anymore for gpu reset case because this way may 3941 * block gpu_recover() routine forever, e.g. this virt_kiq_rreg 3942 * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will 3943 * never return if we keep waiting in virt_kiq_rreg, which cause 3944 * gpu_recover() hang there. 3945 * 3946 * also don't wait anymore for IRQ context 3947 * */ 3948 if (r < 1 && (amdgpu_in_reset(adev))) 3949 goto failed_kiq_read; 3950 3951 might_sleep(); 3952 while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) { 3953 msleep(MAX_KIQ_REG_BAILOUT_INTERVAL); 3954 r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); 3955 } 3956 3957 if (cnt > MAX_KIQ_REG_TRY) 3958 goto failed_kiq_read; 3959 3960 mb(); 3961 value = (uint64_t)adev->wb.wb[reg_val_offs] | 3962 (uint64_t)adev->wb.wb[reg_val_offs + 1 ] << 32ULL; 3963 amdgpu_device_wb_free(adev, reg_val_offs); 3964 return value; 3965 3966 failed_undo: 3967 amdgpu_ring_undo(ring); 3968 failed_unlock: 3969 spin_unlock_irqrestore(&kiq->ring_lock, flags); 3970 failed_kiq_read: 3971 if (reg_val_offs) 3972 amdgpu_device_wb_free(adev, reg_val_offs); 3973 pr_err("failed to read gpu clock\n"); 3974 return ~0; 3975 } 3976 3977 static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev) 3978 { 3979 uint64_t clock, clock_lo, clock_hi, hi_check; 3980 3981 switch (adev->ip_versions[GC_HWIP][0]) { 3982 case IP_VERSION(9, 3, 0): 3983 preempt_disable(); 3984 clock_hi = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir); 3985 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir); 3986 hi_check = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_UPPER_Renoir); 3987 /* The SMUIO TSC clock frequency is 100MHz, which sets 32-bit carry over 3988 * roughly every 42 seconds. 3989 */ 3990 if (hi_check != clock_hi) { 3991 clock_lo = RREG32_SOC15_NO_KIQ(SMUIO, 0, mmGOLDEN_TSC_COUNT_LOWER_Renoir); 3992 clock_hi = hi_check; 3993 } 3994 preempt_enable(); 3995 clock = clock_lo | (clock_hi << 32ULL); 3996 break; 3997 default: 3998 amdgpu_gfx_off_ctrl(adev, false); 3999 mutex_lock(&adev->gfx.gpu_clock_mutex); 4000 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 0, 1) && amdgpu_sriov_runtime(adev)) { 4001 clock = gfx_v9_0_kiq_read_clock(adev); 4002 } else { 4003 WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1); 4004 clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB) | 4005 ((uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL); 4006 } 4007 mutex_unlock(&adev->gfx.gpu_clock_mutex); 4008 amdgpu_gfx_off_ctrl(adev, true); 4009 break; 4010 } 4011 return clock; 4012 } 4013 4014 static void gfx_v9_0_ring_emit_gds_switch(struct amdgpu_ring *ring, 4015 uint32_t vmid, 4016 uint32_t gds_base, uint32_t gds_size, 4017 uint32_t gws_base, uint32_t gws_size, 4018 uint32_t oa_base, uint32_t oa_size) 4019 { 4020 struct amdgpu_device *adev = ring->adev; 4021 4022 /* GDS Base */ 4023 gfx_v9_0_write_data_to_reg(ring, 0, false, 4024 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_BASE) + 2 * vmid, 4025 gds_base); 4026 4027 /* GDS Size */ 4028 gfx_v9_0_write_data_to_reg(ring, 0, false, 4029 SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE) + 2 * vmid, 4030 gds_size); 4031 4032 /* GWS */ 4033 gfx_v9_0_write_data_to_reg(ring, 0, false, 4034 SOC15_REG_OFFSET(GC, 0, mmGDS_GWS_VMID0) + vmid, 4035 gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base); 4036 4037 /* OA */ 4038 gfx_v9_0_write_data_to_reg(ring, 0, false, 4039 SOC15_REG_OFFSET(GC, 0, mmGDS_OA_VMID0) + vmid, 4040 (1 << (oa_size + oa_base)) - (1 << oa_base)); 4041 } 4042 4043 static const u32 vgpr_init_compute_shader[] = 4044 { 4045 0xb07c0000, 0xbe8000ff, 4046 0x000000f8, 0xbf110800, 4047 0x7e000280, 0x7e020280, 4048 0x7e040280, 0x7e060280, 4049 0x7e080280, 0x7e0a0280, 4050 0x7e0c0280, 0x7e0e0280, 4051 0x80808800, 0xbe803200, 4052 0xbf84fff5, 0xbf9c0000, 4053 0xd28c0001, 0x0001007f, 4054 0xd28d0001, 0x0002027e, 4055 0x10020288, 0xb8810904, 4056 0xb7814000, 0xd1196a01, 4057 0x00000301, 0xbe800087, 4058 0xbefc00c1, 0xd89c4000, 4059 0x00020201, 0xd89cc080, 4060 0x00040401, 0x320202ff, 4061 0x00000800, 0x80808100, 4062 0xbf84fff8, 0x7e020280, 4063 0xbf810000, 0x00000000, 4064 }; 4065 4066 static const u32 sgpr_init_compute_shader[] = 4067 { 4068 0xb07c0000, 0xbe8000ff, 4069 0x0000005f, 0xbee50080, 4070 0xbe812c65, 0xbe822c65, 4071 0xbe832c65, 0xbe842c65, 4072 0xbe852c65, 0xb77c0005, 4073 0x80808500, 0xbf84fff8, 4074 0xbe800080, 0xbf810000, 4075 }; 4076 4077 static const u32 vgpr_init_compute_shader_arcturus[] = { 4078 0xd3d94000, 0x18000080, 0xd3d94001, 0x18000080, 0xd3d94002, 0x18000080, 4079 0xd3d94003, 0x18000080, 0xd3d94004, 0x18000080, 0xd3d94005, 0x18000080, 4080 0xd3d94006, 0x18000080, 0xd3d94007, 0x18000080, 0xd3d94008, 0x18000080, 4081 0xd3d94009, 0x18000080, 0xd3d9400a, 0x18000080, 0xd3d9400b, 0x18000080, 4082 0xd3d9400c, 0x18000080, 0xd3d9400d, 0x18000080, 0xd3d9400e, 0x18000080, 4083 0xd3d9400f, 0x18000080, 0xd3d94010, 0x18000080, 0xd3d94011, 0x18000080, 4084 0xd3d94012, 0x18000080, 0xd3d94013, 0x18000080, 0xd3d94014, 0x18000080, 4085 0xd3d94015, 0x18000080, 0xd3d94016, 0x18000080, 0xd3d94017, 0x18000080, 4086 0xd3d94018, 0x18000080, 0xd3d94019, 0x18000080, 0xd3d9401a, 0x18000080, 4087 0xd3d9401b, 0x18000080, 0xd3d9401c, 0x18000080, 0xd3d9401d, 0x18000080, 4088 0xd3d9401e, 0x18000080, 0xd3d9401f, 0x18000080, 0xd3d94020, 0x18000080, 4089 0xd3d94021, 0x18000080, 0xd3d94022, 0x18000080, 0xd3d94023, 0x18000080, 4090 0xd3d94024, 0x18000080, 0xd3d94025, 0x18000080, 0xd3d94026, 0x18000080, 4091 0xd3d94027, 0x18000080, 0xd3d94028, 0x18000080, 0xd3d94029, 0x18000080, 4092 0xd3d9402a, 0x18000080, 0xd3d9402b, 0x18000080, 0xd3d9402c, 0x18000080, 4093 0xd3d9402d, 0x18000080, 0xd3d9402e, 0x18000080, 0xd3d9402f, 0x18000080, 4094 0xd3d94030, 0x18000080, 0xd3d94031, 0x18000080, 0xd3d94032, 0x18000080, 4095 0xd3d94033, 0x18000080, 0xd3d94034, 0x18000080, 0xd3d94035, 0x18000080, 4096 0xd3d94036, 0x18000080, 0xd3d94037, 0x18000080, 0xd3d94038, 0x18000080, 4097 0xd3d94039, 0x18000080, 0xd3d9403a, 0x18000080, 0xd3d9403b, 0x18000080, 4098 0xd3d9403c, 0x18000080, 0xd3d9403d, 0x18000080, 0xd3d9403e, 0x18000080, 4099 0xd3d9403f, 0x18000080, 0xd3d94040, 0x18000080, 0xd3d94041, 0x18000080, 4100 0xd3d94042, 0x18000080, 0xd3d94043, 0x18000080, 0xd3d94044, 0x18000080, 4101 0xd3d94045, 0x18000080, 0xd3d94046, 0x18000080, 0xd3d94047, 0x18000080, 4102 0xd3d94048, 0x18000080, 0xd3d94049, 0x18000080, 0xd3d9404a, 0x18000080, 4103 0xd3d9404b, 0x18000080, 0xd3d9404c, 0x18000080, 0xd3d9404d, 0x18000080, 4104 0xd3d9404e, 0x18000080, 0xd3d9404f, 0x18000080, 0xd3d94050, 0x18000080, 4105 0xd3d94051, 0x18000080, 0xd3d94052, 0x18000080, 0xd3d94053, 0x18000080, 4106 0xd3d94054, 0x18000080, 0xd3d94055, 0x18000080, 0xd3d94056, 0x18000080, 4107 0xd3d94057, 0x18000080, 0xd3d94058, 0x18000080, 0xd3d94059, 0x18000080, 4108 0xd3d9405a, 0x18000080, 0xd3d9405b, 0x18000080, 0xd3d9405c, 0x18000080, 4109 0xd3d9405d, 0x18000080, 0xd3d9405e, 0x18000080, 0xd3d9405f, 0x18000080, 4110 0xd3d94060, 0x18000080, 0xd3d94061, 0x18000080, 0xd3d94062, 0x18000080, 4111 0xd3d94063, 0x18000080, 0xd3d94064, 0x18000080, 0xd3d94065, 0x18000080, 4112 0xd3d94066, 0x18000080, 0xd3d94067, 0x18000080, 0xd3d94068, 0x18000080, 4113 0xd3d94069, 0x18000080, 0xd3d9406a, 0x18000080, 0xd3d9406b, 0x18000080, 4114 0xd3d9406c, 0x18000080, 0xd3d9406d, 0x18000080, 0xd3d9406e, 0x18000080, 4115 0xd3d9406f, 0x18000080, 0xd3d94070, 0x18000080, 0xd3d94071, 0x18000080, 4116 0xd3d94072, 0x18000080, 0xd3d94073, 0x18000080, 0xd3d94074, 0x18000080, 4117 0xd3d94075, 0x18000080, 0xd3d94076, 0x18000080, 0xd3d94077, 0x18000080, 4118 0xd3d94078, 0x18000080, 0xd3d94079, 0x18000080, 0xd3d9407a, 0x18000080, 4119 0xd3d9407b, 0x18000080, 0xd3d9407c, 0x18000080, 0xd3d9407d, 0x18000080, 4120 0xd3d9407e, 0x18000080, 0xd3d9407f, 0x18000080, 0xd3d94080, 0x18000080, 4121 0xd3d94081, 0x18000080, 0xd3d94082, 0x18000080, 0xd3d94083, 0x18000080, 4122 0xd3d94084, 0x18000080, 0xd3d94085, 0x18000080, 0xd3d94086, 0x18000080, 4123 0xd3d94087, 0x18000080, 0xd3d94088, 0x18000080, 0xd3d94089, 0x18000080, 4124 0xd3d9408a, 0x18000080, 0xd3d9408b, 0x18000080, 0xd3d9408c, 0x18000080, 4125 0xd3d9408d, 0x18000080, 0xd3d9408e, 0x18000080, 0xd3d9408f, 0x18000080, 4126 0xd3d94090, 0x18000080, 0xd3d94091, 0x18000080, 0xd3d94092, 0x18000080, 4127 0xd3d94093, 0x18000080, 0xd3d94094, 0x18000080, 0xd3d94095, 0x18000080, 4128 0xd3d94096, 0x18000080, 0xd3d94097, 0x18000080, 0xd3d94098, 0x18000080, 4129 0xd3d94099, 0x18000080, 0xd3d9409a, 0x18000080, 0xd3d9409b, 0x18000080, 4130 0xd3d9409c, 0x18000080, 0xd3d9409d, 0x18000080, 0xd3d9409e, 0x18000080, 4131 0xd3d9409f, 0x18000080, 0xd3d940a0, 0x18000080, 0xd3d940a1, 0x18000080, 4132 0xd3d940a2, 0x18000080, 0xd3d940a3, 0x18000080, 0xd3d940a4, 0x18000080, 4133 0xd3d940a5, 0x18000080, 0xd3d940a6, 0x18000080, 0xd3d940a7, 0x18000080, 4134 0xd3d940a8, 0x18000080, 0xd3d940a9, 0x18000080, 0xd3d940aa, 0x18000080, 4135 0xd3d940ab, 0x18000080, 0xd3d940ac, 0x18000080, 0xd3d940ad, 0x18000080, 4136 0xd3d940ae, 0x18000080, 0xd3d940af, 0x18000080, 0xd3d940b0, 0x18000080, 4137 0xd3d940b1, 0x18000080, 0xd3d940b2, 0x18000080, 0xd3d940b3, 0x18000080, 4138 0xd3d940b4, 0x18000080, 0xd3d940b5, 0x18000080, 0xd3d940b6, 0x18000080, 4139 0xd3d940b7, 0x18000080, 0xd3d940b8, 0x18000080, 0xd3d940b9, 0x18000080, 4140 0xd3d940ba, 0x18000080, 0xd3d940bb, 0x18000080, 0xd3d940bc, 0x18000080, 4141 0xd3d940bd, 0x18000080, 0xd3d940be, 0x18000080, 0xd3d940bf, 0x18000080, 4142 0xd3d940c0, 0x18000080, 0xd3d940c1, 0x18000080, 0xd3d940c2, 0x18000080, 4143 0xd3d940c3, 0x18000080, 0xd3d940c4, 0x18000080, 0xd3d940c5, 0x18000080, 4144 0xd3d940c6, 0x18000080, 0xd3d940c7, 0x18000080, 0xd3d940c8, 0x18000080, 4145 0xd3d940c9, 0x18000080, 0xd3d940ca, 0x18000080, 0xd3d940cb, 0x18000080, 4146 0xd3d940cc, 0x18000080, 0xd3d940cd, 0x18000080, 0xd3d940ce, 0x18000080, 4147 0xd3d940cf, 0x18000080, 0xd3d940d0, 0x18000080, 0xd3d940d1, 0x18000080, 4148 0xd3d940d2, 0x18000080, 0xd3d940d3, 0x18000080, 0xd3d940d4, 0x18000080, 4149 0xd3d940d5, 0x18000080, 0xd3d940d6, 0x18000080, 0xd3d940d7, 0x18000080, 4150 0xd3d940d8, 0x18000080, 0xd3d940d9, 0x18000080, 0xd3d940da, 0x18000080, 4151 0xd3d940db, 0x18000080, 0xd3d940dc, 0x18000080, 0xd3d940dd, 0x18000080, 4152 0xd3d940de, 0x18000080, 0xd3d940df, 0x18000080, 0xd3d940e0, 0x18000080, 4153 0xd3d940e1, 0x18000080, 0xd3d940e2, 0x18000080, 0xd3d940e3, 0x18000080, 4154 0xd3d940e4, 0x18000080, 0xd3d940e5, 0x18000080, 0xd3d940e6, 0x18000080, 4155 0xd3d940e7, 0x18000080, 0xd3d940e8, 0x18000080, 0xd3d940e9, 0x18000080, 4156 0xd3d940ea, 0x18000080, 0xd3d940eb, 0x18000080, 0xd3d940ec, 0x18000080, 4157 0xd3d940ed, 0x18000080, 0xd3d940ee, 0x18000080, 0xd3d940ef, 0x18000080, 4158 0xd3d940f0, 0x18000080, 0xd3d940f1, 0x18000080, 0xd3d940f2, 0x18000080, 4159 0xd3d940f3, 0x18000080, 0xd3d940f4, 0x18000080, 0xd3d940f5, 0x18000080, 4160 0xd3d940f6, 0x18000080, 0xd3d940f7, 0x18000080, 0xd3d940f8, 0x18000080, 4161 0xd3d940f9, 0x18000080, 0xd3d940fa, 0x18000080, 0xd3d940fb, 0x18000080, 4162 0xd3d940fc, 0x18000080, 0xd3d940fd, 0x18000080, 0xd3d940fe, 0x18000080, 4163 0xd3d940ff, 0x18000080, 0xb07c0000, 0xbe8a00ff, 0x000000f8, 0xbf11080a, 4164 0x7e000280, 0x7e020280, 0x7e040280, 0x7e060280, 0x7e080280, 0x7e0a0280, 4165 0x7e0c0280, 0x7e0e0280, 0x808a880a, 0xbe80320a, 0xbf84fff5, 0xbf9c0000, 4166 0xd28c0001, 0x0001007f, 0xd28d0001, 0x0002027e, 0x10020288, 0xb88b0904, 4167 0xb78b4000, 0xd1196a01, 0x00001701, 0xbe8a0087, 0xbefc00c1, 0xd89c4000, 4168 0x00020201, 0xd89cc080, 0x00040401, 0x320202ff, 0x00000800, 0x808a810a, 4169 0xbf84fff8, 0xbf810000, 4170 }; 4171 4172 /* When below register arrays changed, please update gpr_reg_size, 4173 and sec_ded_counter_reg_size in function gfx_v9_0_do_edc_gpr_workarounds, 4174 to cover all gfx9 ASICs */ 4175 static const struct soc15_reg_entry vgpr_init_regs[] = { 4176 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 }, 4177 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 }, 4178 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 }, 4179 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 }, 4180 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x3f }, 4181 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */ 4182 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff }, 4183 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff }, 4184 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff }, 4185 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff }, 4186 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff }, 4187 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff }, 4188 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff }, 4189 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff }, 4190 }; 4191 4192 static const struct soc15_reg_entry vgpr_init_regs_arcturus[] = { 4193 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 }, 4194 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 }, 4195 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 4 }, 4196 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 }, 4197 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0xbf }, 4198 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x400000 }, /* 64KB LDS */ 4199 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0xffffffff }, 4200 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0xffffffff }, 4201 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0xffffffff }, 4202 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0xffffffff }, 4203 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0xffffffff }, 4204 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0xffffffff }, 4205 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0xffffffff }, 4206 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0xffffffff }, 4207 }; 4208 4209 static const struct soc15_reg_entry sgpr1_init_regs[] = { 4210 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 }, 4211 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 }, 4212 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 }, 4213 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 }, 4214 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */ 4215 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 }, 4216 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x000000ff }, 4217 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x000000ff }, 4218 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x000000ff }, 4219 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x000000ff }, 4220 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x000000ff }, 4221 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x000000ff }, 4222 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x000000ff }, 4223 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x000000ff }, 4224 }; 4225 4226 static const struct soc15_reg_entry sgpr2_init_regs[] = { 4227 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_RESOURCE_LIMITS), 0x0000000 }, 4228 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_X), 0x40 }, 4229 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Y), 8 }, 4230 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_NUM_THREAD_Z), 1 }, 4231 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC1), 0x240 }, /* (80 GPRS) */ 4232 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_PGM_RSRC2), 0x0 }, 4233 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE0), 0x0000ff00 }, 4234 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE1), 0x0000ff00 }, 4235 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE2), 0x0000ff00 }, 4236 