drivers/gpu/drm/radeon/r300.c

   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/seq_file.h>
  29 #include <linux/slab.h>
  30 #include "drmP.h"
  31 #include "drm.h"
  32 #include "radeon_reg.h"
  33 #include "radeon.h"
  34 #include "radeon_asic.h"
  35 #include "radeon_drm.h"
  36 #include "r100_track.h"
  37 #include "r300d.h"
  38 #include "rv350d.h"
  39 #include "r300_reg_safe.h"
  40
  41 /* This files gather functions specifics to: r300,r350,rv350,rv370,rv380
  42  *
  43  * GPU Errata:
  44  * - HOST_PATH_CNTL: r300 family seems to dislike write to HOST_PATH_CNTL
  45  *   using MMIO to flush host path read cache, this lead to HARDLOCKUP.
  46  *   However, scheduling such write to the ring seems harmless, i suspect
  47  *   the CP read collide with the flush somehow, or maybe the MC, hard to
  48  *   tell. (Jerome Glisse)
  49  */
  50
  51 /*
  52  * rv370,rv380 PCIE GART
  53  */
  54 static int rv370_debugfs_pcie_gart_info_init(struct radeon_device *rdev);
  55
  56 void rv370_pcie_gart_tlb_flush(struct radeon_device *rdev)
  57 {
  58         uint32_t tmp;
  59         int i;
  60
  61         /* Workaround HW bug do flush 2 times */
  62         for (i = 0; i < 2; i++) {
  63                 tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
  64                 WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp | RADEON_PCIE_TX_GART_INVALIDATE_TLB);
  65                 (void)RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
  66                 WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
  67         }
  68         mb();
  69 }
  70
  71 int rv370_pcie_gart_set_page(struct radeon_device *rdev, int i, uint64_t addr)
  72 {
  73         void __iomem *ptr = (void *)rdev->gart.table.vram.ptr;
  74
  75         if (i < 0 || i > rdev->gart.num_gpu_pages) {
  76                 return -EINVAL;
  77         }
  78         addr = (lower_32_bits(addr) >> 8) |
  79                ((upper_32_bits(addr) & 0xff) << 24) |
  80                0xc;
  81         /* on x86 we want this to be CPU endian, on powerpc
  82          * on powerpc without HW swappers, it'll get swapped on way
  83          * into VRAM - so no need for cpu_to_le32 on VRAM tables */
  84         writel(addr, ((void __iomem *)ptr) + (i * 4));
  85         return 0;
  86 }
  87
  88 int rv370_pcie_gart_init(struct radeon_device *rdev)
  89 {
  90         int r;
  91
  92         if (rdev->gart.table.vram.robj) {
  93                 WARN(1, "RV370 PCIE GART already initialized.\n");
  94                 return 0;
  95         }
  96         /* Initialize common gart structure */
  97         r = radeon_gart_init(rdev);
  98         if (r)
  99                 return r;
 100         r = rv370_debugfs_pcie_gart_info_init(rdev);
 101         if (r)
 102                 DRM_ERROR("Failed to register debugfs file for PCIE gart !\n");
 103         rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
 104         rdev->asic->gart_tlb_flush = &rv370_pcie_gart_tlb_flush;
 105         rdev->asic->gart_set_page = &rv370_pcie_gart_set_page;
 106         return radeon_gart_table_vram_alloc(rdev);
 107 }
 108
 109 int rv370_pcie_gart_enable(struct radeon_device *rdev)
 110 {
 111         uint32_t table_addr;
 112         uint32_t tmp;
 113         int r;
 114
 115         if (rdev->gart.table.vram.robj == NULL) {
 116                 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
 117                 return -EINVAL;
 118         }
 119         r = radeon_gart_table_vram_pin(rdev);
 120         if (r)
 121                 return r;
 122         radeon_gart_restore(rdev);
 123         /* discard memory request outside of configured range */
 124         tmp = RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
 125         WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
 126         WREG32_PCIE(RADEON_PCIE_TX_GART_START_LO, rdev->mc.gtt_start);
 127         tmp = rdev->mc.gtt_end & ~RADEON_GPU_PAGE_MASK;
 128         WREG32_PCIE(RADEON_PCIE_TX_GART_END_LO, tmp);
 129         WREG32_PCIE(RADEON_PCIE_TX_GART_START_HI, 0);
 130         WREG32_PCIE(RADEON_PCIE_TX_GART_END_HI, 0);
 131         table_addr = rdev->gart.table_addr;
 132         WREG32_PCIE(RADEON_PCIE_TX_GART_BASE, table_addr);
 133         /* FIXME: setup default page */
 134         WREG32_PCIE(RADEON_PCIE_TX_DISCARD_RD_ADDR_LO, rdev->mc.vram_start);
 135         WREG32_PCIE(RADEON_PCIE_TX_DISCARD_RD_ADDR_HI, 0);
 136         /* Clear error */
 137         WREG32_PCIE(0x18, 0);
 138         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
 139         tmp |= RADEON_PCIE_TX_GART_EN;
 140         tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
 141         WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
 142         rv370_pcie_gart_tlb_flush(rdev);
 143         DRM_INFO("PCIE GART of %uM enabled (table at 0x%08X).\n",
 144                  (unsigned)(rdev->mc.gtt_size >> 20), table_addr);
 145         rdev->gart.ready = true;
 146         return 0;
 147 }
 148
 149 void rv370_pcie_gart_disable(struct radeon_device *rdev)
 150 {
 151         u32 tmp;
 152         int r;
 153
 154         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
 155         tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
 156         WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp & ~RADEON_PCIE_TX_GART_EN);
 157         if (rdev->gart.table.vram.robj) {
 158                 r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
 159                 if (likely(r == 0)) {
 160                         radeon_bo_kunmap(rdev->gart.table.vram.robj);
 161                         radeon_bo_unpin(rdev->gart.table.vram.robj);
 162                         radeon_bo_unreserve(rdev->gart.table.vram.robj);
 163                 }
 164         }
 165 }
 166
 167 void rv370_pcie_gart_fini(struct radeon_device *rdev)
 168 {
 169         radeon_gart_fini(rdev);
 170         rv370_pcie_gart_disable(rdev);
 171         radeon_gart_table_vram_free(rdev);
 172 }
 173
 174 void r300_fence_ring_emit(struct radeon_device *rdev,
 175                           struct radeon_fence *fence)
 176 {
 177         /* Who ever call radeon_fence_emit should call ring_lock and ask
 178          * for enough space (today caller are ib schedule and buffer move) */
 179         /* Write SC register so SC & US assert idle */
 180         radeon_ring_write(rdev, PACKET0(R300_RE_SCISSORS_TL, 0));
 181         radeon_ring_write(rdev, 0);
 182         radeon_ring_write(rdev, PACKET0(R300_RE_SCISSORS_BR, 0));
 183         radeon_ring_write(rdev, 0);
 184         /* Flush 3D cache */
 185         radeon_ring_write(rdev, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
 186         radeon_ring_write(rdev, R300_RB3D_DC_FLUSH);
 187         radeon_ring_write(rdev, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
 188         radeon_ring_write(rdev, R300_ZC_FLUSH);
 189         /* Wait until IDLE & CLEAN */
 190         radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
 191         radeon_ring_write(rdev, (RADEON_WAIT_3D_IDLECLEAN |
 192                                  RADEON_WAIT_2D_IDLECLEAN |
 193                                  RADEON_WAIT_DMA_GUI_IDLE));
 194         radeon_ring_write(rdev, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 195         radeon_ring_write(rdev, rdev->config.r300.hdp_cntl |
 196                                 RADEON_HDP_READ_BUFFER_INVALIDATE);
