1 /* 2 * Copyright 2014 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 */ 23 24 #include <linux/firmware.h> 25 #include <linux/module.h> 26 27 #include "amdgpu.h" 28 #include "amdgpu_ih.h" 29 #include "amdgpu_gfx.h" 30 #include "cikd.h" 31 #include "cik.h" 32 #include "cik_structs.h" 33 #include "atom.h" 34 #include "amdgpu_ucode.h" 35 #include "clearstate_ci.h" 36 37 #include "dce/dce_8_0_d.h" 38 #include "dce/dce_8_0_sh_mask.h" 39 40 #include "bif/bif_4_1_d.h" 41 #include "bif/bif_4_1_sh_mask.h" 42 43 #include "gca/gfx_7_0_d.h" 44 #include "gca/gfx_7_2_enum.h" 45 #include "gca/gfx_7_2_sh_mask.h" 46 47 #include "gmc/gmc_7_0_d.h" 48 #include "gmc/gmc_7_0_sh_mask.h" 49 50 #include "oss/oss_2_0_d.h" 51 #include "oss/oss_2_0_sh_mask.h" 52 53 #define NUM_SIMD_PER_CU 0x4 /* missing from the gfx_7 IP headers */ 54 55 #define GFX7_NUM_GFX_RINGS 1 56 #define GFX7_MEC_HPD_SIZE 2048 57 58 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev); 59 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev); 60 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev); 61 62 MODULE_FIRMWARE("amdgpu/bonaire_pfp.bin"); 63 MODULE_FIRMWARE("amdgpu/bonaire_me.bin"); 64 MODULE_FIRMWARE("amdgpu/bonaire_ce.bin"); 65 MODULE_FIRMWARE("amdgpu/bonaire_rlc.bin"); 66 MODULE_FIRMWARE("amdgpu/bonaire_mec.bin"); 67 68 MODULE_FIRMWARE("amdgpu/hawaii_pfp.bin"); 69 MODULE_FIRMWARE("amdgpu/hawaii_me.bin"); 70 MODULE_FIRMWARE("amdgpu/hawaii_ce.bin"); 71 MODULE_FIRMWARE("amdgpu/hawaii_rlc.bin"); 72 MODULE_FIRMWARE("amdgpu/hawaii_mec.bin"); 73 74 MODULE_FIRMWARE("amdgpu/kaveri_pfp.bin"); 75 MODULE_FIRMWARE("amdgpu/kaveri_me.bin"); 76 MODULE_FIRMWARE("amdgpu/kaveri_ce.bin"); 77 MODULE_FIRMWARE("amdgpu/kaveri_rlc.bin"); 78 MODULE_FIRMWARE("amdgpu/kaveri_mec.bin"); 79 MODULE_FIRMWARE("amdgpu/kaveri_mec2.bin"); 80 81 MODULE_FIRMWARE("amdgpu/kabini_pfp.bin"); 82 MODULE_FIRMWARE("amdgpu/kabini_me.bin"); 83 MODULE_FIRMWARE("amdgpu/kabini_ce.bin"); 84 MODULE_FIRMWARE("amdgpu/kabini_rlc.bin"); 85 MODULE_FIRMWARE("amdgpu/kabini_mec.bin"); 86 87 MODULE_FIRMWARE("amdgpu/mullins_pfp.bin"); 88 MODULE_FIRMWARE("amdgpu/mullins_me.bin"); 89 MODULE_FIRMWARE("amdgpu/mullins_ce.bin"); 90 MODULE_FIRMWARE("amdgpu/mullins_rlc.bin"); 91 MODULE_FIRMWARE("amdgpu/mullins_mec.bin"); 92 93 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] = { 94 {mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0}, 95 {mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1}, 96 {mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2}, 97 {mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3}, 98 {mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4}, 99 {mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5}, 100 {mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6}, 101 {mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7}, 102 {mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8}, 103 {mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9}, 104 {mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10}, 105 {mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11}, 106 {mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12}, 107 {mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13}, 108 {mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14}, 109 {mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15} 110 }; 111 112 static const u32 spectre_rlc_save_restore_register_list[] = { 113 (0x0e00 << 16) | (0xc12c >> 2), 114 0x00000000, 115 (0x0e00 << 16) | (0xc140 >> 2), 116 0x00000000, 117 (0x0e00 << 16) | (0xc150 >> 2), 118 0x00000000, 119 (0x0e00 << 16) | (0xc15c >> 2), 120 0x00000000, 121 (0x0e00 << 16) | (0xc168 >> 2), 122 0x00000000, 123 (0x0e00 << 16) | (0xc170 >> 2), 124 0x00000000, 125 (0x0e00 << 16) | (0xc178 >> 2), 126 0x00000000, 127 (0x0e00 << 16) | (0xc204 >> 2), 128 0x00000000, 129 (0x0e00 << 16) | (0xc2b4 >> 2), 130 0x00000000, 131 (0x0e00 << 16) | (0xc2b8 >> 2), 132 0x00000000, 133 (0x0e00 << 16) | (0xc2bc >> 2), 134 0x00000000, 135 (0x0e00 << 16) | (0xc2c0 >> 2), 136 0x00000000, 137 (0x0e00 << 16) | (0x8228 >> 2), 138 0x00000000, 139 (0x0e00 << 16) | (0x829c >> 2), 140 0x00000000, 141 (0x0e00 << 16) | (0x869c >> 2), 142 0x00000000, 143 (0x0600 << 16) | (0x98f4 >> 2), 144 0x00000000, 145 (0x0e00 << 16) | (0x98f8 >> 2), 146 0x00000000, 147 (0x0e00 << 16) | (0x9900 >> 2), 148 0x00000000, 149 (0x0e00 << 16) | (0xc260 >> 2), 150 0x00000000, 151 (0x0e00 << 16) | (0x90e8 >> 2), 152 0x00000000, 153 (0x0e00 << 16) | (0x3c000 >> 2), 154 0x00000000, 155 (0x0e00 << 16) | (0x3c00c >> 2), 156 0x00000000, 157 (0x0e00 << 16) | (0x8c1c >> 2), 158 0x00000000, 159 (0x0e00 << 16) | (0x9700 >> 2), 160 0x00000000, 161 (0x0e00 << 16) | (0xcd20 >> 2), 162 0x00000000, 163 (0x4e00 << 16) | (0xcd20 >> 2), 164 0x00000000, 165 (0x5e00 << 16) | (0xcd20 >> 2), 166 0x00000000, 167 (0x6e00 << 16) | (0xcd20 >> 2), 168 0x00000000, 169 (0x7e00 << 16) | (0xcd20 >> 2), 170 0x00000000, 171 (0x8e00 << 16) | (0xcd20 >> 2), 172 0x00000000, 173 (0x9e00 << 16) | (0xcd20 >> 2), 174 0x00000000, 175 (0xae00 << 16) | (0xcd20 >> 2), 176 0x00000000, 177 (0xbe00 << 16) | (0xcd20 >> 2), 178 0x00000000, 179 (0x0e00 << 16) | (0x89bc >> 2), 180 0x00000000, 181 (0x0e00 << 16) | (0x8900 >> 2), 182 0x00000000, 183 0x3, 184 (0x0e00 << 16) | (0xc130 >> 2), 185 0x00000000, 186 (0x0e00 << 16) | (0xc134 >> 2), 187 0x00000000, 188 (0x0e00 << 16) | (0xc1fc >> 2), 189 0x00000000, 190 (0x0e00 << 16) | (0xc208 >> 2), 191 0x00000000, 192 (0x0e00 << 16) | (0xc264 >> 2), 193 0x00000000, 194 (0x0e00 << 16) | (0xc268 >> 2), 195 0x00000000, 196 (0x0e00 << 16) | (0xc26c >> 2), 197 0x00000000, 198 (0x0e00 << 16) | (0xc270 >> 2), 199 0x00000000, 200 (0x0e00 << 16) | (0xc274 >> 2), 201 0x00000000, 202 (0x0e00 << 16) | (0xc278 >> 2), 203 0x00000000, 204 (0x0e00 << 16) | (0xc27c >> 2), 205 0x00000000, 206 (0x0e00 << 16) | (0xc280 >> 2), 207 0x00000000, 208 (0x0e00 << 16) | (0xc284 >> 2), 209 0x00000000, 210 (0x0e00 << 16) | (0xc288 >> 2), 211 0x00000000, 212 (0x0e00 << 16) | (0xc28c >> 2), 213 0x00000000, 214 (0x0e00 << 16) | (0xc290 >> 2), 215 0x00000000, 216 (0x0e00 << 16) | (0xc294 >> 2), 217 0x00000000, 218 (0x0e00 << 16) | (0xc298 >> 2), 219 0x00000000, 220 (0x0e00 << 16) | (0xc29c >> 2), 221 0x00000000, 222 (0x0e00 << 16) | (0xc2a0 >> 2), 223 0x00000000, 224 (0x0e00 << 16) | (0xc2a4 >> 2), 225 0x00000000, 226 (0x0e00 << 16) | (0xc2a8 >> 2), 227 0x00000000, 228 (0x0e00 << 16) | (0xc2ac >> 2), 229 0x00000000, 230 (0x0e00 << 16) | (0xc2b0 >> 2), 231 0x00000000, 232 (0x0e00 << 16) | (0x301d0 >> 2), 233 0x00000000, 234 (0x0e00 << 16) | (0x30238 >> 2), 235 0x00000000, 236 (0x0e00 << 16) | (0x30250 >> 2), 237 0x00000000, 238 (0x0e00 << 16) | (0x30254 >> 2), 239 0x00000000, 240 (0x0e00 << 16) | (0x30258 >> 2), 241 0x00000000, 242 (0x0e00 << 16) | (0x3025c >> 2), 243 0x00000000, 244 (0x4e00 << 16) | (0xc900 >> 2), 245 0x00000000, 246 (0x5e00 << 16) | (0xc900 >> 2), 247 0x00000000, 248 (0x6e00 << 16) | (0xc900 >> 2), 249 0x00000000, 250 (0x7e00 << 16) | (0xc900 >> 2), 251 0x00000000, 252 (0x8e00 << 16) | (0xc900 >> 2), 253 0x00000000, 254 (0x9e00 << 16) | (0xc900 >> 2), 255 0x00000000, 256 (0xae00 << 16) | (0xc900 >> 2), 257 0x00000000, 258 (0xbe00 << 16) | (0xc900 >> 2), 259 0x00000000, 260 (0x4e00 << 16) | (0xc904 >> 2), 261 0x00000000, 262 (0x5e00 << 16) | (0xc904 >> 2), 263 0x00000000, 264 (0x6e00 << 16) | (0xc904 >> 2), 265 0x00000000, 266 (0x7e00 << 16) | (0xc904 >> 2), 267 0x00000000, 268 (0x8e00 << 16) | (0xc904 >> 2), 269 0x00000000, 270 (0x9e00 << 16) | (0xc904 >> 2), 271 0x00000000, 272 (0xae00 << 16) | (0xc904 >> 2), 273 0x00000000, 274 (0xbe00 << 16) | (0xc904 >> 2), 275 0x00000000, 276 (0x4e00 << 16) | (0xc908 >> 2), 277 0x00000000, 278 (0x5e00 << 16) | (0xc908 >> 2), 279 0x00000000, 280 (0x6e00 << 16) | (0xc908 >> 2), 281 0x00000000, 282 (0x7e00 << 16) | (0xc908 >> 2), 283 0x00000000, 284 (0x8e00 << 16) | (0xc908 >> 2), 285 0x00000000, 286 (0x9e00 << 16) | (0xc908 >> 2), 287 0x00000000, 288 (0xae00 << 16) | (0xc908 >> 2), 289 0x00000000, 290 (0xbe00 << 16) | (0xc908 >> 2), 291 0x00000000, 292 (0x4e00 << 16) | (0xc90c >> 2), 293 0x00000000, 294 (0x5e00 << 16) | (0xc90c >> 2), 295 0x00000000, 296 (0x6e00 << 16) | (0xc90c >> 2), 297 0x00000000, 298 (0x7e00 << 16) | (0xc90c >> 2), 299 0x00000000, 300 (0x8e00 << 16) | (0xc90c >> 2), 301 0x00000000, 302 (0x9e00 << 16) | (0xc90c >> 2), 303 0x00000000, 304 (0xae00 << 16) | (0xc90c >> 2), 305 0x00000000, 306 (0xbe00 << 16) | (0xc90c >> 2), 307 0x00000000, 308 (0x4e00 << 16) | (0xc910 >> 2), 309 0x00000000, 310 (0x5e00 << 16) | (0xc910 >> 2), 311 0x00000000, 312 (0x6e00 << 16) | (0xc910 >> 2), 313 0x00000000, 314 (0x7e00 << 16) | (0xc910 >> 2), 315 0x00000000, 316 (0x8e00 << 16) | (0xc910 >> 2), 317 0x00000000, 318 (0x9e00 << 16) | (0xc910 >> 2), 319 0x00000000, 320 (0xae00 << 16) | (0xc910 >> 2), 321 0x00000000, 322 (0xbe00 << 16) | (0xc910 >> 2), 323 0x00000000, 324 (0x0e00 << 16) | (0xc99c >> 2), 325 0x00000000, 326 (0x0e00 << 16) | (0x9834 >> 2), 327 0x00000000, 328 (0x0000 << 16) | (0x30f00 >> 2), 329 0x00000000, 330 (0x0001 << 16) | (0x30f00 >> 2), 331 0x00000000, 332 (0x0000 << 16) | (0x30f04 >> 2), 333 0x00000000, 334 (0x0001 << 16) | (0x30f04 >> 2), 335 0x00000000, 336 (0x0000 << 16) | (0x30f08 >> 2), 337 0x00000000, 338 (0x0001 << 16) | (0x30f08 >> 2), 339 0x00000000, 340 (0x0000 << 16) | (0x30f0c >> 2), 341 0x00000000, 342 (0x0001 << 16) | (0x30f0c >> 2), 343 0x00000000, 344 (0x0600 << 16) | (0x9b7c >> 2), 345 0x00000000, 346 (0x0e00 << 16) | (0x8a14 >> 2), 347 0x00000000, 348 (0x0e00 << 16) | (0x8a18 >> 2), 349 0x00000000, 350 (0x0600 << 16) | (0x30a00 >> 2), 351 0x00000000, 352 (0x0e00 << 16) | (0x8bf0 >> 2), 353 0x00000000, 354 (0x0e00 << 16) | (0x8bcc >> 2), 355 0x00000000, 356 (0x0e00 << 16) | (0x8b24 >> 2), 357 0x00000000, 358 (0x0e00 << 16) | (0x30a04 >> 2), 359 0x00000000, 360 (0x0600 << 16) | (0x30a10 >> 2), 361 0x00000000, 362 (0x0600 << 16) | (0x30a14 >> 2), 363 0x00000000, 364 (0x0600 << 16) | (0x30a18 >> 2), 365 0x00000000, 366 (0x0600 << 16) | (0x30a2c >> 2), 367 0x00000000, 368 (0x0e00 << 16) | (0xc700 >> 2), 369 0x00000000, 370 (0x0e00 << 16) | (0xc704 >> 2), 371 0x00000000, 372 (0x0e00 << 16) | (0xc708 >> 2), 373 0x00000000, 374 (0x0e00 << 16) | (0xc768 >> 2), 375 0x00000000, 376 (0x0400 << 16) | (0xc770 >> 2), 377 0x00000000, 378 (0x0400 << 16) | (0xc774 >> 2), 379 0x00000000, 380 (0x0400 << 16) | (0xc778 >> 2), 381 0x00000000, 382 (0x0400 << 16) | (0xc77c >> 2), 383 0x00000000, 384 (0x0400 << 16) | (0xc780 >> 2), 385 0x00000000, 386 (0x0400 << 16) | (0xc784 >> 2), 387 0x00000000, 388 (0x0400 << 16) | (0xc788 >> 2), 389 0x00000000, 390 (0x0400 << 16) | (0xc78c >> 2), 391 0x00000000, 392 (0x0400 << 16) | (0xc798 >> 2), 393 0x00000000, 394 (0x0400 << 16) | (0xc79c >> 2), 395 0x00000000, 396 (0x0400 << 16) | (0xc7a0 >> 2), 397 0x00000000, 398 (0x0400 << 16) | (0xc7a4 >> 2), 399 0x00000000, 400 (0x0400 << 16) | (0xc7a8 >> 2), 401 0x00000000, 402 (0x0400 << 16) | (0xc7ac >> 2), 403 0x00000000, 404 (0x0400 << 16) | (0xc7b0 >> 2), 405 0x00000000, 406 (0x0400 << 16) | (0xc7b4 >> 2), 407 0x00000000, 408 (0x0e00 << 16) | (0x9100 >> 2), 409 0x00000000, 410 (0x0e00 << 16) | (0x3c010 >> 2), 411 0x00000000, 412 (0x0e00 << 16) | (0x92a8 >> 2), 413 0x00000000, 414 (0x0e00 << 16) | (0x92ac >> 2), 415 0x00000000, 416 (0x0e00 << 16) | (0x92b4 >> 2), 417 0x00000000, 418 (0x0e00 << 16) | (0x92b8 >> 2), 419 0x00000000, 420 (0x0e00 << 16) | (0x92bc >> 2), 421 0x00000000, 422 (0x0e00 << 16) | (0x92c0 >> 2), 423 0x00000000, 424 (0x0e00 << 16) | (0x92c4 >> 2), 425 0x00000000, 426 (0x0e00 << 16) | (0x92c8 >> 2), 427 0x00000000, 428 (0x0e00 << 16) | (0x92cc >> 2), 429 0x00000000, 430 (0x0e00 << 16) | (0x92d0 >> 2), 431 0x00000000, 432 (0x0e00 << 16) | (0x8c00 >> 2), 433 0x00000000, 434 (0x0e00 << 16) | (0x8c04 >> 2), 435 0x00000000, 436 (0x0e00 << 16) | (0x8c20 >> 2), 437 0x00000000, 438 (0x0e00 << 16) | (0x8c38 >> 2), 439 0x00000000, 440 (0x0e00 << 16) | (0x8c3c >> 2), 441 0x00000000, 442 (0x0e00 << 16) | (0xae00 >> 2), 443 0x00000000, 444 (0x0e00 << 16) | (0x9604 >> 2), 445 0x00000000, 446 (0x0e00 << 16) | (0xac08 >> 2), 447 0x00000000, 448 (0x0e00 << 16) | (0xac0c >> 2), 449 0x00000000, 450 (0x0e00 << 16) | (0xac10 >> 2), 451 0x00000000, 452 (0x0e00 << 16) | (0xac14 >> 2), 453 0x00000000, 454 (0x0e00 << 16) | (0xac58 >> 2), 455 0x00000000, 456 (0x0e00 << 16) | (0xac68 >> 2), 457 0x00000000, 458 (0x0e00 << 16) | (0xac6c >> 2), 459 0x00000000, 460 (0x0e00 << 16) | (0xac70 >> 2), 461 0x00000000, 462 (0x0e00 << 16) | (0xac74 >> 2), 463 0x00000000, 464 (0x0e00 << 16) | (0xac78 >> 2), 465 0x00000000, 466 (0x0e00 << 16) | (0xac7c >> 2), 467 0x00000000, 468 (0x0e00 << 16) | (0xac80 >> 2), 469 0x00000000, 470 (0x0e00 << 16) | (0xac84 >> 2), 471 0x00000000, 472 (0x0e00 << 16) | (0xac88 >> 2), 473 0x00000000, 474 (0x0e00 << 16) | (0xac8c >> 2), 475 0x00000000, 476 (0x0e00 << 16) | (0x970c >> 2), 477 0x00000000, 478 (0x0e00 << 16) | (0x9714 >> 2), 479 0x00000000, 480 (0x0e00 << 16) | (0x9718 >> 2), 481 0x00000000, 482 (0x0e00 << 16) | (0x971c >> 2), 483 0x00000000, 484 (0x0e00 << 16) | (0x31068 >> 2), 485 0x00000000, 486 (0x4e00 << 16) | (0x31068 >> 2), 487 0x00000000, 488 (0x5e00 << 16) | (0x31068 >> 2), 489 0x00000000, 490 (0x6e00 << 16) | (0x31068 >> 2), 491 0x00000000, 492 (0x7e00 << 16) | (0x31068 >> 2), 493 0x00000000, 494 (0x8e00 << 16) | (0x31068 >> 2), 495 0x00000000, 496 (0x9e00 << 16) | (0x31068 >> 2), 497 0x00000000, 498 (0xae00 << 16) | (0x31068 >> 2), 499 0x00000000, 500 (0xbe00 << 16) | (0x31068 >> 2), 501 0x00000000, 502 (0x0e00 << 16) | (0xcd10 >> 2), 503 0x00000000, 504 (0x0e00 << 16) | (0xcd14 >> 2), 505 0x00000000, 506 (0x0e00 << 16) | (0x88b0 >> 2), 507 0x00000000, 508 (0x0e00 << 16) | (0x88b4 >> 2), 509 0x00000000, 510 (0x0e00 << 16) | (0x88b8 >> 2), 511 0x00000000, 512 (0x0e00 << 16) | (0x88bc >> 2), 513 0x00000000, 514 (0x0400 << 16) | (0x89c0 >> 2), 515 0x00000000, 516 (0x0e00 << 16) | (0x88c4 >> 2), 517 0x00000000, 518 (0x0e00 << 16) | (0x88c8 >> 2), 519 0x00000000, 520 (0x0e00 << 16) | (0x88d0 >> 2), 521 0x00000000, 522 (0x0e00 << 16) | (0x88d4 >> 2), 523 0x00000000, 524 (0x0e00 << 16) | (0x88d8 >> 2), 525 0x00000000, 526 (0x0e00 << 16) | (0x8980 >> 2), 527 0x00000000, 528 (0x0e00 << 16) | (0x30938 >> 2), 529 0x00000000, 530 (0x0e00 << 16) | (0x3093c >> 2), 531 0x00000000, 532 (0x0e00 << 16) | (0x30940 >> 2), 533 0x00000000, 534 (0x0e00 << 16) | (0x89a0 >> 2), 535 0x00000000, 536 (0x0e00 << 16) | (0x30900 >> 2), 537 0x00000000, 538 (0x0e00 << 16) | (0x30904 >> 2), 539 0x00000000, 540 (0x0e00 << 16) | (0x89b4 >> 2), 541 0x00000000, 542 (0x0e00 << 16) | (0x3c210 >> 2), 543 0x00000000, 544 (0x0e00 << 16) | (0x3c214 >> 2), 545 0x00000000, 546 (0x0e00 << 16) | (0x3c218 >> 2), 547 0x00000000, 548 (0x0e00 << 16) | (0x8904 >> 2), 549 0x00000000, 550 0x5, 551 (0x0e00 << 16) | (0x8c28 >> 2), 552 (0x0e00 << 16) | (0x8c2c >> 2), 553 (0x0e00 << 16) | (0x8c30 >> 2), 554 (0x0e00 << 16) | (0x8c34 >> 2), 555 (0x0e00 << 16) | (0x9600 >> 2), 556 }; 557 558 static const u32 kalindi_rlc_save_restore_register_list[] = { 559 (0x0e00 << 16) | (0xc12c >> 2), 560 0x00000000, 561 (0x0e00 << 16) | (0xc140 >> 2), 562 0x00000000, 563 (0x0e00 << 16) | (0xc150 >> 2), 564 0x00000000, 565 (0x0e00 << 16) | (0xc15c >> 2), 566 0x00000000, 567 (0x0e00 << 16) | (0xc168 >> 2), 568 0x00000000, 569 (0x0e00 << 16) | (0xc170 >> 2), 570 0x00000000, 571 (0x0e00 << 16) | (0xc204 >> 2), 572 0x00000000, 573 (0x0e00 << 16) | (0xc2b4 >> 2), 574 0x00000000, 575 (0x0e00 << 16) | (0xc2b8 >> 2), 576 0x00000000, 577 (0x0e00 << 16) | (0xc2bc >> 2), 578 0x00000000, 579 (0x0e00 << 16) | (0xc2c0 >> 2), 580 0x00000000, 581 (0x0e00 << 16) | (0x8228 >> 2), 582 0x00000000, 583 (0x0e00 << 16) | (0x829c >> 2), 584 0x00000000, 585 (0x0e00 << 16) | (0x869c >> 2), 586 0x00000000, 587 (0x0600 << 16) | (0x98f4 >> 2), 588 0x00000000, 589 (0x0e00 << 16) | (0x98f8 >> 2), 590 0x00000000, 591 (0x0e00 << 16) | (0x9900 >> 2), 592 0x00000000, 593 (0x0e00 << 16) | (0xc260 >> 2), 594 0x00000000, 595 (0x0e00 << 16) | (0x90e8 >> 2), 596 0x00000000, 597 (0x0e00 << 16) | (0x3c000 >> 2), 598 0x00000000, 599 (0x0e00 << 16) | (0x3c00c >> 2), 600 0x00000000, 601 (0x0e00 << 16) | (0x8c1c >> 2), 602 0x00000000, 603 (0x0e00 << 16) | (0x9700 >> 2), 604 0x00000000, 605 (0x0e00 << 16) | (0xcd20 >> 2), 606 0x00000000, 607 (0x4e00 << 16) | (0xcd20 >> 2), 608 0x00000000, 609 (0x5e00 << 16) | (0xcd20 >> 2), 610 0x00000000, 611 (0x6e00 << 16) | (0xcd20 >> 2), 612 0x00000000, 613 (0x7e00 << 16) | (0xcd20 >> 2), 614 0x00000000, 615 (0x0e00 << 16) | (0x89bc >> 2), 616 0x00000000, 617 (0x0e00 << 16) | (0x8900 >> 2), 618 0x00000000, 619 0x3, 620 (0x0e00 << 16) | (0xc130 >> 2), 621 0x00000000, 622 (0x0e00 << 16) | (0xc134 >> 2), 623 0x00000000, 624 (0x0e00 << 16) | (0xc1fc >> 2), 625 0x00000000, 626 (0x0e00 << 16) | (0xc208 >> 2), 627 0x00000000, 628 (0x0e00 << 16) | (0xc264 >> 2), 629 0x00000000, 630 (0x0e00 << 16) | (0xc268 >> 2), 631 0x00000000, 632 (0x0e00 << 16) | (0xc26c >> 2), 633 0x00000000, 634 (0x0e00 << 16) | (0xc270 >> 2), 635 0x00000000, 636 (0x0e00 << 16) | (0xc274 >> 2), 637 0x00000000, 638 (0x0e00 << 16) | (0xc28c >> 2), 639 0x00000000, 640 (0x0e00 << 16) | (0xc290 >> 2), 641 0x00000000, 642 (0x0e00 << 16) | (0xc294 >> 2), 643 0x00000000, 644 (0x0e00 << 16) | (0xc298 >> 2), 645 0x00000000, 646 (0x0e00 << 16) | (0xc2a0 >> 2), 647 0x00000000, 648 (0x0e00 << 16) | (0xc2a4 >> 2), 649 0x00000000, 650 (0x0e00 << 16) | (0xc2a8 >> 2), 651 0x00000000, 652 (0x0e00 << 16) | (0xc2ac >> 2), 653 0x00000000, 654 (0x0e00 << 16) | (0x301d0 >> 2), 655 0x00000000, 656 (0x0e00 << 16) | (0x30238 >> 2), 657 0x00000000, 658 (0x0e00 << 16) | (0x30250 >> 2), 659 0x00000000, 660 (0x0e00 << 16) | (0x30254 >> 2), 661 0x00000000, 662 (0x0e00 << 16) | (0x30258 >> 2), 663 0x00000000, 664 (0x0e00 << 16) | (0x3025c >> 2), 665 0x00000000, 666 (0x4e00 << 16) | (0xc900 >> 2), 667 0x00000000, 668 (0x5e00 << 16) | (0xc900 >> 2), 669 0x00000000, 670 (0x6e00 << 16) | (0xc900 >> 2), 671 0x00000000, 672 (0x7e00 << 16) | (0xc900 >> 2), 673 0x00000000, 674 (0x4e00 << 16) | (0xc904 >> 2), 675 0x00000000, 676 (0x5e00 << 16) | (0xc904 >> 2), 677 0x00000000, 678 (0x6e00 << 16) | (0xc904 >> 2), 679 0x00000000, 680 (0x7e00 << 16) | (0xc904 >> 2), 681 0x00000000, 682 (0x4e00 << 16) | (0xc908 >> 2), 683 0x00000000, 684 (0x5e00 << 16) | (0xc908 >> 2), 685 0x00000000, 686 (0x6e00 << 16) | (0xc908 >> 2), 687 0x00000000, 688 (0x7e00 << 16) | (0xc908 >> 2), 689 0x00000000, 690 (0x4e00 << 16) | (0xc90c >> 2), 691 0x00000000, 692 (0x5e00 << 16) | (0xc90c >> 2), 693 0x00000000, 694 (0x6e00 << 16) | (0xc90c >> 2), 695 0x00000000, 696 (0x7e00 << 16) | (0xc90c >> 2), 697 0x00000000, 698 (0x4e00 << 16) | (0xc910 >> 2), 699 0x00000000, 700 (0x5e00 << 16) | (0xc910 >> 2), 701 0x00000000, 702 (0x6e00 << 16) | (0xc910 >> 2), 703 0x00000000, 704 (0x7e00 << 16) | (0xc910 >> 2), 705 0x00000000, 706 (0x0e00 << 16) | (0xc99c >> 2), 707 0x00000000, 708 (0x0e00 << 16) | (0x9834 >> 2), 709 0x00000000, 710 (0x0000 << 16) | (0x30f00 >> 2), 711 0x00000000, 712 (0x0000 << 16) | (0x30f04 >> 2), 713 0x00000000, 714 (0x0000 << 16) | (0x30f08 >> 2), 715 0x00000000, 716 (0x0000 << 16) | (0x30f0c >> 2), 717 0x00000000, 718 (0x0600 << 16) | (0x9b7c >> 2), 719 0x00000000, 720 (0x0e00 << 16) | (0x8a14 >> 2), 721 0x00000000, 722 (0x0e00 << 16) | (0x8a18 >> 2), 723 0x00000000, 724 (0x0600 << 16) | (0x30a00 >> 2), 725 0x00000000, 726 (0x0e00 << 16) | (0x8bf0 >> 2), 727 0x00000000, 728 (0x0e00 << 16) | (0x8bcc >> 2), 729 0x00000000, 730 (0x0e00 << 16) | (0x8b24 >> 2), 731 0x00000000, 732 (0x0e00 << 16) | (0x30a04 >> 2), 733 0x00000000, 734 (0x0600 << 16) | (0x30a10 >> 2), 735 0x00000000, 736 (0x0600 << 16) | (0x30a14 >> 2), 737 0x00000000, 738 (0x0600 << 16) | (0x30a18 >> 2), 739 0x00000000, 740 (0x0600 << 16) | (0x30a2c >> 2), 741 0x00000000, 742 (0x0e00 << 16) | (0xc700 >> 2), 743 0x00000000, 744 (0x0e00 << 16) | (0xc704 >> 2), 745 0x00000000, 746 (0x0e00 << 16) | (0xc708 >> 2), 747 0x00000000, 748 (0x0e00 << 16) | (0xc768 >> 2), 749 0x00000000, 750 (0x0400 << 16) | (0xc770 >> 2), 751 0x00000000, 752 (0x0400 << 16) | (0xc774 >> 2), 753 0x00000000, 754 (0x0400 << 16) | (0xc798 >> 2), 755 0x00000000, 756 (0x0400 << 16) | (0xc79c >> 2), 757 0x00000000, 758 (0x0e00 << 16) | (0x9100 >> 2), 759 0x00000000, 760 (0x0e00 << 16) | (0x3c010 >> 2), 761 0x00000000, 762 (0x0e00 << 16) | (0x8c00 >> 2), 763 0x00000000, 764 (0x0e00 << 16) | (0x8c04 >> 2), 765 0x00000000, 766 (0x0e00 << 16) | (0x8c20 >> 2), 767 0x00000000, 768 (0x0e00 << 16) | (0x8c38 >> 2), 769 0x00000000, 770 (0x0e00 << 16) | (0x8c3c >> 2), 771 0x00000000, 772 (0x0e00 << 16) | (0xae00 >> 2), 773 0x00000000, 774 (0x0e00 << 16) | (0x9604 >> 2), 775 0x00000000, 776 (0x0e00 << 16) | (0xac08 >> 2), 777 0x00000000, 778 (0x0e00 << 16) | (0xac0c >> 2), 779 0x00000000, 780 (0x0e00 << 16) | (0xac10 >> 2), 781 0x00000000, 782 (0x0e00 << 16) | (0xac14 >> 2), 783 0x00000000, 784 (0x0e00 << 16) | (0xac58 >> 2), 785 0x00000000, 786 (0x0e00 << 16) | (0xac68 >> 2), 787 0x00000000, 788 (0x0e00 << 16) | (0xac6c >> 2), 789 0x00000000, 790 (0x0e00 << 16) | (0xac70 >> 2), 791 0x00000000, 792 (0x0e00 << 16) | (0xac74 >> 2), 793 0x00000000, 794 (0x0e00 << 16) | (0xac78 >> 2), 795 0x00000000, 796 (0x0e00 << 16) | (0xac7c >> 2), 797 0x00000000, 798 (0x0e00 << 16) | (0xac80 >> 2), 799 0x00000000, 800 (0x0e00 << 16) | (0xac84 >> 2), 801 0x00000000, 802 (0x0e00 << 16) | (0xac88 >> 2), 803 0x00000000, 804 (0x0e00 << 16) | (0xac8c >> 2), 805 0x00000000, 806 (0x0e00 << 16) | (0x970c >> 2), 807 0x00000000, 808 (0x0e00 << 16) | (0x9714 >> 2), 809 0x00000000, 810 (0x0e00 << 16) | (0x9718 >> 2), 811 0x00000000, 812 (0x0e00 << 16) | (0x971c >> 2), 813 0x00000000, 814 (0x0e00 << 16) | (0x31068 >> 2), 815 0x00000000, 816 (0x4e00 << 16) | (0x31068 >> 2), 817 0x00000000, 818 (0x5e00 << 16) | (0x31068 >> 2), 819 0x00000000, 820 (0x6e00 << 16) | (0x31068 >> 2), 821 0x00000000, 822 (0x7e00 << 16) | (0x31068 >> 2), 823 0x00000000, 824 (0x0e00 << 16) | (0xcd10 >> 2), 825 0x00000000, 826 (0x0e00 << 16) | (0xcd14 >> 2), 827 0x00000000, 828 (0x0e00 << 16) | (0x88b0 >> 2), 829 0x00000000, 830 (0x0e00 << 16) | (0x88b4 >> 2), 831 0x00000000, 832 (0x0e00 << 16) | (0x88b8 >> 2), 833 0x00000000, 834 (0x0e00 << 16) | (0x88bc >> 2), 835 0x00000000, 836 (0x0400 << 16) | (0x89c0 >> 2), 837 0x00000000, 838 (0x0e00 << 16) | (0x88c4 >> 2), 839 0x00000000, 840 (0x0e00 << 16) | (0x88c8 >> 2), 841 0x00000000, 842 (0x0e00 << 16) | (0x88d0 >> 2), 843 0x00000000, 844 (0x0e00 << 16) | (0x88d4 >> 2), 845 0x00000000, 846 (0x0e00 << 16) | (0x88d8 >> 2), 847 0x00000000, 848 (0x0e00 << 16) | (0x8980 >> 2), 849 0x00000000, 850 (0x0e00 << 16) | (0x30938 >> 2), 851 0x00000000, 852 (0x0e00 << 16) | (0x3093c >> 2), 853 0x00000000, 854 (0x0e00 << 16) | (0x30940 >> 2), 855 0x00000000, 856 (0x0e00 << 16) | (0x89a0 >> 2), 857 0x00000000, 858 (0x0e00 << 16) | (0x30900 >> 2), 859 0x00000000, 860 (0x0e00 << 16) | (0x30904 >> 2), 861 0x00000000, 862 (0x0e00 << 16) | (0x89b4 >> 2), 863 0x00000000, 864 (0x0e00 << 16) | (0x3e1fc >> 2), 865 0x00000000, 866 (0x0e00 << 16) | (0x3c210 >> 2), 867 0x00000000, 868 (0x0e00 << 16) | (0x3c214 >> 2), 869 0x00000000, 870 (0x0e00 << 16) | (0x3c218 >> 2), 871 0x00000000, 872 (0x0e00 << 16) | (0x8904 >> 2), 873 0x00000000, 874 0x5, 875 (0x0e00 << 16) | (0x8c28 >> 2), 876 (0x0e00 << 16) | (0x8c2c >> 2), 877 (0x0e00 << 16) | (0x8c30 >> 2), 878 (0x0e00 << 16) | (0x8c34 >> 2), 879 (0x0e00 << 16) | (0x9600 >> 2), 880 }; 881 882 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev); 883 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, volatile u32 *buffer); 884 static void gfx_v7_0_init_pg(struct amdgpu_device *adev); 885 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev); 886 887 static void gfx_v7_0_free_microcode(struct amdgpu_device *adev) 888 { 889 amdgpu_ucode_release(&adev->gfx.pfp_fw); 890 amdgpu_ucode_release(&adev->gfx.me_fw); 891 amdgpu_ucode_release(&adev->gfx.ce_fw); 892 amdgpu_ucode_release(&adev->gfx.mec_fw); 893 amdgpu_ucode_release(&adev->gfx.mec2_fw); 894 amdgpu_ucode_release(&adev->gfx.rlc_fw); 895 } 896 897 /* 898 * Core functions 899 */ 900 /** 901 * gfx_v7_0_init_microcode - load ucode images from disk 902 * 903 * @adev: amdgpu_device pointer 904 * 905 * Use the firmware interface to load the ucode images into 906 * the driver (not loaded into hw). 