1 /* 2 * Copyright 2011 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/slab.h> 26 #include <linux/module.h> 27 #include <drm/drmP.h> 28 #include "radeon.h" 29 #include "radeon_asic.h" 30 #include "radeon_audio.h" 31 #include <drm/radeon_drm.h> 32 #include "sid.h" 33 #include "atom.h" 34 #include "si_blit_shaders.h" 35 #include "clearstate_si.h" 36 #include "radeon_ucode.h" 37 38 39 MODULE_FIRMWARE("radeon/TAHITI_pfp.bin"); 40 MODULE_FIRMWARE("radeon/TAHITI_me.bin"); 41 MODULE_FIRMWARE("radeon/TAHITI_ce.bin"); 42 MODULE_FIRMWARE("radeon/TAHITI_mc.bin"); 43 MODULE_FIRMWARE("radeon/TAHITI_mc2.bin"); 44 MODULE_FIRMWARE("radeon/TAHITI_rlc.bin"); 45 MODULE_FIRMWARE("radeon/TAHITI_smc.bin"); 46 47 MODULE_FIRMWARE("radeon/tahiti_pfp.bin"); 48 MODULE_FIRMWARE("radeon/tahiti_me.bin"); 49 MODULE_FIRMWARE("radeon/tahiti_ce.bin"); 50 MODULE_FIRMWARE("radeon/tahiti_mc.bin"); 51 MODULE_FIRMWARE("radeon/tahiti_rlc.bin"); 52 MODULE_FIRMWARE("radeon/tahiti_smc.bin"); 53 54 MODULE_FIRMWARE("radeon/PITCAIRN_pfp.bin"); 55 MODULE_FIRMWARE("radeon/PITCAIRN_me.bin"); 56 MODULE_FIRMWARE("radeon/PITCAIRN_ce.bin"); 57 MODULE_FIRMWARE("radeon/PITCAIRN_mc.bin"); 58 MODULE_FIRMWARE("radeon/PITCAIRN_mc2.bin"); 59 MODULE_FIRMWARE("radeon/PITCAIRN_rlc.bin"); 60 MODULE_FIRMWARE("radeon/PITCAIRN_smc.bin"); 61 62 MODULE_FIRMWARE("radeon/pitcairn_pfp.bin"); 63 MODULE_FIRMWARE("radeon/pitcairn_me.bin"); 64 MODULE_FIRMWARE("radeon/pitcairn_ce.bin"); 65 MODULE_FIRMWARE("radeon/pitcairn_mc.bin"); 66 MODULE_FIRMWARE("radeon/pitcairn_rlc.bin"); 67 MODULE_FIRMWARE("radeon/pitcairn_smc.bin"); 68 MODULE_FIRMWARE("radeon/pitcairn_k_smc.bin"); 69 70 MODULE_FIRMWARE("radeon/VERDE_pfp.bin"); 71 MODULE_FIRMWARE("radeon/VERDE_me.bin"); 72 MODULE_FIRMWARE("radeon/VERDE_ce.bin"); 73 MODULE_FIRMWARE("radeon/VERDE_mc.bin"); 74 MODULE_FIRMWARE("radeon/VERDE_mc2.bin"); 75 MODULE_FIRMWARE("radeon/VERDE_rlc.bin"); 76 MODULE_FIRMWARE("radeon/VERDE_smc.bin"); 77 78 MODULE_FIRMWARE("radeon/verde_pfp.bin"); 79 MODULE_FIRMWARE("radeon/verde_me.bin"); 80 MODULE_FIRMWARE("radeon/verde_ce.bin"); 81 MODULE_FIRMWARE("radeon/verde_mc.bin"); 82 MODULE_FIRMWARE("radeon/verde_rlc.bin"); 83 MODULE_FIRMWARE("radeon/verde_smc.bin"); 84 MODULE_FIRMWARE("radeon/verde_k_smc.bin"); 85 86 MODULE_FIRMWARE("radeon/OLAND_pfp.bin"); 87 MODULE_FIRMWARE("radeon/OLAND_me.bin"); 88 MODULE_FIRMWARE("radeon/OLAND_ce.bin"); 89 MODULE_FIRMWARE("radeon/OLAND_mc.bin"); 90 MODULE_FIRMWARE("radeon/OLAND_mc2.bin"); 91 MODULE_FIRMWARE("radeon/OLAND_rlc.bin"); 92 MODULE_FIRMWARE("radeon/OLAND_smc.bin"); 93 94 MODULE_FIRMWARE("radeon/oland_pfp.bin"); 95 MODULE_FIRMWARE("radeon/oland_me.bin"); 96 MODULE_FIRMWARE("radeon/oland_ce.bin"); 97 MODULE_FIRMWARE("radeon/oland_mc.bin"); 98 MODULE_FIRMWARE("radeon/oland_rlc.bin"); 99 MODULE_FIRMWARE("radeon/oland_smc.bin"); 100 MODULE_FIRMWARE("radeon/oland_k_smc.bin"); 101 102 MODULE_FIRMWARE("radeon/HAINAN_pfp.bin"); 103 MODULE_FIRMWARE("radeon/HAINAN_me.bin"); 104 MODULE_FIRMWARE("radeon/HAINAN_ce.bin"); 105 MODULE_FIRMWARE("radeon/HAINAN_mc.bin"); 106 MODULE_FIRMWARE("radeon/HAINAN_mc2.bin"); 107 MODULE_FIRMWARE("radeon/HAINAN_rlc.bin"); 108 MODULE_FIRMWARE("radeon/HAINAN_smc.bin"); 109 110 MODULE_FIRMWARE("radeon/hainan_pfp.bin"); 111 MODULE_FIRMWARE("radeon/hainan_me.bin"); 112 MODULE_FIRMWARE("radeon/hainan_ce.bin"); 113 MODULE_FIRMWARE("radeon/hainan_mc.bin"); 114 MODULE_FIRMWARE("radeon/hainan_rlc.bin"); 115 MODULE_FIRMWARE("radeon/hainan_smc.bin"); 116 MODULE_FIRMWARE("radeon/hainan_k_smc.bin"); 117 MODULE_FIRMWARE("radeon/banks_k_2_smc.bin"); 118 119 MODULE_FIRMWARE("radeon/si58_mc.bin"); 120 121 static u32 si_get_cu_active_bitmap(struct radeon_device *rdev, u32 se, u32 sh); 122 static void si_pcie_gen3_enable(struct radeon_device *rdev); 123 static void si_program_aspm(struct radeon_device *rdev); 124 extern void sumo_rlc_fini(struct radeon_device *rdev); 125 extern int sumo_rlc_init(struct radeon_device *rdev); 126 extern int r600_ih_ring_alloc(struct radeon_device *rdev); 127 extern void r600_ih_ring_fini(struct radeon_device *rdev); 128 extern void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev); 129 extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save); 130 extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save); 131 extern u32 evergreen_get_number_of_dram_channels(struct radeon_device *rdev); 132 extern void evergreen_print_gpu_status_regs(struct radeon_device *rdev); 133 extern bool evergreen_is_display_hung(struct radeon_device *rdev); 134 static void si_enable_gui_idle_interrupt(struct radeon_device *rdev, 135 bool enable); 136 static void si_init_pg(struct radeon_device *rdev); 137 static void si_init_cg(struct radeon_device *rdev); 138 static void si_fini_pg(struct radeon_device *rdev); 139 static void si_fini_cg(struct radeon_device *rdev); 140 static void si_rlc_stop(struct radeon_device *rdev); 141 142 static const u32 crtc_offsets[] = 143 { 144 EVERGREEN_CRTC0_REGISTER_OFFSET, 145 EVERGREEN_CRTC1_REGISTER_OFFSET, 146 EVERGREEN_CRTC2_REGISTER_OFFSET, 147 EVERGREEN_CRTC3_REGISTER_OFFSET, 148 EVERGREEN_CRTC4_REGISTER_OFFSET, 149 EVERGREEN_CRTC5_REGISTER_OFFSET 150 }; 151 152 static const u32 si_disp_int_status[] = 153 { 154 DISP_INTERRUPT_STATUS, 155 DISP_INTERRUPT_STATUS_CONTINUE, 156 DISP_INTERRUPT_STATUS_CONTINUE2, 157 DISP_INTERRUPT_STATUS_CONTINUE3, 158 DISP_INTERRUPT_STATUS_CONTINUE4, 159 DISP_INTERRUPT_STATUS_CONTINUE5 160 }; 161 162 #define DC_HPDx_CONTROL(x) (DC_HPD1_CONTROL + (x * 0xc)) 163 #define DC_HPDx_INT_CONTROL(x) (DC_HPD1_INT_CONTROL + (x * 0xc)) 164 #define DC_HPDx_INT_STATUS_REG(x) (DC_HPD1_INT_STATUS + (x * 0xc)) 165 166 static const u32 verde_rlc_save_restore_register_list[] = 167 { 168 (0x8000 << 16) | (0x98f4 >> 2), 169 0x00000000, 170 (0x8040 << 16) | (0x98f4 >> 2), 171 0x00000000, 172 (0x8000 << 16) | (0xe80 >> 2), 173 0x00000000, 174 (0x8040 << 16) | (0xe80 >> 2), 175 0x00000000, 176 (0x8000 << 16) | (0x89bc >> 2), 177 0x00000000, 178 (0x8040 << 16) | (0x89bc >> 2), 179 0x00000000, 180 (0x8000 << 16) | (0x8c1c >> 2), 181 0x00000000, 182 (0x8040 << 16) | (0x8c1c >> 2), 183 0x00000000, 184 (0x9c00 << 16) | (0x98f0 >> 2), 185 0x00000000, 186 (0x9c00 << 16) | (0xe7c >> 2), 187 0x00000000, 188 (0x8000 << 16) | (0x9148 >> 2), 189 0x00000000, 190 (0x8040 << 16) | (0x9148 >> 2), 191 0x00000000, 192 (0x9c00 << 16) | (0x9150 >> 2), 193 0x00000000, 194 (0x9c00 << 16) | (0x897c >> 2), 195 0x00000000, 196 (0x9c00 << 16) | (0x8d8c >> 2), 197 0x00000000, 198 (0x9c00 << 16) | (0xac54 >> 2), 199 0X00000000, 200 0x3, 201 (0x9c00 << 16) | (0x98f8 >> 2), 202 0x00000000, 203 (0x9c00 << 16) | (0x9910 >> 2), 204 0x00000000, 205 (0x9c00 << 16) | (0x9914 >> 2), 206 0x00000000, 207 (0x9c00 << 16) | (0x9918 >> 2), 208 0x00000000, 209 (0x9c00 << 16) | (0x991c >> 2), 210 0x00000000, 211 (0x9c00 << 16) | (0x9920 >> 2), 212 0x00000000, 213 (0x9c00 << 16) | (0x9924 >> 2), 214 0x00000000, 215 (0x9c00 << 16) | (0x9928 >> 2), 216 0x00000000, 217 (0x9c00 << 16) | (0x992c >> 2), 218 0x00000000, 219 (0x9c00 << 16) | (0x9930 >> 2), 220 0x00000000, 221 (0x9c00 << 16) | (0x9934 >> 2), 222 0x00000000, 223 (0x9c00 << 16) | (0x9938 >> 2), 224 0x00000000, 225 (0x9c00 << 16) | (0x993c >> 2), 226 0x00000000, 227 (0x9c00 << 16) | (0x9940 >> 2), 228 0x00000000, 229 (0x9c00 << 16) | (0x9944 >> 2), 230 0x00000000, 231 (0x9c00 << 16) | (0x9948 >> 2), 232 0x00000000, 233 (0x9c00 << 16) | (0x994c >> 2), 234 0x00000000, 235 (0x9c00 << 16) | (0x9950 >> 2), 236 0x00000000, 237 (0x9c00 << 16) | (0x9954 >> 2), 238 0x00000000, 239 (0x9c00 << 16) | (0x9958 >> 2), 240 0x00000000, 241 (0x9c00 << 16) | (0x995c >> 2), 242 0x00000000, 243 (0x9c00 << 16) | (0x9960 >> 2), 244 0x00000000, 245 (0x9c00 << 16) | (0x9964 >> 2), 246 0x00000000, 247 (0x9c00 << 16) | (0x9968 >> 2), 248 0x00000000, 249 (0x9c00 << 16) | (0x996c >> 2), 250 0x00000000, 251 (0x9c00 << 16) | (0x9970 >> 2), 252 0x00000000, 253 (0x9c00 << 16) | (0x9974 >> 2), 254 0x00000000, 255 (0x9c00 << 16) | (0x9978 >> 2), 256 0x00000000, 257 (0x9c00 << 16) | (0x997c >> 2), 258 0x00000000, 259 (0x9c00 << 16) | (0x9980 >> 2), 260 0x00000000, 261 (0x9c00 << 16) | (0x9984 >> 2), 262 0x00000000, 263 (0x9c00 << 16) | (0x9988 >> 2), 264 0x00000000, 265 (0x9c00 << 16) | (0x998c >> 2), 266 0x00000000, 267 (0x9c00 << 16) | (0x8c00 >> 2), 268 0x00000000, 269 (0x9c00 << 16) | (0x8c14 >> 2), 270 0x00000000, 271 (0x9c00 << 16) | (0x8c04 >> 2), 272 0x00000000, 273 (0x9c00 << 16) | (0x8c08 >> 2), 274 0x00000000, 275 (0x8000 << 16) | (0x9b7c >> 2), 276 0x00000000, 277 (0x8040 << 16) | (0x9b7c >> 2), 278 0x00000000, 279 (0x8000 << 16) | (0xe84 >> 2), 280 0x00000000, 281 (0x8040 << 16) | (0xe84 >> 2), 282 0x00000000, 283 (0x8000 << 16) | (0x89c0 >> 2), 284 0x00000000, 285 (0x8040 << 16) | (0x89c0 >> 2), 286 0x00000000, 287 (0x8000 << 16) | (0x914c >> 2), 288 0x00000000, 289 (0x8040 << 16) | (0x914c >> 2), 290 0x00000000, 291 (0x8000 << 16) | (0x8c20 >> 2), 292 0x00000000, 293 (0x8040 << 16) | (0x8c20 >> 2), 294 0x00000000, 295 (0x8000 << 16) | (0x9354 >> 2), 296 0x00000000, 297 (0x8040 << 16) | (0x9354 >> 2), 298 0x00000000, 299 (0x9c00 << 16) | (0x9060 >> 2), 300 0x00000000, 301 (0x9c00 << 16) | (0x9364 >> 2), 302 0x00000000, 303 (0x9c00 << 16) | (0x9100 >> 2), 304 0x00000000, 305 (0x9c00 << 16) | (0x913c >> 2), 306 0x00000000, 307 (0x8000 << 16) | (0x90e0 >> 2), 308 0x00000000, 309 (0x8000 << 16) | (0x90e4 >> 2), 310 0x00000000, 311 (0x8000 << 16) | (0x90e8 >> 2), 312 0x00000000, 313 (0x8040 << 16) | (0x90e0 >> 2), 314 0x00000000, 315 (0x8040 << 16) | (0x90e4 >> 2), 316 0x00000000, 317 (0x8040 << 16) | (0x90e8 >> 2), 318 0x00000000, 319 (0x9c00 << 16) | (0x8bcc >> 2), 320 0x00000000, 321 (0x9c00 << 16) | (0x8b24 >> 2), 322 0x00000000, 323 (0x9c00 << 16) | (0x88c4 >> 2), 324 0x00000000, 325 (0x9c00 << 16) | (0x8e50 >> 2), 326 0x00000000, 327 (0x9c00 << 16) | (0x8c0c >> 2), 328 0x00000000, 329 (0x9c00 << 16) | (0x8e58 >> 2), 330 0x00000000, 331 (0x9c00 << 16) | (0x8e5c >> 2), 332 0x00000000, 333 (0x9c00 << 16) | (0x9508 >> 2), 334 0x00000000, 335 (0x9c00 << 16) | (0x950c >> 2), 336 0x00000000, 337 (0x9c00 << 16) | (0x9494 >> 2), 338 0x00000000, 339 (0x9c00 << 16) | (0xac0c >> 2), 340 0x00000000, 341 (0x9c00 << 16) | (0xac10 >> 2), 342 0x00000000, 343 (0x9c00 << 16) | (0xac14 >> 2), 344 0x00000000, 345 (0x9c00 << 16) | (0xae00 >> 2), 346 0x00000000, 347 (0x9c00 << 16) | (0xac08 >> 2), 348 0x00000000, 349 (0x9c00 << 16) | (0x88d4 >> 2), 350 0x00000000, 351 (0x9c00 << 16) | (0x88c8 >> 2), 352 0x00000000, 353 (0x9c00 << 16) | (0x88cc >> 2), 354 0x00000000, 355 (0x9c00 << 16) | (0x89b0 >> 2), 356 0x00000000, 357 (0x9c00 << 16) | (0x8b10 >> 2), 358 0x00000000, 359 (0x9c00 << 16) | (0x8a14 >> 2), 360 0x00000000, 361 (0x9c00 << 16) | (0x9830 >> 2), 362 0x00000000, 363 (0x9c00 << 16) | (0x9834 >> 2), 364 0x00000000, 365 (0x9c00 << 16) | (0x9838 >> 2), 366 0x00000000, 367 (0x9c00 << 16) | (0x9a10 >> 2), 368 0x00000000, 369 (0x8000 << 16) | (0x9870 >> 2), 370 0x00000000, 371 (0x8000 << 16) | (0x9874 >> 2), 372 0x00000000, 373 (0x8001 << 16) | (0x9870 >> 2), 374 0x00000000, 375 (0x8001 << 16) | (0x9874 >> 2), 376 0x00000000, 377 (0x8040 << 16) | (0x9870 >> 2), 378 0x00000000, 379 (0x8040 << 16) | (0x9874 >> 2), 380 0x00000000, 381 (0x8041 << 16) | (0x9870 >> 2), 382 0x00000000, 383 (0x8041 << 16) | (0x9874 >> 2), 384 0x00000000, 385 0x00000000 386 }; 387 388 static const u32 tahiti_golden_rlc_registers[] = 389 { 390 0xc424, 0xffffffff, 0x00601005, 391 0xc47c, 0xffffffff, 0x10104040, 392 0xc488, 0xffffffff, 0x0100000a, 393 0xc314, 0xffffffff, 0x00000800, 394 0xc30c, 0xffffffff, 0x800000f4, 395 0xf4a8, 0xffffffff, 0x00000000 396 }; 397 398 static const u32 tahiti_golden_registers[] = 399 { 400 0x9a10, 0x00010000, 0x00018208, 401 0x9830, 0xffffffff, 0x00000000, 402 0x9834, 0xf00fffff, 0x00000400, 403 0x9838, 0x0002021c, 0x00020200, 404 0xc78, 0x00000080, 0x00000000, 405 0xd030, 0x000300c0, 0x00800040, 406 0xd830, 0x000300c0, 0x00800040, 407 0x5bb0, 0x000000f0, 0x00000070, 408 0x5bc0, 0x00200000, 0x50100000, 409 0x7030, 0x31000311, 0x00000011, 410 0x277c, 0x00000003, 0x000007ff, 411 0x240c, 0x000007ff, 0x00000000, 412 0x8a14, 0xf000001f, 0x00000007, 413 0x8b24, 0xffffffff, 0x00ffffff, 414 0x8b10, 0x0000ff0f, 0x00000000, 415 0x28a4c, 0x07ffffff, 0x4e000000, 416 0x28350, 0x3f3f3fff, 0x2a00126a, 417 0x30, 0x000000ff, 0x0040, 418 0x34, 0x00000040, 0x00004040, 419 0x9100, 0x07ffffff, 0x03000000, 420 0x8e88, 0x01ff1f3f, 0x00000000, 421 0x8e84, 0x01ff1f3f, 0x00000000, 422 0x9060, 0x0000007f, 0x00000020, 423 0x9508, 0x00010000, 0x00010000, 424 0xac14, 0x00000200, 0x000002fb, 425 0xac10, 0xffffffff, 0x0000543b, 426 0xac0c, 0xffffffff, 0xa9210876, 427 0x88d0, 0xffffffff, 0x000fff40, 428 0x88d4, 0x0000001f, 0x00000010, 429 0x1410, 0x20000000, 0x20fffed8, 430 0x15c0, 0x000c0fc0, 0x000c0400 431 }; 432 433 static const u32 tahiti_golden_registers2[] = 434 { 435 0xc64, 0x00000001, 0x00000001 436 }; 437 438 static const u32 pitcairn_golden_rlc_registers[] = 439 { 440 0xc424, 0xffffffff, 0x00601004, 441 0xc47c, 0xffffffff, 0x10102020, 442 0xc488, 0xffffffff, 0x01000020, 443 0xc314, 0xffffffff, 0x00000800, 444 0xc30c, 0xffffffff, 0x800000a4 445 }; 446 447 static const u32 pitcairn_golden_registers[] = 448 { 449 0x9a10, 0x00010000, 0x00018208, 450 0x9830, 0xffffffff, 0x00000000, 451 0x9834, 0xf00fffff, 0x00000400, 452 0x9838, 0x0002021c, 0x00020200, 453 0xc78, 0x00000080, 0x00000000, 454 0xd030, 0x000300c0, 0x00800040, 455 0xd830, 0x000300c0, 0x00800040, 456 0x5bb0, 0x000000f0, 0x00000070, 457 0x5bc0, 0x00200000, 0x50100000, 458 0x7030, 0x31000311, 0x00000011, 459 0x2ae4, 0x00073ffe, 0x000022a2, 460 0x240c, 0x000007ff, 0x00000000, 461 0x8a14, 0xf000001f, 0x00000007, 462 0x8b24, 0xffffffff, 0x00ffffff, 463 0x8b10, 0x0000ff0f, 0x00000000, 464 0x28a4c, 0x07ffffff, 0x4e000000, 465 0x28350, 0x3f3f3fff, 0x2a00126a, 466 0x30, 0x000000ff, 0x0040, 467 0x34, 0x00000040, 0x00004040, 468 0x9100, 0x07ffffff, 0x03000000, 469 0x9060, 0x0000007f, 0x00000020, 470 0x9508, 0x00010000, 0x00010000, 471 0xac14, 0x000003ff, 0x000000f7, 472 0xac10, 0xffffffff, 0x00000000, 473 0xac0c, 0xffffffff, 0x32761054, 474 0x88d4, 0x0000001f, 0x00000010, 475 0x15c0, 0x000c0fc0, 0x000c0400 476 }; 477 478 static const u32 verde_golden_rlc_registers[] = 479 { 480 0xc424, 0xffffffff, 0x033f1005, 481 0xc47c, 0xffffffff, 0x10808020, 482 0xc488, 0xffffffff, 0x00800008, 483 0xc314, 0xffffffff, 0x00001000, 484 0xc30c, 0xffffffff, 0x80010014 485 }; 486 487 static const u32 verde_golden_registers[] = 488 { 489 0x9a10, 0x00010000, 0x00018208, 490 0x9830, 0xffffffff, 0x00000000, 491 0x9834, 0xf00fffff, 0x00000400, 492 0x9838, 0x0002021c, 0x00020200, 493 0xc78, 0x00000080, 0x00000000, 494 0xd030, 0x000300c0, 0x00800040, 495 0xd030, 0x000300c0, 0x00800040, 496 0xd830, 0x000300c0, 0x00800040, 497 0xd830, 0x000300c0, 0x00800040, 498 0x5bb0, 0x000000f0, 0x00000070, 499 0x5bc0, 0x00200000, 0x50100000, 500 0x7030, 0x31000311, 0x00000011, 501 0x2ae4, 0x00073ffe, 0x000022a2, 502 0x2ae4, 0x00073ffe, 0x000022a2, 503 0x2ae4, 0x00073ffe, 0x000022a2, 504 0x240c, 0x000007ff, 0x00000000, 505 0x240c, 0x000007ff, 0x00000000, 506 0x240c, 0x000007ff, 0x00000000, 507 0x8a14, 0xf000001f, 0x00000007, 508 0x8a14, 0xf000001f, 0x00000007, 509 0x8a14, 0xf000001f, 0x00000007, 510 0x8b24, 0xffffffff, 0x00ffffff, 511 0x8b10, 0x0000ff0f, 0x00000000, 512 0x28a4c, 0x07ffffff, 0x4e000000, 513 0x28350, 0x3f3f3fff, 0x0000124a, 514 0x28350, 0x3f3f3fff, 0x0000124a, 515 0x28350, 0x3f3f3fff, 0x0000124a, 516 0x30, 0x000000ff, 0x0040, 517 0x34, 0x00000040, 0x00004040, 518 0x9100, 0x07ffffff, 0x03000000, 519 0x9100, 0x07ffffff, 0x03000000, 520 0x8e88, 0x01ff1f3f, 0x00000000, 521 0x8e88, 0x01ff1f3f, 0x00000000, 522 0x8e88, 0x01ff1f3f, 0x00000000, 523 0x8e84, 0x01ff1f3f, 0x00000000, 524 0x8e84, 0x01ff1f3f, 0x00000000, 525 0x8e84, 0x01ff1f3f, 0x00000000, 526 0x9060, 0x0000007f, 0x00000020, 527 0x9508, 0x00010000, 0x00010000, 528 0xac14, 0x000003ff, 0x00000003, 529 0xac14, 0x000003ff, 0x00000003, 530 0xac14, 0x000003ff, 0x00000003, 531 0xac10, 0xffffffff, 0x00000000, 532 0xac10, 0xffffffff, 0x00000000, 533 0xac10, 0xffffffff, 0x00000000, 534 0xac0c, 0xffffffff, 0x00001032, 535 0xac0c, 0xffffffff, 0x00001032, 536 0xac0c, 0xffffffff, 0x00001032, 537 0x88d4, 0x0000001f, 0x00000010, 538 0x88d4, 0x0000001f, 0x00000010, 539 0x88d4, 0x0000001f, 0x00000010, 540 0x15c0, 0x000c0fc0, 0x000c0400 541 }; 542 543 static const u32 oland_golden_rlc_registers[] = 544 { 545 0xc424, 0xffffffff, 0x00601005, 546 0xc47c, 0xffffffff, 0x10104040, 547 0xc488, 0xffffffff, 0x0100000a, 548 0xc314, 0xffffffff, 0x00000800, 549 0xc30c, 0xffffffff, 0x800000f4 550 }; 551 552 static const u32 oland_golden_registers[] = 553 { 554 0x9a10, 0x00010000, 0x00018208, 555 0x9830, 0xffffffff, 0x00000000, 556 0x9834, 0xf00fffff, 0x00000400, 557 0x9838, 0x0002021c, 0x00020200, 558 0xc78, 0x00000080, 0x00000000, 559 0xd030, 0x000300c0, 0x00800040, 560 0xd830, 0x000300c0, 0x00800040, 561 0x5bb0, 0x000000f0, 0x00000070, 562 0x5bc0, 0x00200000, 0x50100000, 563 0x7030, 0x31000311, 0x00000011, 564 0x2ae4, 0x00073ffe, 0x000022a2, 565 0x240c, 0x000007ff, 0x00000000, 566 0x8a14, 0xf000001f, 0x00000007, 567 0x8b24, 0xffffffff, 0x00ffffff, 568 0x8b10, 0x0000ff0f, 0x00000000, 569 0x28a4c, 0x07ffffff, 0x4e000000, 570 0x28350, 0x3f3f3fff, 0x00000082, 571 0x30, 0x000000ff, 0x0040, 572 0x34, 0x00000040, 0x00004040, 573 0x9100, 0x07ffffff, 0x03000000, 574 0x9060, 0x0000007f, 0x00000020, 575 0x9508, 0x00010000, 0x00010000, 576 0xac14, 0x000003ff, 0x000000f3, 577 0xac10, 0xffffffff, 0x00000000, 578 0xac0c, 0xffffffff, 0x00003210, 579 0x88d4, 0x0000001f, 0x00000010, 580 0x15c0, 0x000c0fc0, 0x000c0400 581 }; 582 583 static const u32 hainan_golden_registers[] = 584 { 585 0x9a10, 0x00010000, 0x00018208, 586 0x9830, 0xffffffff, 0x00000000, 587 0x9834, 0xf00fffff, 0x00000400, 588 0x9838, 0x0002021c, 0x00020200, 589 0xd0c0, 0xff000fff, 0x00000100, 590 0xd030, 0x000300c0, 0x00800040, 591 0xd8c0, 0xff000fff, 0x00000100, 592 0xd830, 0x000300c0, 0x00800040, 593 0x2ae4, 0x00073ffe, 0x000022a2, 594 0x240c, 0x000007ff, 0x00000000, 595 0x8a14, 0xf000001f, 0x00000007, 596 0x8b24, 0xffffffff, 0x00ffffff, 597 0x8b10, 0x0000ff0f, 0x00000000, 598 0x28a4c, 0x07ffffff, 0x4e000000, 599 0x28350, 0x3f3f3fff, 0x00000000, 600 0x30, 0x000000ff, 0x0040, 601 0x34, 0x00000040, 0x00004040, 602 0x9100, 0x03e00000, 0x03600000, 603 0x9060, 0x0000007f, 0x00000020, 604 0x9508, 0x00010000, 0x00010000, 605 0xac14, 0x000003ff, 0x000000f1, 606 0xac10, 0xffffffff, 0x00000000, 607 0xac0c, 0xffffffff, 0x00003210, 608 0x88d4, 0x0000001f, 0x00000010, 609 0x15c0, 0x000c0fc0, 0x000c0400 610 }; 611 612 static const u32 hainan_golden_registers2[] = 613 { 614 0x98f8, 0xffffffff, 0x02010001 615 }; 616 617 static const u32 tahiti_mgcg_cgcg_init[] = 618 { 619 0xc400, 0xffffffff, 0xfffffffc, 620 0x802c, 0xffffffff, 0xe0000000, 621 0x9a60, 0xffffffff, 0x00000100, 622 0x92a4, 0xffffffff, 0x00000100, 623 0xc164, 0xffffffff, 0x00000100, 624 0x9774, 0xffffffff, 0x00000100, 625 0x8984, 0xffffffff, 0x06000100, 626 0x8a18, 0xffffffff, 0x00000100, 627 0x92a0, 0xffffffff, 0x00000100, 628 0xc380, 0xffffffff, 0x00000100, 629 0x8b28, 0xffffffff, 0x00000100, 630 0x9144, 0xffffffff, 0x00000100, 631 0x8d88, 0xffffffff, 0x00000100, 632 0x8d8c, 0xffffffff, 0x00000100, 633 0x9030, 0xffffffff, 0x00000100, 634 0x9034, 0xffffffff, 0x00000100, 635 0x9038, 0xffffffff, 0x00000100, 636 0x903c, 0xffffffff, 0x00000100, 637 0xad80, 0xffffffff, 0x00000100, 638 0xac54, 0xffffffff, 0x00000100, 639 0x897c, 0xffffffff, 0x06000100, 640 0x9868, 0xffffffff, 0x00000100, 641 0x9510, 0xffffffff, 0x00000100, 642 0xaf04, 0xffffffff, 0x00000100, 643 0xae04, 0xffffffff, 0x00000100, 644 0x949c, 0xffffffff, 0x00000100, 645 0x802c, 0xffffffff, 0xe0000000, 646 0x9160, 0xffffffff, 0x00010000, 647 0x9164, 0xffffffff, 0x00030002, 648 0x9168, 0xffffffff, 0x00040007, 649 0x916c, 0xffffffff, 0x00060005, 650 0x9170, 0xffffffff, 0x00090008, 651 0x9174, 0xffffffff, 0x00020001, 652 0x9178, 0xffffffff, 0x00040003, 653 0x917c, 0xffffffff, 0x00000007, 654 0x9180, 0xffffffff, 0x00060005, 655 0x9184, 0xffffffff, 0x00090008, 656 0x9188, 0xffffffff, 0x00030002, 657 0x918c, 0xffffffff, 0x00050004, 658 0x9190, 0xffffffff, 0x00000008, 659 0x9194, 0xffffffff, 0x00070006, 660 0x9198, 0xffffffff, 0x000a0009, 661 0x919c, 0xffffffff, 0x00040003, 662 0x91a0, 0xffffffff, 0x00060005, 663 0x91a4, 0xffffffff, 0x00000009, 664 0x91a8, 0xffffffff, 0x00080007, 665 0x91ac, 0xffffffff, 0x000b000a, 666 0x91b0, 0xffffffff, 0x00050004, 667 0x91b4, 0xffffffff, 0x00070006, 668 0x91b8, 0xffffffff, 0x0008000b, 669 0x91bc, 0xffffffff, 0x000a0009, 670 0x91c0, 0xffffffff, 0x000d000c, 671 0x91c4, 0xffffffff, 0x00060005, 672 0x91c8, 0xffffffff, 0x00080007, 673 0x91cc, 0xffffffff, 0x0000000b, 674 0x91d0, 0xffffffff, 0x000a0009, 675 0x91d4, 0xffffffff, 0x000d000c, 676 0x91d8, 0xffffffff, 0x00070006, 677 0x91dc, 0xffffffff, 0x00090008, 678 0x91e0, 0xffffffff, 0x0000000c, 679 0x91e4, 0xffffffff, 0x000b000a, 680 0x91e8, 0xffffffff, 0x000e000d, 681 0x91ec, 0xffffffff, 0x00080007, 682 0x91f0, 0xffffffff, 0x000a0009, 683 0x91f4, 0xffffffff, 0x0000000d, 684 0x91f8, 0xffffffff, 0x000c000b, 685 0x91fc, 0xffffffff, 0x000f000e, 686 0x9200, 0xffffffff, 0x00090008, 687 0x9204, 0xffffffff, 0x000b000a, 688 0x9208, 0xffffffff, 0x000c000f, 689 0x920c, 0xffffffff, 0x000e000d, 690 0x9210, 0xffffffff, 0x00110010, 691 0x9214, 0xffffffff, 0x000a0009, 692 0x9218, 0xffffffff, 0x000c000b, 693 0x921c, 0xffffffff, 0x0000000f, 694 0x9220, 0xffffffff, 0x000e000d, 695 0x9224, 0xffffffff, 0x00110010, 696 0x9228, 0xffffffff, 0x000b000a, 697 0x922c, 0xffffffff, 0x000d000c, 698 0x9230, 0xffffffff, 0x00000010, 699 0x9234, 0xffffffff, 0x000f000e, 700 0x9238, 0xffffffff, 0x00120011, 701 0x923c, 0xffffffff, 0x000c000b, 702 0x9240, 0xffffffff, 0x000e000d, 703 0x9244, 0xffffffff, 0x00000011, 704 0x9248, 0xffffffff, 0x0010000f, 705 0x924c, 0xffffffff, 0x00130012, 706 0x9250, 0xffffffff, 0x000d000c, 707 0x9254, 0xffffffff, 0x000f000e, 708 0x9258, 0xffffffff, 0x00100013, 709 0x925c, 0xffffffff, 0x00120011, 710 0x9260, 0xffffffff, 0x00150014, 711 0x9264, 0xffffffff, 0x000e000d, 712 0x9268, 0xffffffff, 0x0010000f, 713 0x926c, 0xffffffff, 0x00000013, 714 0x9270, 0xffffffff, 0x00120011, 715 0x9274, 0xffffffff, 0x00150014, 716 0x9278, 0xffffffff, 0x000f000e, 717 0x927c, 0xffffffff, 0x00110010, 718 0x9280, 0xffffffff, 0x00000014, 719 0x9284, 0xffffffff, 0x00130012, 720 0x9288, 0xffffffff, 0x00160015, 721 0x928c, 0xffffffff, 0x0010000f, 722 0x9290, 0xffffffff, 0x00120011, 723 0x9294, 0xffffffff, 0x00000015, 724 0x9298, 0xffffffff, 0x00140013, 725 0x929c, 0xffffffff, 0x00170016, 726 0x9150, 0xffffffff, 0x96940200, 727 0x8708, 0xffffffff, 0x00900100, 728 0xc478, 0xffffffff, 0x00000080, 729 0xc404, 0xffffffff, 0x0020003f, 730 0x30, 0xffffffff, 0x0000001c, 731 0x34, 0x000f0000, 0x000f0000, 732 0x160c, 0xffffffff, 0x00000100, 733 0x1024, 0xffffffff, 0x00000100, 734 0x102c, 0x00000101, 0x00000000, 735 0x20a8, 0xffffffff, 0x00000104, 736 0x264c, 0x000c0000, 0x000c0000, 737 0x2648, 0x000c0000, 0x000c0000, 738 0x55e4, 0xff000fff, 0x00000100, 739 0x55e8, 0x00000001, 0x00000001, 740 0x2f50, 0x00000001, 0x00000001, 741 0x30cc, 0xc0000fff, 0x00000104, 742 0xc1e4, 0x00000001, 0x00000001, 743 0xd0c0, 0xfffffff0, 0x00000100, 744 0xd8c0, 0xfffffff0, 0x00000100 745 }; 746 747 static const u32 pitcairn_mgcg_cgcg_init[] = 748 { 749 0xc400, 0xffffffff, 0xfffffffc, 750 0x802c, 0xffffffff, 0xe0000000, 751 0x9a60, 0xffffffff, 0x00000100, 752 0x92a4, 0xffffffff, 0x00000100, 753 0xc164, 0xffffffff, 0x00000100, 754 0x9774, 0xffffffff, 0x00000100, 755 0x8984, 0xffffffff, 0x06000100, 756 0x8a18, 0xffffffff, 0x00000100, 757 0x92a0, 0xffffffff, 0x00000100, 758 0xc380, 0xffffffff, 0x00000100, 759 0x8b28, 0xffffffff, 0x00000100, 760 0x9144, 0xffffffff, 0x00000100, 761 0x8d88, 0xffffffff, 0x00000100, 762 0x8d8c, 0xffffffff, 0x00000100, 763 0x9030, 0xffffffff, 0x00000100, 764 0x9034, 0xffffffff, 0x00000100, 765 0x9038, 0xffffffff, 0x00000100, 766 0x903c, 0xffffffff, 0x00000100, 767 0xad80, 0xffffffff, 0x00000100, 768 0xac54, 0xffffffff, 0x00000100, 769 0x897c, 0xffffffff, 0x06000100, 770 0x9868, 0xffffffff, 0x00000100, 771 0x9510, 0xffffffff, 0x00000100, 772 0xaf04, 0xffffffff, 0x00000100, 773 0xae04, 0xffffffff, 0x00000100, 774 0x949c, 0xffffffff, 0x00000100, 775 0x802c, 0xffffffff, 0xe0000000, 776 0x9160, 0xffffffff, 0x00010000, 777 0x9164, 0xffffffff, 0x00030002, 778 0x9168, 0xffffffff, 0x00040007, 779 0x916c, 0xffffffff, 0x00060005, 780 0x9170, 0xffffffff, 0x00090008, 781 0x9174, 0xffffffff, 0x00020001, 782 0x9178, 0xffffffff, 0x00040003, 783 0x917c, 0xffffffff, 0x00000007, 784 0x9180, 0xffffffff, 0x00060005, 785 0x9184, 0xffffffff, 0x00090008, 786 0x9188, 0xffffffff, 0x00030002, 787 0x918c, 0xffffffff, 0x00050004, 788 0x9190, 0xffffffff, 0x00000008, 789 0x9194, 0xffffffff, 0x00070006, 790 0x9198, 0xffffffff, 0x000a0009, 791 0x919c, 0xffffffff, 0x00040003, 792 0x91a0, 0xffffffff, 0x00060005, 793 0x91a4, 0xffffffff, 0x00000009, 794 0x91a8, 0xffffffff, 0x00080007, 795 0x91ac, 0xffffffff, 0x000b000a, 796 0x91b0, 0xffffffff, 0x00050004, 797 0x91b4, 0xffffffff, 0x00070006, 798 0x91b8, 0xffffffff, 0x0008000b, 799 0x91bc, 0xffffffff, 0x000a0009, 800 0x91c0, 0xffffffff, 0x000d000c, 801 0x9200, 0xffffffff, 0x00090008, 802 0x9204, 0xffffffff, 0x000b000a, 803 0x9208, 0xffffffff, 0x000c000f, 804 0x920c, 0xffffffff, 0x000e000d, 805 0x9210, 0xffffffff, 0x00110010, 806 0x9214, 0xffffffff, 0x000a0009, 807 0x9218, 0xffffffff, 0x000c000b, 808 0x921c, 0xffffffff, 0x0000000f, 809 0x9220, 0xffffffff, 0x000e000d, 810 0x9224, 0xffffffff, 0x00110010, 811 0x9228, 0xffffffff, 0x000b000a, 812 0x922c, 0xffffffff, 0x000d000c, 813 0x9230, 0xffffffff, 0x00000010, 814 0x9234, 0xffffffff, 0x000f000e, 815 0x9238, 0xffffffff, 0x00120011, 816 0x923c, 0xffffffff, 0x000c000b, 817 0x9240, 0xffffffff, 0x000e000d, 818 0x9244, 0xffffffff, 0x00000011, 819 0x9248, 0xffffffff, 0x0010000f, 820 0x924c, 0xffffffff, 0x00130012, 821 0x9250, 0xffffffff, 0x000d000c, 822 0x9254, 0xffffffff, 0x000f000e, 823 0x9258, 0xffffffff, 0x00100013, 824 0x925c, 0xffffffff, 0x00120011, 825 0x9260, 0xffffffff, 0x00150014, 826 0x9150, 0xffffffff, 0x96940200, 827 0x8708, 0xffffffff, 0x00900100, 828 0xc478, 0xffffffff, 0x00000080, 829 0xc404, 0xffffffff, 0x0020003f, 830 0x30, 0xffffffff, 0x0000001c, 831 0x34, 0x000f0000, 0x000f0000, 832 0x160c, 0xffffffff, 0x00000100, 833 0x1024, 0xffffffff, 0x00000100, 834 0x102c, 0x00000101, 0x00000000, 835 0x20a8, 0xffffffff, 0x00000104, 836 0x55e4, 0xff000fff, 0x00000100, 837 0x55e8, 0x00000001, 0x00000001, 838 0x2f50, 0x00000001, 0x00000001, 839 0x30cc, 0xc0000fff, 0x00000104, 840 0xc1e4, 0x00000001, 0x00000001, 841 0xd0c0, 0xfffffff0, 0x00000100, 842 0xd8c0, 0xfffffff0, 0x00000100 843 }; 844 845 static const u32 verde_mgcg_cgcg_init[] = 846 { 847 0xc400, 0xffffffff, 0xfffffffc, 848 0x802c, 0xffffffff, 0xe0000000, 849 0x9a60, 0xffffffff, 0x00000100, 850 0x92a4, 0xffffffff, 0x00000100, 851 0xc164, 0xffffffff, 0x00000100, 852 0x9774, 0xffffffff, 0x00000100, 853 0x8984, 0xffffffff, 0x06000100, 854 0x8a18, 0xffffffff, 0x00000100, 855 0x92a0, 0xffffffff, 0x00000100, 856 0xc380, 0xffffffff, 0x00000100, 857 0x8b28, 0xffffffff, 0x00000100, 858 0x9144, 0xffffffff, 0x00000100, 859 0x8d88, 0xffffffff, 0x00000100, 860 0x8d8c, 0xffffffff, 0x00000100, 861 0x9030, 0xffffffff, 0x00000100, 862 0x9034, 0xffffffff, 0x00000100, 863 0x9038, 0xffffffff, 0x00000100, 864 0x903c, 0xffffffff, 0x00000100, 865 0xad80, 0xffffffff, 0x00000100, 866 0xac54, 0xffffffff, 0x00000100, 867 0x897c, 0xffffffff, 0x06000100, 868 0x9868, 0xffffffff, 0x00000100, 869 0x9510, 0xffffffff, 0x00000100, 870 0xaf04, 0xffffffff, 0x00000100, 871 0xae04, 0xffffffff, 0x00000100, 872 0x949c, 0xffffffff, 0x00000100, 873 0x802c, 0xffffffff, 0xe0000000, 874 0x9160, 0xffffffff, 0x00010000, 875 0x9164, 0xffffffff, 0x00030002, 876 0x9168, 0xffffffff, 0x00040007, 877 0x916c, 0xffffffff, 0x00060005, 878 0x9170, 0xffffffff, 0x00090008, 879 0x9174, 0xffffffff, 0x00020001, 880 0x9178, 0xffffffff, 0x00040003, 881 0x917c, 0xffffffff, 0x00000007, 882 0x9180, 0xffffffff, 0x00060005, 883 0x9184, 0xffffffff, 0x00090008, 884 0x9188, 0xffffffff, 0x00030002, 885 0x918c, 0xffffffff, 0x00050004, 886 0x9190, 0xffffffff, 0x00000008, 887 0x9194, 0xffffffff, 0x00070006, 888 0x9198, 0xffffffff, 0x000a0009, 889 0x919c, 0xffffffff, 0x00040003, 890 0x91a0, 0xffffffff, 0x00060005, 891 0x91a4, 0xffffffff, 0x00000009, 892 0x91a8, 0xffffffff, 0x00080007, 893 0x91ac, 0xffffffff, 0x000b000a, 894 0x91b0, 0xffffffff, 0x00050004, 895 0x91b4, 0xffffffff, 0x00070006, 896 0x91b8, 0xffffffff, 0x0008000b, 897 0x91bc, 0xffffffff, 0x000a0009, 898 0x91c0, 0xffffffff, 0x000d000c, 899 0x9200, 0xffffffff, 0x00090008, 900 0x9204, 0xffffffff, 0x000b000a, 901 0x9208, 0xffffffff, 0x000c000f, 902 0x920c, 0xffffffff, 0x000e000d, 903 0x9210, 0xffffffff, 0x00110010, 904 0x9214, 0xffffffff, 0x000a0009, 905 0x9218, 0xffffffff, 0x000c000b, 906 0x921c, 0xffffffff, 0x0000000f, 907 0x9220, 0xffffffff, 0x000e000d, 908 0x9224, 0xffffffff, 0x00110010, 909 0x9228, 0xffffffff, 0x000b000a, 910 0x922c, 0xffffffff, 0x000d000c, 911 0x9230, 0xffffffff, 0x00000010, 912 0x9234, 0xffffffff, 0x000f000e, 913 0x9238, 0xffffffff, 0x00120011, 914 0x923c, 0xffffffff, 0x000c000b, 915 0x9240, 0xffffffff, 0x000e000d, 916 0x9244, 0xffffffff, 0x00000011, 917 0x9248, 0xffffffff, 0x0010000f, 918 0x924c, 0xffffffff, 0x00130012, 919 0x9250, 0xffffffff, 0x000d000c, 920 0x9254, 0xffffffff, 0x000f000e, 921 0x9258, 0xffffffff, 0x00100013, 922 0x925c, 0xffffffff, 0x00120011, 923 0x9260, 0xffffffff, 0x00150014, 924 0x9150, 0xffffffff, 0x96940200, 925 0x8708, 0xffffffff, 0x00900100, 926 0xc478, 0xffffffff, 0x00000080, 927 0xc404, 0xffffffff, 0x0020003f, 928 0x30, 0xffffffff, 0x0000001c, 929 0x34, 0x000f0000, 0x000f0000, 930 0x160c, 0xffffffff, 0x00000100, 931 0x1024, 0xffffffff, 0x00000100, 932 0x102c, 0x00000101, 0x00000000, 933 0x20a8, 0xffffffff, 0x00000104, 934 0x264c, 