1 /* 2 * Copyright 2010 Advanced Micro Devices, Inc. 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included in 12 * all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 * 22 * Authors: Alex Deucher 23 */ 24 #include <linux/firmware.h> 25 #include <linux/platform_device.h> 26 #include <linux/slab.h> 27 #include <linux/module.h> 28 #include <drm/drmP.h> 29 #include "radeon.h" 30 #include "radeon_asic.h" 31 #include <drm/radeon_drm.h> 32 #include "nid.h" 33 #include "atom.h" 34 #include "ni_reg.h" 35 #include "cayman_blit_shaders.h" 36 37 extern bool evergreen_is_display_hung(struct radeon_device *rdev); 38 extern void evergreen_print_gpu_status_regs(struct radeon_device *rdev); 39 extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save); 40 extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save); 41 extern int evergreen_mc_wait_for_idle(struct radeon_device *rdev); 42 extern void evergreen_mc_program(struct radeon_device *rdev); 43 extern void evergreen_irq_suspend(struct radeon_device *rdev); 44 extern int evergreen_mc_init(struct radeon_device *rdev); 45 extern void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev); 46 extern void evergreen_pcie_gen2_enable(struct radeon_device *rdev); 47 extern void si_rlc_fini(struct radeon_device *rdev); 48 extern int si_rlc_init(struct radeon_device *rdev); 49 50 #define EVERGREEN_PFP_UCODE_SIZE 1120 51 #define EVERGREEN_PM4_UCODE_SIZE 1376 52 #define EVERGREEN_RLC_UCODE_SIZE 768 53 #define BTC_MC_UCODE_SIZE 6024 54 55 #define CAYMAN_PFP_UCODE_SIZE 2176 56 #define CAYMAN_PM4_UCODE_SIZE 2176 57 #define CAYMAN_RLC_UCODE_SIZE 1024 58 #define CAYMAN_MC_UCODE_SIZE 6037 59 60 #define ARUBA_RLC_UCODE_SIZE 1536 61 62 /* Firmware Names */ 63 MODULE_FIRMWARE("radeon/BARTS_pfp.bin"); 64 MODULE_FIRMWARE("radeon/BARTS_me.bin"); 65 MODULE_FIRMWARE("radeon/BARTS_mc.bin"); 66 MODULE_FIRMWARE("radeon/BTC_rlc.bin"); 67 MODULE_FIRMWARE("radeon/TURKS_pfp.bin"); 68 MODULE_FIRMWARE("radeon/TURKS_me.bin"); 69 MODULE_FIRMWARE("radeon/TURKS_mc.bin"); 70 MODULE_FIRMWARE("radeon/CAICOS_pfp.bin"); 71 MODULE_FIRMWARE("radeon/CAICOS_me.bin"); 72 MODULE_FIRMWARE("radeon/CAICOS_mc.bin"); 73 MODULE_FIRMWARE("radeon/CAYMAN_pfp.bin"); 74 MODULE_FIRMWARE("radeon/CAYMAN_me.bin"); 75 MODULE_FIRMWARE("radeon/CAYMAN_mc.bin"); 76 MODULE_FIRMWARE("radeon/CAYMAN_rlc.bin"); 77 MODULE_FIRMWARE("radeon/ARUBA_pfp.bin"); 78 MODULE_FIRMWARE("radeon/ARUBA_me.bin"); 79 MODULE_FIRMWARE("radeon/ARUBA_rlc.bin"); 80 81 82 static const u32 cayman_golden_registers2[] = 83 { 84 0x3e5c, 0xffffffff, 0x00000000, 85 0x3e48, 0xffffffff, 0x00000000, 86 0x3e4c, 0xffffffff, 0x00000000, 87 0x3e64, 0xffffffff, 0x00000000, 88 0x3e50, 0xffffffff, 0x00000000, 89 0x3e60, 0xffffffff, 0x00000000 90 }; 91 92 static const u32 cayman_golden_registers[] = 93 { 94 0x5eb4, 0xffffffff, 0x00000002, 95 0x5e78, 0x8f311ff1, 0x001000f0, 96 0x3f90, 0xffff0000, 0xff000000, 97 0x9148, 0xffff0000, 0xff000000, 98 0x3f94, 0xffff0000, 0xff000000, 99 0x914c, 0xffff0000, 0xff000000, 100 0xc78, 0x00000080, 0x00000080, 101 0xbd4, 0x70073777, 0x00011003, 102 0xd02c, 0xbfffff1f, 0x08421000, 103 0xd0b8, 0x73773777, 0x02011003, 104 0x5bc0, 0x00200000, 0x50100000, 105 0x98f8, 0x33773777, 0x02011003, 106 0x98fc, 0xffffffff, 0x76541032, 107 0x7030, 0x31000311, 0x00000011, 108 0x2f48, 0x33773777, 0x42010001, 109 0x6b28, 0x00000010, 0x00000012, 110 0x7728, 0x00000010, 0x00000012, 111 0x10328, 0x00000010, 0x00000012, 112 0x10f28, 0x00000010, 0x00000012, 113 0x11b28, 0x00000010, 0x00000012, 114 0x12728, 0x00000010, 0x00000012, 115 0x240c, 0x000007ff, 0x00000000, 116 0x8a14, 0xf000001f, 0x00000007, 117 0x8b24, 0x3fff3fff, 0x00ff0fff, 118 0x8b10, 0x0000ff0f, 0x00000000, 119 0x28a4c, 0x07ffffff, 0x06000000, 120 0x10c, 0x00000001, 0x00010003, 121 0xa02c, 0xffffffff, 0x0000009b, 122 0x913c, 0x0000010f, 0x01000100, 123 0x8c04, 0xf8ff00ff, 0x40600060, 124 0x28350, 0x00000f01, 0x00000000, 125 0x9508, 0x3700001f, 0x00000002, 126 0x960c, 0xffffffff, 0x54763210, 127 0x88c4, 0x001f3ae3, 0x00000082, 128 0x88d0, 0xffffffff, 0x0f40df40, 129 0x88d4, 0x0000001f, 0x00000010, 130 0x8974, 0xffffffff, 0x00000000 131 }; 132 133 static const u32 dvst_golden_registers2[] = 134 { 135 0x8f8, 0xffffffff, 0, 136 0x8fc, 0x00380000, 0, 137 0x8f8, 0xffffffff, 1, 138 0x8fc, 0x0e000000, 0 139 }; 140 141 static const u32 dvst_golden_registers[] = 142 { 143 0x690, 0x3fff3fff, 0x20c00033, 144 0x918c, 0x0fff0fff, 0x00010006, 145 0x91a8, 0x0fff0fff, 0x00010006, 146 0x9150, 0xffffdfff, 0x6e944040, 147 0x917c, 0x0fff0fff, 0x00030002, 148 0x9198, 0x0fff0fff, 0x00030002, 149 0x915c, 0x0fff0fff, 0x00010000, 150 0x3f90, 0xffff0001, 0xff000000, 151 0x9178, 0x0fff0fff, 0x00070000, 152 0x9194, 0x0fff0fff, 0x00070000, 153 0x9148, 0xffff0001, 0xff000000, 154 0x9190, 0x0fff0fff, 0x00090008, 155 0x91ac, 0x0fff0fff, 0x00090008, 156 0x3f94, 0xffff0000, 0xff000000, 157 0x914c, 0xffff0000, 0xff000000, 158 0x929c, 0x00000fff, 0x00000001, 159 0x55e4, 0xff607fff, 0xfc000100, 160 0x8a18, 0xff000fff, 0x00000100, 161 0x8b28, 0xff000fff, 0x00000100, 162 0x9144, 0xfffc0fff, 0x00000100, 163 0x6ed8, 0x00010101, 0x00010000, 164 0x9830, 0xffffffff, 0x00000000, 165 0x9834, 0xf00fffff, 0x00000400, 166 0x9838, 0xfffffffe, 0x00000000, 167 0xd0c0, 0xff000fff, 0x00000100, 168 0xd02c, 0xbfffff1f, 0x08421000, 169 0xd0b8, 0x73773777, 0x12010001, 170 0x5bb0, 0x000000f0, 0x00000070, 171 0x98f8, 0x73773777, 0x12010001, 172 0x98fc, 0xffffffff, 0x00000010, 173 0x9b7c, 0x00ff0000, 0x00fc0000, 174 0x8030, 0x00001f0f, 0x0000100a, 175 0x2f48, 0x73773777, 0x12010001, 176 0x2408, 0x00030000, 0x000c007f, 177 0x8a14, 0xf000003f, 0x00000007, 178 0x8b24, 0x3fff3fff, 0x00ff0fff, 179 0x8b10, 0x0000ff0f, 0x00000000, 180 0x28a4c, 0x07ffffff, 0x06000000, 181 0x4d8, 0x00000fff, 0x00000100, 182 0xa008, 0xffffffff, 0x00010000, 183 0x913c, 0xffff03ff, 0x01000100, 184 0x8c00, 0x000000ff, 0x00000003, 185 0x8c04, 0xf8ff00ff, 0x40600060, 186 0x8cf0, 0x1fff1fff, 0x08e00410, 187 0x28350, 0x00000f01, 0x00000000, 188 0x9508, 0xf700071f, 0x00000002, 189 0x960c, 0xffffffff, 0x54763210, 190 0x20ef8, 0x01ff01ff, 0x00000002, 191 0x20e98, 0xfffffbff, 0x00200000, 192 0x2015c, 0xffffffff, 0x00000f40, 193 0x88c4, 0x001f3ae3, 0x00000082, 194 0x8978, 0x3fffffff, 0x04050140, 195 0x88d4, 0x0000001f, 0x00000010, 196 0x8974, 0xffffffff, 0x00000000 197 }; 198 199 static const u32 scrapper_golden_registers[] = 200 { 201 0x690, 0x3fff3fff, 0x20c00033, 202 0x918c, 0x0fff0fff, 0x00010006, 203 0x918c, 0x0fff0fff, 0x00010006, 204 0x91a8, 0x0fff0fff, 0x00010006, 205 0x91a8, 0x0fff0fff, 0x00010006, 206 0x9150, 0xffffdfff, 0x6e944040, 207 0x9150, 0xffffdfff, 0x6e944040, 208 0x917c, 0x0fff0fff, 0x00030002, 209 0x917c, 0x0fff0fff, 0x00030002, 210 0x9198, 0x0fff0fff, 0x00030002, 211 0x9198, 0x0fff0fff, 0x00030002, 212 0x915c, 0x0fff0fff, 0x00010000, 213 0x915c, 0x0fff0fff, 0x00010000, 214 0x3f90, 0xffff0001, 0xff000000, 215 0x3f90, 0xffff0001, 0xff000000, 216 0x9178, 0x0fff0fff, 0x00070000, 217 0x9178, 0x0fff0fff, 0x00070000, 218 0x9194, 0x0fff0fff, 0x00070000, 219 0x9194, 0x0fff0fff, 0x00070000, 220 0x9148, 0xffff0001, 0xff000000, 221 0x9148, 0xffff0001, 0xff000000, 222 0x9190, 0x0fff0fff, 0x00090008, 223 0x9190, 0x0fff0fff, 0x00090008, 224 0x91ac, 0x0fff0fff, 0x00090008, 225 0x91ac, 0x0fff0fff, 0x00090008, 226 0x3f94, 0xffff0000, 0xff000000, 227 0x3f94, 0xffff0000, 0xff000000, 228 0x914c, 0xffff0000, 0xff000000, 229 0x914c, 0xffff0000, 0xff000000, 230 0x929c, 0x00000fff, 0x00000001, 231 0x929c, 0x00000fff, 0x00000001, 232 0x55e4, 0xff607fff, 0xfc000100, 233 0x8a18, 0xff000fff, 0x00000100, 234 0x8a18, 0xff000fff, 0x00000100, 235 0x8b28, 0xff000fff, 0x00000100, 236 0x8b28, 0xff000fff, 0x00000100, 237 0x9144, 0xfffc0fff, 0x00000100, 238 0x9144, 0xfffc0fff, 0x00000100, 239 0x6ed8, 0x00010101, 0x00010000, 240 0x9830, 0xffffffff, 0x00000000, 241 0x9830, 0xffffffff, 0x00000000, 242 0x9834, 0xf00fffff, 0x00000400, 243 0x9834, 0xf00fffff, 0x00000400, 244 0x9838, 0xfffffffe, 0x00000000, 245 0x9838, 0xfffffffe, 0x00000000, 246 0xd0c0, 0xff000fff, 0x00000100, 247 0xd02c, 0xbfffff1f, 0x08421000, 248 0xd02c, 0xbfffff1f, 0x08421000, 249 0xd0b8, 0x73773777, 0x12010001, 250 0xd0b8, 0x73773777, 0x12010001, 251 0x5bb0, 0x000000f0, 0x00000070, 252 0x98f8, 0x73773777, 0x12010001, 253 0x98f8, 0x73773777, 0x12010001, 254 0x98fc, 0xffffffff, 0x00000010, 255 0x98fc, 0xffffffff, 0x00000010, 256 0x9b7c, 0x00ff0000, 0x00fc0000, 257 0x9b7c, 0x00ff0000, 0x00fc0000, 258 0x8030, 0x00001f0f, 0x0000100a, 259 0x8030, 0x00001f0f, 0x0000100a, 260 0x2f48, 0x73773777, 0x12010001, 261 0x2f48, 0x73773777, 0x12010001, 262 0x2408, 0x00030000, 0x000c007f, 263 0x8a14, 0xf000003f, 0x00000007, 264 0x8a14, 0xf000003f, 0x00000007, 265 0x8b24, 0x3fff3fff, 0x00ff0fff, 266 0x8b24, 0x3fff3fff, 0x00ff0fff, 267 0x8b10, 0x0000ff0f, 0x00000000, 268 0x8b10, 0x0000ff0f, 0x00000000, 269 0x28a4c, 0x07ffffff, 0x06000000, 270 0x28a4c, 0x07ffffff, 0x06000000, 271 0x4d8, 0x00000fff, 0x00000100, 272 0x4d8, 0x00000fff, 0x00000100, 273 0xa008, 0xffffffff, 0x00010000, 274 0xa008, 0xffffffff, 0x00010000, 275 0x913c, 0xffff03ff, 0x01000100, 276 0x913c, 0xffff03ff, 0x01000100, 277 0x90e8, 0x001fffff, 0x010400c0, 278 0x8c00, 0x000000ff, 0x00000003, 279 0x8c00, 0x000000ff, 0x00000003, 280 0x8c04, 0xf8ff00ff, 0x40600060, 281 0x8c04, 0xf8ff00ff, 0x40600060, 282 0x8c30, 0x0000000f, 0x00040005, 283 0x8cf0, 0x1fff1fff, 0x08e00410, 284 0x8cf0, 0x1fff1fff, 0x08e00410, 285 0x900c, 0x00ffffff, 0x0017071f, 286 0x28350, 0x00000f01, 0x00000000, 287 0x28350, 0x00000f01, 0x00000000, 288 0x9508, 0xf700071f, 0x00000002, 289 0x9508, 0xf700071f, 0x00000002, 290 0x9688, 0x00300000, 0x0017000f, 291 0x960c, 0xffffffff, 0x54763210, 292 0x960c, 0xffffffff, 0x54763210, 293 0x20ef8, 0x01ff01ff, 0x00000002, 294 0x20e98, 0xfffffbff, 0x00200000, 295 0x2015c, 0xffffffff, 0x00000f40, 296 0x88c4, 0x001f3ae3, 0x00000082, 297 