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 "drmP.h"
  30 #include "drm.h"
  31 #include "radeon_reg.h"
  32 #include "radeon.h"
  33 #include "radeon_drm.h"
  34 #include "r100_track.h"
  35 #include "r300d.h"
  36 #include "rv350d.h"
  37 #include "r300_reg_safe.h"
  38
  39 /* This files gather functions specifics to: r300,r350,rv350,rv370,rv380 */
  40
  41 /*
  42  * rv370,rv380 PCIE GART
  43  */
  44 static int rv370_debugfs_pcie_gart_info_init(struct radeon_device *rdev);
  45
  46 void rv370_pcie_gart_tlb_flush(struct radeon_device *rdev)
  47 {
  48         uint32_t tmp;
  49         int i;
  50
  51         /* Workaround HW bug do flush 2 times */
  52         for (i = 0; i < 2; i++) {
  53                 tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
  54                 WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp | RADEON_PCIE_TX_GART_INVALIDATE_TLB);
  55                 (void)RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
  56                 WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
  57         }
  58         mb();
  59 }
  60
  61 int rv370_pcie_gart_set_page(struct radeon_device *rdev, int i, uint64_t addr)
  62 {
  63         void __iomem *ptr = (void *)rdev->gart.table.vram.ptr;
  64
  65         if (i < 0 || i > rdev->gart.num_gpu_pages) {
  66                 return -EINVAL;
  67         }
  68         addr = (lower_32_bits(addr) >> 8) |
  69                ((upper_32_bits(addr) & 0xff) << 24) |
  70                0xc;
  71         /* on x86 we want this to be CPU endian, on powerpc
  72          * on powerpc without HW swappers, it'll get swapped on way
  73          * into VRAM - so no need for cpu_to_le32 on VRAM tables */
  74         writel(addr, ((void __iomem *)ptr) + (i * 4));
  75         return 0;
  76 }
  77
  78 int rv370_pcie_gart_init(struct radeon_device *rdev)
  79 {
  80         int r;
  81
  82         if (rdev->gart.table.vram.robj) {
  83                 WARN(1, "RV370 PCIE GART already initialized.\n");
  84                 return 0;
  85         }
  86         /* Initialize common gart structure */
  87         r = radeon_gart_init(rdev);
  88         if (r)
  89                 return r;
  90         r = rv370_debugfs_pcie_gart_info_init(rdev);
  91         if (r)
  92                 DRM_ERROR("Failed to register debugfs file for PCIE gart !\n");
  93         rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
  94         rdev->asic->gart_tlb_flush = &rv370_pcie_gart_tlb_flush;
  95         rdev->asic->gart_set_page = &rv370_pcie_gart_set_page;
  96         return radeon_gart_table_vram_alloc(rdev);
  97 }
  98
  99 int rv370_pcie_gart_enable(struct radeon_device *rdev)
 100 {
 101         uint32_t table_addr;
 102         uint32_t tmp;
 103         int r;
 104
 105         if (rdev->gart.table.vram.robj == NULL) {
 106                 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
 107                 return -EINVAL;
 108         }
 109         r = radeon_gart_table_vram_pin(rdev);
 110         if (r)
 111                 return r;
 112         /* discard memory request outside of configured range */
 113         tmp = RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
 114         WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
 115         WREG32_PCIE(RADEON_PCIE_TX_GART_START_LO, rdev->mc.gtt_location);
 116         tmp = rdev->mc.gtt_location + rdev->mc.gtt_size - RADEON_GPU_PAGE_SIZE;
 117         WREG32_PCIE(RADEON_PCIE_TX_GART_END_LO, tmp);
 118         WREG32_PCIE(RADEON_PCIE_TX_GART_START_HI, 0);
 119         WREG32_PCIE(RADEON_PCIE_TX_GART_END_HI, 0);
 120         table_addr = rdev->gart.table_addr;
 121         WREG32_PCIE(RADEON_PCIE_TX_GART_BASE, table_addr);
 122         /* FIXME: setup default page */
 123         WREG32_PCIE(RADEON_PCIE_TX_DISCARD_RD_ADDR_LO, rdev->mc.vram_location);
 124         WREG32_PCIE(RADEON_PCIE_TX_DISCARD_RD_ADDR_HI, 0);
 125         /* Clear error */
 126         WREG32_PCIE(0x18, 0);
 127         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
 128         tmp |= RADEON_PCIE_TX_GART_EN;
 129         tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
 130         WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
 131         rv370_pcie_gart_tlb_flush(rdev);
 132         DRM_INFO("PCIE GART of %uM enabled (table at 0x%08X).\n",
 133                  (unsigned)(rdev->mc.gtt_size >> 20), table_addr);
 134         rdev->gart.ready = true;
 135         return 0;
 136 }
 137
 138 void rv370_pcie_gart_disable(struct radeon_device *rdev)
 139 {
 140         u32 tmp;
 141         int r;
 142
 143         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
 144         tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
 145         WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp & ~RADEON_PCIE_TX_GART_EN);
 146         if (rdev->gart.table.vram.robj) {
 147                 r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
 148                 if (likely(r == 0)) {
 149                         radeon_bo_kunmap(rdev->gart.table.vram.robj);
 150                         radeon_bo_unpin(rdev->gart.table.vram.robj);
 151                         radeon_bo_unreserve(rdev->gart.table.vram.robj);
 152                 }
 153         }
 154 }
 155
 156 void rv370_pcie_gart_fini(struct radeon_device *rdev)
 157 {
 158         rv370_pcie_gart_disable(rdev);
 159         radeon_gart_table_vram_free(rdev);
 160         radeon_gart_fini(rdev);
 161 }
 162
 163 void r300_fence_ring_emit(struct radeon_device *rdev,
 164                           struct radeon_fence *fence)
 165 {
 166         /* Who ever call radeon_fence_emit should call ring_lock and ask
 167          * for enough space (today caller are ib schedule and buffer move) */
 168         /* Write SC register so SC & US assert idle */
 169         radeon_ring_write(rdev, PACKET0(0x43E0, 0));
 170         radeon_ring_write(rdev, 0);
 171         radeon_ring_write(rdev, PACKET0(0x43E4, 0));
 172         radeon_ring_write(rdev, 0);
 173         /* Flush 3D cache */
 174         radeon_ring_write(rdev, PACKET0(0x4E4C, 0));
 175         radeon_ring_write(rdev, (2 << 0));
 176         radeon_ring_write(rdev, PACKET0(0x4F18, 0));
 177         radeon_ring_write(rdev, (1 << 0));
 178         /* Wait until IDLE & CLEAN */
 179         radeon_ring_write(rdev, PACKET0(0x1720, 0));
 180         radeon_ring_write(rdev, (1 << 17) | (1 << 16)  | (1 << 9));
 181         /* Emit fence sequence & fire IRQ */
 182         radeon_ring_write(rdev, PACKET0(rdev->fence_drv.scratch_reg, 0));
 183         radeon_ring_write(rdev, fence->seq);
 184         radeon_ring_write(rdev, PACKET0(RADEON_GEN_INT_STATUS, 0));
 185         radeon_ring_write(rdev, RADEON_SW_INT_FIRE);
 186 }
 187
 188 int r300_copy_dma(struct radeon_device *rdev,
 189                   uint64_t src_offset,
 190                   uint64_t dst_offset,
 191                   unsigned num_pages,
 192                   struct radeon_fence *fence)
 193 {
 194         uint32_t size;
 195         uint32_t cur_size;
 196         int i, num_loops;
 197         int r = 0;
 198
