1 /* 2 * Copyright 2015 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 #include <linux/firmware.h> 24 #include "amdgpu.h" 25 #include "amdgpu_ih.h" 26 #include "amdgpu_gfx.h" 27 #include "amdgpu_ucode.h" 28 #include "clearstate_si.h" 29 #include "bif/bif_3_0_d.h" 30 #include "bif/bif_3_0_sh_mask.h" 31 #include "oss/oss_1_0_d.h" 32 #include "oss/oss_1_0_sh_mask.h" 33 #include "gca/gfx_6_0_d.h" 34 #include "gca/gfx_6_0_sh_mask.h" 35 #include "gmc/gmc_6_0_d.h" 36 #include "gmc/gmc_6_0_sh_mask.h" 37 #include "dce/dce_6_0_d.h" 38 #include "dce/dce_6_0_sh_mask.h" 39 #include "gca/gfx_7_2_enum.h" 40 #include "si_enums.h" 41 42 static void gfx_v6_0_set_ring_funcs(struct amdgpu_device *adev); 43 static void gfx_v6_0_set_irq_funcs(struct amdgpu_device *adev); 44 static void gfx_v6_0_get_cu_info(struct amdgpu_device *adev); 45 46 MODULE_FIRMWARE("radeon/tahiti_pfp.bin"); 47 MODULE_FIRMWARE("radeon/tahiti_me.bin"); 48 MODULE_FIRMWARE("radeon/tahiti_ce.bin"); 49 MODULE_FIRMWARE("radeon/tahiti_rlc.bin"); 50 51 MODULE_FIRMWARE("radeon/pitcairn_pfp.bin"); 52 MODULE_FIRMWARE("radeon/pitcairn_me.bin"); 53 MODULE_FIRMWARE("radeon/pitcairn_ce.bin"); 54 MODULE_FIRMWARE("radeon/pitcairn_rlc.bin"); 55 56 MODULE_FIRMWARE("radeon/verde_pfp.bin"); 57 MODULE_FIRMWARE("radeon/verde_me.bin"); 58 MODULE_FIRMWARE("radeon/verde_ce.bin"); 59 MODULE_FIRMWARE("radeon/verde_rlc.bin"); 60 61 MODULE_FIRMWARE("radeon/oland_pfp.bin"); 62 MODULE_FIRMWARE("radeon/oland_me.bin"); 63 MODULE_FIRMWARE("radeon/oland_ce.bin"); 64 MODULE_FIRMWARE("radeon/oland_rlc.bin"); 65 66 MODULE_FIRMWARE("radeon/hainan_pfp.bin"); 67 MODULE_FIRMWARE("radeon/hainan_me.bin"); 68 MODULE_FIRMWARE("radeon/hainan_ce.bin"); 69 MODULE_FIRMWARE("radeon/hainan_rlc.bin"); 70 71 static u32 gfx_v6_0_get_csb_size(struct amdgpu_device *adev); 72 static void gfx_v6_0_get_csb_buffer(struct amdgpu_device *adev, volatile u32 *buffer); 73 //static void gfx_v6_0_init_cp_pg_table(struct amdgpu_device *adev); 74 static void gfx_v6_0_init_pg(struct amdgpu_device *adev); 75 76 #define ARRAY_MODE(x) ((x) << GB_TILE_MODE0__ARRAY_MODE__SHIFT) 77 #define PIPE_CONFIG(x) ((x) << GB_TILE_MODE0__PIPE_CONFIG__SHIFT) 78 #define TILE_SPLIT(x) ((x) << GB_TILE_MODE0__TILE_SPLIT__SHIFT) 79 #define MICRO_TILE_MODE(x) ((x) << 0) 80 #define SAMPLE_SPLIT(x) ((x) << GB_TILE_MODE0__SAMPLE_SPLIT__SHIFT) 81 #define BANK_WIDTH(x) ((x) << 14) 82 #define BANK_HEIGHT(x) ((x) << 16) 83 #define MACRO_TILE_ASPECT(x) ((x) << 18) 84 #define NUM_BANKS(x) ((x) << 20) 85 86 static const u32 verde_rlc_save_restore_register_list[] = 87 { 88 (0x8000 << 16) | (0x98f4 >> 2), 89 0x00000000, 90 (0x8040 << 16) | (0x98f4 >> 2), 91 0x00000000, 92 (0x8000 << 16) | (0xe80 >> 2), 93 0x00000000, 94 (0x8040 << 16) | (0xe80 >> 2), 95 0x00000000, 96 (0x8000 << 16) | (0x89bc >> 2), 97 0x00000000, 98 (0x8040 << 16) | (0x89bc >> 2), 99 0x00000000, 100 (0x8000 << 16) | (0x8c1c >> 2), 101 0x00000000, 102 (0x8040 << 16) | (0x8c1c >> 2), 103 0x00000000, 104 (0x9c00 << 16) | (0x98f0 >> 2), 105 0x00000000, 106 (0x9c00 << 16) | (0xe7c >> 2), 107 0x00000000, 108 (0x8000 << 16) | (0x9148 >> 2), 109 0x00000000, 110 (0x8040 << 16) | (0x9148 >> 2), 111 0x00000000, 112 (0x9c00 << 16) | (0x9150 >> 2), 113 0x00000000, 114 (0x9c00 << 16) | (0x897c >> 2), 115 0x00000000, 116 (0x9c00 << 16) | (0x8d8c >> 2), 117 0x00000000, 118 (0x9c00 << 16) | (0xac54 >> 2), 119 0X00000000, 120 0x3, 121 (0x9c00 << 16) | (0x98f8 >> 2), 122 0x00000000, 123 (0x9c00 << 16) | (0x9910 >> 2), 124 0x00000000, 125 (0x9c00 << 16) | (0x9914 >> 2), 126 0x00000000, 127 (0x9c00 << 16) | (0x9918 >> 2), 128 0x00000000, 129 (0x9c00 << 16) | (0x991c >> 2), 130 0x00000000, 131 (0x9c00 << 16) | (0x9920 >> 2), 132 0x00000000, 133 (0x9c00 << 16) | (0x9924 >> 2), 134 0x00000000, 135 (0x9c00 << 16) | (0x9928 >> 2), 136 0x00000000, 137 (0x9c00 << 16) | (0x992c >> 2), 138 0x00000000, 139 (0x9c00 << 16) | (0x9930 >> 2), 140 0x00000000, 141 (0x9c00 << 16) | (0x9934 >> 2), 142 0x00000000, 143 (0x9c00 << 16) | (0x9938 >> 2), 144 0x00000000, 145 (0x9c00 << 16) | (0x993c >> 2), 146 0x00000000, 147 (0x9c00 << 16) | (0x9940 >> 2), 148 0x00000000, 149 (0x9c00 << 16) | (0x9944 >> 2), 150 0x00000000, 151 (0x9c00 << 16) | (0x9948 >> 2), 152 0x00000000, 153 (0x9c00 << 16) | (0x994c >> 2), 154 0x00000000, 155 (0x9c00 << 16) | (0x9950 >> 2), 156 0x00000000, 157 (0x9c00 << 16) | (0x9954 >> 2), 158 0x00000000, 159 (0x9c00 << 16) | (0x9958 >> 2), 160 0x00000000, 161 (0x9c00 << 16) | (0x995c >> 2), 162 0x00000000, 163 (0x9c00 << 16) | (0x9960 >> 2), 164 0x00000000, 165 (0x9c00 << 16) | (0x9964 >> 2), 166 0x00000000, 167 (0x9c00 << 16) | (0x9968 >> 2), 168 0x00000000, 169 (0x9c00 << 16) | (0x996c >> 2), 170 0x00000000, 171 (0x9c00 << 16) | (0x9970 >> 2), 172 0x00000000, 173 (0x9c00 << 16) | (0x9974 >> 2), 174 0x00000000, 175 (0x9c00 << 16) | (0x9978 >> 2), 176 0x00000000, 177 (0x9c00 << 16) | (0x997c >> 2), 178 0x00000000, 179 (0x9c00 << 16) | (0x9980 >> 2), 180 0x00000000, 181 (0x9c00 << 16) | (0x9984 >> 2), 182 0x00000000, 183 (0x9c00 << 16) | (0x9988 >> 2), 184 0x00000000, 185 (0x9c00 << 16) | (0x998c >> 2), 186 0x00000000, 187 (0x9c00 << 16) | (0x8c00 >> 2), 188 0x00000000, 189 (0x9c00 << 16) | (0x8c14 >> 2), 190 0x00000000, 191 (0x9c00 << 16) | (0x8c04 >> 2), 192 0x00000000, 193 (0x9c00 << 16) | (0x8c08 >> 2), 194 0x00000000, 195 (0x8000 << 16) | (0x9b7c >> 2), 196 0x00000000, 197 (0x8040 << 16) | (0x9b7c >> 2), 198 0x00000000, 199 (0x8000 << 16) | (0xe84 >> 2), 200 0x00000000, 201 (0x8040 << 16) | (0xe84 >> 2), 202 0x00000000, 203 (0x8000 << 16) | (0x89c0 >> 2), 204 0x00000000, 205 (0x8040 << 16) | (0x89c0 >> 2), 206 0x00000000, 207 (0x8000 << 16) | (0x914c >> 2), 208 0x00000000, 209 (0x8040 << 16) | (0x914c >> 2), 210 0x00000000, 211 (0x8000 << 16) | (0x8c20 >> 2), 212 0x00000000, 213 (0x8040 << 16) | (0x8c20 >> 2), 214 0x00000000, 215 (0x8000 << 16) | (0x9354 >> 2), 216 0x00000000, 217 (0x8040 << 16) | (0x9354 >> 2), 218 0x00000000, 219 (0x9c00 << 16) | (0x9060 >> 2), 220 0x00000000, 221 (0x9c00 << 16) | (0x9364 >> 2), 222 0x00000000, 223 (0x9c00 << 16) | (0x9100 >> 2), 224 0x00000000, 225 (0x9c00 << 16) | (0x913c >> 2), 226 0x00000000, 227 (0x8000 << 16) | (0x90e0 >> 2), 228 0x00000000, 229 (0x8000 << 16) | (0x90e4 >> 2), 230 0x00000000, 231 (0x8000 << 16) | (0x90e8 >> 2), 232 0x00000000, 233 (0x8040 << 16) | (0x90e0 >> 2), 234 0x00000000, 235 (0x8040 << 16) | (0x90e4 >> 2), 236 0x00000000, 237 (0x8040 << 16) | (0x90e8 >> 2), 238 0x00000000, 239 (0x9c00 << 16) | (0x8bcc >> 2), 240 0x00000000, 241 (0x9c00 << 16) | (0x8b24 >> 2), 242 0x00000000, 243 (0x9c00 << 16) | (0x88c4 >> 2), 244 0x00000000, 245 (0x9c00 << 16) | (0x8e50 >> 2), 246 0x00000000, 247 (0x9c00 << 16) | (0x8c0c >> 2), 248 0x00000000, 249 (0x9c00 << 16) | (0x8e58 >> 2), 250 0x00000000, 251 (0x9c00 << 16) | (0x8e5c >> 2), 252 0x00000000, 253 (0x9c00 << 16) | (0x9508 >> 2), 254 0x00000000, 255 (0x9c00 << 16) | (0x950c >> 2), 256 0x00000000, 257 (0x9c00 << 16) | (0x9494 >> 2), 258 0x00000000, 259 (0x9c00 << 16) | (0xac0c >> 2), 260 0x00000000, 261 (0x9c00 << 16) | (0xac10 >> 2), 262 0x00000000, 263 (0x9c00 << 16) | (0xac14 >> 2), 264 0x00000000, 265 (0x9c00 << 16) | (0xae00 >> 2), 266 0x00000000, 267 (0x9c00 << 16) | (0xac08 >> 2), 268 0x00000000, 269 (0x9c00 << 16) | (0x88d4 >> 2), 270 0x00000000, 271 (0x9c00 << 16) | (0x88c8 >> 2), 272 0x00000000, 273 (0x9c00 << 16) | (0x88cc >> 2), 274 0x00000000, 275 (0x9c00 << 16) | (0x89b0 >> 2), 276 0x00000000, 277 (0x9c00 << 16) | (0x8b10 >> 2), 278 0x00000000, 279 (0x9c00 << 16) | (0x8a14 >> 2), 280 0x00000000, 281 (0x9c00 << 16) | (0x9830 >> 2), 282 0x00000000, 283 (0x9c00 << 16) | (0x9834 >> 2), 284 0x00000000, 285 (0x9c00 << 16) | (0x9838 >> 2), 286 0x00000000, 287 (0x9c00 << 16) | (0x9a10 >> 2), 288 0x00000000, 289 (0x8000 << 16) | (0x9870 >> 2), 290 0x00000000, 291 (0x8000 << 16) | (0x9874 >> 2), 292 0x00000000, 293 (0x8001 << 16) | (0x9870 >> 2), 294 0x00000000, 295 (0x8001 << 16) | (0x9874 >> 2), 296 0x00000000, 297 (0x8040 << 16) | (0x9870 >> 2), 298 0x00000000, 299 (0x8040 << 16) | (0x9874 >> 2), 300 0x00000000, 301 (0x8041 << 16) | (0x9870 >> 2), 302 0x00000000, 303 (0x8041 << 16) | (0x9874 >> 2), 304 0x00000000, 305 0x00000000 306 }; 307 308 static int gfx_v6_0_init_microcode(struct amdgpu_device *adev) 309 { 310 const char *chip_name; 311 char fw_name[30]; 312 int err; 313 const struct gfx_firmware_header_v1_0 *cp_hdr; 314 const struct rlc_firmware_header_v1_0 *rlc_hdr; 315 316 DRM_DEBUG("\n"); 317 318 switch (adev->asic_type) { 319 case CHIP_TAHITI: 320 chip_name = "tahiti"; 321 break; 322 case CHIP_PITCAIRN: 323 chip_name = "pitcairn"; 324 break; 325 case CHIP_VERDE: 326 chip_name = "verde"; 327 break; 328 case CHIP_OLAND: 329 chip_name = "oland"; 330 break; 331 case CHIP_HAINAN: 332 chip_name = "hainan"; 333 break; 334 default: BUG(); 335 } 336 337 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name); 338 err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev); 339 if (err) 340 goto out; 341 err = amdgpu_ucode_validate(adev->gfx.pfp_fw); 342 if (err) 343 goto out; 344 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; 345 adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); 346 adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); 347 348 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name); 349 err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev); 350 if (err) 351 goto out; 352 err = amdgpu_ucode_validate(adev->gfx.me_fw); 353 if (err) 354 goto out; 355 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; 356 adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); 357 adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); 358 359 snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name); 360 err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev); 361 if (err) 362 goto out; 363 err = amdgpu_ucode_validate(adev->gfx.ce_fw); 364 if (err) 365 goto out; 366 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; 367 adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); 368 adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); 369 370 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", chip_name); 371 err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev); 372 if (err) 373 goto out; 374 err = amdgpu_ucode_validate(adev->gfx.rlc_fw); 375 rlc_hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data; 376 adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version); 377 adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version); 378 379 out: 380 if (err) { 381 pr_err("gfx6: Failed to load firmware \"%s\"\n", fw_name); 382 release_firmware(adev->gfx.pfp_fw); 383 adev->gfx.pfp_fw = NULL; 384 release_firmware(adev->gfx.me_fw); 385 adev->gfx.me_fw = NULL; 386 release_firmware(adev->gfx.ce_fw); 387 adev->gfx.ce_fw = NULL; 388 release_firmware(adev->gfx.rlc_fw); 389 adev->gfx.rlc_fw = NULL; 390 } 391 return err; 392 } 393 394 static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev) 395 { 396 const u32 num_tile_mode_states = ARRAY_SIZE(adev->gfx.config.tile_mode_array); 397 u32 reg_offset, split_equal_to_row_size, *tilemode; 398 399 memset(adev->gfx.config.tile_mode_array, 0, sizeof(adev->gfx.config.tile_mode_array)); 400 tilemode = adev->gfx.config.tile_mode_array; 401 402 switch (adev->gfx.config.mem_row_size_in_kb) { 403 case 1: 404 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB; 405 break; 406 case 2: 407 default: 408 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB; 409 break; 410 case 4: 411 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB; 412 break; 413 } 414 415 if (adev->asic_type == CHIP_VERDE) { 416 tilemode[0] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 417 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 418 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 419 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 420 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 421 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 422 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 423 NUM_BANKS(ADDR_SURF_16_BANK); 424 tilemode[1] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 425 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 426 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 427 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 428 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 429 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 430 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 431 NUM_BANKS(ADDR_SURF_16_BANK); 432 tilemode[2] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 433 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 434 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 435 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 436 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 437 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 438 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 439 NUM_BANKS(ADDR_SURF_16_BANK); 440 tilemode[3] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 441 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 442 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 443 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 444 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 445 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 446 NUM_BANKS(ADDR_SURF_8_BANK) | 447 TILE_SPLIT(split_equal_to_row_size); 448 tilemode[4] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 449 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 450 PIPE_CONFIG(ADDR_SURF_P4_8x16); 451 tilemode[5] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 452 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 453 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 454 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 455 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 456 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 457 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 458 NUM_BANKS(ADDR_SURF_4_BANK); 459 tilemode[6] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 460 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 461 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 462 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 463 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 464 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 465 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 466 NUM_BANKS(ADDR_SURF_4_BANK); 467 tilemode[7] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 468 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 469 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 470 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 471 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 472 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 473 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 474 NUM_BANKS(ADDR_SURF_2_BANK); 475 tilemode[8] = ARRAY_MODE(ARRAY_LINEAR_ALIGNED); 476 tilemode[9] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 477 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 478 PIPE_CONFIG(ADDR_SURF_P4_8x16); 479 tilemode[10] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 480 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 481 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 482 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 483 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 484 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 485 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 486 NUM_BANKS(ADDR_SURF_16_BANK); 487 tilemode[11] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 488 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 489 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 490 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 491 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 492 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 493 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 494 NUM_BANKS(ADDR_SURF_16_BANK); 495 tilemode[12] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 496 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 497 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 498 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 499 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 500 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 501 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 502 NUM_BANKS(ADDR_SURF_16_BANK); 503 tilemode[13] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 504 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 505 PIPE_CONFIG(ADDR_SURF_P4_8x16); 506 tilemode[14] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 507 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 508 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 509 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 510 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 511 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 512 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 