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 <linux/module.h> 25 26 #include "amdgpu.h" 27 #include "amdgpu_ih.h" 28 #include "amdgpu_gfx.h" 29 #include "amdgpu_ucode.h" 30 #include "clearstate_si.h" 31 #include "bif/bif_3_0_d.h" 32 #include "bif/bif_3_0_sh_mask.h" 33 #include "oss/oss_1_0_d.h" 34 #include "oss/oss_1_0_sh_mask.h" 35 #include "gca/gfx_6_0_d.h" 36 #include "gca/gfx_6_0_sh_mask.h" 37 #include "gmc/gmc_6_0_d.h" 38 #include "gmc/gmc_6_0_sh_mask.h" 39 #include "dce/dce_6_0_d.h" 40 #include "dce/dce_6_0_sh_mask.h" 41 #include "gca/gfx_7_2_enum.h" 42 #include "si_enums.h" 43 #include "si.h" 44 45 static void gfx_v6_0_set_ring_funcs(struct amdgpu_device *adev); 46 static void gfx_v6_0_set_irq_funcs(struct amdgpu_device *adev); 47 static void gfx_v6_0_get_cu_info(struct amdgpu_device *adev); 48 49 MODULE_FIRMWARE("amdgpu/tahiti_pfp.bin"); 50 MODULE_FIRMWARE("amdgpu/tahiti_me.bin"); 51 MODULE_FIRMWARE("amdgpu/tahiti_ce.bin"); 52 MODULE_FIRMWARE("amdgpu/tahiti_rlc.bin"); 53 54 MODULE_FIRMWARE("amdgpu/pitcairn_pfp.bin"); 55 MODULE_FIRMWARE("amdgpu/pitcairn_me.bin"); 56 MODULE_FIRMWARE("amdgpu/pitcairn_ce.bin"); 57 MODULE_FIRMWARE("amdgpu/pitcairn_rlc.bin"); 58 59 MODULE_FIRMWARE("amdgpu/verde_pfp.bin"); 60 MODULE_FIRMWARE("amdgpu/verde_me.bin"); 61 MODULE_FIRMWARE("amdgpu/verde_ce.bin"); 62 MODULE_FIRMWARE("amdgpu/verde_rlc.bin"); 63 64 MODULE_FIRMWARE("amdgpu/oland_pfp.bin"); 65 MODULE_FIRMWARE("amdgpu/oland_me.bin"); 66 MODULE_FIRMWARE("amdgpu/oland_ce.bin"); 67 MODULE_FIRMWARE("amdgpu/oland_rlc.bin"); 68 69 MODULE_FIRMWARE("amdgpu/hainan_pfp.bin"); 70 MODULE_FIRMWARE("amdgpu/hainan_me.bin"); 71 MODULE_FIRMWARE("amdgpu/hainan_ce.bin"); 72 MODULE_FIRMWARE("amdgpu/hainan_rlc.bin"); 73 74 static u32 gfx_v6_0_get_csb_size(struct amdgpu_device *adev); 75 static void gfx_v6_0_get_csb_buffer(struct amdgpu_device *adev, volatile u32 *buffer); 76 //static void gfx_v6_0_init_cp_pg_table(struct amdgpu_device *adev); 77 static void gfx_v6_0_init_pg(struct amdgpu_device *adev); 78 79 #define ARRAY_MODE(x) ((x) << GB_TILE_MODE0__ARRAY_MODE__SHIFT) 80 #define PIPE_CONFIG(x) ((x) << GB_TILE_MODE0__PIPE_CONFIG__SHIFT) 81 #define TILE_SPLIT(x) ((x) << GB_TILE_MODE0__TILE_SPLIT__SHIFT) 82 #define MICRO_TILE_MODE(x) ((x) << 0) 83 #define SAMPLE_SPLIT(x) ((x) << GB_TILE_MODE0__SAMPLE_SPLIT__SHIFT) 84 #define BANK_WIDTH(x) ((x) << 14) 85 #define BANK_HEIGHT(x) ((x) << 16) 86 #define MACRO_TILE_ASPECT(x) ((x) << 18) 87 #define NUM_BANKS(x) ((x) << 20) 88 89 static const u32 verde_rlc_save_restore_register_list[] = 90 { 91 (0x8000 << 16) | (0x98f4 >> 2), 92 0x00000000, 93 (0x8040 << 16) | (0x98f4 >> 2), 94 0x00000000, 95 (0x8000 << 16) | (0xe80 >> 2), 96 0x00000000, 97 (0x8040 << 16) | (0xe80 >> 2), 98 0x00000000, 99 (0x8000 << 16) | (0x89bc >> 2), 100 0x00000000, 101 (0x8040 << 16) | (0x89bc >> 2), 102 0x00000000, 103 (0x8000 << 16) | (0x8c1c >> 2), 104 0x00000000, 105 (0x8040 << 16) | (0x8c1c >> 2), 106 0x00000000, 107 (0x9c00 << 16) | (0x98f0 >> 2), 108 0x00000000, 109 (0x9c00 << 16) | (0xe7c >> 2), 110 0x00000000, 111 (0x8000 << 16) | (0x9148 >> 2), 112 0x00000000, 113 (0x8040 << 16) | (0x9148 >> 2), 114 0x00000000, 115 (0x9c00 << 16) | (0x9150 >> 2), 116 0x00000000, 117 (0x9c00 << 16) | (0x897c >> 2), 118 0x00000000, 119 (0x9c00 << 16) | (0x8d8c >> 2), 120 0x00000000, 121 (0x9c00 << 16) | (0xac54 >> 2), 122 0X00000000, 123 0x3, 124 (0x9c00 << 16) | (0x98f8 >> 2), 125 0x00000000, 126 (0x9c00 << 16) | (0x9910 >> 2), 127 0x00000000, 128 (0x9c00 << 16) | (0x9914 >> 2), 129 0x00000000, 130 (0x9c00 << 16) | (0x9918 >> 2), 131 0x00000000, 132 (0x9c00 << 16) | (0x991c >> 2), 133 0x00000000, 134 (0x9c00 << 16) | (0x9920 >> 2), 135 0x00000000, 136 (0x9c00 << 16) | (0x9924 >> 2), 137 0x00000000, 138 (0x9c00 << 16) | (0x9928 >> 2), 139 0x00000000, 140 (0x9c00 << 16) | (0x992c >> 2), 141 0x00000000, 142 (0x9c00 << 16) | (0x9930 >> 2), 143 0x00000000, 144 (0x9c00 << 16) | (0x9934 >> 2), 145 0x00000000, 146 (0x9c00 << 16) | (0x9938 >> 2), 147 0x00000000, 148 (0x9c00 << 16) | (0x993c >> 2), 149 0x00000000, 150 (0x9c00 << 16) | (0x9940 >> 2), 151 0x00000000, 152 (0x9c00 << 16) | (0x9944 >> 2), 153 0x00000000, 154 (0x9c00 << 16) | (0x9948 >> 2), 155 0x00000000, 156 (0x9c00 << 16) | (0x994c >> 2), 157 0x00000000, 158 (0x9c00 << 16) | (0x9950 >> 2), 159 0x00000000, 160 (0x9c00 << 16) | (0x9954 >> 2), 161 0x00000000, 162 (0x9c00 << 16) | (0x9958 >> 2), 163 0x00000000, 164 (0x9c00 << 16) | (0x995c >> 2), 165 0x00000000, 166 (0x9c00 << 16) | (0x9960 >> 2), 167 0x00000000, 168 (0x9c00 << 16) | (0x9964 >> 2), 169 0x00000000, 170 (0x9c00 << 16) | (0x9968 >> 2), 171 0x00000000, 172 (0x9c00 << 16) | (0x996c >> 2), 173 0x00000000, 174 (0x9c00 << 16) | (0x9970 >> 2), 175 0x00000000, 176 (0x9c00 << 16) | (0x9974 >> 2), 177 0x00000000, 178 (0x9c00 << 16) | (0x9978 >> 2), 179 0x00000000, 180 (0x9c00 << 16) | (0x997c >> 2), 181 0x00000000, 182 (0x9c00 << 16) | (0x9980 >> 2), 183 0x00000000, 184 (0x9c00 << 16) | (0x9984 >> 2), 185 0x00000000, 186 (0x9c00 << 16) | (0x9988 >> 2), 187 0x00000000, 188 (0x9c00 << 16) | (0x998c >> 2), 189 0x00000000, 190 (0x9c00 << 16) | (0x8c00 >> 2), 191 0x00000000, 192 (0x9c00 << 16) | (0x8c14 >> 2), 193 0x00000000, 194 (0x9c00 << 16) | (0x8c04 >> 2), 195 0x00000000, 196 (0x9c00 << 16) | (0x8c08 >> 2), 197 0x00000000, 198 (0x8000 << 16) | (0x9b7c >> 2), 199 0x00000000, 200 (0x8040 << 16) | (0x9b7c >> 2), 201 0x00000000, 202 (0x8000 << 16) | (0xe84 >> 2), 203 0x00000000, 204 (0x8040 << 16) | (0xe84 >> 2), 205 0x00000000, 206 (0x8000 << 16) | (0x89c0 >> 2), 207 0x00000000, 208 (0x8040 << 16) | (0x89c0 >> 2), 209 0x00000000, 210 (0x8000 << 16) | (0x914c >> 2), 211 0x00000000, 212 (0x8040 << 16) | (0x914c >> 2), 213 0x00000000, 214 (0x8000 << 16) | (0x8c20 >> 2), 215 0x00000000, 216 (0x8040 << 16) | (0x8c20 >> 2), 217 0x00000000, 218 (0x8000 << 16) | (0x9354 >> 2), 219 0x00000000, 220 (0x8040 << 16) | (0x9354 >> 2), 221 0x00000000, 222 (0x9c00 << 16) | (0x9060 >> 2), 223 0x00000000, 224 (0x9c00 << 16) | (0x9364 >> 2), 225 0x00000000, 226 (0x9c00 << 16) | (0x9100 >> 2), 227 0x00000000, 228 (0x9c00 << 16) | (0x913c >> 2), 229 0x00000000, 230 (0x8000 << 16) | (0x90e0 >> 2), 231 0x00000000, 232 (0x8000 << 16) | (0x90e4 >> 2), 233 0x00000000, 234 (0x8000 << 16) | (0x90e8 >> 2), 235 0x00000000, 236 (0x8040 << 16) | (0x90e0 >> 2), 237 0x00000000, 238 (0x8040 << 16) | (0x90e4 >> 2), 239 0x00000000, 240 (0x8040 << 16) | (0x90e8 >> 2), 241 0x00000000, 242 (0x9c00 << 16) | (0x8bcc >> 2), 243 0x00000000, 244 (0x9c00 << 16) | (0x8b24 >> 2), 245 0x00000000, 246 (0x9c00 << 16) | (0x88c4 >> 2), 247 0x00000000, 248 (0x9c00 << 16) | (0x8e50 >> 2), 249 0x00000000, 250 (0x9c00 << 16) | (0x8c0c >> 2), 251 0x00000000, 252 (0x9c00 << 16) | (0x8e58 >> 2), 253 0x00000000, 254 (0x9c00 << 16) | (0x8e5c >> 2), 255 0x00000000, 256 (0x9c00 << 16) | (0x9508 >> 2), 257 0x00000000, 258 (0x9c00 << 16) | (0x950c >> 2), 259 0x00000000, 260 (0x9c00 << 16) | (0x9494 >> 2), 261 0x00000000, 262 (0x9c00 << 16) | (0xac0c >> 2), 263 0x00000000, 264 (0x9c00 << 16) | (0xac10 >> 2), 265 0x00000000, 266 (0x9c00 << 16) | (0xac14 >> 2), 267 0x00000000, 268 (0x9c00 << 16) | (0xae00 >> 2), 269 0x00000000, 270 (0x9c00 << 16) | (0xac08 >> 2), 271 0x00000000, 272 (0x9c00 << 16) | (0x88d4 >> 2), 273 0x00000000, 274 (0x9c00 << 16) | (0x88c8 >> 2), 275 0x00000000, 276 (0x9c00 << 16) | (0x88cc >> 2), 277 0x00000000, 278 (0x9c00 << 16) | (0x89b0 >> 2), 279 0x00000000, 280 (0x9c00 << 16) | (0x8b10 >> 2), 281 0x00000000, 282 (0x9c00 << 16) | (0x8a14 >> 2), 283 0x00000000, 284 (0x9c00 << 16) | (0x9830 >> 2), 285 0x00000000, 286 (0x9c00 << 16) | (0x9834 >> 2), 287 0x00000000, 288 (0x9c00 << 16) | (0x9838 >> 2), 289 0x00000000, 290 (0x9c00 << 16) | (0x9a10 >> 2), 291 0x00000000, 292 (0x8000 << 16) | (0x9870 >> 2), 293 0x00000000, 294 (0x8000 << 16) | (0x9874 >> 2), 295 0x00000000, 296 (0x8001 << 16) | (0x9870 >> 2), 297 0x00000000, 298 (0x8001 << 16) | (0x9874 >> 2), 299 0x00000000, 300 (0x8040 << 16) | (0x9870 >> 2), 301 0x00000000, 302 (0x8040 << 16) | (0x9874 >> 2), 303 0x00000000, 304 (0x8041 << 16) | (0x9870 >> 2), 305 0x00000000, 306 (0x8041 << 16) | (0x9874 >> 2), 307 0x00000000, 308 0x00000000 309 }; 310 311 static int gfx_v6_0_init_microcode(struct amdgpu_device *adev) 312 { 313 const char *chip_name; 314 char fw_name[30]; 315 int err; 316 const struct gfx_firmware_header_v1_0 *cp_hdr; 317 const struct rlc_firmware_header_v1_0 *rlc_hdr; 318 319 DRM_DEBUG("\n"); 320 321 switch (adev->asic_type) { 322 case CHIP_TAHITI: 323 chip_name = "tahiti"; 324 break; 325 case CHIP_PITCAIRN: 326 chip_name = "pitcairn"; 327 break; 328 case CHIP_VERDE: 329 chip_name = "verde"; 330 break; 331 case CHIP_OLAND: 332 chip_name = "oland"; 333 break; 334 case CHIP_HAINAN: 335 chip_name = "hainan"; 336 break; 337 default: BUG(); 338 } 339 340 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name); 341 err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw, fw_name); 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), "amdgpu/%s_me.bin", chip_name); 349 err = amdgpu_ucode_request(adev, &adev->gfx.me_fw, fw_name); 350 if (err) 351 goto out; 352 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; 353 adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); 354 adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); 355 356 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name); 357 err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw, fw_name); 358 if (err) 359 goto out; 360 cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; 361 adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version); 362 adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version); 363 364 snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name); 365 err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, fw_name); 366 if (err) 367 goto out; 368 rlc_hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data; 369 adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version); 370 adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version); 371 372 out: 373 if (err) { 374 pr_err("gfx6: Failed to load firmware \"%s\"\n", fw_name); 375 amdgpu_ucode_release(&adev->gfx.pfp_fw); 376 amdgpu_ucode_release(&adev->gfx.me_fw); 377 amdgpu_ucode_release(&adev->gfx.ce_fw); 378 amdgpu_ucode_release(&adev->gfx.rlc_fw); 379 } 380 return err; 381 } 382 383 static void gfx_v6_0_tiling_mode_table_init(struct amdgpu_device *adev) 384 { 385 const u32 num_tile_mode_states = ARRAY_SIZE(adev->gfx.config.tile_mode_array); 386 u32 reg_offset, split_equal_to_row_size, *tilemode; 387 388 memset(adev->gfx.config.tile_mode_array, 0, sizeof(adev->gfx.config.tile_mode_array)); 389 tilemode = adev->gfx.config.tile_mode_array; 390 391 switch (adev->gfx.config.mem_row_size_in_kb) { 392 case 1: 393 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB; 394 break; 395 case 2: 396 default: 397 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB; 398 break; 399 case 4: 400 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB; 401 break; 402 } 403 404 if (adev->asic_type == CHIP_VERDE) { 405 tilemode[0] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 406 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 407 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 408 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 409 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 410 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 411 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 412 NUM_BANKS(ADDR_SURF_16_BANK); 413 tilemode[1] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 414 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 415 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 416 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 417 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 418 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 419 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 420 NUM_BANKS(ADDR_SURF_16_BANK); 421 tilemode[2] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 422 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 423 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 424 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 425 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 426 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 427 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 428 NUM_BANKS(ADDR_SURF_16_BANK); 429 tilemode[3] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 430 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 431 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 432 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 433 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 434 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 435 NUM_BANKS(ADDR_SURF_8_BANK) | 436 TILE_SPLIT(split_equal_to_row_size); 437 tilemode[4] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 438 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 439 PIPE_CONFIG(ADDR_SURF_P4_8x16); 440 tilemode[5] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 441 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 442 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 443 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 444 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 445 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 446 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 447 NUM_BANKS(ADDR_SURF_4_BANK); 448 tilemode[6] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 449 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 450 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 451 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 452 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 453 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 454 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 455 NUM_BANKS(ADDR_SURF_4_BANK); 456 tilemode[7] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 457 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 458 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 459 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 460 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 461 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 462 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 463 NUM_BANKS(ADDR_SURF_2_BANK); 464 tilemode[8] = ARRAY_MODE(ARRAY_LINEAR_ALIGNED); 465 tilemode[9] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 466 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 467 PIPE_CONFIG(ADDR_SURF_P4_8x16); 468 tilemode[10] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 469 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 470 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 471 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 472 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 473 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 474 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 475 NUM_BANKS(ADDR_SURF_16_BANK); 476 tilemode[11] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 477 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 478 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 479 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 480 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 481 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 482 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 483 NUM_BANKS(ADDR_SURF_16_BANK); 484 tilemode[12] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 