1 /* 2 * Copyright 2008 Advanced Micro Devices, Inc. 3 * Copyright 2008 Red Hat Inc. 4 * Copyright 2009 Jerome Glisse. 5 * 6 * Permission is hereby granted, free of charge, to any person obtaining a 7 * copy of this software and associated documentation files (the "Software"), 8 * to deal in the Software without restriction, including without limitation 9 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 10 * and/or sell copies of the Software, and to permit persons to whom the 11 * Software is furnished to do so, subject to the following conditions: 12 * 13 * The above copyright notice and this permission notice shall be included in 14 * all copies or substantial portions of the Software. 15 * 16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 19 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR 20 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 21 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 22 * OTHER DEALINGS IN THE SOFTWARE. 23 * 24 * Authors: Dave Airlie 25 * Alex Deucher 26 * Jerome Glisse 27 */ 28 #include <linux/firmware.h> 29 #include <linux/platform_device.h> 30 #include <linux/slab.h> 31 #include <drm/drmP.h> 32 #include "radeon.h" 33 #include "radeon_asic.h" 34 #include <drm/radeon_drm.h> 35 #include "rv770d.h" 36 #include "atom.h" 37 #include "avivod.h" 38 39 #define R700_PFP_UCODE_SIZE 848 40 #define R700_PM4_UCODE_SIZE 1360 41 42 static void rv770_gpu_init(struct radeon_device *rdev); 43 void rv770_fini(struct radeon_device *rdev); 44 static void rv770_pcie_gen2_enable(struct radeon_device *rdev); 45 46 u32 rv770_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base) 47 { 48 struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id]; 49 u32 tmp = RREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset); 50 int i; 51 52 /* Lock the graphics update lock */ 53 tmp |= AVIVO_D1GRPH_UPDATE_LOCK; 54 WREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset, tmp); 55 56 /* update the scanout addresses */ 57 if (radeon_crtc->crtc_id) { 58 WREG32(D2GRPH_SECONDARY_SURFACE_ADDRESS_HIGH, upper_32_bits(crtc_base)); 59 WREG32(D2GRPH_PRIMARY_SURFACE_ADDRESS_HIGH, upper_32_bits(crtc_base)); 60 } else { 61 WREG32(D1GRPH_SECONDARY_SURFACE_ADDRESS_HIGH, upper_32_bits(crtc_base)); 62 WREG32(D1GRPH_PRIMARY_SURFACE_ADDRESS_HIGH, upper_32_bits(crtc_base)); 63 } 64 WREG32(D1GRPH_SECONDARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset, 65 (u32)crtc_base); 66 WREG32(D1GRPH_PRIMARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset, 67 (u32)crtc_base); 68 69 /* Wait for update_pending to go high. */ 70 for (i = 0; i < rdev->usec_timeout; i++) { 71 if (RREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset) & AVIVO_D1GRPH_SURFACE_UPDATE_PENDING) 72 break; 73 udelay(1); 74 } 75 DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n"); 76 77 /* Unlock the lock, so double-buffering can take place inside vblank */ 78 tmp &= ~AVIVO_D1GRPH_UPDATE_LOCK; 79 WREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset, tmp); 80 81 /* Return current update_pending status: */ 82 return RREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset) & AVIVO_D1GRPH_SURFACE_UPDATE_PENDING; 83 } 84 85 /* get temperature in millidegrees */ 86 int rv770_get_temp(struct radeon_device *rdev) 87 { 88 u32 temp = (RREG32(CG_MULT_THERMAL_STATUS) & ASIC_T_MASK) >> 89 ASIC_T_SHIFT; 90 int actual_temp; 91 92 if (temp & 0x400) 93 actual_temp = -256; 94 else if (temp & 0x200) 95 actual_temp = 255; 96 else if (temp & 0x100) { 97 actual_temp = temp & 0x1ff; 98 actual_temp |= ~0x1ff; 99 } else 100 actual_temp = temp & 0xff; 101 102 return (actual_temp * 1000) / 2; 103 } 104 105 void rv770_pm_misc(struct radeon_device *rdev) 106 { 107 int req_ps_idx = rdev->pm.requested_power_state_index; 108 int req_cm_idx = rdev->pm.requested_clock_mode_index; 109 struct radeon_power_state *ps = &rdev->pm.power_state[req_ps_idx]; 110 struct radeon_voltage *voltage = &ps->clock_info[req_cm_idx].voltage; 111 112 if ((voltage->type == VOLTAGE_SW) && voltage->voltage) { 113 /* 0xff01 is a flag rather then an actual voltage */ 114 if (voltage->voltage == 0xff01) 115 return; 116 if (voltage->voltage != rdev->pm.current_vddc) { 117 radeon_atom_set_voltage(rdev, voltage->voltage, SET_VOLTAGE_TYPE_ASIC_VDDC); 118 rdev->pm.current_vddc = voltage->voltage; 119 DRM_DEBUG("Setting: v: %d\n", voltage->voltage); 120 } 121 } 122 } 123 124 /* 125 * GART 126 */ 127 static int rv770_pcie_gart_enable(struct radeon_device *rdev) 128 { 129 u32 tmp; 130 int r, i; 131 132 if (rdev->gart.robj == NULL) { 133 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n"); 134 return -EINVAL; 135 } 136 r = radeon_gart_table_vram_pin(rdev); 137 if (r) 138 return r; 139 radeon_gart_restore(rdev); 140 /* Setup L2 cache */ 141 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING | 142 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE | 143 EFFECTIVE_L2_QUEUE_SIZE(7)); 144 WREG32(VM_L2_CNTL2, 0); 145 WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2)); 146 /* Setup TLB control */ 147 tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING | 148 SYSTEM_ACCESS_MODE_NOT_IN_SYS | 149 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU | 150 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5); 151 WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp); 152 WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp); 153 WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp); 154 if (rdev->family == CHIP_RV740) 155 WREG32(MC_VM_MD_L1_TLB3_CNTL, tmp); 156 WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp); 157 WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp); 158 WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp); 159 WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp); 160 WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12); 161 WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12); 162 WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12); 163 WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) | 164 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT); 165 WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR, 166 (u32)(rdev->dummy_page.addr >> 12)); 167 for (i = 1; i < 7; i++) 168 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0); 169 170 r600_pcie_gart_tlb_flush(rdev); 171 DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n", 172 (unsigned)(rdev->mc.gtt_size >> 20), 173 (unsigned long long)rdev->gart.table_addr); 174 rdev->gart.ready = true; 175 return 0; 176 } 177 178 static void rv770_pcie_gart_disable(struct radeon_device *rdev) 179 { 180 u32 tmp; 181 int i; 182 183 /* Disable all tables */ 184 for (i = 0; i < 7; i++) 185 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0); 186 187 /* Setup L2 cache */ 188 WREG32(VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING | 189 EFFECTIVE_L2_QUEUE_SIZE(7)); 190 WREG32(VM_L2_CNTL2, 0); 191 WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2)); 192 /* Setup TLB control */ 193 tmp = EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5); 194 WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp); 195 WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp); 196 WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp); 197 WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp); 198 WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp); 199 WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp); 200 WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp); 201 radeon_gart_table_vram_unpin(rdev); 202 } 203 204 static void rv770_pcie_gart_fini(struct radeon_device *rdev) 205 { 206 radeon_gart_fini(rdev); 207 rv770_pcie_gart_disable(rdev); 208 radeon_gart_table_vram_free(rdev); 209 } 210 211 212 static void rv770_agp_enable(struct radeon_device *rdev) 213 { 214 u32 tmp; 215 int i; 216 217 /* Setup L2 cache */ 218 WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING | 219 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE | 220 EFFECTIVE_L2_QUEUE_SIZE(7)); 221 WREG32(VM_L2_CNTL2, 0); 222 WREG32(VM_L2_CNTL3, BANK_SELECT(0) | CACHE_UPDATE_MODE(2)); 223 /* Setup TLB control */ 224 tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING | 225 SYSTEM_ACCESS_MODE_NOT_IN_SYS | 226 SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU | 227 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5); 228 WREG32(MC_VM_MD_L1_TLB0_CNTL, tmp); 229 WREG32(MC_VM_MD_L1_TLB1_CNTL, tmp); 230 WREG32(MC_VM_MD_L1_TLB2_CNTL, tmp); 231 WREG32(MC_VM_MB_L1_TLB0_CNTL, tmp); 232 WREG32(MC_VM_MB_L1_TLB1_CNTL, tmp); 233 WREG32(MC_VM_MB_L1_TLB2_CNTL, tmp); 234 WREG32(MC_VM_MB_L1_TLB3_CNTL, tmp); 235 for (i = 0; i < 7; i++) 236 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0); 237 } 238 239 static void rv770_mc_program(struct radeon_device *rdev) 240 { 241 struct rv515_mc_save save; 242 u32 tmp; 243 int i, j; 244 245 /* Initialize HDP */ 246 for (i = 0, j = 0; i < 32; i++, j += 0x18) { 247 WREG32((0x2c14 + j), 0x00000000); 248 WREG32((0x2c18 + j), 0x00000000); 249 WREG32((0x2c1c + j), 0x00000000); 250 WREG32((0x2c20 + j), 0x00000000); 251 WREG32((0x2c24 + j), 0x00000000); 252 } 253 /* r7xx hw bug. Read from HDP_DEBUG1 rather 254 * than writing to HDP_REG_COHERENCY_FLUSH_CNTL 255 */ 256 tmp = RREG32(HDP_DEBUG1); 257 258 rv515_mc_stop(rdev, &save); 259 if (r600_mc_wait_for_idle(rdev)) { 260 dev_warn(rdev->dev, "Wait for MC idle timedout !\n"); 261 } 262 /* Lockout access through VGA aperture*/ 263 WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE); 264 /* Update configuration */ 265 if (rdev->flags & RADEON_IS_AGP) { 266 if (rdev->mc.vram_start < rdev->mc.gtt_start) { 267 /* VRAM before AGP */ 268 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR, 269 rdev->mc.vram_start >> 12); 270 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR, 271 rdev->mc.gtt_end >> 12); 272 } else { 273 /* VRAM after AGP */ 274 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR, 275 rdev->mc.gtt_start >> 12); 276 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR, 277 rdev->mc.vram_end >> 12); 278 } 279 } else { 280 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR, 281 rdev->mc.vram_start >> 12); 282 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR, 283 rdev->mc.vram_end >> 12); 284 } 285 WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, rdev->vram_scratch.gpu_addr >> 12); 286 tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16; 287 tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF); 288 WREG32(MC_VM_FB_LOCATION, tmp); 289 WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8)); 290 WREG32(HDP_NONSURFACE_INFO, (2 << 7)); 291 WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF); 292 if (rdev->flags & RADEON_IS_AGP) { 293 WREG32(MC_VM_AGP_TOP, rdev->mc.gtt_end >> 16); 294 WREG32(MC_VM_AGP_BOT, rdev->mc.gtt_start >> 16); 295 WREG32(MC_VM_AGP_BASE, rdev->mc.agp_base >> 22); 296 } else { 297 WREG32(MC_VM_AGP_BASE, 0); 298 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF); 299 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF); 300 } 301 if (r600_mc_wait_for_idle(rdev)) { 302 dev_warn(rdev->dev, "Wait for MC idle timedout !