 199         /* radeon pitch is /64 */
 200         size = num_pages << PAGE_SHIFT;
 201         num_loops = DIV_ROUND_UP(size, 0x1FFFFF);
 202         r = radeon_ring_lock(rdev, num_loops * 4 + 64);
 203         if (r) {
 204                 DRM_ERROR("radeon: moving bo (%d).\n", r);
 205                 return r;
 206         }
 207         /* Must wait for 2D idle & clean before DMA or hangs might happen */
 208         radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0 ));
 209         radeon_ring_write(rdev, (1 << 16));
 210         for (i = 0; i < num_loops; i++) {
 211                 cur_size = size;
 212                 if (cur_size > 0x1FFFFF) {
 213                         cur_size = 0x1FFFFF;
 214                 }
 215                 size -= cur_size;
 216                 radeon_ring_write(rdev, PACKET0(0x720, 2));
 217                 radeon_ring_write(rdev, src_offset);
 218                 radeon_ring_write(rdev, dst_offset);
 219                 radeon_ring_write(rdev, cur_size | (1 << 31) | (1 << 30));
 220                 src_offset += cur_size;
 221                 dst_offset += cur_size;
 222         }
 223         radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
 224         radeon_ring_write(rdev, RADEON_WAIT_DMA_GUI_IDLE);
 225         if (fence) {
 226                 r = radeon_fence_emit(rdev, fence);
 227         }
 228         radeon_ring_unlock_commit(rdev);
 229         return r;
 230 }
 231
 232 void r300_ring_start(struct radeon_device *rdev)
 233 {
 234         unsigned gb_tile_config;
 235         int r;
 236
 237         /* Sub pixel 1/12 so we can have 4K rendering according to doc */
 238         gb_tile_config = (R300_ENABLE_TILING | R300_TILE_SIZE_16);
 239         switch(rdev->num_gb_pipes) {
 240         case 2:
 241                 gb_tile_config |= R300_PIPE_COUNT_R300;
 242                 break;
 243         case 3:
 244                 gb_tile_config |= R300_PIPE_COUNT_R420_3P;
 245                 break;
 246         case 4:
 247                 gb_tile_config |= R300_PIPE_COUNT_R420;
 248                 break;
 249         case 1:
 250         default:
 251                 gb_tile_config |= R300_PIPE_COUNT_RV350;
 252                 break;
 253         }
 254
 255         r = radeon_ring_lock(rdev, 64);
 256         if (r) {
 257                 return;
 258         }
 259         radeon_ring_write(rdev, PACKET0(RADEON_ISYNC_CNTL, 0));
 260         radeon_ring_write(rdev,
 261                           RADEON_ISYNC_ANY2D_IDLE3D |
 262                           RADEON_ISYNC_ANY3D_IDLE2D |
 263                           RADEON_ISYNC_WAIT_IDLEGUI |
 264                           RADEON_ISYNC_CPSCRATCH_IDLEGUI);
 265         radeon_ring_write(rdev, PACKET0(R300_GB_TILE_CONFIG, 0));
 266         radeon_ring_write(rdev, gb_tile_config);
 267         radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
 268         radeon_ring_write(rdev,
 269                           RADEON_WAIT_2D_IDLECLEAN |
 270                           RADEON_WAIT_3D_IDLECLEAN);
 271         radeon_ring_write(rdev, PACKET0(0x170C, 0));
 272         radeon_ring_write(rdev, 1 << 31);
 273         radeon_ring_write(rdev, PACKET0(R300_GB_SELECT, 0));
 274         radeon_ring_write(rdev, 0);
 275         radeon_ring_write(rdev, PACKET0(R300_GB_ENABLE, 0));
 276         radeon_ring_write(rdev, 0);
 277         radeon_ring_write(rdev, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
 278         radeon_ring_write(rdev, R300_RB3D_DC_FLUSH | R300_RB3D_DC_FREE);
 279         radeon_ring_write(rdev, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
 280         radeon_ring_write(rdev, R300_ZC_FLUSH | R300_ZC_FREE);
 281         radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
 282         radeon_ring_write(rdev,
 283                           RADEON_WAIT_2D_IDLECLEAN |
 284                           RADEON_WAIT_3D_IDLECLEAN);
 285         radeon_ring_write(rdev, PACKET0(R300_GB_AA_CONFIG, 0));
 286         radeon_ring_write(rdev, 0);
 287         radeon_ring_write(rdev, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
 288         radeon_ring_write(rdev, R300_RB3D_DC_FLUSH | R300_RB3D_DC_FREE);
 289         radeon_ring_write(rdev, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
 290         radeon_ring_write(rdev, R300_ZC_FLUSH | R300_ZC_FREE);
 291         radeon_ring_write(rdev, PACKET0(R300_GB_MSPOS0, 0));
 292         radeon_ring_write(rdev,
 293                           ((6 << R300_MS_X0_SHIFT) |
 294                            (6 << R300_MS_Y0_SHIFT) |
 295                            (6 << R300_MS_X1_SHIFT) |
 296                            (6 << R300_MS_Y1_SHIFT) |
 297                            (6 << R300_MS_X2_SHIFT) |
 298                            (6 << R300_MS_Y2_SHIFT) |
 299                            (6 << R300_MSBD0_Y_SHIFT) |
 300                            (6 << R300_MSBD0_X_SHIFT)));
 301         radeon_ring_write(rdev, PACKET0(R300_GB_MSPOS1, 0));
 302         radeon_ring_write(rdev,
 303                           ((6 << R300_MS_X3_SHIFT) |
 304                            (6 << R300_MS_Y3_SHIFT) |
 305                            (6 << R300_MS_X4_SHIFT) |
 306                            (6 << R300_MS_Y4_SHIFT) |
 307                            (6 << R300_MS_X5_SHIFT) |
 308                            (6 << R300_MS_Y5_SHIFT) |
 309                            (6 << R300_MSBD1_SHIFT)));
 310         radeon_ring_write(rdev, PACKET0(R300_GA_ENHANCE, 0));
 311         radeon_ring_write(rdev, R300_GA_DEADLOCK_CNTL | R300_GA_FASTSYNC_CNTL);
 312         radeon_ring_write(rdev, PACKET0(R300_GA_POLY_MODE, 0));
 313         radeon_ring_write(rdev,
 314                           R300_FRONT_PTYPE_TRIANGE | R300_BACK_PTYPE_TRIANGE);
 315         radeon_ring_write(rdev, PACKET0(R300_GA_ROUND_MODE, 0));
 316         radeon_ring_write(rdev,
 317                           R300_GEOMETRY_ROUND_NEAREST |
 318                           R300_COLOR_ROUND_NEAREST);
 319         radeon_ring_unlock_commit(rdev);
 320 }
 321
 322 void r300_errata(struct radeon_device *rdev)
 323 {
 324         rdev->pll_errata = 0;
 325
 326         if (rdev->family == CHIP_R300 &&
 327             (RREG32(RADEON_CONFIG_CNTL) & RADEON_CFG_ATI_REV_ID_MASK) == RADEON_CFG_ATI_REV_A11) {
 328                 rdev->pll_errata |= CHIP_ERRATA_R300_CG;
 329         }
 330 }
 331
 332 int r300_mc_wait_for_idle(struct radeon_device *rdev)
 333 {
 334         unsigned i;
 335         uint32_t tmp;
 336
 337         for (i = 0; i < rdev->usec_timeout; i++) {
 338                 /* read MC_STATUS */
 339                 tmp = RREG32(0x0150);
 340                 if (tmp & (1 << 4)) {
 341                         return 0;
 342                 }
 343                 DRM_UDELAY(1);
 344         }
 345         return -1;
 346 }
 347
 348 void r300_gpu_init(struct radeon_device *rdev)
 349 {
 350         uint32_t gb_tile_config, tmp;
 351
 352         r100_hdp_reset(rdev);
 353         /* FIXME: rv380 one pipes ? */
 354         if ((rdev->family == CHIP_R300) || (rdev->family == CHIP_R350)) {
 355                 /* r300,r350 */
 356                 rdev->num_gb_pipes = 2;