0x000c0000, 0x000c0000, 935 0x2648, 0x000c0000, 0x000c0000, 936 0x55e4, 0xff000fff, 0x00000100, 937 0x55e8, 0x00000001, 0x00000001, 938 0x2f50, 0x00000001, 0x00000001, 939 0x30cc, 0xc0000fff, 0x00000104, 940 0xc1e4, 0x00000001, 0x00000001, 941 0xd0c0, 0xfffffff0, 0x00000100, 942 0xd8c0, 0xfffffff0, 0x00000100 943 }; 944 945 static const u32 oland_mgcg_cgcg_init[] = 946 { 947 0xc400, 0xffffffff, 0xfffffffc, 948 0x802c, 0xffffffff, 0xe0000000, 949 0x9a60, 0xffffffff, 0x00000100, 950 0x92a4, 0xffffffff, 0x00000100, 951 0xc164, 0xffffffff, 0x00000100, 952 0x9774, 0xffffffff, 0x00000100, 953 0x8984, 0xffffffff, 0x06000100, 954 0x8a18, 0xffffffff, 0x00000100, 955 0x92a0, 0xffffffff, 0x00000100, 956 0xc380, 0xffffffff, 0x00000100, 957 0x8b28, 0xffffffff, 0x00000100, 958 0x9144, 0xffffffff, 0x00000100, 959 0x8d88, 0xffffffff, 0x00000100, 960 0x8d8c, 0xffffffff, 0x00000100, 961 0x9030, 0xffffffff, 0x00000100, 962 0x9034, 0xffffffff, 0x00000100, 963 0x9038, 0xffffffff, 0x00000100, 964 0x903c, 0xffffffff, 0x00000100, 965 0xad80, 0xffffffff, 0x00000100, 966 0xac54, 0xffffffff, 0x00000100, 967 0x897c, 0xffffffff, 0x06000100, 968 0x9868, 0xffffffff, 0x00000100, 969 0x9510, 0xffffffff, 0x00000100, 970 0xaf04, 0xffffffff, 0x00000100, 971 0xae04, 0xffffffff, 0x00000100, 972 0x949c, 0xffffffff, 0x00000100, 973 0x802c, 0xffffffff, 0xe0000000, 974 0x9160, 0xffffffff, 0x00010000, 975 0x9164, 0xffffffff, 0x00030002, 976 0x9168, 0xffffffff, 0x00040007, 977 0x916c, 0xffffffff, 0x00060005, 978 0x9170, 0xffffffff, 0x00090008, 979 0x9174, 0xffffffff, 0x00020001, 980 0x9178, 0xffffffff, 0x00040003, 981 0x917c, 0xffffffff, 0x00000007, 982 0x9180, 0xffffffff, 0x00060005, 983 0x9184, 0xffffffff, 0x00090008, 984 0x9188, 0xffffffff, 0x00030002, 985 0x918c, 0xffffffff, 0x00050004, 986 0x9190, 0xffffffff, 0x00000008, 987 0x9194, 0xffffffff, 0x00070006, 988 0x9198, 0xffffffff, 0x000a0009, 989 0x919c, 0xffffffff, 0x00040003, 990 0x91a0, 0xffffffff, 0x00060005, 991 0x91a4, 0xffffffff, 0x00000009, 992 0x91a8, 0xffffffff, 0x00080007, 993 0x91ac, 0xffffffff, 0x000b000a, 994 0x91b0, 0xffffffff, 0x00050004, 995 0x91b4, 0xffffffff, 0x00070006, 996 0x91b8, 0xffffffff, 0x0008000b, 997 0x91bc, 0xffffffff, 0x000a0009, 998 0x91c0, 0xffffffff, 0x000d000c, 999 0x91c4, 0xffffffff, 0x00060005, 1000 0x91c8, 0xffffffff, 0x00080007, 1001 0x91cc, 0xffffffff, 0x0000000b, 1002 0x91d0, 0xffffffff, 0x000a0009, 1003 0x91d4, 0xffffffff, 0x000d000c, 1004 0x9150, 0xffffffff, 0x96940200, 1005 0x8708, 0xffffffff, 0x00900100, 1006 0xc478, 0xffffffff, 0x00000080, 1007 0xc404, 0xffffffff, 0x0020003f, 1008 0x30, 0xffffffff, 0x0000001c, 1009 0x34, 0x000f0000, 0x000f0000, 1010 0x160c, 0xffffffff, 0x00000100, 1011 0x1024, 0xffffffff, 0x00000100, 1012 0x102c, 0x00000101, 0x00000000, 1013 0x20a8, 0xffffffff, 0x00000104, 1014 0x264c, 0x000c0000, 0x000c0000, 1015 0x2648, 0x000c0000, 0x000c0000, 1016 0x55e4, 0xff000fff, 0x00000100, 1017 0x55e8, 0x00000001, 0x00000001, 1018 0x2f50, 0x00000001, 0x00000001, 1019 0x30cc, 0xc0000fff, 0x00000104, 1020 0xc1e4, 0x00000001, 0x00000001, 1021 0xd0c0, 0xfffffff0, 0x00000100, 1022 0xd8c0, 0xfffffff0, 0x00000100 1023 }; 1024 1025 static const u32 hainan_mgcg_cgcg_init[] = 1026 { 1027 0xc400, 0xffffffff, 0xfffffffc, 1028 0x802c, 0xffffffff, 0xe0000000, 1029 0x9a60, 0xffffffff, 0x00000100, 1030 0x92a4, 0xffffffff, 0x00000100, 1031 0xc164, 0xffffffff, 0x00000100, 1032 0x9774, 0xffffffff, 0x00000100, 1033 0x8984, 0xffffffff, 0x06000100, 1034 0x8a18, 0xffffffff, 0x00000100, 1035 0x92a0, 0xffffffff, 0x00000100, 1036 0xc380, 0xffffffff, 0x00000100, 1037 0x8b28, 0xffffffff, 0x00000100, 1038 0x9144, 0xffffffff, 0x00000100, 1039 0x8d88, 0xffffffff, 0x00000100, 1040 0x8d8c, 0xffffffff, 0x00000100, 1041 0x9030, 0xffffffff, 0x00000100, 1042 0x9034, 0xffffffff, 0x00000100, 1043 0x9038, 0xffffffff, 0x00000100, 1044 0x903c, 0xffffffff, 0x00000100, 1045 0xad80, 0xffffffff, 0x00000100, 1046 0xac54, 0xffffffff, 0x00000100, 1047 0x897c, 0xffffffff, 0x06000100, 1048 0x9868, 0xffffffff, 0x00000100, 1049 0x9510, 0xffffffff, 0x00000100, 1050 0xaf04, 0xffffffff, 0x00000100, 1051 0xae04, 0xffffffff, 0x00000100, 1052 0x949c, 0xffffffff, 0x00000100, 1053 0x802c, 0xffffffff, 0xe0000000, 1054 0x9160, 0xffffffff, 0x00010000, 1055 0x9164, 0xffffffff, 0x00030002, 1056 0x9168, 0xffffffff, 0x00040007, 1057 0x916c, 0xffffffff, 0x00060005, 1058 0x9170, 0xffffffff, 0x00090008, 1059 0x9174, 0xffffffff, 0x00020001, 1060 0x9178, 0xffffffff, 0x00040003, 1061 0x917c, 0xffffffff, 0x00000007, 1062 0x9180, 0xffffffff, 0x00060005, 1063 0x9184, 0xffffffff, 0x00090008, 1064 0x9188, 0xffffffff, 0x00030002, 1065 0x918c, 0xffffffff, 0x00050004, 1066 0x9190, 0xffffffff, 0x00000008, 1067 0x9194, 0xffffffff, 0x00070006, 1068 0x9198, 0xffffffff, 0x000a0009, 1069 0x919c, 0xffffffff, 0x00040003, 1070 0x91a0, 0xffffffff, 0x00060005, 1071 0x91a4, 0xffffffff, 0x00000009, 1072 0x91a8, 0xffffffff, 0x00080007, 1073 0x91ac, 0xffffffff, 0x000b000a, 1074 0x91b0, 0xffffffff, 0x00050004, 1075 0x91b4, 0xffffffff, 0x00070006, 1076 0x91b8, 0xffffffff, 0x0008000b, 1077 0x91bc, 0xffffffff, 0x000a0009, 1078 0x91c0, 0xffffffff, 0x000d000c, 1079 0x91c4, 0xffffffff, 0x00060005, 1080 0x91c8, 0xffffffff, 0x00080007, 1081 0x91cc, 0xffffffff, 0x0000000b, 1082 0x91d0, 0xffffffff, 0x000a0009, 1083 0x91d4, 0xffffffff, 0x000d000c, 1084 0x9150, 0xffffffff, 0x96940200, 1085 0x8708, 0xffffffff, 0x00900100, 1086 0xc478, 0xffffffff, 0x00000080, 1087 0xc404, 0xffffffff, 0x0020003f, 1088 0x30, 0xffffffff, 0x0000001c, 1089 0x34, 0x000f0000, 0x000f0000, 1090 0x160c, 0xffffffff, 0x00000100, 1091 0x1024, 0xffffffff, 0x00000100, 1092 0x20a8, 0xffffffff, 0x00000104, 1093 0x264c, 0x000c0000, 0x000c0000, 1094 0x2648, 0x000c0000, 0x000c0000, 1095 0x2f50, 0x00000001, 0x00000001, 1096 0x30cc, 0xc0000fff, 0x00000104, 1097 0xc1e4, 0x00000001, 0x00000001, 1098 0xd0c0, 0xfffffff0, 0x00000100, 1099 0xd8c0, 0xfffffff0, 0x00000100 1100 }; 1101 1102 static u32 verde_pg_init[] = 1103 { 1104 0x353c, 0xffffffff, 0x40000, 1105 0x3538, 0xffffffff, 0x200010ff, 1106 0x353c, 0xffffffff, 0x0, 1107 0x353c, 0xffffffff, 0x0, 1108 0x353c, 0xffffffff, 0x0, 1109 0x353c, 0xffffffff, 0x0, 1110 0x353c, 0xffffffff, 0x0, 1111 0x353c, 0xffffffff, 0x7007, 1112 0x3538, 0xffffffff, 0x300010ff, 1113 0x353c, 0xffffffff, 0x0, 1114 0x353c, 0xffffffff, 0x0, 1115 0x353c, 0xffffffff, 0x0, 1116 0x353c, 0xffffffff, 0x0, 1117 0x353c, 0xffffffff, 0x0, 1118 0x353c, 0xffffffff, 0x400000, 1119 0x3538, 0xffffffff, 0x100010ff, 1120 0x353c, 0xffffffff, 0x0, 1121 0x353c, 0xffffffff, 0x0, 1122 0x353c, 0xffffffff, 0x0, 1123 0x353c, 0xffffffff, 0x0, 1124 0x353c, 0xffffffff, 0x0, 1125 0x353c, 0xffffffff, 0x120200, 1126 0x3538, 0xffffffff, 0x500010ff, 1127 0x353c, 0xffffffff, 0x0, 1128 0x353c, 0xffffffff, 0x0, 1129 0x353c, 0xffffffff, 0x0, 1130 0x353c, 0xffffffff, 0x0, 1131 0x353c, 0xffffffff, 0x0, 1132 0x353c, 0xffffffff, 0x1e1e16, 1133 0x3538, 0xffffffff, 0x600010ff, 1134 0x353c, 0xffffffff, 0x0, 1135 0x353c, 0xffffffff, 0x0, 1136 0x353c, 0xffffffff, 0x0, 1137 0x353c, 0xffffffff, 0x0, 1138 0x353c, 0xffffffff, 0x0, 1139 0x353c, 0xffffffff, 0x171f1e, 1140 0x3538, 0xffffffff, 0x700010ff, 1141 0x353c, 0xffffffff, 0x0, 1142 0x353c, 0xffffffff, 0x0, 1143 0x353c, 0xffffffff, 0x0, 1144 0x353c, 0xffffffff, 0x0, 1145 0x353c, 0xffffffff, 0x0, 1146 0x353c, 0xffffffff, 0x0, 1147 0x3538, 0xffffffff, 0x9ff, 1148 0x3500, 0xffffffff, 0x0, 1149 0x3504, 0xffffffff, 0x10000800, 1150 0x3504, 0xffffffff, 0xf, 1151 0x3504, 0xffffffff, 0xf, 1152 0x3500, 0xffffffff, 0x4, 1153 0x3504, 0xffffffff, 0x1000051e, 1154 0x3504, 0xffffffff, 0xffff, 1155 0x3504, 0xffffffff, 0xffff, 1156 0x3500, 0xffffffff, 0x8, 1157 0x3504, 0xffffffff, 0x80500, 1158 0x3500, 0xffffffff, 0x12, 1159 0x3504, 0xffffffff, 0x9050c, 1160 0x3500, 0xffffffff, 0x1d, 1161 0x3504, 0xffffffff, 0xb052c, 1162 0x3500, 0xffffffff, 0x2a, 1163 0x3504, 0xffffffff, 0x1053e, 1164 0x3500, 0xffffffff, 0x2d, 1165 0x3504, 0xffffffff, 0x10546, 1166 0x3500, 0xffffffff, 0x30, 1167 0x3504, 0xffffffff, 0xa054e, 1168 0x3500, 0xffffffff, 0x3c, 1169 0x3504, 0xffffffff, 0x1055f, 1170 0x3500, 0xffffffff, 0x3f, 1171 0x3504, 0xffffffff, 0x10567, 1172 0x3500, 0xffffffff, 0x42, 1173 0x3504, 0xffffffff, 0x1056f, 1174 0x3500, 0xffffffff, 0x45, 1175 0x3504, 0xffffffff, 0x10572, 1176 0x3500, 0xffffffff, 0x48, 1177 0x3504, 0xffffffff, 0x20575, 1178 0x3500, 0xffffffff, 0x4c, 1179 0x3504, 0xffffffff, 0x190801, 1180 0x3500, 0xffffffff, 0x67, 1181 0x3504, 0xffffffff, 0x1082a, 1182 0x3500, 0xffffffff, 0x6a, 1183 0x3504, 0xffffffff, 0x1b082d, 1184 0x3500, 0xffffffff, 0x87, 1185 0x3504, 0xffffffff, 0x310851, 1186 0x3500, 0xffffffff, 0xba, 1187 0x3504, 0xffffffff, 0x891, 1188 0x3500, 0xffffffff, 0xbc, 1189 0x3504, 0xffffffff, 0x893, 1190 0x3500, 0xffffffff, 0xbe, 1191 0x3504, 0xffffffff, 0x20895, 1192 0x3500, 0xffffffff, 0xc2, 1193 0x3504, 0xffffffff, 0x20899, 1194 0x3500, 0xffffffff, 0xc6, 1195 0x3504, 0xffffffff, 0x2089d, 1196 0x3500, 0xffffffff, 0xca, 1197 0x3504, 0xffffffff, 0x8a1, 1198 0x3500, 0xffffffff, 0xcc, 1199 0x3504, 0xffffffff, 0x8a3, 1200 0x3500, 0xffffffff, 0xce, 1201 0x3504, 0xffffffff, 0x308a5, 1202 0x3500, 0xffffffff, 0xd3, 1203 0x3504, 0xffffffff, 0x6d08cd, 1204 0x3500, 0xffffffff, 0x142, 1205 0x3504, 0xffffffff, 0x2000095a, 1206 0x3504, 0xffffffff, 0x1, 1207 0x3500, 0xffffffff, 0x144, 1208 0x3504, 0xffffffff, 0x301f095b, 1209 0x3500, 0xffffffff, 0x165, 1210 0x3504, 0xffffffff, 0xc094d, 1211 0x3500, 0xffffffff, 0x173, 1212 0x3504, 0xffffffff, 0xf096d, 1213 0x3500, 0xffffffff, 0x184, 1214 0x3504, 0xffffffff, 0x15097f, 1215 0x3500, 0xffffffff, 0x19b, 1216 0x3504, 0xffffffff, 0xc0998, 1217 0x3500, 0xffffffff, 0x1a9, 1218 0x3504, 0xffffffff, 0x409a7, 1219 0x3500, 0xffffffff, 0x1af, 1220 0x3504, 0xffffffff, 0xcdc, 1221 0x3500, 0xffffffff, 0x1b1, 1222 0x3504, 0xffffffff, 0x800, 1223 0x3508, 0xffffffff, 0x6c9b2000, 1224 0x3510, 0xfc00, 0x2000, 1225 0x3544, 0xffffffff, 0xfc0, 1226 0x28d4, 0x00000100, 0x100 1227 }; 1228 1229 static void si_init_golden_registers(struct radeon_device *rdev) 1230 { 1231 switch (rdev->family) { 1232 case CHIP_TAHITI: 1233 radeon_program_register_sequence(rdev, 1234 tahiti_golden_registers, 1235 (const u32)ARRAY_SIZE(tahiti_golden_registers)); 1236 radeon_program_register_sequence(rdev, 1237 tahiti_golden_rlc_registers, 1238 (const u32)ARRAY_SIZE(tahiti_golden_rlc_registers)); 1239 radeon_program_register_sequence(rdev, 1240 tahiti_mgcg_cgcg_init, 1241 (const u32)ARRAY_SIZE(tahiti_mgcg_cgcg_init)); 1242 radeon_program_register_sequence(rdev, 1243 tahiti_golden_registers2, 1244 (const u32)ARRAY_SIZE(tahiti_golden_registers2)); 1245 break; 1246 case CHIP_PITCAIRN: 1247 radeon_program_register_sequence(rdev, 1248 pitcairn_golden_registers, 1249 (const u32)ARRAY_SIZE(pitcairn_golden_registers)); 1250 radeon_program_register_sequence(rdev, 1251 pitcairn_golden_rlc_registers, 1252 (const u32)ARRAY_SIZE(pitcairn_golden_rlc_registers)); 1253 radeon_program_register_sequence(rdev, 1254 pitcairn_mgcg_cgcg_init, 1255 (const u32)ARRAY_SIZE(pitcairn_mgcg_cgcg_init)); 1256 break; 1257 case CHIP_VERDE: 1258 radeon_program_register_sequence(rdev, 1259 verde_golden_registers, 1260 (const u32)ARRAY_SIZE(verde_golden_registers)); 1261 radeon_program_register_sequence(rdev, 1262 verde_golden_rlc_registers, 1263 (const u32)ARRAY_SIZE(verde_golden_rlc_registers)); 1264 radeon_program_register_sequence(rdev, 1265 verde_mgcg_cgcg_init, 1266 (const u32)ARRAY_SIZE(verde_mgcg_cgcg_init)); 1267 radeon_program_register_sequence(rdev, 1268 verde_pg_init, 1269 (const u32)ARRAY_SIZE(verde_pg_init)); 1270 break; 1271 case CHIP_OLAND: 1272 radeon_program_register_sequence(rdev, 1273 oland_golden_registers, 1274 (const u32)ARRAY_SIZE(oland_golden_registers)); 1275 radeon_program_register_sequence(rdev, 1276 oland_golden_rlc_registers, 1277 (const u32)ARRAY_SIZE(oland_golden_rlc_registers)); 1278 radeon_program_register_sequence(rdev, 1279 oland_mgcg_cgcg_init, 1280 (const u32)ARRAY_SIZE(oland_mgcg_cgcg_init)); 1281 break; 1282 case CHIP_HAINAN: 1283 radeon_program_register_sequence(rdev, 1284 hainan_golden_registers, 1285 (const u32)ARRAY_SIZE(hainan_golden_registers)); 1286 radeon_program_register_sequence(rdev, 1287 hainan_golden_registers2, 1288 (const u32)ARRAY_SIZE(hainan_golden_registers2)); 1289 radeon_program_register_sequence(rdev, 1290 hainan_mgcg_cgcg_init, 1291 (const u32)ARRAY_SIZE(hainan_mgcg_cgcg_init)); 1292 break; 1293 default: 1294 break; 1295 } 1296 } 1297 1298 /** 1299 * si_get_allowed_info_register - fetch the register for the info ioctl 1300 * 1301 * @rdev: radeon_device pointer 1302 * @reg: register offset in bytes 1303 * @val: register value 1304 * 1305 * Returns 0 for success or -EINVAL for an invalid register 1306 * 1307 */ 1308 int si_get_allowed_info_register(struct radeon_device *rdev, 1309 u32 reg, u32 *val) 1310 { 1311 switch (reg) { 1312 case GRBM_STATUS: 1313 case GRBM_STATUS2: 1314 case GRBM_STATUS_SE0: 1315 case GRBM_STATUS_SE1: 1316 case SRBM_STATUS: 1317 case SRBM_STATUS2: 1318 case (DMA_STATUS_REG + DMA0_REGISTER_OFFSET): 1319 case (DMA_STATUS_REG + DMA1_REGISTER_OFFSET): 1320 case UVD_STATUS: 1321 *val = RREG32(reg); 1322 return 0; 1323 default: 1324 return -EINVAL; 1325 } 1326 } 1327 1328 #define PCIE_BUS_CLK 10000 1329 #define TCLK (PCIE_BUS_CLK / 10) 1330 1331 /** 1332 * si_get_xclk - get the xclk 1333 * 1334 * @rdev: radeon_device pointer 1335 * 1336 * Returns the reference clock used by the gfx engine 1337 * (SI). 1338 */ 1339 u32 si_get_xclk(struct radeon_device *rdev) 1340 { 1341 u32 reference_clock = rdev->clock.spll.reference_freq; 1342 u32 tmp; 1343 1344 tmp = RREG32(CG_CLKPIN_CNTL_2); 1345 if (tmp & MUX_TCLK_TO_XCLK) 1346 return TCLK; 1347 1348 tmp = RREG32(CG_CLKPIN_CNTL); 1349 if (tmp & XTALIN_DIVIDE) 1350 return reference_clock / 4; 1351 1352 return reference_clock; 1353 } 1354 1355 /* get temperature in millidegrees */ 1356 int si_get_temp(struct radeon_device *rdev) 1357 { 1358 u32 temp; 1359 int actual_temp = 0; 1360 1361 temp = (RREG32(CG_MULT_THERMAL_STATUS) & CTF_TEMP_MASK) >> 1362 CTF_TEMP_SHIFT; 1363 1364 if (temp & 0x200) 1365 actual_temp = 255; 1366 else 1367 actual_temp = temp & 0x1ff; 1368 1369 actual_temp = (actual_temp * 1000); 1370 1371 return actual_temp; 1372 } 1373 1374 #define TAHITI_IO_MC_REGS_SIZE 36 1375 1376 static const u32 tahiti_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = { 1377 {0x0000006f, 0x03044000}, 1378 {0x00000070, 0x0480c018}, 1379 {0x00000071, 0x00000040}, 1380 {0x00000072, 0x01000000}, 1381 {0x00000074, 0x000000ff}, 1382 {0x00000075, 0x00143400}, 1383 {0x00000076, 0x08ec0800}, 1384 {0x00000077, 0x040000cc}, 1385 {0x00000079, 0x00000000}, 1386 {0x0000007a, 0x21000409}, 1387 {0x0000007c, 0x00000000}, 1388 {0x0000007d, 0xe8000000}, 1389 {0x0000007e, 0x044408a8}, 1390 {0x0000007f, 0x00000003}, 1391 {0x00000080, 0x00000000}, 1392 {0x00000081, 0x01000000}, 1393 {0x00000082, 0x02000000}, 1394 {0x00000083, 0x00000000}, 1395 {0x00000084, 0xe3f3e4f4}, 1396 {0x00000085, 0x00052024}, 1397 {0x00000087, 0x00000000}, 1398 {0x00000088, 0x66036603}, 1399 {0x00000089, 0x01000000}, 1400 {0x0000008b, 0x1c0a0000}, 1401 {0x0000008c, 0xff010000}, 1402 {0x0000008e, 0xffffefff}, 1403 {0x0000008f, 0xfff3efff}, 1404 {0x00000090, 0xfff3efbf}, 1405 {0x00000094, 0x00101101}, 1406 {0x00000095, 0x00000fff}, 1407 {0x00000096, 0x00116fff}, 1408 {0x00000097, 0x60010000}, 1409 {0x00000098, 0x10010000}, 1410 {0x00000099, 0x00006000}, 1411 {0x0000009a, 0x00001000}, 1412 {0x0000009f, 0x00a77400} 1413 }; 1414 1415 static const u32 pitcairn_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = { 1416 {0x0000006f, 0x03044000}, 1417 {0x00000070, 0x0480c018}, 1418 {0x00000071, 0x00000040}, 1419 {0x00000072, 0x01000000}, 1420 {0x00000074, 0x000000ff}, 1421 {0x00000075, 0x00143400}, 1422 {0x00000076, 0x08ec0800}, 1423 {0x00000077, 0x040000cc}, 1424 {0x00000079, 0x00000000}, 1425 {0x0000007a, 0x21000409}, 1426 {0x0000007c, 0x00000000}, 1427 {0x0000007d, 0xe8000000}, 1428 {0x0000007e, 0x044408a8}, 1429 {0x0000007f, 0x00000003}, 1430 {0x00000080, 0x00000000}, 1431 {0x00000081, 0x01000000}, 1432 {0x00000082, 0x02000000}, 1433 {0x00000083, 0x00000000}, 1434 {0x00000084, 0xe3f3e4f4}, 1435 {0x00000085, 0x00052024}, 1436 {0x00000087, 0x00000000}, 1437 {0x00000088, 0x66036603}, 1438 {0x00000089, 0x01000000}, 1439 {0x0000008b, 0x1c0a0000}, 1440 {0x0000008c, 0xff010000}, 1441 {0x0000008e, 0xffffefff}, 1442 {0x0000008f, 0xfff3efff}, 1443 {0x00000090, 0xfff3efbf}, 1444 {0x00000094, 0x00101101}, 1445 {0x00000095, 0x00000fff}, 1446 {0x00000096, 0x00116fff}, 1447 {0x00000097, 0x60010000}, 1448 {0x00000098, 0x10010000}, 1449 {0x00000099, 0x00006000}, 1450 {0x0000009a, 0x00001000}, 1451 {0x0000009f, 0x00a47400} 1452 }; 1453 1454 static const u32 verde_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = { 1455 {0x0000006f, 0x03044000}, 1456 {0x00000070, 0x0480c018}, 1457 {0x00000071, 0x00000040}, 1458 {0x00000072, 0x01000000}, 1459 {0x00000074, 0x000000ff}, 1460 {0x00000075, 0x00143400}, 1461 {0x00000076, 0x08ec0800}, 1462 {0x00000077, 0x040000cc}, 1463 {0x00000079, 0x00000000}, 1464 {0x0000007a, 0x21000409}, 1465 {0x0000007c, 0x00000000}, 1466 {0x0000007d, 0xe8000000}, 1467 {0x0000007e, 0x044408a8}, 1468 {0x0000007f, 0x00000003}, 1469 {0x00000080, 0x00000000}, 1470 {0x00000081, 0x01000000}, 1471 {0x00000082, 0x02000000}, 1472 {0x00000083, 0x00000000}, 1473 {0x00000084, 0xe3f3e4f4}, 1474 {0x00000085, 0x00052024}, 1475 {0x00000087, 0x00000000}, 1476 {0x00000088, 0x66036603}, 1477 {0x00000089, 0x01000000}, 1478 {0x0000008b, 0x1c0a0000}, 1479 {0x0000008c, 0xff010000}, 1480 {0x0000008e, 0xffffefff}, 1481 {0x0000008f, 0xfff3efff}, 1482 {0x00000090, 0xfff3efbf}, 1483 {0x00000094, 0x00101101}, 1484 {0x00000095, 0x00000fff}, 1485 {0x00000096, 0x00116fff}, 1486 {0x00000097, 0x60010000}, 1487 {0x00000098, 0x10010000}, 1488 {0x00000099, 0x00006000}, 1489 {0x0000009a, 0x00001000}, 1490 {0x0000009f, 0x00a37400} 1491 }; 1492 1493 static const u32 oland_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = { 1494 {0x0000006f, 0x03044000}, 1495 {0x00000070, 0x0480c018}, 1496 {0x00000071, 0x00000040}, 1497 {0x00000072, 0x01000000}, 1498 {0x00000074, 0x000000ff}, 1499 {0x00000075, 0x00143400}, 1500 {0x00000076, 0x08ec0800}, 1501 {0x00000077, 0x040000cc}, 1502 {0x00000079, 0x00000000}, 1503 {0x0000007a, 0x21000409}, 1504 {0x0000007c, 0x00000000}, 1505 {0x0000007d, 0xe8000000}, 1506 {0x0000007e, 0x044408a8}, 1507 {0x0000007f, 0x00000003}, 1508 {0x00000080, 0x00000000}, 1509 {0x00000081, 0x01000000}, 1510 {0x00000082, 0x02000000}, 1511 {0x00000083, 0x00000000}, 1512 {0x00000084, 0xe3f3e4f4}, 1513 {0x00000085, 0x00052024}, 1514 {0x00000087, 0x00000000}, 1515 {0x00000088, 0x66036603}, 1516 {0x00000089, 0x01000000}, 1517 {0x0000008b, 0x1c0a0000}, 1518 {0x0000008c, 0xff010000}, 1519 {0x0000008e, 0xffffefff}, 1520 {0x0000008f, 0xfff3efff}, 1521 {0x00000090, 0xfff3efbf}, 1522 {0x00000094, 0x00101101}, 1523 {0x00000095, 0x00000fff}, 1524 {0x00000096, 0x00116fff}, 1525 {0x00000097, 0x60010000}, 1526 {0x00000098, 0x10010000}, 1527 {0x00000099, 0x00006000}, 1528 {0x0000009a, 0x00001000}, 1529 {0x0000009f, 0x00a17730} 1530 }; 1531 1532 static const u32 hainan_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = { 1533 {0x0000006f, 0x03044000}, 1534 {0x00000070, 0x0480c018}, 1535 {0x00000071, 0x00000040}, 1536 {0x00000072, 0x01000000}, 1537 {0x00000074, 0x000000ff}, 1538 {0x00000075, 0x00143400}, 1539 {0x00000076, 0x08ec0800}, 1540 {0x00000077, 0x040000cc}, 1541 {0x00000079, 0x00000000}, 1542 {0x0000007a, 0x21000409}, 1543 {0x0000007c, 0x00000000}, 1544 {0x0000007d, 0xe8000000}, 1545 {0x0000007e, 0x044408a8}, 1546 {0x0000007f, 0x00000003}, 1547 {0x00000080, 0x00000000}, 1548 {0x00000081, 0x01000000}, 1549 {0x00000082, 0x02000000}, 1550 {0x00000083, 0x00000000}, 1551 {0x00000084, 0xe3f3e4f4}, 1552 {0x00000085, 0x00052024}, 1553 {0x00000087, 0x00000000}, 1554 {0x00000088, 0x66036603}, 1555 {0x00000089, 0x01000000}, 1556 {0x0000008b, 0x1c0a0000}, 1557 {0x0000008c, 0xff010000}, 1558 {0x0000008e, 0xffffefff}, 1559 {0x0000008f, 0xfff3efff}, 1560 {0x00000090, 0xfff3efbf}, 1561 {0x00000094, 0x00101101}, 1562 {0x00000095, 0x00000fff}, 1563 {0x00000096, 0x00116fff}, 1564 {0x00000097, 0x60010000}, 1565 {0x00000098, 0x10010000}, 1566 {0x00000099, 0x00006000}, 1567 {0x0000009a, 0x00001000}, 1568 {0x0000009f, 0x00a07730} 1569 }; 1570 1571 /* ucode loading */ 1572 int si_mc_load_microcode(struct radeon_device *rdev) 1573 { 1574 const __be32 *fw_data = NULL; 1575 const __le32 *new_fw_data = NULL; 1576 u32 running; 1577 u32 *io_mc_regs = NULL; 1578 const __le32 *new_io_mc_regs = NULL; 1579 int i, regs_size, ucode_size; 1580 1581 if (!rdev->mc_fw) 1582 return -EINVAL; 1583 1584 if (rdev->new_fw) { 1585 const struct mc_firmware_header_v1_0 *hdr = 1586 (const struct mc_firmware_header_v1_0 *)rdev->mc_fw->data; 1587 1588 radeon_ucode_print_mc_hdr(&hdr->header); 1589 regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2); 1590 new_io_mc_regs = (const __le32 *) 1591 (rdev->mc_fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes)); 1592 ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; 1593 new_fw_data = (const __le32 *) 1594 (rdev->mc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 1595 } else { 1596 ucode_size = rdev->mc_fw->size / 4; 1597 1598 switch (rdev->family) { 1599 case CHIP_TAHITI: 1600 io_mc_regs = (u32 *)&tahiti_io_mc_regs; 1601 regs_size = TAHITI_IO_MC_REGS_SIZE; 1602 break; 1603 case CHIP_PITCAIRN: 1604 io_mc_regs = (u32 *)&pitcairn_io_mc_regs; 1605 regs_size = TAHITI_IO_MC_REGS_SIZE; 1606 break; 1607 case CHIP_VERDE: 1608 default: 1609 io_mc_regs = (u32 *)&verde_io_mc_regs; 1610 regs_size = TAHITI_IO_MC_REGS_SIZE; 1611 break; 1612 case CHIP_OLAND: 1613 io_mc_regs = (u32 *)&oland_io_mc_regs; 1614 regs_size = TAHITI_IO_MC_REGS_SIZE; 1615 break; 1616 case CHIP_HAINAN: 1617 io_mc_regs = (u32 *)&hainan_io_mc_regs; 1618 regs_size = TAHITI_IO_MC_REGS_SIZE; 1619 break; 1620 } 1621 fw_data = (const __be32 *)rdev->mc_fw->data; 1622 } 1623 1624 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK; 1625 1626 if (running == 0) { 1627 /* reset the engine and set to writable */ 1628 WREG32(MC_SEQ_SUP_CNTL, 0x00000008); 1629 WREG32(MC_SEQ_SUP_CNTL, 0x00000010); 1630 1631 /* load mc io regs */ 1632 for (i = 0; i < regs_size; i++) { 1633 if (rdev->new_fw) { 1634 WREG32(MC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(new_io_mc_regs++)); 1635 WREG32(MC_SEQ_IO_DEBUG_DATA, le32_to_cpup(new_io_mc_regs++)); 1636 } else { 1637 WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]); 1638 WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]); 1639 } 1640 } 1641 /* load the MC ucode */ 1642 for (i = 0; i < ucode_size; i++) { 1643 if (rdev->new_fw) 1644 WREG32(MC_SEQ_SUP_PGM, le32_to_cpup(new_fw_data++)); 1645 else 1646 WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++)); 1647 } 1648 1649 /* put the engine back into the active state */ 1650 WREG32(MC_SEQ_SUP_CNTL, 0x00000008); 1651 WREG32(MC_SEQ_SUP_CNTL, 0x00000004); 1652 WREG32(MC_SEQ_SUP_CNTL, 0x00000001); 1653 1654 /* wait for training to complete */ 1655 for (i = 0; i < rdev->usec_timeout; i++) { 1656 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D0) 1657 break; 1658 udelay(1); 1659 } 1660 for (i = 0; i < rdev->usec_timeout; i++) { 1661 if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D1) 1662 break; 1663 udelay(1); 1664 } 1665 } 1666 1667 return 0; 1668 } 1669 1670 static int si_init_microcode(struct radeon_device *rdev) 1671 { 1672 const char *chip_name; 1673 const char *new_chip_name; 1674 size_t pfp_req_size, me_req_size, ce_req_size, rlc_req_size, mc_req_size; 1675 size_t smc_req_size, mc2_req_size; 1676 char fw_name[30]; 1677 int err; 1678 int new_fw = 0; 1679 bool new_smc = false; 1680 bool si58_fw = false; 1681 bool banks2_fw = false; 1682 1683 DRM_DEBUG("\n"); 1684 1685 switch (rdev->family) { 1686 case CHIP_TAHITI: 1687 chip_name = "TAHITI"; 1688 new_chip_name = "tahiti"; 1689 pfp_req_size = SI_PFP_UCODE_SIZE * 4; 1690 me_req_size = SI_PM4_UCODE_SIZE * 4; 1691 ce_req_size = SI_CE_UCODE_SIZE * 4; 1692 rlc_req_size = SI_RLC_UCODE_SIZE * 4; 1693 mc_req_size = SI_MC_UCODE_SIZE * 4; 1694 mc2_req_size = TAHITI_MC_UCODE_SIZE * 4; 1695 smc_req_size = ALIGN(TAHITI_SMC_UCODE_SIZE, 4); 1696 break; 1697 case CHIP_PITCAIRN: 1698 chip_name = "PITCAIRN"; 1699 if ((rdev->pdev->revision == 0x81) && 1700 ((rdev->pdev->device == 0x6810) || 1701 (rdev->pdev->device == 0x6811))) 1702 new_smc = true; 1703 new_chip_name = "pitcairn"; 1704 pfp_req_size = SI_PFP_UCODE_SIZE * 4; 1705 me_req_size = SI_PM4_UCODE_SIZE * 4; 1706 ce_req_size = SI_CE_UCODE_SIZE * 4; 1707 rlc_req_size = SI_RLC_UCODE_SIZE * 4; 1708 mc_req_size = SI_MC_UCODE_SIZE * 4; 1709 mc2_req_size = PITCAIRN_MC_UCODE_SIZE * 4; 1710 smc_req_size = ALIGN(PITCAIRN_SMC_UCODE_SIZE, 4); 1711 break; 1712 case CHIP_VERDE: 1713 chip_name = "VERDE"; 1714 if (((rdev->pdev->device == 0x6820) && 1715 ((rdev->pdev->revision == 0x81) || 1716 (rdev->pdev->revision == 0x83))) || 1717 ((rdev->pdev->device == 0x6821) && 1718 ((rdev->pdev->revision == 0x83) || 1719 (rdev->pdev->revision == 0x87))) || 1720 ((rdev->pdev->revision == 0x87) && 1721 ((rdev->pdev->device == 0x6823) || 1722 (rdev->pdev->device == 0x682b)))) 1723 new_smc = true; 1724 new_chip_name = "verde"; 1725 pfp_req_size = SI_PFP_UCODE_SIZE * 4; 1726 me_req_size = SI_PM4_UCODE_SIZE * 4; 1727 ce_req_size = SI_CE_UCODE_SIZE * 4; 1728 rlc_req_size = SI_RLC_UCODE_SIZE * 4; 1729 mc_req_size = SI_MC_UCODE_SIZE * 4; 1730 mc2_req_size = VERDE_MC_UCODE_SIZE * 4; 1731 smc_req_size = ALIGN(VERDE_SMC_UCODE_SIZE, 4); 1732 break; 1733 case CHIP_OLAND: 1734 chip_name = "OLAND"; 1735 if (((rdev->pdev->revision == 0x81) && 1736 ((rdev->pdev->device == 0x6600) || 1737 (rdev->pdev->device == 0x6604) || 1738 (rdev->pdev->device == 0x6605) || 1739 (rdev->pdev->device == 0x6610))) || 1740 ((rdev->pdev->revision == 0x83) && 1741 (rdev->pdev->device == 0x6610))) 1742 new_smc = true; 1743 new_chip_name = "oland"; 1744 pfp_req_size = SI_PFP_UCODE_SIZE * 4; 1745 me_req_size = SI_PM4_UCODE_SIZE * 4; 1746 ce_req_size = SI_CE_UCODE_SIZE * 4; 1747 rlc_req_size = SI_RLC_UCODE_SIZE * 4; 1748 mc_req_size = mc2_req_size = OLAND_MC_UCODE_SIZE * 4; 1749 smc_req_size = ALIGN(OLAND_SMC_UCODE_SIZE, 4); 1750 break; 1751 case CHIP_HAINAN: 1752 chip_name = "HAINAN"; 1753 if (((rdev->pdev->revision == 0x81) && 1754 (rdev->pdev->device == 0x6660)) || 1755 ((rdev->pdev->revision == 0x83) && 1756 ((rdev->pdev->device == 0x6660) || 1757 (rdev->pdev->device == 0x6663) || 1758 (rdev->pdev->device == 0x6665) || 1759 (rdev->pdev->device == 0x6667)))) 1760 new_smc = true; 1761 else if ((rdev->pdev->revision == 0xc3) && 1762 (rdev->pdev->device == 0x6665)) 1763 banks2_fw = true; 1764 new_chip_name = "hainan"; 1765 pfp_req_size = SI_PFP_UCODE_SIZE * 4; 1766 me_req_size = SI_PM4_UCODE_SIZE * 4; 1767 ce_req_size = SI_CE_UCODE_SIZE * 4; 1768 rlc_req_size = SI_RLC_UCODE_SIZE * 4; 1769 mc_req_size = mc2_req_size = OLAND_MC_UCODE_SIZE * 4; 1770 smc_req_size = ALIGN(HAINAN_SMC_UCODE_SIZE, 4); 1771 break; 1772 default: BUG(); 1773 } 1774 1775 /* this memory configuration requires special firmware */ 1776 if (((RREG32(MC_SEQ_MISC0) & 0xff000000) >> 24) == 0x58) 1777 si58_fw = true; 1778 1779 DRM_INFO("Loading %s Microcode\n", new_chip_name); 1780 1781 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", new_chip_name); 1782 err = request_firmware(&rdev->pfp_fw, fw_name, rdev->dev); 1783 if (err) { 1784 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name); 1785 err = request_firmware(&rdev->pfp_fw, fw_name, rdev->dev); 1786 if (err) 1787 goto out; 1788 if (rdev->pfp_fw->size != pfp_req_size) { 1789 pr_err("si_cp: Bogus length %zu in firmware \"%s\"\n", 1790 rdev->pfp_fw->size, fw_name); 1791 err = -EINVAL; 1792 goto out; 1793 } 1794 } else { 1795 err = radeon_ucode_validate(rdev->pfp_fw); 1796 if (err) { 1797 pr_err("si_cp: validation failed for firmware \"%s\"\n", 1798 fw_name); 1799 goto out; 1800 } else { 1801 new_fw++; 1802 } 1803 } 1804 1805 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", new_chip_name); 1806 err = request_firmware(&rdev->me_fw, fw_name, rdev->dev); 1807 if (err) { 1808 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name); 1809 err = request_firmware(&rdev->me_fw, fw_name, rdev->dev); 1810 if (err) 1811 goto out; 1812 if (rdev->me_fw->size != me_req_size) { 1813 pr_err("si_cp: Bogus length %zu in firmware \"%s\"\n", 1814 rdev->me_fw->size, fw_name); 1815 err = -EINVAL; 1816 } 1817 } else { 1818 err = radeon_ucode_validate(rdev->me_fw); 1819 if (err) { 1820 pr_err("si_cp: validation failed for firmware \"%s\"\n", 1821 fw_name); 1822 goto out; 1823 } else { 1824 new_fw++; 1825 } 1826 } 1827 1828 snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", new_chip_name); 1829 err = request_firmware(&rdev->ce_fw, fw_name, rdev->dev); 1830 if (err) { 1831 snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name); 1832 err = request_firmware(&rdev->ce_fw, fw_name, rdev->dev); 1833 if (err) 1834 goto out; 1835 if (rdev->ce_fw->size != ce_req_size) { 1836 pr_err("si_cp: Bogus length %zu in firmware \"%s\"\n", 1837 rdev->ce_fw->size, fw_name); 1838 err = -EINVAL; 1839 } 1840 } else { 1841 err = radeon_ucode_validate(rdev->ce_fw); 1842 if (err) { 1843 pr_err("si_cp: validation failed for firmware \"%s\"\n", 1844 fw_name); 1845 goto out; 1846 } else { 1847 new_fw++; 1848 } 1849 } 1850 1851 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", new_chip_name); 1852 err = request_firmware(&rdev->rlc_fw, fw_name, rdev->dev); 1853 if (err) { 1854 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", chip_name); 1855 err = request_firmware(&rdev->rlc_fw, fw_name, rdev->dev); 1856 if (err) 1857 goto out; 1858 if (rdev->rlc_fw->size != rlc_req_size) { 1859 pr_err("si_rlc: Bogus length %zu in firmware \"%s\"\n", 1860 rdev->rlc_fw->size, fw_name); 1861 err = -EINVAL; 1862 } 1863 } else { 1864 err = radeon_ucode_validate(rdev->rlc_fw); 1865 if (err) { 1866 pr_err("si_cp: validation failed for firmware \"%s\"\n", 1867 fw_name); 1868 goto out; 1869 } else { 1870 new_fw++; 1871 } 1872 } 1873 1874 if (si58_fw) 1875 snprintf(fw_name, sizeof(fw_name), "radeon/si58_mc.bin"); 1876 else 1877 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", new_chip_name); 1878 err = request_firmware(&rdev->mc_fw, fw_name, rdev->dev); 1879 if (err) { 1880 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc2.bin", chip_name); 1881 err = request_firmware(&rdev->mc_fw, fw_name, rdev->dev); 1882 if (err) { 1883 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name); 1884 err = request_firmware(&rdev->mc_fw, fw_name, rdev->dev); 1885 if (err) 1886 goto out; 1887 } 1888 if ((rdev->mc_fw->size != mc_req_size) && 1889 (rdev->mc_fw->size != mc2_req_size)) { 1890 pr_err("si_mc: Bogus length %zu in firmware \"%s\"\n", 1891 rdev->mc_fw->size, fw_name); 1892 err = -EINVAL; 1893 } 1894 DRM_INFO("%s: %zu bytes\n", fw_name, rdev->mc_fw->size); 1895 } else { 1896 err = radeon_ucode_validate(rdev->mc_fw); 1897 if (err) { 1898 pr_err("si_cp: validation failed for firmware \"%s\"\n", 1899 fw_name); 1900 goto out; 1901 } else { 1902 new_fw++; 1903 } 1904 } 1905 1906 if (banks2_fw) 1907 snprintf(fw_name, sizeof(fw_name), "radeon/banks_k_2_smc.bin"); 1908 else if (new_smc) 1909 snprintf(fw_name, sizeof(fw_name), "radeon/%s_k_smc.bin", new_chip_name); 1910 else 1911 snprintf(fw_name, sizeof(fw_name), "radeon/%s_smc.bin", new_chip_name); 1912 err = request_firmware(&rdev->smc_fw, fw_name, rdev->dev); 1913 if (err) { 1914 snprintf(fw_name, sizeof(fw_name), "radeon/%s_smc.bin", chip_name); 1915 err = request_firmware(&rdev->smc_fw, fw_name, rdev->dev); 1916 if (err) { 1917 pr_err("smc: error loading firmware \"%s\"\n", fw_name); 1918 release_firmware(rdev->smc_fw); 1919 rdev->smc_fw = NULL; 1920 err = 0; 1921 } else if (rdev->smc_fw->size != smc_req_size) { 1922 pr_err("si_smc: Bogus length %zu in firmware \"%s\"\n", 1923 rdev->smc_fw->size, fw_name); 1924 err = -EINVAL; 1925 } 1926 } else { 1927 err = radeon_ucode_validate(rdev->smc_fw); 1928 if (err) { 1929 pr_err("si_cp: validation failed for firmware \"%s\"\n", 1930 fw_name); 1931 goto out; 1932 } else { 1933 new_fw++; 1934 } 1935 } 1936 1937 if (new_fw == 0) { 1938 rdev->new_fw = false; 1939 } else if (new_fw < 6) { 1940 pr_err("si_fw: mixing new and old firmware!