\n"); 303 } 304 rv515_mc_resume(rdev, &save); 305 /* we need to own VRAM, so turn off the VGA renderer here 306 * to stop it overwriting our objects */ 307 rv515_vga_render_disable(rdev); 308 } 309 310 311 /* 312 * CP. 313 */ 314 void r700_cp_stop(struct radeon_device *rdev) 315 { 316 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size); 317 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT)); 318 WREG32(SCRATCH_UMSK, 0); 319 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false; 320 } 321 322 static int rv770_cp_load_microcode(struct radeon_device *rdev) 323 { 324 const __be32 *fw_data; 325 int i; 326 327 if (!rdev->me_fw || !rdev->pfp_fw) 328 return -EINVAL; 329 330 r700_cp_stop(rdev); 331 WREG32(CP_RB_CNTL, 332 #ifdef __BIG_ENDIAN 333 BUF_SWAP_32BIT | 334 #endif 335 RB_NO_UPDATE | RB_BLKSZ(15) | RB_BUFSZ(3)); 336 337 /* Reset cp */ 338 WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP); 339 RREG32(GRBM_SOFT_RESET); 340 mdelay(15); 341 WREG32(GRBM_SOFT_RESET, 0); 342 343 fw_data = (const __be32 *)rdev->pfp_fw->data; 344 WREG32(CP_PFP_UCODE_ADDR, 0); 345 for (i = 0; i < R700_PFP_UCODE_SIZE; i++) 346 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++)); 347 WREG32(CP_PFP_UCODE_ADDR, 0); 348 349 fw_data = (const __be32 *)rdev->me_fw->data; 350 WREG32(CP_ME_RAM_WADDR, 0); 351 for (i = 0; i < R700_PM4_UCODE_SIZE; i++) 352 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++)); 353 354 WREG32(CP_PFP_UCODE_ADDR, 0); 355 WREG32(CP_ME_RAM_WADDR, 0); 356 WREG32(CP_ME_RAM_RADDR, 0); 357 return 0; 358 } 359 360 void r700_cp_fini(struct radeon_device *rdev) 361 { 362 struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 363 r700_cp_stop(rdev); 364 radeon_ring_fini(rdev, ring); 365 radeon_scratch_free(rdev, ring->rptr_save_reg); 366 } 367 368 /* 369 * Core functions 370 */ 371 static void rv770_gpu_init(struct radeon_device *rdev) 372 { 373 int i, j, num_qd_pipes; 374 u32 ta_aux_cntl; 375 u32 sx_debug_1; 376 u32 smx_dc_ctl0; 377 u32 db_debug3; 378 u32 num_gs_verts_per_thread; 379 u32 vgt_gs_per_es; 380 u32 gs_prim_buffer_depth = 0; 381 u32 sq_ms_fifo_sizes; 382 u32 sq_config; 383 u32 sq_thread_resource_mgmt; 384 u32 hdp_host_path_cntl; 385 u32 sq_dyn_gpr_size_simd_ab_0; 386 u32 gb_tiling_config = 0; 387 u32 cc_rb_backend_disable = 0; 388 u32 cc_gc_shader_pipe_config = 0; 389 u32 mc_arb_ramcfg; 390 u32 db_debug4, tmp; 391 u32 inactive_pipes, shader_pipe_config; 392 u32 disabled_rb_mask; 393 unsigned active_number; 394 395 /* setup chip specs */ 396 rdev->config.rv770.tiling_group_size = 256; 397 switch (rdev->family) { 398 case CHIP_RV770: 399 rdev->config.rv770.max_pipes = 4; 400 rdev->config.rv770.max_tile_pipes = 8; 401 rdev->config.rv770.max_simds = 10; 402 rdev->config.rv770.max_backends = 4; 403 rdev->config.rv770.max_gprs = 256; 404 rdev->config.rv770.max_threads = 248; 405 rdev->config.rv770.max_stack_entries = 512; 406 rdev->config.rv770.max_hw_contexts = 8; 407 rdev->config.rv770.max_gs_threads = 16 * 2; 408 rdev->config.rv770.sx_max_export_size = 128; 409 rdev->config.rv770.sx_max_export_pos_size = 16; 410 rdev->config.rv770.sx_max_export_smx_size = 112; 411 rdev->config.rv770.sq_num_cf_insts = 2; 412 413 rdev->config.rv770.sx_num_of_sets = 7; 414 rdev->config.rv770.sc_prim_fifo_size = 0xF9; 415 rdev->config.rv770.sc_hiz_tile_fifo_size = 0x30; 416 rdev->config.rv770.sc_earlyz_tile_fifo_fize = 0x130; 417 break; 418 case CHIP_RV730: 419 rdev->config.rv770.max_pipes = 2; 420 rdev->config.rv770.max_tile_pipes = 4; 421 rdev->config.rv770.max_simds = 8; 422 rdev->config.rv770.max_backends = 2; 423 rdev->config.rv770.max_gprs = 128; 424 rdev->config.rv770.max_threads = 248; 425 rdev->config.rv770.max_stack_entries = 256; 426 rdev->config.rv770.max_hw_contexts = 8; 427 rdev->config.rv770.max_gs_threads = 16 * 2; 428 rdev->config.rv770.sx_max_export_size = 256; 429 rdev->config.rv770.sx_max_export_pos_size = 32; 430 rdev->config.rv770.sx_max_export_smx_size = 224; 431 rdev->config.rv770.sq_num_cf_insts = 2; 432 433 rdev->config.rv770.sx_num_of_sets = 7; 434 rdev->config.rv770.sc_prim_fifo_size = 0xf9; 435 rdev->config.rv770.sc_hiz_tile_fifo_size = 0x30; 436 rdev->config.rv770.sc_earlyz_tile_fifo_fize = 0x130; 437 if (rdev->config.rv770.sx_max_export_pos_size > 16) { 438 rdev->config.rv770.sx_max_export_pos_size -= 16; 439 rdev->config.rv770.sx_max_export_smx_size += 16; 440 } 441 break; 442 case CHIP_RV710: 443 rdev->config.rv770.max_pipes = 2; 444 rdev->config.rv770.max_tile_pipes = 2; 445 rdev->config.rv770.max_simds = 2; 446 rdev->config.rv770.max_backends = 1; 447 rdev->config.rv770.max_gprs = 256; 448 rdev->config.rv770.max_threads = 192; 449 rdev->config.rv770.max_stack_entries = 256; 450 rdev->config.rv770.max_hw_contexts = 4; 451 rdev->config.rv770.max_gs_threads = 8 * 2; 452 rdev->config.rv770.sx_max_export_size = 128; 453 