 197         radeon_ring_write(rdev, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 198         radeon_ring_write(rdev, rdev->config.r300.hdp_cntl);
 199         /* Emit fence sequence & fire IRQ */
 200         radeon_ring_write(rdev, PACKET0(rdev->fence_drv.scratch_reg, 0));
 201         radeon_ring_write(rdev, fence->seq);
 202         radeon_ring_write(rdev, PACKET0(RADEON_GEN_INT_STATUS, 0));
 203         radeon_ring_write(rdev, RADEON_SW_INT_FIRE);
 204 }
 205
 206 void r300_ring_start(struct radeon_device *rdev)
 207 {
 208         unsigned gb_tile_config;
 209         int r;
 210
 211         /* Sub pixel 1/12 so we can have 4K rendering according to doc */
 212         gb_tile_config = (R300_ENABLE_TILING | R300_TILE_SIZE_16);
 213         switch(rdev->num_gb_pipes) {
 214         case 2:
 215                 gb_tile_config |= R300_PIPE_COUNT_R300;
 216                 break;
 217         case 3:
 218                 gb_tile_config |= R300_PIPE_COUNT_R420_3P;
 219                 break;
 220         case 4:
 221                 gb_tile_config |= R300_PIPE_COUNT_R420;
 222                 break;
 223         case 1:
 224         default:
 225                 gb_tile_config |= R300_PIPE_COUNT_RV350;
 226                 break;
 227         }
 228
 229         r = radeon_ring_lock(rdev, 64);
 230         if (r) {
 231                 return;
 232         }
 233         radeon_ring_write(rdev, PACKET0(RADEON_ISYNC_CNTL, 0));
 234         radeon_ring_write(rdev,
 235                           RADEON_ISYNC_ANY2D_IDLE3D |
 236                           RADEON_ISYNC_ANY3D_IDLE2D |
 237                           RADEON_ISYNC_WAIT_IDLEGUI |
 238                           RADEON_ISYNC_CPSCRATCH_IDLEGUI);
 239         radeon_ring_write(rdev, PACKET0(R300_GB_TILE_CONFIG, 0));
 240         radeon_ring_write(rdev, gb_tile_config);
 241         radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
 242         radeon_ring_write(rdev,
 243                           RADEON_WAIT_2D_IDLECLEAN |
 244                           RADEON_WAIT_3D_IDLECLEAN);
 245         radeon_ring_write(rdev, PACKET0(R300_DST_PIPE_CONFIG, 0));
 246         radeon_ring_write(rdev, R300_PIPE_AUTO_CONFIG);
 247         radeon_ring_write(rdev, PACKET0(R300_GB_SELECT, 0));
 248         radeon_ring_write(rdev, 0);
 249         radeon_ring_write(rdev, PACKET0(R300_GB_ENABLE, 0));
 250         radeon_ring_write(rdev, 0);
 251         radeon_ring_write(rdev, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
 252         radeon_ring_write(rdev, R300_RB3D_DC_FLUSH | R300_RB3D_DC_FREE);
 253         radeon_ring_write(rdev, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
 254         radeon_ring_write(rdev, R300_ZC_FLUSH | R300_ZC_FREE);
 255         radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
 256         radeon_ring_write(rdev,
 257                           RADEON_WAIT_2D_IDLECLEAN |
 258                           RADEON_WAIT_3D_IDLECLEAN);
 259         radeon_ring_write(rdev, PACKET0(R300_GB_AA_CONFIG, 0));
 260         radeon_ring_write(rdev, 0);
 261         radeon_ring_write(rdev, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
 262         radeon_ring_write(rdev, R300_RB3D_DC_FLUSH | R300_RB3D_DC_FREE);
 263         radeon_ring_write(rdev, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
 264         radeon_ring_write(rdev, R300_ZC_FLUSH | R300_ZC_FREE);
 265         radeon_ring_write(rdev, PACKET0(R300_GB_MSPOS0, 0));
 266         radeon_ring_write(rdev,
 267                           ((6 << R300_MS_X0_SHIFT) |
 268                            (6 << R300_MS_Y0_SHIFT) |
 269                            (6 << R300_MS_X1_SHIFT) |
 270                            (6 << R300_MS_Y1_SHIFT) |
 271                            (6 << R300_MS_X2_SHIFT) |
 272                            (6 << R300_MS_Y2_SHIFT) |
 273                            (6 << R300_MSBD0_Y_SHIFT) |
 274                            (6 << R300_MSBD0_X_SHIFT)));
 275         radeon_ring_write(rdev, PACKET0(R300_GB_MSPOS1, 0));
 276         radeon_ring_write(rdev,
 277                           ((6 << R300_MS_X3_SHIFT) |
 278                            (6 << R300_MS_Y3_SHIFT) |
 279                            (6 << R300_MS_X4_SHIFT) |
 280                            (6 << R300_MS_Y4_SHIFT) |
 281                            (6 << R300_MS_X5_SHIFT) |
 282                            (6 << R300_MS_Y5_SHIFT) |
 283                            (6 << R300_MSBD1_SHIFT)));
 284         radeon_ring_write(rdev, PACKET0(R300_GA_ENHANCE, 0));
 285         radeon_ring_write(rdev, R300_GA_DEADLOCK_CNTL | R300_GA_FASTSYNC_CNTL);
 286         radeon_ring_write(rdev, PACKET0(R300_GA_POLY_MODE, 0));
 287         radeon_ring_write(rdev,
 288                           R300_FRONT_PTYPE_TRIANGE | R300_BACK_PTYPE_TRIANGE);
 289         radeon_ring_write(rdev, PACKET0(R300_GA_ROUND_MODE, 0));
 290         radeon_ring_write(rdev,
 291                           R300_GEOMETRY_ROUND_NEAREST |
 292                           R300_COLOR_ROUND_NEAREST);
 293         radeon_ring_unlock_commit(rdev);
 294 }
 295
 296 void r300_errata(struct radeon_device *rdev)
 297 {
 298         rdev->pll_errata = 0;
 299
 300         if (rdev->family == CHIP_R300 &&
 301             (RREG32(RADEON_CONFIG_CNTL) & RADEON_CFG_ATI_REV_ID_MASK) == RADEON_CFG_ATI_REV_A11) {
 302                 rdev->pll_errata |= CHIP_ERRATA_R300_CG;
 303         }
 304 }
 305
 306 int r300_mc_wait_for_idle(struct radeon_device *rdev)
 307 {
 308         unsigned i;
 309         uint32_t tmp;
 310
 311         for (i = 0; i < rdev->usec_timeout; i++) {
 312                 /* read MC_STATUS */
 313                 tmp = RREG32(RADEON_MC_STATUS);
 314                 if (tmp & R300_MC_IDLE) {
 315                         return 0;
 316                 }
 317                 DRM_UDELAY(1);
 318         }
 319         return -1;
 320 }
 321
 322 void r300_gpu_init(struct radeon_device *rdev)
 323 {
 324         uint32_t gb_tile_config, tmp;
 325
 326         r100_hdp_reset(rdev);
 327         if ((rdev->family == CHIP_R300 && rdev->pdev->device != 0x4144) ||
 328             (rdev->family == CHIP_R350 && rdev->pdev->device != 0x4148)) {
 329                 /* r300,r350 */
 330                 rdev->num_gb_pipes = 2;
 331         } else {
 332                 /* rv350,rv370,rv380,r300 AD, r350 AH */
 333                 rdev->num_gb_pipes = 1;
 334         }
 335         rdev->num_z_pipes = 1;
 336         gb_tile_config = (R300_ENABLE_TILING | R300_TILE_SIZE_16);
 337         switch (rdev->num_gb_pipes) {
 338         case 2:
 339                 gb_tile_config |= R300_PIPE_COUNT_R300;
 340                 break;
 341         case 3:
 342                 gb_tile_config |= R300_PIPE_COUNT_R420_3P;
 343                 break;
 344         case 4:
 345                 gb_tile_config |= R300_PIPE_COUNT_R420;
 346                 break;
 347         default:
 348         case 1:
 349                 gb_tile_config |= R300_PIPE_COUNT_RV350;
 350                 break;
 351         }
 352         WREG32(R300_GB_TILE_CONFIG, gb_tile_config);
 353
 354         if (r100_gui_wait_for_idle(rdev)) {
 355                 printk(KERN_WARNING "Failed to wait GUI idle while "
 356                        "programming pipes. Bad things might happen.\n");
 357         }
 358
 359         tmp = RREG32(R300_DST_PIPE_CONFIG);
 360         WREG32(R300_DST_PIPE_CONFIG, tmp | R300_PIPE_AUTO_CONFIG);