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE3), 0x0000ff00 }, 4237 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE4), 0x0000ff00 }, 4238 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE5), 0x0000ff00 }, 4239 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE6), 0x0000ff00 }, 4240 { SOC15_REG_ENTRY(GC, 0, mmCOMPUTE_STATIC_THREAD_MGMT_SE7), 0x0000ff00 }, 4241 }; 4242 4243 static const struct soc15_reg_entry gfx_v9_0_edc_counter_regs[] = { 4244 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT), 0, 1, 1}, 4245 { SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT), 0, 1, 1}, 4246 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 0, 1, 1}, 4247 { SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT), 0, 1, 1}, 4248 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 0, 1, 1}, 4249 { SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT), 0, 1, 1}, 4250 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT), 0, 1, 1}, 4251 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT), 0, 1, 1}, 4252 { SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT), 0, 1, 1}, 4253 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 0, 1, 1}, 4254 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_GRBM_CNT), 0, 1, 1}, 4255 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_DED), 0, 1, 1}, 4256 { SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT), 0, 4, 1}, 4257 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 0, 4, 6}, 4258 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_DED_CNT), 0, 4, 16}, 4259 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_INFO), 0, 4, 16}, 4260 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_SEC_CNT), 0, 4, 16}, 4261 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 0, 1, 16}, 4262 { SOC15_REG_ENTRY(GC, 0, mmTCP_ATC_EDC_GATCL1_CNT), 0, 4, 16}, 4263 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT), 0, 4, 16}, 4264 { SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 0, 4, 16}, 4265 { SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 0, 4, 16}, 4266 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 0, 4, 6}, 4267 { SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 0, 4, 16}, 4268 { SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 0, 4, 16}, 4269 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 0, 1, 1}, 4270 { SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 0, 1, 1}, 4271 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 0, 1, 32}, 4272 { SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 0, 1, 32}, 4273 { SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT), 0, 1, 72}, 4274 { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 0, 1, 16}, 4275 { SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 0, 1, 2}, 4276 { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 0, 4, 6}, 4277 }; 4278 4279 static int gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device *adev) 4280 { 4281 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0]; 4282 int i, r; 4283 4284 /* only support when RAS is enabled */ 4285 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) 4286 return 0; 4287 4288 r = amdgpu_ring_alloc(ring, 7); 4289 if (r) { 4290 DRM_ERROR("amdgpu: GDS workarounds failed to lock ring %s (%d).\n", 4291 ring->name, r); 4292 return r; 4293 } 4294 4295 WREG32_SOC15(GC, 0, mmGDS_VMID0_BASE, 0x00000000); 4296 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, adev->gds.gds_size); 4297 4298 amdgpu_ring_write(ring, PACKET3(PACKET3_DMA_DATA, 5)); 4299 amdgpu_ring_write(ring, (PACKET3_DMA_DATA_CP_SYNC | 4300 PACKET3_DMA_DATA_DST_SEL(1) | 4301 PACKET3_DMA_DATA_SRC_SEL(2) | 4302 PACKET3_DMA_DATA_ENGINE(0))); 4303 amdgpu_ring_write(ring, 0); 4304 amdgpu_ring_write(ring, 0); 4305 amdgpu_ring_write(ring, 0); 4306 amdgpu_ring_write(ring, 0); 4307 amdgpu_ring_write(ring, PACKET3_DMA_DATA_CMD_RAW_WAIT | 4308 adev->gds.gds_size); 4309 4310 amdgpu_ring_commit(ring); 4311 4312 for (i = 0; i < adev->usec_timeout; i++) { 4313 if (ring->wptr == gfx_v9_0_ring_get_rptr_compute(ring)) 4314 break; 4315 udelay(1); 4316 } 4317 4318 if (i >= adev->usec_timeout) 4319 r = -ETIMEDOUT; 4320 4321 WREG32_SOC15(GC, 0, mmGDS_VMID0_SIZE, 0x00000000); 4322 4323 return r; 4324 } 4325 4326 static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev) 4327 { 4328 struct amdgpu_ring *ring = &adev->gfx.compute_ring[0]; 4329 struct amdgpu_ib ib; 4330 struct dma_fence *f = NULL; 4331 int r, i; 4332 unsigned total_size, vgpr_offset, sgpr_offset; 4333 u64 gpu_addr; 4334 4335 int compute_dim_x = adev->gfx.config.max_shader_engines * 4336 adev->gfx.config.max_cu_per_sh * 4337 adev->gfx.config.max_sh_per_se; 4338 int sgpr_work_group_size = 5; 4339 int gpr_reg_size = adev->gfx.config.max_shader_engines + 6; 4340 int vgpr_init_shader_size; 4341 const u32 *vgpr_init_shader_ptr; 4342 const struct soc15_reg_entry *vgpr_init_regs_ptr; 4343 4344 /* only support when RAS is enabled */ 4345 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) 4346 return 0; 4347 4348 /* bail if the compute ring is not ready */ 4349 if (!ring->sched.ready) 4350 return 0; 4351 4352 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1)) { 4353 vgpr_init_shader_ptr = vgpr_init_compute_shader_arcturus; 4354 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader_arcturus); 4355 vgpr_init_regs_ptr = vgpr_init_regs_arcturus; 4356 } else { 4357 vgpr_init_shader_ptr = vgpr_init_compute_shader; 4358 vgpr_init_shader_size = sizeof(vgpr_init_compute_shader); 4359 vgpr_init_regs_ptr = vgpr_init_regs; 4360 } 4361 4362 total_size = 4363 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* VGPRS */ 4364 total_size += 4365 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS1 */ 4366 total_size += 4367 (gpr_reg_size * 3 + 4 + 5 + 2) * 4; /* SGPRS2 */ 4368 total_size = ALIGN(total_size, 256); 4369 vgpr_offset = total_size; 4370 total_size += ALIGN(vgpr_init_shader_size, 256); 4371 sgpr_offset = total_size; 4372 total_size += sizeof(sgpr_init_compute_shader); 4373 4374 /* allocate an indirect buffer to put the commands in */ 4375 memset(&ib, 0, sizeof(ib)); 4376 r = amdgpu_ib_get(adev, NULL, total_size, 4377 AMDGPU_IB_POOL_DIRECT, &ib); 4378 if (r) { 4379 DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); 4380 return r; 4381 } 4382 4383 /* load the compute shaders */ 4384 for (i = 0; i < vgpr_init_shader_size/sizeof(u32); i++) 4385 ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_shader_ptr[i]; 4386 4387 for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++) 4388 ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i]; 4389 4390 /* init the ib length to 0 */ 4391 ib.length_dw = 0; 4392 4393 /* VGPR */ 4394 /* write the register state for the compute dispatch */ 4395 for (i = 0; i < gpr_reg_size; i++) { 4396 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); 4397 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(vgpr_init_regs_ptr[i]) 4398 - PACKET3_SET_SH_REG_START; 4399 ib.ptr[ib.length_dw++] = vgpr_init_regs_ptr[i].reg_value; 4400 } 4401 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ 4402 gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8; 4403 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); 4404 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO) 4405 - PACKET3_SET_SH_REG_START; 4406 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); 4407 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); 4408 4409 /* write dispatch packet */ 4410 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); 4411 ib.ptr[ib.length_dw++] = compute_dim_x * 2; /* x */ 4412 ib.ptr[ib.length_dw++] = 1; /* y */ 4413 ib.ptr[ib.length_dw++] = 1; /* z */ 4414 ib.ptr[ib.length_dw++] = 4415 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); 4416 4417 /* write CS partial flush packet */ 4418 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); 4419 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); 4420 4421 /* SGPR1 */ 4422 /* write the register state for the compute dispatch */ 4423 for (i = 0; i < gpr_reg_size; i++) { 4424 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); 4425 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr1_init_regs[i]) 4426 - PACKET3_SET_SH_REG_START; 4427 ib.ptr[ib.length_dw++] = sgpr1_init_regs[i].reg_value; 4428 } 4429 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ 4430 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8; 4431 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); 4432 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO) 4433 - PACKET3_SET_SH_REG_START; 4434 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); 4435 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); 4436 4437 /* write dispatch packet */ 4438 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); 4439 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */ 4440 ib.ptr[ib.length_dw++] = 1; /* y */ 4441 ib.ptr[ib.length_dw++] = 1; /* z */ 4442 ib.ptr[ib.length_dw++] = 4443 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); 4444 4445 /* write CS partial flush packet */ 4446 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); 4447 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); 4448 4449 /* SGPR2 */ 4450 /* write the register state for the compute dispatch */ 4451 for (i = 0; i < gpr_reg_size; i++) { 4452 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1); 4453 ib.ptr[ib.length_dw++] = SOC15_REG_ENTRY_OFFSET(sgpr2_init_regs[i]) 4454 - PACKET3_SET_SH_REG_START; 4455 ib.ptr[ib.length_dw++] = sgpr2_init_regs[i].reg_value; 4456 } 4457 /* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */ 4458 gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8; 4459 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2); 4460 ib.ptr[ib.length_dw++] = SOC15_REG_OFFSET(GC, 0, mmCOMPUTE_PGM_LO) 4461 - PACKET3_SET_SH_REG_START; 4462 ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr); 4463 ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr); 4464 4465 /* write dispatch packet */ 4466 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3); 4467 ib.ptr[ib.length_dw++] = compute_dim_x / 2 * sgpr_work_group_size; /* x */ 4468 ib.ptr[ib.length_dw++] = 1; /* y */ 4469 ib.ptr[ib.length_dw++] = 1; /* z */ 4470 ib.ptr[ib.length_dw++] = 4471 REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1); 4472 4473 /* write CS partial flush packet */ 4474 ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0); 4475 ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4); 4476 4477 /* shedule the ib on the ring */ 4478 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); 4479 if (r) { 4480 DRM_ERROR("amdgpu: ib submit failed (%d).\n", r); 4481 goto fail; 4482 } 4483 4484 /* wait for the GPU to finish processing the IB */ 4485 r = dma_fence_wait(f, false); 4486 if (r) { 4487 DRM_ERROR("amdgpu: fence wait failed (%d).\n", r); 4488 goto fail; 4489 } 4490 4491 fail: 4492 amdgpu_ib_free(adev, &ib, NULL); 4493 dma_fence_put(f); 4494 4495 return r; 4496 } 4497 4498 static int gfx_v9_0_early_init(void *handle) 4499 { 4500 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4501 4502 adev->gfx.funcs = &gfx_v9_0_gfx_funcs; 4503 4504 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1) || 4505 adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2)) 4506 adev->gfx.num_gfx_rings = 0; 4507 else 4508 adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS; 4509 adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev), 4510 AMDGPU_MAX_COMPUTE_RINGS); 4511 gfx_v9_0_set_kiq_pm4_funcs(adev); 4512 gfx_v9_0_set_ring_funcs(adev); 4513 gfx_v9_0_set_irq_funcs(adev); 4514 gfx_v9_0_set_gds_init(adev); 4515 gfx_v9_0_set_rlc_funcs(adev); 4516 4517 /* init rlcg reg access ctrl */ 4518 gfx_v9_0_init_rlcg_reg_access_ctrl(adev); 4519 4520 return gfx_v9_0_init_microcode(adev); 4521 } 4522 4523 static int gfx_v9_0_ecc_late_init(void *handle) 4524 { 4525 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4526 int r; 4527 4528 /* 4529 * Temp workaround to fix the issue that CP firmware fails to 4530 * update read pointer when CPDMA is writing clearing operation 4531 * to GDS in suspend/resume sequence on several cards. So just 4532 * limit this operation in cold boot sequence. 4533 */ 4534 if ((!adev->in_suspend) && 4535 (adev->gds.gds_size)) { 4536 r = gfx_v9_0_do_edc_gds_workarounds(adev); 4537 if (r) 4538 return r; 4539 } 4540 4541 /* requires IBs so do in late init after IB pool is initialized */ 4542 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 2)) 4543 r = gfx_v9_4_2_do_edc_gpr_workarounds(adev); 4544 else 4545 r = gfx_v9_0_do_edc_gpr_workarounds(adev); 4546 4547 if (r) 4548 return r; 4549 4550 if (adev->gfx.ras && 4551 adev->gfx.ras->enable_watchdog_timer) 4552 adev->gfx.ras->enable_watchdog_timer(adev); 4553 4554 return 0; 4555 } 4556 4557 static int gfx_v9_0_late_init(void *handle) 4558 { 4559 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4560 int r; 4561 4562 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0); 4563 if (r) 4564 return r; 4565 4566 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0); 4567 if (r) 4568 return r; 4569 4570 r = gfx_v9_0_ecc_late_init(handle); 4571 if (r) 4572 return r; 4573 4574 return 0; 4575 } 4576 4577 static bool gfx_v9_0_is_rlc_enabled(struct amdgpu_device *adev) 4578 { 4579 uint32_t rlc_setting; 4580 4581 /* if RLC is not enabled, do nothing */ 4582 rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL); 4583 if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK)) 4584 return false; 4585 4586 return true; 4587 } 4588 4589 static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev) 4590 { 4591 uint32_t data; 4592 unsigned i; 4593 4594 data = RLC_SAFE_MODE__CMD_MASK; 4595 data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); 4596 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); 4597 4598 /* wait for RLC_SAFE_MODE */ 4599 for (i = 0; i < adev->usec_timeout; i++) { 4600 if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) 4601 break; 4602 udelay(1); 4603 } 4604 } 4605 4606 static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev) 4607 { 4608 uint32_t data; 4609 4610 data = RLC_SAFE_MODE__CMD_MASK; 4611 WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); 4612 } 4613 4614 static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev, 4615 bool enable) 4616 { 4617 amdgpu_gfx_rlc_enter_safe_mode(adev); 4618 4619 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) { 4620 gfx_v9_0_enable_gfx_cg_power_gating(adev, true); 4621 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE) 4622 gfx_v9_0_enable_gfx_pipeline_powergating(adev, true); 4623 } else { 4624 gfx_v9_0_enable_gfx_cg_power_gating(adev, false); 4625 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE) 4626 gfx_v9_0_enable_gfx_pipeline_powergating(adev, false); 4627 } 4628 4629 amdgpu_gfx_rlc_exit_safe_mode(adev); 4630 } 4631 4632 static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev, 4633 bool enable) 4634 { 4635 /* TODO: double check if we need to perform under safe mode */ 4636 /* gfx_v9_0_enter_rlc_safe_mode(adev); */ 4637 4638 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable) 4639 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, true); 4640 else 4641 gfx_v9_0_enable_gfx_static_mg_power_gating(adev, false); 4642 4643 if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable) 4644 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, true); 4645 else 4646 gfx_v9_0_enable_gfx_dynamic_mg_power_gating(adev, false); 4647 4648 /* gfx_v9_0_exit_rlc_safe_mode(adev); */ 4649 } 4650 4651 static void gfx_v9_0_update_medium_grain_clock_gating(struct amdgpu_device *adev, 4652 bool enable) 4653 { 4654 uint32_t data, def; 4655 4656 amdgpu_gfx_rlc_enter_safe_mode(adev); 4657 4658 /* It is disabled by HW by default */ 4659 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) { 4660 /* 1 - RLC_CGTT_MGCG_OVERRIDE */ 4661 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE); 4662 4663 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1)) 4664 data &= ~RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK; 4665 4666 data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK | 4667 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK | 4668 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK); 4669 4670 /* only for Vega10 & Raven1 */ 4671 data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK; 4672 4673 if (def != data) 4674 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data); 4675 4676 /* MGLS is a global flag to control all MGLS in GFX */ 4677 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) { 4678 /* 2 - RLC memory Light sleep */ 4679 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) { 4680 def = data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL); 4681 data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK; 4682 if (def != data) 4683 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data); 4684 } 4685 /* 3 - CP memory Light sleep */ 4686 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) { 4687 def = data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL); 4688 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 4689 if (def != data) 4690 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data); 4691 } 4692 } 4693 } else { 4694 /* 1 - MGCG_OVERRIDE */ 4695 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE); 4696 4697 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 2, 1)) 4698 data |= RLC_CGTT_MGCG_OVERRIDE__CPF_CGTT_SCLK_OVERRIDE_MASK; 4699 4700 data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK | 4701 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK | 4702 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK | 4703 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK); 4704 4705 if (def != data) 4706 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data); 4707 4708 /* 2 - disable MGLS in RLC */ 4709 data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL); 4710 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) { 4711 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK; 4712 WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data); 4713 } 4714 4715 /* 3 - disable MGLS in CP */ 4716 data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL); 4717 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) { 4718 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 4719 WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data); 4720 } 4721 } 4722 4723 amdgpu_gfx_rlc_exit_safe_mode(adev); 4724 } 4725 4726 static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev, 4727 bool enable) 4728 { 4729 uint32_t data, def; 4730 4731 if (!adev->gfx.num_gfx_rings) 4732 return; 4733 4734 amdgpu_gfx_rlc_enter_safe_mode(adev); 4735 4736 /* Enable 3D CGCG/CGLS */ 4737 if (enable) { 4738 /* write cmd to clear cgcg/cgls ov */ 4739 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE); 4740 /* unset CGCG override */ 4741 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK; 4742 /* update CGCG and CGLS override bits */ 4743 if (def != data) 4744 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data); 4745 4746 /* enable 3Dcgcg FSM(0x0000363f) */ 4747 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D); 4748 4749 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) 4750 data = (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) | 4751 RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK; 4752 else 4753 data = 0x0 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT; 4754 4755 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) 4756 data |= (0x000F << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) | 4757 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK; 4758 if (def != data) 4759 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data); 4760 4761 /* set IDLE_POLL_COUNT(0x00900100) */ 4762 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL); 4763 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) | 4764 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT); 4765 if (def != data) 4766 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data); 4767 } else { 4768 /* Disable CGCG/CGLS */ 4769 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D); 4770 /* disable cgcg, cgls should be disabled */ 4771 data &= ~(RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK | 4772 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK); 4773 /* disable cgcg and cgls in FSM */ 4774 if (def != data) 4775 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data); 4776 } 4777 4778 amdgpu_gfx_rlc_exit_safe_mode(adev); 4779 } 4780 4781 static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev, 4782 bool enable) 4783 { 4784 uint32_t def, data; 4785 4786 amdgpu_gfx_rlc_enter_safe_mode(adev); 4787 4788 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { 4789 def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE); 4790 /* unset CGCG override */ 4791 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK; 4792 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) 4793 data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK; 4794 else 4795 data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK; 4796 /* update CGCG and CGLS override bits */ 4797 if (def != data) 4798 WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data); 4799 4800 /* enable cgcg FSM(0x0000363F) */ 4801 def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL); 4802 4803 if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 1)) 4804 data = (0x2000 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) | 4805 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK; 4806 else 4807 data = (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) | 4808 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK; 4809 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) 4810 data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) | 4811 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK; 4812 if (def != data) 4813 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data); 4814 4815 /* set IDLE_POLL_COUNT(0x00900100) */ 4816 def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL); 4817 data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) | 4818 (0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT); 4819 if (def != data) 4820 WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data); 4821 } else { 4822 def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL); 4823 /* reset CGCG/CGLS bits */ 4824 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK); 4825 /* disable cgcg and cgls in FSM */ 4826 if (def != data) 4827 WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data); 4828 } 4829 4830 amdgpu_gfx_rlc_exit_safe_mode(adev); 4831 } 4832 4833 static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev, 4834 bool enable) 4835 { 4836 if (enable) { 4837 /* CGCG/CGLS should be enabled after MGCG/MGLS 4838 * === MGCG + MGLS === 4839 */ 4840 gfx_v9_0_update_medium_grain_clock_gating(adev, enable); 4841 /* === CGCG /CGLS for GFX 3D Only === */ 4842 gfx_v9_0_update_3d_clock_gating(adev, enable); 4843 /* === CGCG + CGLS === */ 4844 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable); 4845 } else { 4846 /* CGCG/CGLS should be disabled before MGCG/MGLS 4847 * === CGCG + CGLS === 4848 */ 4849 gfx_v9_0_update_coarse_grain_clock_gating(adev, enable); 4850 /* === CGCG /CGLS for GFX 3D Only === */ 4851 gfx_v9_0_update_3d_clock_gating(adev, enable); 4852 /* === MGCG + MGLS === */ 4853 gfx_v9_0_update_medium_grain_clock_gating(adev, enable); 4854 } 4855 return 0; 4856 } 4857 4858 static void gfx_v9_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid) 4859 { 4860 u32 reg, data; 4861 4862 amdgpu_gfx_off_ctrl(adev, false); 4863 4864 reg = SOC15_REG_OFFSET(GC, 0, mmRLC_SPM_MC_CNTL); 4865 if (amdgpu_sriov_is_pp_one_vf(adev)) 4866 data = RREG32_NO_KIQ(reg); 4867 else 4868 data = RREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL); 4869 4870 data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK; 4871 data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT; 4872 4873 if (amdgpu_sriov_is_pp_one_vf(adev)) 4874 WREG32_SOC15_NO_KIQ(GC, 0, mmRLC_SPM_MC_CNTL, data); 4875 else 4876 WREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL, data); 4877 4878 amdgpu_gfx_off_ctrl(adev, true); 4879 } 4880 4881 static bool gfx_v9_0_check_rlcg_range(struct amdgpu_device *adev, 4882 uint32_t offset, 4883 struct soc15_reg_rlcg *entries, int arr_size) 4884 { 4885 int i; 4886 uint32_t reg; 4887 4888 if (!entries) 4889 return false; 4890 4891 for (i = 0; i < arr_size; i++) { 4892 const struct soc15_reg_rlcg *entry; 4893 4894 entry = &entries[i]; 4895 reg = adev->reg_offset[entry->hwip][entry->instance][entry->segment] + entry->reg; 4896 if (offset == reg) 4897 return true; 4898 } 4899 4900 return false; 4901 } 4902 4903 static bool gfx_v9_0_is_rlcg_access_range(struct amdgpu_device *adev, u32 offset) 4904 { 4905 return gfx_v9_0_check_rlcg_range(adev, offset, 4906 (void *)rlcg_access_gc_9_0, 4907 ARRAY_SIZE(rlcg_access_gc_9_0)); 4908 } 4909 4910 static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = { 4911 .is_rlc_enabled = gfx_v9_0_is_rlc_enabled, 4912 .set_safe_mode = gfx_v9_0_set_safe_mode, 4913 .unset_safe_mode = gfx_v9_0_unset_safe_mode, 4914 .init = gfx_v9_0_rlc_init, 4915 .get_csb_size = gfx_v9_0_get_csb_size, 4916 .get_csb_buffer = gfx_v9_0_get_csb_buffer, 4917 .get_cp_table_num = gfx_v9_0_cp_jump_table_num, 4918 .resume = gfx_v9_0_rlc_resume, 4919 .stop = gfx_v9_0_rlc_stop, 4920 .reset = gfx_v9_0_rlc_reset, 4921 .start = gfx_v9_0_rlc_start, 4922 .update_spm_vmid = gfx_v9_0_update_spm_vmid, 4923 .is_rlcg_access_range = gfx_v9_0_is_rlcg_access_range, 4924 }; 4925 4926 static int gfx_v9_0_set_powergating_state(void *handle, 4927 enum amd_powergating_state state) 4928 { 4929 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4930 bool enable = (state == AMD_PG_STATE_GATE); 4931 4932 switch (adev->ip_versions[GC_HWIP][0]) { 4933 case IP_VERSION(9, 2, 2): 4934 case IP_VERSION(9, 1, 0): 4935 case IP_VERSION(9, 3, 0): 4936 if (!enable) 4937 amdgpu_gfx_off_ctrl(adev, false); 4938 4939 if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) { 4940 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, true); 4941 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, true); 4942 } else { 4943 gfx_v9_0_enable_sck_slow_down_on_power_up(adev, false); 4944 gfx_v9_0_enable_sck_slow_down_on_power_down(adev, false); 4945 } 4946 4947 if (adev->pg_flags & AMD_PG_SUPPORT_CP) 4948 gfx_v9_0_enable_cp_power_gating(adev, true); 4949 else 4950 gfx_v9_0_enable_cp_power_gating(adev, false); 4951 4952 /* update gfx cgpg state */ 4953 gfx_v9_0_update_gfx_cg_power_gating(adev, enable); 4954 4955 /* update mgcg state */ 4956 gfx_v9_0_update_gfx_mg_power_gating(adev, enable); 4957 4958 if (enable) 4959 amdgpu_gfx_off_ctrl(adev, true); 4960 break; 4961 case IP_VERSION(9, 2, 1): 4962 amdgpu_gfx_off_ctrl(adev, enable); 4963 break; 4964 default: 4965 break; 4966 } 4967 4968 return 0; 4969 } 4970 4971 static int gfx_v9_0_set_clockgating_state(void *handle, 4972 enum amd_clockgating_state state) 4973 { 4974 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4975 4976 if (amdgpu_sriov_vf(adev)) 4977 return 0; 4978 4979 switch (adev->ip_versions[GC_HWIP][0]) { 4980 case IP_VERSION(9, 0, 1): 4981 case IP_VERSION(9, 2, 1): 4982 case IP_VERSION(9, 4, 0): 4983 case IP_VERSION(9, 2, 2): 4984 case IP_VERSION(9, 1, 0): 4985 case IP_VERSION(9, 4, 1): 4986 case IP_VERSION(9, 3, 0): 4987 case IP_VERSION(9, 4, 2): 4988 gfx_v9_0_update_gfx_clock_gating(adev, 4989 state == AMD_CG_STATE_GATE); 4990 break; 4991 default: 4992 break; 4993 } 4994 return 0; 4995 } 4996 4997 static void gfx_v9_0_get_clockgating_state(void *handle, u64 *flags) 4998 { 4999 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 5000 int data; 5001 5002 if (amdgpu_sriov_vf(adev)) 5003 *flags = 0; 5004 5005 /* AMD_CG_SUPPORT_GFX_MGCG */ 5006 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE)); 5007 if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK)) 5008 *flags |= AMD_CG_SUPPORT_GFX_MGCG; 5009 5010 /* AMD_CG_SUPPORT_GFX_CGCG */ 5011 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL)); 5012 if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK) 5013 *flags |= AMD_CG_SUPPORT_GFX_CGCG; 5014 5015 /* AMD_CG_SUPPORT_GFX_CGLS */ 5016 if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK) 5017 *flags |= AMD_CG_SUPPORT_GFX_CGLS; 5018 5019 /* AMD_CG_SUPPORT_GFX_RLC_LS */ 5020 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_MEM_SLP_CNTL)); 5021 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) 5022 *flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS; 5023 5024 /* AMD_CG_SUPPORT_GFX_CP_LS */ 5025 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmCP_MEM_SLP_CNTL)); 5026 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) 5027 *flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS; 5028 5029 if (adev->ip_versions[GC_HWIP][0] != IP_VERSION(9, 4, 1)) { 5030 /* AMD_CG_SUPPORT_GFX_3D_CGCG */ 5031 data = RREG32_KIQ(SOC15_REG_OFFSET(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D)); 5032 if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK) 5033 *flags |= AMD_CG_SUPPORT_GFX_3D_CGCG; 5034 5035 /* AMD_CG_SUPPORT_GFX_3D_CGLS */ 5036 if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK) 5037 *flags |= AMD_CG_SUPPORT_GFX_3D_CGLS; 5038 } 5039 } 5040 5041 static u64 gfx_v9_0_ring_get_rptr_gfx(struct amdgpu_ring *ring) 5042 { 5043 return *ring->rptr_cpu_addr; /* gfx9 is 32bit rptr*/ 5044 } 5045 5046 static u64 gfx_v9_0_ring_get_wptr_gfx(struct amdgpu_ring *ring) 5047 { 5048 struct amdgpu_device *adev = ring->adev; 5049 u64 wptr; 5050 5051 /* XXX check if swapping is necessary on BE */ 5052 if (ring->use_doorbell) { 5053 wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr); 5054 } else { 5055 wptr = RREG32_SOC15(GC, 0, mmCP_RB0_WPTR); 5056 wptr += (u64)RREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI) << 32; 5057 } 5058 5059 return wptr; 5060 } 5061 5062 static void gfx_v9_0_ring_set_wptr_gfx(struct amdgpu_ring *ring) 5063 { 5064 struct amdgpu_device *adev = ring->adev; 5065 5066 if (ring->use_doorbell) { 5067 /* XXX check if swapping is necessary on BE */ 5068 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr); 5069 WDOORBELL64(ring->doorbell_index, ring->wptr); 5070 } else { 5071 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr)); 5072 WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr)); 5073 } 5074 } 5075 5076 static void gfx_v9_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) 5077 { 5078 struct amdgpu_device *adev = ring->adev; 5079 u32 ref_and_mask, reg_mem_engine; 5080 const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg; 5081 5082 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) { 5083 switch (ring->me) { 5084 case 1: 5085 ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe; 5086 break; 5087 case 2: 5088 ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe; 5089 break; 5090 default: 5091 return; 5092 } 5093 reg_mem_engine = 0; 5094 } else { 5095 ref_and_mask = nbio_hf_reg->ref_and_mask_cp0; 5096 reg_mem_engine = 1; /* pfp */ 5097 } 5098 5099 gfx_v9_0_wait_reg_mem(ring, reg_mem_engine, 0, 1, 5100 adev->nbio.funcs->get_hdp_flush_req_offset(adev), 5101 adev->nbio.funcs->get_hdp_flush_done_offset(adev), 5102 ref_and_mask, ref_and_mask, 0x20); 5103 } 5104 5105 static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, 5106 struct amdgpu_job *job, 5107 struct amdgpu_ib *ib, 5108 uint32_t flags) 5109 { 5110 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 5111 u32 header, control = 0; 5112 5113 if (ib->flags & AMDGPU_IB_FLAG_CE) 5114 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2); 5115 else 5116 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); 5117 5118 control |= ib->length_dw | (vmid << 24); 5119 5120 if (ib->flags & AMDGPU_IB_FLAG_PREEMPT) { 5121 control |= INDIRECT_BUFFER_PRE_ENB(1); 5122 5123 if (flags & AMDGPU_IB_PREEMPTED) 5124 control |= INDIRECT_BUFFER_PRE_RESUME(1); 5125 5126 if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid) 5127 gfx_v9_0_ring_emit_de_meta(ring, 5128 (!amdgpu_sriov_vf(ring->adev) && 5129 flags & AMDGPU_IB_PREEMPTED) ? 