rdev->config.rv770.sx_max_export_pos_size = 16; 454 rdev->config.rv770.sx_max_export_smx_size = 112; 455 rdev->config.rv770.sq_num_cf_insts = 1; 456 457 rdev->config.rv770.sx_num_of_sets = 7; 458 rdev->config.rv770.sc_prim_fifo_size = 0x40; 459 rdev->config.rv770.sc_hiz_tile_fifo_size = 0x30; 460 rdev->config.rv770.sc_earlyz_tile_fifo_fize = 0x130; 461 break; 462 case CHIP_RV740: 463 rdev->config.rv770.max_pipes = 4; 464 rdev->config.rv770.max_tile_pipes = 4; 465 rdev->config.rv770.max_simds = 8; 466 rdev->config.rv770.max_backends = 4; 467 rdev->config.rv770.max_gprs = 256; 468 rdev->config.rv770.max_threads = 248; 469 rdev->config.rv770.max_stack_entries = 512; 470 rdev->config.rv770.max_hw_contexts = 8; 471 rdev->config.rv770.max_gs_threads = 16 * 2; 472 rdev->config.rv770.sx_max_export_size = 256; 473 rdev->config.rv770.sx_max_export_pos_size = 32; 474 rdev->config.rv770.sx_max_export_smx_size = 224; 475 rdev->config.rv770.sq_num_cf_insts = 2; 476 477 rdev->config.rv770.sx_num_of_sets = 7; 478 rdev->config.rv770.sc_prim_fifo_size = 0x100; 479 rdev->config.rv770.sc_hiz_tile_fifo_size = 0x30; 480 rdev->config.rv770.sc_earlyz_tile_fifo_fize = 0x130; 481 482 if (rdev->config.rv770.sx_max_export_pos_size > 16) { 483 rdev->config.rv770.sx_max_export_pos_size -= 16; 484 rdev->config.rv770.sx_max_export_smx_size += 16; 485 } 486 break; 487 default: 488 break; 489 } 490 491 /* Initialize HDP */ 492 j = 0; 493 for (i = 0; i < 32; i++) { 494 WREG32((0x2c14 + j), 0x00000000); 495 WREG32((0x2c18 + j), 0x00000000); 496 WREG32((0x2c1c + j), 0x00000000); 497 WREG32((0x2c20 + j), 0x00000000); 498 WREG32((0x2c24 + j), 0x00000000); 499 j += 0x18; 500 } 501 502 WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff)); 503 504 /* setup tiling, simd, pipe config */ 505 mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG); 506 507 shader_pipe_config = RREG32(CC_GC_SHADER_PIPE_CONFIG); 508 inactive_pipes = (shader_pipe_config & INACTIVE_QD_PIPES_MASK) >> INACTIVE_QD_PIPES_SHIFT; 509 for (i = 0, tmp = 1, active_number = 0; i < R7XX_MAX_PIPES; i++) { 510 if (!(inactive_pipes & tmp)) { 511 active_number++; 512 } 513 tmp <<= 1; 514 } 515 if (active_number == 1) { 516 WREG32(SPI_CONFIG_CNTL, DISABLE_INTERP_1); 517 } else { 518 WREG32(SPI_CONFIG_CNTL, 0); 519 } 520 521 cc_rb_backend_disable = RREG32(CC_RB_BACKEND_DISABLE) & 0x00ff0000; 522 tmp = R7XX_MAX_BACKENDS - r600_count_pipe_bits(cc_rb_backend_disable >> 16); 523 if (tmp < rdev->config.rv770.max_backends) { 524 rdev->config.rv770.max_backends = tmp; 525 } 526 527 cc_gc_shader_pipe_config = RREG32(CC_GC_SHADER_PIPE_CONFIG) & 0xffffff00; 528 tmp = R7XX_MAX_PIPES - r600_count_pipe_bits((cc_gc_shader_pipe_config >> 8) & R7XX_MAX_PIPES_MASK); 529 if (tmp < rdev->config.rv770.max_pipes) { 530 rdev->config.rv770.max_pipes = tmp; 531 } 532 tmp = R7XX_MAX_SIMDS - r600_count_pipe_bits((cc_gc_shader_pipe_config >> 16) & R7XX_MAX_SIMDS_MASK); 533 if (tmp < rdev->config.rv770.max_simds) { 534 rdev->config.rv770.max_simds = tmp; 535 } 536 537 switch (rdev->config.rv770.max_tile_pipes) { 538 case 1: 539 default: 540 gb_tiling_config = PIPE_TILING(0); 541 break; 542 case 2: 543 gb_tiling_config = PIPE_TILING(1); 544 break; 545 case 4: 546 gb_tiling_config = PIPE_TILING(2); 547 break; 548 case 8: 549 gb_tiling_config = PIPE_TILING(3); 550 break; 551 } 552 rdev->config.rv770.tiling_npipes = rdev->config.rv770.max_tile_pipes; 553 554 disabled_rb_mask = (RREG32(CC_RB_BACKEND_DISABLE) >> 16) & R7XX_MAX_BACKENDS_MASK; 555 tmp = (gb_tiling_config & PIPE_TILING__MASK) >> PIPE_TILING__SHIFT; 556 tmp = r6xx_remap_render_backend(rdev, tmp, rdev->config.rv770.max_backends, 557 R7XX_MAX_BACKENDS, disabled_rb_mask); 558 gb_tiling_config |= tmp << 16; 559 rdev->config.rv770.backend_map = tmp; 560 561 if (rdev->family == CHIP_RV770) 562 gb_tiling_config |= BANK_TILING(1); 563 else { 564 if ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) 565 gb_tiling_config |= BANK_TILING(1); 566 else 567 gb_tiling_config |= BANK_TILING(0); 568 } 569 rdev->config.rv770.tiling_nbanks = 4 << ((gb_tiling_config >> 4) & 0x3); 570 gb_tiling_config |= GROUP_SIZE((mc_arb_ramcfg & BURSTLENGTH_MASK) >> BURSTLENGTH_SHIFT); 571 if (((mc_arb_ramcfg & NOOFROWS_MASK) >> NOOFROWS_SHIFT) > 3) { 572 gb_tiling_config |= ROW_TILING(3); 573 gb_tiling_config |= SAMPLE_SPLIT(3); 574 } else { 575 gb_tiling_config |= 576 ROW_TILING(((mc_arb_ramcfg & NOOFROWS_MASK) >> NOOFROWS_SHIFT)); 577 gb_tiling_config |= 578 SAMPLE_SPLIT(((mc_arb_ramcfg & NOOFROWS_MASK) >> NOOFROWS_SHIFT)); 579 } 580 581 gb_tiling_config |= BANK_SWAPS(1); 582 rdev->config.rv770.tile_config = gb_tiling_config; 583 584 WREG32(GB_TILING_CONFIG, gb_tiling_config); 585 WREG32(DCP_TILING_CONFIG, (gb_tiling_config & 0xffff)); 586 WREG32(HDP_TILING_CONFIG, (gb_tiling_config & 0xffff)); 587 WREG32(DMA_TILING_CONFIG, (gb_tiling_config & 0xffff)); 588 WREG32(DMA_TILING_CONFIG2, (gb_tiling_config & 0xffff)); 589 590 WREG32(CGTS_SYS_TCC_DISABLE, 0); 591 WREG32(CGTS_TCC_DISABLE, 0); 592 WREG32(CGTS_USER_SYS_TCC_DISABLE, 0); 593 WREG32(CGTS_USER_TCC_DISABLE, 0); 594 595 596 num_qd_pipes = R7XX_MAX_PIPES - r600_count_pipe_bits((cc_gc_shader_pipe_config & INACTIVE_QD_PIPES_MASK) >> 8); 597 WREG32(VGT_OUT_DEALLOC_CNTL, (num_qd_pipes * 4) & DEALLOC_DIST_MASK); 598 WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL, ((num_qd_pipes * 4) - 2) & VTX_REUSE_DEPTH_MASK); 599 600 /* set HW defaults for 3D engine */ 601 WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) | 602 ROQ_IB2_START(0x2b))); 603 604 WREG32(CP_MEQ_THRESHOLDS, STQ_SPLIT(0x30)); 605 606 ta_aux_cntl = RREG32(TA_CNTL_AUX); 607 WREG32(TA_CNTL_AUX, ta_aux_cntl | DISABLE_CUBE_ANISO); 608 609 sx_debug_1 = RREG32(SX_DEBUG_1); 610 sx_debug_1 |= ENABLE_NEW_SMX_ADDRESS; 611 WREG32(SX_DEBUG_1, sx_debug_1); 612 613 smx_dc_ctl0 = RREG32(SMX_DC_CTL0); 614 smx_dc_ctl0 &= ~CACHE_DEPTH(0x1ff); 615 smx_dc_ctl0 |= CACHE_DEPTH((rdev->config.rv770.sx_num_of_sets * 64) - 1); 616 WREG32(SMX_DC_CTL0, smx_dc_ctl0); 617 618 if (rdev->family != CHIP_RV740) 619 WREG32(SMX_EVENT_CTL, (ES_FLUSH_CTL(4) | 620 GS_FLUSH_CTL(4) | 621 ACK_FLUSH_CTL(3) | 622 SYNC_FLUSH_CTL)); 623 624 if (rdev->family != CHIP_RV770) 625 WREG32(SMX_SAR_CTL0, 0x00003f3f); 626 627 db_debug3 = RREG32(DB_DEBUG3); 628 db_debug3 &= ~DB_CLK_OFF_DELAY(0x1f); 629 switch (rdev->family) { 630 case CHIP_RV770: 631 case CHIP_RV740: 632 db_debug3 |= DB_CLK_OFF_DELAY(0x1f); 633 break; 634 case CHIP_RV710: 635 case CHIP_RV730: 636 default: 637 db_debug3 |= DB_CLK_OFF_DELAY(2); 638 break; 639 } 640 WREG32(DB_DEBUG3, db_debug3); 641 642 if (rdev->family != CHIP_RV770) { 643 db_debug4 = RREG32(DB_DEBUG4); 644 db_debug4 |= DISABLE_TILE_COVERED_FOR_PS_ITER; 645 WREG32(DB_DEBUG4, db_debug4); 646 } 647 648 WREG32(SX_EXPORT_BUFFER_SIZES, (COLOR_BUFFER_SIZE((rdev->config.rv770.sx_max_export_size / 4) - 1) | 649 POSITION_BUFFER_SIZE((rdev->config.rv770.sx_max_export_pos_size / 4) - 1) | 650 SMX_BUFFER_SIZE((rdev->config.rv770.sx_max_export_smx_size / 4) - 1))); 651 652 WREG32(PA_SC_FIFO_SIZE, (SC_PRIM_FIFO_SIZE(rdev->config.rv770.sc_prim_fifo_size) | 653 SC_HIZ_TILE_FIFO_SIZE(rdev->config.rv770.sc_hiz_tile_fifo_size) | 654 SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.rv770.sc_earlyz_tile_fifo_fize))); 655 656 WREG32(PA_SC_MULTI_CHIP_CNTL, 0); 657 658 WREG32(VGT_NUM_INSTANCES, 1); 659 660 WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4)); 661 662 WREG32(CP_PERFMON_CNTL, 0); 663 664 sq_ms_fifo_sizes = (CACHE_FIFO_SIZE(16 * rdev->config.rv770.sq_num_cf_insts) | 665 DONE_FIFO_HIWATER(0xe0) | 666 ALU_UPDATE_FIFO_HIWATER(0x8)); 667 switch (rdev->family) { 668 case CHIP_RV770: 669 case CHIP_RV730: 670 case CHIP_RV710: 671 sq_ms_fifo_sizes |= FETCH_FIFO_HIWATER(0x1); 672 break; 673 case CHIP_RV740: 674 default: 675 sq_ms_fifo_sizes |= FETCH_FIFO_HIWATER(0x4); 676 break; 677 } 678 WREG32(SQ_MS_FIFO_SIZES, sq_ms_fifo_sizes); 679 680 /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT 681 * should be adjusted as needed by the 2D/3D drivers. This just sets default values 682 */ 683 sq_config = RREG32(SQ_CONFIG); 684 sq_config &= ~(PS_PRIO(3) | 685 VS_PRIO(3) | 686 GS_PRIO(3) | 687 ES_PRIO(3)); 688 sq_config |= (DX9_CONSTS | 689 VC_ENABLE | 690 EXPORT_SRC_C | 691 PS_PRIO(0) | 692 VS_PRIO(1) | 693 GS_PRIO(2) | 694 ES_PRIO(3)); 695 if (rdev->family == CHIP_RV710) 696 /* no vertex cache */ 697 sq_config &= ~VC_ENABLE; 698 699 WREG32(SQ_CONFIG, sq_config); 700 701 WREG32(SQ_GPR_RESOURCE_MGMT_1, (NUM_PS_GPRS((rdev->config.rv770.max_gprs * 24)/64) | 702 NUM_VS_GPRS((rdev->config.rv770.max_gprs * 24)/64) | 703 NUM_CLAUSE_TEMP_GPRS(((rdev->config.rv770.max_gprs * 24)/64)/2))); 704 705 WREG32(SQ_GPR_RESOURCE_MGMT_2, (NUM_GS_GPRS((rdev->config.rv770.max_gprs * 7)/64) | 706 NUM_ES_GPRS((rdev->config.rv770.max_gprs * 7)/64))); 707 708 sq_thread_resource_mgmt = (NUM_PS_THREADS((rdev->config.rv770.max_threads * 4)/8) | 709 NUM_VS_THREADS((rdev->config.rv770.max_threads * 2)/8) | 710 NUM_ES_THREADS((rdev->config.rv770.max_threads * 1)/8)); 711 if (((rdev->config.rv770.max_threads * 1) / 8) > rdev->config.rv770.max_gs_threads) 712 sq_thread_resource_mgmt |= NUM_GS_THREADS(rdev->config.rv770.max_gs_threads); 713 else 714 sq_thread_resource_mgmt |= NUM_GS_THREADS((rdev->config.rv770.max_gs_threads * 1)/8); 715 WREG32(SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt); 716 717 WREG32(SQ_STACK_RESOURCE_MGMT_1, (NUM_PS_STACK_ENTRIES((rdev->config.rv770.max_stack_entries * 1)/4) | 718 NUM_VS_STACK_ENTRIES((rdev->config.rv770.max_stack_entries * 1)/4))); 719 720 WREG32(SQ_STACK_RESOURCE_MGMT_2, (NUM_GS_STACK_ENTRIES((rdev->config.rv770.max_stack_entries * 1)/4) | 721 NUM_ES_STACK_ENTRIES((rdev->config.rv770.max_stack_entries * 1)/4))); 722 723 sq_dyn_gpr_size_simd_ab_0 = (SIMDA_RING0((rdev->config.rv770.max_gprs * 38)/64) | 724 SIMDA_RING1((rdev->config.rv770.max_gprs * 38)/64) | 725 SIMDB_RING0((rdev->config.rv770.max_gprs * 38)/64) | 726 SIMDB_RING1((rdev->config.rv770.max_gprs * 38)/64)); 727 728 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_0, sq_dyn_gpr_size_simd_ab_0); 729 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_1, sq_dyn_gpr_size_simd_ab_0); 730 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_2, sq_dyn_gpr_size_simd_ab_0); 731 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_3, sq_dyn_gpr_size_simd_ab_0); 732 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_4, sq_dyn_gpr_size_simd_ab_0); 733 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_5, sq_dyn_gpr_size_simd_ab_0); 734 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_6, sq_dyn_gpr_size_simd_ab_0); 735 WREG32(SQ_DYN_GPR_SIZE_SIMD_AB_7, sq_dyn_gpr_size_simd_ab_0); 736 