907 * Returns 0 on success, error on failure. 908 */ 909 static int gfx_v7_0_init_microcode(struct amdgpu_device *adev) 910 { 911 const char *chip_name; 912 int err; 913 914 DRM_DEBUG("\n"); 915 916 switch (adev->asic_type) { 917 case CHIP_BONAIRE: 918 chip_name = "bonaire"; 919 break; 920 case CHIP_HAWAII: 921 chip_name = "hawaii"; 922 break; 923 case CHIP_KAVERI: 924 chip_name = "kaveri"; 925 break; 926 case CHIP_KABINI: 927 chip_name = "kabini"; 928 break; 929 case CHIP_MULLINS: 930 chip_name = "mullins"; 931 break; 932 default: 933 BUG(); 934 } 935 936 err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw, 937 "amdgpu/%s_pfp.bin", chip_name); 938 if (err) 939 goto out; 940 941 err = amdgpu_ucode_request(adev, &adev->gfx.me_fw, 942 "amdgpu/%s_me.bin", chip_name); 943 if (err) 944 goto out; 945 946 err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw, 947 "amdgpu/%s_ce.bin", chip_name); 948 if (err) 949 goto out; 950 951 err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, 952 "amdgpu/%s_mec.bin", chip_name); 953 if (err) 954 goto out; 955 956 if (adev->asic_type == CHIP_KAVERI) { 957 err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw, 958 "amdgpu/%s_mec2.bin", chip_name); 959 if (err) 960 goto out; 961 } 962 963 err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, 964 "amdgpu/%s_rlc.bin", chip_name); 965 out: 966 if (err) { 967 pr_err("gfx7: Failed to load firmware %s gfx firmware\n", chip_name); 968 gfx_v7_0_free_microcode(adev); 969 } 970 return err; 971 } 972 973 /** 974 * gfx_v7_0_tiling_mode_table_init - init the hw tiling table 975 * 976 * @adev: amdgpu_device pointer 977 * 978 * Starting with SI, the tiling setup is done globally in a 979 * set of 32 tiling modes. Rather than selecting each set of 980 * parameters per surface as on older asics, we just select 981 * which index in the tiling table we want to use, and the 982 * surface uses those parameters (CIK). 983 */ 984 static void gfx_v7_0_tiling_mode_table_init(struct amdgpu_device *adev) 985 { 986 const u32 num_tile_mode_states = 987 ARRAY_SIZE(adev->gfx.config.tile_mode_array); 988 const u32 num_secondary_tile_mode_states = 989 ARRAY_SIZE(adev->gfx.config.macrotile_mode_array); 990 u32 reg_offset, split_equal_to_row_size; 991 uint32_t *tile, *macrotile; 992 993 tile = adev->gfx.config.tile_mode_array; 994 macrotile = adev->gfx.config.macrotile_mode_array; 995 996 switch (adev->gfx.config.mem_row_size_in_kb) { 997 case 1: 998 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB; 999 break; 1000 case 2: 1001 default: 1002 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB; 1003 break; 1004 case 4: 1005 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB; 1006 break; 1007 } 1008 1009 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 1010 tile[reg_offset] = 0; 1011 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 1012 macrotile[reg_offset] = 0; 1013 1014 switch (adev->asic_type) { 1015 case CHIP_BONAIRE: 1016 tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1017 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1018 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 1019 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1020 tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1021 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1022 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 1023 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1024 tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1025 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1026 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1027 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1028 tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1029 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1030 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 1031 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1032 tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1033 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1034 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | 1035 TILE_SPLIT(split_equal_to_row_size)); 1036 tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1037 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1038 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1039 tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1040 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1041 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | 1042 TILE_SPLIT(split_equal_to_row_size)); 1043 tile[7] = (TILE_SPLIT(split_equal_to_row_size)); 1044 tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 1045 PIPE_CONFIG(ADDR_SURF_P4_16x16)); 1046 tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1047 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1048 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); 1049 tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1050 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1051 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 1052 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1053 tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1054 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1055 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 1056 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1057 tile[12] = (TILE_SPLIT(split_equal_to_row_size)); 1058 tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1059 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1060 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); 1061 tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1062 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1063 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1064 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1065 tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | 1066 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1067 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1068 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1069 tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1070 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1071 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1072 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1073 tile[17] = (TILE_SPLIT(split_equal_to_row_size)); 1074 tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 1075 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1076 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1077 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1078 tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 1079 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1080 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); 1081 tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 1082 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1083 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1084 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1085 tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | 1086 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1087 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1088 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1089 tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 1090 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1091 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1092 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1093 tile[23] = (TILE_SPLIT(split_equal_to_row_size)); 1094 tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 1095 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1096 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1097 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1098 tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 1099 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1100 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1101 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1102 tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | 1103 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1104 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1105 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1106 tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1107 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1108 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); 1109 tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1110 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1111 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 1112 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1113 tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1114 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1115 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 1116 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1117 tile[30] = (TILE_SPLIT(split_equal_to_row_size)); 1118 1119 macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1120 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1121 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1122 NUM_BANKS(ADDR_SURF_16_BANK)); 1123 macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1124 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1125 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1126 NUM_BANKS(ADDR_SURF_16_BANK)); 1127 macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1128 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1129 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1130 NUM_BANKS(ADDR_SURF_16_BANK)); 1131 macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1132 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1133 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1134 NUM_BANKS(ADDR_SURF_16_BANK)); 1135 macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1136 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1137 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1138 NUM_BANKS(ADDR_SURF_16_BANK)); 1139 macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1140 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1141 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1142 NUM_BANKS(ADDR_SURF_8_BANK)); 1143 macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1144 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1145 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1146 NUM_BANKS(ADDR_SURF_4_BANK)); 1147 macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 1148 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 1149 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1150 NUM_BANKS(ADDR_SURF_16_BANK)); 1151 macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 1152 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1153 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1154 NUM_BANKS(ADDR_SURF_16_BANK)); 1155 macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1156 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1157 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1158 NUM_BANKS(ADDR_SURF_16_BANK)); 1159 macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1160 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1161 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1162 NUM_BANKS(ADDR_SURF_16_BANK)); 1163 macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1164 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1165 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1166 NUM_BANKS(ADDR_SURF_16_BANK)); 1167 macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1168 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1169 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1170 NUM_BANKS(ADDR_SURF_8_BANK)); 1171 macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1172 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1173 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1174 NUM_BANKS(ADDR_SURF_4_BANK)); 1175 1176 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 1177 WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]); 1178 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 1179 if (reg_offset != 7) 1180 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]); 1181 break; 1182 case CHIP_HAWAII: 1183 tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1184 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1185 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 1186 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1187 tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1188 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1189 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 1190 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1191 tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1192 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1193 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1194 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1195 tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1196 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1197 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 1198 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1199 tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1200 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1201 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | 1202 TILE_SPLIT(split_equal_to_row_size)); 1203 tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1204 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1205 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | 1206 TILE_SPLIT(split_equal_to_row_size)); 1207 tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1208 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1209 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | 1210 TILE_SPLIT(split_equal_to_row_size)); 1211 tile[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1212 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1213 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | 1214 TILE_SPLIT(split_equal_to_row_size)); 1215 tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 1216 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16)); 1217 tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1218 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1219 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); 1220 tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1221 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1222 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 1223 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1224 tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1225 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1226 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 1227 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1228 tile[12] = (ARRAY_MODE(ARRAY_PRT_2D_TILED_THIN1) | 1229 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1230 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 1231 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1232 tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1233 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1234 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); 1235 tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1236 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1237 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1238 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1239 tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | 1240 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1241 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1242 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1243 tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1244 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1245 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1246 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1247 tile[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1248 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1249 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1250 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1251 tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 1252 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1253 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1254 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1255 tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 1256 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1257 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING)); 1258 tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 1259 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1260 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 1261 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1262 tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | 1263 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1264 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 1265 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1266 tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 1267 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1268 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 1269 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1270 tile[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 1271 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1272 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 1273 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1274 tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 1275 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1276 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1277 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1278 tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 1279 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1280 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 1281 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1282 tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | 1283 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1284 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 1285 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1286 tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1287 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1288 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); 1289 tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1290 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1291 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 1292 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1293 tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1294 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | 1295 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 1296 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1297 tile[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1298 PIPE_CONFIG(ADDR_SURF_P4_16x16) | 1299 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 1300 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1301 1302 macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1303 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1304 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1305 NUM_BANKS(ADDR_SURF_16_BANK)); 1306 macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1307 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1308 