\n"); 1941 err = -EINVAL; 1942 } else { 1943 rdev->new_fw = true; 1944 } 1945 out: 1946 if (err) { 1947 if (err != -EINVAL) 1948 pr_err("si_cp: Failed to load firmware \"%s\"\n", 1949 fw_name); 1950 release_firmware(rdev->pfp_fw); 1951 rdev->pfp_fw = NULL; 1952 release_firmware(rdev->me_fw); 1953 rdev->me_fw = NULL; 1954 release_firmware(rdev->ce_fw); 1955 rdev->ce_fw = NULL; 1956 release_firmware(rdev->rlc_fw); 1957 rdev->rlc_fw = NULL; 1958 release_firmware(rdev->mc_fw); 1959 rdev->mc_fw = NULL; 1960 release_firmware(rdev->smc_fw); 1961 rdev->smc_fw = NULL; 1962 } 1963 return err; 1964 } 1965 1966 /* watermark setup */ 1967 static u32 dce6_line_buffer_adjust(struct radeon_device *rdev, 1968 struct radeon_crtc *radeon_crtc, 1969 struct drm_display_mode *mode, 1970 struct drm_display_mode *other_mode) 1971 { 1972 u32 tmp, buffer_alloc, i; 1973 u32 pipe_offset = radeon_crtc->crtc_id * 0x20; 1974 /* 1975 * Line Buffer Setup 1976 * There are 3 line buffers, each one shared by 2 display controllers. 1977 * DC_LB_MEMORY_SPLIT controls how that line buffer is shared between 1978 * the display controllers. The paritioning is done via one of four 1979 * preset allocations specified in bits 21:20: 1980 * 0 - half lb 1981 * 2 - whole lb, other crtc must be disabled 1982 */ 1983 /* this can get tricky if we have two large displays on a paired group 1984 * of crtcs. Ideally for multiple large displays we'd assign them to 1985 * non-linked crtcs for maximum line buffer allocation. 1986 */ 1987 if (radeon_crtc->base.enabled && mode) { 1988 if (other_mode) { 1989 tmp = 0; /* 1/2 */ 1990 buffer_alloc = 1; 1991 } else { 1992 tmp = 2; /* whole */ 1993 buffer_alloc = 2; 1994 } 1995 } else { 1996 tmp = 0; 1997 buffer_alloc = 0; 1998 } 1999 2000 WREG32(DC_LB_MEMORY_SPLIT + radeon_crtc->crtc_offset, 2001 DC_LB_MEMORY_CONFIG(tmp)); 2002 2003 WREG32(PIPE0_DMIF_BUFFER_CONTROL + pipe_offset, 2004 DMIF_BUFFERS_ALLOCATED(buffer_alloc)); 2005 for (i = 0; i < rdev->usec_timeout; i++) { 2006 if (RREG32(PIPE0_DMIF_BUFFER_CONTROL + pipe_offset) & 2007 DMIF_BUFFERS_ALLOCATED_COMPLETED) 2008 break; 2009 udelay(1); 2010 } 2011 2012 if (radeon_crtc->base.enabled && mode) { 2013 switch (tmp) { 2014 case 0: 2015 default: 2016 return 4096 * 2; 2017 case 2: 2018 return 8192 * 2; 2019 } 2020 } 2021 2022 /* controller not enabled, so no lb used */ 2023 return 0; 2024 } 2025 2026 static u32 si_get_number_of_dram_channels(struct radeon_device *rdev) 2027 { 2028 u32 tmp = RREG32(MC_SHARED_CHMAP); 2029 2030 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) { 2031 case 0: 2032 default: 2033 return 1; 2034 case 1: 2035 return 2; 2036 case 2: 2037 return 4; 2038 case 3: 2039 return 8; 2040 case 4: 2041 return 3; 2042 case 5: 2043 return 6; 2044 case 6: 2045 return 10; 2046 case 7: 2047 return 12; 2048 case 8: 2049 return 16; 2050 } 2051 } 2052 2053 struct dce6_wm_params { 2054 u32 dram_channels; /* number of dram channels */ 2055 u32 yclk; /* bandwidth per dram data pin in kHz */ 2056 u32 sclk; /* engine clock in kHz */ 2057 u32 disp_clk; /* display clock in kHz */ 2058 u32 src_width; /* viewport width */ 2059 u32 active_time; /* active display time in ns */ 2060 u32 blank_time; /* blank time in ns */ 2061 bool interlaced; /* mode is interlaced */ 2062 fixed20_12 vsc; /* vertical scale ratio */ 2063 u32 num_heads; /* number of active crtcs */ 2064 u32 bytes_per_pixel; /* bytes per pixel display + overlay */ 2065 u32 lb_size; /* line buffer allocated to pipe */ 2066 u32 vtaps; /* vertical scaler taps */ 2067 }; 2068 2069 static u32 dce6_dram_bandwidth(struct dce6_wm_params *wm) 2070 { 2071 /* Calculate raw DRAM Bandwidth */ 2072 fixed20_12 dram_efficiency; /* 0.7 */ 2073 fixed20_12 yclk, dram_channels, bandwidth; 2074 fixed20_12 a; 2075 2076 a.full = dfixed_const(1000); 2077 yclk.full = dfixed_const(wm->yclk); 2078 yclk.full = dfixed_div(yclk, a); 2079 dram_channels.full = dfixed_const(wm->dram_channels * 4); 2080 a.full = dfixed_const(10); 2081 dram_efficiency.full = dfixed_const(7); 2082 dram_efficiency.full = dfixed_div(dram_efficiency, a); 2083 bandwidth.full = dfixed_mul(dram_channels, yclk); 2084 bandwidth.full = dfixed_mul(bandwidth, dram_efficiency); 2085 2086 return dfixed_trunc(bandwidth); 2087 } 2088 2089 static u32 dce6_dram_bandwidth_for_display(struct dce6_wm_params *wm) 2090 { 2091 /* Calculate DRAM Bandwidth and the part allocated to display. */ 2092 fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */ 2093 fixed20_12 yclk, dram_channels, bandwidth; 2094 fixed20_12 a; 2095 2096 a.full = dfixed_const(1000); 2097 yclk.full = dfixed_const(wm->yclk); 2098 yclk.full = dfixed_div(yclk, a); 2099 dram_channels.full = dfixed_const(wm->dram_channels * 4); 2100 a.full = dfixed_const(10); 2101 disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */ 2102 disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a); 2103 bandwidth.full = dfixed_mul(dram_channels, yclk); 2104 bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation); 2105 2106 return dfixed_trunc(bandwidth); 2107 } 2108 2109 static u32 dce6_data_return_bandwidth(struct dce6_wm_params *wm) 2110 { 2111 /* Calculate the display Data return Bandwidth */ 2112 fixed20_12 return_efficiency; /* 0.8 */ 2113 fixed20_12 sclk, bandwidth; 2114 fixed20_12 a; 2115 2116 a.full = dfixed_const(1000); 2117 sclk.full = dfixed_const(wm->sclk); 2118 sclk.full = dfixed_div(sclk, a); 2119 a.full = dfixed_const(10); 2120 return_efficiency.full = dfixed_const(8); 2121 return_efficiency.full = dfixed_div(return_efficiency, a); 2122 a.full = dfixed_const(32); 2123 bandwidth.full = dfixed_mul(a, sclk); 2124 bandwidth.full = dfixed_mul(bandwidth, return_efficiency); 2125 2126 return dfixed_trunc(bandwidth); 2127 } 2128 2129 static u32 dce6_get_dmif_bytes_per_request(struct dce6_wm_params *wm) 2130 { 2131 return 32; 2132 } 2133 2134 static u32 dce6_dmif_request_bandwidth(struct dce6_wm_params *wm) 2135 { 2136 /* Calculate the DMIF Request Bandwidth */ 2137 fixed20_12 disp_clk_request_efficiency; /* 0.8 */ 2138 fixed20_12 disp_clk, sclk, bandwidth; 2139 fixed20_12 a, b1, b2; 2140 u32 min_bandwidth; 2141 2142 a.full = dfixed_const(1000); 2143 disp_clk.full = dfixed_const(wm->disp_clk); 2144 disp_clk.full = dfixed_div(disp_clk, a); 2145 a.full = dfixed_const(dce6_get_dmif_bytes_per_request(wm) / 2); 2146 b1.full = dfixed_mul(a, disp_clk); 2147 2148 a.full = dfixed_const(1000); 2149 sclk.full = dfixed_const(wm->sclk); 2150 sclk.full = dfixed_div(sclk, a); 2151 a.full = dfixed_const(dce6_get_dmif_bytes_per_request(wm)); 2152 b2.full = dfixed_mul(a, sclk); 2153 2154 a.full = dfixed_const(10); 2155 disp_clk_request_efficiency.full = dfixed_const(8); 2156 disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a); 2157 2158 min_bandwidth = min(dfixed_trunc(b1), dfixed_trunc(b2)); 2159 2160 a.full = dfixed_const(min_bandwidth); 2161 bandwidth.full = dfixed_mul(a, disp_clk_request_efficiency); 2162 2163 return dfixed_trunc(bandwidth); 2164 } 2165 2166 static u32 dce6_available_bandwidth(struct dce6_wm_params *wm) 2167 { 2168 /* Calculate the Available bandwidth. Display can use this temporarily but not in average. */ 2169 u32 dram_bandwidth = dce6_dram_bandwidth(wm); 2170 u32 data_return_bandwidth = dce6_data_return_bandwidth(wm); 2171 u32 dmif_req_bandwidth = dce6_dmif_request_bandwidth(wm); 2172 2173 return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth)); 2174 } 2175 2176 static u32 dce6_average_bandwidth(struct dce6_wm_params *wm) 2177 { 2178 /* Calculate the display mode Average Bandwidth 2179 * DisplayMode should contain the source and destination dimensions, 2180 * timing, etc. 2181 */ 2182 fixed20_12 bpp; 2183 fixed20_12 line_time; 2184 fixed20_12 src_width; 2185 fixed20_12 bandwidth; 2186 fixed20_12 a; 2187 2188 a.full = dfixed_const(1000); 2189 line_time.full = dfixed_const(wm->active_time + wm->blank_time); 2190 line_time.full = dfixed_div(line_time, a); 2191 bpp.full = dfixed_const(wm->bytes_per_pixel); 2192 src_width.full = dfixed_const(wm->src_width); 2193 bandwidth.full = dfixed_mul(src_width, bpp); 2194 bandwidth.full = dfixed_mul(bandwidth, wm->vsc); 2195 bandwidth.full = dfixed_div(bandwidth, line_time); 2196 2197 return dfixed_trunc(bandwidth); 2198 } 2199 2200 static u32 dce6_latency_watermark(struct dce6_wm_params *wm) 2201 { 2202 /* First calcualte the latency in ns */ 2203 u32 mc_latency = 2000; /* 2000 ns. */ 2204 u32 available_bandwidth = dce6_available_bandwidth(wm); 2205 u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth; 2206 u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth; 2207 u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */ 2208 u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) + 2209 (wm->num_heads * cursor_line_pair_return_time); 2210 u32 latency = mc_latency + other_heads_data_return_time + dc_latency; 2211 u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time; 2212 u32 tmp, dmif_size = 12288; 2213 fixed20_12 a, b, c; 2214 2215 if (wm->num_heads == 0) 2216 return 0; 2217 2218 a.full = dfixed_const(2); 2219 b.full = dfixed_const(1); 2220 if ((wm->vsc.full > a.full) || 2221 ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) || 2222 (wm->vtaps >= 5) || 2223 ((wm->vsc.full >= a.full) && wm->interlaced)) 2224 max_src_lines_per_dst_line = 4; 2225 else 2226 max_src_lines_per_dst_line = 2; 2227 2228 a.full = dfixed_const(available_bandwidth); 2229 b.full = dfixed_const(wm->num_heads); 2230 a.full = dfixed_div(a, b); 2231 tmp = div_u64((u64) dmif_size * (u64) wm->disp_clk, mc_latency + 512); 2232 tmp = min(dfixed_trunc(a), tmp); 2233 2234 lb_fill_bw = min(tmp, wm->disp_clk * wm->bytes_per_pixel / 1000); 2235 2236 a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel); 2237 b.full = dfixed_const(1000); 2238 c.full = dfixed_const(lb_fill_bw); 2239 b.full = dfixed_div(c, b); 2240 a.full = dfixed_div(a, b); 2241 line_fill_time = dfixed_trunc(a); 2242 2243 if (line_fill_time < wm->active_time) 2244 return latency; 2245 else 2246 return latency + (line_fill_time - wm->active_time); 2247 2248 } 2249 2250 static bool dce6_average_bandwidth_vs_dram_bandwidth_for_display(struct dce6_wm_params *wm) 2251 { 2252 if (dce6_average_bandwidth(wm) <= 2253 (dce6_dram_bandwidth_for_display(wm) / wm->num_heads)) 2254 return true; 2255 else 2256 return false; 2257 }; 2258 2259 static bool dce6_average_bandwidth_vs_available_bandwidth(struct dce6_wm_params *wm) 2260 { 2261 if (dce6_average_bandwidth(wm) <= 2262 (dce6_available_bandwidth(wm) / wm->num_heads)) 2263 return true; 2264 else 2265 return false; 2266 }; 2267 2268 static bool dce6_check_latency_hiding(struct dce6_wm_params *wm) 2269 { 2270 u32 lb_partitions = wm->lb_size / wm->src_width; 2271 u32 line_time = wm->active_time + wm->blank_time; 2272 u32 latency_tolerant_lines; 2273 u32 latency_hiding; 2274 fixed20_12 a; 2275 2276 a.full = dfixed_const(1); 2277 if (wm->vsc.full > a.full) 2278 latency_tolerant_lines = 1; 2279 else { 2280 if (lb_partitions <= (wm->vtaps + 1)) 2281 latency_tolerant_lines = 1; 2282 else 2283 latency_tolerant_lines = 2; 2284 } 2285 2286 latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time); 2287 2288 if (dce6_latency_watermark(wm) <= latency_hiding) 2289 return true; 2290 else 2291 return false; 2292 } 2293 2294 static void dce6_program_watermarks(struct radeon_device *rdev, 2295 struct radeon_crtc *radeon_crtc, 2296 u32 lb_size, u32 num_heads) 2297 { 2298 struct drm_display_mode *mode = &radeon_crtc->base.mode; 2299 struct dce6_wm_params wm_low, wm_high; 2300 u32 dram_channels; 2301 u32 active_time; 2302 u32 line_time = 0; 2303 u32 latency_watermark_a = 0, latency_watermark_b = 0; 2304 u32 priority_a_mark = 0, priority_b_mark = 0; 2305 u32 priority_a_cnt = PRIORITY_OFF; 2306 u32 priority_b_cnt = PRIORITY_OFF; 2307 u32 tmp, arb_control3; 2308 fixed20_12 a, b, c; 2309 2310 if (radeon_crtc->base.enabled && num_heads && mode) { 2311 active_time = (u32) div_u64((u64)mode->crtc_hdisplay * 1000000, 2312 (u32)mode->clock); 2313 line_time = (u32) div_u64((u64)mode->crtc_htotal * 1000000, 2314 (u32)mode->clock); 2315 line_time = min(line_time, (u32)65535); 2316 priority_a_cnt = 0; 2317 priority_b_cnt = 0; 2318 2319 if (rdev->family == CHIP_ARUBA) 2320 dram_channels = evergreen_get_number_of_dram_channels(rdev); 2321 else 2322 dram_channels = si_get_number_of_dram_channels(rdev); 2323 2324 /* watermark for high clocks */ 2325 if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { 2326 wm_high.yclk = 2327 radeon_dpm_get_mclk(rdev, false) * 10; 2328 wm_high.sclk = 2329 radeon_dpm_get_sclk(rdev, false) * 10; 2330 } else { 2331 wm_high.yclk = rdev->pm.current_mclk * 10; 2332 wm_high.sclk = rdev->pm.current_sclk * 10; 2333 } 2334 2335 wm_high.disp_clk = mode->clock; 2336 wm_high.src_width = mode->crtc_hdisplay; 2337 wm_high.active_time = active_time; 2338 wm_high.blank_time = line_time - wm_high.active_time; 2339 wm_high.interlaced = false; 2340 if (mode->flags & DRM_MODE_FLAG_INTERLACE) 2341 wm_high.interlaced = true; 2342 wm_high.vsc = radeon_crtc->vsc; 2343 wm_high.vtaps = 1; 2344 if (radeon_crtc->rmx_type != RMX_OFF) 2345 wm_high.vtaps = 2; 2346 wm_high.bytes_per_pixel = 4; /* XXX: get this from fb config */ 2347 wm_high.lb_size = lb_size; 2348 wm_high.dram_channels = dram_channels; 2349 wm_high.num_heads = num_heads; 2350 2351 /* watermark for low clocks */ 2352 if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { 2353 wm_low.yclk = 2354 radeon_dpm_get_mclk(rdev, true) * 10; 2355 wm_low.sclk = 2356 radeon_dpm_get_sclk(rdev, true) * 10; 2357 } else { 2358 wm_low.yclk = rdev->pm.current_mclk * 10; 2359 wm_low.sclk = rdev->pm.current_sclk * 10; 2360 } 2361 2362 wm_low.disp_clk = mode->clock; 2363 wm_low.src_width = mode->crtc_hdisplay; 2364 wm_low.active_time = active_time; 2365 wm_low.blank_time = line_time - wm_low.active_time; 2366 wm_low.interlaced = false; 2367 if (mode->flags & DRM_MODE_FLAG_INTERLACE) 2368 wm_low.interlaced = true; 2369 wm_low.vsc = radeon_crtc->vsc; 2370 wm_low.vtaps = 1; 2371 if (radeon_crtc->rmx_type != RMX_OFF) 2372 wm_low.vtaps = 2; 2373 wm_low.bytes_per_pixel = 4; /* XXX: get this from fb config */ 2374 wm_low.lb_size = lb_size; 2375 wm_low.dram_channels = dram_channels; 2376 wm_low.num_heads = num_heads; 2377 2378 /* set for high clocks */ 2379 latency_watermark_a = min(dce6_latency_watermark(&wm_high), (u32)65535); 2380 /* set for low clocks */ 2381 latency_watermark_b = min(dce6_latency_watermark(&wm_low), (u32)65535); 2382 2383 /* possibly force display priority to high */ 2384 /* should really do this at mode validation time... */ 2385 if (!dce6_average_bandwidth_vs_dram_bandwidth_for_display(&wm_high) || 2386 !dce6_average_bandwidth_vs_available_bandwidth(&wm_high) || 2387 !dce6_check_latency_hiding(&wm_high) || 2388 (rdev->disp_priority == 2)) { 2389 DRM_DEBUG_KMS("force priority to high\n"); 2390 priority_a_cnt |= PRIORITY_ALWAYS_ON; 2391 priority_b_cnt |= PRIORITY_ALWAYS_ON; 2392 } 2393 if (!dce6_average_bandwidth_vs_dram_bandwidth_for_display(&wm_low) || 2394 !dce6_average_bandwidth_vs_available_bandwidth(&wm_low) || 2395 !dce6_check_latency_hiding(&wm_low) || 2396 (rdev->disp_priority == 2)) { 2397 DRM_DEBUG_KMS("force priority to high\n"); 2398 priority_a_cnt |= PRIORITY_ALWAYS_ON; 2399 priority_b_cnt |= PRIORITY_ALWAYS_ON; 2400 } 2401 2402 a.full = dfixed_const(1000); 2403 b.full = dfixed_const(mode->clock); 2404 b.full = dfixed_div(b, a); 2405 c.full = dfixed_const(latency_watermark_a); 2406 c.full = dfixed_mul(c, b); 2407 c.full = dfixed_mul(c, radeon_crtc->hsc); 2408 c.full = dfixed_div(c, a); 2409 a.full = dfixed_const(16); 2410 c.full = dfixed_div(c, a); 2411 priority_a_mark = dfixed_trunc(c); 2412 priority_a_cnt |= priority_a_mark & PRIORITY_MARK_MASK; 2413 2414 a.full = dfixed_const(1000); 2415 b.full = dfixed_const(mode->clock); 2416 b.full = dfixed_div(b, a); 2417 c.full = dfixed_const(latency_watermark_b); 2418 c.full = dfixed_mul(c, b); 2419 c.full = dfixed_mul(c, radeon_crtc->hsc); 2420 c.full = dfixed_div(c, a); 2421 a.full = dfixed_const(16); 2422 c.full = dfixed_div(c, a); 2423 priority_b_mark = dfixed_trunc(c); 2424 priority_b_cnt |= priority_b_mark & PRIORITY_MARK_MASK; 2425 2426 /* Save number of lines the linebuffer leads before the scanout */ 2427 radeon_crtc->lb_vblank_lead_lines = DIV_ROUND_UP(lb_size, mode->crtc_hdisplay); 2428 } 2429 2430 /* select wm A */ 2431 arb_control3 = RREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset); 2432 tmp = arb_control3; 2433 tmp &= ~LATENCY_WATERMARK_MASK(3); 2434 tmp |= LATENCY_WATERMARK_MASK(1); 2435 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, tmp); 2436 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset, 2437 (LATENCY_LOW_WATERMARK(latency_watermark_a) | 2438 LATENCY_HIGH_WATERMARK(line_time))); 2439 /* select wm B */ 2440 tmp = RREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset); 2441 tmp &= ~LATENCY_WATERMARK_MASK(3); 2442 tmp |= LATENCY_WATERMARK_MASK(2); 2443 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, tmp); 2444 WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset, 2445 (LATENCY_LOW_WATERMARK(latency_watermark_b) | 2446 LATENCY_HIGH_WATERMARK(line_time))); 2447 /* restore original selection */ 2448 WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, arb_control3); 2449 2450 /* write the priority marks */ 2451 WREG32(PRIORITY_A_CNT + radeon_crtc->crtc_offset, priority_a_cnt); 2452 WREG32(PRIORITY_B_CNT + radeon_crtc->crtc_offset, priority_b_cnt); 2453 2454 /* save values for DPM */ 2455 radeon_crtc->line_time = line_time; 2456 radeon_crtc->wm_high = latency_watermark_a; 2457 radeon_crtc->wm_low = latency_watermark_b; 2458 } 2459 2460 void dce6_bandwidth_update(struct radeon_device *rdev) 2461 { 2462 struct drm_display_mode *mode0 = NULL; 2463 struct drm_display_mode *mode1 = NULL; 2464 u32 num_heads = 0, lb_size; 2465 int i; 2466 2467 if (!rdev->mode_info.mode_config_initialized) 2468 return; 2469 2470 radeon_update_display_priority(rdev); 2471 2472 for (i = 0; i < rdev->num_crtc; i++) { 2473 if (rdev->mode_info.crtcs[i]->base.enabled) 2474 num_heads++; 2475 } 2476 for (i = 0; i < rdev->num_crtc; i += 2) { 2477 mode0 = &rdev->mode_info.crtcs[i]->base.mode; 2478 mode1 = &rdev->mode_info.crtcs[i+1]->base.mode; 2479 lb_size = dce6_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode0, mode1); 2480 dce6_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads); 2481 lb_size = dce6_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i+1], mode1, mode0); 2482 dce6_program_watermarks(rdev, rdev->mode_info.crtcs[i+1], lb_size, num_heads); 2483 } 2484 } 2485 2486 /* 2487 * Core functions 2488 */ 2489 static void si_tiling_mode_table_init(struct radeon_device *rdev) 2490 { 2491 u32 *tile = rdev->config.si.tile_mode_array; 2492 const u32 num_tile_mode_states = 2493 ARRAY_SIZE(rdev->config.si.tile_mode_array); 2494 u32 reg_offset, split_equal_to_row_size; 2495 2496 switch (rdev->config.si.mem_row_size_in_kb) { 2497 case 1: 2498 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB; 2499 break; 2500 case 2: 2501 default: 2502 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB; 2503 break; 2504 case 4: 2505 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB; 2506 break; 2507 } 2508 2509 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 2510 tile[reg_offset] = 0; 2511 2512 switch(rdev->family) { 2513 case CHIP_TAHITI: 2514 case CHIP_PITCAIRN: 2515 /* non-AA compressed depth or any compressed stencil */ 2516 tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2517 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2518 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2519 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2520 NUM_BANKS(ADDR_SURF_16_BANK) | 2521 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2522 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2523 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2524 /* 2xAA/4xAA compressed depth only */ 2525 tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2526 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2527 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2528 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 2529 NUM_BANKS(ADDR_SURF_16_BANK) | 2530 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2531 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2532 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2533 /* 8xAA compressed depth only */ 2534 tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2535 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2536 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2537 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2538 NUM_BANKS(ADDR_SURF_16_BANK) | 2539 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2540 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2541 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2542 /* 2xAA/4xAA compressed depth with stencil (for depth buffer) */ 2543 tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2544 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2545 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2546 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 2547 NUM_BANKS(ADDR_SURF_16_BANK) | 2548 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2549 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2550 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2551 /* Maps w/ a dimension less than the 2D macro-tile dimensions (for mipmapped depth textures) */ 2552 tile[4] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2553 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2554 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2555 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2556 NUM_BANKS(ADDR_SURF_16_BANK) | 2557 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2558 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2559 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2560 /* Uncompressed 16bpp depth - and stencil buffer allocated with it */ 2561 tile[5] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2562 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2563 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2564 TILE_SPLIT(split_equal_to_row_size) | 2565 NUM_BANKS(ADDR_SURF_16_BANK) | 2566 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2567 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2568 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2569 /* Uncompressed 32bpp depth - and stencil buffer allocated with it */ 2570 tile[6] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2571 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2572 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2573 TILE_SPLIT(split_equal_to_row_size) | 2574 NUM_BANKS(ADDR_SURF_16_BANK) | 2575 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2576 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2577 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1)); 2578 /* Uncompressed 8bpp stencil without depth (drivers typically do not use) */ 2579 tile[7] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2580 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2581 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2582 TILE_SPLIT(split_equal_to_row_size) | 2583 NUM_BANKS(ADDR_SURF_16_BANK) | 2584 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2585 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2586 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2587 /* 1D and 1D Array Surfaces */ 2588 tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 2589 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 2590 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2591 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2592 NUM_BANKS(ADDR_SURF_16_BANK) | 2593 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2594 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2595 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2596 /* Displayable maps. */ 2597 tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2598 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 2599 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2600 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2601 NUM_BANKS(ADDR_SURF_16_BANK) | 2602 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2603 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2604 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2605 /* Display 8bpp. */ 2606 tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2607 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 2608 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2609 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2610 NUM_BANKS(ADDR_SURF_16_BANK) | 2611 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2612 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2613 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2614 /* Display 16bpp. */ 2615 tile[11] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2616 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 2617 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2618 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2619 NUM_BANKS(ADDR_SURF_16_BANK) | 2620 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2621 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2622 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2623 /* Display 32bpp. */ 2624 tile[12] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2625 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 2626 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2627 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 2628 NUM_BANKS(ADDR_SURF_16_BANK) | 2629 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2630 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2631 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1)); 2632 /* Thin. */ 2633 tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2634 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2635 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2636 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2637 NUM_BANKS(ADDR_SURF_16_BANK) | 2638 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2639 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2640 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2641 /* Thin 8 bpp. */ 2642 tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2643 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2644 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2645 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2646 NUM_BANKS(ADDR_SURF_16_BANK) | 2647 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2648 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2649 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1)); 2650 /* Thin 16 bpp. */ 2651 tile[15] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2652 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2653 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2654 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2655 NUM_BANKS(ADDR_SURF_16_BANK) | 2656 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2657 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2658 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1)); 2659 /* Thin 32 bpp. */ 2660 tile[16] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2661 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2662 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2663 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 