1 /* 2 * Copyright 2010 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 * Authors: Alex Deucher 23 */ 24 #include <linux/firmware.h> 25 #include <linux/platform_device.h> 26 #include <linux/slab.h> 27 #include <linux/module.h> 28 #include "drmP.h" 29 #include "radeon.h" 30 #include "radeon_asic.h" 31 #include "radeon_drm.h" 32 #include "nid.h" 33 #include "atom.h" 34 #include "ni_reg.h" 35 #include "cayman_blit_shaders.h" 36 37 extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save); 38 extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save); 39 extern int evergreen_mc_wait_for_idle(struct radeon_device *rdev); 40 extern void evergreen_mc_program(struct radeon_device *rdev); 41 extern void evergreen_irq_suspend(struct radeon_device *rdev); 42 extern int evergreen_mc_init(struct radeon_device *rdev); 43 extern void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev); 44 extern void evergreen_pcie_gen2_enable(struct radeon_device *rdev); 45 46 #define EVERGREEN_PFP_UCODE_SIZE 1120 47 #define EVERGREEN_PM4_UCODE_SIZE 1376 48 #define EVERGREEN_RLC_UCODE_SIZE 768 49 #define BTC_MC_UCODE_SIZE 6024 50 51 #define CAYMAN_PFP_UCODE_SIZE 2176 52 #define CAYMAN_PM4_UCODE_SIZE 2176 53 #define CAYMAN_RLC_UCODE_SIZE 1024 54 #define CAYMAN_MC_UCODE_SIZE 6037 55 56 /* Firmware Names */ 57 MODULE_FIRMWARE("radeon/BARTS_pfp.bin"); 58 MODULE_FIRMWARE("radeon/BARTS_me.bin"); 59 MODULE_FIRMWARE("radeon/BARTS_mc.bin"); 60 MODULE_FIRMWARE("radeon/BTC_rlc.bin"); 61 MODULE_FIRMWARE("radeon/TURKS_pfp.bin"); 62 MODULE_FIRMWARE("radeon/TURKS_me.bin"); 63 MODULE_FIRMWARE("radeon/TURKS_mc.bin"); 64 MODULE_FIRMWARE("radeon/CAICOS_pfp.bin"); 65 MODULE_FIRMWARE("radeon/CAICOS_me.bin"); 66 MODULE_FIRMWARE("radeon/CAICOS_mc.bin"); 67 MODULE_FIRMWARE("radeon/CAYMAN_pfp.bin"); 68 MODULE_FIRMWARE("radeon/CAYMAN_me.bin"); 69 MODULE_FIRMWARE("radeon/CAYMAN_mc.bin"); 70 MODULE_FIRMWARE("radeon/CAYMAN_rlc.bin"); 71 72 #define BTC_IO_MC_REGS_SIZE 29 73 74 static const u32 barts_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = { 75 {0x00000077, 0xff010100}, 76 {0x00000078, 0x00000000}, 77 {0x00000079, 0x00001434}, 78 {0x0000007a, 0xcc08ec08}, 79 {0x0000007b, 0x00040000}, 80 {0x0000007c, 0x000080c0}, 81 {0x0000007d, 0x09000000}, 82 {0x0000007e, 0x00210404}, 83 {0x00000081, 0x08a8e800}, 84 {0x00000082, 0x00030444}, 85 {0x00000083, 0x00000000}, 86 {0x00000085, 0x00000001}, 87 {0x00000086, 0x00000002}, 88 {0x00000087, 0x48490000}, 89 {0x00000088, 0x20244647}, 90 {0x00000089, 0x00000005}, 91 {0x0000008b, 0x66030000}, 92 {0x0000008c, 0x00006603}, 93 {0x0000008d, 0x00000100}, 94 {0x0000008f, 0x00001c0a}, 95 {0x00000090, 0xff000001}, 96 {0x00000094, 0x00101101}, 97 {0x00000095, 0x00000fff}, 98 {0x00000096, 0x00116fff}, 99 {0x00000097, 0x60010000}, 100 {0x00000098, 0x10010000}, 101 {0x00000099, 0x00006000}, 102 {0x0000009a, 0x00001000}, 103 {0x0000009f, 0x00946a00} 104 }; 105 106 static const u32 turks_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = { 107 {0x00000077, 0xff010100}, 108 {0x00000078, 0x00000000}, 109 {0x00000079, 0x00001434}, 110 {0x0000007a, 0xcc08ec08}, 111 {0x0000007b, 0x00040000}, 112 {0x0000007c, 0x000080c0}, 113 {0x0000007d, 0x09000000}, 114 {0x0000007e, 0x00210404}, 115 {0x00000081, 0x08a8e800}, 116 {0x00000082, 0x00030444}, 117 {0x00000083, 0x00000000}, 118 {0x00000085, 0x00000001}, 119 {0x00000086, 0x00000002}, 120 {0x00000087, 0x48490000}, 121 {0x00000088, 0x20244647}, 122 {0x00000089, 0x00000005}, 123 {0x0000008b, 0x66030000}, 124 {0x0000008c, 0x00006603}, 125 {0x0000008d, 0x00000100}, 126 {0x0000008f, 0x00001c0a}, 127 {0x00000090, 0xff000001}, 128 {0x00000094, 0x00101101}, 129 {0x00000095, 0x00000fff}, 130 {0x00000096, 0x00116fff}, 131 {0x00000097, 0x60010000}, 132 {0x00000098, 0x10010000}, 133 {0x00000099, 0x00006000}, 134 {0x0000009a, 0x00001000}, 135 {0x0000009f, 0x00936a00} 136 }; 137 138 static const u32 caicos_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = { 139 {0x00000077, 0xff010100}, 140 {0x00000078, 0x00000000}, 141 {0x00000079, 0x00001434}, 142 {0x0000007a, 0xcc08ec08}, 143 {0x0000007b, 0x00040000}, 144 {0x0000007c, 0x000080c0}, 145 {0x0000007d, 0x09000000}, 146 {0x0000007e, 0x00210404}, 147 {0x00000081, 0x08a8e800}, 148 {0x00000082, 0x00030444}, 149 {0x00000083, 0x00000000}, 150 {0x00000085, 0x00000001}, 151 {0x00000086, 0x00000002}, 152 {0x00000087, 0x48490000}, 153 {0x00000088, 0x20244647}, 154 {0x00000089, 0x00000005}, 155 {0x0000008b, 0x66030000}, 156 {0x0000008c, 0x00006603}, 157 {0x0000008d, 0x00000100}, 158 {0x0000008f, 0x00001c0a}, 159 {0x00000090, 0xff000001}, 160 {0x00000094, 0x00101101}, 161 {0x00000095, 0x00000fff}, 162 {0x00000096, 0x00116fff}, 163 {0x00000097, 0x60010000}, 164 {0x00000098, 0x10010000}, 165 {0x00000099, 0x00006000}, 166 {0x0000009a, 0x00001000}, 167 {0x0000009f, 0x00916a00} 168 }; 169 170 static const u32 cayman_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = { 171 {0x00000077, 0xff010100}, 172 {0x00000078, 0x00000000}, 173 {0x00000079, 0x00001434}, 174 {0x0000007a, 0xcc08ec08}, 175 {0x0000007b, 0x00040000}, 176 {0x0000007c, 0x000080c0}, 177 {0x0000007d, 0x09000000}, 178 {0x0000007e, 0x00210404}, 179 {0x00000081, 0x08a8e800}, 180 {0x00000082, 0x00030444}, 181 {0x00000083, 0x00000000}, 182 {0x00000085, 0x00000001}, 183 {0x00000086, 0x00000002}, 184 {0x00000087, 0x48490000}, 185 {0x00000088, 0x20244647}, 186 {0x00000089, 0x00000005}, 187 {0x0000008b, 0x66030000}, 188 {0x0000008c, 0x00006603}, 189 {0x0000008d, 0x00000100}, 190 {0x0000008f, 0x00001c0a}, 191 {0x00000090, 0xff000001}, 192 {0x00000094, 0x00101101}, 193 {0x00000095, 0x00000fff}, 194 {0x00000096, 0x00116fff}, 195 {0x00000097, 0x60010000}, 196 {0x00000098, 0x10010000}, 197 {0x00000099, 0x00006000}, 198 {0x0000009a, 0x00001000}, 199 {0x0000009f, 0x00976b00} 200 }; 201 202 int ni_mc_load_microcode(struct radeon_device *rdev) 203 { 204 const __be32 *fw_data; 205 u32 mem_type, running, blackout = 0; 206 u32 *io_mc_regs; 207 int i, ucode_size, regs_size; 208 209 if (!rdev->mc_fw) 210 return -EINVAL; 211 212 switch (rdev->family) { 213 case CHIP_BARTS: 214 io_mc_regs = (u32 *)&barts_io_mc_regs; 215 ucode_size = BTC_MC_UCODE_SIZE; 216 regs_size = BTC_IO_MC_REGS_SIZE; 217 break; 218 case CHIP_TURKS: 219 io_mc_regs = (u32 *)&turks_io_mc_regs; 220 ucode_size = BTC_MC_UCODE_SIZE; 221 regs_size = BTC_IO_MC_REGS_SIZE; 222 break; 223 case CHIP_CAICOS: 224 default: 225 io_mc_regs = (u32 *)&caicos_io_mc_regs; 226 ucode_size = BTC_MC_UCODE_SIZE; 227 regs_size = BTC_IO_MC_REGS_SIZE; 228 break; 229 case CHIP_CAYMAN: 230 io_mc_regs = (u32 *)&cayman_io_mc_regs; 231 ucode_size = CAYMAN_MC_UCODE_SIZE; 232 regs_size = BTC_IO_MC_REGS_SIZE; 233 break; 234 } 235 236 mem_type = (RREG32(MC_SEQ_MISC0) & MC_SEQ_MISC0_GDDR5_MASK) >> MC_SEQ_MISC0_GDDR5_SHIFT; 237 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK; 238 239 if ((mem_type == MC_SEQ_MISC0_GDDR5_VALUE) && (running == 0)) { 240 if (running) { 241 blackout = RREG32(MC_SHARED_BLACKOUT_CNTL); 242 WREG32(MC_SHARED_BLACKOUT_CNTL, 1); 243 } 244 245 /* reset the engine and set to writable */ 246 WREG32(MC_SEQ_SUP_CNTL, 0x00000008); 247 WREG32(MC_SEQ_SUP_CNTL, 0x00000010); 248 249 /* load mc io regs */ 250 for (i = 0; i < regs_size; i++) { 251 WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]); 252 WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]); 253 } 254 /* load the MC ucode */ 255 fw_data = (const __be32 *)rdev->mc_fw->data; 256 for (i = 0; i < ucode_size; i++) 257 WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++)); 258 259 /* put the engine back into the active state */ 260 WREG32(MC_SEQ_SUP_CNTL, 0x00000008); 261 WREG32(MC_SEQ_SUP_CNTL, 0x00000004); 262 WREG32(MC_SEQ_SUP_CNTL, 0x00000001); 263 264 /* wait for training to complete */ 265 for (i = 0; i < rdev->usec_timeout; i++) { 266 if (RREG32(MC_IO_PAD_CNTL_D0) & MEM_FALL_OUT_CMD) 267 break; 268 udelay(1); 269 } 270 271 if (running) 272 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout); 273 } 274 275 return 0; 276 } 277 278 int ni_init_microcode(struct radeon_device *rdev) 279 { 280 struct platform_device *pdev; 281 const char *chip_name; 282 const char *rlc_chip_name; 283 size_t pfp_req_size, me_req_size, rlc_req_size, mc_req_size; 284 char fw_name[30]; 285 int err; 286 287 DRM_DEBUG("\n"); 288 289 pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0); 290 err = IS_ERR(pdev); 291 if (err) { 292 printk(KERN_ERR "radeon_cp: Failed to register firmware\n"); 293 return -EINVAL; 294 } 295 296 switch (rdev->family) { 297 case CHIP_BARTS: 298 chip_name = "BARTS"; 299 rlc_chip_name = "BTC"; 300 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4; 301 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4; 302 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4; 303 mc_req_size = BTC_MC_UCODE_SIZE * 4; 304 break; 305 case CHIP_TURKS: 306 chip_name = "TURKS"; 307 rlc_chip_name = "BTC"; 308 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4; 309 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4; 310 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4; 311 mc_req_size = BTC_MC_UCODE_SIZE * 4; 312 break; 313 case CHIP_CAICOS: 314 chip_name = "CAICOS"; 315 rlc_chip_name = "BTC"; 316 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4; 317 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4; 318 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4; 319 mc_req_size = BTC_MC_UCODE_SIZE * 4; 320 break; 321 case CHIP_CAYMAN: 322 chip_name = "CAYMAN"; 323 rlc_chip_name = "CAYMAN"; 324 pfp_req_size = CAYMAN_PFP_UCODE_SIZE * 4; 325 me_req_size = CAYMAN_PM4_UCODE_SIZE * 4; 326 rlc_req_size = CAYMAN_RLC_UCODE_SIZE * 4; 327 mc_req_size = CAYMAN_MC_UCODE_SIZE * 4; 328 break; 329 default: BUG(); 330 } 331 332 DRM_INFO("Loading %s Microcode\n", chip_name); 333 334 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name); 335 err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev); 336 if (err) 337 goto out; 338 if (rdev->pfp_fw->size != pfp_req_size) { 339 printk(KERN_ERR 340 "ni_cp: Bogus length %zu in firmware \"%s\"\n", 341 rdev->pfp_fw->size, fw_name); 342 err = -EINVAL; 343 goto out; 344 } 345 346 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name); 347 err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev); 348 if (err) 349 goto out; 350 if (rdev->me_fw->size != me_req_size) { 351 printk(KERN_ERR 352 "ni_cp: Bogus length %zu in firmware \"%s\"\n", 353 rdev->me_fw->size, fw_name); 354 err = -EINVAL; 355 } 356 357 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name); 358 err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev); 359 if (err) 360 goto out; 361 if (rdev->rlc_fw->size != rlc_req_size) { 362 printk(KERN_ERR 363 "ni_rlc: Bogus length %zu in firmware \"%s\"\n", 364 rdev->rlc_fw->size, fw_name); 365 err = -EINVAL; 366 } 367 368 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name); 369 err = request_firmware(&rdev->mc_fw, fw_name, &pdev->dev); 370 if (err) 371 goto out; 372 if (rdev->mc_fw->size != mc_req_size) { 373 printk(KERN_ERR 374 "ni_mc: Bogus length %zu in firmware \"%s\"\n", 375 rdev->mc_fw->size, fw_name); 376 err = -EINVAL; 377 } 378 out: 379 platform_device_unregister(pdev); 380 381 if (err) { 382 if (err != -EINVAL) 383 printk(KERN_ERR 384 "ni_cp: Failed to load firmware \"%s\"\n", 385 fw_name); 386 release_firmware(rdev->pfp_fw); 387 rdev->pfp_fw = NULL; 388 release_firmware(rdev->me_fw); 389 rdev->me_fw = NULL; 390 release_firmware(rdev->rlc_fw); 391 rdev->rlc_fw = NULL; 392 release_firmware(rdev->mc_fw); 393 rdev->mc_fw = NULL; 394 } 395 return err; 396 } 397 398 /* 399 * Core functions 400 */ 401 static u32 cayman_get_tile_pipe_to_backend_map(struct radeon_device *rdev, 402 u32 num_tile_pipes, 403 u32 num_backends_per_asic, 404 u32 *backend_disable_mask_per_asic, 405 u32 num_shader_engines) 406 { 407 u32 backend_map = 0; 408 u32 enabled_backends_mask = 0; 409 u32 enabled_backends_count = 0; 410 u32 num_backends_per_se; 411 u32 cur_pipe; 412 u32 swizzle_pipe[CAYMAN_MAX_PIPES]; 413 u32 cur_backend = 0; 414 u32 i; 415 bool force_no_swizzle; 416 417 /* force legal values */ 418 if (num_tile_pipes < 1) 419 num_tile_pipes = 1; 420 if (num_tile_pipes > rdev->config.cayman.max_tile_pipes) 421 