5130 true : false); 5131 } 5132 5133 amdgpu_ring_write(ring, header); 5134 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */ 5135 amdgpu_ring_write(ring, 5136 #ifdef __BIG_ENDIAN 5137 (2 << 0) | 5138 #endif 5139 lower_32_bits(ib->gpu_addr)); 5140 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 5141 amdgpu_ring_write(ring, control); 5142 } 5143 5144 static void gfx_v9_0_ring_emit_ib_compute(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 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); 5151 5152 /* Currently, there is a high possibility to get wave ID mismatch 5153 * between ME and GDS, leading to a hw deadlock, because ME generates 5154 * different wave IDs than the GDS expects. This situation happens 5155 * randomly when at least 5 compute pipes use GDS ordered append. 5156 * The wave IDs generated by ME are also wrong after suspend/resume. 5157 * Those are probably bugs somewhere else in the kernel driver. 5158 * 5159 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and 5160 * GDS to 0 for this ring (me/pipe). 5161 */ 5162 if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) { 5163 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 5164 amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID); 5165 amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id); 5166 } 5167 5168 amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); 5169 BUG_ON(ib->gpu_addr & 0x3); /* Dword align */ 5170 amdgpu_ring_write(ring, 5171 #ifdef __BIG_ENDIAN 5172 (2 << 0) | 5173 #endif 5174 lower_32_bits(ib->gpu_addr)); 5175 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); 5176 amdgpu_ring_write(ring, control); 5177 } 5178 5179 static void gfx_v9_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, 5180 u64 seq, unsigned flags) 5181 { 5182 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT; 5183 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT; 5184 bool writeback = flags & AMDGPU_FENCE_FLAG_TC_WB_ONLY; 5185 bool exec = flags & AMDGPU_FENCE_FLAG_EXEC; 5186 uint32_t dw2 = 0; 5187 5188 /* RELEASE_MEM - flush caches, send int */ 5189 amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6)); 5190 5191 if (writeback) { 5192 dw2 = EOP_TC_NC_ACTION_EN; 5193 } else { 5194 dw2 = EOP_TCL1_ACTION_EN | EOP_TC_ACTION_EN | 5195 EOP_TC_MD_ACTION_EN; 5196 } 5197 dw2 |= EOP_TC_WB_ACTION_EN | EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | 5198 EVENT_INDEX(5); 5199 if (exec) 5200 dw2 |= EOP_EXEC; 5201 5202 amdgpu_ring_write(ring, dw2); 5203 amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0)); 5204 5205 /* 5206 * the address should be Qword aligned if 64bit write, Dword 5207 * aligned if only send 32bit data low (discard data high) 5208 */ 5209 if (write64bit) 5210 BUG_ON(addr & 0x7); 5211 else 5212 BUG_ON(addr & 0x3); 5213 amdgpu_ring_write(ring, lower_32_bits(addr)); 5214 amdgpu_ring_write(ring, upper_32_bits(addr)); 5215 amdgpu_ring_write(ring, lower_32_bits(seq)); 5216 amdgpu_ring_write(ring, upper_32_bits(seq)); 5217 amdgpu_ring_write(ring, 0); 5218 } 5219 5220 static void gfx_v9_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring) 5221 { 5222 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 5223 uint32_t seq = ring->fence_drv.sync_seq; 5224 uint64_t addr = ring->fence_drv.gpu_addr; 5225 5226 gfx_v9_0_wait_reg_mem(ring, usepfp, 1, 0, 5227 lower_32_bits(addr), upper_32_bits(addr), 5228 seq, 0xffffffff, 4); 5229 } 5230 5231 static void gfx_v9_0_ring_emit_vm_flush(struct amdgpu_ring *ring, 5232 unsigned vmid, uint64_t pd_addr) 5233 { 5234 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 5235 5236 /* compute doesn't have PFP */ 5237 if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) { 5238 /* sync PFP to ME, otherwise we might get invalid PFP reads */ 5239 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0)); 5240 amdgpu_ring_write(ring, 0x0); 5241 } 5242 } 5243 5244 static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring) 5245 { 5246 return *ring->rptr_cpu_addr; /* gfx9 hardware is 32bit rptr */ 5247 } 5248 5249 static u64 gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring *ring) 5250 { 5251 u64 wptr; 5252 5253 /* XXX check if swapping is necessary on BE */ 5254 if (ring->use_doorbell) 5255 wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr); 5256 else 5257 BUG(); 5258 return wptr; 5259 } 5260 5261 static void gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring *ring) 5262 { 5263 struct amdgpu_device *adev = ring->adev; 5264 5265 /* XXX check if swapping is necessary on BE */ 5266 if (ring->use_doorbell) { 5267 atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr); 5268 WDOORBELL64(ring->doorbell_index, ring->wptr); 5269 } else{ 5270 BUG(); /* only DOORBELL method supported on gfx9 now */ 5271 } 5272 } 5273 5274 static void gfx_v9_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr, 5275 u64 seq, unsigned int flags) 5276 { 5277 struct amdgpu_device *adev = ring->adev; 5278 5279 /* we only allocate 32bit for each seq wb address */ 5280 BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT); 5281 5282 /* write fence seq to the "addr" */ 5283 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5284 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 5285 WRITE_DATA_DST_SEL(5) | WR_CONFIRM)); 5286 amdgpu_ring_write(ring, lower_32_bits(addr)); 5287 amdgpu_ring_write(ring, upper_32_bits(addr)); 5288 amdgpu_ring_write(ring, lower_32_bits(seq)); 5289 5290 if (flags & AMDGPU_FENCE_FLAG_INT) { 5291 /* set register to trigger INT */ 5292 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5293 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 5294 WRITE_DATA_DST_SEL(0) | WR_CONFIRM)); 5295 amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, mmCPC_INT_STATUS)); 5296 amdgpu_ring_write(ring, 0); 5297 amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */ 5298 } 5299 } 5300 5301 static void gfx_v9_ring_emit_sb(struct amdgpu_ring *ring) 5302 { 5303 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 5304 amdgpu_ring_write(ring, 0); 5305 } 5306 5307 static void gfx_v9_0_ring_emit_ce_meta(struct amdgpu_ring *ring, bool resume) 5308 { 5309 struct amdgpu_device *adev = ring->adev; 5310 struct v9_ce_ib_state ce_payload = {0}; 5311 uint64_t offset, ce_payload_gpu_addr; 5312 void *ce_payload_cpu_addr; 5313 int cnt; 5314 5315 cnt = (sizeof(ce_payload) >> 2) + 4 - 2; 5316 5317 if (ring->is_mes_queue) { 5318 offset = offsetof(struct amdgpu_mes_ctx_meta_data, 5319 gfx[0].gfx_meta_data) + 5320 offsetof(struct v9_gfx_meta_data, ce_payload); 5321 ce_payload_gpu_addr = 5322 amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset); 5323 ce_payload_cpu_addr = 5324 amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset); 5325 } else { 5326 offset = offsetof(struct v9_gfx_meta_data, ce_payload); 5327 ce_payload_gpu_addr = amdgpu_csa_vaddr(ring->adev) + offset; 5328 ce_payload_cpu_addr = adev->virt.csa_cpu_addr + offset; 5329 } 5330 5331 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt)); 5332 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) | 5333 WRITE_DATA_DST_SEL(8) | 5334 WR_CONFIRM) | 5335 WRITE_DATA_CACHE_POLICY(0)); 5336 amdgpu_ring_write(ring, lower_32_bits(ce_payload_gpu_addr)); 5337 amdgpu_ring_write(ring, upper_32_bits(ce_payload_gpu_addr)); 5338 5339 if (resume) 5340 amdgpu_ring_write_multiple(ring, ce_payload_cpu_addr, 5341 sizeof(ce_payload) >> 2); 5342 else 5343 amdgpu_ring_write_multiple(ring, (void *)&ce_payload, 5344 sizeof(ce_payload) >> 2); 5345 } 5346 5347 static int gfx_v9_0_ring_preempt_ib(struct amdgpu_ring *ring) 5348 { 5349 int i, r = 0; 5350 struct amdgpu_device *adev = ring->adev; 5351 struct amdgpu_kiq *kiq = &adev->gfx.kiq; 5352 struct amdgpu_ring *kiq_ring = &kiq->ring; 5353 unsigned long flags; 5354 5355 if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) 5356 return -EINVAL; 5357 5358 spin_lock_irqsave(&kiq->ring_lock, flags); 5359 5360 if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) { 5361 spin_unlock_irqrestore(&kiq->ring_lock, flags); 5362 return -ENOMEM; 5363 } 5364 5365 /* assert preemption condition */ 5366 amdgpu_ring_set_preempt_cond_exec(ring, false); 5367 5368 ring->trail_seq += 1; 5369 amdgpu_ring_alloc(ring, 13); 5370 gfx_v9_0_ring_emit_fence(ring, ring->trail_fence_gpu_addr, 5371 ring->trail_seq, AMDGPU_FENCE_FLAG_EXEC | AMDGPU_FENCE_FLAG_INT); 5372 /*reset the CP_VMID_PREEMPT after trailing fence*/ 5373 amdgpu_ring_emit_wreg(ring, 5374 SOC15_REG_OFFSET(GC, 0, mmCP_VMID_PREEMPT), 5375 0x0); 5376 5377 /* assert IB preemption, emit the trailing fence */ 5378 kiq->pmf->kiq_unmap_queues(kiq_ring, ring, PREEMPT_QUEUES_NO_UNMAP, 5379 ring->trail_fence_gpu_addr, 5380 ring->trail_seq); 5381 5382 amdgpu_ring_commit(kiq_ring); 5383 spin_unlock_irqrestore(&kiq->ring_lock, flags); 5384 5385 /* poll the trailing fence */ 5386 for (i = 0; i < adev->usec_timeout; i++) { 5387 if (ring->trail_seq == 5388 le32_to_cpu(*ring->trail_fence_cpu_addr)) 5389 break; 5390 udelay(1); 5391 } 5392 5393 if (i >= adev->usec_timeout) { 5394 r = -EINVAL; 5395 DRM_WARN("ring %d timeout to preempt ib\n", ring->idx); 5396 } 5397 5398 amdgpu_ring_commit(ring); 5399 5400 /* deassert preemption condition */ 5401 amdgpu_ring_set_preempt_cond_exec(ring, true); 5402 return r; 5403 } 5404 5405 static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume) 5406 { 5407 struct amdgpu_device *adev = ring->adev; 5408 struct v9_de_ib_state de_payload = {0}; 5409 uint64_t offset, gds_addr, de_payload_gpu_addr; 5410 void *de_payload_cpu_addr; 5411 int cnt; 5412 5413 if (ring->is_mes_queue) { 5414 offset = offsetof(struct amdgpu_mes_ctx_meta_data, 5415 gfx[0].gfx_meta_data) + 5416 offsetof(struct v9_gfx_meta_data, de_payload); 5417 de_payload_gpu_addr = 5418 amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset); 5419 de_payload_cpu_addr = 5420 amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset); 5421 5422 offset = offsetof(struct amdgpu_mes_ctx_meta_data, 5423 gfx[0].gds_backup) + 5424 offsetof(struct v9_gfx_meta_data, de_payload); 5425 gds_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset); 5426 } else { 5427 offset = offsetof(struct v9_gfx_meta_data, de_payload); 5428 de_payload_gpu_addr = amdgpu_csa_vaddr(ring->adev) + offset; 5429 de_payload_cpu_addr = adev->virt.csa_cpu_addr + offset; 5430 5431 gds_addr = ALIGN(amdgpu_csa_vaddr(ring->adev) + 5432 AMDGPU_CSA_SIZE - adev->gds.gds_size, 5433 PAGE_SIZE); 5434 } 5435 5436 de_payload.gds_backup_addrlo = lower_32_bits(gds_addr); 5437 de_payload.gds_backup_addrhi = upper_32_bits(gds_addr); 5438 5439 cnt = (sizeof(de_payload) >> 2) + 4 - 2; 5440 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt)); 5441 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) | 5442 WRITE_DATA_DST_SEL(8) | 5443 WR_CONFIRM) | 5444 WRITE_DATA_CACHE_POLICY(0)); 5445 amdgpu_ring_write(ring, lower_32_bits(de_payload_gpu_addr)); 5446 amdgpu_ring_write(ring, upper_32_bits(de_payload_gpu_addr)); 5447 5448 if (resume) 5449 amdgpu_ring_write_multiple(ring, de_payload_cpu_addr, 5450 sizeof(de_payload) >> 2); 5451 else 5452 amdgpu_ring_write_multiple(ring, (void *)&de_payload, 5453 sizeof(de_payload) >> 2); 5454 } 5455 5456 static void gfx_v9_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start, 5457 bool secure) 5458 { 5459 uint32_t v = secure ? FRAME_TMZ : 0; 5460 5461 amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0)); 5462 amdgpu_ring_write(ring, v | FRAME_CMD(start ? 0 : 1)); 5463 } 5464 5465 static void gfx_v9_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags) 5466 { 5467 uint32_t dw2 = 0; 5468 5469 gfx_v9_0_ring_emit_ce_meta(ring, 5470 (!amdgpu_sriov_vf(ring->adev) && 5471 flags & AMDGPU_IB_PREEMPTED) ? true : false); 5472 5473 dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */ 5474 if (flags & AMDGPU_HAVE_CTX_SWITCH) { 5475 /* set load_global_config & load_global_uconfig */ 5476 dw2 |= 0x8001; 5477 /* set load_cs_sh_regs */ 5478 dw2 |= 0x01000000; 5479 /* set load_per_context_state & load_gfx_sh_regs for GFX */ 5480 dw2 |= 0x10002; 5481 5482 /* set load_ce_ram if preamble presented */ 5483 if (AMDGPU_PREAMBLE_IB_PRESENT & flags) 5484 dw2 |= 0x10000000; 5485 } else { 5486 /* still load_ce_ram if this is the first time preamble presented 5487 * although there is no context switch happens. 5488 */ 5489 if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags) 5490 dw2 |= 0x10000000; 5491 } 5492 5493 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 5494 amdgpu_ring_write(ring, dw2); 5495 amdgpu_ring_write(ring, 0); 5496 } 5497 5498 static unsigned gfx_v9_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring) 5499 { 5500 unsigned ret; 5501 amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3)); 5502 amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr)); 5503 amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr)); 5504 amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */ 5505 ret = ring->wptr & ring->buf_mask; 5506 amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */ 5507 return ret; 5508 } 5509 5510 static void gfx_v9_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset) 5511 { 5512 unsigned cur; 5513 BUG_ON(offset > ring->buf_mask); 5514 BUG_ON(ring->ring[offset] != 0x55aa55aa); 5515 5516 cur = (ring->wptr - 1) & ring->buf_mask; 5517 if (likely(cur > offset)) 5518 ring->ring[offset] = cur - offset; 5519 else 5520 ring->ring[offset] = (ring->ring_size>>2) - offset + cur; 5521 } 5522 5523 static void gfx_v9_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg, 5524 uint32_t reg_val_offs) 5525 { 5526 struct amdgpu_device *adev = ring->adev; 5527 5528 amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4)); 5529 amdgpu_ring_write(ring, 0 | /* src: register*/ 5530 (5 << 8) | /* dst: memory */ 5531 (1 << 20)); /* write confirm */ 5532 amdgpu_ring_write(ring, reg); 5533 amdgpu_ring_write(ring, 0); 5534 amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr + 5535 reg_val_offs * 4)); 5536 amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr + 5537 reg_val_offs * 4)); 5538 } 5539 5540 static void gfx_v9_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, 5541 uint32_t val) 5542 { 5543 uint32_t cmd = 0; 5544 5545 switch (ring->funcs->type) { 5546 case AMDGPU_RING_TYPE_GFX: 5547 cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM; 5548 break; 5549 case AMDGPU_RING_TYPE_KIQ: 5550 cmd = (1 << 16); /* no inc addr */ 5551 break; 5552 default: 5553 cmd = WR_CONFIRM; 5554 break; 5555 } 5556 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5557 amdgpu_ring_write(ring, cmd); 5558 amdgpu_ring_write(ring, reg); 5559 amdgpu_ring_write(ring, 0); 5560 amdgpu_ring_write(ring, val); 5561 } 5562 5563 static void gfx_v9_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg, 5564 uint32_t val, uint32_t mask) 5565 { 5566 gfx_v9_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20); 5567 } 5568 5569 static void gfx_v9_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring, 5570 uint32_t reg0, uint32_t reg1, 5571 uint32_t ref, uint32_t mask) 5572 { 5573 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 5574 struct amdgpu_device *adev = ring->adev; 5575 bool fw_version_ok = (ring->funcs->type == AMDGPU_RING_TYPE_GFX) ? 