2664 NUM_BANKS(ADDR_SURF_16_BANK) | 2665 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2666 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2667 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1)); 2668 /* Thin 64 bpp. */ 2669 tile[17] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2670 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2671 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2672 TILE_SPLIT(split_equal_to_row_size) | 2673 NUM_BANKS(ADDR_SURF_16_BANK) | 2674 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2675 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2676 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1)); 2677 /* 8 bpp PRT. */ 2678 tile[21] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2679 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2680 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2681 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2682 NUM_BANKS(ADDR_SURF_16_BANK) | 2683 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 2684 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2685 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2686 /* 16 bpp PRT */ 2687 tile[22] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2688 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2689 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2690 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2691 NUM_BANKS(ADDR_SURF_16_BANK) | 2692 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2693 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2694 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4)); 2695 /* 32 bpp PRT */ 2696 tile[23] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2697 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2698 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2699 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2700 NUM_BANKS(ADDR_SURF_16_BANK) | 2701 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2702 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2703 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2704 /* 64 bpp PRT */ 2705 tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2706 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2707 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2708 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 2709 NUM_BANKS(ADDR_SURF_16_BANK) | 2710 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2711 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2712 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2713 /* 128 bpp PRT */ 2714 tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2715 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2716 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2717 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 2718 NUM_BANKS(ADDR_SURF_8_BANK) | 2719 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2720 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2721 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1)); 2722 2723 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 2724 WREG32(GB_TILE_MODE0 + (reg_offset * 4), tile[reg_offset]); 2725 break; 2726 2727 case CHIP_VERDE: 2728 case CHIP_OLAND: 2729 case CHIP_HAINAN: 2730 /* non-AA compressed depth or any compressed stencil */ 2731 tile[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2732 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2733 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2734 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2735 NUM_BANKS(ADDR_SURF_16_BANK) | 2736 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2737 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2738 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4)); 2739 /* 2xAA/4xAA compressed depth only */ 2740 tile[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2741 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2742 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2743 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 2744 NUM_BANKS(ADDR_SURF_16_BANK) | 2745 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2746 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2747 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4)); 2748 /* 8xAA compressed depth only */ 2749 tile[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2750 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2751 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2752 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2753 NUM_BANKS(ADDR_SURF_16_BANK) | 2754 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2755 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2756 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4)); 2757 /* 2xAA/4xAA compressed depth with stencil (for depth buffer) */ 2758 tile[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2759 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2760 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2761 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | 2762 NUM_BANKS(ADDR_SURF_16_BANK) | 2763 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2764 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2765 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4)); 2766 /* Maps w/ a dimension less than the 2D macro-tile dimensions (for mipmapped depth textures) */ 2767 tile[4] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2768 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2769 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2770 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2771 NUM_BANKS(ADDR_SURF_16_BANK) | 2772 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2773 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2774 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2775 /* Uncompressed 16bpp depth - and stencil buffer allocated with it */ 2776 tile[5] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2777 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2778 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2779 TILE_SPLIT(split_equal_to_row_size) | 2780 NUM_BANKS(ADDR_SURF_16_BANK) | 2781 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2782 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2783 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2784 /* Uncompressed 32bpp depth - and stencil buffer allocated with it */ 2785 tile[6] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2786 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2787 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2788 TILE_SPLIT(split_equal_to_row_size) | 2789 NUM_BANKS(ADDR_SURF_16_BANK) | 2790 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2791 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2792 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2793 /* Uncompressed 8bpp stencil without depth (drivers typically do not use) */ 2794 tile[7] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2795 MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) | 2796 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2797 TILE_SPLIT(split_equal_to_row_size) | 2798 NUM_BANKS(ADDR_SURF_16_BANK) | 2799 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2800 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2801 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4)); 2802 /* 1D and 1D Array Surfaces */ 2803 tile[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | 2804 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 2805 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2806 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2807 NUM_BANKS(ADDR_SURF_16_BANK) | 2808 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2809 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2810 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2811 /* Displayable maps. */ 2812 tile[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2813 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 2814 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2815 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2816 NUM_BANKS(ADDR_SURF_16_BANK) | 2817 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2818 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2819 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2820 /* Display 8bpp. */ 2821 tile[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2822 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 2823 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2824 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2825 NUM_BANKS(ADDR_SURF_16_BANK) | 2826 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2827 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2828 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4)); 2829 /* Display 16bpp. */ 2830 tile[11] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2831 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 2832 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2833 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2834 NUM_BANKS(ADDR_SURF_16_BANK) | 2835 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2836 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2837 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2838 /* Display 32bpp. */ 2839 tile[12] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2840 MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) | 2841 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2842 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 2843 NUM_BANKS(ADDR_SURF_16_BANK) | 2844 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2845 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2846 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2847 /* Thin. */ 2848 tile[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | 2849 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2850 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2851 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | 2852 NUM_BANKS(ADDR_SURF_16_BANK) | 2853 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2854 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2855 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2856 /* Thin 8 bpp. */ 2857 tile[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2858 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2859 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2860 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2861 NUM_BANKS(ADDR_SURF_16_BANK) | 2862 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2863 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2864 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2865 /* Thin 16 bpp. */ 2866 tile[15] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2867 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2868 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2869 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2870 NUM_BANKS(ADDR_SURF_16_BANK) | 2871 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2872 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2873 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2874 /* Thin 32 bpp. */ 2875 tile[16] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2876 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2877 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2878 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 2879 NUM_BANKS(ADDR_SURF_16_BANK) | 2880 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2881 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2882 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2883 /* Thin 64 bpp. */ 2884 tile[17] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2885 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2886 PIPE_CONFIG(ADDR_SURF_P4_8x16) | 2887 TILE_SPLIT(split_equal_to_row_size) | 2888 NUM_BANKS(ADDR_SURF_16_BANK) | 2889 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2890 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2891 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2892 /* 8 bpp PRT. */ 2893 tile[21] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2894 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2895 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2896 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2897 NUM_BANKS(ADDR_SURF_16_BANK) | 2898 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) | 2899 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2900 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2901 /* 16 bpp PRT */ 2902 tile[22] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2903 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2904 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2905 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2906 NUM_BANKS(ADDR_SURF_16_BANK) | 2907 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2908 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | 2909 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4)); 2910 /* 32 bpp PRT */ 2911 tile[23] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2912 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2913 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2914 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | 2915 NUM_BANKS(ADDR_SURF_16_BANK) | 2916 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2917 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | 2918 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2919 /* 64 bpp PRT */ 2920 tile[24] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2921 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2922 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2923 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | 2924 NUM_BANKS(ADDR_SURF_16_BANK) | 2925 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2926 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2927 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2)); 2928 /* 128 bpp PRT */ 2929 tile[25] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | 2930 MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) | 2931 PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) | 2932 TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) | 2933 NUM_BANKS(ADDR_SURF_8_BANK) | 2934 BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | 2935 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | 2936 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1)); 2937 2938 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) 2939 WREG32(GB_TILE_MODE0 + (reg_offset * 4), tile[reg_offset]); 2940 break; 2941 2942 default: 2943 DRM_ERROR("unknown asic: 0x%x\n", rdev->family); 2944 } 2945 } 2946 2947 static void si_select_se_sh(struct radeon_device *rdev, 2948 u32 se_num, u32 sh_num) 2949 { 2950 u32 data = INSTANCE_BROADCAST_WRITES; 2951 2952 if ((se_num == 0xffffffff) && (sh_num == 0xffffffff)) 2953 data |= SH_BROADCAST_WRITES | SE_BROADCAST_WRITES; 2954 else if (se_num == 0xffffffff) 2955 data |= SE_BROADCAST_WRITES | SH_INDEX(sh_num); 2956 else if (sh_num == 0xffffffff) 2957 data |= SH_BROADCAST_WRITES | SE_INDEX(se_num); 2958 else 2959 data |= SH_INDEX(sh_num) | SE_INDEX(se_num); 2960 WREG32(GRBM_GFX_INDEX, data); 2961 } 2962 2963 static u32 si_create_bitmask(u32 bit_width) 2964 { 2965 u32 i, mask = 0; 2966 2967 for (i = 0; i < bit_width; i++) { 2968 mask <<= 1; 2969 mask |= 1; 2970 } 2971 return mask; 2972 } 2973 2974 static u32 si_get_cu_enabled(struct radeon_device *rdev, u32 cu_per_sh) 2975 { 2976 u32 data, mask; 2977 2978 data = RREG32(CC_GC_SHADER_ARRAY_CONFIG); 2979 if (data & 1) 2980 data &= INACTIVE_CUS_MASK; 2981 else 2982 data = 0; 2983 data |= RREG32(GC_USER_SHADER_ARRAY_CONFIG); 2984 2985 data >>= INACTIVE_CUS_SHIFT; 2986 2987 mask = si_create_bitmask(cu_per_sh); 2988 2989 return ~data & mask; 2990 } 2991 2992 static void si_setup_spi(struct radeon_device *rdev, 2993 u32 se_num, u32 sh_per_se, 2994 u32 cu_per_sh) 2995 { 2996 int i, j, k; 2997 u32 data, mask, active_cu; 2998 2999 for (i = 0; i < se_num; i++) { 3000 for (j = 0; j < sh_per_se; j++) { 3001 si_select_se_sh(rdev, i, j); 3002 data = RREG32(SPI_STATIC_THREAD_MGMT_3); 3003 active_cu = si_get_cu_enabled(rdev, cu_per_sh); 3004 3005 mask = 1; 3006 for (k = 0; k < 16; k++) { 3007 mask <<= k; 3008 if (active_cu & mask) { 3009 data &= ~mask; 3010 WREG32(SPI_STATIC_THREAD_MGMT_3, data); 3011 break; 3012 } 3013 } 3014 } 3015 } 3016 si_select_se_sh(rdev, 0xffffffff, 0xffffffff); 3017 } 3018 3019 static u32 si_get_rb_disabled(struct radeon_device *rdev, 3020 u32 max_rb_num_per_se, 3021 u32 sh_per_se) 3022 { 3023 u32 data, mask; 3024 3025 data = RREG32(CC_RB_BACKEND_DISABLE); 3026 if (data & 1) 3027 data &= BACKEND_DISABLE_MASK; 3028 else 3029 data = 0; 3030 data |= RREG32(GC_USER_RB_BACKEND_DISABLE); 3031 3032 data >>= BACKEND_DISABLE_SHIFT; 3033 3034 mask = si_create_bitmask(max_rb_num_per_se / sh_per_se); 3035 3036 return data & mask; 3037 } 3038 3039 static void si_setup_rb(struct radeon_device *rdev, 3040 u32 se_num, u32 sh_per_se, 3041 u32 max_rb_num_per_se) 3042 { 3043 int i, j; 3044 u32 data, mask; 3045 u32 disabled_rbs = 0; 3046 u32 enabled_rbs = 0; 3047 3048 for (i = 0; i < se_num; i++) { 3049 for (j = 0; j < sh_per_se; j++) { 3050 si_select_se_sh(rdev, i, j); 3051 data = si_get_rb_disabled(rdev, max_rb_num_per_se, sh_per_se); 3052 disabled_rbs |= data << ((i * sh_per_se + j) * TAHITI_RB_BITMAP_WIDTH_PER_SH); 3053 } 3054 } 3055 si_select_se_sh(rdev, 0xffffffff, 0xffffffff); 3056 3057 mask = 1; 3058 for (i = 0; i < max_rb_num_per_se * se_num; i++) { 3059 if (!(disabled_rbs & mask)) 3060 enabled_rbs |= mask; 3061 mask <<= 1; 3062 } 3063 3064 rdev->config.si.backend_enable_mask = enabled_rbs; 3065 3066 for (i = 0; i < se_num; i++) { 3067 si_select_se_sh(rdev, i, 0xffffffff); 3068 data = 0; 3069 for (j = 0; j < sh_per_se; j++) { 3070 switch (enabled_rbs & 3) { 3071 case 1: 3072 data |= (RASTER_CONFIG_RB_MAP_0 << (i * sh_per_se + j) * 2); 3073 break; 3074 case 2: 3075 data |= (RASTER_CONFIG_RB_MAP_3 << (i * sh_per_se + j) * 2); 3076 break; 3077 case 3: 3078 default: 3079 data |= (RASTER_CONFIG_RB_MAP_2 << (i * sh_per_se + j) * 2); 3080 break; 3081 } 3082 enabled_rbs >>= 2; 3083 } 3084 WREG32(PA_SC_RASTER_CONFIG, data); 3085 } 3086 si_select_se_sh(rdev, 0xffffffff, 0xffffffff); 3087 } 3088 3089 static void si_gpu_init(struct radeon_device *rdev) 3090 { 3091 u32 gb_addr_config = 0; 3092 u32 mc_shared_chmap, mc_arb_ramcfg; 3093 u32 sx_debug_1; 3094 u32 hdp_host_path_cntl; 3095 u32 tmp; 3096 int i, j; 3097 3098 switch (rdev->family) { 3099 case CHIP_TAHITI: 3100 rdev->config.si.max_shader_engines = 2; 3101 rdev->config.si.max_tile_pipes = 12; 3102 rdev->config.si.max_cu_per_sh = 8; 3103 rdev->config.si.max_sh_per_se = 2; 3104 rdev->config.si.max_backends_per_se = 4; 3105 rdev->config.si.max_texture_channel_caches = 12; 3106 rdev->config.si.max_gprs = 256; 3107 rdev->config.si.max_gs_threads = 32; 3108 rdev->config.si.max_hw_contexts = 8; 3109 3110 rdev->config.si.sc_prim_fifo_size_frontend = 0x20; 3111 rdev->config.si.sc_prim_fifo_size_backend = 0x100; 3112 rdev->config.si.sc_hiz_tile_fifo_size = 0x30; 3113 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130; 3114 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN; 3115 break; 3116 case CHIP_PITCAIRN: 3117 rdev->config.si.max_shader_engines = 2; 3118 rdev->config.si.max_tile_pipes = 8; 3119 rdev->config.si.max_cu_per_sh = 5; 3120 rdev->config.si.max_sh_per_se = 2; 3121 rdev->config.si.max_backends_per_se = 4; 3122 rdev->config.si.max_texture_channel_caches = 8; 3123 rdev->config.si.max_gprs = 256; 3124 rdev->config.si.max_gs_threads = 32; 3125 rdev->config.si.max_hw_contexts = 8; 3126 3127 rdev->config.si.sc_prim_fifo_size_frontend = 0x20; 3128 rdev->config.si.sc_prim_fifo_size_backend = 0x100; 3129 rdev->config.si.sc_hiz_tile_fifo_size = 0x30; 3130 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130; 3131 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN; 3132 break; 3133 case CHIP_VERDE: 3134 default: 3135 rdev->config.si.max_shader_engines = 1; 3136 rdev->config.si.max_tile_pipes = 4; 3137 rdev->config.si.max_cu_per_sh = 5; 3138 rdev->config.si.max_sh_per_se = 2; 3139 rdev->config.si.max_backends_per_se = 4; 3140 rdev->config.si.max_texture_channel_caches = 4; 3141 rdev->config.si.max_gprs = 256; 3142 rdev->config.si.max_gs_threads = 32; 3143 rdev->config.si.max_hw_contexts = 8; 3144 3145 rdev->config.si.sc_prim_fifo_size_frontend = 0x20; 3146 rdev->config.si.sc_prim_fifo_size_backend = 0x40; 3147 rdev->config.si.sc_hiz_tile_fifo_size = 0x30; 3148 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130; 3149 gb_addr_config = VERDE_GB_ADDR_CONFIG_GOLDEN; 3150 break; 3151 case CHIP_OLAND: 3152 rdev->config.si.max_shader_engines = 1; 3153 rdev->config.si.max_tile_pipes = 4; 3154 rdev->config.si.max_cu_per_sh = 6; 3155 rdev->config.si.max_sh_per_se = 1; 3156 rdev->config.si.max_backends_per_se = 2; 3157 rdev->config.si.max_texture_channel_caches = 4; 3158 rdev->config.si.max_gprs = 256; 3159 rdev->config.si.max_gs_threads = 16; 3160 rdev->config.si.max_hw_contexts = 8; 3161 3162 rdev->config.si.sc_prim_fifo_size_frontend = 0x20; 3163 rdev->config.si.sc_prim_fifo_size_backend = 0x40; 3164 rdev->config.si.sc_hiz_tile_fifo_size = 0x30; 3165 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130; 3166 gb_addr_config = VERDE_GB_ADDR_CONFIG_GOLDEN; 3167 break; 3168 case CHIP_HAINAN: 3169 rdev->config.si.max_shader_engines = 1; 3170 rdev->config.si.max_tile_pipes = 4; 3171 rdev->config.si.max_cu_per_sh = 5; 3172 rdev->config.si.max_sh_per_se = 1; 3173 rdev->config.si.max_backends_per_se = 1; 3174 rdev->config.si.max_texture_channel_caches = 2; 3175 rdev->config.si.max_gprs = 256; 3176 rdev->config.si.max_gs_threads = 16; 3177 rdev->config.si.max_hw_contexts = 8; 3178 3179 rdev->config.si.sc_prim_fifo_size_frontend = 0x20; 3180 rdev->config.si.sc_prim_fifo_size_backend = 0x40; 3181 rdev->config.si.sc_hiz_tile_fifo_size = 0x30; 3182 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130; 3183 gb_addr_config = HAINAN_GB_ADDR_CONFIG_GOLDEN; 3184 break; 3185 } 3186 3187 /* Initialize HDP */ 3188 for (i = 0, j = 0; i < 32; i++, j += 0x18) { 3189 WREG32((0x2c14 + j), 0x00000000); 3190 WREG32((0x2c18 + j), 0x00000000); 3191 WREG32((0x2c1c + j), 0x00000000); 3192 WREG32((0x2c20 + j), 0x00000000); 3193 WREG32((0x2c24 + j), 0x00000000); 3194 } 3195 3196 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff)); 3197 WREG32(SRBM_INT_CNTL, 1); 3198 WREG32(SRBM_INT_ACK, 1); 3199 3200 evergreen_fix_pci_max_read_req_size(rdev); 3201 3202 WREG32(BIF_FB_EN, FB_READ_EN | FB_WRITE_EN); 3203 3204 mc_shared_chmap = RREG32(MC_SHARED_CHMAP); 3205 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG); 3206 3207 rdev->config.si.num_tile_pipes = rdev->config.si.max_tile_pipes; 3208 rdev->config.si.mem_max_burst_length_bytes = 256; 3209 tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT; 3210 rdev->config.si.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; 3211 if (rdev->config.si.mem_row_size_in_kb > 4) 3212 rdev->config.si.mem_row_size_in_kb = 4; 3213 /* XXX use MC settings? */ 3214 rdev->config.si.shader_engine_tile_size = 32; 3215 rdev->config.si.num_gpus = 1; 3216 rdev->config.si.multi_gpu_tile_size = 64; 3217 3218 /* fix up row size */ 3219 gb_addr_config &= ~ROW_SIZE_MASK; 3220 switch (rdev->config.si.mem_row_size_in_kb) { 3221 case 1: 3222 default: 3223 gb_addr_config |= ROW_SIZE(0); 3224 break; 3225 case 2: 3226 gb_addr_config |= ROW_SIZE(1); 3227 break; 3228 case 4: 3229 gb_addr_config |= ROW_SIZE(2); 3230 break; 3231 } 3232 3233 /* setup tiling info dword. gb_addr_config is not adequate since it does 3234 * not have bank info, so create a custom tiling dword. 3235 * bits 3:0 num_pipes 3236 * bits 7:4 num_banks 3237 * bits 11:8 group_size 3238 * bits 15:12 row_size 3239 */ 3240 rdev->config.si.tile_config = 0; 3241 switch (rdev->config.si.num_tile_pipes) { 3242 case 1: 3243 rdev->config.si.tile_config |= (0 << 0); 3244 break; 3245 case 2: 3246 rdev->config.si.tile_config |= (1 << 0); 3247 break; 3248 case 4: 3249 rdev->config.si.tile_config |= (2 << 0); 3250 break; 3251 case 8: 3252 default: 3253 /* XXX what about 12? */ 3254 rdev->config.si.tile_config |= (3 << 0); 3255 break; 3256 } 3257 switch ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) { 3258 case 0: /* four banks */ 3259 rdev->config.si.tile_config |= 0 << 4; 3260 break; 3261 case 1: /* eight banks */ 3262 rdev->config.si.tile_config |= 1 << 4; 3263 break; 3264 case 2: /* sixteen banks */ 3265 default: 3266 rdev->config.si.tile_config |= 2 << 4; 3267 break; 3268 } 3269 rdev->config.si.tile_config |= 3270 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8; 3271 rdev->config.si.tile_config |= 3272 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12; 3273 3274 WREG32(GB_ADDR_CONFIG, gb_addr_config); 3275 WREG32(DMIF_ADDR_CONFIG, gb_addr_config); 3276 WREG32(DMIF_ADDR_CALC, gb_addr_config); 3277 WREG32(HDP_ADDR_CONFIG, gb_addr_config); 3278 WREG32(DMA_TILING_CONFIG + DMA0_REGISTER_OFFSET, gb_addr_config); 3279 WREG32(DMA_TILING_CONFIG + DMA1_REGISTER_OFFSET, gb_addr_config); 3280 if (rdev->has_uvd) { 3281 WREG32(UVD_UDEC_ADDR_CONFIG, gb_addr_config); 3282 WREG32(UVD_UDEC_DB_ADDR_CONFIG, gb_addr_config); 3283 WREG32(UVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config); 3284 } 3285 3286 si_tiling_mode_table_init(rdev); 3287 3288 si_setup_rb(rdev, rdev->config.si.max_shader_engines, 3289 rdev->config.si.max_sh_per_se, 3290 rdev->config.si.max_backends_per_se); 3291 3292 si_setup_spi(rdev, rdev->config.si.max_shader_engines, 3293 rdev->config.si.max_sh_per_se, 3294 rdev->config.si.max_cu_per_sh); 3295 3296 rdev->config.si.active_cus = 0; 3297 for (i = 0; i < rdev->config.si.max_shader_engines; i++) { 3298 for (j = 0; j < rdev->config.si.max_sh_per_se; j++) { 3299 rdev->config.si.active_cus += 3300 hweight32(si_get_cu_active_bitmap(rdev, i, j)); 3301 } 3302 } 3303 3304 /* set HW defaults for 3D engine */ 3305 WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) | 3306 ROQ_IB2_START(0x2b))); 3307 WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60)); 3308 3309 sx_debug_1 = RREG32(SX_DEBUG_1); 3310 WREG32(SX_DEBUG_1, sx_debug_1); 3311 3312 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4)); 3313 3314 WREG32(PA_SC_FIFO_SIZE, (SC_FRONTEND_PRIM_FIFO_SIZE(rdev->config.si.sc_prim_fifo_size_frontend) | 3315 SC_BACKEND_PRIM_FIFO_SIZE(rdev->config.si.sc_prim_fifo_size_backend) | 3316 SC_HIZ_TILE_FIFO_SIZE(rdev->config.si.sc_hiz_tile_fifo_size) | 3317 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.si.sc_earlyz_tile_fifo_size))); 3318 3319 WREG32(VGT_NUM_INSTANCES, 1); 3320 3321 WREG32(CP_PERFMON_CNTL, 0); 3322 3323 WREG32(SQ_CONFIG, 0); 3324 3325 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) | 3326 FORCE_EOV_MAX_REZ_CNT(255))); 3327 3328 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) | 3329 AUTO_INVLD_EN(ES_AND_GS_AUTO)); 3330 3331 WREG32(VGT_GS_VERTEX_REUSE, 16); 3332 WREG32(PA_SC_LINE_STIPPLE_STATE, 0); 3333 3334 WREG32(CB_PERFCOUNTER0_SELECT0, 0); 3335 WREG32(CB_PERFCOUNTER0_SELECT1, 0); 3336 WREG32(CB_PERFCOUNTER1_SELECT0, 0); 3337 WREG32(CB_PERFCOUNTER1_SELECT1, 0); 3338 WREG32(CB_PERFCOUNTER2_SELECT0, 0); 3339 WREG32(CB_PERFCOUNTER2_SELECT1, 0); 3340 WREG32(CB_PERFCOUNTER3_SELECT0, 0); 3341 WREG32(CB_PERFCOUNTER3_SELECT1, 0); 3342 3343 tmp = RREG32(HDP_MISC_CNTL); 3344 tmp |= HDP_FLUSH_INVALIDATE_CACHE; 3345 WREG32(HDP_MISC_CNTL, tmp); 3346 3347 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL); 3348 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl); 3349 3350 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3)); 3351 3352 udelay(50); 3353 } 3354 3355 /* 3356 * GPU scratch registers helpers function. 