num_tile_pipes = rdev->config.cayman.max_tile_pipes; 422 if (num_shader_engines < 1) 423 num_shader_engines = 1; 424 if (num_shader_engines > rdev->config.cayman.max_shader_engines) 425 num_shader_engines = rdev->config.cayman.max_shader_engines; 426 if (num_backends_per_asic < num_shader_engines) 427 num_backends_per_asic = num_shader_engines; 428 if (num_backends_per_asic > (rdev->config.cayman.max_backends_per_se * num_shader_engines)) 429 num_backends_per_asic = rdev->config.cayman.max_backends_per_se * num_shader_engines; 430 431 /* make sure we have the same number of backends per se */ 432 num_backends_per_asic = ALIGN(num_backends_per_asic, num_shader_engines); 433 /* set up the number of backends per se */ 434 num_backends_per_se = num_backends_per_asic / num_shader_engines; 435 if (num_backends_per_se > rdev->config.cayman.max_backends_per_se) { 436 num_backends_per_se = rdev->config.cayman.max_backends_per_se; 437 num_backends_per_asic = num_backends_per_se * num_shader_engines; 438 } 439 440 /* create enable mask and count for enabled backends */ 441 for (i = 0; i < CAYMAN_MAX_BACKENDS; ++i) { 442 if (((*backend_disable_mask_per_asic >> i) & 1) == 0) { 443 enabled_backends_mask |= (1 << i); 444 ++enabled_backends_count; 445 } 446 if (enabled_backends_count == num_backends_per_asic) 447 break; 448 } 449 450 /* force the backends mask to match the current number of backends */ 451 if (enabled_backends_count != num_backends_per_asic) { 452 u32 this_backend_enabled; 453 u32 shader_engine; 454 u32 backend_per_se; 455 456 enabled_backends_mask = 0; 457 enabled_backends_count = 0; 458 *backend_disable_mask_per_asic = CAYMAN_MAX_BACKENDS_MASK; 459 for (i = 0; i < CAYMAN_MAX_BACKENDS; ++i) { 460 /* calc the current se */ 461 shader_engine = i / rdev->config.cayman.max_backends_per_se; 462 /* calc the backend per se */ 463 backend_per_se = i % rdev->config.cayman.max_backends_per_se; 464 /* default to not enabled */ 465 this_backend_enabled = 0; 466 if ((shader_engine < num_shader_engines) && 467 (backend_per_se < num_backends_per_se)) 468 this_backend_enabled = 1; 469 if (this_backend_enabled) { 470 enabled_backends_mask |= (1 << i); 471 *backend_disable_mask_per_asic &= ~(1 << i); 472 ++enabled_backends_count; 473 } 474 } 475 } 476 477 478 memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * CAYMAN_MAX_PIPES); 479 switch (rdev->family) { 480 case CHIP_CAYMAN: 481 force_no_swizzle = true; 482 break; 483 default: 484 force_no_swizzle = false; 485 break; 486 } 487 if (force_no_swizzle) { 488 bool last_backend_enabled = false; 489 490 force_no_swizzle = false; 491 for (i = 0; i < CAYMAN_MAX_BACKENDS; ++i) { 492 if (((enabled_backends_mask >> i) & 1) == 1) { 493 if (last_backend_enabled) 494 force_no_swizzle = true; 495 last_backend_enabled = true; 496 } else 497 last_backend_enabled = false; 498 } 499 } 500 501 switch (num_tile_pipes) { 502 case 1: 503 case 3: 504 case 5: 505 case 7: 506 DRM_ERROR("odd number of pipes!\n"); 507 break; 508 case 2: 509 swizzle_pipe[0] = 0; 510 swizzle_pipe[1] = 1; 511 break; 512 case 4: 513 if (force_no_swizzle) { 514 swizzle_pipe[0] = 0; 515 swizzle_pipe[1] = 1; 516 swizzle_pipe[2] = 2; 517 swizzle_pipe[3] = 3; 518 } else { 519 swizzle_pipe[0] = 0; 520 swizzle_pipe[1] = 2; 521 swizzle_pipe[2] = 1; 522 swizzle_pipe[3] = 3; 523 } 524 break; 525 case 6: 526 if (force_no_swizzle) { 527 swizzle_pipe[0] = 0; 528 swizzle_pipe[1] = 1; 529 swizzle_pipe[2] = 2; 530 swizzle_pipe[3] = 3; 531 swizzle_pipe[4] = 4; 532 swizzle_pipe[5] = 5; 533 } else { 534 swizzle_pipe[0] = 0; 535 swizzle_pipe[1] = 2; 536 swizzle_pipe[2] = 4; 537 swizzle_pipe[3] = 1; 538 swizzle_pipe[4] = 3; 539 swizzle_pipe[5] = 5; 540 } 541 break; 542 case 8: 543 if (force_no_swizzle) { 544 swizzle_pipe[0] = 0; 545 swizzle_pipe[1] = 1; 546 swizzle_pipe[2] = 2; 547 swizzle_pipe[3] = 3; 548 swizzle_pipe[4] = 4; 549 swizzle_pipe[5] = 5; 550 swizzle_pipe[6] = 6; 551 swizzle_pipe[7] = 7; 552 } else { 553 swizzle_pipe[0] = 0; 554 swizzle_pipe[1] = 2; 555 swizzle_pipe[2] = 4; 556 swizzle_pipe[3] = 6; 557 swizzle_pipe[4] = 1; 558 swizzle_pipe[5] = 3; 559 swizzle_pipe[6] = 5; 560 swizzle_pipe[7] = 7; 561 } 562 break; 563 } 564 565 for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) { 566 while (((1 << cur_backend) & enabled_backends_mask) == 0) 567 cur_backend = (cur_backend + 1) % CAYMAN_MAX_BACKENDS; 568 569 backend_map |= (((cur_backend & 0xf) << (swizzle_pipe[cur_pipe] * 4))); 570 571 cur_backend = (cur_backend + 1) % CAYMAN_MAX_BACKENDS; 572 } 573 574 return backend_map; 575 } 576 577 static u32 cayman_get_disable_mask_per_asic(struct radeon_device *rdev, 578 u32 disable_mask_per_se, 579 u32 max_disable_mask_per_se, 580 u32 num_shader_engines) 581 { 582 u32 disable_field_width_per_se = r600_count_pipe_bits(disable_mask_per_se); 583 u32 disable_mask_per_asic = disable_mask_per_se & max_disable_mask_per_se; 584 585 if (num_shader_engines == 1) 586 return disable_mask_per_asic; 587 else if (num_shader_engines == 2) 588 return disable_mask_per_asic | (disable_mask_per_asic << disable_field_width_per_se); 589 else 590 return 0xffffffff; 591 } 592 593 static void cayman_gpu_init(struct radeon_device *rdev) 594 { 595 u32 cc_rb_backend_disable = 0; 596 u32 cc_gc_shader_pipe_config; 597 u32 gb_addr_config = 0; 598 u32 mc_shared_chmap, mc_arb_ramcfg; 599 u32 gb_backend_map; 600 u32 cgts_tcc_disable; 601 u32 sx_debug_1; 602 u32 smx_dc_ctl0; 603 u32 gc_user_shader_pipe_config; 604 u32 gc_user_rb_backend_disable; 605 u32 cgts_user_tcc_disable; 606 u32 cgts_sm_ctrl_reg; 607 u32 hdp_host_path_cntl; 608 u32 tmp; 609 int i, j; 610 611 switch (rdev->family) { 612 case CHIP_CAYMAN: 613 default: 614 rdev->config.cayman.max_shader_engines = 2; 615 rdev->config.cayman.max_pipes_per_simd = 4; 616 rdev->config.cayman.max_tile_pipes = 8; 617 rdev->config.cayman.max_simds_per_se = 12; 618 rdev->config.cayman.max_backends_per_se = 4; 619 rdev->config.cayman.max_texture_channel_caches = 8; 620 rdev->config.cayman.max_gprs = 256; 621 rdev->config.cayman.max_threads = 256; 622 rdev->config.cayman.max_gs_threads = 32; 623 