0x88c4, 0x001f3ae3, 0x00000082, 298 0x8978, 0x3fffffff, 0x04050140, 299 0x8978, 0x3fffffff, 0x04050140, 300 0x88d4, 0x0000001f, 0x00000010, 301 0x88d4, 0x0000001f, 0x00000010, 302 0x8974, 0xffffffff, 0x00000000, 303 0x8974, 0xffffffff, 0x00000000 304 }; 305 306 static void ni_init_golden_registers(struct radeon_device *rdev) 307 { 308 switch (rdev->family) { 309 case CHIP_CAYMAN: 310 radeon_program_register_sequence(rdev, 311 cayman_golden_registers, 312 (const u32)ARRAY_SIZE(cayman_golden_registers)); 313 radeon_program_register_sequence(rdev, 314 cayman_golden_registers2, 315 (const u32)ARRAY_SIZE(cayman_golden_registers2)); 316 break; 317 case CHIP_ARUBA: 318 if ((rdev->pdev->device == 0x9900) || 319 (rdev->pdev->device == 0x9901) || 320 (rdev->pdev->device == 0x9903) || 321 (rdev->pdev->device == 0x9904) || 322 (rdev->pdev->device == 0x9905) || 323 (rdev->pdev->device == 0x9906) || 324 (rdev->pdev->device == 0x9907) || 325 (rdev->pdev->device == 0x9908) || 326 (rdev->pdev->device == 0x9909) || 327 (rdev->pdev->device == 0x990A) || 328 (rdev->pdev->device == 0x990B) || 329 (rdev->pdev->device == 0x990C) || 330 (rdev->pdev->device == 0x990D) || 331 (rdev->pdev->device == 0x990E) || 332 (rdev->pdev->device == 0x990F) || 333 (rdev->pdev->device == 0x9910) || 334 (rdev->pdev->device == 0x9913) || 335 (rdev->pdev->device == 0x9917) || 336 (rdev->pdev->device == 0x9918)) { 337 radeon_program_register_sequence(rdev, 338 dvst_golden_registers, 339 (const u32)ARRAY_SIZE(dvst_golden_registers)); 340 radeon_program_register_sequence(rdev, 341 dvst_golden_registers2, 342 (const u32)ARRAY_SIZE(dvst_golden_registers2)); 343 } else { 344 radeon_program_register_sequence(rdev, 345 scrapper_golden_registers, 346 (const u32)ARRAY_SIZE(scrapper_golden_registers)); 347 radeon_program_register_sequence(rdev, 348 dvst_golden_registers2, 349 (const u32)ARRAY_SIZE(dvst_golden_registers2)); 350 } 351 break; 352 default: 353 break; 354 } 355 } 356 357 #define BTC_IO_MC_REGS_SIZE 29 358 359 static const u32 barts_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = { 360 {0x00000077, 0xff010100}, 361 {0x00000078, 0x00000000}, 362 {0x00000079, 0x00001434}, 363 {0x0000007a, 0xcc08ec08}, 364 {0x0000007b, 0x00040000}, 365 {0x0000007c, 0x000080c0}, 366 {0x0000007d, 0x09000000}, 367 {0x0000007e, 0x00210404}, 368 {0x00000081, 0x08a8e800}, 369 {0x00000082, 0x00030444}, 370 {0x00000083, 0x00000000}, 371 {0x00000085, 0x00000001}, 372 {0x00000086, 0x00000002}, 373 {0x00000087, 0x48490000}, 374 {0x00000088, 0x20244647}, 375 {0x00000089, 0x00000005}, 376 {0x0000008b, 0x66030000}, 377 {0x0000008c, 0x00006603}, 378 {0x0000008d, 0x00000100}, 379 {0x0000008f, 0x00001c0a}, 380 {0x00000090, 0xff000001}, 381 {0x00000094, 0x00101101}, 382 {0x00000095, 0x00000fff}, 383 {0x00000096, 0x00116fff}, 384 {0x00000097, 0x60010000}, 385 {0x00000098, 0x10010000}, 386 {0x00000099, 0x00006000}, 387 {0x0000009a, 0x00001000}, 388 {0x0000009f, 0x00946a00} 389 }; 390 391 static const u32 turks_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = { 392 {0x00000077, 0xff010100}, 393 {0x00000078, 0x00000000}, 394 {0x00000079, 0x00001434}, 395 {0x0000007a, 0xcc08ec08}, 396 {0x0000007b, 0x00040000}, 397 {0x0000007c, 0x000080c0}, 398 {0x0000007d, 0x09000000}, 399 {0x0000007e, 0x00210404}, 400 {0x00000081, 0x08a8e800}, 401 {0x00000082, 0x00030444}, 402 {0x00000083, 0x00000000}, 403 {0x00000085, 0x00000001}, 404 {0x00000086, 0x00000002}, 405 {0x00000087, 0x48490000}, 406 {0x00000088, 0x20244647}, 407 {0x00000089, 0x00000005}, 408 {0x0000008b, 0x66030000}, 409 {0x0000008c, 0x00006603}, 410 {0x0000008d, 0x00000100}, 411 {0x0000008f, 0x00001c0a}, 412 {0x00000090, 0xff000001}, 413 {0x00000094, 0x00101101}, 414 {0x00000095, 0x00000fff}, 415 {0x00000096, 0x00116fff}, 416 {0x00000097, 0x60010000}, 417 {0x00000098, 0x10010000}, 418 {0x00000099, 0x00006000}, 419 {0x0000009a, 0x00001000}, 420 {0x0000009f, 0x00936a00} 421 }; 422 423 static const u32 caicos_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = { 424 {0x00000077, 0xff010100}, 425 {0x00000078, 0x00000000}, 426 {0x00000079, 0x00001434}, 427 {0x0000007a, 0xcc08ec08}, 428 {0x0000007b, 0x00040000}, 429 {0x0000007c, 0x000080c0}, 430 {0x0000007d, 0x09000000}, 431 {0x0000007e, 0x00210404}, 432 {0x00000081, 0x08a8e800}, 433 {0x00000082, 0x00030444}, 434 {0x00000083, 0x00000000}, 435 {0x00000085, 0x00000001}, 436 {0x00000086, 0x00000002}, 437 {0x00000087, 0x48490000}, 438 {0x00000088, 0x20244647}, 439 {0x00000089, 0x00000005}, 440 {0x0000008b, 0x66030000}, 441 {0x0000008c, 0x00006603}, 442 {0x0000008d, 0x00000100}, 443 {0x0000008f, 0x00001c0a}, 444 {0x00000090, 0xff000001}, 445 {0x00000094, 0x00101101}, 446 {0x00000095, 0x00000fff}, 447 {0x00000096, 0x00116fff}, 448 {0x00000097, 0x60010000}, 449 {0x00000098, 0x10010000}, 450 {0x00000099, 0x00006000}, 451 {0x0000009a, 0x00001000}, 452 {0x0000009f, 0x00916a00} 453 }; 454 455 static const u32 cayman_io_mc_regs[BTC_IO_MC_REGS_SIZE][2] = { 456 {0x00000077, 0xff010100}, 457 {0x00000078, 0x00000000}, 458 {0x00000079, 0x00001434}, 459 {0x0000007a, 0xcc08ec08}, 460 {0x0000007b, 0x00040000}, 461 {0x0000007c, 0x000080c0}, 462 {0x0000007d, 0x09000000}, 463 {0x0000007e, 0x00210404}, 464 {0x00000081, 0x08a8e800}, 465 {0x00000082, 0x00030444}, 466 {0x00000083, 0x00000000}, 467 {0x00000085, 0x00000001}, 468 {0x00000086, 0x00000002}, 469 {0x00000087, 0x48490000}, 470 {0x00000088, 0x20244647}, 471 {0x00000089, 0x00000005}, 472 {0x0000008b, 0x66030000}, 473 {0x0000008c, 0x00006603}, 474 {0x0000008d, 0x00000100}, 475 {0x0000008f, 0x00001c0a}, 476 {0x00000090, 0xff000001}, 477 {0x00000094, 0x00101101}, 478 {0x00000095, 0x00000fff}, 479 {0x00000096, 0x00116fff}, 480 {0x00000097, 0x60010000}, 481 {0x00000098, 0x10010000}, 482 {0x00000099, 0x00006000}, 483 {0x0000009a, 0x00001000}, 484 {0x0000009f, 0x00976b00} 485 }; 486 487 int ni_mc_load_microcode(struct radeon_device *rdev) 488 { 489 const __be32 *fw_data; 490 u32 mem_type, running, blackout = 0; 491 u32 *io_mc_regs; 492 int i, ucode_size, regs_size; 493 494 if (!rdev->mc_fw) 495 return -EINVAL; 496 497 switch (rdev->family) { 498 case CHIP_BARTS: 499 io_mc_regs = (u32 *)&barts_io_mc_regs; 500 ucode_size = BTC_MC_UCODE_SIZE; 501 regs_size = BTC_IO_MC_REGS_SIZE; 502 break; 503 case CHIP_TURKS: 504 io_mc_regs = (u32 *)&turks_io_mc_regs; 505 ucode_size = BTC_MC_UCODE_SIZE; 506 regs_size = BTC_IO_MC_REGS_SIZE; 507 break; 508 case CHIP_CAICOS: 509 default: 510 io_mc_regs = (u32 *)&caicos_io_mc_regs; 511 ucode_size = BTC_MC_UCODE_SIZE; 512 regs_size = BTC_IO_MC_REGS_SIZE; 513 break; 514 case CHIP_CAYMAN: 515 io_mc_regs = (u32 *)&cayman_io_mc_regs; 516 ucode_size = CAYMAN_MC_UCODE_SIZE; 517 regs_size = BTC_IO_MC_REGS_SIZE; 518 break; 519 } 520 521 mem_type = (RREG32(MC_SEQ_MISC0) & MC_SEQ_MISC0_GDDR5_MASK) >> MC_SEQ_MISC0_GDDR5_SHIFT; 522 running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK; 523 524 if ((mem_type == MC_SEQ_MISC0_GDDR5_VALUE) && (running == 0)) { 525 if (running) { 526 blackout = RREG32(MC_SHARED_BLACKOUT_CNTL); 527 WREG32(MC_SHARED_BLACKOUT_CNTL, 1); 528 } 529 530 /* reset the engine and set to writable */ 531 WREG32(MC_SEQ_SUP_CNTL, 0x00000008); 532 WREG32(MC_SEQ_SUP_CNTL, 0x00000010); 533 534 /* load mc io regs */ 535 for (i = 0; i < regs_size; i++) { 536 WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]); 537 WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]); 538 } 539 /* load the MC ucode */ 540 fw_data = (const __be32 *)rdev->mc_fw->data; 541 for (i = 0; i < ucode_size; i++) 542 WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++)); 543 544 /* put the engine back into the active state */ 545 WREG32(MC_SEQ_SUP_CNTL, 0x00000008); 546 WREG32(MC_SEQ_SUP_CNTL, 0x00000004); 547 WREG32(MC_SEQ_SUP_CNTL, 0x00000001); 548 549 /* wait for training to complete */ 550 for (i = 0; i < rdev->usec_timeout; i++) { 551 if (RREG32(MC_IO_PAD_CNTL_D0) & MEM_FALL_OUT_CMD) 552 break; 553 udelay(1); 554 } 555 556 if (running) 557 WREG32(MC_SHARED_BLACKOUT_CNTL, blackout); 558 } 559 560 return 0; 561 } 562 563 int ni_init_microcode(struct radeon_device *rdev) 564 { 565 struct platform_device *pdev; 566 const char *chip_name; 567 const char *rlc_chip_name; 568 size_t pfp_req_size, me_req_size, rlc_req_size, mc_req_size; 569 char fw_name[30]; 570 int err; 571 572 DRM_DEBUG("\n"); 573 574 pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0); 575 err = IS_ERR(pdev); 576 if (err) { 577 printk(KERN_ERR "radeon_cp: Failed to register firmware\n"); 578 return -EINVAL; 579 } 580 581 switch (rdev->family) { 582 case CHIP_BARTS: 583 chip_name = "BARTS"; 584 rlc_chip_name = "BTC"; 585 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4; 586 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4; 587 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4; 588 mc_req_size = BTC_MC_UCODE_SIZE * 4; 589 break; 590 case CHIP_TURKS: 591 chip_name = "TURKS"; 592 rlc_chip_name = "BTC"; 593 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4; 594 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4; 595 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4; 596 mc_req_size = BTC_MC_UCODE_SIZE * 4; 597 break; 598 case CHIP_CAICOS: 599 chip_name = "CAICOS"; 600 rlc_chip_name = "BTC"; 601 pfp_req_size = EVERGREEN_PFP_UCODE_SIZE * 4; 602 me_req_size = EVERGREEN_PM4_UCODE_SIZE * 4; 603 rlc_req_size = EVERGREEN_RLC_UCODE_SIZE * 4; 604 mc_req_size = BTC_MC_UCODE_SIZE * 4; 605 break; 606 case CHIP_CAYMAN: 607 chip_name = "CAYMAN"; 608 rlc_chip_name = "CAYMAN"; 609 pfp_req_size = CAYMAN_PFP_UCODE_SIZE * 4; 610 me_req_size = CAYMAN_PM4_UCODE_SIZE * 4; 611 rlc_req_size = CAYMAN_RLC_UCODE_SIZE * 4; 612 mc_req_size = CAYMAN_MC_UCODE_SIZE * 4; 613 break; 614 case CHIP_ARUBA: 615 chip_name = "ARUBA"; 616 rlc_chip_name = "ARUBA"; 617 /* pfp/me same size as CAYMAN */ 618 pfp_req_size = CAYMAN_PFP_UCODE_SIZE * 4; 619 me_req_size = CAYMAN_PM4_UCODE_SIZE * 4; 620 rlc_req_size = ARUBA_RLC_UCODE_SIZE * 4; 621 mc_req_size = 0; 622 break; 623 default: BUG(); 624 } 625 626 DRM_INFO("Loading %s Microcode\n", chip_name); 627 628 snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name); 629 err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev); 630 if (err) 631 goto out; 632 if (rdev->pfp_fw->size != pfp_req_size) { 633 printk(KERN_ERR 634 "ni_cp: Bogus length %zu in firmware \"%s\"\n", 635 rdev->pfp_fw->size, fw_name); 636 err = -EINVAL; 637 goto out; 638 } 639 640 snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name); 641 err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev); 642 if (err) 643 goto out; 644 if (rdev->me_fw->size != me_req_size) { 645 printk(KERN_ERR 646 "ni_cp: Bogus length %zu in firmware \"%s\"\n", 647 rdev->me_fw->size, fw_name); 648 err = -EINVAL; 649 } 650 651 snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name); 652 err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev); 653 if (err) 654 goto out; 655 if (rdev->rlc_fw->size != rlc_req_size) { 656 printk(KERN_ERR 657 "ni_rlc: Bogus length %zu in firmware \"%s\"\n", 658 rdev->rlc_fw->size, fw_name); 659 err = -EINVAL; 660 } 661 662 /* no MC ucode on TN */ 663 if (!