 357         } else {
 358                 /* rv350,rv370,rv380 */
 359                 rdev->num_gb_pipes = 1;
 360         }
 361         rdev->num_z_pipes = 1;
 362         gb_tile_config = (R300_ENABLE_TILING | R300_TILE_SIZE_16);
 363         switch (rdev->num_gb_pipes) {
 364         case 2:
 365                 gb_tile_config |= R300_PIPE_COUNT_R300;
 366                 break;
 367         case 3:
 368                 gb_tile_config |= R300_PIPE_COUNT_R420_3P;
 369                 break;
 370         case 4:
 371                 gb_tile_config |= R300_PIPE_COUNT_R420;
 372                 break;
 373         default:
 374         case 1:
 375                 gb_tile_config |= R300_PIPE_COUNT_RV350;
 376                 break;
 377         }
 378         WREG32(R300_GB_TILE_CONFIG, gb_tile_config);
 379
 380         if (r100_gui_wait_for_idle(rdev)) {
 381                 printk(KERN_WARNING "Failed to wait GUI idle while "
 382                        "programming pipes. Bad things might happen.\n");
 383         }
 384
 385         tmp = RREG32(0x170C);
 386         WREG32(0x170C, tmp | (1 << 31));
 387
 388         WREG32(R300_RB2D_DSTCACHE_MODE,
 389                R300_DC_AUTOFLUSH_ENABLE |
 390                R300_DC_DC_DISABLE_IGNORE_PE);
 391
 392         if (r100_gui_wait_for_idle(rdev)) {
 393                 printk(KERN_WARNING "Failed to wait GUI idle while "
 394                        "programming pipes. Bad things might happen.\n");
 395         }
 396         if (r300_mc_wait_for_idle(rdev)) {
 397                 printk(KERN_WARNING "Failed to wait MC idle while "
 398                        "programming pipes. Bad things might happen.\n");
 399         }
 400         DRM_INFO("radeon: %d quad pipes, %d Z pipes initialized.\n",
 401                  rdev->num_gb_pipes, rdev->num_z_pipes);
 402 }
 403
 404 int r300_ga_reset(struct radeon_device *rdev)
 405 {
 406         uint32_t tmp;
 407         bool reinit_cp;
 408         int i;
 409
 410         reinit_cp = rdev->cp.ready;
 411         rdev->cp.ready = false;
 412         for (i = 0; i < rdev->usec_timeout; i++) {
 413                 WREG32(RADEON_CP_CSQ_MODE, 0);
 414                 WREG32(RADEON_CP_CSQ_CNTL, 0);
 415                 WREG32(RADEON_RBBM_SOFT_RESET, 0x32005);
 416                 (void)RREG32(RADEON_RBBM_SOFT_RESET);
 417                 udelay(200);
 418                 WREG32(RADEON_RBBM_SOFT_RESET, 0);
 419                 /* Wait to prevent race in RBBM_STATUS */
 420                 mdelay(1);
 421                 tmp = RREG32(RADEON_RBBM_STATUS);
 422                 if (tmp & ((1 << 20) | (1 << 26))) {
 423                         DRM_ERROR("VAP & CP still busy (RBBM_STATUS=0x%08X)", tmp);
 424                         /* GA still busy soft reset it */
 425                         WREG32(0x429C, 0x200);
 426                         WREG32(R300_VAP_PVS_STATE_FLUSH_REG, 0);
 427                         WREG32(0x43E0, 0);
 428                         WREG32(0x43E4, 0);
 429                         WREG32(0x24AC, 0);
 430                 }
 431                 /* Wait to prevent race in RBBM_STATUS */
 432                 mdelay(1);
 433                 tmp = RREG32(RADEON_RBBM_STATUS);
 434                 if (!(tmp & ((1 << 20) | (1 << 26)))) {
 435                         break;
 436                 }
 437         }
 438         for (i = 0; i < rdev->usec_timeout; i++) {
 439                 tmp = RREG32(RADEON_RBBM_STATUS);
 440                 if (!(tmp & ((1 << 20) | (1 << 26)))) {
 441                         DRM_INFO("GA reset succeed (RBBM_STATUS=0x%08X)\n",
 442                                  tmp);
 443                         if (reinit_cp) {
 444                                 return r100_cp_init(rdev, rdev->cp.ring_size);
 445                         }
 446                         return 0;
 447                 }
 448                 DRM_UDELAY(1);
 449         }
 450         tmp = RREG32(RADEON_RBBM_STATUS);
 451         DRM_ERROR("Failed to reset GA ! (RBBM_STATUS=0x%08X)\n", tmp);
 452         return -1;
 453 }
 454
 455 int r300_gpu_reset(struct radeon_device *rdev)
 456 {
 457         uint32_t status;
 458
 459         /* reset order likely matter */
 460         status = RREG32(RADEON_RBBM_STATUS);
 461         /* reset HDP */
 462         r100_hdp_reset(rdev);
 463         /* reset rb2d */
 464         if (status & ((1 << 17) | (1 << 18) | (1 << 27))) {
 465                 r100_rb2d_reset(rdev);
 466         }
 467         /* reset GA */
 468         if (status & ((1 << 20) | (1 << 26))) {
 469                 r300_ga_reset(rdev);
 470         }
 471         /* reset CP */
 472         status = RREG32(RADEON_RBBM_STATUS);
 473         if (status & (1 << 16)) {
 474                 r100_cp_reset(rdev);
 475         }
 476         /* Check if GPU is idle */
 477         status = RREG32(RADEON_RBBM_STATUS);
 478         if (status & (1 << 31)) {
 479                 DRM_ERROR("Failed to reset GPU (RBBM_STATUS=0x%08X)\n", status);
 480                 return -1;
 481         }
 482         DRM_INFO("GPU reset succeed (RBBM_STATUS=0x%08X)\n", status);
 483         return 0;
 484 }
 485
 486
 487 /*
 488  * r300,r350,rv350,rv380 VRAM info
 489  */
 490 void r300_vram_info(struct radeon_device *rdev)
 491 {
 492         uint32_t tmp;
 493
 494         /* DDR for all card after R300 & IGP */
 495         rdev->mc.vram_is_ddr = true;
 496         tmp = RREG32(RADEON_MEM_CNTL);
 497         if (tmp & R300_MEM_NUM_CHANNELS_MASK) {
 498                 rdev->mc.vram_width = 128;
 499         } else {
 500                 rdev->mc.vram_width = 64;
 501         }
 502
 503         r100_vram_init_sizes(rdev);
 504 }
 505
 506 void rv370_set_pcie_lanes(struct radeon_device *rdev, int lanes)
 507 {
 508         uint32_t link_width_cntl, mask;
 509
 510         if (rdev->flags & RADEON_IS_IGP)
 511                 return;
 512
 513         if (!(rdev->flags & RADEON_IS_PCIE))
 514                 return;
 515
 516         /* FIXME wait for idle */
 517
 518         switch (lanes) {
 519         case 0:
 520                 mask = RADEON_PCIE_LC_LINK_WIDTH_X0;
 521                 break;
 522         case 1:
 523                 mask = RADEON_PCIE_LC_LINK_WIDTH_X1;
 524                 break;
 525         case 2:
 526                 mask = RADEON_PCIE_LC_LINK_WIDTH_X2;
 527                 break;
 528         case 4:
 529                 mask = RADEON_PCIE_LC_LINK_WIDTH_X4;
 530                 break;
 531         case 8:
 532                 mask = RADEON_PCIE_LC_LINK_WIDTH_X8;
 533                 break;
 534         case 12:
 535                 mask = RADEON_PCIE_LC_LINK_WIDTH_X12;
 536                 break;
 537         case 16:
 538         default:
 539                 mask = RADEON_PCIE_LC_LINK_WIDTH_X16;
 540                 break;
 541         }
 542
 543         link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 544
 545         if ((link_width_cntl & RADEON_PCIE_LC_LINK_WIDTH_RD_MASK) ==
 546             (mask << RADEON_PCIE_LC_LINK_WIDTH_RD_SHIFT))
 547                 return;
 548
 549         link_width_cntl &= ~(RADEON_PCIE_LC_LINK_WIDTH_MASK |