513 NUM_BANKS(ADDR_SURF_16_BANK); 514 tilemode[15] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 515 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 516 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 517 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 518 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 519 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 520 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 521 NUM_BANKS(ADDR_SURF_16_BANK); 522 tilemode[16] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 523 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 524 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 525 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 526 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 527 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 528 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 529 NUM_BANKS(ADDR_SURF_16_BANK); 530 tilemode[17] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 531 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 532 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 533 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 534 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 535 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 536 NUM_BANKS(ADDR_SURF_16_BANK) | 537 TILE_SPLIT(split_equal_to_row_size); 538 tilemode[18] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 539 ARRAY_MODE(ARRAY_1D_TILED_THICK) | 540 PIPE_CONFIG(ADDR_SURF_P4_8x16); 541 tilemode[19] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 542 ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 543 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 544 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 545 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 546 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 547 NUM_BANKS(ADDR_SURF_16_BANK) | 548 TILE_SPLIT(split_equal_to_row_size); 549 tilemode[20] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 550 ARRAY_MODE(ARRAY_2D_TILED_THICK) | 551 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 552 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 553 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 554 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 555 NUM_BANKS(ADDR_SURF_16_BANK) | 556 TILE_SPLIT(split_equal_to_row_size); 557 tilemode[21] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 558 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 559 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 560 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 561 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 562 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 563 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 564 NUM_BANKS(ADDR_SURF_8_BANK); 565 tilemode[22] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 566 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 567 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 568 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 569 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 570 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 571 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 572 NUM_BANKS(ADDR_SURF_8_BANK); 573 tilemode[23] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 574 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 575 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 576 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 577 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 578 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 579 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 580 NUM_BANKS(ADDR_SURF_4_BANK); 581 tilemode[24] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 582 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 583 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 584 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 585 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 586 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 587 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 588 NUM_BANKS(ADDR_SURF_4_BANK); 589 tilemode[25] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 590 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 591 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 592 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 593 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 594 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 595 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 596 NUM_BANKS(ADDR_SURF_2_BANK); 597 tilemode[26] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 598 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 599 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 600 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 601 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 602 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 603 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 604 NUM_BANKS(ADDR_SURF_2_BANK); 605 tilemode[27] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 606 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 607 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 608 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 609 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 610 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 611 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 612 NUM_BANKS(ADDR_SURF_2_BANK); 613 tilemode[28] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 614 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 615 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 616 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 617 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 618 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 619 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 620 NUM_BANKS(ADDR_SURF_2_BANK); 621 tilemode[29] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 622 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 623 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 624 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 625 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 626 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 627 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 628 NUM_BANKS(ADDR_SURF_2_BANK); 629 tilemode[30] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 630 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 631 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 632 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 633 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 634 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 635 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 636 NUM_BANKS(ADDR_SURF_2_BANK); 637 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 638 WREG32(mmGB_TILE_MODE0 + reg_offset, tilemode[reg_offset]); 639 } else if (adev->asic_type == CHIP_OLAND) { 640 tilemode[0] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 641 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 642 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 643 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 644 NUM_BANKS(ADDR_SURF_16_BANK) | 645 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 646 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 647 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 648 tilemode[1] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 649 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 650 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 651 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 652 NUM_BANKS(ADDR_SURF_16_BANK) | 653 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 654 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 655 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 656 tilemode[2] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 657 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 658 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 659 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 660 NUM_BANKS(ADDR_SURF_16_BANK) | 661 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 662 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 663 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 664 tilemode[3] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 665 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 666 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 667 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 668 NUM_BANKS(ADDR_SURF_16_BANK) | 669 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 670 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 671 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 672 tilemode[4] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 673 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 674 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 675 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 676 NUM_BANKS(ADDR_SURF_16_BANK) | 677 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 678 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 679 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 680 tilemode[5] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 681 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 682 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 683 TILE_SPLIT(split_equal_to_row_size) | 684 NUM_BANKS(ADDR_SURF_16_BANK) | 685 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 686 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 687 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 688 tilemode[6] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 689 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 690 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 691 TILE_SPLIT(split_equal_to_row_size) | 692 NUM_BANKS(ADDR_SURF_16_BANK) | 693 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 694 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 695 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 696 tilemode[7] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 697 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 698 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 699 TILE_SPLIT(split_equal_to_row_size) | 700 NUM_BANKS(ADDR_SURF_16_BANK) | 701 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 702 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 703 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 704 tilemode[8] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 705 ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 706 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 707 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 708 NUM_BANKS(ADDR_SURF_16_BANK) | 709 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 710 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 711 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 712 tilemode[9] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 713 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 714 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 715 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 716 NUM_BANKS(ADDR_SURF_16_BANK) | 717 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 718 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 719 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 720 tilemode[10] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 721 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 722 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 723 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 724 NUM_BANKS(ADDR_SURF_16_BANK) | 725 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 726 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 727 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 728 tilemode[11] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 729 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 730 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 731 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 732 NUM_BANKS(ADDR_SURF_16_BANK) | 733 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 734 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 735 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 736 tilemode[12] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 737 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 738 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 739 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 740 NUM_BANKS(ADDR_SURF_16_BANK) | 741 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 742 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 743 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 744 tilemode[13] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 745 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 746 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 747 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 748 NUM_BANKS(ADDR_SURF_16_BANK) | 749 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 750 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 751 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 752 tilemode[14] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 753 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 754 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 755 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 756 NUM_BANKS(ADDR_SURF_16_BANK) | 757 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 758 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 759 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 760 tilemode[15] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 761 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 762 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 763 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 764 NUM_BANKS(ADDR_SURF_16_BANK) | 765 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 766 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 767 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 768 tilemode[16] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 769 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 770 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 771 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 772 NUM_BANKS(ADDR_SURF_16_BANK) | 773 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 774 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 775 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 776 tilemode[17] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 777 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 778 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 779 TILE_SPLIT(split_equal_to_row_size) | 780 NUM_BANKS(ADDR_SURF_16_BANK) | 781 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 782 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 783 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 784 tilemode[21] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 785 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 786 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 787 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 788 NUM_BANKS(ADDR_SURF_16_BANK) | 789 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 790 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 791 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 792 tilemode[22] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 793 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 794 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 795 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 796 NUM_BANKS(ADDR_SURF_16_BANK) | 797 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 798 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 799 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 800 tilemode[23] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 801 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 802 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 803 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 804 NUM_BANKS(ADDR_SURF_16_BANK) | 805 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 806 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 807 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 808 tilemode[24] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 809 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 810 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 811 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 812 NUM_BANKS(ADDR_SURF_16_BANK) | 813 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 814 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 815 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 816 tilemode[25] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 817 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 818 