(rdev->flags & RADEON_IS_IGP)) { 664 snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name); 665 err = request_firmware(&rdev->mc_fw, fw_name, &pdev->dev); 666 if (err) 667 goto out; 668 if (rdev->mc_fw->size != mc_req_size) { 669 printk(KERN_ERR 670 "ni_mc: Bogus length %zu in firmware \"%s\"\n", 671 rdev->mc_fw->size, fw_name); 672 err = -EINVAL; 673 } 674 } 675 out: 676 platform_device_unregister(pdev); 677 678 if (err) { 679 if (err != -EINVAL) 680 printk(KERN_ERR 681 "ni_cp: Failed to load firmware \"%s\"\n", 682 fw_name); 683 release_firmware(rdev->pfp_fw); 684 rdev->pfp_fw = NULL; 685 release_firmware(rdev->me_fw); 686 rdev->me_fw = NULL; 687 release_firmware(rdev->rlc_fw); 688 rdev->rlc_fw = NULL; 689 release_firmware(rdev->mc_fw); 690 rdev->mc_fw = NULL; 691 } 692 return err; 693 } 694 695 /* 696 * Core functions 697 */ 698 static void cayman_gpu_init(struct radeon_device *rdev) 699 { 700 u32 gb_addr_config = 0; 701 u32 mc_shared_chmap, mc_arb_ramcfg; 702 u32 cgts_tcc_disable; 703 u32 sx_debug_1; 704 u32 smx_dc_ctl0; 705 u32 cgts_sm_ctrl_reg; 706 u32 hdp_host_path_cntl; 707 u32 tmp; 708 u32 disabled_rb_mask; 709 int i, j; 710 711 switch (rdev->family) { 712 case CHIP_CAYMAN: 713 rdev->config.cayman.max_shader_engines = 2; 714 rdev->config.cayman.max_pipes_per_simd = 4; 715 rdev->config.cayman.max_tile_pipes = 8; 716 rdev->config.cayman.max_simds_per_se = 12; 717 rdev->config.cayman.max_backends_per_se = 4; 718 rdev->config.cayman.max_texture_channel_caches = 8; 719 rdev->config.cayman.max_gprs = 256; 720 rdev->config.cayman.max_threads = 256; 721 rdev->config.cayman.max_gs_threads = 32; 722 rdev->config.cayman.max_stack_entries = 512; 723 rdev->config.cayman.sx_num_of_sets = 8; 724 rdev->config.cayman.sx_max_export_size = 256; 725 rdev->config.cayman.sx_max_export_pos_size = 64; 726 rdev->config.cayman.sx_max_export_smx_size = 192; 727 rdev->config.cayman.max_hw_contexts = 8; 728 rdev->config.cayman.sq_num_cf_insts = 2; 729 730 rdev->config.cayman.sc_prim_fifo_size = 0x100; 731 rdev->config.cayman.sc_hiz_tile_fifo_size = 0x30; 732 rdev->config.cayman.sc_earlyz_tile_fifo_size = 0x130; 733 gb_addr_config = CAYMAN_GB_ADDR_CONFIG_GOLDEN; 734 break; 735 case CHIP_ARUBA: 736 default: 737 rdev->config.cayman.max_shader_engines = 1; 738 rdev->config.cayman.max_pipes_per_simd = 4; 739 rdev->config.cayman.max_tile_pipes = 2; 740 if ((rdev->pdev->device == 0x9900) || 741 (rdev->pdev->device == 0x9901) || 742 (rdev->pdev->device == 0x9905) || 743 (rdev->pdev->device == 0x9906) || 744 (rdev->pdev->device == 0x9907) || 745 (rdev->pdev->device == 0x9908) || 746 (rdev->pdev->device == 0x9909) || 747 (rdev->pdev->device == 0x990B) || 748 (rdev->pdev->device == 0x990C) || 749 (rdev->pdev->device == 0x990F) || 750 (rdev->pdev->device == 0x9910) || 751 (rdev->pdev->device == 0x9917) || 752 (rdev->pdev->device == 0x9999) || 753 (rdev->pdev->device == 0x999C)) { 754 rdev->config.cayman.max_simds_per_se = 6; 755 rdev->config.cayman.max_backends_per_se = 2; 756 } else if ((rdev->pdev->device == 0x9903) || 757 (rdev->pdev->device == 0x9904) || 758 (rdev->pdev->device == 0x990A) || 759 (rdev->pdev->device == 0x990D) || 760 (rdev->pdev->device == 0x990E) || 761 (rdev->pdev->device == 0x9913) || 762 (rdev->pdev->device == 0x9918) || 763 (rdev->pdev->device == 0x999D)) { 764 rdev->config.cayman.max_simds_per_se = 4; 765 rdev->config.cayman.max_backends_per_se = 2; 766 } else if ((rdev->pdev->device == 0x9919) || 767 (rdev->pdev->device == 0x9990) || 768 (rdev->pdev->device == 0x9991) || 769 (rdev->pdev->device == 0x9994) || 770 (rdev->pdev->device == 0x9995) || 771 (rdev->pdev->device == 0x9996) || 772 (rdev->pdev->device == 0x999A) || 773 (rdev->pdev->device == 0x99A0)) { 774 rdev->config.cayman.max_simds_per_se = 3; 775 rdev->config.cayman.max_backends_per_se = 1; 776 } else { 777 rdev->config.cayman.max_simds_per_se = 2; 778 rdev->config.cayman.max_backends_per_se = 1; 779 } 780 rdev->config.cayman.max_texture_channel_caches = 2; 781 rdev->config.cayman.max_gprs = 256; 782 rdev->config.cayman.max_threads = 256; 783 rdev->config.cayman.max_gs_threads = 32; 784 rdev->config.cayman.max_stack_entries = 512; 785 rdev->config.cayman.sx_num_of_sets = 8; 786 rdev->config.cayman.sx_max_export_size = 256; 787 rdev->config.cayman.sx_max_export_pos_size = 64; 788 rdev->config.cayman.sx_max_export_smx_size = 192; 789 rdev->config.cayman.max_hw_contexts = 8; 790 rdev->config.cayman.sq_num_cf_insts = 2; 791 792 rdev->config.cayman.sc_prim_fifo_size = 0x40; 793 rdev->config.cayman.sc_hiz_tile_fifo_size = 0x30; 794 rdev->config.cayman.sc_earlyz_tile_fifo_size = 0x130; 795 gb_addr_config = ARUBA_GB_ADDR_CONFIG_GOLDEN; 796 break; 797 } 798 799 /* Initialize HDP */ 800 for (i = 0, j = 0; i < 32; i++, j += 0x18) { 801 WREG32((0x2c14 + j), 0x00000000); 802 WREG32((0x2c18 + j), 0x00000000); 803 WREG32((0x2c1c + j), 0x00000000); 804 WREG32((0x2c20 + j), 0x00000000); 805 WREG32((0x2c24 + j), 0x00000000); 806 } 807 808 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff)); 809 810 evergreen_fix_pci_max_read_req_size(rdev); 811 812 mc_shared_chmap = RREG32(MC_SHARED_CHMAP); 813 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG); 814 815 tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT; 816 rdev->config.cayman.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024; 817 if (rdev->config.cayman.mem_row_size_in_kb > 4) 818 rdev->config.cayman.mem_row_size_in_kb = 4; 819 /* XXX use MC settings? */ 820 rdev->config.cayman.shader_engine_tile_size = 32; 821 rdev->config.cayman.num_gpus = 1; 822 rdev->config.cayman.multi_gpu_tile_size = 64; 823 824 tmp = (gb_addr_config & NUM_PIPES_MASK) >> NUM_PIPES_SHIFT; 825 rdev->config.cayman.num_tile_pipes = (1 << tmp); 826 tmp = (gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT; 827 rdev->config.cayman.mem_max_burst_length_bytes = (tmp + 1) * 256; 828 tmp = (gb_addr_config & NUM_SHADER_ENGINES_MASK) >> NUM_SHADER_ENGINES_SHIFT; 829 rdev->config.cayman.num_shader_engines = tmp + 1; 830 tmp = (gb_addr_config & NUM_GPUS_MASK) >> NUM_GPUS_SHIFT; 831 rdev->config.cayman.num_gpus = tmp + 1; 832 tmp = (gb_addr_config & MULTI_GPU_TILE_SIZE_MASK) >> MULTI_GPU_TILE_SIZE_SHIFT; 833 rdev->config.cayman.multi_gpu_tile_size = 1 << tmp; 834 tmp = (gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT; 835 rdev->config.cayman.mem_row_size_in_kb = 1 << tmp; 836 837 838 /* setup tiling info dword. gb_addr_config is not adequate since it does 839 * not have bank info, so create a custom tiling dword. 840 * bits 3:0 num_pipes 841 * bits 7:4 num_banks 842 * bits 11:8 group_size 843 * bits 15:12 row_size 844 */ 845 rdev->config.cayman.tile_config = 0; 846 switch (rdev->config.cayman.num_tile_pipes) { 847 case 1: 848 default: 849 rdev->config.cayman.tile_config |= (0 << 0); 850 break; 851 case 2: 852 rdev->config.cayman.tile_config |= (1 << 0); 853 break; 854 case 4: 855 rdev->config.cayman.tile_config |= (2 << 0); 856 break; 857 case 8: 858 rdev->config.cayman.tile_config |= (3 << 0); 859 break; 860 } 861 862 /* num banks is 8 on all fusion asics. 