485 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 486 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 487 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 488 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 489 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 490 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 491 NUM_BANKS(ADDR_SURF_16_BANK); 492 tilemode[13] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 493 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 494 PIPE_CONFIG(ADDR_SURF_P4_8x16); 495 tilemode[14] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 496 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 497 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 498 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 499 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 500 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 501 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 502 NUM_BANKS(ADDR_SURF_16_BANK); 503 tilemode[15] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 504 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 505 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 506 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 507 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 508 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 509 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 510 NUM_BANKS(ADDR_SURF_16_BANK); 511 tilemode[16] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 512 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 513 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 514 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 515 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 516 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 517 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 518 NUM_BANKS(ADDR_SURF_16_BANK); 519 tilemode[17] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 520 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 521 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 522 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 523 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 524 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 525 NUM_BANKS(ADDR_SURF_16_BANK) | 526 TILE_SPLIT(split_equal_to_row_size); 527 tilemode[18] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 528 ARRAY_MODE(ARRAY_1D_TILED_THICK) | 529 PIPE_CONFIG(ADDR_SURF_P4_8x16); 530 tilemode[19] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 531 ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 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[20] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 539 ARRAY_MODE(ARRAY_2D_TILED_THICK) | 540 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 541 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 542 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 543 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 544 NUM_BANKS(ADDR_SURF_16_BANK) | 545 TILE_SPLIT(split_equal_to_row_size); 546 tilemode[21] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 547 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 548 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 549 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 550 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 551 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 552 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 553 NUM_BANKS(ADDR_SURF_8_BANK); 554 tilemode[22] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 555 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 556 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 557 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 558 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 559 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 560 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 561 NUM_BANKS(ADDR_SURF_8_BANK); 562 tilemode[23] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 563 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 564 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 565 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 566 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 567 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 568 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 569 NUM_BANKS(ADDR_SURF_4_BANK); 570 tilemode[24] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 571 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 572 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 573 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 574 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 575 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 576 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 577 NUM_BANKS(ADDR_SURF_4_BANK); 578 tilemode[25] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 579 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 580 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 581 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 582 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 583 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 584 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 585 NUM_BANKS(ADDR_SURF_2_BANK); 586 tilemode[26] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 587 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 588 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 589 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 590 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 591 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 592 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 593 NUM_BANKS(ADDR_SURF_2_BANK); 594 tilemode[27] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 595 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 596 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 597 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 598 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 599 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 600 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 601 NUM_BANKS(ADDR_SURF_2_BANK); 602 tilemode[28] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 603 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 604 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 605 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 606 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 607 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 608 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 609 NUM_BANKS(ADDR_SURF_2_BANK); 610 tilemode[29] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 611 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 612 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 613 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 614 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 615 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 616 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 617 NUM_BANKS(ADDR_SURF_2_BANK); 618 tilemode[30] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 619 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 620 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 621 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 622 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 623 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 624 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 625 NUM_BANKS(ADDR_SURF_2_BANK); 626 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 627 WREG32(mmGB_TILE_MODE0 + reg_offset, tilemode[reg_offset]); 628 } else if (adev->asic_type == CHIP_OLAND) { 629 tilemode[0] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 630 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 631 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 632 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 633 NUM_BANKS(ADDR_SURF_16_BANK) | 634 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 635 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 636 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 637 tilemode[1] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 638 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 639 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 640 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 641 NUM_BANKS(ADDR_SURF_16_BANK) | 642 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 643 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 644 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 645 tilemode[2] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 646 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 647 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 648 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 649 NUM_BANKS(ADDR_SURF_16_BANK) | 650 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 651 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 652 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 653 tilemode[3] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 654 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 655 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 656 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 657 NUM_BANKS(ADDR_SURF_16_BANK) | 658 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 659 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 660 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 661 tilemode[4] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 662 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 663 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 664 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 665 NUM_BANKS(ADDR_SURF_16_BANK) | 666 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 667 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 668 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 669 tilemode[5] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 670 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 671 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 672 TILE_SPLIT(split_equal_to_row_size) | 673 NUM_BANKS(ADDR_SURF_16_BANK) | 674 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 675 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 676 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 677 tilemode[6] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 678 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 679 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 680 TILE_SPLIT(split_equal_to_row_size) | 681 NUM_BANKS(ADDR_SURF_16_BANK) | 682 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 683 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 684 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 685 tilemode[7] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 686 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 687 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 688 TILE_SPLIT(split_equal_to_row_size) | 689 NUM_BANKS(ADDR_SURF_16_BANK) | 690 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 691 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 692 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 693 tilemode[8] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 694 ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 695 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 696 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 697 NUM_BANKS(ADDR_SURF_16_BANK) | 698 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 699 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 700 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 701 tilemode[9] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 702 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 703 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 704 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 705 NUM_BANKS(ADDR_SURF_16_BANK) | 706 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 707 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 708 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 709 tilemode[10] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 710 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 711 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 712 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 713 NUM_BANKS(ADDR_SURF_16_BANK) | 714 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 715 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 716 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 717 tilemode[11] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 718 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 719 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 720 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 721 NUM_BANKS(ADDR_SURF_16_BANK) | 722 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 723 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 724 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 725 tilemode[12] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 726 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 727 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 728 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 729 NUM_BANKS(ADDR_SURF_16_BANK) | 730 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 731 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 732 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 733 tilemode[13] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 734 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 735 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 736 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 737 NUM_BANKS(ADDR_SURF_16_BANK) | 738 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 739 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 740 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 741 tilemode[14] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 742 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 743 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 744 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 745 NUM_BANKS(ADDR_SURF_16_BANK) | 746 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 747 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 748 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 749 tilemode[15] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 750 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 751 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 752 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 753 NUM_BANKS(ADDR_SURF_16_BANK) | 754 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 755 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 756 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 757 tilemode[16] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 758 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 759 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 760 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 761 NUM_BANKS(ADDR_SURF_16_BANK) | 762 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 763 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 764 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 765 tilemode[17] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 766 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 767 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 