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1309 NUM_BANKS(ADDR_SURF_16_BANK)); 1310 macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1311 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1312 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1313 NUM_BANKS(ADDR_SURF_16_BANK)); 1314 macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1315 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1316 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1317 NUM_BANKS(ADDR_SURF_16_BANK)); 1318 macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1319 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1320 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1321 NUM_BANKS(ADDR_SURF_8_BANK)); 1322 macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1323 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1324 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1325 NUM_BANKS(ADDR_SURF_4_BANK)); 1326 macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1327 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1328 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1329 NUM_BANKS(ADDR_SURF_4_BANK)); 1330 macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1331 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1332 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1333 NUM_BANKS(ADDR_SURF_16_BANK)); 1334 macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1335 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1336 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1337 NUM_BANKS(ADDR_SURF_16_BANK)); 1338 macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1339 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1340 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1341 NUM_BANKS(ADDR_SURF_16_BANK)); 1342 macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1343 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1344 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1345 NUM_BANKS(ADDR_SURF_8_BANK)); 1346 macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1347 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1348 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1349 NUM_BANKS(ADDR_SURF_16_BANK)); 1350 macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1351 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1352 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1353 NUM_BANKS(ADDR_SURF_8_BANK)); 1354 macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1355 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1356 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | 1357 NUM_BANKS(ADDR_SURF_4_BANK)); 1358 1359 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 1360 WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]); 1361 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 1362 if (reg_offset != 7) 1363 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]); 1364 break; 1365 case CHIP_KABINI: 1366 case CHIP_KAVERI: 1367 case CHIP_MULLINS: 1368 default: 1369 tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1370 PIPE_CONFIG(ADDR_SURF_P2) | 1371 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 1372 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1373 tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1374 PIPE_CONFIG(ADDR_SURF_P2) | 1375 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 1376 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1377 tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1378 PIPE_CONFIG(ADDR_SURF_P2) | 1379 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 1380 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1381 tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1382 PIPE_CONFIG(ADDR_SURF_P2) | 1383 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 1384 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1385 tile[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1386 PIPE_CONFIG(ADDR_SURF_P2) | 1387 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | 1388 TILE_SPLIT(split_equal_to_row_size)); 1389 tile[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1390 PIPE_CONFIG(ADDR_SURF_P2) | 1391 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); 1392 tile[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1393 PIPE_CONFIG(ADDR_SURF_P2) | 1394 MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING) | 1395 TILE_SPLIT(split_equal_to_row_size)); 1396 tile[7] = (TILE_SPLIT(split_equal_to_row_size)); 1397 tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 1398 PIPE_CONFIG(ADDR_SURF_P2)); 1399 tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1400 PIPE_CONFIG(ADDR_SURF_P2) | 1401 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING)); 1402 tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1403 PIPE_CONFIG(ADDR_SURF_P2) | 1404 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 1405 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1406 tile[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1407 PIPE_CONFIG(ADDR_SURF_P2) | 1408 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | 1409 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1410 tile[12] = (TILE_SPLIT(split_equal_to_row_size)); 1411 tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1412 PIPE_CONFIG(ADDR_SURF_P2) | 1413 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING)); 1414 tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1415 PIPE_CONFIG(ADDR_SURF_P2) | 1416 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1417 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1418 tile[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | 1419 PIPE_CONFIG(ADDR_SURF_P2) | 1420 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1421 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1422 tile[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1423 PIPE_CONFIG(ADDR_SURF_P2) | 1424 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1425 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1426 tile[17] = (TILE_SPLIT(split_equal_to_row_size)); 1427 tile[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 1428 PIPE_CONFIG(ADDR_SURF_P2) | 1429 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1430 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1431 tile[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | 1432 PIPE_CONFIG(ADDR_SURF_P2) | 1433 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING)); 1434 tile[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 1435 PIPE_CONFIG(ADDR_SURF_P2) | 1436 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 1437 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1438 tile[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | 1439 PIPE_CONFIG(ADDR_SURF_P2) | 1440 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 1441 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1442 tile[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | 1443 PIPE_CONFIG(ADDR_SURF_P2) | 1444 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 1445 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1446 tile[23] = (TILE_SPLIT(split_equal_to_row_size)); 1447 tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | 1448 PIPE_CONFIG(ADDR_SURF_P2) | 1449 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | 1450 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1451 tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | 1452 PIPE_CONFIG(ADDR_SURF_P2) | 1453 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 1454 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1455 tile[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | 1456 PIPE_CONFIG(ADDR_SURF_P2) | 1457 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | 1458 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); 1459 tile[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 1460 PIPE_CONFIG(ADDR_SURF_P2) | 1461 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING)); 1462 tile[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 1463 PIPE_CONFIG(ADDR_SURF_P2) | 1464 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 1465 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); 1466 tile[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | 1467 PIPE_CONFIG(ADDR_SURF_P2) | 1468 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | 1469 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); 1470 tile[30] = (TILE_SPLIT(split_equal_to_row_size)); 1471 1472 macrotile[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1473 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1474 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1475 NUM_BANKS(ADDR_SURF_8_BANK)); 1476 macrotile[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1477 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1478 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1479 NUM_BANKS(ADDR_SURF_8_BANK)); 1480 macrotile[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1481 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1482 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1483 NUM_BANKS(ADDR_SURF_8_BANK)); 1484 macrotile[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1485 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1486 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1487 NUM_BANKS(ADDR_SURF_8_BANK)); 1488 macrotile[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1489 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1490 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1491 NUM_BANKS(ADDR_SURF_8_BANK)); 1492 macrotile[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1493 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1494 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1495 NUM_BANKS(ADDR_SURF_8_BANK)); 1496 macrotile[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1497 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1498 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1499 NUM_BANKS(ADDR_SURF_8_BANK)); 1500 macrotile[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | 1501 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | 1502 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1503 NUM_BANKS(ADDR_SURF_16_BANK)); 1504 macrotile[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) | 1505 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1506 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1507 NUM_BANKS(ADDR_SURF_16_BANK)); 1508 macrotile[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 1509 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 1510 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1511 NUM_BANKS(ADDR_SURF_16_BANK)); 1512 macrotile[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 1513 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1514 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1515 NUM_BANKS(ADDR_SURF_16_BANK)); 1516 macrotile[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1517 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 1518 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1519 NUM_BANKS(ADDR_SURF_16_BANK)); 1520 macrotile[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1521 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1522 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) | 1523 NUM_BANKS(ADDR_SURF_16_BANK)); 1524 macrotile[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 1525 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 1526 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | 1527 NUM_BANKS(ADDR_SURF_8_BANK)); 1528 1529 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 1530 WREG32(mmGB_TILE_MODE0 + reg_offset, tile[reg_offset]); 1531 for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) 1532 if (reg_offset != 7) 1533 WREG32(mmGB_MACROTILE_MODE0 + reg_offset, macrotile[reg_offset]); 1534 break; 1535 } 1536 } 1537 1538 /** 1539 * gfx_v7_0_select_se_sh - select which SE, SH to address 1540 * 1541 * @adev: amdgpu_device pointer 1542 * @se_num: shader engine to address 1543 * @sh_num: sh block to address 1544 * @instance: Certain registers are instanced per SE or SH. 1545 * 0xffffffff means broadcast to all SEs or SHs (CIK). 1546 * @xcc_id: xcc accelerated compute core id 1547 * Select which SE, SH combinations to address. 1548 */ 1549 static void gfx_v7_0_select_se_sh(struct amdgpu_device *adev, 1550 u32 se_num, u32 sh_num, u32 instance, 1551 int xcc_id) 1552 { 1553 u32 data; 1554 1555 if (instance == 0xffffffff) 1556 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1); 1557 else 1558 data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance); 1559 1560 if ((se_num == 0xffffffff) && (sh_num == 0xffffffff)) 1561 data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK | 1562 GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK; 1563 else if (se_num == 0xffffffff) 1564 data |= GRBM_GFX_INDEX__SE_BROADCAST_WRITES_MASK | 1565 (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT); 1566 else if (sh_num == 0xffffffff) 1567 data |= GRBM_GFX_INDEX__SH_BROADCAST_WRITES_MASK | 1568 (se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT); 1569 else 1570 data |= (sh_num << GRBM_GFX_INDEX__SH_INDEX__SHIFT) | 1571 (se_num << GRBM_GFX_INDEX__SE_INDEX__SHIFT); 1572 WREG32(mmGRBM_GFX_INDEX, data); 1573 } 1574 1575 /** 1576 * gfx_v7_0_get_rb_active_bitmap - computes the mask of enabled RBs 1577 * 1578 * @adev: amdgpu_device pointer 1579 * 1580 * Calculates the bitmask of enabled RBs (CIK). 1581 * Returns the enabled RB bitmask. 1582 */ 1583 static u32 gfx_v7_0_get_rb_active_bitmap(struct amdgpu_device *adev) 1584 { 1585 u32 data, mask; 1586 1587 data = RREG32(mmCC_RB_BACKEND_DISABLE); 1588 data |= RREG32(mmGC_USER_RB_BACKEND_DISABLE); 1589 1590 data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK; 1591 data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT; 1592 1593 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se / 1594 adev->gfx.config.max_sh_per_se); 1595 1596 return (~data) & mask; 1597 } 1598 1599 static void 1600 gfx_v7_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1) 1601 { 1602 switch (adev->asic_type) { 1603 case CHIP_BONAIRE: 1604 *rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) | 1605 SE_XSEL(1) | SE_YSEL(1); 1606 *rconf1 |= 0x0; 1607 break; 1608 case CHIP_HAWAII: 1609 *rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) | 1610 RB_XSEL2(1) | PKR_MAP(2) | PKR_XSEL(1) | 1611 PKR_YSEL(1) | SE_MAP(2) | SE_XSEL(2) | 1612 SE_YSEL(3); 1613 *rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) | 1614 SE_PAIR_YSEL(2); 1615 break; 1616 case CHIP_KAVERI: 1617 *rconf |= RB_MAP_PKR0(2); 1618 *rconf1 |= 0x0; 1619 break; 1620 case CHIP_KABINI: 1621 case CHIP_MULLINS: 1622 *rconf |= 0x0; 1623 *rconf1 |= 0x0; 1624 break; 1625 default: 1626 DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type); 1627 break; 1628 } 1629 } 1630 1631 static void 1632 gfx_v7_0_write_harvested_raster_configs(struct amdgpu_device *adev, 1633 u32 raster_config, u32 raster_config_1, 1634 unsigned rb_mask, unsigned num_rb) 1635 { 1636 unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1); 1637 unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1); 1638 unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2); 1639 unsigned rb_per_se = num_rb / num_se; 1640 unsigned se_mask[4]; 1641 unsigned se; 1642 1643 se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask; 1644 se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask; 1645 se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask; 1646 se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask; 1647 1648 WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4)); 1649 WARN_ON(!(sh_per_se == 1 || sh_per_se == 2)); 1650 WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2)); 1651 1652 if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) || 1653 (!se_mask[2] && !se_mask[3]))) { 1654 raster_config_1 &= ~SE_PAIR_MAP_MASK; 1655 1656 if (!se_mask[0] && !se_mask[1]) { 1657 raster_config_1 |= 1658 SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3); 1659 } else { 1660 raster_config_1 |= 1661 SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0); 1662 } 1663 } 1664 1665 for (se = 0; se < num_se; se++) { 1666 unsigned raster_config_se = raster_config; 1667 unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se); 1668 unsigned pkr1_mask = pkr0_mask << rb_per_pkr; 1669 int idx = (se / 2) * 2; 1670 1671 if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) { 1672 raster_config_se &= ~SE_MAP_MASK; 1673 1674 if (!se_mask[idx]) { 1675 raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3); 1676 } else { 1677 raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0); 1678 } 1679 } 1680 1681 pkr0_mask &= rb_mask; 1682 pkr1_mask &= rb_mask; 1683 if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) { 1684 raster_config_se &= ~PKR_MAP_MASK; 1685 1686 if (!pkr0_mask) { 1687 raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3); 1688 } else { 1689 raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0); 1690 } 1691 } 1692 1693 if (rb_per_se >= 2) { 1694 unsigned rb0_mask = 1 << (se * rb_per_se); 1695 unsigned rb1_mask = rb0_mask << 1; 1696 1697 rb0_mask &= rb_mask; 1698 rb1_mask &= rb_mask; 1699 if (!rb0_mask || !rb1_mask) { 1700 raster_config_se &= ~RB_MAP_PKR0_MASK; 1701 1702 if (!rb0_mask) { 1703 raster_config_se |= 1704 RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3); 1705 } else { 1706 raster_config_se |= 1707 RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0); 1708 } 1709 } 1710 1711 if (rb_per_se > 2) { 1712 rb0_mask = 1 << (se * rb_per_se + rb_per_pkr); 1713 rb1_mask = rb0_mask << 1; 1714 rb0_mask &= rb_mask; 1715 rb1_mask &= rb_mask; 1716 if (!rb0_mask || !rb1_mask) { 1717 raster_config_se &= ~RB_MAP_PKR1_MASK; 1718 1719 if (!rb0_mask) { 1720 raster_config_se |= 1721 RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3); 1722 } else { 1723 raster_config_se |= 1724 RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0); 1725 } 1726 } 1727 } 1728 } 1729 1730 /* GRBM_GFX_INDEX has a different offset on CI+ */ 1731 gfx_v7_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff, 0); 1732 WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se); 1733 WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1); 1734 } 1735 1736 /* GRBM_GFX_INDEX has a different offset on CI+ */ 1737 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0); 1738 } 1739 1740 /** 1741 * gfx_v7_0_setup_rb - setup the RBs on the asic 1742 * 1743 * @adev: amdgpu_device pointer 1744 * 1745 * Configures per-SE/SH RB registers (CIK). 1746 */ 1747 static void gfx_v7_0_setup_rb(struct amdgpu_device *adev) 1748 { 1749 int i, j; 1750 u32 data; 1751 u32 raster_config = 0, raster_config_1 = 0; 1752 u32 active_rbs = 0; 1753 u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se / 1754 adev->gfx.config.max_sh_per_se; 1755 unsigned num_rb_pipes; 1756 1757 mutex_lock(&adev->grbm_idx_mutex); 1758 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 1759 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 1760 gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0); 1761 data = gfx_v7_0_get_rb_active_bitmap(adev); 1762 active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) * 1763 rb_bitmap_width_per_sh); 1764 } 1765 } 1766 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0); 1767 1768 adev->gfx.config.backend_enable_mask = active_rbs; 1769 adev->gfx.config.num_rbs = hweight32(active_rbs); 1770 1771 num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se * 1772 adev->gfx.config.max_shader_engines, 16); 1773 1774 gfx_v7_0_raster_config(adev, &raster_config, &raster_config_1); 1775 1776 if (!adev->gfx.config.backend_enable_mask || 1777 adev->gfx.config.num_rbs >= num_rb_pipes) { 1778 WREG32(mmPA_SC_RASTER_CONFIG, raster_config); 1779 WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1); 1780 } else { 1781 gfx_v7_0_write_harvested_raster_configs(adev, raster_config, raster_config_1, 1782 adev->gfx.config.backend_enable_mask, 1783 num_rb_pipes); 1784 } 1785 1786 /* cache the values for userspace */ 1787 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 1788 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 1789 gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0); 1790 adev->gfx.config.rb_config[i][j].rb_backend_disable = 1791 RREG32(mmCC_RB_BACKEND_DISABLE); 1792 adev->gfx.config.rb_config[i][j].user_rb_backend_disable = 1793 RREG32(mmGC_USER_RB_BACKEND_DISABLE); 1794 adev->gfx.config.rb_config[i][j].raster_config = 1795 RREG32(mmPA_SC_RASTER_CONFIG); 1796 adev->gfx.config.rb_config[i][j].raster_config_1 = 1797 RREG32(mmPA_SC_RASTER_CONFIG_1); 1798 } 1799 } 1800 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0); 1801 mutex_unlock(&adev->grbm_idx_mutex); 1802 } 1803 1804 #define DEFAULT_SH_MEM_BASES (0x6000) 1805 /** 1806 * gfx_v7_0_init_compute_vmid - gart enable 1807 * 1808 * @adev: amdgpu_device pointer 1809 * 1810 * Initialize compute vmid sh_mem registers 1811 * 1812 */ 1813 static void gfx_v7_0_init_compute_vmid(struct amdgpu_device *adev) 1814 { 1815 int i; 1816 uint32_t sh_mem_config; 1817 uint32_t sh_mem_bases; 1818 1819 /* 1820 * Configure apertures: 1821 * LDS: 0x60000000'00000000 - 0x60000001'00000000 (4GB) 1822 * Scratch: 0x60000001'00000000 - 0x60000002'00000000 (4GB) 1823 * GPUVM: 0x60010000'00000000 - 0x60020000'00000000 (1TB) 1824 */ 1825 sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16); 1826 sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED << 1827 SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT; 1828 sh_mem_config |= MTYPE_NONCACHED << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT; 1829 mutex_lock(&adev->srbm_mutex); 1830 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) { 1831 cik_srbm_select(adev, 0, 0, 0, i); 1832 /* CP and shaders */ 1833 WREG32(mmSH_MEM_CONFIG, sh_mem_config); 1834 WREG32(mmSH_MEM_APE1_BASE, 1); 1835 WREG32(mmSH_MEM_APE1_LIMIT, 0); 1836 WREG32(mmSH_MEM_BASES, sh_mem_bases); 1837 } 1838 cik_srbm_select(adev, 0, 0, 0, 0); 1839 mutex_unlock(&adev->srbm_mutex); 1840 1841 /* Initialize all compute VMIDs to have no GDS, GWS, or OA 1842 access. These should be enabled by FW for target VMIDs. */ 1843 for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) { 1844 WREG32(amdgpu_gds_reg_offset[i].mem_base, 0); 1845 WREG32(amdgpu_gds_reg_offset[i].mem_size, 0); 1846 WREG32(amdgpu_gds_reg_offset[i].gws, 0); 1847 WREG32(amdgpu_gds_reg_offset[i].oa, 0); 1848 } 1849 } 1850 1851 static void gfx_v7_0_init_gds_vmid(struct amdgpu_device *adev) 1852 { 1853 int vmid; 1854 1855 /* 1856 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA 1857 * access. Compute VMIDs should be enabled by FW for target VMIDs, 1858 * the driver can enable them for graphics. VMID0 should maintain 1859 * access so that HWS firmware can save/restore entries. 