rdev->config.cayman.max_stack_entries = 512; 624 rdev->config.cayman.sx_num_of_sets = 8; 625 rdev->config.cayman.sx_max_export_size = 256; 626 rdev->config.cayman.sx_max_export_pos_size = 64; 627 rdev->config.cayman.sx_max_export_smx_size = 192; 628 rdev->config.cayman.max_hw_contexts = 8; 629 rdev->config.cayman.sq_num_cf_insts = 2; 630 631 rdev->config.cayman.sc_prim_fifo_size = 0x100; 632 rdev->config.cayman.sc_hiz_tile_fifo_size = 0x30; 633 rdev->config.cayman.sc_earlyz_tile_fifo_size = 0x130; 634 break; 635 } 636 637 /* Initialize HDP */ 638 for (i = 0, j = 0; i < 32; i++, j += 0x18) { 639 WREG32((0x2c14 + j), 0x00000000); 640 WREG32((0x2c18 + j), 0x00000000); 641 WREG32((0x2c1c + j), 0x00000000); 642 WREG32((0x2c20 + j), 0x00000000); 643 WREG32((0x2c24 + j), 0x00000000); 644 } 645 646 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff)); 647 648 evergreen_fix_pci_max_read_req_size(rdev); 649 650 mc_shared_chmap = RREG32(MC_SHARED_CHMAP); 651 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG); 652 653 cc_rb_backend_disable = RREG32(CC_RB_BACKEND_DISABLE); 654 cc_gc_shader_pipe_config = RREG32(CC_GC_SHADER_PIPE_CONFIG); 655 cgts_tcc_disable = 0xff000000; 656 gc_user_rb_backend_disable = RREG32(GC_USER_RB_BACKEND_DISABLE); 657 gc_user_shader_pipe_config = RREG32(GC_USER_SHADER_PIPE_CONFIG); 658 cgts_user_tcc_disable = RREG32(CGTS_USER_TCC_DISABLE); 659 660 rdev->config.cayman.num_shader_engines = rdev->config.cayman.max_shader_engines; 661 tmp = ((~gc_user_shader_pipe_config) & INACTIVE_QD_PIPES_MASK) >> INACTIVE_QD_PIPES_SHIFT; 662 rdev->config.cayman.num_shader_pipes_per_simd = r600_count_pipe_bits(tmp); 663 rdev->config.cayman.num_tile_pipes = rdev->config.cayman.max_tile_pipes; 664 tmp = ((~gc_user_shader_pipe_config) & INACTIVE_SIMDS_MASK) >> INACTIVE_SIMDS_SHIFT; 665 rdev->config.cayman.num_simds_per_se = r600_count_pipe_bits(tmp); 666 tmp = ((~gc_user_rb_backend_disable) & BACKEND_DISABLE_MASK) >> BACKEND_DISABLE_SHIFT; 667 rdev->config.cayman.num_backends_per_se = r600_count_pipe_bits(tmp); 668 tmp = (gc_user_rb_backend_disable & BACKEND_DISABLE_MASK) >> BACKEND_DISABLE_SHIFT; 669 rdev->config.cayman.backend_disable_mask_per_asic = 670 cayman_get_disable_mask_per_asic(rdev, tmp, CAYMAN_MAX_BACKENDS_PER_SE_MASK, 671 rdev->config.cayman.num_shader_engines); 672 rdev->config.cayman.backend_map = 673 cayman_get_tile_pipe_to_backend_map(rdev, rdev->config.cayman.num_tile_pipes, 674 rdev->config.cayman.num_backends_per_se * 675 rdev->config.cayman.num_shader_engines, 676 &rdev->config.cayman.backend_disable_mask_per_asic, 677 rdev->config.cayman.num_shader_engines); 678 tmp = ((~cgts_user_tcc_disable) & TCC_DISABLE_MASK) >> TCC_DISABLE_SHIFT; 679 rdev->config.cayman.num_texture_channel_caches = r600_count_pipe_bits(tmp); 680 tmp = (mc_arb_ramcfg & BURSTLENGTH_MASK) >> BURSTLENGTH_SHIFT; 681 rdev->config.cayman.mem_max_burst_length_bytes = (tmp + 1) * 256; 682 if (rdev->config.cayman.mem_max_burst_length_bytes > 512) 683 rdev->config.cayman.mem_max_burst_length_bytes = 512; 684 tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT; 685 rdev->config.cayman.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; 686 if (rdev->config.cayman.mem_row_size_in_kb > 4) 687 rdev->config.cayman.mem_row_size_in_kb = 4; 688 /* XXX use MC settings? */ 689 rdev->config.cayman.shader_engine_tile_size = 32; 690 rdev->config.cayman.num_gpus = 1; 691 rdev->config.cayman.multi_gpu_tile_size = 64; 692 693 //gb_addr_config = 0x02011003 694 #if 0 695 gb_addr_config = RREG32(GB_ADDR_CONFIG); 696 #else 697 gb_addr_config = 0; 698 switch (rdev->config.cayman.num_tile_pipes) { 699 case 1: 700 default: 701 gb_addr_config |= NUM_PIPES(0); 702 break; 703 case 2: 704 gb_addr_config |= NUM_PIPES(1); 705 break; 706 case 4: 707 gb_addr_config |= NUM_PIPES(2); 708 break; 709 case 8: 710 gb_addr_config |= NUM_PIPES(3); 711 break; 712 } 713 714 tmp = (rdev->config.cayman.mem_max_burst_length_bytes / 256) - 1; 715 gb_addr_config |= PIPE_INTERLEAVE_SIZE(tmp); 716 gb_addr_config |= NUM_SHADER_ENGINES(rdev->config.cayman.num_shader_engines - 1); 717 tmp = (rdev->config.cayman.shader_engine_tile_size / 16) - 1; 718 gb_addr_config |= SHADER_ENGINE_TILE_SIZE(tmp); 719 switch (rdev->config.cayman.num_gpus) { 720 case 1: 721 default: 722 gb_addr_config |= NUM_GPUS(0); 723 break; 724 case 2: 725 gb_addr_config |= NUM_GPUS(1); 726 break; 727 case 4: 728 gb_addr_config |= NUM_GPUS(2); 729 break; 730 } 731 switch (rdev->config.cayman.multi_gpu_tile_size) { 732 case 16: 733 gb_addr_config |= MULTI_GPU_TILE_SIZE(0); 734 break; 735 case 32: 736 default: 737 gb_addr_config |= MULTI_GPU_TILE_SIZE(1); 738 break; 739 case 64: 740 gb_addr_config |= MULTI_GPU_TILE_SIZE(2); 741 break; 742 case 128: 743 gb_addr_config |= MULTI_GPU_TILE_SIZE(3); 744 break; 745 } 746 switch (rdev->config.cayman.mem_row_size_in_kb) { 747 case 1: 748 default: 749 gb_addr_config |= ROW_SIZE(0); 750 break; 751 case 2: 752 gb_addr_config |= ROW_SIZE(1); 753 break; 754 case 4: 755 gb_addr_config |= ROW_SIZE(2); 756 break; 757 } 758 #endif 759 760 tmp = (gb_addr_config & NUM_PIPES_MASK) >> NUM_PIPES_SHIFT; 761 rdev->config.cayman.num_tile_pipes = (1 << tmp); 762 tmp = (gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT; 763 rdev->config.cayman.mem_max_burst_length_bytes = (tmp + 1) * 256; 764 tmp = (gb_addr_config & NUM_SHADER_ENGINES_MASK) >> NUM_SHADER_ENGINES_SHIFT; 765 rdev->config.cayman.num_shader_engines = tmp + 1; 766 tmp = (gb_addr_config & NUM_GPUS_MASK) >> NUM_GPUS_SHIFT; 767 rdev->config.cayman.num_gpus = tmp + 1; 768 tmp = (gb_addr_config & MULTI_GPU_TILE_SIZE_MASK) >> MULTI_GPU_TILE_SIZE_SHIFT; 769 rdev->config.cayman.multi_gpu_tile_size = 1 << tmp; 770 tmp = (gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT; 771 rdev->config.cayman.mem_row_size_in_kb = 1 << tmp; 772 773 //gb_backend_map = 0x76541032; 774 #if 0 775 gb_backend_map = RREG32(GB_BACKEND_MAP); 776 #else 777 gb_backend_map = 778 cayman_get_tile_pipe_to_backend_map(rdev, rdev->config.cayman.num_tile_pipes, 779 rdev->config.cayman.num_backends_per_se * 780 rdev->config.cayman.num_shader_engines, 781 &rdev->config.cayman.backend_disable_mask_per_asic, 782 rdev->config.cayman.num_shader_engines); 783 #endif 784 /* setup tiling info dword. gb_addr_config is not adequate since it does 785 * not have bank info, so create a custom tiling dword. 