768 TILE_SPLIT(split_equal_to_row_size) | 769 NUM_BANKS(ADDR_SURF_16_BANK) | 770 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 771 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 772 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 773 tilemode[18] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 774 ARRAY_MODE(ARRAY_1D_TILED_THICK) | 775 PIPE_CONFIG(ADDR_SURF_P4_8x16); 776 tilemode[19] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 777 ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 778 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 779 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 780 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 781 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 782 NUM_BANKS(ADDR_SURF_16_BANK) | 783 TILE_SPLIT(split_equal_to_row_size); 784 tilemode[20] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 785 ARRAY_MODE(ARRAY_2D_TILED_THICK) | 786 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 787 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 788 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 789 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 790 NUM_BANKS(ADDR_SURF_16_BANK) | 791 TILE_SPLIT(split_equal_to_row_size); 792 tilemode[21] = 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_2) | 798 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 799 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 800 tilemode[22] = 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_4) | 807 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4); 808 tilemode[23] = 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_256B) | 812 NUM_BANKS(ADDR_SURF_16_BANK) | 813 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 814 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 815 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2); 816 tilemode[24] = 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_512B) | 820 NUM_BANKS(ADDR_SURF_16_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_2); 824 tilemode[25] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 825 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 826 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 827 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 828 NUM_BANKS(ADDR_SURF_8_BANK) | 829 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 830 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 831 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1); 832 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 833 WREG32(mmGB_TILE_MODE0 + reg_offset, tilemode[reg_offset]); 834 } else if (adev->asic_type == CHIP_HAINAN) { 835 tilemode[0] = 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_64B) | 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[1] = 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_128B) | 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[2] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 852 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 853 PIPE_CONFIG(ADDR_SURF_P2) | 854 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 855 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 856 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 857 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 858 NUM_BANKS(ADDR_SURF_16_BANK); 859 tilemode[3] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 860 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 861 PIPE_CONFIG(ADDR_SURF_P2) | 862 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 863 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 864 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 865 NUM_BANKS(ADDR_SURF_8_BANK) | 866 TILE_SPLIT(split_equal_to_row_size); 867 tilemode[4] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 868 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 869 PIPE_CONFIG(ADDR_SURF_P2); 870 tilemode[5] = 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_512B) | 874 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 875 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 876 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 877 NUM_BANKS(ADDR_SURF_8_BANK); 878 tilemode[6] = 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_256B) | 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_8_BANK); 886 tilemode[7] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 887 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 888 PIPE_CONFIG(ADDR_SURF_P2) | 889 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 890 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 891 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 892 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 893 NUM_BANKS(ADDR_SURF_4_BANK); 894 tilemode[8] = ARRAY_MODE(ARRAY_LINEAR_ALIGNED); 895 tilemode[9] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 896 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 897 PIPE_CONFIG(ADDR_SURF_P2); 898 tilemode[10] = 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_4) | 904 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 905 NUM_BANKS(ADDR_SURF_16_BANK); 906 tilemode[11] = 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_256B) | 910 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 911 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 912 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 913 NUM_BANKS(ADDR_SURF_16_BANK); 914 tilemode[12] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 915 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 916 PIPE_CONFIG(ADDR_SURF_P2) | 917 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 918 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 919 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 920 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 921 NUM_BANKS(ADDR_SURF_16_BANK); 922 tilemode[13] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 923 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 924 PIPE_CONFIG(ADDR_SURF_P2); 925 tilemode[14] = 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_4) | 931 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 932 NUM_BANKS(ADDR_SURF_16_BANK); 933 tilemode[15] = 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_256B) | 937 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 938 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 939 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 940 NUM_BANKS(ADDR_SURF_16_BANK); 941 tilemode[16] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 942 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 943 PIPE_CONFIG(ADDR_SURF_P2) | 944 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 945 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 946 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 947 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 948 NUM_BANKS(ADDR_SURF_16_BANK); 949 tilemode[17] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 950 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 951 PIPE_CONFIG(ADDR_SURF_P2) | 952 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 953 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 954 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 955 NUM_BANKS(ADDR_SURF_16_BANK) | 956 TILE_SPLIT(split_equal_to_row_size); 957 tilemode[18] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 958 ARRAY_MODE(ARRAY_1D_TILED_THICK) | 959 PIPE_CONFIG(ADDR_SURF_P2); 960 tilemode[19] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 961 ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 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[20] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 969 ARRAY_MODE(ARRAY_2D_TILED_THICK) | 970 PIPE_CONFIG(ADDR_SURF_P2) | 971 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 972 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 973 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 974 NUM_BANKS(ADDR_SURF_16_BANK) | 975 TILE_SPLIT(split_equal_to_row_size); 976 tilemode[21] = 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_4) | 982 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 983 NUM_BANKS(ADDR_SURF_8_BANK); 984 tilemode[22] = 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_2) | 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[23] = 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_256B) | 996 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 997 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 998 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 999 NUM_BANKS(ADDR_SURF_8_BANK); 1000 tilemode[24] = 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_512B) | 1004 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1005 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1006 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1007 NUM_BANKS(ADDR_SURF_8_BANK); 1008 tilemode[25] = 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[26] = 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[27] = 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[28] = 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[29] = 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_1KB) | 1044 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1045 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1046 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1047 NUM_BANKS(ADDR_SURF_4_BANK); 1048 tilemode[30] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1049 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1050 PIPE_CONFIG(ADDR_SURF_P2) | 1051 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 1052 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1053 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1054 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1055 NUM_BANKS(ADDR_SURF_4_BANK); 1056 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 1057 WREG32(mmGB_TILE_MODE0 + reg_offset, tilemode[reg_offset]); 1058 } else if ((adev->asic_type == CHIP_TAHITI) || (adev->asic_type == CHIP_PITCAIRN)) { 1059 tilemode[0] = 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_64B) | 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[1] = 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_128B) | 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[2] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1076 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1077 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1078 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1079 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1080 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1081 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1082 NUM_BANKS(ADDR_SURF_16_BANK); 1083 tilemode[3] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1084 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1085 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1086 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1087 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1088 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1089 NUM_BANKS(ADDR_SURF_4_BANK) | 1090 TILE_SPLIT(split_equal_to_row_size); 1091 tilemode[4] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1092 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1093 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16); 1094 tilemode[5] = 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_512B) | 1098 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1099 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1100 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1101 NUM_BANKS(ADDR_SURF_2_BANK); 1102 tilemode[6] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1103 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1104 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1105 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1106 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1107 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 1108 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1109 NUM_BANKS(ADDR_SURF_2_BANK); 1110 tilemode[7] = MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 1111 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1112 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 1113 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 1114 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1115 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1116 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1117 NUM_BANKS(ADDR_SURF_2_BANK); 1118 tilemode[8] = ARRAY_MODE(ARRAY_LINEAR_ALIGNED); 1119 tilemode[9] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 1120 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1121 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16); 1122 tilemode[10] = 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_4) | 1128 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1129 NUM_BANKS(ADDR_SURF_16_BANK); 1130 tilemode[11] = 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_256B) | 1134 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1135 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1136 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1137 NUM_BANKS(ADDR_SURF_16_BANK); 1138 tilemode[12] = MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 1139 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1140 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1141 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 1142 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1143 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1144 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1145 NUM_BANKS(ADDR_SURF_16_BANK); 1146 tilemode[13] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1147 ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1148 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16); 1149 tilemode[14] = 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_4) | 1155 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1156 NUM_BANKS(ADDR_SURF_16_BANK); 1157 tilemode[15] = 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_256B) | 1161 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1162 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1163 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1164 NUM_BANKS(ADDR_SURF_16_BANK); 1165 tilemode[16] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1166 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1167 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1168 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 1169 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1170 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1171 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1172 NUM_BANKS(ADDR_SURF_16_BANK); 1173 tilemode[17] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1174 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1175 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1176 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1177 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1178 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1179 NUM_BANKS(ADDR_SURF_16_BANK) | 1180 TILE_SPLIT(split_equal_to_row_size); 1181 tilemode[18] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1182 ARRAY_MODE(ARRAY_1D_TILED_THICK) | 1183 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16); 1184 tilemode[19] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1185 ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 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[20] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1193 ARRAY_MODE(ARRAY_2D_TILED_THICK) | 1194 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1195 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1196 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1197 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1198 NUM_BANKS(ADDR_SURF_16_BANK) | 1199 TILE_SPLIT(split_equal_to_row_size); 1200 tilemode[21] = 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_8) | 1206 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1207 NUM_BANKS(ADDR_SURF_4_BANK); 1208 tilemode[22] = 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_4) | 1214 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1215 NUM_BANKS(ADDR_SURF_4_BANK); 1216 tilemode[23] = 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_256B) | 1220 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1221 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 1222 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1223 NUM_BANKS(ADDR_SURF_2_BANK); 1224 tilemode[24] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1225 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1226 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 1227 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 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[25] = 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[26] = 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[27] = 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[28] = 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[29] = 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_1KB) | 1268 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1269 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1270 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1271 NUM_BANKS(ADDR_SURF_2_BANK); 1272 tilemode[30] = MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 1273 ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1274 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 1275 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | 1276 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1277 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1278 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1279 NUM_BANKS(ADDR_SURF_2_BANK); 1280 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 1281 WREG32(mmGB_TILE_MODE0 + reg_offset, tilemode[reg_offset]); 1282 } else { 1283 DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type); 1284 } 1285 } 1286 1287 static void gfx_v6_0_select_se_sh(struct amdgpu_device *adev, u32 se_num, 1288 u32 sh_num, u32 instance) 1289 { 1290 u32 data; 1291 1292 if (instance == 0xffffffff) 1293 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1); 1294 else 1295 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance); 1296 1297 if ((se_num == 0xffffffff) && (sh_num == 0xffffffff)) 1298 data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK | 1299 GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK; 1300 else if (se_num == 0xffffffff) 1301 data |= GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK | 1302 (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT); 1303 else if (sh_num == 0xffffffff) 1304 data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK | 1305 (se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT); 1306 else 1307 data |= (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT) | 1308 (se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT); 1309 WREG32(mmGRBM_GFX_INDEX, data); 1310 } 1311 1312 static u32 gfx_v6_0_get_rb_active_bitmap(struct amdgpu_device *adev) 1313 { 1314 u32 data, mask; 1315 1316 data = RREG32(mmCC_RB_BACKEND_DISABLE) | 1317 RREG32(mmGC_USER_RB_BACKEND_DISABLE); 1318 1319 data = REG_GET_FIELD(data, GC_USER_RB_BACKEND_DISABLE, BACKEND_DISABLE); 1320 1321 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se/ 1322 adev->gfx.config.max_sh_per_se); 1323 1324 return ~data & mask; 1325 } 1326 1327 static void gfx_v6_0_raster_config(struct amdgpu_device *adev, u32 *rconf) 1328 { 1329 switch (adev->asic_type) { 1330 case CHIP_TAHITI: 1331 case CHIP_PITCAIRN: 1332 *rconf |= 1333 (2 << PA_SC_RASTER_CONFIG__RB_XSEL2__SHIFT) | 1334 (1 << PA_SC_RASTER_CONFIG__RB_XSEL__SHIFT) | 1335 (2 << PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT) | 1336 (1 << PA_SC_RASTER_CONFIG__PKR_YSEL__SHIFT) | 1337 (2 << PA_SC_RASTER_CONFIG__SE_MAP__SHIFT) | 1338 (2 << PA_SC_RASTER_CONFIG__SE_XSEL__SHIFT) | 1339 (2 << PA_SC_RASTER_CONFIG__SE_YSEL__SHIFT); 1340 break; 1341 case CHIP_VERDE: 1342 *rconf |= 1343 (1 << PA_SC_RASTER_CONFIG__RB_XSEL__SHIFT) | 1344 (2 << PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT) | 1345 (1 << PA_SC_RASTER_CONFIG__PKR_YSEL__SHIFT); 1346 break; 1347 case CHIP_OLAND: 1348 *rconf |= (1 << PA_SC_RASTER_CONFIG__RB_YSEL__SHIFT); 1349 break; 1350 case CHIP_HAINAN: 1351 *rconf |= 0x0; 1352 break; 1353 default: 1354 DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type); 1355 break; 1356 } 1357 } 1358 1359 static void gfx_v6_0_write_harvested_raster_configs(struct amdgpu_device *adev, 1360 u32 raster_config, unsigned rb_mask, 1361 unsigned num_rb) 1362 { 1363 unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1); 1364 unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1); 1365 unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2); 1366 unsigned rb_per_se = num_rb / num_se; 1367 unsigned se_mask[4]; 1368 unsigned se; 1369 1370 se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask; 1371 se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask; 1372 se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask; 1373 se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask; 1374 1375 WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4)); 1376 WARN_ON(!(sh_per_se == 1 || sh_per_se == 2)); 1377 WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2)); 1378 1379 for (se = 0; se < num_se; se++) { 1380 unsigned raster_config_se = raster_config; 1381 unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se); 1382 unsigned pkr1_mask = pkr0_mask << rb_per_pkr; 1383 int idx = (se / 2) * 2; 1384 1385 if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) { 1386 raster_config_se &= ~PA_SC_RASTER_CONFIG__SE_MAP_MASK; 1387 1388 if (!se_mask[idx]) 1389 raster_config_se |= RASTER_CONFIG_SE_MAP_3 << PA_SC_RASTER_CONFIG__SE_MAP__SHIFT; 1390 else 1391 raster_config_se |= RASTER_CONFIG_SE_MAP_0 << PA_SC_RASTER_CONFIG__SE_MAP__SHIFT; 1392 } 1393 1394 pkr0_mask &= rb_mask; 1395 pkr1_mask &= rb_mask; 1396 if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) { 1397 raster_config_se &= ~PA_SC_RASTER_CONFIG__PKR_MAP_MASK; 1398 1399 if (!pkr0_mask) 1400 raster_config_se |= RASTER_CONFIG_PKR_MAP_3 << PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT; 1401 else 1402 raster_config_se |= RASTER_CONFIG_PKR_MAP_0 << PA_SC_RASTER_CONFIG__PKR_MAP__SHIFT; 1403 } 1404 1405 if (rb_per_se >= 2) { 1406 unsigned rb0_mask = 1 << (se * rb_per_se); 1407 unsigned rb1_mask = rb0_mask << 1; 1408 1409 rb0_mask &= rb_mask; 1410 rb1_mask &= rb_mask; 1411 if (!rb0_mask || !rb1_mask) { 1412 raster_config_se &= ~PA_SC_RASTER_CONFIG__RB_MAP_PKR0_MASK; 1413 1414 if (!rb0_mask) 1415 raster_config_se |= 1416 RASTER_CONFIG_RB_MAP_3 << PA_SC_RASTER_CONFIG__RB_MAP_PKR0__SHIFT; 1417 else 1418 raster_config_se |= 1419 RASTER_CONFIG_RB_MAP_0 << PA_SC_RASTER_CONFIG__RB_MAP_PKR0__SHIFT; 1420 } 1421 1422 if (rb_per_se > 2) { 1423 rb0_mask = 1 << (se * rb_per_se + rb_per_pkr); 1424 rb1_mask = rb0_mask << 1; 1425 rb0_mask &= rb_mask; 1426 rb1_mask &= rb_mask; 1427 if (!rb0_mask || !rb1_mask) { 1428 raster_config_se &= ~PA_SC_RASTER_CONFIG__RB_MAP_PKR1_MASK; 1429 1430 if (!rb0_mask) 1431 raster_config_se |= 1432 RASTER_CONFIG_RB_MAP_3 << PA_SC_RASTER_CONFIG__RB_MAP_PKR1__SHIFT; 1433 else 1434 raster_config_se |= 1435 RASTER_CONFIG_RB_MAP_0 << PA_SC_RASTER_CONFIG__RB_MAP_PKR1__SHIFT; 1436 } 1437 } 1438 } 1439 1440 /* GRBM_GFX_INDEX has a different offset on SI */ 1441 gfx_v6_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff); 1442 WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se); 1443 } 1444 1445 /* GRBM_GFX_INDEX has a different offset on SI */ 1446 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 1447 } 1448 1449 static void gfx_v6_0_setup_rb(struct amdgpu_device *adev) 1450 { 1451 int i, j; 1452 u32 data; 1453 u32 raster_config = 0; 1454 u32 active_rbs = 0; 1455 u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se / 1456 adev->gfx.config.max_sh_per_se; 1457 unsigned num_rb_pipes; 1458 1459 mutex_lock(&adev->grbm_idx_mutex); 1460 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 1461 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 1462 gfx_v6_0_select_se_sh(adev, i, j, 0xffffffff); 1463 data = gfx_v6_0_get_rb_active_bitmap(adev); 1464 active_rbs |= data << 1465 ((i * adev->gfx.config.max_sh_per_se + j) * 1466 rb_bitmap_width_per_sh); 1467 } 1468 } 1469 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 1470 1471 adev->gfx.config.backend_enable_mask = active_rbs; 1472 adev->gfx.config.num_rbs = hweight32(active_rbs); 1473 1474 num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se * 1475 adev->gfx.config.max_shader_engines, 16); 1476 1477 gfx_v6_0_raster_config(adev, &raster_config); 1478 1479 if (!adev->gfx.config.backend_enable_mask || 1480 adev->gfx.config.num_rbs >= num_rb_pipes) 1481 WREG32(mmPA_SC_RASTER_CONFIG, raster_config); 1482 else 1483 gfx_v6_0_write_harvested_raster_configs(adev, raster_config, 1484 adev->gfx.config.backend_enable_mask, 1485 num_rb_pipes); 1486 1487 /* cache the values for userspace */ 1488 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 1489 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 1490 gfx_v6_0_select_se_sh(adev, i, j, 0xffffffff); 1491 adev->gfx.config.rb_config[i][j].rb_backend_disable = 1492 RREG32(mmCC_RB_BACKEND_DISABLE); 1493 adev->gfx.config.rb_config[i][j].user_rb_backend_disable = 1494 RREG32(mmGC_USER_RB_BACKEND_DISABLE); 1495 adev->gfx.config.rb_config[i][j].raster_config = 1496 RREG32(mmPA_SC_RASTER_CONFIG); 1497 } 1498 } 1499 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 1500 mutex_unlock(&adev->grbm_idx_mutex); 1501 } 1502 1503 static void gfx_v6_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev, 1504 u32 bitmap) 1505 { 1506 u32 data; 1507 1508 if (!bitmap) 1509 return; 1510 1511 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; 1512 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; 1513 1514 WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data); 1515 } 1516 1517 static u32 gfx_v6_0_get_cu_enabled(struct amdgpu_device *adev) 1518 { 1519 u32 data, mask; 1520 1521 data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG) | 1522 RREG32(mmGC_USER_SHADER_ARRAY_CONFIG); 1523 1524 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh); 1525 return ~REG_GET_FIELD(data, CC_GC_SHADER_ARRAY_CONFIG, INACTIVE_CUS) & mask; 1526 } 1527 1528 1529 static void gfx_v6_0_setup_spi(struct amdgpu_device *adev) 1530 { 1531 int i, j, k; 1532 u32 data, mask; 1533 u32 active_cu = 0; 1534 1535 mutex_lock(&adev->grbm_idx_mutex); 1536 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 1537 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 1538 gfx_v6_0_select_se_sh(adev, i, j, 0xffffffff); 1539 data = RREG32(mmSPI_STATIC_THREAD_MGMT_3); 1540 active_cu = gfx_v6_0_get_cu_enabled(adev); 1541 1542 mask = 1; 1543 for (k = 0; k < 16; k++) { 1544 mask <<= k; 1545 if (active_cu & mask) { 1546 data &= ~mask; 1547 WREG32(mmSPI_STATIC_THREAD_MGMT_3, data); 1548 break; 1549 } 1550 } 1551 } 1552 } 1553 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 1554 mutex_unlock(&adev->grbm_idx_mutex); 1555 } 1556 1557 static void gfx_v6_0_config_init(struct amdgpu_device *adev) 1558 { 1559 adev->gfx.config.double_offchip_lds_buf = 0; 1560 } 1561 1562 static void gfx_v6_0_constants_init(struct amdgpu_device *adev) 1563 { 1564 u32 gb_addr_config = 0; 1565 u32 mc_arb_ramcfg; 1566 u32 sx_debug_1; 1567 u32 hdp_host_path_cntl; 1568 u32 tmp; 1569 1570 switch (adev->asic_type) { 1571 case CHIP_TAHITI: 1572 adev->gfx.config.max_shader_engines = 2; 1573 adev->gfx.config.max_tile_pipes = 12; 1574 adev->gfx.config.max_cu_per_sh = 8; 1575 adev->gfx.config.max_sh_per_se = 2; 1576 adev->gfx.config.max_backends_per_se = 4; 1577 adev->gfx.config.max_texture_channel_caches = 12; 1578 adev->gfx.config.max_gprs = 256; 1579 adev->gfx.config.max_gs_threads = 32; 1580 adev->gfx.config.max_hw_contexts = 8; 1581 1582 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1583 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1584 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1585 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1586 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN; 1587 break; 1588 case CHIP_PITCAIRN: 1589 adev->gfx.config.max_shader_engines = 2; 1590 adev->gfx.config.max_tile_pipes = 8; 1591 adev->gfx.config.max_cu_per_sh = 5; 1592 adev->gfx.config.max_sh_per_se = 2; 1593 adev->gfx.config.max_backends_per_se = 4; 1594 adev->gfx.config.max_texture_channel_caches = 8; 1595 adev->gfx.config.max_gprs = 256; 1596 adev->gfx.config.max_gs_threads = 32; 1597 adev->gfx.config.max_hw_contexts = 8; 1598 1599 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1600 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 1601 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1602 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1603 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN; 1604 break; 1605 case CHIP_VERDE: 1606 adev->gfx.config.max_shader_engines = 1; 1607 adev->gfx.config.max_tile_pipes = 4; 1608 adev->gfx.config.max_cu_per_sh = 5; 1609 adev->gfx.config.max_sh_per_se = 2; 1610 adev->gfx.config.max_backends_per_se = 4; 1611 adev->gfx.config.max_texture_channel_caches = 4; 1612 adev->gfx.config.max_gprs = 256; 1613 adev->gfx.config.max_gs_threads = 32; 1614 adev->gfx.config.max_hw_contexts = 8; 1615 1616 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1617 adev->gfx.config.sc_prim_fifo_size_backend = 0x40; 1618 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1619 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1620 gb_addr_config = VERDE_GB_ADDR_CONFIG_GOLDEN; 1621 break; 1622 case CHIP_OLAND: 1623 adev->gfx.config.max_shader_engines = 1; 1624 adev->gfx.config.max_tile_pipes = 4; 1625 adev->gfx.config.max_cu_per_sh = 6; 1626 adev->gfx.config.max_sh_per_se = 1; 1627 adev->gfx.config.max_backends_per_se = 2; 1628 adev->gfx.config.max_texture_channel_caches = 4; 1629 adev->gfx.config.max_gprs = 256; 1630 adev->gfx.config.max_gs_threads = 16; 1631 adev->gfx.config.max_hw_contexts = 8; 1632 1633 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1634 adev->gfx.config.sc_prim_fifo_size_backend = 0x40; 1635 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1636 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1637 gb_addr_config = VERDE_GB_ADDR_CONFIG_GOLDEN; 1638 break; 1639 case CHIP_HAINAN: 1640 adev->gfx.config.max_shader_engines = 1; 1641 adev->gfx.config.max_tile_pipes = 4; 1642 adev->gfx.config.max_cu_per_sh = 5; 1643 adev->gfx.config.max_sh_per_se = 1; 1644 adev->gfx.config.max_backends_per_se = 1; 1645 adev->gfx.config.max_texture_channel_caches = 2; 1646 adev->gfx.config.max_gprs = 256; 1647 adev->gfx.config.max_gs_threads = 16; 1648 adev->gfx.config.max_hw_contexts = 8; 1649 1650 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 