1860 */ 1861 for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) { 1862 WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0); 1863 WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0); 1864 WREG32(amdgpu_gds_reg_offset[vmid].gws, 0); 1865 WREG32(amdgpu_gds_reg_offset[vmid].oa, 0); 1866 } 1867 } 1868 1869 static void gfx_v7_0_config_init(struct amdgpu_device *adev) 1870 { 1871 adev->gfx.config.double_offchip_lds_buf = 1; 1872 } 1873 1874 /** 1875 * gfx_v7_0_constants_init - setup the 3D engine 1876 * 1877 * @adev: amdgpu_device pointer 1878 * 1879 * init the gfx constants such as the 3D engine, tiling configuration 1880 * registers, maximum number of quad pipes, render backends... 1881 */ 1882 static void gfx_v7_0_constants_init(struct amdgpu_device *adev) 1883 { 1884 u32 sh_mem_cfg, sh_static_mem_cfg, sh_mem_base; 1885 u32 tmp; 1886 int i; 1887 1888 WREG32(mmGRBM_CNTL, (0xff << GRBM_CNTL__READ_TIMEOUT__SHIFT)); 1889 1890 WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config); 1891 WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config); 1892 WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config); 1893 1894 gfx_v7_0_tiling_mode_table_init(adev); 1895 1896 gfx_v7_0_setup_rb(adev); 1897 gfx_v7_0_get_cu_info(adev); 1898 gfx_v7_0_config_init(adev); 1899 1900 /* set HW defaults for 3D engine */ 1901 WREG32(mmCP_MEQ_THRESHOLDS, 1902 (0x30 << CP_MEQ_THRESHOLDS__MEQ1_START__SHIFT) | 1903 (0x60 << CP_MEQ_THRESHOLDS__MEQ2_START__SHIFT)); 1904 1905 mutex_lock(&adev->grbm_idx_mutex); 1906 /* 1907 * making sure that the following register writes will be broadcasted 1908 * to all the shaders 1909 */ 1910 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0); 1911 1912 /* XXX SH_MEM regs */ 1913 /* where to put LDS, scratch, GPUVM in FSA64 space */ 1914 sh_mem_cfg = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE, 1915 SH_MEM_ALIGNMENT_MODE_UNALIGNED); 1916 sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, DEFAULT_MTYPE, 1917 MTYPE_NC); 1918 sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, APE1_MTYPE, 1919 MTYPE_UC); 1920 sh_mem_cfg = REG_SET_FIELD(sh_mem_cfg, SH_MEM_CONFIG, PRIVATE_ATC, 0); 1921 1922 sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG, 1923 SWIZZLE_ENABLE, 1); 1924 sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG, 1925 ELEMENT_SIZE, 1); 1926 sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG, 1927 INDEX_STRIDE, 3); 1928 WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg); 1929 1930 mutex_lock(&adev->srbm_mutex); 1931 for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) { 1932 if (i == 0) 1933 sh_mem_base = 0; 1934 else 1935 sh_mem_base = adev->gmc.shared_aperture_start >> 48; 1936 cik_srbm_select(adev, 0, 0, 0, i); 1937 /* CP and shaders */ 1938 WREG32(mmSH_MEM_CONFIG, sh_mem_cfg); 1939 WREG32(mmSH_MEM_APE1_BASE, 1); 1940 WREG32(mmSH_MEM_APE1_LIMIT, 0); 1941 WREG32(mmSH_MEM_BASES, sh_mem_base); 1942 } 1943 cik_srbm_select(adev, 0, 0, 0, 0); 1944 mutex_unlock(&adev->srbm_mutex); 1945 1946 gfx_v7_0_init_compute_vmid(adev); 1947 gfx_v7_0_init_gds_vmid(adev); 1948 1949 WREG32(mmSX_DEBUG_1, 0x20); 1950 1951 WREG32(mmTA_CNTL_AUX, 0x00010000); 1952 1953 tmp = RREG32(mmSPI_CONFIG_CNTL); 1954 tmp |= 0x03000000; 1955 WREG32(mmSPI_CONFIG_CNTL, tmp); 1956 1957 WREG32(mmSQ_CONFIG, 1); 1958 1959 WREG32(mmDB_DEBUG, 0); 1960 1961 tmp = RREG32(mmDB_DEBUG2) & ~0xf00fffff; 1962 tmp |= 0x00000400; 1963 WREG32(mmDB_DEBUG2, tmp); 1964 1965 tmp = RREG32(mmDB_DEBUG3) & ~0x0002021c; 1966 tmp |= 0x00020200; 1967 WREG32(mmDB_DEBUG3, tmp); 1968 1969 tmp = RREG32(mmCB_HW_CONTROL) & ~0x00010000; 1970 tmp |= 0x00018208; 1971 WREG32(mmCB_HW_CONTROL, tmp); 1972 1973 WREG32(mmSPI_CONFIG_CNTL_1, (4 << SPI_CONFIG_CNTL_1__VTX_DONE_DELAY__SHIFT)); 1974 1975 WREG32(mmPA_SC_FIFO_SIZE, 1976 ((adev->gfx.config.sc_prim_fifo_size_frontend << PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) | 1977 (adev->gfx.config.sc_prim_fifo_size_backend << PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) | 1978 (adev->gfx.config.sc_hiz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) | 1979 (adev->gfx.config.sc_earlyz_tile_fifo_size << PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT))); 1980 1981 WREG32(mmVGT_NUM_INSTANCES, 1); 1982 1983 WREG32(mmCP_PERFMON_CNTL, 0); 1984 1985 WREG32(mmSQ_CONFIG, 0); 1986 1987 WREG32(mmPA_SC_FORCE_EOV_MAX_CNTS, 1988 ((4095 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_CLK_CNT__SHIFT) | 1989 (255 << PA_SC_FORCE_EOV_MAX_CNTS__FORCE_EOV_MAX_REZ_CNT__SHIFT))); 1990 1991 WREG32(mmVGT_CACHE_INVALIDATION, 1992 (VC_AND_TC << VGT_CACHE_INVALIDATION__CACHE_INVALIDATION__SHIFT) | 1993 (ES_AND_GS_AUTO << VGT_CACHE_INVALIDATION__AUTO_INVLD_EN__SHIFT)); 1994 1995 WREG32(mmVGT_GS_VERTEX_REUSE, 16); 1996 WREG32(mmPA_SC_LINE_STIPPLE_STATE, 0); 1997 1998 WREG32(mmPA_CL_ENHANCE, PA_CL_ENHANCE__CLIP_VTX_REORDER_ENA_MASK | 1999 (3 << PA_CL_ENHANCE__NUM_CLIP_SEQ__SHIFT)); 2000 WREG32(mmPA_SC_ENHANCE, PA_SC_ENHANCE__ENABLE_PA_SC_OUT_OF_ORDER_MASK); 2001 2002 tmp = RREG32(mmSPI_ARB_PRIORITY); 2003 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2); 2004 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2); 2005 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2); 2006 tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2); 2007 WREG32(mmSPI_ARB_PRIORITY, tmp); 2008 2009 mutex_unlock(&adev->grbm_idx_mutex); 2010 2011 udelay(50); 2012 } 2013 2014 /** 2015 * gfx_v7_0_ring_test_ring - basic gfx ring test 2016 * 2017 * @ring: amdgpu_ring structure holding ring information 2018 * 2019 * Allocate a scratch register and write to it using the gfx ring (CIK). 2020 * Provides a basic gfx ring test to verify that the ring is working. 2021 * Used by gfx_v7_0_cp_gfx_resume(); 2022 * Returns 0 on success, error on failure. 2023 */ 2024 static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring) 2025 { 2026 struct amdgpu_device *adev = ring->adev; 2027 uint32_t tmp = 0; 2028 unsigned i; 2029 int r; 2030 2031 WREG32(mmSCRATCH_REG0, 0xCAFEDEAD); 2032 r = amdgpu_ring_alloc(ring, 3); 2033 if (r) 2034 return r; 2035 2036 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1)); 2037 amdgpu_ring_write(ring, mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START); 2038 amdgpu_ring_write(ring, 0xDEADBEEF); 2039 amdgpu_ring_commit(ring); 2040 2041 for (i = 0; i < adev->usec_timeout; i++) { 2042 tmp = RREG32(mmSCRATCH_REG0); 2043 if (tmp == 0xDEADBEEF) 2044 break; 2045 udelay(1); 2046 } 2047 if (i >= adev->usec_timeout) 2048 r = -ETIMEDOUT; 2049 return r; 2050 } 2051 2052 /** 2053 * gfx_v7_0_ring_emit_hdp_flush - emit an hdp flush on the cp 2054 * 2055 * @ring: amdgpu_ring structure holding ring information 2056 * 2057 * Emits an hdp flush on the cp. 2058 */ 2059 static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring) 2060 { 2061 u32 ref_and_mask; 2062 int usepfp = ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE ? 0 : 1; 2063 2064 if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) { 2065 switch (ring->me) { 2066 case 1: 2067 ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe; 2068 break; 2069 case 2: 2070 ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe; 2071 break; 2072 default: 2073 return; 2074 } 2075 } else { 2076 ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK; 2077 } 2078 2079 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 2080 amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */ 2081 WAIT_REG_MEM_FUNCTION(3) | /* == */ 2082 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */ 2083 amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ); 2084 amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE); 2085 amdgpu_ring_write(ring, ref_and_mask); 2086 amdgpu_ring_write(ring, ref_and_mask); 2087 amdgpu_ring_write(ring, 0x20); /* poll interval */ 2088 } 2089 2090 static void gfx_v7_0_ring_emit_vgt_flush(struct amdgpu_ring *ring) 2091 { 2092 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0)); 2093 amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) | 2094 EVENT_INDEX(4)); 2095 2096 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0)); 2097 amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) | 2098 EVENT_INDEX(0)); 2099 } 2100 2101 /** 2102 * gfx_v7_0_ring_emit_fence_gfx - emit a fence on the gfx ring 2103 * 2104 * @ring: amdgpu_ring structure holding ring information 2105 * @addr: address 2106 * @seq: sequence number 2107 * @flags: fence related flags 2108 * 2109 * Emits a fence sequence number on the gfx ring and flushes 2110 * GPU caches. 2111 */ 2112 static void gfx_v7_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr, 2113 u64 seq, unsigned flags) 2114 { 2115 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT; 2116 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT; 2117 /* Workaround for cache flush problems. First send a dummy EOP 2118 * event down the pipe with seq one below. 2119 */ 2120 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4)); 2121 amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN | 2122 EOP_TC_ACTION_EN | 2123 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | 2124 EVENT_INDEX(5))); 2125 amdgpu_ring_write(ring, addr & 0xfffffffc); 2126 amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) | 2127 DATA_SEL(1) | INT_SEL(0)); 2128 amdgpu_ring_write(ring, lower_32_bits(seq - 1)); 2129 amdgpu_ring_write(ring, upper_32_bits(seq - 1)); 2130 2131 /* Then send the real EOP event down the pipe. */ 2132 amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4)); 2133 amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN | 2134 EOP_TC_ACTION_EN | 2135 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | 2136 EVENT_INDEX(5))); 2137 amdgpu_ring_write(ring, addr & 0xfffffffc); 2138 amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) | 2139 DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0)); 2140 amdgpu_ring_write(ring, lower_32_bits(seq)); 2141 amdgpu_ring_write(ring, upper_32_bits(seq)); 2142 } 2143 2144 /** 2145 * gfx_v7_0_ring_emit_fence_compute - emit a fence on the compute ring 2146 * 2147 * @ring: amdgpu_ring structure holding ring information 2148 * @addr: address 2149 * @seq: sequence number 2150 * @flags: fence related flags 2151 * 2152 * Emits a fence sequence number on the compute ring and flushes 2153 * GPU caches. 2154 */ 2155 static void gfx_v7_0_ring_emit_fence_compute(struct amdgpu_ring *ring, 2156 u64 addr, u64 seq, 2157 unsigned flags) 2158 { 2159 bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT; 2160 bool int_sel = flags & AMDGPU_FENCE_FLAG_INT; 2161 2162 /* RELEASE_MEM - flush caches, send int */ 2163 amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5)); 2164 amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN | 2165 EOP_TC_ACTION_EN | 2166 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | 2167 EVENT_INDEX(5))); 2168 amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0)); 2169 amdgpu_ring_write(ring, addr & 0xfffffffc); 2170 amdgpu_ring_write(ring, upper_32_bits(addr)); 2171 amdgpu_ring_write(ring, lower_32_bits(seq)); 2172 amdgpu_ring_write(ring, upper_32_bits(seq)); 2173 } 2174 2175 /* 2176 * IB stuff 2177 */ 2178 /** 2179 * gfx_v7_0_ring_emit_ib_gfx - emit an IB (Indirect Buffer) on the ring 2180 * 2181 * @ring: amdgpu_ring structure holding ring information 2182 * @job: job to retrieve vmid from 2183 * @ib: amdgpu indirect buffer object 2184 * @flags: options (AMDGPU_HAVE_CTX_SWITCH) 2185 * 2186 * Emits an DE (drawing engine) or CE (constant engine) IB 2187 * on the gfx ring. IBs are usually generated by userspace 2188 * acceleration drivers and submitted to the kernel for 2189 * scheduling on the ring. This function schedules the IB 2190 * on the gfx ring for execution by the GPU. 2191 */ 2192 static void gfx_v7_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, 2193 struct amdgpu_job *job, 2194 struct amdgpu_ib *ib, 2195 uint32_t flags) 2196 { 2197 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 2198 u32 header, control = 0; 2199 2200 /* insert SWITCH_BUFFER packet before first IB in the ring frame */ 2201 if (flags & AMDGPU_HAVE_CTX_SWITCH) { 2202 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 2203 amdgpu_ring_write(ring, 0); 2204 } 2205 2206 if (ib->flags & AMDGPU_IB_FLAG_CE) 2207 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2); 2208 else 2209 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); 2210 2211 control |= ib->length_dw | (vmid << 24); 2212 2213 amdgpu_ring_write(ring, header); 2214 amdgpu_ring_write(ring, 2215 #ifdef __BIG_ENDIAN 2216 (2 << 0) | 2217 #endif 2218 (ib->gpu_addr & 0xFFFFFFFC)); 2219 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF); 2220 amdgpu_ring_write(ring, control); 2221 } 2222 2223 static void gfx_v7_0_ring_emit_ib_compute(struct amdgpu_ring *ring, 2224 struct amdgpu_job *job, 2225 struct amdgpu_ib *ib, 2226 uint32_t flags) 2227 { 2228 unsigned vmid = AMDGPU_JOB_GET_VMID(job); 2229 u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24); 2230 2231 /* Currently, there is a high possibility to get wave ID mismatch 2232 * between ME and GDS, leading to a hw deadlock, because ME generates 2233 * different wave IDs than the GDS expects. This situation happens 2234 * randomly when at least 5 compute pipes use GDS ordered append. 2235 * The wave IDs generated by ME are also wrong after suspend/resume. 2236 * Those are probably bugs somewhere else in the kernel driver. 2237 * 2238 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and 2239 * GDS to 0 for this ring (me/pipe). 2240 */ 2241 if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) { 2242 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 2243 amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START); 2244 amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id); 2245 } 2246 2247 amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); 2248 amdgpu_ring_write(ring, 2249 #ifdef __BIG_ENDIAN 2250 (2 << 0) | 2251 #endif 2252 (ib->gpu_addr & 0xFFFFFFFC)); 2253 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF); 2254 amdgpu_ring_write(ring, control); 2255 } 2256 2257 static void gfx_v7_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags) 2258 { 2259 uint32_t dw2 = 0; 2260 2261 dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */ 2262 if (flags & AMDGPU_HAVE_CTX_SWITCH) { 2263 gfx_v7_0_ring_emit_vgt_flush(ring); 2264 /* set load_global_config & load_global_uconfig */ 2265 dw2 |= 0x8001; 2266 /* set load_cs_sh_regs */ 2267 dw2 |= 0x01000000; 2268 /* set load_per_context_state & load_gfx_sh_regs */ 2269 dw2 |= 0x10002; 2270 } 2271 2272 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 2273 amdgpu_ring_write(ring, dw2); 2274 amdgpu_ring_write(ring, 0); 2275 } 2276 2277 /** 2278 * gfx_v7_0_ring_test_ib - basic ring IB test 2279 * 2280 * @ring: amdgpu_ring structure holding ring information 2281 * @timeout: timeout value in jiffies, or MAX_SCHEDULE_TIMEOUT 2282 * 2283 * Allocate an IB and execute it on the gfx ring (CIK). 2284 * Provides a basic gfx ring test to verify that IBs are working. 2285 * Returns 0 on success, error on failure. 2286 */ 2287 static int gfx_v7_0_ring_test_ib(struct amdgpu_ring *ring, long timeout) 2288 { 2289 struct amdgpu_device *adev = ring->adev; 2290 struct amdgpu_ib ib; 2291 struct dma_fence *f = NULL; 2292 uint32_t tmp = 0; 2293 long r; 2294 2295 WREG32(mmSCRATCH_REG0, 0xCAFEDEAD); 2296 memset(&ib, 0, sizeof(ib)); 2297 r = amdgpu_ib_get(adev, NULL, 256, AMDGPU_IB_POOL_DIRECT, &ib); 2298 if (r) 2299 return r; 2300 2301 ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1); 2302 ib.ptr[1] = mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START; 2303 ib.ptr[2] = 0xDEADBEEF; 2304 ib.length_dw = 3; 2305 2306 r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f); 2307 if (r) 2308 goto error; 2309 2310 r = dma_fence_wait_timeout(f, false, timeout); 2311 if (r == 0) { 2312 r = -ETIMEDOUT; 2313 goto error; 2314 } else if (r < 0) { 2315 goto error; 2316 } 2317 tmp = RREG32(mmSCRATCH_REG0); 2318 if (tmp == 0xDEADBEEF) 2319 r = 0; 2320 else 2321 r = -EINVAL; 2322 2323 error: 2324 amdgpu_ib_free(adev, &ib, NULL); 2325 dma_fence_put(f); 2326 return r; 2327 } 2328 2329 /* 2330 * CP. 2331 * On CIK, gfx and compute now have independent command processors. 2332 * 2333 * GFX 2334 * Gfx consists of a single ring and can process both gfx jobs and 2335 * compute jobs. The gfx CP consists of three microengines (ME): 2336 * PFP - Pre-Fetch Parser 2337 * ME - Micro Engine 2338 * CE - Constant Engine 2339 * The PFP and ME make up what is considered the Drawing Engine (DE). 2340 * The CE is an asynchronous engine used for updating buffer desciptors 2341 * used by the DE so that they can be loaded into cache in parallel 2342 * while the DE is processing state update packets. 2343 * 2344 * Compute 2345 * The compute CP consists of two microengines (ME): 2346 * MEC1 - Compute MicroEngine 1 2347 * MEC2 - Compute MicroEngine 2 2348 * Each MEC supports 4 compute pipes and each pipe supports 8 queues. 2349 * The queues are exposed to userspace and are programmed directly 2350 * by the compute runtime. 2351 */ 2352 /** 2353 * gfx_v7_0_cp_gfx_enable - enable/disable the gfx CP MEs 2354 * 2355 * @adev: amdgpu_device pointer 2356 * @enable: enable or disable the MEs 2357 * 2358 * Halts or unhalts the gfx MEs. 2359 */ 2360 static void gfx_v7_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable) 2361 { 2362 if (enable) 2363 WREG32(mmCP_ME_CNTL, 0); 2364 else 2365 WREG32(mmCP_ME_CNTL, (CP_ME_CNTL__ME_HALT_MASK | 2366 CP_ME_CNTL__PFP_HALT_MASK | 2367 CP_ME_CNTL__CE_HALT_MASK)); 2368 udelay(50); 2369 } 2370 2371 /** 2372 * gfx_v7_0_cp_gfx_load_microcode - load the gfx CP ME ucode 2373 * 2374 * @adev: amdgpu_device pointer 2375 * 2376 * Loads the gfx PFP, ME, and CE ucode. 2377 * Returns 0 for success, -EINVAL if the ucode is not available. 2378 */ 2379 static int gfx_v7_0_cp_gfx_load_microcode(struct amdgpu_device *adev) 2380 { 2381 const struct gfx_firmware_header_v1_0 *pfp_hdr; 2382 const struct gfx_firmware_header_v1_0 *ce_hdr; 2383 const struct gfx_firmware_header_v1_0 *me_hdr; 2384 const __le32 *fw_data; 2385 unsigned i, fw_size; 2386 2387 if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw) 2388 return -EINVAL; 2389 2390 pfp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; 2391 ce_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; 2392 me_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; 2393 2394 amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header); 2395 amdgpu_ucode_print_gfx_hdr(&ce_hdr->header); 2396 amdgpu_ucode_print_gfx_hdr(&me_hdr->header); 2397 adev->gfx.pfp_fw_version = le32_to_cpu(pfp_hdr->header.ucode_version); 2398 adev->gfx.ce_fw_version = le32_to_cpu(ce_hdr->header.ucode_version); 2399 adev->gfx.me_fw_version = le32_to_cpu(me_hdr->header.ucode_version); 2400 adev->gfx.me_feature_version = le32_to_cpu(me_hdr->ucode_feature_version); 2401 adev->gfx.ce_feature_version = le32_to_cpu(ce_hdr->ucode_feature_version); 2402 adev->gfx.pfp_feature_version = le32_to_cpu(pfp_hdr->ucode_feature_version); 2403 2404 gfx_v7_0_cp_gfx_enable(adev, false); 2405 2406 /* PFP */ 2407 fw_data = (const __le32 *) 2408 (adev->gfx.pfp_fw->data + 2409 le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes)); 2410 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4; 2411 WREG32(mmCP_PFP_UCODE_ADDR, 0); 2412 for (i = 0; i < fw_size; i++) 2413 WREG32(mmCP_PFP_UCODE_DATA, le32_to_cpup(fw_data++)); 2414 WREG32(mmCP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version); 2415 2416 /* CE */ 2417 fw_data = (const __le32 *) 2418 (adev->gfx.ce_fw->data + 2419 le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes)); 2420 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4; 2421 WREG32(mmCP_CE_UCODE_ADDR, 0); 2422 for (i = 0; i < fw_size; i++) 2423 WREG32(mmCP_CE_UCODE_DATA, le32_to_cpup(fw_data++)); 2424 WREG32(mmCP_CE_UCODE_ADDR, adev->gfx.ce_fw_version); 2425 2426 /* ME */ 2427 fw_data = (const __le32 *) 2428 (adev->gfx.me_fw->data + 2429 le32_to_cpu(me_hdr->header.ucode_array_offset_bytes)); 2430 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4; 2431 WREG32(mmCP_ME_RAM_WADDR, 0); 2432 for (i = 0; i < fw_size; i++) 2433 WREG32(mmCP_ME_RAM_DATA, le32_to_cpup(fw_data++)); 2434 WREG32(mmCP_ME_RAM_WADDR, adev->gfx.me_fw_version); 2435 2436 return 0; 2437 } 2438 2439 /** 2440 * gfx_v7_0_cp_gfx_start - start the gfx ring 2441 * 2442 * @adev: amdgpu_device pointer 2443 * 2444 * Enables the ring and loads the clear state context and other 2445 * packets required to init the ring. 2446 * Returns 0 for success, error for failure. 