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 819 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 820 NUM_BANKS(ADDR_SURF_8_BANK) | 821 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 822 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 823 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1); 824 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 825 WREG32(mmGB_TILE_MODE0 + reg_offset, tilemode[reg_offset]); 826 } else if (adev->asic_type == CHIP_HAINAN) { 827 tilemode[0] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 828 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 829 PIPE_CONFIG(ADDR_SURF_P2) | 830 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 831 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 832 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 833 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 834 NUM_BANKS(ADDR_SURF_16_BANK); 835 tilemode[1] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 836 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 837 PIPE_CONFIG(ADDR_SURF_P2) | 838 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 839 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 840 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 841 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 842 NUM_BANKS(ADDR_SURF_16_BANK); 843 tilemode[2] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 844 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 845 PIPE_CONFIG(ADDR_SURF_P2) | 846 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 847 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 848 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 849 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 850 NUM_BANKS(ADDR_SURF_16_BANK); 851 tilemode[3] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 852 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 853 PIPE_CONFIG(ADDR_SURF_P2) | 854 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 855 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 856 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 857 NUM_BANKS(ADDR_SURF_8_BANK) | 858 TILE_SPLIT(split_equal_to_row_size); 859 tilemode[4] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 860 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 861 PIPE_CONFIG(ADDR_SURF_P2); 862 tilemode[5] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 863 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 864 PIPE_CONFIG(ADDR_SURF_P2) | 865 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 866 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 867 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 868 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 869 NUM_BANKS(ADDR_SURF_8_BANK); 870 tilemode[6] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 871 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 872 PIPE_CONFIG(ADDR_SURF_P2) | 873 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 874 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 875 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 876 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 877 NUM_BANKS(ADDR_SURF_8_BANK); 878 tilemode[7] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 879 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 880 PIPE_CONFIG(ADDR_SURF_P2) | 881 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 882 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 883 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 884 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 885 NUM_BANKS(ADDR_SURF_4_BANK); 886 tilemode[8] = ARRAY_MODE(ARRAY_LINEAR_ALIGNED); 887 tilemode[9] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 888 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 889 PIPE_CONFIG(ADDR_SURF_P2); 890 tilemode[10] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 891 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 892 PIPE_CONFIG(ADDR_SURF_P2) | 893 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 894 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 895 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 896 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 897 NUM_BANKS(ADDR_SURF_16_BANK); 898 tilemode[11] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 899 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 900 PIPE_CONFIG(ADDR_SURF_P2) | 901 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 902 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 903 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 904 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 905 NUM_BANKS(ADDR_SURF_16_BANK); 906 tilemode[12] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 907 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 908 PIPE_CONFIG(ADDR_SURF_P2) | 909 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 910 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 911 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 912 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 913 NUM_BANKS(ADDR_SURF_16_BANK); 914 tilemode[13] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 915 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 916 PIPE_CONFIG(ADDR_SURF_P2); 917 tilemode[14] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 918 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 919 PIPE_CONFIG(ADDR_SURF_P2) | 920 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 921 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 922 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 923 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 924 NUM_BANKS(ADDR_SURF_16_BANK); 925 tilemode[15] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 926 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 927 PIPE_CONFIG(ADDR_SURF_P2) | 928 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 929 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 930 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 931 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 932 NUM_BANKS(ADDR_SURF_16_BANK); 933 tilemode[16] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 934 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 935 PIPE_CONFIG(ADDR_SURF_P2) | 936 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 937 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 938 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 939 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 940 NUM_BANKS(ADDR_SURF_16_BANK); 941 tilemode[17] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 942 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 943 PIPE_CONFIG(ADDR_SURF_P2) | 944 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 945 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 946 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 947 NUM_BANKS(ADDR_SURF_16_BANK) | 948 TILE_SPLIT(split_equal_to_row_size); 949 tilemode[18] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 950 ARRAY_MODE(ARRAY_1D_TILED_THICK) | 951 PIPE_CONFIG(ADDR_SURF_P2); 952 tilemode[19] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 953 ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 954 PIPE_CONFIG(ADDR_SURF_P2) | 955 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 956 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 957 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 958 NUM_BANKS(ADDR_SURF_16_BANK) | 959 TILE_SPLIT(split_equal_to_row_size); 960 tilemode[20] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 961 ARRAY_MODE(ARRAY_2D_TILED_THICK) | 962 PIPE_CONFIG(ADDR_SURF_P2) | 963 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 964 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 965 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 966 NUM_BANKS(ADDR_SURF_16_BANK) | 967 TILE_SPLIT(split_equal_to_row_size); 968 tilemode[21] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 969 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 970 PIPE_CONFIG(ADDR_SURF_P2) | 971 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 972 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 973 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 974 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 975 NUM_BANKS(ADDR_SURF_8_BANK); 976 tilemode[22] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 977 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 978 PIPE_CONFIG(ADDR_SURF_P2) | 979 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 980 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 981 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 982 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 983 NUM_BANKS(ADDR_SURF_8_BANK); 984 tilemode[23] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 985 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 986 PIPE_CONFIG(ADDR_SURF_P2) | 987 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 988 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 989 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 990 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 991 NUM_BANKS(ADDR_SURF_8_BANK); 992 tilemode[24] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 993 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 994 PIPE_CONFIG(ADDR_SURF_P2) | 995 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 996 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 997 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 998 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 999 NUM_BANKS(ADDR_SURF_8_BANK); 1000 tilemode[25] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1001 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1002 PIPE_CONFIG(ADDR_SURF_P2) | 1003 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1004 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1005 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1006 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1007 NUM_BANKS(ADDR_SURF_4_BANK); 1008 tilemode[26] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1009 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1010 PIPE_CONFIG(ADDR_SURF_P2) | 1011 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1012 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1013 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1014 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1015 NUM_BANKS(ADDR_SURF_4_BANK); 1016 tilemode[27] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1017 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1018 PIPE_CONFIG(ADDR_SURF_P2) | 1019 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1020 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1021 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1022 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1023 NUM_BANKS(ADDR_SURF_4_BANK); 1024 tilemode[28] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1025 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1026 PIPE_CONFIG(ADDR_SURF_P2) | 1027 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1028 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1029 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1030 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1031 NUM_BANKS(ADDR_SURF_4_BANK); 1032 tilemode[29] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1033 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1034 PIPE_CONFIG(ADDR_SURF_P2) | 1035 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1036 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1037 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1038 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1039 NUM_BANKS(ADDR_SURF_4_BANK); 1040 tilemode[30] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1041 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1042 PIPE_CONFIG(ADDR_SURF_P2) | 1043 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 1044 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1045 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1046 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1047 NUM_BANKS(ADDR_SURF_4_BANK); 1048 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 1049 WREG32(mmGB_TILE_MODE0 + reg_offset, tilemode[reg_offset]); 1050 } else if ((adev->asic_type == CHIP_TAHITI) || (adev->asic_type == CHIP_PITCAIRN)) { 1051 tilemode[0] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1052 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1053 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1054 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 1055 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1056 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1057 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1058 NUM_BANKS(ADDR_SURF_16_BANK); 1059 tilemode[1] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1060 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1061 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1062 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 1063 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1064 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1065 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1066 NUM_BANKS(ADDR_SURF_16_BANK); 1067 tilemode[2] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1068 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1069 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1070 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1071 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1072 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1073 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1074 NUM_BANKS(ADDR_SURF_16_BANK); 1075 tilemode[3] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1076 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1077 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1078 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1079 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1080 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1081 NUM_BANKS(ADDR_SURF_4_BANK) | 1082 TILE_SPLIT(split_equal_to_row_size); 1083 tilemode[4] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1084 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1085 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16); 1086 tilemode[5] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1087 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1088 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1089 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 1090 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1091 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1092 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1093 NUM_BANKS(ADDR_SURF_2_BANK); 1094 tilemode[6] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1095 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1096 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1097 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1098 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1099 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 1100 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1101 NUM_BANKS(ADDR_SURF_2_BANK); 1102 tilemode[7] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1103 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1104 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 1105 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1106 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1107 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1108 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1109 NUM_BANKS(ADDR_SURF_2_BANK); 1110 tilemode[8] = ARRAY_MODE(ARRAY_LINEAR_ALIGNED); 1111 tilemode[9] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 1112 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1113 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16); 1114 tilemode[10] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 1115 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1116 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1117 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1118 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1119 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1120 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1121 NUM_BANKS(ADDR_SURF_16_BANK); 1122 tilemode[11] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 1123 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1124 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1125 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1126 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1127 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1128 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1129 NUM_BANKS(ADDR_SURF_16_BANK); 1130 tilemode[12] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 1131 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1132 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1133 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 