786 * bits 3:0 num_pipes 787 * bits 7:4 num_banks 788 * bits 11:8 group_size 789 * bits 15:12 row_size 790 */ 791 rdev->config.cayman.tile_config = 0; 792 switch (rdev->config.cayman.num_tile_pipes) { 793 case 1: 794 default: 795 rdev->config.cayman.tile_config |= (0 << 0); 796 break; 797 case 2: 798 rdev->config.cayman.tile_config |= (1 << 0); 799 break; 800 case 4: 801 rdev->config.cayman.tile_config |= (2 << 0); 802 break; 803 case 8: 804 rdev->config.cayman.tile_config |= (3 << 0); 805 break; 806 } 807 rdev->config.cayman.tile_config |= 808 ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) << 4; 809 rdev->config.cayman.tile_config |= 810 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8; 811 rdev->config.cayman.tile_config |= 812 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12; 813 814 rdev->config.cayman.backend_map = gb_backend_map; 815 WREG32(GB_BACKEND_MAP, gb_backend_map); 816 WREG32(GB_ADDR_CONFIG, gb_addr_config); 817 WREG32(DMIF_ADDR_CONFIG, gb_addr_config); 818 WREG32(HDP_ADDR_CONFIG, gb_addr_config); 819 820 /* primary versions */ 821 WREG32(CC_RB_BACKEND_DISABLE, cc_rb_backend_disable); 822 WREG32(CC_SYS_RB_BACKEND_DISABLE, cc_rb_backend_disable); 823 WREG32(CC_GC_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config); 824 825 WREG32(CGTS_TCC_DISABLE, cgts_tcc_disable); 826 WREG32(CGTS_SYS_TCC_DISABLE, cgts_tcc_disable); 827 828 /* user versions */ 829 WREG32(GC_USER_RB_BACKEND_DISABLE, cc_rb_backend_disable); 830 WREG32(GC_USER_SYS_RB_BACKEND_DISABLE, cc_rb_backend_disable); 831 WREG32(GC_USER_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config); 832 833 WREG32(CGTS_USER_SYS_TCC_DISABLE, cgts_tcc_disable); 834 WREG32(CGTS_USER_TCC_DISABLE, cgts_tcc_disable); 835 836 /* reprogram the shader complex */ 837 cgts_sm_ctrl_reg = RREG32(CGTS_SM_CTRL_REG); 838 for (i = 0; i < 16; i++) 839 WREG32(CGTS_SM_CTRL_REG, OVERRIDE); 840 WREG32(CGTS_SM_CTRL_REG, cgts_sm_ctrl_reg); 841 842 /* set HW defaults for 3D engine */ 843 WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60)); 844 845 sx_debug_1 = RREG32(SX_DEBUG_1); 846 sx_debug_1 |= ENABLE_NEW_SMX_ADDRESS; 847 WREG32(SX_DEBUG_1, sx_debug_1); 848 849 smx_dc_ctl0 = RREG32(SMX_DC_CTL0); 850 smx_dc_ctl0 &= ~NUMBER_OF_SETS(0x1ff); 851 smx_dc_ctl0 |= NUMBER_OF_SETS(rdev->config.cayman.sx_num_of_sets); 852 WREG32(SMX_DC_CTL0, smx_dc_ctl0); 853 854 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4) | CRC_SIMD_ID_WADDR_DISABLE); 855 856 /* need to be explicitly zero-ed */ 857 WREG32(VGT_OFFCHIP_LDS_BASE, 0); 858 WREG32(SQ_LSTMP_RING_BASE, 0); 859 WREG32(SQ_HSTMP_RING_BASE, 0); 860 WREG32(SQ_ESTMP_RING_BASE, 0); 861 WREG32(SQ_GSTMP_RING_BASE, 0); 862 WREG32(SQ_VSTMP_RING_BASE, 0); 863 WREG32(SQ_PSTMP_RING_BASE, 0); 864 865 WREG32(TA_CNTL_AUX, DISABLE_CUBE_ANISO); 866 867 WREG32(SX_EXPORT_BUFFER_SIZES, (COLOR_BUFFER_SIZE((rdev->config.cayman.sx_max_export_size / 4) - 1) | 868 POSITION_BUFFER_SIZE((rdev->config.cayman.sx_max_export_pos_size / 4) - 1) | 869 SMX_BUFFER_SIZE((rdev->config.cayman.sx_max_export_smx_size / 4) - 1))); 870 871 WREG32(PA_SC_FIFO_SIZE, (SC_PRIM_FIFO_SIZE(rdev->config.cayman.sc_prim_fifo_size) | 872 SC_HIZ_TILE_FIFO_SIZE(rdev->config.cayman.sc_hiz_tile_fifo_size) | 873 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.cayman.sc_earlyz_tile_fifo_size))); 874 875 876 WREG32(VGT_NUM_INSTANCES, 1); 877 878 WREG32(CP_PERFMON_CNTL, 0); 879 880 WREG32(SQ_MS_FIFO_SIZES, (CACHE_FIFO_SIZE(16 * rdev->config.cayman.sq_num_cf_insts) | 881 FETCH_FIFO_HIWATER(0x4) | 882 DONE_FIFO_HIWATER(0xe0) | 883 ALU_UPDATE_FIFO_HIWATER(0x8))); 884 885 WREG32(SQ_GPR_RESOURCE_MGMT_1, NUM_CLAUSE_TEMP_GPRS(4)); 886 WREG32(SQ_CONFIG, (VC_ENABLE | 887 EXPORT_SRC_C | 888 GFX_PRIO(0) | 889 CS1_PRIO(0) | 890 CS2_PRIO(1))); 891 WREG32(SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, DYN_GPR_ENABLE); 892 893 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) | 894 FORCE_EOV_MAX_REZ_CNT(255))); 895 896 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) | 897 AUTO_INVLD_EN(ES_AND_GS_AUTO)); 898 899 WREG32(VGT_GS_VERTEX_REUSE, 16); 900 WREG32(PA_SC_LINE_STIPPLE_STATE, 0); 901 902 WREG32(CB_PERF_CTR0_SEL_0, 0); 903 WREG32(CB_PERF_CTR0_SEL_1, 0); 904 WREG32(CB_PERF_CTR1_SEL_0, 0); 905 WREG32(CB_PERF_CTR1_SEL_1, 0); 906 WREG32(CB_PERF_CTR2_SEL_0, 0); 907 WREG32(CB_PERF_CTR2_SEL_1, 0); 908 WREG32(CB_PERF_CTR3_SEL_0, 0); 909 WREG32(CB_PERF_CTR3_SEL_1, 0); 910 911 tmp = RREG32(HDP_MISC_CNTL); 912 tmp |= HDP_FLUSH_INVALIDATE_CACHE; 913 WREG32(HDP_MISC_CNTL, tmp); 914 915 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL); 916 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl); 917 918 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3)); 919 920 udelay(50); 921 } 922 923 /* 924 * GART 925 */ 926 void cayman_pcie_gart_tlb_flush(struct radeon_device *rdev) 927 { 928 /* flush hdp cache */ 929 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1); 930 931 /* bits 0-7 are the VM contexts0-7 */ 932 WREG32(VM_INVALIDATE_REQUEST, 1); 933 } 934 935 int cayman_pcie_gart_enable(struct radeon_device *rdev) 936 { 937 int r; 938 939 if (rdev->gart.robj == NULL) { 940 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n"); 941 return -EINVAL; 942 } 943 r = radeon_gart_table_vram_pin(rdev); 944 if (r) 945 return r; 946 radeon_gart_restore(rdev); 947 /* Setup TLB control */ 948 WREG32(MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB | 949 ENABLE_L1_FRAGMENT_PROCESSING | 950 SYSTEM_ACCESS_MODE_NOT_IN_SYS | 951 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU); 952 /* Setup L2 cache */ 953 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | 954 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE | 955 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE | 956 EFFECTIVE_L2_QUEUE_SIZE(7) | 957 CONTEXT1_IDENTITY_ACCESS_MODE(1)); 958 