2447 */ 2448 static int gfx_v7_0_cp_gfx_start(struct amdgpu_device *adev) 2449 { 2450 struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0]; 2451 const struct cs_section_def *sect = NULL; 2452 const struct cs_extent_def *ext = NULL; 2453 int r, i; 2454 2455 /* init the CP */ 2456 WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1); 2457 WREG32(mmCP_ENDIAN_SWAP, 0); 2458 WREG32(mmCP_DEVICE_ID, 1); 2459 2460 gfx_v7_0_cp_gfx_enable(adev, true); 2461 2462 r = amdgpu_ring_alloc(ring, gfx_v7_0_get_csb_size(adev) + 8); 2463 if (r) { 2464 DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r); 2465 return r; 2466 } 2467 2468 /* init the CE partitions. CE only used for gfx on CIK */ 2469 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2)); 2470 amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE)); 2471 amdgpu_ring_write(ring, 0x8000); 2472 amdgpu_ring_write(ring, 0x8000); 2473 2474 /* clear state buffer */ 2475 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 2476 amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 2477 2478 amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 2479 amdgpu_ring_write(ring, 0x80000000); 2480 amdgpu_ring_write(ring, 0x80000000); 2481 2482 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 2483 for (ext = sect->section; ext->extent != NULL; ++ext) { 2484 if (sect->id == SECT_CONTEXT) { 2485 amdgpu_ring_write(ring, 2486 PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count)); 2487 amdgpu_ring_write(ring, ext->reg_index - PACKET3_SET_CONTEXT_REG_START); 2488 for (i = 0; i < ext->reg_count; i++) 2489 amdgpu_ring_write(ring, ext->extent[i]); 2490 } 2491 } 2492 } 2493 2494 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2)); 2495 amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START); 2496 amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config); 2497 amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1); 2498 2499 amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 2500 amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE); 2501 2502 amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0)); 2503 amdgpu_ring_write(ring, 0); 2504 2505 amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2)); 2506 amdgpu_ring_write(ring, 0x00000316); 2507 amdgpu_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */ 2508 amdgpu_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */ 2509 2510 amdgpu_ring_commit(ring); 2511 2512 return 0; 2513 } 2514 2515 /** 2516 * gfx_v7_0_cp_gfx_resume - setup the gfx ring buffer registers 2517 * 2518 * @adev: amdgpu_device pointer 2519 * 2520 * Program the location and size of the gfx ring buffer 2521 * and test it to make sure it's working. 2522 * Returns 0 for success, error for failure. 2523 */ 2524 static int gfx_v7_0_cp_gfx_resume(struct amdgpu_device *adev) 2525 { 2526 struct amdgpu_ring *ring; 2527 u32 tmp; 2528 u32 rb_bufsz; 2529 u64 rb_addr, rptr_addr; 2530 int r; 2531 2532 WREG32(mmCP_SEM_WAIT_TIMER, 0x0); 2533 if (adev->asic_type != CHIP_HAWAII) 2534 WREG32(mmCP_SEM_INCOMPLETE_TIMER_CNTL, 0x0); 2535 2536 /* Set the write pointer delay */ 2537 WREG32(mmCP_RB_WPTR_DELAY, 0); 2538 2539 /* set the RB to use vmid 0 */ 2540 WREG32(mmCP_RB_VMID, 0); 2541 2542 WREG32(mmSCRATCH_ADDR, 0); 2543 2544 /* ring 0 - compute and gfx */ 2545 /* Set ring buffer size */ 2546 ring = &adev->gfx.gfx_ring[0]; 2547 rb_bufsz = order_base_2(ring->ring_size / 8); 2548 tmp = (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 2549 #ifdef __BIG_ENDIAN 2550 tmp |= 2 << CP_RB0_CNTL__BUF_SWAP__SHIFT; 2551 #endif 2552 WREG32(mmCP_RB0_CNTL, tmp); 2553 2554 /* Initialize the ring buffer's read and write pointers */ 2555 WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK); 2556 ring->wptr = 0; 2557 WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr)); 2558 2559 /* set the wb address wether it's enabled or not */ 2560 rptr_addr = ring->rptr_gpu_addr; 2561 WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr)); 2562 WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF); 2563 2564 /* scratch register shadowing is no longer supported */ 2565 WREG32(mmSCRATCH_UMSK, 0); 2566 2567 mdelay(1); 2568 WREG32(mmCP_RB0_CNTL, tmp); 2569 2570 rb_addr = ring->gpu_addr >> 8; 2571 WREG32(mmCP_RB0_BASE, rb_addr); 2572 WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr)); 2573 2574 /* start the ring */ 2575 gfx_v7_0_cp_gfx_start(adev); 2576 r = amdgpu_ring_test_helper(ring); 2577 if (r) 2578 return r; 2579 2580 return 0; 2581 } 2582 2583 static u64 gfx_v7_0_ring_get_rptr(struct amdgpu_ring *ring) 2584 { 2585 return *ring->rptr_cpu_addr; 2586 } 2587 2588 static u64 gfx_v7_0_ring_get_wptr_gfx(struct amdgpu_ring *ring) 2589 { 2590 struct amdgpu_device *adev = ring->adev; 2591 2592 return RREG32(mmCP_RB0_WPTR); 2593 } 2594 2595 static void gfx_v7_0_ring_set_wptr_gfx(struct amdgpu_ring *ring) 2596 { 2597 struct amdgpu_device *adev = ring->adev; 2598 2599 WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr)); 2600 (void)RREG32(mmCP_RB0_WPTR); 2601 } 2602 2603 static u64 gfx_v7_0_ring_get_wptr_compute(struct amdgpu_ring *ring) 2604 { 2605 /* XXX check if swapping is necessary on BE */ 2606 return *ring->wptr_cpu_addr; 2607 } 2608 2609 static void gfx_v7_0_ring_set_wptr_compute(struct amdgpu_ring *ring) 2610 { 2611 struct amdgpu_device *adev = ring->adev; 2612 2613 /* XXX check if swapping is necessary on BE */ 2614 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr); 2615 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); 2616 } 2617 2618 /** 2619 * gfx_v7_0_cp_compute_enable - enable/disable the compute CP MEs 2620 * 2621 * @adev: amdgpu_device pointer 2622 * @enable: enable or disable the MEs 2623 * 2624 * Halts or unhalts the compute MEs. 2625 */ 2626 static void gfx_v7_0_cp_compute_enable(struct amdgpu_device *adev, bool enable) 2627 { 2628 if (enable) 2629 WREG32(mmCP_MEC_CNTL, 0); 2630 else 2631 WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | 2632 CP_MEC_CNTL__MEC_ME2_HALT_MASK)); 2633 udelay(50); 2634 } 2635 2636 /** 2637 * gfx_v7_0_cp_compute_load_microcode - load the compute CP ME ucode 2638 * 2639 * @adev: amdgpu_device pointer 2640 * 2641 * Loads the compute MEC1&2 ucode. 2642 * Returns 0 for success, -EINVAL if the ucode is not available. 2643 */ 2644 static int gfx_v7_0_cp_compute_load_microcode(struct amdgpu_device *adev) 2645 { 2646 const struct gfx_firmware_header_v1_0 *mec_hdr; 2647 const __le32 *fw_data; 2648 unsigned i, fw_size; 2649 2650 if (!adev->gfx.mec_fw) 2651 return -EINVAL; 2652 2653 mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; 2654 amdgpu_ucode_print_gfx_hdr(&mec_hdr->header); 2655 adev->gfx.mec_fw_version = le32_to_cpu(mec_hdr->header.ucode_version); 2656 adev->gfx.mec_feature_version = le32_to_cpu( 2657 mec_hdr->ucode_feature_version); 2658 2659 gfx_v7_0_cp_compute_enable(adev, false); 2660 2661 /* MEC1 */ 2662 fw_data = (const __le32 *) 2663 (adev->gfx.mec_fw->data + 2664 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes)); 2665 fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes) / 4; 2666 WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0); 2667 for (i = 0; i < fw_size; i++) 2668 WREG32(mmCP_MEC_ME1_UCODE_DATA, le32_to_cpup(fw_data++)); 2669 WREG32(mmCP_MEC_ME1_UCODE_ADDR, 0); 2670 2671 if (adev->asic_type == CHIP_KAVERI) { 2672 const struct gfx_firmware_header_v1_0 *mec2_hdr; 2673 2674 if (!adev->gfx.mec2_fw) 2675 return -EINVAL; 2676 2677 mec2_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; 2678 amdgpu_ucode_print_gfx_hdr(&mec2_hdr->header); 2679 adev->gfx.mec2_fw_version = le32_to_cpu(mec2_hdr->header.ucode_version); 2680 adev->gfx.mec2_feature_version = le32_to_cpu( 2681 mec2_hdr->ucode_feature_version); 2682 2683 /* MEC2 */ 2684 fw_data = (const __le32 *) 2685 (adev->gfx.mec2_fw->data + 2686 le32_to_cpu(mec2_hdr->header.ucode_array_offset_bytes)); 2687 fw_size = le32_to_cpu(mec2_hdr->header.ucode_size_bytes) / 4; 2688 WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0); 2689 for (i = 0; i < fw_size; i++) 2690 WREG32(mmCP_MEC_ME2_UCODE_DATA, le32_to_cpup(fw_data++)); 2691 WREG32(mmCP_MEC_ME2_UCODE_ADDR, 0); 2692 } 2693 2694 return 0; 2695 } 2696 2697 /** 2698 * gfx_v7_0_cp_compute_fini - stop the compute queues 2699 * 2700 * @adev: amdgpu_device pointer 2701 * 2702 * Stop the compute queues and tear down the driver queue 2703 * info. 2704 */ 2705 static void gfx_v7_0_cp_compute_fini(struct amdgpu_device *adev) 2706 { 2707 int i; 2708 2709 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 2710 struct amdgpu_ring *ring = &adev->gfx.compute_ring[i]; 2711 2712 amdgpu_bo_free_kernel(&ring->mqd_obj, NULL, NULL); 2713 } 2714 } 2715 2716 static void gfx_v7_0_mec_fini(struct amdgpu_device *adev) 2717 { 2718 amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL); 2719 } 2720 2721 static int gfx_v7_0_mec_init(struct amdgpu_device *adev) 2722 { 2723 int r; 2724 u32 *hpd; 2725 size_t mec_hpd_size; 2726 2727 bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); 2728 2729 /* take ownership of the relevant compute queues */ 2730 amdgpu_gfx_compute_queue_acquire(adev); 2731 2732 /* allocate space for ALL pipes (even the ones we don't own) */ 2733 mec_hpd_size = adev->gfx.mec.num_mec * adev->gfx.mec.num_pipe_per_mec 2734 * GFX7_MEC_HPD_SIZE * 2; 2735 2736 r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE, 2737 AMDGPU_GEM_DOMAIN_VRAM | 2738 AMDGPU_GEM_DOMAIN_GTT, 2739 &adev->gfx.mec.hpd_eop_obj, 2740 &adev->gfx.mec.hpd_eop_gpu_addr, 2741 (void **)&hpd); 2742 if (r) { 2743 dev_warn(adev->dev, "(%d) create, pin or map of HDP EOP bo failed\n", r); 2744 gfx_v7_0_mec_fini(adev); 2745 return r; 2746 } 2747 2748 /* clear memory. Not sure if this is required or not */ 2749 memset(hpd, 0, mec_hpd_size); 2750 2751 amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj); 2752 amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj); 2753 2754 return 0; 2755 } 2756 2757 static void gfx_v7_0_compute_pipe_init(struct amdgpu_device *adev, 2758 int mec, int pipe) 2759 { 2760 u64 eop_gpu_addr; 2761 u32 tmp; 2762 size_t eop_offset = (mec * adev->gfx.mec.num_pipe_per_mec + pipe) 2763 * GFX7_MEC_HPD_SIZE * 2; 2764 2765 mutex_lock(&adev->srbm_mutex); 2766 eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr + eop_offset; 2767 2768 cik_srbm_select(adev, mec + 1, pipe, 0, 0); 2769 2770 /* write the EOP addr */ 2771 WREG32(mmCP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8); 2772 WREG32(mmCP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8); 2773 2774 /* set the VMID assigned */ 2775 WREG32(mmCP_HPD_EOP_VMID, 0); 2776 2777 /* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */ 2778 tmp = RREG32(mmCP_HPD_EOP_CONTROL); 2779 tmp &= ~CP_HPD_EOP_CONTROL__EOP_SIZE_MASK; 2780 tmp |= order_base_2(GFX7_MEC_HPD_SIZE / 8); 2781 WREG32(mmCP_HPD_EOP_CONTROL, tmp); 2782 2783 cik_srbm_select(adev, 0, 0, 0, 0); 2784 mutex_unlock(&adev->srbm_mutex); 2785 } 2786 2787 static int gfx_v7_0_mqd_deactivate(struct amdgpu_device *adev) 2788 { 2789 int i; 2790 2791 /* disable the queue if it's active */ 2792 if (RREG32(mmCP_HQD_ACTIVE) & 1) { 2793 WREG32(mmCP_HQD_DEQUEUE_REQUEST, 1); 2794 for (i = 0; i < adev->usec_timeout; i++) { 2795 if (!(RREG32(mmCP_HQD_ACTIVE) & 1)) 2796 break; 2797 udelay(1); 2798 } 2799 2800 if (i == adev->usec_timeout) 2801 return -ETIMEDOUT; 2802 2803 WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0); 2804 WREG32(mmCP_HQD_PQ_RPTR, 0); 2805 WREG32(mmCP_HQD_PQ_WPTR, 0); 2806 } 2807 2808 return 0; 2809 } 2810 2811 static void gfx_v7_0_mqd_init(struct amdgpu_device *adev, 2812 struct cik_mqd *mqd, 2813 uint64_t mqd_gpu_addr, 2814 struct amdgpu_ring *ring) 2815 { 2816 u64 hqd_gpu_addr; 2817 u64 wb_gpu_addr; 2818 2819 /* init the mqd struct */ 2820 memset(mqd, 0, sizeof(struct cik_mqd)); 2821 2822 mqd->header = 0xC0310800; 2823 mqd->compute_static_thread_mgmt_se0 = 0xffffffff; 2824 mqd->compute_static_thread_mgmt_se1 = 0xffffffff; 2825 mqd->compute_static_thread_mgmt_se2 = 0xffffffff; 2826 mqd->compute_static_thread_mgmt_se3 = 0xffffffff; 2827 2828 /* enable doorbell? */ 2829 mqd->cp_hqd_pq_doorbell_control = 2830 RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL); 2831 if (ring->use_doorbell) 2832 mqd->cp_hqd_pq_doorbell_control |= CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK; 2833 else 2834 mqd->cp_hqd_pq_doorbell_control &= ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK; 2835 2836 /* set the pointer to the MQD */ 2837 mqd->cp_mqd_base_addr_lo = mqd_gpu_addr & 0xfffffffc; 2838 mqd->cp_mqd_base_addr_hi = upper_32_bits(mqd_gpu_addr); 2839 2840 /* set MQD vmid to 0 */ 2841 mqd->cp_mqd_control = RREG32(mmCP_MQD_CONTROL); 2842 mqd->cp_mqd_control &= ~CP_MQD_CONTROL__VMID_MASK; 2843 2844 /* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */ 2845 hqd_gpu_addr = ring->gpu_addr >> 8; 2846 mqd->cp_hqd_pq_base_lo = hqd_gpu_addr; 2847 mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr); 2848 2849 /* set up the HQD, this is similar to CP_RB0_CNTL */ 2850 mqd->cp_hqd_pq_control = RREG32(mmCP_HQD_PQ_CONTROL); 2851 mqd->cp_hqd_pq_control &= 2852 ~(CP_HQD_PQ_CONTROL__QUEUE_SIZE_MASK | 2853 CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE_MASK); 2854 2855 mqd->cp_hqd_pq_control |= 2856 order_base_2(ring->ring_size / 8); 2857 mqd->cp_hqd_pq_control |= 2858 (order_base_2(AMDGPU_GPU_PAGE_SIZE/8) << 8); 2859 #ifdef __BIG_ENDIAN 2860 mqd->cp_hqd_pq_control |= 2861 2 << CP_HQD_PQ_CONTROL__ENDIAN_SWAP__SHIFT; 2862 #endif 2863 mqd->cp_hqd_pq_control &= 2864 ~(CP_HQD_PQ_CONTROL__UNORD_DISPATCH_MASK | 2865 CP_HQD_PQ_CONTROL__ROQ_PQ_IB_FLIP_MASK | 2866 CP_HQD_PQ_CONTROL__PQ_VOLATILE_MASK); 2867 mqd->cp_hqd_pq_control |= 2868 CP_HQD_PQ_CONTROL__PRIV_STATE_MASK | 2869 CP_HQD_PQ_CONTROL__KMD_QUEUE_MASK; /* assuming kernel queue control */ 2870 2871 /* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */ 2872 wb_gpu_addr = ring->wptr_gpu_addr; 2873 mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc; 2874 mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff; 2875 2876 /* set the wb address wether it's enabled or not */ 2877 wb_gpu_addr = ring->rptr_gpu_addr; 2878 mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc; 2879 mqd->cp_hqd_pq_rptr_report_addr_hi = 2880 upper_32_bits(wb_gpu_addr) & 0xffff; 2881 2882 /* enable the doorbell if requested */ 2883 if (ring->use_doorbell) { 2884 mqd->cp_hqd_pq_doorbell_control = 2885 RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL); 2886 mqd->cp_hqd_pq_doorbell_control &= 2887 ~CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET_MASK; 2888 mqd->cp_hqd_pq_doorbell_control |= 2889 (ring->doorbell_index << 2890 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT); 2891 mqd->cp_hqd_pq_doorbell_control |= 2892 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_EN_MASK; 2893 mqd->cp_hqd_pq_doorbell_control &= 2894 ~(CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_SOURCE_MASK | 2895 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_HIT_MASK); 2896 2897 } else { 2898 mqd->cp_hqd_pq_doorbell_control = 0; 2899 } 2900 2901 /* read and write pointers, similar to CP_RB0_WPTR/_RPTR */ 2902 ring->wptr = 0; 2903 mqd->cp_hqd_pq_wptr = lower_32_bits(ring->wptr); 2904 mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR); 2905 2906 /* set the vmid for the queue */ 2907 mqd->cp_hqd_vmid = 0; 2908 2909 /* defaults */ 2910 mqd->cp_hqd_ib_control = RREG32(mmCP_HQD_IB_CONTROL); 2911 mqd->cp_hqd_ib_base_addr_lo = RREG32(mmCP_HQD_IB_BASE_ADDR); 2912 mqd->cp_hqd_ib_base_addr_hi = RREG32(mmCP_HQD_IB_BASE_ADDR_HI); 2913 mqd->cp_hqd_ib_rptr = RREG32(mmCP_HQD_IB_RPTR); 2914 mqd->cp_hqd_persistent_state = RREG32(mmCP_HQD_PERSISTENT_STATE); 2915 mqd->cp_hqd_sema_cmd = RREG32(mmCP_HQD_SEMA_CMD); 2916 mqd->cp_hqd_msg_type = RREG32(mmCP_HQD_MSG_TYPE); 2917 mqd->cp_hqd_atomic0_preop_lo = RREG32(mmCP_HQD_ATOMIC0_PREOP_LO); 2918 mqd->cp_hqd_atomic0_preop_hi = RREG32(mmCP_HQD_ATOMIC0_PREOP_HI); 2919 mqd->cp_hqd_atomic1_preop_lo = RREG32(mmCP_HQD_ATOMIC1_PREOP_LO); 2920 mqd->cp_hqd_atomic1_preop_hi = RREG32(mmCP_HQD_ATOMIC1_PREOP_HI); 2921 mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR); 2922 mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM); 2923 mqd->cp_hqd_pipe_priority = RREG32(mmCP_HQD_PIPE_PRIORITY); 2924 mqd->cp_hqd_queue_priority = RREG32(mmCP_HQD_QUEUE_PRIORITY); 2925 mqd->cp_hqd_iq_rptr = RREG32(mmCP_HQD_IQ_RPTR); 2926 2927 /* activate the queue */ 2928 mqd->cp_hqd_active = 1; 2929 } 2930 2931 static int gfx_v7_0_mqd_commit(struct amdgpu_device *adev, struct cik_mqd *mqd) 2932 { 2933 uint32_t tmp; 2934 uint32_t mqd_reg; 2935 uint32_t *mqd_data; 2936 2937 /* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_MQD_CONTROL */ 2938 mqd_data = &mqd->cp_mqd_base_addr_lo; 2939 2940 /* disable wptr polling */ 2941 tmp = RREG32(mmCP_PQ_WPTR_POLL_CNTL); 2942 tmp = REG_SET_FIELD(tmp, CP_PQ_WPTR_POLL_CNTL, EN, 0); 2943 WREG32(mmCP_PQ_WPTR_POLL_CNTL, tmp); 2944 2945 /* program all HQD registers */ 2946 for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_MQD_CONTROL; mqd_reg++) 2947 WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]); 2948 2949 /* activate the HQD */ 2950 for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++) 2951 WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]); 2952 2953 return 0; 2954 } 2955 2956 static int gfx_v7_0_compute_queue_init(struct amdgpu_device *adev, int ring_id) 2957 { 2958 int r; 2959 u64 mqd_gpu_addr; 2960 struct cik_mqd *mqd; 2961 struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id]; 2962 2963 r = amdgpu_bo_create_reserved(adev, sizeof(struct cik_mqd), PAGE_SIZE, 2964 AMDGPU_GEM_DOMAIN_GTT, &ring->mqd_obj, 2965 &mqd_gpu_addr, (void **)&mqd); 2966 if (r) { 2967 dev_warn(adev->dev, "(%d) create MQD bo failed\n", r); 2968 return r; 2969 } 2970 2971 mutex_lock(&adev->srbm_mutex); 2972 cik_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0); 2973 2974 gfx_v7_0_mqd_init(adev, mqd, mqd_gpu_addr, ring); 2975 gfx_v7_0_mqd_deactivate(adev); 2976 gfx_v7_0_mqd_commit(adev, mqd); 2977 2978 cik_srbm_select(adev, 0, 0, 0, 0); 2979 mutex_unlock(&adev->srbm_mutex); 2980 2981 amdgpu_bo_kunmap(ring->mqd_obj); 2982 amdgpu_bo_unreserve(ring->mqd_obj); 2983 return 0; 2984 } 2985 2986 /** 2987 * gfx_v7_0_cp_compute_resume - setup the compute queue registers 2988 * 2989 * @adev: amdgpu_device pointer 2990 * 2991 * Program the compute queues and test them to make sure they 2992 * are working. 2993 * Returns 0 for success, error for failure. 2994 */ 2995 static int gfx_v7_0_cp_compute_resume(struct amdgpu_device *adev) 2996 { 2997 int r, i, j; 2998 u32 tmp; 2999 struct amdgpu_ring *ring; 3000 3001 /* fix up chicken bits */ 3002 tmp = RREG32(mmCP_CPF_DEBUG); 3003 tmp |= (1 << 23); 3004 WREG32(mmCP_CPF_DEBUG, tmp); 3005 3006 /* init all pipes (even the ones we don't own) */ 3007 for (i = 0; i < adev->gfx.mec.num_mec; i++) 3008 for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) 3009 gfx_v7_0_compute_pipe_init(adev, i, j); 3010 3011 /* init the queues */ 3012 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 3013 r = gfx_v7_0_compute_queue_init(adev, i); 3014 if (r) { 3015 gfx_v7_0_cp_compute_fini(adev); 3016 return r; 3017 } 3018 } 3019 3020 gfx_v7_0_cp_compute_enable(adev, true); 3021 3022 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 3023 ring = &adev->gfx.compute_ring[i]; 3024 amdgpu_ring_test_helper(ring); 3025 } 3026 3027 return 0; 3028 } 3029 3030 static void gfx_v7_0_cp_enable(struct amdgpu_device *adev, bool enable) 3031 { 3032 gfx_v7_0_cp_gfx_enable(adev, enable); 3033 gfx_v7_0_cp_compute_enable(adev, enable); 3034 } 3035 3036 static int gfx_v7_0_cp_load_microcode(struct amdgpu_device *adev) 3037 { 3038 int r; 3039 3040 r = gfx_v7_0_cp_gfx_load_microcode(adev); 3041 if (r) 3042 return r; 3043 r = gfx_v7_0_cp_compute_load_microcode(adev); 3044 if (r) 3045 return r; 3046 3047 return 0; 3048 } 3049 3050 static void gfx_v7_0_enable_gui_idle_interrupt(struct amdgpu_device *adev, 3051 bool enable) 3052 { 3053 u32 tmp = RREG32(mmCP_INT_CNTL_RING0); 3054 3055 if (enable) 3056 tmp |= (CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK | 3057 CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK); 3058 else 3059 tmp &= ~(CP_INT_CNTL_RING0__CNTX_BUSY_INT_ENABLE_MASK | 3060 CP_INT_CNTL_RING0__CNTX_EMPTY_INT_ENABLE_MASK); 3061 WREG32(mmCP_INT_CNTL_RING0, tmp); 3062 } 3063 3064 static int gfx_v7_0_cp_resume(struct amdgpu_device *adev) 3065 { 3066 int r; 3067 3068 gfx_v7_0_enable_gui_idle_interrupt(adev, false); 3069 3070 r = gfx_v7_0_cp_load_microcode(adev); 3071 if (r) 3072 return r; 3073 3074 r = gfx_v7_0_cp_gfx_resume(adev); 3075 if (r) 3076 return r; 3077 r = gfx_v7_0_cp_compute_resume(adev); 3078 if (r) 3079 return r; 3080 3081 gfx_v7_0_enable_gui_idle_interrupt(adev, true); 3082 3083 return 0; 3084 } 3085 3086 /** 3087 * gfx_v7_0_ring_emit_pipeline_sync - cik vm flush using the CP 3088 * 3089 * @ring: the ring to emit the commands to 3090 * 3091 * Sync the command pipeline with the PFP. E.g. wait for everything 3092 * to be completed. 3093 */ 3094 static void gfx_v7_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring) 3095 { 3096 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 3097 uint32_t seq = ring->fence_drv.sync_seq; 3098 uint64_t addr = ring->fence_drv.gpu_addr; 3099 3100 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 3101 amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */ 3102 WAIT_REG_MEM_FUNCTION(3) | /* equal */ 3103 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */ 3104 amdgpu_ring_write(ring, addr & 0xfffffffc); 3105 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff); 3106 amdgpu_ring_write(ring, seq); 3107 amdgpu_ring_write(ring, 0xffffffff); 3108 amdgpu_ring_write(ring, 4); /* poll interval */ 3109 3110 if (usepfp) { 3111 /* sync CE with ME to prevent CE fetch CEIB before context switch done */ 3112 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 3113 amdgpu_ring_write(ring, 0); 3114 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 3115 amdgpu_ring_write(ring, 0); 3116 } 3117 } 3118 3119 /* 3120 * vm 3121 * VMID 0 is the physical GPU addresses as used by the kernel. 