1651 adev->gfx.config.sc_prim_fifo_size_backend = 0x40; 1652 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 1653 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 1654 gb_addr_config = HAINAN_GB_ADDR_CONFIG_GOLDEN; 1655 break; 1656 default: 1657 BUG(); 1658 break; 1659 } 1660 1661 WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT)); 1662 WREG32(mmSRBM_INT_CNTL, 1); 1663 WREG32(mmSRBM_INT_ACK, 1); 1664 1665 WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK); 1666 1667 adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG); 1668 mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg; 1669 1670 adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes; 1671 adev->gfx.config.mem_max_burst_length_bytes = 256; 1672 tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT; 1673 adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; 1674 if (adev->gfx.config.mem_row_size_in_kb > 4) 1675 adev->gfx.config.mem_row_size_in_kb = 4; 1676 adev->gfx.config.shader_engine_tile_size = 32; 1677 adev->gfx.config.num_gpus = 1; 1678 adev->gfx.config.multi_gpu_tile_size = 64; 1679 1680 gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK; 1681 switch (adev->gfx.config.mem_row_size_in_kb) { 1682 case 1: 1683 default: 1684 gb_addr_config |= 0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT; 1685 break; 1686 case 2: 1687 gb_addr_config |= 1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT; 1688 break; 1689 case 4: 1690 gb_addr_config |= 2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT; 1691 break; 1692 } 1693 gb_addr_config &= ~GB_ADDR_CONFIG__NUM_SHADER_ENGINES_MASK; 1694 if (adev->gfx.config.max_shader_engines == 2) 1695 gb_addr_config |= 1 << GB_ADDR_CONFIG__NUM_SHADER_ENGINES__SHIFT; 1696 adev->gfx.config.gb_addr_config = gb_addr_config; 1697 1698 WREG32(mmGB_ADDR_CONFIG, gb_addr_config); 1699 WREG32(mmDMIF_ADDR_CONFIG, gb_addr_config); 1700 WREG32(mmDMIF_ADDR_CALC, gb_addr_config); 1701 WREG32(mmHDP_ADDR_CONFIG, gb_addr_config); 1702 WREG32(mmDMA_TILING_CONFIG + DMA0_REGISTER_OFFSET, gb_addr_config); 1703 WREG32(mmDMA_TILING_CONFIG + DMA1_REGISTER_OFFSET, gb_addr_config); 1704 1705 #if 0 1706 if (adev->has_uvd) { 1707 WREG32(mmUVD_UDEC_ADDR_CONFIG, gb_addr_config); 1708 WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, gb_addr_config); 1709 WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config); 1710 } 1711 #endif 1712 gfx_v6_0_tiling_mode_table_init(adev); 1713 1714 gfx_v6_0_setup_rb(adev); 1715 1716 gfx_v6_0_setup_spi(adev); 1717 1718 gfx_v6_0_get_cu_info(adev); 1719 gfx_v6_0_config_init(adev); 1720 1721 WREG32(mmCP_QUEUE_THRESHOLDS, ((0x16 << CP_QUEUE_THRESHOLDS__ROQ_IB1_START__SHIFT) | 1722 (0x2b << CP_QUEUE_THRESHOLDS__ROQ_IB2_START__SHIFT))); 1723 WREG32(mmCP_MEQ_THRESHOLDS, (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) | 1724 (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT)); 1725 1726 sx_debug_1 = RREG32(mmSX_DEBUG_1); 1727 WREG32(mmSX_DEBUG_1, sx_debug_1); 1728 1729 WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT)); 1730 1731 WREG32(mmPA_SC_FIFO_SIZE, ((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) | 1732 (adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) | 1733 (adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) | 1734 (adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT))); 1735 1736 WREG32(mmVGT_NUM_INSTANCES, 1); 1737 WREG32(mmCP_PERFMON_CNTL, 0); 1738 WREG32(mmSQ_CONFIG, 0); 1739 WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS, ((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) | 1740 (255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT))); 1741 1742 WREG32(mmVGT_CACHE_INVALIDATION, 1743 (VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) | 1744 (ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT)); 1745 1746 WREG32(mmVGT_GS_VERTEX_REUSE, 16); 1747 WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0); 1748 1749 WREG32(mmCB_PERFCOUNTER0_SELECT0, 0); 1750 WREG32(mmCB_PERFCOUNTER0_SELECT1, 0); 1751 WREG32(mmCB_PERFCOUNTER1_SELECT0, 0); 1752 WREG32(mmCB_PERFCOUNTER1_SELECT1, 0); 1753 WREG32(mmCB_PERFCOUNTER2_SELECT0, 0); 1754 WREG32(mmCB_PERFCOUNTER2_SELECT1, 0); 1755 WREG32(mmCB_PERFCOUNTER3_SELECT0, 0); 1756 WREG32(mmCB_PERFCOUNTER3_SELECT1, 0); 1757 1758 hdp_host_path_cntl = RREG32(mmHDP_HOST_PATH_CNTL); 1759 WREG32(mmHDP_HOST_PATH_CNTL, hdp_host_path_cntl); 1760 1761 WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK | 1762 (3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT)); 1763 1764 udelay(50); 1765 } 1766 1767 static int gfx_v6_0_ring_test_ring(struct amdgpu_ring *ring) 1768 { 1769 struct amdgpu_device *adev = ring->adev; 1770 uint32_t tmp = 0; 1771 unsigned i; 1772 int r; 1773 1774 WREG32(mmSCRATCH_REG0, 0xCAFEDEAD); 1775 1776 r = amdgpu_ring_alloc(ring, 3); 1777 if (r) 1778 return r; 1779 1780 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 1781 amdgpu_ring_write(ring, mmSCRATCH_REG0 - PACKET3_SET_CONFIG_REG_START); 1782 amdgpu_ring_write(ring, 0xDEADBEEF); 1783 amdgpu_ring_commit(ring); 1784 1785 for (i = 0; i < adev->usec_timeout; i++) { 1786 tmp = RREG32(mmSCRATCH_REG0); 1787 if (tmp == 0xDEADBEEF) 1788 break; 1789 udelay(1); 1790 } 1791 1792 if (i >= adev->usec_timeout) 1793 r = -ETIMEDOUT; 1794 return r; 1795 } 1796 1797 static void gfx_v6_0_ring_emit_vgt_flush(struct amdgpu_ring *ring) 1798 { 1799 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0)); 1800 amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) | 1801 EVENT_INDEX(0)); 1802 } 1803 1804 static void gfx_v6_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, 1805 u64 seq, unsigned flags) 1806 { 1807 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT; 1808 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT; 1809 /* flush read cache over gart */ 1810 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 1811 amdgpu_ring_write(ring, (mmCP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START)); 1812 amdgpu_ring_write(ring, 0); 1813 amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3)); 1814 amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA | 1815 PACKET3_TC_ACTION_ENA | 1816 PACKET3_SH_KCACHE_ACTION_ENA | 1817 PACKET3_SH_ICACHE_ACTION_ENA); 1818 amdgpu_ring_write(ring, 0xFFFFFFFF); 1819 amdgpu_ring_write(ring, 0); 1820 amdgpu_ring_write(ring, 10); /* poll interval */ 1821 /* EVENT_WRITE_EOP - flush caches, send int */ 1822 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4)); 1823 amdgpu_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5)); 1824 amdgpu_ring_write(ring, addr & 0xfffffffc); 1825 amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) | 1826 ((write64bit ? 2 : 1) << CP_EOP_DONE_DATA_CNTL__DATA_SEL__SHIFT) | 1827 ((int_sel ? 2 : 0) << CP_EOP_DONE_DATA_CNTL__INT_SEL__SHIFT)); 1828 amdgpu_ring_write(ring, lower_32_bits(seq)); 1829 amdgpu_ring_write(ring, upper_32_bits(seq)); 1830 } 1831 1832 static void gfx_v6_0_ring_emit_ib(struct amdgpu_ring *ring, 1833 struct amdgpu_job *job, 1834 struct amdgpu_ib *ib, 1835 uint32_t flags) 1836 { 1837 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 1838 u32 header, control = 0; 1839 1840 /* insert SWITCH_BUFFER packet before first IB in the ring frame */ 1841 if (flags & AMDGPU_HAVE_CTX_SWITCH) { 1842 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 1843 amdgpu_ring_write(ring, 0); 1844 } 1845 1846 if (ib->flags & AMDGPU_IB_FLAG_CE) 1847 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2); 1848 else 1849 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); 1850 1851 control |= ib->length_dw | (vmid << 24); 1852 1853 amdgpu_ring_write(ring, header); 1854 amdgpu_ring_write(ring, 1855 #ifdef __BIG_ENDIAN 1856 (2 << 0) | 1857 #endif 1858 (ib->gpu_addr & 0xFFFFFFFC)); 1859 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF); 1860 amdgpu_ring_write(ring, control); 1861 } 1862 1863 /** 1864 * gfx_v6_0_ring_test_ib - basic ring IB test 1865 * 1866 * @ring: amdgpu_ring structure holding ring information 1867 * @timeout: timeout value in jiffies, or MAX_SCHEDULE_TIMEOUT 1868 * 1869 * Allocate an IB and execute it on the gfx ring (SI). 1870 * Provides a basic gfx ring test to verify that IBs are working. 1871 * Returns 0 on success, error on failure. 1872 */ 1873 static int gfx_v6_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) 1874 { 1875 struct amdgpu_device *adev = ring->adev; 1876 struct dma_fence *f = NULL; 1877 struct amdgpu_ib ib; 1878 uint32_t tmp = 0; 1879 long r; 1880 1881 WREG32(mmSCRATCH_REG0, 0xCAFEDEAD); 1882 memset(&ib, 0, sizeof(ib)); 1883 r = amdgpu_ib_get(adev, NULL, 256, AMDGPU_IB_POOL_DIRECT, &ib); 1884 if (r) 1885 return r; 1886 1887 ib.ptr[0] = PACKET3(PACKET3_SET_CONFIG_REG, 1); 1888 ib.ptr[1] = mmSCRATCH_REG0 - PACKET3_SET_CONFIG_REG_START; 1889 ib.ptr[2] = 0xDEADBEEF; 1890 ib.length_dw = 3; 1891 1892 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); 1893 if (r) 1894 goto error; 1895 1896 r = dma_fence_wait_timeout(f, false, timeout); 1897 if (r == 0) { 1898 r = -ETIMEDOUT; 1899 goto error; 1900 } else if (r < 0) { 1901 goto error; 1902 } 1903 tmp = RREG32(mmSCRATCH_REG0); 1904 if (tmp == 0xDEADBEEF) 1905 r = 0; 1906 else 1907 r = -EINVAL; 1908 1909 error: 1910 amdgpu_ib_free(adev, &ib, NULL); 1911 dma_fence_put(f); 1912 return r; 1913 } 1914 1915 static void gfx_v6_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable) 1916 { 1917 if (enable) { 1918 WREG32(mmCP_ME_CNTL, 0); 1919 } else { 1920 WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK | 1921 CP_ME_CNTL__PFP_HALT_MASK | 1922 CP_ME_CNTL__CE_HALT_MASK)); 1923 WREG32(mmSCRATCH_UMSK, 0); 1924 } 1925 udelay(50); 1926 } 1927 1928 static int gfx_v6_0_cp_gfx_load_microcode(struct amdgpu_device *adev) 1929 { 1930 unsigned i; 1931 const struct gfx_firmware_header_v1_0 *pfp_hdr; 1932 const struct gfx_firmware_header_v1_0 *ce_hdr; 1933 const struct gfx_firmware_header_v1_0 *me_hdr; 1934 const __le32 *fw_data; 1935 u32 fw_size; 1936 1937 if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw) 1938 return -EINVAL; 1939 1940 gfx_v6_0_cp_gfx_enable(adev, false); 1941 pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; 1942 ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; 1943 me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; 1944 1945 amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header); 1946 amdgpu_ucode_print_gfx_hdr(&ce_hdr->header); 1947 amdgpu_ucode_print_gfx_hdr(&me_hdr->header); 1948 1949 /* PFP */ 1950 fw_data = (const __le32 *) 1951 (adev->gfx.pfp_fw->data + le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes)); 1952 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4; 1953 WREG32(mmCP_PFP_UCODE_ADDR, 0); 1954 for (i = 0; i < fw_size; i++) 1955 WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++)); 1956 WREG32(mmCP_PFP_UCODE_ADDR, 0); 1957 1958 /* CE */ 1959 fw_data = (const __le32 *) 1960 (adev->gfx.ce_fw->data + le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes)); 1961 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4; 1962 WREG32(mmCP_CE_UCODE_ADDR, 0); 1963 for (i = 0; i < fw_size; i++) 1964 WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++)); 1965 WREG32(mmCP_CE_UCODE_ADDR, 0); 1966 1967 /* ME */ 1968 fw_data = (const __be32 *) 1969 (adev->gfx.me_fw->data + le32_to_cpu(me_hdr->header.ucode_array_offset_bytes)); 1970 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4; 1971 WREG32(mmCP_ME_RAM_WADDR, 0); 1972 for (i = 0; i < fw_size; i++) 1973 WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++)); 1974 WREG32(mmCP_ME_RAM_WADDR, 0); 1975 1976 WREG32(mmCP_PFP_UCODE_ADDR, 0); 1977 WREG32(mmCP_CE_UCODE_ADDR, 0); 1978 WREG32(mmCP_ME_RAM_WADDR, 0); 1979 WREG32(mmCP_ME_RAM_RADDR, 0); 1980 return 0; 1981 } 1982 1983 static int gfx_v6_0_cp_gfx_start(struct amdgpu_device *adev) 1984 { 1985 const struct cs_section_def *sect = NULL; 1986 const struct cs_extent_def *ext = NULL; 1987 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0]; 1988 int r, i; 1989 1990 r = amdgpu_ring_alloc(ring, 7 + 4); 1991 if (r) { 1992 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r); 1993 return r; 1994 } 1995 amdgpu_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5)); 1996 amdgpu_ring_write(ring, 0x1); 1997 amdgpu_ring_write(ring, 0x0); 1998 amdgpu_ring_write(ring, adev->gfx.config.max_hw_contexts - 1); 1999 amdgpu_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1)); 2000 amdgpu_ring_write(ring, 0); 2001 amdgpu_ring_write(ring, 0); 2002 2003 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2)); 2004 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE)); 2005 amdgpu_ring_write(ring, 0xc000); 2006 amdgpu_ring_write(ring, 0xe000); 2007 amdgpu_ring_commit(ring); 2008 2009 gfx_v6_0_cp_gfx_enable(adev, true); 2010 2011 r = amdgpu_ring_alloc(ring, gfx_v6_0_get_csb_size(adev) + 10); 2012 if (r) { 2013 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r); 2014 return r; 2015 } 2016 2017 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 2018 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 2019 2020 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 2021 for (ext = sect->section; ext->extent != NULL; ++ext) { 2022 if (sect->id == SECT_CONTEXT) { 2023 amdgpu_ring_write(ring, 2024 PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count)); 2025 amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START); 2026 for (i = 0; i < ext->reg_count; i++) 2027 amdgpu_ring_write(ring, ext->extent[i]); 2028 } 2029 } 2030 } 2031 2032 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 2033 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE); 2034 2035 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0)); 2036 amdgpu_ring_write(ring, 0); 2037 2038 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2)); 2039 amdgpu_ring_write(ring, 0x00000316); 2040 amdgpu_ring_write(ring, 0x0000000e); 2041 amdgpu_ring_write(ring, 0x00000010); 2042 2043 amdgpu_ring_commit(ring); 2044 2045 return 0; 2046 } 2047 2048 static int gfx_v6_0_cp_gfx_resume(struct amdgpu_device *adev) 2049 { 2050 struct amdgpu_ring *ring; 2051 u32 tmp; 2052 u32 rb_bufsz; 2053 int r; 2054 u64 rptr_addr; 2055 2056 WREG32(mmCP_SEM_WAIT_TIMER, 0x0); 2057 WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0); 2058 2059 /* Set the write pointer delay */ 2060 WREG32(mmCP_RB_WPTR_DELAY, 0); 2061 2062 WREG32(mmCP_DEBUG, 0); 2063 WREG32(mmSCRATCH_ADDR, 0); 2064 2065 /* ring 0 - compute and gfx */ 2066 /* Set ring buffer size */ 2067 ring = &adev->gfx.gfx_ring[0]; 2068 rb_bufsz = order_base_2(ring->ring_size / 8); 2069 tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 2070 2071 #ifdef __BIG_ENDIAN 2072 tmp |= BUF_SWAP_32BIT; 2073 #endif 2074 WREG32(mmCP_RB0_CNTL, tmp); 2075 2076 /* Initialize the ring buffer's read and write pointers */ 2077 WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK); 2078 ring->wptr = 0; 2079 WREG32(mmCP_RB0_WPTR, ring->wptr); 2080 2081 /* set the wb address whether it's enabled or not */ 2082 rptr_addr = ring->rptr_gpu_addr; 2083 WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr)); 2084 WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF); 2085 2086 WREG32(mmSCRATCH_UMSK, 0); 2087 2088 mdelay(1); 2089 WREG32(mmCP_RB0_CNTL, tmp); 2090 2091 WREG32(mmCP_RB0_BASE, ring->gpu_addr >> 8); 2092 2093 /* start the rings */ 2094 gfx_v6_0_cp_gfx_start(adev); 2095 r = amdgpu_ring_test_helper(ring); 2096 if (r) 2097 return r; 2098 2099 return 0; 2100 } 2101 2102 static u64 gfx_v6_0_ring_get_rptr(struct amdgpu_ring *ring) 2103 { 2104 return *ring->rptr_cpu_addr; 2105 } 2106 2107 static u64 