1134 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1135 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1136 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1137 NUM_BANKS(ADDR_SURF_16_BANK); 1138 tilemode[13] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1139 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1140 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16); 1141 tilemode[14] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1142 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1143 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1144 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1145 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1146 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1147 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1148 NUM_BANKS(ADDR_SURF_16_BANK); 1149 tilemode[15] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1150 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1151 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1152 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1153 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1154 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1155 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1156 NUM_BANKS(ADDR_SURF_16_BANK); 1157 tilemode[16] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1158 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1159 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1160 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 1161 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1162 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1163 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1164 NUM_BANKS(ADDR_SURF_16_BANK); 1165 tilemode[17] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1166 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1167 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1168 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1169 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1170 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1171 NUM_BANKS(ADDR_SURF_16_BANK) | 1172 TILE_SPLIT(split_equal_to_row_size); 1173 tilemode[18] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1174 ARRAY_MODE(ARRAY_1D_TILED_THICK) | 1175 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16); 1176 tilemode[19] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1177 ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 1178 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1179 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1180 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1181 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1182 NUM_BANKS(ADDR_SURF_16_BANK) | 1183 TILE_SPLIT(split_equal_to_row_size); 1184 tilemode[20] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1185 ARRAY_MODE(ARRAY_2D_TILED_THICK) | 1186 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1187 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1188 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1189 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1190 NUM_BANKS(ADDR_SURF_16_BANK) | 1191 TILE_SPLIT(split_equal_to_row_size); 1192 tilemode[21] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1193 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1194 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1195 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1196 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1197 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 1198 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1199 NUM_BANKS(ADDR_SURF_4_BANK); 1200 tilemode[22] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1201 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1202 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1203 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1204 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1205 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1206 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1207 NUM_BANKS(ADDR_SURF_4_BANK); 1208 tilemode[23] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1209 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1210 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1211 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1212 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1213 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 1214 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1215 NUM_BANKS(ADDR_SURF_2_BANK); 1216 tilemode[24] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1217 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1218 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1219 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 1220 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1221 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1222 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1223 NUM_BANKS(ADDR_SURF_2_BANK); 1224 tilemode[25] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1225 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1226 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 1227 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1228 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1229 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1230 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1231 NUM_BANKS(ADDR_SURF_2_BANK); 1232 tilemode[26] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1233 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1234 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 1235 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1236 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1237 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1238 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1239 NUM_BANKS(ADDR_SURF_2_BANK); 1240 tilemode[27] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1241 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1242 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 1243 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1244 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1245 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1246 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1247 NUM_BANKS(ADDR_SURF_2_BANK); 1248 tilemode[28] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1249 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1250 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 1251 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1252 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1253 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1254 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1255 NUM_BANKS(ADDR_SURF_2_BANK); 1256 tilemode[29] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1257 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1258 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 1259 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1260 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1261 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1262 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1263 NUM_BANKS(ADDR_SURF_2_BANK); 1264 tilemode[30] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1265 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1266 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 1267 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 1268 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1269 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1270 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1271 NUM_BANKS(ADDR_SURF_2_BANK); 1272 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 1273 WREG32(mmGB_TILE_MODE0 + reg_offset, tilemode[reg_offset]); 1274 } else { 1275 DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type); 1276 } 1277 } 1278 1279 static void gfx_v6_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, 1280 u32 sh_num, u32 instance) 1281 { 1282 u32 data; 1283 1284 if (instance == 0xffffffff) 1285 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1); 1286 else 1287 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance); 1288 1289 if ((se_num == 0xffffffff) && (sh_num == 0xffffffff)) 1290 data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK | 1291 GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK; 1292 else if (se_num == 0xffffffff) 1293 data |= GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK | 1294 (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT); 1295 else if (sh_num == 0xffffffff) 1296 data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK | 1297 (se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT); 1298 else 1299 data |= (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT) | 1300 (se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT); 1301 WREG32(mmGRBM_GFX_INDEX, data); 1302 } 1303 1304 static u32 gfx_v6_0_get_rb_active_bitmap(struct amdgpu_device *adev) 1305 { 1306 u32 data, mask; 1307 1308 data = RREG32(mmCC_RB_BACKEND_DISABLE) | 1309 RREG32(mmGC_USER_RB_BACKEND_DISABLE); 1310 1311 data = REG_GET_FIELD(data, GC_USER_RB_BACKEND_DISABLE, BACKEND_DISABLE); 1312 1313 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se/ 1314 adev->gfx.config.max_sh_per_se); 1315 1316 return ~data & mask; 1317 } 1318 1319 static void gfx_v6_0_raster_config(struct amdgpu_device *adev, u32 *rconf) 1320 { 1321 switch (adev->asic_type) { 1322 case CHIP_TAHITI: 1323 case CHIP_PITCAIRN: 1324 *rconf |= 1325 (2 << PA_SC_RASTER_CONFIG__RB_XSEL2__SHIFT) | 1326 (1 << PA_SC_RASTER_CONFIG__RB_XSEL__SHIFT) | 1327 (2 << PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT) | 1328 (1 << PA_SC_RASTER_CONFIG__PKR_YSEL__SHIFT) | 1329 (2 << PA_SC_RASTER_CONFIG__SE_MAP__SHIFT) | 1330 (2 << PA_SC_RASTER_CONFIG__SE_XSEL__SHIFT) | 1331 (2 << PA_SC_RASTER_CONFIG__SE_YSEL__SHIFT); 1332 break; 1333 case CHIP_VERDE: 1334 *rconf |= 1335 (1 << PA_SC_RASTER_CONFIG__RB_XSEL__SHIFT) | 1336 (2 << PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT) | 1337 (1 << PA_SC_RASTER_CONFIG__PKR_YSEL__SHIFT); 1338 break; 1339 case CHIP_OLAND: 1340 *rconf |= (1 << PA_SC_RASTER_CONFIG__RB_YSEL__SHIFT); 1341 break; 1342 case CHIP_HAINAN: 1343 *rconf |= 0x0; 1344 break; 1345 default: 1346 DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type); 1347 break; 1348 } 1349 } 1350 1351 static void gfx_v6_0_write_harvested_raster_configs(struct amdgpu_device *adev, 1352 u32 raster_config, unsigned rb_mask, 1353 unsigned num_rb) 1354 { 1355 unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1); 1356 unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1); 1357 unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2); 1358 unsigned rb_per_se = num_rb / num_se; 1359 unsigned se_mask[4]; 1360 unsigned se; 1361 1362 se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask; 1363 se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask; 1364 se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask; 1365 se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask; 1366 1367 WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4)); 1368 WARN_ON(!(sh_per_se == 1 || sh_per_se == 2)); 1369 WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2)); 1370 1371 for (se = 0; se < num_se; se++) { 1372 unsigned raster_config_se = raster_config; 1373 unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se); 1374 unsigned pkr1_mask = pkr0_mask << rb_per_pkr; 1375 int idx = (se / 2) * 2; 1376 1377 if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) { 1378 raster_config_se &= ~PA_SC_RASTER_CONFIG__SE_MAP_MASK; 1379 1380 if (!se_mask[idx]) 1381 raster_config_se |= RASTER_CONFIG_SE_MAP_3 << PA_SC_RASTER_CONFIG__SE_MAP__SHIFT; 1382 else 1383 raster_config_se |= RASTER_CONFIG_SE_MAP_0 << PA_SC_RASTER_CONFIG__SE_MAP__SHIFT; 1384 } 1385 1386 pkr0_mask &= rb_mask; 1387 pkr1_mask &= rb_mask; 1388 if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) { 1389 raster_config_se &= ~PA_SC_RASTER_CONFIG__PKR_MAP_MASK; 1390 1391 if (!pkr0_mask) 1392 raster_config_se |= RASTER_CONFIG_PKR_MAP_3 << PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT; 1393 else 1394 raster_config_se |= RASTER_CONFIG_PKR_MAP_0 << PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT; 1395 } 1396 1397 if (rb_per_se >= 2) { 1398 unsigned rb0_mask = 1 << (se * rb_per_se); 1399 unsigned rb1_mask = rb0_mask << 1; 1400 1401 rb0_mask &= rb_mask; 1402 rb1_mask &= rb_mask; 1403 if (!rb0_mask || !rb1_mask) { 1404 raster_config_se &= ~PA_SC_RASTER_CONFIG__RB_MAP_PKR0_MASK; 1405 1406 if (!rb0_mask) 1407 raster_config_se |= 1408 RASTER_CONFIG_RB_MAP_3 << PA_SC_RASTER_CONFIG__RB_MAP_PKR0__SHIFT; 1409 else 1410 raster_config_se |= 1411 RASTER_CONFIG_RB_MAP_0 << PA_SC_RASTER_CONFIG__RB_MAP_PKR0__SHIFT; 1412 } 1413 1414 if (rb_per_se > 2) { 1415 rb0_mask = 1 << (se * rb_per_se + rb_per_pkr); 1416 rb1_mask = rb0_mask << 1; 1417 rb0_mask &= rb_mask; 1418 rb1_mask &= rb_mask; 1419 if (!rb0_mask || !rb1_mask) { 1420 raster_config_se &= ~PA_SC_RASTER_CONFIG__RB_MAP_PKR1_MASK; 1421 1422 if (!rb0_mask) 1423 raster_config_se |= 1424 RASTER_CONFIG_RB_MAP_3 << PA_SC_RASTER_CONFIG__RB_MAP_PKR1__SHIFT; 1425 else 1426 raster_config_se |= 1427 RASTER_CONFIG_RB_MAP_0 << PA_SC_RASTER_CONFIG__RB_MAP_PKR1__SHIFT; 1428 } 1429 } 1430 } 1431 1432 /* GRBM_GFX_INDEX has a different offset on SI */ 1433 gfx_v6_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff); 1434 WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se); 1435 } 1436 1437 /* GRBM_GFX_INDEX has a different offset on SI */ 1438 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 1439 } 1440 1441 static void gfx_v6_0_setup_rb(struct amdgpu_device *adev) 1442 { 1443 int i, j; 1444 u32 data; 1445 u32 raster_config = 0; 1446 u32 active_rbs = 0; 1447 u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se / 1448 adev->gfx.config.max_sh_per_se; 1449 unsigned num_rb_pipes; 1450 1451 mutex_lock(&adev->grbm_idx_mutex); 1452 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 1453 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 1454 gfx_v6_0_select_se_sh(adev, i, j, 0xffffffff); 1455 data = gfx_v6_0_get_rb_active_bitmap(adev); 1456 active_rbs |= data << 1457 ((i * adev->gfx.config.max_sh_per_se + j) * 1458 rb_bitmap_width_per_sh); 1459 } 1460 } 1461 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 1462 1463 adev->gfx.config.backend_enable_mask = active_rbs; 1464 adev->gfx.config.num_rbs = hweight32(active_rbs); 1465 1466 num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se * 1467 adev->gfx.config.max_shader_engines, 16); 1468 1469 gfx_v6_0_raster_config(adev, &raster_config); 1470 1471 if (!adev->gfx.config.backend_enable_mask || 1472 adev->gfx.config.num_rbs >= num_rb_pipes) 1473 WREG32(mmPA_SC_RASTER_CONFIG, raster_config); 1474 else 1475 gfx_v6_0_write_harvested_raster_configs(adev, raster_config, 1476 adev->gfx.config.backend_enable_mask, 1477 num_rb_pipes); 1478 1479 /* cache the values for userspace */ 1480 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 1481 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 1482 gfx_v6_0_select_se_sh(adev, i, j, 0xffffffff); 1483 adev->gfx.config.rb_config[i][j].rb_backend_disable = 1484 RREG32(mmCC_RB_BACKEND_DISABLE); 1485 adev->gfx.config.rb_config[i][j].user_rb_backend_disable = 1486 RREG32(mmGC_USER_RB_BACKEND_DISABLE); 1487 adev->gfx.config.rb_config[i][j].raster_config = 1488 RREG32(mmPA_SC_RASTER_CONFIG); 1489 } 1490 } 1491 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 1492 mutex_unlock(&adev->grbm_idx_mutex); 1493 } 1494 1495 static void gfx_v6_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev, 1496 u32 bitmap) 1497 { 1498 u32 data; 1499 1500 if (!bitmap) 1501 return; 1502 1503 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; 1504 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; 1505 1506 WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data); 1507 } 1508 1509 static u32 gfx_v6_0_get_cu_enabled(struct amdgpu_device *adev) 1510 { 1511 u32 data, mask; 1512 1513 data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG) | 1514 RREG32(mmGC_USER_SHADER_ARRAY_CONFIG); 1515 1516 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh); 1517 return ~REG_GET_FIELD(data, CC_GC_SHADER_ARRAY_CONFIG, INACTIVE_CUS) & mask; 1518 } 1519 1520 1521 static void gfx_v6_0_setup_spi(struct amdgpu_device *adev) 1522 { 1523 int i, j, k; 1524 u32 data, mask; 1525 u32 active_cu = 0; 1526 1527 mutex_lock(&adev->grbm_idx_mutex); 1528 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 1529 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 1530 gfx_v6_0_select_se_sh(adev, i, j, 0xffffffff); 1531 data = RREG32(mmSPI_STATIC_THREAD_MGMT_3); 1532 active_cu = gfx_v6_0_get_cu_enabled(adev); 1533 1534 mask = 1; 1535 for (k = 0; k < 16; k++) { 1536 mask <<= k; 1537 if (active_cu & mask) { 1538 data &= ~mask; 1539 WREG32(mmSPI_STATIC_THREAD_MGMT_3, data); 1540 break; 1541 } 1542 } 1543 } 1544 } 1545 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 1546 mutex_unlock(&adev->grbm_idx_mutex); 1547 } 1548 1549 static void gfx_v6_0_config_init(struct amdgpu_device *adev) 1550 { 1551 adev->gfx.config.double_offchip_lds_buf = 0; 1552 } 1553 1554 static void gfx_v6_0_gpu_init(struct amdgpu_device *adev) 1555 { 1556 u32 gb_addr_config = 0; 1557 u32 mc_shared_chmap, mc_arb_ramcfg; 1558 u32 sx_debug_1; 1559 u32 hdp_host_path_cntl; 1560 u32 tmp; 1561 1562 switch (adev->asic_type) { 1563 case CHIP_TAHITI: 1564 adev->gfx.config.max_shader_engines = 2; 1565 adev->gfx.config.max_tile_pipes = 12; 1566 adev->gfx.config.max_cu_per_sh = 8; 1567 adev->gfx.config.max_sh_per_se = 2; 1568 adev->gfx.config.max_backends_per_se = 4; 1569 adev->gfx.config.max_texture_channel_caches = 12; 1570 adev->gfx.config.max_gprs = 256; 1571 adev->gfx.config.max_gs_threads = 32; 1572 adev->gfx.config.max_hw_contexts = 8; 1573 1574 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1575 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1576 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1577 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1578 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN; 1579 break; 1580 case CHIP_PITCAIRN: 1581 adev->gfx.config.max_shader_engines = 2; 1582 adev->gfx.config.max_tile_pipes = 8; 1583 adev->gfx.config.max_cu_per_sh = 5; 1584 adev->gfx.config.max_sh_per_se = 2; 1585 adev->gfx.config.max_backends_per_se = 4; 1586 adev->gfx.config.max_texture_channel_caches = 8; 1587 adev->gfx.config.max_gprs = 256; 1588 adev->gfx.config.max_gs_threads = 32; 1589 adev->gfx.config.max_hw_contexts = 8; 1590 1591 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1592 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1593 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1594 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1595 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN; 1596 break; 1597 case CHIP_VERDE: 1598 adev->gfx.config.max_shader_engines = 1; 1599 adev->gfx.config.max_tile_pipes = 4; 1600 adev->gfx.config.max_cu_per_sh = 5; 1601 adev->gfx.config.max_sh_per_se = 2; 1602 adev->gfx.config.max_backends_per_se = 4; 1603 adev->gfx.config.max_texture_channel_caches = 4; 1604 adev->gfx.config.max_gprs = 256; 1605 adev->gfx.config.max_gs_threads = 32; 1606 adev->gfx.config.max_hw_contexts = 8; 1607 1608 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1609 adev->gfx.config.sc_prim_fifo_size_backend = 0x40; 1610 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1611 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1612 gb_addr_config = VERDE_GB_ADDR_CONFIG_GOLDEN; 1613 break; 1614 case CHIP_OLAND: 1615 adev->gfx.config.max_shader_engines = 1; 1616 adev->gfx.config.max_tile_pipes = 4; 1617 adev->gfx.config.max_cu_per_sh = 6; 1618 adev->gfx.config.max_sh_per_se = 1; 1619 adev->gfx.config.max_backends_per_se = 2; 1620 adev->gfx.config.max_texture_channel_caches = 4; 1621 adev->gfx.config.max_gprs = 256; 1622 adev->gfx.config.max_gs_threads = 16; 1623 adev->gfx.config.max_hw_contexts = 8; 1624 1625 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1626 adev->gfx.config.sc_prim_fifo_size_backend = 0x40; 1627 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1628 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1629 gb_addr_config = VERDE_GB_ADDR_CONFIG_GOLDEN; 1630 break; 1631 case CHIP_HAINAN: 1632 adev->gfx.config.max_shader_engines = 1; 1633 adev->gfx.config.max_tile_pipes = 4; 1634 adev->gfx.config.max_cu_per_sh = 5; 1635 adev->gfx.config.max_sh_per_se = 1; 1636 adev->gfx.config.max_backends_per_se = 1; 1637 adev->gfx.config.max_texture_channel_caches = 2; 1638 adev->gfx.config.max_gprs = 256; 1639 adev->gfx.config.max_gs_threads = 16; 1640 adev->gfx.config.max_hw_contexts = 8; 1641 1642 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1643 adev->gfx.config.sc_prim_fifo_size_backend = 0x40; 1644 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1645 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1646 gb_addr_config = HAINAN_GB_ADDR_CONFIG_GOLDEN; 1647 break; 1648 default: 1649 BUG(); 1650 break; 1651 } 1652 1653 WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT)); 1654 WREG32(mmSRBM_INT_CNTL, 1); 1655 WREG32(mmSRBM_INT_ACK, 1); 1656 1657 WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK); 1658 1659 mc_shared_chmap = RREG32(mmMC_SHARED_CHMAP); 1660 adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG); 1661 mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg; 1662 1663 adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes; 1664 adev->gfx.config.mem_max_burst_length_bytes = 256; 1665 tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT; 1666 adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; 1667 if (adev->gfx.config.mem_row_size_in_kb > 4) 1668 adev->gfx.config.mem_row_size_in_kb = 4; 1669 adev->gfx.config.shader_engine_tile_size = 32; 1670 adev->gfx.config.num_gpus = 1; 1671 adev->gfx.config.multi_gpu_tile_size = 64; 1672 1673 gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK; 1674 switch (adev->gfx.config.mem_row_size_in_kb) { 1675 case 1: 1676 default: 1677 gb_addr_config |= 0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT; 1678 break; 1679 case 2: 1680 gb_addr_config |= 1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT; 1681 break; 1682 case 4: 1683 gb_addr_config |= 2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT; 1684 break; 1685 } 1686 gb_addr_config &= ~GB_ADDR_CONFIG__NUM_SHADER_ENGINES_MASK; 1687 if (adev->gfx.config.max_shader_engines == 2) 1688 gb_addr_config |= 1 << GB_ADDR_CONFIG__NUM_SHADER_ENGINES__SHIFT; 1689 adev->gfx.config.gb_addr_config = gb_addr_config; 1690 1691 WREG32(mmGB_ADDR_CONFIG, gb_addr_config); 1692 WREG32(mmDMIF_ADDR_CONFIG, gb_addr_config); 1693 WREG32(mmDMIF_ADDR_CALC, gb_addr_config); 1694 WREG32(mmHDP_ADDR_CONFIG, gb_addr_config); 1695 WREG32(mmDMA_TILING_CONFIG + DMA0_REGISTER_OFFSET, gb_addr_config); 1696 WREG32(mmDMA_TILING_CONFIG + DMA1_REGISTER_OFFSET, gb_addr_config); 1697 1698 #if 0 1699 if (adev->has_uvd) { 1700 WREG32(mmUVD_UDEC_ADDR_CONFIG, gb_addr_config); 1701 WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, gb_addr_config); 1702 WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config); 1703 } 1704 #endif 1705 gfx_v6_0_tiling_mode_table_init(adev); 1706 1707 gfx_v6_0_setup_rb(adev); 1708 1709 gfx_v6_0_setup_spi(adev); 1710 1711 gfx_v6_0_get_cu_info(adev); 1712 gfx_v6_0_config_init(adev); 1713 1714 WREG32(mmCP_QUEUE_THRESHOLDS, ((0x16 << CP_QUEUE_THRESHOLDS__ROQ_IB1_START__SHIFT) | 1715 (0x2b << CP_QUEUE_THRESHOLDS__ROQ_IB2_START__SHIFT))); 1716 WREG32(mmCP_MEQ_THRESHOLDS, (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) | 1717 (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT)); 1718 1719 sx_debug_1 = RREG32(mmSX_DEBUG_1); 1720 WREG32(mmSX_DEBUG_1, sx_debug_1); 1721 1722 WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT)); 1723 1724 WREG32(mmPA_SC_FIFO_SIZE, ((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) | 1725 (adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) | 1726 (adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) | 1727 (adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT))); 1728 1729 WREG32(mmVGT_NUM_INSTANCES, 1); 1730 WREG32(mmCP_PERFMON_CNTL, 0); 1731 WREG32(mmSQ_CONFIG, 0); 1732 WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS, ((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) | 1733 (255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT))); 1734 1735 WREG32(mmVGT_CACHE_INVALIDATION, 1736 (VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) | 1737 (ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT)); 1738 1739 WREG32(mmVGT_GS_VERTEX_REUSE, 16); 1740 WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0); 1741 1742 WREG32(mmCB_PERFCOUNTER0_SELECT0, 0); 1743 WREG32(mmCB_PERFCOUNTER0_SELECT1, 0); 1744 WREG32(mmCB_PERFCOUNTER1_SELECT0, 0); 1745 WREG32(mmCB_PERFCOUNTER1_SELECT1, 0); 1746 WREG32(mmCB_PERFCOUNTER2_SELECT0, 0); 1747 WREG32(mmCB_PERFCOUNTER2_SELECT1, 0); 1748 WREG32(mmCB_PERFCOUNTER3_SELECT0, 0); 1749 WREG32(mmCB_PERFCOUNTER3_SELECT1, 0); 1750 1751 hdp_host_path_cntl = RREG32(mmHDP_HOST_PATH_CNTL); 1752 WREG32(mmHDP_HOST_PATH_CNTL, hdp_host_path_cntl); 1753 1754 WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK | 1755 (3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT)); 1756 1757 udelay(50); 1758 } 1759 1760 1761 static void gfx_v6_0_scratch_init(struct amdgpu_device *adev) 1762 { 1763 adev->gfx.scratch.num_reg = 8; 1764 adev->gfx.scratch.reg_base = mmSCRATCH_REG0; 1765 adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1; 1766 } 1767 1768 static int gfx_v6_0_ring_test_ring(struct amdgpu_ring *ring) 1769 { 1770 struct amdgpu_device *adev = ring->adev; 1771 uint32_t scratch; 1772 uint32_t tmp = 0; 1773 unsigned i; 1774 int r; 1775 1776 r = amdgpu_gfx_scratch_get(adev, &scratch); 1777 if (r) { 1778 DRM_ERROR("amdgpu: cp failed to get scratch reg (%d).\n", r); 1779 return r; 1780 } 1781 WREG32(scratch, 0xCAFEDEAD); 1782 1783 r = amdgpu_ring_alloc(ring, 3); 1784 if (r) { 1785 DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n", ring->idx, r); 1786 amdgpu_gfx_scratch_free(adev, scratch); 1787 return r; 1788 } 1789 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 1790 amdgpu_ring_write(ring, (scratch - PACKET3_SET_CONFIG_REG_START)); 1791 amdgpu_ring_write(ring, 0xDEADBEEF); 1792 amdgpu_ring_commit(ring); 1793 1794 for (i = 0; i < adev->usec_timeout; i++) { 1795 tmp = RREG32(scratch); 1796 if (tmp == 0xDEADBEEF) 1797 break; 1798 DRM_UDELAY(1); 1799 } 1800 if (i < adev->usec_timeout) { 1801 DRM_DEBUG("ring test on %d succeeded in %d usecs\n", ring->idx, i); 1802 } else { 1803 DRM_ERROR("amdgpu: ring %d test failed (scratch(0x%04X)=0x%08X)\n", 1804 ring->idx, scratch, tmp); 1805 r = -EINVAL; 1806 } 1807 amdgpu_gfx_scratch_free(adev, scratch); 1808 return r; 1809 } 1810 1811 static void gfx_v6_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) 1812 { 1813 /* flush hdp cache */ 1814 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 1815 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) | 1816 WRITE_DATA_DST_SEL(0))); 1817 amdgpu_ring_write(ring, mmHDP_MEM_COHERENCY_FLUSH_CNTL); 1818 amdgpu_ring_write(ring, 0); 1819 amdgpu_ring_write(ring, 0x1); 1820 } 1821 1822 static void gfx_v6_0_ring_emit_vgt_flush(struct amdgpu_ring *ring) 1823 { 1824 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0)); 1825 amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) | 1826 EVENT_INDEX(0)); 1827 } 1828 1829 /** 1830 * gfx_v6_0_ring_emit_hdp_invalidate - emit an hdp invalidate on the cp 1831 * 1832 * @adev: amdgpu_device pointer 1833 * @ridx: amdgpu ring index 1834 * 1835 * Emits an hdp invalidate on the cp. 1836 */ 1837 static void gfx_v6_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring) 1838 { 1839 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 1840 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) | 1841 WRITE_DATA_DST_SEL(0))); 1842 amdgpu_ring_write(ring, mmHDP_DEBUG0); 1843 amdgpu_ring_write(ring, 0); 1844 amdgpu_ring_write(ring, 0x1); 1845 } 1846 1847 static void gfx_v6_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, 1848 u64 seq, unsigned flags) 1849 { 1850 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT; 1851 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT; 1852 /* flush read cache over gart */ 1853 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 1854 amdgpu_ring_write(ring, (mmCP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START)); 1855 amdgpu_ring_write(ring, 0); 1856 amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3)); 1857 amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA | 1858 PACKET3_TC_ACTION_ENA | 1859 PACKET3_SH_KCACHE_ACTION_ENA | 1860 PACKET3_SH_ICACHE_ACTION_ENA); 1861 amdgpu_ring_write(ring, 0xFFFFFFFF); 1862 amdgpu_ring_write(ring, 0); 1863 amdgpu_ring_write(ring, 10); /* poll interval */ 1864 /* EVENT_WRITE_EOP - flush caches, send int */ 1865 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4)); 1866 amdgpu_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5)); 1867 amdgpu_ring_write(ring, addr & 0xfffffffc); 1868 amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) | 1869 ((write64bit ? 2 : 1) << CP_EOP_DONE_DATA_CNTL__DATA_SEL__SHIFT) | 1870 ((int_sel ? 2 : 0) << CP_EOP_DONE_DATA_CNTL__INT_SEL__SHIFT)); 1871 amdgpu_ring_write(ring, lower_32_bits(seq)); 1872 amdgpu_ring_write(ring, upper_32_bits(seq)); 1873 } 1874 1875 static void gfx_v6_0_ring_emit_ib(struct amdgpu_ring *ring, 1876 struct amdgpu_ib *ib, 1877 unsigned vmid, bool ctx_switch) 1878 { 1879 u32 header, control = 0; 1880 1881 /* insert SWITCH_BUFFER packet before first IB in the ring frame */ 1882 if (ctx_switch) { 1883 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 1884 amdgpu_ring_write(ring, 0); 1885 } 1886 1887 if (ib->flags & AMDGPU_IB_FLAG_CE) 1888 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2); 1889 else 1890 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); 1891 1892 control |= ib->length_dw | (vmid << 24); 1893 1894 amdgpu_ring_write(ring, header); 1895 amdgpu_ring_write(ring, 1896 #ifdef __BIG_ENDIAN 1897 (2 << 0) | 1898 #endif 1899 (ib->gpu_addr & 0xFFFFFFFC)); 1900 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF); 1901 amdgpu_ring_write(ring, control); 1902 } 1903 1904 /** 1905 * gfx_v6_0_ring_test_ib - basic ring IB test 1906 * 1907 * @ring: amdgpu_ring structure holding ring information 1908 * 1909 * Allocate an IB and execute it on the gfx ring (SI). 1910 * Provides a basic gfx ring test to verify that IBs are working. 1911 * Returns 0 on success, error on failure. 1912 */ 1913 static int gfx_v6_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) 1914 { 1915 struct amdgpu_device *adev = ring->adev; 1916 struct amdgpu_ib ib; 1917 struct dma_fence *f = NULL; 1918 uint32_t scratch; 1919 uint32_t tmp = 0; 1920 long r; 1921 1922 r = amdgpu_gfx_scratch_get(adev, &scratch); 1923 if (r) { 1924 DRM_ERROR("amdgpu: failed to get scratch reg (%ld).\n", r); 1925 return r; 1926 } 1927 WREG32(scratch, 0xCAFEDEAD); 1928 memset(&ib, 0, sizeof(ib)); 1929 r = amdgpu_ib_get(adev, NULL, 256, &ib); 1930 if (r) { 1931 DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r); 1932 goto err1; 1933 } 1934 ib.ptr[0] = PACKET3(PACKET3_SET_CONFIG_REG, 1); 1935 ib.ptr[1] = ((scratch - PACKET3_SET_CONFIG_REG_START)); 1936 ib.ptr[2] = 0xDEADBEEF; 1937 ib.length_dw = 3; 1938 1939 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); 1940 if (r) 1941 goto err2; 1942 1943 r = dma_fence_wait_timeout(f, false, timeout); 1944 if (r == 0) { 1945 DRM_ERROR("amdgpu: IB test timed out\n"); 1946 r = -ETIMEDOUT; 1947 goto err2; 1948 } else if (r < 0) { 1949 DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r); 1950 goto err2; 1951 } 1952 tmp = RREG32(scratch); 1953 if (tmp == 0xDEADBEEF) { 1954 DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx); 1955 r = 0; 1956 } else { 1957 DRM_ERROR("amdgpu: ib test failed (scratch(0x%04X)=0x%08X)\n", 1958 scratch, tmp); 1959 r = -EINVAL; 1960 } 1961 1962 err2: 1963 amdgpu_ib_free(adev, &ib, NULL); 1964 dma_fence_put(f); 1965 err1: 1966 amdgpu_gfx_scratch_free(adev, scratch); 1967 return r; 1968 } 1969 1970 static void gfx_v6_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable) 1971 { 1972 int i; 1973 if (enable) { 1974 WREG32(mmCP_ME_CNTL, 0); 1975 } else { 1976 WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK | 1977 CP_ME_CNTL__PFP_HALT_MASK | 1978 CP_ME_CNTL__CE_HALT_MASK)); 1979 WREG32(mmSCRATCH_UMSK, 0); 1980 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 1981 adev->gfx.gfx_ring[i].ready = false; 1982 for (i = 0; i < adev->gfx.num_compute_rings; i++) 1983 adev->gfx.compute_ring[i].ready = false; 1984 } 1985 udelay(50); 1986 } 1987 1988 static int gfx_v6_0_cp_gfx_load_microcode(struct amdgpu_device *adev) 1989 { 1990 unsigned i; 1991 const struct gfx_firmware_header_v1_0 *pfp_hdr; 1992 const struct gfx_firmware_header_v1_0 *ce_hdr; 1993 const struct gfx_firmware_header_v1_0 *me_hdr; 1994 const __le32 *fw_data; 1995 u32 fw_size; 1996 1997 if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw) 1998 return -EINVAL; 1999 2000 gfx_v6_0_cp_gfx_enable(adev, false); 2001 pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; 2002 ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; 2003 me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; 2004 2005 amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header); 2006 amdgpu_ucode_print_gfx_hdr(&ce_hdr->header); 2007 amdgpu_ucode_print_gfx_hdr(&me_hdr->header); 2008 2009 /* PFP */ 2010 fw_data = (const __le32 *) 2011 (adev->gfx.pfp_fw->data + le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes)); 2012 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4; 2013 WREG32(mmCP_PFP_UCODE_ADDR, 0); 2014 for (i = 0; i < fw_size; i++) 2015 WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++)); 2016 WREG32(mmCP_PFP_UCODE_ADDR, 0); 2017 2018 /* CE */ 2019 fw_data = (const __le32 *) 2020 (adev->gfx.ce_fw->data + le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes)); 2021 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4; 2022 WREG32(mmCP_CE_UCODE_ADDR, 0); 2023 for (i = 0; i < fw_size; i++) 2024 WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++)); 2025 WREG32(mmCP_CE_UCODE_ADDR, 0); 2026 2027 /* ME */ 2028 fw_data = (const __be32 *) 2029 (adev->gfx.me_fw->data + le32_to_cpu(me_hdr->header.ucode_array_offset_bytes)); 2030 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4; 2031 WREG32(mmCP_ME_RAM_WADDR, 0); 2032 for (i = 0; i < fw_size; i++) 2033 WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++)); 2034 WREG32(mmCP_ME_RAM_WADDR, 0); 2035 2036 WREG32(mmCP_PFP_UCODE_ADDR, 0); 2037 WREG32(mmCP_CE_UCODE_ADDR, 0); 2038 WREG32(mmCP_ME_RAM_WADDR, 0); 2039 WREG32(mmCP_ME_RAM_RADDR, 0); 2040 return 0; 2041 } 2042 2043 static int gfx_v6_0_cp_gfx_start(struct amdgpu_device *adev) 2044 { 2045 const struct cs_section_def *sect = NULL; 2046 const struct cs_extent_def *ext = NULL; 2047 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0]; 2048 int r, i; 2049 2050 r = amdgpu_ring_alloc(ring, 7 + 4); 2051 if (r) { 2052 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r); 2053 return r; 2054 } 2055 amdgpu_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5)); 2056 amdgpu_ring_write(ring, 0x1); 2057 amdgpu_ring_write(ring, 0x0); 2058 amdgpu_ring_write(ring, adev->gfx.config.max_hw_contexts - 1); 2059 amdgpu_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1)); 2060 amdgpu_ring_write(ring, 0); 2061 amdgpu_ring_write(ring, 0); 2062 2063 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2)); 2064 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE)); 2065 amdgpu_ring_write(ring, 0xc000); 2066 amdgpu_ring_write(ring, 0xe000); 2067 amdgpu_ring_commit(ring); 2068 2069 gfx_v6_0_cp_gfx_enable(adev, true); 2070 2071 r = amdgpu_ring_alloc(ring, gfx_v6_0_get_csb_size(adev) + 10); 2072 if (r) { 2073 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r); 2074 return r; 2075 } 2076 2077 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 2078 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 2079 2080 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 2081 for (ext = sect->section; ext->extent != NULL; ++ext) { 2082 if (sect->id == SECT_CONTEXT) { 2083 amdgpu_ring_write(ring, 2084 PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count)); 2085 amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START); 2086 for (i = 0; i < ext->reg_count; i++) 2087 amdgpu_ring_write(ring, ext->extent[i]); 2088 } 2089 } 2090 } 2091 2092 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 2093 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE); 2094 2095 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0)); 2096 amdgpu_ring_write(ring, 0); 2097 2098 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2)); 2099 amdgpu_ring_write(ring, 0x00000316); 2100 amdgpu_ring_write(ring, 0x0000000e); 2101 amdgpu_ring_write(ring, 0x00000010); 2102 2103 amdgpu_ring_commit(ring); 2104 2105 return 0; 2106 } 2107 2108 static int gfx_v6_0_cp_gfx_resume(struct amdgpu_device *adev) 2109 { 2110 struct amdgpu_ring *ring; 2111 u32 tmp; 2112 u32 rb_bufsz; 2113 int r; 2114 u64 rptr_addr; 2115 2116 WREG32(mmCP_SEM_WAIT_TIMER, 0x0); 2117 WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0); 2118 2119 /* Set the write pointer delay */ 2120 WREG32(mmCP_RB_WPTR_DELAY, 0); 2121 2122 WREG32(mmCP_DEBUG, 0); 2123 WREG32(mmSCRATCH_ADDR, 0); 2124 2125 /* ring 0 - compute and gfx */ 2126 /* Set ring buffer size */ 2127 ring = &adev->gfx.gfx_ring[0]; 2128 rb_bufsz = order_base_2(ring->ring_size / 8); 2129 tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 2130 2131 #ifdef __BIG_ENDIAN 2132 tmp |= BUF_SWAP_32BIT; 2133 #endif 2134 WREG32(mmCP_RB0_CNTL, tmp); 2135 2136 /* Initialize the ring buffer's read and write pointers */ 2137 WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK); 2138 ring->wptr = 0; 2139 WREG32(mmCP_RB0_WPTR, ring->wptr); 2140 2141 /* set the wb address whether it's enabled or not */ 2142 rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4); 2143 WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr)); 2144 WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF); 2145 2146 WREG32(mmSCRATCH_UMSK, 0); 2147 2148 mdelay(1); 2149 WREG32(mmCP_RB0_CNTL, tmp); 2150 2151 WREG32(mmCP_RB0_BASE, ring->gpu_addr >> 8); 2152 2153 /* start the rings */ 2154 gfx_v6_0_cp_gfx_start(adev); 2155 ring->ready = true; 2156 r = amdgpu_ring_test_ring(ring); 2157 if (r) { 2158 ring->ready = false; 2159 return r; 2160 } 2161 2162 return 0; 2163 } 2164 2165 static u64 gfx_v6_0_ring_get_rptr(struct amdgpu_ring *ring) 2166 { 2167 return ring->adev->wb.wb[ring->rptr_offs]; 2168 } 2169 2170 static u64 gfx_v6_0_ring_get_wptr(struct amdgpu_ring *ring) 2171 { 2172 struct amdgpu_device *adev = ring->adev; 2173 2174 if (ring == &adev->gfx.gfx_ring[0]) 2175 return RREG32(mmCP_RB0_WPTR); 2176 else if (ring == &adev->gfx.compute_ring[0]) 2177 return RREG32(mmCP_RB1_WPTR); 2178 else if (ring == &adev->gfx.compute_ring[1]) 2179 return RREG32(mmCP_RB2_WPTR); 2180 else 2181 BUG(); 2182 } 2183 2184 static void gfx_v6_0_ring_set_wptr_gfx(struct amdgpu_ring *ring) 2185 { 2186 struct amdgpu_device *adev = ring->adev; 2187 2188 WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr)); 2189 (void)RREG32(mmCP_RB0_WPTR); 2190 } 2191 2192 static void gfx_v6_0_ring_set_wptr_compute(struct amdgpu_ring *ring) 2193 { 2194 struct amdgpu_device *adev = ring->adev; 2195 2196 if (ring == &adev->gfx.compute_ring[0]) { 2197 WREG32(mmCP_RB1_WPTR, lower_32_bits(ring->wptr)); 2198 (void)RREG32(mmCP_RB1_WPTR); 2199 } else if (ring == &adev->gfx.compute_ring[1]) { 2200 WREG32(mmCP_RB2_WPTR, lower_32_bits(ring->wptr)); 2201 (void)RREG32(mmCP_RB2_WPTR); 2202 } else { 2203 BUG(); 2204 } 2205 2206 } 2207 2208 static int gfx_v6_0_cp_compute_resume(struct amdgpu_device *adev) 2209 { 2210 struct amdgpu_ring *ring; 2211 u32 tmp; 2212 u32 rb_bufsz; 2213 int i, r; 2214 u64 rptr_addr; 2215 2216 /* ring1 - compute only */ 2217 /* Set ring buffer size */ 2218 2219 ring = &adev->gfx.compute_ring[0]; 2220 rb_bufsz = order_base_2(ring->ring_size / 8); 2221 tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 2222 #ifdef __BIG_ENDIAN 2223 tmp |= BUF_SWAP_32BIT; 2224 #endif 2225 WREG32(mmCP_RB1_CNTL, tmp); 2226 2227 WREG32(mmCP_RB1_CNTL, tmp | CP_RB1_CNTL__RB_RPTR_WR_ENA_MASK); 2228 ring->wptr = 0; 2229 WREG32(mmCP_RB1_WPTR, ring->wptr); 2230 2231 rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4); 2232 WREG32(mmCP_RB1_RPTR_ADDR, lower_32_bits(rptr_addr)); 2233 WREG32(mmCP_RB1_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF); 2234 2235 mdelay(1); 2236 WREG32(mmCP_RB1_CNTL, tmp); 2237 WREG32(mmCP_RB1_BASE, ring->gpu_addr >> 8); 2238 2239 ring = &adev->gfx.compute_ring[1]; 2240 rb_bufsz = order_base_2(ring->ring_size / 8); 2241 tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 2242 #ifdef __BIG_ENDIAN 2243 tmp |= BUF_SWAP_32BIT; 2244 #endif 2245 WREG32(mmCP_RB2_CNTL, tmp); 2246 2247 WREG32(mmCP_RB2_CNTL, tmp | CP_RB2_CNTL__RB_RPTR_WR_ENA_MASK); 2248 ring->wptr = 0; 2249 WREG32(mmCP_RB2_WPTR, ring->wptr); 2250 rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4); 2251 WREG32(mmCP_RB2_RPTR_ADDR, lower_32_bits(rptr_addr)); 2252 WREG32(mmCP_RB2_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF); 2253 2254 mdelay(1); 2255 WREG32(mmCP_RB2_CNTL, tmp); 2256 WREG32(mmCP_RB2_BASE, ring->gpu_addr >> 8); 2257 2258 adev->gfx.compute_ring[0].ready = false; 2259 adev->gfx.compute_ring[1].ready = false; 2260 2261 for (i = 0; i < 2; i++) { 2262 r = amdgpu_ring_test_ring(&adev->gfx.compute_ring[i]); 2263 if (r) 2264 return r; 2265 adev->gfx.compute_ring[i].ready = true; 2266 } 2267 2268 return 0; 2269 } 2270 2271 static void gfx_v6_0_cp_enable(struct amdgpu_device *adev, bool enable) 2272 { 2273 gfx_v6_0_cp_gfx_enable(adev, enable); 2274 } 2275 2276 static int gfx_v6_0_cp_load_microcode(struct amdgpu_device *adev) 2277 { 2278 return gfx_v6_0_cp_gfx_load_microcode(adev); 2279 } 2280 2281 static void gfx_v6_0_enable_gui_idle_interrupt(struct amdgpu_device *adev, 2282 bool enable) 2283 { 2284 u32 tmp = RREG32(mmCP_INT_CNTL_RING0); 2285 u32 mask; 2286 int i; 2287 2288 if (enable) 2289 tmp |= (CP_INT_CNTL__CNTX_BUSY_INT_ENABLE_MASK | 2290 CP_INT_CNTL__CNTX_EMPTY_INT_ENABLE_MASK); 2291 else 2292 tmp &= ~(CP_INT_CNTL__CNTX_BUSY_INT_ENABLE_MASK | 2293 CP_INT_CNTL__CNTX_EMPTY_INT_ENABLE_MASK); 2294 WREG32(mmCP_INT_CNTL_RING0, tmp); 2295 2296 if (!enable) { 2297 /* read a gfx register */ 2298 tmp = RREG32(mmDB_DEPTH_INFO); 2299 2300 mask = RLC_BUSY_STATUS | GFX_POWER_STATUS | GFX_CLOCK_STATUS | GFX_LS_STATUS; 2301 for (i = 0; i < adev->usec_timeout; i++) { 2302 if ((RREG32(mmRLC_STAT) & mask) == (GFX_CLOCK_STATUS | GFX_POWER_STATUS)) 2303 break; 2304 udelay(1); 2305 } 2306 } 2307 } 2308 2309 static int gfx_v6_0_cp_resume(struct amdgpu_device *adev) 2310 { 2311 int r; 2312 2313 gfx_v6_0_enable_gui_idle_interrupt(adev, false); 2314 2315 r = gfx_v6_0_cp_load_microcode(adev); 2316 if (r) 2317 return r; 2318 2319 r = gfx_v6_0_cp_gfx_resume(adev); 2320 if (r) 2321 return r; 2322 r = gfx_v6_0_cp_compute_resume(adev); 2323 if (r) 2324 return r; 2325 2326 gfx_v6_0_enable_gui_idle_interrupt(adev, true); 2327 2328 return 0; 2329 } 2330 2331 static void gfx_v6_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring) 2332 { 2333 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 2334 uint32_t seq = ring->fence_drv.sync_seq; 2335 uint64_t addr = ring->fence_drv.gpu_addr; 2336 2337 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 2338 amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */ 2339 WAIT_REG_MEM_FUNCTION(3) | /* equal */ 2340 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */ 2341 amdgpu_ring_write(ring, addr & 0xfffffffc); 2342 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); 2343 amdgpu_ring_write(ring, seq); 2344 amdgpu_ring_write(ring, 0xffffffff); 2345 amdgpu_ring_write(ring, 4); /* poll interval */ 2346 2347 if (usepfp) { 2348 /* synce CE with ME to prevent CE fetch CEIB before context switch done */ 2349 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 2350 amdgpu_ring_write(ring, 0); 2351 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 2352 amdgpu_ring_write(ring, 0); 2353 } 2354 } 2355 2356 static void gfx_v6_0_ring_emit_vm_flush(struct amdgpu_ring *ring, 2357 unsigned vmid, uint64_t pd_addr) 2358 { 2359 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 2360 2361 /* write new base address */ 2362 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 2363 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) | 2364 WRITE_DATA_DST_SEL(0))); 2365 if (vmid < 8) { 2366 amdgpu_ring_write(ring, (mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid )); 2367 } else { 2368 amdgpu_ring_write(ring, (mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + (vmid - 8))); 2369 } 2370 amdgpu_ring_write(ring, 0); 2371 amdgpu_ring_write(ring, pd_addr >> 12); 2372 2373 /* bits 0-15 are the VM contexts0-15 */ 2374 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 2375 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) | 2376 WRITE_DATA_DST_SEL(0))); 2377 amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST); 2378 amdgpu_ring_write(ring, 0); 2379 amdgpu_ring_write(ring, 1 << vmid); 2380 2381 /* wait for the invalidate to complete */ 2382 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 2383 amdgpu_ring_write(ring, (WAIT_REG_MEM_FUNCTION(0) | /* always */ 2384 WAIT_REG_MEM_ENGINE(0))); /* me */ 2385 amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST); 2386 amdgpu_ring_write(ring, 0); 2387 amdgpu_ring_write(ring, 0); /* ref */ 2388 amdgpu_ring_write(ring, 0); /* mask */ 2389 amdgpu_ring_write(ring, 0x20); /* poll interval */ 2390 2391 if (usepfp) { 2392 /* sync PFP to ME, otherwise we might get invalid PFP reads */ 2393 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0)); 2394 amdgpu_ring_write(ring, 0x0); 2395 2396 /* synce CE with ME to prevent CE fetch CEIB before context switch done */ 2397 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 2398 amdgpu_ring_write(ring, 0); 2399 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 2400 amdgpu_ring_write(ring, 0); 2401 } 2402 } 2403 2404 2405 static void gfx_v6_0_rlc_fini(struct amdgpu_device *adev) 2406 { 2407 amdgpu_bo_free_kernel(&adev->gfx.rlc.save_restore_obj, NULL, NULL); 2408 amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj, NULL, NULL); 2409 amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj, NULL, NULL); 2410 } 2411 2412 static int gfx_v6_0_rlc_init(struct amdgpu_device *adev) 2413 { 2414 const u32 *src_ptr; 2415 volatile u32 *dst_ptr; 2416 u32 dws, i; 2417 u64 reg_list_mc_addr; 2418 const struct cs_section_def *cs_data; 2419 int r; 2420 2421 adev->gfx.rlc.reg_list = verde_rlc_save_restore_register_list; 2422 adev->gfx.rlc.reg_list_size = 2423 (u32)ARRAY_SIZE(verde_rlc_save_restore_register_list); 2424 2425 adev->gfx.rlc.cs_data = si_cs_data; 2426 src_ptr = adev->gfx.rlc.reg_list; 2427 dws = adev->gfx.rlc.reg_list_size; 2428 cs_data = adev->gfx.rlc.cs_data; 2429 2430 if (src_ptr) { 2431 /* save restore block */ 2432 r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, 2433 AMDGPU_GEM_DOMAIN_VRAM, 2434 &adev->gfx.rlc.save_restore_obj, 2435 &adev->gfx.rlc.save_restore_gpu_addr, 2436 (void **)&adev->gfx.rlc.sr_ptr); 2437 if (r) { 2438 dev_warn(adev->dev, "(%d) create RLC sr bo failed\n", 2439 r); 2440 gfx_v6_0_rlc_fini(adev); 2441 return r; 2442 } 2443 2444 /* write the sr buffer */ 2445 dst_ptr = adev->gfx.rlc.sr_ptr; 2446 for (i = 0; i < adev->gfx.rlc.reg_list_size; i++) 2447 dst_ptr[i] = cpu_to_le32(src_ptr[i]); 2448 2449 amdgpu_bo_kunmap(adev->gfx.rlc.save_restore_obj); 2450 amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj); 2451 } 2452 2453 if (cs_data) { 2454 /* clear state block */ 2455 adev->gfx.rlc.clear_state_size = gfx_v6_0_get_csb_size(adev); 2456 dws = adev->gfx.rlc.clear_state_size + (256 / 4); 2457 2458 r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, 2459 AMDGPU_GEM_DOMAIN_VRAM, 2460 &adev->gfx.rlc.clear_state_obj, 2461 &adev->gfx.rlc.clear_state_gpu_addr, 2462 (void **)&adev->gfx.rlc.cs_ptr); 2463 if (r) { 2464 dev_warn(adev->dev, "(%d) create RLC c bo failed\n", r); 2465 gfx_v6_0_rlc_fini(adev); 2466 return r; 2467 } 2468 2469 /* set up the cs buffer */ 2470 dst_ptr = adev->gfx.rlc.cs_ptr; 2471 reg_list_mc_addr = adev->gfx.rlc.clear_state_gpu_addr + 256; 2472 dst_ptr[0] = cpu_to_le32(upper_32_bits(reg_list_mc_addr)); 2473 dst_ptr[1] = cpu_to_le32(lower_32_bits(reg_list_mc_addr)); 2474 dst_ptr[2] = cpu_to_le32(adev->gfx.rlc.clear_state_size); 2475 gfx_v6_0_get_csb_buffer(adev, &dst_ptr[(256/4)]); 2476 amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj); 2477 amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj); 2478 } 2479 2480 return 0; 2481 } 2482 2483 static void gfx_v6_0_enable_lbpw(struct amdgpu_device *adev, bool enable) 2484 { 2485 WREG32_FIELD(RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0); 2486 2487 if (!enable) { 2488 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 2489 WREG32(mmSPI_LB_CU_MASK, 0x00ff); 2490 } 2491 } 2492 2493 static void gfx_v6_0_wait_for_rlc_serdes(struct amdgpu_device *adev) 2494 { 2495 int i; 2496 2497 for (i = 0; i < adev->usec_timeout; i++) { 2498 if (RREG32(mmRLC_SERDES_MASTER_BUSY_0) == 0) 2499 break; 2500 udelay(1); 2501 } 2502 2503 for (i = 0; i < adev->usec_timeout; i++) { 2504 if (RREG32(mmRLC_SERDES_MASTER_BUSY_1) == 0) 2505 break; 2506 udelay(1); 2507 } 2508 } 2509 2510 static void gfx_v6_0_update_rlc(struct amdgpu_device *adev, u32 rlc) 2511 { 2512 u32 tmp; 2513 2514 tmp = RREG32(mmRLC_CNTL); 2515 if (tmp != rlc) 2516 WREG32(mmRLC_CNTL, rlc); 2517 } 2518 2519 static u32 gfx_v6_0_halt_rlc(struct amdgpu_device *adev) 2520 { 2521 u32 data, orig; 2522 2523 orig = data = RREG32(mmRLC_CNTL); 2524 2525 if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) { 2526 data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK; 2527 WREG32(mmRLC_CNTL, data); 2528 2529 gfx_v6_0_wait_for_rlc_serdes(adev); 2530 } 2531 2532 return orig; 2533 } 2534 2535 static void gfx_v6_0_rlc_stop(struct amdgpu_device *adev) 2536 { 2537 WREG32(mmRLC_CNTL, 0); 2538 2539 gfx_v6_0_enable_gui_idle_interrupt(adev, false); 2540 gfx_v6_0_wait_for_rlc_serdes(adev); 2541 } 2542 2543 static void gfx_v6_0_rlc_start(struct amdgpu_device *adev) 2544 { 2545 WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK); 2546 2547 gfx_v6_0_enable_gui_idle_interrupt(adev, true); 2548 2549 udelay(50); 2550 } 2551 2552 static void gfx_v6_0_rlc_reset(struct amdgpu_device *adev) 2553 { 2554 WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 1); 2555 udelay(50); 2556 WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 0); 2557 udelay(50); 2558 } 2559 2560 static bool gfx_v6_0_lbpw_supported(struct amdgpu_device *adev) 2561 { 2562 u32 tmp; 2563 2564 /* Enable LBPW only