3122 * VMIDs 1-15 are used for userspace clients and are handled 3123 * by the amdgpu vm/hsa code. 3124 */ 3125 /** 3126 * gfx_v7_0_ring_emit_vm_flush - cik vm flush using the CP 3127 * 3128 * @ring: amdgpu_ring pointer 3129 * @vmid: vmid number to use 3130 * @pd_addr: address 3131 * 3132 * Update the page table base and flush the VM TLB 3133 * using the CP (CIK). 3134 */ 3135 static void gfx_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring, 3136 unsigned vmid, uint64_t pd_addr) 3137 { 3138 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 3139 3140 amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr); 3141 3142 /* wait for the invalidate to complete */ 3143 amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 3144 amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */ 3145 WAIT_REG_MEM_FUNCTION(0) | /* always */ 3146 WAIT_REG_MEM_ENGINE(0))); /* me */ 3147 amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST); 3148 amdgpu_ring_write(ring, 0); 3149 amdgpu_ring_write(ring, 0); /* ref */ 3150 amdgpu_ring_write(ring, 0); /* mask */ 3151 amdgpu_ring_write(ring, 0x20); /* poll interval */ 3152 3153 /* compute doesn't have PFP */ 3154 if (usepfp) { 3155 /* sync PFP to ME, otherwise we might get invalid PFP reads */ 3156 amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0)); 3157 amdgpu_ring_write(ring, 0x0); 3158 3159 /* synce CE with ME to prevent CE fetch CEIB before context switch done */ 3160 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 3161 amdgpu_ring_write(ring, 0); 3162 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 3163 amdgpu_ring_write(ring, 0); 3164 } 3165 } 3166 3167 static void gfx_v7_0_ring_emit_wreg(struct amdgpu_ring *ring, 3168 uint32_t reg, uint32_t val) 3169 { 3170 int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX); 3171 3172 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 3173 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(usepfp) | 3174 WRITE_DATA_DST_SEL(0))); 3175 amdgpu_ring_write(ring, reg); 3176 amdgpu_ring_write(ring, 0); 3177 amdgpu_ring_write(ring, val); 3178 } 3179 3180 /* 3181 * RLC 3182 * The RLC is a multi-purpose microengine that handles a 3183 * variety of functions. 3184 */ 3185 static int gfx_v7_0_rlc_init(struct amdgpu_device *adev) 3186 { 3187 const u32 *src_ptr; 3188 u32 dws; 3189 const struct cs_section_def *cs_data; 3190 int r; 3191 3192 /* allocate rlc buffers */ 3193 if (adev->flags & AMD_IS_APU) { 3194 if (adev->asic_type == CHIP_KAVERI) { 3195 adev->gfx.rlc.reg_list = spectre_rlc_save_restore_register_list; 3196 adev->gfx.rlc.reg_list_size = 3197 (u32)ARRAY_SIZE(spectre_rlc_save_restore_register_list); 3198 } else { 3199 adev->gfx.rlc.reg_list = kalindi_rlc_save_restore_register_list; 3200 adev->gfx.rlc.reg_list_size = 3201 (u32)ARRAY_SIZE(kalindi_rlc_save_restore_register_list); 3202 } 3203 } 3204 adev->gfx.rlc.cs_data = ci_cs_data; 3205 adev->gfx.rlc.cp_table_size = ALIGN(CP_ME_TABLE_SIZE * 5 * 4, 2048); /* CP JT */ 3206 adev->gfx.rlc.cp_table_size += 64 * 1024; /* GDS */ 3207 3208 src_ptr = adev->gfx.rlc.reg_list; 3209 dws = adev->gfx.rlc.reg_list_size; 3210 dws += (5 * 16) + 48 + 48 + 64; 3211 3212 cs_data = adev->gfx.rlc.cs_data; 3213 3214 if (src_ptr) { 3215 /* init save restore block */ 3216 r = amdgpu_gfx_rlc_init_sr(adev, dws); 3217 if (r) 3218 return r; 3219 } 3220 3221 if (cs_data) { 3222 /* init clear state block */ 3223 r = amdgpu_gfx_rlc_init_csb(adev); 3224 if (r) 3225 return r; 3226 } 3227 3228 if (adev->gfx.rlc.cp_table_size) { 3229 r = amdgpu_gfx_rlc_init_cpt(adev); 3230 if (r) 3231 return r; 3232 } 3233 3234 /* init spm vmid with 0xf */ 3235 if (adev->gfx.rlc.funcs->update_spm_vmid) 3236 adev->gfx.rlc.funcs->update_spm_vmid(adev, NULL, 0xf); 3237 3238 return 0; 3239 } 3240 3241 static void gfx_v7_0_enable_lbpw(struct amdgpu_device *adev, bool enable) 3242 { 3243 u32 tmp; 3244 3245 tmp = RREG32(mmRLC_LB_CNTL); 3246 if (enable) 3247 tmp |= RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK; 3248 else 3249 tmp &= ~RLC_LB_CNTL__LOAD_BALANCE_ENABLE_MASK; 3250 WREG32(mmRLC_LB_CNTL, tmp); 3251 } 3252 3253 static void gfx_v7_0_wait_for_rlc_serdes(struct amdgpu_device *adev) 3254 { 3255 u32 i, j, k; 3256 u32 mask; 3257 3258 mutex_lock(&adev->grbm_idx_mutex); 3259 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 3260 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 3261 gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0); 3262 for (k = 0; k < adev->usec_timeout; k++) { 3263 if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0) 3264 break; 3265 udelay(1); 3266 } 3267 } 3268 } 3269 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0); 3270 mutex_unlock(&adev->grbm_idx_mutex); 3271 3272 mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK | 3273 RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK | 3274 RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK | 3275 RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK; 3276 for (k = 0; k < adev->usec_timeout; k++) { 3277 if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0) 3278 break; 3279 udelay(1); 3280 } 3281 } 3282 3283 static void gfx_v7_0_update_rlc(struct amdgpu_device *adev, u32 rlc) 3284 { 3285 u32 tmp; 3286 3287 tmp = RREG32(mmRLC_CNTL); 3288 if (tmp != rlc) 3289 WREG32(mmRLC_CNTL, rlc); 3290 } 3291 3292 static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev) 3293 { 3294 u32 data, orig; 3295 3296 orig = data = RREG32(mmRLC_CNTL); 3297 3298 if (data & RLC_CNTL__RLC_ENABLE_F32_MASK) { 3299 u32 i; 3300 3301 data &= ~RLC_CNTL__RLC_ENABLE_F32_MASK; 3302 WREG32(mmRLC_CNTL, data); 3303 3304 for (i = 0; i < adev->usec_timeout; i++) { 3305 if ((RREG32(mmRLC_GPM_STAT) & RLC_GPM_STAT__RLC_BUSY_MASK) == 0) 3306 break; 3307 udelay(1); 3308 } 3309 3310 gfx_v7_0_wait_for_rlc_serdes(adev); 3311 } 3312 3313 return orig; 3314 } 3315 3316 static bool gfx_v7_0_is_rlc_enabled(struct amdgpu_device *adev) 3317 { 3318 return true; 3319 } 3320 3321 static void gfx_v7_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id) 3322 { 3323 u32 tmp, i, mask; 3324 3325 tmp = 0x1 | (1 << 1); 3326 WREG32(mmRLC_GPR_REG2, tmp); 3327 3328 mask = RLC_GPM_STAT__GFX_POWER_STATUS_MASK | 3329 RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK; 3330 for (i = 0; i < adev->usec_timeout; i++) { 3331 if ((RREG32(mmRLC_GPM_STAT) & mask) == mask) 3332 break; 3333 udelay(1); 3334 } 3335 3336 for (i = 0; i < adev->usec_timeout; i++) { 3337 if ((RREG32(mmRLC_GPR_REG2) & 0x1) == 0) 3338 break; 3339 udelay(1); 3340 } 3341 } 3342 3343 static void gfx_v7_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id) 3344 { 3345 u32 tmp; 3346 3347 tmp = 0x1 | (0 << 1); 3348 WREG32(mmRLC_GPR_REG2, tmp); 3349 } 3350 3351 /** 3352 * gfx_v7_0_rlc_stop - stop the RLC ME 3353 * 3354 * @adev: amdgpu_device pointer 3355 * 3356 * Halt the RLC ME (MicroEngine) (CIK). 3357 */ 3358 static void gfx_v7_0_rlc_stop(struct amdgpu_device *adev) 3359 { 3360 WREG32(mmRLC_CNTL, 0); 3361 3362 gfx_v7_0_enable_gui_idle_interrupt(adev, false); 3363 3364 gfx_v7_0_wait_for_rlc_serdes(adev); 3365 } 3366 3367 /** 3368 * gfx_v7_0_rlc_start - start the RLC ME 3369 * 3370 * @adev: amdgpu_device pointer 3371 * 3372 * Unhalt the RLC ME (MicroEngine) (CIK). 3373 */ 3374 static void gfx_v7_0_rlc_start(struct amdgpu_device *adev) 3375 { 3376 WREG32(mmRLC_CNTL, RLC_CNTL__RLC_ENABLE_F32_MASK); 3377 3378 gfx_v7_0_enable_gui_idle_interrupt(adev, true); 3379 3380 udelay(50); 3381 } 3382 3383 static void gfx_v7_0_rlc_reset(struct amdgpu_device *adev) 3384 { 3385 u32 tmp = RREG32(mmGRBM_SOFT_RESET); 3386 3387 tmp |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK; 3388 WREG32(mmGRBM_SOFT_RESET, tmp); 3389 udelay(50); 3390 tmp &= ~GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK; 3391 WREG32(mmGRBM_SOFT_RESET, tmp); 3392 udelay(50); 3393 } 3394 3395 /** 3396 * gfx_v7_0_rlc_resume - setup the RLC hw 3397 * 3398 * @adev: amdgpu_device pointer 3399 * 3400 * Initialize the RLC registers, load the ucode, 3401 * and start the RLC (CIK). 3402 * Returns 0 for success, -EINVAL if the ucode is not available. 3403 */ 3404 static int gfx_v7_0_rlc_resume(struct amdgpu_device *adev) 3405 { 3406 const struct rlc_firmware_header_v1_0 *hdr; 3407 const __le32 *fw_data; 3408 unsigned i, fw_size; 3409 u32 tmp; 3410 3411 if (!adev->gfx.rlc_fw) 3412 return -EINVAL; 3413 3414 hdr = (const struct rlc_firmware_header_v1_0 *)adev->gfx.rlc_fw->data; 3415 amdgpu_ucode_print_rlc_hdr(&hdr->header); 3416 adev->gfx.rlc_fw_version = le32_to_cpu(hdr->header.ucode_version); 3417 adev->gfx.rlc_feature_version = le32_to_cpu( 3418 hdr->ucode_feature_version); 3419 3420 adev->gfx.rlc.funcs->stop(adev); 3421 3422 /* disable CG */ 3423 tmp = RREG32(mmRLC_CGCG_CGLS_CTRL) & 0xfffffffc; 3424 WREG32(mmRLC_CGCG_CGLS_CTRL, tmp); 3425 3426 adev->gfx.rlc.funcs->reset(adev); 3427 3428 gfx_v7_0_init_pg(adev); 3429 3430 WREG32(mmRLC_LB_CNTR_INIT, 0); 3431 WREG32(mmRLC_LB_CNTR_MAX, 0x00008000); 3432 3433 mutex_lock(&adev->grbm_idx_mutex); 3434 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0); 3435 WREG32(mmRLC_LB_INIT_CU_MASK, 0xffffffff); 3436 WREG32(mmRLC_LB_PARAMS, 0x00600408); 3437 WREG32(mmRLC_LB_CNTL, 0x80000004); 3438 mutex_unlock(&adev->grbm_idx_mutex); 3439 3440 WREG32(mmRLC_MC_CNTL, 0); 3441 WREG32(mmRLC_UCODE_CNTL, 0); 3442 3443 fw_data = (const __le32 *) 3444 (adev->gfx.rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 3445 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; 3446 WREG32(mmRLC_GPM_UCODE_ADDR, 0); 3447 for (i = 0; i < fw_size; i++) 3448 WREG32(mmRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++)); 3449 WREG32(mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version); 3450 3451 /* XXX - find out what chips support lbpw */ 3452 gfx_v7_0_enable_lbpw(adev, false); 3453 3454 if (adev->asic_type == CHIP_BONAIRE) 3455 WREG32(mmRLC_DRIVER_CPDMA_STATUS, 0); 3456 3457 adev->gfx.rlc.funcs->start(adev); 3458 3459 return 0; 3460 } 3461 3462 static void gfx_v7_0_update_spm_vmid(struct amdgpu_device *adev, struct amdgpu_ring *ring, unsigned vmid) 3463 { 3464 u32 data; 3465 3466 amdgpu_gfx_off_ctrl(adev, false); 3467 3468 data = RREG32(mmRLC_SPM_VMID); 3469 3470 data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK; 3471 data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT; 3472 3473 WREG32(mmRLC_SPM_VMID, data); 3474 3475 amdgpu_gfx_off_ctrl(adev, true); 3476 } 3477 3478 static void gfx_v7_0_enable_cgcg(struct amdgpu_device *adev, bool enable) 3479 { 3480 u32 data, orig, tmp, tmp2; 3481 3482 orig = data = RREG32(mmRLC_CGCG_CGLS_CTRL); 3483 3484 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { 3485 gfx_v7_0_enable_gui_idle_interrupt(adev, true); 3486 3487 tmp = gfx_v7_0_halt_rlc(adev); 3488 3489 mutex_lock(&adev->grbm_idx_mutex); 3490 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0); 3491 WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff); 3492 WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff); 3493 tmp2 = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | 3494 RLC_SERDES_WR_CTRL__CGCG_OVERRIDE_0_MASK | 3495 RLC_SERDES_WR_CTRL__CGLS_ENABLE_MASK; 3496 WREG32(mmRLC_SERDES_WR_CTRL, tmp2); 3497 mutex_unlock(&adev->grbm_idx_mutex); 3498 3499 gfx_v7_0_update_rlc(adev, tmp); 3500 3501 data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK; 3502 if (orig != data) 3503 WREG32(mmRLC_CGCG_CGLS_CTRL, data); 3504 3505 } else { 3506 gfx_v7_0_enable_gui_idle_interrupt(adev, false); 3507 3508 RREG32(mmCB_CGTT_SCLK_CTRL); 3509 RREG32(mmCB_CGTT_SCLK_CTRL); 3510 RREG32(mmCB_CGTT_SCLK_CTRL); 3511 RREG32(mmCB_CGTT_SCLK_CTRL); 3512 3513 data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK); 3514 if (orig != data) 3515 WREG32(mmRLC_CGCG_CGLS_CTRL, data); 3516 3517 gfx_v7_0_enable_gui_idle_interrupt(adev, true); 3518 } 3519 } 3520 3521 static void gfx_v7_0_enable_mgcg(struct amdgpu_device *adev, bool enable) 3522 { 3523 u32 data, orig, tmp = 0; 3524 3525 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) { 3526 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) { 3527 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) { 3528 orig = data = RREG32(mmCP_MEM_SLP_CNTL); 3529 data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 3530 if (orig != data) 3531 WREG32(mmCP_MEM_SLP_CNTL, data); 3532 } 3533 } 3534 3535 orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 3536 data |= 0x00000001; 3537 data &= 0xfffffffd; 3538 if (orig != data) 3539 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data); 3540 3541 tmp = gfx_v7_0_halt_rlc(adev); 3542 3543 mutex_lock(&adev->grbm_idx_mutex); 3544 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0); 3545 WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff); 3546 WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff); 3547 data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | 3548 RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_0_MASK; 3549 WREG32(mmRLC_SERDES_WR_CTRL, data); 3550 mutex_unlock(&adev->grbm_idx_mutex); 3551 3552 gfx_v7_0_update_rlc(adev, tmp); 3553 3554 if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) { 3555 orig = data = RREG32(mmCGTS_SM_CTRL_REG); 3556 data &= ~CGTS_SM_CTRL_REG__SM_MODE_MASK; 3557 data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT); 3558 data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK; 3559 data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK; 3560 if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) && 3561 (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS)) 3562 data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK; 3563 data &= ~CGTS_SM_CTRL_REG__ON_MONITOR_ADD_MASK; 3564 data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK; 3565 data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT); 3566 if (orig != data) 3567 WREG32(mmCGTS_SM_CTRL_REG, data); 3568 } 3569 } else { 3570 orig = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); 3571 data |= 0x00000003; 3572 if (orig != data) 3573 WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data); 3574 3575 data = RREG32(mmRLC_MEM_SLP_CNTL); 3576 if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) { 3577 data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK; 3578 WREG32(mmRLC_MEM_SLP_CNTL, data); 3579 } 3580 3581 data = RREG32(mmCP_MEM_SLP_CNTL); 3582 if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) { 3583 data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK; 3584 WREG32(mmCP_MEM_SLP_CNTL, data); 3585 } 3586 3587 orig = data = RREG32(mmCGTS_SM_CTRL_REG); 3588 data |= CGTS_SM_CTRL_REG__OVERRIDE_MASK | CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK; 3589 if (orig != data) 3590 WREG32(mmCGTS_SM_CTRL_REG, data); 3591 3592 tmp = gfx_v7_0_halt_rlc(adev); 3593 3594 mutex_lock(&adev->grbm_idx_mutex); 3595 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0); 3596 WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff); 3597 WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff); 3598 data = RLC_SERDES_WR_CTRL__BPM_ADDR_MASK | RLC_SERDES_WR_CTRL__MGCG_OVERRIDE_1_MASK; 3599 WREG32(mmRLC_SERDES_WR_CTRL, data); 3600 mutex_unlock(&adev->grbm_idx_mutex); 3601 3602 gfx_v7_0_update_rlc(adev, tmp); 3603 } 3604 } 3605 3606 static void gfx_v7_0_update_cg(struct amdgpu_device *adev, 3607 bool enable) 3608 { 3609 gfx_v7_0_enable_gui_idle_interrupt(adev, false); 3610 /* order matters! */ 3611 if (enable) { 3612 gfx_v7_0_enable_mgcg(adev, true); 3613 gfx_v7_0_enable_cgcg(adev, true); 3614 } else { 3615 gfx_v7_0_enable_cgcg(adev, false); 3616 gfx_v7_0_enable_mgcg(adev, false); 3617 } 3618 gfx_v7_0_enable_gui_idle_interrupt(adev, true); 3619 } 3620 3621 static void gfx_v7_0_enable_sclk_slowdown_on_pu(struct amdgpu_device *adev, 3622 bool enable) 3623 { 3624 u32 data, orig; 3625 3626 orig = data = RREG32(mmRLC_PG_CNTL); 3627 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS)) 3628 data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK; 3629 else 3630 data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PU_ENABLE_MASK; 3631 if (orig != data) 3632 WREG32(mmRLC_PG_CNTL, data); 3633 } 3634 3635 static void gfx_v7_0_enable_sclk_slowdown_on_pd(struct amdgpu_device *adev, 3636 bool enable) 3637 { 3638 u32 data, orig; 3639 3640 orig = data = RREG32(mmRLC_PG_CNTL); 3641 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS)) 3642 data |= RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK; 3643 else 3644 data &= ~RLC_PG_CNTL__SMU_CLK_SLOWDOWN_ON_PD_ENABLE_MASK; 3645 if (orig != data) 3646 WREG32(mmRLC_PG_CNTL, data); 3647 } 3648 3649 static void gfx_v7_0_enable_cp_pg(struct amdgpu_device *adev, bool enable) 3650 { 3651 u32 data, orig; 3652 3653 orig = data = RREG32(mmRLC_PG_CNTL); 3654 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_CP)) 3655 data &= ~0x8000; 3656 else 3657 data |= 0x8000; 3658 if (orig != data) 3659 WREG32(mmRLC_PG_CNTL, data); 3660 } 3661 3662 static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable) 3663 { 3664 u32 data, orig; 3665 3666 orig = data = RREG32(mmRLC_PG_CNTL); 3667 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GDS)) 3668 data &= ~0x2000; 3669 else 3670 data |= 0x2000; 3671 if (orig != data) 3672 WREG32(mmRLC_PG_CNTL, data); 3673 } 3674 3675 static int gfx_v7_0_cp_pg_table_num(struct amdgpu_device *adev) 3676 { 3677 if (adev->asic_type == CHIP_KAVERI) 3678 return 5; 3679 else 3680 return 4; 3681 } 3682 3683 static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev, 3684 bool enable) 3685 { 3686 u32 data, orig; 3687 3688 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) { 3689 orig = data = RREG32(mmRLC_PG_CNTL); 3690 data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK; 3691 if (orig != data) 3692 WREG32(mmRLC_PG_CNTL, data); 3693 3694 orig = data = RREG32(mmRLC_AUTO_PG_CTRL); 3695 data |= RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK; 3696 if (orig != data) 3697 WREG32(mmRLC_AUTO_PG_CTRL, data); 3698 } else { 3699 orig = data = RREG32(mmRLC_PG_CNTL); 3700 data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK; 3701 if (orig != data) 3702 WREG32(mmRLC_PG_CNTL, data); 3703 3704 orig = data = RREG32(mmRLC_AUTO_PG_CTRL); 3705 data &= ~RLC_AUTO_PG_CTRL__AUTO_PG_EN_MASK; 3706 if (orig != data) 3707 WREG32(mmRLC_AUTO_PG_CTRL, data); 3708 3709 data = RREG32(mmDB_RENDER_CONTROL); 3710 } 3711 } 3712 3713 static void gfx_v7_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev, 3714 u32 bitmap) 3715 { 3716 u32 data; 3717 3718 if (!bitmap) 3719 return; 3720 3721 data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; 3722 data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; 3723 3724 WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data); 3725 } 3726 3727 static u32 gfx_v7_0_get_cu_active_bitmap(struct amdgpu_device *adev) 3728 { 3729 u32 data, mask; 3730 3731 data = RREG32(mmCC_GC_SHADER_ARRAY_CONFIG); 3732 data |= RREG32(mmGC_USER_SHADER_ARRAY_CONFIG); 3733 3734 data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK; 3735 data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT; 3736 3737 mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh); 3738 3739 return (~data) & mask; 3740 } 3741 3742 static void gfx_v7_0_init_ao_cu_mask(struct amdgpu_device *adev) 3743 { 3744 u32 tmp; 3745 3746 WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask); 3747 3748 tmp = RREG32(mmRLC_MAX_PG_CU); 3749 tmp &= ~RLC_MAX_PG_CU__MAX_POWERED_UP_CU_MASK; 3750 tmp |= (adev->gfx.cu_info.number << RLC_MAX_PG_CU__MAX_POWERED_UP_CU__SHIFT); 3751 WREG32(mmRLC_MAX_PG_CU, tmp); 3752 } 3753 3754 static void gfx_v7_0_enable_gfx_static_mgpg(struct amdgpu_device *adev, 3755 bool enable) 3756 { 3757 u32 data, orig; 3758 3759 orig = data = RREG32(mmRLC_PG_CNTL); 3760 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG)) 3761 data |= RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK; 3762 else 3763 data &= ~RLC_PG_CNTL__STATIC_PER_CU_PG_ENABLE_MASK; 3764 if (orig != data) 3765 WREG32(mmRLC_PG_CNTL, data); 3766 } 3767 3768 static void gfx_v7_0_enable_gfx_dynamic_mgpg(struct amdgpu_device *adev, 3769 bool enable) 3770 { 3771 u32 data, orig; 3772 3773 orig = data = RREG32(mmRLC_PG_CNTL); 3774 if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG)) 3775 data |= RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK; 3776 else 3777 data &= ~RLC_PG_CNTL__DYN_PER_CU_PG_ENABLE_MASK; 3778 if (orig != data) 3779 WREG32(mmRLC_PG_CNTL, data); 3780 } 3781 3782 #define RLC_SAVE_AND_RESTORE_STARTING_OFFSET 0x90 3783 #define RLC_CLEAR_STATE_DESCRIPTOR_OFFSET 0x3D 3784 3785 static void gfx_v7_0_init_gfx_cgpg(struct amdgpu_device *adev) 3786 { 3787 u32 data, orig; 3788 u32 i; 3789 3790 if (adev->gfx.rlc.cs_data) { 3791 WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET); 3792 WREG32(mmRLC_GPM_SCRATCH_DATA, upper_32_bits(adev->gfx.rlc.clear_state_gpu_addr)); 3793 WREG32(mmRLC_GPM_SCRATCH_DATA, lower_32_bits(adev->gfx.rlc.clear_state_gpu_addr)); 3794 WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.clear_state_size); 3795 } else { 3796 WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_CLEAR_STATE_DESCRIPTOR_OFFSET); 3797 for (i = 0; i < 3; i++) 3798 WREG32(mmRLC_GPM_SCRATCH_DATA, 0); 3799 } 3800 if (adev->gfx.rlc.reg_list) { 3801 WREG32(mmRLC_GPM_SCRATCH_ADDR, RLC_SAVE_AND_RESTORE_STARTING_OFFSET); 3802 for (i = 0; i < adev->gfx.rlc.reg_list_size; i++) 3803 WREG32(mmRLC_GPM_SCRATCH_DATA, adev->gfx.rlc.reg_list[i]); 3804 } 3805 3806 orig = data = RREG32(mmRLC_PG_CNTL); 3807 data |= RLC_PG_CNTL__GFX_POWER_GATING_SRC_MASK; 3808 if (orig != data) 3809 WREG32(mmRLC_PG_CNTL, data); 3810 3811 WREG32(mmRLC_SAVE_AND_RESTORE_BASE, adev->gfx.rlc.save_restore_gpu_addr >> 8); 3812 WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8); 3813 3814 data = RREG32(mmCP_RB_WPTR_POLL_CNTL); 3815 data &= ~CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK; 3816 data |= (0x60 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT); 3817 WREG32(mmCP_RB_WPTR_POLL_CNTL, data); 3818 3819 data = 0x10101010; 3820 WREG32(mmRLC_PG_DELAY, data); 3821 3822 data = RREG32(mmRLC_PG_DELAY_2); 3823 data &= ~0xff; 3824 data |= 0x3; 3825 WREG32(mmRLC_PG_DELAY_2, data); 3826 3827 data = RREG32(mmRLC_AUTO_PG_CTRL); 3828 data &= ~RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD_MASK; 3829 data |= (0x700 << RLC_AUTO_PG_CTRL__GRBM_REG_SAVE_GFX_IDLE_THRESHOLD__SHIFT); 3830 WREG32(mmRLC_AUTO_PG_CTRL, data); 3831 3832 } 3833 3834 static void gfx_v7_0_update_gfx_pg(struct amdgpu_device *adev, bool enable) 3835 { 3836 gfx_v7_0_enable_gfx_cgpg(adev, enable); 3837 gfx_v7_0_enable_gfx_static_mgpg(adev, enable); 3838 gfx_v7_0_enable_gfx_dynamic_mgpg(adev, enable); 3839 } 3840 3841 static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev) 3842 { 3843 u32 count = 0; 3844 const struct cs_section_def *sect = NULL; 3845 const struct cs_extent_def *ext = NULL; 3846 3847 if (adev->gfx.rlc.cs_data == NULL) 3848 return 0; 3849 3850 /* begin clear state */ 3851 count += 2; 3852 /* context control state */ 3853 count += 3; 3854 3855 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 3856 for (ext = sect->section; ext->extent != NULL; ++ext) { 3857 if (sect->id == SECT_CONTEXT) 3858 count += 2 + ext->reg_count; 3859 else 3860 return 0; 3861 } 3862 } 3863 /* pa_sc_raster_config/pa_sc_raster_config1 */ 3864 count += 4; 3865 /* end clear state */ 3866 count += 2; 3867 /* clear state */ 3868 count += 2; 3869 3870 return count; 3871 } 3872 3873 static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, 3874 volatile u32 *buffer) 3875 { 3876 u32 count = 0, i; 3877 const struct cs_section_def *sect = NULL; 3878 const struct cs_extent_def *ext = NULL; 3879 3880 if (adev->gfx.rlc.cs_data == NULL) 3881 return; 3882 if (buffer == NULL) 3883 return; 3884 3885 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 3886 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 3887 3888 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 3889 buffer[count++] = cpu_to_le32(0x80000000); 3890 buffer[count++] = cpu_to_le32(0x80000000); 3891 3892 for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) { 3893 for (ext = sect->section; ext->extent != NULL; ++ext) { 3894 if (sect->id == SECT_CONTEXT) { 3895 buffer[count++] = 3896 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count)); 3897 buffer[count++] = cpu_to_le32(ext->reg_index - PACKET3_SET_CONTEXT_REG_START); 3898 for (i = 0; i < ext->reg_count; i++) 3899 buffer[count++] = cpu_to_le32(ext->extent[i]); 3900 } else { 3901 return; 3902 } 3903 } 3904 } 3905 3906 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2)); 3907 buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START); 3908 switch (adev->asic_type) { 3909 case CHIP_BONAIRE: 3910 buffer[count++] = cpu_to_le32(0x16000012); 3911 buffer[count++] = cpu_to_le32(0x00000000); 3912 break; 3913 case CHIP_KAVERI: 3914 buffer[count++] = cpu_to_le32(0x00000000); /* XXX */ 3915 buffer[count++] = cpu_to_le32(0x00000000); 3916 break; 3917 case CHIP_KABINI: 3918 case CHIP_MULLINS: 3919 buffer[count++] = cpu_to_le32(0x00000000); /* XXX */ 3920 buffer[count++] = cpu_to_le32(0x00000000); 3921 break; 3922 case CHIP_HAWAII: 3923 buffer[count++] = cpu_to_le32(0x3a00161a); 3924 buffer[count++] = cpu_to_le32(0x0000002e); 3925 break; 3926 default: 3927 buffer[count++] = cpu_to_le32(0x00000000); 3928 buffer[count++] = cpu_to_le32(0x00000000); 3929 break; 3930 } 3931 3932 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 3933 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE); 3934 3935 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0)); 3936 buffer[count++] = cpu_to_le32(0); 3937 } 3938 3939 static void gfx_v7_0_init_pg(struct amdgpu_device *adev) 3940 { 3941 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | 3942 AMD_PG_SUPPORT_GFX_SMG | 3943 AMD_PG_SUPPORT_GFX_DMG | 3944 AMD_PG_SUPPORT_CP | 3945 AMD_PG_SUPPORT_GDS | 3946 AMD_PG_SUPPORT_RLC_SMU_HS)) { 3947 gfx_v7_0_enable_sclk_slowdown_on_pu(adev, true); 3948 gfx_v7_0_enable_sclk_slowdown_on_pd(adev, true); 3949 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { 3950 gfx_v7_0_init_gfx_cgpg(adev); 3951 gfx_v7_0_enable_cp_pg(adev, true); 3952 gfx_v7_0_enable_gds_pg(adev, true); 3953 } 3954 gfx_v7_0_init_ao_cu_mask(adev); 3955 gfx_v7_0_update_gfx_pg(adev, true); 3956 } 3957 } 3958 3959 static void gfx_v7_0_fini_pg(struct amdgpu_device *adev) 3960 { 3961 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | 3962 AMD_PG_SUPPORT_GFX_SMG | 3963 AMD_PG_SUPPORT_GFX_DMG | 3964 AMD_PG_SUPPORT_CP | 3965 AMD_PG_SUPPORT_GDS | 3966 AMD_PG_SUPPORT_RLC_SMU_HS)) { 3967 gfx_v7_0_update_gfx_pg(adev, false); 3968 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { 3969 gfx_v7_0_enable_cp_pg(adev, false); 3970 gfx_v7_0_enable_gds_pg(adev, false); 3971 } 3972 } 3973 } 3974 3975 /** 3976 * gfx_v7_0_get_gpu_clock_counter - return GPU clock counter snapshot 3977 * 3978 * @adev: amdgpu_device pointer 3979 * 3980 * Fetches a GPU clock counter snapshot (SI). 3981 * Returns the 64 bit clock counter snapshot. 3982 */ 3983 static uint64_t gfx_v7_0_get_gpu_clock_counter(struct amdgpu_device *adev) 3984 { 3985 uint64_t clock; 3986 3987 mutex_lock(&adev->gfx.gpu_clock_mutex); 3988 WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1); 3989 clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) | 3990 ((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL); 3991 mutex_unlock(&adev->gfx.gpu_clock_mutex); 3992 return clock; 3993 } 3994 3995 static void gfx_v7_0_ring_emit_gds_switch(struct amdgpu_ring *ring, 3996 uint32_t vmid, 3997 uint32_t gds_base, uint32_t gds_size, 3998 uint32_t gws_base, uint32_t gws_size, 3999 uint32_t oa_base, uint32_t oa_size) 4000 { 4001 /* GDS Base */ 4002 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 4003 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 4004 WRITE_DATA_DST_SEL(0))); 4005 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base); 4006 amdgpu_ring_write(ring, 0); 4007 amdgpu_ring_write(ring, gds_base); 4008 4009 /* GDS Size */ 4010 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 4011 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 4012 WRITE_DATA_DST_SEL(0))); 4013 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size); 4014 amdgpu_ring_write(ring, 0); 4015 amdgpu_ring_write(ring, gds_size); 4016 4017 /* GWS */ 4018 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 4019 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 4020 WRITE_DATA_DST_SEL(0))); 4021 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws); 4022 amdgpu_ring_write(ring, 0); 4023 amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base); 4024 4025 /* OA */ 4026 amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 4027 amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) | 4028 WRITE_DATA_DST_SEL(0))); 4029 amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa); 4030 amdgpu_ring_write(ring, 0); 4031 amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base)); 4032 } 4033 4034 static void gfx_v7_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid) 4035 { 4036 struct amdgpu_device *adev = ring->adev; 4037 uint32_t value = 0; 4038 4039 value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03); 4040 value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01); 4041 value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1); 4042 value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid); 4043 WREG32(mmSQ_CMD, value); 4044 } 4045 4046 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address) 4047 { 4048 WREG32(mmSQ_IND_INDEX, 4049 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 4050 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 4051 (address << SQ_IND_INDEX__INDEX__SHIFT) | 4052 (SQ_IND_INDEX__FORCE_READ_MASK)); 4053 return RREG32(mmSQ_IND_DATA); 4054 } 4055 4056 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd, 4057 uint32_t wave, uint32_t thread, 4058 uint32_t regno, uint32_t num, uint32_t *out) 4059 { 4060 WREG32(mmSQ_IND_INDEX, 4061 (wave << SQ_IND_INDEX__WAVE_ID__SHIFT) | 4062 (simd << SQ_IND_INDEX__SIMD_ID__SHIFT) | 4063 (regno << SQ_IND_INDEX__INDEX__SHIFT) | 4064 (thread << SQ_IND_INDEX__THREAD_ID__SHIFT) | 4065 (SQ_IND_INDEX__FORCE_READ_MASK) | 4066 (SQ_IND_INDEX__AUTO_INCR_MASK)); 4067 while (num--) 4068 *(out++) = RREG32(mmSQ_IND_DATA); 4069 } 4070 4071 static void gfx_v7_0_read_wave_data(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields) 4072 { 4073 /* type 0 wave data */ 4074 dst[(*no_fields)++] = 0; 4075 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS); 4076 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO); 4077 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI); 4078 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO); 4079 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI); 4080 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID); 4081 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0); 4082 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1); 4083 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC); 4084 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC); 4085 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS); 4086 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS); 4087 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO); 4088 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI); 4089 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO); 4090 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI); 4091 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0); 4092 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0); 4093 dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE); 4094 } 4095 4096 static void gfx_v7_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, 4097 uint32_t wave, uint32_t start, 4098 uint32_t size, uint32_t *dst) 4099 { 4100 wave_read_regs( 4101 adev, simd, wave, 0, 4102 start + SQIND_WAVE_SGPRS_OFFSET, size, dst); 4103 } 4104 4105 static void gfx_v7_0_select_me_pipe_q(struct amdgpu_device *adev, 4106 u32 me, u32 pipe, u32 q, u32 vm, u32 xcc_id) 4107 { 4108 cik_srbm_select(adev, me, pipe, q, vm); 4109 } 4110 4111 static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = { 4112 .get_gpu_clock_counter = &gfx_v7_0_get_gpu_clock_counter, 4113 .select_se_sh = &gfx_v7_0_select_se_sh, 4114 .read_wave_data = &gfx_v7_0_read_wave_data, 4115 .read_wave_sgprs = &gfx_v7_0_read_wave_sgprs, 4116 .select_me_pipe_q = &gfx_v7_0_select_me_pipe_q 4117 }; 4118 4119 static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = { 4120 .is_rlc_enabled = gfx_v7_0_is_rlc_enabled, 4121 .set_safe_mode = gfx_v7_0_set_safe_mode, 4122 .unset_safe_mode = gfx_v7_0_unset_safe_mode, 4123 .init = gfx_v7_0_rlc_init, 4124 .get_csb_size = gfx_v7_0_get_csb_size, 4125 .get_csb_buffer = gfx_v7_0_get_csb_buffer, 4126 .get_cp_table_num = gfx_v7_0_cp_pg_table_num, 4127 .resume = gfx_v7_0_rlc_resume, 4128 .stop = gfx_v7_0_rlc_stop, 4129 .reset = gfx_v7_0_rlc_reset, 4130 .start = gfx_v7_0_rlc_start, 4131 .update_spm_vmid = gfx_v7_0_update_spm_vmid 4132 }; 4133 4134 static int gfx_v7_0_early_init(void *handle) 4135 { 4136 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4137 4138 adev->gfx.xcc_mask = 1; 4139 adev->gfx.num_gfx_rings = GFX7_NUM_GFX_RINGS; 4140 adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev), 4141 AMDGPU_MAX_COMPUTE_RINGS); 4142 adev->gfx.funcs = &gfx_v7_0_gfx_funcs; 4143 adev->gfx.rlc.funcs = &gfx_v7_0_rlc_funcs; 4144 gfx_v7_0_set_ring_funcs(adev); 4145 gfx_v7_0_set_irq_funcs(adev); 4146 gfx_v7_0_set_gds_init(adev); 4147 4148 return 0; 4149 } 4150 4151 static int gfx_v7_0_late_init(void *handle) 4152 { 4153 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4154 int r; 4155 4156 r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0); 4157 if (r) 4158 return r; 4159 4160 r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0); 4161 if (r) 4162 return r; 4163 4164 return 0; 4165 } 4166 4167 static void gfx_v7_0_gpu_early_init(struct amdgpu_device *adev) 4168 { 4169 u32 gb_addr_config; 4170 u32 mc_arb_ramcfg; 4171 u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map; 4172 u32 tmp; 4173 4174 switch (adev->asic_type) { 4175 case CHIP_BONAIRE: 4176 adev->gfx.config.max_shader_engines = 2; 4177 adev->gfx.config.max_tile_pipes = 4; 4178 adev->gfx.config.max_cu_per_sh = 7; 4179 adev->gfx.config.max_sh_per_se = 1; 4180 adev->gfx.config.max_backends_per_se = 2; 4181 adev->gfx.config.max_texture_channel_caches = 4; 4182 adev->gfx.config.max_gprs = 256; 4183 adev->gfx.config.max_gs_threads = 32; 4184 adev->gfx.config.max_hw_contexts = 8; 4185 4186 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 4187 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 4188 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 4189 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 4190 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN; 4191 break; 4192 case CHIP_HAWAII: 4193 adev->gfx.config.max_shader_engines = 4; 4194 adev->gfx.config.max_tile_pipes = 16; 4195 adev->gfx.config.max_cu_per_sh = 11; 4196 adev->gfx.config.max_sh_per_se = 1; 4197 adev->gfx.config.max_backends_per_se = 4; 4198 adev->gfx.config.max_texture_channel_caches = 16; 4199 adev->gfx.config.max_gprs = 256; 4200 adev->gfx.config.max_gs_threads = 32; 4201 adev->gfx.config.max_hw_contexts = 8; 4202 4203 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 4204 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 4205 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 4206 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 4207 gb_addr_config = HAWAII_GB_ADDR_CONFIG_GOLDEN; 4208 break; 4209 case CHIP_KAVERI: 4210 adev->gfx.config.max_shader_engines = 1; 4211 adev->gfx.config.max_tile_pipes = 4; 4212 adev->gfx.config.max_cu_per_sh = 8; 4213 adev->gfx.config.max_backends_per_se = 2; 4214 adev->gfx.config.max_sh_per_se = 1; 4215 adev->gfx.config.max_texture_channel_caches = 4; 4216 adev->gfx.config.max_gprs = 256; 4217 adev->gfx.config.max_gs_threads = 16; 4218 adev->gfx.config.max_hw_contexts = 8; 4219 4220 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 4221 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 4222 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 4223 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 4224 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN; 4225 break; 4226 case CHIP_KABINI: 4227 case CHIP_MULLINS: 4228 default: 4229 adev->gfx.config.max_shader_engines = 1; 4230 adev->gfx.config.max_tile_pipes = 2; 4231 adev->gfx.config.max_cu_per_sh = 2; 4232 adev->gfx.config.max_sh_per_se = 1; 4233 adev->gfx.config.max_backends_per_se = 1; 4234 adev->gfx.config.max_texture_channel_caches = 2; 4235 adev->gfx.config.max_gprs = 256; 4236 adev->gfx.config.max_gs_threads = 16; 4237 adev->gfx.config.max_hw_contexts = 8; 4238 4239 adev->gfx.config.sc_prim_fifo_size_frontend = 0x20; 4240 adev->gfx.config.sc_prim_fifo_size_backend = 0x100; 4241 adev->gfx.config.sc_hiz_tile_fifo_size = 0x30; 4242 adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130; 4243 gb_addr_config = BONAIRE_GB_ADDR_CONFIG_GOLDEN; 4244 break; 4245 } 4246 4247 adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG); 4248 mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg; 4249 4250 adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg, 4251 MC_ARB_RAMCFG, NOOFBANK); 4252 adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg, 4253 MC_ARB_RAMCFG, NOOFRANKS); 4254 4255 adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes; 4256 adev->gfx.config.mem_max_burst_length_bytes = 256; 4257 if (adev->flags & AMD_IS_APU) { 4258 /* Get memory bank mapping mode. */ 4259 tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING); 4260 dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP); 4261 dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP); 4262 4263 tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING); 4264 dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP); 4265 dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP); 4266 4267 /* Validate settings in case only one DIMM installed. */ 4268 if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12)) 4269 dimm00_addr_map = 0; 4270 if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12)) 4271 dimm01_addr_map = 0; 4272 if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12)) 4273 dimm10_addr_map = 0; 4274 if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12)) 4275 dimm11_addr_map = 0; 4276 4277 /* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */ 4278 /* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */ 4279 if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11)) 4280 adev->gfx.config.mem_row_size_in_kb = 2; 4281 else 4282 adev->gfx.config.mem_row_size_in_kb = 1; 4283 } else { 4284 tmp = (mc_arb_ramcfg & MC_ARB_RAMCFG__NOOFCOLS_MASK) >> MC_ARB_RAMCFG__NOOFCOLS__SHIFT; 4285 adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; 4286 if (adev->gfx.config.mem_row_size_in_kb > 4) 4287 adev->gfx.config.mem_row_size_in_kb = 4; 4288 } 4289 /* XXX use MC settings? */ 4290 adev->gfx.config.shader_engine_tile_size = 32; 4291 adev->gfx.config.num_gpus = 1; 4292 adev->gfx.config.multi_gpu_tile_size = 64; 4293 4294 /* fix up row size */ 4295 gb_addr_config &= ~GB_ADDR_CONFIG__ROW_SIZE_MASK; 4296 switch (adev->gfx.config.mem_row_size_in_kb) { 4297 case 1: 4298 default: 4299 gb_addr_config |= (0 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT); 4300 break; 4301 case 2: 4302 gb_addr_config |= (1 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT); 4303 break; 4304 case 4: 4305 gb_addr_config |= (2 << GB_ADDR_CONFIG__ROW_SIZE__SHIFT); 4306 break; 4307 } 4308 adev->gfx.config.gb_addr_config = gb_addr_config; 4309 } 4310 4311 static int gfx_v7_0_compute_ring_init(struct amdgpu_device *adev, int ring_id, 4312 int mec, int pipe, int queue) 4313 { 4314 int r; 4315 unsigned irq_type; 4316 struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id]; 4317 4318 /* mec0 is me1 */ 4319 ring->me = mec + 1; 4320 ring->pipe = pipe; 4321 ring->queue = queue; 4322 4323 ring->ring_obj = NULL; 4324 ring->use_doorbell = true; 4325 ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id; 4326 sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue); 4327 4328 irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP 4329 + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec) 4330 + ring->pipe; 4331 4332 /* type-2 packets are deprecated on MEC, use type-3 instead */ 4333 r = amdgpu_ring_init(adev, ring, 1024, 4334 &adev->gfx.eop_irq, irq_type, 4335 AMDGPU_RING_PRIO_DEFAULT, NULL); 4336 if (r) 4337 return r; 4338 4339 4340 return 0; 4341 } 4342 4343 static int gfx_v7_0_sw_init(void *handle) 4344 { 4345 struct amdgpu_ring *ring; 4346 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4347 int i, j, k, r, ring_id; 4348 4349 switch (adev->asic_type) { 4350 case CHIP_KAVERI: 4351 adev->gfx.mec.num_mec = 2; 4352 break; 4353 case CHIP_BONAIRE: 4354 case CHIP_HAWAII: 4355 case CHIP_KABINI: 4356 case CHIP_MULLINS: 4357 default: 4358 adev->gfx.mec.num_mec = 1; 4359 break; 4360 } 4361 adev->gfx.mec.num_pipe_per_mec = 4; 4362 adev->gfx.mec.num_queue_per_pipe = 8; 4363 4364 /* EOP Event */ 4365 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 181, &adev->gfx.eop_irq); 4366 if (r) 4367 return r; 4368 4369 /* Privileged reg */ 4370 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 184, 4371 &adev->gfx.priv_reg_irq); 4372 if (r) 4373 return r; 4374 4375 /* Privileged inst */ 4376 r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 185, 4377 &adev->gfx.priv_inst_irq); 4378 if (r) 4379 return r; 4380 4381 r = gfx_v7_0_init_microcode(adev); 4382 if (r) { 4383 DRM_ERROR("Failed to load gfx firmware!\n"); 4384 return r; 4385 } 4386 4387 r = adev->gfx.rlc.funcs->init(adev); 4388 if (r) { 4389 DRM_ERROR("Failed to init rlc BOs!\n"); 4390 return r; 4391 } 4392 4393 /* allocate mec buffers */ 4394 r = gfx_v7_0_mec_init(adev); 4395 if (r) { 4396 DRM_ERROR("Failed to init MEC BOs!