gfx_v6_0_ring_get_wptr(struct amdgpu_ring *ring) 2108 { 2109 struct amdgpu_device *adev = ring->adev; 2110 2111 if (ring == &adev->gfx.gfx_ring[0]) 2112 return RREG32(mmCP_RB0_WPTR); 2113 else if (ring == &adev->gfx.compute_ring[0]) 2114 return RREG32(mmCP_RB1_WPTR); 2115 else if (ring == &adev->gfx.compute_ring[1]) 2116 return RREG32(mmCP_RB2_WPTR); 2117 else 2118 BUG(); 2119 } 2120 2121 static void gfx_v6_0_ring_set_wptr_gfx(struct amdgpu_ring *ring) 2122 { 2123 struct amdgpu_device *adev = ring->adev; 2124 2125 WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr)); 2126 (void)RREG32(mmCP_RB0_WPTR); 2127 } 2128 2129 static void gfx_v6_0_ring_set_wptr_compute(struct amdgpu_ring *ring) 2130 { 2131 struct amdgpu_device *adev = ring->adev; 2132 2133 if (ring == &adev->gfx.compute_ring[0]) { 2134 WREG32(mmCP_RB1_WPTR, lower_32_bits(ring->wptr)); 2135 (void)RREG32(mmCP_RB1_WPTR); 2136 } else if (ring == &adev->gfx.compute_ring[1]) { 2137 WREG32(mmCP_RB2_WPTR, lower_32_bits(ring->wptr)); 2138 (void)RREG32(mmCP_RB2_WPTR); 2139 } else { 2140 BUG(); 2141 } 2142 2143 } 2144 2145 static int gfx_v6_0_cp_compute_resume(struct amdgpu_device *adev) 2146 { 2147 struct amdgpu_ring *ring; 2148 u32 tmp; 2149 u32 rb_bufsz; 2150 int i, r; 2151 u64 rptr_addr; 2152 2153 /* ring1 - compute only */ 2154 /* Set ring buffer size */ 2155 2156 ring = &adev->gfx.compute_ring[0]; 2157 rb_bufsz = order_base_2(ring->ring_size / 8); 2158 tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 2159 #ifdef __BIG_ENDIAN 2160 tmp |= BUF_SWAP_32BIT; 2161 #endif 2162 WREG32(mmCP_RB1_CNTL, tmp); 2163 2164 WREG32(mmCP_RB1_CNTL, tmp | CP_RB1_CNTL__RB_RPTR_WR_ENA_MASK); 2165 ring->wptr = 0; 2166 WREG32(mmCP_RB1_WPTR, ring->wptr); 2167 2168 rptr_addr = ring->rptr_gpu_addr; 2169 WREG32(mmCP_RB1_RPTR_ADDR, lower_32_bits(rptr_addr)); 2170 WREG32(mmCP_RB1_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF); 2171 2172 mdelay(1); 2173 WREG32(mmCP_RB1_CNTL, tmp); 2174 WREG32(mmCP_RB1_BASE, ring->gpu_addr >> 8); 2175 2176 ring = &adev->gfx.compute_ring[1]; 2177 rb_bufsz = order_base_2(ring->ring_size / 8); 2178 tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 2179 #ifdef __BIG_ENDIAN 2180 tmp |= BUF_SWAP_32BIT; 2181 #endif 2182 WREG32(mmCP_RB2_CNTL, tmp); 2183 2184 WREG32(mmCP_RB2_CNTL, tmp | CP_RB2_CNTL__RB_RPTR_WR_ENA_MASK); 2185 ring->wptr = 0; 2186 WREG32(mmCP_RB2_WPTR, ring->wptr); 2187 rptr_addr = ring->rptr_gpu_addr; 2188 WREG32(mmCP_RB2_RPTR_ADDR, lower_32_bits(rptr_addr)); 2189 WREG32(mmCP_RB2_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF); 2190 2191 mdelay(1); 2192 WREG32(mmCP_RB2_CNTL, tmp); 2193 WREG32(mmCP_RB2_BASE, ring->gpu_addr >> 8); 2194 2195 2196 for (i = 0; i < 2; i++) { 2197 r = amdgpu_ring_test_helper(&adev->gfx.compute_ring[i]); 2198 if (r) 2199 return r; 2200 } 2201 2202 return 0; 2203 } 2204 2205 static void gfx_v6_0_cp_enable(struct amdgpu_device *adev, bool enable) 2206 { 2207 gfx_v6_0_cp_gfx_enable(adev, enable); 2208 } 2209 2210 static int gfx_v6_0_cp_load_microcode(struct amdgpu_device *adev) 2211 { 2212 return gfx_v6_0_cp_gfx_load_microcode(adev); 2213 } 2214 2215 static void gfx_v6_0_enable_gui_idle_interrupt(struct amdgpu_device *adev, 2216 bool enable) 2217 { 2218 u32 tmp = RREG32(mmCP_INT_CNTL_RING0); 2219 u32 mask; 2220 int i; 2221 2222 if (enable) 2223 tmp |= (CP_INT_CNTL__CNTX_BUSY_INT_ENABLE_MASK | 2224 CP_INT_CNTL__CNTX_EMPTY_INT_ENABLE_MASK); 2225 else 2226 tmp &= ~(CP_INT_CNTL__CNTX_BUSY_INT_ENABLE_MASK | 2227 CP_INT_CNTL__CNTX_EMPTY_INT_ENABLE_MASK); 2228 WREG32(mmCP_INT_CNTL_RING0, tmp); 2229 2230 if (!enable) { 2231 /* read a gfx register */ 2232 tmp = RREG32(mmDB_DEPTH_INFO); 2233 2234 mask = RLC_BUSY_STATUS | GFX_POWER_STATUS | GFX_CLOCK_STATUS | GFX_LS_STATUS; 2235 for (i = 0; i < adev->usec_timeout; i++) { 2236 if ((RREG32(mmRLC_STAT) & mask) == (GFX_CLOCK_STATUS | GFX_POWER_STATUS)) 2237 break; 2238 udelay(1); 2239 } 2240 } 2241 } 2242 2243 static int gfx_v6_0_cp_resume(struct amdgpu_device *adev) 2244 { 2245 int r; 2246 2247 gfx_v6_0_enable_gui_idle_interrupt(adev, false); 2248 2249 r = gfx_v6_0_cp_load_microcode(adev); 2250 if (r) 2251 return r; 2252 2253 r = gfx_v6_0_cp_gfx_resume(adev); 2254 if (r) 2255 return r; 2256 r = gfx_v6_0_cp_compute_resume(adev); 2257 if (r) 2258 return r; 2259 2260 gfx_v6_0_enable_gui_idle_interrupt(adev, true); 2261 2262 return 0; 2263 } 2264 2265 static void gfx_v6_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring) 2266 { 2267 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 2268 uint32_t seq = ring->fence_drv.sync_seq; 2269 uint64_t addr = ring->fence_drv.gpu_addr; 2270 2271 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 2272 amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */ 2273 WAIT_REG_MEM_FUNCTION(3) | /* equal */ 2274 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */ 2275 amdgpu_ring_write(ring, addr & 0xfffffffc); 2276 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); 2277 amdgpu_ring_write(ring, seq); 2278 amdgpu_ring_write(ring, 0xffffffff); 2279 amdgpu_ring_write(ring, 4); /* poll interval */ 2280 2281 if (usepfp) { 2282 /* synce CE with ME to prevent CE fetch CEIB before context switch done */ 2283 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 2284 amdgpu_ring_write(ring, 0); 2285 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 2286 amdgpu_ring_write(ring, 0); 2287 } 2288 } 2289 2290 static void gfx_v6_0_ring_emit_vm_flush(struct amdgpu_ring *ring, 2291 unsigned vmid, uint64_t pd_addr) 2292 { 2293 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 2294 2295 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 2296 2297 /* wait for the invalidate to complete */ 2298 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 2299 amdgpu_ring_write(ring, (WAIT_REG_MEM_FUNCTION(0) | /* always */ 2300 WAIT_REG_MEM_ENGINE(0))); /* me */ 2301 amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST); 2302 amdgpu_ring_write(ring, 0); 2303 amdgpu_ring_write(ring, 0); /* ref */ 2304 amdgpu_ring_write(ring, 0); /* mask */ 2305 amdgpu_ring_write(ring, 0x20); /* poll interval */ 2306 2307 if (usepfp) { 2308 /* sync PFP to ME, otherwise we might get invalid PFP reads */ 2309 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0)); 2310 amdgpu_ring_write(ring, 0x0); 2311 2312 /* synce CE with ME to prevent CE fetch CEIB before context switch done */ 2313 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 2314 amdgpu_ring_write(ring, 0); 2315 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 2316 amdgpu_ring_write(ring, 0); 2317 } 2318 } 2319 2320 static void gfx_v6_0_ring_emit_wreg(struct amdgpu_ring *ring, 2321 uint32_t reg, uint32_t val) 2322 { 2323 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 2324 2325 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 2326 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) | 2327 WRITE_DATA_DST_SEL(0))); 2328 amdgpu_ring_write(ring, reg); 2329 amdgpu_ring_write(ring, 0); 2330 amdgpu_ring_write(ring, val); 2331 } 2332 2333 static int gfx_v6_0_rlc_init(struct amdgpu_device *adev) 2334 { 2335 const u32 *src_ptr; 2336 volatile u32 *dst_ptr; 2337 u32 dws; 2338 u64 reg_list_mc_addr; 2339 const struct cs_section_def *cs_data; 2340 int r; 2341 2342 adev->gfx.rlc.reg_list = verde_rlc_save_restore_register_list; 2343 adev->gfx.rlc.reg_list_size = 2344 (u32)ARRAY_SIZE(verde_rlc_save_restore_register_list); 2345 2346 adev->gfx.rlc.cs_data = si_cs_data; 2347 src_ptr = adev->gfx.rlc.reg_list; 2348 dws = adev->gfx.rlc.reg_list_size; 2349 cs_data = adev->gfx.rlc.cs_data; 2350 2351 if (src_ptr) { 2352 /* init save restore block */ 2353 r = amdgpu_gfx_rlc_init_sr(adev, dws); 2354 if (r) 2355 return r; 2356 } 2357 2358 if (cs_data) { 2359 /* clear state block */ 2360 adev->gfx.rlc.clear_state_size = gfx_v6_0_get_csb_size(adev); 2361 dws = adev->gfx.rlc.clear_state_size + (256 / 4); 2362 2363 r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, 2364 AMDGPU_GEM_DOMAIN_VRAM | 2365 AMDGPU_GEM_DOMAIN_GTT, 2366 &adev->gfx.rlc.clear_state_obj, 2367 &adev->gfx.rlc.clear_state_gpu_addr, 2368 (void **)&adev->gfx.rlc.cs_ptr); 2369 if (r) { 2370 dev_warn(adev->dev, "(%d) create RLC c bo failed\n", r); 2371 amdgpu_gfx_rlc_fini(adev); 2372 return r; 2373 } 2374 2375 /* set up the cs buffer */ 2376 dst_ptr = adev->gfx.rlc.cs_ptr; 2377 reg_list_mc_addr = adev->gfx.rlc.clear_state_gpu_addr + 256; 2378 dst_ptr[0] = cpu_to_le32(upper_32_bits(reg_list_mc_addr)); 2379 dst_ptr[1] = cpu_to_le32(lower_32_bits(reg_list_mc_addr)); 2380 dst_ptr[2] = cpu_to_le32(adev->gfx.rlc.clear_state_size); 2381 gfx_v6_0_get_csb_buffer(adev, &dst_ptr[(256/4)]); 2382 amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj); 2383 amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj); 2384 } 2385 2386 return 0; 2387 } 2388 2389 static void gfx_v6_0_enable_lbpw(struct amdgpu_device *adev, bool enable) 2390 { 2391 WREG32_FIELD(RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0); 2392 2393 if (!enable) { 2394 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 2395 WREG32(mmSPI_LB_CU_MASK, 0x00ff); 2396 } 2397 } 2398 2399 static void gfx_v6_0_wait_for_rlc_serdes(struct amdgpu_device *adev) 2400 { 2401 int i; 2402 2403 for (i = 0; i < adev->usec_timeout; i++) { 2404 if (RREG32(mmRLC_SERDES_MASTER_BUSY_0) == 0) 2405 break; 2406 udelay(1); 2407 } 2408 2409 for (i = 0; i < adev->usec_timeout; i++) { 2410 if (RREG32(mmRLC_SERDES_MASTER_BUSY_1) == 0) 2411 break; 2412 udelay(1); 2413 } 2414 } 2415 2416 static void gfx_v6_0_update_rlc(struct amdgpu_device *adev, u32 rlc) 2417 { 2418 u32 tmp; 2419 2420 tmp = RREG32(mmRLC_CNTL); 2421 if (tmp != rlc) 2422 WREG32(mmRLC_CNTL, rlc); 2423 } 2424 2425 static u32 gfx_v6_0_halt_rlc(struct amdgpu_device *adev) 2426 { 2427 u32 data, orig; 2428 2429 orig = data = RREG32(mmRLC_CNTL); 2430 2431 if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) { 2432 data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK; 2433 WREG32(mmRLC_CNTL, data); 2434 2435 gfx_v6_0_wait_for_rlc_serdes(adev); 2436 } 2437 2438 return orig; 2439 } 2440 2441 static void gfx_v6_0_rlc_stop(struct amdgpu_device *adev) 2442 { 2443 WREG32(mmRLC_CNTL, 0); 2444 2445 gfx_v6_0_enable_gui_idle_interrupt(adev, false); 2446 gfx_v6_0_wait_for_rlc_serdes(adev); 2447 } 2448 2449 static void gfx_v6_0_rlc_start(struct amdgpu_device *adev) 2450 { 2451 WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK); 2452 2453 gfx_v6_0_enable_gui_idle_interrupt(adev, true); 2454 2455 udelay(50); 2456 } 2457 2458 static void gfx_v6_0_rlc_reset(struct amdgpu_device *adev) 2459 { 2460 WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 1); 2461 udelay(50); 2462 WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 0); 2463 udelay(50); 2464 } 2465 2466 static bool gfx_v6_0_lbpw_supported(struct amdgpu_device *adev) 2467 { 2468 u32 tmp; 2469 2470 /* Enable LBPW only for DDR3 */ 2471 tmp = RREG32(mmMC_SEQ_MISC0); 2472 if ((tmp & 0xF0000000) == 0xB0000000) 2473 return true; 2474 return false; 2475 } 2476 2477 static void gfx_v6_0_init_cg(struct amdgpu_device *adev) 2478 { 2479 } 2480 2481 static int gfx_v6_0_rlc_resume(struct amdgpu_device *adev) 2482 { 2483 u32 i; 2484 const struct rlc_firmware_header_v1_0 *hdr; 2485 const __le32 *fw_data; 2486 u32 fw_size; 2487 2488 2489 if (!adev->gfx.rlc_fw) 2490 return -EINVAL; 2491 2492 adev->gfx.rlc.funcs->stop(adev); 2493 adev->gfx.rlc.funcs->reset(adev); 2494 gfx_v6_0_init_pg(adev); 2495 gfx_v6_0_init_cg(adev); 2496 2497 WREG32(mmRLC_RL_BASE, 0); 2498 WREG32(mmRLC_RL_SIZE, 0); 2499 WREG32(mmRLC_LB_CNTL, 0); 2500 WREG32(mmRLC_LB_CNTR_MAX, 0xffffffff); 2501 WREG32(mmRLC_LB_CNTR_INIT, 0); 2502 WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff); 2503 2504 WREG32(mmRLC_MC_CNTL, 0); 2505 WREG32(mmRLC_UCODE_CNTL, 0); 2506 2507 hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data; 2508 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; 2509 fw_data = (const __le32 *) 2510 (adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2511 2512 amdgpu_ucode_print_rlc_hdr(&hdr->header); 2513 2514 for (i = 0; i < fw_size; i++) { 2515 WREG32(mmRLC_UCODE_ADDR, i); 2516 WREG32(mmRLC_UCODE_DATA, le32_to_cpup(fw_data++)); 2517 } 2518 WREG32(mmRLC_UCODE_ADDR, 0); 2519 2520 gfx_v6_0_enable_lbpw(adev, gfx_v6_0_lbpw_supported(adev)); 2521 adev->gfx.rlc.funcs->start(adev); 2522 2523 return 0; 2524 } 2525 2526 static void gfx_v6_0_enable_cgcg(struct amdgpu_device *adev, bool enable) 2527 { 2528 u32 data, orig, tmp; 2529 2530 orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL); 2531 2532 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { 2533 gfx_v6_0_enable_gui_idle_interrupt(adev, true); 2534 2535 WREG32(mmRLC_GCPM_GENERAL_3, 0x00000080); 2536 2537 tmp = gfx_v6_0_halt_rlc(adev); 2538 2539 WREG32(mmRLC_SERDES_WR_MASTER_MASK_0, 0xffffffff); 2540 WREG32(mmRLC_SERDES_WR_MASTER_MASK_1, 0xffffffff); 2541 WREG32(mmRLC_SERDES_WR_CTRL, 0x00b000ff); 2542 2543 gfx_v6_0_wait_for_rlc_serdes(adev); 2544 gfx_v6_0_update_rlc(adev, tmp); 2545 2546 WREG32(mmRLC_SERDES_WR_CTRL, 0x007000ff); 2547 2548 data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK; 2549 } else { 2550 gfx_v6_0_enable_gui_idle_interrupt(adev, false); 2551 2552 RREG32(mmCB_CGTT_SCLK_CTRL); 2553 RREG32(mmCB_CGTT_SCLK_CTRL); 2554 RREG32(mmCB_CGTT_SCLK_CTRL); 2555 RREG32(mmCB_CGTT_SCLK_CTRL); 2556 2557 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK); 2558 } 2559 2560 if (orig != data) 2561 WREG32(mmRLC_CGCG_CGLS_CTRL, data); 2562 2563 } 2564 2565 static void gfx_v6_0_enable_mgcg(struct amdgpu_device *adev, bool enable) 2566 { 2567 2568 u32 data, orig, tmp = 0; 2569 2570 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) { 2571 orig = data = RREG32(mmCGTS_SM_CTRL_REG); 2572 data = 0x96940200; 2573 if (orig != data) 2574 WREG32(mmCGTS_SM_CTRL_REG, data); 2575 2576 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) { 2577 orig = data = RREG32(mmCP_MEM_SLP_CNTL); 2578 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 2579 if (orig != data) 2580 WREG32(mmCP_MEM_SLP_CNTL, data); 2581 } 2582 2583 orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 2584 data &= 0xffffffc0; 2585 if (orig != data) 2586 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data); 2587 2588 tmp = gfx_v6_0_halt_rlc(adev); 2589 2590 WREG32(mmRLC_SERDES_WR_MASTER_MASK_0, 0xffffffff); 2591 WREG32(mmRLC_SERDES_WR_MASTER_MASK_1, 0xffffffff); 2592 WREG32(mmRLC_SERDES_WR_CTRL, 0x00d000ff); 2593 2594 gfx_v6_0_update_rlc(adev, tmp); 2595 } else { 2596 orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 2597 data |= 0x00000003; 2598 if (orig != data) 2599 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data); 2600 2601 data = RREG32(mmCP_MEM_SLP_CNTL); 2602 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) { 2603 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 2604 WREG32(mmCP_MEM_SLP_CNTL, data); 2605 } 2606 orig = data = RREG32(mmCGTS_SM_CTRL_REG); 2607 data |= CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK | CGTS_SM_CTRL_REG__OVERRIDE_MASK; 2608 if (orig != data) 2609 WREG32(mmCGTS_SM_CTRL_REG, data); 2610 2611 tmp = gfx_v6_0_halt_rlc(adev); 2612 2613 WREG32(mmRLC_SERDES_WR_MASTER_MASK_0, 0xffffffff); 2614 WREG32(mmRLC_SERDES_WR_MASTER_MASK_1, 0xffffffff); 2615 WREG32(mmRLC_SERDES_WR_CTRL, 0x00e000ff); 2616 2617 gfx_v6_0_update_rlc(adev, tmp); 2618 } 2619 } 2620 /* 2621 static void gfx_v6_0_update_cg(struct amdgpu_device *adev, 2622 bool enable) 2623 { 2624 gfx_v6_0_enable_gui_idle_interrupt(adev, false); 2625 if (enable) { 2626 gfx_v6_0_enable_mgcg(adev, true); 2627 