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE); 959 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY | 960 L2_CACHE_BIGK_FRAGMENT_SIZE(6)); 961 /* setup context0 */ 962 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12); 963 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12); 964 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12); 965 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR, 966 (u32)(rdev->dummy_page.addr >> 12)); 967 WREG32(VM_CONTEXT0_CNTL2, 0); 968 WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) | 969 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT); 970 /* disable context1-7 */ 971 WREG32(VM_CONTEXT1_CNTL2, 0); 972 WREG32(VM_CONTEXT1_CNTL, 0); 973 974 cayman_pcie_gart_tlb_flush(rdev); 975 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", 976 (unsigned)(rdev->mc.gtt_size >> 20), 977 (unsigned long long)rdev->gart.table_addr); 978 rdev->gart.ready = true; 979 return 0; 980 } 981 982 void cayman_pcie_gart_disable(struct radeon_device *rdev) 983 { 984 /* Disable all tables */ 985 WREG32(VM_CONTEXT0_CNTL, 0); 986 WREG32(VM_CONTEXT1_CNTL, 0); 987 /* Setup TLB control */ 988 WREG32(MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING | 989 SYSTEM_ACCESS_MODE_NOT_IN_SYS | 990 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU); 991 /* Setup L2 cache */ 992 WREG32(VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE | 993 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE | 994 EFFECTIVE_L2_QUEUE_SIZE(7) | 995 CONTEXT1_IDENTITY_ACCESS_MODE(1)); 996 WREG32(VM_L2_CNTL2, 0); 997 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY | 998 L2_CACHE_BIGK_FRAGMENT_SIZE(6)); 999 radeon_gart_table_vram_unpin(rdev); 1000 } 1001 1002 void cayman_pcie_gart_fini(struct radeon_device *rdev) 1003 { 1004 cayman_pcie_gart_disable(rdev); 1005 radeon_gart_table_vram_free(rdev); 1006 radeon_gart_fini(rdev); 1007 } 1008 1009 /* 1010 * CP. 1011 */ 1012 static void cayman_cp_enable(struct radeon_device *rdev, bool enable) 1013 { 1014 if (enable) 1015 WREG32(CP_ME_CNTL, 0); 1016 else { 1017 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size); 1018 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT)); 1019 WREG32(SCRATCH_UMSK, 0); 1020 } 1021 } 1022 1023 static int cayman_cp_load_microcode(struct radeon_device *rdev) 1024 { 1025 const __be32 *fw_data; 1026 int i; 1027 1028 if (!rdev->me_fw || !rdev->pfp_fw) 1029 return -EINVAL; 1030 1031 cayman_cp_enable(rdev, false); 1032 1033 fw_data = (const __be32 *)rdev->pfp_fw->data; 1034 WREG32(CP_PFP_UCODE_ADDR, 0); 1035 for (i = 0; i < CAYMAN_PFP_UCODE_SIZE; i++) 1036 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++)); 1037 WREG32(CP_PFP_UCODE_ADDR, 0); 1038 1039 fw_data = (const __be32 *)rdev->me_fw->data; 1040 WREG32(CP_ME_RAM_WADDR, 0); 1041 for (i = 0; i < CAYMAN_PM4_UCODE_SIZE; i++) 1042 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++)); 1043 1044 WREG32(CP_PFP_UCODE_ADDR, 0); 1045 WREG32(CP_ME_RAM_WADDR, 0); 1046 WREG32(CP_ME_RAM_RADDR, 0); 1047 return 0; 1048 } 1049 1050 static int cayman_cp_start(struct radeon_device *rdev) 1051 { 1052 struct radeon_cp *cp = &rdev->cp[RADEON_RING_TYPE_GFX_INDEX]; 1053 int r, i; 1054 1055 r = radeon_ring_lock(rdev, cp, 7); 1056 if (r) { 1057 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r); 1058 return r; 1059 } 1060 radeon_ring_write(cp, PACKET3(PACKET3_ME_INITIALIZE, 5)); 1061 radeon_ring_write(cp, 0x1); 1062 radeon_ring_write(cp, 0x0); 1063 radeon_ring_write(cp, rdev->config.cayman.max_hw_contexts - 1); 1064 radeon_ring_write(cp, PACKET3_ME_INITIALIZE_DEVICE_ID(1)); 1065 radeon_ring_write(cp, 0); 1066 radeon_ring_write(cp, 0); 1067 radeon_ring_unlock_commit(rdev, cp); 1068 1069 cayman_cp_enable(rdev, true); 1070 1071 r = radeon_ring_lock(rdev, cp, cayman_default_size + 19); 1072 if (r) { 1073 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r); 1074 return r; 1075 } 1076 1077 /* setup clear context state */ 1078 radeon_ring_write(cp, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 1079 radeon_ring_write(cp, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 1080 1081 for (i = 0; i < cayman_default_size; i++) 1082 radeon_ring_write(cp, cayman_default_state[i]); 1083 1084 radeon_ring_write(cp, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 1085 radeon_ring_write(cp, PACKET3_PREAMBLE_END_CLEAR_STATE); 1086 1087 /* set clear context state */ 1088 radeon_ring_write(cp, PACKET3(PACKET3_CLEAR_STATE, 0)); 1089 radeon_ring_write(cp, 0); 1090 1091 /* SQ_VTX_BASE_VTX_LOC */ 1092 radeon_ring_write(cp, 0xc0026f00); 1093 radeon_ring_write(cp, 0x00000000); 1094 radeon_ring_write(cp, 0x00000000); 1095 radeon_ring_write(cp, 0x00000000); 1096 1097 /* Clear consts */ 1098 radeon_ring_write(cp, 0xc0036f00); 1099 radeon_ring_write(cp, 0x00000bc4); 1100 radeon_ring_write(cp, 0xffffffff); 1101 radeon_ring_write(cp, 0xffffffff); 1102 radeon_ring_write(cp, 0xffffffff); 1103 1104 radeon_ring_write(cp, 0xc0026900); 1105 radeon_ring_write(cp, 0x00000316); 1106 radeon_ring_write(cp, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */ 1107 radeon_ring_write(cp, 0x00000010); /* */ 1108 1109 radeon_ring_unlock_commit(rdev, cp); 1110 1111 /* XXX init other rings */ 1112 1113 return 0; 1114 } 1115 1116 static void cayman_cp_fini(struct radeon_device *rdev) 1117 { 1118 cayman_cp_enable(rdev, false); 1119 radeon_ring_fini(rdev, &rdev->cp[RADEON_RING_TYPE_GFX_INDEX]); 1120 } 1121 1122 int cayman_cp_resume(struct radeon_device *rdev) 1123 { 1124 struct radeon_cp *cp; 1125 u32 tmp; 1126 u32 rb_bufsz; 1127 int r; 1128 1129 /* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */ 1130 WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP | 1131 SOFT_RESET_PA | 1132 SOFT_RESET_SH | 1133 SOFT_RESET_VGT | 1134 SOFT_RESET_SPI | 1135 SOFT_RESET_SX)); 1136 RREG32(GRBM_SOFT_RESET); 1137 mdelay(15); 1138 WREG32(GRBM_SOFT_RESET, 0); 1139 RREG32(GRBM_SOFT_RESET); 1140 1141 WREG32(CP_SEM_WAIT_TIMER, 0x0); 1142 1143 /* Set the write pointer delay */ 1144 WREG32(CP_RB_WPTR_DELAY, 0); 1145 1146 WREG32(CP_DEBUG, (1 << 27)); 1147 1148 /* ring 0 - compute and gfx */ 1149 /* Set ring buffer size */ 1150 cp = &rdev->cp[RADEON_RING_TYPE_GFX_INDEX]; 1151 rb_bufsz = drm_order(cp->ring_size / 8); 1152 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 1153 #ifdef __BIG_ENDIAN 1154 tmp |= BUF_SWAP_32BIT; 1155 #endif 1156 WREG32(CP_RB0_CNTL, tmp); 1157 1158 /* Initialize the ring buffer's read and write pointers */ 1159 WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA); 1160 cp->wptr = 0; 1161 WREG32(CP_RB0_WPTR, cp->wptr); 1162 1163 /* set the wb address wether it's enabled or not */ 1164 WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC); 1165 WREG32(CP_RB0_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF); 1166 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF); 1167 1168 if (rdev->wb.enabled) 1169 WREG32(SCRATCH_UMSK, 0xff); 1170 else { 1171 tmp |= RB_NO_UPDATE; 1172 WREG32(SCRATCH_UMSK, 0); 1173 } 1174 1175 mdelay(1); 1176 WREG32(CP_RB0_CNTL, tmp); 1177 1178 WREG32(CP_RB0_BASE, cp->gpu_addr >> 8); 1179 1180 cp->rptr = RREG32(CP_RB0_RPTR); 1181 1182 /* ring1 - compute only */ 1183 /* Set ring buffer size */ 1184 cp = &rdev->cp[CAYMAN_RING_TYPE_CP1_INDEX]; 1185 rb_bufsz = drm_order(cp->ring_size / 8); 1186 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 1187 #ifdef __BIG_ENDIAN 1188 tmp |= BUF_SWAP_32BIT; 1189 #endif 1190 WREG32(CP_RB1_CNTL, tmp); 1191 1192 /* Initialize the ring buffer's read and write pointers */ 1193 WREG32(CP_RB1_CNTL, tmp | RB_RPTR_WR_ENA); 1194 cp->wptr = 0; 1195 WREG32(CP_RB1_WPTR, cp->wptr); 1196 1197 /* set the wb address wether it's enabled or not */ 1198 WREG32(CP_RB1_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFFFFFFFC); 1199 WREG32(CP_RB1_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFF); 1200 1201 mdelay(1); 1202 WREG32(CP_RB1_CNTL, tmp); 1203 1204 WREG32(CP_RB1_BASE, cp->gpu_addr >> 8); 1205 1206 cp->rptr = RREG32(CP_RB1_RPTR); 1207 1208 /* ring2 - compute only */ 1209 /* Set ring buffer size */ 1210 cp = &rdev->cp[CAYMAN_RING_TYPE_CP2_INDEX]; 1211 rb_bufsz = drm_order(cp->ring_size / 8); 1212 tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 1213 #ifdef __BIG_ENDIAN 1214 tmp |= BUF_SWAP_32BIT; 1215 #endif 1216 WREG32(CP_RB2_CNTL, tmp); 1217 1218 /* Initialize the ring buffer's read and write pointers */ 1219 WREG32(CP_RB2_CNTL, tmp | RB_RPTR_WR_ENA); 1220 cp->wptr = 0; 1221 WREG32(CP_RB2_WPTR, cp->wptr); 1222 1223 /* set the wb address wether it's enabled or not */ 1224 WREG32(CP_RB2_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFFFFFFFC); 1225 WREG32(CP_RB2_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFF); 1226 1227 mdelay(1); 1228 WREG32(CP_RB2_CNTL, tmp); 1229 1230 WREG32(CP_RB2_BASE, cp->gpu_addr >> 8); 1231 1232 cp->rptr = RREG32(CP_RB2_RPTR); 1233 1234 /* start the rings */ 1235 cayman_cp_start(rdev); 1236 rdev->cp[RADEON_RING_TYPE_GFX_INDEX].ready = true; 1237 rdev->cp[CAYMAN_RING_TYPE_CP1_INDEX].ready = false; 1238 rdev->cp[CAYMAN_RING_TYPE_CP2_INDEX].ready = false; 1239 /* this only test cp0 */ 1240 r = radeon_ring_test(rdev, &rdev->cp[RADEON_RING_TYPE_GFX_INDEX]); 1241 if (r) { 1242 rdev->cp[RADEON_RING_TYPE_GFX_INDEX].ready = false; 1243 rdev->cp[CAYMAN_RING_TYPE_CP1_INDEX].ready = false; 1244 rdev->cp[CAYMAN_RING_TYPE_CP2_INDEX].ready = false; 1245 return r; 1246 } 1247 1248 return 0; 1249 } 1250 1251 bool cayman_gpu_is_lockup(struct radeon_device *rdev, struct radeon_cp *cp) 1252 { 1253 u32 srbm_status; 1254 u32 grbm_status; 1255 u32 grbm_status_se0, grbm_status_se1; 1256 struct r100_gpu_lockup *lockup = &rdev->config.cayman.lockup; 1257 int r; 1258 1259 srbm_status = RREG32(SRBM_STATUS); 1260 grbm_status = RREG32(GRBM_STATUS); 1261 grbm_status_se0 = RREG32(GRBM_STATUS_SE0); 1262 grbm_status_se1 = RREG32(GRBM_STATUS_SE1); 1263 if (!(grbm_status & GUI_ACTIVE)) { 1264 r100_gpu_lockup_update(lockup, cp); 1265 return false; 1266 } 1267 /* force CP activities */ 1268 r = radeon_ring_lock(rdev, cp, 2); 1269 if (!r) { 1270 /* PACKET2 NOP */ 1271 radeon_ring_write(cp, 0x80000000); 1272 radeon_ring_write(cp, 0x80000000); 1273 radeon_ring_unlock_commit(rdev, cp); 1274 } 1275 /* XXX deal with CP0,1,2 */ 1276 cp->rptr = RREG32(cp->rptr_reg); 1277 return r100_gpu_cp_is_lockup(rdev, lockup, cp); 1278 } 1279 1280 static int cayman_gpu_soft_reset(struct radeon_device *rdev) 1281 { 1282 struct evergreen_mc_save save; 1283 u32 grbm_reset = 0; 1284 1285 if (!(RREG32(GRBM_STATUS) & GUI_ACTIVE)) 1286 return 0; 1287 1288 dev_info(rdev->dev, "GPU softreset \n"); 1289 dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n", 1290 RREG32(GRBM_STATUS)); 1291 dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n", 1292 RREG32(GRBM_STATUS_SE0)); 1293 dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n", 1294 RREG32(GRBM_STATUS_SE1)); 1295 dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n", 1296 RREG32(SRBM_STATUS)); 1297 evergreen_mc_stop(rdev, &save); 1298 if (evergreen_mc_wait_for_idle(rdev)) { 1299 dev_warn(rdev->dev, "Wait for MC idle timedout !\n"); 1300 } 1301 /* Disable CP parsing/prefetching */ 1302 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT); 1303 1304 /* reset all the gfx blocks */ 1305 grbm_reset = (SOFT_RESET_CP | 1306 SOFT_RESET_CB | 1307 SOFT_RESET_DB | 1308 SOFT_RESET_GDS | 1309 SOFT_RESET_PA | 1310 SOFT_RESET_SC | 1311 SOFT_RESET_SPI | 1312 SOFT_RESET_SH | 1313 SOFT_RESET_SX | 1314 SOFT_RESET_TC | 1315 SOFT_RESET_TA | 1316 SOFT_RESET_VGT | 1317 SOFT_RESET_IA); 1318 1319 dev_info(rdev->dev, " GRBM_SOFT_RESET=0x%08X\n", grbm_reset); 1320 WREG32(GRBM_SOFT_RESET, grbm_reset); 1321 (void)RREG32(GRBM_SOFT_RESET); 1322 udelay(50); 1323 WREG32(GRBM_SOFT_RESET, 0); 1324 (void)RREG32(GRBM_SOFT_RESET); 1325 /* Wait a little for things to settle down */ 1326 udelay(50); 1327 dev_info(rdev->dev, " GRBM_STATUS=0x%08X\n", 1328 RREG32(GRBM_STATUS)); 1329 dev_info(rdev->dev, " GRBM_STATUS_SE0=0x%08X\n", 1330 RREG32(GRBM_STATUS_SE0)); 1331 dev_info(rdev->dev, " GRBM_STATUS_SE1=0x%08X\n", 1332 RREG32(GRBM_STATUS_SE1)); 1333 dev_info(rdev->dev, " SRBM_STATUS=0x%08X\n", 1334 RREG32(SRBM_STATUS)); 1335 evergreen_mc_resume(rdev, &save); 1336 return 0; 1337 } 1338 1339 int cayman_asic_reset(struct radeon_device *rdev) 1340 { 1341 return cayman_gpu_soft_reset(rdev); 1342 } 1343 1344 static int cayman_startup(struct radeon_device *rdev) 1345 { 1346 struct radeon_cp *cp = &rdev->cp[RADEON_RING_TYPE_GFX_INDEX]; 1347 int r; 1348 1349 /* enable pcie gen2 link */ 1350 evergreen_pcie_gen2_enable(rdev); 1351 1352 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw || !rdev->mc_fw) { 1353 r = ni_init_microcode(rdev); 1354 if (r) { 1355 DRM_ERROR("Failed to load firmware!