0 = 4, 1 = 8, 2 = 16 */ 863 if (rdev->flags & RADEON_IS_IGP) 864 rdev->config.cayman.tile_config |= 1 << 4; 865 else { 866 switch ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) { 867 case 0: /* four banks */ 868 rdev->config.cayman.tile_config |= 0 << 4; 869 break; 870 case 1: /* eight banks */ 871 rdev->config.cayman.tile_config |= 1 << 4; 872 break; 873 case 2: /* sixteen banks */ 874 default: 875 rdev->config.cayman.tile_config |= 2 << 4; 876 break; 877 } 878 } 879 rdev->config.cayman.tile_config |= 880 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8; 881 rdev->config.cayman.tile_config |= 882 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12; 883 884 tmp = 0; 885 for (i = (rdev->config.cayman.max_shader_engines - 1); i >= 0; i--) { 886 u32 rb_disable_bitmap; 887 888 WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i)); 889 WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_INDEX(i)); 890 rb_disable_bitmap = (RREG32(CC_RB_BACKEND_DISABLE) & 0x00ff0000) >> 16; 891 tmp <<= 4; 892 tmp |= rb_disable_bitmap; 893 } 894 /* enabled rb are just the one not disabled :) */ 895 disabled_rb_mask = tmp; 896 tmp = 0; 897 for (i = 0; i < (rdev->config.cayman.max_backends_per_se * rdev->config.cayman.max_shader_engines); i++) 898 tmp |= (1 << i); 899 /* if all the backends are disabled, fix it up here */ 900 if ((disabled_rb_mask & tmp) == tmp) { 901 for (i = 0; i < (rdev->config.cayman.max_backends_per_se * rdev->config.cayman.max_shader_engines); i++) 902 disabled_rb_mask &= ~(1 << i); 903 } 904 905 WREG32(GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES); 906 WREG32(RLC_GFX_INDEX, INSTANCE_BROADCAST_WRITES | SE_BROADCAST_WRITES); 907 908 WREG32(GB_ADDR_CONFIG, gb_addr_config); 909 WREG32(DMIF_ADDR_CONFIG, gb_addr_config); 910 if (ASIC_IS_DCE6(rdev)) 911 WREG32(DMIF_ADDR_CALC, gb_addr_config); 912 WREG32(HDP_ADDR_CONFIG, gb_addr_config); 913 WREG32(DMA_TILING_CONFIG + DMA0_REGISTER_OFFSET, gb_addr_config); 914 WREG32(DMA_TILING_CONFIG + DMA1_REGISTER_OFFSET, gb_addr_config); 915 WREG32(UVD_UDEC_ADDR_CONFIG, gb_addr_config); 916 WREG32(UVD_UDEC_DB_ADDR_CONFIG, gb_addr_config); 917 WREG32(UVD_UDEC_DBW_ADDR_CONFIG, gb_addr_config); 918 919 if ((rdev->config.cayman.max_backends_per_se == 1) && 920 (rdev->flags & RADEON_IS_IGP)) { 921 if ((disabled_rb_mask & 3) == 1) { 922 /* RB0 disabled, RB1 enabled */ 923 tmp = 0x11111111; 924 } else { 925 /* RB1 disabled, RB0 enabled */ 926 tmp = 0x00000000; 927 } 928 } else { 929 tmp = gb_addr_config & NUM_PIPES_MASK; 930 tmp = r6xx_remap_render_backend(rdev, tmp, 931 rdev->config.cayman.max_backends_per_se * 932 rdev->config.cayman.max_shader_engines, 933 CAYMAN_MAX_BACKENDS, disabled_rb_mask); 934 } 935 WREG32(GB_BACKEND_MAP, tmp); 936 937 cgts_tcc_disable = 0xffff0000; 938 for (i = 0; i < rdev->config.cayman.max_texture_channel_caches; i++) 939 cgts_tcc_disable &= ~(1 << (16 + i)); 940 WREG32(CGTS_TCC_DISABLE, cgts_tcc_disable); 941 WREG32(CGTS_SYS_TCC_DISABLE, cgts_tcc_disable); 942 WREG32(CGTS_USER_SYS_TCC_DISABLE, cgts_tcc_disable); 943 WREG32(CGTS_USER_TCC_DISABLE, cgts_tcc_disable); 944 945 /* reprogram the shader complex */ 946 cgts_sm_ctrl_reg = RREG32(CGTS_SM_CTRL_REG); 947 for (i = 0; i < 16; i++) 948 WREG32(CGTS_SM_CTRL_REG, OVERRIDE); 949 WREG32(CGTS_SM_CTRL_REG, cgts_sm_ctrl_reg); 950 951 /* set HW defaults for 3D engine */ 952 WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60)); 953 954 sx_debug_1 = RREG32(SX_DEBUG_1); 955 sx_debug_1 |= ENABLE_NEW_SMX_ADDRESS; 956 WREG32(SX_DEBUG_1, sx_debug_1); 957 958 smx_dc_ctl0 = RREG32(SMX_DC_CTL0); 959 smx_dc_ctl0 &= ~NUMBER_OF_SETS(0x1ff); 960 smx_dc_ctl0 |= NUMBER_OF_SETS(rdev->config.cayman.sx_num_of_sets); 961 WREG32(SMX_DC_CTL0, smx_dc_ctl0); 962 963 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4) | CRC_SIMD_ID_WADDR_DISABLE); 964 965 /* need to be explicitly zero-ed */ 966 WREG32(VGT_OFFCHIP_LDS_BASE, 0); 967 WREG32(SQ_LSTMP_RING_BASE, 0); 968 WREG32(SQ_HSTMP_RING_BASE, 0); 969 WREG32(SQ_ESTMP_RING_BASE, 0); 970 WREG32(SQ_GSTMP_RING_BASE, 0); 971 WREG32(SQ_VSTMP_RING_BASE, 0); 972 WREG32(SQ_PSTMP_RING_BASE, 0); 973 974 WREG32(TA_CNTL_AUX, DISABLE_CUBE_ANISO); 975 976 WREG32(SX_EXPORT_BUFFER_SIZES, (COLOR_BUFFER_SIZE((rdev->config.cayman.sx_max_export_size / 4) - 1) | 977 POSITION_BUFFER_SIZE((rdev->config.cayman.sx_max_export_pos_size / 4) - 1) | 978 SMX_BUFFER_SIZE((rdev->config.cayman.sx_max_export_smx_size / 4) - 1))); 979 980 WREG32(PA_SC_FIFO_SIZE, (SC_PRIM_FIFO_SIZE(rdev->config.cayman.sc_prim_fifo_size) | 981 SC_HIZ_TILE_FIFO_SIZE(rdev->config.cayman.sc_hiz_tile_fifo_size) | 982 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.cayman.sc_earlyz_tile_fifo_size))); 983 984 985 WREG32(VGT_NUM_INSTANCES, 1); 986 987 WREG32(CP_PERFMON_CNTL, 0); 988 989 WREG32(SQ_MS_FIFO_SIZES, (CACHE_FIFO_SIZE(16 * rdev->config.cayman.sq_num_cf_insts) | 990 FETCH_FIFO_HIWATER(0x4) | 991 DONE_FIFO_HIWATER(0xe0) | 992 ALU_UPDATE_FIFO_HIWATER(0x8))); 993 994 WREG32(SQ_GPR_RESOURCE_MGMT_1, NUM_CLAUSE_TEMP_GPRS(4)); 995 WREG32(SQ_CONFIG, (VC_ENABLE | 996 EXPORT_SRC_C | 997 GFX_PRIO(0) | 998 CS1_PRIO(0) | 999 CS2_PRIO(1))); 1000 WREG32(SQ_DYN_GPR_CNTL_PS_FLUSH_REQ, DYN_GPR_ENABLE); 1001 1002 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) | 1003 FORCE_EOV_MAX_REZ_CNT(255))); 1004 1005 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) | 1006 AUTO_INVLD_EN(ES_AND_GS_AUTO)); 1007 1008 WREG32(VGT_GS_VERTEX_REUSE, 16); 1009 WREG32(PA_SC_LINE_STIPPLE_STATE, 0); 1010 1011 WREG32(CB_PERF_CTR0_SEL_0, 0); 1012 WREG32(CB_PERF_CTR0_SEL_1, 0); 1013 WREG32(CB_PERF_CTR1_SEL_0, 0); 1014 WREG32(CB_PERF_CTR1_SEL_1, 0); 1015 WREG32(CB_PERF_CTR2_SEL_0, 0); 1016 WREG32(CB_PERF_CTR2_SEL_1, 0); 1017 WREG32(CB_PERF_CTR3_SEL_0, 0); 1018 WREG32(CB_PERF_CTR3_SEL_1, 0); 1019 1020 tmp = RREG32(HDP_MISC_CNTL); 1021 tmp |= HDP_FLUSH_INVALIDATE_CACHE; 1022 WREG32(HDP_MISC_CNTL, tmp); 1023 1024 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL); 1025 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl); 1026 1027 WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3)); 1028 1029 udelay(50); 1030 } 1031 1032 /* 1033 * GART 1034 */ 1035 void cayman_pcie_gart_tlb_flush(struct radeon_device *rdev) 1036 { 1037 /* flush hdp cache */ 1038 WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1); 1039 1040 /* bits 0-7 are the VM contexts0-7 */ 1041 WREG32(VM_INVALIDATE_REQUEST, 1); 1042 } 1043 1044 static int cayman_pcie_gart_enable(struct radeon_device *rdev) 1045 { 1046 int i, r; 1047 1048 if (rdev->gart.robj == NULL) { 1049 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n"); 1050 return -EINVAL; 1051 } 1052 r = radeon_gart_table_vram_pin(rdev); 1053 if (r) 1054 return r; 1055 radeon_gart_restore(rdev); 1056 /* Setup TLB control */ 1057 WREG32(MC_VM_MX_L1_TLB_CNTL, 1058 (0xA << 7) | 1059 ENABLE_L1_TLB | 1060 ENABLE_L1_FRAGMENT_PROCESSING | 1061 SYSTEM_ACCESS_MODE_NOT_IN_SYS | 1062 ENABLE_ADVANCED_DRIVER_MODEL | 1063 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU); 1064 /* Setup L2 cache */ 1065 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | 1066 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE | 1067 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE | 1068 EFFECTIVE_L2_QUEUE_SIZE(7) | 1069 CONTEXT1_IDENTITY_ACCESS_MODE(1)); 1070 WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE); 1071 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY | 1072 L2_CACHE_BIGK_FRAGMENT_SIZE(6)); 1073 /* setup context0 */ 1074 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12); 1075 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12); 1076 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12); 1077 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR, 1078 (u32)(rdev->dummy_page.addr >> 12)); 1079 WREG32(VM_CONTEXT0_CNTL2, 0); 1080 WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) | 1081 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT); 1082 1083 WREG32(0x15D4, 0); 1084 WREG32(0x15D8, 0); 1085 WREG32(0x15DC, 0); 1086 1087 /* empty context1-7 */ 1088 /* Assign the pt base to something valid for now; the pts used for 1089 * the VMs are determined by the application and setup and assigned 1090 * on the fly in the vm part of radeon_gart.c 1091 */ 1092 for (i = 1; i < 8; i++) { 1093 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR + (i << 2), 0); 1094 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR + (i << 2), rdev->vm_manager.max_pfn); 1095 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2), 1096 rdev->gart.table_addr >> 12); 1097 } 1098 1099 /* enable context1-7 */ 1100 WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR, 1101 (u32)(rdev->dummy_page.addr >> 12)); 1102 WREG32(VM_CONTEXT1_CNTL2, 4); 1103 WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) | 1104 RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT | 1105 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT | 1106 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT | 1107 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT | 1108 PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT | 1109 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT | 1110 VALID_PROTECTION_FAULT_ENABLE_INTERRUPT | 1111 VALID_PROTECTION_FAULT_ENABLE_DEFAULT | 1112 READ_PROTECTION_FAULT_ENABLE_INTERRUPT | 1113 READ_PROTECTION_FAULT_ENABLE_DEFAULT | 1114 WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT | 1115 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT); 1116 1117 cayman_pcie_gart_tlb_flush(rdev); 1118 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", 1119 (unsigned)(rdev->mc.gtt_size >> 20), 1120 (unsigned long long)rdev->gart.table_addr); 1121 rdev->gart.ready = true; 1122 return 0; 1123 } 1124 1125 static void cayman_pcie_gart_disable(struct radeon_device *rdev) 1126 { 1127 /* Disable all tables */ 1128 WREG32(VM_CONTEXT0_CNTL, 0); 1129 WREG32(VM_CONTEXT1_CNTL, 0); 1130 /* Setup TLB control */ 1131 WREG32(MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING | 1132 SYSTEM_ACCESS_MODE_NOT_IN_SYS | 1133 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU); 1134 /* Setup L2 cache */ 1135 WREG32(VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE | 1136 ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE | 1137 EFFECTIVE_L2_QUEUE_SIZE(7) | 1138 CONTEXT1_IDENTITY_ACCESS_MODE(1)); 1139 WREG32(VM_L2_CNTL2, 0); 1140 WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY | 1141 L2_CACHE_BIGK_FRAGMENT_SIZE(6)); 1142 radeon_gart_table_vram_unpin(rdev); 1143 } 1144 1145 static void cayman_pcie_gart_fini(struct radeon_device *rdev) 1146 { 1147 cayman_pcie_gart_disable(rdev); 1148 radeon_gart_table_vram_free(rdev); 1149 radeon_gart_fini(rdev); 1150 } 1151 1152 void cayman_cp_int_cntl_setup(struct radeon_device *rdev, 1153 int ring, u32 cp_int_cntl) 1154 { 1155 u32 srbm_gfx_cntl = RREG32(SRBM_GFX_CNTL) & ~3; 1156 1157 WREG32(SRBM_GFX_CNTL, srbm_gfx_cntl | (ring & 3)); 1158 WREG32(CP_INT_CNTL, cp_int_cntl); 1159 } 1160 1161 /* 1162 * CP. 1163 */ 1164 void cayman_fence_ring_emit(struct radeon_device *rdev, 1165 struct radeon_fence *fence) 1166 { 1167 struct radeon_ring *ring = &rdev->ring[fence->ring]; 1168 u64 addr = rdev->fence_drv[fence->ring].gpu_addr; 1169 1170 /* flush read cache over gart for this vmid */ 1171 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 