gfx_v6_0_enable_cgcg(adev, true); 2628 } else { 2629 gfx_v6_0_enable_cgcg(adev, false); 2630 gfx_v6_0_enable_mgcg(adev, false); 2631 } 2632 gfx_v6_0_enable_gui_idle_interrupt(adev, true); 2633 } 2634 */ 2635 2636 static void gfx_v6_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev, 2637 bool enable) 2638 { 2639 } 2640 2641 static void gfx_v6_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev, 2642 bool enable) 2643 { 2644 } 2645 2646 static void gfx_v6_0_enable_cp_pg(struct amdgpu_device *adev, bool enable) 2647 { 2648 u32 data, orig; 2649 2650 orig = data = RREG32(mmRLC_PG_CNTL); 2651 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP)) 2652 data &= ~0x8000; 2653 else 2654 data |= 0x8000; 2655 if (orig != data) 2656 WREG32(mmRLC_PG_CNTL, data); 2657 } 2658 2659 static void gfx_v6_0_enable_gds_pg(struct amdgpu_device *adev, bool enable) 2660 { 2661 } 2662 /* 2663 static void gfx_v6_0_init_cp_pg_table(struct amdgpu_device *adev) 2664 { 2665 const __le32 *fw_data; 2666 volatile u32 *dst_ptr; 2667 int me, i, max_me = 4; 2668 u32 bo_offset = 0; 2669 u32 table_offset, table_size; 2670 2671 if (adev->asic_type == CHIP_KAVERI) 2672 max_me = 5; 2673 2674 if (adev->gfx.rlc.cp_table_ptr == NULL) 2675 return; 2676 2677 dst_ptr = adev->gfx.rlc.cp_table_ptr; 2678 for (me = 0; me < max_me; me++) { 2679 if (me == 0) { 2680 const struct gfx_firmware_header_v1_0 *hdr = 2681 (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; 2682 fw_data = (const __le32 *) 2683 (adev->gfx.ce_fw->data + 2684 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2685 table_offset = le32_to_cpu(hdr->jt_offset); 2686 table_size = le32_to_cpu(hdr->jt_size); 2687 } else if (me == 1) { 2688 const struct gfx_firmware_header_v1_0 *hdr = 2689 (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; 2690 fw_data = (const __le32 *) 2691 (adev->gfx.pfp_fw->data + 2692 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2693 table_offset = le32_to_cpu(hdr->jt_offset); 2694 table_size = le32_to_cpu(hdr->jt_size); 2695 } else if (me == 2) { 2696 const struct gfx_firmware_header_v1_0 *hdr = 2697 (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; 2698 fw_data = (const __le32 *) 2699 (adev->gfx.me_fw->data + 2700 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2701 table_offset = le32_to_cpu(hdr->jt_offset); 2702 table_size = le32_to_cpu(hdr->jt_size); 2703 } else if (me == 3) { 2704 const struct gfx_firmware_header_v1_0 *hdr = 2705 (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; 2706 fw_data = (const __le32 *) 2707 (adev->gfx.mec_fw->data + 2708 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2709 table_offset = le32_to_cpu(hdr->jt_offset); 2710 table_size = le32_to_cpu(hdr->jt_size); 2711 } else { 2712 const struct gfx_firmware_header_v1_0 *hdr = 2713 (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; 2714 fw_data = (const __le32 *) 2715 (adev->gfx.mec2_fw->data + 2716 le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 2717 table_offset = le32_to_cpu(hdr->jt_offset); 2718 table_size = le32_to_cpu(hdr->jt_size); 2719 } 2720 2721 for (i = 0; i < table_size; i ++) { 2722 dst_ptr[bo_offset + i] = 2723 cpu_to_le32(le32_to_cpu(fw_data[table_offset + i])); 2724 } 2725 2726 bo_offset += table_size; 2727 } 2728 } 2729 */ 2730 static void gfx_v6_0_enable_gfx_cgpg(struct amdgpu_device *adev, 2731 bool enable) 2732 { 2733 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) { 2734 WREG32(mmRLC_TTOP_D, RLC_PUD(0x10) | RLC_PDD(0x10) | RLC_TTPD(0x10) | RLC_MSD(0x10)); 2735 WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_ENABLE, 1); 2736 WREG32_FIELD(RLC_AUTO_PG_CTRL, AUTO_PG_EN, 1); 2737 } else { 2738 WREG32_FIELD(RLC_AUTO_PG_CTRL, AUTO_PG_EN, 0); 2739 (void)RREG32(mmDB_RENDER_CONTROL); 2740 } 2741 } 2742 2743 static void gfx_v6_0_init_ao_cu_mask(struct amdgpu_device *adev) 2744 { 2745 u32 tmp; 2746 2747 WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask); 2748 2749 tmp = RREG32(mmRLC_MAX_PG_CU); 2750 tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK; 2751 tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT); 2752 WREG32(mmRLC_MAX_PG_CU, tmp); 2753 } 2754 2755 static void gfx_v6_0_enable_gfx_static_mgpg(struct amdgpu_device *adev, 2756 bool enable) 2757 { 2758 u32 data, orig; 2759 2760 orig = data = RREG32(mmRLC_PG_CNTL); 2761 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG)) 2762 data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK; 2763 else 2764 data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK; 2765 if (orig != data) 2766 WREG32(mmRLC_PG_CNTL, data); 2767 } 2768 2769 static void gfx_v6_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev, 2770 bool enable) 2771 { 2772 u32 data, orig; 2773 2774 orig = data = RREG32(mmRLC_PG_CNTL); 2775 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG)) 2776 data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK; 2777 else 2778 data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK; 2779 if (orig != data) 2780 WREG32(mmRLC_PG_CNTL, data); 2781 } 2782 2783 static void gfx_v6_0_init_gfx_cgpg(struct amdgpu_device *adev) 2784 { 2785 u32 tmp; 2786 2787 WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8); 2788 WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_SRC, 1); 2789 WREG32(mmRLC_CLEAR_STATE_RESTORE_BASE, adev->gfx.rlc.clear_state_gpu_addr >> 8); 2790 2791 tmp = RREG32(mmRLC_AUTO_PG_CTRL); 2792 tmp &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK; 2793 tmp |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT); 2794 tmp &= ~RLC_AUTO_PG_CTRL__PG_AFTER_GRBM_REG_SAVE_THRESHOLD_MASK; 2795 WREG32(mmRLC_AUTO_PG_CTRL, tmp); 2796 } 2797 2798 static void gfx_v6_0_update_gfx_pg(struct amdgpu_device *adev, bool enable) 2799 { 2800 gfx_v6_0_enable_gfx_cgpg(adev, enable); 2801 gfx_v6_0_enable_gfx_static_mgpg(adev, enable); 2802 gfx_v6_0_enable_gfx_dynamic_mgpg(adev, enable); 2803 } 2804 2805 static u32 gfx_v6_0_get_csb_size(struct amdgpu_device *adev) 2806 { 2807 u32 count = 0; 2808 const struct cs_section_def *sect = NULL; 2809 const struct cs_extent_def *ext = NULL; 2810 2811 if (adev->gfx.rlc.cs_data == NULL) 2812 return 0; 2813 2814 /* begin clear state */ 2815 count += 2; 2816 /* context control state */ 2817 count += 3; 2818 2819 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 2820 for (ext = sect->section; ext->extent != NULL; ++ext) { 2821 if (sect->id == SECT_CONTEXT) 2822 count += 2 + ext->reg_count; 2823 else 2824 return 0; 2825 } 2826 } 2827 /* pa_sc_raster_config */ 2828 count += 3; 2829 /* end clear state */ 2830 count += 2; 2831 /* clear state */ 2832 count += 2; 2833 2834 return count; 2835 } 2836 2837 static void gfx_v6_0_get_csb_buffer(struct amdgpu_device *adev, 2838 volatile u32 *buffer) 2839 { 2840 u32 count = 0, i; 2841 const struct cs_section_def *sect = NULL; 2842 const struct cs_extent_def *ext = NULL; 2843 2844 if (adev->gfx.rlc.cs_data == NULL) 2845 return; 2846 if (buffer == NULL) 2847 return; 2848 2849 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 2850 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 2851 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 2852 buffer[count++] = cpu_to_le32(0x80000000); 2853 buffer[count++] = cpu_to_le32(0x80000000); 2854 2855 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 2856 for (ext = sect->section; ext->extent != NULL; ++ext) { 2857 if (sect->id == SECT_CONTEXT) { 2858 buffer[count++] = 2859 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count)); 2860 buffer[count++] = cpu_to_le32(ext->reg_index - 0xa000); 2861 for (i = 0; i < ext->reg_count; i++) 2862 buffer[count++] = cpu_to_le32(ext->extent[i]); 2863 } else { 2864 return; 2865 } 2866 } 2867 } 2868 2869 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 1)); 2870 buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START); 2871 buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config); 2872 2873 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 2874 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE); 2875 2876 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0)); 2877 buffer[count++] = cpu_to_le32(0); 2878 } 2879 2880 static void gfx_v6_0_init_pg(struct amdgpu_device *adev) 2881 { 2882 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | 2883 AMD_PG_SUPPORT_GFX_SMG | 2884 AMD_PG_SUPPORT_GFX_DMG | 2885 AMD_PG_SUPPORT_CP | 2886 AMD_PG_SUPPORT_GDS | 2887 AMD_PG_SUPPORT_RLC_SMU_HS)) { 2888 gfx_v6_0_enable_sclk_slowdown_on_pu(adev, true); 2889 gfx_v6_0_enable_sclk_slowdown_on_pd(adev, true); 2890 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { 2891 gfx_v6_0_init_gfx_cgpg(adev); 2892 gfx_v6_0_enable_cp_pg(adev, true); 2893 gfx_v6_0_enable_gds_pg(adev, true); 2894 } else { 2895 WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8); 2896 WREG32(mmRLC_CLEAR_STATE_RESTORE_BASE, adev->gfx.rlc.clear_state_gpu_addr >> 8); 2897 2898 } 2899 gfx_v6_0_init_ao_cu_mask(adev); 2900 gfx_v6_0_update_gfx_pg(adev, true); 2901 } else { 2902 2903 WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8); 2904 WREG32(mmRLC_CLEAR_STATE_RESTORE_BASE, adev->gfx.rlc.clear_state_gpu_addr >> 8); 2905 } 2906 } 2907 2908 static void gfx_v6_0_fini_pg(struct amdgpu_device *adev) 2909 { 2910 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | 2911 AMD_PG_SUPPORT_GFX_SMG | 2912 AMD_PG_SUPPORT_GFX_DMG | 2913 AMD_PG_SUPPORT_CP | 2914 AMD_PG_SUPPORT_GDS | 2915 AMD_PG_SUPPORT_RLC_SMU_HS)) { 2916 gfx_v6_0_update_gfx_pg(adev, false); 2917 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { 2918 gfx_v6_0_enable_cp_pg(adev, false); 2919 gfx_v6_0_enable_gds_pg(adev, false); 2920 } 2921 } 2922 } 2923 2924 static uint64_t gfx_v6_0_get_gpu_clock_counter(struct amdgpu_device *adev) 2925 { 2926 uint64_t clock; 2927 2928 mutex_lock(&adev->gfx.gpu_clock_mutex); 2929 WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1); 2930 clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) | 2931 ((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL); 2932 mutex_unlock(&adev->gfx.gpu_clock_mutex); 2933 return clock; 2934 } 2935 2936 static void gfx_v6_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags) 2937 { 2938 if (flags & AMDGPU_HAVE_CTX_SWITCH) 2939 gfx_v6_0_ring_emit_vgt_flush(ring); 2940 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 2941 amdgpu_ring_write(ring, 0x80000000); 2942 amdgpu_ring_write(ring, 0); 2943 } 2944 2945 2946 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address) 2947 { 2948 WREG32(mmSQ_IND_INDEX, 2949 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 2950 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 2951 (address << SQ_IND_INDEX__INDEX__SHIFT) | 2952 (SQ_IND_INDEX__FORCE_READ_MASK)); 2953 return RREG32(mmSQ_IND_DATA); 2954 } 2955 2956 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd, 2957 uint32_t wave, uint32_t thread, 2958 uint32_t regno, uint32_t num, uint32_t *out) 2959 { 2960 WREG32(mmSQ_IND_INDEX, 2961 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 2962 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 2963 (regno << SQ_IND_INDEX__INDEX__SHIFT) | 2964 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) | 2965 (SQ_IND_INDEX__FORCE_READ_MASK) | 2966 (SQ_IND_INDEX__AUTO_INCR_MASK)); 2967 while (num--) 2968 *(out++) = RREG32(mmSQ_IND_DATA); 2969 } 2970 2971 static void gfx_v6_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields) 2972 { 2973 /* type 0 wave data */ 2974 dst[(*no_fields)++] = 0; 2975 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS); 2976 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO); 2977 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI); 2978 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO); 2979 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI); 2980 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID); 2981 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0); 2982 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1); 2983 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC); 2984 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC); 2985 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS); 2986 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS); 2987 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO); 2988 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI); 2989 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO); 2990 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI); 2991 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0); 2992 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0); 2993 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE); 2994 } 2995 2996 static void gfx_v6_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd, 2997 uint32_t wave, uint32_t start, 2998 uint32_t size, uint32_t *dst) 2999 { 3000 wave_read_regs( 3001 adev, simd, wave, 0, 3002 start + SQIND_WAVE_SGPRS_OFFSET, size, dst); 3003 } 3004 3005 static void gfx_v6_0_select_me_pipe_q(struct amdgpu_device *adev, 3006 u32 me, u32 pipe, u32 q, u32 vm) 3007 { 3008 DRM_INFO("Not implemented\n"); 3009 } 3010 3011 static const struct amdgpu_gfx_funcs gfx_v6_0_gfx_funcs = { 3012 .get_gpu_clock_counter = &gfx_v6_0_get_gpu_clock_counter, 3013 .select_se_sh = &gfx_v6_0_select_se_sh, 3014 .read_wave_data = &gfx_v6_0_read_wave_data, 3015 .read_wave_sgprs = &gfx_v6_0_read_wave_sgprs, 3016 .select_me_pipe_q = &gfx_v6_0_select_me_pipe_q 3017 }; 3018 3019 static const struct amdgpu_rlc_funcs gfx_v6_0_rlc_funcs = { 3020 .init = gfx_v6_0_rlc_init, 3021 .resume = gfx_v6_0_rlc_resume, 3022 .stop = gfx_v6_0_rlc_stop, 3023 .reset = gfx_v6_0_rlc_reset, 3024 .start = gfx_v6_0_rlc_start 3025 }; 3026 3027 static int gfx_v6_0_early_init(void *handle) 3028 { 3029 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3030 3031 adev->gfx.num_gfx_rings = GFX6_NUM_GFX_RINGS; 3032 adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev), 3033 GFX6_NUM_COMPUTE_RINGS); 3034 adev->gfx.funcs = &gfx_v6_0_gfx_funcs; 3035 adev->gfx.rlc.funcs = &gfx_v6_0_rlc_funcs; 3036 gfx_v6_0_set_ring_funcs(adev); 3037 gfx_v6_0_set_irq_funcs(adev); 3038 3039 return 0; 3040 } 3041 3042 static int gfx_v6_0_sw_init(void *handle) 3043 { 3044 struct amdgpu_ring *ring; 3045 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3046 int i, r; 3047 3048 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 181, &adev->gfx.eop_irq); 3049 if (r) 3050 return r; 3051 3052 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 184, &adev->gfx.priv_reg_irq); 3053 if (r) 3054 return r; 3055 3056 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 185, &adev->gfx.priv_inst_irq); 3057 if (r) 3058 return r; 3059 3060 r = gfx_v6_0_init_microcode(adev); 3061 if (r) { 3062 DRM_ERROR("Failed to load gfx firmware!\n"); 3063 return r; 3064 } 3065 3066 r = adev->gfx.rlc.funcs->init(adev); 3067 if (r) { 3068 DRM_ERROR("Failed to init rlc BOs!\n"); 3069 return r; 3070 } 3071 3072 for (i = 0; i < adev->gfx.num_gfx_rings; i++) { 3073 ring = &adev->gfx.gfx_ring[i]; 3074 ring->ring_obj = NULL; 3075 sprintf(ring->name, "gfx"); 3076 r = amdgpu_ring_init(adev, ring, 1024, 3077 &adev->gfx.eop_irq, 3078 AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP, 3079 AMDGPU_RING_PRIO_DEFAULT, NULL); 3080 if (r) 3081 return r; 3082 } 3083 3084 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 3085 unsigned irq_type; 3086 3087 if ((i >= 32) || (i >= AMDGPU_MAX_COMPUTE_RINGS)) { 3088 DRM_ERROR("Too many (%d) compute rings!