\n"); 4397 return r; 4398 } 4399 4400 for (i = 0; i < adev->gfx.num_gfx_rings; i++) { 4401 ring = &adev->gfx.gfx_ring[i]; 4402 ring->ring_obj = NULL; 4403 sprintf(ring->name, "gfx"); 4404 r = amdgpu_ring_init(adev, ring, 1024, 4405 &adev->gfx.eop_irq, 4406 AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP, 4407 AMDGPU_RING_PRIO_DEFAULT, NULL); 4408 if (r) 4409 return r; 4410 } 4411 4412 /* set up the compute queues - allocate horizontally across pipes */ 4413 ring_id = 0; 4414 for (i = 0; i < adev->gfx.mec.num_mec; ++i) { 4415 for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) { 4416 for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) { 4417 if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i, 4418 k, j)) 4419 continue; 4420 4421 r = gfx_v7_0_compute_ring_init(adev, 4422 ring_id, 4423 i, k, j); 4424 if (r) 4425 return r; 4426 4427 ring_id++; 4428 } 4429 } 4430 } 4431 4432 adev->gfx.ce_ram_size = 0x8000; 4433 4434 gfx_v7_0_gpu_early_init(adev); 4435 4436 return r; 4437 } 4438 4439 static int gfx_v7_0_sw_fini(void *handle) 4440 { 4441 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4442 int i; 4443 4444 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 4445 amdgpu_ring_fini(&adev->gfx.gfx_ring[i]); 4446 for (i = 0; i < adev->gfx.num_compute_rings; i++) 4447 amdgpu_ring_fini(&adev->gfx.compute_ring[i]); 4448 4449 gfx_v7_0_cp_compute_fini(adev); 4450 amdgpu_gfx_rlc_fini(adev); 4451 gfx_v7_0_mec_fini(adev); 4452 amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj, 4453 &adev->gfx.rlc.clear_state_gpu_addr, 4454 (void **)&adev->gfx.rlc.cs_ptr); 4455 if (adev->gfx.rlc.cp_table_size) { 4456 amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj, 4457 &adev->gfx.rlc.cp_table_gpu_addr, 4458 (void **)&adev->gfx.rlc.cp_table_ptr); 4459 } 4460 gfx_v7_0_free_microcode(adev); 4461 4462 return 0; 4463 } 4464 4465 static int gfx_v7_0_hw_init(void *handle) 4466 { 4467 int r; 4468 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4469 4470 gfx_v7_0_constants_init(adev); 4471 4472 /* init CSB */ 4473 adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr); 4474 /* init rlc */ 4475 r = adev->gfx.rlc.funcs->resume(adev); 4476 if (r) 4477 return r; 4478 4479 r = gfx_v7_0_cp_resume(adev); 4480 if (r) 4481 return r; 4482 4483 return r; 4484 } 4485 4486 static int gfx_v7_0_hw_fini(void *handle) 4487 { 4488 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4489 4490 amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); 4491 amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); 4492 gfx_v7_0_cp_enable(adev, false); 4493 adev->gfx.rlc.funcs->stop(adev); 4494 gfx_v7_0_fini_pg(adev); 4495 4496 return 0; 4497 } 4498 4499 static int gfx_v7_0_suspend(void *handle) 4500 { 4501 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4502 4503 return gfx_v7_0_hw_fini(adev); 4504 } 4505 4506 static int gfx_v7_0_resume(void *handle) 4507 { 4508 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4509 4510 return gfx_v7_0_hw_init(adev); 4511 } 4512 4513 static bool gfx_v7_0_is_idle(void *handle) 4514 { 4515 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4516 4517 if (RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK) 4518 return false; 4519 else 4520 return true; 4521 } 4522 4523 static int gfx_v7_0_wait_for_idle(void *handle) 4524 { 4525 unsigned i; 4526 u32 tmp; 4527 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4528 4529 for (i = 0; i < adev->usec_timeout; i++) { 4530 /* read MC_STATUS */ 4531 tmp = RREG32(mmGRBM_STATUS) & GRBM_STATUS__GUI_ACTIVE_MASK; 4532 4533 if (!tmp) 4534 return 0; 4535 udelay(1); 4536 } 4537 return -ETIMEDOUT; 4538 } 4539 4540 static int gfx_v7_0_soft_reset(void *handle) 4541 { 4542 u32 grbm_soft_reset = 0, srbm_soft_reset = 0; 4543 u32 tmp; 4544 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4545 4546 /* GRBM_STATUS */ 4547 tmp = RREG32(mmGRBM_STATUS); 4548 if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK | 4549 GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK | 4550 GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK | 4551 GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK | 4552 GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK | 4553 GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) 4554 grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK | 4555 GRBM_SOFT_RESET__SOFT_RESET_GFX_MASK; 4556 4557 if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) { 4558 grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_CP_MASK; 4559 srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK; 4560 } 4561 4562 /* GRBM_STATUS2 */ 4563 tmp = RREG32(mmGRBM_STATUS2); 4564 if (tmp & GRBM_STATUS2__RLC_BUSY_MASK) 4565 grbm_soft_reset |= GRBM_SOFT_RESET__SOFT_RESET_RLC_MASK; 4566 4567 /* SRBM_STATUS */ 4568 tmp = RREG32(mmSRBM_STATUS); 4569 if (tmp & SRBM_STATUS__GRBM_RQ_PENDING_MASK) 4570 srbm_soft_reset |= SRBM_SOFT_RESET__SOFT_RESET_GRBM_MASK; 4571 4572 if (grbm_soft_reset || srbm_soft_reset) { 4573 /* disable CG/PG */ 4574 gfx_v7_0_fini_pg(adev); 4575 gfx_v7_0_update_cg(adev, false); 4576 4577 /* stop the rlc */ 4578 adev->gfx.rlc.funcs->stop(adev); 4579 4580 /* Disable GFX parsing/prefetching */ 4581 WREG32(mmCP_ME_CNTL, CP_ME_CNTL__ME_HALT_MASK | CP_ME_CNTL__PFP_HALT_MASK | CP_ME_CNTL__CE_HALT_MASK); 4582 4583 /* Disable MEC parsing/prefetching */ 4584 WREG32(mmCP_MEC_CNTL, CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK); 4585 4586 if (grbm_soft_reset) { 4587 tmp = RREG32(mmGRBM_SOFT_RESET); 4588 tmp |= grbm_soft_reset; 4589 dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp); 4590 WREG32(mmGRBM_SOFT_RESET, tmp); 4591 tmp = RREG32(mmGRBM_SOFT_RESET); 4592 4593 udelay(50); 4594 4595 tmp &= ~grbm_soft_reset; 4596 WREG32(mmGRBM_SOFT_RESET, tmp); 4597 tmp = RREG32(mmGRBM_SOFT_RESET); 4598 } 4599 4600 if (srbm_soft_reset) { 4601 tmp = RREG32(mmSRBM_SOFT_RESET); 4602 tmp |= srbm_soft_reset; 4603 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp); 4604 WREG32(mmSRBM_SOFT_RESET, tmp); 4605 tmp = RREG32(mmSRBM_SOFT_RESET); 4606 4607 udelay(50); 4608 4609 tmp &= ~srbm_soft_reset; 4610 WREG32(mmSRBM_SOFT_RESET, tmp); 4611 tmp = RREG32(mmSRBM_SOFT_RESET); 4612 } 4613 /* Wait a little for things to settle down */ 4614 udelay(50); 4615 } 4616 return 0; 4617 } 4618 4619 static void gfx_v7_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev, 4620 enum amdgpu_interrupt_state state) 4621 { 4622 u32 cp_int_cntl; 4623 4624 switch (state) { 4625 case AMDGPU_IRQ_STATE_DISABLE: 4626 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 4627 cp_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK; 4628 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 4629 break; 4630 case AMDGPU_IRQ_STATE_ENABLE: 4631 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 4632 cp_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK; 4633 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 4634 break; 4635 default: 4636 break; 4637 } 4638 } 4639 4640 static void gfx_v7_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev, 4641 int me, int pipe, 4642 enum amdgpu_interrupt_state state) 4643 { 4644 u32 mec_int_cntl, mec_int_cntl_reg; 4645 4646 /* 4647 * amdgpu controls only the first MEC. That's why this function only 4648 * handles the setting of interrupts for this specific MEC. All other 4649 * pipes' interrupts are set by amdkfd. 4650 */ 4651 4652 if (me == 1) { 4653 switch (pipe) { 4654 case 0: 4655 mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL; 4656 break; 4657 case 1: 4658 mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL; 4659 break; 4660 case 2: 4661 mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL; 4662 break; 4663 case 3: 4664 mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL; 4665 break; 4666 default: 4667 DRM_DEBUG("invalid pipe %d\n", pipe); 4668 return; 4669 } 4670 } else { 4671 DRM_DEBUG("invalid me %d\n", me); 4672 return; 4673 } 4674 4675 switch (state) { 4676 case AMDGPU_IRQ_STATE_DISABLE: 4677 mec_int_cntl = RREG32(mec_int_cntl_reg); 4678 mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK; 4679 WREG32(mec_int_cntl_reg, mec_int_cntl); 4680 break; 4681 case AMDGPU_IRQ_STATE_ENABLE: 4682 mec_int_cntl = RREG32(mec_int_cntl_reg); 4683 mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK; 4684 WREG32(mec_int_cntl_reg, mec_int_cntl); 4685 break; 4686 default: 4687 break; 4688 } 4689 } 4690 4691 static int gfx_v7_0_set_priv_reg_fault_state(struct amdgpu_device *adev, 4692 struct amdgpu_irq_src *src, 4693 unsigned type, 4694 enum amdgpu_interrupt_state state) 4695 { 4696 u32 cp_int_cntl; 4697 4698 switch (state) { 4699 case AMDGPU_IRQ_STATE_DISABLE: 4700 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 4701 cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK; 4702 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 4703 break; 4704 case AMDGPU_IRQ_STATE_ENABLE: 4705 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 4706 cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_REG_INT_ENABLE_MASK; 4707 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 4708 break; 4709 default: 4710 break; 4711 } 4712 4713 return 0; 4714 } 4715 4716 static int gfx_v7_0_set_priv_inst_fault_state(struct amdgpu_device *adev, 4717 struct amdgpu_irq_src *src, 4718 unsigned type, 4719 enum amdgpu_interrupt_state state) 4720 { 4721 u32 cp_int_cntl; 4722 4723 switch (state) { 4724 case AMDGPU_IRQ_STATE_DISABLE: 4725 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 4726 cp_int_cntl &= ~CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK; 4727 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 4728 break; 4729 case AMDGPU_IRQ_STATE_ENABLE: 4730 cp_int_cntl = RREG32(mmCP_INT_CNTL_RING0); 4731 cp_int_cntl |= CP_INT_CNTL_RING0__PRIV_INSTR_INT_ENABLE_MASK; 4732 WREG32(mmCP_INT_CNTL_RING0, cp_int_cntl); 4733 break; 4734 default: 4735 break; 4736 } 4737 4738 return 0; 4739 } 4740 4741 static int gfx_v7_0_set_eop_interrupt_state(struct amdgpu_device *adev, 4742 struct amdgpu_irq_src *src, 4743 unsigned type, 4744 enum amdgpu_interrupt_state state) 4745 { 4746 switch (type) { 4747 case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP: 4748 gfx_v7_0_set_gfx_eop_interrupt_state(adev, state); 4749 break; 4750 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP: 4751 gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 0, state); 4752 break; 4753 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP: 4754 gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 1, state); 4755 break; 4756 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP: 4757 gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 2, state); 4758 break; 4759 case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP: 4760 gfx_v7_0_set_compute_eop_interrupt_state(adev, 1, 3, state); 4761 break; 4762 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP: 4763 gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 0, state); 4764 break; 4765 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP: 4766 gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 1, state); 4767 break; 4768 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP: 4769 gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 2, state); 4770 break; 4771 case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP: 4772 gfx_v7_0_set_compute_eop_interrupt_state(adev, 2, 3, state); 4773 break; 4774 default: 4775 break; 4776 } 4777 return 0; 4778 } 4779 4780 static int gfx_v7_0_eop_irq(struct amdgpu_device *adev, 4781 struct amdgpu_irq_src *source, 4782 struct amdgpu_iv_entry *entry) 4783 { 4784 u8 me_id, pipe_id; 4785 struct amdgpu_ring *ring; 4786 int i; 4787 4788 DRM_DEBUG("IH: CP EOP\n"); 4789 me_id = (entry->ring_id & 0x0c) >> 2; 4790 pipe_id = (entry->ring_id & 0x03) >> 0; 4791 switch (me_id) { 4792 case 0: 4793 amdgpu_fence_process(&adev->gfx.gfx_ring[0]); 4794 break; 4795 case 1: 4796 case 2: 4797 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 4798 ring = &adev->gfx.compute_ring[i]; 4799 if ((ring->me == me_id) && (ring->pipe == pipe_id)) 4800 amdgpu_fence_process(ring); 4801 } 4802 break; 4803 } 4804 return 0; 4805 } 4806 4807 static void gfx_v7_0_fault(struct amdgpu_device *adev, 4808 struct amdgpu_iv_entry *entry) 4809 { 4810 struct amdgpu_ring *ring; 4811 u8 me_id, pipe_id; 4812 int i; 4813 4814 me_id = (entry->ring_id & 0x0c) >> 2; 4815 pipe_id = (entry->ring_id & 0x03) >> 0; 4816 switch (me_id) { 4817 case 0: 4818 drm_sched_fault(&adev->gfx.gfx_ring[0].sched); 4819 break; 4820 case 1: 4821 case 2: 4822 for (i = 0; i < adev->gfx.num_compute_rings; i++) { 4823 ring = &adev->gfx.compute_ring[i]; 4824 if ((ring->me == me_id) && (ring->pipe == pipe_id)) 4825 drm_sched_fault(&ring->sched); 4826 } 4827 break; 4828 } 4829 } 4830 4831 static int gfx_v7_0_priv_reg_irq(struct amdgpu_device *adev, 4832 struct amdgpu_irq_src *source, 4833 struct amdgpu_iv_entry *entry) 4834 { 4835 DRM_ERROR("Illegal register access in command stream\n"); 4836 gfx_v7_0_fault(adev, entry); 4837 return 0; 4838 } 4839 4840 static int gfx_v7_0_priv_inst_irq(struct amdgpu_device *adev, 4841 struct amdgpu_irq_src *source, 4842 struct amdgpu_iv_entry *entry) 4843 { 4844 DRM_ERROR("Illegal instruction in command stream\n"); 4845 // XXX soft reset the gfx block only 4846 gfx_v7_0_fault(adev, entry); 4847 return 0; 4848 } 4849 4850 static int gfx_v7_0_set_clockgating_state(void *handle, 4851 enum amd_clockgating_state state) 4852 { 4853 bool gate = false; 4854 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4855 4856 if (state == AMD_CG_STATE_GATE) 4857 gate = true; 4858 4859 gfx_v7_0_enable_gui_idle_interrupt(adev, false); 4860 /* order matters! */ 4861 if (gate) { 4862 gfx_v7_0_enable_mgcg(adev, true); 4863 gfx_v7_0_enable_cgcg(adev, true); 4864 } else { 4865 gfx_v7_0_enable_cgcg(adev, false); 4866 gfx_v7_0_enable_mgcg(adev, false); 4867 } 4868 gfx_v7_0_enable_gui_idle_interrupt(adev, true); 4869 4870 return 0; 4871 } 4872 4873 static int gfx_v7_0_set_powergating_state(void *handle, 4874 enum amd_powergating_state state) 4875 { 4876 bool gate = false; 4877 struct amdgpu_device *adev = (struct amdgpu_device *)handle; 4878 4879 if (state == AMD_PG_STATE_GATE) 4880 gate = true; 4881 4882 if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_PG | 4883 AMD_PG_SUPPORT_GFX_SMG | 4884 AMD_PG_SUPPORT_GFX_DMG | 4885 AMD_PG_SUPPORT_CP | 4886 AMD_PG_SUPPORT_GDS | 4887 AMD_PG_SUPPORT_RLC_SMU_HS)) { 4888 gfx_v7_0_update_gfx_pg(adev, gate); 4889 if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) { 4890 gfx_v7_0_enable_cp_pg(adev, gate); 4891 gfx_v7_0_enable_gds_pg(adev, gate); 4892 } 4893 } 4894 4895 return 0; 4896 } 4897 4898 static void gfx_v7_0_emit_mem_sync(struct amdgpu_ring *ring) 4899 { 4900 amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3)); 4901 amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA | 4902 PACKET3_TC_ACTION_ENA | 4903 PACKET3_SH_KCACHE_ACTION_ENA | 4904 PACKET3_SH_ICACHE_ACTION_ENA); /* CP_COHER_CNTL */ 4905 amdgpu_ring_write(ring, 0xffffffff); /* CP_COHER_SIZE */ 4906 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */ 4907 amdgpu_ring_write(ring, 0x0000000A); /* poll interval */ 4908 } 4909 4910 static void gfx_v7_0_emit_mem_sync_compute(struct amdgpu_ring *ring) 4911 { 4912 amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5)); 4913 amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA | 4914 PACKET3_TC_ACTION_ENA | 4915 PACKET3_SH_KCACHE_ACTION_ENA | 4916 PACKET3_SH_ICACHE_ACTION_ENA); /* CP_COHER_CNTL */ 4917 amdgpu_ring_write(ring, 0xffffffff); /* CP_COHER_SIZE */ 4918 amdgpu_ring_write(ring, 0xff); /* CP_COHER_SIZE_HI */ 4919 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */ 4920 amdgpu_ring_write(ring, 0); /* CP_COHER_BASE_HI */ 4921 amdgpu_ring_write(ring, 0x0000000A); /* poll interval */ 4922 } 4923 4924 static const struct amd_ip_funcs gfx_v7_0_ip_funcs = { 4925 .name = "gfx_v7_0", 4926 .early_init = gfx_v7_0_early_init, 4927 .late_init = gfx_v7_0_late_init, 4928 .sw_init = gfx_v7_0_sw_init, 4929 .sw_fini = gfx_v7_0_sw_fini, 4930 .hw_init = gfx_v7_0_hw_init, 4931 .hw_fini = gfx_v7_0_hw_fini, 4932 .suspend = gfx_v7_0_suspend, 4933 .resume = gfx_v7_0_resume, 4934 .is_idle = gfx_v7_0_is_idle, 4935 .wait_for_idle = gfx_v7_0_wait_for_idle, 4936 .soft_reset = gfx_v7_0_soft_reset, 4937 .set_clockgating_state = gfx_v7_0_set_clockgating_state, 4938 .set_powergating_state = gfx_v7_0_set_powergating_state, 4939 .dump_ip_state = NULL, 4940 .print_ip_state = NULL, 4941 }; 4942 4943 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = { 4944 .type = AMDGPU_RING_TYPE_GFX, 4945 .align_mask = 0xff, 4946 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 4947 .support_64bit_ptrs = false, 4948 .get_rptr = gfx_v7_0_ring_get_rptr, 4949 .get_wptr = gfx_v7_0_ring_get_wptr_gfx, 4950 .set_wptr = gfx_v7_0_ring_set_wptr_gfx, 4951 .emit_frame_size = 4952 20 + /* gfx_v7_0_ring_emit_gds_switch */ 4953 7 + /* gfx_v7_0_ring_emit_hdp_flush */ 4954 5 + /* hdp invalidate */ 4955 12 + 12 + 12 + /* gfx_v7_0_ring_emit_fence_gfx x3 for user fence, vm fence */ 4956 7 + 4 + /* gfx_v7_0_ring_emit_pipeline_sync */ 4957 CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + 6 + /* gfx_v7_0_ring_emit_vm_flush */ 4958 3 + 4 + /* gfx_v7_ring_emit_cntxcntl including vgt flush*/ 4959 5, /* SURFACE_SYNC */ 4960 .emit_ib_size = 4, /* gfx_v7_0_ring_emit_ib_gfx */ 4961 .emit_ib = gfx_v7_0_ring_emit_ib_gfx, 4962 .emit_fence = gfx_v7_0_ring_emit_fence_gfx, 4963 .emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync, 4964 .emit_vm_flush = gfx_v7_0_ring_emit_vm_flush, 4965 .emit_gds_switch = gfx_v7_0_ring_emit_gds_switch, 4966 .emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush, 4967 .test_ring = gfx_v7_0_ring_test_ring, 4968 .test_ib = gfx_v7_0_ring_test_ib, 4969 .insert_nop = amdgpu_ring_insert_nop, 4970 .pad_ib = amdgpu_ring_generic_pad_ib, 4971 .emit_cntxcntl = gfx_v7_ring_emit_cntxcntl, 4972 .emit_wreg = gfx_v7_0_ring_emit_wreg, 4973 .soft_recovery = gfx_v7_0_ring_soft_recovery, 4974 .emit_mem_sync = gfx_v7_0_emit_mem_sync, 4975 }; 4976 4977 static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_compute = { 4978 .type = AMDGPU_RING_TYPE_COMPUTE, 4979 .align_mask = 0xff, 4980 .nop = PACKET3(PACKET3_NOP, 0x3FFF), 4981 .support_64bit_ptrs = false, 4982 .get_rptr = gfx_v7_0_ring_get_rptr, 4983 .get_wptr = gfx_v7_0_ring_get_wptr_compute, 4984 .set_wptr = gfx_v7_0_ring_set_wptr_compute, 4985 .emit_frame_size = 4986 20 + /* gfx_v7_0_ring_emit_gds_switch */ 4987 7 + /* gfx_v7_0_ring_emit_hdp_flush */ 4988 5 + /* hdp invalidate */ 4989 7 + /* gfx_v7_0_ring_emit_pipeline_sync */ 4990 CIK_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v7_0_ring_emit_vm_flush */ 4991 7 + 7 + 7 + /* gfx_v7_0_ring_emit_fence_compute x3 for user fence, vm fence */ 4992 7, /* gfx_v7_0_emit_mem_sync_compute */ 4993 .emit_ib_size = 7, /* gfx_v7_0_ring_emit_ib_compute */ 4994 .emit_ib = gfx_v7_0_ring_emit_ib_compute, 4995 .emit_fence = gfx_v7_0_ring_emit_fence_compute, 4996 .emit_pipeline_sync = gfx_v7_0_ring_emit_pipeline_sync, 4997 .emit_vm_flush = gfx_v7_0_ring_emit_vm_flush, 4998 .emit_gds_switch = gfx_v7_0_ring_emit_gds_switch, 4999 .emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush, 5000 .test_ring = gfx_v7_0_ring_test_ring, 5001 .test_ib = gfx_v7_0_ring_test_ib, 5002 .insert_nop = amdgpu_ring_insert_nop, 5003 .pad_ib = amdgpu_ring_generic_pad_ib, 5004 .emit_wreg = gfx_v7_0_ring_emit_wreg, 5005 .emit_mem_sync = gfx_v7_0_emit_mem_sync_compute, 5006 }; 5007 5008 static void gfx_v7_0_set_ring_funcs(struct amdgpu_device *adev) 5009 { 5010 int i; 5011 5012 for (i = 0; i < adev->gfx.num_gfx_rings; i++) 5013 adev->gfx.gfx_ring[i].funcs = &gfx_v7_0_ring_funcs_gfx; 5014 for (i = 0; i < adev->gfx.num_compute_rings; i++) 5015 adev->gfx.compute_ring[i].funcs = &gfx_v7_0_ring_funcs_compute; 5016 } 5017 5018 static const struct amdgpu_irq_src_funcs gfx_v7_0_eop_irq_funcs = { 5019 .set = gfx_v7_0_set_eop_interrupt_state, 5020 .process = gfx_v7_0_eop_irq, 5021 }; 5022 5023 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_reg_irq_funcs = { 5024 .set = gfx_v7_0_set_priv_reg_fault_state, 5025 .process = gfx_v7_0_priv_reg_irq, 5026 }; 5027 5028 static const struct amdgpu_irq_src_funcs gfx_v7_0_priv_inst_irq_funcs = { 5029 .set = gfx_v7_0_set_priv_inst_fault_state, 5030 .process = gfx_v7_0_priv_inst_irq, 5031 }; 5032 5033 static void gfx_v7_0_set_irq_funcs(struct amdgpu_device *adev) 5034 { 5035 adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST; 5036 adev->gfx.eop_irq.funcs = &gfx_v7_0_eop_irq_funcs; 5037 5038 adev->gfx.priv_reg_irq.num_types = 1; 5039 adev->gfx.priv_reg_irq.funcs = &gfx_v7_0_priv_reg_irq_funcs; 5040 5041 adev->gfx.priv_inst_irq.num_types = 1; 5042 adev->gfx.priv_inst_irq.funcs = &gfx_v7_0_priv_inst_irq_funcs; 5043 } 5044 5045 static void gfx_v7_0_set_gds_init(struct amdgpu_device *adev) 5046 { 5047 /* init asci gds info */ 5048 adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE); 5049 adev->gds.gws_size = 64; 5050 adev->gds.oa_size = 16; 5051 adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID); 5052 } 5053 5054 5055 static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev) 5056 { 5057 int i, j, k, counter, active_cu_number = 0; 5058 u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0; 5059 struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info; 5060 unsigned disable_masks[4 * 2]; 5061 u32 ao_cu_num; 5062 5063 if (adev->flags & AMD_IS_APU) 5064 ao_cu_num = 2; 5065 else 5066 ao_cu_num = adev->gfx.config.max_cu_per_sh; 5067 5068 memset(cu_info, 0, sizeof(*cu_info)); 5069 5070 amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2); 5071 5072 mutex_lock(&adev->grbm_idx_mutex); 5073 for (i = 0; i < adev->gfx.config.max_shader_engines; i++) { 5074 for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) { 5075 mask = 1; 5076 ao_bitmap = 0; 5077 counter = 0; 5078 gfx_v7_0_select_se_sh(adev, i, j, 0xffffffff, 0); 5079 if (i < 4 && j < 2) 5080 gfx_v7_0_set_user_cu_inactive_bitmap( 5081 adev, disable_masks[i * 2 + j]); 5082 bitmap = gfx_v7_0_get_cu_active_bitmap(adev); 5083 cu_info->bitmap[0][i][j] = bitmap; 5084 5085 for (k = 0; k < adev->gfx.config.max_cu_per_sh; k++) { 5086 if (bitmap & mask) { 5087 if (counter < ao_cu_num) 5088 ao_bitmap |= mask; 5089 counter++; 5090 } 5091 mask <<= 1; 5092 } 5093 active_cu_number += counter; 5094 if (i < 2 && j < 2) 5095 ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8)); 5096 cu_info->ao_cu_bitmap[i][j] = ao_bitmap; 5097 } 5098 } 5099 gfx_v7_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0); 5100 mutex_unlock(&adev->grbm_idx_mutex); 5101 5102 cu_info->number = active_cu_number; 5103 cu_info->ao_cu_mask = ao_cu_mask; 5104 cu_info->simd_per_cu = NUM_SIMD_PER_CU; 5105 cu_info->max_waves_per_simd = 10; 5106 cu_info->max_scratch_slots_per_cu = 32; 5107 cu_info->wave_front_size = 64; 5108 cu_info->lds_size = 64; 5109 } 5110 5111 const struct amdgpu_ip_block_version gfx_v7_1_ip_block = { 5112 .type = AMD_IP_BLOCK_TYPE_GFX, 5113 .major = 7, 5114 .minor = 1, 5115 .rev = 0, 5116 .funcs = &gfx_v7_0_ip_funcs, 5117 }; 5118 5119 const struct amdgpu_ip_block_version gfx_v7_2_ip_block = { 5120 .type = AMD_IP_BLOCK_TYPE_GFX, 5121 .major = 7, 5122 .minor = 2, 5123 .rev = 0, 5124 .funcs = &gfx_v7_0_ip_funcs, 5125 }; 5126 5127 const struct amdgpu_ip_block_version gfx_v7_3_ip_block = { 5128 .type = AMD_IP_BLOCK_TYPE_GFX, 5129 .major = 7, 5130 .minor = 3, 5131 .rev = 0, 5132 .funcs = &gfx_v7_0_ip_funcs, 5133 }; 5134