5576 adev->gfx.me_fw_write_wait : adev->gfx.mec_fw_write_wait; 5577 5578 if (fw_version_ok) 5579 gfx_v9_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1, 5580 ref, mask, 0x20); 5581 else 5582 amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1, 5583 ref, mask); 5584 } 5585 5586 static void gfx_v9_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid) 5587 { 5588 struct amdgpu_device *adev = ring->adev; 5589 uint32_t value = 0; 5590 5591 value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03); 5592 value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01); 5593 value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1); 5594 value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid); 5595 WREG32_SOC15(GC, 0, mmSQ_CMD, value); 5596 } 5597 5598 static void gfx_v9_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev, 5599 enum amdgpu_interrupt_state state) 5600 { 5601 switch (state) { 5602 case AMDGPU_IRQ_STATE_DISABLE: 5603 case AMDGPU_IRQ_STATE_ENABLE: 5604 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0, 5605 TIME_STAMP_INT_ENABLE, 5606 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); 5607 break; 5608 default: 5609 break; 5610 } 5611 } 5612 5613 static void gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev, 5614 int me, int pipe, 5615 enum amdgpu_interrupt_state state) 5616 { 5617 u32 mec_int_cntl, mec_int_cntl_reg; 5618 5619 /* 5620 * amdgpu controls only the first MEC. That's why this function only 5621 * handles the setting of interrupts for this specific MEC. All other 5622 * pipes' interrupts are set by amdkfd. 5623 */ 5624 5625 if (me == 1) { 5626 switch (pipe) { 5627 case 0: 5628 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL); 5629 break; 5630 case 1: 5631 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL); 5632 break; 5633 case 2: 5634 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL); 5635 break; 5636 case 3: 5637 mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL); 5638 break; 5639 default: 5640 DRM_DEBUG("invalid pipe %d\n", pipe); 5641 return; 5642 } 5643 } else { 5644 DRM_DEBUG("invalid me %d\n", me); 5645 return; 5646 } 5647 5648 switch (state) { 5649 case AMDGPU_IRQ_STATE_DISABLE: 5650 mec_int_cntl = RREG32_SOC15_IP(GC,mec_int_cntl_reg); 5651 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL, 5652 TIME_STAMP_INT_ENABLE, 0); 5653 WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl); 5654 break; 5655 case AMDGPU_IRQ_STATE_ENABLE: 5656 mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg); 5657 mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL, 5658 TIME_STAMP_INT_ENABLE, 1); 5659 WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl); 5660 break; 5661 default: 5662 break; 5663 } 5664 } 5665 5666 static int gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device *adev, 5667 struct amdgpu_irq_src *source, 5668 unsigned type, 5669 enum amdgpu_interrupt_state state) 5670 { 5671 switch (state) { 5672 case AMDGPU_IRQ_STATE_DISABLE: 5673 case AMDGPU_IRQ_STATE_ENABLE: 5674 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0, 5675 PRIV_REG_INT_ENABLE, 5676 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); 5677 break; 5678 default: 5679 break; 5680 } 5681 5682 return 0; 5683 } 5684 5685 static int gfx_v9_0_set_priv_inst_fault_state(struct amdgpu_device *adev, 5686 struct amdgpu_irq_src *source, 5687 unsigned type, 5688 enum amdgpu_interrupt_state state) 5689 { 5690 switch (state) { 5691 case AMDGPU_IRQ_STATE_DISABLE: 5692 case AMDGPU_IRQ_STATE_ENABLE: 5693 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0, 5694 PRIV_INSTR_INT_ENABLE, 5695 state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); 5696 break; 5697 default: 5698 break; 5699 } 5700 5701 return 0; 5702 } 5703 5704 #define ENABLE_ECC_ON_ME_PIPE(me, pipe) \ 5705 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\ 5706 CP_ECC_ERROR_INT_ENABLE, 1) 5707 5708 #define DISABLE_ECC_ON_ME_PIPE(me, pipe) \ 5709 WREG32_FIELD15(GC, 0, CP_ME##me##_PIPE##pipe##_INT_CNTL,\ 5710 CP_ECC_ERROR_INT_ENABLE, 0) 5711 5712 static int gfx_v9_0_set_cp_ecc_error_state(struct amdgpu_device *adev, 5713 struct amdgpu_irq_src *source, 5714 unsigned type, 5715 enum amdgpu_interrupt_state state) 5716 { 5717 switch (state) { 5718 case AMDGPU_IRQ_STATE_DISABLE: 5719 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0, 5720 CP_ECC_ERROR_INT_ENABLE, 0); 5721 DISABLE_ECC_ON_ME_PIPE(1, 0); 5722 DISABLE_ECC_ON_ME_PIPE(1, 1); 5723 DISABLE_ECC_ON_ME_PIPE(1, 2); 5724 DISABLE_ECC_ON_ME_PIPE(1, 3); 5725 break; 5726 5727 case AMDGPU_IRQ_STATE_ENABLE: 5728 WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0, 5729 CP_ECC_ERROR_INT_ENABLE, 1); 5730 ENABLE_ECC_ON_ME_PIPE(1, 0); 5731 ENABLE_ECC_ON_ME_PIPE(1, 1); 5732 ENABLE_ECC_ON_ME_PIPE(1, 2); 5733 ENABLE_ECC_ON_ME_PIPE(1, 3); 5734 break; 5735 default: 5736 break; 5737 } 5738 5739 return 0; 5740 } 5741 5742 5743 static int gfx_v9_0_set_eop_interrupt_state(struct amdgpu_device *adev, 5744 struct amdgpu_irq_src *src, 5745 unsigned type, 5746 enum amdgpu_interrupt_state state) 5747 { 5748 switch (type) { 5749 case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP: 5750 gfx_v9_0_set_gfx_eop_interrupt_state(adev, state); 5751 break; 5752 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP: 5753 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 0, state); 5754 break; 5755 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP: 5756 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 1, state); 5757 break; 5758 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP: 5759 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 2, state); 5760 break; 5761 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP: 5762 gfx_v9_0_set_compute_eop_interrupt_state(adev, 1, 3, state); 5763 break; 5764 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP: 5765 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 0, state); 5766 break; 5767 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP: 5768 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 1, state); 5769 break; 5770 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP: 5771 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 2, state); 5772 break; 5773 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP: 5774 gfx_v9_0_set_compute_eop_interrupt_state(adev, 2, 3, state); 5775 break; 5776 default: 5777 break; 5778 } 5779 return 0; 5780 } 5781 5782 static int gfx_v9_0_eop_irq(struct amdgpu_device *adev, 5783 struct amdgpu_irq_src *source, 5784 struct amdgpu_iv_entry *entry) 5785 { 5786 int i; 5787 u8 me_id, pipe_id, queue_id; 5788 struct amdgpu_ring *ring; 5789 5790 DRM_DEBUG("IH: CP EOP\n"); 5791 me_id = (entry->ring_id & 0x0c) >> 2; 5792 pipe_id = (entry->ring_id & 0x03) >> 0; 5793 queue_id = (entry->ring_id & 0x70) >> 4; 5794 5795 switch (me_id) { 5796 case 0: 5797 if (adev->gfx.num_gfx_rings && 5798 !amdgpu_mcbp_handle_trailing_fence_irq(&adev->gfx.muxer)) { 5799 /* Fence signals are handled on the software rings*/ 5800 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++) 5801 amdgpu_fence_process(&adev->gfx.sw_gfx_ring[i]); 5802 } 5803 break; 5804 case 1: 5805 case 2: 5806 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 5807 ring = &adev->gfx.compute_ring[i]; 5808 /* Per-queue interrupt is supported for MEC starting from VI. 5809 * The interrupt can only be enabled/disabled per pipe instead of per queue. 5810 */ 5811 if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id)) 5812 amdgpu_fence_process(ring); 5813 } 5814 break; 5815 } 5816 return 0; 5817 } 5818 5819 static void gfx_v9_0_fault(struct amdgpu_device *adev, 5820 struct amdgpu_iv_entry *entry) 5821 { 5822 u8 me_id, pipe_id, queue_id; 5823 struct amdgpu_ring *ring; 5824 int i; 5825 5826 me_id = (entry->ring_id & 0x0c) >> 2; 5827 pipe_id = (entry->ring_id & 0x03) >> 0; 5828 queue_id = (entry->ring_id & 0x70) >> 4; 5829 5830 switch (me_id) { 5831 case 0: 5832 drm_sched_fault(&adev->gfx.gfx_ring[0].sched); 5833 break; 5834 case 1: 5835 case 2: 5836 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 5837 ring = &adev->gfx.compute_ring[i]; 5838 if (ring->me == me_id && ring->pipe == pipe_id && 5839 ring->queue == queue_id) 5840 drm_sched_fault(&ring->sched); 5841 } 5842 break; 5843 } 5844 } 5845 5846 static int gfx_v9_0_priv_reg_irq(struct amdgpu_device *adev, 5847 struct amdgpu_irq_src *source, 5848 struct amdgpu_iv_entry *entry) 5849 { 5850 DRM_ERROR("Illegal register access in command stream\n"); 5851 gfx_v9_0_fault(adev, entry); 5852 return 0; 5853 } 5854 5855 static int gfx_v9_0_priv_inst_irq(struct amdgpu_device *adev, 5856 struct amdgpu_irq_src *source, 5857 struct amdgpu_iv_entry *entry) 5858 { 5859 DRM_ERROR("Illegal instruction in command stream\n"); 5860 gfx_v9_0_fault(adev, entry); 5861 return 0; 5862 } 5863 5864 5865 static const struct soc15_ras_field_entry gfx_v9_0_ras_fields[] = { 5866 { "CPC_SCRATCH", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_SCRATCH_CNT), 5867 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, SEC_COUNT), 5868 SOC15_REG_FIELD(CPC_EDC_SCRATCH_CNT, DED_COUNT) 5869 }, 5870 { "CPC_UCODE", SOC15_REG_ENTRY(GC, 0, mmCPC_EDC_UCODE_CNT), 5871 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, SEC_COUNT), 5872 SOC15_REG_FIELD(CPC_EDC_UCODE_CNT, DED_COUNT) 5873 }, 5874 { "CPF_ROQ_ME1", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 5875 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME1), 5876 0, 0 5877 }, 5878 { "CPF_ROQ_ME2", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_ROQ_CNT), 5879 SOC15_REG_FIELD(CPF_EDC_ROQ_CNT, COUNT_ME2), 5880 0, 0 5881 }, 5882 { "CPF_TAG", SOC15_REG_ENTRY(GC, 0, mmCPF_EDC_TAG_CNT), 5883 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, SEC_COUNT), 5884 SOC15_REG_FIELD(CPF_EDC_TAG_CNT, DED_COUNT) 5885 }, 5886 { "CPG_DMA_ROQ", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 5887 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, ROQ_COUNT), 5888 0, 0 5889 }, 5890 { "CPG_DMA_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_DMA_CNT), 5891 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_SEC_COUNT), 5892 SOC15_REG_FIELD(CPG_EDC_DMA_CNT, TAG_DED_COUNT) 5893 }, 5894 { "CPG_TAG", SOC15_REG_ENTRY(GC, 0, mmCPG_EDC_TAG_CNT), 5895 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, SEC_COUNT), 5896 SOC15_REG_FIELD(CPG_EDC_TAG_CNT, DED_COUNT) 5897 }, 5898 { "DC_CSINVOC", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_CSINVOC_CNT), 5899 SOC15_REG_FIELD(DC_EDC_CSINVOC_CNT, COUNT_ME1), 5900 0, 0 5901 }, 5902 { "DC_RESTORE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_RESTORE_CNT), 5903 SOC15_REG_FIELD(DC_EDC_RESTORE_CNT, COUNT_ME1), 5904 0, 0 5905 }, 5906 { "DC_STATE", SOC15_REG_ENTRY(GC, 0, mmDC_EDC_STATE_CNT), 5907 SOC15_REG_FIELD(DC_EDC_STATE_CNT, COUNT_ME1), 5908 0, 0 5909 }, 5910 { "GDS_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 5911 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_SEC), 5912 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_MEM_DED) 5913 }, 5914 { "GDS_INPUT_QUEUE", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_CNT), 5915 SOC15_REG_FIELD(GDS_EDC_CNT, GDS_INPUT_QUEUE_SED), 5916 0, 0 5917 }, 5918 { "GDS_ME0_CS_PIPE_MEM", SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 5919 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_SEC), 5920 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, ME0_CS_PIPE_MEM_DED) 5921 }, 5922 { "GDS_OA_PHY_PHY_CMD_RAM_MEM", 5923 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 5924 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_SEC), 5925 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_CMD_RAM_MEM_DED) 5926 }, 5927 { "GDS_OA_PHY_PHY_DATA_RAM_MEM", 5928 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PHY_CNT), 5929 SOC15_REG_FIELD(GDS_EDC_OA_PHY_CNT, PHY_DATA_RAM_MEM_SED), 5930 0, 0 5931 }, 5932 { "GDS_OA_PIPE_ME1_PIPE0_PIPE_MEM", 5933 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 5934 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_SEC), 5935 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE0_PIPE_MEM_DED) 5936 }, 5937 { "GDS_OA_PIPE_ME1_PIPE1_PIPE_MEM", 5938 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 5939 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_SEC), 5940 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE1_PIPE_MEM_DED) 5941 }, 5942 { "GDS_OA_PIPE_ME1_PIPE2_PIPE_MEM", 5943 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 5944 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_SEC), 5945 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE2_PIPE_MEM_DED) 5946 }, 5947 { "GDS_OA_PIPE_ME1_PIPE3_PIPE_MEM", 5948 SOC15_REG_ENTRY(GC, 0, mmGDS_EDC_OA_PIPE_CNT), 5949 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_SEC), 5950 SOC15_REG_FIELD(GDS_EDC_OA_PIPE_CNT, ME1_PIPE3_PIPE_MEM_DED) 5951 }, 5952 { "SPI_SR_MEM", SOC15_REG_ENTRY(GC, 0, mmSPI_EDC_CNT), 5953 SOC15_REG_FIELD(SPI_EDC_CNT, SPI_SR_MEM_SED_COUNT), 5954 0, 0 5955 }, 5956 { "TA_FS_DFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 5957 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_SEC_COUNT), 5958 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_DFIFO_DED_COUNT) 5959 }, 5960 { "TA_FS_AFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 5961 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_AFIFO_SED_COUNT), 5962 0, 0 5963 }, 5964 { "TA_FL_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 5965 SOC15_REG_FIELD(TA_EDC_CNT, TA_FL_LFIFO_SED_COUNT), 5966 0, 0 5967 }, 5968 { "TA_FX_LFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 5969 SOC15_REG_FIELD(TA_EDC_CNT, TA_FX_LFIFO_SED_COUNT), 5970 0, 0 5971 }, 5972 { "TA_FS_CFIFO", SOC15_REG_ENTRY(GC, 0, mmTA_EDC_CNT), 5973 SOC15_REG_FIELD(TA_EDC_CNT, TA_FS_CFIFO_SED_COUNT), 5974 0, 0 5975 }, 5976 { "TCA_HOLE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 5977 SOC15_REG_FIELD(TCA_EDC_CNT, HOLE_FIFO_SED_COUNT), 5978 0, 0 5979 }, 5980 { "TCA_REQ_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 5981 SOC15_REG_FIELD(TCA_EDC_CNT, REQ_FIFO_SED_COUNT), 5982 0, 0 5983 }, 5984 { "TCC_CACHE_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5985 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_SEC_COUNT), 5986 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DATA_DED_COUNT) 5987 }, 5988 { "TCC_CACHE_DIRTY", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5989 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_SEC_COUNT), 5990 SOC15_REG_FIELD(TCC_EDC_CNT, CACHE_DIRTY_DED_COUNT) 5991 }, 5992 { "TCC_HIGH_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5993 