3357 */ 3358 static void si_scratch_init(struct radeon_device *rdev) 3359 { 3360 int i; 3361 3362 rdev->scratch.num_reg = 7; 3363 rdev->scratch.reg_base = SCRATCH_REG0; 3364 for (i = 0; i < rdev->scratch.num_reg; i++) { 3365 rdev->scratch.free[i] = true; 3366 rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4); 3367 } 3368 } 3369 3370 void si_fence_ring_emit(struct radeon_device *rdev, 3371 struct radeon_fence *fence) 3372 { 3373 struct radeon_ring *ring = &rdev->ring[fence->ring]; 3374 u64 addr = rdev->fence_drv[fence->ring].gpu_addr; 3375 3376 /* flush read cache over gart */ 3377 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 3378 radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2); 3379 radeon_ring_write(ring, 0); 3380 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3)); 3381 radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA | 3382 PACKET3_TC_ACTION_ENA | 3383 PACKET3_SH_KCACHE_ACTION_ENA | 3384 PACKET3_SH_ICACHE_ACTION_ENA); 3385 radeon_ring_write(ring, 0xFFFFFFFF); 3386 radeon_ring_write(ring, 0); 3387 radeon_ring_write(ring, 10); /* poll interval */ 3388 /* EVENT_WRITE_EOP - flush caches, send int */ 3389 radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4)); 3390 radeon_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5)); 3391 radeon_ring_write(ring, lower_32_bits(addr)); 3392 radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | DATA_SEL(1) | INT_SEL(2)); 3393 radeon_ring_write(ring, fence->seq); 3394 radeon_ring_write(ring, 0); 3395 } 3396 3397 /* 3398 * IB stuff 3399 */ 3400 void si_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib) 3401 { 3402 struct radeon_ring *ring = &rdev->ring[ib->ring]; 3403 unsigned vm_id = ib->vm ? ib->vm->ids[ib->ring].id : 0; 3404 u32 header; 3405 3406 if (ib->is_const_ib) { 3407 /* set switch buffer packet before const IB */ 3408 radeon_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0)); 3409 radeon_ring_write(ring, 0); 3410 3411 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2); 3412 } else { 3413 u32 next_rptr; 3414 if (ring->rptr_save_reg) { 3415 next_rptr = ring->wptr + 3 + 4 + 8; 3416 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 3417 radeon_ring_write(ring, ((ring->rptr_save_reg - 3418 PACKET3_SET_CONFIG_REG_START) >> 2)); 3419 radeon_ring_write(ring, next_rptr); 3420 } else if (rdev->wb.enabled) { 3421 next_rptr = ring->wptr + 5 + 4 + 8; 3422 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 3423 radeon_ring_write(ring, (1 << 8)); 3424 radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc); 3425 radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr)); 3426 radeon_ring_write(ring, next_rptr); 3427 } 3428 3429 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); 3430 } 3431 3432 radeon_ring_write(ring, header); 3433 radeon_ring_write(ring, 3434 #ifdef __BIG_ENDIAN 3435 (2 << 0) | 3436 #endif 3437 (ib->gpu_addr & 0xFFFFFFFC)); 3438 radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF); 3439 radeon_ring_write(ring, ib->length_dw | (vm_id << 24)); 3440 3441 if (!ib->is_const_ib) { 3442 /* flush read cache over gart for this vmid */ 3443 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 3444 radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2); 3445 radeon_ring_write(ring, vm_id); 3446 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3)); 3447 radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA | 3448 PACKET3_TC_ACTION_ENA | 3449 PACKET3_SH_KCACHE_ACTION_ENA | 3450 PACKET3_SH_ICACHE_ACTION_ENA); 3451 radeon_ring_write(ring, 0xFFFFFFFF); 3452 radeon_ring_write(ring, 0); 3453 radeon_ring_write(ring, 10); /* poll interval */ 3454 } 3455 } 3456 3457 /* 3458 * CP. 3459 */ 3460 static void si_cp_enable(struct radeon_device *rdev, bool enable) 3461 { 3462 if (enable) 3463 WREG32(CP_ME_CNTL, 0); 3464 else { 3465 if (rdev->asic->copy.copy_ring_index == RADEON_RING_TYPE_GFX_INDEX) 3466 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size); 3467 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT)); 3468 WREG32(SCRATCH_UMSK, 0); 3469 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false; 3470 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false; 3471 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false; 3472 } 3473 udelay(50); 3474 } 3475 3476 static int si_cp_load_microcode(struct radeon_device *rdev) 3477 { 3478 int i; 3479 3480 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw) 3481 return -EINVAL; 3482 3483 si_cp_enable(rdev, false); 3484 3485 if (rdev->new_fw) { 3486 const struct gfx_firmware_header_v1_0 *pfp_hdr = 3487 (const struct gfx_firmware_header_v1_0 *)rdev->pfp_fw->data; 3488 const struct gfx_firmware_header_v1_0 *ce_hdr = 3489 (const struct gfx_firmware_header_v1_0 *)rdev->ce_fw->data; 3490 const struct gfx_firmware_header_v1_0 *me_hdr = 3491 (const struct gfx_firmware_header_v1_0 *)rdev->me_fw->data; 3492 const __le32 *fw_data; 3493 u32 fw_size; 3494 3495 radeon_ucode_print_gfx_hdr(&pfp_hdr->header); 3496 radeon_ucode_print_gfx_hdr(&ce_hdr->header); 3497 radeon_ucode_print_gfx_hdr(&me_hdr->header); 3498 3499 /* PFP */ 3500 fw_data = (const __le32 *) 3501 (rdev->pfp_fw->data + le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes)); 3502 fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes) / 4; 3503 WREG32(CP_PFP_UCODE_ADDR, 0); 3504 for (i = 0; i < fw_size; i++) 3505 WREG32(CP_PFP_UCODE_DATA, le32_to_cpup(fw_data++)); 3506 WREG32(CP_PFP_UCODE_ADDR, 0); 3507 3508 /* CE */ 3509 fw_data = (const __le32 *) 3510 (rdev->ce_fw->data + le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes)); 3511 fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes) / 4; 3512 WREG32(CP_CE_UCODE_ADDR, 0); 3513 for (i = 0; i < fw_size; i++) 3514 WREG32(CP_CE_UCODE_DATA, le32_to_cpup(fw_data++)); 3515 WREG32(CP_CE_UCODE_ADDR, 0); 3516 3517 /* ME */ 3518 fw_data = (const __be32 *) 3519 (rdev->me_fw->data + le32_to_cpu(me_hdr->header.ucode_array_offset_bytes)); 3520 fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes) / 4; 3521 WREG32(CP_ME_RAM_WADDR, 0); 3522 for (i = 0; i < fw_size; i++) 3523 WREG32(CP_ME_RAM_DATA, le32_to_cpup(fw_data++)); 3524 WREG32(CP_ME_RAM_WADDR, 0); 3525 } else { 3526 const __be32 *fw_data; 3527 3528 /* PFP */ 3529 fw_data = (const __be32 *)rdev->pfp_fw->data; 3530 WREG32(CP_PFP_UCODE_ADDR, 0); 3531 for (i = 0; i < SI_PFP_UCODE_SIZE; i++) 3532 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++)); 3533 WREG32(CP_PFP_UCODE_ADDR, 0); 3534 3535 /* CE */ 3536 fw_data = (const __be32 *)rdev->ce_fw->data; 3537 WREG32(CP_CE_UCODE_ADDR, 0); 3538 for (i = 0; i < SI_CE_UCODE_SIZE; i++) 3539 WREG32(CP_CE_UCODE_DATA, be32_to_cpup(fw_data++)); 3540 WREG32(CP_CE_UCODE_ADDR, 0); 3541 3542 /* ME */ 3543 fw_data = (const __be32 *)rdev->me_fw->data; 3544 WREG32(CP_ME_RAM_WADDR, 0); 3545 for (i = 0; i < SI_PM4_UCODE_SIZE; i++) 3546 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++)); 3547 WREG32(CP_ME_RAM_WADDR, 0); 3548 } 3549 3550 WREG32(CP_PFP_UCODE_ADDR, 0); 3551 WREG32(CP_CE_UCODE_ADDR, 0); 3552 WREG32(CP_ME_RAM_WADDR, 0); 3553 WREG32(CP_ME_RAM_RADDR, 0); 3554 return 0; 3555 } 3556 3557 static int si_cp_start(struct radeon_device *rdev) 3558 { 3559 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 3560 int r, i; 3561 3562 r = radeon_ring_lock(rdev, ring, 7 + 4); 3563 if (r) { 3564 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r); 3565 return r; 3566 } 3567 /* init the CP */ 3568 radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5)); 3569 radeon_ring_write(ring, 0x1); 3570 radeon_ring_write(ring, 0x0); 3571 radeon_ring_write(ring, rdev->config.si.max_hw_contexts - 1); 3572 radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1)); 3573 radeon_ring_write(ring, 0); 3574 radeon_ring_write(ring, 0); 3575 3576 /* init the CE partitions */ 3577 radeon_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2)); 3578 radeon_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE)); 3579 radeon_ring_write(ring, 0xc000); 3580 radeon_ring_write(ring, 0xe000); 3581 radeon_ring_unlock_commit(rdev, ring, false); 3582 3583 si_cp_enable(rdev, true); 3584 3585 r = radeon_ring_lock(rdev, ring, si_default_size + 10); 3586 if (r) { 3587 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r); 3588 return r; 3589 } 3590 3591 /* setup clear context state */ 3592 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 3593 radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 3594 3595 for (i = 0; i < si_default_size; i++) 3596 radeon_ring_write(ring, si_default_state[i]); 3597 3598 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 3599 radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE); 3600 3601 /* set clear context state */ 3602 radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0)); 3603 radeon_ring_write(ring, 0); 3604 3605 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2)); 3606 radeon_ring_write(ring, 0x00000316); 3607 radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */ 3608 radeon_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */ 3609 3610 radeon_ring_unlock_commit(rdev, ring, false); 3611 3612 for (i = RADEON_RING_TYPE_GFX_INDEX; i <= CAYMAN_RING_TYPE_CP2_INDEX; ++i) { 3613 ring = &rdev->ring[i]; 3614 r = radeon_ring_lock(rdev, ring, 2); 3615 3616 /* clear the compute context state */ 3617 radeon_ring_write(ring, PACKET3_COMPUTE(PACKET3_CLEAR_STATE, 0)); 3618 radeon_ring_write(ring, 0); 3619 3620 radeon_ring_unlock_commit(rdev, ring, false); 3621 } 3622 3623 return 0; 3624 } 3625 3626 static void si_cp_fini(struct radeon_device *rdev) 3627 { 3628 struct radeon_ring *ring; 3629 si_cp_enable(rdev, false); 3630 3631 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 3632 radeon_ring_fini(rdev, ring); 3633 radeon_scratch_free(rdev, ring->rptr_save_reg); 3634 3635 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX]; 3636 radeon_ring_fini(rdev, ring); 3637 radeon_scratch_free(rdev, ring->rptr_save_reg); 3638 3639 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX]; 3640 radeon_ring_fini(rdev, ring); 3641 radeon_scratch_free(rdev, ring->rptr_save_reg); 3642 } 3643 3644 static int si_cp_resume(struct radeon_device *rdev) 3645 { 3646 struct radeon_ring *ring; 3647 u32 tmp; 3648 u32 rb_bufsz; 3649 int r; 3650 3651 si_enable_gui_idle_interrupt(rdev, false); 3652 3653 WREG32(CP_SEM_WAIT_TIMER, 0x0); 3654 WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0); 3655 3656 /* Set the write pointer delay */ 3657 WREG32(CP_RB_WPTR_DELAY, 0); 3658 3659 WREG32(CP_DEBUG, 0); 3660 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF); 3661 3662 /* ring 0 - compute and gfx */ 3663 /* Set ring buffer size */ 3664 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 3665 rb_bufsz = order_base_2(ring->ring_size / 8); 3666 tmp = (order_base_2(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 3667 #ifdef __BIG_ENDIAN 3668 tmp |= BUF_SWAP_32BIT; 3669 #endif 3670 WREG32(CP_RB0_CNTL, tmp); 3671 3672 /* Initialize the ring buffer's read and write pointers */ 3673 WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA); 3674 ring->wptr = 0; 3675 WREG32(CP_RB0_WPTR, ring->wptr); 3676 3677 /* set the wb address whether it's enabled or not */ 3678 WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC); 3679 WREG32(CP_RB0_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF); 3680 3681 if (rdev->wb.enabled) 3682 WREG32(SCRATCH_UMSK, 0xff); 3683 else { 3684 tmp |= RB_NO_UPDATE; 3685 WREG32(SCRATCH_UMSK, 0); 3686 } 3687 3688 mdelay(1); 3689 WREG32(CP_RB0_CNTL, tmp); 3690 3691 WREG32(CP_RB0_BASE, ring->gpu_addr >> 8); 3692 3693 /* ring1 - compute only */ 3694 /* Set ring buffer size */ 3695 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX]; 3696 rb_bufsz = order_base_2(ring->ring_size / 8); 3697 tmp = (order_base_2(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 3698 #ifdef __BIG_ENDIAN 3699 tmp |= BUF_SWAP_32BIT; 3700 #endif 3701 WREG32(CP_RB1_CNTL, tmp); 3702 3703 /* Initialize the ring buffer's read and write pointers */ 3704 WREG32(CP_RB1_CNTL, tmp | RB_RPTR_WR_ENA); 3705 ring->wptr = 0; 3706 WREG32(CP_RB1_WPTR, ring->wptr); 3707 3708 /* set the wb address whether it's enabled or not */ 3709 WREG32(CP_RB1_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFFFFFFFC); 3710 WREG32(CP_RB1_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFF); 3711 3712 mdelay(1); 3713 WREG32(CP_RB1_CNTL, tmp); 3714 3715 WREG32(CP_RB1_BASE, ring->gpu_addr >> 8); 3716 3717 /* ring2 - compute only */ 3718 /* Set ring buffer size */ 3719 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX]; 3720 rb_bufsz = order_base_2(ring->ring_size / 8); 3721 tmp = (order_base_2(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz; 3722 #ifdef __BIG_ENDIAN 3723 tmp |= BUF_SWAP_32BIT; 3724 #endif 3725 WREG32(CP_RB2_CNTL, tmp); 3726 3727 /* Initialize the ring buffer's read and write pointers */ 3728 WREG32(CP_RB2_CNTL, tmp | RB_RPTR_WR_ENA); 3729 ring->wptr = 0; 3730 WREG32(CP_RB2_WPTR, ring->wptr); 3731 3732 /* set the wb address whether it's enabled or not */ 3733 WREG32(CP_RB2_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFFFFFFFC); 3734 WREG32(CP_RB2_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFF); 3735 3736 mdelay(1); 3737 WREG32(CP_RB2_CNTL, tmp); 3738 3739 WREG32(CP_RB2_BASE, ring->gpu_addr >> 8); 3740 3741 /* start the rings */ 3742 si_cp_start(rdev); 3743 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true; 3744 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = true; 3745 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = true; 3746 r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]); 3747 if (r) { 3748 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false; 3749 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false; 3750 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false; 3751 return r; 3752 } 3753 r = radeon_ring_test(rdev, CAYMAN_RING_TYPE_CP1_INDEX, &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX]); 3754 if (r) { 3755 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false; 3756 } 3757 r = radeon_ring_test(rdev, CAYMAN_RING_TYPE_CP2_INDEX, &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX]); 3758 if (r) { 3759 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false; 3760 } 3761 3762 si_enable_gui_idle_interrupt(rdev, true); 3763 3764 if (rdev->asic->copy.copy_ring_index == RADEON_RING_TYPE_GFX_INDEX) 3765 radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size); 3766 3767 return 0; 3768 } 3769 3770 u32 si_gpu_check_soft_reset(struct radeon_device *rdev) 3771 { 3772 u32 reset_mask = 0; 3773 u32 tmp; 3774 3775 /* GRBM_STATUS */ 3776 tmp = RREG32(GRBM_STATUS); 3777 if (tmp & (PA_BUSY | SC_BUSY | 3778 BCI_BUSY | SX_BUSY | 3779 TA_BUSY | VGT_BUSY | 3780 DB_BUSY | CB_BUSY | 3781 GDS_BUSY | SPI_BUSY | 3782 IA_BUSY | IA_BUSY_NO_DMA)) 3783 reset_mask |= RADEON_RESET_GFX; 3784 3785 if (tmp & (CF_RQ_PENDING | PF_RQ_PENDING | 3786 CP_BUSY | CP_COHERENCY_BUSY)) 3787 reset_mask |= RADEON_RESET_CP; 3788 3789 if (tmp & GRBM_EE_BUSY) 3790 reset_mask |= RADEON_RESET_GRBM | RADEON_RESET_GFX | RADEON_RESET_CP; 3791 3792 /* GRBM_STATUS2 */ 3793 tmp = RREG32(GRBM_STATUS2); 3794 if (tmp & (RLC_RQ_PENDING | RLC_BUSY)) 3795 reset_mask |= RADEON_RESET_RLC; 3796 3797 /* DMA_STATUS_REG 0 */ 3798 tmp = RREG32(DMA_STATUS_REG + DMA0_REGISTER_OFFSET); 3799 if (!(tmp & DMA_IDLE)) 3800 reset_mask |= RADEON_RESET_DMA; 3801 3802 /* DMA_STATUS_REG 1 */ 3803 tmp = RREG32(DMA_STATUS_REG + DMA1_REGISTER_OFFSET); 3804 if (!(tmp & DMA_IDLE)) 3805 reset_mask |= RADEON_RESET_DMA1; 3806 3807 /* SRBM_STATUS2 */ 3808 tmp = RREG32(SRBM_STATUS2); 3809 if (tmp & DMA_BUSY) 3810 reset_mask |= RADEON_RESET_DMA; 3811 3812 if (tmp & DMA1_BUSY) 3813 reset_mask |= RADEON_RESET_DMA1; 3814 3815 /* SRBM_STATUS */ 3816 tmp = RREG32(SRBM_STATUS); 3817 3818 if (tmp & IH_BUSY) 3819 reset_mask |= RADEON_RESET_IH; 3820 3821 if (tmp & SEM_BUSY) 3822 reset_mask |= RADEON_RESET_SEM; 3823 3824 if (tmp & GRBM_RQ_PENDING) 3825 reset_mask |= RADEON_RESET_GRBM; 3826 3827 if (tmp & VMC_BUSY) 3828 reset_mask |= RADEON_RESET_VMC; 3829 3830 if (tmp & (MCB_BUSY | MCB_NON_DISPLAY_BUSY | 3831 MCC_BUSY | MCD_BUSY)) 3832 reset_mask |= RADEON_RESET_MC; 3833 3834 if (evergreen_is_display_hung(rdev)) 3835 reset_mask |= RADEON_RESET_DISPLAY; 3836 3837 /* VM_L2_STATUS */ 3838 tmp = RREG32(VM_L2_STATUS); 3839 if (tmp & L2_BUSY) 3840 reset_mask |= RADEON_RESET_VMC; 3841 3842 /* Skip MC reset as it's mostly likely not hung, just busy */ 3843 if (reset_mask & RADEON_RESET_MC) { 3844 DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask); 3845 reset_mask &= ~RADEON_RESET_MC; 3846 } 3847 3848 return reset_mask; 3849 } 3850 3851 static void si_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask) 3852 { 3853 struct evergreen_mc_save save; 3854 u32 grbm_soft_reset = 0, srbm_soft_reset = 0; 3855 u32 tmp; 3856 3857 if (reset_mask == 0) 3858 return; 3859 3860 dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask); 3861 3862 evergreen_print_gpu_status_regs(rdev); 3863 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n", 3864 RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR)); 3865 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n", 3866 RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS)); 3867 3868 /* disable PG/CG */ 3869 si_fini_pg(rdev); 3870 si_fini_cg(rdev); 3871 3872 /* stop the rlc */ 3873 si_rlc_stop(rdev); 3874 3875 /* Disable CP parsing/prefetching */ 3876 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT); 3877 3878 if (reset_mask & RADEON_RESET_DMA) { 3879 /* dma0 */ 3880 tmp = RREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET); 3881 tmp &= ~DMA_RB_ENABLE; 3882 WREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET, tmp); 3883 } 3884 if (reset_mask & RADEON_RESET_DMA1) { 3885 /* dma1 */ 3886 tmp = RREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET); 3887 tmp &= ~DMA_RB_ENABLE; 3888 WREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET, tmp); 3889 } 3890 3891 udelay(50); 3892 3893 evergreen_mc_stop(rdev, &save); 3894 if (evergreen_mc_wait_for_idle(rdev)) { 3895 dev_warn(rdev->dev, "Wait for MC idle timedout !\n"); 3896 } 3897 3898 if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE | RADEON_RESET_CP)) { 3899 grbm_soft_reset = SOFT_RESET_CB | 3900 SOFT_RESET_DB | 3901 SOFT_RESET_GDS | 3902 SOFT_RESET_PA | 3903 SOFT_RESET_SC | 3904 SOFT_RESET_BCI | 3905 SOFT_RESET_SPI | 3906 SOFT_RESET_SX | 3907 SOFT_RESET_TC | 3908 SOFT_RESET_TA | 3909 SOFT_RESET_VGT | 3910 SOFT_RESET_IA; 3911 } 3912 3913 if (reset_mask & RADEON_RESET_CP) { 3914 grbm_soft_reset |= SOFT_RESET_CP | SOFT_RESET_VGT; 3915 3916 srbm_soft_reset |= SOFT_RESET_GRBM; 3917 } 3918 3919 if (reset_mask & RADEON_RESET_DMA) 3920 srbm_soft_reset |= SOFT_RESET_DMA; 3921 3922 if (reset_mask & RADEON_RESET_DMA1) 3923 srbm_soft_reset |= SOFT_RESET_DMA1; 3924 3925 if (reset_mask & RADEON_RESET_DISPLAY) 3926 srbm_soft_reset |= SOFT_RESET_DC; 3927 3928 if (reset_mask & RADEON_RESET_RLC) 3929 grbm_soft_reset |= SOFT_RESET_RLC; 3930 3931 if (reset_mask & RADEON_RESET_SEM) 3932 srbm_soft_reset |= SOFT_RESET_SEM; 3933 3934 if (reset_mask & RADEON_RESET_IH) 3935 srbm_soft_reset |= SOFT_RESET_IH; 3936 3937 if (reset_mask & RADEON_RESET_GRBM) 3938 srbm_soft_reset |= SOFT_RESET_GRBM; 3939 3940 if (reset_mask & RADEON_RESET_VMC) 3941 srbm_soft_reset |= SOFT_RESET_VMC; 3942 3943 if (reset_mask & RADEON_RESET_MC) 3944 srbm_soft_reset |= SOFT_RESET_MC; 3945 3946 if (grbm_soft_reset) { 3947 tmp = RREG32(GRBM_SOFT_RESET); 3948 tmp |= grbm_soft_reset; 3949 dev_info(rdev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp); 3950 WREG32(GRBM_SOFT_RESET, tmp); 3951 tmp = RREG32(GRBM_SOFT_RESET); 3952 3953 udelay(50); 3954 3955 tmp &= ~grbm_soft_reset; 3956 WREG32(GRBM_SOFT_RESET, tmp); 3957 tmp = RREG32(GRBM_SOFT_RESET); 3958 } 3959 3960 if (srbm_soft_reset) { 3961 tmp = RREG32(SRBM_SOFT_RESET); 3962 tmp |= srbm_soft_reset; 3963 dev_info(rdev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp); 3964 WREG32(SRBM_SOFT_RESET, tmp); 3965 tmp = RREG32(SRBM_SOFT_RESET); 3966 3967 udelay(50); 3968 3969 tmp &= ~srbm_soft_reset; 3970 WREG32(SRBM_SOFT_RESET, tmp); 3971 tmp = RREG32(SRBM_SOFT_RESET); 3972 } 3973 3974 /* Wait a little for things to settle down */ 3975 udelay(50); 3976 3977 evergreen_mc_resume(rdev, &save); 3978 udelay(50); 3979 3980 evergreen_print_gpu_status_regs(rdev); 3981 } 3982 3983 static void si_set_clk_bypass_mode(struct radeon_device *rdev) 3984 { 3985 u32 tmp, i; 3986 3987 tmp = RREG32(CG_SPLL_FUNC_CNTL); 3988 tmp |= SPLL_BYPASS_EN; 3989 WREG32(CG_SPLL_FUNC_CNTL, tmp); 3990 3991 tmp = RREG32(CG_SPLL_FUNC_CNTL_2); 3992 tmp |= SPLL_CTLREQ_CHG; 3993 WREG32(CG_SPLL_FUNC_CNTL_2, tmp); 3994 3995 for (i = 0; i < rdev->usec_timeout; i++) { 3996 if (RREG32(SPLL_STATUS) & SPLL_CHG_STATUS) 3997 break; 3998 udelay(1); 3999 } 4000 4001 tmp = RREG32(CG_SPLL_FUNC_CNTL_2); 4002 tmp &= ~(SPLL_CTLREQ_CHG | SCLK_MUX_UPDATE); 4003 WREG32(CG_SPLL_FUNC_CNTL_2, tmp); 4004 4005 tmp = RREG32(MPLL_CNTL_MODE); 4006 tmp &= ~MPLL_MCLK_SEL; 4007 WREG32(MPLL_CNTL_MODE, tmp); 4008 } 4009 4010 static void si_spll_powerdown(struct radeon_device *rdev) 4011 { 4012 u32 tmp; 4013 4014 tmp = RREG32(SPLL_CNTL_MODE); 4015 tmp |= SPLL_SW_DIR_CONTROL; 4016 WREG32(SPLL_CNTL_MODE, tmp); 4017 4018 tmp = RREG32(CG_SPLL_FUNC_CNTL); 4019 tmp |= SPLL_RESET; 4020 WREG32(CG_SPLL_FUNC_CNTL, tmp); 4021 4022 tmp = RREG32(CG_SPLL_FUNC_CNTL); 4023 tmp |= SPLL_SLEEP; 4024 WREG32(CG_SPLL_FUNC_CNTL, tmp); 4025 4026 tmp = RREG32(SPLL_CNTL_MODE); 4027 tmp &= ~SPLL_SW_DIR_CONTROL; 4028 WREG32(SPLL_CNTL_MODE, tmp); 4029 } 4030 4031 static void si_gpu_pci_config_reset(struct radeon_device *rdev) 4032 { 4033 struct evergreen_mc_save save; 4034 u32 tmp, i; 4035 4036 dev_info(rdev->dev, "GPU pci config reset\n"); 4037 4038 /* disable dpm? */ 4039 4040 /* disable cg/pg */ 4041 si_fini_pg(rdev); 4042 si_fini_cg(rdev); 4043 4044 /* Disable CP parsing/prefetching */ 4045 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT); 4046 /* dma0 */ 4047 tmp = RREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET); 4048 tmp &= ~DMA_RB_ENABLE; 4049 WREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET, tmp); 4050 /* dma1 */ 4051 tmp = RREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET); 4052 tmp &= ~DMA_RB_ENABLE; 4053 WREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET, tmp); 4054 /* XXX other engines? */ 4055 4056 /* halt the rlc, disable cp internal ints */ 4057 si_rlc_stop(rdev); 4058 4059 udelay(50); 4060 4061 /* disable mem access */ 4062 evergreen_mc_stop(rdev, &save); 4063 if (evergreen_mc_wait_for_idle(rdev)) { 4064 dev_warn(rdev->dev, "Wait for MC idle timed out !\n"); 4065 } 4066 4067 /* set mclk/sclk to bypass */ 4068 si_set_clk_bypass_mode(rdev); 4069 /* powerdown spll */ 4070 si_spll_powerdown(rdev); 4071 /* disable BM */ 4072 pci_clear_master(rdev->pdev); 4073 /* reset */ 4074 radeon_pci_config_reset(rdev); 4075 /* wait for asic to come out of reset */ 4076 for (i = 0; i < rdev->usec_timeout; i++) { 4077 if (RREG32(CONFIG_MEMSIZE) != 0xffffffff) 4078 break; 4079 udelay(1); 4080 } 4081 } 4082 4083 int si_asic_reset(struct radeon_device *rdev, bool hard) 4084 { 4085 u32 reset_mask; 4086 4087 if (hard) { 4088 si_gpu_pci_config_reset(rdev); 4089 return 0; 4090 } 4091 4092 reset_mask = si_gpu_check_soft_reset(rdev); 4093 4094 if (reset_mask) 4095 r600_set_bios_scratch_engine_hung(rdev, true); 4096 4097 /* try soft reset */ 4098 si_gpu_soft_reset(rdev, reset_mask); 4099 4100 reset_mask = si_gpu_check_soft_reset(rdev); 4101 4102 /* try pci config reset */ 4103 if (reset_mask && radeon_hard_reset) 4104 si_gpu_pci_config_reset(rdev); 4105 4106 reset_mask = si_gpu_check_soft_reset(rdev); 4107 4108 if (!reset_mask) 4109 r600_set_bios_scratch_engine_hung(rdev, false); 4110 4111 return 0; 4112 } 4113 4114 /** 4115 * si_gfx_is_lockup - Check if the GFX engine is locked up 4116 * 4117 * @rdev: radeon_device pointer 4118 * @ring: radeon_ring structure holding ring information 4119 * 4120 * Check if the GFX engine is locked up. 4121 * Returns true if the engine appears to be locked up, false if not. 4122 */ 4123 bool si_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring) 4124 { 4125 u32 reset_mask = si_gpu_check_soft_reset(rdev); 4126 4127 if (!(reset_mask & (RADEON_RESET_GFX | 4128 RADEON_RESET_COMPUTE | 4129 RADEON_RESET_CP))) { 4130 radeon_ring_lockup_update(rdev, ring); 4131 return false; 4132 } 4133 return radeon_ring_test_lockup(rdev, ring); 4134 } 4135 4136 /* MC */ 4137 static void si_mc_program(struct radeon_device *rdev) 4138 { 4139 struct evergreen_mc_save save; 4140 u32 tmp; 4141 int i, j; 4142 4143 /* Initialize HDP */ 4144 for (i = 0, j = 0; i < 32; i++, j += 0x18) { 4145 WREG32((0x2c14 + j), 0x00000000); 4146 WREG32((0x2c18 + j), 0x00000000); 4147 WREG32((0x2c1c + j), 0x00000000); 4148 WREG32((0x2c20 + j), 0x00000000); 4149 WREG32((0x2c24 + j), 0x00000000); 4150 } 4151 WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0); 4152 4153 evergreen_mc_stop(rdev, &save); 4154 if (radeon_mc_wait_for_idle(rdev)) { 4155 dev_warn(rdev->dev, "Wait for MC idle timedout !\n"); 4156 } 4157 if (!ASIC_IS_NODCE(rdev)) 4158 /* Lockout access through VGA aperture*/ 4159 WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE); 4160 /* Update configuration */ 4161 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR, 4162 rdev->mc.vram_start >> 12); 4163 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR, 4164 rdev->mc.vram_end >> 12); 4165 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, 4166 rdev->vram_scratch.gpu_addr >> 12); 4167 tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16; 4168 tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF); 4169 WREG32(MC_VM_FB_LOCATION, tmp); 4170 /* XXX double check these! */ 4171 WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8)); 4172 WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30)); 4173 WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF); 4174 WREG32(MC_VM_AGP_BASE, 0); 4175 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF); 4176 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF); 4177 if (radeon_mc_wait_for_idle(rdev)) { 4178 dev_warn(rdev->dev, "Wait for MC idle timedout !\n"); 4179 } 4180 evergreen_mc_resume(rdev, &save); 4181 if (!ASIC_IS_NODCE(rdev)) { 4182 /* we need to own VRAM, so turn off the VGA renderer here 4183 * to stop it overwriting our objects */ 4184 rv515_vga_render_disable(rdev); 4185 } 4186 } 4187 4188 void si_vram_gtt_location(struct radeon_device *rdev, 4189 struct radeon_mc *mc) 4190 { 4191 if (mc->mc_vram_size > 0xFFC0000000ULL) { 4192 /* leave room for at least 1024M GTT */ 4193 dev_warn(rdev->dev, "limiting VRAM\n"); 4194 mc->real_vram_size = 0xFFC0000000ULL; 4195 mc->mc_vram_size = 0xFFC0000000ULL; 4196 } 4197 radeon_vram_location(rdev, &rdev->mc, 0); 4198 rdev->mc.gtt_base_align = 0; 4199 radeon_gtt_location(rdev, mc); 4200 } 4201 4202 static int si_mc_init(struct radeon_device *rdev) 4203 { 4204 u32 tmp; 4205 int chansize, numchan; 4206 4207 /* Get VRAM informations */ 4208 rdev->mc.vram_is_ddr = true; 4209 tmp = RREG32(MC_ARB_RAMCFG); 4210 if (tmp & CHANSIZE_OVERRIDE) { 4211 chansize = 16; 4212 } else if (tmp & CHANSIZE_MASK) { 4213 chansize = 64; 4214 } else { 4215 chansize = 32; 4216 } 4217 tmp = RREG32(MC_SHARED_CHMAP); 4218 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) { 4219 case 0: 4220 default: 4221 numchan = 1; 4222 break; 4223 case 1: 4224 numchan = 2; 4225 break; 4226 case 2: 4227 numchan = 4; 4228 break; 4229 case 3: 4230 numchan = 8; 4231 break; 4232 case 4: 4233 numchan = 3; 4234 break; 4235 case 5: 4236 numchan = 6; 4237 break; 4238 case 6: 4239 numchan = 10; 4240 break; 4241 case 7: 4242 numchan = 12; 4243 break; 4244 case 8: 4245 numchan = 16; 4246 break; 4247 } 4248 rdev->mc.vram_width = numchan * chansize; 4249 /* Could aper size report 0 ? */ 4250 rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0); 4251 rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0); 4252 /* size in MB on si */ 4253 tmp = RREG32(CONFIG_MEMSIZE); 4254 /* some boards may have garbage in the upper 16 bits */ 4255 if (tmp & 0xffff0000) { 4256 DRM_INFO("Probable bad vram size: 0x%08x\n", tmp); 4257 if (tmp & 0xffff) 4258 tmp &= 0xffff; 4259 } 4260 rdev->mc.mc_vram_size = tmp * 1024ULL * 1024ULL; 4261 rdev->mc.real_vram_size = rdev->mc.mc_vram_size; 4262 rdev->mc.visible_vram_size = rdev->mc.aper_size; 4263 si_vram_gtt_location(rdev, &rdev->mc); 4264 radeon_update_bandwidth_info(rdev); 4265 4266 return 0; 4267 } 4268 4269 /* 4270 * GART 4271 */ 4272 void si_pcie_gart_tlb_flush(struct radeon_device *rdev) 4273 { 4274 /* flush hdp cache */ 4275 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1); 4276 4277 /* bits 0-15 are the VM contexts0-15 */ 4278 WREG32(VM_INVALIDATE_REQUEST, 1); 4279 } 4280 4281 static int si_pcie_gart_enable(struct radeon_device *rdev) 4282 { 4283 int r, i; 4284 4285 if (rdev->gart.robj == NULL) { 4286 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n"); 4287 return -EINVAL; 4288 } 4289 r = radeon_gart_table_vram_pin(rdev); 4290 if (r) 4291 return r; 4292 /* Setup TLB control */ 4293 WREG32(MC_VM_MX_L1_TLB_CNTL, 4294 (0xA << 7) | 4295 ENABLE_L1_TLB | 4296 ENABLE_L1_FRAGMENT_PROCESSING | 4297 SYSTEM_ACCESS_MODE_NOT_IN_SYS | 4298 ENABLE_ADVANCED_DRIVER_MODEL | 4299 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU); 4300 /* Setup L2 cache */ 4301 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | 4302 ENABLE_L2_FRAGMENT_PROCESSING | 4303 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE | 4304 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE | 4305 EFFECTIVE_L2_QUEUE_SIZE(7) | 4306 CONTEXT1_IDENTITY_ACCESS_MODE(1)); 4307 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE); 4308 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY | 4309 BANK_SELECT(4) | 4310 L2_CACHE_BIGK_FRAGMENT_SIZE(4)); 4311 /* setup context0 */ 4312 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12); 4313 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12); 4314 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12); 4315 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR, 4316 (u32)(rdev->dummy_page.addr >> 12)); 4317 WREG32(VM_CONTEXT0_CNTL2, 0); 4318 WREG32(VM_CONTEXT0_CNTL, (ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) | 4319 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT)); 4320 4321 WREG32(0x15D4, 0); 4322 WREG32(0x15D8, 0); 4323 WREG32(0x15DC, 0); 4324 4325 /* empty context1-15 */ 4326 /* set vm size, must be a multiple of 4 */ 4327 WREG32(VM_CONTEXT1_PAGE_TABLE_START_ADDR, 0); 4328 WREG32(VM_CONTEXT1_PAGE_TABLE_END_ADDR, rdev->vm_manager.max_pfn - 1); 4329 /* Assign the pt base to something valid for now; the pts used for 4330 * the VMs are determined by the application and setup and assigned 4331 * on the fly in the vm part of radeon_gart.c 4332 */ 4333 for (i = 1; i < 16; i++) { 4334 if (i < 8) 4335 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2), 4336 rdev->vm_manager.saved_table_addr[i]); 4337 else 4338 WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((i - 8) << 2), 4339 rdev->vm_manager.saved_table_addr[i]); 4340 } 4341 4342 /* enable context1-15 */ 4343 WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR, 4344 (u32)(rdev->dummy_page.addr >> 12)); 4345 WREG32(VM_CONTEXT1_CNTL2, 4); 4346 WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) | 4347 PAGE_TABLE_BLOCK_SIZE(radeon_vm_block_size - 9) | 4348 RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT | 4349 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT | 4350 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT | 4351 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT | 4352 PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT | 4353 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT | 4354 VALID_PROTECTION_FAULT_ENABLE_INTERRUPT | 4355 VALID_PROTECTION_FAULT_ENABLE_DEFAULT | 4356 READ_PROTECTION_FAULT_ENABLE_INTERRUPT | 4357 READ_PROTECTION_FAULT_ENABLE_DEFAULT | 4358 WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT | 4359 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT); 4360 4361 si_pcie_gart_tlb_flush(rdev); 4362 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", 4363 (unsigned)(rdev->mc.gtt_size >> 20), 4364 (unsigned long long)rdev->gart.table_addr); 4365 rdev->gart.ready = true; 4366 return 0; 4367 } 4368 4369 static void si_pcie_gart_disable(struct radeon_device *rdev) 4370 { 4371 unsigned i; 4372 4373 for (i = 1; i < 16; ++i) { 4374 uint32_t reg; 4375 if (i < 8) 4376 reg = VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2); 4377 else 4378 reg = VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((i - 8) << 2); 4379 rdev->vm_manager.saved_table_addr[i] = RREG32(reg); 4380 } 4381 4382 /* Disable all tables */ 4383 WREG32(VM_CONTEXT0_CNTL, 0); 4384 WREG32(VM_CONTEXT1_CNTL, 0); 4385 /* Setup TLB control */ 4386 WREG32(MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE_NOT_IN_SYS | 4387 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU); 4388 /* Setup L2 cache */ 4389 WREG32(VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE | 4390 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE | 4391 EFFECTIVE_L2_QUEUE_SIZE(7) | 4392 CONTEXT1_IDENTITY_ACCESS_MODE(1)); 4393 WREG32(VM_L2_CNTL2, 0); 4394 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY | 4395 L2_CACHE_BIGK_FRAGMENT_SIZE(0)); 4396 radeon_gart_table_vram_unpin(rdev); 4397 } 4398 4399 static void si_pcie_gart_fini(struct radeon_device *rdev) 4400 { 4401 si_pcie_gart_disable(rdev); 4402 radeon_gart_table_vram_free(rdev); 4403 radeon_gart_fini(rdev); 4404 } 4405 4406 /* vm parser */ 4407 static bool si_vm_reg_valid(u32 reg) 4408 { 4409 /* context regs are fine */ 4410 if (reg >= 0x28000) 4411 return true; 4412 4413 /* shader regs are also fine */ 4414 if (reg >= 0xB000 && reg < 0xC000) 4415 return true; 4416 4417 /* check config regs */ 4418 switch (reg) { 4419 case GRBM_GFX_INDEX: 4420 case CP_STRMOUT_CNTL: 4421 case VGT_VTX_VECT_EJECT_REG: 4422 case VGT_CACHE_INVALIDATION: 4423 case VGT_ESGS_RING_SIZE: 4424 case VGT_GSVS_RING_SIZE: 4425 case VGT_GS_VERTEX_REUSE: 4426 case VGT_PRIMITIVE_TYPE: 4427 case VGT_INDEX_TYPE: 4428 case VGT_NUM_INDICES: 4429 case VGT_NUM_INSTANCES: 4430 case VGT_TF_RING_SIZE: 4431 case VGT_HS_OFFCHIP_PARAM: 4432 case VGT_TF_MEMORY_BASE: 4433 case PA_CL_ENHANCE: 4434 case PA_SU_LINE_STIPPLE_VALUE: 4435 case PA_SC_LINE_STIPPLE_STATE: 4436 case PA_SC_ENHANCE: 4437 case SQC_CACHES: 4438 case SPI_STATIC_THREAD_MGMT_1: 4439 case SPI_STATIC_THREAD_MGMT_2: 4440 case SPI_STATIC_THREAD_MGMT_3: 4441 case SPI_PS_MAX_WAVE_ID: 4442 case SPI_CONFIG_CNTL: 4443 case SPI_CONFIG_CNTL_1: 4444 case TA_CNTL_AUX: 4445 case TA_CS_BC_BASE_ADDR: 4446 return true; 4447 default: 4448 DRM_ERROR("Invalid register 0x%x in CS\n", reg); 4449 return false; 4450 } 4451 } 4452 4453 static int si_vm_packet3_ce_check(struct radeon_device *rdev, 4454 u32 *ib, struct radeon_cs_packet *pkt) 4455 { 4456 switch (pkt->opcode) { 4457 case PACKET3_NOP: 4458 case PACKET3_SET_BASE: 4459 case PACKET3_SET_CE_DE_COUNTERS: 4460 case PACKET3_LOAD_CONST_RAM: 4461 case PACKET3_WRITE_CONST_RAM: 4462 case PACKET3_WRITE_CONST_RAM_OFFSET: 4463 case PACKET3_DUMP_CONST_RAM: 4464 case PACKET3_INCREMENT_CE_COUNTER: 4465 case PACKET3_WAIT_ON_DE_COUNTER: 4466 case PACKET3_CE_WRITE: 4467 break; 4468 default: 4469 DRM_ERROR("Invalid CE packet3: 0x%x\n", pkt->opcode); 4470 return -EINVAL; 4471 } 4472 return 0; 4473 } 4474 4475 static int si_vm_packet3_cp_dma_check(u32 *ib, u32 idx) 4476 { 4477 u32 start_reg, reg, i; 4478 u32 command = ib[idx + 4]; 4479 u32 info = ib[idx + 1]; 4480 u32 idx_value = ib[idx]; 4481 if (command & PACKET3_CP_DMA_CMD_SAS) { 4482 /* src address space is register */ 4483 if (((info & 0x60000000) >> 29) == 0) { 4484 start_reg = idx_value << 2; 4485 if (command & PACKET3_CP_DMA_CMD_SAIC) { 4486 reg = start_reg; 4487 if (!si_vm_reg_valid(reg)) { 4488 DRM_ERROR("CP DMA Bad SRC register\n"); 4489 return -EINVAL; 4490 } 4491 } else { 4492 for (i = 0; i < (command & 0x1fffff); i++) { 4493 reg = start_reg + (4 * i); 4494 if (!si_vm_reg_valid(reg)) { 4495 DRM_ERROR("CP DMA Bad SRC register\n"); 4496 return -EINVAL; 4497 } 4498 } 4499 } 4500 } 4501 } 4502 if (command & PACKET3_CP_DMA_CMD_DAS) { 4503 /* dst address space is register */ 4504 if (((info & 0x00300000) >> 20) == 0) { 4505 start_reg = ib[idx + 2]; 4506 if (command & PACKET3_CP_DMA_CMD_DAIC) { 4507 reg = start_reg; 4508 if (!si_vm_reg_valid(reg)) { 4509 DRM_ERROR("CP DMA Bad DST register\n"); 4510 return -EINVAL; 4511 } 4512 } else { 4513 for (i = 0; i < (command & 0x1fffff); i++) { 4514 reg = start_reg + (4 * i); 4515 if (!si_vm_reg_valid(reg)) { 4516 DRM_ERROR("CP DMA Bad DST register\n"); 4517 return -EINVAL; 4518 } 4519 } 4520 } 4521 } 4522 } 4523 return 0; 4524 } 4525 4526 static int si_vm_packet3_gfx_check(struct radeon_device *rdev, 4527 u32 *ib, struct radeon_cs_packet *pkt) 4528 { 4529 int r; 4530 u32 idx = pkt->idx + 1; 4531 u32 idx_value = ib[idx]; 4532 u32 start_reg, end_reg, reg, i; 4533 4534 switch (pkt->opcode) { 4535 case PACKET3_NOP: 4536 case PACKET3_SET_BASE: 4537 case PACKET3_CLEAR_STATE: 4538 case PACKET3_INDEX_BUFFER_SIZE: 4539 case PACKET3_DISPATCH_DIRECT: 4540 case PACKET3_DISPATCH_INDIRECT: 4541 case PACKET3_ALLOC_GDS: 4542 case PACKET3_WRITE_GDS_RAM: 4543 case PACKET3_ATOMIC_GDS: 4544 case PACKET3_ATOMIC: 4545 case PACKET3_OCCLUSION_QUERY: 4546 case PACKET3_SET_PREDICATION: 4547 case PACKET3_COND_EXEC: 4548 case PACKET3_PRED_EXEC: 4549 case PACKET3_DRAW_INDIRECT: 4550 case PACKET3_DRAW_INDEX_INDIRECT: 4551 case PACKET3_INDEX_BASE: 4552 case PACKET3_DRAW_INDEX_2: 4553 case PACKET3_CONTEXT_CONTROL: 4554 case PACKET3_INDEX_TYPE: 4555 case PACKET3_DRAW_INDIRECT_MULTI: 4556 case PACKET3_DRAW_INDEX_AUTO: 4557 case PACKET3_DRAW_INDEX_IMMD: 4558 case PACKET3_NUM_INSTANCES: 4559 case PACKET3_DRAW_INDEX_MULTI_AUTO: 4560 case PACKET3_STRMOUT_BUFFER_UPDATE: 4561 case PACKET3_DRAW_INDEX_OFFSET_2: 4562 case PACKET3_DRAW_INDEX_MULTI_ELEMENT: 4563 case PACKET3_DRAW_INDEX_INDIRECT_MULTI: 4564 case PACKET3_MPEG_INDEX: 4565 case PACKET3_WAIT_REG_MEM: 4566 case PACKET3_MEM_WRITE: 4567 case PACKET3_PFP_SYNC_ME: 4568 case PACKET3_SURFACE_SYNC: 4569 case PACKET3_EVENT_WRITE: 4570 case PACKET3_EVENT_WRITE_EOP: 4571 case PACKET3_EVENT_WRITE_EOS: 4572 case PACKET3_SET_CONTEXT_REG: 4573 case PACKET3_SET_CONTEXT_REG_INDIRECT: 4574 case PACKET3_SET_SH_REG: 4575 case PACKET3_SET_SH_REG_OFFSET: 4576 case PACKET3_INCREMENT_DE_COUNTER: 4577 case PACKET3_WAIT_ON_CE_COUNTER: 4578 case PACKET3_WAIT_ON_AVAIL_BUFFER: 4579 case PACKET3_ME_WRITE: 4580 break; 4581 case PACKET3_COPY_DATA: 4582 if ((idx_value & 0xf00) == 0) { 4583 reg = ib[idx + 3] * 4; 4584 if (!si_vm_reg_valid(reg)) 4585 return -EINVAL; 4586 } 4587 break; 4588 case PACKET3_WRITE_DATA: 4589 if ((idx_value & 0xf00) == 0) { 4590 start_reg = ib[idx + 1] * 4; 4591 if (idx_value & 0x10000) { 4592 if (!si_vm_reg_valid(start_reg)) 4593 return -EINVAL; 4594 } else { 4595 for (i = 0; i < (pkt->count - 2); i++) { 4596 reg = start_reg + (4 * i); 4597 if (!si_vm_reg_valid(reg)) 4598 return -EINVAL; 4599 } 4600 } 4601 } 4602 break; 4603 case PACKET3_COND_WRITE: 4604 if (idx_value & 0x100) { 4605 reg = ib[idx + 5] * 4; 4606 if (!si_vm_reg_valid(reg)) 4607 return -EINVAL; 4608 } 4609 break; 4610 case PACKET3_COPY_DW: 4611 if (idx_value & 0x2) { 4612 reg = ib[idx + 3] * 4; 4613 if (!si_vm_reg_valid(reg)) 4614 return -EINVAL; 4615 } 4616 break; 4617 case PACKET3_SET_CONFIG_REG: 4618 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START; 4619 end_reg = 4 * pkt->count + start_reg - 4; 4620 if ((start_reg < PACKET3_SET_CONFIG_REG_START) || 4621 (start_reg >= PACKET3_SET_CONFIG_REG_END) || 4622 (end_reg >= PACKET3_SET_CONFIG_REG_END)) { 4623 DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n"); 4624 return -EINVAL; 4625 } 4626 for (i = 0; i < pkt->count; i++) { 4627 reg = start_reg + (4 * i); 4628 if (!si_vm_reg_valid(reg)) 4629 return -EINVAL; 4630 } 4631 break; 4632 case PACKET3_CP_DMA: 4633 r = si_vm_packet3_cp_dma_check(ib, idx); 4634 if (r) 4635 return r; 4636 break; 4637 default: 4638 DRM_ERROR("Invalid GFX packet3: 0x%x\n", pkt->opcode); 4639 return -EINVAL; 4640 } 4641 return 0; 4642 } 4643 4644 static int si_vm_packet3_compute_check(struct radeon_device *rdev, 4645 u32 *ib, struct radeon_cs_packet *pkt) 4646 { 4647 int r; 4648 u32 idx = pkt->idx + 1; 4649 u32 idx_value = ib[idx]; 4650 u32 start_reg, reg, i; 4651 4652 switch (pkt->opcode) { 4653 case PACKET3_NOP: 4654 case PACKET3_SET_BASE: 4655 case PACKET3_CLEAR_STATE: 4656 case PACKET3_DISPATCH_DIRECT: 4657 case PACKET3_DISPATCH_INDIRECT: 4658 case PACKET3_ALLOC_GDS: 4659 case PACKET3_WRITE_GDS_RAM: 4660 case PACKET3_ATOMIC_GDS: 4661 case PACKET3_ATOMIC: 4662 case PACKET3_OCCLUSION_QUERY: 4663 case PACKET3_SET_PREDICATION: 4664 case PACKET3_COND_EXEC: 4665 case PACKET3_PRED_EXEC: 4666 case PACKET3_CONTEXT_CONTROL: 4667 case PACKET3_STRMOUT_BUFFER_UPDATE: 4668 case PACKET3_WAIT_REG_MEM: 4669 case PACKET3_MEM_WRITE: 4670 case PACKET3_PFP_SYNC_ME: 4671 case PACKET3_SURFACE_SYNC: 4672 case PACKET3_EVENT_WRITE: 4673 case PACKET3_EVENT_WRITE_EOP: 4674 case PACKET3_EVENT_WRITE_EOS: 4675 case PACKET3_SET_CONTEXT_REG: 4676 case PACKET3_SET_CONTEXT_REG_INDIRECT: 4677 case PACKET3_SET_SH_REG: 4678 case PACKET3_SET_SH_REG_OFFSET: 4679 case PACKET3_INCREMENT_DE_COUNTER: 4680 case PACKET3_WAIT_ON_CE_COUNTER: 4681 case PACKET3_WAIT_ON_AVAIL_BUFFER: 4682 case PACKET3_ME_WRITE: 4683 break; 4684 case PACKET3_COPY_DATA: 4685 if ((idx_value & 0xf00) == 0) { 4686 reg = ib[idx + 3] * 4; 4687 if (!si_vm_reg_valid(reg)) 4688 return -EINVAL; 4689 } 4690 break; 4691 case PACKET3_WRITE_DATA: 4692 if ((idx_value & 0xf00) == 0) { 4693 start_reg = ib[idx + 1] * 4; 4694 if (idx_value & 0x10000) { 4695 if (!si_vm_reg_valid(start_reg)) 4696 return -EINVAL; 4697 } else { 4698 for (i = 0; i < (pkt->count - 2); i++) { 4699 reg = start_reg + (4 * i); 4700 if (!si_vm_reg_valid(reg)) 4701 return -EINVAL; 4702 } 4703 } 4704 } 4705 break; 4706 case PACKET3_COND_WRITE: 4707 if (idx_value & 0x100) { 4708 reg = ib[idx + 5] * 4; 4709 if (!si_vm_reg_valid(reg)) 4710 return -EINVAL; 4711 } 4712 break; 4713 case PACKET3_COPY_DW: 4714 if (idx_value & 0x2) { 4715 reg = ib[idx + 3] * 4; 4716 if (!si_vm_reg_valid(reg)) 4717 return -EINVAL; 4718 } 4719 break; 4720 case PACKET3_CP_DMA: 4721 r = si_vm_packet3_cp_dma_check(ib, idx); 4722 if (r) 4723 return r; 4724 break; 4725 default: 4726 DRM_ERROR("Invalid Compute packet3: 0x%x\n", pkt->opcode); 4727 return -EINVAL; 4728 } 4729 return 0; 4730 } 4731 4732 int si_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib) 4733 { 4734 int ret = 0; 4735 u32 idx = 0, i; 4736 struct radeon_cs_packet pkt; 4737 4738 do { 4739 pkt.idx = idx; 4740 pkt.type = RADEON_CP_PACKET_GET_TYPE(ib->ptr[idx]); 4741 pkt.count = RADEON_CP_PACKET_GET_COUNT(ib->ptr[idx]); 4742 pkt.one_reg_wr = 0; 4743 switch (pkt.type) { 4744 case RADEON_PACKET_TYPE0: 4745 dev_err(rdev->dev, "Packet0 not allowed!\n"); 4746 ret = -EINVAL; 4747 break; 4748 case RADEON_PACKET_TYPE2: 4749 idx += 1; 4750 break; 4751 case RADEON_PACKET_TYPE3: 4752 pkt.opcode = RADEON_CP_PACKET3_GET_OPCODE(ib->ptr[idx]); 4753 if (ib->is_const_ib) 4754 ret = si_vm_packet3_ce_check(rdev, ib->ptr, &pkt); 4755 else { 4756 switch (ib->ring) { 4757 case RADEON_RING_TYPE_GFX_INDEX: 4758 ret = si_vm_packet3_gfx_check(rdev, ib->ptr, &pkt); 4759 break; 4760 case CAYMAN_RING_TYPE_CP1_INDEX: 4761 case CAYMAN_RING_TYPE_CP2_INDEX: 4762 ret = si_vm_packet3_compute_check(rdev, ib->ptr, &pkt); 4763 break; 4764 default: 4765 dev_err(rdev->dev, "Non-PM4 ring %d !\n", ib->ring); 4766 ret = -EINVAL; 4767 break; 4768 } 4769 } 4770 idx += pkt.count + 2; 4771 break; 4772 default: 4773 dev_err(rdev->dev, "Unknown packet type %d !\n", pkt.type); 4774 ret = -EINVAL; 4775 break; 4776 } 4777 if (ret) { 4778 for (i = 0; i < ib->length_dw; i++) { 4779 if (i == idx) 4780 printk("\t0x%08x <---\n", ib->ptr[i]); 4781 else 4782 printk("\t0x%08x\n", ib->ptr[i]); 4783 } 4784 break; 4785 } 4786 } while (idx < ib->length_dw); 4787 4788 return ret; 4789 } 4790 4791 /* 4792 * vm 4793 */ 4794 int si_vm_init(struct radeon_device *rdev) 4795 { 4796 /* number of VMs */ 4797 rdev->vm_manager.nvm = 16; 4798 /* base offset of vram pages */ 4799 rdev->vm_manager.vram_base_offset = 0; 4800 4801 return 0; 4802 } 4803 4804 void si_vm_fini(struct radeon_device *rdev) 4805 { 4806 } 4807 4808 /** 4809 * si_vm_decode_fault - print human readable fault info 4810 * 4811 * @rdev: radeon_device pointer 4812 * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value 4813 * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value 4814 * 4815 * Print human readable fault information (SI). 4816 */ 4817 static void si_vm_decode_fault(struct radeon_device *rdev, 4818 u32 status, u32 addr) 4819 { 4820 u32 mc_id = (status & MEMORY_CLIENT_ID_MASK) >> MEMORY_CLIENT_ID_SHIFT; 4821 u32 vmid = (status & FAULT_VMID_MASK) >> FAULT_VMID_SHIFT; 4822 u32 protections = (status & PROTECTIONS_MASK) >> PROTECTIONS_SHIFT; 4823 char *block; 4824 4825 if (rdev->family == CHIP_TAHITI) { 4826 switch (mc_id) { 4827 case 160: 4828 case 144: 4829 case 96: 4830 case 80: 4831 case 224: 4832 case 208: 4833 case 32: 4834 case 16: 4835 block = "CB"; 4836 break; 4837 case 161: 4838 case 145: 4839 case 97: 4840 case 81: 4841 case 225: 4842 case 209: 4843 case 33: 4844 case 17: 4845 block = "CB_FMASK"; 4846 break; 4847 case 162: 4848 case 146: 4849 case 98: 4850 case 82: 4851 case 226: 4852 case 210: 4853 case 34: 4854 case 18: 4855 block = "CB_CMASK"; 4856 break; 4857 case 163: 4858 case 147: 4859 case 99: 4860 case 83: 4861 case 227: 4862 case 211: 4863 case 35: 4864 case 19: 4865 block = "CB_IMMED"; 4866 break; 4867 case 164: 4868 case 148: 4869 case 100: 4870 case 84: 4871 case 228: 4872 case 212: 4873 case 36: 4874 case 20: 4875 block = "DB"; 4876 break; 4877 case 165: 4878 case 149: 4879 case 101: 4880 case 85: 4881 case 229: 4882 case 213: 4883 case 37: 4884 case 21: 4885 block = "DB_HTILE"; 4886 break; 4887 case 167: 4888 case 151: 4889 case 103: 4890 case 87: 4891 case 231: 4892 case 215: 4893 case 39: 4894 case 23: 4895 block = "DB_STEN"; 4896 break; 4897 case 72: 4898 case 68: 4899 case 64: 4900 case 8: 4901 case 4: 4902 case 0: 4903 case 136: 4904 case 132: 4905 case 128: 4906 case 200: 4907 case 196: 4908 case 192: 4909 block = "TC"; 4910 break; 4911 case 112: 4912 case 48: 4913 block = "CP"; 4914 break; 4915 case 49: 4916 case 177: 4917 case 50: 4918 case 178: 4919 block = "SH"; 4920 break; 4921 case 53: 4922 case 190: 4923 block = "VGT"; 4924 break; 4925 case 117: 4926 block = "IH"; 4927 break; 4928 case 51: 4929 case 115: 4930 block = "RLC"; 4931 break; 4932 case 119: 4933 case 183: 4934 block = "DMA0"; 4935 break; 4936 case 61: 4937 block = "DMA1"; 4938 break; 4939 case 248: 4940 case 120: 4941 block = "HDP"; 4942 break; 4943 default: 4944 block = "unknown"; 4945 break; 4946 } 4947 } else { 4948 switch (mc_id) { 4949 case 32: 4950 case 16: 4951 case 96: 4952 case 80: 4953 case 160: 4954 case 144: 4955 case 224: 4956 case 208: 4957 block = "CB"; 4958 break; 4959 case 33: 4960 case 17: 4961 case 97: 4962 case 81: 4963 case 161: 4964 case 145: 4965 case 225: 4966 case 209: 4967 block = "CB_FMASK"; 4968 break; 4969 case 34: 4970 case 18: 4971 case 98: 4972 case 82: 4973 case 162: 4974 case 146: 4975 case 226: 4976 case 210: 4977 block = "CB_CMASK"; 4978 break; 4979 case 35: 4980 case 19: 4981 case 99: 4982 case 83: 4983 case 163: 4984 case 147: 4985 case 227: 4986 case 211: 4987 block = "CB_IMMED"; 4988 break; 4989 case 36: 4990 case 20: 4991 case 100: 4992 case 84: 4993 case 164: 4994 case 148: 4995 case 228: 4996 case 212: 4997 block = "DB"; 4998 break; 4999 case 37: 5000 case 21: 5001 case 101: 5002 case 85: 5003 case 165: 5004 case 149: 5005 case 229: 5006 case 213: 5007 block = "DB_HTILE"; 5008 break; 5009 case 39: 5010 case 23: 5011 case 103: 5012 case 87: 5013 case 167: 5014 case 151: 5015 case 231: 5016 case 215: 5017 block = "DB_STEN"; 5018 break; 5019 case 72: 5020 case 68: 5021 case 8: 5022 case 4: 5023 case 136: 5024 case 132: 5025 case 200: 5026 case 196: 5027 block = "TC"; 5028 break; 5029 case 112: 5030 case 48: 5031 block = "CP"; 5032 break; 5033 case 49: 5034 case 177: 5035 case 50: 5036 case 178: 5037 block = "SH"; 5038 break; 5039 case 53: 5040 block = "VGT"; 5041 break; 5042 case 117: 5043 block = "IH"; 5044 break; 5045 case 51: 5046 case 115: 5047 block = "RLC"; 5048 break; 5049 case 119: 5050 case 183: 5051 block = "DMA0"; 5052 break; 5053 case 61: 5054 block = "DMA1"; 5055 break; 5056 case 248: 5057 case 120: 5058 block = "HDP"; 5059 break; 5060 default: 5061 block = "unknown"; 5062 break; 5063 } 5064 } 5065 5066 printk("VM fault (0x%02x, vmid %d) at page %u, %s from %s (%d)\n", 5067 protections, vmid, addr, 5068 (status & MEMORY_CLIENT_RW_MASK) ? "write" : "read", 5069 block, mc_id); 5070 } 5071 5072 void si_vm_flush(struct radeon_device *rdev, struct radeon_ring *ring, 5073 unsigned vm_id, uint64_t pd_addr) 5074 { 5075 /* write new base address */ 5076 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5077 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) | 5078 WRITE_DATA_DST_SEL(0))); 5079 5080 if (vm_id < 8) { 5081 radeon_ring_write(ring, 5082 (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm_id << 2)) >> 2); 5083 } else { 5084 radeon_ring_write(ring, 5085 (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm_id - 8) << 2)) >> 2); 5086 } 5087 radeon_ring_write(ring, 0); 5088 radeon_ring_write(ring, pd_addr >> 12); 5089 5090 /* flush hdp cache */ 5091 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5092 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) | 5093 WRITE_DATA_DST_SEL(0))); 5094 radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2); 5095 radeon_ring_write(ring, 0); 5096 radeon_ring_write(ring, 0x1); 5097 5098 /* bits 0-15 are the VM contexts0-15 */ 5099 radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3)); 5100 radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) | 5101 WRITE_DATA_DST_SEL(0))); 5102 radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2); 5103 radeon_ring_write(ring, 0); 5104 radeon_ring_write(ring, 1 << vm_id); 5105 5106 /* wait for the invalidate to complete */ 5107 radeon_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5)); 5108 radeon_ring_write(ring, (WAIT_REG_MEM_FUNCTION(0) | /* always */ 5109 WAIT_REG_MEM_ENGINE(0))); /* me */ 5110 radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2); 5111 radeon_ring_write(ring, 0); 5112 radeon_ring_write(ring, 0); /* ref */ 5113 radeon_ring_write(ring, 0); /* mask */ 5114 radeon_ring_write(ring, 0x20); /* poll interval */ 5115 5116 /* sync PFP to ME, otherwise we might get invalid PFP reads */ 5117 radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0)); 5118 radeon_ring_write(ring, 0x0); 5119 } 5120 5121 /* 5122 * Power and clock gating 5123 */ 5124 static void si_wait_for_rlc_serdes(struct radeon_device *rdev) 5125 { 5126 int i; 5127 5128 for (i = 0; i < rdev->usec_timeout; i++) { 5129 if (RREG32(RLC_SERDES_MASTER_BUSY_0) == 0) 5130 break; 5131 udelay(1); 5132 } 5133 5134 for (i = 0; i < rdev->usec_timeout; i++) { 5135 if (RREG32(RLC_SERDES_MASTER_BUSY_1) == 0) 5136 break; 5137 udelay(1); 5138 } 5139 } 5140 5141 static void si_enable_gui_idle_interrupt(struct radeon_device *rdev, 5142 bool enable) 5143 { 5144 u32 tmp = RREG32(CP_INT_CNTL_RING0); 5145 u32 mask; 5146 int i; 5147 5148 if (enable) 5149 tmp |= (CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE); 5150 else 5151 tmp &= ~(CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE); 5152 WREG32(CP_INT_CNTL_RING0, tmp); 5153 5154 if (!enable) { 5155 /* read a gfx register */ 5156 tmp = RREG32(DB_DEPTH_INFO); 5157 5158 mask = RLC_BUSY_STATUS | GFX_POWER_STATUS | GFX_CLOCK_STATUS | GFX_LS_STATUS; 5159 for (i = 0; i < rdev->usec_timeout; i++) { 5160 if ((RREG32(RLC_STAT) & mask) == (GFX_CLOCK_STATUS | GFX_POWER_STATUS)) 5161 break; 5162 udelay(1); 5163 } 5164 } 5165 } 5166 5167 static void si_set_uvd_dcm(struct radeon_device *rdev, 5168 bool sw_mode) 5169 { 5170 u32 tmp, tmp2; 5171 5172 tmp = RREG32(UVD_CGC_CTRL); 5173 tmp &= ~(CLK_OD_MASK | CG_DT_MASK); 5174 tmp |= DCM | CG_DT(1) | CLK_OD(4); 5175 5176 if (sw_mode) { 5177 tmp &= ~0x7ffff800; 5178 tmp2 = DYN_OR_EN | DYN_RR_EN | G_DIV_ID(7); 5179 } else { 5180 tmp |= 0x7ffff800; 5181 tmp2 = 0; 5182 } 5183 5184 WREG32(UVD_CGC_CTRL, tmp); 5185 WREG32_UVD_CTX(UVD_CGC_CTRL2, tmp2); 5186 } 5187 5188 void si_init_uvd_internal_cg(struct radeon_device *rdev) 5189 { 5190 bool hw_mode = true; 5191 5192 if (hw_mode) { 5193 si_set_uvd_dcm(rdev, false); 5194 } else { 5195 u32 tmp = RREG32(UVD_CGC_CTRL); 5196 tmp &= ~DCM; 5197 WREG32(UVD_CGC_CTRL, tmp); 5198 } 5199 } 5200 5201 static u32 si_halt_rlc(struct radeon_device *rdev) 5202 { 5203 u32 data, orig; 5204 5205 orig = data = RREG32(RLC_CNTL); 5206 5207 if (data & RLC_ENABLE) { 5208 data &= ~RLC_ENABLE; 5209 WREG32(RLC_CNTL, data); 5210 5211 si_wait_for_rlc_serdes(rdev); 5212 } 5213 5214 return orig; 5215 } 5216 5217 static void si_update_rlc(struct radeon_device *rdev, u32 rlc) 5218 { 5219 u32 tmp; 5220 5221 tmp = RREG32(RLC_CNTL); 5222 if (tmp != rlc) 5223 WREG32(RLC_CNTL, rlc); 5224 } 5225 5226 static void si_enable_dma_pg(struct radeon_device *rdev, bool enable) 5227 { 5228 u32 data, orig; 5229 5230 orig = data = RREG32(DMA_PG); 5231 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_SDMA)) 5232 data |= PG_CNTL_ENABLE; 5233 else 5234 data &= ~PG_CNTL_ENABLE; 5235 if (orig != data) 5236 WREG32(DMA_PG, data); 5237 } 5238 5239 static void si_init_dma_pg(struct radeon_device *rdev) 5240 { 5241 u32 tmp; 5242 5243 WREG32(DMA_PGFSM_WRITE, 0x00002000); 5244 WREG32(DMA_PGFSM_CONFIG, 0x100010ff); 5245 5246 for (tmp = 0; tmp < 5; tmp++) 5247 WREG32(DMA_PGFSM_WRITE, 0); 5248 } 5249 5250 static void si_enable_gfx_cgpg(struct radeon_device *rdev, 5251 bool enable) 5252 { 5253 u32 tmp; 5254 5255 if (enable && (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_PG)) { 5256 tmp = RLC_PUD(0x10) | RLC_PDD(0x10) | RLC_TTPD(0x10) | RLC_MSD(0x10); 5257 WREG32(RLC_TTOP_D, tmp); 5258 5259 tmp = RREG32(RLC_PG_CNTL); 5260 tmp |= GFX_PG_ENABLE; 5261 WREG32(RLC_PG_CNTL, tmp); 5262 5263 tmp = RREG32(RLC_AUTO_PG_CTRL); 5264 tmp |= AUTO_PG_EN; 5265 WREG32(RLC_AUTO_PG_CTRL, tmp); 5266 } else { 5267 tmp = RREG32(RLC_AUTO_PG_CTRL); 5268 tmp &= ~AUTO_PG_EN; 5269 WREG32(RLC_AUTO_PG_CTRL, tmp); 5270 5271 tmp = RREG32(DB_RENDER_CONTROL); 5272 } 5273 } 5274 5275 static void si_init_gfx_cgpg(struct radeon_device *rdev) 5276 { 5277 u32 tmp; 5278 5279 WREG32(RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8); 5280 5281 tmp = RREG32(RLC_PG_CNTL); 5282 tmp |= GFX_PG_SRC; 5283 WREG32(RLC_PG_CNTL, tmp); 5284 5285 WREG32(RLC_CLEAR_STATE_RESTORE_BASE, rdev->rlc.clear_state_gpu_addr >> 8); 5286 5287 tmp = RREG32(RLC_AUTO_PG_CTRL); 5288 5289 tmp &= ~GRBM_REG_SGIT_MASK; 5290 tmp |= GRBM_REG_SGIT(0x700); 5291 tmp &= ~PG_AFTER_GRBM_REG_ST_MASK; 5292 WREG32(RLC_AUTO_PG_CTRL, tmp); 5293 } 5294 5295 static u32 si_get_cu_active_bitmap(struct radeon_device *rdev, u32 se, u32 sh) 5296 { 5297 u32 mask = 0, tmp, tmp1; 5298 int i; 5299 5300 si_select_se_sh(rdev, se, sh); 5301 tmp = RREG32(CC_GC_SHADER_ARRAY_CONFIG); 5302 tmp1 = RREG32(GC_USER_SHADER_ARRAY_CONFIG); 5303 si_select_se_sh(rdev, 0xffffffff, 0xffffffff); 5304 5305 tmp &= 0xffff0000; 5306 5307 tmp |= tmp1; 5308 tmp >>= 16; 5309 5310 for (i = 0; i < rdev->config.si.max_cu_per_sh; i ++) { 5311 mask <<= 1; 5312 mask |= 1; 5313 } 5314 5315 return (~tmp) & mask; 5316 } 5317 5318 static void si_init_ao_cu_mask(struct radeon_device *rdev) 5319 { 5320 u32 i, j, k, active_cu_number = 0; 5321 u32 mask, counter, cu_bitmap; 5322 u32 tmp = 0; 5323 5324 for (i = 0; i < rdev->config.si.max_shader_engines; i++) { 5325 for (j = 0; j < rdev->config.si.max_sh_per_se; j++) { 5326 mask = 1; 5327 cu_bitmap = 0; 5328 counter = 0; 5329 for (k = 0; k < rdev->config.si.max_cu_per_sh; k++) { 5330 if (si_get_cu_active_bitmap(rdev, i, j) & mask) { 5331 if (counter < 2) 5332 cu_bitmap |= mask; 5333 counter++; 5334 } 5335 mask <<= 1; 5336 } 5337 5338 active_cu_number += counter; 5339 tmp |= (cu_bitmap << (i * 16 + j * 8)); 5340 } 5341 } 5342 5343 WREG32(RLC_PG_AO_CU_MASK, tmp); 5344 5345 tmp = RREG32(RLC_MAX_PG_CU); 5346 tmp &= ~MAX_PU_CU_MASK; 5347 tmp |= MAX_PU_CU(active_cu_number); 5348 WREG32(RLC_MAX_PG_CU, tmp); 5349 } 5350 5351 static void si_enable_cgcg(struct radeon_device *rdev, 5352 bool enable) 5353 { 5354 u32 data, orig, tmp; 5355 5356 orig = data = RREG32(RLC_CGCG_CGLS_CTRL); 5357 5358 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_CGCG)) { 5359 si_enable_gui_idle_interrupt(rdev, true); 5360 5361 WREG32(RLC_GCPM_GENERAL_3, 0x00000080); 5362 5363 tmp = si_halt_rlc(rdev); 5364 5365 WREG32(RLC_SERDES_WR_MASTER_MASK_0, 0xffffffff); 5366 WREG32(RLC_SERDES_WR_MASTER_MASK_1, 0xffffffff); 5367 WREG32(RLC_SERDES_WR_CTRL, 0x00b000ff); 5368 5369 si_wait_for_rlc_serdes(rdev); 5370 5371 si_update_rlc(rdev, tmp); 5372 5373 WREG32(RLC_SERDES_WR_CTRL, 0x007000ff); 5374 5375 data |= CGCG_EN | CGLS_EN; 5376 } else { 5377 si_enable_gui_idle_interrupt(rdev, false); 5378 5379 RREG32(CB_CGTT_SCLK_CTRL); 5380 RREG32(CB_CGTT_SCLK_CTRL); 5381 RREG32(CB_CGTT_SCLK_CTRL); 5382 RREG32(CB_CGTT_SCLK_CTRL); 5383 5384 data &= ~(CGCG_EN | CGLS_EN); 5385 } 5386 5387 if (orig != data) 5388 WREG32(RLC_CGCG_CGLS_CTRL, data); 5389 } 5390 5391 static void si_enable_mgcg(struct radeon_device *rdev, 5392 bool enable) 5393 { 5394 u32 data, orig, tmp = 0; 5395 5396 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_MGCG)) { 5397 orig = data = RREG32(CGTS_SM_CTRL_REG); 5398 data = 0x96940200; 5399 if (orig != data) 5400 WREG32(CGTS_SM_CTRL_REG, data); 5401 5402 if (rdev->cg_flags & RADEON_CG_SUPPORT_GFX_CP_LS) { 5403 orig = data = RREG32(CP_MEM_SLP_CNTL); 5404 data |= CP_MEM_LS_EN; 5405 if (orig != data) 5406 WREG32(CP_MEM_SLP_CNTL, data); 5407 } 5408 5409 orig = data = RREG32(RLC_CGTT_MGCG_OVERRIDE); 5410 data &= 0xffffffc0; 5411 if (orig != data) 5412 WREG32(RLC_CGTT_MGCG_OVERRIDE, data); 5413 5414 tmp = si_halt_rlc(rdev); 5415 5416 WREG32(RLC_SERDES_WR_MASTER_MASK_0, 0xffffffff); 5417 WREG32(RLC_SERDES_WR_MASTER_MASK_1, 0xffffffff); 5418 WREG32(RLC_SERDES_WR_CTRL, 0x00d000ff); 5419 5420 si_update_rlc(rdev, tmp); 5421 } else { 5422 orig = data = RREG32(RLC_CGTT_MGCG_OVERRIDE); 5423 data |= 0x00000003; 5424 if (orig != data) 5425 WREG32(RLC_CGTT_MGCG_OVERRIDE, data); 5426 5427 data = RREG32(CP_MEM_SLP_CNTL); 5428 if (data & CP_MEM_LS_EN) { 5429 data &= ~CP_MEM_LS_EN; 5430 WREG32(CP_MEM_SLP_CNTL, data); 5431 } 5432 orig = data = RREG32(CGTS_SM_CTRL_REG); 5433 data |= LS_OVERRIDE | OVERRIDE; 5434 if (orig != data) 5435 WREG32(CGTS_SM_CTRL_REG, data); 5436 5437 tmp = si_halt_rlc(rdev); 5438 5439 WREG32(RLC_SERDES_WR_MASTER_MASK_0, 0xffffffff); 5440 WREG32(RLC_SERDES_WR_MASTER_MASK_1, 0xffffffff); 5441 WREG32(RLC_SERDES_WR_CTRL, 0x00e000ff); 5442 5443 si_update_rlc(rdev, tmp); 5444 } 5445 } 5446 5447 static void si_enable_uvd_mgcg(struct radeon_device *rdev, 5448 bool enable) 5449 { 5450 u32 orig, data, tmp; 5451 5452 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_UVD_MGCG)) { 5453 tmp = RREG32_UVD_CTX(UVD_CGC_MEM_CTRL); 5454 tmp |= 0x3fff; 5455 WREG32_UVD_CTX(UVD_CGC_MEM_CTRL, tmp); 5456 5457 orig = data = RREG32(UVD_CGC_CTRL); 5458 data |= DCM; 5459 if (orig != data) 5460 WREG32(UVD_CGC_CTRL, data); 5461 5462 WREG32_SMC(SMC_CG_IND_START + CG_CGTT_LOCAL_0, 0); 5463 WREG32_SMC(SMC_CG_IND_START + CG_CGTT_LOCAL_1, 0); 5464 } else { 5465 tmp = RREG32_UVD_CTX(UVD_CGC_MEM_CTRL); 5466 tmp &= ~0x3fff; 5467 WREG32_UVD_CTX(UVD_CGC_MEM_CTRL, tmp); 5468 5469 orig = data = RREG32(UVD_CGC_CTRL); 5470 data &= ~DCM; 5471 if (orig != data) 5472 WREG32(UVD_CGC_CTRL, data); 5473 5474 WREG32_SMC(SMC_CG_IND_START + CG_CGTT_LOCAL_0, 0xffffffff); 5475 WREG32_SMC(SMC_CG_IND_START + CG_CGTT_LOCAL_1, 0xffffffff); 5476 } 5477 } 5478 5479 static const u32 mc_cg_registers[] = 5480 { 5481 MC_HUB_MISC_HUB_CG, 5482 MC_HUB_MISC_SIP_CG, 5483 MC_HUB_MISC_VM_CG, 5484 MC_XPB_CLK_GAT, 5485 ATC_MISC_CG, 5486 MC_CITF_MISC_WR_CG, 5487 MC_CITF_MISC_RD_CG, 5488 MC_CITF_MISC_VM_CG, 5489 VM_L2_CG, 5490 }; 5491 5492 static void si_enable_mc_ls(struct radeon_device *rdev, 5493 bool enable) 5494 { 5495 int i; 5496 u32 orig, data; 5497 5498 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) { 5499 orig = data = RREG32(mc_cg_registers[i]); 5500 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_MC_LS)) 5501 data |= MC_LS_ENABLE; 5502 else 5503 data &= ~MC_LS_ENABLE; 5504 if (data != orig) 5505 WREG32(mc_cg_registers[i], data); 5506 } 5507 } 5508 5509 static void si_enable_mc_mgcg(struct radeon_device *rdev, 5510 bool enable) 5511 { 5512 int i; 5513 u32 orig, data; 5514 5515 for (i = 0; i < ARRAY_SIZE(mc_cg_registers); i++) { 5516 orig = data = RREG32(mc_cg_registers[i]); 5517 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_MC_MGCG)) 5518 data |= MC_CG_ENABLE; 5519 else 5520 data &= ~MC_CG_ENABLE; 5521 if (data != orig) 5522 WREG32(mc_cg_registers[i], data); 5523 } 5524 } 5525 5526 static void si_enable_dma_mgcg(struct radeon_device *rdev, 5527 bool enable) 5528 { 5529 u32 orig, data, offset; 5530 int i; 5531 5532 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_SDMA_MGCG)) { 5533 for (i = 0; i < 2; i++) { 5534 if (i == 0) 5535 offset = DMA0_REGISTER_OFFSET; 5536 else 5537 offset = DMA1_REGISTER_OFFSET; 5538 orig = data = RREG32(DMA_POWER_CNTL + offset); 5539 data &= ~MEM_POWER_OVERRIDE; 5540 if (data != orig) 5541 WREG32(DMA_POWER_CNTL + offset, data); 5542 WREG32(DMA_CLK_CTRL + offset, 0x00000100); 5543 } 5544 } else { 5545 for (i = 0; i < 2; i++) { 5546 if (i == 0) 5547 offset = DMA0_REGISTER_OFFSET; 5548 else 5549 offset = DMA1_REGISTER_OFFSET; 5550 orig = data = RREG32(DMA_POWER_CNTL + offset); 5551 data |= MEM_POWER_OVERRIDE; 5552 if (data != orig) 5553 WREG32(DMA_POWER_CNTL + offset, data); 5554 5555 orig = data = RREG32(DMA_CLK_CTRL + offset); 5556 data = 0xff000000; 5557 if (data != orig) 5558 WREG32(DMA_CLK_CTRL + offset, data); 5559 } 5560 } 5561 } 5562 5563 static void si_enable_bif_mgls(struct radeon_device *rdev, 5564 bool enable) 5565 { 5566 u32 orig, data; 5567 5568 orig = data = RREG32_PCIE(PCIE_CNTL2); 5569 5570 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_BIF_LS)) 5571 data |= SLV_MEM_LS_EN | MST_MEM_LS_EN | 5572 REPLAY_MEM_LS_EN | SLV_MEM_AGGRESSIVE_LS_EN; 5573 else 5574 data &= ~(SLV_MEM_LS_EN | MST_MEM_LS_EN | 5575 REPLAY_MEM_LS_EN | SLV_MEM_AGGRESSIVE_LS_EN); 5576 5577 if (orig != data) 5578 WREG32_PCIE(PCIE_CNTL2, data); 5579 } 5580 5581 static void si_enable_hdp_mgcg(struct radeon_device *rdev, 5582 bool enable) 5583 { 5584 u32 orig, data; 5585 5586 orig = data = RREG32(HDP_HOST_PATH_CNTL); 5587 5588 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_HDP_MGCG)) 5589 data &= ~CLOCK_GATING_DIS; 5590 else 5591 data |= CLOCK_GATING_DIS; 5592 5593 if (orig != data) 5594 WREG32(HDP_HOST_PATH_CNTL, data); 5595 } 5596 5597 static void si_enable_hdp_ls(struct radeon_device *rdev, 5598 bool enable) 5599 { 5600 u32 orig, data; 5601 5602 orig = data = RREG32(HDP_MEM_POWER_LS); 5603 5604 if (enable && (rdev->cg_flags & RADEON_CG_SUPPORT_HDP_LS)) 5605 data |= HDP_LS_ENABLE; 5606 else 5607 data &= ~HDP_LS_ENABLE; 5608 5609 if (orig != data) 5610 WREG32(HDP_MEM_POWER_LS, data); 5611 } 5612 5613 static void si_update_cg(struct radeon_device *rdev, 5614 u32 block, bool enable) 5615 { 5616 if (block & RADEON_CG_BLOCK_GFX) { 5617 si_enable_gui_idle_interrupt(rdev, false); 5618 /* order matters! */ 5619 if (enable) { 5620 si_enable_mgcg(rdev, true); 5621 si_enable_cgcg(rdev, true); 5622 } else { 5623 si_enable_cgcg(rdev, false); 5624 si_enable_mgcg(rdev, false); 5625 } 5626 si_enable_gui_idle_interrupt(rdev, true); 5627 } 5628 5629 if (block & RADEON_CG_BLOCK_MC) { 5630 si_enable_mc_mgcg(rdev, enable); 5631 si_enable_mc_ls(rdev, enable); 5632 } 5633 5634 if (block & RADEON_CG_BLOCK_SDMA) { 5635 si_enable_dma_mgcg(rdev, enable); 5636 } 5637 5638 if (block & RADEON_CG_BLOCK_BIF) { 5639 si_enable_bif_mgls(rdev, enable); 5640 } 5641 5642 if (block & RADEON_CG_BLOCK_UVD) { 5643 if (rdev->has_uvd) { 5644 si_enable_uvd_mgcg(rdev, enable); 5645 } 5646 } 5647 5648 if (block & RADEON_CG_BLOCK_HDP) { 5649 si_enable_hdp_mgcg(rdev, enable); 5650 si_enable_hdp_ls(rdev, enable); 5651 } 5652 } 5653 5654 static void si_init_cg(struct radeon_device *rdev) 5655 { 5656 si_update_cg(rdev, (RADEON_CG_BLOCK_GFX | 5657 RADEON_CG_BLOCK_MC | 5658 RADEON_CG_BLOCK_SDMA | 5659 RADEON_CG_BLOCK_BIF | 5660 RADEON_CG_BLOCK_HDP), true); 5661 if (rdev->has_uvd) { 5662 si_update_cg(rdev, RADEON_CG_BLOCK_UVD, true); 5663 si_init_uvd_internal_cg(rdev); 5664 } 5665 } 5666 5667 static void si_fini_cg(struct radeon_device *rdev) 5668 { 5669 if (rdev->has_uvd) { 5670 si_update_cg(rdev, RADEON_CG_BLOCK_UVD, false); 5671 } 5672 si_update_cg(rdev, (RADEON_CG_BLOCK_GFX | 5673 RADEON_CG_BLOCK_MC | 5674 RADEON_CG_BLOCK_SDMA | 5675 RADEON_CG_BLOCK_BIF | 5676 RADEON_CG_BLOCK_HDP), false); 5677 } 5678 5679 u32 si_get_csb_size(struct radeon_device *rdev) 5680 { 5681 u32 count = 0; 5682 const struct cs_section_def *sect = NULL; 5683 const struct cs_extent_def *ext = NULL; 5684 5685 if (rdev->rlc.cs_data == NULL) 5686 return 0; 5687 5688 /* begin clear state */ 5689 count += 2; 5690 /* context control state */ 5691 count += 3; 5692 5693 for (sect = rdev->rlc.cs_data; sect->section != NULL; ++sect) { 5694 for (ext = sect->section; ext->extent != NULL; ++ext) { 5695 if (sect->id == SECT_CONTEXT) 5696 count += 2 + ext->reg_count; 5697 else 5698 return 0; 5699 } 5700 } 5701 /* pa_sc_raster_config */ 5702 count += 3; 5703 /* end clear state */ 5704 count += 2; 5705 /* clear state */ 5706 count += 2; 5707 5708 return count; 5709 } 5710 5711 void si_get_csb_buffer(struct radeon_device *rdev, volatile u32 *buffer) 5712 { 5713 u32 count = 0, i; 5714 const struct cs_section_def *sect = NULL; 5715 const struct cs_extent_def *ext = NULL; 5716 5717 if (rdev->rlc.cs_data == NULL) 5718 return; 5719 if (buffer == NULL) 5720 return; 5721 5722 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 5723 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 5724 5725 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1)); 5726 buffer[count++] = cpu_to_le32(0x80000000); 5727 buffer[count++] = cpu_to_le32(0x80000000); 5728 5729 for (sect = rdev->rlc.cs_data; sect->section != NULL; ++sect) { 5730 for (ext = sect->section; ext->extent != NULL; ++ext) { 5731 if (sect->id == SECT_CONTEXT) { 5732 buffer[count++] = 5733 cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count)); 5734 buffer[count++] = cpu_to_le32(ext->reg_index - 0xa000); 5735 for (i = 0; i < ext->reg_count; i++) 5736 buffer[count++] = cpu_to_le32(ext->extent[i]); 5737 } else { 5738 return; 5739 } 5740 } 5741 } 5742 5743 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 1)); 5744 buffer[count++] = cpu_to_le32(PA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START); 5745 switch (rdev->family) { 5746 case CHIP_TAHITI: 5747 case CHIP_PITCAIRN: 5748 buffer[count++] = cpu_to_le32(0x2a00126a); 5749 break; 5750 case CHIP_VERDE: 5751 buffer[count++] = cpu_to_le32(0x0000124a); 5752 break; 5753 case CHIP_OLAND: 5754 buffer[count++] = cpu_to_le32(0x00000082); 5755 break; 5756 case CHIP_HAINAN: 5757 buffer[count++] = cpu_to_le32(0x00000000); 5758 break; 5759 default: 5760 buffer[count++] = cpu_to_le32(0x00000000); 5761 break; 5762 } 5763 5764 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 5765 buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE); 5766 5767 buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0)); 5768 buffer[count++] = cpu_to_le32(0); 5769 } 5770 5771 static void si_init_pg(struct radeon_device *rdev) 5772 { 5773 if (rdev->pg_flags) { 5774 if (rdev->pg_flags & RADEON_PG_SUPPORT_SDMA) { 5775 si_init_dma_pg(rdev); 5776 } 5777 si_init_ao_cu_mask(rdev); 5778 if (rdev->pg_flags & RADEON_PG_SUPPORT_GFX_PG) { 5779 si_init_gfx_cgpg(rdev); 5780 } else { 5781 WREG32(RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8); 5782 WREG32(RLC_CLEAR_STATE_RESTORE_BASE, rdev->rlc.clear_state_gpu_addr >> 8); 5783 } 5784 si_enable_dma_pg(rdev, true); 5785 si_enable_gfx_cgpg(rdev, true); 5786 } else { 5787 WREG32(RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8); 5788 WREG32(RLC_CLEAR_STATE_RESTORE_BASE, rdev->rlc.clear_state_gpu_addr >> 8); 5789 } 5790 } 5791 5792 static void si_fini_pg(struct radeon_device *rdev) 5793 { 5794 if (rdev->pg_flags) { 5795 si_enable_dma_pg(rdev, false); 5796 si_enable_gfx_cgpg(rdev, false); 5797 } 5798 } 5799 5800 /* 5801 * RLC 5802 */ 5803 void si_rlc_reset(struct radeon_device *rdev) 5804 { 5805 u32 tmp = RREG32(GRBM_SOFT_RESET); 5806 5807 tmp |= SOFT_RESET_RLC; 5808 WREG32(GRBM_SOFT_RESET, tmp); 5809 udelay(50); 5810 tmp &= ~SOFT_RESET_RLC; 5811 WREG32(GRBM_SOFT_RESET, tmp); 5812 udelay(50); 5813 } 5814 5815 static void si_rlc_stop(struct radeon_device *rdev) 5816 { 5817 WREG32(RLC_CNTL, 0); 5818 5819 si_enable_gui_idle_interrupt(rdev, false); 5820 5821 si_wait_for_rlc_serdes(rdev); 5822 } 5823 5824 static void si_rlc_start(struct radeon_device *rdev) 5825 { 5826 WREG32(RLC_CNTL, RLC_ENABLE); 5827 5828 si_enable_gui_idle_interrupt(rdev, true); 5829 5830 udelay(50); 5831 } 5832 5833 static bool si_lbpw_supported(struct radeon_device *rdev) 5834 { 5835 u32 tmp; 5836 5837 /* Enable LBPW only for DDR3 */ 5838 tmp = RREG32(MC_SEQ_MISC0); 5839 if ((tmp & 0xF0000000) == 0xB0000000) 5840 return true; 5841 return false; 5842 } 5843 5844 static void si_enable_lbpw(struct radeon_device *rdev, bool enable) 5845 { 5846 u32 tmp; 5847 5848 tmp = RREG32(RLC_LB_CNTL); 5849 if (enable) 5850 tmp |= LOAD_BALANCE_ENABLE; 5851 else 5852 tmp &= ~LOAD_BALANCE_ENABLE; 5853 WREG32(RLC_LB_CNTL, tmp); 5854 5855 if (!enable) { 5856 si_select_se_sh(rdev, 0xffffffff, 0xffffffff); 5857 WREG32(SPI_LB_CU_MASK, 0x00ff); 5858 } 5859 } 5860 5861 static int si_rlc_resume(struct radeon_device *rdev) 5862 { 5863 u32 i; 5864 5865 if (!rdev->rlc_fw) 5866 return -EINVAL; 5867 5868 si_rlc_stop(rdev); 5869 5870 si_rlc_reset(rdev); 5871 5872 si_init_pg(rdev); 5873 5874 si_init_cg(rdev); 5875 5876 WREG32(RLC_RL_BASE, 0); 5877 WREG32(RLC_RL_SIZE, 0); 5878 WREG32(RLC_LB_CNTL, 0); 5879 WREG32(RLC_LB_CNTR_MAX, 0xffffffff); 5880 WREG32(RLC_LB_CNTR_INIT, 0); 5881 WREG32(RLC_LB_INIT_CU_MASK, 0xffffffff); 5882 5883 WREG32(RLC_MC_CNTL, 0); 5884 WREG32(RLC_UCODE_CNTL, 0); 5885 5886 if (rdev->new_fw) { 5887 const struct rlc_firmware_header_v1_0 *hdr = 5888 (const struct rlc_firmware_header_v1_0 *)rdev->rlc_fw->data; 5889 u32 fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4; 5890 const __le32 *fw_data = (const __le32 *) 5891 (rdev->rlc_fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); 5892 5893 radeon_ucode_print_rlc_hdr(&hdr->header); 5894 5895 for (i = 0; i < fw_size; i++) { 5896 WREG32(RLC_UCODE_ADDR, i); 5897 WREG32(RLC_UCODE_DATA, le32_to_cpup(fw_data++)); 5898 } 5899 } else { 5900 const __be32 *fw_data = 5901 (const __be32 *)rdev->rlc_fw->data; 5902 for (i = 0; i < SI_RLC_UCODE_SIZE; i++) { 5903 WREG32(RLC_UCODE_ADDR, i); 5904 WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++)); 5905 } 5906 } 5907 WREG32(RLC_UCODE_ADDR, 0); 5908 5909 si_enable_lbpw(rdev, si_lbpw_supported(rdev)); 5910 5911 si_rlc_start(rdev); 5912 5913 return 0; 5914 } 5915 5916 static void si_enable_interrupts(struct radeon_device *rdev) 5917 { 5918 u32 ih_cntl = RREG32(IH_CNTL); 5919 u32 ih_rb_cntl = RREG32(IH_RB_CNTL); 5920 5921 ih_cntl |= ENABLE_INTR; 5922 ih_rb_cntl |= IH_RB_ENABLE; 5923 WREG32(IH_CNTL, ih_cntl); 5924 WREG32(IH_RB_CNTL, ih_rb_cntl); 5925 rdev->ih.enabled = true; 5926 } 5927 5928 static void si_disable_interrupts(struct radeon_device *rdev) 5929 { 5930 u32 ih_rb_cntl = RREG32(IH_RB_CNTL); 5931 u32 ih_cntl = RREG32(IH_CNTL); 5932 5933 ih_rb_cntl &= ~IH_RB_ENABLE; 5934 ih_cntl &= ~ENABLE_INTR; 5935 WREG32(IH_RB_CNTL, ih_rb_cntl); 5936 WREG32(IH_CNTL, ih_cntl); 5937 /* set rptr, wptr to 0 */ 5938 WREG32(IH_RB_RPTR, 0); 5939 WREG32(IH_RB_WPTR, 0); 5940 rdev->ih.enabled = false; 5941 rdev->ih.rptr = 0; 5942 } 5943 5944 static void si_disable_interrupt_state(struct radeon_device *rdev) 5945 { 5946 int i; 5947 u32 tmp; 5948 5949 tmp = RREG32(CP_INT_CNTL_RING0) & 5950 (CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE); 5951 WREG32(CP_INT_CNTL_RING0, tmp); 5952 WREG32(CP_INT_CNTL_RING1, 0); 5953 WREG32(CP_INT_CNTL_RING2, 0); 5954 tmp = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET) & ~TRAP_ENABLE; 5955 WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, tmp); 5956 tmp = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET) & ~TRAP_ENABLE; 5957 WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, tmp); 5958 WREG32(GRBM_INT_CNTL, 0); 5959 WREG32(SRBM_INT_CNTL, 0); 5960 for (i = 0; i < rdev->num_crtc; i++) 5961 WREG32(INT_MASK + crtc_offsets[i], 0); 5962 for (i = 0; i < rdev->num_crtc; i++) 5963 WREG32(GRPH_INT_CONTROL + crtc_offsets[i], 0); 5964 5965 if (!ASIC_IS_NODCE(rdev)) { 5966 WREG32(DAC_AUTODETECT_INT_CONTROL, 0); 5967 5968 for (i = 0; i < 6; i++) 5969 WREG32_AND(DC_HPDx_INT_CONTROL(i), 5970 DC_HPDx_INT_POLARITY); 5971 } 5972 } 5973 5974 static int si_irq_init(struct radeon_device *rdev) 5975 { 5976 int ret = 0; 5977 int rb_bufsz; 5978 u32 interrupt_cntl, ih_cntl, ih_rb_cntl; 5979 5980 /* allocate ring */ 5981 ret = r600_ih_ring_alloc(rdev); 5982 if (ret) 5983 return ret; 5984 5985 /* disable irqs */ 5986 si_disable_interrupts(rdev); 5987 5988 /* init rlc */ 5989 ret = si_rlc_resume(rdev); 5990 if (ret) { 5991 r600_ih_ring_fini(rdev); 5992 return ret; 5993 } 5994 5995 /* setup interrupt control */ 5996 /* set dummy read address to ring address */ 5997 WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8); 5998 interrupt_cntl = RREG32(INTERRUPT_CNTL); 5999 /* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi 6000 * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN 6001 */ 6002 interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE; 6003 /* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */ 6004 interrupt_cntl &= ~IH_REQ_NONSNOOP_EN; 6005 WREG32(INTERRUPT_CNTL, interrupt_cntl); 6006 6007 WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8); 6008 rb_bufsz = order_base_2(rdev->ih.ring_size / 4); 6009 6010 ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE | 6011 IH_WPTR_OVERFLOW_CLEAR | 6012 (rb_bufsz << 1)); 6013 6014 if (rdev->wb.enabled) 6015 ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE; 6016 6017 /* set the writeback address whether it's enabled or not */ 6018 WREG32(IH_RB_WPTR_ADDR_LO, (rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFFFFFFFC); 6019 WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFF); 6020 6021 WREG32(IH_RB_CNTL, ih_rb_cntl); 6022 6023 /* set rptr, wptr to 0 */ 6024 WREG32(IH_RB_RPTR, 0); 6025 WREG32(IH_RB_WPTR, 0); 6026 6027 /* Default settings for IH_CNTL (disabled at first) */ 6028 ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10) | MC_VMID(0); 6029 /* RPTR_REARM only works if msi's are enabled */ 6030 if (rdev->msi_enabled) 6031 ih_cntl |= RPTR_REARM; 6032 WREG32(IH_CNTL, ih_cntl); 6033 6034 /* force the active interrupt state to all disabled */ 6035 si_disable_interrupt_state(rdev); 6036 6037 pci_set_master(rdev->pdev); 6038 6039 /* enable irqs */ 6040 si_enable_interrupts(rdev); 6041 6042 return ret; 6043 } 6044 6045 /* The order we write back each register here is important */ 6046 int si_irq_set(struct radeon_device *rdev) 6047 { 6048 int i; 6049 u32 cp_int_cntl; 6050 u32 cp_int_cntl1 = 0, cp_int_cntl2 = 0; 6051 u32 grbm_int_cntl = 0; 6052 u32 dma_cntl, dma_cntl1; 6053 u32 thermal_int = 0; 6054 6055 if (!rdev->irq.installed) { 6056 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n"); 6057 return -EINVAL; 6058 } 6059 /* don't enable anything if the ih is disabled */ 6060 if (!rdev->ih.enabled) { 6061 si_disable_interrupts(rdev); 6062 /* force the active interrupt state to all disabled */ 6063 si_disable_interrupt_state(rdev); 6064 return 0; 6065 } 6066 6067 cp_int_cntl = RREG32(CP_INT_CNTL_RING0) & 6068 (CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE); 6069 6070 dma_cntl = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET) & ~TRAP_ENABLE; 6071 dma_cntl1 = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET) & ~TRAP_ENABLE; 6072 6073 thermal_int = RREG32(CG_THERMAL_INT) & 6074 ~(THERM_INT_MASK_HIGH | THERM_INT_MASK_LOW); 6075 6076 /* enable CP interrupts on all rings */ 6077 if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) { 6078 DRM_DEBUG("si_irq_set: sw int gfx\n"); 6079 cp_int_cntl |= TIME_STAMP_INT_ENABLE; 6080 } 6081 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP1_INDEX])) { 6082 DRM_DEBUG("si_irq_set: sw int cp1\n"); 6083 cp_int_cntl1 |= TIME_STAMP_INT_ENABLE; 6084 } 6085 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP2_INDEX])) { 6086 DRM_DEBUG("si_irq_set: sw int cp2\n"); 6087 cp_int_cntl2 |= TIME_STAMP_INT_ENABLE; 6088 } 6089 if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) { 6090 DRM_DEBUG("si_irq_set: sw int dma\n"); 6091 dma_cntl |= TRAP_ENABLE; 6092 } 6093 6094 if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_DMA1_INDEX])) { 6095 DRM_DEBUG("si_irq_set: sw int dma1\n"); 6096 dma_cntl1 |= TRAP_ENABLE; 6097 } 6098 6099 WREG32(CP_INT_CNTL_RING0, cp_int_cntl); 6100 WREG32(CP_INT_CNTL_RING1, cp_int_cntl1); 6101 WREG32(CP_INT_CNTL_RING2, cp_int_cntl2); 6102 6103 WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, dma_cntl); 6104 WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, dma_cntl1); 6105 6106 WREG32(GRBM_INT_CNTL, grbm_int_cntl); 6107 6108 if (rdev->irq.dpm_thermal) { 6109 DRM_DEBUG("dpm thermal\n"); 6110 thermal_int |= THERM_INT_MASK_HIGH | THERM_INT_MASK_LOW; 6111 } 6112 6113 for (i = 0; i < rdev->num_crtc; i++) { 6114 radeon_irq_kms_set_irq_n_enabled( 6115 rdev, INT_MASK + crtc_offsets[i], VBLANK_INT_MASK, 6116 rdev->irq.crtc_vblank_int[i] || 6117 atomic_read(&rdev->irq.pflip[i]), "vblank", i); 6118 } 6119 6120 for (i = 0; i < rdev->num_crtc; i++) 6121 WREG32(GRPH_INT_CONTROL + crtc_offsets[i], GRPH_PFLIP_INT_MASK); 6122 6123 if (!ASIC_IS_NODCE(rdev)) { 6124 for (i = 0; i < 6; i++) { 6125 radeon_irq_kms_set_irq_n_enabled( 6126 rdev, DC_HPDx_INT_CONTROL(i), 6127 DC_HPDx_INT_EN | DC_HPDx_RX_INT_EN, 6128 rdev->irq.hpd[i], "HPD", i); 6129 } 6130 } 6131 6132 WREG32(CG_THERMAL_INT, thermal_int); 6133 6134 /* posting read */ 6135 RREG32(SRBM_STATUS); 6136 6137 return 0; 6138 } 6139 6140 /* The order we write back each register here is important */ 6141 static inline void si_irq_ack(struct radeon_device *rdev) 6142 { 6143 int i, j; 6144 u32 *disp_int = rdev->irq.stat_regs.evergreen.disp_int; 6145 u32 *grph_int = rdev->irq.stat_regs.evergreen.grph_int; 6146 6147 if (ASIC_IS_NODCE(rdev)) 6148 return; 6149 6150 for (i = 0; i < 6; i++) { 6151 disp_int[i] = RREG32(si_disp_int_status[i]); 6152 if (i < rdev->num_crtc) 6153 grph_int[i] = RREG32(GRPH_INT_STATUS + crtc_offsets[i]); 6154 } 6155 6156 /* We write back each interrupt register in pairs of two */ 6157 for (i = 0; i < rdev->num_crtc; i += 2) { 6158 for (j = i; j < (i + 2); j++) { 6159 if (grph_int[j] & GRPH_PFLIP_INT_OCCURRED) 6160 WREG32(GRPH_INT_STATUS + crtc_offsets[j], 6161 GRPH_PFLIP_INT_CLEAR); 6162 } 6163 6164 for (j = i; j < (i + 2); j++) { 6165 if (disp_int[j] & LB_D1_VBLANK_INTERRUPT) 6166 WREG32(VBLANK_STATUS + crtc_offsets[j], 6167 VBLANK_ACK); 6168 if (disp_int[j] & LB_D1_VLINE_INTERRUPT) 6169 WREG32(VLINE_STATUS + crtc_offsets[j], 6170 VLINE_ACK); 6171 } 6172 } 6173 6174 for (i = 0; i < 6; i++) { 6175 if (disp_int[i] & DC_HPD1_INTERRUPT) 6176 WREG32_OR(DC_HPDx_INT_CONTROL(i), DC_HPDx_INT_ACK); 6177 } 6178 6179 for (i = 0; i < 6; i++) { 6180 if (disp_int[i] & DC_HPD1_RX_INTERRUPT) 6181 WREG32_OR(DC_HPDx_INT_CONTROL(i), DC_HPDx_RX_INT_ACK); 6182 } 6183 } 6184 6185 static void si_irq_disable(struct radeon_device *rdev) 6186 { 6187 si_disable_interrupts(rdev); 6188 /* Wait and acknowledge irq */ 6189 mdelay(1); 6190 si_irq_ack(rdev); 6191 si_disable_interrupt_state(rdev); 6192 } 6193 6194 static void si_irq_suspend(struct radeon_device *rdev) 6195 { 6196 si_irq_disable(rdev); 6197 si_rlc_stop(rdev); 6198 } 6199 6200 static void si_irq_fini(struct radeon_device *rdev) 6201 { 6202 si_irq_suspend(rdev); 6203 r600_ih_ring_fini(rdev); 6204 } 6205 6206 static inline u32 si_get_ih_wptr(struct radeon_device *rdev) 6207 { 6208 u32 wptr, tmp; 6209 6210 if (rdev->wb.enabled) 6211 wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]); 6212 else 6213 wptr = RREG32(IH_RB_WPTR); 6214 6215 if (wptr & RB_OVERFLOW) { 6216 wptr &= ~RB_OVERFLOW; 6217 /* When a ring buffer overflow happen start parsing interrupt 6218 * from the last not overwritten vector (wptr + 16). Hopefully 6219 * this should allow us to catchup. 6220 */ 6221 dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, 0x%08X, 0x%08X)\n", 6222 wptr, rdev->ih.rptr, (wptr + 16) & rdev->ih.ptr_mask); 6223 rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask; 6224 tmp = RREG32(IH_RB_CNTL); 6225 tmp |= IH_WPTR_OVERFLOW_CLEAR; 6226 WREG32(IH_RB_CNTL, tmp); 6227 } 6228 return (wptr & rdev->ih.ptr_mask); 6229 } 6230 6231 /* SI IV Ring 6232 * Each IV ring entry is 128 bits: 6233 * [7:0] - interrupt source id 6234 * [31:8] - reserved 6235 * [59:32] - interrupt source data 6236 * [63:60] - reserved 6237 * [71:64] - RINGID 6238 * [79:72] - VMID 6239 * [127:80] - reserved 6240 */ 6241 int si_irq_process(struct radeon_device *rdev) 6242 { 6243 u32 *disp_int = rdev->irq.stat_regs.evergreen.disp_int; 6244 u32 crtc_idx, hpd_idx; 6245 u32 mask; 6246 u32 wptr; 6247 u32 rptr; 6248 u32 src_id, src_data, ring_id; 6249 u32 ring_index; 6250 bool queue_hotplug = false; 6251 bool queue_dp = false; 6252 bool queue_thermal = false; 6253 u32 status, addr; 6254 const char *event_name; 6255 6256 if (!rdev->ih.enabled || rdev->shutdown) 6257 return IRQ_NONE; 6258 6259 wptr = si_get_ih_wptr(rdev); 6260 6261 restart_ih: 6262 /* is somebody else already processing irqs? */ 6263 if (atomic_xchg(&rdev->ih.lock, 1)) 6264 return IRQ_NONE; 6265 6266 rptr = rdev->ih.rptr; 6267 DRM_DEBUG("si_irq_process start: rptr %d, wptr %d\n", rptr, wptr); 6268 6269 /* Order reading of wptr vs. reading of IH ring data */ 6270 rmb(); 6271 6272 /* display interrupts */ 6273 si_irq_ack(rdev); 6274 6275 while (rptr != wptr) { 6276 /* wptr/rptr are in bytes! */ 6277 ring_index = rptr / 4; 6278 src_id = le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff; 6279 src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff; 6280 ring_id = le32_to_cpu(rdev->ih.ring[ring_index + 2]) & 0xff; 6281 6282 switch (src_id) { 6283 case 1: /* D1 vblank/vline */ 6284 case 2: /* D2 vblank/vline */ 6285 case 3: /* D3 vblank/vline */ 6286 case 4: /* D4 vblank/vline */ 6287 case 5: /* D5 vblank/vline */ 6288 case 6: /* D6 vblank/vline */ 6289 crtc_idx = src_id - 1; 6290 6291 if (src_data == 0) { /* vblank */ 6292 mask = LB_D1_VBLANK_INTERRUPT; 6293 event_name = "vblank"; 6294 6295 if (rdev->irq.crtc_vblank_int[crtc_idx]) { 6296 drm_handle_vblank(rdev->ddev, crtc_idx); 6297 rdev->pm.vblank_sync = true; 6298 wake_up(&rdev->irq.vblank_queue); 6299 } 6300 if (atomic_read(&rdev->irq.pflip[crtc_idx])) { 6301 radeon_crtc_handle_vblank(rdev, 6302 crtc_idx); 6303 } 6304 6305 } else if (src_data == 1) { /* vline */ 6306 mask = LB_D1_VLINE_INTERRUPT; 6307 event_name = "vline"; 6308 } else { 6309 DRM_DEBUG("Unhandled interrupt: %d %d\n", 6310 src_id, src_data); 6311 break; 6312 } 6313 6314 if (!(disp_int[crtc_idx] & mask)) { 6315 DRM_DEBUG("IH: D%d %s - IH event w/o asserted irq bit?\n", 6316 crtc_idx + 1, event_name); 6317 } 6318 6319 disp_int[crtc_idx] &= ~mask; 6320 DRM_DEBUG("IH: D%d %s\n", crtc_idx + 1, event_name); 6321 6322 break; 6323 case 8: /* D1 page flip */ 6324 case 10: /* D2 page flip */ 6325 case 12: /* D3 page flip */ 6326 case 14: /* D4 page flip */ 6327 case 16: /* D5 page flip */ 6328 case 18: /* D6 page flip */ 6329 DRM_DEBUG("IH: D%d flip\n", ((src_id - 8) >> 1) + 1); 6330 if (radeon_use_pflipirq > 0) 6331 radeon_crtc_handle_flip(rdev, (src_id - 8) >> 1); 6332 break; 6333 case 42: /* HPD hotplug */ 6334 if (src_data <= 5) { 6335 hpd_idx = src_data; 6336 mask = DC_HPD1_INTERRUPT; 6337 queue_hotplug = true; 6338 event_name = "HPD"; 6339 6340 } else if (src_data <= 11) { 6341 hpd_idx = src_data - 6; 6342 mask = DC_HPD1_RX_INTERRUPT; 6343 queue_dp = true; 6344 event_name = "HPD_RX"; 6345 6346 } else { 6347 DRM_DEBUG("Unhandled interrupt: %d %d\n", 6348 src_id, src_data); 6349 break; 6350 } 6351 6352 if (!(disp_int[hpd_idx] & mask)) 6353 DRM_DEBUG("IH: IH event w/o asserted irq bit?\n"); 6354 6355 disp_int[hpd_idx] &= ~mask; 6356 DRM_DEBUG("IH: %s%d\n", event_name, hpd_idx + 1); 6357 break; 6358 case 96: 6359 DRM_ERROR("SRBM_READ_ERROR: 0x%x\n", RREG32(SRBM_READ_ERROR)); 6360 WREG32(SRBM_INT_ACK, 0x1); 6361 break; 6362 case 124: /* UVD */ 6363 DRM_DEBUG("IH: UVD int: 0x%08x\n", src_data); 6364 radeon_fence_process(rdev, R600_RING_TYPE_UVD_INDEX); 6365 break; 6366 case 146: 6367 case 147: 6368 addr = RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR); 6369 status = RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS); 6370 /* reset addr and status */ 6371 WREG32_P(VM_CONTEXT1_CNTL2, 1, ~1); 6372 if (addr == 0x0 && status == 0x0) 6373 break; 6374 dev_err(rdev->dev, "GPU fault detected: %d 0x%08x\n", src_id, src_data); 6375 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n", 6376 addr); 6377 dev_err(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n", 6378 status); 6379 si_vm_decode_fault(rdev, status, addr); 6380 break; 6381 case 176: /* RINGID0 CP_INT */ 6382 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX); 6383 break; 6384 case 177: /* RINGID1 CP_INT */ 6385 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX); 6386 break; 6387 case 178: /* RINGID2 CP_INT */ 6388 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX); 6389 break; 6390 case 181: /* CP EOP event */ 6391 DRM_DEBUG("IH: CP EOP\n"); 6392 switch (ring_id) { 6393 case 0: 6394 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX); 6395 break; 6396 case 1: 6397 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX); 6398 break; 6399 case 2: 6400 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX); 6401 break; 6402 } 6403 break; 6404 case 224: /* DMA trap event */ 6405 DRM_DEBUG("IH: DMA trap\n"); 6406 radeon_fence_process(rdev, R600_RING_TYPE_DMA_INDEX); 6407 break; 6408 case 230: /* thermal low to high */ 6409 DRM_DEBUG("IH: thermal low to high\n"); 6410 rdev->pm.dpm.thermal.high_to_low = false; 6411 queue_thermal = true; 6412 break; 6413 case 231: /* thermal high to low */ 6414 DRM_DEBUG("IH: thermal high to low\n"); 6415 rdev->pm.dpm.thermal.high_to_low = true; 6416 queue_thermal = true; 6417 break; 6418 case 233: /* GUI IDLE */ 6419 DRM_DEBUG("IH: GUI idle\n"); 6420 break; 6421 case 244: /* DMA trap event */ 6422 DRM_DEBUG("IH: DMA1 trap\n"); 6423 radeon_fence_process(rdev, CAYMAN_RING_TYPE_DMA1_INDEX); 6424 break; 6425 default: 6426 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data); 6427 break; 6428 } 6429 6430 /* wptr/rptr are in bytes! */ 6431 rptr += 16; 6432 rptr &= rdev->ih.ptr_mask; 6433 WREG32(IH_RB_RPTR, rptr); 6434 } 6435 if (queue_dp) 6436 schedule_work(&rdev->dp_work); 6437 if (queue_hotplug) 6438 schedule_delayed_work(&rdev->hotplug_work, 0); 6439 if (queue_thermal && rdev->pm.dpm_enabled) 6440 schedule_work(&rdev->pm.dpm.thermal.work); 6441 rdev->ih.rptr = rptr; 6442 atomic_set(&rdev->ih.lock, 0); 6443 6444 /* make sure wptr hasn't changed while processing */ 6445 wptr = si_get_ih_wptr(rdev); 6446 if (wptr != rptr) 6447 goto restart_ih; 6448 6449 return IRQ_HANDLED; 6450 } 6451 6452 /* 6453 * startup/shutdown callbacks 6454 */ 6455 static void si_uvd_init(struct radeon_device *rdev) 6456 { 6457 int r; 6458 6459 if (!rdev->has_uvd) 6460 return; 6461 6462 r = radeon_uvd_init(rdev); 6463 if (r) { 6464 dev_err(rdev->dev, "failed UVD (%d) init.\n", r); 6465 /* 6466 * At this point rdev->uvd.vcpu_bo is NULL which trickles down 6467 * to early fails uvd_v2_2_resume() and thus nothing happens 6468 * there. So it is pointless to try to go through that code 6469 * hence why we disable uvd here. 6470 */ 6471 rdev->has_uvd = 0; 6472 return; 6473 } 6474 rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_obj = NULL; 6475 r600_ring_init(rdev, &rdev->ring[R600_RING_TYPE_UVD_INDEX], 4096); 6476 } 6477 6478 static void si_uvd_start(struct radeon_device *rdev) 6479 { 6480 int r; 6481 6482 if (!rdev->has_uvd) 6483 return; 6484 6485 r = uvd_v2_2_resume(rdev); 6486 if (r) { 6487 dev_err(rdev->dev, "failed UVD resume (%d).\n", r); 6488 goto error; 6489 } 6490 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_UVD_INDEX); 6491 if (r) { 6492 dev_err(rdev->dev, "failed initializing UVD fences (%d).\n", r); 6493 goto error; 6494 } 6495 return; 6496 6497 error: 6498 rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size = 0; 6499 } 6500 6501 static void si_uvd_resume(struct radeon_device *rdev) 6502 { 6503 struct radeon_ring *ring; 6504 int r; 6505 6506 if (!rdev->has_uvd || !rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size) 6507 return; 6508 6509 ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX]; 6510 r = radeon_ring_init(rdev, ring, ring->ring_size, 0, PACKET0(UVD_NO_OP, 0)); 6511 if (r) { 6512 dev_err(rdev->dev, "failed initializing UVD ring (%d).\n", r); 6513 return; 6514 } 6515 r = uvd_v1_0_init(rdev); 6516 if (r) { 6517 dev_err(rdev->dev, "failed initializing UVD (%d).\n", r); 6518 return; 6519 } 6520 } 6521 6522 static void si_vce_init(struct radeon_device *rdev) 6523 { 6524 int r; 6525 6526 if (!rdev->has_vce) 6527 return; 6528 6529 r = radeon_vce_init(rdev); 6530 if (r) { 6531 dev_err(rdev->dev, "failed VCE (%d) init.\n", r); 6532 /* 6533 * At this point rdev->vce.vcpu_bo is NULL which trickles down 6534 * to early fails si_vce_start() and thus nothing happens 6535 * there. So it is pointless to try to go through that code 6536 * hence why we disable vce here. 6537 */ 6538 rdev->has_vce = 0; 6539 return; 6540 } 6541 rdev->ring[TN_RING_TYPE_VCE1_INDEX].ring_obj = NULL; 6542 r600_ring_init(rdev, &rdev->ring[TN_RING_TYPE_VCE1_INDEX], 4096); 6543 rdev->ring[TN_RING_TYPE_VCE2_INDEX].ring_obj = NULL; 6544 r600_ring_init(rdev, &rdev->ring[TN_RING_TYPE_VCE2_INDEX], 4096); 6545 } 6546 6547 static void si_vce_start(struct radeon_device *rdev) 6548 { 6549 int r; 6550 6551 if (!rdev->has_vce) 6552 return; 6553 6554 r = radeon_vce_resume(rdev); 6555 if (r) { 6556 dev_err(rdev->dev, "failed VCE resume (%d).\n", r); 6557 goto error; 6558 } 6559 r = vce_v1_0_resume(rdev); 6560 if (r) { 6561 dev_err(rdev->dev, "failed VCE resume (%d).\n", r); 6562 goto error; 6563 } 6564 r = radeon_fence_driver_start_ring(rdev, TN_RING_TYPE_VCE1_INDEX); 6565 if (r) { 6566 dev_err(rdev->dev, "failed initializing VCE1 fences (%d).\n", r); 6567 goto error; 6568 } 6569 r = radeon_fence_driver_start_ring(rdev, TN_RING_TYPE_VCE2_INDEX); 6570 if (r) { 6571 dev_err(rdev->dev, "failed initializing VCE2 fences (%d).\n", r); 6572 goto error; 6573 } 6574 return; 6575 6576 error: 6577 rdev->ring[TN_RING_TYPE_VCE1_INDEX].ring_size = 0; 6578 rdev->ring[TN_RING_TYPE_VCE2_INDEX].ring_size = 0; 6579 } 6580 6581 static void si_vce_resume(struct radeon_device *rdev) 6582 { 6583 struct radeon_ring *ring; 6584 int r; 6585 6586 if (!rdev->has_vce || !rdev->ring[TN_RING_TYPE_VCE1_INDEX].ring_size) 6587 return; 6588 6589 ring = &rdev->ring[TN_RING_TYPE_VCE1_INDEX]; 6590 r = radeon_ring_init(rdev, ring, ring->ring_size, 0, VCE_CMD_NO_OP); 6591 if (r) { 6592 dev_err(rdev->dev, "failed initializing VCE1 ring (%d).