 361
 362         WREG32(R300_RB2D_DSTCACHE_MODE,
 363                R300_DC_AUTOFLUSH_ENABLE |
 364                R300_DC_DC_DISABLE_IGNORE_PE);
 365
 366         if (r100_gui_wait_for_idle(rdev)) {
 367                 printk(KERN_WARNING "Failed to wait GUI idle while "
 368                        "programming pipes. Bad things might happen.\n");
 369         }
 370         if (r300_mc_wait_for_idle(rdev)) {
 371                 printk(KERN_WARNING "Failed to wait MC idle while "
 372                        "programming pipes. Bad things might happen.\n");
 373         }
 374         DRM_INFO("radeon: %d quad pipes, %d Z pipes initialized.\n",
 375                  rdev->num_gb_pipes, rdev->num_z_pipes);
 376 }
 377
 378 int r300_ga_reset(struct radeon_device *rdev)
 379 {
 380         uint32_t tmp;
 381         bool reinit_cp;
 382         int i;
 383
 384         reinit_cp = rdev->cp.ready;
 385         rdev->cp.ready = false;
 386         for (i = 0; i < rdev->usec_timeout; i++) {
 387                 WREG32(RADEON_CP_CSQ_MODE, 0);
 388                 WREG32(RADEON_CP_CSQ_CNTL, 0);
 389                 WREG32(RADEON_RBBM_SOFT_RESET, 0x32005);
 390                 (void)RREG32(RADEON_RBBM_SOFT_RESET);
 391                 udelay(200);
 392                 WREG32(RADEON_RBBM_SOFT_RESET, 0);
 393                 /* Wait to prevent race in RBBM_STATUS */
 394                 mdelay(1);
 395                 tmp = RREG32(RADEON_RBBM_STATUS);
 396                 if (tmp & ((1 << 20) | (1 << 26))) {
 397                         DRM_ERROR("VAP & CP still busy (RBBM_STATUS=0x%08X)", tmp);
 398                         /* GA still busy soft reset it */
 399                         WREG32(0x429C, 0x200);
 400                         WREG32(R300_VAP_PVS_STATE_FLUSH_REG, 0);
 401                         WREG32(R300_RE_SCISSORS_TL, 0);
 402                         WREG32(R300_RE_SCISSORS_BR, 0);
 403                         WREG32(0x24AC, 0);
 404                 }
 405                 /* Wait to prevent race in RBBM_STATUS */
 406                 mdelay(1);
 407                 tmp = RREG32(RADEON_RBBM_STATUS);
 408                 if (!(tmp & ((1 << 20) | (1 << 26)))) {
 409                         break;
 410                 }
 411         }
 412         for (i = 0; i < rdev->usec_timeout; i++) {
 413                 tmp = RREG32(RADEON_RBBM_STATUS);
 414                 if (!(tmp & ((1 << 20) | (1 << 26)))) {
 415                         DRM_INFO("GA reset succeed (RBBM_STATUS=0x%08X)\n",
 416                                  tmp);
 417                         if (reinit_cp) {
 418                                 return r100_cp_init(rdev, rdev->cp.ring_size);
 419                         }
 420                         return 0;
 421                 }
 422                 DRM_UDELAY(1);
 423         }
 424         tmp = RREG32(RADEON_RBBM_STATUS);
 425         DRM_ERROR("Failed to reset GA ! (RBBM_STATUS=0x%08X)\n", tmp);
 426         return -1;
 427 }
 428
 429 int r300_gpu_reset(struct radeon_device *rdev)
 430 {
 431         uint32_t status;
 432
 433         /* reset order likely matter */
 434         status = RREG32(RADEON_RBBM_STATUS);
 435         /* reset HDP */
 436         r100_hdp_reset(rdev);
 437         /* reset rb2d */
 438         if (status & ((1 << 17) | (1 << 18) | (1 << 27))) {
 439                 r100_rb2d_reset(rdev);
 440         }
 441         /* reset GA */
 442         if (status & ((1 << 20) | (1 << 26))) {
 443                 r300_ga_reset(rdev);
 444         }
 445         /* reset CP */
 446         status = RREG32(RADEON_RBBM_STATUS);
 447         if (status & (1 << 16)) {
 448                 r100_cp_reset(rdev);
 449         }
 450         /* Check if GPU is idle */
 451         status = RREG32(RADEON_RBBM_STATUS);
 452         if (status & RADEON_RBBM_ACTIVE) {
 453                 DRM_ERROR("Failed to reset GPU (RBBM_STATUS=0x%08X)\n", status);
 454                 return -1;
 455         }
 456         DRM_INFO("GPU reset succeed (RBBM_STATUS=0x%08X)\n", status);
 457         return 0;
 458 }
 459
 460
 461 /*
 462  * r300,r350,rv350,rv380 VRAM info
 463  */
 464 void r300_mc_init(struct radeon_device *rdev)
 465 {
 466         u64 base;
 467         u32 tmp;
 468
 469         /* DDR for all card after R300 & IGP */
 470         rdev->mc.vram_is_ddr = true;
 471         tmp = RREG32(RADEON_MEM_CNTL);
 472         tmp &= R300_MEM_NUM_CHANNELS_MASK;
 473         switch (tmp) {
 474         case 0: rdev->mc.vram_width = 64; break;
 475         case 1: rdev->mc.vram_width = 128; break;
 476         case 2: rdev->mc.vram_width = 256; break;
 477         default:  rdev->mc.vram_width = 128; break;
 478         }
 479         r100_vram_init_sizes(rdev);
 480         base = rdev->mc.aper_base;
 481         if (rdev->flags & RADEON_IS_IGP)
 482                 base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
 483         radeon_vram_location(rdev, &rdev->mc, base);
 484         if (!(rdev->flags & RADEON_IS_AGP))
 485                 radeon_gtt_location(rdev, &rdev->mc);
 486         radeon_update_bandwidth_info(rdev);
 487 }
 488
 489 void rv370_set_pcie_lanes(struct radeon_device *rdev, int lanes)
 490 {
 491         uint32_t link_width_cntl, mask;
 492
 493         if (rdev->flags & RADEON_IS_IGP)
 494                 return;
 495
 496         if (!(rdev->flags & RADEON_IS_PCIE))
 497                 return;
 498
 499         /* FIXME wait for idle */
 500
 501         switch (lanes) {
 502         case 0:
 503                 mask = RADEON_PCIE_LC_LINK_WIDTH_X0;
 504                 break;
 505         case 1:
 506                 mask = RADEON_PCIE_LC_LINK_WIDTH_X1;
 507                 break;
 508         case 2:
 509                 mask = RADEON_PCIE_LC_LINK_WIDTH_X2;
 510                 break;
 511         case 4:
 512                 mask = RADEON_PCIE_LC_LINK_WIDTH_X4;
 513                 break;
 514         case 8:
 515                 mask = RADEON_PCIE_LC_LINK_WIDTH_X8;
 516                 break;
 517         case 12:
 518                 mask = RADEON_PCIE_LC_LINK_WIDTH_X12;
 519                 break;
 520         case 16:
 521         default:
 522                 mask = RADEON_PCIE_LC_LINK_WIDTH_X16;
 523                 break;
 524         }
 525
 526         link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 527
 528         if ((link_width_cntl & RADEON_PCIE_LC_LINK_WIDTH_RD_MASK) ==
 529             (mask << RADEON_PCIE_LC_LINK_WIDTH_RD_SHIFT))
 530                 return;
 531
 532         link_width_cntl &= ~(RADEON_PCIE_LC_LINK_WIDTH_MASK |
 533                              RADEON_PCIE_LC_RECONFIG_NOW |
 534                              RADEON_PCIE_LC_RECONFIG_LATER |
 535                              RADEON_PCIE_LC_SHORT_RECONFIG_EN);
 536         link_width_cntl |= mask;
 537         WREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
 538         WREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL, (link_width_cntl |
 539                                                      RADEON_PCIE_LC_RECONFIG_NOW));
 540
 541         /* wait for lane set to complete */
 542         link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 543         while (link_width_cntl == 0xffffffff)
 544                 link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 545
 546 }
 547
 548 int rv370_get_pcie_lanes(struct radeon_device *rdev)