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_SEC_COUNT), 5994 SOC15_REG_FIELD(TCC_EDC_CNT, HIGH_RATE_TAG_DED_COUNT) 5995 }, 5996 { "TCC_LOW_RATE_TAG", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 5997 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_SEC_COUNT), 5998 SOC15_REG_FIELD(TCC_EDC_CNT, LOW_RATE_TAG_DED_COUNT) 5999 }, 6000 { "TCC_SRC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 6001 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_SEC_COUNT), 6002 SOC15_REG_FIELD(TCC_EDC_CNT, SRC_FIFO_DED_COUNT) 6003 }, 6004 { "TCC_IN_USE_DEC", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 6005 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_DEC_SED_COUNT), 6006 0, 0 6007 }, 6008 { "TCC_IN_USE_TRANSFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 6009 SOC15_REG_FIELD(TCC_EDC_CNT, IN_USE_TRANSFER_SED_COUNT), 6010 0, 0 6011 }, 6012 { "TCC_LATENCY_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 6013 SOC15_REG_FIELD(TCC_EDC_CNT, LATENCY_FIFO_SED_COUNT), 6014 0, 0 6015 }, 6016 { "TCC_RETURN_DATA", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 6017 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_DATA_SED_COUNT), 6018 0, 0 6019 }, 6020 { "TCC_RETURN_CONTROL", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 6021 SOC15_REG_FIELD(TCC_EDC_CNT, RETURN_CONTROL_SED_COUNT), 6022 0, 0 6023 }, 6024 { "TCC_UC_ATOMIC_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT), 6025 SOC15_REG_FIELD(TCC_EDC_CNT, UC_ATOMIC_FIFO_SED_COUNT), 6026 0, 0 6027 }, 6028 { "TCC_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 6029 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_RETURN_SED_COUNT), 6030 0, 0 6031 }, 6032 { "TCC_WRITE_CACHE_READ", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 6033 SOC15_REG_FIELD(TCC_EDC_CNT2, WRITE_CACHE_READ_SED_COUNT), 6034 0, 0 6035 }, 6036 { "TCC_SRC_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 6037 SOC15_REG_FIELD(TCC_EDC_CNT2, SRC_FIFO_NEXT_RAM_SED_COUNT), 6038 0, 0 6039 }, 6040 { "TCC_LATENCY_FIFO_NEXT_RAM", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 6041 SOC15_REG_FIELD(TCC_EDC_CNT2, LATENCY_FIFO_NEXT_RAM_SED_COUNT), 6042 0, 0 6043 }, 6044 { "TCC_CACHE_TAG_PROBE_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 6045 SOC15_REG_FIELD(TCC_EDC_CNT2, CACHE_TAG_PROBE_FIFO_SED_COUNT), 6046 0, 0 6047 }, 6048 { "TCC_WRRET_TAG_WRITE_RETURN", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 6049 SOC15_REG_FIELD(TCC_EDC_CNT2, WRRET_TAG_WRITE_RETURN_SED_COUNT), 6050 0, 0 6051 }, 6052 { "TCC_ATOMIC_RETURN_BUFFER", SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 6053 SOC15_REG_FIELD(TCC_EDC_CNT2, ATOMIC_RETURN_BUFFER_SED_COUNT), 6054 0, 0 6055 }, 6056 { "TCI_WRITE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCI_EDC_CNT), 6057 SOC15_REG_FIELD(TCI_EDC_CNT, WRITE_RAM_SED_COUNT), 6058 0, 0 6059 }, 6060 { "TCP_CACHE_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 6061 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_SEC_COUNT), 6062 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CACHE_RAM_DED_COUNT) 6063 }, 6064 { "TCP_LFIFO_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 6065 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_SEC_COUNT), 6066 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, LFIFO_RAM_DED_COUNT) 6067 }, 6068 { "TCP_CMD_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 6069 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, CMD_FIFO_SED_COUNT), 6070 0, 0 6071 }, 6072 { "TCP_VM_FIFO", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 6073 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, VM_FIFO_SEC_COUNT), 6074 0, 0 6075 }, 6076 { "TCP_DB_RAM", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 6077 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, DB_RAM_SED_COUNT), 6078 0, 0 6079 }, 6080 { "TCP_UTCL1_LFIFO0", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 6081 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_SEC_COUNT), 6082 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO0_DED_COUNT) 6083 }, 6084 { "TCP_UTCL1_LFIFO1", SOC15_REG_ENTRY(GC, 0, mmTCP_EDC_CNT_NEW), 6085 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_SEC_COUNT), 6086 SOC15_REG_FIELD(TCP_EDC_CNT_NEW, UTCL1_LFIFO1_DED_COUNT) 6087 }, 6088 { "TD_SS_FIFO_LO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 6089 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_SEC_COUNT), 6090 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_LO_DED_COUNT) 6091 }, 6092 { "TD_SS_FIFO_HI", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 6093 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_SEC_COUNT), 6094 SOC15_REG_FIELD(TD_EDC_CNT, SS_FIFO_HI_DED_COUNT) 6095 }, 6096 { "TD_CS_FIFO", SOC15_REG_ENTRY(GC, 0, mmTD_EDC_CNT), 6097 SOC15_REG_FIELD(TD_EDC_CNT, CS_FIFO_SED_COUNT), 6098 0, 0 6099 }, 6100 { "SQ_LDS_D", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6101 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_SEC_COUNT), 6102 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_D_DED_COUNT) 6103 }, 6104 { "SQ_LDS_I", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6105 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_SEC_COUNT), 6106 SOC15_REG_FIELD(SQ_EDC_CNT, LDS_I_DED_COUNT) 6107 }, 6108 { "SQ_SGPR", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6109 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_SEC_COUNT), 6110 SOC15_REG_FIELD(SQ_EDC_CNT, SGPR_DED_COUNT) 6111 }, 6112 { "SQ_VGPR0", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6113 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_SEC_COUNT), 6114 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR0_DED_COUNT) 6115 }, 6116 { "SQ_VGPR1", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6117 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_SEC_COUNT), 6118 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR1_DED_COUNT) 6119 }, 6120 { "SQ_VGPR2", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6121 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_SEC_COUNT), 6122 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR2_DED_COUNT) 6123 }, 6124 { "SQ_VGPR3", SOC15_REG_ENTRY(GC, 0, mmSQ_EDC_CNT), 6125 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_SEC_COUNT), 6126 SOC15_REG_FIELD(SQ_EDC_CNT, VGPR3_DED_COUNT) 6127 }, 6128 { "SQC_DATA_CU0_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6129 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_SEC_COUNT), 6130 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_WRITE_DATA_BUF_DED_COUNT) 6131 }, 6132 { "SQC_DATA_CU0_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6133 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_SEC_COUNT), 6134 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU0_UTCL1_LFIFO_DED_COUNT) 6135 }, 6136 { "SQC_DATA_CU1_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6137 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_SEC_COUNT), 6138 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_WRITE_DATA_BUF_DED_COUNT) 6139 }, 6140 { "SQC_DATA_CU1_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6141 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_SEC_COUNT), 6142 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU1_UTCL1_LFIFO_DED_COUNT) 6143 }, 6144 { "SQC_DATA_CU2_WRITE_DATA_BUF", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6145 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_SEC_COUNT), 6146 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_WRITE_DATA_BUF_DED_COUNT) 6147 }, 6148 { "SQC_DATA_CU2_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT), 6149 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_SEC_COUNT), 6150 SOC15_REG_FIELD(SQC_EDC_CNT, DATA_CU2_UTCL1_LFIFO_DED_COUNT) 6151 }, 6152 { "SQC_INST_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6153 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_SEC_COUNT), 6154 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_TAG_RAM_DED_COUNT) 6155 }, 6156 { "SQC_INST_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6157 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_SEC_COUNT), 6158 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_BANK_RAM_DED_COUNT) 6159 }, 6160 { "SQC_DATA_BANKA_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6161 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_SEC_COUNT), 6162 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_TAG_RAM_DED_COUNT) 6163 }, 6164 { "SQC_DATA_BANKA_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6165 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_SEC_COUNT), 6166 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_BANK_RAM_DED_COUNT) 6167 }, 6168 { "SQC_INST_BANKA_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6169 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_UTCL1_MISS_FIFO_SED_COUNT), 6170 0, 0 6171 }, 6172 { "SQC_INST_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6173 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_BANKA_MISS_FIFO_SED_COUNT), 6174 0, 0 6175 }, 6176 { "SQC_DATA_BANKA_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6177 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_HIT_FIFO_SED_COUNT), 6178 0, 0 6179 }, 6180 { "SQC_DATA_BANKA_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6181 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_MISS_FIFO_SED_COUNT), 6182 0, 0 6183 }, 6184 { "SQC_DATA_BANKA_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6185 SOC15_REG_FIELD(SQC_EDC_CNT2, DATA_BANKA_DIRTY_BIT_RAM_SED_COUNT), 6186 0, 0 6187 }, 6188 { "SQC_INST_UTCL1_LFIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT2), 6189 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_SEC_COUNT), 6190 SOC15_REG_FIELD(SQC_EDC_CNT2, INST_UTCL1_LFIFO_DED_COUNT) 6191 }, 6192 { "SQC_INST_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6193 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_SEC_COUNT), 6194 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_TAG_RAM_DED_COUNT) 6195 }, 6196 { "SQC_INST_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6197 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_SEC_COUNT), 6198 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_BANK_RAM_DED_COUNT) 6199 }, 6200 { "SQC_DATA_BANKB_TAG_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6201 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_SEC_COUNT), 6202 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_TAG_RAM_DED_COUNT) 6203 }, 6204 { "SQC_DATA_BANKB_BANK_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6205 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_SEC_COUNT), 6206 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_BANK_RAM_DED_COUNT) 6207 }, 6208 { "SQC_INST_BANKB_UTCL1_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6209 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_UTCL1_MISS_FIFO_SED_COUNT), 6210 0, 0 6211 }, 6212 { "SQC_INST_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6213 SOC15_REG_FIELD(SQC_EDC_CNT3, INST_BANKB_MISS_FIFO_SED_COUNT), 6214 0, 0 6215 }, 6216 { "SQC_DATA_BANKB_HIT_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6217 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_HIT_FIFO_SED_COUNT), 6218 0, 0 6219 }, 6220 { "SQC_DATA_BANKB_MISS_FIFO", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6221 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_MISS_FIFO_SED_COUNT), 6222 0, 0 6223 }, 6224 { "SQC_DATA_BANKB_DIRTY_BIT_RAM", SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 6225 SOC15_REG_FIELD(SQC_EDC_CNT3, DATA_BANKB_DIRTY_BIT_RAM_SED_COUNT), 6226 0, 0 6227 }, 6228 { "EA_DRAMRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6229 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_SEC_COUNT), 6230 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_CMDMEM_DED_COUNT) 6231 }, 6232 { "EA_DRAMWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6233 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_SEC_COUNT), 6234 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_CMDMEM_DED_COUNT) 6235 }, 6236 { "EA_DRAMWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6237 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_SEC_COUNT), 6238 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_DATAMEM_DED_COUNT) 6239 }, 6240 { "EA_RRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6241 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_SEC_COUNT), 6242 SOC15_REG_FIELD(GCEA_EDC_CNT, RRET_TAGMEM_DED_COUNT) 6243 }, 6244 { "EA_WRET_TAGMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6245 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_SEC_COUNT), 6246 SOC15_REG_FIELD(GCEA_EDC_CNT, WRET_TAGMEM_DED_COUNT) 6247 }, 6248 { "EA_DRAMRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6249 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMRD_PAGEMEM_SED_COUNT), 6250 0, 0 6251 }, 6252 { "EA_DRAMWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6253 SOC15_REG_FIELD(GCEA_EDC_CNT, DRAMWR_PAGEMEM_SED_COUNT), 6254 0, 0 6255 }, 6256 { "EA_IORD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6257 SOC15_REG_FIELD(GCEA_EDC_CNT, IORD_CMDMEM_SED_COUNT), 6258 0, 0 6259 }, 6260 { "EA_IOWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6261 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_CMDMEM_SED_COUNT), 6262 0, 0 6263 }, 6264 { "EA_IOWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT), 6265 SOC15_REG_FIELD(GCEA_EDC_CNT, IOWR_DATAMEM_SED_COUNT), 6266 0, 0 6267 }, 6268 { "GMIRD_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6269 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_SEC_COUNT), 6270 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_CMDMEM_DED_COUNT) 6271 }, 6272 { "GMIWR_CMDMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6273 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_SEC_COUNT), 6274 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_CMDMEM_DED_COUNT) 6275 }, 6276 { "GMIWR_DATAMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6277 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_SEC_COUNT), 6278 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_DATAMEM_DED_COUNT) 6279 }, 6280 { "GMIRD_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6281 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIRD_PAGEMEM_SED_COUNT), 6282 0, 0 6283 }, 6284 { "GMIWR_PAGEMEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6285 SOC15_REG_FIELD(GCEA_EDC_CNT2, GMIWR_PAGEMEM_SED_COUNT), 6286 0, 0 6287 }, 6288 { "MAM_D0MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6289 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D0MEM_SED_COUNT), 6290 0, 0 6291 }, 6292 { "MAM_D1MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6293 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D1MEM_SED_COUNT), 6294 0, 0 6295 }, 6296 { "MAM_D2MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6297 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D2MEM_SED_COUNT), 6298 0, 0 6299 }, 6300 { "MAM_D3MEM", SOC15_REG_ENTRY(GC, 0, mmGCEA_EDC_CNT2), 6301 SOC15_REG_FIELD(GCEA_EDC_CNT2, MAM_D3MEM_SED_COUNT), 6302 0, 0 6303 } 6304 }; 6305 6306 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev, 6307 void *inject_if) 6308 { 6309 struct ras_inject_if *info = (struct ras_inject_if *)inject_if; 6310 int ret; 6311 struct ta_ras_trigger_error_input block_info = { 0 }; 6312 6313 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) 6314 return -EINVAL; 6315 6316 if (info->head.sub_block_index >= ARRAY_SIZE(ras_gfx_subblocks)) 6317 return -EINVAL; 6318 6319 if (!ras_gfx_subblocks[info->head.sub_block_index].name) 6320 return -EPERM; 6321 6322 if (!