1172 radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2); 1173 radeon_ring_write(ring, 0); 1174 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3)); 1175 radeon_ring_write(ring, PACKET3_TC_ACTION_ENA | PACKET3_SH_ACTION_ENA); 1176 radeon_ring_write(ring, 0xFFFFFFFF); 1177 radeon_ring_write(ring, 0); 1178 radeon_ring_write(ring, 10); /* poll interval */ 1179 /* EVENT_WRITE_EOP - flush caches, send int */ 1180 radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4)); 1181 radeon_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_EVENT_TS) | EVENT_INDEX(5)); 1182 radeon_ring_write(ring, addr & 0xffffffff); 1183 radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | DATA_SEL(1) | INT_SEL(2)); 1184 radeon_ring_write(ring, fence->seq); 1185 radeon_ring_write(ring, 0); 1186 } 1187 1188 void cayman_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib) 1189 { 1190 struct radeon_ring *ring = &rdev->ring[ib->ring]; 1191 1192 /* set to DX10/11 mode */ 1193 radeon_ring_write(ring, PACKET3(PACKET3_MODE_CONTROL, 0)); 1194 radeon_ring_write(ring, 1); 1195 1196 if (ring->rptr_save_reg) { 1197 uint32_t next_rptr = ring->wptr + 3 + 4 + 8; 1198 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 1199 radeon_ring_write(ring, ((ring->rptr_save_reg - 1200 PACKET3_SET_CONFIG_REG_START) >> 2)); 1201 radeon_ring_write(ring, next_rptr); 1202 } 1203 1204 radeon_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2)); 1205 radeon_ring_write(ring, 1206 #ifdef __BIG_ENDIAN 1207 (2 << 0) | 1208 #endif 1209 (ib->gpu_addr & 0xFFFFFFFC)); 1210 radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFF); 1211 radeon_ring_write(ring, ib->length_dw | 1212 (ib->vm ? (ib->vm->id << 24) : 0)); 1213 1214 /* flush read cache over gart for this vmid */ 1215 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); 1216 radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2); 1217 radeon_ring_write(ring, ib->vm ? ib->vm->id : 0); 1218 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3)); 1219 radeon_ring_write(ring, PACKET3_TC_ACTION_ENA | PACKET3_SH_ACTION_ENA); 1220 radeon_ring_write(ring, 0xFFFFFFFF); 1221 radeon_ring_write(ring, 0); 1222 radeon_ring_write(ring, 10); /* poll interval */ 1223 } 1224 1225 void cayman_uvd_semaphore_emit(struct radeon_device *rdev, 1226 struct radeon_ring *ring, 1227 struct radeon_semaphore *semaphore, 1228 bool emit_wait) 1229 { 1230 uint64_t addr = semaphore->gpu_addr; 1231 1232 radeon_ring_write(ring, PACKET0(UVD_SEMA_ADDR_LOW, 0)); 1233 radeon_ring_write(ring, (addr >> 3) & 0x000FFFFF); 1234 1235 radeon_ring_write(ring, PACKET0(UVD_SEMA_ADDR_HIGH, 0)); 1236 radeon_ring_write(ring, (addr >> 23) & 0x000FFFFF); 1237 1238 radeon_ring_write(ring, PACKET0(UVD_SEMA_CMD, 0)); 1239 radeon_ring_write(ring, 0x80 | (emit_wait ? 1 : 0)); 1240 } 1241 1242 static void cayman_cp_enable(struct radeon_device *rdev, bool enable) 1243 { 1244 if (enable) 1245 WREG32(CP_ME_CNTL, 0); 1246 else { 1247 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size); 1248 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT)); 1249 WREG32(SCRATCH_UMSK, 0); 1250 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false; 1251 } 1252 } 1253 1254 static int cayman_cp_load_microcode(struct radeon_device *rdev) 1255 { 1256 const __be32 *fw_data; 1257 int i; 1258 1259 if (!rdev->me_fw || !rdev->pfp_fw) 1260 return -EINVAL; 1261 1262 cayman_cp_enable(rdev, false); 1263 1264 fw_data = (const __be32 *)rdev->pfp_fw->data; 1265 WREG32(CP_PFP_UCODE_ADDR, 0); 1266 for (i = 0; i < CAYMAN_PFP_UCODE_SIZE; i++) 1267 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++)); 1268 WREG32(CP_PFP_UCODE_ADDR, 0); 1269 1270 fw_data = (const __be32 *)rdev->me_fw->data; 1271 WREG32(CP_ME_RAM_WADDR, 0); 1272 for (i = 0; i < CAYMAN_PM4_UCODE_SIZE; i++) 1273 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++)); 1274 1275 WREG32(CP_PFP_UCODE_ADDR, 0); 1276 WREG32(CP_ME_RAM_WADDR, 0); 1277 WREG32(CP_ME_RAM_RADDR, 0); 1278 return 0; 1279 } 1280 1281 static int cayman_cp_start(struct radeon_device *rdev) 1282 { 1283 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 1284 int r, i; 1285 1286 r = radeon_ring_lock(rdev, ring, 7); 1287 if (r) { 1288 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r); 1289 return r; 1290 } 1291 radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5)); 1292 radeon_ring_write(ring, 0x1); 1293 radeon_ring_write(ring, 0x0); 1294 radeon_ring_write(ring, rdev->config.cayman.max_hw_contexts - 1); 1295 radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1)); 1296 radeon_ring_write(ring, 0); 1297 radeon_ring_write(ring, 0); 1298 radeon_ring_unlock_commit(rdev, ring); 1299 1300 cayman_cp_enable(rdev, true); 1301 1302 r = radeon_ring_lock(rdev, ring, cayman_default_size + 19); 1303 if (r) { 1304 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r); 1305 return r; 1306 } 1307 1308 /* setup clear context state */ 1309 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 1310 radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE); 1311 1312 for (i = 0; i < cayman_default_size; i++) 1313 radeon_ring_write(ring, cayman_default_state[i]); 1314 1315 radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0)); 1316 radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE); 1317 1318 /* set clear context state */ 1319 radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0)); 1320 radeon_ring_write(ring, 0); 1321 1322 /* SQ_VTX_BASE_VTX_LOC */ 1323 radeon_ring_write(ring, 0xc0026f00); 1324 radeon_ring_write(ring, 0x00000000); 1325 radeon_ring_write(ring, 0x00000000); 1326 radeon_ring_write(ring, 0x00000000); 1327 1328 /* Clear consts */ 1329 radeon_ring_write(ring, 0xc0036f00); 1330 radeon_ring_write(ring, 0x00000bc4); 1331 radeon_ring_write(ring, 0xffffffff); 1332 radeon_ring_write(ring, 0xffffffff); 1333 radeon_ring_write(ring, 0xffffffff); 1334 1335 radeon_ring_write(ring, 0xc0026900); 1336 radeon_ring_write(ring, 0x00000316); 1337 radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */ 1338 radeon_ring_write(ring, 0x00000010); /* */ 1339 1340 radeon_ring_unlock_commit(rdev, ring); 1341 1342 /* XXX init other rings */ 1343 1344 return 0; 1345 } 1346 1347 static void cayman_cp_fini(struct radeon_device *rdev) 1348 { 1349 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 1350 cayman_cp_enable(rdev, false); 1351 radeon_ring_fini(rdev, ring); 1352 radeon_scratch_free(rdev, ring->rptr_save_reg); 1353 } 1354 1355 static int cayman_cp_resume(struct radeon_device *rdev) 1356 { 1357 static const int ridx[] = { 1358 RADEON_RING_TYPE_GFX_INDEX, 1359 CAYMAN_RING_TYPE_CP1_INDEX, 1360 CAYMAN_RING_TYPE_CP2_INDEX 1361 }; 1362 static const unsigned cp_rb_cntl[] = { 1363 CP_RB0_CNTL, 1364 CP_RB1_CNTL, 1365 CP_RB2_CNTL, 1366 }; 1367 static const unsigned cp_rb_rptr_addr[] = { 1368 CP_RB0_RPTR_ADDR, 1369 CP_RB1_RPTR_ADDR, 1370 CP_RB2_RPTR_ADDR 1371 }; 1372 static const unsigned cp_rb_rptr_addr_hi[] = { 1373 CP_RB0_RPTR_ADDR_HI, 1374 CP_RB1_RPTR_ADDR_HI, 1375 CP_RB2_RPTR_ADDR_HI 1376 }; 1377 static const unsigned cp_rb_base[] = { 1378 CP_RB0_BASE, 1379 CP_RB1_BASE, 1380 CP_RB2_BASE 1381 }; 1382 struct radeon_ring *ring; 1383 int i, r; 1384 1385 /* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */ 1386 WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP | 1387 SOFT_RESET_PA | 1388 SOFT_RESET_SH | 1389 SOFT_RESET_VGT | 1390 SOFT_RESET_SPI | 1391 SOFT_RESET_SX)); 1392 RREG32(GRBM_SOFT_RESET); 1393 mdelay(15); 1394 WREG32(GRBM_SOFT_RESET, 0); 1395 RREG32(GRBM_SOFT_RESET); 1396 1397 WREG32(CP_SEM_WAIT_TIMER, 0x0); 1398 WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0); 1399 1400 /* Set the write pointer delay */ 1401 WREG32(CP_RB_WPTR_DELAY, 0); 1402 1403 WREG32(CP_DEBUG, (1 << 27)); 1404 1405 /* set the wb address whether it's enabled or not */ 1406 WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF); 1407 WREG32(SCRATCH_UMSK, 0xff); 1408 1409 for (i = 0; i < 3; ++i) { 1410 uint32_t rb_cntl; 1411 uint64_t addr; 1412 1413 /* Set ring buffer size */ 1414 ring = &rdev->ring[ridx[i]]; 1415 rb_cntl = drm_order(ring->ring_size / 8); 1416 rb_cntl |= drm_order(RADEON_GPU_PAGE_SIZE/8) << 8; 1417 #ifdef __BIG_ENDIAN 1418 rb_cntl |= BUF_SWAP_32BIT; 1419 #endif 1420 WREG32(cp_rb_cntl[i], rb_cntl); 1421 1422 /* set the wb address whether it's enabled or not */ 1423 addr = rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET; 1424 WREG32(cp_rb_rptr_addr[i], addr & 0xFFFFFFFC); 1425 WREG32(cp_rb_rptr_addr_hi[i], upper_32_bits(addr) & 0xFF); 1426 } 1427 1428 /* set the rb base addr, this causes an internal reset of ALL rings */ 1429 for (i = 0; i < 3; ++i) { 1430 ring = &rdev->ring[ridx[i]]; 1431 WREG32(cp_rb_base[i], ring->gpu_addr >> 8); 1432 } 1433 1434 for (i = 0; i < 3; ++i) { 1435 /* Initialize the ring buffer's read and write pointers */ 1436 ring = &rdev->ring[ridx[i]]; 1437 WREG32_P(cp_rb_cntl[i], RB_RPTR_WR_ENA, ~RB_RPTR_WR_ENA); 1438 1439 ring->rptr = ring->wptr = 0; 1440 WREG32(ring->rptr_reg, ring->rptr); 1441 WREG32(ring->wptr_reg, ring->wptr); 1442 1443 mdelay(1); 1444 WREG32_P(cp_rb_cntl[i], 0, ~RB_RPTR_WR_ENA); 1445 } 1446 1447 /* start the rings */ 1448 cayman_cp_start(rdev); 1449 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true; 1450 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false; 1451 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false; 1452 /* this only test cp0 */ 1453 r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]); 1454 if (r) { 1455 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false; 1456 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false; 1457 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false; 1458 return r; 1459 } 1460 1461 return 0; 1462 } 1463 1464 /* 1465 * DMA 1466 * Starting with R600, the GPU has an asynchronous 1467 * DMA engine. The programming model is very similar 1468 * to the 3D engine (ring buffer, IBs, etc.), but the 1469 * DMA controller has it's own packet format that is 1470 * different form the PM4 format used by the 3D engine. 