 549 {
 550         u32 link_width_cntl;
 551
 552         if (rdev->flags & RADEON_IS_IGP)
 553                 return 0;
 554
 555         if (!(rdev->flags & RADEON_IS_PCIE))
 556                 return 0;
 557
 558         /* FIXME wait for idle */
 559
 560         if (rdev->family < CHIP_R600)
 561                 link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 562         else
 563                 link_width_cntl = RREG32_PCIE_P(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 564
 565         switch ((link_width_cntl & RADEON_PCIE_LC_LINK_WIDTH_RD_MASK) >> RADEON_PCIE_LC_LINK_WIDTH_RD_SHIFT) {
 566         case RADEON_PCIE_LC_LINK_WIDTH_X0:
 567                 return 0;
 568         case RADEON_PCIE_LC_LINK_WIDTH_X1:
 569                 return 1;
 570         case RADEON_PCIE_LC_LINK_WIDTH_X2:
 571                 return 2;
 572         case RADEON_PCIE_LC_LINK_WIDTH_X4:
 573                 return 4;
 574         case RADEON_PCIE_LC_LINK_WIDTH_X8:
 575                 return 8;
 576         case RADEON_PCIE_LC_LINK_WIDTH_X16:
 577         default:
 578                 return 16;
 579         }
 580 }
 581
 582 #if defined(CONFIG_DEBUG_FS)
 583 static int rv370_debugfs_pcie_gart_info(struct seq_file *m, void *data)
 584 {
 585         struct drm_info_node *node = (struct drm_info_node *) m->private;
 586         struct drm_device *dev = node->minor->dev;
 587         struct radeon_device *rdev = dev->dev_private;
 588         uint32_t tmp;
 589
 590         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
 591         seq_printf(m, "PCIE_TX_GART_CNTL 0x%08x\n", tmp);
 592         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_BASE);
 593         seq_printf(m, "PCIE_TX_GART_BASE 0x%08x\n", tmp);
 594         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_START_LO);
 595         seq_printf(m, "PCIE_TX_GART_START_LO 0x%08x\n", tmp);
 596         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_START_HI);
 597         seq_printf(m, "PCIE_TX_GART_START_HI 0x%08x\n", tmp);
 598         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_END_LO);
 599         seq_printf(m, "PCIE_TX_GART_END_LO 0x%08x\n", tmp);
 600         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_END_HI);
 601         seq_printf(m, "PCIE_TX_GART_END_HI 0x%08x\n", tmp);
 602         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_ERROR);
 603         seq_printf(m, "PCIE_TX_GART_ERROR 0x%08x\n", tmp);
 604         return 0;
 605 }
 606
 607 static struct drm_info_list rv370_pcie_gart_info_list[] = {
 608         {"rv370_pcie_gart_info", rv370_debugfs_pcie_gart_info, 0, NULL},
 609 };
 610 #endif
 611
 612 static int rv370_debugfs_pcie_gart_info_init(struct radeon_device *rdev)
 613 {
 614 #if defined(CONFIG_DEBUG_FS)
 615         return radeon_debugfs_add_files(rdev, rv370_pcie_gart_info_list, 1);
 616 #else
 617         return 0;
 618 #endif
 619 }
 620
 621 static int r300_packet0_check(struct radeon_cs_parser *p,
 622                 struct radeon_cs_packet *pkt,
 623                 unsigned idx, unsigned reg)
 624 {
 625         struct radeon_cs_reloc *reloc;
 626         struct r100_cs_track *track;
 627         volatile uint32_t *ib;
 628         uint32_t tmp, tile_flags = 0;
 629         unsigned i;
 630         int r;
 631         u32 idx_value;
 632
 633         ib = p->ib->ptr;
 634         track = (struct r100_cs_track *)p->track;
 635         idx_value = radeon_get_ib_value(p, idx);
 636
 637         switch(reg) {
 638         case AVIVO_D1MODE_VLINE_START_END:
 639         case RADEON_CRTC_GUI_TRIG_VLINE:
 640                 r = r100_cs_packet_parse_vline(p);
 641                 if (r) {
 642                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 643                                         idx, reg);
 644                         r100_cs_dump_packet(p, pkt);
 645                         return r;
 646                 }
 647                 break;
 648         case RADEON_DST_PITCH_OFFSET:
 649         case RADEON_SRC_PITCH_OFFSET:
 650                 r = r100_reloc_pitch_offset(p, pkt, idx, reg);
 651                 if (r)
 652                         return r;
 653                 break;
 654         case R300_RB3D_COLOROFFSET0:
 655         case R300_RB3D_COLOROFFSET1:
 656         case R300_RB3D_COLOROFFSET2:
 657         case R300_RB3D_COLOROFFSET3:
 658                 i = (reg - R300_RB3D_COLOROFFSET0) >> 2;
 659                 r = r100_cs_packet_next_reloc(p, &reloc);
 660                 if (r) {
 661                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 662                                         idx, reg);
 663                         r100_cs_dump_packet(p, pkt);
 664                         return r;
 665                 }
 666                 track->cb[i].robj = reloc->robj;
 667                 track->cb[i].offset = idx_value;
 668                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
 669                 break;
 670         case R300_ZB_DEPTHOFFSET:
 671                 r = r100_cs_packet_next_reloc(p, &reloc);
 672                 if (r) {
 673                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 674                                         idx, reg);
 675                         r100_cs_dump_packet(p, pkt);
 676                         return r;
 677                 }
 678                 track->zb.robj = reloc->robj;
 679                 track->zb.offset = idx_value;
 680                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
 681                 break;
 682         case R300_TX_OFFSET_0:
 683         case R300_TX_OFFSET_0+4:
 684         case R300_TX_OFFSET_0+8:
 685         case R300_TX_OFFSET_0+12:
 686         case R300_TX_OFFSET_0+16:
 687         case R300_TX_OFFSET_0+20:
 688         case R300_TX_OFFSET_0+24:
 689         case R300_TX_OFFSET_0+28:
 690         case R300_TX_OFFSET_0+32:
 691         case R300_TX_OFFSET_0+36:
 692         case R300_TX_OFFSET_0+40:
 693         case R300_TX_OFFSET_0+44:
 694         case R300_TX_OFFSET_0+48:
 695         case R300_TX_OFFSET_0+52:
 696         case R300_TX_OFFSET_0+56:
 697         case R300_TX_OFFSET_0+60:
 698                 i = (reg - R300_TX_OFFSET_0) >> 2;
 699                 r = r100_cs_packet_next_reloc(p, &reloc);
 700                 if (r) {
 701                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 702                                         idx, reg);
 703                         r100_cs_dump_packet(p, pkt);
 704                         return r;
 705                 }
 706
 707                 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
 708                         tile_flags |= R300_TXO_MACRO_TILE;
 709                 if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
 710                         tile_flags |= R300_TXO_MICRO_TILE;
 711                 else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO_SQUARE)
 712                         tile_flags |= R300_TXO_MICRO_TILE_SQUARE;
 713
 714                 tmp = idx_value + ((u32)reloc->lobj.gpu_offset);
 715                 tmp |= tile_flags;
 716                 ib[idx] = tmp;
 717                 track->textures[i].robj = reloc->robj;
 718                 break;
 719         /* Tracked registers */
 720         case 0x2084:
 721                 /* VAP_VF_CNTL */
 722                 track->vap_vf_cntl = idx_value;