737 WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) | 738 FORCE_EOV_MAX_REZ_CNT(255))); 739 740 if (rdev->family == CHIP_RV710) 741 WREG32(VGT_CACHE_INVALIDATION, (CACHE_INVALIDATION(TC_ONLY) | 742 AUTO_INVLD_EN(ES_AND_GS_AUTO))); 743 else 744 WREG32(VGT_CACHE_INVALIDATION, (CACHE_INVALIDATION(VC_AND_TC) | 745 AUTO_INVLD_EN(ES_AND_GS_AUTO))); 746 747 switch (rdev->family) { 748 case CHIP_RV770: 749 case CHIP_RV730: 750 case CHIP_RV740: 751 gs_prim_buffer_depth = 384; 752 break; 753 case CHIP_RV710: 754 gs_prim_buffer_depth = 128; 755 break; 756 default: 757 break; 758 } 759 760 num_gs_verts_per_thread = rdev->config.rv770.max_pipes * 16; 761 vgt_gs_per_es = gs_prim_buffer_depth + num_gs_verts_per_thread; 762 /* Max value for this is 256 */ 763 if (vgt_gs_per_es > 256) 764 vgt_gs_per_es = 256; 765 766 WREG32(VGT_ES_PER_GS, 128); 767 WREG32(VGT_GS_PER_ES, vgt_gs_per_es); 768 WREG32(VGT_GS_PER_VS, 2); 769 770 /* more default values. 2D/3D driver should adjust as needed */ 771 WREG32(VGT_GS_VERTEX_REUSE, 16); 772 WREG32(PA_SC_LINE_STIPPLE_STATE, 0); 773 WREG32(VGT_STRMOUT_EN, 0); 774 WREG32(SX_MISC, 0); 775 WREG32(PA_SC_MODE_CNTL, 0); 776 WREG32(PA_SC_EDGERULE, 0xaaaaaaaa); 777 WREG32(PA_SC_AA_CONFIG, 0); 778 WREG32(PA_SC_CLIPRECT_RULE, 0xffff); 779 WREG32(PA_SC_LINE_STIPPLE, 0); 780 WREG32(SPI_INPUT_Z, 0); 781 WREG32(SPI_PS_IN_CONTROL_0, NUM_INTERP(2)); 782 WREG32(CB_COLOR7_FRAG, 0); 783 784 /* clear render buffer base addresses */ 785 WREG32(CB_COLOR0_BASE, 0); 786 WREG32(CB_COLOR1_BASE, 0); 787 WREG32(CB_COLOR2_BASE, 0); 788 WREG32(CB_COLOR3_BASE, 0); 789 WREG32(CB_COLOR4_BASE, 0); 790 WREG32(CB_COLOR5_BASE, 0); 791 WREG32(CB_COLOR6_BASE, 0); 792 WREG32(CB_COLOR7_BASE, 0); 793 794 WREG32(TCP_CNTL, 0); 795 796 hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL); 797 WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl); 798 799 WREG32(PA_SC_MULTI_CHIP_CNTL, 0); 800 801 WREG32(PA_CL_ENHANCE, (CLIP_VTX_REORDER_ENA | 802 NUM_CLIP_SEQ(3))); 803 WREG32(VC_ENHANCE, 0); 804 } 805 806 void r700_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc) 807 { 808 u64 size_bf, size_af; 809 810 if (mc->mc_vram_size > 0xE0000000) { 811 /* leave room for at least 512M GTT */ 812 dev_warn(rdev->dev, "limiting VRAM\n"); 813 mc->real_vram_size = 0xE0000000; 814 mc->mc_vram_size = 0xE0000000; 815 } 816 if (rdev->flags & RADEON_IS_AGP) { 817 size_bf = mc->gtt_start; 818 size_af = 0xFFFFFFFF - mc->gtt_end; 819 if (size_bf > size_af) { 820 if (mc->mc_vram_size > size_bf) { 821 dev_warn(rdev->dev, "limiting VRAM\n"); 822 mc->real_vram_size = size_bf; 823 mc->mc_vram_size = size_bf; 824 } 825 mc->vram_start = mc->gtt_start - mc->mc_vram_size; 826 } else { 827 if (mc->mc_vram_size > size_af) { 828 dev_warn(rdev->dev, "limiting VRAM\n"); 829 mc->real_vram_size = size_af; 830 mc->mc_vram_size = size_af; 831 } 832 mc->vram_start = mc->gtt_end + 1; 833 } 834 mc->vram_end = mc->vram_start + mc->mc_vram_size - 1; 835 dev_info(rdev->dev, "VRAM: %lluM 0x%08llX - 0x%08llX (%lluM used)\n", 836 mc->mc_vram_size >> 20, mc->vram_start, 837 mc->vram_end, mc->real_vram_size >> 20); 838 } else { 839 radeon_vram_location(rdev, &rdev->mc, 0); 840 rdev->mc.gtt_base_align = 0; 841 radeon_gtt_location(rdev, mc); 842 } 843 } 844 845 static int rv770_mc_init(struct radeon_device *rdev) 846 { 847 u32 tmp; 848 int chansize, numchan; 849 850 /* Get VRAM informations */ 851 rdev->mc.vram_is_ddr = true; 852 tmp = RREG32(MC_ARB_RAMCFG); 853 if (tmp & CHANSIZE_OVERRIDE) { 854 chansize = 16; 855 } else if (tmp & CHANSIZE_MASK) { 856 chansize = 64; 857 } else { 858 chansize = 32; 859 } 860 tmp = RREG32(MC_SHARED_CHMAP); 861 switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) { 862 case 0: 863 default: 864 numchan = 1; 865 break; 866 case 1: 867 numchan = 2; 868 break; 869 case 2: 870 numchan = 4; 871 break; 872 case 3: 873 numchan = 8; 874 break; 875 } 876 rdev->mc.vram_width = numchan * chansize; 877 /* Could aper size report 0 ? */ 878 rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0); 879 rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0); 880 /* Setup GPU memory space */ 881 rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE); 882 rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE); 883 rdev->mc.visible_vram_size = rdev->mc.aper_size; 884 r700_vram_gtt_location(rdev, &rdev->mc); 885 radeon_update_bandwidth_info(rdev); 886 887 return 0; 888 } 889 890 /** 891 * rv770_copy_dma - copy pages using the DMA engine 892 * 893 * @rdev: radeon_device pointer 894 * @src_offset: src GPU address 895 * @dst_offset: dst GPU address 896 * @num_gpu_pages: number of GPU pages to xfer 897 * @fence: radeon fence object 898 * 899 * Copy GPU paging using the DMA engine (r7xx). 900 * Used by the radeon ttm implementation to move pages if 901 * registered as the asic copy callback. 902 */ 903 int rv770_copy_dma(struct radeon_device *rdev, 904 uint64_t src_offset, uint64_t dst_offset, 905 unsigned num_gpu_pages, 906 struct radeon_fence **fence) 907 { 908 struct radeon_semaphore *sem = NULL; 909 int ring_index = rdev->asic->copy.dma_ring_index; 910 struct radeon_ring *ring = &rdev->ring[ring_index]; 911 u32 size_in_dw, cur_size_in_dw; 912 int i, num_loops; 913 int r = 0; 914 915 r = radeon_semaphore_create(rdev, &sem); 916 if (r) { 917 DRM_ERROR("radeon: moving bo (%d).