1471 * It supports copying data, writing embedded data, 1472 * solid fills, and a number of other things. It also 1473 * has support for tiling/detiling of buffers. 1474 * Cayman and newer support two asynchronous DMA engines. 1475 */ 1476 /** 1477 * cayman_dma_ring_ib_execute - Schedule an IB on the DMA engine 1478 * 1479 * @rdev: radeon_device pointer 1480 * @ib: IB object to schedule 1481 * 1482 * Schedule an IB in the DMA ring (cayman-SI). 1483 */ 1484 void cayman_dma_ring_ib_execute(struct radeon_device *rdev, 1485 struct radeon_ib *ib) 1486 { 1487 struct radeon_ring *ring = &rdev->ring[ib->ring]; 1488 1489 if (rdev->wb.enabled) { 1490 u32 next_rptr = ring->wptr + 4; 1491 while ((next_rptr & 7) != 5) 1492 next_rptr++; 1493 next_rptr += 3; 1494 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 1)); 1495 radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc); 1496 radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xff); 1497 radeon_ring_write(ring, next_rptr); 1498 } 1499 1500 /* The indirect buffer packet must end on an 8 DW boundary in the DMA ring. 1501 * Pad as necessary with NOPs. 1502 */ 1503 while ((ring->wptr & 7) != 5) 1504 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0)); 1505 radeon_ring_write(ring, DMA_IB_PACKET(DMA_PACKET_INDIRECT_BUFFER, ib->vm ? ib->vm->id : 0, 0)); 1506 radeon_ring_write(ring, (ib->gpu_addr & 0xFFFFFFE0)); 1507 radeon_ring_write(ring, (ib->length_dw << 12) | (upper_32_bits(ib->gpu_addr) & 0xFF)); 1508 1509 } 1510 1511 /** 1512 * cayman_dma_stop - stop the async dma engines 1513 * 1514 * @rdev: radeon_device pointer 1515 * 1516 * Stop the async dma engines (cayman-SI). 1517 */ 1518 void cayman_dma_stop(struct radeon_device *rdev) 1519 { 1520 u32 rb_cntl; 1521 1522 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size); 1523 1524 /* dma0 */ 1525 rb_cntl = RREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET); 1526 rb_cntl &= ~DMA_RB_ENABLE; 1527 WREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET, rb_cntl); 1528 1529 /* dma1 */ 1530 rb_cntl = RREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET); 1531 rb_cntl &= ~DMA_RB_ENABLE; 1532 WREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET, rb_cntl); 1533 1534 rdev->ring[R600_RING_TYPE_DMA_INDEX].ready = false; 1535 rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX].ready = false; 1536 } 1537 1538 /** 1539 * cayman_dma_resume - setup and start the async dma engines 1540 * 1541 * @rdev: radeon_device pointer 1542 * 1543 * Set up the DMA ring buffers and enable them. (cayman-SI). 1544 * Returns 0 for success, error for failure. 1545 */ 1546 int cayman_dma_resume(struct radeon_device *rdev) 1547 { 1548 struct radeon_ring *ring; 1549 u32 rb_cntl, dma_cntl, ib_cntl; 1550 u32 rb_bufsz; 1551 u32 reg_offset, wb_offset; 1552 int i, r; 1553 1554 /* Reset dma */ 1555 WREG32(SRBM_SOFT_RESET, SOFT_RESET_DMA | SOFT_RESET_DMA1); 1556 RREG32(SRBM_SOFT_RESET); 1557 udelay(50); 1558 WREG32(SRBM_SOFT_RESET, 0); 1559 1560 for (i = 0; i < 2; i++) { 1561 if (i == 0) { 1562 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX]; 1563 reg_offset = DMA0_REGISTER_OFFSET; 1564 wb_offset = R600_WB_DMA_RPTR_OFFSET; 1565 } else { 1566 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]; 1567 reg_offset = DMA1_REGISTER_OFFSET; 1568 wb_offset = CAYMAN_WB_DMA1_RPTR_OFFSET; 1569 } 1570 1571 WREG32(DMA_SEM_INCOMPLETE_TIMER_CNTL + reg_offset, 0); 1572 WREG32(DMA_SEM_WAIT_FAIL_TIMER_CNTL + reg_offset, 0); 1573 1574 /* Set ring buffer size in dwords */ 1575 rb_bufsz = drm_order(ring->ring_size / 4); 1576 rb_cntl = rb_bufsz << 1; 1577 #ifdef __BIG_ENDIAN 1578 rb_cntl |= DMA_RB_SWAP_ENABLE | DMA_RPTR_WRITEBACK_SWAP_ENABLE; 1579 #endif 1580 WREG32(DMA_RB_CNTL + reg_offset, rb_cntl); 1581 1582 /* Initialize the ring buffer's read and write pointers */ 1583 WREG32(DMA_RB_RPTR + reg_offset, 0); 1584 WREG32(DMA_RB_WPTR + reg_offset, 0); 1585 1586 /* set the wb address whether it's enabled or not */ 1587 WREG32(DMA_RB_RPTR_ADDR_HI + reg_offset, 1588 upper_32_bits(rdev->wb.gpu_addr + wb_offset) & 0xFF); 1589 WREG32(DMA_RB_RPTR_ADDR_LO + reg_offset, 1590 ((rdev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC)); 1591 1592 if (rdev->wb.enabled) 1593 rb_cntl |= DMA_RPTR_WRITEBACK_ENABLE; 1594 1595 WREG32(DMA_RB_BASE + reg_offset, ring->gpu_addr >> 8); 1596 1597 /* enable DMA IBs */ 1598 ib_cntl = DMA_IB_ENABLE | CMD_VMID_FORCE; 1599 #ifdef __BIG_ENDIAN 1600 ib_cntl |= DMA_IB_SWAP_ENABLE; 1601 #endif 1602 WREG32(DMA_IB_CNTL + reg_offset, ib_cntl); 1603 1604 dma_cntl = RREG32(DMA_CNTL + reg_offset); 1605 dma_cntl &= ~CTXEMPTY_INT_ENABLE; 1606 WREG32(DMA_CNTL + reg_offset, dma_cntl); 1607 1608 ring->wptr = 0; 1609 WREG32(DMA_RB_WPTR + reg_offset, ring->wptr << 2); 1610 1611 ring->rptr = RREG32(DMA_RB_RPTR + reg_offset) >> 2; 1612 1613 WREG32(DMA_RB_CNTL + reg_offset, rb_cntl | DMA_RB_ENABLE); 1614 1615 ring->ready = true; 1616 1617 r = radeon_ring_test(rdev, ring->idx, ring); 1618 if (r) { 1619 ring->ready = false; 1620 return r; 1621 } 1622 } 1623 1624 radeon_ttm_set_active_vram_size(rdev, rdev->mc.real_vram_size); 1625 1626 return 0; 1627 } 1628 1629 /** 1630 * cayman_dma_fini - tear down the async dma engines 1631 * 1632 * @rdev: radeon_device pointer 1633 * 1634 * Stop the async dma engines and free the rings (cayman-SI). 1635 */ 1636 void cayman_dma_fini(struct radeon_device *rdev) 1637 { 1638 cayman_dma_stop(rdev); 1639 radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX]); 1640 radeon_ring_fini(rdev, &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]); 1641 } 1642 1643 static u32 cayman_gpu_check_soft_reset(struct radeon_device *rdev) 1644 { 1645 u32 reset_mask = 0; 1646 u32 tmp; 1647 1648 /* GRBM_STATUS */ 1649 tmp = RREG32(GRBM_STATUS); 1650 if (tmp & (PA_BUSY | SC_BUSY | 1651 SH_BUSY | SX_BUSY | 1652 TA_BUSY | VGT_BUSY | 1653 DB_BUSY | CB_BUSY | 1654 GDS_BUSY | SPI_BUSY | 1655 IA_BUSY | IA_BUSY_NO_DMA)) 1656 reset_mask |= RADEON_RESET_GFX; 1657 1658 if (tmp & (CF_RQ_PENDING | PF_RQ_PENDING | 1659 CP_BUSY | CP_COHERENCY_BUSY)) 1660 reset_mask |= RADEON_RESET_CP; 1661 1662 if (tmp & GRBM_EE_BUSY) 1663 reset_mask |= RADEON_RESET_GRBM | RADEON_RESET_GFX | RADEON_RESET_CP; 1664 1665 /* DMA_STATUS_REG 0 */ 1666 tmp = RREG32(DMA_STATUS_REG + DMA0_REGISTER_OFFSET); 1667 if (!(tmp & DMA_IDLE)) 1668 reset_mask |= RADEON_RESET_DMA; 1669 1670 /* DMA_STATUS_REG 1 */ 1671 tmp = RREG32(DMA_STATUS_REG + DMA1_REGISTER_OFFSET); 1672 if (!(tmp & DMA_IDLE)) 1673 reset_mask |= RADEON_RESET_DMA1; 1674 1675 /* SRBM_STATUS2 */ 1676 tmp = RREG32(SRBM_STATUS2); 1677 if (tmp & DMA_BUSY) 1678 reset_mask |= RADEON_RESET_DMA; 1679 1680 if (tmp & DMA1_BUSY) 1681 reset_mask |= RADEON_RESET_DMA1; 1682 1683 /* SRBM_STATUS */ 1684 tmp = RREG32(SRBM_STATUS); 1685 if (tmp & (RLC_RQ_PENDING | RLC_BUSY)) 1686 reset_mask |= RADEON_RESET_RLC; 1687 1688 if (tmp & IH_BUSY) 1689 reset_mask |= RADEON_RESET_IH; 1690 1691 if (tmp & SEM_BUSY) 1692 reset_mask |= RADEON_RESET_SEM; 1693 1694 if (tmp & GRBM_RQ_PENDING) 1695 reset_mask |= RADEON_RESET_GRBM; 1696 1697 if (tmp & VMC_BUSY) 1698 reset_mask |= RADEON_RESET_VMC; 1699 1700 if (tmp & (MCB_BUSY | MCB_NON_DISPLAY_BUSY | 1701 MCC_BUSY | MCD_BUSY)) 1702 reset_mask |= RADEON_RESET_MC; 1703 1704 if (evergreen_is_display_hung(rdev)) 1705 reset_mask |= RADEON_RESET_DISPLAY; 1706 1707 /* VM_L2_STATUS */ 1708 tmp = RREG32(VM_L2_STATUS); 1709 if (tmp & L2_BUSY) 1710 reset_mask |= RADEON_RESET_VMC; 1711 1712 /* Skip MC reset as it's mostly likely not hung, just busy */ 1713 if (reset_mask & RADEON_RESET_MC) { 1714 DRM_DEBUG("MC busy: 0x%08X, clearing.\n", reset_mask); 1715 reset_mask &= ~RADEON_RESET_MC; 1716 } 1717 1718 return reset_mask; 1719 } 1720 1721 static void cayman_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask) 1722 { 1723 struct evergreen_mc_save save; 1724 u32 grbm_soft_reset = 0, srbm_soft_reset = 0; 1725 u32 tmp; 1726 1727 if (reset_mask == 0) 1728 return; 1729 1730 dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask); 1731 1732 evergreen_print_gpu_status_regs(rdev); 1733 dev_info(rdev->dev, " VM_CONTEXT0_PROTECTION_FAULT_ADDR 0x%08X\n", 1734 RREG32(0x14F8)); 1735 dev_info(rdev->dev, " VM_CONTEXT0_PROTECTION_FAULT_STATUS 0x%08X\n", 1736 RREG32(0x14D8)); 1737 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_ADDR 0x%08X\n", 1738 RREG32(0x14FC)); 1739 dev_info(rdev->dev, " VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n", 1740 RREG32(0x14DC)); 1741 1742 /* Disable CP parsing/prefetching */ 1743 WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT); 1744 1745 if (reset_mask & RADEON_RESET_DMA) { 1746 /* dma0 */ 1747 tmp = RREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET); 1748 tmp &= ~DMA_RB_ENABLE; 1749 WREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET, tmp); 1750 } 1751 1752 if (reset_mask & RADEON_RESET_DMA1) { 1753 /* dma1 */ 1754 tmp = RREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET); 1755 tmp &= ~DMA_RB_ENABLE; 1756 WREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET, tmp); 1757 } 1758 1759 udelay(50); 1760 1761 evergreen_mc_stop(rdev, &save); 1762 if (evergreen_mc_wait_for_idle(rdev)) { 1763 dev_warn(rdev->dev, "Wait for MC idle timedout !