 723                 break;
 724         case 0x20B4:
 725                 /* VAP_VTX_SIZE */
 726                 track->vtx_size = idx_value & 0x7F;
 727                 break;
 728         case 0x2134:
 729                 /* VAP_VF_MAX_VTX_INDX */
 730                 track->max_indx = idx_value & 0x00FFFFFFUL;
 731                 break;
 732         case 0x2088:
 733                 /* VAP_ALT_NUM_VERTICES - only valid on r500 */
 734                 if (p->rdev->family < CHIP_RV515)
 735                         goto fail;
 736                 track->vap_alt_nverts = idx_value & 0xFFFFFF;
 737                 break;
 738         case 0x43E4:
 739                 /* SC_SCISSOR1 */
 740                 track->maxy = ((idx_value >> 13) & 0x1FFF) + 1;
 741                 if (p->rdev->family < CHIP_RV515) {
 742                         track->maxy -= 1440;
 743                 }
 744                 break;
 745         case 0x4E00:
 746                 /* RB3D_CCTL */
 747                 track->num_cb = ((idx_value >> 5) & 0x3) + 1;
 748                 break;
 749         case 0x4E38:
 750         case 0x4E3C:
 751         case 0x4E40:
 752         case 0x4E44:
 753                 /* RB3D_COLORPITCH0 */
 754                 /* RB3D_COLORPITCH1 */
 755                 /* RB3D_COLORPITCH2 */
 756                 /* RB3D_COLORPITCH3 */
 757                 r = r100_cs_packet_next_reloc(p, &reloc);
 758                 if (r) {
 759                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 760                                   idx, reg);
 761                         r100_cs_dump_packet(p, pkt);
 762                         return r;
 763                 }
 764
 765                 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
 766                         tile_flags |= R300_COLOR_TILE_ENABLE;
 767                 if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
 768                         tile_flags |= R300_COLOR_MICROTILE_ENABLE;
 769                 else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO_SQUARE)
 770                         tile_flags |= R300_COLOR_MICROTILE_SQUARE_ENABLE;
 771
 772                 tmp = idx_value & ~(0x7 << 16);
 773                 tmp |= tile_flags;
 774                 ib[idx] = tmp;
 775                 i = (reg - 0x4E38) >> 2;
 776                 track->cb[i].pitch = idx_value & 0x3FFE;
 777                 switch (((idx_value >> 21) & 0xF)) {
 778                 case 9:
 779                 case 11:
 780                 case 12:
 781                         track->cb[i].cpp = 1;
 782                         break;
 783                 case 3:
 784                 case 4:
 785                 case 13:
 786                 case 15:
 787                         track->cb[i].cpp = 2;
 788                         break;
 789                 case 6:
 790                         track->cb[i].cpp = 4;
 791                         break;
 792                 case 10:
 793                         track->cb[i].cpp = 8;
 794                         break;
 795                 case 7:
 796                         track->cb[i].cpp = 16;
 797                         break;
 798                 default:
 799                         DRM_ERROR("Invalid color buffer format (%d) !\n",
 800                                   ((idx_value >> 21) & 0xF));
 801                         return -EINVAL;
 802                 }
 803                 break;
 804         case 0x4F00:
 805                 /* ZB_CNTL */
 806                 if (idx_value & 2) {
 807                         track->z_enabled = true;
 808                 } else {
 809                         track->z_enabled = false;
 810                 }
 811                 break;
 812         case 0x4F10:
 813                 /* ZB_FORMAT */
 814                 switch ((idx_value & 0xF)) {
 815                 case 0:
 816                 case 1:
 817                         track->zb.cpp = 2;
 818                         break;
 819                 case 2:
 820                         track->zb.cpp = 4;
 821                         break;
 822                 default:
 823                         DRM_ERROR("Invalid z buffer format (%d) !\n",
 824                                   (idx_value & 0xF));
 825                         return -EINVAL;
 826                 }
 827                 break;
 828         case 0x4F24:
 829                 /* ZB_DEPTHPITCH */
 830                 r = r100_cs_packet_next_reloc(p, &reloc);
 831                 if (r) {
 832                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 833                                   idx, reg);
 834                         r100_cs_dump_packet(p, pkt);
 835                         return r;
 836                 }
 837
 838                 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
 839                         tile_flags |= R300_DEPTHMACROTILE_ENABLE;
 840                 if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
 841                         tile_flags |= R300_DEPTHMICROTILE_TILED;
 842                 else if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO_SQUARE)
 843                         tile_flags |= R300_DEPTHMICROTILE_TILED_SQUARE;
 844
 845                 tmp = idx_value & ~(0x7 << 16);
 846                 tmp |= tile_flags;
 847                 ib[idx] = tmp;
 848
 849                 track->zb.pitch = idx_value & 0x3FFC;
 850                 break;
 851         case 0x4104:
 852                 for (i = 0; i < 16; i++) {
 853                         bool enabled;
 854
 855                         enabled = !!(idx_value & (1 << i));
 856                         track->textures[i].enabled = enabled;
 857                 }
 858                 break;
 859         case 0x44C0:
 860         case 0x44C4:
 861         case 0x44C8:
 862         case 0x44CC:
 863         case 0x44D0:
 864         case 0x44D4:
 865         case 0x44D8:
 866         case 0x44DC:
 867         case 0x44E0:
 868         case 0x44E4:
 869         case 0x44E8:
 870         case 0x44EC:
 871         case 0x44F0:
 872         case 0x44F4:
 873         case 0x44F8:
 874         case 0x44FC:
 875                 /* TX_FORMAT1_[0-15] */
 876                 i = (reg - 0x44C0) >> 2;
 877                 tmp = (idx_value >> 25) & 0x3;
 878                 track->textures[i].tex_coord_type = tmp;
 879                 switch ((idx_value & 0x1F)) {
 880                 case R300_TX_FORMAT_X8:
 881                 case R300_TX_FORMAT_Y4X4:
 882                 case R300_TX_FORMAT_Z3Y3X2:
 883                         track->textures[i].cpp = 1;
 884                         break;
 885                 case R300_TX_FORMAT_X16:
 886                 case R300_TX_FORMAT_Y8X8:
 887                 case R300_TX_FORMAT_Z5Y6X5:
 888                 case R300_TX_FORMAT_Z6Y5X5:
 889                 case R300_TX_FORMAT_W4Z4Y4X4:
 890                 case R300_TX_FORMAT_W1Z5Y5X5:
 891                 case R300_TX_FORMAT_D3DMFT_CxV8U8:
 892                 case R300_TX_FORMAT_B8G8_B8G8:
 893                 case R300_TX_FORMAT_G8R8_G8B8:
 894                         track->textures[i].cpp = 2;
 895                         break;
 896                 case R300_TX_FORMAT_Y16X16:
 897                 case R300_TX_FORMAT_Z11Y11X10:
 898                 case R300_TX_FORMAT_Z10Y11X11:
 899                 case R300_TX_FORMAT_W8Z8Y8X8:
 900                 case R300_TX_FORMAT_W2Z10Y10X10:
 901                 case 0x17:
 902                 case R300_TX_FORMAT_FL_I32:
 903                 case 0x1e:
 904                         track->textures[i].cpp = 4;
 905                         break;
 906                 case R300_TX_FORMAT_W16Z16Y16X16:
 907                 case R300_TX_FORMAT_FL_R16G16B16A16:
 908                 case R300_TX_FORMAT_FL_I32A32:
 909                         track->textures[i].cpp = 8;
 910                         break;
 911                 case R300_TX_FORMAT_FL_R32G32B32A32:
 912                         track->textures[i].cpp = 16;
 913                         break;
 914                 case R300_TX_FORMAT_DXT1:
 915                         track->textures[i].cpp = 1;
 916                         track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
 917                         break;
 918                 case R300_TX_FORMAT_ATI2N:
 919                         if (p->rdev->family < CHIP_R420) {
 920                                 DRM_ERROR("Invalid texture format %u\n",
 921                                           (idx_value & 0x1F));
 922                                 return -EINVAL;
 923                         }
 924                         /* The same rules apply as for DXT3/5. */
 925                         /* Pass through. */
 926                 case R300_TX_FORMAT_DXT3:
 927                 case R300_TX_FORMAT_DXT5:
 928                         track->textures[i].cpp = 1;
 929                         track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
 930                         break;
 931                 default:
 932                         DRM_ERROR("Invalid texture format %u\n",
 933                                   (idx_value & 0x1F));
 934                         return -EINVAL;
 935                         break;
 936                 }
 937                 break;
 938         case 0x4400:
 939         case 0x4404:
 940         case 0x4408:
 941         case 0x440C:
 942         case 0x4410:
 943         case 0x4414:
 944         case 0x4418:
 945         case 0x441C:
 946         case 0x4420:
 947         case 0x4424:
 948         case 0x4428:
 949         case 0x442C:
 950         case 0x4430:
 951         case 0x4434:
 952         case 0x4438:
 953         case 0x443C:
 954                 /* TX_FILTER0_[0-15] */
 955                 i = (reg - 0x4400) >> 2;
 956                 tmp = idx_value & 0x7;
 957                 if (tmp == 2 || tmp == 4 || tmp == 6) {
 958                         track->textures[i].roundup_w = false;
 959                 }
 960                 tmp = (idx_value >> 3) & 0x7;
 961                 if (tmp == 2 || tmp == 4 || tmp == 6) {
 962                         track->textures[i].roundup_h = false;
 963                 }
 964                 break;
 965         case 0x4500:
 966         case 0x4504:
 967         case 0x4508:
 968         case 0x450C:
 969         case 0x4510:
 970         case 0x4514:
 971         case 0x4518:
 972         case 0x451C:
 973         case 0x4520:
 974         case 0x4524:
 975         case 0x4528:
 976         case 0x452C:
 977         case 0x4530:
 978         case 0x4534:
 979         case 0x4538:
 980         case 0x453C:
 981                 /* TX_FORMAT2_[0-15] */
 982                 i = (reg - 0x4500) >> 2;
 983                 tmp = idx_value & 0x3FFF;
 984                 track->textures[i].pitch = tmp + 1;
 985                 if (p->rdev->family >= CHIP_RV515) {
 986                         tmp = ((idx_value >> 15) & 1) << 11;
 987                         track->textures[i].width_11 = tmp;
 988                         tmp = ((idx_value >> 16) & 1) << 11;
 989                         track->textures[i].height_11 = tmp;
 990
 991                         /* ATI1N */
 992                         if (idx_value & (1 << 14)) {
 993                                 /* The same rules apply as for DXT1. */
 994                                 track->textures[i].compress_format =
 995                                         R100_TRACK_COMP_DXT1;
 996                         }
 997                 } else if (idx_value & (1 << 14)) {
 998                         DRM_ERROR("Forbidden bit TXFORMAT_MSB\n");
 999                         return -EINVAL;
1000                 }
1001                 break;
1002         case 0x4480:
1003         case 0x4484:
1004         case 0x4488:
1005         case 0x448C:
1006         case 0x4490:
1007         case 0x4494:
1008         case 0x4498:
1009         case 0x449C:
1010         case 0x44A0:
1011         case 0x44A4:
1012         case 0x44A8:
1013         case 0x44AC:
1014         case 0x44B0:
1015         case 0x44B4:
1016         case 0x44B8:
1017         case 0x44BC:
1018                 /* TX_FORMAT0_[0-15] */
1019                 i = (reg - 0x4480) >> 2;
1020                 tmp = idx_value & 0x7FF;
1021                 track->textures[i].width = tmp + 1;
1022                 tmp = (idx_value >> 11) & 0x7FF;
1023                 track->textures[i].height = tmp + 1;
1024                 tmp = (idx_value >> 26) & 0xF;
1025                 track->textures[i].num_levels = tmp;
1026                 tmp = idx_value & (1 << 31);
1027                 track->textures[i].use_pitch = !!tmp;
1028                 tmp = (idx_value >> 22) & 0xF;
1029                 track->textures[i].txdepth = tmp;
1030                 break;
1031         case R300_ZB_ZPASS_ADDR:
1032                 r = r100_cs_packet_next_reloc(p, &reloc);
1033                 if (r) {
1034                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1035                                         idx, reg);
1036                         r100_cs_dump_packet(p, pkt);
1037                         return r;
1038                 }
1039                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1040                 break;
1041         case 0x4e0c:
1042                 /* RB3D_COLOR_CHANNEL_MASK */
1043                 track->color_channel_mask = idx_value;
1044                 break;
1045         case 0x4d1c:
1046                 /* ZB_BW_CNTL */
1047                 track->fastfill = !!(idx_value & (1 << 2));
1048                 break;
1049         case 0x4e04:
1050                 /* RB3D_BLENDCNTL */
1051                 track->blend_read_enable = !!(idx_value & (1 << 2));
1052                 break;
1053         case 0x4be8:
1054                 /* valid register only on RV530 */
1055                 if (p->rdev->family == CHIP_RV530)
1056                         break;
1057                 /* fallthrough do not move */
1058         default:
1059                 goto fail;
1060         }
1061         return 0;
1062 fail:
1063         printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1064                reg, idx);
1065         return -EINVAL;
1066 }
1067
1068 static int r300_packet3_check(struct radeon_cs_parser *p,
1069                               struct radeon_cs_packet *pkt)
1070 {
1071         struct radeon_cs_reloc *reloc;
1072         struct r100_cs_track *track;
1073         volatile uint32_t *ib;
1074         unsigned idx;
1075         int r;
1076
1077         ib = p->ib->ptr;
1078         idx = pkt->idx + 1;
1079         track = (struct r100_cs_track *)p->track;
1080         switch(pkt->opcode) {
1081         case PACKET3_3D_LOAD_VBPNTR:
1082                 r = r100_packet3_load_vbpntr(p, pkt, idx);
1083                 if (r)
1084                         return r;