\n", i); 3089 break; 3090 } 3091 ring = &adev->gfx.compute_ring[i]; 3092 ring->ring_obj = NULL; 3093 ring->use_doorbell = false; 3094 ring->doorbell_index = 0; 3095 ring->me = 1; 3096 ring->pipe = i; 3097 ring->queue = i; 3098 sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue); 3099 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP + ring->pipe; 3100 r = amdgpu_ring_init(adev, ring, 1024, 3101 &adev->gfx.eop_irq, irq_type, 3102 AMDGPU_RING_PRIO_DEFAULT, NULL); 3103 if (r) 3104 return r; 3105 } 3106 3107 return r; 3108 } 3109 3110 static int gfx_v6_0_sw_fini(void *handle) 3111 { 3112 int i; 3113 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3114 3115 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 3116 amdgpu_ring_fini(&adev->gfx.gfx_ring[i]); 3117 for (i = 0; i < adev->gfx.num_compute_rings; i++) 3118 amdgpu_ring_fini(&adev->gfx.compute_ring[i]); 3119 3120 amdgpu_gfx_rlc_fini(adev); 3121 3122 return 0; 3123 } 3124 3125 static int gfx_v6_0_hw_init(void *handle) 3126 { 3127 int r; 3128 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3129 3130 gfx_v6_0_constants_init(adev); 3131 3132 r = adev->gfx.rlc.funcs->resume(adev); 3133 if (r) 3134 return r; 3135 3136 r = gfx_v6_0_cp_resume(adev); 3137 if (r) 3138 return r; 3139 3140 adev->gfx.ce_ram_size = 0x8000; 3141 3142 return r; 3143 } 3144 3145 static int gfx_v6_0_hw_fini(void *handle) 3146 { 3147 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3148 3149 gfx_v6_0_cp_enable(adev, false); 3150 adev->gfx.rlc.funcs->stop(adev); 3151 gfx_v6_0_fini_pg(adev); 3152 3153 return 0; 3154 } 3155 3156 static int gfx_v6_0_suspend(void *handle) 3157 { 3158 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3159 3160 return gfx_v6_0_hw_fini(adev); 3161 } 3162 3163 static int gfx_v6_0_resume(void *handle) 3164 { 3165 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3166 3167 return gfx_v6_0_hw_init(adev); 3168 } 3169 3170 static bool gfx_v6_0_is_idle(void *handle) 3171 { 3172 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3173 3174 if (RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK) 3175 return false; 3176 else 3177 return true; 3178 } 3179 3180 static int gfx_v6_0_wait_for_idle(void *handle) 3181 { 3182 unsigned i; 3183 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3184 3185 for (i = 0; i < adev->usec_timeout; i++) { 3186 if (gfx_v6_0_is_idle(handle)) 3187 return 0; 3188 udelay(1); 3189 } 3190 return -ETIMEDOUT; 3191 } 3192 3193 static int gfx_v6_0_soft_reset(void *handle) 3194 { 3195 return 0; 3196 } 3197 3198 static void gfx_v6_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev, 3199 enum amdgpu_interrupt_state state) 3200 { 3201 u32 cp_int_cntl; 3202 3203 switch (state) { 3204 case AMDGPU_IRQ_STATE_DISABLE: 3205 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3206 cp_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK; 3207 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3208 break; 3209 case AMDGPU_IRQ_STATE_ENABLE: 3210 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3211 cp_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK; 3212 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3213 break; 3214 default: 3215 break; 3216 } 3217 } 3218 3219 static void gfx_v6_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev, 3220 int ring, 3221 enum amdgpu_interrupt_state state) 3222 { 3223 u32 cp_int_cntl; 3224 switch (state){ 3225 case AMDGPU_IRQ_STATE_DISABLE: 3226 if (ring == 0) { 3227 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING1); 3228 cp_int_cntl &= ~CP_INT_CNTL_RING1__TIME_STAMP_INT_ENABLE_MASK; 3229 WREG32(mmCP_INT_CNTL_RING1, cp_int_cntl); 3230 break; 3231 } else { 3232 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING2); 3233 cp_int_cntl &= ~CP_INT_CNTL_RING2__TIME_STAMP_INT_ENABLE_MASK; 3234 WREG32(mmCP_INT_CNTL_RING2, cp_int_cntl); 3235 break; 3236 3237 } 3238 case AMDGPU_IRQ_STATE_ENABLE: 3239 if (ring == 0) { 3240 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING1); 3241 cp_int_cntl |= CP_INT_CNTL_RING1__TIME_STAMP_INT_ENABLE_MASK; 3242 WREG32(mmCP_INT_CNTL_RING1, cp_int_cntl); 3243 break; 3244 } else { 3245 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING2); 3246 cp_int_cntl |= CP_INT_CNTL_RING2__TIME_STAMP_INT_ENABLE_MASK; 3247 WREG32(mmCP_INT_CNTL_RING2, cp_int_cntl); 3248 break; 3249 3250 } 3251 3252 default: 3253 BUG(); 3254 break; 3255 3256 } 3257 } 3258 3259 static int gfx_v6_0_set_priv_reg_fault_state(struct amdgpu_device *adev, 3260 struct amdgpu_irq_src *src, 3261 unsigned type, 3262 enum amdgpu_interrupt_state state) 3263 { 3264 u32 cp_int_cntl; 3265 3266 switch (state) { 3267 case AMDGPU_IRQ_STATE_DISABLE: 3268 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3269 cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK; 3270 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3271 break; 3272 case AMDGPU_IRQ_STATE_ENABLE: 3273 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3274 cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK; 3275 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3276 break; 3277 default: 3278 break; 3279 } 3280 3281 return 0; 3282 } 3283 3284 static int gfx_v6_0_set_priv_inst_fault_state(struct amdgpu_device *adev, 3285 struct amdgpu_irq_src *src, 3286 unsigned type, 3287 enum amdgpu_interrupt_state state) 3288 { 3289 u32 cp_int_cntl; 3290 3291 switch (state) { 3292 case AMDGPU_IRQ_STATE_DISABLE: 3293 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3294 cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK; 3295 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3296 break; 3297 case AMDGPU_IRQ_STATE_ENABLE: 3298 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 3299 cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK; 3300 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 3301 break; 3302 default: 3303 break; 3304 } 3305 3306 return 0; 3307 } 3308 3309 static int gfx_v6_0_set_eop_interrupt_state(struct amdgpu_device *adev, 3310 struct amdgpu_irq_src *src, 3311 unsigned type, 3312 enum amdgpu_interrupt_state state) 3313 { 3314 switch (type) { 3315 case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP: 3316 gfx_v6_0_set_gfx_eop_interrupt_state(adev, state); 3317 break; 3318 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP: 3319 gfx_v6_0_set_compute_eop_interrupt_state(adev, 0, state); 3320 break; 3321 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP: 3322 gfx_v6_0_set_compute_eop_interrupt_state(adev, 1, state); 3323 break; 3324 default: 3325 break; 3326 } 3327 return 0; 3328 } 3329 3330 static int gfx_v6_0_eop_irq(struct amdgpu_device *adev, 3331 struct amdgpu_irq_src *source, 3332 struct amdgpu_iv_entry *entry) 3333 { 3334 switch (entry->ring_id) { 3335 case 0: 3336 amdgpu_fence_process(&adev->gfx.gfx_ring[0]); 3337 break; 3338 case 1: 3339 case 2: 3340 amdgpu_fence_process(&adev->gfx.compute_ring[entry->ring_id - 1]); 3341 break; 3342 default: 3343 break; 3344 } 3345 return 0; 3346 } 3347 3348 static void gfx_v6_0_fault(struct amdgpu_device *adev, 3349 struct amdgpu_iv_entry *entry) 3350 { 3351 struct amdgpu_ring *ring; 3352 3353 switch (entry->ring_id) { 3354 case 0: 3355 ring = &adev->gfx.gfx_ring[0]; 3356 break; 3357 case 1: 3358 case 2: 3359 ring = &adev->gfx.compute_ring[entry->ring_id - 1]; 3360 break; 3361 default: 3362 return; 3363 } 3364 drm_sched_fault(&ring->sched); 3365 } 3366 3367 static int gfx_v6_0_priv_reg_irq(struct amdgpu_device *adev, 3368 struct amdgpu_irq_src *source, 3369 struct amdgpu_iv_entry *entry) 3370 { 3371 DRM_ERROR("Illegal register access in command stream\n"); 3372 gfx_v6_0_fault(adev, entry); 3373 return 0; 3374 } 3375 3376 static int gfx_v6_0_priv_inst_irq(struct amdgpu_device *adev, 3377 struct amdgpu_irq_src *source, 3378 struct amdgpu_iv_entry *entry) 3379 { 3380 DRM_ERROR("Illegal instruction in command stream\n"); 3381 gfx_v6_0_fault(adev, entry); 3382 return 0; 3383 } 3384 3385 static int gfx_v6_0_set_clockgating_state(void *handle, 3386 enum amd_clockgating_state state) 3387 { 3388 bool gate = false; 3389 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3390 3391 if (state == AMD_CG_STATE_GATE) 3392 gate = true; 3393 3394 gfx_v6_0_enable_gui_idle_interrupt(adev, false); 3395 if (gate) { 3396 gfx_v6_0_enable_mgcg(adev, true); 3397 gfx_v6_0_enable_cgcg(adev, true); 3398 } else { 3399 gfx_v6_0_enable_cgcg(adev, false); 3400 gfx_v6_0_enable_mgcg(adev, false); 3401 } 3402 gfx_v6_0_enable_gui_idle_interrupt(adev, true); 3403 3404 return 0; 3405 } 3406 3407 static int gfx_v6_0_set_powergating_state(void *handle, 3408 enum amd_powergating_state state) 3409 { 3410 bool gate = false; 3411 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 3412 3413 if (state == AMD_PG_STATE_GATE) 3414 gate = true; 3415 3416 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | 3417 AMD_PG_SUPPORT_GFX_SMG | 3418 AMD_PG_SUPPORT_GFX_DMG | 3419 AMD_PG_SUPPORT_CP | 3420 AMD_PG_SUPPORT_GDS | 3421 AMD_PG_SUPPORT_RLC_SMU_HS)) { 3422 gfx_v6_0_update_gfx_pg(adev, gate); 3423 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { 3424 gfx_v6_0_enable_cp_pg(adev, gate); 3425 gfx_v6_0_enable_gds_pg(adev, gate); 3426 } 3427 } 3428 3429 return 0; 3430 } 3431 3432 static void gfx_v6_0_emit_mem_sync(struct amdgpu_ring *ring) 3433 { 3434 amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3)); 3435 amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA | 3436 PACKET3_TC_ACTION_ENA | 3437 PACKET3_SH_KCACHE_ACTION_ENA | 3438 PACKET3_SH_ICACHE_ACTION_ENA); /* CP_COHER_CNTL */ 3439 amdgpu_ring_write(ring, 0xffffffff); /* CP_COHER_SIZE */ 3440 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */ 3441 amdgpu_ring_write(ring, 0x0000000A); /* poll interval */ 3442 } 3443 3444 static const struct amd_ip_funcs gfx_v6_0_ip_funcs = { 3445 .name = "gfx_v6_0", 3446 .early_init = gfx_v6_0_early_init, 3447 .late_init = NULL, 3448 .sw_init = gfx_v6_0_sw_init, 3449 .sw_fini = gfx_v6_0_sw_fini, 3450 .hw_init = gfx_v6_0_hw_init, 3451 .hw_fini = gfx_v6_0_hw_fini, 3452 .suspend = gfx_v6_0_suspend, 3453 .resume = gfx_v6_0_resume, 3454 .is_idle = gfx_v6_0_is_idle, 3455 .wait_for_idle = gfx_v6_0_wait_for_idle, 3456 .soft_reset = gfx_v6_0_soft_reset, 3457 .set_clockgating_state = gfx_v6_0_set_clockgating_state, 3458 .set_powergating_state = gfx_v6_0_set_powergating_state, 3459 }; 3460 3461 static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_gfx = { 3462 .type = AMDGPU_RING_TYPE_GFX, 3463 .align_mask = 0xff, 3464 .nop = 0x80000000, 3465 .support_64bit_ptrs = false, 3466 .get_rptr = gfx_v6_0_ring_get_rptr, 3467 .get_wptr = gfx_v6_0_ring_get_wptr, 3468 .set_wptr = gfx_v6_0_ring_set_wptr_gfx, 3469 .emit_frame_size = 3470 5 + 5 + /* hdp flush / invalidate */ 3471 14 + 14 + 14 + /* gfx_v6_0_ring_emit_fence x3 for user fence, vm fence */ 3472 7 + 4 + /* gfx_v6_0_ring_emit_pipeline_sync */ 3473 SI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v6_0_ring_emit_vm_flush */ 3474 3 + 2 + /* gfx_v6_ring_emit_cntxcntl including vgt flush */ 3475 5, /* SURFACE_SYNC */ 3476 .emit_ib_size = 6, /* gfx_v6_0_ring_emit_ib */ 3477 .emit_ib = gfx_v6_0_ring_emit_ib, 3478 .emit_fence = gfx_v6_0_ring_emit_fence, 3479 .emit_pipeline_sync = gfx_v6_0_ring_emit_pipeline_sync, 3480 .emit_vm_flush = gfx_v6_0_ring_emit_vm_flush, 3481 .test_ring = gfx_v6_0_ring_test_ring, 3482 .test_ib = gfx_v6_0_ring_test_ib, 3483 .insert_nop = amdgpu_ring_insert_nop, 3484 .emit_cntxcntl = gfx_v6_ring_emit_cntxcntl, 3485 .emit_wreg = gfx_v6_0_ring_emit_wreg, 3486 .emit_mem_sync = gfx_v6_0_emit_mem_sync, 3487 }; 3488 3489 static const struct amdgpu_ring_funcs gfx_v6_0_ring_funcs_compute = { 3490 .type = AMDGPU_RING_TYPE_COMPUTE, 3491 .align_mask = 0xff, 3492 .nop = 0x80000000, 3493 .get_rptr = gfx_v6_0_ring_get_rptr, 3494 .get_wptr = gfx_v6_0_ring_get_wptr, 3495 .set_wptr = gfx_v6_0_ring_set_wptr_compute, 3496 .emit_frame_size = 3497 5 + 5 + /* hdp flush / invalidate */ 3498 7 + /* gfx_v6_0_ring_emit_pipeline_sync */ 3499 SI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v6_0_ring_emit_vm_flush */ 3500 14 + 14 + 14 + /* gfx_v6_0_ring_emit_fence x3 for user fence, vm fence */ 3501 5, /* SURFACE_SYNC */ 3502 .emit_ib_size = 6, /* gfx_v6_0_ring_emit_ib */ 3503 .emit_ib = gfx_v6_0_ring_emit_ib, 3504 .emit_fence = gfx_v6_0_ring_emit_fence, 3505 .emit_pipeline_sync = gfx_v6_0_ring_emit_pipeline_sync, 3506 .emit_vm_flush = gfx_v6_0_ring_emit_vm_flush, 3507 .test_ring = gfx_v6_0_ring_test_ring, 3508 .test_ib = gfx_v6_0_ring_test_ib, 3509 .insert_nop = amdgpu_ring_insert_nop, 3510 .emit_wreg = gfx_v6_0_ring_emit_wreg, 3511 .emit_mem_sync = gfx_v6_0_emit_mem_sync, 3512 }; 3513 3514 static void gfx_v6_0_set_ring_funcs(struct amdgpu_device *adev) 3515 { 3516 int i; 3517 3518 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 3519 adev->gfx.gfx_ring[i].funcs = &gfx_v6_0_ring_funcs_gfx; 3520 for (i = 0; i < adev->gfx.num_compute_rings; i++) 3521 adev->gfx.compute_ring[i].funcs = &gfx_v6_0_ring_funcs_compute; 3522 } 3523 3524 static const struct amdgpu_irq_src_funcs gfx_v6_0_eop_irq_funcs = { 3525 .set = gfx_v6_0_set_eop_interrupt_state, 3526 .process = gfx_v6_0_eop_irq, 3527 }; 3528 3529 static const struct amdgpu_irq_src_funcs gfx_v6_0_priv_reg_irq_funcs = { 3530 .set = gfx_v6_0_set_priv_reg_fault_state, 3531 .process = gfx_v6_0_priv_reg_irq, 3532 }; 3533 3534 static const struct amdgpu_irq_src_funcs gfx_v6_0_priv_inst_irq_funcs = { 3535 .set = gfx_v6_0_set_priv_inst_fault_state, 3536 .process = gfx_v6_0_priv_inst_irq, 3537 }; 3538 3539 static void gfx_v6_0_set_irq_funcs(struct amdgpu_device *adev) 3540 { 3541 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST; 3542 adev->gfx.eop_irq.funcs = &gfx_v6_0_eop_irq_funcs; 3543 3544 adev->gfx.priv_reg_irq.num_types = 1; 3545 adev->gfx.priv_reg_irq.funcs = &gfx_v6_0_priv_reg_irq_funcs; 3546 3547 adev->gfx.priv_inst_irq.num_types = 1; 3548 adev->gfx.priv_inst_irq.funcs = &gfx_v6_0_priv_inst_irq_funcs; 3549 } 3550 3551 static void gfx_v6_0_get_cu_info(struct amdgpu_device *adev) 3552 { 3553 int i, j, k, counter, active_cu_number = 0; 3554 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0; 3555 struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info; 3556 unsigned disable_masks[4 * 2]; 3557 u32 ao_cu_num; 3558 3559 if (adev->flags & AMD_IS_APU) 3560 ao_cu_num = 2; 3561 else 3562 ao_cu_num = adev->gfx.config.max_cu_per_sh; 3563 3564 memset(cu_info, 0, sizeof(*cu_info)); 3565 3566 amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2); 3567 3568 mutex_lock(&adev->grbm_idx_mutex); 3569 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 3570 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 3571 mask = 1; 3572 ao_bitmap = 0; 3573 counter = 0; 3574 gfx_v6_0_select_se_sh(adev, i, j, 0xffffffff); 3575 if (i < 4 && j < 2) 3576 gfx_v6_0_set_user_cu_inactive_bitmap( 3577 adev, disable_masks[i * 2 + j]); 3578 bitmap = gfx_v6_0_get_cu_enabled(adev); 3579 cu_info->bitmap[i][j] = bitmap; 3580 3581 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k++) { 3582 if (bitmap & mask) { 3583 if (counter < ao_cu_num) 3584 ao_bitmap |= mask; 3585 counter ++; 3586 } 3587 mask <<= 1; 3588 } 3589 active_cu_number += counter; 3590 if (i < 2 && j < 2) 3591 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8)); 3592 cu_info->ao_cu_bitmap[i][j] = ao_bitmap; 3593 } 3594 } 3595 3596 gfx_v6_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff); 3597 mutex_unlock(&adev->grbm_idx_mutex); 3598 3599 cu_info->number = active_cu_number; 3600 cu_info->ao_cu_mask = ao_cu_mask; 3601 } 3602 3603 const struct amdgpu_ip_block_version gfx_v6_0_ip_block = 3604 { 3605 .type = AMD_IP_BLOCK_TYPE_GFX, 3606 .major = 6, 3607 .minor = 0, 3608 .rev = 0, 3609 .funcs = &gfx_v6_0_ip_funcs, 3610 }; 3611