for DDR3 */ 2565 tmp = RREG32(mmMC_SEQ_MISC0); 2566 if ((tmp & 0xF0000000) == 0xB0000000) 2567 return true; 2568 return false; 2569 } 2570 2571 static void gfx_v6_0_init_cg(struct amdgpu_device *adev) 2572 { 2573 } 2574 2575 static int gfx_v6_0_rlc_resume(struct amdgpu_device *adev) 2576 { 2577 u32 i; 2578 const struct rlc_firmware_header_v1_0 *hdr; 2579 const __le32 *fw_data; 2580 u32 fw_size; 2581 2582 2583 if (!adev->gfx.rlc_fw) 2584 return -EINVAL; 2585 2586 gfx_v6_0_rlc_stop(adev); 2587 gfx_v6_0_rlc_reset(adev); 2588 gfx_v6_0_init_pg(adev); 2589 gfx_v6_0_init_cg(adev); 2590 2591 WREG32(mmRLC_RL_BASE, 0); 2592 WREG32(mmRLC_RL_SIZE, 0); 2593 WREG32(mmRLC_LB_CNTL, 0); 2594 WREG32(mmRLC_LB_CNTR_MAX, 0xffffffff); 2595 WREG32(mmRLC_LB_CNTR_INIT, 0); 2596 WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff); 2597 2598 WREG32(mmRLC_MC_CNTL, 0); 2599 WREG32(mmRLC_UCODE_CNTL, 0); 2600 2601 hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data; 2602 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; 2603 fw_data = (const __le32 *) 2604 (adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2605 2606 amdgpu_ucode_print_rlc_hdr(&hdr->header); 2607 2608 for (i = 0; i < fw_size; i++) { 2609 WREG32(mmRLC_UCODE_ADDR, i); 2610 WREG32(mmRLC_UCODE_DATA, le32_to_cpup(fw_data++)); 2611 } 2612 WREG32(mmRLC_UCODE_ADDR, 0); 2613 2614 gfx_v6_0_enable_lbpw(adev, gfx_v6_0_lbpw_supported(adev)); 2615 gfx_v6_0_rlc_start(adev); 2616 2617 return 0; 2618 } 2619 2620 static void gfx_v6_0_enable_cgcg(struct amdgpu_device *adev, bool enable) 2621 { 2622 u32 data, orig, tmp; 2623 2624 orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL); 2625 2626 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { 2627 gfx_v6_0_enable_gui_idle_interrupt(adev, true); 2628 2629 WREG32(mmRLC_GCPM_GENERAL_3, 0x00000080); 2630 2631 tmp = gfx_v6_0_halt_rlc(adev); 2632 2633 WREG32(mmRLC_SERDES_WR_MASTER_MASK_0, 0xffffffff); 2634 WREG32(mmRLC_SERDES_WR_MASTER_MASK_1, 0xffffffff); 2635 WREG32(mmRLC_SERDES_WR_CTRL, 0x00b000ff); 2636 2637 gfx_v6_0_wait_for_rlc_serdes(adev); 2638 gfx_v6_0_update_rlc(adev, tmp); 2639 2640 WREG32(mmRLC_SERDES_WR_CTRL, 0x007000ff); 2641 2642 data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK; 2643 } else { 2644 gfx_v6_0_enable_gui_idle_interrupt(adev, false); 2645 2646 RREG32(mmCB_CGTT_SCLK_CTRL); 2647 RREG32(mmCB_CGTT_SCLK_CTRL); 2648 RREG32(mmCB_CGTT_SCLK_CTRL); 2649 RREG32(mmCB_CGTT_SCLK_CTRL); 2650 2651 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK); 2652 } 2653 2654 if (orig != data) 2655 WREG32(mmRLC_CGCG_CGLS_CTRL, data); 2656 2657 } 2658 2659 static void gfx_v6_0_enable_mgcg(struct amdgpu_device *adev, bool enable) 2660 { 2661 2662 u32 data, orig, tmp = 0; 2663 2664 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) { 2665 orig = data = RREG32(mmCGTS_SM_CTRL_REG); 2666 data = 0x96940200; 2667 if (orig != data) 2668 WREG32(mmCGTS_SM_CTRL_REG, data); 2669 2670 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) { 2671 orig = data = RREG32(mmCP_MEM_SLP_CNTL); 2672 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 2673 if (orig != data) 2674 WREG32(mmCP_MEM_SLP_CNTL, data); 2675 } 2676 2677 orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 2678 data &= 0xffffffc0; 2679 if (orig != data) 2680 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data); 2681 2682 tmp = gfx_v6_0_halt_rlc(adev); 2683 2684 WREG32(mmRLC_SERDES_WR_MASTER_MASK_0, 0xffffffff); 2685 WREG32(mmRLC_SERDES_WR_MASTER_MASK_1, 0xffffffff); 2686 WREG32(mmRLC_SERDES_WR_CTRL, 0x00d000ff); 2687 2688 gfx_v6_0_update_rlc(adev, tmp); 2689 } else { 2690 orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 2691 data |= 0x00000003; 2692 if (orig != data) 2693 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data); 2694 2695 data = RREG32(mmCP_MEM_SLP_CNTL); 2696 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) { 2697 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 2698 WREG32(mmCP_MEM_SLP_CNTL, data); 2699 } 2700 orig = data = RREG32(mmCGTS_SM_CTRL_REG); 2701 data |= CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK | CGTS_SM_CTRL_REG__OVERRIDE_MASK; 2702 if (orig != data) 2703 WREG32(mmCGTS_SM_CTRL_REG, data); 2704 2705 tmp = gfx_v6_0_halt_rlc(adev); 2706 2707 WREG32(mmRLC_SERDES_WR_MASTER_MASK_0, 0xffffffff); 2708 WREG32(mmRLC_SERDES_WR_MASTER_MASK_1, 0xffffffff); 2709 WREG32(mmRLC_SERDES_WR_CTRL, 0x00e000ff); 2710 2711 gfx_v6_0_update_rlc(adev, tmp); 2712 } 2713 } 2714 /* 2715 static void gfx_v6_0_update_cg(struct amdgpu_device *adev, 2716 bool enable) 2717 { 2718 gfx_v6_0_enable_gui_idle_interrupt(adev, false); 2719 if (enable) { 2720 gfx_v6_0_enable_mgcg(adev, true); 2721 gfx_v6_0_enable_cgcg(adev, true); 2722 } else { 2723 gfx_v6_0_enable_cgcg(adev, false); 2724 gfx_v6_0_enable_mgcg(adev, false); 2725 } 2726 gfx_v6_0_enable_gui_idle_interrupt(adev, true); 2727 } 2728 */ 2729 2730 static void gfx_v6_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev, 2731 bool enable) 2732 { 2733 } 2734 2735 static void gfx_v6_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev, 2736 bool enable) 2737 { 2738 } 2739 2740 static void gfx_v6_0_enable_cp_pg(struct amdgpu_device *adev, bool enable) 2741 { 2742 u32 data, orig; 2743 2744 orig = data = RREG32(mmRLC_PG_CNTL); 2745 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP)) 2746 data &= ~0x8000; 2747 else 2748 data |= 0x8000; 2749 if (orig != data) 2750 WREG32(mmRLC_PG_CNTL, data); 2751 } 2752 2753 static void gfx_v6_0_enable_gds_pg(struct amdgpu_device *adev, bool enable) 2754 { 2755 } 2756 /* 2757 static void gfx_v6_0_init_cp_pg_table(struct amdgpu_device *adev) 2758 { 2759 const __le32 *fw_data; 2760 volatile u32 *dst_ptr; 2761 int me, i, max_me = 4; 2762 u32 bo_offset = 0; 2763 u32 table_offset, table_size; 2764 2765 if (adev->asic_type == CHIP_KAVERI) 2766 max_me = 5; 2767 2768 if (adev->gfx.rlc.cp_table_ptr == NULL) 2769 return; 2770 2771 dst_ptr = adev->gfx.rlc.cp_table_ptr; 2772 for (me = 0; me < max_me; me++) { 2773 if (me == 0) { 2774 const struct gfx_firmware_header_v1_0 *hdr = 2775 (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; 2776 fw_data = (const __le32 *) 2777 (adev->gfx.ce_fw->data + 2778 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2779 table_offset = le32_to_cpu(hdr->jt_offset); 2780 table_size = le32_to_cpu(hdr->jt_size); 2781 } else if (me == 1) { 2782 const struct gfx_firmware_header_v1_0 *hdr = 2783 (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; 2784 fw_data = (const __le32 *) 2785 (adev->gfx.pfp_fw->data + 2786 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2787 table_offset = le32_to_cpu(hdr->jt_offset); 2788 table_size = le32_to_cpu(hdr->jt_size); 2789 } else if (me == 2) { 2790 const struct gfx_firmware_header_v1_0 *hdr = 2791 (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; 2792 fw_data = (const __le32 *) 2793 (adev->gfx.me_fw->data + 2794 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2795 table_offset = le32_to_cpu(hdr->jt_offset); 2796 table_size = le32_to_cpu(hdr->jt_size); 2797 } else if (me == 3) { 2798 const struct gfx_firmware_header_v1_0 *hdr = 2799 (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; 2800 fw_data = (const __le32 *) 2801 (adev->gfx.mec_fw->data + 2802 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2803 table_offset = le32_to_cpu(hdr->jt_offset); 2804 table_size = le32_to_cpu(hdr->jt_size); 2805 } else { 2806 const struct gfx_firmware_header_v1_0 *hdr = 2807 (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; 2808 fw_data = (const __le32 *) 2809 (adev->gfx.mec2_fw->data + 2810 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2811 table_offset = le32_to_cpu(hdr->jt_offset); 2812 table_size = le32_to_cpu(hdr->jt_size); 2813 } 2814 2815 for (i = 0; i < table_size; i ++) { 2816 dst_ptr[bo_offset + i] = 2817 cpu_to_le32(le32_to_cpu(fw_data[table_offset + i])); 2818 } 2819 2820 bo_offset += table_size; 2821 } 2822 } 2823 */ 2824 static void gfx_v6_0_enable_gfx_cgpg(struct amdgpu_device *adev, 2825 bool enable) 2826 { 2827 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) { 2828 WREG32(mmRLC_TTOP_D, RLC_PUD(0x10) | RLC_PDD(0x10) | RLC_TTPD(0x10) | RLC_MSD(0x10)); 2829 WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_ENABLE, 1); 2830 WREG32_FIELD(RLC_AUTO_PG_CTRL, AUTO_PG_EN, 1); 2831 } else { 2832 WREG32_FIELD(RLC_AUTO_PG_CTRL, AUTO_PG_EN, 0); 2833 (void)RREG32(mmDB_RENDER_CONTROL); 2834 } 2835 } 2836 2837 static void gfx_v6_0_init_ao_cu_mask(struct amdgpu_device *adev) 2838 { 2839 u32 tmp; 2840 2841 WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask); 2842 2843 tmp = RREG32(mmRLC_MAX_PG_CU); 2844 tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK; 2845 tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT); 2846 WREG32(mmRLC_MAX_PG_CU, tmp); 2847 } 2848 2849 static void gfx_v6_0_enable_gfx_static_mgpg(struct amdgpu_device *adev, 2850 bool enable) 2851 { 2852 u32 data, orig; 2853 2854 orig = data = RREG32(mmRLC_PG_CNTL); 2855 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG)) 2856 data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK; 2857 else 2858 data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK; 2859 if (orig != data) 2860 WREG32(mmRLC_PG_CNTL, data); 2861 } 2862 2863 static void gfx_v6_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev, 2864 bool enable) 2865 { 2866 u32 data, orig; 2867 2868 orig = data = RREG32(mmRLC_PG_CNTL); 2869 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG)) 2870 data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK; 2871 else 2872 data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK; 2873 if (orig != data) 2874 WREG32(mmRLC_PG_CNTL, data); 2875 } 2876 2877 static void gfx_v6_0_init_gfx_cgpg(struct amdgpu_device *adev) 2878 { 2879 u32 tmp; 2880 2881 WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8); 2882 WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_SRC, 1); 2883 WREG32(mmRLC_CLEAR_STATE_RESTORE_BASE, adev->gfx.rlc.clear_state_gpu_addr >> 8); 2884 2885 tmp = RREG32(mmRLC_AUTO_PG_CTRL); 2886 tmp &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK; 2887 tmp |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT); 2888 tmp &= ~RLC_AUTO_PG_CTRL__PG_AFTER_GRBM_REG_SAVE_THRESHOLD_MASK; 2889 WREG32(mmRLC_AUTO_PG_CTRL, tmp); 2890 } 2891 2892 static void gfx_v6_0_update_gfx_pg(struct amdgpu_device *adev, bool enable) 2893 { 2894 gfx_v6_0_enable_gfx_cgpg(adev, enable); 2895 gfx_v6_0_enable_gfx_static_mgpg(adev, enable); 2896 gfx_v6_0_enable_gfx_dynamic_mgpg(adev, enable); 2897 } 2898 2899 static u32 gfx_v6_0_get_csb_size(struct amdgpu_device *adev) 2900 { 2901 u32 count = 0; 2902 const struct cs_section_def *sect = NULL; 2903 const struct cs_extent_def *ext = NULL; 2904 2905 if (adev->gfx.rlc.cs_data == NULL) 2906 return 0; 2907 2908 /* begin clear state */ 2909 count += 2; 2910 /* context control state */ 2911 count += 3; 2912 2913 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 2914 for (ext = sect->section; ext->extent != NULL; ++ext) { 2915 if (sect->id == SECT_CONTEXT) 2916 count += 2 + ext->reg_count; 2917 else 2918 return 0; 2919 } 2920 } 2921 /* pa_sc_raster_config */ 2922 count += 3; 2923 /* end clear state */ 2924 count += 2; 2925 /* clear state */ 2926 count += 2; 2927 2928 return count; 2929 } 2930 2931 static void gfx_v6_0_get_csb_buffer(struct amdgpu_device *adev, 2932 volatile u32 *buffer) 2933 { 2934 u32 count = 0, i; 2935 const struct cs_section_def *sect = NULL; 2936 const struct cs_extent_def *ext = NULL; 2937 2938 if (adev->gfx.rlc.cs_data == NULL) 2939 return; 2940 if (buffer == NULL) 2941 return; 2942 2943 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 2944 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 2945 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 2946 buffer[count++] = cpu_to_le32(0x80000000); 2947 buffer[count++] = cpu_to_le32(0x80000000); 2948 2949 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 2950 for (ext = sect->section; ext->extent != NULL; ++ext) { 2951 if (sect->id == SECT_CONTEXT) { 2952 buffer[count++] = 2953 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count)); 2954 buffer[count++] = cpu_to_le32(ext->reg_index - 0xa000); 2955 for (i = 0; i < ext->reg_count; i++) 2956 buffer[count++] = cpu_to_le32(ext->extent[i]); 2957 } else { 2958 return; 2959 } 2960 } 2961 } 2962 2963 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 1)); 2964 buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START); 2965 buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config); 2966 2967 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 2968 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE); 2969 2970 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0)); 2971 buffer[count++] = cpu_to_le32(0); 2972 } 2973 2974 static void gfx_v6_0_init_pg(struct amdgpu_device *adev) 2975 { 2976 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | 2977 AMD_PG_SUPPORT_GFX_SMG | 2978 AMD_PG_SUPPORT_GFX_DMG | 2979 AMD_PG_SUPPORT_CP | 2980 AMD_PG_SUPPORT_GDS | 2981 AMD_PG_SUPPORT_RLC_SMU_HS)) { 2982 gfx_v6_0_enable_sclk_slowdown_on_pu(adev, true); 2983 gfx_v6_0_enable_sclk_slowdown_on_pd(adev, true); 2984 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { 2985 gfx_v6_0_init_gfx_cgpg(adev); 2986 gfx_v6_0_enable_cp_pg(adev, true); 2987 gfx_v6_0_enable_gds_pg(adev, true); 2988 } else { 2989 WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8); 2990 WREG32(mmRLC_CLEAR_STATE_RESTORE_BASE, adev->gfx.rlc.clear_state_gpu_addr >> 8); 2991 2992 } 2993 gfx_v6_0_init_ao_cu_mask(adev); 2994 gfx_v6_0_update_gfx_pg(adev, true); 2995 } else { 2996 2997 WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8); 2998 WREG32(mmRLC_CLEAR_STATE_RESTORE_BASE, adev->gfx.rlc.clear_state_gpu_addr >> 8); 2999 } 3000 } 3001 3002 static void gfx_v6_0_fini_pg(struct amdgpu_device *adev) 3003 { 3004 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | 3005 AMD_PG_SUPPORT_GFX_SMG | 3006 AMD_PG_SUPPORT_GFX_DMG | 3007 AMD_PG_SUPPORT_CP | 3008 AMD_PG_SUPPORT_GDS | 3009 AMD_PG_SUPPORT_RLC_SMU_HS)) { 3010 gfx_v6_0_update_gfx_pg(adev, false); 3011 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { 3012 gfx_v6_0_enable_cp_pg(adev, false); 3013 gfx_v6_0_enable_gds_pg(adev, false); 3014 } 3015 } 3016 } 3017 3018 static uint64_t gfx_v6_0_get_gpu_clock_counter(struct amdgpu_device *adev) 3019 { 3020 uint64_t clock; 3021 3022 mutex_lock(&adev->gfx.gpu_clock_mutex); 3023 WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1); 3024 clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) | 3025 ((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL); 3026 mutex_unlock(&adev->gfx.gpu_clock_mutex); 3027 return clock; 3028 } 3029 3030 static void gfx_v6_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags) 3031 { 3032 if (flags & AMDGPU_HAVE_CTX_SWITCH) 3033 gfx_v6_0_ring_emit_vgt_flush(ring); 3034 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 3035 amdgpu_ring_write(ring, 0x80000000); 3036 amdgpu_ring_write(ring, 0); 3037 } 3038 3039 3040 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address) 3041 { 3042 WREG32(mmSQ_IND_INDEX, 3043 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 3044 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 3045 (address << SQ_IND_INDEX__INDEX__SHIFT) | 3046 (SQ_IND_INDEX__FORCE_READ_MASK)); 3047 return RREG32(mmSQ_IND_DATA); 3048 } 3049 3050 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd, 3051 uint32_t wave, uint32_t thread, 3052 uint32_t regno, uint32_t num, uint32_t *out) 3053 { 3054 WREG32(mmSQ_IND_INDEX, 3055 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 3056 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 3057 (regno << SQ_IND_INDEX__INDEX__SHIFT) | 3058 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) | 3059 (SQ_IND_INDEX__FORCE_READ_MASK) | 3060 (SQ_IND_INDEX__AUTO_INCR_MASK)); 3061 while (num--) 3062 *(out++) = RREG32(mmSQ_IND_DATA); 3063 } 3064 3065 static void gfx_v6_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields) 3066 { 3067 /* type 0 wave data */ 3068 dst[(*no_fields)++] = 0; 3069 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS); 3070 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO); 3071 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI); 3072 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO); 3073 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI); 3074 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID); 3075 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0); 3076 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1); 3077 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC); 3078 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC); 3079 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS); 3080 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS); 3081 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO); 3082 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI); 3083 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO); 3084 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI); 3085 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0); 3086 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0); 3087 } 3088 3089 static void gfx_v6_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd, 3090 uint32_t wave, uint32_t start, 3091 uint32_t size, uint32_t *dst) 3092 { 3093 wave_read_regs( 3094 adev, simd, wave, 0, 3095 start + SQIND_WAVE_SGPRS_OFFSET, size, dst); 3096 } 3097 3098 static const struct amdgpu_gfx_funcs gfx_v6_0_gfx_funcs = { 3099 .get_gpu_clock_counter = &gfx_v6_0_get_gpu_clock_counter, 3100 .select_se_sh = &gfx_v6_0_select_se_sh, 3101 .read_wave_data = &gfx_v6_0_read_wave_data, 3102 .read_wave_sgprs = &gfx_v6_0_read_wave_sgprs, 3103 }; 3104 3105 static int gfx_v6_0_early_init(void *handle) 3106 { 3107 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3108 3109 adev->gfx.num_gfx_rings = GFX6_NUM_GFX_RINGS; 3110 adev->gfx.num_compute_rings = GFX6_NUM_COMPUTE_RINGS; 3111 adev->gfx.funcs = &gfx_v6_0_gfx_funcs; 3112 gfx_v6_0_set_ring_funcs(adev); 3113 gfx_v6_0_set_irq_funcs(adev); 3114 3115 return 0; 3116 } 3117 3118 static int gfx_v6_0_sw_init(void *handle) 3119 { 3120 struct amdgpu_ring *ring; 3121 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3122 int i, r; 3123 3124 r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 181, &adev->gfx.eop_irq); 3125 if (r) 3126 return r; 3127 3128 r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 184, &adev->gfx.priv_reg_irq); 3129 if (r) 3130 return r; 3131 3132 r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_LEGACY, 185, &adev->gfx.priv_inst_irq); 3133 if (r) 3134 return r; 3135 3136 gfx_v6_0_scratch_init(adev); 3137 3138 r = gfx_v6_0_init_microcode(adev); 3139 if (r) { 3140 DRM_ERROR("Failed to load gfx firmware!