 550                              RADEON_PCIE_LC_RECONFIG_NOW |
 551                              RADEON_PCIE_LC_RECONFIG_LATER |
 552                              RADEON_PCIE_LC_SHORT_RECONFIG_EN);
 553         link_width_cntl |= mask;
 554         WREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
 555         WREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL, (link_width_cntl |
 556                                                      RADEON_PCIE_LC_RECONFIG_NOW));
 557
 558         /* wait for lane set to complete */
 559         link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 560         while (link_width_cntl == 0xffffffff)
 561                 link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 562
 563 }
 564
 565 #if defined(CONFIG_DEBUG_FS)
 566 static int rv370_debugfs_pcie_gart_info(struct seq_file *m, void *data)
 567 {
 568         struct drm_info_node *node = (struct drm_info_node *) m->private;
 569         struct drm_device *dev = node->minor->dev;
 570         struct radeon_device *rdev = dev->dev_private;
 571         uint32_t tmp;
 572
 573         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
 574         seq_printf(m, "PCIE_TX_GART_CNTL 0x%08x\n", tmp);
 575         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_BASE);
 576         seq_printf(m, "PCIE_TX_GART_BASE 0x%08x\n", tmp);
 577         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_START_LO);
 578         seq_printf(m, "PCIE_TX_GART_START_LO 0x%08x\n", tmp);
 579         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_START_HI);
 580         seq_printf(m, "PCIE_TX_GART_START_HI 0x%08x\n", tmp);
 581         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_END_LO);
 582         seq_printf(m, "PCIE_TX_GART_END_LO 0x%08x\n", tmp);
 583         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_END_HI);
 584         seq_printf(m, "PCIE_TX_GART_END_HI 0x%08x\n", tmp);
 585         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_ERROR);
 586         seq_printf(m, "PCIE_TX_GART_ERROR 0x%08x\n", tmp);
 587         return 0;
 588 }
 589
 590 static struct drm_info_list rv370_pcie_gart_info_list[] = {
 591         {"rv370_pcie_gart_info", rv370_debugfs_pcie_gart_info, 0, NULL},
 592 };
 593 #endif
 594
 595 static int rv370_debugfs_pcie_gart_info_init(struct radeon_device *rdev)
 596 {
 597 #if defined(CONFIG_DEBUG_FS)
 598         return radeon_debugfs_add_files(rdev, rv370_pcie_gart_info_list, 1);
 599 #else
 600         return 0;
 601 #endif
 602 }
 603
 604 static int r300_packet0_check(struct radeon_cs_parser *p,
 605                 struct radeon_cs_packet *pkt,
 606                 unsigned idx, unsigned reg)
 607 {
 608         struct radeon_cs_reloc *reloc;
 609         struct r100_cs_track *track;
 610         volatile uint32_t *ib;
 611         uint32_t tmp, tile_flags = 0;
 612         unsigned i;
 613         int r;
 614         u32 idx_value;
 615
 616         ib = p->ib->ptr;
 617         track = (struct r100_cs_track *)p->track;
 618         idx_value = radeon_get_ib_value(p, idx);
 619
 620         switch(reg) {
 621         case AVIVO_D1MODE_VLINE_START_END:
 622         case RADEON_CRTC_GUI_TRIG_VLINE:
 623                 r = r100_cs_packet_parse_vline(p);
 624                 if (r) {
 625                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 626                                         idx, reg);
 627                         r100_cs_dump_packet(p, pkt);
 628                         return r;
 629                 }
 630                 break;
 631         case RADEON_DST_PITCH_OFFSET:
 632         case RADEON_SRC_PITCH_OFFSET:
 633                 r = r100_reloc_pitch_offset(p, pkt, idx, reg);
 634                 if (r)
 635                         return r;
 636                 break;
 637         case R300_RB3D_COLOROFFSET0:
 638         case R300_RB3D_COLOROFFSET1:
 639         case R300_RB3D_COLOROFFSET2:
 640         case R300_RB3D_COLOROFFSET3:
 641                 i = (reg - R300_RB3D_COLOROFFSET0) >> 2;
 642                 r = r100_cs_packet_next_reloc(p, &reloc);
 643                 if (r) {
 644                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 645                                         idx, reg);
 646                         r100_cs_dump_packet(p, pkt);
 647                         return r;
 648                 }
 649                 track->cb[i].robj = reloc->robj;
 650                 track->cb[i].offset = idx_value;
 651                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
 652                 break;
 653         case R300_ZB_DEPTHOFFSET:
 654                 r = r100_cs_packet_next_reloc(p, &reloc);
 655                 if (r) {
 656                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 657                                         idx, reg);
 658                         r100_cs_dump_packet(p, pkt);
 659                         return r;
 660                 }
 661                 track->zb.robj = reloc->robj;
 662                 track->zb.offset = idx_value;
 663                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
 664                 break;
 665         case R300_TX_OFFSET_0:
 666         case R300_TX_OFFSET_0+4:
 667         case R300_TX_OFFSET_0+8:
 668         case R300_TX_OFFSET_0+12:
 669         case R300_TX_OFFSET_0+16:
 670         case R300_TX_OFFSET_0+20:
 671         case R300_TX_OFFSET_0+24:
 672         case R300_TX_OFFSET_0+28:
 673         case R300_TX_OFFSET_0+32:
 674         case R300_TX_OFFSET_0+36:
 675         case R300_TX_OFFSET_0+40:
 676         case R300_TX_OFFSET_0+44:
 677         case R300_TX_OFFSET_0+48:
 678         case R300_TX_OFFSET_0+52:
 679         case R300_TX_OFFSET_0+56:
 680         case R300_TX_OFFSET_0+60:
 681                 i = (reg - R300_TX_OFFSET_0) >> 2;
 682                 r = r100_cs_packet_next_reloc(p, &reloc);
 683                 if (r) {
 684                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 685                                         idx, reg);
 686                         r100_cs_dump_packet(p, pkt);
 687                         return r;
 688                 }
 689
 690                 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
 691                         tile_flags |= R300_TXO_MACRO_TILE;
 692                 if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
 693                         tile_flags |= R300_TXO_MICRO_TILE;
 694
 695                 tmp = idx_value + ((u32)reloc->lobj.gpu_offset);
 696                 tmp |= tile_flags;
 697                 ib[idx] = tmp;
 698                 track->textures[i].robj = reloc->robj;
 699                 break;
 700         /* Tracked registers */
 701         case 0x2084:
 702                 /* VAP_VF_CNTL */
 703                 track->vap_vf_cntl = idx_value;
 704                 break;
 705         case 0x20B4:
 706                 /* VAP_VTX_SIZE */
 707                 track->vtx_size = idx_value & 0x7F;
 708                 break;
 709         case 0x2134:
 710                 /* VAP_VF_MAX_VTX_INDX */