\n", r); 918 return r; 919 } 920 921 size_in_dw = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT) / 4; 922 num_loops = DIV_ROUND_UP(size_in_dw, 0xFFFF); 923 r = radeon_ring_lock(rdev, ring, num_loops * 5 + 8); 924 if (r) { 925 DRM_ERROR("radeon: moving bo (%d).\n", r); 926 radeon_semaphore_free(rdev, &sem, NULL); 927 return r; 928 } 929 930 if (radeon_fence_need_sync(*fence, ring->idx)) { 931 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring, 932 ring->idx); 933 radeon_fence_note_sync(*fence, ring->idx); 934 } else { 935 radeon_semaphore_free(rdev, &sem, NULL); 936 } 937 938 for (i = 0; i < num_loops; i++) { 939 cur_size_in_dw = size_in_dw; 940 if (cur_size_in_dw > 0xFFFF) 941 cur_size_in_dw = 0xFFFF; 942 size_in_dw -= cur_size_in_dw; 943 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 0, 0, cur_size_in_dw)); 944 radeon_ring_write(ring, dst_offset & 0xfffffffc); 945 radeon_ring_write(ring, src_offset & 0xfffffffc); 946 radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff); 947 radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff); 948 src_offset += cur_size_in_dw * 4; 949 dst_offset += cur_size_in_dw * 4; 950 } 951 952 r = radeon_fence_emit(rdev, fence, ring->idx); 953 if (r) { 954 radeon_ring_unlock_undo(rdev, ring); 955 return r; 956 } 957 958 radeon_ring_unlock_commit(rdev, ring); 959 radeon_semaphore_free(rdev, &sem, *fence); 960 961 return r; 962 } 963 964 static int rv770_startup(struct radeon_device *rdev) 965 { 966 struct radeon_ring *ring; 967 int r; 968 969 /* enable pcie gen2 link */ 970 rv770_pcie_gen2_enable(rdev); 971 972 if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) { 973 r = r600_init_microcode(rdev); 974 if (r) { 975 DRM_ERROR("Failed to load firmware!\n"); 976 return r; 977 } 978 } 979 980 r = r600_vram_scratch_init(rdev); 981 if (r) 982 return r; 983 984 rv770_mc_program(rdev); 985 if (rdev->flags & RADEON_IS_AGP) { 986 rv770_agp_enable(rdev); 987 } else { 988 r = rv770_pcie_gart_enable(rdev); 989 if (r) 990 return r; 991 } 992 993 rv770_gpu_init(rdev); 994 r = r600_blit_init(rdev); 995 if (r) { 996 r600_blit_fini(rdev); 997 rdev->asic->copy.copy = NULL; 998 dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r); 999 } 1000 1001 /* allocate wb buffer */ 1002 r = radeon_wb_init(rdev); 1003 if (r) 1004 return r; 1005 1006 r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX); 1007 if (r) { 1008 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r); 1009 return r; 1010 } 1011 1012 r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX); 1013 if (r) { 1014 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r); 1015 return r; 1016 } 1017 1018 /* Enable IRQ */ 1019 r = r600_irq_init(rdev); 1020 if (r) { 1021 DRM_ERROR("radeon: IH init failed (%d).\n", r); 1022 radeon_irq_kms_fini(rdev); 1023 return r; 1024 } 1025 r600_irq_set(rdev); 1026 1027 ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]; 1028 r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET, 1029 R600_CP_RB_RPTR, R600_CP_RB_WPTR, 1030 0, 0xfffff, RADEON_CP_PACKET2); 1031 if (r) 1032 return r; 1033 1034 ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX]; 1035 r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET, 1036 DMA_RB_RPTR, DMA_RB_WPTR, 1037 2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0)); 1038 if (r) 1039 return r; 1040 1041 r = rv770_cp_load_microcode(rdev); 1042 if (r) 1043 return r; 1044 r = r600_cp_resume(rdev); 1045 if (r) 1046 return r; 1047 1048 r = r600_dma_resume(rdev); 1049 if (r) 1050 return r; 1051 1052 r = radeon_ib_pool_init(rdev); 1053 if (r) { 1054 dev_err(rdev->dev, "IB initialization failed (%d).\n", r); 1055 return r; 1056 } 1057 1058 r = r600_audio_init(rdev); 1059 if (r) { 1060 DRM_ERROR("radeon: audio init failed\n"); 1061 return r; 1062 } 1063 1064 return 0; 1065 } 1066 1067 int rv770_resume(struct radeon_device *rdev) 1068 { 1069 int r; 1070 1071 /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw, 1072 * posting will perform necessary task to bring back GPU into good 1073 * shape. 1074 */ 1075 /* post card */ 1076 atom_asic_init(rdev->mode_info.atom_context); 1077 1078 rdev->accel_working = true; 1079 r = rv770_startup(rdev); 1080 if (r) { 1081 DRM_ERROR("r600 startup failed on resume\n"); 1082 rdev->accel_working = false; 1083 return r; 1084 } 1085 1086 return r; 1087 1088 } 1089 1090 int rv770_suspend(struct radeon_device *rdev) 1091 { 1092 r600_audio_fini(rdev); 1093 r700_cp_stop(rdev); 1094 r600_dma_stop(rdev); 1095 r600_irq_suspend(rdev); 1096 radeon_wb_disable(rdev); 1097 rv770_pcie_gart_disable(rdev); 1098 1099 return 0; 1100 } 1101 1102 /* Plan is to move initialization in that function and use 1103 * helper function so that radeon_device_init pretty much 1104 * do nothing more than calling asic specific function. This 1105 * should also allow to remove a bunch of callback function 1106 * like vram_info. 1107 */ 1108 int rv770_init(struct radeon_device *rdev) 1109 { 1110 int r; 1111 1112 /* Read BIOS */ 1113 if (!radeon_get_bios(rdev)) { 1114 if (ASIC_IS_AVIVO(rdev)) 1115 return -EINVAL; 1116 } 1117 /* Must be an ATOMBIOS */ 1118 if (!rdev->is_atom_bios) { 1119 dev_err(rdev->dev, "Expecting atombios for R600 GPU\n"); 1120 return -EINVAL; 1121 } 1122 r = radeon_atombios_init(rdev); 1123 if (r) 1124 return r; 1125 /* Post card if necessary */ 1126 if (!radeon_card_posted(rdev)) { 1127 if (!rdev->bios) { 1128 dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n"); 1129 return -EINVAL; 1130 } 1131 DRM_INFO("GPU not posted. posting now...