\n"); 1764 } 1765 1766 if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE)) { 1767 grbm_soft_reset = SOFT_RESET_CB | 1768 SOFT_RESET_DB | 1769 SOFT_RESET_GDS | 1770 SOFT_RESET_PA | 1771 SOFT_RESET_SC | 1772 SOFT_RESET_SPI | 1773 SOFT_RESET_SH | 1774 SOFT_RESET_SX | 1775 SOFT_RESET_TC | 1776 SOFT_RESET_TA | 1777 SOFT_RESET_VGT | 1778 SOFT_RESET_IA; 1779 } 1780 1781 if (reset_mask & RADEON_RESET_CP) { 1782 grbm_soft_reset |= SOFT_RESET_CP | SOFT_RESET_VGT; 1783 1784 srbm_soft_reset |= SOFT_RESET_GRBM; 1785 } 1786 1787 if (reset_mask & RADEON_RESET_DMA) 1788 srbm_soft_reset |= SOFT_RESET_DMA; 1789 1790 if (reset_mask & RADEON_RESET_DMA1) 1791 srbm_soft_reset |= SOFT_RESET_DMA1; 1792 1793 if (reset_mask & RADEON_RESET_DISPLAY) 1794 srbm_soft_reset |= SOFT_RESET_DC; 1795 1796 if (reset_mask & RADEON_RESET_RLC) 1797 srbm_soft_reset |= SOFT_RESET_RLC; 1798 1799 if (reset_mask & RADEON_RESET_SEM) 1800 srbm_soft_reset |= SOFT_RESET_SEM; 1801 1802 if (reset_mask & RADEON_RESET_IH) 1803 srbm_soft_reset |= SOFT_RESET_IH; 1804 1805 if (reset_mask & RADEON_RESET_GRBM) 1806 srbm_soft_reset |= SOFT_RESET_GRBM; 1807 1808 if (reset_mask & RADEON_RESET_VMC) 1809 srbm_soft_reset |= SOFT_RESET_VMC; 1810 1811 if (!(rdev->flags & RADEON_IS_IGP)) { 1812 if (reset_mask & RADEON_RESET_MC) 1813 srbm_soft_reset |= SOFT_RESET_MC; 1814 } 1815 1816 if (grbm_soft_reset) { 1817 tmp = RREG32(GRBM_SOFT_RESET); 1818 tmp |= grbm_soft_reset; 1819 dev_info(rdev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp); 1820 WREG32(GRBM_SOFT_RESET, tmp); 1821 tmp = RREG32(GRBM_SOFT_RESET); 1822 1823 udelay(50); 1824 1825 tmp &= ~grbm_soft_reset; 1826 WREG32(GRBM_SOFT_RESET, tmp); 1827 tmp = RREG32(GRBM_SOFT_RESET); 1828 } 1829 1830 if (srbm_soft_reset) { 1831 tmp = RREG32(SRBM_SOFT_RESET); 1832 tmp |= srbm_soft_reset; 1833 dev_info(rdev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp); 1834 WREG32(SRBM_SOFT_RESET, tmp); 1835 tmp = RREG32(SRBM_SOFT_RESET); 1836 1837 udelay(50); 1838 1839 tmp &= ~srbm_soft_reset; 1840 WREG32(SRBM_SOFT_RESET, tmp); 1841 tmp = RREG32(SRBM_SOFT_RESET); 1842 } 1843 1844 /* Wait a little for things to settle down */ 1845 udelay(50); 1846 1847 evergreen_mc_resume(rdev, &save); 1848 udelay(50); 1849 1850 evergreen_print_gpu_status_regs(rdev); 1851 } 1852 1853 int cayman_asic_reset(struct radeon_device *rdev) 1854 { 1855 u32 reset_mask; 1856 1857 reset_mask = cayman_gpu_check_soft_reset(rdev); 1858 1859 if (reset_mask) 1860 r600_set_bios_scratch_engine_hung(rdev, true); 1861 1862 cayman_gpu_soft_reset(rdev, reset_mask); 1863 1864 reset_mask = cayman_gpu_check_soft_reset(rdev); 1865 1866 if (!reset_mask) 1867 r600_set_bios_scratch_engine_hung(rdev, false); 1868 1869 return 0; 1870 } 1871 1872 /** 1873 * cayman_gfx_is_lockup - Check if the GFX engine is locked up 1874 * 1875 * @rdev: radeon_device pointer 1876 * @ring: radeon_ring structure holding ring information 1877 * 1878 * Check if the GFX engine is locked up. 1879 * Returns true if the engine appears to be locked up, false if not. 1880 */ 1881 bool cayman_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring) 1882 { 1883 u32 reset_mask = cayman_gpu_check_soft_reset(rdev); 1884 1885 if (!(reset_mask & (RADEON_RESET_GFX | 1886 RADEON_RESET_COMPUTE | 1887 RADEON_RESET_CP))) { 1888 radeon_ring_lockup_update(ring); 1889 return false; 1890 } 1891 /* force CP activities */ 1892 radeon_ring_force_activity(rdev, ring); 1893 return radeon_ring_test_lockup(rdev, ring); 1894 } 1895 1896 /** 1897 * cayman_dma_is_lockup - Check if the DMA engine is locked up 1898 * 1899 * @rdev: radeon_device pointer 1900 * @ring: radeon_ring structure holding ring information 1901 * 1902 * Check if the async DMA engine is locked up. 1903 * Returns true if the engine appears to be locked up, false if not. 1904 */ 1905 bool cayman_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring) 1906 { 1907 u32 reset_mask = cayman_gpu_check_soft_reset(rdev); 1908 u32 mask; 1909 1910 if (ring->idx == R600_RING_TYPE_DMA_INDEX) 1911 mask = RADEON_RESET_DMA; 1912 else 1913 mask = RADEON_RESET_DMA1; 1914 1915 if (!(reset_mask & mask)) { 1916 radeon_ring_lockup_update(ring); 1917 return false; 1918 } 1919 /* force ring activities */ 1920 radeon_ring_force_activity(rdev, ring); 1921 return radeon_ring_test_lockup(rdev, ring); 1922 } 1923 1924 static int cayman_startup(struct radeon_device *rdev) 1925 { 1926 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 1927 int r; 1928 1929 /* enable pcie gen2 link */ 1930 evergreen_pcie_gen2_enable(rdev); 1931 1932 if (rdev->flags & RADEON_IS_IGP) { 1933 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) { 1934 r = ni_init_microcode(rdev); 1935 if (r) { 1936 DRM_ERROR("Failed to load firmware!\n"); 1937 return r; 1938 } 1939 } 1940 } else { 1941 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw || !rdev->mc_fw) { 1942 r = ni_init_microcode(rdev); 1943 if (r) { 1944 DRM_ERROR("Failed to load firmware!\n"); 1945 return r; 1946 } 1947 } 1948 1949 r = ni_mc_load_microcode(rdev); 1950 if (r) { 1951 DRM_ERROR("Failed to load MC firmware!\n"); 1952 return r; 1953 } 1954 } 1955 1956 r = r600_vram_scratch_init(rdev); 1957 if (r) 1958 return r; 1959 1960 evergreen_mc_program(rdev); 1961 r = cayman_pcie_gart_enable(rdev); 1962 if (r) 1963 return r; 1964 cayman_gpu_init(rdev); 1965 1966 r = evergreen_blit_init(rdev); 1967 if (r) { 1968 r600_blit_fini(rdev); 1969 rdev->asic->copy.copy = NULL; 1970 dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r); 1971 } 1972 1973 /* allocate rlc buffers */ 1974 if (rdev->flags & RADEON_IS_IGP) { 1975 r = si_rlc_init(rdev); 1976 if (r) { 1977 DRM_ERROR("Failed to init rlc BOs!\n"); 1978 return r; 1979 } 1980 } 1981 1982 /* allocate wb buffer */ 1983 r = radeon_wb_init(rdev); 1984 if (r) 1985 return r; 1986 1987 r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX); 1988 if (r) { 1989 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r); 1990 return r; 1991 } 1992 1993 r = rv770_uvd_resume(rdev); 1994 if (!r) { 1995 r = radeon_fence_driver_start_ring(rdev, 1996 R600_RING_TYPE_UVD_INDEX); 1997 if (r) 1998 dev_err(rdev->dev, "UVD fences init error (%d).\n", r); 1999 } 2000 if (r) 2001 rdev->ring[R600_RING_TYPE_UVD_INDEX].ring_size = 0; 2002 2003 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP1_INDEX); 2004 if (r) { 2005 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r); 2006 return r; 2007 } 2008 2009 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP2_INDEX); 2010 if (r) { 2011 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r); 2012 return r; 2013 } 2014 2015 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX); 2016 if (r) { 2017 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r); 2018 return r; 2019 } 2020 2021 r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX); 2022 if (r) { 2023 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r); 2024 return r; 2025 } 2026 2027 /* Enable IRQ */ 2028 if (!rdev->irq.installed) { 2029 r = radeon_irq_kms_init(rdev); 2030 if (r) 2031 return r; 2032 } 2033 2034 r = r600_irq_init(rdev); 2035 if (r) { 2036 DRM_ERROR("radeon: IH init failed (%d).\n", r); 2037 radeon_irq_kms_fini(rdev); 2038 return r; 2039 } 2040 evergreen_irq_set(rdev); 2041 2042 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET, 2043 CP_RB0_RPTR, CP_RB0_WPTR, 2044 0, 0xfffff, RADEON_CP_PACKET2); 2045 if (r) 2046 return r; 2047 2048 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX]; 2049 r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET, 2050 DMA_RB_RPTR + DMA0_REGISTER_OFFSET, 2051 DMA_RB_WPTR + DMA0_REGISTER_OFFSET, 2052 2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0)); 2053 if (r) 2054 return r; 2055 2056 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]; 2057 r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET, 2058 DMA_RB_RPTR + DMA1_REGISTER_OFFSET, 2059 DMA_RB_WPTR + DMA1_REGISTER_OFFSET, 2060 2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0)); 2061 if (r) 2062 return r; 2063 2064 r = cayman_cp_load_microcode(rdev); 2065 if (r) 2066 return r; 2067 r = cayman_cp_resume(rdev); 2068 if (r) 2069 return r; 2070 2071 r = cayman_dma_resume(rdev); 2072 if (r) 2073 return r; 2074 2075 ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX]; 2076 if (ring->ring_size) { 2077 r = radeon_ring_init(rdev, ring, ring->ring_size, 2078 R600_WB_UVD_RPTR_OFFSET, 2079 UVD_RBC_RB_RPTR, UVD_RBC_RB_WPTR, 2080 0, 0xfffff, RADEON_CP_PACKET2); 2081 if (!r) 2082 r = r600_uvd_init(rdev); 2083 if (r) 2084 DRM_ERROR("radeon: failed initializing UVD (%d).\n", r); 2085 } 2086 2087 r = radeon_ib_pool_init(rdev); 2088 if (r) { 2089 dev_err(rdev->dev, "IB initialization failed (%d).\n", r); 2090 return r; 2091 } 2092 2093 r = radeon_vm_manager_init(rdev); 2094 if (r) { 2095 dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r); 2096 return r; 2097 } 2098 2099 r = r600_audio_init(rdev); 2100 if (r) 2101 return r; 2102 2103 return 0; 2104 } 2105 2106 int cayman_resume(struct radeon_device *rdev) 2107 { 2108 int r; 2109 2110 /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw, 2111 * posting will perform necessary task to bring back GPU into good 2112 * shape. 2113 */ 2114 /* post card */ 2115 atom_asic_init(rdev->mode_info.atom_context); 2116 2117 /* init golden registers */ 2118 ni_init_golden_registers(rdev); 2119 2120 rdev->accel_working = true; 2121 r = cayman_startup(rdev); 2122 if (r) { 2123 DRM_ERROR("cayman startup failed on resume\n"); 2124 rdev->accel_working = false; 2125 return r; 2126 } 2127 return r; 2128 } 2129 2130 int cayman_suspend(struct radeon_device *rdev) 2131 { 2132 r600_audio_fini(rdev); 2133 radeon_vm_manager_fini(rdev); 2134 cayman_cp_enable(rdev, false); 2135 cayman_dma_stop(rdev); 2136 r600_uvd_rbc_stop(rdev); 2137 radeon_uvd_suspend(rdev); 2138 evergreen_irq_suspend(rdev); 2139 radeon_wb_disable(rdev); 2140 cayman_pcie_gart_disable(rdev); 2141 return 0; 2142 } 2143 2144 /* Plan is to move initialization in that function and use 2145 * helper function so that radeon_device_init pretty much 2146 * do nothing more than calling asic specific function. This 2147 * should also allow to remove a bunch of callback function 2148 * like vram_info. 2149 */ 2150 int cayman_init(struct radeon_device *rdev) 2151 { 2152 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 2153 int r; 2154 2155 /* Read BIOS */ 2156 if (!radeon_get_bios(rdev)) { 2157 if (ASIC_IS_AVIVO(rdev)) 2158 return -EINVAL; 2159 } 2160 /* Must be an ATOMBIOS */ 2161 if (!rdev->is_atom_bios) { 2162 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n"); 2163 return -EINVAL; 2164 } 2165 r = radeon_atombios_init(rdev); 2166 if (r) 2167 return r; 2168 2169 /* Post card if necessary */ 2170 if (!radeon_card_posted(rdev)) { 2171 if (!rdev->bios) { 2172 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n"); 2173 return -EINVAL; 2174 } 2175 DRM_INFO("GPU not posted. posting now...