\n"); 3141 return r; 3142 } 3143 3144 r = gfx_v6_0_rlc_init(adev); 3145 if (r) { 3146 DRM_ERROR("Failed to init rlc BOs!\n"); 3147 return r; 3148 } 3149 3150 for (i = 0; i < adev->gfx.num_gfx_rings; i++) { 3151 ring = &adev->gfx.gfx_ring[i]; 3152 ring->ring_obj = NULL; 3153 sprintf(ring->name, "gfx"); 3154 r = amdgpu_ring_init(adev, ring, 1024, 3155 &adev->gfx.eop_irq, AMDGPU_CP_IRQ_GFX_EOP); 3156 if (r) 3157 return r; 3158 } 3159 3160 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 3161 unsigned irq_type; 3162 3163 if ((i >= 32) || (i >= AMDGPU_MAX_COMPUTE_RINGS)) { 3164 DRM_ERROR("Too many (%d) compute rings!\n", i); 3165 break; 3166 } 3167 ring = &adev->gfx.compute_ring[i]; 3168 ring->ring_obj = NULL; 3169 ring->use_doorbell = false; 3170 ring->doorbell_index = 0; 3171 ring->me = 1; 3172 ring->pipe = i; 3173 ring->queue = i; 3174 sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue); 3175 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP + ring->pipe; 3176 r = amdgpu_ring_init(adev, ring, 1024, 3177 &adev->gfx.eop_irq, irq_type); 3178 if (r) 3179 return r; 3180 } 3181 3182 return r; 3183 } 3184 3185 static int gfx_v6_0_sw_fini(void *handle) 3186 { 3187 int i; 3188 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3189 3190 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 3191 amdgpu_ring_fini(&adev->gfx.gfx_ring[i]); 3192 for (i = 0; i < adev->gfx.num_compute_rings; i++) 3193 amdgpu_ring_fini(&adev->gfx.compute_ring[i]); 3194 3195 gfx_v6_0_rlc_fini(adev); 3196 3197 return 0; 3198 } 3199 3200 static int gfx_v6_0_hw_init(void *handle) 3201 { 3202 int r; 3203 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3204 3205 gfx_v6_0_gpu_init(adev); 3206 3207 r = gfx_v6_0_rlc_resume(adev); 3208 if (r) 3209 return r; 3210 3211 r = gfx_v6_0_cp_resume(adev); 3212 if (r) 3213 return r; 3214 3215 adev->gfx.ce_ram_size = 0x8000; 3216 3217 return r; 3218 } 3219 3220 static int gfx_v6_0_hw_fini(void *handle) 3221 { 3222 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3223 3224 gfx_v6_0_cp_enable(adev, false); 3225 gfx_v6_0_rlc_stop(adev); 3226 gfx_v6_0_fini_pg(adev); 3227 3228 return 0; 3229 } 3230 3231 static int gfx_v6_0_suspend(void *handle) 3232 { 3233 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3234 3235 return gfx_v6_0_hw_fini(adev); 3236 } 3237 3238 static int gfx_v6_0_resume(void *handle) 3239 { 3240 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3241 3242 return gfx_v6_0_hw_init(adev); 3243 } 3244 3245 static bool gfx_v6_0_is_idle(void *handle) 3246 { 3247 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3248 3249 if (RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK) 3250 return false; 3251 else 3252 return true; 3253 } 3254 3255 static int gfx_v6_0_wait_for_idle(void *handle) 3256 { 3257 unsigned i; 3258 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3259 3260 for (i = 0; i < adev->usec_timeout; i++) { 3261 if (gfx_v6_0_is_idle(handle)) 3262 return 0; 3263 udelay(1); 3264 } 3265 return -ETIMEDOUT; 3266 } 3267 3268 static int gfx_v6_0_soft_reset(void *handle) 3269 { 3270 return 0; 3271 } 3272 3273 static void gfx_v6_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev, 3274 enum amdgpu_interrupt_state state) 3275 { 3276 u32 cp_int_cntl; 3277 3278 switch (state) { 3279 case AMDGPU_IRQ_STATE_DISABLE: 3280 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3281 cp_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK; 3282 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3283 break; 3284 case AMDGPU_IRQ_STATE_ENABLE: 3285 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3286 cp_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK; 3287 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3288 break; 3289 default: 3290 break; 3291 } 3292 } 3293 3294 static void gfx_v6_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev, 3295 int ring, 3296 enum amdgpu_interrupt_state state) 3297 { 3298 u32 cp_int_cntl; 3299 switch (state){ 3300 case AMDGPU_IRQ_STATE_DISABLE: 3301 if (ring == 0) { 3302 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING1); 3303 cp_int_cntl &= ~CP_INT_CNTL_RING1__TIME_STAMP_INT_ENABLE_MASK; 3304 WREG32(mmCP_INT_CNTL_RING1, cp_int_cntl); 3305 break; 3306 } else { 3307 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING2); 3308 cp_int_cntl &= ~CP_INT_CNTL_RING2__TIME_STAMP_INT_ENABLE_MASK; 3309 WREG32(mmCP_INT_CNTL_RING2, cp_int_cntl); 3310 break; 3311 3312 } 3313 case AMDGPU_IRQ_STATE_ENABLE: 3314 if (ring == 0) { 3315 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING1); 3316 cp_int_cntl |= CP_INT_CNTL_RING1__TIME_STAMP_INT_ENABLE_MASK; 3317 WREG32(mmCP_INT_CNTL_RING1, cp_int_cntl); 3318 break; 3319 } else { 3320 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING2); 3321 cp_int_cntl |= CP_INT_CNTL_RING2__TIME_STAMP_INT_ENABLE_MASK; 3322 WREG32(mmCP_INT_CNTL_RING2, cp_int_cntl); 3323 break; 3324 3325 } 3326 3327 default: 3328 BUG(); 3329 break; 3330 3331 } 3332 } 3333 3334 static int gfx_v6_0_set_priv_reg_fault_state(struct amdgpu_device *adev, 3335 struct amdgpu_irq_src *src, 3336 unsigned type, 3337 enum amdgpu_interrupt_state state) 3338 { 3339 u32 cp_int_cntl; 3340 3341 switch (state) { 3342 case AMDGPU_IRQ_STATE_DISABLE: 3343 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3344 cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK; 3345 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3346 break; 3347 case AMDGPU_IRQ_STATE_ENABLE: 3348 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3349 cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK; 3350 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3351 break; 3352 default: 3353 break; 3354 } 3355 3356 return 0; 3357 } 3358 3359 static int gfx_v6_0_set_priv_inst_fault_state(struct amdgpu_device *adev, 3360 struct amdgpu_irq_src *src, 3361 unsigned type, 3362 enum amdgpu_interrupt_state state) 3363 { 3364 u32 cp_int_cntl; 3365 3366 switch (state) { 3367 case AMDGPU_IRQ_STATE_DISABLE: 3368 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3369 cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK; 3370 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3371 break; 3372 case AMDGPU_IRQ_STATE_ENABLE: 3373 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3374 cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK; 3375 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3376 break; 3377 default: 3378 break; 3379 } 3380 3381 return 0; 3382 } 3383 3384 static int gfx_v6_0_set_eop_interrupt_state(struct amdgpu_device *adev, 3385 struct amdgpu_irq_src *src, 3386 unsigned type, 3387 enum amdgpu_interrupt_state state) 3388 { 3389 switch (type) { 3390 case AMDGPU_CP_IRQ_GFX_EOP: 3391 gfx_v6_0_set_gfx_eop_interrupt_state(adev, state); 3392 break; 3393 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP: 3394 gfx_v6_0_set_compute_eop_interrupt_state(adev, 0, state); 3395 break; 3396 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP: 3397 gfx_v6_0_set_compute_eop_interrupt_state(adev, 1, state); 3398 break; 3399 default: 3400 break; 3401 } 3402 return 0; 3403 } 3404 3405 static int gfx_v6_0_eop_irq(struct amdgpu_device *adev, 3406 struct amdgpu_irq_src *source, 3407 struct amdgpu_iv_entry *entry) 3408 { 3409 switch (entry->ring_id) { 3410 case 0: 3411 amdgpu_fence_process(&adev->gfx.gfx_ring[0]); 3412 break; 3413 case 1: 3414 case 2: 3415 amdgpu_fence_process(&adev->gfx.compute_ring[entry->ring_id - 1]); 3416 break; 3417 default: 3418 break; 3419 } 3420 return 0; 3421 } 3422 3423 static int gfx_v6_0_priv_reg_irq(struct amdgpu_device *adev, 3424 struct amdgpu_irq_src *source, 3425 struct amdgpu_iv_entry *entry) 3426 { 3427 DRM_ERROR("Illegal register access in command stream\n"); 3428 schedule_work(&adev->reset_work); 3429 return 0; 3430 } 3431 3432 static int gfx_v6_0_priv_inst_irq(struct amdgpu_device *adev, 3433 struct amdgpu_irq_src *source, 3434 struct amdgpu_iv_entry *entry) 3435 { 3436 DRM_ERROR("Illegal instruction in command stream\n"); 3437 schedule_work(&adev->reset_work); 3438 return 0; 3439 } 3440 3441 static int gfx_v6_0_set_clockgating_state(void *handle, 3442 enum amd_clockgating_state state) 3443 { 3444 bool gate = false; 3445 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3446 3447 if (state == AMD_CG_STATE_GATE) 3448 gate = true; 3449 3450 gfx_v6_0_enable_gui_idle_interrupt(adev, false); 3451 if (gate) { 3452 gfx_v6_0_enable_mgcg(adev, true); 3453 gfx_v6_0_enable_cgcg(adev, true); 3454 } else { 3455 gfx_v6_0_enable_cgcg(adev, false); 3456 gfx_v6_0_enable_mgcg(adev, false); 3457 } 3458 gfx_v6_0_enable_gui_idle_interrupt(adev, true); 3459 3460 return 0; 3461 } 3462 3463 static int gfx_v6_0_set_powergating_state(void *handle, 3464 enum amd_powergating_state state) 3465 { 3466 bool gate = false; 3467 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3468 3469 if (state == AMD_PG_STATE_GATE) 3470 gate = true; 3471 3472 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | 3473 AMD_PG_SUPPORT_GFX_SMG | 3474 AMD_PG_SUPPORT_GFX_DMG | 3475 AMD_PG_SUPPORT_CP | 3476 AMD_PG_SUPPORT_GDS | 3477 AMD_PG_SUPPORT_RLC_SMU_HS)) { 3478 gfx_v6_0_update_gfx_pg(adev, gate); 3479 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { 3480 gfx_v6_0_enable_cp_pg(adev, gate); 3481 gfx_v6_0_enable_gds_pg(adev, gate); 3482 } 3483 } 3484 3485 return 0; 3486 } 3487 3488 static const struct amd_ip_funcs gfx_v6_0_ip_funcs = { 3489 .name = "gfx_v6_0", 3490 .early_init = gfx_v6_0_early_init, 3491 .late_init = NULL, 3492 .sw_init = gfx_v6_0_sw_init, 3493 .sw_fini = gfx_v6_0_sw_fini, 3494 .hw_init = gfx_v6_0_hw_init, 3495 .hw_fini = gfx_v6_0_hw_fini, 3496 .suspend = gfx_v6_0_suspend, 3497 .resume = gfx_v6_0_resume, 3498 .is_idle = gfx_v6_0_is_idle, 3499 .wait_for_idle = gfx_v6_0_wait_for_idle, 3500 .soft_reset = gfx_v6_0_soft_reset, 3501 .set_clockgating_state = gfx_v6_0_set_clockgating_state, 3502 .set_powergating_state = gfx_v6_0_set_powergating_state, 3503 }; 3504 3505 static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_gfx = { 3506 .type = AMDGPU_RING_TYPE_GFX, 3507 .align_mask = 0xff, 3508 .nop = 0x80000000, 3509 .support_64bit_ptrs = false, 3510 .get_rptr = gfx_v6_0_ring_get_rptr, 3511 .get_wptr = gfx_v6_0_ring_get_wptr, 3512 .set_wptr = gfx_v6_0_ring_set_wptr_gfx, 3513 .emit_frame_size = 3514 5 + /* gfx_v6_0_ring_emit_hdp_flush */ 3515 5 + /* gfx_v6_0_ring_emit_hdp_invalidate */ 3516 14 + 14 + 14 + /* gfx_v6_0_ring_emit_fence x3 for user fence, vm fence */ 3517 7 + 4 + /* gfx_v6_0_ring_emit_pipeline_sync */ 3518 17 + 6 + /* gfx_v6_0_ring_emit_vm_flush */ 3519 3 + 2, /* gfx_v6_ring_emit_cntxcntl including vgt flush */ 3520 .emit_ib_size = 6, /* gfx_v6_0_ring_emit_ib */ 3521 .emit_ib = gfx_v6_0_ring_emit_ib, 3522 .emit_fence = gfx_v6_0_ring_emit_fence, 3523 .emit_pipeline_sync = gfx_v6_0_ring_emit_pipeline_sync, 3524 .emit_vm_flush = gfx_v6_0_ring_emit_vm_flush, 3525 .emit_hdp_flush = gfx_v6_0_ring_emit_hdp_flush, 3526 .emit_hdp_invalidate = gfx_v6_0_ring_emit_hdp_invalidate, 3527 .test_ring = gfx_v6_0_ring_test_ring, 3528 .test_ib = gfx_v6_0_ring_test_ib, 3529 .insert_nop = amdgpu_ring_insert_nop, 3530 .emit_cntxcntl = gfx_v6_ring_emit_cntxcntl, 3531 }; 3532 3533 static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_compute = { 3534 .type = AMDGPU_RING_TYPE_COMPUTE, 3535 .align_mask = 0xff, 3536 .nop = 0x80000000, 3537 .get_rptr = gfx_v6_0_ring_get_rptr, 3538 .get_wptr = gfx_v6_0_ring_get_wptr, 3539 .set_wptr = gfx_v6_0_ring_set_wptr_compute, 3540 .emit_frame_size = 3541 5 + /* gfx_v6_0_ring_emit_hdp_flush */ 3542 5 + /* gfx_v6_0_ring_emit_hdp_invalidate */ 3543 7 + /* gfx_v6_0_ring_emit_pipeline_sync */ 3544 17 + /* gfx_v6_0_ring_emit_vm_flush */ 3545 14 + 14 + 14, /* gfx_v6_0_ring_emit_fence x3 for user fence, vm fence */ 3546 .emit_ib_size = 6, /* gfx_v6_0_ring_emit_ib */ 3547 .emit_ib = gfx_v6_0_ring_emit_ib, 3548 .emit_fence = gfx_v6_0_ring_emit_fence, 3549 .emit_pipeline_sync = gfx_v6_0_ring_emit_pipeline_sync, 3550 .emit_vm_flush = gfx_v6_0_ring_emit_vm_flush, 3551 .emit_hdp_flush = gfx_v6_0_ring_emit_hdp_flush, 3552 .emit_hdp_invalidate = gfx_v6_0_ring_emit_hdp_invalidate, 3553 .test_ring = gfx_v6_0_ring_test_ring, 3554 .test_ib = gfx_v6_0_ring_test_ib, 3555 .insert_nop = amdgpu_ring_insert_nop, 3556 }; 3557 3558 static void gfx_v6_0_set_ring_funcs(struct amdgpu_device *adev) 3559 { 3560 int i; 3561 3562 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 3563 adev->gfx.gfx_ring[i].funcs = &gfx_v6_0_ring_funcs_gfx; 3564 for (i = 0; i < adev->gfx.num_compute_rings; i++) 3565 adev->gfx.compute_ring[i].funcs = &gfx_v6_0_ring_funcs_compute; 3566 } 3567 3568 static const struct amdgpu_irq_src_funcs gfx_v6_0_eop_irq_funcs = { 3569 .set = gfx_v6_0_set_eop_interrupt_state, 3570 .process = gfx_v6_0_eop_irq, 3571 }; 3572 3573 static const struct amdgpu_irq_src_funcs gfx_v6_0_priv_reg_irq_funcs = { 3574 .set = gfx_v6_0_set_priv_reg_fault_state, 3575 .process = gfx_v6_0_priv_reg_irq, 3576 }; 3577 3578 static const struct amdgpu_irq_src_funcs gfx_v6_0_priv_inst_irq_funcs = { 3579 .set = gfx_v6_0_set_priv_inst_fault_state, 3580 .process = gfx_v6_0_priv_inst_irq, 3581 }; 3582 3583 static void gfx_v6_0_set_irq_funcs(struct amdgpu_device *adev) 3584 { 3585 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST; 3586 adev->gfx.eop_irq.funcs = &gfx_v6_0_eop_irq_funcs; 3587 3588 adev->gfx.priv_reg_irq.num_types = 1; 3589 adev->gfx.priv_reg_irq.funcs = &gfx_v6_0_priv_reg_irq_funcs; 3590 3591 adev->gfx.priv_inst_irq.num_types = 1; 3592 adev->gfx.priv_inst_irq.funcs = &gfx_v6_0_priv_inst_irq_funcs; 3593 } 3594 3595 static void gfx_v6_0_get_cu_info(struct amdgpu_device *adev) 3596 { 3597 int i, j, k, counter, active_cu_number = 0; 3598 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0; 3599 struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info; 3600 unsigned disable_masks[4 * 2]; 3601 u32 ao_cu_num; 3602 3603 if (adev->flags & AMD_IS_APU) 3604 ao_cu_num = 2; 3605 else 3606 ao_cu_num = adev->gfx.config.max_cu_per_sh; 3607 3608 memset(cu_info, 0, sizeof(*cu_info)); 3609 3610 amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2); 3611 3612 mutex_lock(&adev->grbm_idx_mutex); 3613 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 3614 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 3615 mask = 1; 3616 ao_bitmap = 0; 3617 counter = 0; 3618 gfx_v6_0_select_se_sh(adev, i, j, 0xffffffff); 3619 if (i < 4 && j < 2) 3620 gfx_v6_0_set_user_cu_inactive_bitmap( 3621 adev, disable_masks[i * 2 + j]); 3622 bitmap = gfx_v6_0_get_cu_enabled(adev); 3623 cu_info->bitmap[i][j] = bitmap; 3624 3625 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k++) { 3626 if (bitmap & mask) { 3627 if (counter < ao_cu_num) 3628 ao_bitmap |= mask; 3629 counter ++; 3630 } 3631 mask <<= 1; 3632 } 3633 active_cu_number += counter; 3634 if (i < 2 && j < 2) 3635 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8)); 3636 cu_info->ao_cu_bitmap[i][j] = ao_bitmap; 3637 } 3638 } 3639 3640 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 3641 mutex_unlock(&adev->grbm_idx_mutex); 3642 3643 cu_info->number = active_cu_number; 3644 cu_info->ao_cu_mask = ao_cu_mask; 3645 } 3646 3647 const struct amdgpu_ip_block_version gfx_v6_0_ip_block = 3648 { 3649 .type = AMD_IP_BLOCK_TYPE_GFX, 3650 .major = 6, 3651 .minor = 0, 3652 .rev = 0, 3653 .funcs = &gfx_v6_0_ip_funcs, 3654 }; 3655