 711                 track->max_indx = idx_value & 0x00FFFFFFUL;
 712                 break;
 713         case 0x43E4:
 714                 /* SC_SCISSOR1 */
 715                 track->maxy = ((idx_value >> 13) & 0x1FFF) + 1;
 716                 if (p->rdev->family < CHIP_RV515) {
 717                         track->maxy -= 1440;
 718                 }
 719                 break;
 720         case 0x4E00:
 721                 /* RB3D_CCTL */
 722                 track->num_cb = ((idx_value >> 5) & 0x3) + 1;
 723                 break;
 724         case 0x4E38:
 725         case 0x4E3C:
 726         case 0x4E40:
 727         case 0x4E44:
 728                 /* RB3D_COLORPITCH0 */
 729                 /* RB3D_COLORPITCH1 */
 730                 /* RB3D_COLORPITCH2 */
 731                 /* RB3D_COLORPITCH3 */
 732                 r = r100_cs_packet_next_reloc(p, &reloc);
 733                 if (r) {
 734                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 735                                   idx, reg);
 736                         r100_cs_dump_packet(p, pkt);
 737                         return r;
 738                 }
 739
 740                 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
 741                         tile_flags |= R300_COLOR_TILE_ENABLE;
 742                 if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
 743                         tile_flags |= R300_COLOR_MICROTILE_ENABLE;
 744
 745                 tmp = idx_value & ~(0x7 << 16);
 746                 tmp |= tile_flags;
 747                 ib[idx] = tmp;
 748
 749                 i = (reg - 0x4E38) >> 2;
 750                 track->cb[i].pitch = idx_value & 0x3FFE;
 751                 switch (((idx_value >> 21) & 0xF)) {
 752                 case 9:
 753                 case 11:
 754                 case 12:
 755                         track->cb[i].cpp = 1;
 756                         break;
 757                 case 3:
 758                 case 4:
 759                 case 13:
 760                 case 15:
 761                         track->cb[i].cpp = 2;
 762                         break;
 763                 case 6:
 764                         track->cb[i].cpp = 4;
 765                         break;
 766                 case 10:
 767                         track->cb[i].cpp = 8;
 768                         break;
 769                 case 7:
 770                         track->cb[i].cpp = 16;
 771                         break;
 772                 default:
 773                         DRM_ERROR("Invalid color buffer format (%d) !\n",
 774                                   ((idx_value >> 21) & 0xF));
 775                         return -EINVAL;
 776                 }
 777                 break;
 778         case 0x4F00:
 779                 /* ZB_CNTL */
 780                 if (idx_value & 2) {
 781                         track->z_enabled = true;
 782                 } else {
 783                         track->z_enabled = false;
 784                 }
 785                 break;
 786         case 0x4F10:
 787                 /* ZB_FORMAT */
 788                 switch ((idx_value & 0xF)) {
 789                 case 0:
 790                 case 1:
 791                         track->zb.cpp = 2;
 792                         break;
 793                 case 2:
 794                         track->zb.cpp = 4;
 795                         break;
 796                 default:
 797                         DRM_ERROR("Invalid z buffer format (%d) !\n",
 798                                   (idx_value & 0xF));
 799                         return -EINVAL;
 800                 }
 801                 break;
 802         case 0x4F24:
 803                 /* ZB_DEPTHPITCH */
 804                 r = r100_cs_packet_next_reloc(p, &reloc);
 805                 if (r) {
 806                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 807                                   idx, reg);
 808                         r100_cs_dump_packet(p, pkt);
 809                         return r;
 810                 }
 811
 812                 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
 813                         tile_flags |= R300_DEPTHMACROTILE_ENABLE;
 814                 if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
 815                         tile_flags |= R300_DEPTHMICROTILE_TILED;;
 816
 817                 tmp = idx_value & ~(0x7 << 16);
 818                 tmp |= tile_flags;
 819                 ib[idx] = tmp;
 820
 821                 track->zb.pitch = idx_value & 0x3FFC;
 822                 break;
 823         case 0x4104:
 824                 for (i = 0; i < 16; i++) {
 825                         bool enabled;
 826
 827                         enabled = !!(idx_value & (1 << i));
 828                         track->textures[i].enabled = enabled;
 829                 }
 830                 break;
 831         case 0x44C0:
 832         case 0x44C4:
 833         case 0x44C8:
 834         case 0x44CC:
 835         case 0x44D0:
 836         case 0x44D4:
 837         case 0x44D8:
 838         case 0x44DC:
 839         case 0x44E0:
 840         case 0x44E4:
 841         case 0x44E8:
 842         case 0x44EC:
 843         case 0x44F0:
 844         case 0x44F4:
 845         case 0x44F8:
 846         case 0x44FC:
 847                 /* TX_FORMAT1_[0-15] */
 848                 i = (reg - 0x44C0) >> 2;
 849                 tmp = (idx_value >> 25) & 0x3;
 850                 track->textures[i].tex_coord_type = tmp;
 851                 switch ((idx_value & 0x1F)) {
 852                 case R300_TX_FORMAT_X8:
 853                 case R300_TX_FORMAT_Y4X4:
 854                 case R300_TX_FORMAT_Z3Y3X2:
 855                         track->textures[i].cpp = 1;
 856                         break;
 857                 case R300_TX_FORMAT_X16:
 858                 case R300_TX_FORMAT_Y8X8:
 859                 case R300_TX_FORMAT_Z5Y6X5:
 860                 case R300_TX_FORMAT_Z6Y5X5:
 861                 case R300_TX_FORMAT_W4Z4Y4X4:
 862                 case R300_TX_FORMAT_W1Z5Y5X5:
 863                 case R300_TX_FORMAT_D3DMFT_CxV8U8:
 864                 case R300_TX_FORMAT_B8G8_B8G8:
 865                 case R300_TX_FORMAT_G8R8_G8B8:
 866                         track->textures[i].cpp = 2;
 867                         break;
 868                 case R300_TX_FORMAT_Y16X16:
 869                 case R300_TX_FORMAT_Z11Y11X10:
 870                 case R300_TX_FORMAT_Z10Y11X11:
 871                 case R300_TX_FORMAT_W8Z8Y8X8:
 872                 case R300_TX_FORMAT_W2Z10Y10X10:
 873                 case 0x17:
 874                 case R300_TX_FORMAT_FL_I32:
 875                 case 0x1e:
 876                         track->textures[i].cpp = 4;
 877                         break;
 878                 case R300_TX_FORMAT_W16Z16Y16X16:
 879                 case R300_TX_FORMAT_FL_R16G16B16A16:
 880                 case R300_TX_FORMAT_FL_I32A32:
 881                         track->textures[i].cpp = 8;
 882                         break;
 883                 case R300_TX_FORMAT_FL_R32G32B32A32:
 884                         track->textures[i].cpp = 16;