\n", r); 6593 return; 6594 } 6595 ring = &rdev->ring[TN_RING_TYPE_VCE2_INDEX]; 6596 r = radeon_ring_init(rdev, ring, ring->ring_size, 0, VCE_CMD_NO_OP); 6597 if (r) { 6598 dev_err(rdev->dev, "failed initializing VCE1 ring (%d).\n", r); 6599 return; 6600 } 6601 r = vce_v1_0_init(rdev); 6602 if (r) { 6603 dev_err(rdev->dev, "failed initializing VCE (%d).\n", r); 6604 return; 6605 } 6606 } 6607 6608 static int si_startup(struct radeon_device *rdev) 6609 { 6610 struct radeon_ring *ring; 6611 int r; 6612 6613 /* enable pcie gen2/3 link */ 6614 si_pcie_gen3_enable(rdev); 6615 /* enable aspm */ 6616 si_program_aspm(rdev); 6617 6618 /* scratch needs to be initialized before MC */ 6619 r = r600_vram_scratch_init(rdev); 6620 if (r) 6621 return r; 6622 6623 si_mc_program(rdev); 6624 6625 if (!rdev->pm.dpm_enabled) { 6626 r = si_mc_load_microcode(rdev); 6627 if (r) { 6628 DRM_ERROR("Failed to load MC firmware!\n"); 6629 return r; 6630 } 6631 } 6632 6633 r = si_pcie_gart_enable(rdev); 6634 if (r) 6635 return r; 6636 si_gpu_init(rdev); 6637 6638 /* allocate rlc buffers */ 6639 if (rdev->family == CHIP_VERDE) { 6640 rdev->rlc.reg_list = verde_rlc_save_restore_register_list; 6641 rdev->rlc.reg_list_size = 6642 (u32)ARRAY_SIZE(verde_rlc_save_restore_register_list); 6643 } 6644 rdev->rlc.cs_data = si_cs_data; 6645 r = sumo_rlc_init(rdev); 6646 if (r) { 6647 DRM_ERROR("Failed to init rlc BOs!\n"); 6648 return r; 6649 } 6650 6651 /* allocate wb buffer */ 6652 r = radeon_wb_init(rdev); 6653 if (r) 6654 return r; 6655 6656 r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX); 6657 if (r) { 6658 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r); 6659 return r; 6660 } 6661 6662 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP1_INDEX); 6663 if (r) { 6664 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r); 6665 return r; 6666 } 6667 6668 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP2_INDEX); 6669 if (r) { 6670 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r); 6671 return r; 6672 } 6673 6674 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX); 6675 if (r) { 6676 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r); 6677 return r; 6678 } 6679 6680 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX); 6681 if (r) { 6682 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r); 6683 return r; 6684 } 6685 6686 si_uvd_start(rdev); 6687 si_vce_start(rdev); 6688 6689 /* Enable IRQ */ 6690 if (!rdev->irq.installed) { 6691 r = radeon_irq_kms_init(rdev); 6692 if (r) 6693 return r; 6694 } 6695 6696 r = si_irq_init(rdev); 6697 if (r) { 6698 DRM_ERROR("radeon: IH init failed (%d).\n", r); 6699 radeon_irq_kms_fini(rdev); 6700 return r; 6701 } 6702 si_irq_set(rdev); 6703 6704 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 6705 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET, 6706 RADEON_CP_PACKET2); 6707 if (r) 6708 return r; 6709 6710 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX]; 6711 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP1_RPTR_OFFSET, 6712 RADEON_CP_PACKET2); 6713 if (r) 6714 return r; 6715 6716 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX]; 6717 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP2_RPTR_OFFSET, 6718 RADEON_CP_PACKET2); 6719 if (r) 6720 return r; 6721 6722 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX]; 6723 r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET, 6724 DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0)); 6725 if (r) 6726 return r; 6727 6728 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]; 6729 r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET, 6730 DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0)); 6731 if (r) 6732 return r; 6733 6734 r = si_cp_load_microcode(rdev); 6735 if (r) 6736 return r; 6737 r = si_cp_resume(rdev); 6738 if (r) 6739 return r; 6740 6741 r = cayman_dma_resume(rdev); 6742 if (r) 6743 return r; 6744 6745 si_uvd_resume(rdev); 6746 si_vce_resume(rdev); 6747 6748 r = radeon_ib_pool_init(rdev); 6749 if (r) { 6750 dev_err(rdev->dev, "IB initialization failed (%d).\n", r); 6751 return r; 6752 } 6753 6754 r = radeon_vm_manager_init(rdev); 6755 if (r) { 6756 dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r); 6757 return r; 6758 } 6759 6760 r = radeon_audio_init(rdev); 6761 if (r) 6762 return r; 6763 6764 return 0; 6765 } 6766 6767 int si_resume(struct radeon_device *rdev) 6768 { 6769 int r; 6770 6771 /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw, 6772 * posting will perform necessary task to bring back GPU into good 6773 * shape. 6774 */ 6775 /* post card */ 6776 atom_asic_init(rdev->mode_info.atom_context); 6777 6778 /* init golden registers */ 6779 si_init_golden_registers(rdev); 6780 6781 if (rdev->pm.pm_method == PM_METHOD_DPM) 6782 radeon_pm_resume(rdev); 6783 6784 rdev->accel_working = true; 6785 r = si_startup(rdev); 6786 if (r) { 6787 DRM_ERROR("si startup failed on resume\n"); 6788 rdev->accel_working = false; 6789 return r; 6790 } 6791 6792 return r; 6793 6794 } 6795 6796 int si_suspend(struct radeon_device *rdev) 6797 { 6798 radeon_pm_suspend(rdev); 6799 radeon_audio_fini(rdev); 6800 radeon_vm_manager_fini(rdev); 6801 si_cp_enable(rdev, false); 6802 cayman_dma_stop(rdev); 6803 if (rdev->has_uvd) { 6804 uvd_v1_0_fini(rdev); 6805 radeon_uvd_suspend(rdev); 6806 } 6807 if (rdev->has_vce) 6808 radeon_vce_suspend(rdev); 6809 si_fini_pg(rdev); 6810 si_fini_cg(rdev); 6811 si_irq_suspend(rdev); 6812 radeon_wb_disable(rdev); 6813 si_pcie_gart_disable(rdev); 6814 return 0; 6815 } 6816 6817 /* Plan is to move initialization in that function and use 6818 * helper function so that radeon_device_init pretty much 6819 * do nothing more than calling asic specific function. This 6820 * should also allow to remove a bunch of callback function 6821 * like vram_info. 6822 */ 6823 int si_init(struct radeon_device *rdev) 6824 { 6825 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 6826 int r; 6827 6828 /* Read BIOS */ 6829 if (!radeon_get_bios(rdev)) { 6830 if (ASIC_IS_AVIVO(rdev)) 6831 return -EINVAL; 6832 } 6833 /* Must be an ATOMBIOS */ 6834 if (!rdev->is_atom_bios) { 6835 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n"); 6836 return -EINVAL; 6837 } 6838 r = radeon_atombios_init(rdev); 6839 if (r) 6840 return r; 6841 6842 /* Post card if necessary */ 6843 if (!radeon_card_posted(rdev)) { 6844 if (!rdev->bios) { 6845 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n"); 6846 return -EINVAL; 6847 } 6848 DRM_INFO("GPU not posted. posting now...\n"); 6849 atom_asic_init(rdev->mode_info.atom_context); 6850 } 6851 /* init golden registers */ 6852 si_init_golden_registers(rdev); 6853 /* Initialize scratch registers */ 6854 si_scratch_init(rdev); 6855 /* Initialize surface registers */ 6856 radeon_surface_init(rdev); 6857 /* Initialize clocks */ 6858 radeon_get_clock_info(rdev->ddev); 6859 6860 /* Fence driver */ 6861 r = radeon_fence_driver_init(rdev); 6862 if (r) 6863 return r; 6864 6865 /* initialize memory controller */ 6866 r = si_mc_init(rdev); 6867 if (r) 6868 return r; 6869 /* Memory manager */ 6870 r = radeon_bo_init(rdev); 6871 if (r) 6872 return r; 6873 6874 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw || 6875 !rdev->rlc_fw || !rdev->mc_fw) { 6876 r = si_init_microcode(rdev); 6877 if (r) { 6878 DRM_ERROR("Failed to load firmware!\n"); 6879 return r; 6880 } 6881 } 6882 6883 /* Initialize power management */ 6884 radeon_pm_init(rdev); 6885 6886 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 6887 ring->ring_obj = NULL; 6888 r600_ring_init(rdev, ring, 1024 * 1024); 6889 6890 ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX]; 6891 ring->ring_obj = NULL; 6892 r600_ring_init(rdev, ring, 1024 * 1024); 6893 6894 ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX]; 6895 ring->ring_obj = NULL; 6896 r600_ring_init(rdev, ring, 1024 * 1024); 6897 6898 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX]; 6899 ring->ring_obj = NULL; 6900 r600_ring_init(rdev, ring, 64 * 1024); 6901 6902 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]; 6903 ring->ring_obj = NULL; 6904 r600_ring_init(rdev, ring, 64 * 1024); 6905 6906 si_uvd_init(rdev); 6907 si_vce_init(rdev); 6908 6909 rdev->ih.ring_obj = NULL; 6910 r600_ih_ring_init(rdev, 64 * 1024); 6911 6912 r = r600_pcie_gart_init(rdev); 6913 if (r) 6914 return r; 6915 6916 rdev->accel_working = true; 6917 r = si_startup(rdev); 6918 if (r) { 6919 dev_err(rdev->dev, "disabling GPU acceleration\n"); 6920 si_cp_fini(rdev); 6921 cayman_dma_fini(rdev); 6922 si_irq_fini(rdev); 6923 sumo_rlc_fini(rdev); 6924 radeon_wb_fini(rdev); 6925 radeon_ib_pool_fini(rdev); 6926 radeon_vm_manager_fini(rdev); 6927 radeon_irq_kms_fini(rdev); 6928 si_pcie_gart_fini(rdev); 6929 rdev->accel_working = false; 6930 } 6931 6932 /* Don't start up if the MC ucode is missing. 6933 * The default clocks and voltages before the MC ucode 6934 * is loaded are not suffient for advanced operations. 6935 */ 6936 if (!rdev->mc_fw) { 6937 DRM_ERROR("radeon: MC ucode required for NI+.\n"); 6938 return -EINVAL; 6939 } 6940 6941 return 0; 6942 } 6943 6944 void si_fini(struct radeon_device *rdev) 6945 { 6946 radeon_pm_fini(rdev); 6947 si_cp_fini(rdev); 6948 cayman_dma_fini(rdev); 6949 si_fini_pg(rdev); 6950 si_fini_cg(rdev); 6951 si_irq_fini(rdev); 6952 sumo_rlc_fini(rdev); 6953 radeon_wb_fini(rdev); 6954 radeon_vm_manager_fini(rdev); 6955 radeon_ib_pool_fini(rdev); 6956 radeon_irq_kms_fini(rdev); 6957 if (rdev->has_uvd) { 6958 uvd_v1_0_fini(rdev); 6959 radeon_uvd_fini(rdev); 6960 } 6961 if (rdev->has_vce) 6962 radeon_vce_fini(rdev); 6963 si_pcie_gart_fini(rdev); 6964 r600_vram_scratch_fini(rdev); 6965 radeon_gem_fini(rdev); 6966 radeon_fence_driver_fini(rdev); 6967 radeon_bo_fini(rdev); 6968 radeon_atombios_fini(rdev); 6969 kfree(rdev->bios); 6970 rdev->bios = NULL; 6971 } 6972 6973 /** 6974 * si_get_gpu_clock_counter - return GPU clock counter snapshot 6975 * 6976 * @rdev: radeon_device pointer 6977 * 6978 * Fetches a GPU clock counter snapshot (SI). 6979 * Returns the 64 bit clock counter snapshot. 6980 */ 6981 uint64_t si_get_gpu_clock_counter(struct radeon_device *rdev) 6982 { 6983 uint64_t clock; 6984 6985 mutex_lock(&rdev->gpu_clock_mutex); 6986 WREG32(RLC_CAPTURE_GPU_CLOCK_COUNT, 1); 6987 clock = (uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_LSB) | 6988 ((uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_MSB) << 32ULL); 6989 mutex_unlock(&rdev->gpu_clock_mutex); 6990 return clock; 6991 } 6992 6993 int si_set_uvd_clocks(struct radeon_device *rdev, u32 vclk, u32 dclk) 6994 { 6995 unsigned fb_div = 0, vclk_div = 0, dclk_div = 0; 6996 int r; 6997 6998 /* bypass vclk and dclk with bclk */ 6999 WREG32_P(CG_UPLL_FUNC_CNTL_2, 7000 VCLK_SRC_SEL(1) | DCLK_SRC_SEL(1), 7001 ~(VCLK_SRC_SEL_MASK | DCLK_SRC_SEL_MASK)); 7002 7003 /* put PLL in bypass mode */ 7004 WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_BYPASS_EN_MASK, ~UPLL_BYPASS_EN_MASK); 7005 7006 if (!vclk || !dclk) { 7007 /* keep the Bypass mode */ 7008 return 0; 7009 } 7010 7011 r = radeon_uvd_calc_upll_dividers(rdev, vclk, dclk, 125000, 250000, 7012 16384, 0x03FFFFFF, 0, 128, 5, 7013 &fb_div, &vclk_div, &dclk_div); 7014 if (r) 7015 return r; 7016 7017 /* set RESET_ANTI_MUX to 0 */ 7018 WREG32_P(CG_UPLL_FUNC_CNTL_5, 0, ~RESET_ANTI_MUX_MASK); 7019 7020 /* set VCO_MODE to 1 */ 7021 WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_VCO_MODE_MASK, ~UPLL_VCO_MODE_MASK); 7022 7023 /* disable sleep mode */ 7024 WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_SLEEP_MASK); 7025 7026 /* deassert UPLL_RESET */ 7027 WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_RESET_MASK); 7028 7029 mdelay(1); 7030 7031 r = radeon_uvd_send_upll_ctlreq(rdev, CG_UPLL_FUNC_CNTL); 7032 if (r) 7033 return r; 7034 7035 /* assert UPLL_RESET again */ 7036 WREG32_P(CG_UPLL_FUNC_CNTL, UPLL_RESET_MASK, ~UPLL_RESET_MASK); 7037 7038 /* disable spread spectrum. */ 7039 WREG32_P(CG_UPLL_SPREAD_SPECTRUM, 0, ~SSEN_MASK); 7040 7041 /* set feedback divider */ 7042 WREG32_P(CG_UPLL_FUNC_CNTL_3, UPLL_FB_DIV(fb_div), ~UPLL_FB_DIV_MASK); 7043 7044 /* set ref divider to 0 */ 7045 WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_REF_DIV_MASK); 7046 7047 if (fb_div < 307200) 7048 WREG32_P(CG_UPLL_FUNC_CNTL_4, 0, ~UPLL_SPARE_ISPARE9); 7049 else 7050 WREG32_P(CG_UPLL_FUNC_CNTL_4, UPLL_SPARE_ISPARE9, ~UPLL_SPARE_ISPARE9); 7051 7052 /* set PDIV_A and PDIV_B */ 7053 WREG32_P(CG_UPLL_FUNC_CNTL_2, 7054 UPLL_PDIV_A(vclk_div) | UPLL_PDIV_B(dclk_div), 7055 ~(UPLL_PDIV_A_MASK | UPLL_PDIV_B_MASK)); 7056 7057 /* give the PLL some time to settle */ 7058 mdelay(15); 7059 7060 /* deassert PLL_RESET */ 7061 WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_RESET_MASK); 7062 7063 mdelay(15); 7064 7065 /* switch from bypass mode to normal mode */ 7066 WREG32_P(CG_UPLL_FUNC_CNTL, 0, ~UPLL_BYPASS_EN_MASK); 7067 7068 r = radeon_uvd_send_upll_ctlreq(rdev, CG_UPLL_FUNC_CNTL); 7069 if (r) 7070 return r; 7071 7072 /* switch VCLK and DCLK selection */ 7073 WREG32_P(CG_UPLL_FUNC_CNTL_2, 7074 VCLK_SRC_SEL(2) | DCLK_SRC_SEL(2), 7075 ~(VCLK_SRC_SEL_MASK | DCLK_SRC_SEL_MASK)); 7076 7077 mdelay(100); 7078 7079 return 0; 7080 } 7081 7082 static void si_pcie_gen3_enable(struct radeon_device *rdev) 7083 { 7084 struct pci_dev *root = rdev->pdev->bus->self; 7085 enum pci_bus_speed speed_cap; 7086 int bridge_pos, gpu_pos; 7087 u32 speed_cntl, current_data_rate; 7088 int i; 7089 u16 tmp16; 7090 7091 if (pci_is_root_bus(rdev->pdev->bus)) 7092 return; 7093 7094 if (radeon_pcie_gen2 == 0) 7095 return; 7096 7097 if (rdev->flags & RADEON_IS_IGP) 7098 return; 7099 7100 if (!(rdev->flags & RADEON_IS_PCIE)) 7101 return; 7102 7103 speed_cap = pcie_get_speed_cap(root); 7104 if (speed_cap == PCI_SPEED_UNKNOWN) 7105 return; 7106 7107 if ((speed_cap != PCIE_SPEED_8_0GT) && 7108 (speed_cap != PCIE_SPEED_5_0GT)) 7109 return; 7110 7111 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL); 7112 current_data_rate = (speed_cntl & LC_CURRENT_DATA_RATE_MASK) >> 7113 LC_CURRENT_DATA_RATE_SHIFT; 7114 if (speed_cap == PCIE_SPEED_8_0GT) { 7115 if (current_data_rate == 2) { 7116 DRM_INFO("PCIE gen 3 link speeds already enabled\n"); 7117 return; 7118 } 7119 DRM_INFO("enabling PCIE gen 3 link speeds, disable with radeon.pcie_gen2=0\n"); 7120 } else if (speed_cap == PCIE_SPEED_5_0GT) { 7121 if (current_data_rate == 1) { 7122 DRM_INFO("PCIE gen 2 link speeds already enabled\n"); 7123 return; 7124 } 7125 DRM_INFO("enabling PCIE gen 2 link speeds, disable with radeon.pcie_gen2=0\n"); 7126 } 7127 7128 bridge_pos = pci_pcie_cap(root); 7129 if (!bridge_pos) 7130 return; 7131 7132 gpu_pos = pci_pcie_cap(rdev->pdev); 7133 if (!gpu_pos) 7134 return; 7135 7136 if (speed_cap == PCIE_SPEED_8_0GT) { 7137 /* re-try equalization if gen3 is not already enabled */ 7138 if (current_data_rate != 2) { 7139 u16 bridge_cfg, gpu_cfg; 7140 u16 bridge_cfg2, gpu_cfg2; 7141 u32 max_lw, current_lw, tmp; 7142 7143 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &bridge_cfg); 7144 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &gpu_cfg); 7145 7146 tmp16 = bridge_cfg | PCI_EXP_LNKCTL_HAWD; 7147 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL, tmp16); 7148 7149 tmp16 = gpu_cfg | PCI_EXP_LNKCTL_HAWD; 7150 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, tmp16); 7151 7152 tmp = RREG32_PCIE(PCIE_LC_STATUS1); 7153 max_lw = (tmp & LC_DETECTED_LINK_WIDTH_MASK) >> LC_DETECTED_LINK_WIDTH_SHIFT; 7154 current_lw = (tmp & LC_OPERATING_LINK_WIDTH_MASK) >> LC_OPERATING_LINK_WIDTH_SHIFT; 7155 7156 if (current_lw < max_lw) { 7157 tmp = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL); 7158 if (tmp & LC_RENEGOTIATION_SUPPORT) { 7159 tmp &= ~(LC_LINK_WIDTH_MASK | LC_UPCONFIGURE_DIS); 7160 tmp |= (max_lw << LC_LINK_WIDTH_SHIFT); 7161 tmp |= LC_UPCONFIGURE_SUPPORT | LC_RENEGOTIATE_EN | LC_RECONFIG_NOW; 7162 WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, tmp); 7163 } 7164 } 7165 7166 for (i = 0; i < 10; i++) { 7167 /* check status */ 7168 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_DEVSTA, &tmp16); 7169 if (tmp16 & PCI_EXP_DEVSTA_TRPND) 7170 break; 7171 7172 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &bridge_cfg); 7173 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &gpu_cfg); 7174 7175 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, &bridge_cfg2); 7176 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &gpu_cfg2); 7177 7178 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4); 7179 tmp |= LC_SET_QUIESCE; 7180 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp); 7181 7182 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4); 7183 tmp |= LC_REDO_EQ; 7184 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp); 7185 7186 msleep(100); 7187 7188 /* linkctl */ 7189 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL, &tmp16); 7190 tmp16 &= ~PCI_EXP_LNKCTL_HAWD; 7191 tmp16 |= (bridge_cfg & PCI_EXP_LNKCTL_HAWD); 7192 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL, tmp16); 7193 7194 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, &tmp16); 7195 tmp16 &= ~PCI_EXP_LNKCTL_HAWD; 7196 tmp16 |= (gpu_cfg & PCI_EXP_LNKCTL_HAWD); 7197 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL, tmp16); 7198 7199 /* linkctl2 */ 7200 pci_read_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, &tmp16); 7201 tmp16 &= ~((1 << 4) | (7 << 9)); 7202 tmp16 |= (bridge_cfg2 & ((1 << 4) | (7 << 9))); 7203 pci_write_config_word(root, bridge_pos + PCI_EXP_LNKCTL2, tmp16); 7204 7205 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &tmp16); 7206 tmp16 &= ~((1 << 4) | (7 << 9)); 7207 tmp16 |= (gpu_cfg2 & ((1 << 4) | (7 << 9))); 7208 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, tmp16); 7209 7210 tmp = RREG32_PCIE_PORT(PCIE_LC_CNTL4); 7211 tmp &= ~LC_SET_QUIESCE; 7212 WREG32_PCIE_PORT(PCIE_LC_CNTL4, tmp); 7213 } 7214 } 7215 } 7216 7217 /* set the link speed */ 7218 speed_cntl |= LC_FORCE_EN_SW_SPEED_CHANGE | LC_FORCE_DIS_HW_SPEED_CHANGE; 7219 speed_cntl &= ~LC_FORCE_DIS_SW_SPEED_CHANGE; 7220 WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl); 7221 7222 pci_read_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, &tmp16); 7223 tmp16 &= ~0xf; 7224 if (speed_cap == PCIE_SPEED_8_0GT) 7225 tmp16 |= 3; /* gen3 */ 7226 else if (speed_cap == PCIE_SPEED_5_0GT) 7227 tmp16 |= 2; /* gen2 */ 7228 else 7229 tmp16 |= 1; /* gen1 */ 7230 pci_write_config_word(rdev->pdev, gpu_pos + PCI_EXP_LNKCTL2, tmp16); 7231 7232 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL); 7233 speed_cntl |= LC_INITIATE_LINK_SPEED_CHANGE; 7234 WREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL, speed_cntl); 7235 7236 for (i = 0; i < rdev->usec_timeout; i++) { 7237 speed_cntl = RREG32_PCIE_PORT(PCIE_LC_SPEED_CNTL); 7238 if ((speed_cntl & LC_INITIATE_LINK_SPEED_CHANGE) == 0) 7239 break; 7240 udelay(1); 7241 } 7242 } 7243 7244 static void si_program_aspm(struct radeon_device *rdev) 7245 { 7246 u32 data, orig; 7247 bool disable_l0s = false, disable_l1 = false, disable_plloff_in_l1 = false; 7248 bool disable_clkreq = false; 7249 7250 if (radeon_aspm == 0) 7251 return; 7252 7253 if (!(rdev->flags & RADEON_IS_PCIE)) 7254 return; 7255 7256 orig = data = RREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL); 7257 data &= ~LC_XMIT_N_FTS_MASK; 7258 data |= LC_XMIT_N_FTS(0x24) | LC_XMIT_N_FTS_OVERRIDE_EN; 7259 if (orig != data) 7260 WREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL, data); 7261 7262 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL3); 7263 data |= LC_GO_TO_RECOVERY; 7264 if (orig != data) 7265 WREG32_PCIE_PORT(PCIE_LC_CNTL3, data); 7266 7267 orig = data = RREG32_PCIE(PCIE_P_CNTL); 7268 data |= P_IGNORE_EDB_ERR; 7269 if (orig != data) 7270 WREG32_PCIE(PCIE_P_CNTL, data); 7271 7272 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL); 7273 data &= ~(LC_L0S_INACTIVITY_MASK | LC_L1_INACTIVITY_MASK); 7274 data |= LC_PMI_TO_L1_DIS; 7275 if (!disable_l0s) 7276 data |= LC_L0S_INACTIVITY(7); 7277 7278 if (!disable_l1) { 7279 data |= LC_L1_INACTIVITY(7); 7280 data &= ~LC_PMI_TO_L1_DIS; 7281 if (orig != data) 7282 WREG32_PCIE_PORT(PCIE_LC_CNTL, data); 7283 7284 if (!disable_plloff_in_l1) { 7285 bool clk_req_support; 7286 7287 orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0); 7288 data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK); 7289 data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7); 7290 if (orig != data) 7291 WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0, data); 7292 7293 orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1); 7294 data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK); 7295 data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7); 7296 if (orig != data) 7297 WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1, data); 7298 7299 orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0); 7300 data &= ~(PLL_POWER_STATE_IN_OFF_0_MASK | PLL_POWER_STATE_IN_TXS2_0_MASK); 7301 data |= PLL_POWER_STATE_IN_OFF_0(7) | PLL_POWER_STATE_IN_TXS2_0(7); 7302 if (orig != data) 7303 WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0, data); 7304 7305 orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1); 7306 data &= ~(PLL_POWER_STATE_IN_OFF_1_MASK | PLL_POWER_STATE_IN_TXS2_1_MASK); 7307 data |= PLL_POWER_STATE_IN_OFF_1(7) | PLL_POWER_STATE_IN_TXS2_1(7); 7308 if (orig != data) 7309 WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1, data); 7310 7311 if ((rdev->family != CHIP_OLAND) && (rdev->family != CHIP_HAINAN)) { 7312 orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0); 7313 data &= ~PLL_RAMP_UP_TIME_0_MASK; 7314 if (orig != data) 7315 WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_0, data); 7316 7317 orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1); 7318 data &= ~PLL_RAMP_UP_TIME_1_MASK; 7319 if (orig != data) 7320 WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_1, data); 7321 7322 orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_2); 7323 data &= ~PLL_RAMP_UP_TIME_2_MASK; 7324 if (orig != data) 7325 WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_2, data); 7326 7327 orig = data = RREG32_PIF_PHY0(PB0_PIF_PWRDOWN_3); 7328 data &= ~PLL_RAMP_UP_TIME_3_MASK; 7329 if (orig != data) 7330 WREG32_PIF_PHY0(PB0_PIF_PWRDOWN_3, data); 7331 7332 orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0); 7333 data &= ~PLL_RAMP_UP_TIME_0_MASK; 7334 if (orig != data) 7335 WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_0, data); 7336 7337 orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1); 7338 data &= ~PLL_RAMP_UP_TIME_1_MASK; 7339 if (orig != data) 7340 WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_1, data); 7341 7342 orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_2); 7343 data &= ~PLL_RAMP_UP_TIME_2_MASK; 7344 if (orig != data) 7345 WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_2, data); 7346 7347 orig = data = RREG32_PIF_PHY1(PB1_PIF_PWRDOWN_3); 7348 data &= ~PLL_RAMP_UP_TIME_3_MASK; 7349 if (orig != data) 7350 WREG32_PIF_PHY1(PB1_PIF_PWRDOWN_3, data); 7351 } 7352 orig = data = RREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL); 7353 data &= ~LC_DYN_LANES_PWR_STATE_MASK; 7354 data |= LC_DYN_LANES_PWR_STATE(3); 7355 if (orig != data) 7356 WREG32_PCIE_PORT(PCIE_LC_LINK_WIDTH_CNTL, data); 7357 7358 orig = data = RREG32_PIF_PHY0(PB0_PIF_CNTL); 7359 data &= ~LS2_EXIT_TIME_MASK; 7360 if ((rdev->family == CHIP_OLAND) || (rdev->family == CHIP_HAINAN)) 7361 data |= LS2_EXIT_TIME(5); 7362 if (orig != data) 7363 WREG32_PIF_PHY0(PB0_PIF_CNTL, data); 7364 7365 orig = data = RREG32_PIF_PHY1(PB1_PIF_CNTL); 7366 data &= ~LS2_EXIT_TIME_MASK; 7367 if ((rdev->family == CHIP_OLAND) || (rdev->family == CHIP_HAINAN)) 7368 data |= LS2_EXIT_TIME(5); 7369 if (orig != data) 7370 WREG32_PIF_PHY1(PB1_PIF_CNTL, data); 7371 7372 if (!disable_clkreq && 7373 !pci_is_root_bus(rdev->pdev->bus)) { 7374 struct pci_dev *root = rdev->pdev->bus->self; 7375 u32 lnkcap; 7376 7377 clk_req_support = false; 7378 pcie_capability_read_dword(root, PCI_EXP_LNKCAP, &lnkcap); 7379 if (lnkcap & PCI_EXP_LNKCAP_CLKPM) 7380 clk_req_support = true; 7381 } else { 7382 clk_req_support = false; 7383 } 7384 7385 if (clk_req_support) { 7386 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL2); 7387 data |= LC_ALLOW_PDWN_IN_L1 | LC_ALLOW_PDWN_IN_L23; 7388 if (orig != data) 7389 WREG32_PCIE_PORT(PCIE_LC_CNTL2, data); 7390 7391 orig = data = RREG32(THM_CLK_CNTL); 7392 data &= ~(CMON_CLK_SEL_MASK | TMON_CLK_SEL_MASK); 7393 data |= CMON_CLK_SEL(1) | TMON_CLK_SEL(1); 7394 if (orig != data) 7395 WREG32(THM_CLK_CNTL, data); 7396 7397 orig = data = RREG32(MISC_CLK_CNTL); 7398 data &= ~(DEEP_SLEEP_CLK_SEL_MASK | ZCLK_SEL_MASK); 7399 data |= DEEP_SLEEP_CLK_SEL(1) | ZCLK_SEL(1); 7400 if (orig != data) 7401 WREG32(MISC_CLK_CNTL, data); 7402 7403 orig = data = RREG32(CG_CLKPIN_CNTL); 7404 data &= ~BCLK_AS_XCLK; 7405 if (orig != data) 7406 WREG32(CG_CLKPIN_CNTL, data); 7407 7408 orig = data = RREG32(CG_CLKPIN_CNTL_2); 7409 data &= ~FORCE_BIF_REFCLK_EN; 7410 if (orig != data) 7411 WREG32(CG_CLKPIN_CNTL_2, data); 7412 7413 orig = data = RREG32(MPLL_BYPASSCLK_SEL); 7414 data &= ~MPLL_CLKOUT_SEL_MASK; 7415 data |= MPLL_CLKOUT_SEL(4); 7416 if (orig != data) 7417 WREG32(MPLL_BYPASSCLK_SEL, data); 7418 7419 orig = data = RREG32(SPLL_CNTL_MODE); 7420 data &= ~SPLL_REFCLK_SEL_MASK; 7421 if (orig != data) 7422 WREG32(SPLL_CNTL_MODE, data); 7423 } 7424 } 7425 } else { 7426 if (orig != data) 7427 WREG32_PCIE_PORT(PCIE_LC_CNTL, data); 7428 } 7429 7430 orig = data = RREG32_PCIE(PCIE_CNTL2); 7431 data |= SLV_MEM_LS_EN | MST_MEM_LS_EN | REPLAY_MEM_LS_EN; 7432 if (orig != data) 7433 WREG32_PCIE(PCIE_CNTL2, data); 7434 7435 if (!disable_l0s) { 7436 data = RREG32_PCIE_PORT(PCIE_LC_N_FTS_CNTL); 7437 if((data & LC_N_FTS_MASK) == LC_N_FTS_MASK) { 7438 data = RREG32_PCIE(PCIE_LC_STATUS1); 7439 if ((data & LC_REVERSE_XMIT) && (data & LC_REVERSE_RCVR)) { 7440 orig = data = RREG32_PCIE_PORT(PCIE_LC_CNTL); 7441 data &= ~LC_L0S_INACTIVITY_MASK; 7442 if (orig != data) 7443 WREG32_PCIE_PORT(PCIE_LC_CNTL, data); 7444 } 7445 } 7446 } 7447 } 7448 7449 static int si_vce_send_vcepll_ctlreq(struct radeon_device *rdev) 7450 { 7451 unsigned i; 7452 7453 /* make sure VCEPLL_CTLREQ is deasserted */ 7454 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~UPLL_CTLREQ_MASK); 7455 7456 mdelay(10); 7457 7458 /* assert UPLL_CTLREQ */ 7459 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, UPLL_CTLREQ_MASK, ~UPLL_CTLREQ_MASK); 7460 7461 /* wait for CTLACK and CTLACK2 to get asserted */ 7462 for (i = 0; i < 100; ++i) { 7463 uint32_t mask = UPLL_CTLACK_MASK | UPLL_CTLACK2_MASK; 7464 if ((RREG32_SMC(CG_VCEPLL_FUNC_CNTL) & mask) == mask) 7465 break; 7466 mdelay(10); 7467 } 7468 7469 /* deassert UPLL_CTLREQ */ 7470 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~UPLL_CTLREQ_MASK); 7471 7472 if (i == 100) { 7473 DRM_ERROR("Timeout setting UVD clocks!\n"); 7474 return -ETIMEDOUT; 7475 } 7476 7477 return 0; 7478 } 7479 7480 int si_set_vce_clocks(struct radeon_device *rdev, u32 evclk, u32 ecclk) 7481 { 7482 unsigned fb_div = 0, evclk_div = 0, ecclk_div = 0; 7483 int r; 7484 7485 /* bypass evclk and ecclk with bclk */ 7486 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL_2, 7487 EVCLK_SRC_SEL(1) | ECCLK_SRC_SEL(1), 7488 ~(EVCLK_SRC_SEL_MASK | ECCLK_SRC_SEL_MASK)); 7489 7490 /* put PLL in bypass mode */ 7491 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, VCEPLL_BYPASS_EN_MASK, 7492 ~VCEPLL_BYPASS_EN_MASK); 7493 7494 if (!evclk || !ecclk) { 7495 /* keep the Bypass mode, put PLL to sleep */ 7496 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, VCEPLL_SLEEP_MASK, 7497 ~VCEPLL_SLEEP_MASK); 7498 return 0; 7499 } 7500 7501 r = radeon_uvd_calc_upll_dividers(rdev, evclk, ecclk, 125000, 250000, 7502 16384, 0x03FFFFFF, 0, 128, 5, 7503 &fb_div, &evclk_div, &ecclk_div); 7504 if (r) 7505 return r; 7506 7507 /* set RESET_ANTI_MUX to 0 */ 7508 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL_5, 0, ~RESET_ANTI_MUX_MASK); 7509 7510 /* set VCO_MODE to 1 */ 7511 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, VCEPLL_VCO_MODE_MASK, 7512 ~VCEPLL_VCO_MODE_MASK); 7513 7514 /* toggle VCEPLL_SLEEP to 1 then back to 0 */ 7515 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, VCEPLL_SLEEP_MASK, 7516 ~VCEPLL_SLEEP_MASK); 7517 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~VCEPLL_SLEEP_MASK); 7518 7519 /* deassert VCEPLL_RESET */ 7520 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~VCEPLL_RESET_MASK); 7521 7522 mdelay(1); 7523 7524 r = si_vce_send_vcepll_ctlreq(rdev); 7525 if (r) 7526 return r; 7527 7528 /* assert VCEPLL_RESET again */ 7529 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, VCEPLL_RESET_MASK, ~VCEPLL_RESET_MASK); 7530 7531 /* disable spread spectrum. */ 7532 WREG32_SMC_P(CG_VCEPLL_SPREAD_SPECTRUM, 0, ~SSEN_MASK); 7533 7534 /* set feedback divider */ 7535 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL_3, VCEPLL_FB_DIV(fb_div), ~VCEPLL_FB_DIV_MASK); 7536 7537 /* set ref divider to 0 */ 7538 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~VCEPLL_REF_DIV_MASK); 7539 7540 /* set PDIV_A and PDIV_B */ 7541 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL_2, 7542 VCEPLL_PDIV_A(evclk_div) | VCEPLL_PDIV_B(ecclk_div), 7543 ~(VCEPLL_PDIV_A_MASK | VCEPLL_PDIV_B_MASK)); 7544 7545 /* give the PLL some time to settle */ 7546 mdelay(15); 7547 7548 /* deassert PLL_RESET */ 7549 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~VCEPLL_RESET_MASK); 7550 7551 mdelay(15); 7552 7553 /* switch from bypass mode to normal mode */ 7554 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL, 0, ~VCEPLL_BYPASS_EN_MASK); 7555 7556 r = si_vce_send_vcepll_ctlreq(rdev); 7557 if (r) 7558 return r; 7559 7560 /* switch VCLK and DCLK selection */ 7561 WREG32_SMC_P(CG_VCEPLL_FUNC_CNTL_2, 7562 EVCLK_SRC_SEL(16) | ECCLK_SRC_SEL(16), 7563 ~(EVCLK_SRC_SEL_MASK | ECCLK_SRC_SEL_MASK)); 7564 7565 mdelay(100); 7566 7567 return 0; 7568 } 7569