\n"); 1356 return r; 1357 } 1358 } 1359 r = ni_mc_load_microcode(rdev); 1360 if (r) { 1361 DRM_ERROR("Failed to load MC firmware!\n"); 1362 return r; 1363 } 1364 1365 r = r600_vram_scratch_init(rdev); 1366 if (r) 1367 return r; 1368 1369 evergreen_mc_program(rdev); 1370 r = cayman_pcie_gart_enable(rdev); 1371 if (r) 1372 return r; 1373 cayman_gpu_init(rdev); 1374 1375 r = evergreen_blit_init(rdev); 1376 if (r) { 1377 r600_blit_fini(rdev); 1378 rdev->asic->copy = NULL; 1379 dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r); 1380 } 1381 1382 /* allocate wb buffer */ 1383 r = radeon_wb_init(rdev); 1384 if (r) 1385 return r; 1386 1387 /* Enable IRQ */ 1388 r = r600_irq_init(rdev); 1389 if (r) { 1390 DRM_ERROR("radeon: IH init failed (%d).\n", r); 1391 radeon_irq_kms_fini(rdev); 1392 return r; 1393 } 1394 evergreen_irq_set(rdev); 1395 1396 r = radeon_ring_init(rdev, cp, cp->ring_size, RADEON_WB_CP_RPTR_OFFSET, 1397 CP_RB0_RPTR, CP_RB0_WPTR); 1398 if (r) 1399 return r; 1400 r = cayman_cp_load_microcode(rdev); 1401 if (r) 1402 return r; 1403 r = cayman_cp_resume(rdev); 1404 if (r) 1405 return r; 1406 1407 return 0; 1408 } 1409 1410 int cayman_resume(struct radeon_device *rdev) 1411 { 1412 int r; 1413 1414 /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw, 1415 * posting will perform necessary task to bring back GPU into good 1416 * shape. 1417 */ 1418 /* post card */ 1419 atom_asic_init(rdev->mode_info.atom_context); 1420 1421 r = cayman_startup(rdev); 1422 if (r) { 1423 DRM_ERROR("cayman startup failed on resume\n"); 1424 return r; 1425 } 1426 1427 r = r600_ib_test(rdev, RADEON_RING_TYPE_GFX_INDEX); 1428 if (r) { 1429 DRM_ERROR("radeon: failled testing IB (%d).\n", r); 1430 return r; 1431 } 1432 1433 return r; 1434 1435 } 1436 1437 int cayman_suspend(struct radeon_device *rdev) 1438 { 1439 /* FIXME: we should wait for ring to be empty */ 1440 cayman_cp_enable(rdev, false); 1441 rdev->cp[RADEON_RING_TYPE_GFX_INDEX].ready = false; 1442 evergreen_irq_suspend(rdev); 1443 radeon_wb_disable(rdev); 1444 cayman_pcie_gart_disable(rdev); 1445 r600_blit_suspend(rdev); 1446 1447 return 0; 1448 } 1449 1450 /* Plan is to move initialization in that function and use 1451 * helper function so that radeon_device_init pretty much 1452 * do nothing more than calling asic specific function. This 1453 * should also allow to remove a bunch of callback function 1454 * like vram_info. 1455 */ 1456 int cayman_init(struct radeon_device *rdev) 1457 { 1458 struct radeon_cp *cp = &rdev->cp[RADEON_RING_TYPE_GFX_INDEX]; 1459 int r; 1460 1461 /* This don't do much */ 1462 r = radeon_gem_init(rdev); 1463 if (r) 1464 return r; 1465 /* Read BIOS */ 1466 if (!radeon_get_bios(rdev)) { 1467 if (ASIC_IS_AVIVO(rdev)) 1468 return -EINVAL; 1469 } 1470 /* Must be an ATOMBIOS */ 1471 if (!rdev->is_atom_bios) { 1472 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n"); 1473 return -EINVAL; 1474 } 1475 r = radeon_atombios_init(rdev); 1476 if (r) 1477 return r; 1478 1479 /* Post card if necessary */ 1480 if (!radeon_card_posted(rdev)) { 1481 if (!rdev->bios) { 1482 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n"); 1483 return -EINVAL; 1484 } 1485 DRM_INFO("GPU not posted. posting now...\n"); 1486 atom_asic_init(rdev->mode_info.atom_context); 1487 } 1488 /* Initialize scratch registers */ 1489 r600_scratch_init(rdev); 1490 /* Initialize surface registers */ 1491 radeon_surface_init(rdev); 1492 /* Initialize clocks */ 1493 radeon_get_clock_info(rdev->ddev); 1494 /* Fence driver */ 1495 r = radeon_fence_driver_init(rdev, 3); 1496 if (r) 1497 return r; 1498 /* initialize memory controller */ 1499 r = evergreen_mc_init(rdev); 1500 if (r) 1501 return r; 1502 /* Memory manager */ 1503 r = radeon_bo_init(rdev); 1504 if (r) 1505 return r; 1506 1507 r = radeon_irq_kms_init(rdev); 1508 if (r) 1509 return r; 1510 1511 cp->ring_obj = NULL; 1512 r600_ring_init(rdev, cp, 1024 * 1024); 1513 1514 rdev->ih.ring_obj = NULL; 1515 r600_ih_ring_init(rdev, 64 * 1024); 1516 1517 r = r600_pcie_gart_init(rdev); 1518 if (r) 1519 return r; 1520 1521 rdev->accel_working = true; 1522 r = cayman_startup(rdev); 1523 if (r) { 1524 dev_err(rdev->dev, "disabling GPU acceleration\n"); 1525 cayman_cp_fini(rdev); 1526 r600_irq_fini(rdev); 1527 radeon_wb_fini(rdev); 1528 radeon_irq_kms_fini(rdev); 1529 cayman_pcie_gart_fini(rdev); 1530 rdev->accel_working = false; 1531 } 1532 if (rdev->accel_working) { 1533 r = radeon_ib_pool_init(rdev); 1534 if (r) { 1535 DRM_ERROR("radeon: failed initializing IB pool (%d).\n", r); 1536 rdev->accel_working = false; 1537 } 1538 r = r600_ib_test(rdev, RADEON_RING_TYPE_GFX_INDEX); 1539 if (r) { 1540 DRM_ERROR("radeon: failed testing IB (%d).\n", r); 1541 rdev->accel_working = false; 1542 } 1543 } 1544 1545 /* Don't start up if the MC ucode is missing. 1546 * The default clocks and voltages before the MC ucode 1547 * is loaded are not suffient for advanced operations. 1548 */ 1549 if (!rdev->mc_fw) { 1550 DRM_ERROR("radeon: MC ucode required for NI+.\n"); 1551 return -EINVAL; 1552 } 1553 1554 return 0; 1555 } 1556 1557 void cayman_fini(struct radeon_device *rdev) 1558 { 1559 r600_blit_fini(rdev); 1560 cayman_cp_fini(rdev); 1561 r600_irq_fini(rdev); 1562 radeon_wb_fini(rdev); 1563 radeon_ib_pool_fini(rdev); 1564 radeon_irq_kms_fini(rdev); 1565 cayman_pcie_gart_fini(rdev); 1566 r600_vram_scratch_fini(rdev); 1567 radeon_gem_fini(rdev); 1568 radeon_semaphore_driver_fini(rdev); 1569 radeon_fence_driver_fini(rdev); 1570 radeon_bo_fini(rdev); 1571 radeon_atombios_fini(rdev); 1572 kfree(rdev->bios); 1573 rdev->bios = NULL; 1574 } 1575 1576