 885                         break;
 886                 case R300_TX_FORMAT_DXT1:
 887                         track->textures[i].cpp = 1;
 888                         track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
 889                         break;
 890                 case R300_TX_FORMAT_ATI2N:
 891                         if (p->rdev->family < CHIP_R420) {
 892                                 DRM_ERROR("Invalid texture format %u\n",
 893                                           (idx_value & 0x1F));
 894                                 return -EINVAL;
 895                         }
 896                         /* The same rules apply as for DXT3/5. */
 897                         /* Pass through. */
 898                 case R300_TX_FORMAT_DXT3:
 899                 case R300_TX_FORMAT_DXT5:
 900                         track->textures[i].cpp = 1;
 901                         track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
 902                         break;
 903                 default:
 904                         DRM_ERROR("Invalid texture format %u\n",
 905                                   (idx_value & 0x1F));
 906                         return -EINVAL;
 907                         break;
 908                 }
 909                 break;
 910         case 0x4400:
 911         case 0x4404:
 912         case 0x4408:
 913         case 0x440C:
 914         case 0x4410:
 915         case 0x4414:
 916         case 0x4418:
 917         case 0x441C:
 918         case 0x4420:
 919         case 0x4424:
 920         case 0x4428:
 921         case 0x442C:
 922         case 0x4430:
 923         case 0x4434:
 924         case 0x4438:
 925         case 0x443C:
 926                 /* TX_FILTER0_[0-15] */
 927                 i = (reg - 0x4400) >> 2;
 928                 tmp = idx_value & 0x7;
 929                 if (tmp == 2 || tmp == 4 || tmp == 6) {
 930                         track->textures[i].roundup_w = false;
 931                 }
 932                 tmp = (idx_value >> 3) & 0x7;
 933                 if (tmp == 2 || tmp == 4 || tmp == 6) {
 934                         track->textures[i].roundup_h = false;
 935                 }
 936                 break;
 937         case 0x4500:
 938         case 0x4504:
 939         case 0x4508:
 940         case 0x450C:
 941         case 0x4510:
 942         case 0x4514:
 943         case 0x4518:
 944         case 0x451C:
 945         case 0x4520:
 946         case 0x4524:
 947         case 0x4528:
 948         case 0x452C:
 949         case 0x4530:
 950         case 0x4534:
 951         case 0x4538:
 952         case 0x453C:
 953                 /* TX_FORMAT2_[0-15] */
 954                 i = (reg - 0x4500) >> 2;
 955                 tmp = idx_value & 0x3FFF;
 956                 track->textures[i].pitch = tmp + 1;
 957                 if (p->rdev->family >= CHIP_RV515) {
 958                         tmp = ((idx_value >> 15) & 1) << 11;
 959                         track->textures[i].width_11 = tmp;
 960                         tmp = ((idx_value >> 16) & 1) << 11;
 961                         track->textures[i].height_11 = tmp;
 962
 963                         /* ATI1N */
 964                         if (idx_value & (1 << 14)) {
 965                                 /* The same rules apply as for DXT1. */
 966                                 track->textures[i].compress_format =
 967                                         R100_TRACK_COMP_DXT1;
 968                         }
 969                 } else if (idx_value & (1 << 14)) {
 970                         DRM_ERROR("Forbidden bit TXFORMAT_MSB\n");
 971                         return -EINVAL;
 972                 }
 973                 break;
 974         case 0x4480:
 975         case 0x4484:
 976         case 0x4488:
 977         case 0x448C:
 978         case 0x4490:
 979         case 0x4494:
 980         case 0x4498:
 981         case 0x449C:
 982         case 0x44A0:
 983         case 0x44A4:
 984         case 0x44A8:
 985         case 0x44AC:
 986         case 0x44B0:
 987         case 0x44B4:
 988         case 0x44B8:
 989         case 0x44BC:
 990                 /* TX_FORMAT0_[0-15] */
 991                 i = (reg - 0x4480) >> 2;
 992                 tmp = idx_value & 0x7FF;
 993                 track->textures[i].width = tmp + 1;
 994                 tmp = (idx_value >> 11) & 0x7FF;
 995                 track->textures[i].height = tmp + 1;
 996                 tmp = (idx_value >> 26) & 0xF;
 997                 track->textures[i].num_levels = tmp;
 998                 tmp = idx_value & (1 << 31);
 999                 track->textures[i].use_pitch = !!tmp;
1000                 tmp = (idx_value >> 22) & 0xF;
1001                 track->textures[i].txdepth = tmp;
1002                 break;
1003         case R300_ZB_ZPASS_ADDR:
1004                 r = r100_cs_packet_next_reloc(p, &reloc);
1005                 if (r) {
1006                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1007                                         idx, reg);
1008                         r100_cs_dump_packet(p, pkt);
1009                         return r;
1010                 }
1011                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1012                 break;
1013         case 0x4e0c:
1014                 /* RB3D_COLOR_CHANNEL_MASK */
1015                 track->color_channel_mask = idx_value;
1016                 break;
1017         case 0x4d1c:
1018                 /* ZB_BW_CNTL */
1019                 track->fastfill = !!(idx_value & (1 << 2));
1020                 break;
1021         case 0x4e04:
1022                 /* RB3D_BLENDCNTL */
1023                 track->blend_read_enable = !!(idx_value & (1 << 2));
1024                 break;
1025         case 0x4be8:
1026                 /* valid register only on RV530 */
1027                 if (p->rdev->family == CHIP_RV530)
1028                         break;
1029                 /* fallthrough do not move */
1030         default:
1031                 printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1032                        reg, idx);
1033                 return -EINVAL;
1034         }
1035         return 0;
1036 }
1037
1038 static int r300_packet3_check(struct radeon_cs_parser *p,
1039                               struct radeon_cs_packet *pkt)
1040 {
1041         struct radeon_cs_reloc *reloc;
1042         struct r100_cs_track *track;
1043         volatile uint32_t *ib;
1044         unsigned idx;
1045         int r;
1046
1047         ib = p->ib->ptr;
1048         idx = pkt->idx + 1;
1049         track = (struct r100_cs_track *)p->track;
1050         switch(pkt->opcode) {
1051         case PACKET3_3D_LOAD_VBPNTR:
1052                 r = r100_packet3_load_vbpntr(p, pkt, idx);
1053                 if (r)
1054                         return r;
1055                 break;
1056         case PACKET3_INDX_BUFFER:
1057                 r = r100_cs_packet_next_reloc(p, &reloc);
1058                 if (r) {
1059                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1060                         r100_cs_dump_packet(p, pkt);
1061                         return r;