(ras_gfx_subblocks[info->head.sub_block_index].hw_supported_error_type & 6323 info->head.type)) { 6324 DRM_ERROR("GFX Subblock %s, hardware do not support type 0x%x\n", 6325 ras_gfx_subblocks[info->head.sub_block_index].name, 6326 info->head.type); 6327 return -EPERM; 6328 } 6329 6330 if (!(ras_gfx_subblocks[info->head.sub_block_index].sw_supported_error_type & 6331 info->head.type)) { 6332 DRM_ERROR("GFX Subblock %s, driver do not support type 0x%x\n", 6333 ras_gfx_subblocks[info->head.sub_block_index].name, 6334 info->head.type); 6335 return -EPERM; 6336 } 6337 6338 block_info.block_id = amdgpu_ras_block_to_ta(info->head.block); 6339 block_info.sub_block_index = 6340 ras_gfx_subblocks[info->head.sub_block_index].ta_subblock; 6341 block_info.inject_error_type = amdgpu_ras_error_to_ta(info->head.type); 6342 block_info.address = info->address; 6343 block_info.value = info->value; 6344 6345 mutex_lock(&adev->grbm_idx_mutex); 6346 ret = psp_ras_trigger_error(&adev->psp, &block_info); 6347 mutex_unlock(&adev->grbm_idx_mutex); 6348 6349 return ret; 6350 } 6351 6352 static const char *vml2_mems[] = { 6353 "UTC_VML2_BANK_CACHE_0_BIGK_MEM0", 6354 "UTC_VML2_BANK_CACHE_0_BIGK_MEM1", 6355 "UTC_VML2_BANK_CACHE_0_4K_MEM0", 6356 "UTC_VML2_BANK_CACHE_0_4K_MEM1", 6357 "UTC_VML2_BANK_CACHE_1_BIGK_MEM0", 6358 "UTC_VML2_BANK_CACHE_1_BIGK_MEM1", 6359 "UTC_VML2_BANK_CACHE_1_4K_MEM0", 6360 "UTC_VML2_BANK_CACHE_1_4K_MEM1", 6361 "UTC_VML2_BANK_CACHE_2_BIGK_MEM0", 6362 "UTC_VML2_BANK_CACHE_2_BIGK_MEM1", 6363 "UTC_VML2_BANK_CACHE_2_4K_MEM0", 6364 "UTC_VML2_BANK_CACHE_2_4K_MEM1", 6365 "UTC_VML2_BANK_CACHE_3_BIGK_MEM0", 6366 "UTC_VML2_BANK_CACHE_3_BIGK_MEM1", 6367 "UTC_VML2_BANK_CACHE_3_4K_MEM0", 6368 "UTC_VML2_BANK_CACHE_3_4K_MEM1", 6369 }; 6370 6371 static const char *vml2_walker_mems[] = { 6372 "UTC_VML2_CACHE_PDE0_MEM0", 6373 "UTC_VML2_CACHE_PDE0_MEM1", 6374 "UTC_VML2_CACHE_PDE1_MEM0", 6375 "UTC_VML2_CACHE_PDE1_MEM1", 6376 "UTC_VML2_CACHE_PDE2_MEM0", 6377 "UTC_VML2_CACHE_PDE2_MEM1", 6378 "UTC_VML2_RDIF_LOG_FIFO", 6379 }; 6380 6381 static const char *atc_l2_cache_2m_mems[] = { 6382 "UTC_ATCL2_CACHE_2M_BANK0_WAY0_MEM", 6383 "UTC_ATCL2_CACHE_2M_BANK0_WAY1_MEM", 6384 "UTC_ATCL2_CACHE_2M_BANK1_WAY0_MEM", 6385 "UTC_ATCL2_CACHE_2M_BANK1_WAY1_MEM", 6386 }; 6387 6388 static const char *atc_l2_cache_4k_mems[] = { 6389 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM0", 6390 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM1", 6391 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM2", 6392 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM3", 6393 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM4", 6394 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM5", 6395 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM6", 6396 "UTC_ATCL2_CACHE_4K_BANK0_WAY0_MEM7", 6397 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM0", 6398 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM1", 6399 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM2", 6400 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM3", 6401 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM4", 6402 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM5", 6403 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM6", 6404 "UTC_ATCL2_CACHE_4K_BANK0_WAY1_MEM7", 6405 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM0", 6406 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM1", 6407 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM2", 6408 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM3", 6409 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM4", 6410 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM5", 6411 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM6", 6412 "UTC_ATCL2_CACHE_4K_BANK1_WAY0_MEM7", 6413 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM0", 6414 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM1", 6415 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM2", 6416 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM3", 6417 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM4", 6418 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM5", 6419 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM6", 6420 "UTC_ATCL2_CACHE_4K_BANK1_WAY1_MEM7", 6421 }; 6422 6423 static int gfx_v9_0_query_utc_edc_status(struct amdgpu_device *adev, 6424 struct ras_err_data *err_data) 6425 { 6426 uint32_t i, data; 6427 uint32_t sec_count, ded_count; 6428 6429 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255); 6430 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0); 6431 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255); 6432 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0); 6433 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255); 6434 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0); 6435 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255); 6436 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0); 6437 6438 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) { 6439 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i); 6440 data = RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT); 6441 6442 sec_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, SEC_COUNT); 6443 if (sec_count) { 6444 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6445 "SEC %d\n", i, vml2_mems[i], sec_count); 6446 err_data->ce_count += sec_count; 6447 } 6448 6449 ded_count = REG_GET_FIELD(data, VM_L2_MEM_ECC_CNT, DED_COUNT); 6450 if (ded_count) { 6451 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6452 "DED %d\n", i, vml2_mems[i], ded_count); 6453 err_data->ue_count += ded_count; 6454 } 6455 } 6456 6457 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) { 6458 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i); 6459 data = RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT); 6460 6461 sec_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT, 6462 SEC_COUNT); 6463 if (sec_count) { 6464 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6465 "SEC %d\n", i, vml2_walker_mems[i], sec_count); 6466 err_data->ce_count += sec_count; 6467 } 6468 6469 ded_count = REG_GET_FIELD(data, VM_L2_WALKER_MEM_ECC_CNT, 6470 DED_COUNT); 6471 if (ded_count) { 6472 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6473 "DED %d\n", i, vml2_walker_mems[i], ded_count); 6474 err_data->ue_count += ded_count; 6475 } 6476 } 6477 6478 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) { 6479 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i); 6480 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT); 6481 6482 sec_count = (data & 0x00006000L) >> 0xd; 6483 if (sec_count) { 6484 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6485 "SEC %d\n", i, atc_l2_cache_2m_mems[i], 6486 sec_count); 6487 err_data->ce_count += sec_count; 6488 } 6489 } 6490 6491 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) { 6492 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i); 6493 data = RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT); 6494 6495 sec_count = (data & 0x00006000L) >> 0xd; 6496 if (sec_count) { 6497 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6498 "SEC %d\n", i, atc_l2_cache_4k_mems[i], 6499 sec_count); 6500 err_data->ce_count += sec_count; 6501 } 6502 6503 ded_count = (data & 0x00018000L) >> 0xf; 6504 if (ded_count) { 6505 dev_info(adev->dev, "Instance[%d]: SubBlock %s, " 6506 "DED %d\n", i, atc_l2_cache_4k_mems[i], 6507 ded_count); 6508 err_data->ue_count += ded_count; 6509 } 6510 } 6511 6512 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255); 6513 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255); 6514 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255); 6515 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255); 6516 6517 return 0; 6518 } 6519 6520 static int gfx_v9_0_ras_error_count(struct amdgpu_device *adev, 6521 const struct soc15_reg_entry *reg, 6522 uint32_t se_id, uint32_t inst_id, uint32_t value, 6523 uint32_t *sec_count, uint32_t *ded_count) 6524 { 6525 uint32_t i; 6526 uint32_t sec_cnt, ded_cnt; 6527 6528 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_ras_fields); i++) { 6529 if(gfx_v9_0_ras_fields[i].reg_offset != reg->reg_offset || 6530 gfx_v9_0_ras_fields[i].seg != reg->seg || 6531 gfx_v9_0_ras_fields[i].inst != reg->inst) 6532 continue; 6533 6534 sec_cnt = (value & 6535 gfx_v9_0_ras_fields[i].sec_count_mask) >> 6536 gfx_v9_0_ras_fields[i].sec_count_shift; 6537 if (sec_cnt) { 6538 dev_info(adev->dev, "GFX SubBlock %s, " 6539 "Instance[%d][%d], SEC %d\n", 6540 gfx_v9_0_ras_fields[i].name, 6541 se_id, inst_id, 6542 sec_cnt); 6543 *sec_count += sec_cnt; 6544 } 6545 6546 ded_cnt = (value & 6547 gfx_v9_0_ras_fields[i].ded_count_mask) >> 6548 gfx_v9_0_ras_fields[i].ded_count_shift; 6549 if (ded_cnt) { 6550 dev_info(adev->dev, "GFX SubBlock %s, " 6551 "Instance[%d][%d], DED %d\n", 6552 gfx_v9_0_ras_fields[i].name, 6553 se_id, inst_id, 6554 ded_cnt); 6555 *ded_count += ded_cnt; 6556 } 6557 } 6558 6559 return 0; 6560 } 6561 6562 static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev) 6563 { 6564 int i, j, k; 6565 6566 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) 6567 return; 6568 6569 /* read back registers to clear the counters */ 6570 mutex_lock(&adev->grbm_idx_mutex); 6571 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) { 6572 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) { 6573 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) { 6574 amdgpu_gfx_select_se_sh(adev, j, 0x0, k); 6575 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i])); 6576 } 6577 } 6578 } 6579 WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, 0xe0000000); 6580 mutex_unlock(&adev->grbm_idx_mutex); 6581 6582 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255); 6583 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT, 0); 6584 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255); 6585 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT, 0); 6586 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255); 6587 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT, 0); 6588 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255); 6589 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT, 0); 6590 6591 for (i = 0; i < ARRAY_SIZE(vml2_mems); i++) { 6592 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, i); 6593 RREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_CNT); 6594 } 6595 6596 for (i = 0; i < ARRAY_SIZE(vml2_walker_mems); i++) { 6597 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, i); 6598 RREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_CNT); 6599 } 6600 6601 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_2m_mems); i++) { 6602 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, i); 6603 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_CNT); 6604 } 6605 6606 for (i = 0; i < ARRAY_SIZE(atc_l2_cache_4k_mems); i++) { 6607 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, i); 6608 RREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_CNT); 6609 } 6610 6611 WREG32_SOC15(GC, 0, mmVM_L2_MEM_ECC_INDEX, 255); 6612 WREG32_SOC15(GC, 0, mmVM_L2_WALKER_MEM_ECC_INDEX, 255); 6613 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_2M_EDC_INDEX, 255); 6614 WREG32_SOC15(GC, 0, mmATC_L2_CACHE_4K_EDC_INDEX, 255); 6615 } 6616 6617 static void gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev, 6618 void *ras_error_status) 6619 { 6620 struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status; 6621 uint32_t sec_count = 0, ded_count = 0; 6622 uint32_t i, j, k; 6623 uint32_t reg_value; 6624 6625 if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX)) 6626 return; 6627 6628 err_data->ue_count = 0; 6629 err_data->ce_count = 0; 6630 6631 mutex_lock(&adev->grbm_idx_mutex); 6632 6633 for (i = 0; i < ARRAY_SIZE(gfx_v9_0_edc_counter_regs); i++) { 6634 for (j = 0; j < gfx_v9_0_edc_counter_regs[i].se_num; j++) { 6635 for (k = 0; k < gfx_v9_0_edc_counter_regs[i].instance; k++) { 6636 amdgpu_gfx_select_se_sh(adev, j, 0, k); 6637 reg_value = 6638 RREG32(SOC15_REG_ENTRY_OFFSET(gfx_v9_0_edc_counter_regs[i])); 6639 if (reg_value) 6640 gfx_v9_0_ras_error_count(adev, 6641 &gfx_v9_0_edc_counter_regs[i], 6642 j, k, reg_value, 6643 &sec_count, &ded_count); 6644 } 6645 } 6646 } 6647 6648 err_data->ce_count += sec_count; 6649 err_data->ue_count += ded_count; 6650 6651 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 6652 mutex_unlock(&adev->grbm_idx_mutex); 6653 6654 gfx_v9_0_query_utc_edc_status(adev, err_data); 6655 } 6656 6657 static void gfx_v9_0_emit_mem_sync(struct amdgpu_ring *ring) 6658 { 6659 const unsigned int cp_coher_cntl = 6660 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_ICACHE_ACTION_ENA(1) | 6661 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_KCACHE_ACTION_ENA(1) | 6662 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_ACTION_ENA(1) | 6663 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TCL1_ACTION_ENA(1) | 6664 PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_WB_ACTION_ENA(1); 6665 6666 /* ACQUIRE_MEM -make one or more surfaces valid for use by the subsequent operations */ 6667 amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5)); 6668 amdgpu_ring_write(ring, cp_coher_cntl); /* CP_COHER_CNTL */ 6669 amdgpu_ring_write(ring, 0xffffffff); /* CP_COHER_SIZE */ 6670 amdgpu_ring_write(ring, 0xffffff); /* CP_COHER_SIZE_HI */ 6671 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */ 6672 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE_HI */ 6673 amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */ 6674 } 6675 6676 static void gfx_v9_0_emit_wave_limit_cs(struct amdgpu_ring *ring, 6677 uint32_t pipe, bool enable) 6678 { 6679 struct amdgpu_device *adev = ring->adev; 6680 uint32_t val; 6681 uint32_t wcl_cs_reg; 6682 6683 /* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */ 6684 val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS0_DEFAULT; 6685 6686 switch (pipe) { 6687 case 0: 6688 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS0); 6689 break; 6690 case 1: 6691 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS1); 6692 break; 6693 case 2: 6694 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS2); 6695 break; 6696 case 3: 6697 wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS3); 6698 break; 6699 default: 6700 DRM_DEBUG("invalid pipe %d\n", pipe); 6701 return; 6702 } 6703 6704 amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val); 6705 6706 } 6707 static void gfx_v9_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable) 6708 { 6709 struct amdgpu_device *adev = ring->adev; 6710 uint32_t val; 6711 int i; 6712 6713 6714 /* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit 6715 * number of gfx waves. Setting 5 bit will make sure gfx only gets 6716 * around 25% of gpu resources. 6717 */ 6718 val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT; 6719 amdgpu_ring_emit_wreg(ring, 6720 SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_GFX), 6721 val); 6722 6723 /* Restrict waves for normal/low priority compute queues as well 6724 * to get best QoS for high priority compute jobs. 6725 * 6726 * amdgpu controls only 1st ME(0-3 CS pipes). 