\n"); 2176 atom_asic_init(rdev->mode_info.atom_context); 2177 } 2178 /* init golden registers */ 2179 ni_init_golden_registers(rdev); 2180 /* Initialize scratch registers */ 2181 r600_scratch_init(rdev); 2182 /* Initialize surface registers */ 2183 radeon_surface_init(rdev); 2184 /* Initialize clocks */ 2185 radeon_get_clock_info(rdev->ddev); 2186 /* Fence driver */ 2187 r = radeon_fence_driver_init(rdev); 2188 if (r) 2189 return r; 2190 /* initialize memory controller */ 2191 r = evergreen_mc_init(rdev); 2192 if (r) 2193 return r; 2194 /* Memory manager */ 2195 r = radeon_bo_init(rdev); 2196 if (r) 2197 return r; 2198 2199 ring->ring_obj = NULL; 2200 r600_ring_init(rdev, ring, 1024 * 1024); 2201 2202 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX]; 2203 ring->ring_obj = NULL; 2204 r600_ring_init(rdev, ring, 64 * 1024); 2205 2206 ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX]; 2207 ring->ring_obj = NULL; 2208 r600_ring_init(rdev, ring, 64 * 1024); 2209 2210 r = radeon_uvd_init(rdev); 2211 if (!r) { 2212 ring = &rdev->ring[R600_RING_TYPE_UVD_INDEX]; 2213 ring->ring_obj = NULL; 2214 r600_ring_init(rdev, ring, 4096); 2215 } 2216 2217 rdev->ih.ring_obj = NULL; 2218 r600_ih_ring_init(rdev, 64 * 1024); 2219 2220 r = r600_pcie_gart_init(rdev); 2221 if (r) 2222 return r; 2223 2224 rdev->accel_working = true; 2225 r = cayman_startup(rdev); 2226 if (r) { 2227 dev_err(rdev->dev, "disabling GPU acceleration\n"); 2228 cayman_cp_fini(rdev); 2229 cayman_dma_fini(rdev); 2230 r600_irq_fini(rdev); 2231 if (rdev->flags & RADEON_IS_IGP) 2232 si_rlc_fini(rdev); 2233 radeon_wb_fini(rdev); 2234 radeon_ib_pool_fini(rdev); 2235 radeon_vm_manager_fini(rdev); 2236 radeon_irq_kms_fini(rdev); 2237 cayman_pcie_gart_fini(rdev); 2238 rdev->accel_working = false; 2239 } 2240 2241 /* Don't start up if the MC ucode is missing. 2242 * The default clocks and voltages before the MC ucode 2243 * is loaded are not suffient for advanced operations. 2244 * 2245 * We can skip this check for TN, because there is no MC 2246 * ucode. 2247 */ 2248 if (!rdev->mc_fw && !(rdev->flags & RADEON_IS_IGP)) { 2249 DRM_ERROR("radeon: MC ucode required for NI+.\n"); 2250 return -EINVAL; 2251 } 2252 2253 return 0; 2254 } 2255 2256 void cayman_fini(struct radeon_device *rdev) 2257 { 2258 r600_blit_fini(rdev); 2259 cayman_cp_fini(rdev); 2260 cayman_dma_fini(rdev); 2261 r600_irq_fini(rdev); 2262 if (rdev->flags & RADEON_IS_IGP) 2263 si_rlc_fini(rdev); 2264 radeon_wb_fini(rdev); 2265 radeon_vm_manager_fini(rdev); 2266 radeon_ib_pool_fini(rdev); 2267 radeon_irq_kms_fini(rdev); 2268 radeon_uvd_fini(rdev); 2269 cayman_pcie_gart_fini(rdev); 2270 r600_vram_scratch_fini(rdev); 2271 radeon_gem_fini(rdev); 2272 radeon_fence_driver_fini(rdev); 2273 radeon_bo_fini(rdev); 2274 radeon_atombios_fini(rdev); 2275 kfree(rdev->bios); 2276 rdev->bios = NULL; 2277 } 2278 2279 /* 2280 * vm 2281 */ 2282 int cayman_vm_init(struct radeon_device *rdev) 2283 { 2284 /* number of VMs */ 2285 rdev->vm_manager.nvm = 8; 2286 /* base offset of vram pages */ 2287 if (rdev->flags & RADEON_IS_IGP) { 2288 u64 tmp = RREG32(FUS_MC_VM_FB_OFFSET); 2289 tmp <<= 22; 2290 rdev->vm_manager.vram_base_offset = tmp; 2291 } else 2292 rdev->vm_manager.vram_base_offset = 0; 2293 return 0; 2294 } 2295 2296 void cayman_vm_fini(struct radeon_device *rdev) 2297 { 2298 } 2299 2300 #define R600_ENTRY_VALID (1 << 0) 2301 #define R600_PTE_SYSTEM (1 << 1) 2302 #define R600_PTE_SNOOPED (1 << 2) 2303 #define R600_PTE_READABLE (1 << 5) 2304 #define R600_PTE_WRITEABLE (1 << 6) 2305 2306 uint32_t cayman_vm_page_flags(struct radeon_device *rdev, uint32_t flags) 2307 { 2308 uint32_t r600_flags = 0; 2309 r600_flags |= (flags & RADEON_VM_PAGE_VALID) ? R600_ENTRY_VALID : 0; 2310 r600_flags |= (flags & RADEON_VM_PAGE_READABLE) ? R600_PTE_READABLE : 0; 2311 r600_flags |= (flags & RADEON_VM_PAGE_WRITEABLE) ? R600_PTE_WRITEABLE : 0; 2312 if (flags & RADEON_VM_PAGE_SYSTEM) { 2313 r600_flags |= R600_PTE_SYSTEM; 2314 r600_flags |= (flags & RADEON_VM_PAGE_SNOOPED) ? R600_PTE_SNOOPED : 0; 2315 } 2316 return r600_flags; 2317 } 2318 2319 /** 2320 * cayman_vm_set_page - update the page tables using the CP 2321 * 2322 * @rdev: radeon_device pointer 2323 * @ib: indirect buffer to fill with commands 2324 * @pe: addr of the page entry 2325 * @addr: dst addr to write into pe 2326 * @count: number of page entries to update 2327 * @incr: increase next addr by incr bytes 2328 * @flags: access flags 2329 * 2330 * Update the page tables using the CP (cayman/TN). 2331 */ 2332 void cayman_vm_set_page(struct radeon_device *rdev, 2333 struct radeon_ib *ib, 2334 uint64_t pe, 2335 uint64_t addr, unsigned count, 2336 uint32_t incr, uint32_t flags) 2337 { 2338 uint32_t r600_flags = cayman_vm_page_flags(rdev, flags); 2339 uint64_t value; 2340 unsigned ndw; 2341 2342 if (rdev->asic->vm.pt_ring_index == RADEON_RING_TYPE_GFX_INDEX) { 2343 while (count) { 2344 ndw = 1 + count * 2; 2345 if (ndw > 0x3FFF) 2346 ndw = 0x3FFF; 2347 2348 ib->ptr[ib->length_dw++] = PACKET3(PACKET3_ME_WRITE, ndw); 2349 ib->ptr[ib->length_dw++] = pe; 2350 ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff; 2351 for (; ndw > 1; ndw -= 2, --count, pe += 8) { 2352 if (flags & RADEON_VM_PAGE_SYSTEM) { 2353 value = radeon_vm_map_gart(rdev, addr); 2354 value &= 0xFFFFFFFFFFFFF000ULL; 2355 } else if (flags & RADEON_VM_PAGE_VALID) { 2356 value = addr; 2357 } else { 2358 value = 0; 2359 } 2360 addr += incr; 2361 value |= r600_flags; 2362 ib->ptr[ib->length_dw++] = value; 2363 ib->ptr[ib->length_dw++] = upper_32_bits(value); 2364 } 2365 } 2366 } else { 2367 if ((flags & RADEON_VM_PAGE_SYSTEM) || 2368 (count == 1)) { 2369 while (count) { 2370 ndw = count * 2; 2371 if (ndw > 0xFFFFE) 2372 ndw = 0xFFFFE; 2373 2374 /* for non-physically contiguous pages (system) */ 2375 ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, ndw); 2376 ib->ptr[ib->length_dw++] = pe; 2377 ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff; 2378 for (; ndw > 0; ndw -= 2, --count, pe += 8) { 2379 if (flags & RADEON_VM_PAGE_SYSTEM) { 2380 value = radeon_vm_map_gart(rdev, addr); 2381 value &= 0xFFFFFFFFFFFFF000ULL; 2382 } else if (flags & RADEON_VM_PAGE_VALID) { 2383 value = addr; 2384 } else { 2385 value = 0; 2386 } 2387 addr += incr; 2388 value |= r600_flags; 2389 ib->ptr[ib->length_dw++] = value; 2390 ib->ptr[ib->length_dw++] = upper_32_bits(value); 2391 } 2392 } 2393 while (ib->length_dw & 0x7) 2394 ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0); 2395 } else { 2396 while (count) { 2397 ndw = count * 2; 2398 if (ndw > 0xFFFFE) 2399 ndw = 0xFFFFE; 2400 2401 if (flags & RADEON_VM_PAGE_VALID) 2402 value = addr; 2403 else 2404 value = 0; 2405 /* for physically contiguous pages (vram) */ 2406 ib->ptr[ib->length_dw++] = DMA_PTE_PDE_PACKET(ndw); 2407 ib->ptr[ib->length_dw++] = pe; /* dst addr */ 2408 ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff; 2409 ib->ptr[ib->length_dw++] = r600_flags; /* mask */ 2410 ib->ptr[ib->length_dw++] = 0; 2411 ib->ptr[ib->length_dw++] = value; /* value */ 2412 ib->ptr[ib->length_dw++] = upper_32_bits(value); 2413 ib->ptr[ib->length_dw++] = incr; /* increment size */ 2414 ib->ptr[ib->length_dw++] = 0; 2415 pe += ndw * 4; 2416 addr += (ndw / 2) * incr; 2417 count -= ndw / 2; 2418 } 2419 } 2420 while (ib->length_dw & 0x7) 2421 ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0); 2422 } 2423 } 2424 2425 /** 2426 * cayman_vm_flush - vm flush using the CP 2427 * 2428 * @rdev: radeon_device pointer 2429 * 2430 * Update the page table base and flush the VM TLB 2431 * using the CP (cayman-si). 2432 */ 2433 void cayman_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm) 2434 { 2435 struct radeon_ring *ring = &rdev->ring[ridx]; 2436 2437 if (vm == NULL) 2438 return; 2439 2440 radeon_ring_write(ring, PACKET0(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2), 0)); 2441 radeon_ring_write(ring, vm->pd_gpu_addr >> 12); 2442 2443 /* flush hdp cache */ 2444 radeon_ring_write(ring, PACKET0(HDP_MEM_COHERENCY_FLUSH_CNTL, 0)); 2445 radeon_ring_write(ring, 0x1); 2446 2447 /* bits 0-7 are the VM contexts0-7 */ 2448 radeon_ring_write(ring, PACKET0(VM_INVALIDATE_REQUEST, 0)); 2449 radeon_ring_write(ring, 1 << vm->id); 2450 2451 /* sync PFP to ME, otherwise we might get invalid PFP reads */ 2452 radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0)); 2453 radeon_ring_write(ring, 0x0); 2454 } 2455 2456 void cayman_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm) 2457 { 2458 struct radeon_ring *ring = &rdev->ring[ridx]; 2459 2460 if (vm == NULL) 2461 return; 2462 2463 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0)); 2464 radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2)); 2465 radeon_ring_write(ring, vm->pd_gpu_addr >> 12); 2466 2467 /* flush hdp cache */ 2468 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0)); 2469 radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2)); 2470 radeon_ring_write(ring, 1); 2471 2472 /* bits 0-7 are the VM contexts0-7 */ 2473 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0)); 2474 radeon_ring_write(ring, (0xf << 16) | (VM_INVALIDATE_REQUEST >> 2)); 2475 radeon_ring_write(ring, 1 << vm->id); 2476 } 2477 2478