1062                 }
1063                 ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->lobj.gpu_offset);
1064                 r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1065                 if (r) {
1066                         return r;
1067                 }
1068                 break;
1069         /* Draw packet */
1070         case PACKET3_3D_DRAW_IMMD:
1071                 /* Number of dwords is vtx_size * (num_vertices - 1)
1072                  * PRIM_WALK must be equal to 3 vertex data in embedded
1073                  * in cmd stream */
1074                 if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1075                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1076                         return -EINVAL;
1077                 }
1078                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1079                 track->immd_dwords = pkt->count - 1;
1080                 r = r100_cs_track_check(p->rdev, track);
1081                 if (r) {
1082                         return r;
1083                 }
1084                 break;
1085         case PACKET3_3D_DRAW_IMMD_2:
1086                 /* Number of dwords is vtx_size * (num_vertices - 1)
1087                  * PRIM_WALK must be equal to 3 vertex data in embedded
1088                  * in cmd stream */
1089                 if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1090                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1091                         return -EINVAL;
1092                 }
1093                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1094                 track->immd_dwords = pkt->count;
1095                 r = r100_cs_track_check(p->rdev, track);
1096                 if (r) {
1097                         return r;
1098                 }
1099                 break;
1100         case PACKET3_3D_DRAW_VBUF:
1101                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1102                 r = r100_cs_track_check(p->rdev, track);
1103                 if (r) {
1104                         return r;
1105                 }
1106                 break;
1107         case PACKET3_3D_DRAW_VBUF_2:
1108                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1109                 r = r100_cs_track_check(p->rdev, track);
1110                 if (r) {
1111                         return r;
1112                 }
1113                 break;
1114         case PACKET3_3D_DRAW_INDX:
1115                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1116                 r = r100_cs_track_check(p->rdev, track);
1117                 if (r) {
1118                         return r;
1119                 }
1120                 break;
1121         case PACKET3_3D_DRAW_INDX_2:
1122                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1123                 r = r100_cs_track_check(p->rdev, track);
1124                 if (r) {
1125                         return r;
1126                 }
1127                 break;
1128         case PACKET3_NOP:
1129                 break;
1130         default:
1131                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1132                 return -EINVAL;
1133         }
1134         return 0;
1135 }
1136
1137 int r300_cs_parse(struct radeon_cs_parser *p)
1138 {
1139         struct radeon_cs_packet pkt;
1140         struct r100_cs_track *track;
1141         int r;
1142
1143         track = kzalloc(sizeof(*track), GFP_KERNEL);
1144         r100_cs_track_clear(p->rdev, track);
1145         p->track = track;
1146         do {
1147                 r = r100_cs_packet_parse(p, &pkt, p->idx);
1148                 if (r) {
1149                         return r;
1150                 }
1151                 p->idx += pkt.count + 2;
1152                 switch (pkt.type) {
1153                 case PACKET_TYPE0:
1154                         r = r100_cs_parse_packet0(p, &pkt,
1155                                                   p->rdev->config.r300.reg_safe_bm,
1156                                                   p->rdev->config.r300.reg_safe_bm_size,
1157                                                   &r300_packet0_check);
1158                         break;
1159                 case PACKET_TYPE2:
1160                         break;
1161                 case PACKET_TYPE3:
1162                         r = r300_packet3_check(p, &pkt);
1163                         break;
1164                 default:
1165                         DRM_ERROR("Unknown packet type %d !\n", pkt.type);
1166                         return -EINVAL;
1167                 }
1168                 if (r) {
1169                         return r;
1170                 }
1171         } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1172         return 0;
1173 }
1174
1175 void r300_set_reg_safe(struct radeon_device *rdev)
1176 {
1177         rdev->config.r300.reg_safe_bm = r300_reg_safe_bm;
1178         rdev->config.r300.reg_safe_bm_size = ARRAY_SIZE(r300_reg_safe_bm);
1179 }
1180
1181 void r300_mc_program(struct radeon_device *rdev)
1182 {
1183         struct r100_mc_save save;
1184         int r;
1185
1186         r = r100_debugfs_mc_info_init(rdev);
1187         if (r) {
1188                 dev_err(rdev->dev, "Failed to create r100_mc debugfs file.\n");
1189         }
1190
1191         /* Stops all mc clients */
1192         r100_mc_stop(rdev, &save);
1193         if (rdev->flags & RADEON_IS_AGP) {
1194                 WREG32(R_00014C_MC_AGP_LOCATION,
1195                         S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
1196                         S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
1197                 WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
1198                 WREG32(R_00015C_AGP_BASE_2,
1199                         upper_32_bits(rdev->mc.agp_base) & 0xff);
1200         } else {
1201                 WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
1202                 WREG32(R_000170_AGP_BASE, 0);
1203                 WREG32(R_00015C_AGP_BASE_2, 0);
1204         }
1205         /* Wait for mc idle */
1206         if (r300_mc_wait_for_idle(rdev))
1207                 DRM_INFO("Failed to wait MC idle before programming MC.\n");
1208         /* Program MC, should be a 32bits limited address space */
1209         WREG32(R_000148_MC_FB_LOCATION,
1210                 S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
1211                 S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
1212         r100_mc_resume(rdev, &save);
1213 }
1214
1215 void r300_clock_startup(struct radeon_device *rdev)
1216 {
1217         u32 tmp;
1218
1219         if (radeon_dynclks != -1 && radeon_dynclks)
1220                 radeon_legacy_set_clock_gating(rdev, 1);
1221         /* We need to force on some of the block */
1222         tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
1223         tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
1224         if ((rdev->family == CHIP_RV350) || (rdev->family == CHIP_RV380))
1225                 tmp |= S_00000D_FORCE_VAP(1);
1226         WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