6727 */ 6728 for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) { 6729 if (i != ring->pipe) 6730 gfx_v9_0_emit_wave_limit_cs(ring, i, enable); 6731 6732 } 6733 } 6734 6735 static const struct amd_ip_funcs gfx_v9_0_ip_funcs = { 6736 .name = "gfx_v9_0", 6737 .early_init = gfx_v9_0_early_init, 6738 .late_init = gfx_v9_0_late_init, 6739 .sw_init = gfx_v9_0_sw_init, 6740 .sw_fini = gfx_v9_0_sw_fini, 6741 .hw_init = gfx_v9_0_hw_init, 6742 .hw_fini = gfx_v9_0_hw_fini, 6743 .suspend = gfx_v9_0_suspend, 6744 .resume = gfx_v9_0_resume, 6745 .is_idle = gfx_v9_0_is_idle, 6746 .wait_for_idle = gfx_v9_0_wait_for_idle, 6747 .soft_reset = gfx_v9_0_soft_reset, 6748 .set_clockgating_state = gfx_v9_0_set_clockgating_state, 6749 .set_powergating_state = gfx_v9_0_set_powergating_state, 6750 .get_clockgating_state = gfx_v9_0_get_clockgating_state, 6751 }; 6752 6753 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_gfx = { 6754 .type = AMDGPU_RING_TYPE_GFX, 6755 .align_mask = 0xff, 6756 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 6757 .support_64bit_ptrs = true, 6758 .secure_submission_supported = true, 6759 .get_rptr = gfx_v9_0_ring_get_rptr_gfx, 6760 .get_wptr = gfx_v9_0_ring_get_wptr_gfx, 6761 .set_wptr = gfx_v9_0_ring_set_wptr_gfx, 6762 .emit_frame_size = /* totally 242 maximum if 16 IBs */ 6763 5 + /* COND_EXEC */ 6764 7 + /* PIPELINE_SYNC */ 6765 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 6766 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 6767 2 + /* VM_FLUSH */ 6768 8 + /* FENCE for VM_FLUSH */ 6769 20 + /* GDS switch */ 6770 4 + /* double SWITCH_BUFFER, 6771 the first COND_EXEC jump to the place just 6772 prior to this double SWITCH_BUFFER */ 6773 5 + /* COND_EXEC */ 6774 7 + /* HDP_flush */ 6775 4 + /* VGT_flush */ 6776 14 + /* CE_META */ 6777 31 + /* DE_META */ 6778 3 + /* CNTX_CTRL */ 6779 5 + /* HDP_INVL */ 6780 8 + 8 + /* FENCE x2 */ 6781 2 + /* SWITCH_BUFFER */ 6782 7, /* gfx_v9_0_emit_mem_sync */ 6783 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_gfx */ 6784 .emit_ib = gfx_v9_0_ring_emit_ib_gfx, 6785 .emit_fence = gfx_v9_0_ring_emit_fence, 6786 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync, 6787 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush, 6788 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch, 6789 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush, 6790 .test_ring = gfx_v9_0_ring_test_ring, 6791 .insert_nop = amdgpu_ring_insert_nop, 6792 .pad_ib = amdgpu_ring_generic_pad_ib, 6793 .emit_switch_buffer = gfx_v9_ring_emit_sb, 6794 .emit_cntxcntl = gfx_v9_ring_emit_cntxcntl, 6795 .init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec, 6796 .patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec, 6797 .preempt_ib = gfx_v9_0_ring_preempt_ib, 6798 .emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl, 6799 .emit_wreg = gfx_v9_0_ring_emit_wreg, 6800 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait, 6801 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait, 6802 .soft_recovery = gfx_v9_0_ring_soft_recovery, 6803 .emit_mem_sync = gfx_v9_0_emit_mem_sync, 6804 }; 6805 6806 static const struct amdgpu_ring_funcs gfx_v9_0_sw_ring_funcs_gfx = { 6807 .type = AMDGPU_RING_TYPE_GFX, 6808 .align_mask = 0xff, 6809 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 6810 .support_64bit_ptrs = true, 6811 .secure_submission_supported = true, 6812 .get_rptr = amdgpu_sw_ring_get_rptr_gfx, 6813 .get_wptr = amdgpu_sw_ring_get_wptr_gfx, 6814 .set_wptr = amdgpu_sw_ring_set_wptr_gfx, 6815 .emit_frame_size = /* totally 242 maximum if 16 IBs */ 6816 5 + /* COND_EXEC */ 6817 7 + /* PIPELINE_SYNC */ 6818 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 6819 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 6820 2 + /* VM_FLUSH */ 6821 8 + /* FENCE for VM_FLUSH */ 6822 20 + /* GDS switch */ 6823 4 + /* double SWITCH_BUFFER, 6824 * the first COND_EXEC jump to the place just 6825 * prior to this double SWITCH_BUFFER 6826 */ 6827 5 + /* COND_EXEC */ 6828 7 + /* HDP_flush */ 6829 4 + /* VGT_flush */ 6830 14 + /* CE_META */ 6831 31 + /* DE_META */ 6832 3 + /* CNTX_CTRL */ 6833 5 + /* HDP_INVL */ 6834 8 + 8 + /* FENCE x2 */ 6835 2 + /* SWITCH_BUFFER */ 6836 7, /* gfx_v9_0_emit_mem_sync */ 6837 .emit_ib_size = 4, /* gfx_v9_0_ring_emit_ib_gfx */ 6838 .emit_ib = gfx_v9_0_ring_emit_ib_gfx, 6839 .emit_fence = gfx_v9_0_ring_emit_fence, 6840 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync, 6841 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush, 6842 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch, 6843 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush, 6844 .test_ring = gfx_v9_0_ring_test_ring, 6845 .test_ib = gfx_v9_0_ring_test_ib, 6846 .insert_nop = amdgpu_sw_ring_insert_nop, 6847 .pad_ib = amdgpu_ring_generic_pad_ib, 6848 .emit_switch_buffer = gfx_v9_ring_emit_sb, 6849 .emit_cntxcntl = gfx_v9_ring_emit_cntxcntl, 6850 .init_cond_exec = gfx_v9_0_ring_emit_init_cond_exec, 6851 .patch_cond_exec = gfx_v9_0_ring_emit_patch_cond_exec, 6852 .emit_frame_cntl = gfx_v9_0_ring_emit_frame_cntl, 6853 .emit_wreg = gfx_v9_0_ring_emit_wreg, 6854 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait, 6855 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait, 6856 .soft_recovery = gfx_v9_0_ring_soft_recovery, 6857 .emit_mem_sync = gfx_v9_0_emit_mem_sync, 6858 }; 6859 6860 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = { 6861 .type = AMDGPU_RING_TYPE_COMPUTE, 6862 .align_mask = 0xff, 6863 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 6864 .support_64bit_ptrs = true, 6865 .get_rptr = gfx_v9_0_ring_get_rptr_compute, 6866 .get_wptr = gfx_v9_0_ring_get_wptr_compute, 6867 .set_wptr = gfx_v9_0_ring_set_wptr_compute, 6868 .emit_frame_size = 6869 20 + /* gfx_v9_0_ring_emit_gds_switch */ 6870 7 + /* gfx_v9_0_ring_emit_hdp_flush */ 6871 5 + /* hdp invalidate */ 6872 7 + /* gfx_v9_0_ring_emit_pipeline_sync */ 6873 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 6874 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 6875 2 + /* gfx_v9_0_ring_emit_vm_flush */ 6876 8 + 8 + 8 + /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */ 6877 7 + /* gfx_v9_0_emit_mem_sync */ 6878 5 + /* gfx_v9_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */ 6879 15, /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */ 6880 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */ 6881 .emit_ib = gfx_v9_0_ring_emit_ib_compute, 6882 .emit_fence = gfx_v9_0_ring_emit_fence, 6883 .emit_pipeline_sync = gfx_v9_0_ring_emit_pipeline_sync, 6884 .emit_vm_flush = gfx_v9_0_ring_emit_vm_flush, 6885 .emit_gds_switch = gfx_v9_0_ring_emit_gds_switch, 6886 .emit_hdp_flush = gfx_v9_0_ring_emit_hdp_flush, 6887 .test_ring = gfx_v9_0_ring_test_ring, 6888 .test_ib = gfx_v9_0_ring_test_ib, 6889 .insert_nop = amdgpu_ring_insert_nop, 6890 .pad_ib = amdgpu_ring_generic_pad_ib, 6891 .emit_wreg = gfx_v9_0_ring_emit_wreg, 6892 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait, 6893 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait, 6894 .emit_mem_sync = gfx_v9_0_emit_mem_sync, 6895 .emit_wave_limit = gfx_v9_0_emit_wave_limit, 6896 }; 6897 6898 static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = { 6899 .type = AMDGPU_RING_TYPE_KIQ, 6900 .align_mask = 0xff, 6901 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 6902 .support_64bit_ptrs = true, 6903 .get_rptr = gfx_v9_0_ring_get_rptr_compute, 6904 .get_wptr = gfx_v9_0_ring_get_wptr_compute, 6905 .set_wptr = gfx_v9_0_ring_set_wptr_compute, 6906 .emit_frame_size = 6907 20 + /* gfx_v9_0_ring_emit_gds_switch */ 6908 7 + /* gfx_v9_0_ring_emit_hdp_flush */ 6909 5 + /* hdp invalidate */ 6910 7 + /* gfx_v9_0_ring_emit_pipeline_sync */ 6911 SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 + 6912 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 6913 2 + /* gfx_v9_0_ring_emit_vm_flush */ 6914 8 + 8 + 8, /* gfx_v9_0_ring_emit_fence_kiq x3 for user fence, vm fence */ 6915 .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */ 6916 .emit_fence = gfx_v9_0_ring_emit_fence_kiq, 6917 .test_ring = gfx_v9_0_ring_test_ring, 6918 .insert_nop = amdgpu_ring_insert_nop, 6919 .pad_ib = amdgpu_ring_generic_pad_ib, 6920 .emit_rreg = gfx_v9_0_ring_emit_rreg, 6921 .emit_wreg = gfx_v9_0_ring_emit_wreg, 6922 .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait, 6923 .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait, 6924 }; 6925 6926 static void gfx_v9_0_set_ring_funcs(struct amdgpu_device *adev) 6927 { 6928 int i; 6929 6930 adev->gfx.kiq.ring.funcs = &gfx_v9_0_ring_funcs_kiq; 6931 6932 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 6933 adev->gfx.gfx_ring[i].funcs = &gfx_v9_0_ring_funcs_gfx; 6934 6935 if (adev->gfx.num_gfx_rings) { 6936 for (i = 0; i < GFX9_NUM_SW_GFX_RINGS; i++) 6937 adev->gfx.sw_gfx_ring[i].funcs = &gfx_v9_0_sw_ring_funcs_gfx; 6938 } 6939 6940 for (i = 0; i < adev->gfx.num_compute_rings; i++) 6941 adev->gfx.compute_ring[i].funcs = &gfx_v9_0_ring_funcs_compute; 6942 } 6943 6944 static const struct amdgpu_irq_src_funcs gfx_v9_0_eop_irq_funcs = { 6945 .set = gfx_v9_0_set_eop_interrupt_state, 6946 .process = gfx_v9_0_eop_irq, 6947 }; 6948 6949 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_reg_irq_funcs = { 6950 .set = gfx_v9_0_set_priv_reg_fault_state, 6951 .process = gfx_v9_0_priv_reg_irq, 6952 }; 6953 6954 static const struct amdgpu_irq_src_funcs gfx_v9_0_priv_inst_irq_funcs = { 6955 .set = gfx_v9_0_set_priv_inst_fault_state, 6956 .process = gfx_v9_0_priv_inst_irq, 6957 }; 6958 6959 static const struct amdgpu_irq_src_funcs gfx_v9_0_cp_ecc_error_irq_funcs = { 6960 .set = gfx_v9_0_set_cp_ecc_error_state, 6961 .process = amdgpu_gfx_cp_ecc_error_irq, 6962 }; 6963 6964 6965 static void gfx_v9_0_set_irq_funcs(struct amdgpu_device *adev) 6966 { 6967 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST; 6968 adev->gfx.eop_irq.funcs = &gfx_v9_0_eop_irq_funcs; 6969 6970 adev->gfx.priv_reg_irq.num_types = 1; 6971 adev->gfx.priv_reg_irq.funcs = &gfx_v9_0_priv_reg_irq_funcs; 6972 6973 adev->gfx.priv_inst_irq.num_types = 1; 6974 adev->gfx.priv_inst_irq.funcs = &gfx_v9_0_priv_inst_irq_funcs; 6975 6976 adev->gfx.cp_ecc_error_irq.num_types = 2; /*C5 ECC error and C9 FUE error*/ 6977 adev->gfx.cp_ecc_error_irq.funcs = &gfx_v9_0_cp_ecc_error_irq_funcs; 6978 } 6979 6980 static void gfx_v9_0_set_rlc_funcs(struct amdgpu_device *adev) 6981 { 6982 switch (adev->ip_versions[GC_HWIP][0]) { 6983 case IP_VERSION(9, 0, 1): 6984 case IP_VERSION(9, 2, 1): 6985 case IP_VERSION(9, 4, 0): 6986 case IP_VERSION(9, 2, 2): 6987 case IP_VERSION(9, 1, 0): 6988 case IP_VERSION(9, 4, 1): 6989 case IP_VERSION(9, 3, 0): 6990 case IP_VERSION(9, 4, 2): 6991 adev->gfx.rlc.funcs = &gfx_v9_0_rlc_funcs; 6992 break; 6993 default: 6994 break; 6995 } 6996 } 6997 6998 static void gfx_v9_0_set_gds_init(struct amdgpu_device *adev) 6999 { 7000 /* init asci gds info */ 7001 switch (adev->ip_versions[GC_HWIP][0]) { 7002 case IP_VERSION(9, 0, 1): 7003 case IP_VERSION(9, 2, 1): 7004 case IP_VERSION(9, 4, 0): 7005 adev->gds.gds_size = 0x10000; 7006 break; 7007 case IP_VERSION(9, 2, 2): 7008 case IP_VERSION(9, 1, 0): 7009 case IP_VERSION(9, 4, 1): 7010 adev->gds.gds_size = 0x1000; 7011 break; 7012 case IP_VERSION(9, 4, 2): 7013 /* aldebaran removed all the GDS internal memory, 7014 * only support GWS opcode in kernel, like barrier 7015 * semaphore.etc */ 7016 adev->gds.gds_size = 0; 7017 break; 7018 default: 7019 adev->gds.gds_size = 0x10000; 7020 break; 7021 } 7022 7023 switch (adev->ip_versions[GC_HWIP][0]) { 7024 case IP_VERSION(9, 0, 1): 7025 case IP_VERSION(9, 4, 0): 7026 adev->gds.gds_compute_max_wave_id = 0x7ff; 7027 break; 7028 case IP_VERSION(9, 2, 1): 7029 adev->gds.gds_compute_max_wave_id = 0x27f; 7030 break; 7031 case IP_VERSION(9, 2, 2): 7032 case IP_VERSION(9, 1, 0): 7033 if (adev->apu_flags & AMD_APU_IS_RAVEN2) 7034 adev->gds.gds_compute_max_wave_id = 0x77; /* raven2 */ 7035 else 7036 adev->gds.gds_compute_max_wave_id = 0x15f; /* raven1 */ 7037 break; 7038 case IP_VERSION(9, 4, 1): 7039 adev->gds.gds_compute_max_wave_id = 0xfff; 7040 break; 7041 case IP_VERSION(9, 4, 2): 7042 /* deprecated for Aldebaran, no usage at all */ 7043 adev->gds.gds_compute_max_wave_id = 0; 7044 break; 7045 default: 7046 /* this really depends on the chip */ 7047 adev->gds.gds_compute_max_wave_id = 0x7ff; 7048 break; 7049 } 7050 7051 adev->gds.gws_size = 64; 7052 adev->gds.oa_size = 16; 7053 } 7054 7055 static void gfx_v9_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev, 7056 u32 bitmap) 7057 { 7058 u32 data; 7059 7060 if (!bitmap) 7061 return; 7062 7063 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; 7064 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; 7065 7066 WREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG, data); 7067 } 7068 7069 static u32 gfx_v9_0_get_cu_active_bitmap(struct amdgpu_device *adev) 7070 { 7071 u32 data, mask; 7072 7073 data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG); 7074 data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG); 7075 7076 data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; 7077 data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; 7078 7079 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh); 7080 7081 return (~data) & mask; 7082 } 7083 7084 static int gfx_v9_0_get_cu_info(struct amdgpu_device *adev, 7085 struct amdgpu_cu_info *cu_info) 7086 { 7087 int i, j, k, counter, active_cu_number = 0; 7088 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0; 7089 unsigned disable_masks[4 * 4]; 7090 7091 if (!adev || !cu_info) 7092 return -EINVAL; 7093 7094 /* 7095 * 16 comes from bitmap array size 4*4, and it can cover all gfx9 ASICs 7096 */ 7097 if (adev->gfx.config.max_shader_engines * 7098 adev->gfx.config.max_sh_per_se > 16) 7099 return -EINVAL; 7100 7101 amdgpu_gfx_parse_disable_cu(disable_masks, 7102 adev->gfx.config.max_shader_engines, 7103 adev->gfx.config.max_sh_per_se); 7104 7105 mutex_lock(&adev->grbm_idx_mutex); 7106 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 7107 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 7108 mask = 1; 7109 ao_bitmap = 0; 7110 counter = 0; 7111 amdgpu_gfx_select_se_sh(adev, i, j, 0xffffffff); 7112 gfx_v9_0_set_user_cu_inactive_bitmap( 7113 adev, disable_masks[i * adev->gfx.config.max_sh_per_se + j]); 7114 bitmap = gfx_v9_0_get_cu_active_bitmap(adev); 7115 7116 /* 7117 * The bitmap(and ao_cu_bitmap) in cu_info structure is 7118 * 4x4 size array, and it's usually suitable for Vega 7119 * ASICs which has 4*2 SE/SH layout. 7120 * But for Arcturus, SE/SH layout is changed to 8*1. 7121 * To mostly reduce the impact, we make it compatible 7122 * with current bitmap array as below: 7123 * SE4,SH0 --> bitmap[0][1] 7124 * SE5,SH0 --> bitmap[1][1] 7125 * SE6,SH0 --> bitmap[2][1] 7126 * SE7,SH0 --> bitmap[3][1] 7127 */ 7128 cu_info->bitmap[i % 4][j + i / 4] = bitmap; 7129 7130 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) { 7131 if (bitmap & mask) { 7132 if (counter < adev->gfx.config.max_cu_per_sh) 7133 ao_bitmap |= mask; 7134 counter ++; 7135 } 7136 mask <<= 1; 7137 } 7138 active_cu_number += counter; 7139 if (i < 2 && j < 2) 7140 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8)); 7141 cu_info->ao_cu_bitmap[i % 4][j + i / 4] = ao_bitmap; 7142 } 7143 } 7144 amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 7145 mutex_unlock(&adev->grbm_idx_mutex); 7146 7147 cu_info->number = active_cu_number; 7148 cu_info->ao_cu_mask = ao_cu_mask; 7149 cu_info->simd_per_cu = NUM_SIMD_PER_CU; 7150 7151 return 0; 7152 } 7153 7154 const struct amdgpu_ip_block_version gfx_v9_0_ip_block = 7155 { 7156 .type = AMD_IP_BLOCK_TYPE_GFX, 7157 .major = 9, 7158 .minor = 0, 7159 .rev = 0, 7160 .funcs = &gfx_v9_0_ip_funcs, 7161 }; 7162