1085                 break;
1086         case PACKET3_INDX_BUFFER:
1087                 r = r100_cs_packet_next_reloc(p, &reloc);
1088                 if (r) {
1089                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1090                         r100_cs_dump_packet(p, pkt);
1091                         return r;
1092                 }
1093                 ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
1094                 r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1095                 if (r) {
1096                         return r;
1097                 }
1098                 break;
1099         /* Draw packet */
1100         case PACKET3_3D_DRAW_IMMD:
1101                 /* Number of dwords is vtx_size * (num_vertices - 1)
1102                  * PRIM_WALK must be equal to 3 vertex data in embedded
1103                  * in cmd stream */
1104                 if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1105                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1106                         return -EINVAL;
1107                 }
1108                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1109                 track->immd_dwords = pkt->count - 1;
1110                 r = r100_cs_track_check(p->rdev, track);
1111                 if (r) {
1112                         return r;
1113                 }
1114                 break;
1115         case PACKET3_3D_DRAW_IMMD_2:
1116                 /* Number of dwords is vtx_size * (num_vertices - 1)
1117                  * PRIM_WALK must be equal to 3 vertex data in embedded
1118                  * in cmd stream */
1119                 if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1120                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1121                         return -EINVAL;
1122                 }
1123                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1124                 track->immd_dwords = pkt->count;
1125                 r = r100_cs_track_check(p->rdev, track);
1126                 if (r) {
1127                         return r;
1128                 }
1129                 break;
1130         case PACKET3_3D_DRAW_VBUF:
1131                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1132                 r = r100_cs_track_check(p->rdev, track);
1133                 if (r) {
1134                         return r;
1135                 }
1136                 break;
1137         case PACKET3_3D_DRAW_VBUF_2:
1138                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1139                 r = r100_cs_track_check(p->rdev, track);
1140                 if (r) {
1141                         return r;
1142                 }
1143                 break;
1144         case PACKET3_3D_DRAW_INDX:
1145                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1146                 r = r100_cs_track_check(p->rdev, track);
1147                 if (r) {
1148                         return r;
1149                 }
1150                 break;
1151         case PACKET3_3D_DRAW_INDX_2:
1152                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1153                 r = r100_cs_track_check(p->rdev, track);
1154                 if (r) {
1155                         return r;
1156                 }
1157                 break;
1158         case PACKET3_NOP:
1159                 break;
1160         default:
1161                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1162                 return -EINVAL;
1163         }
1164         return 0;
1165 }
1166
1167 int r300_cs_parse(struct radeon_cs_parser *p)
1168 {
1169         struct radeon_cs_packet pkt;
1170         struct r100_cs_track *track;
1171         int r;
1172
1173         track = kzalloc(sizeof(*track), GFP_KERNEL);
1174         r100_cs_track_clear(p->rdev, track);
1175         p->track = track;
1176         do {
1177                 r = r100_cs_packet_parse(p, &pkt, p->idx);
1178                 if (r) {
1179                         return r;
1180                 }
1181                 p->idx += pkt.count + 2;
1182                 switch (pkt.type) {
1183                 case PACKET_TYPE0:
1184                         r = r100_cs_parse_packet0(p, &pkt,
1185                                                   p->rdev->config.r300.reg_safe_bm,
1186                                                   p->rdev->config.r300.reg_safe_bm_size,
1187                                                   &r300_packet0_check);
1188                         break;
1189                 case PACKET_TYPE2:
1190                         break;
1191                 case PACKET_TYPE3:
1192                         r = r300_packet3_check(p, &pkt);
1193                         break;
1194                 default:
1195                         DRM_ERROR("Unknown packet type %d !\n", pkt.type);
1196                         return -EINVAL;
1197                 }
1198                 if (r) {
1199                         return r;
1200                 }
1201         } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1202         return 0;
1203 }
1204
1205 void r300_set_reg_safe(struct radeon_device *rdev)
1206 {
1207         rdev->config.r300.reg_safe_bm = r300_reg_safe_bm;
1208         rdev->config.r300.reg_safe_bm_size = ARRAY_SIZE(r300_reg_safe_bm);
1209 }
1210
1211 void r300_mc_program(struct radeon_device *rdev)
1212 {
1213         struct r100_mc_save save;
1214         int r;
1215
1216         r = r100_debugfs_mc_info_init(rdev);
1217         if (r) {
1218                 dev_err(rdev->dev, "Failed to create r100_mc debugfs file.\n");
1219         }
1220
1221         /* Stops all mc clients */
1222         r100_mc_stop(rdev, &save);
1223         if (rdev->flags & RADEON_IS_AGP) {
1224                 WREG32(R_00014C_MC_AGP_LOCATION,
1225                         S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
1226                         S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
1227                 WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
1228                 WREG32(R_00015C_AGP_BASE_2,
1229                         upper_32_bits(rdev->mc.agp_base) & 0xff);
1230         } else {
1231                 WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
1232                 WREG32(R_000170_AGP_BASE, 0);
1233                 WREG32(R_00015C_AGP_BASE_2, 0);
1234         }
1235         /* Wait for mc idle */
1236         if (r300_mc_wait_for_idle(rdev))
1237                 DRM_INFO("Failed to wait MC idle before programming MC.\n");
1238         /* Program MC, should be a 32bits limited address space */
1239         WREG32(R_000148_MC_FB_LOCATION,
1240                 S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
1241                 S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
1242         r100_mc_resume(rdev, &save);
1243 }
1244
1245 void r300_clock_startup(struct radeon_device *rdev)
1246 {
1247         u32 tmp;
1248
1249         if (radeon_dynclks != -1 && radeon_dynclks)
1250                 radeon_legacy_set_clock_gating(rdev, 1);
1251         /* We need to force on some of the block */
1252         tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
1253         tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
1254         if ((rdev->family == CHIP_RV350) || (rdev->family == CHIP_RV380))
1255                 tmp |= S_00000D_FORCE_VAP(1);