1227 }
1228
1229 static int r300_startup(struct radeon_device *rdev)
1230 {
1231         int r;
1232
1233         /* set common regs */
1234         r100_set_common_regs(rdev);
1235         /* program mc */
1236         r300_mc_program(rdev);
1237         /* Resume clock */
1238         r300_clock_startup(rdev);
1239         /* Initialize GPU configuration (# pipes, ...) */
1240         r300_gpu_init(rdev);
1241         /* Initialize GART (initialize after TTM so we can allocate
1242          * memory through TTM but finalize after TTM) */
1243         if (rdev->flags & RADEON_IS_PCIE) {
1244                 r = rv370_pcie_gart_enable(rdev);
1245                 if (r)
1246                         return r;
1247         }
1248
1249         if (rdev->family == CHIP_R300 ||
1250             rdev->family == CHIP_R350 ||
1251             rdev->family == CHIP_RV350)
1252                 r100_enable_bm(rdev);
1253
1254         if (rdev->flags & RADEON_IS_PCI) {
1255                 r = r100_pci_gart_enable(rdev);
1256                 if (r)
1257                         return r;
1258         }
1259         /* Enable IRQ */
1260         r100_irq_set(rdev);
1261         /* 1M ring buffer */
1262         r = r100_cp_init(rdev, 1024 * 1024);
1263         if (r) {
1264                 dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
1265                 return r;
1266         }
1267         r = r100_wb_init(rdev);
1268         if (r)
1269                 dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
1270         r = r100_ib_init(rdev);
1271         if (r) {
1272                 dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
1273                 return r;
1274         }
1275         return 0;
1276 }
1277
1278 int r300_resume(struct radeon_device *rdev)
1279 {
1280         /* Make sur GART are not working */
1281         if (rdev->flags & RADEON_IS_PCIE)
1282                 rv370_pcie_gart_disable(rdev);
1283         if (rdev->flags & RADEON_IS_PCI)
1284                 r100_pci_gart_disable(rdev);
1285         /* Resume clock before doing reset */
1286         r300_clock_startup(rdev);
1287         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
1288         if (radeon_gpu_reset(rdev)) {
1289                 dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
1290                         RREG32(R_000E40_RBBM_STATUS),
1291                         RREG32(R_0007C0_CP_STAT));
1292         }
1293         /* post */
1294         radeon_combios_asic_init(rdev->ddev);
1295         /* Resume clock after posting */
1296         r300_clock_startup(rdev);
1297         /* Initialize surface registers */
1298         radeon_surface_init(rdev);
1299         return r300_startup(rdev);
1300 }
1301
1302 int r300_suspend(struct radeon_device *rdev)
1303 {
1304         r100_cp_disable(rdev);
1305         r100_wb_disable(rdev);
1306         r100_irq_disable(rdev);
1307         if (rdev->flags & RADEON_IS_PCIE)
1308                 rv370_pcie_gart_disable(rdev);
1309         if (rdev->flags & RADEON_IS_PCI)
1310                 r100_pci_gart_disable(rdev);
1311         return 0;
1312 }
1313
1314 void r300_fini(struct radeon_device *rdev)
1315 {
1316         r300_suspend(rdev);
1317         r100_cp_fini(rdev);
1318         r100_wb_fini(rdev);
1319         r100_ib_fini(rdev);
1320         radeon_gem_fini(rdev);
1321         if (rdev->flags & RADEON_IS_PCIE)
1322                 rv370_pcie_gart_fini(rdev);
1323         if (rdev->flags & RADEON_IS_PCI)
1324                 r100_pci_gart_fini(rdev);
1325         radeon_irq_kms_fini(rdev);
1326         radeon_fence_driver_fini(rdev);
1327         radeon_bo_fini(rdev);
1328         radeon_atombios_fini(rdev);
1329         kfree(rdev->bios);
1330         rdev->bios = NULL;
1331 }
1332
1333 int r300_init(struct radeon_device *rdev)
1334 {
1335         int r;
1336
1337         /* Disable VGA */
1338         r100_vga_render_disable(rdev);
1339         /* Initialize scratch registers */
1340         radeon_scratch_init(rdev);
1341         /* Initialize surface registers */
1342         radeon_surface_init(rdev);
1343         /* TODO: disable VGA need to use VGA request */
1344         /* BIOS*/
1345         if (!radeon_get_bios(rdev)) {
1346                 if (ASIC_IS_AVIVO(rdev))
1347                         return -EINVAL;
1348         }
1349         if (rdev->is_atom_bios) {
1350                 dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
1351                 return -EINVAL;
1352         } else {
1353                 r = radeon_combios_init(rdev);
1354                 if (r)
1355                         return r;
1356         }
1357         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
1358         if (radeon_gpu_reset(rdev)) {
1359                 dev_warn(rdev->dev,
1360                         "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
1361                         RREG32(R_000E40_RBBM_STATUS),
1362                         RREG32(R_0007C0_CP_STAT));
1363         }
1364         /* check if cards are posted or not */
1365         if (radeon_boot_test_post_card(rdev) == false)
1366                 return -EINVAL;
1367         /* Set asic errata */
1368         r300_errata(rdev);
1369         /* Initialize clocks */
1370         radeon_get_clock_info(rdev->ddev);
1371         /* Initialize power management */
1372         radeon_pm_init(rdev);
1373         /* Get vram informations */
1374         r300_vram_info(rdev);
1375         /* Initialize memory controller (also test AGP) */
1376         r = r420_mc_init(rdev);
1377         if (r)
1378                 return r;
1379         /* Fence driver */
1380         r = radeon_fence_driver_init(rdev);
1381         if (r)
1382                 return r;
1383         r = radeon_irq_kms_init(rdev);
1384         if (r)
1385                 return r;
1386         /* Memory manager */
1387         r = radeon_bo_init(rdev);
1388         if (r)
1389                 return r;
1390         if (rdev->flags & RADEON_IS_PCIE) {
1391                 r = rv370_pcie_gart_init(rdev);
1392                 if (r)
1393                         return r;
1394         }
1395         if (rdev->flags & RADEON_IS_PCI) {
1396                 r = r100_pci_gart_init(rdev);
1397                 if (r)
1398                         return r;
1399         }
1400         r300_set_reg_safe(rdev);
1401         rdev->accel_working = true;
1402         r = r300_startup(rdev);
1403         if (r) {
1404                 /* Somethings want wront with the accel init stop accel */
1405                 dev_err(rdev->dev, "Disabling GPU acceleration\n");
1406                 r300_suspend(rdev);
1407                 r100_cp_fini(rdev);
1408                 r100_wb_fini(rdev);
1409                 r100_ib_fini(rdev);
1410                 if (rdev->flags & RADEON_IS_PCIE)
1411                         rv370_pcie_gart_fini(rdev);
1412                 if (rdev->flags & RADEON_IS_PCI)
1413                         r100_pci_gart_fini(rdev);
1414                 radeon_irq_kms_fini(rdev);
1415                 rdev->accel_working = false;
1416         }
1417         return 0;
1418 }