\n"); 1132 atom_asic_init(rdev->mode_info.atom_context); 1133 } 1134 /* Initialize scratch registers */ 1135 r600_scratch_init(rdev); 1136 /* Initialize surface registers */ 1137 radeon_surface_init(rdev); 1138 /* Initialize clocks */ 1139 radeon_get_clock_info(rdev->ddev); 1140 /* Fence driver */ 1141 r = radeon_fence_driver_init(rdev); 1142 if (r) 1143 return r; 1144 /* initialize AGP */ 1145 if (rdev->flags & RADEON_IS_AGP) { 1146 r = radeon_agp_init(rdev); 1147 if (r) 1148 radeon_agp_disable(rdev); 1149 } 1150 r = rv770_mc_init(rdev); 1151 if (r) 1152 return r; 1153 /* Memory manager */ 1154 r = radeon_bo_init(rdev); 1155 if (r) 1156 return r; 1157 1158 r = radeon_irq_kms_init(rdev); 1159 if (r) 1160 return r; 1161 1162 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ring_obj = NULL; 1163 r600_ring_init(rdev, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX], 1024 * 1024); 1164 1165 rdev->ring[R600_RING_TYPE_DMA_INDEX].ring_obj = NULL; 1166 r600_ring_init(rdev, &rdev->ring[R600_RING_TYPE_DMA_INDEX], 64 * 1024); 1167 1168 rdev->ih.ring_obj = NULL; 1169 r600_ih_ring_init(rdev, 64 * 1024); 1170 1171 r = r600_pcie_gart_init(rdev); 1172 if (r) 1173 return r; 1174 1175 rdev->accel_working = true; 1176 r = rv770_startup(rdev); 1177 if (r) { 1178 dev_err(rdev->dev, "disabling GPU acceleration\n"); 1179 r700_cp_fini(rdev); 1180 r600_dma_fini(rdev); 1181 r600_irq_fini(rdev); 1182 radeon_wb_fini(rdev); 1183 radeon_ib_pool_fini(rdev); 1184 radeon_irq_kms_fini(rdev); 1185 rv770_pcie_gart_fini(rdev); 1186 rdev->accel_working = false; 1187 } 1188 1189 return 0; 1190 } 1191 1192 void rv770_fini(struct radeon_device *rdev) 1193 { 1194 r600_blit_fini(rdev); 1195 r700_cp_fini(rdev); 1196 r600_dma_fini(rdev); 1197 r600_irq_fini(rdev); 1198 radeon_wb_fini(rdev); 1199 radeon_ib_pool_fini(rdev); 1200 radeon_irq_kms_fini(rdev); 1201 rv770_pcie_gart_fini(rdev); 1202 r600_vram_scratch_fini(rdev); 1203 radeon_gem_fini(rdev); 1204 radeon_fence_driver_fini(rdev); 1205 radeon_agp_fini(rdev); 1206 radeon_bo_fini(rdev); 1207 radeon_atombios_fini(rdev); 1208 kfree(rdev->bios); 1209 rdev->bios = NULL; 1210 } 1211 1212 static void rv770_pcie_gen2_enable(struct radeon_device *rdev) 1213 { 1214 u32 link_width_cntl, lanes, speed_cntl, tmp; 1215 u16 link_cntl2; 1216 u32 mask; 1217 int ret; 1218 1219 if (radeon_pcie_gen2 == 0) 1220 return; 1221 1222 if (rdev->flags & RADEON_IS_IGP) 1223 return; 1224 1225 if (!(rdev->flags & RADEON_IS_PCIE)) 1226 return; 1227 1228 /* x2 cards have a special sequence */ 1229 if (ASIC_IS_X2(rdev)) 1230 return; 1231 1232 ret = drm_pcie_get_speed_cap_mask(rdev->ddev, &mask); 1233 if (ret != 0) 1234 return; 1235 1236 if (!(mask & DRM_PCIE_SPEED_50)) 1237 return; 1238 1239 DRM_INFO("enabling PCIE gen 2 link speeds, disable with radeon.pcie_gen2=0\n"); 1240 1241 /* advertise upconfig capability */ 1242 link_width_cntl = RREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL); 1243 link_width_cntl &= ~LC_UPCONFIGURE_DIS; 1244 WREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl); 1245 link_width_cntl = RREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL); 1246 if (link_width_cntl & LC_RENEGOTIATION_SUPPORT) { 1247 lanes = (link_width_cntl & LC_LINK_WIDTH_RD_MASK) >> LC_LINK_WIDTH_RD_SHIFT; 1248 link_width_cntl &= ~(LC_LINK_WIDTH_MASK | 1249 LC_RECONFIG_ARC_MISSING_ESCAPE); 1250 link_width_cntl |= lanes | LC_RECONFIG_NOW | 1251 LC_RENEGOTIATE_EN | LC_UPCONFIGURE_SUPPORT; 1252 WREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl); 1253 } else { 1254 link_width_cntl |= LC_UPCONFIGURE_DIS; 1255 WREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl); 1256 } 1257 1258 speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL); 1259 if ((speed_cntl & LC_OTHER_SIDE_EVER_SENT_GEN2) && 1260 (speed_cntl & LC_OTHER_SIDE_SUPPORTS_GEN2)) { 1261 1262 tmp = RREG32(0x541c); 1263 WREG32(0x541c, tmp | 0x8); 1264 WREG32(MM_CFGREGS_CNTL, MM_WR_TO_CFG_EN); 1265 link_cntl2 = RREG16(0x4088); 1266 link_cntl2 &= ~TARGET_LINK_SPEED_MASK; 1267 link_cntl2 |= 0x2; 1268 WREG16(0x4088, link_cntl2); 1269 WREG32(MM_CFGREGS_CNTL, 0); 1270 1271 speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL); 1272 speed_cntl &= ~LC_TARGET_LINK_SPEED_OVERRIDE_EN; 1273 WREG32_PCIE_P(PCIE_LC_SPEED_CNTL, speed_cntl); 1274 1275 speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL); 1276 speed_cntl |= LC_CLR_FAILED_SPD_CHANGE_CNT; 1277 WREG32_PCIE_P(PCIE_LC_SPEED_CNTL, speed_cntl); 1278 1279 speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL); 1280 speed_cntl &= ~LC_CLR_FAILED_SPD_CHANGE_CNT; 1281 WREG32_PCIE_P(PCIE_LC_SPEED_CNTL, speed_cntl); 1282 1283 speed_cntl = RREG32_PCIE_P(PCIE_LC_SPEED_CNTL); 1284 speed_cntl |= LC_GEN2_EN_STRAP; 1285 WREG32_PCIE_P(PCIE_LC_SPEED_CNTL, speed_cntl); 1286 1287 } else { 1288 link_width_cntl = RREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL); 1289 /* XXX: only disable it if gen1 bridge vendor == 0x111d or 0x1106 */ 1290 if (1) 1291 link_width_cntl |= LC_UPCONFIGURE_DIS; 1292 else 1293 link_width_cntl &= ~LC_UPCONFIGURE_DIS; 1294 WREG32_PCIE_P(PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl); 1295 } 1296 } 1297