1256         WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
1257 }
1258
1259 static int r300_startup(struct radeon_device *rdev)
1260 {
1261         int r;
1262
1263         /* set common regs */
1264         r100_set_common_regs(rdev);
1265         /* program mc */
1266         r300_mc_program(rdev);
1267         /* Resume clock */
1268         r300_clock_startup(rdev);
1269         /* Initialize GPU configuration (# pipes, ...) */
1270         r300_gpu_init(rdev);
1271         /* Initialize GART (initialize after TTM so we can allocate
1272          * memory through TTM but finalize after TTM) */
1273         if (rdev->flags & RADEON_IS_PCIE) {
1274                 r = rv370_pcie_gart_enable(rdev);
1275                 if (r)
1276                         return r;
1277         }
1278
1279         if (rdev->family == CHIP_R300 ||
1280             rdev->family == CHIP_R350 ||
1281             rdev->family == CHIP_RV350)
1282                 r100_enable_bm(rdev);
1283
1284         if (rdev->flags & RADEON_IS_PCI) {
1285                 r = r100_pci_gart_enable(rdev);
1286                 if (r)
1287                         return r;
1288         }
1289         /* Enable IRQ */
1290         r100_irq_set(rdev);
1291         rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
1292         /* 1M ring buffer */
1293         r = r100_cp_init(rdev, 1024 * 1024);
1294         if (r) {
1295                 dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
1296                 return r;
1297         }
1298         r = r100_wb_init(rdev);
1299         if (r)
1300                 dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
1301         r = r100_ib_init(rdev);
1302         if (r) {
1303                 dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
1304                 return r;
1305         }
1306         return 0;
1307 }
1308
1309 int r300_resume(struct radeon_device *rdev)
1310 {
1311         /* Make sur GART are not working */
1312         if (rdev->flags & RADEON_IS_PCIE)
1313                 rv370_pcie_gart_disable(rdev);
1314         if (rdev->flags & RADEON_IS_PCI)
1315                 r100_pci_gart_disable(rdev);
1316         /* Resume clock before doing reset */
1317         r300_clock_startup(rdev);
1318         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
1319         if (radeon_gpu_reset(rdev)) {
1320                 dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
1321                         RREG32(R_000E40_RBBM_STATUS),
1322                         RREG32(R_0007C0_CP_STAT));
1323         }
1324         /* post */
1325         radeon_combios_asic_init(rdev->ddev);
1326         /* Resume clock after posting */
1327         r300_clock_startup(rdev);
1328         /* Initialize surface registers */
1329         radeon_surface_init(rdev);
1330         return r300_startup(rdev);
1331 }
1332
1333 int r300_suspend(struct radeon_device *rdev)
1334 {
1335         r100_cp_disable(rdev);
1336         r100_wb_disable(rdev);
1337         r100_irq_disable(rdev);
1338         if (rdev->flags & RADEON_IS_PCIE)
1339                 rv370_pcie_gart_disable(rdev);
1340         if (rdev->flags & RADEON_IS_PCI)
1341                 r100_pci_gart_disable(rdev);
1342         return 0;
1343 }
1344
1345 void r300_fini(struct radeon_device *rdev)
1346 {
1347         radeon_pm_fini(rdev);
1348         r100_cp_fini(rdev);
1349         r100_wb_fini(rdev);
1350         r100_ib_fini(rdev);
1351         radeon_gem_fini(rdev);
1352         if (rdev->flags & RADEON_IS_PCIE)
1353                 rv370_pcie_gart_fini(rdev);
1354         if (rdev->flags & RADEON_IS_PCI)
1355                 r100_pci_gart_fini(rdev);
1356         radeon_agp_fini(rdev);
1357         radeon_irq_kms_fini(rdev);
1358         radeon_fence_driver_fini(rdev);
1359         radeon_bo_fini(rdev);
1360         radeon_atombios_fini(rdev);
1361         kfree(rdev->bios);
1362         rdev->bios = NULL;
1363 }
1364
1365 int r300_init(struct radeon_device *rdev)
1366 {
1367         int r;
1368
1369         /* Disable VGA */
1370         r100_vga_render_disable(rdev);
1371         /* Initialize scratch registers */
1372         radeon_scratch_init(rdev);
1373         /* Initialize surface registers */
1374         radeon_surface_init(rdev);
1375         /* TODO: disable VGA need to use VGA request */
1376         /* BIOS*/
1377         if (!radeon_get_bios(rdev)) {
1378                 if (ASIC_IS_AVIVO(rdev))
1379                         return -EINVAL;
1380         }
1381         if (rdev->is_atom_bios) {
1382                 dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
1383                 return -EINVAL;
1384         } else {
1385                 r = radeon_combios_init(rdev);
1386                 if (r)
1387                         return r;
1388         }
1389         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
1390         if (radeon_gpu_reset(rdev)) {
1391                 dev_warn(rdev->dev,
1392                         "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
1393                         RREG32(R_000E40_RBBM_STATUS),
1394                         RREG32(R_0007C0_CP_STAT));
1395         }
1396         /* check if cards are posted or not */
1397         if (radeon_boot_test_post_card(rdev) == false)
1398                 return -EINVAL;
1399         /* Set asic errata */
1400         r300_errata(rdev);
1401         /* Initialize clocks */
1402         radeon_get_clock_info(rdev->ddev);
1403         /* Initialize power management */
1404         radeon_pm_init(rdev);
1405         /* initialize AGP */
1406         if (rdev->flags & RADEON_IS_AGP) {
1407                 r = radeon_agp_init(rdev);
1408                 if (r) {
1409                         radeon_agp_disable(rdev);
1410                 }
1411         }
1412         /* initialize memory controller */
1413         r300_mc_init(rdev);
1414         /* Fence driver */
1415         r = radeon_fence_driver_init(rdev);
1416         if (r)
1417                 return r;
1418         r = radeon_irq_kms_init(rdev);
1419         if (r)
1420                 return r;
1421         /* Memory manager */
1422         r = radeon_bo_init(rdev);
1423         if (r)
1424                 return r;
1425         if (rdev->flags & RADEON_IS_PCIE) {
1426                 r = rv370_pcie_gart_init(rdev);
1427                 if (r)
1428                         return r;
1429         }
1430         if (rdev->flags & RADEON_IS_PCI) {
1431                 r = r100_pci_gart_init(rdev);
1432                 if (r)
1433                         return r;
1434         }
1435         r300_set_reg_safe(rdev);
1436         rdev->accel_working = true;
1437         r = r300_startup(rdev);
1438         if (r) {
1439                 /* Somethings want wront with the accel init stop accel */
1440                 dev_err(rdev->dev, "Disabling GPU acceleration\n");
1441                 r100_cp_fini(rdev);
1442                 r100_wb_fini(rdev);
1443                 r100_ib_fini(rdev);
1444                 radeon_irq_kms_fini(rdev);
1445                 if (rdev->flags & RADEON_IS_PCIE)
1446                         rv370_pcie_gart_fini(rdev);
1447                 if (rdev->flags & RADEON_IS_PCI)
1448                         r100_pci_gart_fini(rdev);
1449                 radeon_agp_fini(rdev);
1450                 rdev->accel_working = false;
1451         }
1452         return 0;
1453 }